Stress

The years since the financial crisis have seen The book includes several of the high profile

Stress Testing
stress testing emerge as a cornerstone of both system-wide stress-testing exercises that were
modern risk management and regulatory good essential elements of the public sector response
practice. to the financial crisis.

Testing
Acknowledged as an essential part of the risk Key chapters focus on:
manager’s armoury, increasingly regulators • Counterparty exposures
and supervisors see stress tests as key to • Validation and independent review
understanding the risks an institution faces and • Scenario design and model estimation
also, critically, those macroprudential risks with • Capital ratios
systemic impact. • CCAR stress testing

Stress testing is the common currency of the Contributors include Paolo Bisio,
regulation shaping the future of the financial Demelze Jurcevic and Mario Quagliariello Approaches, Methods

Edited by Akhtar Siddique and Iftekhar Hasan

system from Basel III and the Dodd–Frank Act, (European Banking Authority), David Lynch
to Interagency Stress Testing guidance in the US (Federal Reserve Board), Olivier de Bandt, and Applications
and CRD-IV in Europe. Nicolas Dumontaux, Vincent Martin and
Denys Médée (Autorité de Contrôle Prudentiel),
In Stress Testing: Approaches, Methods and Benjamin M. Tabak, Solange M. Guerra,
Applications, editors Akhtar Siddique (Office Sergio R. S. Souza and Rodrigo C. C. Miranda
of the Comptroller of Currency) and Iftekhar (Central Bank of Brazil) and David E. Palmer
Hasan (Fordham University) have assembled (Federal Reserve Board).
contributions from key figures directly involved
in the measurement and application of these new “With this book as a guide, both financial EDITED BY AKHTAR SIDDIQUE
stress-testing practices. regulators and financial practitioners can gain AND IFTEKHAR HASAN
insights that will help them do their jobs more
This new book sets out and analyses the effectively.”
techniques used, providing key insight into MARK LEVONIAN, Senior Deputy Comptroller,
supervisory perspectives and expectations. Office of the Comptroller of the Currency

PEFC Certified

This book has been

produced entirely from
sustainable papers that
are accredited as PEFC
compliant.
[Link]

Stress Testing v4_2.indd 1 21/05/2013 10:08

 Stress Testing: Approaches,
Methods and Applications

stress_testing.indd 1 20/05/2013 18:45

stress_testing.indd 2 20/05/2013 18:45
 Stress Testing: Approaches,
Methods and Applications

Edited by Akhtar Siddique and Iftekhar Hasan

stress_testing.indd 3 20/05/2013 18:45

 Published by Risk Books, a Division of Incisive Media Investments Ltd

Incisive Media
32–34 Broadwick Street
London W1A 2HG
Tel: +44(0) 20 7316 9000
E-mail: books@[Link]
Sites: [Link]
[Link]

© 2013 Incisive Media

ISBN 978-1-78272-008-9
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library

Publisher: Nick Carver

Editorial Development: Amy Jordan
Commissioning Editor: Sarah Hastings
Managing Editor: Lewis O’Sullivan
Designer: Lisa Ling

Copy-edited by Laurie Donaldson

Typeset by Tricolour Design

Printed and bound in the UK by Berforts Group

Conditions of sale
All rights reserved. No part of this publication may be reproduced in any material form whether
by photocopying or storing in any medium by electronic means whether or not transiently or
incidentally to some other use for this publication without the prior written consent of the copy-
right owner except in accordance with the provisions of the Copyright, Designs and
Patents Act 1988 or under the terms of a licence issued by the Copyright Licensing Agency
Limited of Saffron House, 6–10 Kirby Street, London EC1N 8TS, UK.

Warning: the doing of any unauthorised act in relation to this work may result in both civil
and criminal liability.

Every effort has been made to ensure the accuracy of the text at the time of publication, this
includes efforts to contact each author to ensure the accuracy of their details at publication is
correct. However, no responsibility for loss occasioned to any person acting or refraining from
acting as a result of the material contained in this publication will be accepted by the
copyright owner, the editor, the authors or Incisive Media.

Many of the product names contained in this publication are registered trade marks, and Risk
Books has made every effort to print them with the capitalisation and punctuation used by the
trademark owner. For reasons of textual clarity, it is not our house style to use symbols such
as TM, ®, etc. However, the absence of such symbols should not be taken to indicate absence
of trademark protection; anyone wishing to use product names in the public domain should first
clear such use with the product owner.

While best efforts have been intended for the preparation of this book, neither the publisher, the
editor nor any of the potentially implicitly affiliated organisations accept responsibility for any
errors, mistakes and or omissions it may provide or for any losses howsoever arising from or in
reliance upon its information, meanings and interpretations by any parties.

stress_testing.indd 4 20/05/2013 18:45

 Contents

Acknowledgements from the Editors vii

Foreword xi

About the Editors xv

About the Authors xvii

Introduction xxv
Akhtar Siddique; Iftekhar Hasan
Office of the Comptroller of Currency; Fordham University

1 Governance over Stress Testing 1

David E. Palmer
Federal Reserve Board

2 Stress Testing and Other Risk-Management Tools 15

Akhtar Siddique; Iftekhar Hasan
Office of the Comptroller of Currency; Fordham University

3 Stress Testing for Market Risk 25

Dilip K. Patro, Akhtar Siddique; Xian Sun
Office of the Comptroller of the Currency; Johns Hopkins University

4 The Evolution of Stress Testing Counterparty Exposures 37

David Lynch
Federal Reserve Board

5 Operational Risk: An Overview of Stress-Testing Methodologies 57

Brian Clark; Bakhodir Ergashev
Office of the Comptroller of Currency; Federal Reserve Bank of Richmond

6 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool 71

Paul Calem, Arden Hall
Federal Reserve Bank of Philadelphia

stress_testing.indd 5 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

7 Stress-Test Modelling for Loan Losses and Reserves 89

Michael Carhill, Jonathan Jones
Office of the Comptroller of the Currency

8 A Framework for Stress Testing Banks’ Corporate Credit Portfolio 109

Olivier de Bandt, Nicolas Dumontaux, Vincent Martin, Denys Médée
Autorité de Contrôle Prudentiel (French Prudential Supervision Authority)

9 EU-Wide Stress Test: The Experience of the EBA 127

Paolo Bisio, Demelza Jurcevic and Mario Quagliariello
European Banking Authority

10 Stress Testing Across International Exposures and Activities 143

Robert Scavotto, Robert H. Skinkle
Office of the Comptroller of the Currency

11 Liquidity Risk: The Case of the Brazilian Banking System 161

Benjamin M. Tabak, Solange M. Guerra, Sergio R. S. Souza,
Rodrigo C. C. Miranda
Banco Central do Brasil

12 Determining the Severity of Macroeconomic Stress Scenarios 193

Kapo Yuen
Federal Reserve Bank of New York

Index 225

vi

stress_testing.indd 6 20/05/2013 18:45

 Acknowledgements from the Editors

We would like to acknowledge advice from and discussions with

colleagues in various settings. Akhtar would like to acknowledge
discussions with colleagues from the OCC, Federal Reserve Banks
and the FDIC during examinations as well as in policy settings. En-
couragement from Mark Levonian and Mike Carhill are particu-
larly acknowledged. We would also like to thank Clayton Lamar
for research assistance and help.

vii

stress_testing.indd 7 20/05/2013 18:45

stress_testing.indd 8 20/05/2013 18:45
 To the people in our lives. For Akhtar, the children in his life, Ayaan, Aahil and his
nieces Babu and Putul but above all, his parents, wife: Deepa, sister: Moon, and
his in-laws. For Iftekhar, his parents, parents in-law, siblings, nephews, nieces,
and his spouse.

stress_testing.indd 9 20/05/2013 18:45

stress_testing.indd 10 20/05/2013 18:45
 Foreword

Stress testing is a tool. It is a tool used in many fields, particularly

where failure of some component of a complex system can have se-
riously adverse consequences and high costs: aircraft wings, high-
way bridges, coronary arteries.
And financial institutions.
Financial firms have long used stress testing as a tool in various
aspects of financial management. Firms use stress tests to assess the
impact of adverse events on various elements of the organisation,
or on the organisation as a whole, to gauge their ability to absorb
those stresses and carry on with core activities. Stress testing can
reveal vulnerabilities and point the way to actions that reduce those
vulnerabilities.
But disruptions in financial markets and continuing economic
weakness around the world have focused greater attention on
stress testing and its potential value. A widely acknowledged les-
son of the global financial crisis was that unusually stressful events
or combinations of events can and do occur, and that stress testing
can help assess the potential impact of those events and guide ap-
propriate preparation and response. Apart from its role as a tool for
management, stress testing fosters a culture of conscientious atten-
tion to risk. A common theme in post-crisis discussions with senior
management has been that stress testing helps create a robust dia-
logue around risks and risk management inside any firm.
“Stress testing” really describes a class of tools rather than a sin-
gle tool. Variations in stress-testing tools – the methods – roughly
match the variation in the specific ways stress testing is used – the
applications. Chapters of this book highlight distinct but related
aspects of stress testing in both methods and applications. Some
of the chapters emphasise risk management in the areas of credit
risk, market risk, operational risk, liquidity risk and others; other
chapters address the role of stress testing within the broader man-

xi

stress_testing.indd 11 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

agement environment of financial firms, as well as how to ensure

that stress testing remains a relevant and effective tool.
With such variety in methods and applications, it is perhaps no
surprise that overall approaches to stress testing vary as well. Re-
sponsibilities for the many different aspects of stress testing may be
centralised or distributed; specific management uses of the results
may make sense in certain environments but not in others. There is
no secret formula – many different approaches can succeed. Indeed,
approaches to stress testing should be expected to vary, because
stress tests add their greatest value when they are organised, ex-
ecuted and used in ways that take into account the unique charac-
teristics, operating environment and management style of the firms
or institutions using them.
Several chapters also address the increasingly prominent use of
stress testing in the supervision and regulation of financial firms.
This has been one of the most notable developments in interna-
tional financial supervision over the last few years, as the renewed
focus on stress testing within financial firms has been matched by a
simultaneous increase in attention from the supervisory communi-
ty. Financial authorities around the world have gravitated towards
stress testing as a powerful way to promote the health of the firms
they regulate. Institutional supervisors tend to focus on effective
use of stress testing for institutional risk management, consistent
with the ongoing use of stress testing as a risk-management tool.
Macroprudential regulators also have recognised the potential val-
ue of stress testing, with an emphasis on financial stability appli-
cations. Several highly visible system-wide stress-testing exercises
– some of which are discussed in this book – were key elements of
the public-sector response to the financial crisis; they were used to
assess the scope of the problems, evaluate the resiliency of large
financial firms and provide transparency and assurance to financial
market participants and others.
Effective stress testing relies on three major elements. First comes
the selection of meaningful stresses, either hypothetical or actual
scenarios, or more abstract approaches based on simulation of risk
factors. Second is the translation of those stresses into the impact
on the firm, or on the part of the firm subject to stress testing. And
third is the assessment of the stress-testing results, the evaluation

xii

stress_testing.indd 12 20/05/2013 18:45

 FOREWORD

of the impact of the stress. Relevant measures of impact vary de-

pending on the focus of a specific stress test – liquidity, earnings,
credit losses, capital, financial stability – but ideally point the way
towards meaningful response and action.
In scenario development, stress translation or impact assess-
ment, stress testing remains a tool in evolution. In particular, while
stress-testing practices may converge over time towards a some-
what more standard set of methods and applications, the end state
is unlikely ever to evolve to a single, uniform design. This is one of
those cases in which details must be tailored to the specific use; no
single stress-testing methodology will be appropriate for all appli-
cations at all times.
The combination of rapid change in stress-testing methods and
increased supervisory focus makes this volume especially relevant,
timely and useful. The book rewards readers with comprehensive
and thoughtful discussions of stress testing in its many forms,
covering a broad range of real-world applications and real-world
concerns, and drawing on the knowledge of an impressive set of
contributors. Its emphasis on both methods and applications, com-
bined with a consideration of multiple approaches, sets it apart
from other offerings with narrower focus.
Many of the contributors are quantitative experts from across the
global regulatory community. Readers anxious for insight into su-
pervisory perspectives and expectations can find much to consider
in these chapters. A window into the thinking of the supervisory
community is always valuable for regulated firms. But supervisory
authorship has additional value in this case. Individual financial
practitioners know a great deal about practices at their own firms,
but less about current approaches, methods and applications else-
where. In such a rapidly changing field, a narrow focus on practices
at any single firm can be misleading. Best practices diffuse across
the industry over time, but supervisors have a unique and explic-
itly horizontal perspective on stress testing, combining insights
from the many variations in stress-testing practices they are able to
observe across the regulated financial spectrum, and speeding the
recognition and adoption of best practices.
As readers consider the chapters of this book, they should not be
blind to the fact that, as valuable as the stress-testing tool may be,

xiii

stress_testing.indd 13 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

it is still only one of many tools that can and should be used. There
are other valid ways to measure or assess risk. There are also many
ways to control that risk. Any tool requires the hand of a skilled
user to make it effective. Ultimately, tools, even good ones, are no
substitute for sound judgement based on experience. Stress testing
must be intelligently applied, with expertise and insight.
With this book as a guide, both financial regulators and finan-
cial practitioners can gain insights that will help them do their jobs
more effectively. If broader and more thoughtful use of stress test-
ing reduces the impact of future stress events, then time spent with
this volume will be time well spent indeed.

Mark Levonian

Senior Deputy Comptroller, US Office of the

Comptroller of the Currency
March 2013

xiv

stress_testing.indd 14 20/05/2013 18:45

 About the Editors

Iftekhar Hasan is the E. Gerald Corrigan Chair in International

Business and Finance at the Schools of Business of Fordham Uni-
versity, New York and concurrently serves as a scientific advisor to
the Bank of Finland in Helsinki. His research focus is primarily in
the areas of financial intermediation, capital market and corporate
finance. He is also affiliated as a research associate at the Berkley
Center of the Stern School of New York University. Hasan has held
visiting positions at universities and organisations in several coun-
tries. He is the managing editor of the Journal of Financial Stability
and an associate editor in several other journals. Hasan is currently
the president of the Eurasia Business and Economics Society. A Ful-
bright Specialist Scholar, Hasan has over 250 publications in print,
including 12 books and edited volumes and over 160 peer-reviewed
journal articles in reputed finance, economics, management, opera-
tion research, accounting, and management information system
journals. He received his PhD from the University of Houston and
also received a “Doctor Honoris Causa” from the Romanian Ameri-
can University in Bucharest.

Akhtar Siddique is the deputy director of the Enterprise Risk

Analysis Division of the US Office of the Comptroller of the Cur-
rency (OCC), where he has worked from 2003. He manages a staff
of financial economists who provide technical assistance for ex-
aminations of national banks in Pillar II, operational risk, ALLL,
economic capital and enterprise-wide stress testing; participate in
policy initiatives; and conduct independent research. He directly
participates in examinations and intra- and interagency superviso-
ry and policy initiatives, particularly related to counterparty credit
risk, economic capital, valuation issues, ALLL, stress testing and
Pillar II. He has represented the OCC on the Risk Measurement
Group and Interaction of Market and Credit Risk Group of the Ba-

xv

stress_testing.indd 15 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

sel Committee on Banking Supervision. He has also represented the

OCC on the drafting of US interagency rules and guidance relating
to stress testing, Pillar II, etc. His research has spanned financial
econometrics, asset pricing, corporate finance and numerical meth-
ods. Siddique has authored numerous papers published in peer-
reviewed journals including the Journal of Finance, the Review of
Financial Studies, Management Science and the Journal of Account-
ing Research. He holds a PhD in Finance from Duke University and
taught finance at Georgetown University prior to joining the OCC.

xvi

stress_testing.indd 16 20/05/2013 18:45

 About the Authors

Paolo Bisio is a principal banks expert at the European Banking

Authority. He previously served in the Secretariat of the Committee
of the European Banking Supervisors and as an analyst in the Bank-
ing Supervision Department of Banca d’Italia, the Italian central
bank. He has been deeply involved in the supervisory review pro-
cess of the largest Italian banking groups, the validation of internal
models, Pillar II topics and stress testing. In 2010, he coordinated
the CEBS task forces on market and credit risks that developed the
methodology for the EU-wide stress tests, and played a pivotal role
in the different editions of the exercises. He holds a Master’s degree
in economics from the University of Rome.

Paul Calem is assistant vice president and chief of the Retail Risk
Analysis section in the Supervision, Regulation, and Credit Depart-
ment at the Federal Reserve Bank of Philadelphia. Previously, he
was a senior economist in the Division of Banking Supervision and
Regulation at the Federal Reserve Board. Calem moved into bank-
ing supervision after several years in the private sector at Freddie
Mac and the mortgage data and analytics company LoanPerfor-
mance, and a prior position in the Division of Research and Statis-
tics at the Federal Reserve Board. He has a PhD in economics from
Brown University and a BA in mathematics from Duke University.
His current responsibilities include implementing annual supervi-
sory stress testing of large banks’ retail portfolios; quantitative sup-
port of bank examinations; and policy analysis and research.

Mike Carhill has, since 2003, served as a director in the Risk Analysis
Division (RAD) of the US Office of the Comptroller of the Currency
(OCC). RAD employs quantitative modelling experts who specialise
in one or another of about a dozen lines of business to advise bank
examiners, bankers and policymakers on the state-of-the-art in risk-

xvii

stress_testing.indd 17 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

management-information systems, asset-pricing models and statisti-

cal modelling of fair-lending compliance. Carhill joined the OCC as
a staff economist in 1991, and was deputy director for market risk
modelling from 1995 to 2003. Before joining the OCC, he was a staff
economist with the Federal Home Loan Bank of Atlanta, engaged in
researching the role that interest-rate risk plays in thrift profitability
and as a consultant on risk-management issues. He received a PhD in
economics in May 1988 from Washington University.

Brian Clark is a senior financial economist in the Enterprise Risk

Analysis Division of the US Office of the Comptroller of the Cur-
rency. His current research focuses on areas of operational risk and
topics in corporate finance and banking. Clark received his PhD in
finance from Rensselaer Polytechnic Institute in 2010.

Olivier de Bandt is head of the Research Directorate at the ACP, the

French Prudential Supervision Authority, in charge of the supervi-
sion of banks and insurance companies. He was previously director
of macroeconomic forecasting at the Banque de France. He is also
an associate professor at the University of Paris Ouest. De Bandt
earned a PhD from the Economics Department of the University
of Chicago. His research interests include notably the economics of
banking and insurance, stress-testing methods and the analysis of
systemic risk.

Nicolas Dumontaux works for the French Prudential Supervision

Authority (ACP), as an onsite banking supervisor. He graduated from
l’École Nationale de la Statistique et de l’Administration Economique
and holds a Master in Economics degree from the Paris School of Eco-
nomics. Since joining the ACP in 2008, he has been involved in several
stress-testing exercises coordinated by the European Banking Author-
ity and has worked on several impact studies such as corporate credit
risk and countercyclical capital buffer. Dumontaux began his career as
a statistician at the French Ministry of Health.

Bakhodir Ergashev is a lead financial economist in the Supervision,

Regulation and Credit Department of the Federal Reserve Bank of
Richmond. He heads a team of financial economists responsible for

xviii

stress_testing.indd 18 20/05/2013 18:45

 about the authors

supervisory research and the supervision of risk models at large bank-

ing institutions. Ergashev’s research has focused on stress-testing risk
models, modelling operational risk in the presence of censored losses,
incorporating scenarios into operational risk models, and the role of
asset commonality in systemic risk. Ergashev joined the Federal Re-
serve Bank of Richmond in 2006 after earning his PhD in Economics
from Washington University. He also has a PhD in Mathematical Sta-
tistics from Steklov Institute of Mathematics, Moscow.

Solange M. Guerra works in the Financial Stability Group of the

Economic Research Department at the Central Bank of Brazil. She
is involved in developing indicators to assess the soundness of the
Brazilian banking system, such as default probabilities, governance,
concentration and efficiency indexes, and stress tests. Guerra holds
an MA in Mathematics and an MA and a PhD in Economics, both
from the Brasilia University.

Arden Hall is a special advisory at the Federal Reserve Bank of

Philadelphia, specialising in the modelling and risk management
of retail credit products. Before this, he held positions in risk man-
agement at the Federal Home Loan Bank of San Francisco, Bank of
America and Wells Fargo. He holds a MA in Statistics and a PhD in
Economics from the University of California, Berkeley.

Jonathan Jones is a senior financial economist in the Risk Analysis

Division at the US Office of the Comptroller of the Currency (OCC).
He received his PhD in Economics from the University of Colorado.
Before joining the OCC, he worked as a senior economist at the Of-
fice of Thrift Supervision, the Securities and Exchange Commission
and the Office of Tax Analysis at the US Treasury Department, and
taught at Vassar College and the Catholic University of America.
He has articles in journals such as Applied Financial Economics, the
Journal of Corporate Finance, the Journal of Empirical Finance, the
Journal of Finance and the Journal of Financial Research.

Demelza Jurcevic is a policy adviser at De Nederlandsche Bank,

where she is working in the Financial Risk Management Banking
department. She was on secondment for a year (2011–12) at the Eu-

xix

stress_testing.indd 19 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

ropean Banking Authority. Her main fields of expertise are finan-

cial risk management, stress testing, asset-quality review and the
supervisory review and evaluation process. In particular, Jurcevic
has been intimately involved in developing the CEBS guidelines
for stress testing, which were published in summer 2010, and in the
EU-wide stress-test exercise of CEBS/EBA and the IMF Financial
Sector Assessment Programme for the Netherlands. She is a mem-
ber of the EBA Stress Test Taskforce as well as the BCBS Pillar II
network. Jurcevic received a BA in Economics and an MSc in Finan-
cial Economics from Erasmus University of Rotterdam. She is also
a certified financial risk manager (GARP).

David Lynch is assistant director for quantitative risk management

in banking supervision and regulation at the Board of Governors
of the Federal Reserve. He joined the Board in 2005. His areas of
responsibility include oversight of models for market-risk capital
and counterparty-risk capital. Lynch is a representative of the Risk
Measurement Group of the Basel Committee on Banking Supervi-
sion. He has worked at the US Securities and Exchange Commis-
sion in broker-dealer finance. He holds a PhD in Economics from
the University of Maryland.

Vincent Martin works as a senior economist in the macroprudential

division of the Research Directorate of the French Prudential Super-
vision Authority (ACP), and is in charge of solvency and liquidity
stress-testing exercises for the French banking sector. He has been
involved in stress-testing exercises led by the European Banking
Authority and the IMF. Martin joined ACP in 2011, and previously
worked for Credit Agricole Corporate and Investment Bank (CACIB)
as a quantitative credit risk analyst. He graduated from l’École Na-
tionale de la Statistique et de l’Analyse de l’Information (ENSAI).

Denys Médée works for the French Prudential Supervision Authori-

ty (ACP), within the Research Directorate. He graduated from l’École
Polytechnique and holds a Master’s degree in Economics from Paris
University. He previously worked as a modelling expert for the Risk
Division of a French bank. Since joining ACP in 2009, he has been
involved in onsite supervision, participating in model-validation

xx

stress_testing.indd 20 20/05/2013 18:45

 about the authors

missions (such as credit risk, VaR, ALM). He has also participated

in coordinated top-down/bottom-up stress-testing exercises, and
has been appointed as an expert for several IMF technical assistance
missions, where he helped to implement and enhance stress-testing
frameworks for emerging countries’ banking supervisors.

Rodrigo C. C. Miranda works in the Financial Stability Group of the

Economic Research Department at the Central Bank of Brazil. He
joined the Economic Research Department in 2010 after four years
in the Central Bank’s IT department. He holds an MSc degree in
Computer Science from the University of Brasília.

David Palmer is a senior supervisory financial analyst in the Risk

Section of the Division of Banking Supervision and Regulation at
the Federal Reserve Board. He focuses on several primary topic
areas, including banks’ and supervisors’ stress-testing activities,
banks’ model-risk-management practices, banks’ internal process-
es to assess capital adequacy, and banks’ credit-risk capital models.
He engages in both policy-related projects and onsite examinations.
Palmer was a primary author of US supervisory guidance on stress-
testing for large banking organisations issued in May 2012. He was
also a key contributor to the development of the Federal Reserve’s
final rules to implement Dodd–Frank capital stress-testing require-
ments, issued in October 2012. In addition, he was a primary author
of the Federal Reserve’s supervisory guidance on model risk man-
agement issued in April 2011 jointly with the Office of the Comp-
troller of the Currency. He also serves in a leadership position in the
Federal Reserve for evaluating firms’ capital-adequacy processes
for Pillar II and for the CCAR. Palmer has a Bachelor’s degree from
Oberlin College and a Master’s from Georgetown University.

Dilip K. Patro is the deputy director of the Market Risk Analysis

Division of the US Office of the Comptroller of the Currency (OCC).
He was an assistant professor of finance before joining the OCC,
where he serves as a modelling and model-risk management ex-
pert, reviewing and evaluating models used for derivatives pricing,
market risk, counterparty credit risk, stress testing and regulatory
capital calculations at large national banks. He also represents the

xxi

stress_testing.indd 21 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

OCC in various policy initiatives of the Basel Committee, including

the Macroprudential Supervision Group. Patro is a graduate of IIT,
Delhi and a CFA charter holder, and he has a PhD from the Univer-
sity of Maryland at College Park.

Mario Quagliariello is the head of the Risk Analysis Unit at the Eu-
ropean Banking Authority. He previously served as a senior econ-
omist in the Regulation and Supervisory Policies Department of
Banca d’Italia, the Italian central bank. His interests concern macro-
prudential analysis and stress tests, Basel II and procyclicality, and
the economics of financial regulation. Quagliariello has published
several articles in Italian and international journals, including the
Journal of Banking and Finance, the Journal of Financial Services
Research and the and Risk. He edited Stress Testing the Banking
System: Methodologies and Applications, published by Cambridge
University Press, and co-edited Basel III and Beyond: A Guide to
Banking Regulation after the Crisis published by Risk Books. He
holds a PhD in Economics from the University of York in the UK.

Robert Scavotto is the lead international expert in the Economics

Department of the US Office of the Comptroller of the Currency
(OCC). He has over 15 years of experience in country risk analysis,
including developing stress-test applications. He has worked on
stress-testing policy, including guidance on implementing the Basel
stress-testing principles as well as writing the OCC’s Annual Stress
Test rule. He received his PhD and MA in Economics from Penn-
sylvania State University and a BA in Economics and Government/
International Relations from Clark University

Robert H. Skinkle is a senior international advisory in the Econom-

ics Department of the US Office of the Comptroller of the Currency
(OCC). He covers the Emerging Markets portfolio, with an emphasis
on Asia. Special projects include developing early-warning identifi-
cation systems, assessing country-risk-management techniques and
stress-testing bank portfolios. Prior to this, Skinkle specialised in
commercial real estate and systemic risk analysis of the US economy.
Before joining the OCC in 1991, he was the regional economist for
Wells Fargo Bank in San Francisco. His undergraduate degree in Eco-

xxii

stress_testing.indd 22 20/05/2013 18:45

 about the authors

nomics is from the University of California, Berkeley, and he has a

Master’s in Economics from the University of California, San Diego.

Sergio R. S. Souza works as a researcher in the Financial Stability

team in the Research Department at Central Bank of Brazil. He re-
ceived his BE degree in Mechanical Engineering at the University of
São Paulo and MSc and PhD degrees in Economics at the University
of Brasília. He joined the Central Bank in January 1998, having ini-
tially worked several years developing and implementing systems
forming the IT infrastructure for the International Reserves opera-
tions conducted by the bank. As a researcher, he has been working
with financial systems models, mainly network models and agent-
based models, with a focus on financial stability.

Xian Sun is an assistant professor of Finance at Carey Business

School at Johns Hopkins University. Before joining Carey Busi-
ness School, she worked for the US Office of the Comptroller of the
Currency (OCC) for three years as a senior financial economist in
the Market Risk Division. Her research interests include emerging
markets, political institutions and creditor rights. Her papers have
been published in prestigious journals such as the Journal of Finan-
cial and Quantitative Analysis, the Journal of Banking and Finance
and the Journal of International Money and Finance.

Benjamin M. Tabak holds a PhD in Economics from University of

Brasília. He works at the Central Bank of Brazil and is a professor at
Catholic University of Brasília. His research interests concern banking,
financial stability, law and economics, and behavioural modelling.

Kapo Yuen is a supervising bank examiner in the Financial Institu-

tion Supervision Group at the Federal Reserve Bank of New York,
where he leads a team of quantitative analysts and modellers to
support the supervision of systemically important financial insti-
tutions. His supervision responsibilities include evaluating the ap-
propriateness of firms’ stress-testing models. He has participated
in many of the Federal Reserve’s Comprehensive Capital Analysis
and Review (CCAR) exercises and was one of the authors of the
Basel Committee on Banking Supervision’s article “Principles for

xxiii

stress_testing.indd 23 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

sound stress testing practices and supervision”, published in May

2009. Before joining the Federal Reserve Bank of New York in 2006,
Yuen worked in several large financial institutions in the US and
overseas. He has more than 20 years of financial industry experi-
ence, mostly in credit risk management, predictive modelling, cred-
it-scoring and decision strategy. He holds an MS degree in Statistics
from the State University of New York at Stony Brook.

xxiv

stress_testing.indd 24 20/05/2013 18:45

 Introduction

Akhtar Siddique; Iftekhar Hasan

Office of the Comptroller of Currency; Fordham University

Banks across the globe faced significant challenges during the

2007–9 financial crisis. Declining capital ratios, equity prices, gov-
ernment takeovers and subsidies by the public sector have been
norms rather than special cases in many healthy “free-market”
economies. The crisis caused the confidence of investors, market
makers and lenders to plummet. Local and international regulators
attempted to restore confidence across the banking system in gen-
eral, but particularly with the goal of boosting lending and capital
structure. Among other initiatives, regulators are trying to imple-
ment stress testing of banks in order to lower or even prevent the
possibility of a future banking crisis. The exact details of these tests
vary according to specific economies and regulators, but all are in-
tended to examine how banks perform under an adverse or unex-
pected economic environment, with regard to both macroeconomic
uncertainty or bank-specific asset-liability and off-balance-sheet
activities. These tests are designed not only to understand the vul-
nerability of banks, but also to make sure that banks are prepared to
face adverse situations with the capital to mitigate potential losses
or non-performing assets.
This book considers the stress testing of financial institutions. In
the wake of the crisis, stress testing has received prominent media
coverage. Many research papers and several books have been pub-
lished on stress testing. The primary goal of all these initiatives, in-
cluding the efforts undertaken by this book, is to better understand

xxv

stress_testing.indd 25 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

effective stress testing. As per the subtitle, this book focuses on stress-
testing approaches and methods, as well as applications. This book
takes a comprehensive approach and covers a wide array of stress-
testing methods and scenarios. It makes a significant contribution, as
the authors are primarily drawn from regulatory bodies around the
world. The authors have considerable experience in the field, as they
have been studying and engaged in stress testing at financial institu-
tions for years, and many have been involved in the policy conse-
quences and development that stem from effective stress testing.
Regulators maintain an external view, albeit an external view that
is very often closely focused on the financial institutions. In compari-
son, “quants” at a bank can focus more on the implementation details.
Nevertheless, the regulatory view has the advantage of being exposed
to different approaches, methods and applications at different institu-
tions. Regulators can therefore weigh in on the relative strengths and
weaknesses of different banks. This comparative ability has become
particularly important, as smaller institutions that did not typically
engage in significant stress testing are now expected to do so. More-
over, stress testing has been evolving very rapidly, and, given that the
contributors to this book have been a part of this evolution, they can
provide insight into the development and situation of stress testing.
Stress testing has many approaches. For example, at many insti-
tutions, finance and treasury have had ownership of the budgeting
process. Hence, centralised stress testing could be housed in such
a central function. Other institutions have taken a more decentral-
ised approach, wherein the central function has only acted as an
aggregator. Effective stress testing does not need a given organisa-
tional form or approach. However, a given organisational form will
require certain aspects to ensure effectiveness. For example, in a
decentralised approach towards stress testing, consistency across
the organisation becomes important, and the institution will need
to find ways of ensuring that consistency.
An important element to stress testing is that the processes in
it need to be effective. That generally has required a build-out of
processes and functions. Stress testing includes not only the enter-
prise-wide stress tests that look at the bank’s projected capital with
scenarios based on macroeconomic variables, but also other types
of stress testing, such as portfolio and transaction-level stress tests.

xxvi

stress_testing.indd 26 20/05/2013 18:45

 introduction

Given that stress testing is generally conducted for distinct risks,

such as credit, counterparty credit, market, liquidity and so forth, the
contributions to the book start with governance and controls, and the
bulk of the book is devoted to separate chapters on the different risks.
In Chapter 1, David Palmer considers governance and controls
for the stress testing of financial institutions. Stress testing is highly
technical, makes many assumptions and entails much uncertainty,
and so appropriate governance and controls are crucial. Proper
stress-testing governance and controls not only confirm that stress
tests are conducted in a rigorous manner, but also help ensure that
stress tests and their outcomes are subject to oversight.
Differentiation between the roles of the senior management and
the board of directors is considered in this chapter. The role of an
internal audit should not be forgotten in concerns about gover-
nance. The role of validation is also discussed, both in terms of how
validation needs to examine data and inputs and how to ensure
the integrity of the process. What differentiates validation of stress-
testing from validation of models is the emphasis on the process,
as well as the need to validate the framework. Very usefully, the
chapter provides pointers on the policies, procedures, and docu-
mentation for stress testing. Given the recent implementation of
enterprise-wide stress testing at most financial institutions, such
policies, procedures and documentation are often less developed.
Chapter 2 examines the relationship between the various risk-
management tools used by financial institutions and stress testing,
particularly the value-at-risk-type, enterprise-wide risk measures.
Akhtar Siddique and Iftekhar Hasan discuss consistency between
the other risk metrics and stress tests. They then discuss how stress-
testing has begun to impact other risk measures.
It is important to note that stress testing has played two distinct
roles in the regulatory responses to the financial crisis. The first is
the stress tests conducted by the regulators, for example the SCAP
in the US and the EBA’s stress tests in Europe. Another role has
been to increase the amount of capital required of financial institu-
tions by incorporating stressed inputs into the capital calculations.
This has happened for market risk in Basel 2.5, as well as for coun-
terparty credit risk in Basel III. This chapter discusses both variants
of the interaction that stress testing has had with capital regulation.

xxvii

stress_testing.indd 27 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Dilip Patro, Akhtar Siddique and Sun Xian discuss stress testing for
market risk in Chapter 3. Given the very large number of positions as
well as the large number of inputs required, market risk poses challeng-
es for stress testing. The authors discuss the differences between stress
testing for market and credit risk (equivalently between trading book
and banking book) The difficulty in utilising macroeconomic scenarios
into market risk stress tests is another point they focus on. Horizon,
and the difficulty in incorporating the aftershock from market events
as well as the impact of hedging, are then discussed. The last section
focuses on revaluation of the values and/or computation of P&L under
the stressed scenarios. This discusses the challenges that banks face in
the revaluation as well as what to be concerned about in these areas.
David Lynch, in Chapter 4, discusses stress testing for counterpar-
ty credit risk. Counterparty credit risk (CCR) was one of the key av-
enues through which the financial crisis manifested itself. The mea-
surement and management of counterparty credit risk has become
greatly sophisticated over the past 20 or so years. However, stress
testing of counterparty credit risk has not kept pace with the evolu-
tion of counterparty credit risk measurement and management.
An important aspect of counterparty credit risk that needs to be rec-
ognised is that it may be approached by either a credit or market risk
perspective. The duality of CCR has led to the adoption of measures
that manage to capture some facet of CCR. In terms of credit risk: cur-
rent exposure, peak exposure and expected exposure are all impor-
tant measures. In terms of market risk: credit valuation adjustment’s
(CVA), valuation and the risk generated by changes in CVA are impor-
tant. Although treating CCR either way is a valid method for portfolio
management, failing to consider both opens an institution to risks from
the unconsidered perspective, and a single-view approach is better for
trading activities and risk-management discipline. Furthermore, there
is an unusual problem associated with CCR: wrong-way risk. Wrong-
way risk is a type of risk that occurs when exposure is correlated with a
counterparty’s credit quality, so that exposure increases as the counter-
party becomes likelier to default. This does not happen when exposure
is fixed, as with a loan, so the application of banking-book risk metrics
to CCR can cause considerable difficulties.
Given such a large amount of information, understanding and
interpreting stress tests at a portfolio and a counterparty level be-

xxviii

stress_testing.indd 28 20/05/2013 18:45

 introduction

comes difficult. The sheer variety of risk measures ensures that

stress testing for CCR is a complicated task, but financial institu-
tions are beginning to address these complications and are finally
conducting stress tests that go beyond simple tests of current expo-
sure. This is an exciting moment for CCR stress testing, as the best-
practice methodologies are only just now being developed.
In Chapter 5, Bakhodir Ergashev and Brian Clark provide back-
ground on the current state of operational risk modelling and con-
sider the challenges in the field today. Although data limitations
are one of these challenges, there are several areas where merely
increasing the quality and quantity of operational loss data is not
enough. Significant research is required to address these challenges
and to further the field.
Stress testing operational risk increases these challenges even
more. Most of these challenges are a consequence of the fact that
operational risk exposure is driven by infrequent large events. To
model the aggregate loss distribution, for instance, we need addi-
tional loss data. Stress testing insufficiently accurate static models
only amplifies the model’s risk, and modelling dependence with
the macroeconomic environment is even more challenging, due to
the lack of strong supportive evidence.
Despite the many challenges and limitations associated with op-
erational risk modelling, the authors have offered a variety of fruit-
ful possibilities for institutions to test operational risk exposure. We
are currently in a nascent moment for the development of method-
ology, and, although some methods may appear more useful than
others, additional research will be required to determine the utility
of any particular one. The authors currently recommend that banks
employ an array of models, thereby both providing a variety of per-
spectives and contributing to possible future research.
Paul Calem and Arden Hall offer an overview of the current
regulatory stress-testing environment in Chapter 6. They provide
a particularly useful perspective on the limitations of stress testing.
The authors focus specifically on the stress testing of balance-sheet
loan losses, but much of their work is applicable to bank capital
stress tests generally. Given that a balance-sheet loan-loss stress test
is based on predictions of borrower repayment under extreme con-
ditions that are generally outside the scope of historical precedents,

xxix

stress_testing.indd 29 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

such testing presents major modelling challenges. Yet, despite such

challenges, balance-sheet stress testing can offer a useful view of a
financial institution’s sensitivity to adverse shocks, and Calem and
Hall propose a framework for such testing. A careful and controlled
balance-sheet loan-loss stress test can pinpoint sources of risk, or-
der institutions according to risk, indicate the quality of a bank’s
risk position over time and provide benchmarks for understanding
an institution’s risk-management processes and capital adequacy.
Stress testing can also evaluate the risk sensitivity or risk composi-
tion of an institution’s loan portfolio.
In Chapter 7, Michael Carhill and Jonathan Jones examine stress-
test modelling for loan losses and reserves. Their contribution is
driven by their experience of looking at the stress-testing practices of
institutions. They observed the decision-making processes at large fi-
nancial institutions as executives decided that models capable of esti-
mating credit losses conditional on economic scenarios were required
for enterprise-wide capital planning and stress testing. A number of
supervisory and regulatory developments account for this interest
in deploying economic factors in stress testing. Such developments
include: the Advanced Internal Ratings Based (AIRB) approach of
Basel II; the recommendations of the Basel Committee on Banking
Supervision; the Federal Reserve’s Supervisory Capital Analysis
Program (SCAP) exercise and the following Comprehensive Capital
Analysis and Review (CCAR) bank stress tests; and the Dodd–Frank
Wall Street Reform and Consumer Protection Act of 2010.
Carhill and Jones present their first-hand observations in this
chapter, including their evaluation of the strengths and weakness-
es of the approaches they witnessed. They also identify potential
challenges and pitfalls that need to be addressed when developing
macroeconomic-based credit risk stress-testing models. However,
given that this field is rapidly evolving and no best practices have
yet been established, the reliability of these models is still question-
able. Nonetheless, the financial crisis has made clear the need for
banks to include economic and market factors in credit risk models
in order to produce accurate stress loan-loss estimates.
Olivier de Bandt, Nicolas Dumontaux, Vincent Martin, and De-
nys Médée examine in Chapter 8 stress testing for credit risk and
focus on risks arising from corporate loans and other credit expo-

xxx

stress_testing.indd 30 20/05/2013 18:45

 introduction

sures. Corporate credit risk – known also as wholesale credit risk –

is highly important when stress testing global firms. Credit risk is a
huge source of risk for banks and is very relevant to an institution’s
financial solvency. The crisis has underscored the need for stress-
testing banks’ portfolios, as institutions have incurred significant
losses from structured US subprime-related assets. This chapter
develops the foundations of a Basel II-type modelling effort to per-
form credit stress-test scenarios through credit-migration matrices
(or transition matrices), which have already been implemented in
France and are currently used for top-down stress-test exercises.
Credit risk stems from either actual defaults or migrations of credit
ratings, and several model approaches are available to quantify this
type of risk. These models may be structural, in which a firm’s value
and capital structure are modelled, or reduced, in which credit events
are exogenous to a firm. The authors use a reduced form in which
events are triggered by macroeconomic shock. The authors also focus
on credit migrations. The model rests on the premise that the evolution
of rating transitions can be linked to a synthetic credit indicator.
In Chapter 9, Paolo Bisio, Demelza Jurcevic, and Mario Quagliari-
ello examine and evaluate the 2011 EU-wide stress tests conducted
by the European Banking Authority (EBA) among 91 banks. The ob-
jective of the stress tests was to evaluate the EU banking system and
the solvency of institutions under scenarios imposed by supervisors.
After describing the sample selection, macro scenario and methodol-
ogy, the authors note that the EBA is satisfied with the progress made
by banks in fulfilling the recommendations to ensure appropriate
mitigating actions based on the exercise’s results.
Needless to say, the overall assessment was not without its criticisms
– the EBA received negative feedback on some of its assumptions – but
the 2011 exercise was a partial success when assessed against the bench-
marks of analytical rigour, communication and resulting actions. The
development of a stress-test methodology and its early publication was
received positively. The presence of an organised quality-assurance
and robust peer review was perceived as helpful as regarding interpre-
tation of the methodology and the development of micro parameters
from the macro scenario. The coordinated and extensive disclosure of
results was seen as a significant victory for proponents of transparency
and was welcomed by market participants and the public.

xxxi

stress_testing.indd 31 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Robert Scavotto and Robert Skinkle argue in Chapter 10 that, dur-

ing stress testing of internationally active financial institutions, in-
dividual-country characteristics can produce great variation in the
stress-test results. Therefore, to perform reliable stress testing we
must first approach the estimation of regional or global regressions
with caution. Many challenges emerge in cross-country testing, in-
cluding very basic factors, such as the development of datasets and
the acquisition of an understanding of qualitative factors that may
be relevant to certain countries. Drawing upon their own experi-
ences in cross-county stress testing, the authors determine that a
best-practice methodology entails pursuing an individual-country
approach with specific stratifications. The authors’ examination of
the Korean consumer market as a case study is a well-reasoned ap-
proach and provides a novel methodology for any researchers in-
terested in performing cross-country systemic stress tests.
In Chapter 11, Benjamin M. Tabak, Solange M. Guerra, Sergio Ru-
bens Stancato de Souza and Rodrigo Cesar de Castro Miranda discuss
the effects of the financial crisis on the Brazilian banking system and
liquidity risk. Liquidity risks have risen in the wake of the financial cri-
sis, and there is a need for measures to restore confidence and increase
liquidity to enable financial institutions to handle additional risks. Li-
quidity crises are less frequent than other types, but their impacts can
be very significant. Such low-frequency, high-impact events are trou-
bling, insofar as they are not often easily planned for, yet can have as
comparably devastating results as more common crises. As expected,
liquidity stress tests are not currently as well developed as credit and
market risk stress tests, although there is now increased interest in the
threats posed by liquidity risks as a result of the financial crisis
Most central banks do not publish the results from their liquidity
stress tests, a fact that is indicative of liquidity-modelling complex-
ity and lack of data. The Central Bank of Brazil, however, has pub-
lished its results since 2009, and provides a useful case study for
discussing liquidity stress testing in general. The authors examine
the liquidity stress-testing approach that is under use in the Central
Bank of Brazil and their methodology and results provide a crucial
foray into a field that is in dire need of quality research.
Kapo Yuen discusses in Chapter 12 the question of the severity of
supervisory adverse scenarios and provides a methodology to com-

xxxii

stress_testing.indd 32 20/05/2013 18:45

 introduction

pare the severity of different adverse macroeconomic scenarios. Since

2009, the Federal Reserve System has conducted an annual stress test
on the US banking system, called the Comprehensive Capital Anal-
ysis and Review (CCAR). During this annual exercise, the Federal
Reserve Board provides a supervisory macro scenario for stress test-
ing, but also requires that each financial institution submit an indi-
vidualised scenario that is particularly relevant to that institution’s
specific features. The instructions for that individualised scenario are
somewhat abstract: they indicate that the scenario must reflect “a se-
verely adverse economic and financial market environment”, but the
question of what constitutes “appropriate severity” is left unclear.
Yuen poses several questions in attempting to determine the severity
of a stress scenario: How can we measure severity? How severe is
an individualised scenario versus the supervisory macro scenario?
Are the individualised scenarios credibly severe? The financial crisis
may be a useful starting point to measure severity, but Yuen won-
ders whether banks have actually changed; severe stress scenarios
must be developed that will result in estimates of losses that demon-
strate vulnerabilities. In posing these questions, Yuen demonstrates
a methodology that can be used to assess the severity of a particular
scenario and offers some possible alternate methodologies, as well.
The editors hope readers will learn about both the practical ele-
ments of stress testing and stress-testing principles from this book.
More practically, they will also learn the pitfalls to avoid while con-
ducting stress testing. The authors have been involved in matters
relating to stress testing for many years and have also seen many
other good books on stress testing. However, what those books ap-
peared to lack was the discussion of the comparative merits and de-
merits of various approaches. The regulatory community (at least
those members engaged in active conduct or examinations of stress
testing) is uniquely suited to this role given its view of stress-testing
practices across different institutions. This insight motivated the
editors to bring these chapters together.

The views expressed in this chapter are those of the authors alone and do not
necessarily represent those of the Comptroller of the Currency or Bank of Finland.

xxxiii

stress_testing.indd 33 20/05/2013 18:45

stress_testing.indd 34 20/05/2013 18:45
 1
Governance over Stress Testing

David E. Palmer
Federal Reserve Board

Governance and controls are a very important aspect of stress test-

ing, yet are sometimes overlooked or given insufficient attention by
financial institutions.1 Proper governance and controls over stress-
testing not only confirm that stress tests are conducted in a rigorous
manner, but also help ensure that stress tests and their outcomes are
subject to an appropriately critical eye. Governance and controls
are particularly needed in the area of stress testing given the highly
technical nature of many stress-testing activities, the generally large
number of assumptions in stress-testing exercises and the inher-
ent uncertainty in estimating the nature, likelihood and impact of
stressful events and conditions.
While the exact form of governance and controls over stress-test-
ing activities can and should vary across countries and financial in-
stitutions, there are some general principles, expectations and rec-
ommendations that financial institutions can follow. The manner in
which the principles, expectations and recommendations outlined
in this chapter are applied at any given financial institution should
involve a “tailored” approach that is specifically tied to the size,
complexity, risk profile, culture and individual characteristics of
that institution.
This chapter discusses key elements of effective governance over
stress testing, including: governance structure; policies, procedures
and documentation; validation and independent review; and inter-
nal audit. It also discusses other aspects of stress-testing activities

stress_testing.indd 1 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

that should be considered and reviewed as part of the stress testing

governance process.

GOVERNANCE STRUCTURE
Governance structure is one of the primary elements for sound
governance over stress testing. While institutions may have differ-
ent structures based on the legal, regulatory or cultural norms in
their countries, it is generally expected that every institution has
separation of duties between a board of directors and senior man-
agement. This separation of duties is equally important for stress-
testing activities, as it helps ensure there is proper oversight and ac-
tion taken on an ongoing basis. The board and senior management
should share some responsibilities – albeit to varying degrees of
detail – but also have distinct responsibilities in other cases. Togeth-
er, an institution’s board and senior management should establish
comprehensive, integrated and effective stress testing that fits into
the broader risk management of the institution.

Board of directors
In general, the board of directors has ultimate oversight responsi-
bility and accountability for the entire organisation. It should be
responsible for key strategies and decisions, define the culture of
the organisation and set the “tone at the top”. This applies to stress-
testing as well, as the board is ultimately responsible for the insti-
tution’s stress-testing activities, even if the board is not intimate-
ly involved in the details. Board members should be sufficiently
knowledgeable about stress-testing activities to ask informed ques-
tions, even if they are not experts in the technical details. The board
should actively evaluate and discuss information received from
senior management about stress testing, ensuring that the stress-
testing activities are in line with the institution’s risk appetite, over-
all strategies and business plans, and contingency plans – directing
changes where appropriate.2 Board members should also ensure
they review that information with an appropriately critical eye,
challenging key assumptions, ensuring that there is sufficient infor-
mation with appropriate detail and supplementing the information
with their own views and perspectives.
Stress-testing results should be used, along with other informa-

stress_testing.indd 2 20/05/2013 18:45

 Governance over Stress Testing

tion, to inform the board about alignment of the institution’s risk

profile with the board’s chosen risk appetite, as well as inform op-
erating and strategic decisions. Stress-testing results should be con-
sidered directly for decisions relating to capital and liquidity ade-
quacy, including capital contingency plans and contingency funding
plans. While stress-testing exercises can be very helpful in provid-
ing a forward-looking assessment of the potential impact of ad-
verse outcomes, board members should ensure they use the results
of the stress tests with an appropriate degree of scepticism, given
the assumptions, limitations and uncertainties inherent in any type
of stress testing. In general, the board should not rely on just one
stress-test exercise in making key decisions, but should aim to have
it supplemented with other tests and other quantitative and qualita-
tive information. The board should be able to take action based on its
review of stress-test results and accompanying information, which
could include changing capital levels, bolstering liquidity, reducing
risk, adjusting exposures, altering strategies or withdrawing from
certain activities. In many cases, stress-testing activities can serve as
a useful “early-warning” mechanism for the board, especially during
benign times (ie, non-stress periods), and thus can be useful in guid-
ing the overall direction and strategy for the institution.

Senior management
Senior management has the responsibility of ensuring that stress-
testing activities authorised by the board are implemented in a sat-
isfactory manner, and is accountable to the board for the effective-
ness of those activities. That is, senior management should execute
on the overall stress-testing strategy determined by the board. Se-
nior management duties should include establishing adequate poli-
cies and procedures and ensuring compliance with them, allocating
appropriate resources and assigning competent staff, overseeing
stress-test development and implementation, evaluating stress-test
results, reviewing any findings related to the functioning of stress-
test processes and taking prompt remedial action where necessary.
In addition, whether directly or through relevant committees,
senior management should be responsible for regularly reporting
to the board on stress-testing developments (including the process
to design tests and develop scenarios) and on stress-testing results

stress_testing.indd 3 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

(including those from individual tests, where material), as well

as on compliance with stress-testing policies. Senior management
should ensure there is appropriate buy-in at different levels of the
institution, and that stress-testing activities are appropriately coor-
dinated. Such coordination does not have to mean that all stress-
testing exercises are built on the same assumptions or use the same
information. Indeed, it can be very useful to conduct different types
of stress tests to achieve a wide perspective. But senior manage-
ment should be mindful of potential inconsistencies, contradictions
or gaps among its stress tests and assess what actions should be
taken as a result. At a minimum, this means that assumptions are
transparent and that results are not used in a contradictory manner.
Senior management, in consultation with the board, should en-
sure that stress-testing activities include a sufficient range of stress-
testing activities applied at the appropriate levels of the institution
(ie, not just one single stress test). Another key task is to ensure
that stress-test results are appropriately aggregated, particularly for
enterprise-wide tests. Senior management should maintain an in-
ternal summary of test results to document at a high level the range
of its stress-testing activities and outcomes, as well as proposed
follow-up actions. Sound governance at this level also includes
using stress testing to consider the effectiveness of an institution’s
risk-mitigation techniques for various risk types over their respec-
tive time horizons, such as to explore what could occur if expected
mitigation techniques break down during stressful periods.
Stress-test results should inform management’s analysis and
decision-making related to business strategies, limits, capital and
liquidity, risk profile and other aspects of risk management, con-
sistent with the institution’s established risk appetite. Wherever
possible, benchmarking or other comparative analysis should be
used to evaluate the stress-testing results relative to other tools and
measures – both internal and external to the institution – to provide
proper context and a check on results. Just as at the board level,
senior management should challenge the results and workings of
stress-testing exercises. In fact, senior management should be much
more well versed in the details of stress testing and be able to drill
down in many cases to discuss technical issues and challenge re-
sults on a granular level.

stress_testing.indd 4 20/05/2013 18:45

 Governance over Stress Testing

Senior management can and should use stress testing to supple-

ment other information it develops and provides to the board, such
as other risk metrics or measures of capital and liquidity adequa-
cy. When reporting stress-testing information to the board, senior
management should be able to explain the key elements of stress-
testing activities, including assumptions, limitations and uncertain-
ties. Reports from senior management to the board should be clear,
comprehensive and current, providing a good balance of succinct-
ness and detail. Those reports should include information about
the extent to which stress test models are appropriately governed,
including the extent to which they have been subject to validation
or other type of independent review (see below). Senior manage-
ment, as part of its overall efforts to ensure proper governance and
controls, is also responsible for ensuring that staff involved in stress
testing operate under the proper incentives. Finally, senior man-
agement should ensure that there is a regular assessment of stress-
testing activities across the institution by an independent, unbiased
party (such as internal audit – see below).
Senior management should ensure that stress-testing activities
are updated in light of new risks, better understanding of the insti-
tution’s exposures and activities, new stress-testing techniques, up-
dated data sources and any changes in its operating structure and
its internal and external environment. An institution’s stress-testing
development should be iterative, with ongoing adjustments and
refinements to better calibrate the tests to provide current and rel-
evant information. In addition, management should review stress-
testing activities on a regular basis to confirm the general appropri-
ateness of, among other things, the validity of the assumptions, the
severity of tests, the robustness of the estimates, the performance of
any underlying models and the stability and reasonableness of the
results. In addition to conducting formal, routine stress tests, man-
agement should ensure the institution has the flexibility to conduct
new or ad hoc stress tests in a timely manner to address rapidly
emerging risks and vulnerabilities.

POLICIES, PROCEDURES AND DOCUMENTATION

Having clear and comprehensive policies, procedures and docu-
mentation is integral to sound stress-testing governance. These

stress_testing.indd 5 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

areas provide the important codification of an institution’s prac-

tices and allow the institution as a whole to follow the same gen-
eral principles and standards for its stress-testing activities. Thus,
in order to promote a sound control environment and allow for
consistency and repeatability in stress-testing activities across the
entity, the institution should have written policies that direct and
govern the implementation of stress-testing activities in a clear and
comprehensive manner. It is generally expected that these policies
would be approved and annually reassessed by the board. Stress-
testing policies, along with procedures to implement them, should:

o describe the overall purpose of stress-testing activities;

o articulate consistent and sufficiently rigorous stress-testing prac-
tices across the entire institution;
o indicate stress-testing roles and responsibilities, including con-
trols over external resources used for any part of stress testing
(such as vendors and data providers);
o describe the frequency and priority with which stress-testing ac-
tivities should be conducted;
o outline the process for choosing appropriately stressful condi-
tions for tests, including the manner in which scenarios are de-
signed and selected;
o include information about validation and independent review of
stress tests;
o provide transparency to third parties for their understanding of
an institution’s stress-testing activities;
o indicate how stress-test results are used and by whom, and out-
line instances in which remedial actions should be taken; and
o be reviewed and updated as necessary to ensure that stress-
testing practices remain appropriate and keep up to date with
changes in market conditions, the institution’s products and
strategies, its risks, exposures and activities, its established risk
appetite and industry stress-testing practices.

In addition to having clear and comprehensive policies and pro-

cedures, an institution should ensure that its stress tests are doc-
umented appropriately, including a description of the types of
stress tests and methodologies used, test results, key assumptions,

stress_testing.indd 6 20/05/2013 18:45

 Governance over Stress Testing

limitations and uncertainties, and suggested actions. Among other

things, documentation:

o allows management to track results and analyse differences over

time, including changes due to methodologies and assumptions
as well as changes due to market conditions or other external
factors;
o is a vital aspect of stress-testing governance as it allows third
parties to evaluate stress tests and their components, including
for validation and internal audit review;
o provides for continuity of operations, makes compliance with
policy transparent and helps track recommendations, responses
and exceptions; and
o is a useful tool for stress-test developers, as it forces them to think
clearly about their stress tests, categorise the components of the
tests and describe choices made and assumptions used.

Documenting stress tests takes time and effort, so institutions should

therefore provide incentives to produce effective and complete
documentation. Developers of stress tests should be given explicit
responsibility during development for thorough documentation,
which should be kept up to date as stress testing and application
environment change. In addition, the institution should ensure that
other participants document their work related to stress-testing ac-
tivities, including validators, reviewers and senior management.
For cases in which a bank uses stress tests from a vendor or other
third party, it should ensure that appropriate documentation of the
third-party approach is available so that the stress test can be ap-
propriately understood, validated, reviewed, approved and used.

VALIDATION AND INDEPENDENT REVIEW

Another key element of governance over stress testing is validation
and independent review. Stress-testing governance should incor-
porate validation or other type of independent review to ensure the
integrity of stress-testing processes and results. Such unbiased, criti-
cal review of stress-testing activities gives additional assurance that
the stress tests are functioning as intended. In general, validation
and independent review of stress-testing activities should be con-

stress_testing.indd 7 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

ducted on an ongoing basis, not just as a single event. In addition,

validation and review work for stress testing should be integrated
with an institution’s general approach to validation and indepen-
dent review of its quantitative estimation tools – although stress
tests may need to be validated and reviewed in a particular man-
ner. Specifically, because stress tests by definition aim to estimate
the potential impact of rare events and circumstances, conducting
more traditional outcomes analysis used in a more data-rich envi-
ronment may not be possible. For instance, statistical backtesting of
stress-test estimates against realised outcomes may not be feasible.
To address challenges associated with validating stress tests,
some institutions may try to test their models using data from non-
stress periods, ie, during “good times” or in a “baseline” setting.
Such testing can be beneficial to determine whether the stress test
generally functions as a predictive model under those conditions.
If the stress test does not perform well in a more data-rich envi-
ronment, that would certainly raise questions about its usefulness.
However, while “baseline” outcomes showing good test perfor-
mance can provide some additional confidence in the stress test,
those outcomes should not be interpreted as sufficient for the desig-
nated task of estimating stress outcomes. For instance, markets and
market actors can behave quite differently in stress environments,
and assumed interactions among variables can change markedly
(such as higher incidence of nonlinearities). Thus, the model used
in a baseline situation may actually require a different specification
to properly estimate stress outcomes (or an entirely new model may
be needed for stress periods). There can be additional challenges
when upgrades or enhancements are made to stress tests, because
it may not be immediately clear that the upgraded or enhanced
model actually performs better. Here, too, assessing the baseline
outcomes can provide some assurance about such changes, but can-
not offer full confirmation. In sum, even with rigorous quantita-
tive analytics, there can remain very real limitations in the extent
to which stress tests can be formally validated or otherwise fully
assessed in terms of quantitative performance.
As an additional response to these validation issues, given the
limitations of relying on outcomes analysis, an institution may
need to rely on other aspects of validation and independent review

stress_testing.indd 8 20/05/2013 18:45

 Governance over Stress Testing

of stress tests – such as a greater emphasis on conceptual sound-

ness of the stress test, additional sensitivity testing, and simulation
techniques. Or an institution may choose to create holdout sam-
ple portfolios and run them through its stress-test model. Bench-
marking to internal or external models, tools or results can also be
beneficial, but institutions should be careful that the benchmarks
appropriately fit the institution’s risks, exposures and activities. Fi-
nally, expert-based judgement should be applied to ensure that test
results are intuitive and logical, and to add additional perspective
on stress-test performance.
Despite these additional efforts, institutions may continue to be
challenged in trying to fully validate their stress tests to the same
extent as other models, given the limitations in conducting perfor-
mance testing. Such limitations do not mean that those stress tests
cannot be used, but there should be transparency about validation
status, and information about the lack of full validation should be
communicated to users of stress-test results. For cases in which
validation and independent review have identified material defi-
ciencies or limitations in a stress test, there should be a remediation
plan to explain how the stress test will be enhanced or its use lim-
ited, or both. Identified deficiencies in stress tests should be com-
municated to all stress-test users.
Additional areas of validation and independent review for stress
tests that require attention from a governance perspective include:

o ensuring that there is appropriate independence and effective

challenge in the validation and review process;
o including validation and independent review of the qualitative
or judgemental aspects of a stress test – such aspects can be an
integral component of a stress test and thus should be reviewed
in some manner, even if they cannot be tested in a quantitative/
statistical sense;
o ensuring that stress tests are subject to appropriate development
standards, including a clear statement of purpose, proper theory
and design, sound methodologies and processing components,
and developmental testing (including testing of assumptions);
o acknowledging limitations in stress-testing methodologies, even
if they represent best practices;

stress_testing.indd 9 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

o recognising any data limitations or weaknesses in data quality;

o ensuring that stress tests are implemented in a rigorous manner that
is appropriate for the stated use, and accounting for any changes to
the developed stress test that occur during implementation;
o monitoring performance on an ongoing basis and assessing any
degradation in performance (where possible);
o expressing stress-test uncertainty and inaccuracy, including in
the form of confidence bands around estimates and/or factors
not observable or not fully incorporated; and
o ensuring that vendor or other third-party models are sufficiently
validated, including their implementation, to ensure they func-
tion as intended and are appropriate for the institution’s use.

INTERNAL AUDIT
An additional aspect of governance and controls is the role of inter-
nal audit. An institution’s internal audit function evaluates practic-
es in a range of risk-management areas, and stress-testing activities
should be among them. Internal audit should provide independent
evaluation of the ongoing performance, integrity and reliability of
stress-testing activities. It is not expected that internal audit will
have full knowledge of all stress-test details, or will have to inde-
pendently assess each stress test used. Rather, internal audit should
look across the firm’s stress-testing activities and ensure that, as a
whole, they are being conducted in a sound manner, are appropri-
ate for the intended purpose and remain current. There should also
be an assessment of the staff involved in stress-testing activities re-
garding their expertise and roles and responsibilities.
Internal audit should also check that the manner in which all ma-
terial changes to stress tests and their components are appropriately
documented, reviewed and approved. In addition, it should eval-
uate the validation and independent review conducted for stress
tests, including all the items listed above relating to validation.
In order to conduct such evaluations, internal audit staff should
possess sufficient technical expertise to understand the stress tests
and challenge their processes and results. It is also important to
review the manner in which stress-testing deficiencies are identi-
fied, tracked and remediated. On the whole, internal audit serves
the valuable task of assessing the full suite of stress-testing activi-

10

stress_testing.indd 10 20/05/2013 18:45

 Governance over Stress Testing

ties across the institution on a regular basis to evaluate whether, as

a whole, such activities are functioning as intended, in adherence
with policies and procedures and serving the institution properly.

OTHER KEY ASPECTS OF STRESS-TESTING GOVERNANCE

This final section outlines some key aspects of stress-testing gov-
ernance that should also receive attention, and areas in which gov-
ernance should be exercised. These include stress-testing coverage,
stress-testing types and approaches and capital/liquidity stress-
testing. The manner in which these areas are addressed can and
should vary across institutions. But, at a minimum, these are areas
of stress testing that should be addressed in some way by senior
management, evaluated as part of the internal control framework
and summarised for review by the board.

Stress-testing coverage

o Appropriate coverage in stress-testing activities is important, as

stress-testing results could give a false sense of comfort if certain
portfolios, exposures, liabilities or business-line activities are not
included; this underscores the need to document clearly what is
included in each stress test and what is not being covered.
o Effective stress testing should be applied at various levels in the
institution, such as business line, portfolio and risk type, as well
as on an enterprise-wide basis; in some cases, stress testing can
also be applied to individual exposures or instruments (eg, struc-
tured products).
o Stress testing should capture the interplay among different expo-
sures, activities and risks and their combined effects; while stress
testing several types of risks or business lines simultaneously
may prove operationally challenging, an institution should aim
to identify concentrations and common risk drivers across risk
types and business lines that can adversely affect its financial
condition – including those not readily apparent during more
benign periods.
o Stress testing should be conducted over various relevant time ho-
rizons to adequately capture both conditions that may materialise
in the near term and adverse situations that take longer to develop.

11

stress_testing.indd 11 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Stress-testing types and approaches

o For any scenario analysis conducted, the scenarios used should

be relevant to the direction and strategy set by its board of direc-
tors, as well as sufficiently severe to be credible to internal and
external stakeholders; at least some scenarios should be of suf-
ficient severity to challenge the viability of the institution.
o Scenarios should consider the impact of both firm-specific and
systemic stress events and circumstances that are based on his-
torical experience as well as on hypothetical occurrences that
could have an adverse impact on an institution’s operations and
financial condition.
o An institution should carefully consider the incremental and cu-
mulative effects of stress conditions, particularly with respect to
potential interactions among exposures, activities, and risks and
possible second-order or “knock-on” effects.
o For an enterprise-wide stress test, institutions should take care in
aggregating results across the firm, and business lines and risk
areas should use the same assumptions for the chosen scenario,
since the objective is to see how the institution as a whole will be
affected by a common scenario.
o Consideration should be given to reverse stress tests that “break the
bank” to help an institution consider scenarios beyond its normal
business expectations and see what kinds of events could threaten
its viability (even if it is difficult to estimate their likelihood).

Capital and liquidity stress testing

o Stress testing for capital and liquidity adequacy should be con-

ducted in coordination with an institution’s overall strategy and
annual planning cycles; results should be refreshed in the event
of major strategic decisions, or other decisions that can materi-
ally impact capital or liquidity.
o An institution’s capital and liquidity stress testing should consid-
er how losses, earnings, cashflows, capital and liquidity would
be affected in an environment in which multiple risks manifest
themselves at the same time – for example, an increase in credit
losses during an adverse interest-rate environment.

12

stress_testing.indd 12 20/05/2013 18:45

 Governance over Stress Testing

o Stress testing can aid contingency planning by helping manage-

ment identify exposures or risks in advance that would need to
be reduced and actions that could be taken to bolster capital and
liquidity positions or otherwise maintain capital and liquidity
adequacy, as well as actions that in times of stress might not be
possible – such as raising capital or accessing debt markets.
o Capital and liquidity stress testing should assess the potential
impact of an institution’s material subsidiaries suffering capital
and liquidity problems on their own, even if the consolidated
institution is not encountering problems.
o Effective stress testing should explore the potential for capital
and liquidity problems to arise at the same time or exacerbate
one another; for example, an institution in a stressed liquidity
position is often required to take actions that have a negative di-
rect or indirect capital impact (eg, selling assets at a loss or incur-
ring funding costs at above market rates to meet funding needs),
which can then further exacerbate liquidity problems.
o For capital and liquidity stress tests, it is beneficial for an institu-
tion to articulate clearly its objectives for a post-stress outcome,
for instance to remain a viable financial market participant that
is able to meet its existing and prospective obligations and com-
mitments.

CONCLUSION
Similar to other aspects of risk management, an institution’s stress-
testing will be effective only if it is subject to strong governance
and effective internal controls to ensure the stress-testing activities
are functioning as intended. Strong governance and effective inter-
nal controls help ensure that stress-testing activities contain core
elements, from clearly defined stress-testing objectives to recom-
mended actions. There are many elements that contribute to effec-
tive stress-testing governance, foremost being the role of the board
and senior management. Stress testing can be a very powerful risk-
management tool, but the board and senior management should
challenge stress-testing processes and results, demonstrating a
solid understanding of their assumptions, limitations and uncer-
tainties. Additionally, strong governance helps ensure that stress
testing is not isolated within its risk-management function, but

13

stress_testing.indd 13 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

is firmly integrated into business lines, capital and asset–liability

committees and other decision-making bodies. Finally, strong gov-
ernance can help institutions continue to recognise the difficulty in
estimating the impact of stressful events and circumstances, there-
by acknowledging that stress-test results should be used only with
sound judgement and a healthy degree of scepticism.

The views expressed in this chapter do not necessarily represent the

views of the Federal Reserve Board or the Federal Reserve System.

1 For the purposes of this chapter, the term “stress testing” is defined as exercises used to
conduct a forward-looking assessment of the potential impact of various adverse events and
circumstances on a banking institution.
2 Risk appetite is defined as the level and type of risk an institution is able and willing to as-
sume in its exposures and business activities, given its business objectives and obligations
to stakeholders.

14

stress_testing.indd 14 20/05/2013 18:45

 2
Stress Testing and
Other Risk-Management Tools

Akhtar Siddique; Iftekhar Hasan

Office of the Comptroller of Currency; Fordham University

The later chapters in this book are focused on various elements and
aspects of stress testing. Stress tests have gained in prominence
since the financial crisis of 2007-9. However, stress testing existed
in the arsenal of risk managers well before the financial crisis. But
it has not existed in isolation: along with stress tests, risk managers
have always used other tools.
In our experience, quite sophisticated stress testing existed in
many banks’ management of market risk before the 2007–9 crisis,
and it often focused on the trading book. This included both trans-
action and portfolio-level stress testing. In contrast, stress testing of
credit risk was more likely to be at a transaction level. Portfolio-lev-
el stress testing was often rudimentary, if it existed all. Enterprise-
wide stress tests tended to be rudimentary (with one or two notable
exceptions), as well, especially for institutions that had large bank-
ing books.
Risk management in financial institutions has always relied on a
panoply of tools and measures. Textbooks on risk management at
financial institutions describe various other tools such as position
limits and exposure limits, as well as limits on the Greeks, such as
on delta or vega.1
In this chapter, we discuss the relationship between those other
tools and stress testing. We first focus on similarities, differences
and consistencies between them. We then discuss the ongoing evo-
lution whereby stress testing has affected other risk-management

15

stress_testing.indd 15 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

tools. We also discuss how other risk-management tools are affect-

ing stress testing.
Of the other risk measures, we focus on the value-at-risk (VaR) mea-
sures. These include the economic capital (EC) measures. This choice
is motivated by the fact that such metrics are designed to capture risk
across different types (such as market, credit, interest rate, etc.) in a
manner similar to stress testing. Additionally, regulatory capital mod-
els as used in Basel II/III can also be viewed as akin to EC models.
Enterprise-wide risk limits have often been based on value at risk or its
variants. More concretely, many institutions have expressed their risk
appetite in terms of a very high percentile such as 99.97% EC.

ENTERPRISE-WIDE STRESS TESTING

As is well known, an important use of stress testing has been to
acquire enterprise-wide views of risk, especially in the supervisory
stress tests run by regulators around the world. These are the enter-
prise-wide stress tests.
At a basic level, different risk-management tools can produce
different results because of differences in the inputs. For both VaR
measures and stress tests, the inputs are data and scenarios.
A stress test may be viewed as translation of a scenario into a
loss estimate. In a similar vein, EC or VaR methods also involve
translation of scenarios into loss estimates. The distribution of the
loss estimates are then used to derive the VaR at a high percentile
such as 99% or 99.9%. In practice, stress tests usually focus on a few
scenarios, whereas VaR measures commonly utilise a very large
number of scenarios.
Hence, as long as identical inputs and similar definitions of loss
estimates are used between stress tests and EC/VaR methods, there
can be consistency between stress tests and EC/VaR methods, at
least when identical scenarios are used.
However, in practice, the loss estimates are often defined quite
differently between stress tests and EC methods. In particular, a
significant difference is that losses in stress tests have more often
than not taken an accounting view rather than a “market” view
commonly attempted in EC methods.
The second significant difference has been the horizon. Enter-
prise-wide stress tests have often examined a long period such as

16

stress_testing.indd 16 20/05/2013 18:45

 Stress Testing and Other Risk-Management Tools

losses over nine quarters in the Dodd–Frank stress tests in the US.
In contrast, EC models have focused on losses at a point in time,
such as the loss in value at the end of a year.
The final significant difference is the role of probabilities. Scenar-
ios for stress tests can sometimes be generated using distributions
of the macroeconomic variables. Therefore, the results of a scenario
in a stress test can be assigned a probability, ie, the probability of
that scenario. However, probabilities have not played a promi-
nent role in stress tests. For many stress tests conducted around
the world, ordinal rank assignments such as “base”, “adverse” and
“severely adverse” have been done, but with little discussion of the
cardinal probabilities attached to them. In contrast, cardinal prob-
abilities generally play a large role in the VaR-type models. For the
VaR/EC models using Monte Carlo simulation, there exist complex
statistical models underneath. For the VaR models using histori-
cal simulation, the history has been viewed as the distribution to
draw from. More importantly, in the interpretation and use of the
VaR/EC model results, probabilities have played a very large role.
A 99.9% VaR loss has often been viewed as a 1-in-1,000 event, albeit
with uncertainty (or standard errors) around it.
The last difference has been the approach to scenarios. Stress-test
scenarios are often ad hoc and conditional, rather than the uncondi-
tional scenarios typically generated in VaR-type metrics. Especially,
for the regulatory stress tests, the scenario-generation process has
looked at the present period as the starting point and then gener-
ated two or three hypothetical scenarios from that starting point.

A simple example: stress test

A concrete example can be given for a wholesale portfolio. It is a very
simplified example designed to get the idea across rather than provide
a guideline to follow. Let us assume the bank is using a two-year sce-
nario that consists of GDP growth and unemployment (see Table 2.1).

Table 2.1 Hypothetical scenarios for two macro variables

Macro variable 1st-year change from base 2nd-year change from base
GDP −1% −0.5%
Unemployment +1% 0%

17

stress_testing.indd 17 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

For wholesale exposures, let us assume that the bank has chosen
to model at a portfolio (top-down) level rather than a loan level. At
a basic level, the bank needs to estimate the sensitivities of losses
in this portfolio to the changes in the two macro variables: GDP
growth and unemployment.
Let us assume the following information on the bank’s wholesale
portfolio (see Table 2.2).

Table 2.2 Portfolio composition for the hypothetical bank

Rating bucket Balance 1-year default 2-year default

rate (%) rate (%)
1 200 0.00 0.00
2 350 0.01 0.02
3 400 0.02 0.10
4 500 0.18 0.53
5 100 1.23 3.31
6 10 5.65 12.35
7 0 21.12 33.53

Let us assume that the bank chooses to use a PD LGD approach. There-
fore, the bank needs to compute what the PD is in the stress scenario
for each of the two years. Additionally, the bank needs to model which
of the exposures transition to a lower rating. Finally, the bank needs to
understand what new wholesale loans the bank will generate in the
two years and what rating buckets (and PD) the new loans will be in.2
Based on historical experience, the bank establishes the following first-
year and second-year stressed PDs. This may be based on the bank’s
own historical experience or on industry data. The exposures (EAD)
are not expected to change. However, the LGD does change. Using
the experience of 2008, the bank finds that, according to Moody’s URD
data, the LGD for senior unsecured increases from 53% to 63%. The
bank chooses to increase its LGD by 10% for all rating buckets. Table
2.3 presents the balances and the stressed parameters for the bank’s
wholesale portfolio. We are assuming no new business and are not tak-
ing into account migration between the two years.

18

stress_testing.indd 18 20/05/2013 18:45

 Stress Testing and Other Risk-Management Tools

Table 2.3 Projected stressed parameters for the bank’s portfolio

Rating bucket Balance Stressed LGD 1st-year 2nd-year

stressed PD stressed PD
(%) (%)
1 100 60% 0.02 0.00
2 200 60% 0.03 0.02
3 400 70% 0.04 0.03
4 500 70% 0.25 0.20
5 100 80% 1.85 1.50
6 200 80% 8.00 8.50
7 500 90% 25.00 20.00

The two-year cumulative loss rate comes out to be 7.44% with these
assumptions.

A simple example, continued: EC/VaR.

In the implementation of EC models, banks commonly use a Mer-
ton model framework to simulate the defaults and credit quality.
In this framework, asset returns are simulated using a factor model
framework, and default occurs when the simulated asset value is
below a threshold (generally tied to the leverage of the borrower)
at the one-year horizon.
In a multifactor setup, for a borrower i with default probability PD

Zi < N-1(PDi)) where

Zi = βi1GDP + βi 2Unemployment +( 1 – βi21 – β i22 )ηi

where Zi is a unit normal variable and GDP and unemployment

are simulated values for the two macroeconomic factors. For credit
quality, the simulated asset value (and by extension the simulated
leverage) is used to impute a spread. It is common for the shock to
the spreads to be modelled as a function of Zi as well. Banks gener-
ally generate the asset values using a correlation matrix using cor-
relations between industries and countries.

19

stress_testing.indd 19 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Banks run these simulations a number of times, sort the losses from
the draws, and arrive at the 99th or 99.9th percentile of the loss dis-
tribution as the 99th or 99.9th percentile VaR.
The loss for the stress test may correspond to one of the losses and
can allow the user to roughly gauge the severity of the stress test.

USE OF VaR MODELS IN STRESS TESTS

Since the VaR models provide a mechanism for computing loss via

Loss=PD×LGD×EAD

One approach that some institutions have taken is to assess where

the losses based on stress tests lie in the loss distribution used in the
VaR/EC estimation. This process has been one mechanism to associ-
ate probability with a given hypothetical or historical stress scenario.
Going one step further, some institutions have also used such mecha-
nisms to tie together scenarios across disparate lines of business.
As an example, if a scenario’s loss magnitude translates into a
90th percentile loss on the loss distribution for VaR, the bank may
take the 90th percentile loss in the EC model as an approximation
to the stressed loss for market risk.
No financial institution can be run with zero risk tolerance, nor
can all sources of risk be eliminated. However, clearly, some losses
are unacceptable because of their magnitudes, irrespective of the sce-
narios. For such losses, the likelihood (or probability) of the scenario
is not that material. However, for most scenarios, the output tends to
be used as the loss in that scenario and the likelihood of that scenario.
The assignment of probability via “matching” the stressed loss to
a point on the loss distribution serves the useful purpose of coming
up with the probability of that scenario. Since, for the practical imple-
mentation of stress tests in risk management, assignment of prob-
abilities to the outcomes is important, the probability arrived via the
loss distribution can help make the stress tests more actionable.3

STRESSED CALIBRATION OF VALUE AT RISK MEASURES

Another approach to incorporating stress into risk measurement
methodologies has been the use of stressed inputs. There have been
quite a few variants. This has been particularly useful in the mar-

20

stress_testing.indd 20 20/05/2013 18:45

 Stress Testing and Other Risk-Management Tools

ket risk area. The incorporation of stress into the risk measurement
as well as capital metrics has occurred in both the supervisory ap-
proaches and the many banks’ internal approaches.
On the supervisory approaches, the new market risk rule re-
quires banks to use stressed inputs, ie, the revisions to the market
risk capital framework (BCBS 2011a) states,

In addition, a bank must calculate a “stressed value-at-risk” measure.

This measure is intended to replicate a value-at-risk calculation that
would be generated on the bank’s current portfolio if the relevant
market factors were experiencing a period of stress; and should there-
fore be based on the 10-day, 99th percentile, one-tailed confidence
interval value-at-risk measure of the current portfolio, with model
inputs calibrated to historical data from a continuous 12-month
period of significant financial stress relevant to the bank’s portfolio.

The revisions to the market risk capital framework also explicitly

require the use of stress tests: “Banks that use the internal models
approach for meeting market risk capital requirements must have
in place a rigorous and comprehensive stress-testing program.”
Similarly, in the revisions to Basel III (BCBS 2011b), stressed param-
eters are required: “To determine the default risk capital charge for
counterparty credit risk as defined in paragraph 105, banks must use
the greater of the portfolio-level capital charge (not including the CVA
charge in paragraphs 97–104) based on Effective EPE using current
market data and the portfolio-level capital charge based on Effective
EPE using a stress calibration. The stress calibration should be a single
consistent stress calibration for the whole portfolio of counterparties.”
As an illustration, we present some results from Siddique (2010),
with six risk factors to simulate the exposures. These are: (1) Three
month LIBOR (LIBOR3M); (2) the yield on BAA-rated bonds
(BAA); (3) the spread between yields on BAA- and AAA-rated
bonds (BAA–AAA); (4) the return on the S&P 500 index (SPX); (5)
the change in the volatility option index (VIX); and (6) contract in-
terest rates on commitments for fixed-rate first mortgages (from the
Freddie Mac survey) (MORTG). MORTG is in a weekly frequency
that is converted to daily data through imputation using a Mar-
kov chain Monte Carlo. There are a total of 2,103 daily observations
over the period January 2, 2002, through May 10, 2010.

21

stress_testing.indd 21 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

With Monte Carlo, the stressed VaR as 99.9th percentile of a dis-

tribution of P&Ls generated using stressed parameters can be con-
structed. Two separate sets of moments, (1) using the previous 180
days or 750 days’ history of the risk factors and (2) the stress period
(180 days or 750 days ending in 30.06.09), are used to simulate the
risk factors. The 99.9th percentile of the portfolio value is then the
99.9th regular VaR or stressed VaR based on which sets of moments
are used. Figure 2.1 illustrates VaR and stressed VaR with a bal-
anced portfolio.

Figure 2.1 VaR and stressed VaR with a balanced portfolio

2.5

1.5
VaR

1
VaR
0.5 Stressed VaR
0
17/01/2003
20/05/2003

20/01/2004
19/05/2004
20/09/2004
19/01/2005
19/05/2005
19/09/2005
19/01/2006
19/05/2006
19/09/2006
22/01/2007
22/05/2007
20/09/2007
22/01/2008
21/05/2008
19/09/2008
21/01/2009
21/05/2009
21/09/2009
21/01/2010
18/09/2002

18/09/2003

Source: Siddique (2010)

Figure 2.2 Comparison of VaR & stressed VaR of CVA: different stressed
periods

1.20E-02
1.00E-02
8.00E-03
CVA VaR stressed 180 days
6.00E-03
4.00E-03 CVA VaR regular 180 days
2.00E-03
CVA VaR stressed 2008 180 days
0.00E+00
11/07/2003
04/12/2003

20/07/2005
11/05/2006
05/10/2006
06/03/2007
31/07/2007
24/12/2007
21/05/2008
15/10/2008
07/08/2009
04/01/2010
13/02/2003

03/05/2004
23/02/2005
13/12/2005

13/03/2009
18/09/2002

28/09/2004

Source: Siddique (2010)

22

stress_testing.indd 22 20/05/2013 18:45

 Stress Testing and Other Risk-Management Tools

Stressed inputs are also used in the capital charge for credit valua-
tion adjustment (CVA) as mentioned above. To assess the impact of
the use of stressed inputs for those metrics, Siddique (2010) carries
out some other simulations whose results are presented in Figure 2.2.
Two separate periods are used to compute the stressed calibra-
tion: (1) 180 days ending 30.09.08; and (2) 180 days ending 30.06.09.
The impact of a stressed calibration appears in the early period in
the data, where the CVA VaR is substantially higher than the un-
stressed (regular) CVA VaR. However, in the latter period the un-
stressed and stressed VaR are identical. It is important to note that
an incorrect stress period (ie, ending 30.09.08) can actually produce
VaR lower than an unstressed CVA VaR.
There are both advantages and disadvantage of such stressed risk
metrics. An obvious advantage is that, with capital for unexpected
losses taking into account stressed environments, capital should be
adequate when the next stress or shock occurs. That is, a risk metric
with a stressed input is usually going to be more conservative.
However, given that the inputs are always stressed, the risk met-
ric will no longer be responsive to the current market conditions,
but primarily depend on the portfolio composition.
Only time will tell what the final impact of the incorporation of
stress-testing elements into risk management and capital adequacy
metrics will be.

CONCLUSION
Stress testing has played a very large role in the assessment of capi-
tal adequacy. It has always played a role in risk management as
well, which has become much larger as a result of the 2007–9 fi-
nancial crisis. However, banks have continued to use other risk-
management tools such as VaR as well. Nevertheless, stress testing
has influenced those tools and those tools have also been used in
stress testing.

The views expressed in this chapter are those of the authors alone and do not
necessarily represent those of the Comptroller of the Currency or the Bank of
Finland.

23

stress_testing.indd 23 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

REFERENCES
Basel Committee on Banking Supervision, 2011a, “Revisions to the Basel II mar-
ket risk framework”, available at [Link]

Basel Committee on Banking Supervision, 2011b, “Basel III: A global regulatory

framework for more resilient banks and banking system”, available at http://
[Link]/publ/[Link].

Hull, John, 2012, Risk Management and Financial Institutions, 3rd edn (New York:
Wiley Books).

Siddique, Akhtar, 2010, “Stressed versus unstressed calibration”, Unpublished

Manuscript,, Office of the Comptroller of the Currency

1 See, for example, Hull (2012).

2 For the purposes of this simplified example, we are aware that we are making very strong
assumptions and simplifications in this example and are ignoring many elements that banks
take into account. For example, banks can find that the underwriting of new loans can actu-
ally be stricter in a recession, resulting in a lower PD for new business compared with the
existing book.
3 Action triggers (when actions need to be taken) for stress tests can be tied to either the out-
put – for example, if the losses exceed a certain level. Alternatively, they can be tied to the
input, ie, if the realised input into a test is below/above a threshold. As an example, a stress
test can involve a scenario for GDP growth. If GDP growth in a quarter is below a trigger
such as −2%, actions can be taken.

24

stress_testing.indd 24 20/05/2013 18:45

 3
Stress Testing for Market Risk

Dilip K. Patro, Akhtar Siddique; Xian Sun

Office of the Comptroller of the Currency; Johns Hopkins University

Stress testing has received increased attention from both financial

institutions and regulators since the 2007–9 financial crisis. What
constitutes reliable and relevant stress tests, however, still stirs a lot
of debate among practitioners, stakeholders and researchers. The
development and evolving practices of stress testing in the area of
market risk management are reviewed in this chapter. In addition,
the ways in which stress-testing methodology can be improved for
risk management is presented.
Generally speaking, market risk refers to changes in a financial
institution’s portfolio values due to unanticipated changes in mar-
ket risk factors such as the price and volatility of equities, interest
rates, credit spreads, the price of commodities and foreign-exchange
rates. Stress testing for market risk has been an important compo-
nent of stress tests, both in the internal stress tests run by bank-
ing organisations and in the stress tests run by financial regulators.
During the financial crisis of 2007–9, the largest losses were often
in portfolios sensitive to market risk. Traditionally, stress tests for
market risk have been conducted for portfolios in banks’ trading
books, by using scenarios of possible states for market risk factors.
In order to do so, severe but plausible scenarios are often chosen.
The choice of scenarios, therefore, is crucial but sometimes inevi-
tably subjective as well. The traditional approaches for scenarios
utilise one of the three methods: standard scenarios, historical sce-
narios and worst-case scenarios.

25

stress_testing.indd 25 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

In the analysis using standard scenarios, the values of portfolios

are estimated by stressing the market risk factors by pre-specified
shocks, such as changing equity prices by some standard deviations
or increasing oil prices to a certain level. These are also referred
to as hypothetical scenarios. In the historical-scenarios analysis,
market states for a particular historical time period relevant for the
bank’s portfolio are used. This may include, for example, the 1987
equity crash, the 2007–9 credit crisis, etc. In the worst-case scenario
analysis, an automated search over prospective changes in market
states is conducted in order to evaluate profits and losses under
that scenario.
The traditional stress tests are easy to conduct but have impor-
tant limitations. For example, the historical scenarios offer reliable
but likely less relevant information for risk management in the fu-
ture; the standard analysis may not be based on the changes in the
market states that are close to a stress event; and the worst-case
analysis looks at the impacts of changes that are unlikely to occur.
Given the drawbacks of traditional stress tests, risk managers have
been compelled to use expert judgement along with the results of
the stress analyses before taking actions based on results.
The literature on stress testing has grown rapidly, incorporat-
ing many suggestions for improvements in stress-testing methods.
For example, to integrate stress testing into formal risk modelling,
Berkowitz (1999) proposed a solution that includes assigning prob-
abilities to stress-test scenarios; Artzner et al (1999) suggested esti-
mating the expected tail loss to address the drawbacks of market
value-at-risk (VaR); and Kupiec (1998) discussed the importance of
allowing changes in the market states to be correlated. However, it
was the financial crisis of 2007–9 that highlighted the importance
of stress testing as an important risk-management and regulatory
tool, and, as a result, many large banks have incorporated such
practices as part of their enterprise-wide risk management. Fur-
thermore, there have been many statutory and regulatory reforms
that have required such practices. There has also been an increas-
ing consensus among practitioners, regulators and researchers that
stress tests are important because they help banking organisations
to understand market risk exposures that methods such as VaR
may miss, depending on how VaR models are developed and im-

26

stress_testing.indd 26 20/05/2013 18:45

 Stress Testing for Market Risk

plemented. The hope is that such information can be used by ex-

ecutive management for strategic planning, capital adequacy and
capital allocation, and other major decisions. Stress tests can also
help regulators identify risk concentrations and systemic risk, and
take mitigating actions.
This chapter discusses several important aspects of stress testing
for market risk. First, we focus on the distinction between stress test-
ing for market risk and the more common stress testing for credit
risk. After that, we discuss the role of scenarios in market risk and
how they may differ from other types of stress tests. A discussion of
the horizon for the stress tests follows. The last section focuses on re-
valuation of the values and computation of profits and losses under
the stressed scenarios followed by concluding remarks.

DISTINGUISHING BETWEEN STRESS TESTING FOR MARKET RISK

AND CREDIT RISK
Stress testing for market risk and that for credit risk have the com-
mon ultimate goal of estimating the impacts on portfolio values of
given plausible and severe events. Stress testing for market risk has
become almost synonymous with stress testing of the trading book.
The trading book, in contrast with the banking book, is character-
ised by a large number of instruments as well as a larger number of
positions and risk factors. Further, mark-to-market losses for trad-
ing books are usually measured at a point in time. In particular,
derivative positions that are the result of off-balance-sheet activities
by financial institutions are often a major focus of market risk stress
tests. Probabilistic risk measures, such as VaR, are the primary tool
for measuring risk capital as well as computing regulatory capital
for such activities. Basel 2.5 has also introduced a “stressed VaR”.
On the other hand, stress testing for credit risk is mostly con-
ducted for portfolios in banking books, which primarily consist of
portfolios of loans, securities or positions held as investments as
well as direct equity investments. The output for stress testing for
credit risk could be accounting losses over a period of time due
to the impacts on rating changes, probability of default (PD), and
loss-given default (LGD). The selection of scenarios could also be
more nuanced for the stress testing of credit risks because different
segments (such as utility, banking, real estate) in a portfolio may

27

stress_testing.indd 27 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

be exposed to different market states. However, trading-book posi-

tions are also subject to credit risk if the risk factors include credit
spreads. Furthermore, both trading-book and banking-book posi-
tions are subject to default of the counterparty for derivatives or se-
curities financing transactions, which is referred to as counterparty
credit risk and is discussed in Chapter 4.
Another source of difference between these two types of risk comes
from liquidity risk. The impact of changes in liquidity may have an
immediate and severe impact on the value of a trading book position.
The 2007–9 financial crisis highlighted the severe impacts of extreme
illiquidity, which had received little attention in stress testing before
the crisis. While banks, regulators and researchers alike have realised
the importance of liquidity risk management, incorporating it in the
formal stress testing is less straightforward. Liquidity and stress test-
ing are discussed in more detail in Chapter 11.

Aggregation of results from stress tests

Although researchers have proposed sophisticated approaches in
aggregating the results of stress tests from various risk sources, con-
cerns remain in bottom-up approaches because it may significantly
underestimate the true total risk (Breuer et al 2010). For example,
Rosenberg and Schuermann (2006) used an approach to aggregate
different risk types such as market, credit and operational. They
used a bottom-up approach in which various risks (market, credit
and operational) are separately analysed and estimated, and then
aggregated to produce the total risk. This aggregation relied on the
assumption that risk types are subadditive. For example, as Rosen-
berg and Schuermann (2006) showed, risks can aggregate with per-
fect correlation (“add-VaR”), which placed an upper bound on the
economic capital a bank would need at a given risk-tolerance or
confidence level. Risks could also be aggregated using a copula-
VaR approach, in which a joint distribution of loss can be estimated
by using the shape and location of each of the risk distributions and
a dependence function. In general, the aggregated risk estimated
by copula-VaR models has tended to be lower than risk estimated
by add-VaR. Regardless of whether researchers/practitioners have
used an add-VaR or a copula-VaR method to aggregate risk, each
of the risk distributions has been separately estimated. The loca-

28

stress_testing.indd 28 20/05/2013 18:45

 Stress Testing for Market Risk

tion and shape of the loss distribution for each risk type remain
unchanged under summing (with a correlation coefficient of one,
or with some “diversification” discount). However, bank losses
associated with different risk types may not be additive. In other
words, when the worst-case scenario for market risk has occurred,
the chance of the worst-case scenario for credit risk may also have
increased. This is plausible because financial markets/instruments
have become so developed that credit risk is now borne opaquely
by capital markets.

STRESS TESTING MARKET RISK AND CHOICE OF SCENARIOS

As in the stress tests for other risks, market risk stress tests have
also relied on hypothetical and historical scenarios. There have
been other approaches used in the selection of scenarios, such as via
mechanically driven searching. Scenarios used in enterprise-wide
stress tests have tended to be focused on macroeconomic and finan-
cial variables. The scenarios are often specified as vectors of macro-
economic variables such as GDP, unemployment, inflation, house
prices and interest rates. The losses resulting from market risk in
such scenarios are often difficult to estimate directly. Financial insti-
tutions have generally needed to translate the macroeconomic sce-
narios into changes in market risk factors such as changes in credit
spreads, commodity prices and volatilities of the other factors that
are used to compute the losses for the market shocks.
As an example, the valuation of a credit default swap depends on
the recovery rate and the default probability of the reference asset.
If a macroeconomic scenario has been specified in terms of GDP,
the unemployment rate and HPI (House Price Index), the bank
has created an auxiliary model that translated the macroeconomic
variables into default probabilities and recovery rates. Similarly,
for financial products such as securitisations, banks generally use
systems such as Intex to value the securities.1 The inputs for Intex
are computed from the macroeconomic scenarios using auxiliary
models. For a large proportion, if not all, of the exposures subject
to market risk, the necessary variables are the inputs for the pricing
models used for the trading-book positions. These inputs are pric-
es, volatilities and other parameters such as correlations. Generally,
the stress is modelled as very large movements in these inputs.

29

stress_testing.indd 29 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Given the fairly large flexibility in how the macroeconomic vari-

ables are translated into the variables more commonly used in mar-
ket risk stress testing, comparability and consistency within and
across institutions can be a challenge in market risk stress testing as
compared with credit risk stress testing. Therefore, in many of the
regulatory stress tests based on the Dodd–Frank Act or the Federal
Reserve’s CCAR (Comprehensive Capital Analysis and Review) in
the United States and the supervisory stress tests run by the Euro-
pean Banking Authority (EBA) in Europe, regulators have speci-
fied the values for many of the variables that would constitute the
inputs for the market risk valuations systems. However, even in
those cases, institutions frequently had to come up with inputs for
risk factors that the regulators had not specified.
An additional consideration has been that, unlike with the bank-
ing book, where a bank is naturally “long credit”, for trading-book
portfolios a bank may actually have positions that have gains in an
adverse scenario. That imposes additional pressures on how (and
how well) the results of a market risk stress test are verified. Given
that the scenarios provided by regulators may have failed to en-
compass all possible risk factors that an institution uses, a greater
degree of inconsistency in the market risk stress tests compared
with credit risk stress tests occurs. This highlights the importance
of effective monitoring by the regulator. Although the regulator’s
goal of specifying the values for the inputs of the market risk valua-
tions is to enhance consistency and comparability within and across
the banks, it encounters problems such as “one size does not fit all”
and “catch up with the changes”. In order to ensure the market risk
stress tests correctly reflect an individual bank’s riskiness, the regu-
lator needs to subject the inputs, especially those generated by the
institutions, to greater scrutiny to validate their representativeness
and comprehensiveness.

TIME HORIZON IN MARKET RISK STRESS TESTS

Stress tests for other risk areas such as credit risk have generally
focused on the losses over a long horizon such as the nine quarters
used for the Dodd–Frank Act and CCAR stress tests in the United
States. An important element of these stress tests has been how the
portfolio may have changed in the stressed period, such as new

30

stress_testing.indd 30 20/05/2013 18:45

 Stress Testing for Market Risk

business or changes in the value of the existing portfolio even if

they do not default. Historically, stress tests for market risk, such as
the stress tests focused on trading portfolios, have had very short
horizons, such as ten days or even instantaneous, depending on the
nature of the trading frequency of the products in the portfolios.
The shorter horizon was motivated by more frequent trading and
a shorter holding period that has characterised the trading book.
The choice of horizon not only had a direct impact on the valu-
ation of the portfolios under stress tests, but also had significant
implications for how banks took actions (such as reducing expo-
sures) based on the outputs of the stress tests. For example, an im-
portant aspect of the choice of horizon is what happens after the
initial shock, that is to say the aftershock. Alexander and Sheedy
(2008) pointed out that the consequences of a shock event can in-
clude some or all of the following: further large moves in the same
market (as predicted by volatility clustering); large moves in other
markets and higher correlations between markets; and increased
implied volatility in option markets and reduced market liquidity.
In a portfolio-level stress test of market risk, the horizon used may
be the same as the horizon for the VaR, such as ten days. In such a
scenario, the aftershock may not materialise.
However, for enterprise-wide stress tests the market risk stress
tests generally need to be the same horizon as the stress tests for the
other risks, such as credit risk, because a common horizon is needed
to put the various risks on a common footing. Two features then can
become important. The first is that the aforementioned aftershock is
then relevant. Simply, taking the 10-day loss and expanding it to a
one-year horizon via a method such as multiplying by the square root
of time may misstate the losses, since that ignores the aftershocks.2
The second is the impact of hedging, other management actions and
what assumptions are made regarding hedging. Whether hedging
is taken into account and how the hedging is accommodated has
varied among institutions. Banks have traditionally argued that they
can implement dynamic hedging strategies that can end up reducing
the unexpected losses of a portfolio in a significant way. However, in
a severely stressed environment it is not clear if the hedging instru-
ments remain available with enough liquidity. The experiences with
the 1987 crisis showed that dynamic hedging strategies that worked

31

stress_testing.indd 31 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

in normal environments failed to take into account the positive feed-

back of hedging demand in a stressed environment especially where
there very large discontinuities.3 When lengthening the horizon for
the market stress tests while conducting enterprise-wide stress tests,
it may be problematic to assume that dynamic hedging is still as ef-
fective as in normal markets when everything else is also stressed.

REVALUATIONS AND COMPUTATION OF P&L UNDER THE

STRESSED SCENARIOS
Once market stress scenarios are specified, the next step is to re-
value the positions in the portfolio under the stressed scenarios.
The difference in the value of the positions under the stressed sce-
nario and the current value is the profit or loss (generally referred
to as P&L). There are several steps in this process of estimation of
stressed P&L that may involve approximations and calibrations.
These steps are discussed below.

Mark-to-market versus market-to-model valuations

Valuations for routine risk management and reporting purposes
rely on either mark-to-market valuations that use closing market
prices from exchanges/market consensus prices from a pricing ven-
dor, or mark-to-model valuations based on analytical or numerical
models. These models often have input parameters that are calibrat-
ed using current market prices. These pricing models, or “pricers”,
form the backbone of revaluations or repricing of positions under
stressed scenarios. Whether the positions are marked to market or
marked to model, the pricing models are used to generate risk sen-
sitivities to risk factors. For example, changes in prices to spot prices
are often referred to as delta, while changes in prices to changes in
volatility are referred to as vega etc. These risk sensitivities, such as
delta, gamma and vega, are based on Taylor-series representation of
changes in valuations for changes in risk factors. Banks often use a
“bump-and-reprice” approach to estimate these sensitivities.4 These
risk sensitivities are aggregated by product, business unit, legal en-
tity and so forth, based on the granularity of desired risk measure-
ment and risk reporting. Once we have the pricers and/or the risk
sensitivities for the positions in the portfolio, the market shocks un-
der stress are applied to reprice those positions.

32

stress_testing.indd 32 20/05/2013 18:45

 Stress Testing for Market Risk

While in many cases the market shocks may be applied directly to

the risk factors, in many other cases it is necessary to apply them as a
relative or absolute percentage of the current level of the risk factors.
Further, based on how the pricing model takes market inputs, there
may be a need for transformation of the stressed scenario risk factor
inputs to inputs that can be used in the pricing models. This may also
be necessitated by the need to avoid hitting boundary conditions,
negative rates or negative forward volatilities.

Revaluations: sensitivity-based, grid-based or full revaluations

When repricing positions under specified scenarios, banks may use
risk sensitivities that are usually generated in front-office pricing
systems to revalue the positions. Such approximations are reason-
able when the risk factor shocks are small. For linear instruments
such as cash equity, use of risk sensitivities can be exact. However
when revaluing positions for which there is a non-linear relation-
ship between prices and risk factor moves, use of full revaluations
using the front-office pricers is recommended, since the Taylor-se-
ries approximation using risk sensitivities will not perform well for
large moves in risk factors. Apart from use of risk sensitivities using
Taylor-series approximations or use of full revaluations where all
relevant risk factors are shocked simultaneously, banks may also
use what are called valuation grids. These are pre-estimated full
revaluations for specified moves in risk factors, and can be for one
risk factor or for joint moves in risk factors (two or multidimen-
sional grids). If the stress scenario is in between or outside what
are the prespecified grid points (say 1%, 5%, 10% and so forth), in-
terpolation and extrapolation are used to estimate P&L for those
scenarios. Although not as reliable as full revaluations using front
office systems, grid-based approximation has the advantage of
computational speed, especially when dealing with large numbers
of positions and/or scenarios.

Revaluations in practice
Banks may use a mix of full revaluation or revaluations using ap-
proximations for the various positions. For positions and products
that have a non-linear relationship between the prices and the risk
factors, full revaluation using front-office systems is the best op-

33

stress_testing.indd 33 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

tion. In some cases banks may use middle-office versions of the

front-office systems. As long as these are calibrated as frequently
as the front-office systems and have the exact implementation,
that may be sufficient. Banks may sometimes also have pricers
in risk systems that are different from front-office pricers. In such
cases it is important to ensure that the valuations from the two
systems are consistent and that these models have gone through
the model-validation process that is expected of other pricing and
risk models.

Model failures, cross-effects, approximations, specific risk and use

of proxies
Sometimes valuation models that are designed for normal market
conditions may fail to calibrate or price under extreme scenarios.
This could be due to things such as negative interest rates or for-
ward rates or approximations such as moment-matching condi-
tions that do not perform well under higher volatilities. Further,
the practice of revaluating positions using risk sensitivities or
grids may fail to capture the cross effect of shocks across differ-
ent asset classes (for example, equities and currencies prices and
volatilities). A full revaluation by design may capture such effects
while use of risk sensitivities and grids may not. In such cases the
banks must estimate the impact of such omissions separately and,
if found material, make a conservative adjustment to the loss es-
timates. Similarly, shocks designed for broad market risk factors
will not be sufficient for issuer-specific risk, especially if the bank
has a concentration in such positions. In such cases, use of name-
specific shocks may be necessary. Furthermore, when proxies are
used, it must be noted that there will be a basis between the risk
factor and the proxy, and the basis may get exacerbated during pe-
riods of stress. There do not exist standard fixes for such issues, and
management needs to monitor for such failures and deal with them
on a case-by-case basis and apply conservative adjustments as nec-
essary. These issues highlight the need for an effective model risk
management process at the institution, which should also scope in
use of models for stress testing.5

34

stress_testing.indd 34 20/05/2013 18:45

 Stress Testing for Market Risk

CONCLUSION
Stress testing for market risk is an important tool for risk manage-
ment, capital adequacy and bank supervision. This chapter sum-
marises the important elements of stress testing for market risk
at financial institutions, where different scenarios of market risk
factors are developed and portfolios sensitive to market risk are
revalued under those scenarios to estimate potential losses. This
chapter distinguished stress testing for market risk from stress test-
ing for credit risk, discussed development of stress-test scenarios,
importance of time horizons in stress testing and challenges with
aggregation of results for various types of risks. Finally, this chap-
ter discussed methods for revaluation of the portfolios under stress
scenarios and some things to consider as part of an effective model
risk management for stress testing.

The views expressed in this chapter are those of the authors alone and do not
necessarily represent those of the Comptroller of the Currency. The authors
would like to thank Jonathan Jones and Wenling Lin for helpful comments.

REFERENCES
Alexander, Carol, and Elisabeth Sheedy, 2008, “Developing a stress-testing
framework based on market risk models”, Journal of Banking & Finance 32, pp.
2220–36.

Artzner, Philippe, et al, 1999, “Coherent Measures of Risk.” Mathematical Finance

9, pp. 203–28.

Berkowitz, Jeremy, 1999, “A Coherent Framework for Stress-Testing”, manuscript,

Board of Governors of the Federal Reserve.

Breuer, T., et al, 2010, “Does adding up of economic capital for market- and credit
risk amount to conservative risk measurement?”, Journal of Banking and Finance 34,
pp. 703–712.

Kambhu, John, 1997, “The size of hedge adjustments of derivatives dealers’ US

dollar interest rate options”, manuscript, Federal Reserve Bank of New York.

Kupiec, Paul, 1998, “Stress-testing in a Value at Risk Framework.” Journal of

Derivatives 6, pp. 7–24.

Rosenberg, J. V., and T. Schuermann, 2006, “A general approach to integrated

risk management with skewed, fat-tailed risks”, Journal of Financial Economics 79,
pp. 569–614.

35

stress_testing.indd 35 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

1 Intex is software widely used by many banks. It has provided deal cashflow models, ana-
lytics and structuring software for RMBSs, ABSs, CMBSs, CDOs, CLNs and covered bond
securities.
2 Whether applying a simple square-root-of-time rule understates or overstates the losses in
such a situation is unclear in the authors’ experience, and depends on the assumptions and
the markets. In some cases, mean reversion over a one-year period means that a straight
square root of time overstates losses.
3 Kambhu (1997) assesses the magnitude of hedging demand from dealers in fixed-income
markets.
4 A Taylor series is a series expansion of a function from the values of the function’s deriva-
tives at single point.
5 See OCC bulletin 2011–12 for principles of effective model risk management.

36

stress_testing.indd 36 20/05/2013 18:45

 4
The Evolution of Stress Testing
Counterparty Exposures

David Lynch
Federal Reserve Board

The call for better stress testing of counterparty credit risk exposures
has been a common occurrence from both regulators and industry
in response to financial crises (CRMPG I 1999; CRMPG II 2005; FRB
2011). Despite this call, statistical measures have progressed more rap-
idly than stress testing. In this chapter we examine how stress testing
may be improved by building off the development of the statistical
measures. We begin by describing how the measurement of counter-
party risk has developed by viewing the risk as a credit risk and as a
market risk. The problems this creates for a risk manager who is devel-
oping a stress-testing framework for counterparty risk are then identi-
fied. Methods to stress-test counterparty risk are described from both
a credit risk perspective and from a market risk perspective, starting
with the simple case of stressing current exposures to a counterparty.
These stress tests are considered from both a portfolio perspective and
individual counterparty perspective. Last, some common pitfalls in
stress testing counterparty exposures are identified.

The evolution of counterparty credit risk management

The measurement and management of counterparty credit risk
(CCR) has evolved rapidly since the late 1990s. CCR may well be
the fastest-changing part of financial risk management over the time
period. This is especially true of the statistical measures used in CCR.
Despite this quick progress in the evolution of statistical measures of
CCR, stress testing of CCR has not evolved nearly as quickly.

37

stress_testing.indd 37 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

In the 1990s a large part of counterparty credit management in-

volved evaluation of the creditworthiness of an institution’s de-
rivatives counterparties and tracking the current exposure of the
counterparty. In the wake of the Long-Term Capital Management
crisis, the Counterparty Risk Management Policy Group cited de-
ficiencies in these areas and also called for use of better measures
of CCR. Regulatory capital for CCR consisted of add-ons to current
exposure measures (BCBS 1988.) The add-ons were a percentage of
the gross notional of derivative transactions with a counterparty.
As computer technology has advanced, the ability to model CCR
developed quickly and allowed assessments of how the risk would
change in the future.
The fast pace of change in CCR modelling can be seen in the pro-
gression of statistical measures used to gauge counterparty credit
risk. First, potential-exposure models were developed to measure
and limit counterparty risk. Second, the potential-exposure models
were adapted to expected positive-exposure models that allowed
derivatives to be placed in portfolio credit risk models similar to
loans (Canabarro, Picoult and Wilde 2003). These two types of
models are the hallmark of treating CCR as a credit risk. Pykhtin
and Zhu (2007) provide an introduction to these models. The treat-
ment of CCR as credit risk was the predominant framework for
measuring and managing CCR from 2000 to 2006 and was estab-
lished as the basis for regulatory capital as part of Basel II (BCBS
2005). During this time, risk mitigants such as netting agreements
and margining were incorporated into the modelling of CCR. The
definitions of these exposure measures used in this chapter follow
those in BCBS (2005).

o Current exposure is the larger of zero and the market value of a

transaction or portfolio of transactions within a netting set, with a
counterparty that would be lost upon the default of the counter-
party, assuming no recovery on the value of those transactions in
bankruptcy. Current exposure is often also called replacement cost.
o Peak exposure is a high-percentile (typically 95% or 99%) of the
distribution of exposures at any particular future date before the
maturity date of the longest transaction in the netting set. A peak
exposure value is typically generated for many future dates up
until the longest maturity date of transactions in the netting set.

38

stress_testing.indd 38 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

o Expected exposure is the mean (average) of the distribution of

exposures at any particular future date before the longest-matu-
rity transaction in the netting set matures. An expected exposure
value is typically generated for many future dates up until the
longest maturity date of transactions in the netting set.
o Expected positive exposure (EPE) is the weighted average over
time of expected exposures where the weights are the propor-
tion that an individual expected exposure represents of the entire
time interval. When calculating the minimum capital require-
ment, the average is taken over the first year or over the time
period of the longest-maturity contract in the netting set.

Furthermore, an unusual problem associated with CCR, that of

wrong-way risk, has been identified (Levin and Levy 1999; Finger
2000). Wrong-way risk occurs when the credit quality of the coun-
terparty is correlated with the exposure, so that exposure grows
when the counterparty is most likely to default. When exposure is
fixed, as is the case for a loan, this does not occur, so adaptation of
techniques used in other areas of risk management is more difficult.
At the same time, the treatment of CCR as a market risk was
developing, but was largely relegated to pricing in a credit valua-
tion adjustment (CVA), prior to the financial crisis of 2007–9. This
was first described for Swaps (Sorensen and Bollier 1994; Duffie
and Huang 1996) and has since become widespread due to the ac-
counting requirement of FAS 157 (FASB 2006). The complexities of
risk-managing this price aspect of a derivatives portfolio did not
become apparent until the crisis. Prior to the crisis, credit spreads
for financial institutions were relatively stable and the CVA was a
small portion of the valuation of banks’ derivatives portfolios. Dur-
ing the crisis, both credit spreads and exposure amounts for deriva-
tive transactions experienced wide swings, and the combined ef-
fect resulted in both large losses and large, unusual gains. Financial
institutions are just now beginning to develop their frameworks to
risk-manage CVA. The regulatory capital framework has adopted a
CVA charge to account for this source of risk (BCBS 2011).
The treatment of CCR as a credit risk or CCR as a market risk has
implications for the organisation of a financial institution’s trading
activities and the risk-management disciplines (Picoult 2005; Cana-

39

stress_testing.indd 39 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

barro 2009). Both treatments are valid ways to manage the portfolio,
but adoption of one view alone leaves a financial institution blind
to the risk from the other view. If CCR is treated as a credit risk, a
bank can still be exposed to changes in CVA. A financial institution
may establish PFE limits and manage its default risk through collat-
eral and netting, but it still must include CVA in the valuation of its
derivatives portfolio. Inattention to this could lead to balance-sheet
surprises. If CCR is treated as a market risk, dynamically hedging
its CVA to limit its market risk losses, it remains exposed to large
drops in creditworthiness or the sudden default of one of its coun-
terparties. A derivatives dealer is forced to consider both aspects.
The view of CCR has implications for how the risk is managed
as well. The traditional credit risk view is that the credit risk of the
counterparty can be managed at inception or through collateral ar-
rangements set up in advance, but there is little that can be done
once the trades are in place. At default the financial institution must
replace the trades of the defaulting counterparty in the market all
at once in order to rebalance its book. A large emphasis is placed on
risk mitigants and credit evaluation as a result.
The view of CCR as a market risk allows that its counterparty
credit risk can be hedged. Instead of waiting until the counterparty
defaults to replace the contracts, the financial institution will replace
the trades with a counterparty in the market before it defaults by
buying the positions in proportion to the counterparty’s probability
of default. Thus a counterparty with a low probability of default will
have little of its trades replaced in advance by the financial institu-
tion, but, as its credit quality deteriorates, a larger proportion of those
trades will be replaced by moving them to other counterparties. At
default, the financial institution will have already replaced the trades
and the default itself would be a non-event.

Implications for stress testing

The dual nature of CCR leads to many measures that capture some
important aspect of CCR. On the credit risk side, there are the im-
portant measures of exposure: current exposure, peak exposure
and expected exposure. On the market risk side there is the valu-
ation aspect coming from CVA, and there is the risk generated by
changes in the CVA, as measured by VaR of CVA, for example. This

40

stress_testing.indd 40 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

creates a dazzling array of information that can be difficult to inter-

pret and understand at both portfolio and counterparty levels. The
search for a concise answer to the question “What is my counter-
party credit risk?” is difficult enough, but an equally difficult ques-
tion is “What CCR measures should I stress?”
When confronted with the question of stress testing for CCR, the
multiplicity of risk measures means that stress testing is a compli-
cated endeavour. To illustrate this complexity we can compare the
number of stresses that a bank may run on its market risk portfolio
with the number of similar stresses a bank would run on its coun-
terparty credit risk portfolio. In market risk, running an equity crash
stress test may result in one or two stress numbers: an instantaneous
loss on the current portfolio and potentially a stress VaR loss. A risk
manager can easily consider the implications of this stress.
In contrast, the CCR manager would have to run this stress at
the portfolio level and at the counterparty level, and would have to
consider CCR as both a credit risk and a market risk. The number
of stress-test results would be at least twice the number of counter-
parties plus one.1 The number of stress-test results would at least
double again if the risk manager stressed risk measures in addition
to considering instantaneous shocks.2 The number of stress values
that can be produced can bewilder even the most diligent risk man-
ager, and overwhelm IT resources.
Despite this array of potential stress results, a risk manager must
stress-test counterparty exposures to arrive at a comprehensive view
of the risk of the financial institution’s portfolio.3 This chapter pro-
vides a description of the types of stress tests that can be run to get a
picture of the CCR in a financial institution’s derivative portfolio.

Stress testing current exposure

The most common stress tests used in counterparty credit are stress-
es of current exposure. To create a stressed current value, the bank as-
sumes a scenario of underlying risk-factor changes and reprices the
portfolio under that scenario. Generally speaking, a financial institu-
tion applies these stresses to each counterparty. It is common practice
for banks to report their top counterparties with the largest current
exposure to senior management in one table, and then follow that
table with their top counterparties, with the largest stressed current

41

stress_testing.indd 41 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

exposure placed under each scenario in separate tables.

For example, Table 4.1 shows an example of what a financial in-
stitution’s report on its equity crash stress test for current exposure
might look like. The table lists the top 10 counterparties by their
exposure to an equity market crash of 25%. It shows the following
categories: the counterparty rating, market value of the trades with
the counterparty, collateral, current exposure, and stressed current
exposure after the stress is applied but before any collateral is col-
lected. This provides a snapshot of which counterparties a CCR
manager should be concerned about in the event of a large drop
in equity markets. A financial institution would construct similar
tables for other stresses representing credit events or interest-rate
shocks. These tables would likely list different counterparties as be-
ing exposed to the stress scenario, since it is unlikely that the coun-
terparty with the most exposure to an equity crash is the same as
the counterparty with the most exposure to a shock in interest rates.

Table 4.1 Current exposure stress test: equity crash

Scenario: Equity market down 25%

Stressed
Current Current
($MM) Rating MtM Collateral Exposure Exposure

Counterparty A A 0.5 0 0.5 303

Counterparty B AA 100 0 100 220

Counterparty C AA 35 0 35 119

Counterparty D BBB 20 20 0 76

Counterparty E BBB 600 600 0 75

Counterparty F A -5 0 0 68

Counterparty G A -10 0 0 50

Counterparty H BB -50 0 0 24

Counterparty I A 35 20 15 17

Counterparty J BB 24 24 0 11

42

stress_testing.indd 42 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

This type of stress testing is quite useful, and financial institutions

have been conducting it for some time. It allows the bank to identi-
fy which counterparties would be of concern in such a stress event,
and also how much the counterparty would owe the financial insti-
tution under the scenario. However, stress tests of current exposure
has a few problems. First, aggregation of the results is problematic,
and, second, it does not account for the credit quality of the coun-
terparties. Also, it provides no information on wrong-way risk.
While the individual counterparty results are meaningful, there is
no meaningful way to aggregate these stress exposures without in-
corporating further information. If we were to sum the exposures to
arrive at an aggregate stress exposure, this would represent the loss
that would occur if every counterparty defaulted in the stress sce-
nario. Unless the scenario were the Apocalypse, this would clearly
be an exaggeration of the losses. Other attempts to aggregate these
results are also flawed. For example, running the stressed current ex-
posure through a portfolio credit risk model would also be incorrect,
since expected exposures, not current exposures, should go through
a portfolio credit risk model (Canabarro, Picoult, Wilde 2003). Table
4.1 does not provide an aggregate stressed amount as a result.
The stressed current exposures also do not take into account the
credit quality of the counterparty. This should be clear from the
outset, since it accounts only for the value of the trades with the
counterparty and not the counterparty’s willingness or ability to
pay. This is an important deficiency since a US$200 million expo-
sure to a start-up hedge fund is very different from a US$200 mil-
lion exposure to an AAA corporate. While we could imagine a limit
structure for stressed current exposure that takes into account the
credit quality of the counterparty, most financial institutions have
not gone down this path for stressed current exposure. The degree
of difficulty involved in doing this for each scenario and each rating
category is daunting, mostly because the statistical measures such
peak exposure provide a more consistent way to limit exposure by
counterparties who may be exposed to different scenarios. From
Table 4.1, it is unclear whether the CCR manager should be more
concerned about Counterparty C or Counterparty D in the stress
event. While Counterparty C has a larger stressed current exposure
than Counterparty D, Counterparty C has a better credit quality.

43

stress_testing.indd 43 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Last, stress tests of current exposure provide little insight into

wrong-way risk. As a measure of exposure that omits the credit
quality of the counterparty, these stress tests without additional
information cannot provide any insight into the correlation of ex-
posure with credit quality. Stresses of current exposure are useful
for monitoring exposures to individual counterparties, but do not
provide either a portfolio outlook or incorporate a credit quality.

Stress testing the loan equivalent

To stress-test in the credit framework for CCR, we first have to de-
scribe a typical stress test that would be performed on a loan port-
folio. The typical framework for loans is to analyse how expected
losses would change under a stress.
For credit provisioning, we might look at an unconditional expected
loss across a pool of loan counterparties. Expected loss for any one
counterparty is the product of the probability of default, pi, where this
may depend on other variables, exposure at default, eadi, and loss-giv-
en default, lgdi. The expected loss for the pool of loan counterparties is:

= ∑ . .
=1
A stress test could take exposure at default and loss-given default as
deterministic and focus on stresses where the probability of default
is subject to a stress. In this case, the probability of default is taken
to be a function of other variables; these variables may represent
an important exchange rate or an unemployment rate, for example.
In this case, the stressed expected loss is calculated conditional on
some of the variables affecting the probability of default being set to
their stressed values; the stressed probability of default is denoted
pis ; and the stressed expected loss is:
= ∑ =1 . .

The stress loss for the loan portfolio is ELs-EL. A financial institu-
tion can generate stress tests in this framework rather easily. It can
simply increase the probability of defaults, or it can stress the vari-
ables that these probabilities of defaults depend on. These variables
are typically macroeconomic variables or balance-sheet items for
the counterparty. The stress losses can be generated for individual

44

stress_testing.indd 44 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

loan counterparties as well as at an aggregate level.

This framework can be adapted for CCR treated as a credit risk. In
this case the probability of default and loss-given default of the coun-
terparty are treated the same, but now exposure at default is stochastic
and depends on the levels of market variables. EPE multiplied by an
alpha factor (Picoult 2005; Wilde 2005) is the value that allows CCR
exposures to be placed in a portfolio credit model along with loans and
arrive at a high-percentile loss for the portfolio of exposures (both loan
and derivatives).4 The same procedure is applied here and EPE is used
in an expected-loss model. In this case expected loss and expected loss
conditional on a stress for derivatives counterparties are:

.
= ∑ . .
=1

= ∑ =1 . . .

Stress losses on the derivatives portfolio can be calculated similarly

to the loan portfolio case. A financial institution can stress the prob-
ability of default similarly to the loan case by stressing probability
of default or the variables that affect probability of default, includ-
ing company balance-sheet values, macroeconomic indicators and
values of financial instruments. It can also combine the stress losses
on the loan portfolio and the stress losses on its derivatives portfo-
lio by adding these stress losses together.
Table 4.2 shows the results of a typical stress test that could be
run that would shock the probability of default of counterparties in
a derivatives portfolio. The stress test might parallel the increase in
PD by industry after the dotcom crash in 2001–2. The expected loss,
stressed expected loss and the stress loss may all be aggregated and
even combined with similar values from the loan portfolio.
In addition, a financial institution has a new set of variables to
stress. Exposure, as measured by EPE, depends on market vari-
ables such as equity prices and swap rates. A financial institution
can stress these market variables and see their impact. It should be
noted that it is not clear whether a stress will, in aggregate, increase
or decrease expected losses. This will depend on a whole host of
factors, including the directional bias of the bank’s portfolio, which
counterparties are margined and which have excess margin. This is

45

stress_testing.indd 45 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

in marked contrast to the case where stresses of the probabilities of

default are considered. Stresses to the variables affecting the prob-
ability of default generally have similar effects and the effects are in
the same direction across counterparties. When conducting stresses
to EPE, a bank need not consider aggregation with its loan portfo-
lio.5 Loans are insensitive to the market variables and thus will not
have any change in exposure due to changes in market variables.
There are a whole host of stresses that can be considered. Typically
a financial institution will use an instantaneous shock of market vari-
ables, these are often the same current exposure shocks from the previ-
ous section. In principle, we could shock these variables at some future
point in their evolution or create a series of shocks over time. This is not
common, however, and shocks to current exposure are the norm. In the
performance of these instantaneous shocks, the initial market value of
the derivatives is shocked prior to running the simulation to calculate
EPE. How this shock affects EPE depends on the degree of collateralisa-
tion and the “moneyness” of the portfolio, among other things.
Table 4.3 shows how a financial institution might reconsider its
stress test of current exposure in an expected-loss framework. Now, in
addition to considering just current exposure, the financial institution
must consider including the probability of default over the time hori-
zon and the expected positive exposure in its stress-test framework. In
this case we are looking at changes to current exposures and thus EPE.
We hold the PD constant here. The expected loss, even under stress, is
small and measured in thousands. This is due to the rather small prob-
abilities of default that we are considering. We are able to aggregate
expected losses and stress losses by simply adding them up.
A financial institution can consider joint stresses of credit quality
and market variables as well. Conceptually, this is a straightforward
exercise, but, in practice, deciding how changes in macroeconomic
variables or balance-sheet variables are consistent with changes in
market variables can be daunting. There is very little that neces-
sarily connects these variables. Equity-based approaches (Merton
1974; Kealhofer 2003) come close to providing a link; however, it
remains unclear how to link an instantaneous shock of exposure
to the equity-based probability of default. While exposure can and
should react immediately, it is unclear whether equity-based prob-
abilities of default should react so quickly.

46

stress_testing.indd 46 20/05/2013 18:45

stress_testing.indd 47

Table 4.2 PD stress: dotcom crash

Stressed PD Stressed EL Stress loss
PD (%) EPE (US$m) LGD (%) EL (US$m) (%) (US$m) (US$m)

Counterparty AA 0.05 213.00 0.70 0.08 0.50 0.77 0.69

Counterparty BB 0.03 202.50 0.60 0.04 0.30 0.38 0.34

Counterparty CC 0.45 75.00 0.70 0.24 0.62 0.34 0.09

Counterparty DD 0.90 30.00 0.65 0.18 1.20 0.24 0.06

The Evolution of Stress Testing Counterparty Exposures

Counterparty EE 1.05 10.00 0.75 0.08 1.40 0.11 0.03

Counterparty FF 0.09 157.00 0.50 0.07 0.12 0.10 0.02

Counterparty GG 0.98 68.00 0.70 0.48 1.02 0.50 0.02

Counterparty HH 2.17 3.00 0.34 0.02 3.00 0.03 0.01

Counterparty II 0.03 150.00 0.20 0.01 0.05 0.02 0.01

Counterparty JJ 0.50 50.00 0.60 0.15 0.50 0.15 0.00

Aggregate 1.36 2.63 1.27

47
20/05/2013 18:45
stress_testing.indd 48

48

Stress Testing: Approaches, Methods and Applications

Table 4.3 Expected-loss stress test in a credit framework
Scenario: Equity market down 25%
MtM Collateral EPE EL stress EPE stress EL stress loss
PD (%) (US$m) (US$m) CE (US$m) (US$m) (US$000) (US$m) (US$000) (US$000)

Counterparty A 0.03 0.5 0 0.5 4.37 0.09 303.00 6.09 6.00

Counterparty B 0.02 100 0 100 100.00 1.34 220.00 2.95 1.61

Counterparty C 0.02 35 0 35 35.16 0.47 119.00 1.59 1.12

Counterparty D 0.18 20 20 0 3.99 0.48 76.00 9.16 8.68

Counterparty E 0.18 600 600 0 3.99 0.48 75.00 9.04 8.56

Counterparty F 0.03 -5 0 0 2.86 0.06 68.00 1.37 1.31

Counterparty G 0.03 -10 0 0 1.98 0.04 50.04 1.00 0.96

Counterparty H 1.2 -50 0 0 0.02 0.02 25.12 19.73 19.72

Counterparty I 0.03 35 20 15 16.31 0.33 19.20 0.36 0.04

Counterparty J 0.12 24 24 0 3.99 0.32 14.66 1.03 0.71

aggregate 3.62 52.32 48.70

20/05/2013 18:45
 The Evolution of Stress Testing Counterparty Exposures

This leads to another drawback: the difficulty of capturing the con-

nection between the probability of default and exposure that is
often of concern in CCR. There are many attempts to capture the
wrong-way risk, but most are ad hoc. At present the best approach
to identifying wrong-way risk in the credit framework is to stress
the current exposure, identify those counterparties that are most
exposed to the stress and then carefully consider whether the coun-
terparty is also subject to wrong-way risk.
Stress tests of CCR as a credit risk allow a financial institution
to advance beyond simple stresses of current exposure. They al-
low aggregation of losses with loan portfolios, and also allow con-
sideration of the quality of the counterparty. These are important
improvements that allow a financial institution to better manage
its portfolio of derivatives. Treating CCR as a market risk allows
further improvements (notably, the probability of default will be
inferred from market variables), and it will be easier to consider
joint stresses of credit quality and exposure.

Stress testing CVA

When stress testing CCR in a market risk context, we are usually
concerned with the market value of the counterparty credit risk
and the losses that could result due to changes in market variables,
including the credit spread of the counterparty. In many cases a
financial institution will consider its unilateral CVA for stress test-
ing. Here, the financial institution is concerned with the fact that
its counterparties could default under various market scenarios. In
addition, we might consider not only that a financial institution’s
counterparty could default, but also that the financial institution in
question could default to its counterparty. In this case, the financial
institution is considering its bilateral CVA. Initially we just consider
stress testing the unilateral CVA.
First we use a common simplified formula for CVA to a counter-
party that omits wrong-way risk (Gregory 2010).

= ∗ . ∑ ∗
( ). ∗
( −1 , )
=1

Where:

49

stress_testing.indd 49 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

𝐸𝐸𝑛∗ (𝑡𝑗 ) is the discounted expected exposure during the jth time
period calculated under a risk-neutral measure for counterparty n.

𝑞𝑛∗ (𝑡𝑗−1 , 𝑡𝑗 ) is the risk-neutral marginal default probability for coun-

terparty n in the time interval from 𝑡𝑗−1to 𝑡𝑗 and T is the final maturity.

𝐿𝐺𝐷𝑛∗ is the risk-neutral loss-given default for counterparty n.

Aggregating across N counterparties:

= ∑ ∗ . ∑ ∗
( ). ∗
( −1 , )
=1 =1

Implicit in this description is that the key components all depend

on values of market variables. 𝑞𝑛∗ (𝑡𝑗−1 , 𝑡𝑗 ) is derived from credit
spreads of the counterparty, 𝐿𝐺𝐷𝑛∗ is generally set by convention or
from market spreads and 𝐸𝐸𝑛∗(𝑡𝑗 ) depends on the values of deriva-
tive transactions with the counterparty. To calculate a stressed CVA
we would apply an instantaneous shock to some of these market
variables. The stresses could affect 𝐸𝐸𝑛∗(𝑡𝑗 ) or 𝑞𝑛∗ (𝑡𝑗−1 , 𝑡𝑗 ).

Stressed CVA is given by:

= ∑ ∗
∙ ∑ ( )∙ ( −1 , )
=1 =1

And the stress loss is CVAs-CVA.

Stressing current exposure, as described previously, has similar ef-

fects. An instantaneous shock will have some impact on the expected
exposure calculated in later time periods, so all of the expected expo-
sures will have to be recalculated. Stresses to the marginal probabil-
ity of default are usually derived from credit spread shocks.
Similarities can be seen between stress testing CCR in a credit risk
framework and doing so in market risk framework. There is a reli-
ance in both cases on expected losses being the product of loss-given
default, exposure and the probability of default. However, these val-
ues will be quite different, depending on the view of CCR as a mar-

50

stress_testing.indd 50 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

ket risk or credit risk. The reasons for the differences are many, and
the use of risk-neutral values for CVA as opposed to physical values
for expected losses is the most prominent. In addition, CVA uses
expected losses over the life of the transactions, whereas expected
losses use a specified time horizon, and the model for determining
the probability of default is market-based in CVA.
Using a market-based measure for the probability of default pro-
vides some benefits. It is possible in these circumstances to incorpo-
rate a correlation between the probability of default and the expo-
sure. Hull and White (2012) describe methods to do this. They also
demonstrate an important stress test that is available, a stress of the
correlation between exposure and the probability of default. They
show that the correlation can have an important effect on the mea-
sured CVA. Since there is likely to be a high degree of uncertainty
around the correlation, a financial institution should run stress tests
to determine the impact on profit and loss if the correlation is wrong.
To capture the full impact of various scenarios on CVA profit and
loss, a financial institution should include the liability side effects
in the stress as well. This part of the bilateral CVA (BCVA), often
called DVA, captures the value of the financial institution’s option
to default on its counterparties. The formula for DVA is similar to
the formula for CVA except for two changes. First, instead of ex-
pected exposure, we have to calculate the negative expected expo-
sure (NEE). This is expected exposure calculated from the point of
view of the counterparty. Second, the value of the option to default
for the financial institution is dependent on the survival of the
counterparty, so the probability that the counterparty has survived
must enter into the calculation as SI. A similar change must be made
to the CVA portion, since the loss due to the counterparty default-
ing now depends on the financial institution not defaulting first.
The bilateral CVA formula is (Gregory 2010):

= ∑ ∗ . ∑ ∗
( ). ∗
( −1 , ). ∗
( −1 ) − ∑ ∗ . ∑ ∗
( ). ∗
( −1 , ).
=1 =1 =1 =1

∗
( ). ∗
( −1 , ). ∗
( −1 ) − ∑ ∗ . ∑ ∗
( ). ∗
( −1 , ). ∗
( −1 )
=1 =1

51

stress_testing.indd 51 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

The subscript I refers to the financial institution. Notable in this

formulation is that the survival probabilities also depend on CDS
spreads and now the losses depend on the firm’s own credit spread.
This may lead to counterintuitive results such as losses occurring
because the firm’s own credit quality improves. When looking at
stress tests from a bilateral perspective, the financial institution will
also have to consider how its own credit spread is correlated with
its counterparties’ credit spread. Stress losses can be calculated in
a similar way as for CVA losses by calculating a stress BCVA and
subtracting the current BCVA.
BCVA allows CCR to be treated as a market risk. This means CCR
can be incorporated into market risk stress testing in a coherent
manner. The gains or losses from the BCVA stress loss can be add-
ed to the firm’s stress tests from market risk. As long as the same
shocks to market variables are applied to the trading portfolio and
to the BCVA results, they can be aggregated by simple addition.

Common pitfalls in stress testing CCR

Financial institutions are only beginning to conduct a level of stress-
testing beyond stressing current exposure. The methodologies to
conduct these tests are only just being developed. It is also rare for
CCR to be aggregated with either stress tests of the loan portfolio
or with trading-position stress testing results in a consistent frame-
work. With better modelling of CCR exposures and CVA, it is pos-
sible to begin aggregating stress tests of CCR with either the loan
portfolio or trading positions.
Since most financial institutions will do some form of stressing
current exposure, it is tempting to use those stresses of current ex-
posure when combining the losses with loans or trading positions.
The analysis above shows that expected exposure or expected posi-
tive exposure should be used as the exposure amount, and that us-
ing current exposure instead would be a mistake.
In fact, the use of current exposure instead of expected exposure
can lead to substantial errors. This can be shown using a normal
approximation (Gregory 2010) to expected exposures, which is ac-
curate for linear derivatives with no intermediate payments. Figure
4.1 plots current exposure and expected exposure after a million-
dollar shock to the market value of the derivative. For at-the-money

52

stress_testing.indd 52 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

exposures, the difference between current exposure and expected

exposure is almost half the value of the shock.

Figure 4.1 Current exposure and expected exposure after US$1m shock

0.8

0.6

Postshock exposure
0.4

0.2

0
-6 -4 -2 0 2 4 6
change in MtM change in CE change in EE
Starting MtM

Use of delta sensitivities to calculate changes in exposures is also

especially problematic for CCR, since it is highly nonlinear. While
this can save on computational resources, the errors introduced are
not obvious and the linearisation can be highly misleading. At-the-
money portfolios with large price moves applied to the portfolio
are especially prone to errors from using delta approximations.

Conclusion
A counterparty credit risk manager now has a multiplicity of stress
tests to consider. Too many stress tests can hide the risk of a portfo-
lio, but a fair number of stresses is important to develop a compre-
hensive view of the risks in the portfolio. Both the credit risk and
market risk views are important since both fair-value losses and de-
fault losses can occur no matter how a financial institution manages
its CCR. More integrated stress tests can be generated by combining

53

stress_testing.indd 53 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

the credit risk view with the loan portfolio, or the market risk view
of CCR can be combined with the trading book. The true difficulty
remains combining the default stresses and the fair-value stresses
to get a single comprehensive stress test. This difficulty aside, coun-
terparty credit risk managers now have more tools at their disposal
to measure and manage CCR. The irony is that regulators have be-
gun to move derivative transactions to central clearing to reduce
the counterparty credit risk problem just as the ability to manage
counterparty credit risk is making major advances.

The views expressed in this article are the author’s own and do not represent
the views of the Board of Governors of the Federal Reserve System or its staff.

REFERENCES
Basel Committee on Banking Supervision, 1988, “The International Convergence
of Capital Measurement and Capital Standards” July.

Basel Committee on Banking Supervision, 2005, “The Application of Basel II to

Trading Activities and the Treatment of Double Default Effects” July.

Basel Committee on Banking Supervision, 2011, “Basel III: A Global Regulatory

Framework for More Resilient Banks and Banking Systems” June.

Canabarro, E., 2009, “Pricing and Hedging Counterparty Risk: Lessons Re-
learned?”, in Canabarro, E., Counterparty Credit Risk Measurement, Pricing and
Hedging (London: Risk Books).

Canabarro, E., E. Picoult and T. Wilde, 2003, “Analyzing Counterparty Risk”, Risk
16(9), pp. 117–22.

Counterparty Risk Management Policy Group I, 1999, “Improving Counterparty

Risk Management Practices”, June.

Counterparty Risk Management Policy Group II, 2005, “Toward Greater Finan-
cial Stability: A Private Sector Perspective”, July.

Duffie, D., and M. Huang, 1996, “Swap Rates and Credit Quality”, Journal of
Finance 51, pp. 921–49.

Federal Reserve Board, 2011, “Interagency Counterparty Credit Risk Manage-

ment Guidance”, SR 11-10, July.

Financial Accounting Standards Board, 2006, “Statement of Financial Accounting

Standards No. 157 – Fair Value Measurements”, September.

54

stress_testing.indd 54 20/05/2013 18:45

 The Evolution of Stress Testing Counterparty Exposures

Finger, C., 2000, “Toward a Better Estimation of Wrong-Way Credit Exposure”,

Journal of Risk Finance 1(3), pp. 43–51.

Gregory, J., 2010, Counterparty Credit Risk: The New Challenge for Global Financial
Markets (London: John Wiley and Sons).

Hull, J., and A. White, 2012, “CVA and Wrong Way Risk”, Financial Analysts Jour-
nal 68(5), September–October, pp. 38–56.

Kealhofer, S., 2003, “Quantifying Credit Risk I: Default Prediction”, Financial

Analysts Journal, January–February, pp. 30-=44.

Levin, R., and A. Levy, 1999, “Wrong Way Exposure – Are Firms Underestimating
Their Credit Risk?”, Risk, July, pp. 52–5.

Merton, R. C., 1974, “On the Pricing of Corporate Debt: The Risk Structure of
Interest Rates”, Journal of Finance 29, pp. 449–70.

Picoult, E., 2005, “Calculating and Hedging Exposure, Credit Valuation Adjust-
ment, and Economic Capital for Counterparty Credit Risk”, in Pykhtin, M., Coun-
terparty Credit Risk Modelling: Risk Management, Pricing and Regulation (London:
Risk Books).

Pykhtin, M., and S. Zhu, 2007 “A Guide to modelling counterparty Credit Risk”,
GARP Risk Review, July–August.

Sorensen, E., and T. Bollier, 1994, “Pricing Swap Default Risk”, Financial Analysts
Journal 50, May–June, pp. 23–33.

Wilde, T., 2005, “Analytic Methods for Portfolio Counterparty Credit Risk”, in
Pykhtin, M., Counterparty Credit Risk Modelling: Risk Management, Pricing and
Regulation (London: Risk Books).

1 The stresses are run for each counterparty and at the aggregate portfolio level. The stress
may also be run for various subportfolios, divided by region or industry, for example. These
would have to be run in both a credit and market risk context.
2 It might increase even more since there are multiple risk measures of importance in CCR.
3 This is included in regulatory guidance on stress testing for counterparty credit risk, for
example in SR 11-10 (Federal Reserve Board 2011).
4 Alpha typically depends on the quantile at which we measure economic capital. In this case
it would be the alpha calculated at the expected loss. For this reason it may differ from the
alpha used for economic or regulatory capital calculations.
5 Although exposure for loans is insensitive to market variables for the most part, there can
still be some increase in expected losses if probabilities of default are correlated with mar-
ket variables. Furthermore, loan commitments and some other loan products can have a
stochastic exposure.

55

stress_testing.indd 55 20/05/2013 18:45

stress_testing.indd 56 20/05/2013 18:45
 5
Operational Risk: An Overview
of Stress-Testing Methodologies

Brian Clark; Bakhodir Ergashev

Office of the Comptroller of Currency; Federal Reserve Bank of Richmond

Numerous international regulatory standards require the imple-

mentation of stress testing as a risk-management tool. The Basel
Committee on Banking Supervision (2009) (henceforth BCBS), a key
international regulatory guidance on stress testing, recommends
including stress tests in a bank’s overall risk-management toolkit.
The document broadly refers to stress tests as “the evaluation of a
bank’s financial position under a severe but plausible scenario”. It
provides general principles for stress testing practices, while allow-
ing banks ample discretion in choosing stress-test methodologies.
However, the document refrains from prescribing any particular
stress-testing approach, thereby leaving banking institutions with
broad discretion in choosing stress-testing methodologies.
The purpose of stress testing is often viewed by regulatory bodies
and financial institutions as a means to determine how a financial
institution’s capital or financial position would be impacted on by an
adverse scenario. In most applications, this requires modelling a link
between a macroeconomic event or series of macroeconomic events
and the performance of a bank’s portfolio of assets. In the context of
credit and market portfolios, this is often an extension of the models
banks already have in place to model and manage the risk of these
portfolios because they often have factor models in place.
The state-of-the-art operational risk models, however, do not
readily lend themselves to similar stress-testing applications be-
cause they tend to be less risk-sensitive compared with most other

57

stress_testing.indd 57 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

types of risk models. In other words, correlations between opera-

tional risk losses and the macroeconomic or internal business envi-
ronment variables are generally not explicitly modelled. This poses
obvious challenges for stress testing operational risk. As we expand
on below, the primary reasons for this include the relatively short
history of operational risk modelling, limited historical loss data
and the fact that most operational risk models have been developed
with the explicit purpose of capital modelling.1
In this chapter, we begin with a brief description of the evolu-
tion of operational risk models to provide the reader with a better
understanding of the unique challenges inherent to modelling and
thus stress testing operational risk. We then describe the implica-
tions of these unique challenges to stress testing and provide sev-
eral potential solutions.

MODELLING OPERATIONAL RISK

Although modelling operational risk is a relatively new discipline
for banks, it is an important part of the Basel Final Rule (2007). The
Final Rule allows banks to use internally developed, risk-based ap-
proaches to measure credit and operational risk. For operational risk,
the Rule allows banks to follow an Advanced Measurement Ap-
proach (AMA) and states the following (Final Rule 2007, p. 69294):

Given the complexities involved in measuring operational risk, the

AMA provides banks with substantial flexibility and, therefore, does
not require a bank to use specific methodologies or distributional
assumptions. Nevertheless, a bank using the AMA must demonstrate
to the satisfaction of its primary Federal supervisor that its systems
for managing and measuring operational risk meet established stan-
dards, including producing an estimate of operational risk exposure
that meets a one-year, 99.9th percentile soundness standard.

The Rule also specifies that banks incorporate four elements into
the capital quantification model: (1) internal loss data, (2) external
loss data, (3) scenario analysis and (4) business environment and
internal control factors (BEICFs). Despite the fact that the Final Rule
effectively does not specify how banks must measure or model ex-
posure to operational risk, most large US banks have settled on the
loss-distributional approach (LDA) as the primary quantification

58

stress_testing.indd 58 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

methodology. Given this widespread use of the LDA across the in-
dustry, we now describe the basic approach.

The loss-distribution approach

The goal of the LDA is to model an aggregate loss distribution in
a value-at-risk (VaR) framework. The aggregate loss distribution is
defined as the total dollar amount of operational risk losses over a
given time horizon. Banks typically use historical loss data to sepa-
rately model severity and frequency distributions and then calcu-
late the aggregate loss distribution through a convolution of these
two distributions. In a VaR application, the bank would define a
capital charge as a high percentile of this distribution. As an ex-
ample, the Final Rule (2007) defines the 99.9th percentile of the ag-
gregate loss distribution over a one-year horizon as the minimum
regulatory capital requirement.
Because of the complexity of operational risk and the various
types of loss events that are defined as operational losses, banks
tend to use the LDA to model operational risk for various risk cells,
which are commonly referred to as units of measure. Each unit of
measure is meant to capture the operational risk losses stemming
from a homogeneous data-generating process. Although the specif-
ic unit-of-measure definitions vary across banks, most define units
of measure along business lines and event types. In the Final Rule,
BCBS (pages 69314–15) explicitly defines seven event types:

1. INTERNAL FRAUD: Operational losses resulting from an act in-

volving at least one internal party of a type intended to defraud,
misappropriate property or circumvent regulations, the law or
company policy, excluding diversity and discrimination.
2. EXTERNAL FRAUD: Operational losses resulting from an act
by a third party of a type intended to defraud, misappropriate
property or circumvent the law.
3. EMPLOYMENT PRACTICES AND WORKPLACE SAFETY: Op-
erational losses resulting from an act inconsistent with employ-
ment, health or safety laws or agreements, payment of personal
injury claims or payment arising from diversity or discrimina-
tion events.

59

stress_testing.indd 59 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

4. CLIENTS, PRODUCTS, AND BUSINESS PRACTICES: Opera-

tional losses resulting from the nature or design of a product or
from an unintentional or negligent failure to meet a professional
obligation to specific clients (including fiduciary and suitability
requirements).
5. DAMAGE TO PHYSICAL ASSETS: Operational losses resulting
from the loss of or damage to physical assets from natural disas-
ters or other events.
6. BUSINESS DISRUPTION AND SYSTEM FAILURES: Operation-
al losses resulting from disruption of business or system failures.
7. EXECUTION, DELIVER AND PROCESS MANAGEMENT: Op-
erational losses resulting from failed transaction processing or
process management or losses arising from relationships with
trade parties or vendors.

In practice, banks have anywhere from a handful to the order of

several dozen units of measure. Even with a careful choice of units
of measure, a bank may not be able to fully achieve the homogene-
ity within each unit. The main reasons for the remaining heteroge-
neity include data limitations and the size and scope of the bank.
Regardless of the number of units of measure, banks are adopting
the LDA approach independently to model each risk cell.
Because it is a common belief that losses of different units of mea-
sure might exhibit certain levels of dependence, a bank’s overall
capital charge should be aggregated from individual capital charg-
es of its units of measure through, for example, the use of a copula.
Copulas are designed to capture various dependence structures
within marginal distributions (ie, the aggregate loss distributions
of individual units of measure) of a continuous multivariate distri-
bution (ie, the overall aggregate loss distribution at the bank level).
Once an appropriate dependence structure is established with the
use of copulas, the operational risk capital charge would then be
equivalent to the 99.9th percentile of the overall aggregated loss
distribution in the case of the Basel Final Rule.
While the LDA may be an appropriate framework for modelling
operational risk capital, especially at the high percentile the Final
Rule requires, it is often criticised for its lack of sensitivity to risk
factors. Furthermore, since the standard LDA framework does not

60

stress_testing.indd 60 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

explicitly model the link between macroeconomic variables and op-

erational risk losses, it poses obvious challenges for macro stress-
testing operational risk. Without going into the details of these
criticisms, suffice it to say at this point that the general framework
poses several challenges for stress testing operational risk, which
we expand on below.

Challenges to modelling operational risk

Regardless of the methodology used, practitioners are faced with sig-
nificant challenges when modelling operational risk. Several of these
challenges are unique to operational risk because of the nature of
operational losses, and many are due to the relatively short history
of modelling operational risk. In any case, each of the following chal-
lenges impacts on how we can stress-test operational risk.
The prevailing majority of risk models employed in measuring
operational risk are statistical models that are static in nature. Dy-
namic factor models that are supposed to link operational risk with
its drivers are still, at the time of writing, under development. Since
static statistical models are extreme simplifications of real-world
relationships, they are not capable of capturing the true nature of
risk. Below, we list several reasons why dynamic models are still in
the development phase and, in doing so, outline the challenges of
modelling operational risk in general.
First, banks have a limited amount of operational risk data, es-
pecially large-tail events that tend to drive operational risk capi-
tal. This creates challenges, as even the largest US institutions have
been collecting reliable and comprehensive operational risk data
only since about 2000. Adding even more complexity is the fact that
most banks collect detailed operational loss data only above a cer-
tain collection threshold and thus are left with censored data, which
makes fitting severity distributions all the more challenging.
Second, because the LDA approach originated in the insurance
and actuarial industries, it was developed with the explicit intent of
modelling maximum exposure or worst-case scenarios. Similarly,
in operational risk applications, the LDA approach is commonly
used for capital-modelling purposes and thus the goal is to fit the
tail of the distribution. Therefore, modellers tend to focus on higher
moments of the distributions. In terms of stress testing, the implica-

61

stress_testing.indd 61 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

tion is that there is less focus on fitting the mean of the distribution.
Therefore, estimating the base case, or expected loss, is not the pri-
mary goal. The implication is that even if operational losses could
be linked to risk factors in an LDA framework, it is not clear what
would be gained by stressing the expected loss.
Third, the nature of operational risk loss events makes it especially
challenging to establish a clear link between macroeconomic events
and loss severity. In particular, operational risk exposure tends to
be driven by low-frequency, high-severity events generated by fat-
tailed, sub-exponential distributions, and, unlike with a portfolio of
loans or equity securities, the maximum exposure is essentially un-
bounded. For example, single events such as rogue trades, class-ac-
tion lawsuits and natural disasters that cost banks billions of dollars
each year are very difficult to model, especially with the relatively
small sample sizes available. And it is these large, infrequent losses
that tend to drive operational risk exposure. This creates obvious
challenges for stress testing because the stress distribution should
be realistic enough to assign plausible likelihoods to such losses to
ensure a sufficient capital buffer. On the one hand, the stress distribu-
tion should ensure a bank has sufficient capital to cover any severe
but plausible stress event or a set of events. On the other hand, if not
careful, we might end up with a stress distribution assigning unreal-
istically high probabilities to single events, each individually capable
of putting a bank out of business.
Finally, the timing of operational risk events and operational risk
losses can be complex. This further complicates estimation of de-
pendence between operational losses and macroeconomic events.
Most notably, large losses tend to be legal suits where the control
failure that caused the event precedes the legal reserve and settle-
ment or fine by several years. One implication is that it is not clear
what date to use in the estimation of the relationship between the
macroeconomic events and the operational loss events. On one
hand, from a risk-management perspective, it may be best to know
the correlation between the timing of control failures and the mac-
roeconomic environment, so managers can prevent future control
breakdowns. On the other hand, from a capital-impact perspective,
we would wish to model the relationship between the macroeco-
nomic environment and the loss-realisation date.

62

stress_testing.indd 62 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

All of these factors combine to pose significant challenges for stress-

testing operational risk. In particular, the LDA approach is not easily
amenable to a stress-testing framework. However, despite these chal-
lenges, there are several methodologies banks can use to stress-test
operational risk, including several within the basic LDA framework.
In the remainder of the chapter, we discuss these approaches.

APPROACHES TO STRESS TESTING OPERATIONAL RISK

In this section, we outline several methodologies for stress testing
operational risk. We also note that a bank need not use any one
methodology in isolation and would likely benefit from imple-
menting a combination of the approaches to better understand its
exposure to operational risk.

Stress testing frequency distribution within the LDA

Banks tend to use Poisson or negative binomial distributions to
model frequency and some have had success using factor models.
If a link between risk factors and operational risk-loss frequency
could be established, then stress testing via the frequency distribu-
tion would be relatively straightforward. For example, Chernobai,
Jorion and Yu (2011) use publicly reported loss data for a large set of
financial institutions and develop a Poisson panel regression model
for the mean annual frequency of loss arrival. In their model, the
Poisson frequency parameter is regressed against a set of firm-spe-
cific and macroeconomic variables.
Such a model could be used to stress-test operational risk. Name-
ly, by supplying the model with stressed values of the macroeco-
nomic and firm-specific variables we could extrapolate stressed
frequency values and use them to calculate estimated stress values
of capital. Stressed values of macroeconomic and firm-specific vari-
ables could be generated in an integrated framework under differ-
ent historic or hypothetical scenarios affecting both sets of variables
simultaneously. One caveat to this approach is that it assumes that
estimated relationships between the frequency and the explanatory
variables of the model as well as the loss severity distribution are
not affected by stressed conditions.
An important downside to this approach is that operational risk
exposure tends to be driven by low-frequency, high-severity losses.

63

stress_testing.indd 63 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Consequently, the impact of stressing the frequency distribution on

the overall operational risk exposure is small compared with stress-
ing severity.

Stress testing severity distribution within the LDA

Stress testing via the severity distributions in the LDA approach is
challenging for the opposite reasons to stressing the frequency dis-
tribution. While the shape of the severity distribution is extremely
important for measuring operational risk exposure, it is all the more
difficult to establish a link between operational risk and macroeco-
nomic risk factors.
In addition to the heavy-tailed nature of the losses, the timing
between the events and realisation of losses also makes this chal-
lenging. More research needs to be done in this direction. We do not
foresee that stress testing severity distributions without bringing
any additional information about the tail behaviour (ie, information
going beyond what is contained in historically observed losses) is
the most fruitful way of approaching stress testing of operational
risk exposure.

Stress testing using scenarios

The Final Supervisory Guidance (2012), which was developed
to provide a guidance on stress testing for large banks in the US,
recommends that the banks consider using several stress-testing
approaches, one of which is scenario analysis. In addition to this
regulatory document, there are several risk-specific regulatory doc-
uments related to stress-testing requirements. For example, the Fi-
nal Rule (2007) requires Basel II financial institutions to incorporate
scenario analysis into their operational risk assessment and quan-
tification systems. Scenario analysis refers to the application of a
broad range of historical and/or hypothetical scenarios to various
levels of the banking organisations, in order to assess their vulner-
ability to adverse circumstances.
Perhaps Kupiec (1998) is among the earliest papers focusing on
stress testing VaR models through the use of scenarios. While Ku-
piec uses historical scenarios to stress-test risk models, we could
also use hypothetical expert-generated scenarios. In the context of
stress testing market risk models, Berkowitz (2000) proposes a mix-

64

stress_testing.indd 64 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

ture approach to stress testing VaR models in which a probability

distribution and a probability of occurrence are assigned to each
scenario. The stressed VaR model is derived as a probability mix-
ture of the original VaR model and the set of identified scenarios.
Taking into account well-known drawbacks of VaR as a risk mea-
sure, Aragones, Blanco and Dowd (2001) propose to complement
Bertkowitz’s framework by estimating expected tail loss (ETL), an
alternative to VaR. Although ETL possess superior properties rela-
tive to VaR (Artzner et al 1999), one important drawback of ETL
is that it does not exist if the loss distribution does not have finite
first moment. VaR can still be calculated for such distributions. In
addition, quantifying the ETL involves integration, which makes it
more difficult to estimate than the VaR. Also, regulatory rules re-
quire the use of the VaR measure to quantify risk. For these reasons,
we focus on VaR in this chapter.
Incorporating scenarios in a mixture framework for the purpose
of stress testing an operational risk model is difficult, at least for
the following reason. In operational risk, capital is determined by
the VaR risk measure as a fixed and prespecified level of VaR (ie,
the 99.9th percentile), which is in the far tail of the loss distribu-
tion. Therefore, the effect of each scenario on the original VaR is
not always obvious under the mixture framework. Scenario realisa-
tions exceeding the original VaR might lead to a higher stress VaR
measure conforming with the main purpose of stress testing, while
those falling short of the original VaR may even reduce the stress
VaR. The net effect is not clear. If line-of-business managers are
responsible for generating operational loss scenarios, they might
avoid generating a reasonable number of scenario losses from the
far tail if doing so will negatively affect their performance. If, as a
result, too few scenarios have been generated from the far tail rela-
tive to the rest of the severity distribution’s domain, then this situa-
tion might lead to a reduction in capital. Furthermore, whether the
reduction in capital happened due to expert opinions or simply be-
cause of a disproportionately lower number of far-tail scenarios is
not clear. In other words, the mixture approach, in its original form,
does not possess a built-in protection against producing stress risk
VaR measures that fall short of the original risk measures.
To address the above-described shortcoming of the mixture ap-

65

stress_testing.indd 65 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

proach, Ergashev (2012) proposes an alternative theoretical frame-

work for incorporating scenario losses into operational risk model-
ling. The basis of this framework is the idea that we need to focus
on worst-case scenarios, because only those scenarios contain valu-
able information about the tail behaviour of operational losses. A
simple rule for identifying the worst-case scenarios from the pool
of all scenarios is proposed. Only the information contained in the
worst-case scenarios enters the quantification process in the form of
lower bound constraints on the specific quantile levels of the sever-
ity distribution. Therefore, the stress VaR values never fall below the
original VaR values. This framework could be used to stress-test the
original severity distribution without any need for re-estimating the
stress severity distribution, because the last distribution is explicitly
derived during the process of incorporating the worst-case scenarios.

Converting the macroeconomic scenario into a 1-in-N-year event

As discussed above, the most significant challenge in terms of stress
testing operational risk is that, to date, the relationship between
operational losses and macroeconomic factors has not been clearly
modelled. Regardless of the reason for this, it poses challenges for
managers trying to use macroeconomic scenarios to stress opera-
tional risk exposure. Ideally, we could resolve these issues and de-
velop risk-sensitive operational risk models. We also need to con-
sider the possibility that the industry and academic literature have
not had much success in modelling these relationships because
they might not exist.
Since it is difficult to establish a modelled relationship between
macroeconomic factors and operational risk losses, in this section
we simply assume that there is a one-to-one correspondence be-
tween quantiles of both distributions. To the extent that this as-
sumption is true, we could convert the macroeconomic scenario to
a 1-in-N-year event and simply compute the 1 − 1/N quantile of the
aggregate operational loss distribution as the corresponding stress
event. More specifically, we assume that, if a specific macroeco-
nomic scenario corresponds to 1-in-N-year event in the historically
observed distribution of the macroeconomic variable(s), then this
scenario would trigger 1-in-N-year loss event in the operational
loss distribution as well.

66

stress_testing.indd 66 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

This is an appealing approach to stressing operational risk capi-

tal for banks that already have an established AMA model based
on the LDA approach simply because it would not require any ad-
ditional model development.
One downside of this approach, however, is that the link be-
tween operational risk exposure and the specific scenario is estab-
lished only through the likelihood of the scenario occurring. The
implication is that stress testing operational risk in this setting does
not allow us to distinguish between banks with different charac-
teristics.2 Therefore, it is essentially equivalent to the BCBS Final
Rule (although the percentiles and time horizon may be different).
Another potential drawback relates to our above discussion regard-
ing the challenges associated with using LDA to estimate expected
losses, and specifically the fact that LDA is designed to model the
tail of the loss distribution. Of course, this concern could be miti-
gated to a certain extent by estimating the LDA model with a focus
on the percentile of interest.

Stress testing the dependence structure

Another potential way of stress testing in the LDA framework is to
stress the dependence structure. The idea is that, during stressed
periods, internal controls may be more prone to breakdown and
the link between losses across units of measure may intensify. As
discussed above, banks generally model the aggregate loss distri-
bution for each unit of measure and then combine the units of mea-
sure using a copula-based approach.
There are two basic parts to the dependence model which could
be stressed: (1) the choice of copula and (2) the dependence structure
embedded in a particular copula modelling the dependence between
units of measure. Either, or both simultaneously, could be stressed.
The appealing aspect of stress testing the dependence model is
that most experts agree that dependency structures across finan-
cial assets tend to change in periods of stress. The financial crisis of
2007–9 also highlighted the importance of taking into account pos-
sible changes in dependence structure. However, we should also
emphasise that stress-testing dependence is among the most dif-
ficult approaches to stress testing. First, justifying the copula choice
is not an easy task with a limited number of observed losses. Sec-

67

stress_testing.indd 67 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

ond, the embedded dependence structure is difficult to estimate.

Copulas involve a substantial number of parameters to estimate,
which is difficult to accomplish given the fact that the sample sizes
of observed losses are not large in practical applications. Third, as
per the above discussion regarding the timing of operational loss
event dates, the calibration of even a simple correlation between
events is not straightforward.

Stress testing and model risk

Model risk is present in all financial models. In the case of opera-
tional risk, it is especially relevant for reasons including those dis-
cussed above and the fact that the field is still in its infancy.3 This
makes stress testing even more important in operational risk than
in the other risk area. However, stress testing a model with a po-
tentially substantial model risk is not an easy task. Here, sensitivity
analysis, as a special case of stress testing, can be used in conjunc-
tion with the above approaches to demonstrate that a model is not
overly sensitive to certain reasonable changes to the parameter val-
ues or the assumptions of the model.

CONCLUSION
We began this chapter by providing the reader with a high-level
background to state-of-the-art operational risk modelling. Within
this discussion, we have made explicit reference to the challenges
inherent in modelling operational risk. While data limitations are at
or near the top of the list of challenges, there are several areas where
simply increasing the quality and quantity of operational loss data
alone will likely not be sufficient. In other words, these challenges
are not going away any time soon, and further research is needed
to advance the field.
In terms of stress testing operational risk, the challenges are mag-
nified. Most of them arise from the fact that operational risk exposure
tends to be driven by infrequent large events. For example, model-
ling the aggregate loss distribution accurately requires additional
loss data, especially in the tail of the distribution. Stress testing insuf-
ficiently accurate static models could amplify the model risk. Mod-
elling dependence with the macroeconomic environment is equally
challenging due to the lack of strong supportive evidence.

68

stress_testing.indd 68 20/05/2013 18:45

 Operational Risk: An Overview of Stress-Testing Methodologies

Despite these challenges and limitations associated with opera-

tional risk modelling, we have laid out several potentially fruitful
methodologies for banks to stress-test their operational risk expo-
sure. While some appear more promising than others, each of the
methodologies is still in its infancy and more research is needed to
decide on the appropriateness of any one method. Therefore, we
recommend a bank to use a variety of alternative models to esti-
mate the impact of a stress scenario on operational risk capital.

The views expressed in this chapter are those of the authors and do not neces-
sarily reflect the position of the Federal Reserve Bank of Richmond, the Office
of the Comptroller of Currency, the US Department of the Treasury or the
Federal Reserve System. Bakhodir Ergashev is at the Federal Reserve Bank of
Richmond, Charlotte Office, 530 East Trade Street, Charlotte, NC 28202; email:
[Link]@[Link]. Brian Clark is at the Enterprise Risk Analysis
Division of the Office of the Comptroller of Currency, 400 7th Street SW, Wash-
ington, DC 20219; email: [Link]@[Link].

REFERENCES
Aragones, J. R., C. Blanco and D. Dowd, 2001, “Incorporating stress tests into
market risk models,” Derivatives Quarterly, issue 7, pp. 44–9.

Artzner, P., et al, 1999, “Coherent Measures of Risk”, Mathematical Finance 9(3), pp.
203–28.

Basel Committee on Banking Supervision, 2009, “Principles for sound stress test-
ing practices and supervision”.

Berkowitz, J., 2000, “A coherent framework for stress testing,” Journal of Risk, 2(2), 1-11.

Chernobai, A., P. Jorion and F. Yu, 2011, “The determinants of operational risk in
U.S. financial institutions”, Journal of Financial and Quantitative Analysis 46(6), pp.
1683–725.

Ergashev, B., 2012, “A theoretical framework for incorporating scenario analysis

into operational risk modeling”, Journal of Financial Services Research 41(3), pp.
145–161.

Final Rule, 2007, “Federal Register, Vol. 72, No. 235, Rules and Regulations Mod-
elling operational risk”, US Department of Treasury, Federal Reserve System, and
Federal Deposit Insurance Corporation.

Final Supervisory Guidance, 2012, “Supervisory Guidance on Stress Testing for

Banking Organizations with More Than $10 Billion In Total Consolidated Assets”,
US Department of Treasury, Federal Reserve System, and Federal Deposit Insur-
ance Corporation.

69

stress_testing.indd 69 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Kupiec, P., 1998, “Stress Testing in a Value at Risk Framework”, Journal of Deriva-
tives 6, pp. 7–24.

Model Risk Management Guidance, 2011, “Supervisory Guidance on Model Risk

Management”, Board of Governors of the Federal Reserve System and Office of
the Comptroller of the Currency.

1 Hence, the focus is on the tail of the loss distributions as opposed to credit and market risk
models that have been developed with the purpose of modelling expected losses.
2 For example, consider the case of a housing price shock as the stress scenario. In this case,
a bank with no exposure to mortgages would be subject to the same level of stressing its
operational risk capital as a bank with a significant amount of exposure to mortgages.
3 Model Risk Management Guidance (2011) defines model risk as “the potential for adverse
consequences from decisions based on incorrect or misused model outputs and reports”.

70

stress_testing.indd 70 20/05/2013 18:45

 6
Stress Testing of Bank Loan
Portfolios as a Diagnostic Tool

Paul Calem, Arden Hall

Federal Reserve Bank of Philadelphia

In the aftermath of the financial crisis and its attendant government

interventions to stabilise financial markets, stress testing of bank capi-
tal adequacy has taken on a prominent role in monitoring risk and
capital adequacy at large banking organisations.1 The stress tests are
designed to provide a comprehensive view of bank risk exposure, in-
cluding exposure to balance-sheet loan losses; revenue declines; coun-
terparty credit risk; trading and market risk; and operational risk.
In February 2009, the federal banking agencies – led by the Fed-
eral Reserve – created a stress test and required the nation’s 19 larg-
est bank holding companies to apply it as part of the Supervisory
Capital Assessment Program (SCAP).2 The immediate motivation
for the 2009 stress test was to determine how much additional capi-
tal a bank holding company would need to ensure that it would
remain a viable financial intermediary even in the adverse scenario,
with the Treasury Department prepared to provide capital to any
bank that could not raise the required amount from private sourc-
es.3 The SCAP experience demonstrated the value of a simultane-
ous, forward-looking projection of potential losses and revenue ef-
fects based on each bank’s own portfolio and circumstances.
In November 2011, the Federal Reserve issued a new regulation
requiring banking organisations with consolidated assets of US$50
billion or more to submit an annual capital plan. Under the rule,
the Federal Reserve annually evaluates each institution’s capital
adequacy, internal capital adequacy assessment processes and their

71

stress_testing.indd 71 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

plans to make capital distributions such as dividend payments or

stock repurchases. This annual exercise, named the Comprehensive
Capital Analysis and Review (CCAR), is designed to ensure that in-
stitutions have robust, forward-looking capital planning processes
that account for their unique risks, and sufficient capital to con-
tinue operations throughout times of economic and financial stress.
Company-run and supervisory stress tests are a critical part of this
annual capital review. They are used to determine whether a firm’s
capital distribution plans are consistent with remaining a viable fi-
nancial intermediary even in an adverse scenario.4
Congress drew on the lessons of the 2009 exercise by including
a requirement for stress testing in the 2010 Dodd–Frank Wall Street
Reform and Consumer Protection Act. Bank holding companies with
total consolidated assets of US$50 billion or more and non-bank fi-
nancial companies that the Financial Stability Oversight Council has
designated for supervision by the Federal Reserve began annual su-
pervisory stress testing and semi-annual company-run stress testing
under DFA rules in 2012; these also serve the purpose of CCAR stress
testing.5 Bank holding companies with total consolidated assets be-
tween US$10 billion and US$50 billion and savings and loan hold-
ing companies and state member banks with total consolidated assets
of more than US$10 billion will also undergo annual company-run
stress tests. Stress tests for these firms were slated to begin in 2013.
In May 2012, the regulatory agencies published a “Supervisory
Guidance on Stress Testing for Banking Organizations with More
Than $10 Billion in Total Consolidated Assets”. As defined by the
guidance, stress testing consists of “exercises to conduct a forward-
looking assessment of the potential impact of various adverse
events and circumstances”. This guidance covers a much wider ap-
plication of stress testing than those conducted for CCAR and DFA.
An important component of the CCAR and DFA stress tests is
exposure to credit losses from commercial and consumer loans held
on balance sheet, particularly losses from high-risk residential and
commercial mortgage lending. These have figured prominently in
overall stress projections due to the lingering impacts of high-risk
mortgage lending and to the positing of stress scenarios involving
a further steep decline in house prices and rise in unemployment.
Discussion in this chapter is two-tiered. It includes high-level dis-

72

stress_testing.indd 72 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

cussion of principles applicable to bank capital stress tests gener-

ally, concerning objectives and limitations of stress testing, design
of appropriate scenarios, and the role of models. Where the discus-
sion turns to particular issues related to model estimation, how-
ever, we narrow the discussion to modelling of balance sheet loan
loss, which is our particular area of experience.
A balance-sheet loan-loss stress projection is distinguished by
quantification of losses arising directly from a borrower’s failure
to repay a loan held on the bank’s books. Generally, banks have ac-
cess to historical data on the repayment performance of commercial
and retail credits from which statistical inferences can be obtained
regarding borrower performance under varying economic condi-
tions. A loan-loss stress test, however, is based on modelling of bor-
rower repayment performance under conditions at the extremes,
generally outside the scope of historical experience. As such, it en-
tails major modelling challenges.
In this chapter we propose a framework for bank capital stress-
testing based on recognising inherent limitations of the process; set-
ting objectives appropriate to these limitations; designing appropriate
scenarios; and applying credible models. The discussion emphasises
that, despite the challenges and limitations, bank capital stress testing
in 2012 provides a useful diagnostic analysis of a banking institution’s
sensitivity to adverse shocks impacting on the loan portfolio. Stress
testing can help pinpoint sources of risk; rank order institutions ac-
cording to risk; indicate improvement or deterioration in a bank’s risk
position over time; and provide benchmarks for evaluating an institu-
tion’s risk-management processes and capital adequacy.
The chapter is organised as follows. The next section discusses
the purposes and limitations of stress testing and suggests a set
of suitable objectives. The following section describes appropriate
scenario design and the elements of a credible model estimation
process. We then conclude.

DEFINING AND ACHIEVING STRESS-TEST GOALS

Ideally, the goal of bank capital sheet stress tests is to alert the bank or
bank regulator to an unacceptable level of insolvency risk. Of course,
the level of risk that is unacceptable is subjective, and judgements
about it are likely to differ among investors, bank management and

73

stress_testing.indd 73 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

the regulators. But, even with a specified a level of risk, this goal,
while simple to state, is difficult to operationalise for several reasons.
First, the connection to stress scenarios is not straightforward.
Risk is multidimensional. The translation of a particular type of risk
into a scenario seems feasible, but risks are correlated, so isolat-
ing a single type of risk may simplify the scenario unrealistically.
However, designing a scenario based on correlations among types
of risk poses implementation challenges and complicates the in-
terpretation of stress-test results. If risks are best represented by a
multivariate probability distribution, this seems to imply that no
single scenario or small number of scenarios can adequately rep-
resent the tails of the distribution. And, if correlations are not ac-
curately known, there can be little assurance that selected scenarios
adequately represent tail risks.
Second, stress tests rely on models or constructs that may generate
erroneous predictions. In other words, they are subject to substantial
model risk and statistical uncertainty. Stress-test models of necessity
are simplified representations of reality. They are generally based on
past experience as embodied in historical data, which may not be
relevant to the next stress event. Moreover, historical data typical-
ly covers relatively short periods and are often incomplete, lacking
information on important predictors of loss. Even with a correctly
specified model and adequate data, estimation of model parameters
is subject to statistical error. Finally, because bank balance sheets, fi-
nancial activities and operational structures are complex, stress-loss
predictions depend on combinations of models, making it exceed-
ingly difficult to measure the estimation error for aggregate losses.
Despite these problems, we maintain the view that bank capital
stress tests can provide valuable information to bank managers and
regulators. We also believe, however, that, while the information
obtainable from stress tests is useful, it is inherently limited. The
practical value of stress tests depends first and foremost on recog-
nising their limitations and defining appropriate objectives.

Limitations of stress tests

As described above, stress testing is fraught with limitations if ap-
proached with the goal of accurately forecasting losses or precisely
quantifying risk. To succeed at this level, analysts would need to

74

stress_testing.indd 74 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

work out the underlying causes of default as applicable to unprec-

edented or evolving conditions, rather than just find historical cor-
relations. Statistical techniques and even economic theory may pro-
vide little guidance here.
Precision is unattainable with respect to both assignment of prob-
abilities to any particular scenario and estimation of credit losses
conditional on assigned scenarios. The question of scenario design
is addressed in some detail below; for now, suffice it to say that sce-
nario selection is critical to the information content of a stress test.
Scenarios that are too mild could lead to failure to identify at-risk
banks; those that are too harsh may generate excessive concerns
and will not differentiate risk levels among banks.
Model risk and statistical uncertainty impede predictive accura-
cy of stress-test models. Model risk arises when the circumstances
or scenarios of the stress test diverge from the historical experience
on which the model is based, or when data limitations preclude
consideration of relevant predictors of credit loss. Statistical uncer-
tainty arises even in the extraordinary case of a fully and correctly
specified empirical model, because model coefficient estimates will
have a range of uncertainty (confidence interval) around them.
Balance-sheet loan-loss models being built for stress testing sure-
ly benefit from the availability of data from the credit crisis and sub-
sequent recession and housing-value decline in 2007 through 2009.
Models built with data covering a narrower range of economic con-
ditions cannot be expected to extrapolate as well. Models built with
data from the much wider range of economic conditions should
do better in deriving estimates based on past experience. However,
they remain vulnerable to error when extrapolated beyond their
range of economic conditions, to unprecedented situations such as
a “double-dip” housing recession.
The CCAR exercise conducted at the end of 2011 to evaluate
banks’ ability to absorb stress losses during 2012 through 2013 il-
lustrates this aspect of model risk. The unemployment scenario
selected for this exercise has the national unemployment rate ris-
ing above 13%, well outside the realm of recent historical experi-
ence regarding both level and one-year change. Thus, determining
the impact of the unemployment scenario on retail and wholesale
balance-sheet losses requires a significant degree of extrapolation

75

stress_testing.indd 75 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

of fitted historical relationships. Moreover, the CCAR loan-loss

models are based on previously observed relationships between
borrower repayment performance and macroeconomic variables,
which may not generalise to the next downturn.
Future relationships may diverge from those implied by a balance-
sheet loan-loss model due to inadequacies of historical data or limits
to our understanding of behavioural aspects of repayment. Borrower
behaviour around “turning points” in the economic environment
or in response to adverse economic events is particularly difficult
to accurately predict. For example, credit performance on first-lien
mortgages as reflected in first-time delinquency has been improv-
ing somewhat faster than models generally predict; see, for instance,
Goodman et al 2012. The moderating delinquency is partly explained
by macroeconomic variables and observable dimensions of borrower
credit quality. However, given the large percentage of borrowers who
continue to have little or no equity in their homes, and the modest
pace of economic recovery, models have tended to predict somewhat
slower improvement in mortgage repayment performance. Thus,
unobservable factors appear to be at play.6 The impact of reversion to
an adverse housing market and unemployment situation, as would
be posited for a stress test, similarly could depend on such unobserv-
able factors, implying substantial uncertainty around predictions of
mortgage credit performance for the stress test.
When statistical models are estimated using large datasets, the
amount of uncertainty around coefficient estimates generally is
small, so that statistical uncertainty tends to be less of a concern
than model risk. Even within large datasets, however, statistical un-
certainty is aggravated when the number of observations available
for inferring the impact of a key risk driver of the stress test is small,
or when important variables are measured with error. For example,
within the population of mortgage borrowers with negative equity
in 2012, the number of observations thins out at higher levels of
negative equity, leading to uncertainty around the impact of a fur-
ther, large decline in home values on this population of borrowers.
An additional, potential limitation of bank capital stress tests
pertains to the context in which they are implemented, as distinct
from any inherent statistical or methodological weaknesses. Stress-
test implementation carries a potential for excessive focus on mod-

76

stress_testing.indd 76 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

el-building and model application, at the expense of gathering and

analysing other relevant information or of recognising risks that
lie outside the bounds of the model. A related problem is the po-
tential for excessive focus on process and methodology, whereby
the stress test becomes a mechanical or rigid process of risk assess-
ment, displacing relevant qualitative analysis and impeding mod-
eller creativity. Indeed, there may be circumstances where, due to
the available data or the nature of the emerging risk, back-of-the-
envelope-type calculations based on plausible assumptions may
provide better insight than a statistical model.7

Practical objectives for stress tests

Given the major limitations around predictive accuracy of stress-
test models, what constitutes reasonable objectives for a stress-test
exercise? We believe that the following are practical and effective
uses of bank capital stress testing:

o establishing a standard for an acceptable level of post-loss capital;

o differentiating banks by relative risk to identify and remediate
banks that lie outside of an acceptable range;
o providing benchmarks for evaluating other quantitative loss
models;
o probing the risk-sensitivity of the loan portfolio or other bank
portfolios to relevant economic variables;
o probing the risk composition of the loan portfolio or other bank
portfolios; and
o delineating the bounds of statistical modelling – highlighting
risks outside the realm of historical data.

In establishing a standard for capital adequacy, the stress test would

be applied on a pass/not pass basis. In this context, the stress test
would be based on a single stress scenario or a small set of scenarios
that represent an appropriate level of risk.
This use of the stress test may seem somewhat arbitrary, given
the limitations of stress-test models, although no less arbitrary than
other regulatory capital standards and offering the advantage of a
relatively strong empirical basis for quantifying risk. The decision
rule can be made less arbitrary by combining the stress-test results

77

stress_testing.indd 77 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

with other relevant information by experienced managers or regu-

lators to reach conclusions about the level of risk and the actions
that may be needed to reduce risk.
Using stress tests to rank banks according to risk potentially allows
greater flexibility around scenario design, because this does not in-
volve identifying the single, specific stress threshold that corresponds
to an appropriate capital standard. While scenarios would need to be
plausible, the range of admissible scenarios is determined by what
would be informative about relative risk exposures across banks.
This use of stress tests clearly would be confined to bank regulators.
However, an analogous, internal use of bank stress tests would be to
evaluate relative risk across exposure categories or credit products.
Benchmarking is a natural application of stress tests. For in-
stance, supervisory stress tests can be used as benchmarks for as-
sessing banks’ internal loss estimates, while banks’ models can pro-
vide benchmarks for assessing the reasonableness of loss reserving
or economic capital models.
Probing the risk sensitivity and risk composition of bank loan,
investment, trading or other portfolios is also a very natural ap-
plication. Analysis of bank performance under a variety of stresses
can reveal weaknesses and anticipate developments likely to raise
the bank’s level of risk. The goal of probing risk composition is to
identify at-risk portfolios or exposure categories.
These objectives pertain to use of a stress-test model to directly quan-
tify risk. Stress-testing exercises can also be informative to the extent
that they focus attention on the distinction between risks that are mod-
elled and those not amenable to modelling. Thus, indirectly, the stress-
testing process can help identify emerging risks, such as those arising
from product innovation or loan origination channels or technologies,
outside the bounds of the historical data and current stress-test models.
For the sake of brevity, we forgo more extended discussion of
these objectives. It should be self-evident that they are consistent
with effective risk management or supervision of banking institu-
tions and that model risk is not an impediment to their implemen-
tation. They depend only on the design of appropriate scenarios
and the development of logically and statistically sound and rea-
sonably robust loss models, topics we address next.

78

stress_testing.indd 78 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

SCENARIO DESIGN AND MODEL ESTIMATION

After specifying an appropriate objective, the next step is to design
stress scenarios that are consistent with the objective. The ability to
achieve the objective additionally depends on adequacy and cred-
ibility of the models and the model-estimation process.

Scenario design
The choice of stress scenarios potentially introduces error distinct
from model risk or statistical uncertainty: it may illuminate the
wrong region of the risk distribution, leaving the region that should
be the focus of bank supervision unobserved. One step towards a
better process should be to describe explicitly the goal of the stress-
testing exercise, and the way in which stress-test results will be used.
At its most basic, stress-test design depends on the type of stress it
aims to measure. For instance, the Supervisory Capital Assessment
Program (SCAP) and Comprehensive Capital Analysis and Review
(CCAR) exercises aimed at measuring classic financial stress for sys-
temically important banks – are there sets of plausible if unlikely
events that would produce credit losses on a bank’s assets, and sup-
press its future income to the extent that its capital level would fall
to a dangerous level? Such a focus on individual bank risk ignores
network effects – the risk that one bank’s capital will fall to an inad-
equate level because of the failure of another systemically important
bank. It also ignores risk of adverse liquidity events such as deposi-
tor runs or collapse of a loan sale or securitisation market.
The scenarios that the Federal Reserve employed for SCAP and
CCAR represent one way that financial stress can be created: through
a serious recession. These scenarios take the existing economic con-
ditions as a starting point and then hypothesise an immediate sharp
global decline in economic activity, represented through forecast time
series on a few fundamental domestic and international macroeco-
nomic drivers. These include GDP and disposable-income growth,
unemployment, inflation and interest rates, and stock, commercial
real-estate and house prices for the US, plus a more limited set of
drivers for major international trading partners.
The Federal Reserve scenarios from the latest round of CCAR
certainly represent a stressful episode for banks, but does forecast
performance under these scenarios provide all the information

79

stress_testing.indd 79 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

needed to asses banks’ systemic risk? In particular, are there im-

portant dimensions of risk that are being overlooked? Are there
plausible scenarios along other dimensions that would produce a
different pattern of capital impacts across banks than the dimen-
sions included in the Federal Reserve scenarios?
Certainly if Bank A were well capitalised in Scenario 1, and Bank
B were undercapitalised, but in equally likely Scenario 2 the re-
sults were reversed, then regulators should be aware of both re-
sults. Finding whether other plausible scenarios besides a sharp
recession would produce distinctly different patterns of outcomes
would require loss forecasts under a large number of scenarios, a
potentially daunting task.
If stresses that are different in kind represent one source of ad-
ditional information, then what about stresses of varying severity?
The answer to this question would seem to depend on the objective
of the stress test. If the objective is a single pass/fail rule, then a
single, specific severity threshold is necessary. However, if banks
will be ranked, with decisions about capital adequacy made based
not only on stress-test results but also on additional factors such
as quality of risk management, then a wide range of scenario se-
verities could provide additional information and a more nuanced
view of capital adequacy. For instance, an institution viewed as par-
ticularly well managed might be subjected to less severe scenarios
for determining capital adequacy.
The Dodd–Frank Act requires banks to model at least three sce-
narios, including baseline and adverse scenarios as well as a severely
adverse scenario all provided by the Federal Reserve. The severely
adverse scenario is used to assess the adequacy of bank capital. The
baseline and adverse scenarios provide additional information about
the sensitivity of bank performance to stress and may be useful for
more general risk assessment, but are not used to assess capital ad-
equacy at the time of writing.
Creating and forecasting an even more severe version of the
Federal Reserve’s severely adverse scenario would be relatively
straightforward, but would it provide additional useful informa-
tion? Is there something to be learned from a scenario in which
many or all banks become undercapitalised? Potentially, the an-
swer is yes. If increasing the severity of the scenario substantially

80

stress_testing.indd 80 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

lowered a bank’s ranking among its peers, it could indicate a risk

exposure not indicated in less severe scenarios. However, the scope
for more adverse scenarios should be limited to those with a non-
negligible likelihood of occurring.
If stress testing is to be used as a tool to explore the risk structure
of the systemically important banks, then stress scenarios focused
on specific types of exposures could be useful. For example, if the
regulator wants to identify banks particularly exposed to losses in
commercial real estate, then a scenario simulating distress in that
particular market will more precisely identify at-risk banks than a
general recession scenario that includes a downturn in commercial
real estate among several stressors.
Careful exploration of targeted stress scenarios is likely to give
the regulator a useful picture of the sources and distribution of risk,
but, again, carrying out such an analysis would be a large task. It
would also generate multiple views of the riskiness of individu-
al banks, which then must be synthesised into an overall view of
the adequacy of the bank’s capital plan. Too great a proliferation
of stress-test results could just make that task harder, rather than
increase its accuracy.
In the discussion to this point, scenarios appear as completely
exogenous events. However, the entire rationale for stress testing
systemically important banks is that they have the capacity to feed
the stress they receive back into the economy. So scenario analy-
sis as described above implicitly assumes either that there are no
feedbacks or that the scenario in some sense represents the impact
of both the initial economic stress and the net effect of all the feed-
backs from systemically important banks. If, as a result of regula-
tion based on the stress test, systemically important banks remain
well capitalised through the stress scenario, then an assumption of
no feedbacks may be reasonable. However, if feedbacks are expect-
ed, then the stress scenario must become dynamic. Initial economic
stress impacts the large banks, is distributed within their network,
and then feeds back to the economy as a whole, which then deter-
mines the economic stress for the next period. A structure like this,
incorporating all potential feedbacks, is much more complex than
anything so far attempted.
A final issue relating to stress scenarios relates to their variabil-

81

stress_testing.indd 81 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

ity through time. As with any type of capital regulation, the use of
stress testing implicitly assigns capital weights to asset classes and
off-balance-sheet exposures. These can reasonably be expected to
affect future bank asset choices. Assuming that loss modelling will
never become an exact science, bank optimisation with respect to
implicit stress test risk weights could increase risk in ways difficult
to detect a priori. With this concern in mind a case might be made
for some variation in stress scenarios over time simply to avoid
over reliance on a single view of bank risk, and thereby reduce un-
measured risk in the banking system.

Model estimation
Models are the means of translating economic scenarios into loss
forecasts. The goals of stress testing articulated above require that the
model provide an accurate representation of historical variation of
losses in relation to economic conditions (dynamic relationships). The
model also should provide credible projections of future losses under
severely adverse economic conditions. In addition, the model should
accurately capture relevant (cross-sectional) relationships across dif-
ferent types of exposures within a portfolio, such as between loan and
borrower characteristics and credit risk in the case of balance-sheet
loan-loss models. Finally, the model should incorporate conservatism
in choice of model structure and variable specification as a counter-
weight to the inevitable data limitations and model risk.
At the same time, practical considerations dictate that the models
be as simple and transparent as possible. For instance, in the case of
supervisory models, the models must be amenable to relatively quick
application to the portfolios of multiple institutions during the annual
stress-testing cycle. They must also be conducive to annual updating
and validation and submission to some level of public scrutiny.
Accuracy in capturing historical variation requires testing alter-
native model specifications with the aim of improving the model’s
fit over the economic cycle. Similarly, accuracy in capturing cross-
sectional relationships requires investigating alternative model
specifications. Specification testing, of course, is limited by the set
of variables within the historical, development database and by the
set of economic variables for which scenarios are obtainable.
In building balance-sheet loan-loss models for stress tests, there

82

stress_testing.indd 82 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

tend to be more rapidly diminishing returns to specification testing

compared with other activities, for two reasons. First, the objectives
of stress testing are relatively broad. Assessing the bank’s ability
to withstand a severe economic downturn generally requires less
granular attention to individual borrower characteristics than other
applications such as credit scoring, for example, where the focus
is on each borrower’s marginal contribution to credit risk. Second,
the marginal improvements in a model’s fit that might be achieved
through extensive specification testing will do little to offset the
substantial model risk inherent in a stress-testing exercise.
In stress testing, as in modelling generally, improving a model’s
fit historically does not necessarily improve the accuracy of pre-
dictions. One potential complication is “overfitting” of historical
relationships, whereby a model produces estimates based on tem-
porary or spurious correlations rather than stable causal relation-
ships. An overfitted model may match historical variation but at
the expense of eroding the model’s predictive accuracy, reducing its
transparency or weakening its theoretical or intuitive foundation.
In stress testing, this concern is exacerbated by the possibility
that improving the model’s accuracy in capturing historical varia-
tion may involve weakening estimated relationships to economic
variables. This could lead to less conservative stress-loss predic-
tions, diminishing the credibility of the stress test. For example, in
a time series or dynamic panel-type loan loss model, inclusion of a
lagged loss rate may substantially improve the historical fit of the
model as well as the accuracy of short-term forecasts. However the
autoregressive term might partly capture the impact of economic
conditions, causing weaker estimated economic relationships and
less conservative stress-loss forecasts.
One way to limit the potential for overfitting is to apply econom-
ic theory to guide the selection of variables for the model. A second
is to be guided by the dictum “less can be more”: improvements to
historical fit accorded by increasing the complexity of a model or
number of included variables should be balanced against the in-
creased potential for overfitting. Particular caution should be ap-
plied with respect to inclusion of potentially endogenous variables
that improve the fit to historical data but weaken the estimated re-
lationship between default and macroeconomic variables.

83

stress_testing.indd 83 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Another potential problem is that a model may be robust in-sam-

ple but not with respect to loss prediction under stress scenarios.
In other words, minor alterations may have little impact on model
goodness of fit but imply substantially different losses under stress
scenarios. The preferred specification should be robust not only
with respect to in-sample fit but also robust or at least conservative
(in the sense of providing an upper bound on loss estimates) with
respect to stress-loss prediction.
Even out-of-sample testing or backfitting is of more limited value
in the stress testing arena compared with other modelling contexts,
and should be interpreted with caution. A model may predict well
out-of-sample under non-stress conditions, but have limited cred-
ibility for stress-loss prediction. Thus, for example, a balance-sheet
loan-loss model that is predicting higher-than-observed delinquen-
cy rates during a favourable or improving economic environment
may be preferable to one that has been re-estimated to provide a
better fit. While it is important to monitor the performance of a
statistical prediction model, including a stress-test model, through
backtesting missing the mark is not necessarily an indication of a
need to re-estimate. Rather, especially in the context of a stress test,
off-the-mark prediction is an issue to investigate and seek to under-
stand, with the goal of improving the model as needed.
In the realm of stress testing, the goal for the development and
validation of balance-sheet loan-loss models is not forecast accura-
cy per se. Rather, the goal is to provide credible loss forecasts under
the assumption that borrowers’ responses to the stress scenarios
will resemble historical performance under broadly comparable
conditions. There can be no assurance either that future behaviour
under stress conditions will resemble past performance, or that the
historical data incorporate a sufficiently diverse range of economic
conditions to guarantee accuracy of the prediction. Moreover, the
stress scenarios are unlikely to actually materialise (hopefully they
will not), ruling out a true backtest of the model. Hence, robustness
of stress-loss predictions and appropriate conservatism of model
assumptions are equal if not more important considerations than
model performance in backtesting.

84

stress_testing.indd 84 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

CONCLUSIONS
Bank capital sheet stress testing is a potentially useful tool for bank
risk management and supervisory risk assessment. However, loss
predictions under stress scenarios necessarily represent extrapola-
tions of experience and knowledge and should be considered as
such. Model risk and statistical uncertainty impede predictive ac-
curacy of stress-test models. Thus, a bank capital stress test can pro-
vide a benchmark, but not a forecast.
As such, stress testing can be used to establish a standard for capi-
tal adequacy; to identify and remediate banks that lie outside of an
acceptable range of exposure to credit loss; and to provide bench-
marks for evaluating other quantitative loss models. Stress testing is
also potentially useful as an analytical tool to probe the risk sensitivi-
ty or risk composition of the loan portfolio. Somewhat paradoxically,
it is also useful for illuminating the limits of statistical modelling;
that is, for highlighting risks outside the realm of historical data.
While using stress-test results in a pass/not-pass mode is
straightforward, well-executed stress tests should be able to sup-
port a more nuanced view of bank risk. For example, stress tests
could highlight specific activities that make a disproportionate
contribution to overall risk for the institution and deserve deeper
review. However, stress testing can potentially produce very large
amounts of information. To be used effectively, stress-test results
must be conveyed in a form that can be effectively understood and
acted on by bank managements and regulators.
The design of both individual stress scenarios and an effective
set of scenarios for a stress test are problems that are ripe for further
research and development. For example, the Federal Reserve sce-
narios from the 2011 CCAR represent a stressful episode for banks,
yet it is still important to consider whether the forecast performance
under these scenarios provides all the information needed to assess
banks’ systemic risk.
To achieve effective stress tests, institutions must recognise the
specific demands of forecasting for stress scenarios, and design
their model development process appropriately. Loss modelling for
stress testing puts a premium on ability to extrapolate reasonably.
The preferred modelling approach or model specification should
be robust not only with respect to in-sample fit but also robust or at

85

stress_testing.indd 85 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

least conservative with respect to stress-loss prediction.

With respect to balance-sheet loan-loss models in particular,
there tend to be more rapidly diminishing returns to specification
testing and out-of-sample validation efforts in stress testing com-
pared with other modelling contexts. Pursuing marginal gains in
model fit will do little to offset the substantial model risk inherent
in the stress-testing exercise, but may lead to overfitting, which will
reduce a model’s reliability.
Likewise, a loan-loss model may predict well out-of-sample un-
der non-stress conditions, but have limited credibility for stress-
loss prediction. Robustness of stress-loss predictions and appropri-
ate conservatism of model assumptions are equally important to
consider along with model performance in backtesting
When implementing a bank capital stress-test process, it is im-
portant to be mindful of the potential for the process to promote
inefficient use of resources and impede creativity in assessing risk.
Excessive focus on methodological issues in model-building and
model application can divert attention from gathering and analys-
ing other relevant information. Likewise, excessive focus on process
and methodology may cause the stress test to become a mechanical
risk assessment exercise, impeding creativity and hampering rec-
ognition of emerging risks.
If undertaken without a clear sense of the limitations, and without
clear objectives or a context for judging model effectiveness, bank
capital stress testing can end up becoming an open-ended activity
and drain on resources with no satisfactory conclusion. Effective and
efficient application of bank capital stress tests requires acknowledg-
ing the limitations of the exercise; setting feasible objectives; design-
ing scenarios consistent with the articulated goals; and developing
loss models subject to sensible and cost-effective robustness criteria.

The views expressed in this chapter are those of the authors and do not reflect
the views of the Federal Reserve Bank of Philadelphia, Federal Reserve Sys-
tem or Office of the Comptroller of the Currency. We thank William Lang and
Amy Jordan for many helpful comments.

86

stress_testing.indd 86 20/05/2013 18:45

 Stress Testing of Bank Loan Portfolios as a Diagnostic Tool

REFERENCES
Goodman, Laurie S., et al, 2010, “Negative Equity Trumps Unemployment in
Predicting Defaults”, Journal of Fixed Income 19(4), Spring, pp. 67–72.

1 For additional historical perspective, see the speech “Developing Tools for Dynamic Capital
Supervision” by Federal Reserve Governor Daniel K. Tarullo at the Federal Reserve Bank of
Chicago Annual Risk Conference, April 10, 2012.
2 The test involved two scenarios – one based on the consensus forecast of professional fore-
casters and the other based on a severe, but plausible, economic situation – with specified
macroeconomic variables such as GDP growth, employment and house prices. Each partici-
pating institution was asked to supply, in a standardised format, detailed information on
portfolio risk factors and revenue drivers that supervisors could use to estimate losses and
revenues over a two-year period conditional on these scenarios.
3 The Federal Reserve’s decision to disclose the results of the test on a firm-specific basis
served a second purpose: to provide investors, and markets more generally, with informa-
tion that would help them form their own judgements on the condition of US banking in-
stitutions.
4 In the first quarter of 2011, the Federal Reserve had conducted a similar review of the 19
firms that had participated in the SCAP. As part of the CCAR, the Federal Reserve evalu-
ates institutions’ capital adequacy, internal capital adequacy assessment processes, and their
plans to make capital distributions, such as dividend payments or stock repurchases. The
stress test is only one of several essential components of the capital review. The Federal Re-
serve may object to a capital plan because of significant deficiencies in the capital planning
process, as well as because one or more relevant capital ratios would fall below required
levels under the assumptions of stress and planned capital distributions.
5 There are two sets of instructions: one for the 19 firms that participated in the CCAR in 2011,
the other for 12 additional firms with at least US$50 billion in assets that have not previously
participated in a supervisory stress-test exercise. The level of detail and analysis expected in
each institution’s capital plan will vary based on the company’s size, complexity, risk profile
and scope of operations. The instructions include a supervisory stress scenario that will be
used by all of the firms and the Federal Reserve to analyse firms’ capital needs to withstand
such a scenario while continuing to act as a financial intermediary. For the 19 firms that par-
ticipated in the CCAR in 2011, the Federal Reserve will also conduct a supervisory stress test
using internally developed models to generate loss estimates and post-stress capital ratios.
6 In short, several years of high default rates on mortgages may have filtered out higher-risk
borrowers, including along unmeasured dimensions of credit quality. Unobserved hetero-
geneity may be tied to the household’s overall balance-sheet composition and vulnerability
to income or wealth shocks.
7 For an example of such a rough calculation or scenario-based analysis that provided warning
as early as 2006 of the potential impact of payment shocks from hybrid ARM resets, see http://
[Link]/infocenter/whitepaper/FARES_resets_whitepaper_021406.pdf

87

stress_testing.indd 87 20/05/2013 18:45

stress_testing.indd 88 20/05/2013 18:45
 7
Stress-Test Modelling for Loan
Losses and Reserves

Michael Carhill, Jonathan Jones

Office of the Comptroller of the Currency

The widespread use of macroeconomic and financial factors in the

quantitative models that banks use to forecast their credit losses has
been an important development. The financial crisis of 2007–9, and
the associated severe recession, underscored the need for banks to
incorporate economic and market conditions into their retail and
wholesale credit risk models in order to produce credible stress
loan-loss estimates. Prior to the crisis, banks were unable to esti-
mate, and apparently uninterested in estimating, the credit losses
that would result from a recession, probably because a genera-
tion of bank executives had never experienced a severe recession.
This left banks unprepared for the severe loan losses that occurred
between 2008 and 2010. In onsite supervision of national banks’
credit-risk management systems during the past several years, the
authors have observed that bank executives at the larger banks, in
response to the 2007–9 crisis, have determined that models capable
of estimating credit losses conditional on stress economic scenarios
were necessary for enterprise-wide capital planning and stress test-
ing, and so directed their staffs to begin model-development efforts.
Several recent supervisory and regulatory developments prob-
ably also have accentuated this trend. First, the largest US banks
that are subject to the Advanced Internal Ratings Based (AIRB)
approach of Basel II (and Basel III) are required to conduct a cy-
clicality stress test in Pillar I and a forward-looking stress test of
credit risk as part of the Internal Capital Adequacy Assessment

89

stress_testing.indd 89 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Process in Pillar II. Second, the Basel Committee on Banking Su-

pervision issued a consultative paper in January 2009 that recom-
mended that banks use enterprise-wide stress tests to provide a
forward-looking assessment of their risk profile and capital ad-
equacy (Basel Committee 2009). Third, in an attempt to quell rap-
idly escalating fears about the solvency of the US banking sys-
tem during the 2007-9 crisis, the Federal Reserve conducted its
Supervisory Capital Analysis Program (SCAP) in spring 2009.
The SCAP, a macroprudential supervisory bank stress test, used
a stressful macroeconomic scenario to assess the capital adequacy
of the 19 largest US bank holding companies.1 Building on the suc-
cess of the SCAP exercise, the Federal Reserve conducted its Com-
prehensive Capital Analysis and Review (CCAR) bank stress tests
in 2011, 2012 and 2013 using macroeconomic stress scenarios that
involved a much larger number of macroeconomic and financial
factors than were used in the SCAP.2
Finally, the Dodd–Frank Wall Street Reform and Consumer Pro-
tection Act of 2010 required certain financial companies with total
consolidated assets of more than US$10 billion to conduct annual
stress tests using a minimum of three macroeconomic scenarios
(baseline, adverse and severely adverse) provided by the compa-
ny’s primary federal regulator. As a result, enterprise-wide stress
testing (and an associated formal capital-planning process) has
become a regulatory requirement for all banks and federal sav-
ings associations with more than US$10 billion in total assets.
Since 2008, substantial work has been conducted to develop mac-
ro-forecasting models for retail and wholesale credit risk, especially
at the largest banks. In this chapter, some of the internal modelling
practices that the authors have observed at national banks, where
macroeconomic and financial risk drivers have been incorporated
into the quantitative models used to generate retail and wholesale
credit loss forecasts and loan-loss reserve estimates, are discussed.
Some of the strengths and weaknesses of the various modelling ap-
proaches are also presented. Finally, several econometric specifica-
tion and estimation issues the authors have observed that need to
be addressed by banks in developing their macroeconomic-based
credit risk stress-testing models are discussed.

90

stress_testing.indd 90 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

BANK MODELS FOR LOAN LOSSES AND RESERVES

Most of the risk-assessment models needed for enterprise-wide
stress tests (eg, trading risk and banking-book interest-rate risk mod-
els) have been in place at banks for quite some time. Therefore, the
major modelling challenge has involved the development of credit-
risk models that incorporate macroeconomic and financial variables.
Prior to the crisis of 2007–9, banks had relied on a variety of quantita-
tive and expert-judgement approaches in their credit-risk modelling.
For retail credit risk, banks relied on credit and behavioural scoring
models, matrix models, roll-rate and Markovian-chain models, and
vintage models.3 With very few exceptions, none of these models
incorporated macroeconomic conditions, but instead rank-ordered
credit applicants by their unconditional probability of default (PD).
While most of the models can be used to generate a borrower-specific
PD, the relationship between macroeconomic factors and defaults
rates was, for the most part, not taken into account.
For wholesale credit risk, loan officers’ expert judgement was
used in credit-grading systems to assign each borrower a risk
grade. Similar to retail credit risk, historical data can be used to pro-
vide estimates of the PD for each risk grade. However, macroeco-
nomic factors were not used prior to the financial crisis of 2007–9,
except perhaps indirectly, since economic conditions were viewed
by banks as tending to unduly influence the loan officers’ credit-
risk judgement.

ALLOWANCE FOR LOAN AND LEASE LOSS MODELS

All banks subject to credit risk must establish an adequate allow-
ance for loan and lease losses (ALLL), which is intended to cover
expected credit losses (measured in the form of charge-offs) over a
given horizon.4 The ALLL is a stock and recorded as a contra-asset
(ie, an asset with a negative balance) on the bank’s balance sheet. In
the financial reports banks issue each quarter, changes in the ALLL
are referred to as loan-loss provisions, and included in non-interest
expense and deducted from net income. When banks take charge-
offs, the amount is subtracted from the ALLL. Although net loan-
loss provisions and net charge-offs (ie, gross charge-offs less recov-
eries) can be negative, this typically happens infrequently. Under
the uniform capital standards of US federal banking regulators,

91

stress_testing.indd 91 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

loan-loss reserves up to a value of 1.25% of risk-weighted assets are

included in Tier 2 regulatory capital.
While accounting standards are vague on the issue, the horizon
used for charge-off forecasts in setting loan-loss reserves is typi-
cally one year for retail loans, and one or two years for wholesale
loans. At smaller banks, the method used for determining the ALLL
is typically based on expert judgement. At larger banks, a combina-
tion of quantitative models and expert judgement is used in deter-
mining reserves.
The total ALLL consists of a quantitative reserve component and
a qualitative reserve component. The quantitative component is
based on historical charge-offs, while the qualitative component
typically has reflected an expert-judgement-based adjustment that
captured factors such as changes in lending policies and proce-
dures and changes in international, national and local economic
and business conditions that are not reflected in historical charge-
offs. Since 2008, however, the ALLL has become more quantitative
and forward looking, at least at larger banks, many of which now
use statistical and econometric methods to capture the sensitivity
of charge-offs to changes in macroeconomic and financial vari-
ables. For example, ALLL qualitative models may have a macro-
economic modelling component with two or three macroeconomic
variables lagged several quarters up to a year. Moreover, many
banks have begun to conduct stress tests of the qualitative com-
ponent of their loan-loss reserves using multiple macroeconomic
scenarios (Pocock 2012).
The overall aggregate predictive accuracy of banks’ ALLL mod-
els can be evaluated by comparing the ratio of the subsequent
year’s charge-offs to the beginning-of-year loan-loss reserves. Table
7.1 presents data taken from the December 31 financial statements
over the period 2005–11.

Table 7.1 Charge-offs as a percentage of beginning of year ALLL,

commercial banks, 2005–11 (unweighted means)
Year 2005 2006 2007 2008 2009 2010 2011

15 largest banks 22% 36% 48% 108% 103% 79% 56%

All banks 19% 18% 25% 48% 73% 58% 50%

92

stress_testing.indd 92 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

As shown in Table 7.1, assuming the typical one-year charge-off

coverage target for reserves, banks dramatically overstated their
reserves in the boom years that ended in 2007. This result shows
the near-total lack of a relation between ALLL and charge-offs dur-
ing the good years, suggesting banks perhaps had motives other
than credit losses in setting their loan-loss reserves such as income
smoothing, regulatory capital management and private informa-
tion signalling.5 In sharp contrast, the largest 15 banks understated
their reserves in 2008 and 2009, a period that covered the height of
the 2007–9 crisis, and the associated recession, which was the worst
since the Great Depression.
The authors’ loan-loss reserve results for the 15 largest banks re-
ported in Table 7.1 are consistent with those of Furlong and Knight
(2010), who have argued that the unusually low levels of reserves
compared with eventual loan losses in 2008 and 2009 reflected two
underlying causes. First, accounting standards limit banks’ ability
to set aside provisions for loans that are performing, but that might
become delinquent in the future, without evidence that charge-offs
would increase. Second, banks’ forecasts of the severity of reces-
sions and subsequent future loan losses are characterised by large
errors. Given the extreme severity of the 2007–9 crisis, and the un-
precedented decline in real-estate values, it was not surprising that
the largest banks substantially underreserved in 2008 and 2009. As
such, then, banks that forecast losses and set provisions based on
past experience would have severely underestimated the credit
losses that eventually occurred during the crisis of 2007–9.

POOL-LEVEL LOSS MODELS

Pool-level models are top-down models used by banks to forecast
charge-off rates by retail and wholesale loan type as a function of
macroeconomic and financial variables. The typical pool-level mod-
el has defined the loan type broadly. For example, all commercial
real-estate (CRE) loans may be defined as the pool, or, alternatively,
CRE loans may be divided into their major subtypes, eg, industrial,
hospitality, retail. For the large banks subject to the CCAR stress
test, the level of granularity for the portfolios has reflected the gran-
ularity requested in the CCAR reporting templates. At some banks,
the charge-off time series might extend back five years, while at

93

stress_testing.indd 93 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

other banks the charge-off history might extend back twenty years
– a period that would capture the past three US recessions. Even at
the pool level, however, many banks have found it difficult to cre-
ate internal datasets for charge-offs that are long enough for cred-
ible estimation. Those banks that cannot create sufficiently long
time series usually resort to having an external vendor create the
charge-off time-series datasets.
The charge-off rates used by banks in their top-down regression
models are measured quarterly, which means that a five-year time se-
ries contains only 20 observations, and a 20-year time series has only 80
observations. This paucity of data creates a premium on parsimony in
model development. Therefore, most banks use only from one to four
macroeconomic and financial risk drivers as explanatory variables,
to preserve degrees of freedom in their regression-based forecasting
models. At most banks, the macroeconomic and financial variables
that are considered for use in the charge-off regressions are usually
determined by an executive-level committee that has been assigned
to specify the various economic scenarios. Banks’ modellers then use
a combination of statistical significance tests and management’s ex-
pert judgement to arrive at the final set of macroeconomic and finan-
cial variables that are used in the regressions for each loan type. Some
banks have decided to use the same set of variables for each loan type,
sometimes sacrificing predictive accuracy for the sake of consistency.
The primary advantage of pool-level models has been the fo-
cus on the charge-off rate instead of loan-loss reserves.6 While still
an estimate, charge-offs are taken by banks at a time that more
closely aligns with the actual realised losses, and thus are much
more accurate and closer to the specific quantity of interest than
are reserves. The primary disadvantage of pool-level models is that
borrower-specific characteristics are generally not used as explana-
tory variables, except at the aggregate level using pool averages.
For example, for first-lien residential mortgage loans, defaults are
extremely rare on loans for which the loan-to-value (LTV) ratio is
less than 100%, but rise at an increasing rate as the current LTV ratio
goes above 100% (Goodman et al 2010; UBS 2005). This asymmetry
means that the pool average LTV ratio generally has not helped to
predict mortgage defaults. To illustrate, assume a mortgage pool
for which half of the mortgages have an LTV ratio of 50% and the

94

stress_testing.indd 94 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

other half have an LTV ratio of 130%. Given such a mortgage pool,
the average LTV ratio would be 90%, and a zero default rate would
probably be predicted, even though borrowers whose loans have
an LTV ratio of 130% would default at a high rate.
Banks typically have used single-equation, reduced-form regres-
sion models to forecast the annualised net charge-off rate quarterly for
each loan-type portfolio. For the most part, these regression models
are dynamic and specified as autoregressive models of order p, where
p lags of the charge-off rate (the dependent variable in these regres-
sions) are used as explanatory variables to account for the substantial
autocorrelation displayed by charge-off rates. Besides the reduced-
form autoregressive econometric models, a small number of banks
have been developing and using simple error-correction models to
forecast annualised net charge-off rates (Crook and Banasik 2012; As-
souan 2012). For all top-down models, quarterly net charge-offs are
derived by multiplying the net charge-off rate forecasts by balance
forecasts provided by the banks’ corporate treasuries.
A potential concern associated with banks’ use of autoregressive
forecasting models is the significant correlation between the mac-
roeconomic and financial risk drivers. This multicollinearity can
serve to mute the effect of shocks to the macroeconomic and finan-
cial variables on the charge-off forecasts. This is an important issue
in conducting effective and meaningful stress tests, since it may not
be possible to quantify the full impact of macrofinancial shocks on
credit losses when there is significant correlation among the macro-
economic and financial variables, as well as between these variables
and lagged values of the charge-off rate. Also, the use of an autore-
gressive model may prevent the charge-off rate (the dependent vari-
able) from responding quickly to a shock in the macroeconomic and
financial variables due to including lags of the charge-off rate as ex-
planatory variables.
As far as the development of the top-down charge-off models is
concerned, smaller mid-size banks (those close to the $10 billion as-
set threshold for banks required to conduct enterprise-wide stress
testing under the Dodd–Frank Act) may not have the financial re-
sources necessary to attract skilled statisticians and econometricians.
For these banks, then, an attraction of the pool-level models is that
ordinary least squares (OLS) estimation typically can be used.

95

stress_testing.indd 95 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

LOAN-LEVEL LOSS MODELS

Loan-level models are bottom-up models used by banks to forecast
the expected loss by retail and wholesale loan type for each loan.
The expected loss is calculated for each loan. In these loan-level
models, the sum of expected losses across all loans provides an es-
timate of portfolio losses.
The large US banks that are subject to the AIRB approach typically
have used the Pillar I definition of default in calculating PD, loss-
given default (LGD), and exposure at default (EAD) in the bottom-
up models. For the most part, account-level PDs are estimated using
logistic regressions that include both internal borrower-specific char-
acteristics and macroeconomic and financial variables as predictors.
In contrast, LGDs are usually estimated using a regression model at
the portfolio level that includes macroeconomic and financial risk
drivers. Finally, EAD for term loans equals the principal balance at
default, including currently undrawn commitments. In order to cap-
ture the larger positive correlation between PD and LGD during pe-
riods of stress, many banks use the same macroeconomic risk drivers
in their PD and LGD regression models. (Technically, AIRB models
use pools of loans, but Basel II specifies that the loans in each pool
must be essentially identical to each other in terms of credit risk, so
the AIRB models are defined as loan-level here.)
In contrast to the relative simplicity of top-down charge-off mod-
els, there are a variety of loan-level methodologies that can be used,
but these models are much more complex to specify and estimate.
Loan-level methodologies generally require more sophisticated
econometric and simulation techniques. Also, the macroeconomic
and financial risk drivers are often correlated with borrower-spe-
cific characteristics, and similar to the pool-level autoregressive
models discussed above, the multicollinearity among the regres-
sors serves to mute the effect of shocks to the macroeconomic and
financial variables on expected losses. This is an important issue in
conducting effective and meaningful stress tests, since it may not
be possible to quantify the full impact of macrofinancial shocks on
credit losses.
For many banks, loan-level models require data that are not
available, particularly as long historical time series. US banks that
are subject to the AIRB approach of Basel II have been maintaining

96

stress_testing.indd 96 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

detailed data on borrower-specific characteristics since the early

2000s. However, other banks generally find it to be a big challenge
to create historical data on borrower-specific characteristics. As not-
ed above, some banks have resorted to using vendors to provide
pool-level charge-off data. However, vendors may not be able to
provide borrower-specific data at a reasonable cost for these banks.
As a result, apart from the Basel II AIRB banks, the authors gener-
ally have observed historical loan-level data at other banks of at
most five years in length.
In principle, loan-level models would appear to be superior to
pool-level models in terms of forecast accuracy, since they use both
macroeconomic factors and loan-specific characteristics as predic-
tors of expected losses. However, the forecasting superiority of
loan-level models remains to be adequately demonstrated by out-
of-sample backtesting. Hughes and Stewart (2008) empirically ad-
dressed the issue of whether aggregate models of credit risk yield-
ed better forecasts of portfolio aggregates than loan-level models.
Using simulation techniques, they found that pool-level models
performed better than loan-level models.

Credit and behavioural scoring models with macroeconomic factors

A simple model used by banks combines credit scores, such as the
Vantage or FICO score, and macroeconomic factors to predict de-
faults or severe delinquencies, and then produces default probabili-
ties at the loan level. These loan-level estimates are then aggregated
to produce a pool-level default rate. An example of this approach
was provided by Hughes (2009), who showed that macroeconomic
factors had predictive power for defaults and delinquencies. Other
work also has documented the importance of macroeconomic factors
as predictors of loan-level defaults, deliquencies and credit scores
(Hughes 2008; Thomas 2000; Bellotti and Crook 2009).

Credit migration analyses

Many banks’ stress-testing models use credit-migration or roll-rate
analysis. Migration in this context refers to the movement of a loan
through a transition matrix of some kind, where the rows indicate
some measure of current credit quality, and the columns indicate
the next period’s credit quality.

97

stress_testing.indd 97 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Retail migration matrices are usually based on delinquency sta-

tus. For each loan type, the bank defines the number of days de-
linquent, at which point the loan is deemed to be in default. The
matrix is typically terminated at that point, although it can include
a column that shows the loan to have been converted to collateral.
A stylised migration matrix for first-lien residential mortgages is
shown in Table 7.2.
The transition of a loan from current to 1–30 days past due is
probably the most important metric for measuring the sensitivity to
macroeconomic factors. However, at least for retail credit, 1–30-day
delinquencies are very volatile. They display a significant seasonal
component, and consist of payments that are missed accidentally,
but then quickly cured, and so-called “rolling 30s”, where the bor-
rower has missed one payment that is never made up. As a result,
few, if any, banks use this delinquency bucket in their retail credit
modelling.

Table 7.2 A stylised retail transition matrix

To Current 30–60 60–90 days Default (90+ days
From days past past due past due or real-
due estate-owned)
Current 98% 2% 0 0
30–60 DPD 40% 20% 40% 0
60–90 DPD 20% 20% 20% 40%
Default 2% 4% 8% 20%

Note: The last row does not add to 100% due to the conversion of REO to cash.

The time-and-state invariant transition probabilities assumed for

the typical first-order Markovian transition matrix used by banks
are restrictive and not the most realistic default-modelling approach
(van Deventer 2009; Kiefer and Larson 2004). However, most banks
interpret the conditions loosely and calculate the PD as the product
of adverse credit migrations. In Table 7.2, assuming static transi-
tion probabilities, the PD of a current loan at the 90-day horizon is
(0.02*0.4*0.4), reflecting the probability of transition from “current”
(actually 1–30 days past due) to 30–60 days past due multiplied by
the probability of transitioning from 30–60 to 60–90 days past due

98

stress_testing.indd 98 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

multiplied by the probability of transitioning from that bucket to

default. Similarly, the default probability of a 30-day delinquent
loan at the 90-day horizon is (0.2*0.4*0.4)+(0.4*0.4), ie, the probabil-
ity that the borrower makes a payment but then misses the next two
payments plus the probability that it misses the next two payments,
and so on. There are several ways to convert the 90-day default rate
to the 12-month default rate typically used for retail loans.
The transition-matrix approach produces only the probabilities
of default, and not the LGD. As a result, modellers must estimate
the LGD separately. For most types of retail loans, the LGDs are
close to 100%, with the exception of first-lien residential mortgages,
where substantial recoveries are typically expected. Prior to the
housing bubble of the 2000s, the rule of thumb was an LGD of 20%
to 30% for prime residential mortgages (Qi and Yang 2009) and an
LGD of 40% to 50% for subprime mortgages.7 In the aftermath of
the crisis of 2007–9, however, that rule of thumb has become too
simplistic to apply. The alternative approach has been to calculate
the LGD of a loan as a function of the loan-to-value ratio at the time
of default plus an add-on for recovery costs. Frequently, the loan-
to-value ratio is modelled as a function of a scenario-specific home-
price index.
Wholesale credit migration matrices are usually based on the
credit ratings assigned to the obligors by the bank. Banks typically
have assigned 10 to 20 grades to wholesale debtors, rank-ordering
them in terms of their PD. For each rating grade, the PD at any ho-
rizon can be calculated as a historical average. Usually, the internal
ratings are based on the expert judgement of the loan officers, al-
though quantitative models are increasingly being used to supple-
ment the expert judgements.
Unlike for retail loans, most wholesale loans keep the same grade
from month to month. While most of the stability in ratings is prob-
ably real, some is undoubtedly due to inertia on the part of the loan
officers. Moreover, banks probably vary considerably in the dili-
gence of their loan officers in keeping the internal credit ratings up
to date.
For a given relationship between internal loan grade and default
rating, the stress-testing challenge for banks has been to estimate
how changes in economic conditions correlate with changes in loan

99

stress_testing.indd 99 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

grade. In the first years after 2008, the impact of changing economic
conditions tended to be based on the expert judgement of the loan
officers, but there has been movement towards the use of macro-
economic-based models. Since the macroeconomic models require
a history of internal loan-grade migrations, the macroeconomic
sensitivities of these migrations can be affected by the loan officers’
diligence and ability in maintaining up-to-date loan ratings.
For wholesale LGDs, many banks assign a rating analogous to
the PD ratings. Some banks combine the PD and LGD ratings into
one rating, which effectively produces an expected-loss rating. Oth-
er banks simply assign a historical-average LGD, which could be
model-based. Unlike retail LGDs, wholesale LGDs are highly sensi-
tive to macroeconomic and financial factors (Frye and Jacobs 2012).
In order to account for the changing risk profile of both retail
and wholesale portfolios at the loan level, the authors have ob-
served some banks using dynamic credit transition matrices that
are conditional on stressed economic scenarios (Bangia et al 2002).
Retail transition matrices are based on delinquency status, while
wholesale transition matrices are based on internal risk ratings.
These dynamic transition matrices, which are conditional on mac-
roeconomic and financial variables, require nonstationary Markov
chains (Grimshaw and Alexander 2011).
There are several different econometric approaches for estimating
transition probabilities that the authors have observed. The simplest
was to estimate the relationship between macroeconomic and finan-
cial variables and the PD using OLS. Technically, this is not a valid
approach, since the migration probabilities are bounded by zero and
one, while OLS assumes an unbounded distribution. However, for
most of the intermediate cells in a transition matrix, probabilities are
reasonably close to 50%, so the boundary issue is less of an issue.
However, the transition from current to 30-days past due is close to a
zero probability, while transitions from the late delinquency buckets
to default are close to a probability of one.
There are two possible solutions to this problem. One approach
is to find an alternative method for estimating initial and late-stage
transitions, but the authors have not seen such an approach imple-
mented by banks. A second method would be to standardise the
transition probabilities (ie, characterise observations by the number

100

stress_testing.indd 100 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

of standard deviations from the mean), thereby converting the tran-

sition probabilities from a bounded to an unbounded distribution.
The authors have seen this approach implemented, but, as of late
2012, there has not been sufficient time for banks to compare fore-
casts with actual results.
The preferred econometric approach is to use a logistic regres-
sion to combine borrower-specific characteristics with macroeco-
nomic variables to predict credit migration as a loan-level binary
outcome. Chen et al (2011) used this approach, and showed that
macroeconomic variables appeared to outperform borrower char-
acteristics in predicting credit migrations. They also found that
generalised maximum entropy outperformed a multinomial logit
model. However, as with the other models discussed here, they
did not include a substantial out-of-sample comparison of fore-
casts with actual results.

A variant on the use of migration matrices

In this approach, the modeller predicts the proportion of loans in
each migration bucket as a function of macroeconomic and finan-
cial variables, rather than using the transition probabilities as the
dependent variable. In other respects, such as estimating LGDs and
draws on unfunded commitments, the approach is the same as for
migration analyses. An alternative implementation would use a
multinomial logit or ordered probit regression model.

Charge-off rates across loan types

Since economic risk factors change at the same time for all banks, it
would be convenient for modellers if charge-offs for different loan
types responded with the same lag to changes in macroeconomic
and financial variables. Table 7.3 shows charge-off rates across loan
types, by year.
Table 7.3 shows that CRE charge-off rates tend to lag other
lending types, such as C&I, Consumer and HELOC loans, by
about one year. Otherwise, charge-off rates for all loan types in-
creased for the most part much in tandem. For example, charge-
off rates for both C&I and residential construction loans have
peaked at the same time.

101

stress_testing.indd 101 20/05/2013 18:45

stress_testing.indd 102

102

Stress Testing: Approaches, Methods and Applications

Table 7.3 Annual percentage charge-off rates, by loan type, industry aggregates (weighted means)

C&I Consumer Credit card HELOC Res. const. Owner- CRE 2nd-lien, 1st-lien
Occ. CRE closed-end residential
residential
2005 0.7 1.8 5.9 0.1 0.0 0.1 0.1 0.3 0.1

2006 0.6 1.5 4.3 0.2 0.1 0.1 0.1 0.4 0.1

2007 0.8 2.0 5.0 0.5 0.5 0.1 0.2 0.7 0.2

2008 1.4 2.7 6.3 1.9 4.2 0.1 0.5 3.1 0.9

2009 3.1 3.6 10.2 3.0 7.6 0.6 1.5 5.8 1.4

2010 2.4 2.7 10.9 2.8 7.8 0.9 2.4 5.2 1.4

2011 1.3 2.4 6.7 2.2 5.2 0.8 1.8 4.5 1.1

2012 (1H) 0.9 2.0 5.2 2.0 2.9 0.6 1.2 4.5 0.9

Qrtly peak 2009 Q4, 2009 Q2, 2010 Q1, 2009 Q4, 2009 Q4, 2010 Q4, 2010 Q4, 2009 Q4, 2009 Q4,
4.0 3.6+ 14.5 2010 Q1, 9.9 1.2 2.7 6.6 1.9
3.3
Source: Call Reports. The charge-off rates for 2012 are through June 30. The peak in credit-card charge-off rates was probably artificial, as FAS-167
required banks to put revolving-trust securitisations back onto the balance sheet, with a resulting charge-off surge
20/05/2013 18:45
 Stress-Test Modelling for Loan Losses and Reserves

ECONOMETRIC MODELLING ISSUES

In conducting onsite supervisory review during the past several
years, the authors have observed a common set of econometric is-
sues that arise when model developers incorporate macroeconomic
and financial variables into regression-based forecasting models.
These issues have generally fallen into those related to regression
specification and those related to estimation. Potential remedies
for the econometric issues identified below can be found in many
econometrics textbooks.

Specification issues
First, in choosing the set of macroeconomic and financial variables
to be included as explanatory variables in the regression models,
model developers have frequently used pairwise Pearson corre-
lation coefficients between the dependent variable, eg, charge-off
rates, and the various macroeconomic and financial explanatory
variables to determine the subset of economic variables considered
for the final econometric models. Typically, the correlation coeffi-
cients are rank-ordered and the three or four economic variables
with the largest correlation coefficients are considered for use in the
regression specification. This is a rather narrow approach to select-
ing the final explanatory variables. It is well known, for example,
that Pearson correlation coefficients can only detect linear associa-
tion between variables, and they do not capture significant dynam-
ic nonlinear relationships that could be present. They are also likely
to pick up spurious relationships that are driven by a third common
variable that is omitted from consideration.
Second, the lag lengths for the macroeconomic and financial
variables are typically chosen in an ad hoc manner. Although there
are optimal lag-length search procedures that could be used in
specifying the lag lengths, these very often are not used. This issue
would also apply to the choice of lag length for the dependent vari-
able used in autoregressive models. The choice of lag length for the
macroeconomic variables is important, since the use of lags that are
too short results in changes to the dependent variable that do not
capture the full impact of a macroeconomic shock.
Third, modellers typically have not engaged in a careful and
well-documented, empirically-based approach to the choice be-

103

stress_testing.indd 103 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

tween levels, log-levels or first differences of the charge-off rate

variables and the macroeconomic and financial variables in the
top-down regression models. This issue has also been observed for
bottom-up loan-level models. Generally, it would be preferable to
test the sensitivity of stress-test results to the choice of functional
form and variable transformations.
Fourth, there generally has been little or no attention paid to the
difference between trend-stationary series and difference-station-
ary series (ie, those time series with a unit root or stochastic trend).
This is considered to be an important specification issue for time-
series regression modelling. For most banks, it appeared that only
the possibility of difference-stationary data was considered in spec-
ifying the regression models. Spurious autocorrelations can easily
be induced in a time series showing trend, either by mistakenly
removing a deterministic trend from difference-stationary data or
by differencing trend-stationary data. Because of this, careful atten-
tion should be paid to the type of non-stationarity characterised by
the time-series data.
Fifth, the regression models used are for the most part speci-
fied as being linear in the macroeconomic and financial variables.
The assumption of linearity can impose important restrictions on
the responses of the dependent variable for stress-testing analysis.
For example, the linearity restriction imposes the following severe
properties: symmetry, ie, the magnitude of the responses is the
same regardless of whether the macro shock is positive or negative;
proportionality, ie, responses are proportional to the change in the
macro factor; and history independence, ie, the shape of responses
is independent of initial conditions of the macroeconomic variables
(Misina and Tessier 2008).
Sixth, the potential for a significant seasonal component in the
dependent and predictor variables has been typically ignored. For
example, charge-off rates display significant quarterly variation
that could be taken into account by including quarterly dummy
variables in the top-down regression specifications.

Estimation issues
First, there has been a lack of comprehensive diagnostics to assess
the validity of the estimated regression models, including func-

104

stress_testing.indd 104 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

tional form, variable selection and lag-length choice. For example,

there has typically been inadequate attention paid to the properties
of the regression residuals, such as serial correlation and heterosce-
dasticity, and the presence of outliers, influential observations and
structural breaks. Also problematic has been the use of the Durbin–
Watson (DW) statistic to test for serial correlation when lagged de-
pendent variables are used as regressors. The DW test is biased to-
wards accepting the null hypothesis of no serial correlation in such
cases, and therefore should not be used.
Second, outlier observations are frequently deleted from the da-
tasets used for estimation. Instead of doing so, robust estimation
techniques such as median regression should be explored to ad-
dress the issue.
Even when model development is sound and robust, the most im-
portant test of a model comes when out-of-sample forecasts are com-
pared to actual values in evaluating a model’s predictive accuracy.

CONCLUSION
The financial crisis of 2007–9, and the associated severe recession,
underscored the need for banks to incorporate economic and mar-
ket conditions into their retail and wholesale credit risk models in
order to produce credible stress loan-loss forecasts. While substan-
tial work has been conducted to develop macro-forecasting models
for retail and wholesale credit risk, especially at the largest banks,
the industry has not yet established what could be viewed as best
modelling practices. In this chapter, some of the internal modelling
practices that the authors have observed at national banks, where
macroeconomic and financial risk drivers have been incorporated
into the quantitative models used to generate retail and wholesale
credit loss forecasts and loan-loss reserve estimates, have been dis-
cussed. Some of the strengths and weaknesses of the various mod-
elling approaches have also been presented.

The views in this chapter are those of the authors and do not necessarily rep-
resent the views of the Office of the Comptroller of the Currency.

105

stress_testing.indd 105 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

REFERENCES
Assouan, Steeve, 2012, “Stress Testing a Retail Loan Portfolio: An Error Correction
Model Approach”, Journal of Risk Model Validation 6(1), pp. 3–25.

Bangia, Anil, Francis Diebold, André Kronimus, Christian Schagen, and Til
Schuermann, 2002, “Rating Migration and the Business Cycle, with Application to
Credit Protfolio Stress Testing”, Journal of Banking and Finance, 26, pp. 445-474.

Basel Committee on Banking Supervision, 2009, “Principles for Sound Stress

Testing Practices and Supervision”, Bank for International Settlements, May.

Basel Committee on Banking Supervision, 2012, “Models and Tools for Macropru-
dential Analysis”, Bank for International Settlements, Working Paper No. 21, May.

Bellotti, Tony, and Jonathan Crook, 2009, “Credit scoring with macroeconomic
variables using survival analysis”, Journal of the Operational Research Society 60, pp.
1699–707.

Chen, Qing Qing, Dennis Glennon and Amos Golan, 2011, “Estimating Condi-
tional Mortgage Delinquency Transition Matrices”, working paper, Office of the
Comptroller of the Currency, Washington, DC.

Choy, Murphy, and Ma Nang Laik, 2011, “A Markov Chain Approach to Deter-
mine the Optimal Performance Period and Bad Definition for Credit Scorecard”,
Research Journal of Social Science & Management 1(6), October, pp. 227–34.

Crook, Jonathan, and John Banasik, 2012, “Forecasting and explaining aggregate
consumer credit delinquency behaviour”, International Journal of Forecasting 28, pp.
145–60.

Floro, Danvee, 2010, “Loan Loss Provisioning and the Business Cycle: Does Capi-
tal Matter?: Evidence from Philippine Banks”, Bangko Sentral ng Pilinas, working
paper, March.

Foglia, Antonella, 2009, “Stress Testing Credit Risk: A Survey of Authorities’ Ap-
proaches”, International Journal of Central Banking 5(3) September, pp. 9–42.

Frye, John, and Michael Jacobs Jr., 2012, “Credit Loss and Systematic Loss Given
Default”, Journal of Credit Risk 8(1).

Furlong, Fred, and Zena Knight, 2010, “Loss Provisions and Bank Charge-offs in
the Financial Crisis: Lessons Learned”, FRBSF Economic Letter 2010–16, May.

Goodman, Laurie S., Roger Ashworth, Brian Landy, and Ke Yin, 2010, “Negative
Equity Trumps Unemployment in Predicting Defaults”, Journal of Fixed Income
19(4), Spring, pp. 67–72.

Grant Thornton LLP, 2012, “Allowance for Loan and Lease Losses (ALLL) Adjust-
ment Factors”.

Grimshaw, Scott, and William Alexander, 2011, “Markov Chain Models for De-
linquency: Transition Matrix Estimation and Forecasting”, DMM Forecasting.

Henderson, Christopher, 2009, “Retail Credit Risk Models: What Do These Mod-
els Look Like and How Did They Fare in the Crisis?”, FRB of Philadelphia, June.

106

stress_testing.indd 106 20/05/2013 18:45

 Stress-Test Modelling for Loan Losses and Reserves

Hughes, Tony, and Robert Stewart, 2008, “Forecasting and Stress Testing Using
Model Pool Level Data”, Moody’s Analytics, August.

Hughes, Tony, 2008, “The Macroeconomics of Credit Scores”, Moody’s Analytics.

Hughes, Tony, 2009, “The Economic Credit Score”, Moody’s [Link].

Kiefer, Nicholas M., and C. Erik Larson, 2004, “Testing Simple Markov Structures
for Credit Rating Transitions”, OCC Working Paper Series, 2004-3.

Malik, Madhur, and Lyn Thomas, undated, “Modelling Credit Risk in Portfolios
of Consumer Loans: Transition Matrix Model for Consumer Credit Ratings”,
working paper, University of Southampton, United Kingdom.

Misina, Miroslav, and David Tessier, 2008, “Non-Linearities, Model Uncertainty,

and Macro Stress Testing”, working paper, Bank of Canada.

Pocock, Mark, 2012, “Perspectives on Recent Allowance for Loan and Lease
Loss Modeling Issues”, Office of the Comptroller of the Currency, working paper
(preliminary), Washington, DC.

Qi, Min, and Xiaolong Yang, 2009, “Loss Given Default of High Loan-to-Value
Residential Mortgages”, Journal of Banking and Finance 33, pp. 788–99.

Qu, Yiping, 2008, “Macro Economic Factors and Probability of Default”, European
Journal of Economics, Finance and Administrative Sciences (13), pp. 192–215.

Schuermann, Til, 2012, “Stress Testing Banks”, Wharton Financial Institutions

Center, working paper, April.

Tanta, 2007, “Delinquencies and Defaults for UberNerds”, CalculatedRiskBBlog.

com, July.

Thomas, Lyn, 2000, “Survey of Credit and Behavioural Scoring: Forecasting Financial
Risk of Lending to Consumers”, International Journal of Forecasting 16, pp. 149–72.

UBS, 2005, Mortgage Strategist, December 13.

Van Deventer, Donald, 2009, “Reduced Form Macro Factor and Roll Rate Models of
Mortgage Default: An Introduction and Application”, Kamakura Corporation, October.

1 Macroprudential bank stress tests are used to assess the key vulnerabilities of the banking
system as a whole to macroeconomic and financial shocks. See Foglia (2009), BCBS (2012)
and Schuermann (2012) for details.
2 The 2009 SCAP used three macroeconomic variables in its state space for the banking-book
stress test: GDP growth, unemployment and house price index. The CCAR bank stress tests
in 2011, 2012 and 2013 focused on estimates of projected revenues, losses, reserves and pro
forma capital levels under a baseline, a severe and an adversely severe stress economic sce-
narios. See Schuermann (2012) for a more detailed discussion.
3 Henderson (2009) discussed banks’ retail credit risk models, their features and how the
models performed during the financial crisis of 2007–9.

107

stress_testing.indd 107 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

4 See SFAS 5, SFAS 11, SFAS 112, and SFAS 114. While those in the industry still referred to
these as “FAS-5” and “FAS-114”, SFAS 5 has become ASC 450 and SFAS 114 has become
ASC 310. The primary components of a bank’s ALLL consist of loans collectively evaluated
for impairment (the FAS 5 part), loans individually evaluated for impairment (the FAS 114
part) and loans acquired by the bank with decreased credit quality (the SOP 03-3 part). See
Thornton (2012) for a detailed discussion.
5 The academic literature has distinguished between non-discretionary and discretionary fac-
tors used by banks in setting their loan-loss provisions and reserves. Non-discretionary fac-
tors refer to those related to changes in loan credit quality and economic conditions, while
discretionary factors refer to those related to income smoothing, regulatory capital manage-
ment and private information signalling. Floro (2010) conducted an empirical study of the
two sets of factors used by banks in setting their loan-loss reserves.
6 For nearly all the types of models discussed here, many model developers have used the
Pillar I convention of Basel II and have modelled exposure at default, LGD and probability
of default separately in estimating expected losses. These models are more complex and less
parsimonious than those that forecast charge-offs directly and, therefore, subject to a higher
degree of modelling uncertainty. Despite the loss of parsimony, however, there is evidence
that the pool-level charge-off models perform no worse than those using a more granular
expected loss approach (Frye and Jacobs 2012).
7 The latter is based on the authors’ conversations with industry experts in residential-mort-
gage recoveries.

108

stress_testing.indd 108 20/05/2013 18:45

 8
A Framework for Stress Testing
Banks’ Corporate Credit Portfolio

Olivier de Bandt, Nicolas Dumontaux,

Vincent Martin, Denys Médée
Autorité de Contrôle Prudentiel (French Prudential Supervision Authority)

Since the financial crisis of 2007 and beyond, which drew unprec-
edented attention to the stress testing of financial institutions,
stress-tests exercises have become a central risk-management tool
to assess the potential impact of extreme events on banks’ P&L and
balance-sheet structures.
Stress tests are viewed as complementary to traditional risk-
measurement metrics such as value-at-risk (VaR), as they are an
important mechanism for detecting weaknesses of both a single
financial institution and threats to financial stability. Nowadays, fi-
nancial institutions are required to perform regular exercises within
Pillar II of the regulatory framework of the Basel Accord in order
to assess the global impact of adverse events or changes in market
conditions on banks’ capital adequacy. Supervisory authorities as
well are used to leading such exercises: the International Monetary
Fund (IMF) with its regular Financial Sector Assessment Program,
the European Banking Authority (EBA) with its European “bottom-
up” stress tests, including a disclosure step, and national supervi-
sory authorities, which all have built dedicated tools, especially for
regular top-down exercises. The scope of stress testing includes
traditional credit risks, market risks, operational risks, interest-rate
risks and, since the financial crisis, liquidity risks.
Stress testing corporate credit risk, also known as “wholesale
credit risk”, is a key component of stress testing for global institu-
tions. Credit risk in itself (including retail credit risk) is indeed one

109

stress_testing.indd 109 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

of the major sources of risk for banks, judging by the extent of banks’
credit risk-weighted assets (RWAs),1 and, accordingly, may have a
major impact on the solvency of financial institutions. The subprime
crisis has highlighted the need for stress testing banks’ portfolios as
numerous credit institutions incurred major losses and write-downs
from structured US subprime related assets since mid-2007.
This chapter examines stress testing for credit risk, focusing on
risks arising from corporate loans and other credit exposures.2 It
aims at introducing a Basel II-type modelling framework to per-
form credit stress-test scenarios through credit migration matrices
(or transition matrices), which has been implemented by French
authorities and is currently used as a tool for top-down stress-test
exercises. This approach is still relevant under the Basel III frame-
work, since nothing new has been introduced in the Basel III frame-
work with respect to the assessment of credit risks of banks’ corpo-
rate portfolios.
The chapter is organised as follows: the first section briefly de-
picts the model our stress-test framework rests on, which is largely
based on the Merton model; then we introduce the way this frame-
work is implemented to conduct top-down stress-test exercises; fi-
nally, we comment on a few outputs of the stress tests.

THE MODEL SPECIFICATION

Several models are available for quantifying credit risk, this risk
stemming either directly from actual defaults of credit exposures,
or indirectly from migrations of credit ratings, taking into account
their prudential treatment.3 These models may be either structural
(modelling of firms’ value and capital structure) or reduced forms,
where credit events are exogenous to the firms. Here we rely on
the latter approach, with credit events triggered by macroeconomic
shocks, and focus on credit migrations (see ECB, 2007, for alterna-
tive industry credit models).
The model we introduce in this section relies on the basic idea
that the evolution of rating transitions can easily be linked to a syn-
thetic credit indicator.4 The main hypothesis is that the underlying
asset value of a firm evolves over time, through a simple diffusion
process, and that default is triggered by a drop in firm’s asset value
below the value of its callable liabilities. In the Merton framework,

110

stress_testing.indd 110 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

shareholders actually hold a call option on the firm, while debt-

holders hold a put option.

General specification
Based on the assumption that changes in the logarithm of the firm’s
asset value (ΔlogAi) can be related to both a systemic factor (Z, the
credit index) and an idiosyncratic factor (𝜀𝑖) via a factor model, the
specification is the following for firm i:

∆log𝐴𝑖,𝑡 = –√𝜌𝑍𝑡+√1−𝜌𝜀𝑖,𝑡 + c𝑖

where ci is the long-run growth of firm i’s asset value and r is the as-
set correlation between any two firms in the portfolio. All firms are
supposed to have identical characteristics (eg, their correlation to
Z) with respect to their credit rating, which then leads us to identify
i as a credit class rather than an entity.
Assuming, then, that changes in asset value are normally distrib-
uted (Z and 𝜀𝑖 are mutually standard normal variables and mutu-
ally independent), the default probability may be expressed as the
probability of a standard normal variable falling below a critical
value, defined as the different ratings (with a total of n classes, with
n=8). Similarly, thresholds can be set up for rating migrations, as
graphically represented in Figure 8.1.

Figure 8. 1 Asset return distribution with rating thresholds for BBB issuers

2.5%

2.0%
ZBB ZBBB
Probability

1.5%

ZA
1.0%

0.5%
ZB ZAA
ZCCC
ZC
0.0%
-4

-3

-2

-1

4
-3.75

-3.25

-2.75

-2.25

-1.75

-1.25

-0.75

-0.25
-3.5

-2.5

-1.5

-0.5

0.25

0.75

1.25

1.75

2.25

2.75

3.25

3.75
0.5

1.5

2.5

3.5

Asset return over 1 year

111

stress_testing.indd 111 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

This framework, on which is also based the Basel II Asymptotic

Single Risk Factor (ASRF) model, results in the following relations
with respect to default probabilities:

 Φ −1 ( pi ) + ρ Φ −1 (α )
PDi Zt −1
α  
 1− ρ 

where PDi |{...} is the (conditional) default probability in state Z, F is

the Gaussian cumulative distribution function, pi is the long term av-
erage probability of default (PD) of class i (i=1,…n) (or unconditional
PD) and α is the probability that the value Z (or below) occurs. The
closer α is to 0, the less frequent is the crisis and the greater its severity.
In the Basel II framework, α is equal to 0.1% (this corresponds to the
regulatory confidence level of 99.9%, which is the required confidence
level to compute regulatory capital requirements under Basel II & III
frameworks). In our application we use n=8 risk classes, where class
8 stands for the default class. Default probabilities along with rating
migrations depend on a sole parameter (Zt, the credit latent index
discussed below), meaning that migrations matrices can be modelled
through one macro factor (ASRF: Asymptotic Single Risk Factor).
This formula, expanded to every component of the nxn (here
8x8) transition matrix, is:

 Φ−1 ( Pi 8 + ... + Pij + ρ Zt 

Pijt = Φ   − Pi 8t − ... − Pi , j +1 ,t
 1− ρ 

This approach, which aims at representing transition matrices by

a single parameter, was studied by Belkin, B., Forest, L. R. Jr and
Suchower, S. J. (1998). They follow the CreditMetrics framework pro-
posed by Gupton, Finger and Bhatia (1997).

Estimation of the correlation factor

The correlation factor 𝜌 between the obligor i and the general state
of the economy (in the ASRF model, all obligors are linked to each
other by this single risk factor Z, which reflects the general state of
the economy) has been estimated in order to obtain the best pos-
sible fit of historical data by the model, under the hypothesis that
the correlation is the same for all obligors (and thus rating classes).

112

stress_testing.indd 112 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

THE STRESS-TESTING FRAMEWORK

The following framework is based on a relationship between the
latent credit index and the macroeconomic situation. This link is
indeed fundamental, since most stress-test exercises start with the
choice of a set of macroeconomic stress scenarios. Those stress sce-
narios are then linked to risk parameters – default rates (DRs), loss
rates, regulatory PDs, regulatory LGDs, transition matrices – which
will, in the end, affect banks’ solvency.
Our approach is based on an intermediary variable, namely the
aggregate DR. First, we measure the link between GDP (or the mac-
roeconomic scenario) and the DR, then between the DR and the la-
tent credit index, in order to compute the stressed transition matrix.
The different steps described here are based on relationships un-
covered for France, in particular regarding the link between credit
risk and the macroeconomy, but we explain how they may be repli-
cated for other institutions/countries.

The data
Adequate data is of course necessary for calibrating the model. While
specific information on the loan portfolios of institutions are essen-
tial, we show how to rely on S&P transition matrices,5 a method that
can be to some extent transferable to other institutions/countries as
long as they have a similar global portfolio of corporate loans.

Prudential Common Reporting

Our framework basically requires information on the structure of
banks’ portfolios by types of rating. They are available from banks’
Prudential Common Reporting (COREP).6 COREP is a set of Eu-
ropean harmonised data on solvency issues (own-funds adequacy,
credit RWAs, market RWAs, operational RWAs) handled by the
EBA. It intends to enhance the level of harmonisation of the su-
pervisory reporting. A specific COREP template is dedicated to the
credit risk of IRB corporate portfolios, in which the regulatory PDs
for each class of risk are reported to the French authorities by banks
in the quarterly prudential COREP templates.
This information is actually combined with mappings (provided
by the onsite-inspections division) that convert the internal rating
system of each bank into the S&P rating scale. This step is facilitated

113

stress_testing.indd 113 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

by the fact that most banks use such a conversion scale to compute,
when possible, a distance between their internal rating and agency
ratings, as an indicator for assessing the performance of their inter-
nal models. This is a necessary step in order to stress banks’ portfo-
lios by using S&P transition matrices.

The S&P transition matrices

Our framework takes advantage of the S&P CreditPro database, which
contains issuer ratings history for 15,726 obligors over the 1981–2011 pe-
riod, of which 2,127 ended in default. The obligors are mainly large cor-
porate institutions – sovereigns and municipals are excluded – and pools
include both US and non-US industrials, utilities, insurance companies,
banks and other financial institutions, and real-estate companies.7
Over the past two decades from 1990 to 2011, three major business
cycles can be distinguished: (i) the recession that took place in the
wake of the First Gulf war at the beginning of the 1990s (GDP growth
dropped to −0.3% in 1991 in the US and to −0,7% in 1993 in France);
(ii) the burst of the Internet bubble (GDP growth dropped in the US
to 1.1% in 2001 from over 4% from 1998 to 2000, while in France GDP
growth declined to 0,9% in 2002 and 2003 from over 3.5% in 2000);
and (iii) the 2007-and-beyond subprime crisis (both American and
French GDP growth dropped to less than −3% in 2009).

Figure 8.2 Default rate according to S&P (left scale) and GDP growth
(right scale) over the 1990–2011 period

7.0% 6.0%
5.0%
6.0%
4.0%
5.0% 3.0%
2.0%
4.0%
1.0%
3.0%
0.0%

2.0% -1.0%
-2.0%
1.0%
-3.0%
0.0% -4.0%
1990
1991
1992

1994
1995
1996
1997

1999
2000
2001
2002

2004
2005
2006
2007

2009
2010
2011
1993

2003
1998

2008

United States Europe GDPg Fr GDPg US

114

stress_testing.indd 114 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

During each of these periods, DRs surged both in Europe and the
US (see Figure 8.2). It is especially striking after the burst of the
Internet bubble, during which (i) the DR of American corporates
reached the level of 4.5% (2% in Europe) and (ii) the total amount of
debt-defaulting was historically high due to failures of major com-
panies (Enron, WorldCom, Parmalat and so on). DRs during the
subprime crisis surged even higher in the US (5.70% in 2009).

Figure 8.3 Default rate according to S&P over the 1990-2011 period for
both speculative-grade (SG) and investment-grade (IG) obligors

9.0%
8.0%
7.0%
6.0%
5.0%
4.0%
3.0%
2.0%
1.0%
0.0%
1990
1991
1992

1994
1995
1996
1997

1999
2000
2001
2002

2004
2005
2006
2007

2009
2010
2011
1993

2003
1998

2008

US IG US SG EU IG EU SG

If DRs, and, more globally, credit migrations, are therefore clearly

linked to the economic context, it turns out that default events main-
ly involve speculative-grade obligors (rated BBB and below). Invest-
ment-grade obligors are much less sensitive to the business cycle,
underlining two different dynamics for investment-grade corporates
on the one hand, and for speculative-grade (which is actually the
main driver of the global DR) on the other hand (see figure 8.3).
In Table 8.1, annual S&P transition matrices displaying probabili-
ties to migrate from one rating to another are based on a “static pool
approach”. Credit-migration rates are computed by comparing rat-
ings on the first and last days of the year to construct the migra-
tion rates. Rating movements within the year are accordingly not
counted. This estimation approach, based on average behaviour,
does not actually capture rare events such as back-and-forth transi-
tions or series of consecutive downgrades within the year. Default

115

stress_testing.indd 115 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

is considered to be an absorbing risk class: if a recovery from default

may be observed, it is extremely rare. Usually, firms having defaulted
are excluded from the pool the following year, which prevents the re-
covery trajectory to ever be caught in a one year transition matrix.

Table 8.1 S&P average credit rating transition matrix (1990–2011)

AAA AA A BBB BB B CCC, D
CC, C
AAA 90.2% 8.9% 0.5% 0.2% 0.0% 0.0% 0.2% 0.0%

AA 0.3% 89.3% 9.8% 0.6% 0.0% 0.0% 0.0% 0.0%

A 0.0% 2.1% 92.3% 5.1% 0.2% 0.1% 0.0% 0.0%

BBB 0.0% 0.1% 4.2% 91.2% 3.4% 0.7% 0.1% 0.2%

BB 0.0% 0.1% 0.4% 5.1% 86.9% 6.4% 0.5% 0.6%

B 0.0% 0.0% 0.2% 0.5% 7.1% 82.7% 4.8% 4.7%

CCC,
0.0% 0.0% 0.3% 0.6% 1.1% 13.0% 58.1% 27.1%
CC, C
D 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 0.0% 100.0%

As we have seen, migration matrices may be driven by a standard

normal distribution – (almost) without losing any information. Mi-
grations are then not depicted by migration rates but through a set
of thresholds that depend on the latent variable Z.
As an example, we suppose that an issuer is currently rated A.
Table 8.2 and Figure 8.4 show the migration probability from the
current A rating to any of the other eight ratings, together with the
corresponding threshold from a standard normal distribution.

Table 8.2 Migration rates and scores for an A-rated issuer (1990–2011)
AAA AA A BBB BB B CCC, D
CC, C
Migration
0.0% 2.1% 92.3% 5.1% 0.2% 0.1% 10.0% 0.0%
rates
]3.61 ; ]2.03 ; ]-1.60 ; ]-2.64 ; ]-2.88 ; ]-3.25 ; ]-3.30 ; ]-∞ ;
Score bins
+∞[ 3.61] 2.03] -1.60] -2.64] -2.88] -3.25] -3.30]

116

stress_testing.indd 116 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

Figure 8.4 Rating thresholds and transition probability distribution for

an A-rated issuer

0.45

0.4

0.35
D B BBB A AA
0.3

0.25

0.2

0.15

0.1

0.05

0
-4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

Under adverse economic conditions, the normal distribution of rat-

ing migration would shift to the left, implying worse ratings levels
(see figure 8.5), meaning that the probability of downgrade and de-
fault increases. As the whole credit-migration matrices are driven
by a single-parameter Z, which depicts the average financial health
of corporate institutions (credit index), this shift corresponds to a
simple change in the value of Z.

Figure 8.5 Rating thresholds and transition probability distribution for

an A-rated issuer after a shift to the left (−0.5)

0.45

0.4

0.35
D B BBB A AA
0.3

0.25

0.2

0.15

0.1

0.05

0
-4.0 -3.5 -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

117

stress_testing.indd 117 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

The economic situation: linking GDP to the aggregate default rate

In order to calibrate the model, we need to measure the link be-
tween the macroeconomic environment and defaults. The results
presented here are specific to the French situation, hence would
need to be re-estimated to implement it inn other countries. How-
ever, the method is quite general and can be replicated, following
the same steps. This implies:

o defining an economic situation scale, as a percentage deviation

from maximum defaults;
o linking defaults to (national) macroeconomic determinants; and
o mapping the aggregate DR into the latent credit index, and we
offer a numerical method for doing that.

The model we present is mainly designed to compute RWAs,

through PDs linked to the macroeconomic environment. The im-
pact of stressed scenarios on P&Ls, including credit losses, is com-
puted through another model (Coffinet, Lin, Martin, 2009; Coffinet
and Lin, 2010).

The economic situation scale

Our economic situation index is the DR since this is both highly cor-
related to macro-variables like the GDP and directly linked to the
situation of corporate institutions.
In our stress-test framework, the state of the economy is accord-
ingly measured on the following scale:
DR t DR
λt =
DRcrisis DR
where DR t isDR the DR forecast at DR t and
t DR is the average DR over
λ = λt =
thet sample period
DRcrisis DR and DRcrisis
is the DR reached during the worst
DRcrisis DR
crisis observed over the period under observation (in our example,
this is computed as a mean of yearly DR for the years 1991, 2001,
2002, 2009). lt equals 0 on average over the business cycle, 1 when
the reference crisis is reached. If lt equals 0.33 for example, the eco-
nomic situation would be one-third of the maximum historical de-
viation from the average DR.8 This scale is unbounded so as to suit
stress scenarios which never occurred previously.

118

stress_testing.indd 118 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

Forecasting the default rate in a stress-testing exercise

The link between the DR and the macroeconomic environment may
obviously be subjected to more discussions than those presented
below. We aim through this couple of equations to provide some
alternative specifications for the link between the DR and the eco-
nomic situation, using different macroeconomic variables. These
equation can be used in order to project the DR over the simulation
horizon of the stress tests (each scenario would consist of time se-
ries of GDP, inflation, interest rates and so forth) that would be fed
into the equation to get a DR scenario. The equations are estimated
by Ordinary Least Squares. These equations should be re-estimated
for implementing our model to other countries.

DRt = 2.565*** + 0.579 *** DRt−1 − 0.379 ***GDPgt − 0.887 *** INFLt−1
( 0.0003 ) ( 0.0038 ) ( 0.0010 ) ( 0.0021 )

R = 0.67 , DW = 2.17
2
(8.1)

Where DR is the default rate, GDPg is GDP growth, INFL is the inflation

DRt = 2.673 *** − 0.372*** GDPgt − 0.705 ** INFLt−1

( 0.0019 ) ( 0.0077 ) ( 0.0345 )

R 2 = 0.39 , DW = 1.05 (8.2)

DRt = 3.584 *** + 0.478 ** DRt−1 − 0.348*** GDPgt − 0.798*** INFLt−1 − 0.125 URt
( 0.0077 ) ( 0.0286 ) ( 0.0032 ) ( 0.0074 ) ( 0.3278 )

R 2 = 0.70 , DW = 2.15 (8.3)

Where 𝑈𝑅𝑡 stand for the unemployment rate at t.

DRt = 2.4512*** + 0.5718 *** DRt−1 − 0.3744 ***GDPgt − 0.8508 *** INFLt−1 + 0.050 7 SPREAD
( 0.0012 ) ( 0.0055 ) ( 0.0017 ) ( 0.0050 ) ( 0.6097 )

R 2 = 0.68 , DW = 2.12
(8.4)

Where SPREAD is the spread between the interest rates on the ten-
year French Treasury note and the three-month Euribor.
Among the prominent points highlighted by these equations are
as follows.

o THE INERTIA OF THE DR: The DR inertia, ie, the autoregressive

coefficient in equations (1), (3) and (4), is both strong and signifi-
cant: past values of the DR provide good forecasting results.

119

stress_testing.indd 119 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

o THE INDICATORS OF THE STATE OF THE ECONOMY: Ac-

cording to the range of econometric tests, led by the ACP, the
most relevant economic variables with respect to DR forecasting
are GDP growth and the unemployment rate.
o THE FINANCIAL ENVIRONMENT. The spread between long-
term interest rates (10y) and short-term ones (3m) used in equa-
tion (4) is both classical and relevant. However, its impact is often
mild and positive.9

In addition, it might be interesting to integrate feedback effects

between defaults and the business cycle (Bruneau, de Bandt and
El Amri 2012).

The “conversion” function: mapping the aggregate default rate into

the latent credit index to generate the stressed transition matrix
The final step is to map our time series of defaults (more precisely of
our economic situation scale, which is a simple transformation of de-
faults, as indicated earlier) into our latent macroeconomic systemic
factor on which the transition matrices are based. We provide here a
numerical method for doing that, which could be easily replicated.
We thus define a second conversion scale in order to convert the cri-
sis percentage into the corresponding stressed transition matrix.
For that purpose, we consider the actual transition matrix observed
when the economy enters into the recession periods mentioned above
(namely in 1991, 2001–2 and 2009).10 We compare that matrix to the un-
conditional transition matrix over the sample period. The latter matrix
can be viewed as a through-the-cycle transition matrix.
More precisely, we define a scale based on the couple 𝑍0%𝑐𝑟𝑖𝑠𝑖𝑠 and
𝑍100%𝑐𝑟𝑖𝑠𝑖𝑠,which are the two credit indexes that respectively best fit
(i) the through-the-cycle transition matrix (𝑇𝑀𝑇𝑇𝐶) as the average
of transition matrices observed over the period and (ii) the “crisis”
transition matrix (𝑇𝑀𝑐𝑟𝑖𝑠𝑖𝑠) as an average of the transition matrices
observed in 1991, 2001, 2002 and 2009. Note that we chose a simple
average transition matrix conditional on exceeding the probability of
82% (which would be equivalent to the 4 worst observations in a 22-
year sample). However, we could have used variable weights over
time to compute an average transition matrix highlighting different
crisis episodes (banking- or industry-related crisis).

120

stress_testing.indd 120 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

From the comparison of the observed and asymptotic single-

risk-factor (ASRF) transition matrices we derive the value of the
latent macroeconomic systemic factor corresponding, respectively,
to “normal times” and “crisis”:

Zˆ 0%crisis = arg min TMTTC − TM( Zt)

 Zt
ˆ
Z100%crisis = arg min TMcrisis − TM( Zt)
 Zt

Several kinds of matricial norms can be used: we chose the Euclid-

ian norm, typically used in linear optimisation problems. Assuming
the relationships 𝑍𝑡 = 𝑍0% ⟺lt = 0 and 𝑍𝑡 = 𝑍100% ⟺lt= 1 between the
state of the economy and the credit index, the final step is to com-
pute the value of the macroeconomic systemic risk, which comes
from the following “conversion” function:

Zt = [(Z100%𝑐𝑟𝑖𝑠𝑖𝑠 – Z0%𝑐𝑟𝑖𝑠𝑖𝑠) lt + Z0%𝑐𝑟𝑖𝑠𝑖𝑠]

+

where lt depends on the spread between the long-term average DR

and the one forecast over the stress horizon. The stressed transition
matrix is then used to compute the level of RWAs (under the large
corporate parameters of the Basel II formula, using regulatory PDs).
We should mention that, in our stress-test framework, the transi-
tion matrices do not depend on one parameter Z but on two pa-
rameters (𝑍𝑖𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡 𝑔𝑟𝑎𝑑𝑒; 𝑍𝑠𝑝𝑒𝑐𝑢𝑙𝑎𝑡𝑖𝑣𝑒 𝑔𝑟𝑎𝑑𝑒) in order (i) to stick to a sim-
ple approach – we could have as many credit indexes as notches
within S&P transition matrices – and (ii) to reflect the two distinct
regimes followed by investment-grade and speculative-grade ob-
ligors. Indeed, as depicted earlier in this chapter, global DRs are
largely driven by the credit quality of speculative-grade counter-
parties, so that the identified crisis periods are periods of crisis for
speculative-grade obligors, rather than for investment-grade ones.
Practically, this means that two sets of parameters (𝑍0%𝑐𝑟𝑖𝑠𝑖𝑠, 𝑍100%𝑐𝑟𝑖𝑠𝑖𝑠)
have been estimated, one on the investment sub-part of the TTC
and crisis matrices, and one on the speculative sub-part.

121

stress_testing.indd 121 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

NUMERICAL APPLICATION
We present now a few outputs of the model for stress testing. First,
we provide more detailed information on necessary inputs, namely
banks’ exposures. We then use the model presented in the section
“The stress-testing framework” above to compute ratings migra-
tion on banks’ portfolios, hence to compute the level of RWAs un-
der stress. We present the aggregate results for our sample of five of
the largest French banks in a baseline and a stressed scenario.

Composition of French large banks’ corporate credit portfolio and

evolution over time of exposures at default
An initial portfolio is made up of corporate credit exposures of the
five largest French banks. Information on exposures and risk pro-
file11 is available in COREP reports. Banks’ portfolios are relatively
diversified in terms of sectors. Most exposures are investment-
grade and are mainly located in Europe and North America.
Based on this information, we need to compute the evolution
over time of exposures at default (EAD).
Let us assume that the horizon of the following exercise is two
years, so, starting from end-year 0, the stress test ends Year 2.
Starting with a portfolio of assets in different rating categories,
we compute how the portfolio changes over time following a shock.
This implies computing a stressed transition (or migration) matrix
with the probability of moving from one rating category to another.
Technically, this is a Markov chain matrix, assuming that all infor-
mation at time t+1 is contained at time t.
Furthermore, banks’ balance sheets are supposed to be static (as
opposed to dynamic), meaning that the total non-defaulted expo-
sures remain stable over the stress period. The assumptions made
in our example could very well be modified.

Calculation of risk-weighted assets and capital requirements for

credit risk.
Formally, considering an initial portfolio with a given risk structure
𝐸𝐴𝐷0,𝑖 at time 0 with i=AAA, AA,…,D, the dynamic behaviour of
the portfolio has the following form:

𝐸𝐴𝐷𝑡,𝑖=𝐸𝐴𝐷𝑡−1,𝑖[𝑇𝑀𝑡𝑠𝑡𝑟𝑒𝑠𝑠(𝑍𝐼𝐺;𝑍𝑆𝐺)+∆𝐸𝐴𝐷𝑡]

122

stress_testing.indd 122 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

The portfolio risk structure depends on the credit migration matrix,

which is a function of macroeconomic and financial factors . ∆𝐸𝐴𝐷𝑡
is the growth of new loans; it is adjusted so as to comply with the
static balance-sheet constraint. Regulatory capital requirements are
then calculated according to the Basel II formula.
We consider the two hypothetical following scenarios:
o a baseline scenario based on GDP growth projections from the
IMF’s World Economic Outlook (WEO); and
o an adverse scenario that is supposed to lead to a maximum cu-
mulated deviation from baseline of two standard deviations of
GDP growth for 2012–13.

Table 8.3 shows the main key macroeconomic factors that drive our
two scenarios.

Table 8.3 Macroeconomic variables’ forecast used for a stress-test

simulation
Baseline Adverse
2012 2013 2012 2013
Inflation (%) 1.7 1.5 1.3 0.2

GDP real growth (%) 0.5 1.0 −1.9 0.0

The main outcomes under these two scenarios, in terms of risk pa-
rameters (regulatory PDs) and capital requirements (RWA levels),
which are the main final output of our stress-testing framework, are
displayed in Table 8.4. Changes in RWAs in the table are computed
as the sum of changes in RWAs over five of the largest French banks.

Table 8.4 Main outcomes under the baseline and adverse scenarios

Baseline Adverse
2012 2013 2012 2013
Stressed regulatory PDs +2% +12% +15% +11%
(annual rate of change)
RWAs (annual rate of +2.6% +5.3% +12.9% +5.9%
change)

123

stress_testing.indd 123 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Table 8.4 shows the outcome of this sensitivity analysis in which

regulatory PDs12 and migration rates were stressed over the 2012–
13 period. It is worth noticing that other regulatory parameters,
such as LGD and correlations, have not been stressed in this par-
ticular exercise. The results in Table 8.4 illustrate the existence of
a smoothing effect of the stressed scenario on RWAs, due to the
negative relationship between regulatory PDs and the correlation
with the credit index (𝜌), as assumed in the internal-ratings-based
(IRB) model.
The outcome of this simulation shows that the sensitivity of IRB
minimum capital requirements to increases in regulatory PDs and
credit migrations is significant. Indeed, an increase of PD by 15%
in 2012 (resp. 13% in 2013) in the adverse scenario raises capital re-
quirements by about 11% (resp. 5.9% in 2013). Moreover, the initial
shock, a deep recession in 2012, raises both risk parameters and
capital requirements at least up to 2013. This is consistent with our
expectations, since Basel II formulas rely on through-the-cycle PDs,
which tend to smooth the impact of the shock at the very beginning
of the stress but makes it last longer.

CONCLUSION
Credit risk remains one of the most important risks faced by com-
mercial banks. This chapter provides a stress-testing framework for
banks’ corporate credit portfolios, a framework that is currently
used by French authorities to perform biannual top-down exercises.
This framework is therefore appropriate for data available at a
supervisory authority level and aims to achieve the best trade-off
between simplicity and robustness. Our stress-test framework takes
advantage of the quarterly prudential COREP templates and of the
S&P CreditPro database, which provides statistics over the previ-
ous two decades regarding credit migration of more than 10,000
American and European companies.
The calibrations proposed – namely AR models for observed
DRs, which assume stationary explanatory variables as well as
mean reversion dynamics – are consistent with the Basel II frame-
work, which relies on through-the-cycle risk parameters (PDs). This
framework is therefore fully relevant for benchmarking bottom-up
exercises. Furthermore, from a regulatory point of view, this frame-

124

stress_testing.indd 124 20/05/2013 18:45

 A Framework for Stress Testing Banks’ Corporate Credit Portfolio

work is a realistic approach to how banks compute their RWAs:

regulatory parameters, such as PDs and LGDs, are estimated as
through-the-cycle parameters, possibly with an add-on coefficient
for prudence (taking into account downturn economic conditions
for LGD). As a consequence, they tend to become less and less sen-
sitive to a given stress period, given that they are based on ever-in-
creasing historical datasets. It is indeed important in our view that
the stress-testing framework be as close as possible to the actual
regulation governing the computation of RWAs.

Views expressed do not necessarily correspond to those of the Auto-

rité de Contrôle Prudentiel.

REFERENCES
Belkin, B., Forest, L. R. Jr and S. J. Suchower, 1998, “A one-parameter representa-
tion of credit risk and transition matrixes”, CreditMetrics Monitor (3rd Quarter.

Bruneau, C., O. de Bandt and W. El Amri, 2012, “Macroeconomic fluctuations and

corporate financial fragility”, Journal of Financial Stability, 8, pp. 219–35.

Coffinet, J., and S. Lin, 2010, “Stress-testing banks’ profitability”, Banque de France
Working Paper 306.

Coffinet, J., S. Lin and C. Martin, 2009, “Stress-testing French banks’ subcompo-
nents”, Banque de France Working Paper 242.

Dumontaux, N., and D. Médée, 2009, “Prévision de matrixes de transition en fonc-

tion d’un scénario macroéconomique”, Autorité de Contrôle Prudentiel, mimeo.

Gupton, G., C. Finger and M. Bhatia, 1997, “CreditMetrics-Technical Document”,

Morgan Guaranty Trust Co.

Merton, R., 1974, “On the Pricing of Corporate Debt: the Risk Structure of Interest
Rates”, Journal of Finance, 29, 449-470

Standard & Poor’s, 2012, “Annual 2011 Global Corporate Default Study and Rating
Transitions”, Global Fixed Income Research.

Standard & Poor’s, 2008, “Annual 2007 Global Corporate Default Study and Rating
Transitions”, Global Fixed Income Research.

Vasicek, O., 1991, “Limiting Loan Loss Probability Distribution”, KMV Corporation.

125

stress_testing.indd 125 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

1 In the case of France, credit-risk RWAs represents more than 75% of total RWAs.
2 Exposures from structured credit products or from over-the-counter (OTC) derivatives ex-
posures are not covered here.
3 Prudential filters may sometimes dampen the effect of marked-to-market gains or losses.
4 See Dumontaux and Médée, 2009.
5 See Standard and Poor’s (2012) and Standard and Poor’s (2007).
6 For any further information, see [Link]
[Link]
7 The structure of the corporate portfolio of French banks, dominated by international groups,
allows the use of such a reference sample to calibrate their stress-testing framework. It could
therefore be extended to other global banks, once we are ready to assume that all global banks
tap the same markets, in terms of risk characteristics, but differ in terms of portfolio composition.
8 Actually, it is one-third of the deviation between the average default rate and DRcrisis (the
mean of yearly DRs for the years 1991, 2001, 2002, 2009).
9 There is a vast literature on the forecasting properties of the slope of the yield curve.
10 Notice that we consider the recession dates in the US since the database we used is based on
a sample of US and European firms, also assuming that large corporates are global compa-
nies significantly affected by the US business cycle. In the practical implementation of stress
tests, however, we assume that this calibration also holds for the portfolio of corporate assets
held by French banking groups. Such an assumption is imposed by the data constraints (rat-
ings on corporate assets as provided in the Banque de France FIBEN database are available
only with a lag, preventing their use in real-time stress testing).
11 The breakdown by rating is given by the S&P equivalent of internal rating.
12 Regulatory PDs, consistent with Basel II regulations, are estimated as a long-term moving
average of observed default rates; a stressed default rate, which is produced by our model, is
therefore included in the new time window at the end of each year, thus yielding a stressed
regulatory PD.

126

stress_testing.indd 126 20/05/2013 18:45

 9
EU-Wide Stress Test:
The Experience of the EBA

Paolo Bisio, Demelza Jurcevic and Mario Quagliariello

European Banking Authority

In the midst of the financial crisis, the European Banking Authority

(EBA) was established on January 1, 2011, with a broad remit that in-
cluded safeguarding the stability of the EU financial system.1 Accord-
ing to its founding regulation,2 the EBA is required, in cooperation with
the European Systemic Risk Board (ESRB), to initiate and coordinate
EU-wide stress tests to assess the resilience of financial institutions to
adverse market developments. While the stress test is a key compo-
nent of the EBA’s toolkit, it is only one of a range of supervisory tools
used by the EBA for assessing the resilience of individual institutions
as well as the overall resilience of the European banking system.
The first EU-wide stress test of 22 banks was performed by the
Committee of European Banking Supervisors (CEBS), the EBA’s
predecessor, in 2009. The individual results of the stress test were
kept confidential. Only a press release was published with the key
results. Some details of the macroeconomic scenario were pub-
lished (GDP, unemployment and property prices). Furthermore,
the results were published in a very concise manner:

Under such adverse scenario, the potential credit and trading losses
over the years 2009–2010 could amount to almost €400 bn. However,
the financial position and expected results of banks are sufficient to
maintain an adequate level of capital also under such negative cir-
cumstances. Notably, the aggregate Tier 1 ratio for the banks in the
sample would remain above 8% and no bank would see its Tier 1
ratio falling under 6% as a result of the adverse scenario.3

127

stress_testing.indd 127 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

In 2010, CEBS performed another EU-wide stress test among 91

banks; an aggregate report was published as well as individual
bank results. The individual bank results consisted of a single page
with the level of Tier 1 capital, risk-weighted assets (RWAs), losses
and loss rates for both scenarios (baseline and adverse).
Building on the experience of two previous EU-wide stress tests
undertaken by CEBS, the EBA conducted a stress test among 91
banks in 2011. This exercise was undertaken in coordination with
National Supervisory Authorities (NSAs), the ESRB, the European
Central Bank (ECB) and the European Commission (EC).
The exercise was conducted on a bank-by-bank basis, on the high-
est level of consolidation of the banking group. The objective of the
stress test was to assess the resilience of the EU banking system, and
the specific solvency of individual institutions, to hypothetical stress
events under certain severe scenarios decided on by supervisors in
conjunction with the ESRB/ECB. It was a microprudential stress test
focused primarily on assessing banks in a bottom-up manner in a
way that is conservative and consistent across the EU.
In this chapter we describe the EBA’s experience in carrying out
the EU-wide stress test, with a focus on 2011, and provide some
insights into the design, organisation and management of such a
complex exercise. After a first, introductory section, we move to the
summary of the main findings of the stress test and conclude with
our views on the lessons learnt from the 2011 exercise.

KEY CHARACTERISTICS OF THE EBA 2011 STRESS-TEST EXERCISE4

The 2011 EU-wide stress-test exercise was characterised as a con-
strained bottom-up stress test. While banks are required to use
their internal models for estimating possible losses in a stress sce-
nario, the EBA identified common minimum methodological as-
sumptions, imposed a single adverse scenario and carried out an
in-depth quality-assurance process. This is probably the most im-
portant feature of the EBA’s stress test, a precondition for ensuring
comparability of the results across institutions from different coun-
tries and a level playing field in the implementation of supervisory
measures following the stress test.
Therefore, the organisation of an EU-wide stress test requires
complex preparatory work, involving both methodological and

128

stress_testing.indd 128 20/05/2013 18:45

 EU-Wide Stress Test: The Experience of the EBA

procedural aspects. For the 2011 exercise, the preparatory phase

started in September 2010, and 10 months later the bank-by-bank
results were published.
The first step in running an EU-wide stress-test exercise was to
define the sample of banks to be involved and objective of the ex-
ercise. As for the former, the 2011 exercise was carried out among
91 banks at consolidated bank group level. The selection of the
sample was based on representativeness (50% of national banking
sector and 65% of the European banking sector), with the possibil-
ity for NSAs to add additional banks when deemed relevant from
a financial stability perspective. The objective was to assess the
overall resilience of the EU banking system and to identify pos-
sible capital needs at specific institutions.
In fact, the EU stress test has been – like similar exercises carried
out in other countries during the financial crisis – a pass/fail test:
banks that proved to be not able to maintain enough capital in the
adverse scenario have been requested to raise new capital or to ac-
tivate mitigating measures, if available.
The second step in the process is developing the macro scenario.
The baseline scenario was based on the European Commission fore-
casts, while the adverse scenario has been developed by the ECB in
cooperation with the ESRB. The stress-test horizon was set at two
years, as in many regulatory stress tests. The adverse scenario was
composed of three elements: (1) a set of EU-shocks mostly tied to
the persistence of the sovereign debt crisis, (2) a global negative de-
mand shock originating in the US and (3) a USD depreciation vis-à-
vis all currencies. In the adverse scenario the cumulative GDP drop
in the EU was 4%, with a fall of the residential house price by 15%
at EU level. Besides the macro scenario for positions in the bank-
ing book, market-risk shocks were designed for stress testing the
positions in the trading book as well. The market-risk shocks were
aligned with the macro scenario. For example, share prices within
Europe decreased by 15% instantaneously in the adverse scenario.
Furthermore, a specific stress scenario was applied for securitisa-
tion positions in the banking and trading book. To that end, pre-
defined migration matrices for the baseline and adverse scenario
were used, based on historical experience.
The third step in the process is designing the stress-test meth-

129

stress_testing.indd 129 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

odology. As mentioned above, the EBA 2011 stress-test exercise

was carried out by following a constrained bottom-up approach. A
bottom-up stress test is a microprudential test that relies on specific
idiosyncratic data and approaches (for example, banks used their
internal models – where available – to calculate the impact of the
stress test on the balance sheets and the profit-and-loss accounts).
But the exercise was constrained by a series of EBA-prescribed re-
strictive assumptions and a common macro scenario. The design of
the methodology was a challenging task aimed at striking the right
balance between realism and conservatism, respecting the banks’
specificities and ensuring consistency of treatment and compara-
bility of the results. While this is common to different system-wide
exercises, the EU setting required extra effort in order to ensure
consistency across banks belonging to different jurisdictions and
subject to not-fully harmonised rules and supervisory practices.
In general, the methodology was designed in respect of the exist-
ing and forthcoming prudential and accounting rules applicable in
the time horizon of the exercise, notably the International Account-
ing Standards (IAS) and the Capital Requirement Directive (CRD)
and amendments. Two main features were thus the application of
the prudential filters and the application of the amortised cost for
the assets booked in the held-to-maturity portfolio (EBA 2011a).
Another important element for the credibility of the entire ex-
ercise was to define the capital threshold in terms of Core Tier 1
ratio (CT1). Notwithstanding the lack of a consistent definition of
CT1 capital in the pre-Basel III regulation, the EBA decided to de-
duct hybrid capital instruments from the Tier 1 capital, in order to
identify a common definition in line with – even though not equal
to – the Basel concept of common equity Tier 1. Government sup-
port measures were recognised as eligible CT1 capital. By setting a
common definition of CT1 capital, comparability across the banks
involved in the exercise was ensured. The threshold was set at 5%,
which banks were expected to meet after taking into account the
impact of the adverse scenario.
A key assumption designed to ensure consistency in the con-
strained bottom-up approach, and which had a large impact on the
results of the stress test, was considering banks’ balance sheets as
static over the time horizon of the exercise, ie, no management ac-

130

stress_testing.indd 130 20/05/2013 18:45

 EU-Wide Stress Test: The Experience of the EBA

tions were allowed for mitigating the impact of the adverse sce-
nario (static balance-sheet assumption). In practice, this implied a
constant business mix, constant funding mix, a zero growth rate,
the rollover of maturing assets and liabilities and no workout of
defaulted assets. This approach guaranteed comparability of the re-
sults across banks and a level playing field, at the cost of reducing
the realism of the exercise (see also the last section).
The methodology also imposed some constraints on the stress
test’s starting point. In particular, banks were requested to subtract
the PL-impact of the market risk shocks on the trading book portfolio
from the average net trading income of a bank over the last five years.
Since estimating net trading income before stress would have been
difficult, using a historical average represented a simple and conser-
vative benchmark. Furthermore in order to incorporate funding risk,
the EBA prescribed to a certain extent the evolution in the cost of
funding. For example, the interest expenses to be paid for wholesale
funding had to increase according to the evolution of the interest rate
envisaged in the macroeconomic scenario. Another assumption pre-
scribed was that there is a perfect correlation between the evolution
in the sovereign credit spread and that in the bank’s credit spread.
Furthermore, banks have only been allowed to pass up to 50% of the
cost of funding increase to the clients through a potential adjustment
to the credit spread on the maturing loans.
The reliability of the results and the comparability across banks
were then assured through three lines of defence (EBA 2011c): (i)
the banks’ own internal controls, (ii) the consistency checks carried
out by the national supervisors and (iii) a quality assurance process
carried out by the EBA.
This last line of defence was a new aspect of the 2011 stress test,
with an ad hoc task force in charge of performing a thorough peer
review and assessing the proper application of the common meth-
odology. This level of cooperation, with national experts coming
to join EBA staff for a prolonged period in assessing results, was a
step forward in stress-test cooperation in the EU. Banks’ results, in
particular in terms of risk parameters for credit risk, have also been
compared and outliers identified. The computation of benchmarks
has been performed using the granular data collected by the EBA,
including exposures at default (EADs), probabilities of default

131

stress_testing.indd 131 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

(PDs) and loss-given defaults (LGDs) with asset-class and coun-

terparty countries’ breakdowns. However, it is worth underlining
that no asset quality review was undertaken ahead of or during the
stress-test exercise. This is in the remit of national supervisors.
In some cases, banks were required to revise their estimates in
order to bring them into a more consistent range. Indeed, although
the EBA did not request an automatic application of the benchmarks,
banks were asked in some cases to revaluate their initial estima-
tions, providing explanations in case of significant deviation and re-
submitting the results if explanations were considered inadequate.
As the result of peer review, additional guidance was provided to
banks for clarification purposes and in order to obtain consistency
across banks. In this additional guidance (see, EBA, 2011), the EBA
prescribed for example an approach to determining regulatory risk
parameters and thus additional provisions to be held by banks for
sovereign and institutional exposures (see Panel 9.1).

PANEL 9.1 Treatment of sovereign and financial

institutions’ exposures in the banking book
Due to discrepancies between the additional provisions for sovereign
and financial institutions across banks, the EBA prescribed the addition-
al provisions to be held by banks by setting the additional provisions
equal to the increase in expected loss. The EBA developed an approach
based on probabilities of default attributed by rating agencies and as-
suming downgrades depending on the conditions of the specific coun-
tries. The first step was to identify the degree of the downgrade (number
of notches), which was linked to the initial rating. For exposures with
AAA ratings, no downgrade was needed. For AA exposures a two-notch
downgraded needed to be applied. At last, for exposures with BBB+
rating, a four-notch downgrade was requested (floor at CCC). After the
application of the downgrade to the initial rating, this rating after stress
needed to be mapped to a PD. For simplicity and consistency purposes,
the EBA used the two years’ cumulative default rates observed by the
three well-known rating agencies Fitch, Moody’s and S&P (see Table
9.1). The LGD to be used for the calculation of the impairments was
set by the EBA at 40%. The increased provisions amount is equal to
the multiplication of the PD after the downgrade and the LGD of 40%.

132

stress_testing.indd 132 20/05/2013 18:45

 EU-Wide Stress Test: The Experience of the EBA

Further benchmarks were also identified for the computation of the

increase in the cost of funding, the impact caused by the transition
to the Basel 2.5 framework for market risk capital requirements.

RESULTS AND DISCLOSURE

One interesting point to note is that, from the announcement of the
start of the stress test, banks raced to strengthen their capital base
in order to ensure they were deemed strong in anticipation of it. To
this end, while the EBA, together with NSAs, ECB, ESRB and EC,
were preparing for the launch of the EBA 2011 stress-test exercise
in March 2011, some banks that were selected to participate in the
exercise were already increasing their capital base. In total, the capi-
tal base of the banks participating in the exercise increased by €50
billion between January and April 2011.
The average CT1 ratio of the banks decreased from 8.9% to 7.7%
after two years of stress. In total, eight banks had a CT1 ratio below
5%, leading to a total capital deficit of €2.5 billion. Table 9.1, pub-
lished by the EBA (2011c) provides an overview of the results of the
EBA 2011 stress test exercise.
In total, eight banks had a CT1 ratio under the adverse scenario
below the set 5% benchmark.
Following the publication of the 2011 EU-wide stress test results
in July 2011, the EBA issued a recommendation (EBA 2011e) to na-
tional supervisory authorities (NSAs) to ensure that appropriate
mitigating actions were put in place with respect to (i) banks with
a CT1 ratio below 5% in the adverse scenario and (ii) banks with
a CT1 ratio close to 5% in the adverse scenario and with sizeable
exposures to sovereigns under stress.
Along with the bank-by-bank stress-test results, the EBA also pub-
lished detailed information about banks’ actual exposures. This was
part of an unprecedented effort for increasing transparency and reduc-
ing uncertainty over EU banks’ financial conditions. In total 3,200 data
points were published per bank resulting in an unprecedented level of
harmonised disclosure of the results and banks’ exposure data. In fact,
disclosure has represented a complement to the analysis conducted
by the EBA itself, allowing market participants to make their own as-
sumptions and possibly testing further scenarios beyond the EBA base-
line and adverse scenarios for the banking book and the trading book.

133

stress_testing.indd 133 20/05/2013 18:45

stress_testing.indd 134

134

Stress Testing: Approaches, Methods and Applications

Table 9.1 Results of the 2011 EU-wide stress test – country data
Adverse scenario

2010 2012 < 2% < 3% < 4% < 5% < 6% < 7% < 8% < 9% < 10% > 10%
AT 8.2% 7.6% 0 0 0 1 0 0 1 1 0 0
BE 11.4% 10.2% 0 0 0 0 0 0 0 0 0 2
CY 7.7% 5.7% 0 0 0 0 1 1 0 0 0 0
DE 9.4% 6.8% 0 0 0 0 2 4 2 1 1 2
DK 9.8% 11.9% 0 0 0 0 0 0 0 0 1 3
ES 7.4% 7.3% 0 0 3 2 7 5 1 3 2 2
FI 12.2% 11.6% 0 0 0 0 0 0 0 0 0 1
FR 8.4% 7.5% 0 0 0 0 0 2 1 1 0 0
GB 10.1% 7.6% 0 0 0 0 0 1 2 1 0 0
GR 10.2% 6.1% 1 0 0 1 2 0 2 0 0 0
HU 12.3% 13.6% 0 0 0 0 0 0 0 0 0 1
IE 6.2% 9.8% 0 0 0 0 0 0 1 0 0 2
IT 7.4% 7.3% 0 0 0 0 1 2 1 1 0 0
LU 12.0% 13.3% 0 0 0 0 0 0 0 0 0 1
MT 10.5% 10.4% 0 0 0 0 0 0 0 0 0 1
NL 10.6% 9.4% 0 0 0 0 0 1 0 1 1 1
NO 8.3% 9.0% 0 0 0 0 0 0 0 1 0 0
PL 11.8% 12.2% 0 0 0 0 0 0 0 0 0 1
PT 7.1% 5.7% 0 0 0 0 2 2 0 0 0 0
SE 9.0% 9.5% 0 0 0 0 0 0 0 1 2 1
SI 5.7% 6.0% 0 0 0 0 1 0 0 1 0 0

Total 8.9% 7.7% 1 0 3 4 16 18 11 12 7 18

Source: EBA (2011c)

20/05/2013 18:45
 EU-Wide Stress Test: The Experience of the EBA

LESSON LEARNED FROM THE EBA 2011 STRESS TEST

The EBA 2011 stress-test exercise has represented an important step
towards an EU-wide forward-looking assessment of risks and vul-
nerabilities in the European banking sector. The development of a
common methodology and its early publication was very well com-
mented on by the different stakeholders. This ensured transparency
for market participants as regards the assumptions and the mechanics
of the exercise as well as consistency across banks in conducting the
stress-test exercise. At the same time, the commitment for future ex-
ercises is to further anticipate the publication of the methodology, in
order to enhance the interaction with the industry ahead of the launch
of the exercise and, ideally, testing the templates for data collection.
Furthermore, the decision to set the capital threshold at CT1, and
communicating the capital threshold (5% CT1) in advance of publish-
ing the stress-test results, guaranteed a credible and disclosed capital
benchmark. Early disclosure also encouraged banks to strengthen
their capital positions ahead of the test by raising €50 billion be-
tween January and April 2011.
The quality assurance and peer review were also beneficial and
added value in terms of ensuring adequacy in the interpretation of
the methodology and the mapping of the macroeconomic scenario
into micro-parameters. Looking ahead, further work on the early
definitions of benchmarks to be used in this process is needed and
would lead to even more robust results.
Last, but not least, the unprecedented level of harmonised disclosure
of the results and banks’ exposure data disclosure was perceived a sig-
nificant step in the direction of greater transparency. In particular, dis-
closure on banks’ actual exposures and starting points has added value
and contributed to reducing uncertainty about the state of health of the
EU banking sector. It also allowed analysts to perform their own assess-
ment and complement the EBA’s scenarios with further sensitivities.
Along with significant pros, the EBA stress test has been also chal-
lenged. Some criticisms relate to the process and the ongoing fine tun-
ing of templates and benchmarks during the exercise itself as more
information became available. The process for future stress tests will
attempt to overcome this by road-testing templates and gathering in-
formation on benchmarks ex ante. However, particular criticism was in
terms of the outcome of the exercise and of some of its assumptions.

135

stress_testing.indd 135 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Market concerns about the EU banking system and the financial sta-
bility implications thereof were deemed not to be directly addressed
by the stress-test exercise. A first problem was probably linked to the
design of the macroeconomic scenario. Indeed, the adverse scenario
agreed when the exercise was launched was taken over by later events,
with a further deterioration of the economic environment (vis-à-vis an
expected modest recovery) and the eruption of the sovereign crisis. In
addition, funding problems and, more importantly, liquidity squeeze
pointed to the lack of a review of banks’ liquidity positions. Finally,
increasing pressure on asset quality, especially in some jurisdictions
and for some asset classes, reduced the credibility of the results. In
particular the capital shortfall has been perceived as far too optimistic,
also in relation to the lack of severe sovereign stress for exposures in
the banking book. This criticism was, however, partly tackled by the
EBA by running a capital exercise in Q4 2011 (see Panel 9.2).
It should be noted, however, that this criticism is only partially
grounded. In some cases, observers have not been able to read more
carefully the published results and correctly interpret the data dis-
closed by the EBA. It has been partly a communication problem linked
to the misperception of what a stress test can (and cannot) deliver.
It is beyond the purpose of this chapter to analyse this issue in
depth, but it is interesting to show a different way of presenting the
stress-test results in terms of incremental impact instead of overall
level of CT1 ratio.
Figures 9.1 to 9.3, based on data published by the EBA, show for
instance the impact of the stress scenario on banks’ loss rates for
the retail and corporate exposures and represent an intuitive way
for identifying banks more affected, also in relation to the country-
specific change in GDP.
In Table 9.2, also based on public data for a sub-sample of banks,
we ranked banks by impact of the adverse scenario on some P&L
flows, capital position and RWAs.
Table 9.2 tells us quite a different story with some banks that dealt
with significant problems that the stress test had actually identified
as more sensitive (or less resilient) to the EBA’s adverse scenario. In
our view, the message is clear: the mechanics of the stress test have
some shortcomings that cannot be easily avoided, but a more in-
depth and imaginative analysis of the results would have provided
interesting insights.

136

stress_testing.indd 136 20/05/2013 18:45

 EU-Wide Stress Test: The Experience of the EBA

Figure 9.1 Multiple Loss rates, retail scenario/Dec 2010

6
5
4
3
2
1
0
AT001
AT002
BE004
BE005
CY006
CY007
DE017
DE018
DK008
DK009
ES059
ES060
FI012
FR013
FR014
GB088
GB089
GR030
GR031
HU036
IE037
IE038
IT040
IT041
LU045
MT046
NL047
NL048
NO051
PL052
PT053
PT054
SE084
SE085
SI057
SI058
Figure 9.2 Multiple loss rates, corporate scenario/Dec 2010

7
6
5
4
3
2
1
0
Adverse 2011 Adverse 2012 Baseline 2011 Baseline 2012

Figure 9.3 Percentage change in GDP in the scenarios (home country banks)

5
4
3
2
1
0
-1
-2 AT AT BE BE CY CY DE DE DK DK ES ES FI FR FR GB GB GR GR HU IE IE IT IT LU MT NL NL NO PL PT PT SE SE SI SI
-3
-4
-5
Adverse 2011 Adverse 2012 Baseline 2011 Baseline 2012

A last point is related to the preparation of future EBA stress tests. An

important aspect to manage is the expectations from the market. A stress
test is not a forecast of the future: it is just the assessment of what would
happen should an event, subject to a probability distribution, materialise.
A stress scenario needs to be severe but plausible. The sovereign crisis
has broadened the perspective with respect to plausibility and sever-
ity. The objective of an EU-wide stress-test exercise needs to be clarified
in advance. Moreover, it needs to be communicated what stress testing
does and does not do. So for the next EU-wide stress-test exercise, which
was planned for 2013, the lesson learned from the EBA 2011 stress-test
exercise would be incorporated as much as possible.

137

stress_testing.indd 137 20/05/2013 18:45

stress_testing.indd 138

138

Stress Testing: Approaches, Methods and Applications

Table 9.2 Ranking of banks according to the impact of the 2011 EU-wide stress test
Bank code Bank name Operative Trading Cost Impairments RWA CT1 CT1R
Income book funding Adv-Bas
AT001 Erste Bank Group (EBG) 4 3 5 3 5 3 3

AT002 Raiffeisen Bank International (RBI) 5 4 3 2 4 4 2

BE004 Dexia 7 4 6 6 6 5 5

BE005 KBC Bank 7 7 6 3 7 5 5

CY006 Marfin Popular Bank Public Co Ltd 2 7 2 7 3 7 6

CY007 Bank of Cyprus Public Co Ltd 6 5 3 5 2 6 6

DE017 Deutsche Bank AG 3 6 7 2 7 2 4

DE018 Commerzbank AG 5 6 3 4 7 5 5

DK008 Danske Bank 1 6 4 1 1 2 1

DK009 Jyske Bank 2 3 7 1 4 1 7

ES059 Banco Santander S.A. 1 3 1 3 5 1 2

ES060 Banco Bilbao Vizcaya Argentaria S.A. (BBVA) 2 3 6 3 3 1 4

20/05/2013 18:45
stress_testing.indd 139

Table 9.2 (continued)

FI012 Op-Pohjola Group 5 5 7 6 5 4 3

FR013 Bnp Paribas 7 2 2 4 6 4 4

FR014 Credit Agricole 6 5 4 4 1 4 2

GB088 Royal Bank of Scotland Group Plc 7 3 7 1 5 6 6

GB089 Hsbc Holdings Plc 6 2 7 2 7 3 5

GR030 Efg Eurobank Ergasias S.A. 4 1 2 5 3 7 7

GR031 National Bank of Greece 1 7 2 7 4 7 7

HU036 Otp Bank Nyrt. 2 4 5 3 3 1 7

EU-Wide Stress Test: The Experience of the EBA

IE037 Allied Irish Banks Plc 7 7 1 1 6 7 6

IE038 Bank of Ireland 7 5 2 2 1 7 7

IT040 Intesa Sanpaolo S.P.A 2 1 6 4 4 4 3

IT041 Unicredit S.P.A 3 4 6 4 6 5 2

LU045 Banque et Caisse d’Epargne de l’Etat 4 6 5 7 1 1 1

MT046 Bank of Valletta (Bov) 4 7 7 7 1 5 3

139
20/05/2013 18:45
stress_testing.indd 140

140

Stress Testing: Approaches, Methods and Applications

Table 9.2 (continued)
NL047 Ing Bank NV 5 4 1 5 7 2 7

NL048 Rabobank Nederland 3 6 3 5 7 2 6

NO051 Dnb Nor Bank ASA 3 1 5 7 1 2 1

PL052 Powszechna Kasa Oszczednosci Bank Polski 1 1 4 6 3 3 1

S.A. (Pko Bank Polski)

PT053 Caixa Geral de Depósitos, SA 6 7 5 6 4 6 5

PT054 Banco Comercial Português, SA (BCP Or 6 1 3 5 6 7 4

Millennium Bcp)

SE084 Nordea Bank AB (Publ) 5 2 1 6 1 3 1

SE085 Skandinaviska Enskilda Banken AB (Publ) (Seb) 3 2 7 7 5 3 3

SI057 Nova Ljubljanska Banka D.D. (NLB D.D.) 1 2 1 1 2 6 2

SI058 Nova Kreditna Banka Maribor D.d. (NKBM D.D.) 4 5 4 2 2 6 4

20/05/2013 18:45
 EU-Wide Stress Test: The Experience of the EBA

Panel 9.2 EBA 2011 capital exercise

In December 2011, the EBA launched a capital exercise in order to re-
assure market participants on the resilience of EU banks to a sovereign
shock. This was part of a more general plan to break the link between
banks and sovereigns via (i) direct capital injections into banks from
EU bodies, (ii) effective EU-wide bank term debt guarantees, and (iii)
higher capital buffers across the entire banking system.
In this context, in December 2011 (EBA 2011d), the EBA issued
a recommendation that banks raise their CT1 capital to 9% after ac-
counting for an additional buffer against stressed sovereign risk hold-
ings. The capital exercise was not a stress-test exercise, since no stress
scenario was prescribed. In the capital exercise banks were required to
evaluate, in a prudent manner, their sovereign exposures.
The recommendation requires banks to strengthen their capital
positions by building up an exceptional and temporary capital buffer
against sovereign debt exposures to reflect market prices as at the end
of September 2011. In addition, banks were required to establish an
exceptional and temporary buffer such that the Core Tier 1 capital ratio
would reach a level of 9% by the end of June 2012.
The buffer requirement was not designed to cover losses in sover-
eigns, but to provide a reassurance to markets about the banks’ ability
to withstand a range of shocks and maintain adequate capital levels.
The sovereign capital buffer has been clearly established as a one-off
measure and not as a permanent requirement.

CONCLUSION
This chapter described the EBA experience in running the EU-wide
stress test, focusing on the 2011 exercise. In particular, we tried to
highlight what the challenges were – in terms of methodology, gov-
ernance of the process and communication – of a stress test cover-
ing an ample sample of banks from different jurisdictions which are
subject to banking regulations that are not fully harmonised.
There are various peculiarities in handling this kind of exercise,
but we would like to conclude with three main messages.
The first is that a choice needs to be made between comparabil-
ity and realism in designing the stress test. The EBA privileged the
former and opted for a static balance-sheet assumption and a con-
strained bottom-up setting. This may be criticised from a theoreti-
cal perspective, but it is the only manageable approach in practice
and contributes to ensure a level playing field.
Second, the results of the stress test cannot be interpreted in a

141

stress_testing.indd 141 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

mechanistic way, but should be read carefully and subject to further

sensitivities. In that respect, the disclosure exercise that accompa-
nied the release of the stress-test results greatly contributed to re-
duce uncertainty on banks’ exposures to different sources of risk.
Finally, communication is key and it is important to increase
awareness of what a stress test can and cannot deliver. In that re-
spect, managing expectations is part of the preparation for a stress
test, particularly if individual results are disclosed. This avoids the
spreading of a false sense of security as well as of complacency.

The opinions expressed in this chapter are those of the authors and do not in-
volve the EBA and its Members. Useful comments and suggestions from Piers
Haben are gratefully acknowledged.

REFERENCES
EBA, 2010, “Aggregate outcome of the 2010 EU-wide stress test exercise coordi-
nated by CEBS in cooperation with the ECB”, July.

EBA, 2011a, “EU-wide stress test Methodological note”, Version 1.1, March.

EBA, 2011b, “EU-wide stress test: Methodological note – Additional guidance”,

June 9.

EBA, 2011c, “EU-wide stress test aggregate report”, July 15.

EBA, 2011d, “Capital buffers for addressing market concerns over sovereign expo-
sures – Methodological Note”, December 8.

EBA, 2011e, “Recommendation on the creation and supervisory oversight of tempo-

rary capital buffers to restore market confidence” (EBA/REC/2011/1), December 8.

EBA, 2012, “Report on the fulfilment of the EBA Recommendation following the
2011 EU-wide stress test”, April 30.

1 High Level Group on Financial Supervision in the EU published a report in February 2009,
the so-called De Larosière Report. The aim of the report was to lay out a framework to take
the EU further in its process of integration, which includes (i) a new regulatory agenda (Ba-
sel III), (ii) stronger coordinated supervision (eg, EBA) and (iii) effective crisis-management
procedures.
2 Regulation (EU) No. 1093/2010 of the European Parliament and of the Council of November
24, 2010.
3 CEBS’s press release on the results of the EU-wide stress testing exercise (2009).
4 This section analyses the results presented in EBA (2011a and 2011b).

142

stress_testing.indd 142 20/05/2013 18:45

 10
Stress Testing Across
International Exposures and Activities

Robert Scavotto, Robert H. Skinkle

Office of the Comptroller of the Currency

Stress testing foreign exposures of internationally active financial

institutions presents an array of challenges from developing data-
sets to understanding and factoring in qualitative factors that may
drive loss rates. These factors can be a function of the government
or more simply related to developments in lending practices within
a given market. Our work to stress test consumer loan portfolios
across a group of countries led us to conclude that a stress-test ap-
proach for consumer portfolios in Asia is best pursued on an indi-
vidual-country basis with a stratification of the consumer portfolio
by secured and unsecured credits. This chapter describes our work
in this area and uses the Korean consumer market as a case study.

CONDUCTING STRESS TESTS OF INTERNATIONAL ACTIVITIES

IN AN IDIOSYNCRATIC WORLD
Financial institutions with international business activities, includ-
ing operations in foreign markets, have a number of questions to
consider in designing stress tests that cover these activities, expo-
sures and risks. The first and most important is deciding on the pur-
pose of the stress test, since international risks can overlie a number
of different risk types, such as credit, market and operational. In-
ternational risks can be viewed as an overlay since there are deter-
minants of loss rates to consider that are additional to those in a
more standard functional form of a model estimated using data and
exposures strictly within a single country. For example, if corporate

143

stress_testing.indd 143 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

default rates were simply determined by the economic growth rate

within a domestic economy, this function would have to be altered
if the corporate obligor was also exposed to risks from government
intervention in the provision of foreign exchange.
In foreign markets, banks and borrowers are subject to country
risk, which is the risk that economic, social and political condi-
tions and events in a foreign country will affect an institution (OCC
2008). Further, these conditions can manifest themselves in the
form of a crisis, such as a sovereign-default, exchange-rate or bank-
ing-system crisis. These crises can happen in any combination and
in any order and need to be considered in the design of an interna-
tional stress test. Emerging-market countries have been more prone
to these crises, but, as evidenced by Iceland (banking-system and
exchange-rate crises), Ireland (banking-system crisis) and Greece
(sovereign default, as evidenced by a forced restructuring), devel-
oped countries are also vulnerable to these crises. Thus, depending
on the purpose and scope of the stress test, factoring the potential
for one or more crises into the downside scenario is an important
consideration when stress testing international portfolios. Being
subject to these types of risks also requires that the stress-testing
framework be forward-looking and flexible so that stress tests can
be executed should sudden changes in economic, financial, political
or social conditions take place.
Given the wide range of risk types and potential crises that a
bank is exposed to in foreign markets, the scope of the stress test
needs to capture the majority of a bank’s foreign exposures, activi-
ties and risks. The scope of the stress test can also pre-identify trans-
mission channels for some of the losses, but, as managers and mod-
ellers think through the scenario under consideration, additional
risk channels may be identified and incorporated into the stress
test. For example, a recession scenario may open up a liquidity risk
channel if the country is highly indebted, has a low level of interna-
tional reserves and is reliant on foreign capital inflows to finance its
debt. These channels may surface as the scenario, and incorporated
shocks in the stress test will be driven by internal or external fac-
tors. For example, does the scenario incorporate an internal shock,
such as a change in government policies, eg, nationalisation of an
industry that triggers a sharp decrease in investment activity, or an

144

stress_testing.indd 144 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

external shock where the scenario envisions a sovereign default by

a major trading partner? Given the complexities that international
activities introduce, there is a much wider range of factors and prin-
ciples that firms need to account for in the design of the stress test.
In particular, firms should check to see if Principles 7–15 of the Ba-
sel Committee on Banking Supervision’s (BCBS) May 2009 report1
have been implemented and are guiding the stress test.
One way to consider and track the risks that the firm could be
exposed to is through a table, such as Table 10.1.

Table 10.1 Potential risks to factor into stress tests for international
activities

Country Transfer Nationali- Sovereign Operational … Bank …

Risk2 sation default risk from holiday
civil strife
Argentina X X
… X X X X
Zambia X

A traditional recession-based scenario should be considered a mini-

mum stress test for most foreign-country exposures and activities.
The level of complexity of the stress test would vary substantial-
ly for a firm with limited exposures in a few countries compared
with a firm operating in more than 100 countries using complex
products and risk-mitigation strategies. A firm with more complex
operations may at the outset of developing its stress-test platform
identify a set of countries of highest priority, eg, through the use of
a simple calculation of risk-adjusted exposures, known concentra-
tions, or a contagion scenario set of identified countries. The firm
may also be able to group countries with similar economic struc-
tures (commodity-based, fixed exchange rate) or risks (highly in-
debted with both current account and fiscal deficits) until the full
stress-testing framework can be built. Proceeding in this manner
also may help overcome some of the data challenges that the mod-
ellers may face, such as generating or obtaining a consistent set of
data elements across all of the countries the firm operates in.

145

stress_testing.indd 145 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Some of the risks are idiosyncratic to individual countries, such

as operational risks to the firm from civil strife or imposition of
regulatory controls, which imposes challenges to doing a cross-
country systemic stress test. However, in the knowledge that these
factors are present, the model can control for those elements using
various econometric techniques, such as fixed-effects regressions
that introduce country dummy variables. These factors can then be
tested separately through sensitivity or country-specific stress tests
to determine whether material risks were masked by the modelling
technique employed. This is likely to involve considerable expert
judgement that senior management should be made aware of dur-
ing the design, testing and reporting on the stress tests.
The final critical elements of the stress test will be the measurement(s)
chosen for its objective – capital, earnings, informing bank strategies –
and determining a set of potential action points for monitoring and ex-
ecuting risk-mitigation strategies should the scenario materialise. The
governance of the stress-testing framework should provide direction
for determining these elements; but, at a minimum, the results should
be clear, actionable and well supported, and inform decision-making
vertically within the firm as well as horizontally across all business
lines and functions that manage international activities or would be
exposed to those risks.

DATA CHALLENGES IN ESTIMATING INTERNATIONAL STRESS TESTS

The measurement(s) chosen for assessing the international stress
test (in accordance with its purpose) will depend in part on the data
available for the model estimation – a significant challenge for some
countries and variables. Cross-country data has been facilitated by
the Basel Committee’s efforts to standardise measures for capital;
the international accounting standards for capturing key financial
indicators for corporations and financial institutions; and the IMF’s
standardisation of basic macroeconomic data. However, the avail-
ability of comparable, cross-country data can diminish significantly
depending on the risk type(s) assessed and model strategy chosen,
particularly at more granular levels, and may require the financial
institution to start building a dataset. Significant data issues will
need to be conveyed to senior management, risk managers and
business lines, since the interpretation of stress-test results needs

146

stress_testing.indd 146 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

to be pragmatic and viewed in light of data limitations relative to

home country stress tests.
There are some common public sources for data, including the
IMF’s international financial statistics and trade data; the World
Bank’s external debt and governance indicators; and central bank,
supervisory and ministry-of-finance websites. Private-sector firms
may collect additional elements. The multilateral organisations have
worked to make the measurement of the data consistent across coun-
tries; however, this may not be the case when collecting individual
country data, so it is critical to understand how variables are defined
and measured and not assume that, because the name of the variable
is similar to that used in other countries, they are indeed identical or
consistently measured. Delinquency rates and non-performing loan
ratios are good examples of variables that may be defined differently
across countries. In Russia, for example, the non-performing assets
ratio includes just the past due portion of the loan. Such definitional
variations can distort comparisons of portfolio credit quality across
countries. This is true whether the data is obtained directly from in-
country sources or from vendors that have collected it and packaged
it for end-users. The data limitations can be particularly acute and
constrain the ability to implement consistent, granular stress-test
analyses across countries. However, using proxies, stress tests on a
country-by-country basis may be feasible.
There may be a more basic problem when what would normally
be considered a standard variable is simply not collected by one or
more countries. For example, the unemployment rate is a signifi-
cant explanatory variable for retail loan performance across coun-
tries. However, in India, the government has not collected monthly
or quarterly unemployment-rate data across the country, since the
logistical difficulties are enormous and the cost prohibitive, or in
some countries the unofficial labour market is so large as to ren-
der the official unemployment numbers meaningless. Therefore,
when a model is being developed, a starting point is to inventory
the available data series for each country the bank is exposed to and
then consider possible proxies for those data series that cannot be
obtained. Some proxies can be developed from what theoretically
should be highly correlated variables, such as the level of industrial
production and the level of employment, or there is the possibility

147

stress_testing.indd 147 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

of obtaining and using proxies from structurally similar economies

and banking systems.
There are also some basic data considerations and challenges for
banks’ internal datasets, including identifying the country of risk,
the location of collateral and the reliability of guarantees.

STRESS TESTING INTERNATIONAL RETAIL PORTFOLIOS

A good stress-testing framework employs multiple conceptually
sound stress-testing approaches. This is particularly true with inter-
national activities, as data limitations may force some programmes
to be basic sensitivity analyses while some more advanced banks
could do full-scope, enterprise-wide stress tests with multiple sce-
narios. In this section we will focus on stress testing international
retail portfolios to estimate expected losses on bank exposures to
foreign consumers. Retail models are more challenging as, to our
knowledge, there are no existing comprehensive, comparable cross-
country datasets on which to base models. Further, significant varia-
tions in laws and culture can affect retail default rates and, due to dif-
ferences in creditor rights, the level of loss-given default (LGD). This
can cause the relationship of economic variables to loss rates to vary
from country to country and from product to product. However, on
a country-by-country basis for most developed countries and a num-
ber of emerging-market countries, data can be gathered on retail loan
performance and basic macroeconomic and governance variables.
The loan performance data can be obtained from a bank’s own
loss history or from the central bank/supervisors, which report
aggregated data on loan-loss rates, even possibly segmented by
consumer product type. Portfolio segmentation may be neces-
sary when building models for a bank operating across countries
or with multiple loan types. For example, within the portfolio of
countries that we have modelled, secured portfolios did not yield
significant coefficients or meaningful regression equations, while
there was success with unsecured lending. An important lesson
from our modelling experience was that international retail models
are not necessarily the same as modelling home country portfolios
or even across countries within a region, eg, Asia, and that basic
portfolio diagnostics and more in-depth examination of the coun-
try’s consumer market was required to develop a sound model.

148

stress_testing.indd 148 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

Segmenting the portfolios within and across countries has the

benefit of isolating potential correlations and eliminating noise and
offsetting impacts. With so many moving parts in a multiple-coun-
try, multiple-product-type model, results can be watered down or
rendered insignificant. Indeed, in our modelling work, the assess-
ment of an aggregated portfolio of secured and unsecured products
for individual countries did not yield robust results. For example, a
retail portfolio in the Eurozone could include Greece and Germany
with both credit cards and first-lien home mortgages. If this were
the case, then it would seem that the most logical approach for this
retail portfolio would be a segmented portfolio approach, ie, one
that isolates, to the extent possible, the retail portfolio’s various
moving parts on both the left and right sides of the equation:

Country X Losses (Mortgages) = f (home prices, interest rates,

unemployment)

Country X Losses (Credit Cards) = f (unemployment, inflation,

interest rates, credit growth)

It is possible that data diagnostics and evaluation of portfolio char-

acteristics of the retail portfolios reveal that some countries could
be grouped into a homogeneous portfolio, barring the presence of
idiosyncratic risks as discussed in earlier.

STRUCTURAL AND CYCLICAL ISSUES

For each county, a number of structural and cyclical considerations had
to be considered and factored into the models. Structural considerations
include the structure of the capital markets and legal framework for the
institution’s products and services as well as creditor rights, which af-
fect LGD estimates. For emerging markets, there are structural changes
occurring to both the economy and banking system (such as financial
deepening) that need to be taken into consideration. These structural
changes can cause correlations and variable relationships between de-
pendent variables and prospective independent variables to change or
break down. The chart in Figure 10.1 showing Brazil’s unemployment
rate juxtaposed with the percentage of banking system loans to indi-
viduals that are past due by 90 days or more highlights this point.

149

stress_testing.indd 149 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Figure 10.1 Brazil banking system: loans to Individuals 90+ days past
due versus unemployment rate

14
13
12
11
10
9
8
7
6
5
4
01/01/

01/01/

01/01/

01/01/

01/01/

01/01/
2001

2003

2005

2007

2009

2011
90+ days PD % Unemployment Rate %

Source: Banco Central do Brasil and Instituto Brasileiro de Geografia e Estatística

As Figure 10.1 indicates, in the years from 2002 to 2012, Brazil has
experienced a long-run decline in its unemployment rate, reflect-
ing, in part, a period of structural change and rapid development
for the economy. As a result, the predictive power of the unemploy-
ment rate for the consumer loan portfolio performance had deterio-
rated. In fact, the correlation between changes in past due loans and
the unemployment rate became negative in 2011, a counterintuitive
result. Over this same period, bank lending expanded rapidly with
institutions targeting new (and untested) consumers with an array
of products from credit cards to payday lending (an example of fi-
nancial deepening and the evolution of the banking system).
The introduction of new products and other financial innovations
is an important consideration when designing stress tests. In such
countries as Hungary and Poland during the early 2000s, banks start-
ed marketing Swiss-franc-linked mortgage products, which offered
interest rates significantly lower than comparable local currency
loans. By the summer of 2008, according to the IMF, the foreign-cur-
rency share of new house-purchase loans in Hungary hit 70%. The
global financial crisis of 2008 and 2009 resulted in a surging value
for the Swiss franc, precipitating a repayment crisis for Hungarian
homeowners. This underscores the importance of understanding the
portfolio characteristics in the design of stress tests.

150

stress_testing.indd 150 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

Not only is segmenting by product type important, but segmenting

by classes of borrowers, as well as by loan vintage, could yield meaning-
ful differences in the results of the stress tests. For example, countries
with a new generation or economic class of borrowers, loan-loss experi-
ence may be difficult to predict, as the loss experience is likely to dif-
fer from the loan-loss history for borrowers with longer credit histories.
Many countries have established credit bureaus and that can be useful
for segmenting different borrower classes (if those registries are made
publicly available). The existence and quality of local credit bureaus is a
function of the length of their existence, the type of information they col-
lected and the quality controls over the registries. In Australia, the credit
bureau contains only negative information. However, “comprehensive
credit reporting” will begin in 2014. In 2003, Hong Kong’s credit bureau,
which was established in the early 1980s, expanded the type of informa-
tion provided, in the wake of changes to Hong Kong’s privacy laws.
Modellers need to determine whether there are significant variations
in credit registries across countries (eg, some reported only negative in-
formation) that can also contribute to bank underwriting decisions and
ultimately have a bearing on cross-country loss rates.
Cyclical considerations entail not only debt levels and servicing,
but also credit growth relative to the economic cycle, particularly
if the real rate of growth for the banking system has been rapid.
If the rate of real credit growth is rapid, eg, greater than 10%, then
the bank has to consider not only the condition of its own portfolio
but the potential condition of the portfolios of its competitors as
well, since the possibility of widespread degradation of underwrit-
ing standards could lead to regulatory actions if not a full-blown
banking crisis. If a crisis is a possibility, then the scenario under
consideration may need to be re-scoped or an additional, more se-
vere downside scenario may need to be executed.

FACTORING IN FOREIGN REGULATORY ACTIONS

Regulatory actions complicate the predictive power of past portfolio
performance under forward-looking stress tests. Supervisory authori-
ties have implemented a range of macroprudential regulations aimed
at controlling loan growth and striving to prevent asset bubbles. Such
measures include lowering the maximum loan-to-value (LTV) ratios
for mortgage loans as well as implementing requirements on debt ser-

151

stress_testing.indd 151 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

vice capabilities for consumer loans. Such measures have been widely
adopted across Asian consumer lending markets, from Korea to Sin-
gapore, though there are significant variations in the stringency of the
measures. While largely a positive for loan performance, these mea-
sures will potentially affect the performance of loan-loss models.
If the assessment is for possible regulatory actions, then more
typical administrative measures such as increasing LTV require-
ments and lowering debt-to-income measures could affect the reve-
nue stream of the bank, while increased provisioning requirements
could affect net income and the market’s perception of the bank. As
such, not only should these actions be factored into scenarios but
models must also account for these actions in bank performance
and the wider affect on the macroeconomy.
In addition, regulatory actions can have negative impacts on mar-
kets through moral hazard and forbearance. Taiwan’s experience in
2004–6 is a case in point. During this period, banks aggressively ex-
panded unsecured consumer loans. As charge-off rates spiked, the
government responded with restructuring measures offering easier
repayment terms that covered 30% of outstanding credit-card bal-
ances, according to IMF estimates (Laeven and Laryea 2009). These
restructured loans were for the most part reclassified as performing.
Such regulatory moves can have a large effect on model performance,
which needs to be accounted for in the evaluation of the results and
possibly require modification to and re-estimation of the model.
Quantitative proxies can be developed for these seemingly qualita-
tive factors. For example, the potential for a wider range of regulatory
actions should be considered in the scenario if the risk of a political back-
lash from rising defaults is a possibility. In addition, potential changes in
regulatory requirements depend on the quality of the regulator, the ma-
turity of the banking sector and validity of the legal system. Quantitative
measures for these elements are available and could be factored into the
modelling process. For example, the World Bank publishes numerical
estimates for six governance indicators, which could then be incorpo-
rated into the model directly or transformed into an index.3

ASIAN CONSUMER PORTFOLIOS: STRATIFICATION REQUIRED

Stress testing retail portfolios may require stratification between se-
cured and unsecured lending, due to wide variances in loss rates

152

stress_testing.indd 152 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

between these two broad types of lending. An analysis of loss rates

across Asian countries underscores the need to differentiate be-
tween secured and unsecured retail portfolios. As highlighted in
Figure 10.2, mortgages and car loans involve collateral and tend to
pose lower risk than unsecured lines of credit, such as credit cards.
In Singapore for example, Moody’s (2009) has an expected loss for
housing loans of 0.05% under their base case, compared with 5.3%
for other individual loans. The expected loss under the stress case is
2.5% for housing and 17.0% for other loans to individuals.

Figure 10.2 A simple stratification of a retail portfolio

Portfolio

Secured Unsecured

Lines of
Mortgage Auto
Credit

In Asia, historical loss rates on mortgages are extremely low primar-

ily due to conservative underwriting standards that are typical in
Asia (for example, low LTV ratios, and lenders typically have full
recourse). Even in extreme downside scenarios – such as in the Hong
Kong home price collapse during the late 1990s – mortgage loss rates
typically do not spike in this region. Stress tests of the mortgage port-
folios in Asia confirmed this, as the results did not show material loss
rates under a basic recession downside scenario. This does not mean

153

stress_testing.indd 153 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

that periodic stress tests for mortgage portfolios are not required or
stressed using alternative scenarios – such as a disaster-based sce-
nario – just that other portfolios should receive higher priority until
the full enterprise-wide stress-testing framework is operational. In
contrast, more meaningful and material results were obtained when
assessing unsecured and credit-card lending.

Case study: Korea’s credit-card crisis and structural changes in the

consumer market
Korea was chosen as a case study since we anticipated collecting a
relatively rich dataset, given that Korea experienced three crises in
10 years:

1. 1997–8 Asian crisis;

2. 2002–4 credit card crisis; and
3. 2007–9 global credit crisis.

The Korean credit-card crisis of the early 2000s provided a good

cycle to stress-test. The crisis was precipitated by a rapid increase in
credit growth and surging consumer debt levels. In particular, the
surge in Korean credit-card debt relative to GDP was closely fol-
lowed by a spike in impaired assets (Kang and Ma 2007). This crisis
provided a key ingredient for model development: a time series
that included a complete credit cycle to evaluate.
But, even with the Korean crisis providing a useful cycle, many
changes have occurred in the Korean consumer credit market since
2003, and these changes had to be taken into account when specifying
models for the post-credit-card-crisis period. In particular, the tighten-
ing of macroprudential regulations, stronger bank risk management
and controls (underwriting), better credit infrastructure and a less
crowded, more competitive banking system affected the environment.
Another important change in the Korean credit-card market
since the crisis was the change in product composition, from high-
er-risk cash-advance products to less risky instalment products. Ac-
cording to the Korean Financial Supervisory Service, cash advances
accounted for 65% of credit-card loans in 2002, compared with less
than 20% by 2011. These changes have contributed to the develop-
ment of a vastly different Korean consumer credit market, and this
structural change needed to be accounted for in the model.

154

stress_testing.indd 154 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

STRUCTURAL BREAKS AND DUMMY VARIABLES

Another potential structural break was from industry consolida-
tion, since a merger between a strong and weak institution can in-
troduce a structural break in the time series data for the combined
institution. This poses a decision point for the sample size for the
analysis – typically, a longer time series is desirable to ensure the
largest possible number of observations, but this must be weighed
against the need to control for structural breaks in the data series.
In the Korean estimations, a dummy variable for the credit-card
crisis was included in the credit-card model. The dates for the dum-
my variable were chosen based on the Quandt Andrews Breakpoint
test for structural breaks. Breaks were identified at December 2002
and April 2005.
Indeed, there is a broader point at work here, particularly when
considering emerging-market economies and other banking systems.
The emerging-market banking systems can be dynamic, rapidly
evolving systems, with new products being introduced routinely and
new borrowers with little to no experience with credit. Depending on
the magnitude and the size of the concentration the bank has, these el-
ements can introduce sizeable stress losses for the bank. These factors
can also introduce noise into time series data and need to be assessed
and potentially addressed within the model specifications.

ESTIMATING LOSSES ON KOREAN CREDIT CARDS AND

UNSECURED LOANS
A primary goal in the design of these models was to identify how
macroeconomic shocks affect portfolio credit quality. There are at
least two approaches for imputing a shock: one would be a straight-
forward application of a historical shock, while a second approach
would be to apply a combination of the most extreme performance,
to date, in terms of both depth and duration, for key economic vari-
ables from each of the crises. To have this stress – a low-probability
but high-impact event – Korea would have to experience internal
and external economic, financial market shocks simultaneously. It
is important to think through and estimate a wide range of sce-
narios given the potential for contagion. The focus of the discussion
below is on identifying the underlying risk factors that could cause
a substantive deterioration in the asset quality of the portfolio.

155

stress_testing.indd 155 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

To capture turning or inflection points in asset quality and pro-

vide a larger number of observations, monthly data was favoured
over quarterly data. For the credit-card market, bank-level data was
used, as the Korean supervisors did not start posting monthly per-
formance data until 2005. In many countries, the regressions will
have to be estimated using quarterly data, since that is the only
publicly available data, or banks may be able to use internal data-
bases that have higher-frequency data.
A decision on whether to estimate the level of losses or the ratio of
losses relative to credit-card receivables was also required in laying out
the approach. Additionally, the issue of modelling gross or net losses
was considered. The choice between the gross and net losses can have
dramatic effects on estimation results as recoveries (the difference be-
tween gross and net) can be lumpy on a monthly basis. In addition,
the pattern and behaviour of recoveries (and therefore net losses) will
vary by product type and country. Bankruptcy laws and other local
provisions (recourse versus non-recourse lending) and secured versus
unsecured lending (as well as the value of collateral) will all have an
impact on recovery assumptions. If using internal data, historical ex-
perience with recoveries can typically be extrapolated and evaluated
to determine the most appropriate dependent variable.

INDEPENDENT VARIABLES: KOREAN CREDIT CARDS AND

UNSECURED LOANS
In general, the performance of the consumer credit markets was
assumed to be thus:
To determine the drivers of credit-card loss rates, a large number
of variables that could cause cashflow problems and drive delin-
quencies and defaults were considered:

o economic growth (GDP) and related measures such as exports,

industrial production and the unemployment rate;
o monetary indicators such as inflation rates, credit growth and
interest rates; and
o wealth measures such as home prices and stock indexes.
Not all contemporaneous measures showed clear predictive power
for credit card loss rates

156

stress_testing.indd 156 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

DATA DIAGNOSTICS MAY ALTER THE MODEL SPECIFICATION

As part of data diagnostics, the data was reviewed graphically, and
statistical tests were conducted to determine the underlying prop-
erties for such things as stationarity. Stationarity, whereby statisti-
cal parameters do not change with time, is a required property for
time series regressions, which is often obtained by first differenc-
ing the variable, ie, subtracting the prior period’s observation from
the current period’s observation. We also checked for correlation
among the variables.

Table 10.2 Correlation matrix of Korean macroeconomic factors

unem- credit exports unem- unem- credit credit exports
ploy- growth index ploy- ploy- growth growth index
ment ment ment (-3) (-24) (-6)
(-2) (-3)
unem- 1.00
ploy-
ment
credit -0.42 1.00
growth
exports 0.08 -0.20 1.00
index
unem- 0.81 -0.50 0.20 1.00
ploy-
ment
(-2)
unem- 0.77 -0.51 0.24 0.92 1.00
ploy-
ment
(-3)
credit -0.29 0.89 -0.09 -0.48 -0.41 1.00
growth
(-3)
credit -0.03 -0.12 -0.48 0.00 0.02 -0.10 1.00
growth
(-24)
exports -0.06 -0.19 -0.04 0.00 0.00 -0.21 -0.19 1.00
index
(-6)

Numbers in parentheses indicate the lag of that variable

157

stress_testing.indd 157 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Table 10.2 shows that the unemployment rates are highly correlated
with one another (0.81 and 0.77) as is credit growth with the credit
growth from three months back (0.89). Any time a variable is highly
correlated (0.80 or greater), the model developer should consider
dropping one of the two variables.

ESTIMATION RESULTS
Given the likelihood of borrowers drawing on savings to stave off
defaulting, different lags of the independent variables were also
included in the regressions. Our estimations showed that lag struc-
tures can vary dramatically from indicator to indicator across coun-
tries. The following variables were important risk factors for Kore-
an credit-card loss rates: the change in the unemployment rate and
credit growth with 3-month and 24-month lags. The regressions did
incorporate a dummy variable identifying the credit-card crisis and
an autoregressive (AR) term. The AR term was used to correct for
serial correlation. For unsecured lending, the change in the unem-
ployment rate, credit growth and an export index were significant
in the estimation of monthly Korean unsecured loan loss rates over
the sample period 2000–10 (using monthly data).
As noted above in the general discussion on structural and cycli-
cal issues, rapid consumer credit growth is frequently a significant
factor (with a several-quarter lag) in contributing to future credit
losses, particularly if the seasoning of these new credits coincides
with an economic event. Various lags for the real credit growth
variable were significant but the lag length was purely a function of
the data as opposed to a point that would be predicted by theory,
ie, 6-, 12-, 18-, 24-month lags were not necessarily significant since
the timing of the cycle was not readily apparent. Akaike Informa-
tion Criterion (AIC) was used to determine the best lags of each
independent variable, but this raises the probability that the out-
of-sample predictive power of the model will be weak and require
further analysis of the model specification.
For example, several consumer credit risk managers have point-
ed out that, in emerging-market economies, consumer loan perfor-
mance can be dramatically affected by inflation (food prices, for
instance). A key consideration in this regard is the income segment
of the institution’s portfolio: lower-income borrowers are more vul-

158

stress_testing.indd 158 20/05/2013 18:45

 Stress Testing Across International Exposures and Activities

nerable to rising inflation rates than the more affluent consumers.

So the model could possibly be improved if this data were available
or could be constructed or proxied.

SUMMARY
Although the modelling of unsecured portfolios proved promising
as changes in macroeconomic variables showed a causal relation-
ship in the changes in Korean credit-card loss rates, the truly im-
portant element of this work was the required study of the Korean
credit cycles, consumer product markets, regulatory actions and
economic shocks arising from multiple quarters – both internal and
external – and that these factors differed to varying degrees across
countries in Asia. This underscored the need to evaluate markets
on an individual basis and to proceed with caution when trying to
estimate regressions across a region or globally. This is not a small
undertaking, but it yields a large benefit by being able to anticipate
potential drivers of future portfolio losses.

The views expressed in this chapter are those of the authors and do not neces-
sarily reflect those of the Office of the Comptroller of the Currency or the US
Department of the Treasury.

REFERENCES
IMF, 2012, “Macrofinancial Stress Testing – Principles and Practices”, working pa-
per, August 22, “Best Practice” principles 1–7, pp. 19–45.

Kang, Taesoo, and Guonan Ma, 2007, “Credit card lending distress in Korea in
2003”, Bank for International Settlements.

Laeven, Luc, and Thomas Laryea, 2009, “Principles of Household Debt Restructur-
ing”, IMF Staff Position Note, June.

Moody’s, 2009, “Approach to Estimating Singaporean Banks’ Credit Losses”.

OCC, 2008, “Country Risk Management, Comptroller’s Handbook”, p. 1.

159

stress_testing.indd 159 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

1 See [Link] See also IMF 2012, pp. 19–45, for an alternative
formulation of principles.
2 “Transfer risk is the possibility that an asset cannot be serviced in the currency of payment
because of a lack of, or restraints on the availability of, needed foreign exchange in the coun-
try of the obligor” – “Guide to the Interagency Country Exposure Review Committee Pro-
cess”, 2008, p. 1.
3 The six World Bank indicators are: voice and accountability; political stability/absence of
violence; government effectiveness; regulatory quality; rule of law; and control of corrup-
tion (see [Link]

160

stress_testing.indd 160 20/05/2013 18:45

 11
Liquidity Risk: The Case of the
Brazilian Banking System

Benjamin M. Tabak, Solange M. Guerra,

Sergio R. S. Souza, Rodrigo C. C. Miranda
Banco Central do Brasil

Stress tests are already a widely used tool for risk management of
financial institutions. Central banks and individual banks run these
tests for determining potential risk sources that they might encoun-
ter in scenarios of severe change in the macroeconomic situation
and assessing their resilience to such events. By testing themselves
or the financial system as a whole beyond normal operational ca-
pacity, they can quantify vulnerabilities, and the stability of the
given system or entity may be studied and pursued more easily
(Vazquez, Tabak and Souto 2012).
To design and apply a stress test, many important assumptions
should be taken. The first step must be identifying the specific risk
and vulnerability of concern. In the literature about stress testing of
banking risks, the most common type of risks considered are credit,
market and liquidity. The majority of papers have focused on as-
sessing credit risk, since this is the bank’s most important risk com-
ponent. However, liquidity stress testing is getting more visibility
and importance.
Although liquidity crises are not so frequent, their impacts are
high (low-frequency, high-impact events), especially due to their
contagious effects and to the consequences of the interaction be-
tween the banking risk factors. After the global financial crisis of
2007–9 there is an increasing interest in studying the vulnerabilities
provided by liquidity risks. From this important event many les-
sons can be taken. The De Larosière Group (2009) points out the key

161

stress_testing.indd 161 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

lesson that regulators paid little attention to the system as a whole,

while too much focus was given to microprudential supervision of
individual institutions.
The crisis served to show weakness in the stress-testing exercises
performed on financial institutions and systems around the world
(Ong and Čihák 2010). It also showed how the vicious dynamics
of liquidity risk can undermine the stability of the financial system
(Van den End 2010). To Aikman et al (2009), the crisis illustrates the
importance of modelling the closure of funding markets to finan-
cial institutions and accounting for liquidity feedbacks within any
model of systemic risk. In sum, the ongoing crisis serves as an alert
to the importance of managing liquidity risk and, therefore, it un-
derscores the need to explicitly take into account liquidity risk in
stress-testing frameworks (Van den End 2009).
Once they had understood the importance of stress testing li-
quidity risk, researchers working for different financial institutions
around the world started to develop methods to endogenise liquidi-
ty risk in a stress-testing framework. This task is quite complex, since
a method has to be developed that has the ability to quantify depen-
dences and interactions between the various types of risk. Wong and
Hui (2009) suggest that for banking stability it is important to assess
the extent to which a banking system is exposed to the interaction of
risks. In their paper, the stress-testing framework explicitly captures
the link between default risk and deposit outflows. Not only is the
interaction of the risks incorporated but also their contagious effects.
The framework presented by Aikman et al (2009) also attempts to
fully integrate funding risks and solvency risk.
In a framework of stress testing for liquidity risk, two compo-
nents are important: (i) funding liquidity risk (concerning the
bank’s balance-sheet liability side: there may be a bank run by de-
positors or the bank may be unable to rollover liabilities) and (ii)
market liquidity risk (asset side: illiquidity in the market for the
bank’s assets, when the bank needs to sell them). An example of
a stress-test model that involves both components is the one pre-
sented by Van den End (2009). By considering the first and second
rounds (feedback) of shocks, the model presented endogenises
market and funding liquidity risks and captures, as second-round
effects, the collective response of heterogeneous banks and reputa-

162

stress_testing.indd 162 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

tional effects. The IMF originally centred its liquidity tests on the
paper of Čihák (2007) using bank balance-sheet data to perform
bank-run-type stress tests on a bank-by-bank level. Aikman et al
(2009), on the other hand, focused on the role of asset-side (market
liquidity) feedbacks.
Some papers innovate with their stress-testing models. One topic
that motivated some interesting material was the establishment of
minimum standards for liquidity risk (Liquidity Coverage Ratio,
or LCR, and Net Stable Funding Ratio, or NFSR) by Basel III (BCBS
2010). To study the effects of these new minimum standards, Van
den End (2010) developed a stress study that linked funding cost li-
quidity to regulation and central bank operations. The conclusions
from its model outcomes support policy initiatives such as the ones
proposed by the Basel Committee (BCBS 2010). By testing scenarios
of stress, the paper finds that banks that adjust to the Basel III estab-
lishments (such as by holding a higher stock of liquid assets) have
substantially lower second-round effects and tail risks. These find-
ings highlight the importance of defining a sufficiently high-quality
level of liquid assets to limit the idiosyncratic risks to a bank. The
outcomes of the tests also evidence the important role of stronger li-
quidity profiles in reducing the risk of collective reactions by banks
and therefore in preventing second-round effects and instability of
the financial system as a whole.
Van den End and Kruidhof (2012) simulate the systemic implica-
tions of the LCR using a liquidity stress-testing model. The authors
model the LCR as a macroprudential instrument that can be used
to moderate the adverse side effects that arise due to interactions of
bank behaviour with the regulatory liquidity constraint. The authors
applied tests with different switching rules and banking sector struc-
tures. By testing the reduction of the minimum LCR requirements,
the paper finds that a flexible approach of the LCR in stressed times
reduces the number of bank reactions and associated negative side
effects. Another rule tested was the widening of the buffer definition,
and the measure was found to be effective in limiting the interaction
between the minimum requirement and bank reactions. At extreme
stress levels, the paper finds that the LCR becomes ineffective as a
macroprudential instrument and, in order to maintain the stability of
the system, a lender of last resort is requested.

163

stress_testing.indd 163 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

The development of the framework that endogenises liquidity

risk into stress tests is an essential stage of the stress-testing ex-
ercise. Maybe just as important are the stages of data-collecting,
information-processing and numerical analysis. The top-down
and bottom-up approaches are the two strategies of information
processing that can be applied to stress testing bank risks. The ad-
vantage of running a test with the bottom-up approach is the use
of more detailed data and less complexity in modelling liquidity
shocks. The disadvantage is that, unlike the top-down approach,
these tests are less consistent. The advantages of the top-down ap-
proach include more consistent results and more flexibility to simu-
late different scenarios of shocks. According to Čihák (2007), the
majority of stress tests presented in financial-stability reports are
based on bank-by-bank data. Central banks that are not involved
in microprudential supervision and do not have access to more de-
tailed data rely on top-down approaches.
The BCBS (2011) published a document on the progress in imple-
menting macroprudential policy frameworks, gathering information
on the subject. Although the document discussed macroprudential
policy and systemic risk, many of its considerations can be readily
applied to liquidity risks, given their systemic nature. According
to the document, risk measures should be able to capture the time
and cross-sectional dimensions of risk, which define requirements
for metrics to be created and monitored. The main measurement ap-
proaches that apply, extracted from a survey conducted by the IMF
(2011), includes the following: indicators of imbalances (eg, of bank
credit, liquidity and maturity mismatch, and currency risk), indica-
tors of liquidity market conditions, metrics of concentration of risk
within the system and stress testing. The metrics of concentration
of risk within the system (eg, the network-theory-related measure-
ments, like connectivity and centrality, or the results of default cas-
cade simulations) are related to the cross-sectional dimension of risk,
focusing on the channels of contagion and amplification, and could
be used to determine the systemically important institutions, while
the stress testing is used to evaluate the resilience of individual banks
and of the banking system as a whole.
Stress tests for liquidity are not so developed as stress tests for
credit and market risks. However, important works have been done

164

stress_testing.indd 164 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

by central banks researchers, as the already mentioned Van den End

(2010), Wong and Hui (2009), and Aikman et al (2009). These models
are usually integrated with credit or market risk. This feature is the
main difference between these models and the approach for liquid-
ity stress testing at the Central Bank of Brazil.
Despite the importance of liquidity risk stress testing, most central
banks do not publish results from liquidity stress tests. This reflects
the liquidity-modelling complexity and the need of more detailed
and high-frequency data. The Central Bank of Brazil has published
liquidity stress test results since 2009. From the side of the banks,
a survey with Brazilian banks indicates that their risk-management
policies have been improved to account for possible liquidity prob-
lems. Many banks have started to run liquidity stress tests after the
financial crisis. However, it is not usual to disclose the results.
Given the importance of developing liquidity stress-test models
we focus on the Brazilian banking system and how it has been im-
pacted on by the financial crisis, focusing on liquidity issues. We
then present the Brazilian banking system, before discussing the
impacts of the crisis on the system, employing contagion tests to
show that banks may have heterogeneous responses to liquidity
shocks. The following section presents a discussion on liquidity
stress testing performed by the Central Bank of Brazil, as well as
results from a survey of banks that operate in the Brazilian banking
system. The last section draws our conclusions.

THE BRAZILIAN BANKING SYSTEM

Banks are financial institutions with a major role in a capitalist
economy. Their importance is a consequence of their roles as money
creators, as managers of the payments system and as financiers of
economic activities. At the same time, as rational agents, banks take
actions to maximise their profits. Restrictions can come from the mac-
roeconomic environment and from the banking system microstruc-
ture. This condition gives a hint on the risks involving the banking
system and why they are potentially dangerous. Fragility in banks’
individual accounting and management, combined with macroeco-
nomic shocks, can lead to crises in the system. When such crises hap-
pen, the consequences can be great and will include impacts on the
economies’ credit situation, interest rates and investments, and bring

165

stress_testing.indd 165 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

about negative changes in the levels of economic activity. To main-

tain a solid and healthy banking system, it is essential to establish
bank regulations supplemented by constant supervision.
The Brazilian banking system began its trajectory at the beginning
of the 19th century, when the Banco do Brasil (Bank of Brazil) was
founded and later was partially considered a monetary authority (da
Costa, 2012). However, it was only between 1930 and 1945 that the
most important banks were founded and the Brazilian banking sys-
tem effectively began to grow, reaching the total of 644 banks in 1944.
Since its start, Brazil’s banking system went through various transfor-
mations, mainly adaptations to the various changes in national and
international politics and economic scenarios. These transformations
led to a system with solid regulations, supervised by the Central Bank
of Brazil (created in 1964). Based on the Federal Constitution of 1988,
some of the Central Bank functions as a monetary authority are the is-
sue of money, the determination of the reserves requirements of banks
and controlling liquidity with open market operations.
The banking system’s condition in the late 20th and early 21st
centuries was shaped by important structural transformations
that occurred in the early 1990s. These transformations were con-
sequences of the implementation of measures of monetary policy
in 1994 and 1995. The 1994 measure is known as “Plano Real”, in
which the exchange rate between the Brazilian currency (real) and
the American dollar was initially set at 1 to 1. This measure was
used by the government to stabilise the economy, which had been
passing through a long period of high inflation rates initiated in
1964, during the military regime (1964–85). The impact of this plan
on the banking system was deep. One of the major changes faced
by banks was on their profits’ sources. During the high inflation
period, banks took advantage of the condition to profit from float-
ing; however, after the currency stabilisation, this kind of revenue
vanished. Banks found an alternative source of profit by charging
their customers fees for services provided. The demand for credit
also increased, given the increase of the predictability horizons al-
lowed by the stabilisation of the economy and the more optimistic
associated expectations. The banks’ profit with services, which rep-
resented only 8% of the GDP in 1990, reached 10.5% in 1993 and
21.5% in 1995.

166

stress_testing.indd 166 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Important policies were also implemented in 1995. Programmes

were created to restructure and fortify the financial system, pre-
venting liquidity crises and stabilising the system. Another relevant
measure taken that year were the incentives given to the opening
of the Brazilian financial system to foreign capital and banks. The
objective of this action was to attract foreign banks to the national
system and expand the credit supply. This would increase banks’
competition, forcing them to reduce costs by improving their man-
agement to become more efficient. To the population and firms, the
benefits would be a greater variety of banks and lower interest rates.
The measure was indeed effective in attracting more foreign banks,
but the concentration increased. This happened because mergers
and acquisitions not only involved the new-entrant foreign banks
but also were performed by domestic institutions among them-
selves. During the first four years after the “Plano Real”, 104 fi-
nancial institutions suffered some kind of adjustment. The number
of domestic private banks and state-owned banks reduced while
the number of foreign-controlled banks increased by less than the
reduction in the number of domestic banks. In 1993, foreign-con-
trolled banks owned 7.28% of the total financial system, reaching
25.91% by 1999.
From 1999, Brazilian banks began an innovation process. They
developed techniques of fundraising and asset management, in-
creasing their loans-to-reserves ratios. The efficient use of the in-
terbank market can be considered the key innovation (da Costa
2012). The period between 2003 and 2006 was marked by greater
access by the population to banks and credit. Between 2001 and
2006, the number of accounts in the banking system increased by
52%. Savings accounts were the most popular service provided and
increased 50%, while the growth of current accounts was 37%. The
popularisation of banking services became possible due to tech-
nological advances, which included the installation of ATMs and
credit-card readers in busy areas and retail outlets. These indirect
banking facilities increased the supply of banking services with-
out the need for an increase in the number of bank agencies. Popu-
lar credit programmes offered by commercial banks also brought
about an expansion in the economy’s consumption demands.
By 2012, the Brazilian financial system had 2,218 operating finan-

167

stress_testing.indd 167 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

cial institutions. In December 2011, the total assets of Brazil’s whole

financial system exceeded BRL5,135 billion. Its stock of credit op-
erations reached BRL2 trillion, which corresponds to 49% of the
country’s GDP in the same period. The banking system is a part
of the financial system, composed of independent institutions and
financial conglomerates, which must contain at least one institution
from a commercial bank, a savings bank or a multiple bank, since it
is authorised to receive demand deposits. In March 2012, the bank-
ing system’s assets totalled BRL4,486 billion, a share of 84% of the
whole financial system’s total assets. These institutions’ evolution
of total assets is illustrated by Figure 11.1.

Figure 11.1 Banking-sector asset evolution

BRL Billion
5000

4500

4000

3500

3000

2500

2000
Jun Dec Jun Dec Jun Dec Jun Dec
2008 2008 2009 2009 2010 2010 2011 2011

Total assets

The Brazilian banking system’s history shows periods of concentra-

tion alternating with periods of increase in the number of banks. In
the early 1990s, as we have seen, the banking system went through
a structural transformation. In that period of banking crisis, priva-
tisations and incentives to the entry of foreign banks, the domestic
banking system went through a decrease. In 1994, there were 271
commercial and multiple banks; by 2002, they were 167. In 2012,
there were 160 banks.

168

stress_testing.indd 168 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Figure 11.2 Distribution of the banking sector by capital source

(December 2011)

5.6%
3.8%

Private - national (89)

Private - national with

foreign control (56)
55%

Private - foreign branch (6)

%
35%
State-owned (9)

Figure 11.3 Distribution of the banking sector by assets and credit operations

60%
52.98%
50%
43.02%
39.56%
40%

29.11%
30%

20% 17.80% 17.39%

10%
0.11% 0.03%
0%
Private - national Private - national Private - national State-owned
with foreign branch
control

Total assets Credit operations

Figure 11.2 shows the division of the Brazilian banking system by

type of control (in number of banks and percentages). The participa-
tions of each of these categories in the aggregated credit operations
and total assets in the same period are shown in Figure 11.3. This fig-
ure shows that the state-owned banks increased their participation

169

stress_testing.indd 169 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

in credit operations as a result of government measures to maintain

the level of economic activity after the crisis. Total credit operations
grew between the second half of 2009 and the same half of 2011, as
shown in Figure 11.6. Additional indicators of the financial system’s
concentration are the HHI and CR4 indexes. HHI is the Herfindahl–
Hirschman Index. According to the Horizontal Merger Guidelines,
published by the Department of Justice and Federal Trade Commis-
sion (EUA), HHI of less than 0.15 means that the market is not con-
centrated; HHI between 0.15 and 0.25 means that the market is mod-
erately concentrated; and HHI above 0.25 indicates that the market
is highly concentrated. CR4 stands for four-firm concentration ratio.
It measures the market share of the four largest banks in the system.
Values between 50% and 80% indicate medium concentration. The
levels of concentration in the Brazilian financial system are moni-
tored by the Central Bank of Brazil. These indexes are calculated for
asset totals, credit operations and deposit totals. The HHI values of
the system for the second half of 2011 indicate that the Brazilian bank-
ing system is non-concentrated to moderately concentrated from the
point of view of all these three quantities (respectively 0.13, 0.14 and
0.1509). The CR4 readings for the three quantities in the same period
were 67.21%, 69.2% and 72.55%, respectively, which show that the
four largest banks, in total assets, hold a slightly higher total depos-
its ratio, with approximately the same total credit ratio. This means
these banks, as a group, do not have a special participation in the
system: as credit providers or deposit holders, their participation cor-
responds to their size.

EFFECTS OF THE SUBPRIME CRISIS ON BRAZILIAN BANKING

SYSTEM
Since 2007, important events have been taking place in interna-
tional banking. The American banking crisis of 2008 spilled over
into economies around the world. The collapse of large banks in
the US had a domino effect that led to the collapse of even entire
economies, as occurred in Greece in the late 2000s. The impact on
the Brazilian banking system and economy wasn’t as catastrophic.
Due stricter regulations and controls imposed in 1995, the Brazilian
banking system has remained relatively solid when facing the in-
ternational crises and has been preserved without much loss. Also,

170

stress_testing.indd 170 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

the macroeconomic politics of fiscal austerity and the regime of in-

flation targeting adopted allowed Brazil to stand out among emer-
gent countries and continue to attract foreign interest. The liquidity
situation of Brazilian banks can be shown by the evolution of the
system-wide liquidity index in Figure 11.4.

Figure 11.4 System-wide liquidity index

3.0
2.5
2.0
1.5
1.0
0.5
02/01/2008 02/07/2008 02/01/2009 02/07/2009 02/01/2010 02/07/2010 02/01/2011 02/07/2011
Liquidity index (5 days moving average)

Source: Financial Stability Report - March 2012

The index is calculated by the Central Bank and is the ratio between
(a) institutions’ total liquid assets available to honour their obliga-
tions and (b) the possible losses in liquidity that institutions would
be subject to in stress situations. Such situations include unexpected
withdrawals and sudden changes in the market scenario. The BCB
publishes an aggregated liquidity index for the whole banking sec-
tor in the financial-stability report, along with a detailed liquidity
analysis of the financial system. More details on calculation of the
liquidity index and its use to monitor the financial system liquidity
will be presented later.
Volatility in exchange and interest rates usually increases the
liquidity required in the case of stress situations (has a negative
impact on the index). Figure 11.5 illustrates the behaviour of the
volatility of these rates since 2008. The highest volatility occurred,
of course, when the crises began. Figure 11.6 shows the finance of
credit expansion and liquid assets since 2008.
From the Figures 11.4 and 11.5, it can be observed that the interna-
tional financial crisis had a greater impact on the Brazilian financial
system’s liquidity during 2008. During the period, the volatility of the
exchange and interest rates was very high and certainly increased the

171

stress_testing.indd 171 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

possible losses that the institutions would be subjected to in concrete

stress situations. This is related to a decrease in the liquidity index
during the period. After 2008, the trajectory of the system’s liquidity
had a recovery and the trend for the following year was of healthy li-
quidity conditions. According to the Financial Stability Report (FSR)
from the Central Bank in the second half of 2009, the banking system
presented an large amount of high-quality liquid assets and had low
dependency on foreign resources. These conditions reduced Brazil’s
vulnerability to liquidity risks and international turbulence. By the
first half of 2010, the liquidity of Brazilian financial institutions had
returned to the pre-crisis level (BCB 2010).

Figure 11.5 Exchange and interest-rate volatility

0.007 0.060

0.006 0.050

Exchange rate volatility

Interest rate volatility

0.005
0.040
0.004
0.030
0.003
0.020
0.002

0.001 0.010

- -
Jul-08 Jan-09 Jul-09 Jan-10 Jul-10 Jan-11 Jul-11 Jan-12

Interest rate short-term Interest rate long-term Exchange rate

Source: Financial Stability Report - March 2012

The Financial Stability Report (BCB 2012) concerning the year of 2011
concludes that the banking system’s liquidity is in a very favourable
situation. The Brazilian banking system had the ability to finance its
own operations, mostly with funds raised in the domestic market.
In the first half of 2011, the system’s funding increased by BRL186.1
billion, representing a 9.1% increase compared with the previous se-
mester. The growth in liquid assets (composed basically of federal
government bonds) was remarkable, favoured by the slower growth
of the volume of credits (BCB 2011). In the second half of 2011, the
system’s funding increased by BRL246.9 billion (an 11.2% increase).
In this period, the credit expansion was reduced by the available re-

172

stress_testing.indd 172 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

sources from domestic and foreign markets. The liquidity index re-
mained at a good level even after the negative shock caused by the
volatility of the interest and exchange rates (BCB 2012).

Figure 11.6 Credit operations and liquid assets growth

BRL billion
280

210

140

70

-70
2º/2009 1º/2010 2º/2010 1º/2011 2º/2011

Net funding Credit operations Liquid assets

Source: Financial Stability Report 2012

Deposits (including savings, on-demand and time deposits) have

been the major funding source for Brazilian banks. Deposits pre-
sented a declining trend in terms of the share of total funding from
December 2005 to December 2007. However, in 2008, this trend was
reversed due to the crisis’s effects on time deposits. Time deposits’
interest rate increased to attract funds, which would compensate for
the reduction in other sources of liquidity, especially foreign funding.
Consequently, the amount of time deposits increased 31.1% in the
second half 2008 (BCB 2008 and 2009). Savings deposits have been
tracking the funding growth, remaining stable in terms of relative
shares. From 2008 to 2011, savings accounted for 17% to about 20%
of total funding (see Figure 11.7).1 Regarding on-demand deposits,
we can see in Figure 11.7 that their relative shares declined from
10% in December 2009 to 7.8% in December 2011. On average, these

173

stress_testing.indd 173 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

three types of deposits account for more than 60% of total funding
between 2008 and 2011. Deposits up to BRL70,000 are guaranteed
by the Credit Guarantee Fund (the Fundo Garantidor de Créditos
(FGC)). From the total amount of time deposits, the largest holders
are households, followed by legal entities (see Figure 11.8).

Figure 11.7 Funding sources

%
100
90
80
70
60
50
40
30
20
10
0
Jun Dec Jun Dec Jun Dec Jun Dec
2008 2008 2009 2009 2010 2010 2011 2011

Savings On demand deposits

Time deposits Repo with corporate bonds
Liabilities on loans Financial bills
Others

Source: Financial Stability Report - March 2012

Financial bills became a more interesting source of funding due to an

exemption on reserves requirements for their holders, which became
effective in late 2010. Since then, financial bills have presented a growth
trend; however, they account for only a small share of the system’s to-
tal liabilities (BCB 2012). Financial bills are a source of long-term fund-
ing and have contributed to the lengthening of the banking system’s li-
abilities profile, as they cannot be redeemed in total or partially before
their maturity date (according to the Resolução BCB No. 3.836/2010).
This is desirable since the loan’s average term has increased due to an
increased number of mortgages in the credit portfolios (BCB 2011).

174

stress_testing.indd 174 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Figure 11.8 Aggregated time deposits

BRL billion
400

350

300

250

200

150

100

50

0
Jan May Sep Jan May Sep Jan May Sep Jan May Sep Jan May Sep Jan May Sep Jan May
2006 2007 2008 2009 2010 2011 2012

Financial institutions Institutional investors Legal entities Households

Between 2008 and 2011, liabilities on loans have accounted, on aver-

age, for about 16% of total funding. These sources of funding include
foreign funding. Most banks that use foreign funding are small foreign
banks, whose business model is not related to credit. Nevertheless, only
a small part of these banks’ funding comes from abroad. The large for-
eign banks rely mainly on domestic funding. Besides this, liabilities in
foreign currency have reduced since the subprime crises (see Figure
11.9). Thus, turbulences in international markets have had a limited im-
pact on the Brazilian banking system (BCB 2012). Liabilities on loans
also include loans from the Brazilian Development Bank (BNDES).
A general look at banks’ funding structures highlights the exis-
tence of institutional differences. Large banks have more diversified
sources of funding, and, due to a wide network of branches, these
banks have more access to retail deposits. This source of funding is
more stable, reducing the liquidity risk. On the other hand, small-
er banks rely mainly on time deposits and have a less diversified
funding structure. Financial bills provide a more stable source of
funding to their issuers, while being more attractive for their hold-
ers, especially if they are large banks, due to a reserve requirement
exemption associated with it, which is larger than the one related
to time deposits. As smaller banks already have an exemption from

175

stress_testing.indd 175 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

reserve requirements due to their low amount of funding, holding

financial bills is not as interesting for these banks (see Figure 11.10).

Figure 11.9 External liabilities

11

10

4
Mar Jun Sep Dec Mar Jun Sep Dec Mar Jun Sep Dec Mar
2008 2009 2010 2011

Note: Operations with foreign currency or foreign counterpart.

Source: Financial Stability Report - September 2011

Figure 11.10 Funding distribution by bank size

%
100
90
80
70
60
50
40
30
20
10
0
Big Medium Small Micro

On demand deposits Time deposits

Savings Repo with corporate bonds
Liabilities on loans Financial bills
Others

Source: Financial Stability Report - March 2012

176

stress_testing.indd 176 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Funding-source patterns also differ sharply among the types of con-

trol segments. Foreign banks concentrate their funding on time de-
posits, while public banks tend to emphasise savings. On the other
hand, private banks are more focused on repo operations (BCB 2008).
Altogether, the Brazilian banking system relies mainly on domes-
tic sources of funding and is prepared to cope with an occasional
liquidity stress. However, liquidity is not equally distributed among
banks. Smaller banks that rely on time deposits from large custom-
ers are subject to higher liquidity risks during stress periods. These
banks can get funding from credit assignment transactions. In Au-
gust 2011, the Central Bank of Brazil created the Credit Transfer Bu-
reau, in which banks must register credits’ assignments (BCB 2011).
Brazilian banks have a more stable funding source, as the reli-
ance on retail deposits is higher than on wholesale funding. None-
theless, for some banks, institutional investors may represent an
important funding source. Therefore, in moments of stress, these
banks may incur in liquidity problems. To overcome these prob-
lems during the financial crisis the Central Bank of Brazil created
a new type of deposit that is guaranteed by the Credit Guarantee
Fund (the Fundo Garantidor de Créditos (FGC).
The FGC has an important role in the security of the national
financial system. In March 2009, the FGC’s Special Guarantee of
Time Deposits (DPGE) was implemented. This measure helped the
smaller institutions to recover their funding (the amount in the term
deposits of small banks grew by about 24% from March to May of
the same year). An improvement in the rediscount regulation was
also implemented. The deadlines of the rediscount operations were
extended and the central bank was authorised to impose restrictive
prudential measures to manage the financial institutions (Mesquita
and Torós 2010).
The BCB took measures to address the liquidity constraint both
in domestic and foreign currencies: bank reserve requirements
were lowered; lines of credit in foreign exchange were provided to
the private sector; the central bank offered US dollars in spot mar-
ket auctions and foreign-exchange swap contracts.2
Figure 11.11 shows an increase in time deposits, from BRL270
billion in December 2007 to over BRL500 billion in October 2008.
The largest growth in deposits was observed in financial institu-

177

stress_testing.indd 177 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

tions and households, followed by companies with a more modest

growth. Institutional investors’ deposits were maintained at about
the same level. That signals a shift towards seemingly less risky
investments (that is, time deposits).

Figure 11.11 Time deposits by client type

BRL billion
400

350

300

250

200

150

100

50

0
Jan Jun Nov Apr Sep Fev Jul Dec May Oct Mar Aug Jan Jun Nov Apr
2006 2007 2008 2009 2010 2011

Financial institutions Institutional investors Households Companies

Figure 11.12 Time deposits by institutional investors

BRL billion

25

20

15

10

0
Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan Jul Jan
2006 2007 2008 2009 2010 2011 2012

Large Medium Small Micro

178

stress_testing.indd 178 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Conventional wisdom would expect more informed investors to

seek better rates (and therefore riskier investments) in tranquil times,
and safer investments in riskier times. That was also evident in time
deposits from institutional investors in Brazil. In 2006, large banking
institutions and conglomerates had the lion’s share of time deposits
from institutional investors, about BRL15 billion, whereas the medi-
um-sized banks’ share was about BRL6 billion. From January 2007 to
December 2007 this was reversed, with about BRL19 billion in time
deposits in medium banks and about BRL3 billion in large banks. In
January 2008 the situation was once again reversed, as time deposits
shifted towards large banks, which are usually regarded as less risky.
That movement can be seen quite clearly in Figure 11.12 and signals
a flight-to-quality movement in time deposits.

Figure 11.13 Banks and financial conglomerates affected by contagion

in time deposits - according to ownership

Number of institutions
50
45
40
35
30
25
20
15
10
5
0
Feb Aug Jan Jun Nov May Mar Sep Feb Aug
2006 2007 2008 2009 2010

Foreign Private national State-owned

This flight to quality is also evident in correlation contagion tests.

Regarding financial contagion as “a significant increase in correla-
tions after a shock to one institution”, a series of correlation-change
tests was run based on the Forbes and Rigobon (2002) statistic (FR)
devised by Fry, Martin and Tang (2008). In these tests, the log dif-
ference of the weekly time deposits’ stock was tested for contagion
using a vector auto-regressive model and 30-week crisis windows.

179

stress_testing.indd 179 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

The test results indicate how many institutions were affected by

contagion within each crisis window, and are summarised in Fig-
ures 11.13 and 11.14. The contagion results indicate that most banks
affected by contagion were small and medium-sized banks. Regard-
ing ownership, foreign institutions were the most affected by con-
tagion, followed by Brazilian private domestic banks. The peak of
contagion was in the windows starting in September to November
2007 and ending in March to May 2008. These time-deposit move-
ments and time-deposit contagions seem to indicate that, during
the financial crisis, there was an investor movement towards assets
such as guaranteed time deposits or time deposits at large national
institutions that were deemed safer at the time. During the crisis,
the market perception was that the resilience of small and medium-
sized banks reduced, which was translated into the transfer of time
deposits from these institutions to large ones, corresponding to a
liquidity transfer between them.

Figure 11.14 Banks and financial conglomerates affected by contagion

in time deposits – according to size

Number of institutions
30

25

20

15

10

0
Feb Jun Sep Dec Mar Jun Sep Dec Apr Aug Mar Jul Oct Jan Apr Aug Nov
2006 2007 2008 2009 2010
Large Medium Small Micro

LIQUIDITY STRESS TESTS IN BRAZIL

Central bank’s approach
In Brazil, liquidity risk monitoring is part of the banking supervi-
sory process and includes a continuous follow-up of the systemi-

180

stress_testing.indd 180 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

cally important financial institutions and a liquidity stress test. The

liquidity stress test combines the bottom-up and top-down ap-
proaches. It considers, for each individual financial institution, the
different classes of assets and raised funds, but does not take into
account the linkages among institutions, resulting in a liquidity in-
dex for each institution. This index is a short-term liquidity index
similar to Basel III’s LCR (see Figure 11.4).
The liquidity index is the ratio between the total liquidity and the
estimated liquidity needs. The total liquidity is the amount of liq-
uid assets each institution can dispose of to meet its obligations. It
is calculated as the sum of active market operations with maturity
on the next day (eg, involving federal securities, active interbank
deposits and bank deposit certificates maturing the next day), with
active interbank deposits and bank deposit certificates maturing af-
ter the next day, weighted by coefficients associated with a possible
early redemption of these instruments. The calculation of the total
liquidity also considers the balance of other accounting assets: cash,
shares, foreign currencies and investments in mutual funds, gold
and foreign federal securities.
The estimated liquidity needs is the liquidity level an institu-
tion needs to keep to withstand funding volatility and losses under
market stress. It is calculated from:
o deposits’ volatility under stress on a two-week horizon;
o deposits’ concentration index (excluding interbank deposits), tak-
ing into account value ranges and client profiles (individuals, firms,
financial institutions and institutional investors);
o Interbank deposits raised maturing after the next working day,
considering, for short-term Interbank deposits, that they will not
be renewed, and a possible early redemption for the remaining
Interbank deposits;
o remaining liabilities on the balance sheet; and
o stressed market net positions.

Liquidity stress tests are very useful to assess whether specific banks
have liquidity vulnerabilities. In this case, bank supervision can fol-
low up a bank’s risks and make accurate interventions. Furthermore,
it is useful to design proper public policies to reduce shocks that stem
from systemic liquidity problems. The Central Bank of Brazil has

181

stress_testing.indd 181 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

performed macroeconomic stress tests and monitored the liquidity

of financial institutions to identify possible sources of liquidity stress.
Additionally, it has sought to enhance the efficiency and stabil-
ity of the Brazilian financial system by improvements to the ac-
counting procedures for registering credit assignment operations,
by the reduction of the issuance limits of DPGEs by the FGC, by
changes on the balance-sheet items subject to reserve require-
ments and by the gradual introduction of Basel III’s recommenda-
tions (BCB 2012).
An important measure in the case of Brazil has been the DPGE,
which has helped create a liquidity cushion for medium-sized
banks, which were suffering from a liquidity shortage immediately
after the crisis. These measures have proven to be very successful at
relatively low cost, and have increased confidence in the financial
system, which is crucial in the middle of a crisis.
The main lesson to be drawn from the financial crisis is that li-
quidity is crucial. Evaluating it on a continual basis is important
and the results from liquidity stress tests can be a very useful moni-
toring tool and suggest whether liquidity problems are local, spe-
cific to certain banks or systemic, in which case public policies can
be triggered to help circumvent these problems.

Individual banks’ approaches

This section presents some results from the liquidity stress-testing
survey carried out by the Central Bank of Brazil on June 2012. The
survey aimed to better understand the methods and scenarios that
banks used in their liquidity stress tests. It is similar to the survey
applied to European banks (ECB 2008).
To mitigate liquidity risks, banks need effective risk manage-
ment. Fundamental to this task, liquidity stress tests allow banks
to assess the possible impact of exceptional but plausible stress sce-
narios on their liquidity position and can help them to determine
the size of liquidity buffers.
The respondent banks say liquidity risk is considered the second
most important type of risk in their risk management; the most im-
portant is credit risk. Although some banks have been performing
liquidity stress tests since about 2002, the majority of them began to
perform these tests after 2008.

182

stress_testing.indd 182 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

A total of 46 large banks received the survey and 27 banks pro-

vided information about their liquidity stress tests, including the
largest Brazilian banks. From these banks, 23 perform internal li-
quidity stress tests while 4 of them use vendors’ models.
All banks but one in the survey quantify liquidity risk toler-
ance. In the sample, 17 banks affirm they quantify risk tolerance
by a system of limit settings. These limits are usually defined
based on expert judgement. In eight banks, the quantification of
liquidity risk tolerance is based on stress tests. Other forms, less
frequently used, to quantify liquidity risk tolerance are: cashflow
forecast (6 banks), concentration of the liquidity sources (2) and
survival horizon (1). The ECB (2008) affirms that banks focus on
risk containment – systems of limits interrelated with liquidity
risk tolerance – rather than the quantification of liquidity risk tol-
erance per se. The explanation for this is that the quantification of
liquidity risk tolerance is a difficult task. The major problem in
the area of liquidity risk management is that liquidity risk events
are of low probability but high impact, which implies that is not
feasible to assign probabilities to all (reasonably well-defined)
possible liquidity shocks. It seems that Brazilian banks have the
same focus.

Time horizon
The time horizon of liquidity stress tests indicates the horizon the
bank considers necessary to take corrective measures to mitigate
liquidity risks occasionally detected. The majority of banks (17)
perform liquidity stress tests monthly and some banks (8) perform
them daily. Time horizons for stress-test scenarios mainly vary be-
tween four weeks and three months, although longer time horizons
are also cited (see Figure 11.15). Almost every bank uses short or
medium time horizons to perform their stress tests. However, the
period considered as short, medium or long is not uniform among
banks. A short period may comprise from one to twelve weeks,
while a medium period comprises from two to twelve months.
The most commonly time horizons considered are four weeks for a
short period and three months, for a medium period.

183

stress_testing.indd 183 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Figure 11.15 Time horizon

no replay 5% 1 week 5%
3 years 5%

2 years 2% 4 weeks
29%

1 year 7%

9 months 2%

6 months 2%

3 months
38% 2 months
5%

Source: BCB survey

Scenarios
The scenarios banks use consider the risk sources and magnitudes
they relate to their business. Most banks (15) perform tests under
market-wide stress scenarios, but only six banks use idiosyncratic
scenarios. A considerable number of the surveyed banks (13) use a
combination of adverse market conditions and idiosyncratic shocks
to their institutions. Of these banks, only 9 run the combined sce-
nario, while 3 also run both market and idiosyncratic scenarios
separately and one bank also runs the market scenario. Of those
banks that do not run tests with combined scenarios, the majority
(15) rely exclusively on either tests with market stress scenarios (9)
or tests with a firm-specific stress scenario (1). Five banks declared
that they considered other types of stress test scenarios and one did
not respond to the question (Figure 11.16).

184

stress_testing.indd 184 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Figure 11.16 Types of stress-test scenario

Adverse market
conditions (I) and
No replay
idiosyncratic shocks
1
(II) 2
Adverse market
conditions (I) and
combinations of (I)
Adverse market
and (II) 1 conditions (I)
9

Other
scenarios
5

Adverse market
conditions (I),
idiosyncratic Idiosyncratic
shocks (II) and shocks (II)
combinations of (I) 1
Combinations
and (II) of (I) and (II)
3 9

Source: BCB survey

The surveyed banks described a multiplicity of scenarios with dif-

ferent sets of assumptions concerning the effect that these scenarios
were expected to have on both the assets and liabilities sides of
their balance sheets. However, there are some sources of stress that
are common in most scenarios:

o reduction in asset prices;

o increased collateral and margin calls;
o increased delinquency;
o reduced access to funding markets;
o increased deposits withdrawals;
o non-rollover of term deposits; and
o utilisation of credit lines previously approved.

Although most banks claim they perform a combination of adverse

market conditions and idiosyncratic-scenarios tests, it is not defined
which shocks they consider as coming from market conditions or from
bank-specific situations. Only a few banks consider bank-specific sourc-
es of stress, such as downgrade and liquidity problems in the group.

185

stress_testing.indd 185 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Most banks make assumptions about deposits in their stress sce-

narios. These assumptions are consistent with the structure of the
BBS and the economic outlook. In the BBS, banks rely on domestic
funds provided mainly by deposits. Another source of stress, usual
for more than one-third of the banks, is an increase in delinquency.
Brazil has experienced a fast credit growth, so banks seem to be
aware of the effects of credit risk on liquidity.
According to the IMF (2012), the Brazilian financial sector is ex-
posed to international commodities and capital markets’ volatility
effects, but the risks related to them are significantly mitigated by
a flexible exchange rate; strong macro- and microprudential policy
frameworks; and financial institutions’ sound balance sheets, high
capital, profitability and abundant liquid assets. However, banks
consider this source of stress by means of assumptions about col-
lateral or margin calls.
The vast majority of banks consider prospective approaches. Most
banks forecast their cashflows by means of assumptions about the
impact on inflows and outflows. However, it is not clear how these
assumptions are made. It seems to be based on expert judgement.
Concerning scenario revisions, all respondents reported that li-
quidity stress scenarios are revised, with 19 banking conducting
revisions regularly. From these banks, eight review their scenarios
annually and six do it monthly. The events that trigger adjustments
in the stress scenarios include, in no particular order of impor-
tance, changes in the macroeconomic scenario; changes in policies,
guidelines and practices at the group level; changes in regulations;
changes in monetary policy; business developments; changes in the
levels of delinquency; changes in markets. From the 27 banks, 23
need either approval from an asset-and-liability committee, a risk
committee or a board of directors for significant adjustments to the
liquidity stress test scenarios. Regarding the stress-test level, eleven
banks perform them at the entity level, while eight perform stress
tests at the group level. However, only six banks perform stress
tests at both levels. Five banks perform liquidity stress tests at other
levels, such as the currency level.

186

stress_testing.indd 186 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Figure 11.17 Measurement approach to liquidity stress tests

Number of banks

30

25

20

15

10

1 2 3 4 5 6 7 8

No Yes

1 - Cash flow maturity mismatch

2 - Liquidity stock approach
3 - Balance-sheet maturity mismatch
4 - Mixture of cash-flow and liquidity-stock approach
5 - Other cash-flow analysis
6 - Liquidity coverage ratio
7 - Current liquidity ratio
8 - Other

Source: BCB survey

Banks use more than one approach to quantify their liquidity risk
exposure (see Figure 11.17). According to the survey, the most com-
mon type of measurement approach (23 banks) is the cashflow ma-
turity mismatch, followed by other cashflows analyses (18). The
main advantages of the cashflow maturity mismatch seem to be
that it is transparent, flexible and simple and gives a general over-
view of risk (ECB 2008). Matz and Neu (2007) argue that measures
built on maturity mismatch and cashflow modelling help to reflect
the dynamic nature of liquidity. The main disadvantage is that it is
considered to be a short-term tool that does not reveal long-term
liquidity problems (ECB 2008).
In summary, the respondent banks consider liquidity risk an
important source of risk. The importance attributed to it increased

187

stress_testing.indd 187 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

after 2008, which can be related to the start of performing liquidity

stress tests by the majority of institutions. The time horizon and
sources of stress considered reflect the threats against which the
individual institutions want to be protected, which depend on the
particularities of their business.
Banks are reluctant to disclose the results of their stress tests, do-
ing so, on demand, mainly to rating agencies and supervisors. Only
a few banks frequently disclose their stress-test results to auditing
firms, committees and boards. What possible reasons do banks give
for this reluctance? Although most banks agree that disclosure would
enhance market discipline in liquidity risk management and see val-
ue added in disclosing the results of liquidity stress tests, all banks
agree strongly (19) or agree (8) that the results of liquidity stress tests
cannot be interpreted without a detailed understanding of the sce-
narios and the considerations underlying them (Figure 11.18).

Figure 11.18 Disclosure policy of stress testing

Number of banks
30
25
20
15
10
5
0
A B C D
Strongly agree Agree Disagree Strongly disagree

A - Results cannot be interpreted without detailed understanding of the scenarios

and the considerations underlying them
B - Lack of comparability across banks
C- Our bank does not see value added in disclosing liquidity stress test results
D - Disclosure would not enhance market discipline

Source: BCB survey

CONCLUSIONS
This chapter discusses the effects of the financial crisis on the Bra-
zilian banking system. The financial crisis has had major impacts
worldwide, and liquidity risks have risen accordingly. There was
an urgent need for macroprudential measures to help banking sys-

188

stress_testing.indd 188 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

tems regain confidence and increase their liquidity to cope with ad-
ditional risks.
We have presented the liquidity stress-testing approach in use
in the Central Bank of Brazil and the results of a survey on liquid-
ity stress testing that has been applied to banks that operate in the
Brazilian banking system.
Overall, the Brazilian Banking system has experienced a small
impact from the financial crisis due to several macroprudential
measures and strong bank supervision and regulation. This impact
affects banks differently. Medium-sized banks experienced a strong
liquidity constraint due to a fly-to-quality movement in time de-
posits. Regarding ownership, foreign banks were the most affected
by contagion. To avoid a confidence crisis, the BCB took measures
both in domestic and foreign currencies that helped banks to over-
come liquidity problems.
The survey applied to the largest Brazilian banks showed that li-
quidity risk is the second most important risk in their risk manage-
ment. It seems that the crisis led to an improvement in the banks’
risk management, since most of them started to perform liquidity
stress tests after the period of turbulences starting in 2007. There is
a considerable diversity in liquidity stress-test scenarios. However,
most banks use a combination of adverse market conditions and
tests of idiosyncratic-shock scenarios. The findings show that banks
do not rely on any single measure of liquidity but they have a pref-
erence for measurements related to cashflows.

Benjamin M. Tabak gratefully acknowledges financial support from CNPQ

Foundation. The opinions expressed in this chapter are those of the authors
and do not necessarily reflect those of the Central Bank of Brazil.

REFERENCES
Aikman, D., et al, 2009, “Funding liquidity risk in a quantitative model of systemic
stability”, Bank of England Working Papers 372, Bank of England.

BCB, 2007, “Financial Stability Report”, Banco Central do Brasil 6(2).

BCB, 2008, “Financial Stability Report”, Banco Central do Brasil 7(2).

189

stress_testing.indd 189 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

BCB, 2009, “Financial Stability Report”, Banco Central do Brasil 8(1).

BCB, 2010, “Financial Stability Report”, Banco Central do Brasil 9(2).

BCB, 2011, “Financial Stability Report”, Banco Central do Brasil 10(2).

BCB, 2012, “Financial Stability Report”, Banco Central do Brasil 11(1).

BCBS, 2010, “Basel III: International framework for liquidity risk measurement,
standards and monitoring”, Bank for International Settlements, December.

BCBS, 2011, “Macroprudential Policy Tools and Frameworks” Bank for Interna-
tional Settlements, October.

Čihák, Martin, 2007, “Introduction to Applied Stress-testing”, IMF Working Papers

07/59, International Monetary Fund.

Da Costa, F. N., 2012, “Brasil dos Bancos”, University of São Paulo Press, São
Paulo, Brazil.

ECB, 2008, “EU banks’ liquidity stress-testing and contingency funding plans”
([Link]

Forbes, K. J., and R. Rigobon, 2002, “No contagion, only interdependence: Measur-
ing stock market comovements, Journal of Finance 57(5), pp. 2223–61.

Fry, R., V. L. Martin and C. Tang, 2008, “A new class of tests of contagion with appli-
cations to real estate markets” CAMA Working Papers 2008-01, Australian National
University, Centre for Applied Macroeconomic Analysis.

IMF, 2011, “Macroprudential Policy: An Organizing Framework”, background pa-

per, March.

IMF, 2012, “Brazil, Financial System Stability Assessment”, June 20.

Matz, L., and P. Neu, 2007, “Liquidity Risk Measurement and Management: A Practi-
tioner’s Guide to Global Best Practices” (Wiley Finance Series, Wiley & Sons (Asia)).

Mesquita, M., and M. Torós, 2010, “Considerações sobre a Atuação do Banco Cen-
tral na Crise de 2008”, Banco Central do Brasil, Working Paper Series, March.

Ong, Li L., and Martin Čihák, 2010, “Of Runes and Sagas: Perspectives on Liquid-
ity Stress-testing Using an Iceland Example,” IMF Working Papers 10/156, Interna-
tional Monetary Fund.

Silva, L. A. P., and R. E. Harris, 2012, “Sailing through the Global Financial Storm:
Brazil’s recent experience with monetary and macroprudential policies to lean
against the financial cycle and deal with systemic risks“, Banco Central do Brasil,
Working Paper Series, August.

De Larosière Group, 2009, “The de Larosière Report”, February.

Van den End, J. W., 2009, “Liquidity Stress-Tester: A Model for Stress-testing Banks’
Liquidity Risk,” CESifo Economic Studies 56(1), pp. 38–69, April.

190

stress_testing.indd 190 20/05/2013 18:45

 Liquidity Risk: The Case of the Brazilian Banking System

Van den End, J. W., 2010, “Liquidity Stress-Tester: Do Basel III and Unconventional
Monetary Policy Work?”, DNB Working Papers 269, Netherlands Central Bank Re-
search Department.

Van den End, J. W., and M. Kruidhof, 2012, “Modelling the liquidity ratio as
macroprudential instrument”, DNB Working Papers 342, Netherlands Central
Bank, Research Department.

Vazquez, F., B. M. Tabak and M. Souto, 2012, “A macro stress test model of credit
risk for the Brazilian banking sector”, Journal of Financial Stability, 8, pp. 69-83.

Wong, E., and C. Hui, 2009, “A Liquidity Risk Stress-Testing Framework with In-
teraction between Market and Credit Risks,” Working Papers 0906, Hong Kong
Monetary Authority.

1 An analysis of the distribution of these deposits among its holders shows that, in 2006,
54.1% of the total amount was concentrated at the level of up to BRL100 per account, with
41,565,238 depositors (BCB 2007).
2 For more details about the measures taken by Brazilian authorities, see Mesquita and Torós
(2010) and Silva and Harris (2012).
3 Adapted from “a significant increase in cross-market linkages after a shock to one country
(or group of countries)”, by Forbes and Rigobon (2002).

191

stress_testing.indd 191 20/05/2013 18:45

stress_testing.indd 192 20/05/2013 18:45
 12
Determining the Severity of
Macroeconomic Stress Scenarios

Kapo Yuen
Federal Reserve Bank of New York

In the midst of the 2008 financial panic caused by the collapse of

the subprime housing market, the US government responded with
unprecedented measures, including liquidity provision through
various funding programmes, debt and deposit guarantees and
large-scale asset purchases. In February 2009, the US banking su-
pervisors conducted the first-ever system-wide stress test on 19
of the largest US bank holding companies (BHCs), known as the
Supervisory Capital Assessment Program (SCAP) (Federal Reserve
2009a). The stress test required these 19 BHCs to undergo simulta-
neous, forward-looking exercises designed to determine whether
they would have adequate capital to sustain lending to the econo-
my in the event of an unexpectedly adverse scenario. By conduct-
ing this SCAP exercise, the supervisors hoped that it would reduce
uncertainty and restore confidence in the US financial institutions.
In their 2010 staff reports, Peristian, Morgan and Savino (2010), of
the Federal Reserve Bank of New York, concluded that the SCAP
might have helped to quell the financial panic by releasing vital in-
formation about the BHCs. They claimed, “While investors did not
need supervisors to tell them which banks had capital deficiencies,
they were surprised by the size of the capital gaps and they used
that information to revalue banks.”
Since conducting the SCAP in 2009, the Federal Reserve Sys-
tem has conducted annual stress tests on the US banking system,
called the Comprehensive Capital Analysis and Review (CCAR).

193

stress_testing.indd 193 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Additionally, the Dodd–Frank Wall Street Reform and Consumer

Protection Act requires the Federal Reserve Board to conduct an-
nual stress tests of bank holding companies with total consolidat-
ed assets of US$50 billion or more. In each CCAR, the Federal Re-
serve Board generates an adverse macroeconomic scenario (two
adverse scenarios in 2013) and requires BHCs to submit at least
one adverse scenario that is related to their own specific portfolios
and risk profiles. While the 2013 instructions for CCAR indicate
that the adverse scenario developed by the BHC must reflect “a
severely adverse economic and financial market environment”, it
does not specify what should be the “appropriate severity” of an
adverse scenario used for the capital planning. This chapter will
discuss the severity of the supervisory adverse scenarios and pro-
vide a simple methodology to compare the severity of different
adverse macroeconomic scenarios. In particular, the key questions
we aim to answer are:

o how to measure the severity of a firm’s BHC macro stress scenario;

o how severe the BHC stress scenario is when compared with the
supervisory stress scenarios; and
o what implications can be drawn about the credibility of the
BHC’s stress scenario.

THE US SUPERVISORY STRESS SCENARIOS

As part of the Federal Reserve System’s Comprehensive Capital
Analysis and Review (CCAR), US domiciled top-tier bank holding
companies (BHCs) are required to submit comprehensive capital
plans, including pro forma capital analyses, based on at least one
BHC defined adverse scenario. The adverse scenario is described
by quarterly trajectories for key macroeconomic variables (MVs)
over the next nine quarters (or longer, as in 2013). In addition,
the Federal Reserve generates its own supervisory stress scenari-
os so that firms are expected to apply both BHC and supervisory
stress scenarios to all exposures to estimate potential losses under
stressed operating conditions. Separately, firms with significant
trading activity are asked to estimate a one-time potential trading-
related market and counterparty credit losses under their own BHC
scenarios and market stress scenarios provided by the supervisors.1

194

stress_testing.indd 194 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

For the supervisory stress scenarios, the Federal Reserve provides

firms with global market shock components that are one-time,
hypothetical shocks to a large set of risk factors. For the last two
CCAR exercises, these shocks involved large and sudden changes
in asset prices, rates and CDS spreads that mirrored the severe mar-
ket condition in the second half of 2008.
Since CCAR is a comprehensive assessment of a firm’s capi-
tal plan, the BHCs are asked to conduct an assessment of the ex-
pected uses and sources of capital over a planning horizon. In
the 2009 SCAP, firms were asked to submit stress losses over the
next two years, on a yearly basis. Since then, the planning horizon
has changed to nine quarters. For the last three CCAR exercises,
a BHC is asked to submit its pro forma, post-stress capital pro-
jections in its capital plan beginning with data as of September
30, and spans the nine-quarter planning horizon. The projections
begin in the fourth quarter of the current year and conclude at the
end of the fourth quarter two years forward. Hence, for defining
BHC stress scenarios, firms are asked to project the movements of
key macroeconomic variables over the planning horizon of nine
quarters. Our analysis on determining the severity of macro stress
scenarios is based on the movements of the macroeconomic vari-
ables in these nine quarters. As for determining the severity of
the global market shock components for trading and counterparty
credit losses, it will not be discussed in this chapter because it is
a one-time shock and the evaluation will be on the movements of
the market risk factors rather the macroeconomic variables. This
is examined in Chapter 3.
In the 2011 CCAR, the Federal Reserve defined the stress super-
visory scenario using nine macroeconomic variables: Real GDP,
Consumer Price Index (CPI), Real Disposable Personal Income,
Unemployment Rate, Three-Month Treasury Bill Rate, 10-Year
Treasury Bond Rate, BBB Corporate Rate, Dow Jones Index and
National House Price Index (Federal Reserve 2011a). In CCAR
2012, the number of macroeconomic variables that defined the
supervisory stress scenario increased to 14. Besides the original
nine variables, the added variables were Nominal GDP Growth,
Nominal Disposable Income Growth, Mortgage Rate, Market
Volatility Index and Commercial Real Estate Price Index (Federal

195

stress_testing.indd 195 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Reserve 2011b) Additionally, there is another set of 12 internation-

al macroeconomic variables, three macroeconomic variables and
four countries/country blocks, included in the supervisory stress
scenario. As for CCAR 2013, the Federal Reserve System uses the
same set of variables to define the supervisory adverse scenario
(Federal Reserve 2012) as in 2012. Since the BHCs are required
to define their own adverse scenarios of “a severely adverse eco-
nomic environment”, one way to determine the “appropriate se-
verity” of the BHCs’ stress scenarios is to compare them with the
supervisory adverse scenarios. Although it is stated that the BHC
stress scenarios should reflect the BHC’s unique vulnerabilities to
factors that affect its exposures, activities and risks, by compar-
ing the severity of the BHC’s and supervisory stress scenarios we
can determine whether the estimates of the losses are consistent
with the relative severity of the stress scenarios. For example, if
the BHC’s own stress scenario is determined to be more severe
than the supervisory scenario, but the estimated losses from the
supervisory scenario are larger, then we would need to examine
the details of the stress loss estimation methodology to determine
the causes of this inconsistency.

ALIGNING THE US SUPERVISORY STRESS SCENARIOS

Let us consider the two CCAR supervisory stress scenarios in 2011
and 2012 and the supervisory severely adverse scenario in 2013.
By comparing the forecast of the macroeconomic variables over the
next nine quarters, we will try to determine which scenarios are the
most and least severe.
In general, comparing severity of stress scenarios is relative. That
is, we can usually deduce with relative confidence that Scenario A
is more severe than Scenario B, but it is much more challenging to
quantify how much more severe is Scenario A over Scenario B. In
other words, it is much more difficult to define a metric to measure
the severity of a stress scenario. Later in this chapter, we will at-
tempt to use a historical event as a reference to give some sights on
the measurement of severity.

196

stress_testing.indd 196 20/05/2013 18:45

stress_testing.indd 197

Table 12.1 The Federal Reserve supervisory adverse scenarios with nine quarters of projections
CCAR 2011 Common Variables Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012
Real GDP 13,261 13,332 13,393 13,255 13,206 13,138 13,178 13,229 13,343 13,453
Real Disposable Personal Income 10,237 10,271 10,299 10,318 10,236 10,179 10,081 10,054 10,047 10,066
Unemployment Rate 9.6 9.6 9.6 10.1 10.6 11.0 11.1 11.0 10.9 10.6
CPI 218.0 219.0 219.9 220.9 221.7 222.3 222.9 223.4 224.0 224.7
3-Month Treasury Yield 0.16 0.16 0.19 0.07 0.13 0.13 0.13 0.13 0.13 0.13

Determining the Severity of Macroeconomic Stress Scenarios

10-Year Treasury Yield 2.90 2.57 2.64 2.66 2.79 2.77 2.71 2.98 3.12 3.35
BBB corporate yield 5.07 4.69 4.86 5.88 6.26 6.46 6.16 6.27 6.22 6.25
Dow Jones Total Stock Market Index 11,947 12,069 11,822 9,116 8,809 8,716 10,682 11,083 11,498 11,930
National House Price Index 142 140 139 137 134 132 130 128 127 126

CCAR 2012 Common Variables Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012 Q1 2013 Q2 2013 Q3 2013 Q4 2013
Real GDP growth 2.46 -4.84 -7.98 -4.23 -3.51 0.00 0.72 2.21 2.32 3.45
Real Disposable Personal Income -1.73 -6.02 -6.81 -4.29 -3.16 -0.57 0.74 1.66 2.69 2.27
growth
Unemployment Rate 9.09 9.68 10.58 11.40 12.16 12.76 13.00 13.05 12.96 12.76
CPI inflation rate 3.09 2.21 1.78 1.02 0.89 0.35 0.23 0.21 0.30 0.32
3-Month Treasury Yield 0.02 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10
197
20/05/2013 18:45
stress_testing.indd 198

198

Stress Testing: Approaches, Methods and Applications

Table 12.1 (continued)
10-Year Treasury Yield 2.48 2.07 1.94 1.76 1.67 1.76 1.74 1.84 1.98 1.98
BBB corporate yield 4.87 5.65 6.83 6.81 6.75 6.45 6.07 5.83 5.74 5.51
Dow Jones Total Stock Market Index 11,771.86 9,501.48 7,576.38 7,089.87 5,705.55 5,668.34 6,082.47 6,384.32 7,084.65 7,618.89
National House Price Index 136.86 135.13 131.61 127.50 123.12 119.08 115.15 111.92 109.77 108.48

CCAR 2013 Common Variables Q3 2012 Q4 2012 Q1 2013 Q2 2013 Q3 2013 Q4 2013 Q1 2014 Q2 2014 Q3 2014 Q4 2014
Real GDP growth 2.0 -3.5 -6.1 -4.4 -4.2 -1.2 0.0 2.2 2.6 3.8
Real Disposable Personal Income 0.8 -3.8 -6.7 -4.6 -3.2 -1.5 0.8 0.9 2.5 2.8
growth
Unemployment Rate 8.1 8.9 10.0 10.7 11.5 11.9 12.0 12.1 12.0 11.9
CPI inflation rate 2.3 1.8 1.4 1.1 1.0 0.3 1.0 0.9 0.7 0.6
3-Month Treasury Yield 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
10-Year Treasury Yield 1.6 1.4 1.2 1.2 1.2 1.2 1.2 1.5 1.7 1.9
BBB corporate yield 4.2 5.6 6.4 6.7 6.8 6.5 6.2 6.2 6.0 5.9
Dow Jones Total Stock Market Index 14997.8 12105.2 9652.6 9032.8 7269.1 7221.7 7749.3 8133.9 9026.1 9706.7
National House Price Index 143.4 141.6 137.9 133.6 129.0 124.7 120.6 117.2 115.0 113.6
20/05/2013 18:45
 Determining the Severity of Macroeconomic Stress Scenarios

Before we begin to examine the three supervisory scenarios, we

have to make the following key assumptions.

o All scenarios start on the same quarter, Q3 2010, and the projec-
tions are over the nine quarters from Q4 2010 to Q4 2012. Hence
the scenarios are compared by measuring the change over the
nine quarters on the macroeconomic variables.
o The scenarios are compared on the set of common macroeco-
nomic variables. For example, in 2012 and 2013, the Market
Volatility Index is included in the supervisory stress scenario,
but not in 2011, thus this variable is excluded. Therefore, the
set of common macroeconomic variables for comparison are
all the variables that are defined in the supervisory scenario of
2011 CCAR.

Of the nine macroeconomic variables, Real GDP, CPI, and Real

Disposable Personal Income are expressed as growth rates in 2012
and 2013 (see Table 12.1). To convert them back to Real GDP, CPI
and Real Disposable Personal Income, we use Equation 12.1 to
convert growth rates into actual values, and align all the start-
ing values using the Q3 2010 actual values. For variables such as
Unemployment Rate (UR), we first align all the starting values
to be Q3 2010, then we use the percentage change over the pe-
riod (Equation 12.2) to convert the 2012 and 2013 projections to
the projections with the same starting values. For example, the
Real GDP Growth Rate (Real_GDPGR) is converted to Real GDP
(Real_GDP) by the following equation:

_ 0.25
+1
_ +1 = _ ∗{ 1+ }
100 (12.1)

For other variables, the alignment is just the percentage change

over the period,
2012 +1 − 2012
_2011 +1 = _2011 1+
2012 (12.2)

199

stress_testing.indd 199 20/05/2013 18:45

stress_testing.indd 200

200

Stress Testing: Approaches, Methods and Applications

Table 12.2 The macroeconomic variables with nine quarters of projections on the four scenarios
Scenario CCAR Common Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
Variables 2010 2010 2011 2011 2011 2011 2012 2012 2012 2012
2011 Real GDP 13,261 13,332 13,393 13,255 13,206 13,138 13,178 13,229 13,343 13,453
2012 Real GDP 13,261 13,097 12,828 12,690 12,577 12,577 12,599 12,668 12,741 12,849
2013 Real GDP 13,261 13,143 12,938 12,793 12,657 12,619 12,619 12,687 12,769 12,889
Hypothetical Real GDP 13,261 12,690 12,690 12,690 12,690 12,690 12,690 12,690 12,690 12,690
2011 Real Disposable 10,237 10,271 10,299 10,318 10,236 10,179 10,081 10,054 10,047 10,066
Personal Income
2012 Real Disposable 10,237 10,079 9,903 9,795 9,717 9,703 9,721 9,761 9,826 9,881
Personal Income
2013 Real Disposable 10,237 10,138 9,964 9,847 9,768 9,731 9,750 9,772 9,832 9,901
Personal Income
Hypothetical Real Disposable 10,237 9,700 9,700 9,700 9,700 9,700 9,700 9,700 9,700 9,700
Personal Income
2011 Unemployment Rate 9.58 9.61 9.62 10.11 10.56 11.02 11.12 11.04 10.87 10.64
2012 Unemployment Rate 9.58 10.20 11.15 12.02 12.82 13.45 13.70 13.75 13.66 13.45
2013 Unemployment Rate 9.58 10.53 11.83 12.66 13.60 14.07 14.19 14.31 14.19 14.07
Hypothetical Unemployment Rate 9.58 12.66 12.66 12.66 12.66 12.66 12.66 12.66 12.66 12.66
2011 10-Year Treasury Yield 2.90 2.57 2.64 2.66 2.79 2.77 2.71 2.98 3.12 3.35
2012 10-Year Treasury Yield 2.90 2.42 2.27 2.05 1.95 2.05 2.04 2.15 2.31 2.32
20/05/2013 18:45
stress_testing.indd 201

Table 12.2 (continued)

2013 10-Year Treasury Yield 2.90 2.54 2.18 2.18 2.18 2.18 2.18 2.72 3.08 3.44
Hypothetical 10-Year Treasury Yield 2.90 2.80 2.80 2.80 2.80 2.80 2.80 2.80 2.80 2.80
2011 3-Month Treasury Yield 0.16 0.16 0.19 0.07 0.13 0.13 0.13 0.13 0.13 0.13
2012 3-Month Treasury Yield 0.16 0.69 0.69 0.69 0.69 0.69 0.69 0.69 0.69 0.69
2013 3-Month Treasury Yield 0.16 0.16 0.16 0.16 0.16 0.16 0.16 0.16 0.16 0.16
Hypothetical 3-month Treasury yield 0.16 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50 0.50

Determining the Severity of Macroeconomic Stress Scenarios

2011 BBB corporate yield 5.07 4.69 4.86 5.88 6.26 6.46 6.16 6.27 6.22 6.25
2012 BBB corporate yield 5.07 5.87 7.11 7.09 7.03 6.71 6.31 6.07 5.97 5.73
2013 BBB corporate yield 5.07 6.76 7.73 8.09 8.21 7.85 7.48 7.48 7.24 7.12
Hypothetical BBB corporate yield 5.07 8.09 8.09 8.09 8.09 8.09 8.09 8.09 8.09 8.09
2011 CPI 218.0 219.0 219.9 220.9 221.7 222.3 222.9 223.4 224.0 224.7
2012 CPI 218.0 219.2 220.2 220.8 221.3 221.4 221.6 221.7 221.9 222.0
2013 CPI 218.0 219.0 219.8 220.4 220.9 221.1 221.6 222.1 222.5 222.9
Hypothetical CPI 218.0 222.0 222.0 222.0 222.0 222.0 222.0 222.0 222.0 222.0
2011 Dow Jones Total Stock 11,947 12,069 11,822 9,116 8,809 8,716 10,682 11,083 11,498 11,930
Market Index
2012 Dow Jones Total Stock 11,947 9,643 7,689 7,195 5,791 5,753 6,173 6,479 7,190 7,732
Market Index
2013 Dow Jones Total Stock 11,947 9,643 7,689 7,195 5,791 5,753 6,173 6,479 7,190 7,732
Market Index
201
20/05/2013 18:45
stress_testing.indd 202

202

Stress Testing: Approaches, Methods and Applications

Table 12.2 (continued)
Hypothetical Dow Jones Total Stock 11,947 7,200 7,200 7,200 7,200 7,200 7,200 7,200 7,200 7,200
Market Index
2011 National House Price 142.4 140.4 139.1 136.8 134.3 131.8 129.6 127.8 126.8 126.4
Index
2012 National House Price 142.4 140.6 136.9 132.6 128.1 123.9 119.8 116.4 114.2 112.8
Index
2013 National House Price 142.4 140.6 136.9 132.6 128.1 123.8 119.7 116.3 114.2 112.8
Index
Hypothetical National House Price 142.4 120.0 120.0 120.0 120.0 120.0 120.0 120.0 120.0 120.0
Index
20/05/2013 18:45
 Determining the Severity of Macroeconomic Stress Scenarios

After aligning the supervisory stress scenarios from CCAR 2011,

2012 and 2013, we also create an additional hypothetical stress sce-
nario to illustrate how the severity of stress scenarios can be com-
pared between the supervisory scenarios and a “BHC-developed”
stress scenario, as part of the BHC’s requirements under CCAR. Ta-
ble 12.2 shows the three supervisory and the hypothetical scenarios
as they are all aligned at Q3 2010. All the scenarios have the same
set of macroeconomic variables and they are all aligned to the same
value as of Q3 2010. Hence, by comparing the changes of the mac-
roeconomic variables over the next nine quarters, we can examine
the severity of each scenario relative the others.

HISTORICAL TREND OF THE MACROECONOMIC VARIABLES

In generating the supervisory adverse scenarios, the Federal Re-
serve emphasises that the scenarios are not economic forecasts,
but rather hypothetical scenarios that show significant contrac-
tion in economic activities. A contraction in economic activities
means macroeconomic indicators such as GDP, employment, stock
indexes, investment spending, capacity utilisation, household in-
come, housing prices and inflation fall, while the unemployment
rate and personal and corporate bankruptcies rise. Of the nine com-
mon macroeconomic variables, the one that is most related to stress
economic conditions is a drop in Real GDP. As for the rest of the
common variables, most economists would agree that a stress eco-
nomic condition will associate with a decrease in Real Disposable
Personal Income, Dow Jones Index and House Price Index, and an
increase in Unemployment Rate. However, for CPI, BBB Corporate
Bond Rate, Three-Month Treasury Yield and the 10-Year Treasury
Yield, it is not so clear that an increase or decrease in any one of
these variables will indicate a stress economic condition.
We will use historical data to examine each of the common vari-
ables to understand its relationship with historical recessions. All
the historical values are obtained from the Board of Governors’
document on Supervisory Scenarios2.
In the US, from 1980 to 2013, there have been nine periods of
negative economic growth over one fiscal quarter or more (see Fig-
ure 12.1). According to the National Bureau of Economic Research
(NBER), there have been five periods considered recessions:3

203

stress_testing.indd 203 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

o January 1980–July 1980: 6 months;

o July 1981–November 1982: 16 months;
o July 1990–March 1991: 8 months;
o March 2001–November 2001: 8 months; and
o December 2007–June 2009: 18 months.

Figure 12.1 US Real GDP growth, 1980–2012

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007
-5.00

-10.00
US real GDP growth

Figure 12.2 US Real GDP growth and income, 1980–2012

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007

-5.00

-10.00
US real GDP growth Disposable Income

204

stress_testing.indd 204 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

From Figure 12.2, we can see that a drop in GDP Growth Rate is usually
associated with a drop in the Real Disposable Personal Income Growth
Rate. Thus, we can state, in general, a stress economic condition is as-
sociated with a drop in Real Disposable Personal Income Growth.

Figure 12.3 US Real GDP growth and Dow Jones Index

20.00

15.00

10.00

5.00
%

0.00
Q1 1980
Q1 1981
Q1 1982
Q1 1983
Q1 1984
Q1 1985
Q1 1986
Q1 1987
Q1 1988
Q1 1989
Q1 1990
Q1 1991
Q1 1992
Q1 1993
Q1 1994
Q1 1995
Q1 1996
Q1 1997
Q1 1998
Q1 1999
Q1 2000
Q1 2001
Q1 2002
Q1 2003
Q1 2004
Q1 2005
Q1 2006
Q1 2007
Q1 2008
Q1 2009
Q1 2010
Q1 2011
Q1 2012
-5.00

-10.00

-15.00
US real GDP growth Dow Jones Index (,000)

Figure 12.4 US Real GDP growth and House Price Index

25.00

20.00

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007

-5.00

-10.00

-15.00
US real GDP growth House Price Index

205

stress_testing.indd 205 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

From Figure 12.3, we can see that the last two recessions are associ-
ated with significant drops in the Dow Jones Index, and from Fig-
ure 12.4 we see that the most severe recession followed a huge drop
in the House Price Index. Hence, we can also confidently claim that
a stress economic condition is associated with a drop in Dow Jones
Index or House Price Index.

Figure 12.5 US Real GDP growth and unemployment rate

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007
-5.00

-10.00
US real GDP growth Unemployment Rate

Figure 12.6 US Real GDP growth and BBB bond rate

20.00

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007

-5.00

-10.00

-15.00
US real GDP growth BBB Corporate Bond Rate

206

stress_testing.indd 206 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

Figure 12.5 clearly shows that each recession was associated with a rise
in the unemployment rate. However, for BBB Corporate Bond Rate –
although for the recessions in 1981 and 2008 we see a high increase in
the BBB Rate – the rest of the data does not show a high association
between GDP Growth Rate and BBB Bond Rate (Figure 12.6).

Figure 12.7 US Real GDP growth and CPI percentage change

20.00

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007
-5.00

-10.00

-15.00
US real GDP growth CPI Inflation Rate

Figure 12.8 US Real GDP Growth and 3 months T bill

20.00

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007

-5.00

-10.00

-15.00
US real GDP growth 3MoTBill

207

stress_testing.indd 207 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

From Figure 12.7, we see only that a sharp drop in CPI is associated
with the last recession, but, for the rest of the historical data, there is
no obvious association between change in CPI and GDP Growth Rate.
As for the Three-Month Treasury Yield and the US 10-Year Trea-
sury Yield, from Figures 12.8 and 12.9, there is no clear association
between Real GDP Growth Rate and Treasury Yields. From examin-
ing the historical data, of the nine common macroeconomic variables
defined in the stress scenarios we have seen that four of them – CPI,
Three-Month Treasury Yield, 10-Year Treasury Yield and BBB Corpo-
rate Bond Rate – do not show any “directional” indication that either
an increase or decrease in value is necessarily associated with a stress
economic condition (“directional” means that, if a macroeconomic
variable increases in value, then, historically, the economic condition
always reacts the same way, either less or more severe in the same
direction). Thus, these four variables will not be further considered
in our discussion for measuring the severity of a stress scenario.

Figure 12.9 US Real GDP Growth and US 10-Year Treasury Yield

20.00

15.00

10.00

5.00
%

0.00
Q4 1980
Q3 1981
Q2 1982

Q4 1983
Q3 1984
Q2 1985

Q4 1986

Q2 1988

Q4 1989
Q3 1990
Q2 1991

Q4 1992
Q3 1993
Q2 1994

Q4 1995
Q3 1996

Q4 1998
Q3 1999
Q2 2000

Q4 2001
Q3 2002
Q2 2003

Q4 2004
Q3 2005
Q2 2006

Q3 2008
Q2 2009

Q4 2010
Q3 2011
Q2 2012
Q3 1987

Q2 1997

Q4 2007

-5.00

-10.00

-15.00
US real GDP growth US 10 Years Treasury

NINE-QUARTERS PROJECTIONS OF THE MACROECONOMIC

VARIABLES
We will now examine the projections of the remaining five macro-
economic variables in each of the four stress scenarios. From Figures
12.10 and 12.11, we can clearly see that the 2011 supervisory scenario

208

stress_testing.indd 208 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

is the least severe in terms of GDP and Real Disposable Personal In-
come. The 2013 scenario follows the same pattern as the 2012 sce-
nario, but the drop at every quarter is only slightly less than that of
2012. As for the Hypothetical portfolio, it has the biggest drop in dis-
posable income. However, the severity in Real GDP is not conclusive
because the projected GDP in the Hypothetical are higher than sce-
narios 2012 and 2013 in some quarters, but lower in other quarters.

Figure 12.10 Projections of the four scenarios – Real GDP

13,600

13,400

13,200

13,000
Real GDP

12,800

12,600

12,400

12,200

12,000
Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012

Scenario 2011 Scenario 2012 Scenario 2013 Hypothetical scenario

Figure 12.11 Projections – Real Disposable Personal Income

10,400

10,300

10,200

10,100
Real Disposable Income

10,000

9,900

9,800

9,700

9,600

9,500

9,400
Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012

Scenario 2011 Scenario 2012 Scenario 2013 Hypothetical scenario

209

stress_testing.indd 209 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Figure 12.12 Projections – Dow Jones Index

14,000

12,000
Dow Jones Index

10,000

8,000

6,000

4,000

2,000
Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012

Scenario 2011 Scenario 2012 Scenario 2013 Hypothetical scenario

Figure 12.13 Projections – House Price Index

150.0

140.0
House Price Index

130.0

120.0

110.0

100.0
Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012

Scenario 2011 Scenario 2012 Scenario 2013 Hypothetical scenario

We now focus on the next two “directional” macroeconomic vari-

ables: Dow Jones Index and House Price Index. From Figures 12.12
and 12.13, we can see that scenarios 2012 and 2013 are exactly the
same. The three supervisory scenarios follow similar patterns. On
the Dow Jones Index, there is a sharp drop in the beginning, followed
by an increase after the fourth quarter, and scenario 2011 shows that

210

stress_testing.indd 210 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

the drop is the mildest. As for the House Price Index, the projections
are all declining quarter after quarter, and the decline in scenario
2011 is also the mildest. Since the Hypothetical cut across the curve
of scenarios 2012 and 2013, by examining the charts, apart from sce-
nario 2011, it is not conclusive which one is the most severe scenario.
Finally, on examining the Unemployment Rate (Figure 12.14), we
see that the three supervisory scenarios all project that the unem-
ployment rate will go up over the next six quarters, then plateau,
and finally drop slightly at the end. It is also obvious that scenario
2011 is the mildest, and it is inconclusive between the Hypothetical
scenario and scenarios 2012 and 2013. By now, we have established
that for the three supervisory scenarios – since each of the variables
we have seen follows the same pattern, on the whole – scenario
2011 is the mildest, and scenario 2012 and 2013 are almost the same
except that scenario 2012 is slightly more severe.

Figure 12.14 Projections – Unemployment Rate

15.00

14.00

13.00

12.00
Unemployment Rate

11.00

10.00

9.00

8.00

7.00

6.00
Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012

Scenario 2011 Scenario 2012 Scenario 2013 Hypothetical scenario

When comparing the Hypothetical scenario with the 2012 and 2013
scenarios in Unemployment Rate, there are some quarters in which
the hypothetical scenario has the higher Unemployment Rate, and
some quarters in which the 2012 and 2013 scenarios have the higher
Unemployment Rate. Therefore, in order to make an overall com-
parison, it is necessary to develop a statistic to summarise, measure
and standardise the severity of each variable over the nine quarters.

211

stress_testing.indd 211 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Since we have aligned all the scenarios to the same starting point,
the severity of each quarter is determined by the percentage change
with respect to the starting quarter (Q3 2010). Thus, a probable sum-
mary statistic is the average of the percentage change with respect to
the starting quarter. Another choice is the maximum change over the
nine quarters, but the maximum value ignores the projections with
recovery at the end of the nine quarters. The average of the percent-
age change is computed by Equation 12.3 below, where i is the index
for the four scenarios, and j is the index for each of the projections.

(( ,3 2010 )
)
100 −
% ℎ = ∑9=1 ,
9 ,3 2010 (12.3)

This summary statistic can be computed for the five “directional”

macroeconomic variables in each scenario. We now have a measure
to compare the severity of the scenarios on each of the variables
independently of the other variables. From Figure 12.15, we can see
that the longer the vertical bar is away from zero, the more sever-
ity it indicates. Thus, for Unemployment Rate, the longest bar is
in 2013, which means the most severe scenario is 2013. For GDP,
Disposable Income, and HPI, the Hypothetical scenario is the most
severe. For Dow Jones Index, the 2012, 2013 and Hypothetical sce-
narios are more or less the same. Viewing across the chart, we see
that the 2011 scenario is the least severe in each of the five variables.

Figure 12.15 Average change in macroeconomic variables from Q3 2010

50.00%
Unemployment Rate
40.00%

30.00%

20.00%

10.00%
GDP Income Dow Jones Index HPI
0.00%
2011
2012
2013
Hypothetical
2011
2012
2013
Hypothetical
2011
2012
2013
Hypothetical
2011
2012
2013
Hypothetical
2011
2012
2013
Hypothetical

-10.00%

-20.00%

-30.00%

-40.00%

-50.00%

212

stress_testing.indd 212 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

In terms of the magnitude of severity, we see that Unemployment

Rate and Dow Jones Index have 30% to 40% average deterioration
in scenarios 2012, 2013 and the Hypothetical, whereas the other
variables are much less severe. Hence, we have a measurement of
severity for each macroeconomic variable. In the next section, we
will suggest some ways to combine all these measurement to give
an overall view of the severity of a stress scenario.

AN OVERALL ASSESSMENT OF THE SEVERITY

At the start of our discussion, we mentioned that measuring sever-
ity is, in general, relative. However, some economists (Edge 2012)
have tried to use historical recessions as anchors, and relate the
stress scenarios to historical recessions to get some sense of mag-
nitude as compared with past recessions. We will discuss other
methodologies in more detail in the next section. For now, we will
propose a simple methodology to summarise the severity of each
scenario with reference to a historical recession.
For each macroeconomic variable, we have four scenarios. We
will use an ordinal ranking to assign a rank to each of the scenar-
io by using the Average % Change, giving rankings of 1–4 to in-
dicate least to most severity. When the Average % Change shows
that two scenarios are very similar (within 1% of each other), we
assign the average of the two ranks to both scenarios. Since sever-
ity in Unemployment Rate is associated with increasing value,
the lowest rank is given to the lowest Average % Change. Table
12.3 gives the ranking of each variable and the total ranking. By
summing across the rank of each variable, we arrive at a statis-
tic that gives an overall ranking (Total Rank) of the scenarios.
With this simple methodology, we have determined the severity
of the four macro stress scenarios. In order of severity, the most
severe scenario is Hypothetical, followed by 2013, 2012 and 2011.
However, on further examination, the differences in Total Rank
between the Hypothetical, 2013 and 2012 scenarios are not that
significant (<20%), hence these three scenarios are comparable in
terms of severity.

213

stress_testing.indd 213 20/05/2013 18:45

stress_testing.indd 214

214

Stress Testing: Approaches, Methods and Applications

Table 12.3 Ranking of each macroeconomic variable and the total rank of the scenario
Real GDP Disposable Income Unemployment Rate Dow Jones HPI
Scenario Ave % RANK Ave % RANK Ave % RANK Ave % RANK Ave % RANK Total
Change Change Change Change Change Rank
2011 0.2 1 -0.6 1 9.7 1 -11.0 1 -6.88 1 5.0

2012 -4.0 3 -4.1 2.5 32.4 2.5 -40.8 3 -12.2 2.5 13.5

2013 -3.5 3 -3.7 2.5 38.5 4 -40.8 3 -12.2 2.5 15.0

Hypothetical -4.3 3 -5.2 4 32.2 2.5 -39.7 3 -15.7 4 16.5

Table 12.4 Nine-quarters values of changes for 2008 Recession4

Common Variables Q3 2010 Q1 2008 Q2 2008 Q3 2008 Q4 2008 Q1 2009 Q2 2009 Q3 2009 Q4 2009 Q1 2010

Real GDP 13,261 13,202 13,245 13,122 12,820 12,649 12,639 12,684 12,810 12,884

Real Disposable Personal 10,237 10,385 10,592 10,350 10,344 10,220 10,208 10,048 10,033 10,172
Income
Unemployment Rate 9.58 9.98 10.64 11.98 13.70 16.50 18.49 19.23 19.83 19.49

Dow Jones Index 11,947 10,748 10,540 9,574 7,326 6,541 7,598 8,797 9,269 9,804

House Price Index 142 137 131 126 120 114 112 113 114 115
20/05/2013 18:45
 Determining the Severity of Macroeconomic Stress Scenarios

Since all these four scenarios are hypothetical forward-looking sce-

narios, and we now have some understanding of the severity among
them, the next logical question should be how they are related to our
historical experience. Looking back at the past postwar recessions,
we see that the most severe recession is the 2008 recession (December
2007 to June 2009), which lasted 18 months. In order to align with the
nine projected quarters of our scenarios, we will use the NBER data
and choose the consecutive nine quarters of Real GDP Growth Rate
when the recession began in the first quarter of 2008. After aligning
for Q3 2010 as the starting quarter and the conversion of the values as
described in the last section on the remaining nine quarters, we have
the similar values on the macroeconomic variables for comparison in
Table 12.4. The first column has the actual values as in Q3 2010. This is
used as the anchor to align all the scenarios that have the same starting
values. The rest of the nine columns are from Q1 2008 to Q1 2010 (pro-
jections of nine quarters as in all supervisory scenarios). The values in
these nine columns are scaled to the starting values of the first column,
and so the 2008 Recession values start from the second column.
After we calculate the average percentage change of the five vari-
ables over the nine quarters, we find that the Average % Change on
Real Disposable Personal Income is slightly positive (0.2%). Thus,
during the 2008 Recession, the Real Disposable Personal Income is
not “directional”, as we once thought. We have come to realise that
it is not necessarily true that Real Disposable Personal Income will
decrease in a severe stress economic condition, especially when we
are looking at the overall change in nine quarters. Therefore, includ-
ing the Real Disposable Personal Income will taint our measurement
of severity. Hence we will drop the Real Disposable Personal Income
from our ranking on the overall severity. We can now put the Average
% Change of the rest of the variables back into the ranking methodol-
ogy and observe the ranking of the recession with other scenarios.
From Table 12.5, we observe that 2008 Recession is ranked as
the most severe, followed by Hypothetical, 2013, 2012 and 2011.
However, apart from the unemployment rate in 2008 Recession,
the differences among the four most stressful scenarios are not so
significant that they can be easily separated. In fact, for the Dow
Jones Index, 2008 Recession is the second least severe of the five
scenarios. Hence we conclude that the Hypothetical, 2013 and 2013
scenarios have similar severity to 2008 Recession.

215

stress_testing.indd 215 20/05/2013 18:45

stress_testing.indd 216

216

Stress Testing: Approaches, Methods and Applications

Table 12.5 Ranking of each macroeconomic variable and the total rank of the scenario and recession
Real GDP Unemployment Rate Dow Jones HPI

Scenario Ave % RANK Ave % RANK Ave % RANK Ave % RANK Total Rank
Change Change Change Change
2011 0.2 1 9.7 1 -11.0 1 -6.88 1 4.0

2012 -4.0 3.5 32.4 2.5 -40.8 4 -12.2 2.5 12.5

2013 -3.5 3.5 38.5 4 -40.8 4 -12.2 2.5 14.0

Hypothetical -4.3 3.5 32.2 2.5 -39.7 4 -15.7 4.5 14.5

08 Recession -2.8 3.5 62.2 5 -25.4 2 -15.5 4.5 15.0

20/05/2013 18:45
 Determining the Severity of Macroeconomic Stress Scenarios

In all our discussion so far, we have assumed that all the directional
variables are of the same importance, hence we have not assigned
different weights when aggregating the ranks of each variable.
However, we know different macroeconomic variables will af-
fect different banking institutions. For example, banks with large
credit-card portfolios are more sensitive to the unemployment rate;
banks with large mortgage portfolios are more sensitive to house
prices; and banks with large corporate portfolios are more sensitive
to real GDP and equity indexes. In addition, the severity of the loss
is related to the credit quality of the portfolio. Thus, the severity of
the stress scenario also depends on the risk profile and the busi-
nesses of a bank. Based on the bank’s experience, different weights
can be assigned to different variables to emphasise the importance
of certain variables.

Panel 12.1 Case study: Banco de España stress scenario

In this chapter, we have demonstrated the use of a simple method to
assess the relative severity of a stress macroeconomic scenario. To il-
lustrate how simple that methodology is, let us try to assess the sever-
ity of the stress-test exercise (Wyman 2012) that was conducted by
Oliver Wyman in 2012 on behalf of the Banco de España. Page 83 of
the report describes the adverse scenario with eight macroeconomic
variables, and four of them, Real GDP, Unemployment Rate, Housing
Prices and Madrid Stock Exchange Index, are similar to the variables
we have studied. After aligning the starting values, the average per-
centage change over the two-year periods for the four variables are,
Real GDP −5.1%, Unemployment Rate 19.9%, Madrid Stock Exchange
Index −52.5% and Housing Prices −21.7%. Compared with the scenar-
ios in Table 12.5, except for the Unemployment Rate, the other three
variables are the most extreme among all the scenarios examined. If we
use the ranking assignment as above, this adverse scenario is the most
severe among all the six scenarios.
However, comparing the severity of these scenarios this way is not
without problems. First, there is no reason to assume that the macro-
economic variables will affect the economies of Spain and the US in
the same way. In addition, Spain’s unemployment rate is already at
21.6% to begin with, and 21.6% is double what we saw in the 2008
recession. We can say Spain is already under an economic stress that
we have not seen in the US since the Great Depression. Thus, we face
additional challenges when comparing scenarios across different coun-
tries with different starting values on the macroeconomic variables.

217

stress_testing.indd 217 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

OTHER METHODOLOGIES OF MEASURING SEVERITY

Our approach to assessing severity is entirely ordinal and so the
relative ranks do not reflect anything about the magnitudes of the
average percentage change, or the degree of severities. Ranking
the scenarios will necessarily give some scenarios high rankings
even if all the scenarios are quite mild, and some scenarios low
rankings even if all the scenarios are quite severe. As we saw ear-
lier, our approach is to give relative ranking to the scenarios under
consideration. By comparing the scenarios with 2008 Recession,
we have attempted to give the four scenarios a reference point
besides comparing them with each other.
There have been other attempts to quantify the severity of
stress scenarios. One of them is to determine the probability of oc-
currence of a stress (or worst) scenario. The general belief is that
the less likelihood there is of an occurrence of a stress scenario,
the more severe the stress scenario will be. In the 2009 SCAP ex-
ercise, in Footnotes 3 and 4 of Federal Reserve (2009b), the su-
pervisors attempted to assign a probability of occurrence to the
adverse scenario by claiming that “the likelihood that the aver-
age unemployment rate in 2010 could be at least as high as in the
alternative more adverse scenario is roughly 10 percent. In addi-
tion, the subjective probability assessments... imply a roughly 15
percent chance that real GDP growth could be at least as low, and
unemployment at least as high... [and] there is roughly a 10 per-
cent probability that house prices will be 10 percent lower than in
the baseline by 2010”. Another example appears in an article on
[Link] in which Ed Friedman (2012) assigned probabili-
ties to the Fed’s 2013 CCAR scenarios. He first put the probabil-
ity for each baseline scenario at around 50%. Then he claimed,
“The Fed’s Severely Adverse scenario is comparable to one that
Moody’s Analytics terms S4. We currently see a 4% chance that
this scenario will occur. The Fed’s Adverse scenario is more puz-
zling... we believe [it] has about a 10% chance of occurring.” It
seems to us that most of this assigning of probabilities is based on
judgement rather than any empirical derivation.

218

stress_testing.indd 218 20/05/2013 18:45

stress_testing.indd 219

Table 12.6 Score of each macroeconomic variable and the average score of the scenario and recession
Real GDP Unemployment Rate Dow Jones HPI

Scenario Ave % SCORE Ave % SCORE Ave % SCORE Ave % SCORE Average
Change Change Change Change SCORE
2011 0.2 0 9.7 16 -11.0 43 -6.88 44 26

2012 -4.0 100 32.4 52 -40.8 100 -12.2 78 83

Determining the Severity of Macroeconomic Stress Scenarios

2013 -3.5 100 38.5 62 -40.8 100 -12.2 78 85

Hypothetical -4.3 100 32.2 52 -39.7 100 -15.7 100 88

08 Recession -2.8 100 62.2 100 -25.4 100 -15.5 100 100
219
20/05/2013 18:45
 Stress Testing: Approaches, Methods and Applications

One of the more interesting approaches to quantifying severity is

proposed by two Federal Reserve Board economists, Rochelle Edge
and Sam Rosen. Their simple approach (2012) is for each scenario,
they will score the Average % Change of the common variables by
assigning a value of 100 to the variable if the deterioration in the
variable equals what occurred in 2008 Recession (most severe), and
by assigning a value of 0 to the variable if the deterioration equals
what occurred on the average in the two recessions (mild reces-
sions) before 2008 Recession. Below, we use a modified version of
their approach to illustrate how severity can be quantified.
For each common variable in 2008 Recession, we assign a score
of 100 for the Average % Change, and 0 if the Average % Change is
0. In addition, we cap the score of each variable at 100 and assign
a floor of 0. We then use a simple linear interpolation to convert
each variable of a scenario into a score. For example, the Average
% Change in UR in 2008 Recession is 62.3%, and in Scenario 2013
is 38.5%. Thus the score for 2008 Recession in UR is 100, and for
Scenario 2013 is given by the following equation:

_2013 = 100 1 +( (38.5−62.2)

62.2 ) = 62 (12.4)

Table 12.6 gives the results of the scoring approach. Using 2008 Re-
cession as a reference point, Scenario Hypothetical, 2013 and 2012
are very similar to and a bit less severe than the 2008 recession. The
2011 scenario is the mildest and its average score is quite different
from the rest. Although our choices in this approach are sometimes
arbitrary, nonetheless the approach is intuitive and informative,
and gives a good sense of how the scenarios are compared.
So far, the approaches we have discussed do not assume any cor-
relations among the four common macroeconomic variables. The
overall assessment is made by summing each variable indepen-
dently. A statistical approach proposed by Debashish Sarkar (2012)
attempted to solve the major problems of the aggregating of infor-
mation across different variables and across time. He suggested us-
ing the Mahalanobis distance to solve these problems with a set of
weights. The distance, D, is defined by the following equation:

= √ ( − )′ −1 ( − ) (12.5)

220

stress_testing.indd 220 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

In the formula, X is a vector that stacks different variables in a sce-

nario through time. The vector µ stacks the same variables for a ref-
erence scenario. The matrix W collects the weights. This approach
is to construct W based on the variance–covariance matrix of the
out-of-sample forecast errors. While the main challenge of this ap-
proach is that the distance measure is not directional, it is also very
sensitive to the choice of W and the reference scenario. Sarkar’s re-
sults showed that this approach is best used in conjunction with
some other methods and is more efficient for identifying outlier/
problem scenarios. Similarly, in a somewhat related paper, Breuer et
al (2009) also suggested using the Mahalanobis distance to quantify
the plausibility of a severe stress scenario.
Lastly, we mention another approach that aggregates the infor-
mation in different macroeconomic variables through a simple fore-
casting framework. In their 2012 paper, Guerrieri and Welch (2012)
derived several forecasting models to predict some key metrics that
measure the “health” of a bank holding company. The key metrics
that they wanted to forecast were: net charge-offs on loans and leas-
es, pre-provision net revenue (PPNR), net interest margin (NIM)
and the Tier 1 regulatory capital ratio. The forecast is based on a
set of macroeconomic variables: Real GDP Growth, Unemployment
Rate, the growth rate of the national house price index, the term
spread, the growth rate of the S&P 500 index, the implied volatility
of the S&P 500 index options, and the real interest rate. For each
macroeconomic variable Vi and for each key banking metric, C,
they used a simple (lag) regression that takes the form:

= + −1 + 1 −1 + 2 −2 + 3 −3 + 4 −4 +
(12.6)

They used the Consolidated Reports of Condition and Income (Call

Report) of the Federal Deposit Insurance Corporation to develop
their forecasting models. By applying the historical information
from the Call Report data and the changes in the macroeconomic
variables in the stress scenarios to the forecasting models, we can
obtain the forecast estimates of the key metrics. Consequently, we
can then use the results, eg, average forecast charge-offs, to quan-
tify the severity of the scenarios. The usefulness of this approach
depends highly on the accuracy of the forecasting models, especial-

221

stress_testing.indd 221 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

ly on the later forecasting quarters. However, Guerrieri and Welch

found large root-mean-square errors for the forecasts of all the met-
rics, and their best-performing model did not beat a random walk
at all horizons for forecasting pre-provision net income.

CONCLUSION
Our approach is based on examining the extremities of each of the
directional variables, and adding up the extremities without con-
sidering the correlation and the timing of the macroeconomic vari-
ables. We have avoided trying to quantify the severity and used or-
dinal ranking to smooth out the “noises” of the variations. There is
no error estimate in our assessment, nor do we give any confidence
levels on our assessment: there is a danger of pseudo-accuracy
when we try to find precision, and precision is difficult to define. It
is exceptionally difficult to validate a model to assess the severity of
a stress scenario. In the previous section, we mention that economy.
com states that the 2013 Fed Severely Adverse scenario has a 4%
chance that it will occur. The interesting question is why it is 4%
and not 10%? The figure seems arbitrary. No one will dispute that
the 2008 recession is the most stressful economic period since the
Great Depression. Using the 2008 recession as a severe stress sce-
nario, someone might say it is a 1-in-80-year event because it is the
most severe recession for 80 years. However, if another recession
as severe as or more severe than the 2008 recession happens in the
next 10 years after 2008, then it reduces the occurrence of such a se-
vere event to a 1-in-45-year event. Thus, it is quite difficult to quan-
tify such a rare event with certainty because any occurrence of a
similar event in the future will render the estimate to be inaccurate.
One of the principles of stress testing listed in the 2009 BIS
(BIS BCBS 2009) paper is, “Stress tests should feature a range of
severities, including events capable of generating the most dam-
age whether through size of loss or through loss of reputation.”
The challenge is how to define an event that will generate the most
damage to a bank. In their paper, Borio, Drehmann and Tsatsaronis
(2012) concluded that “stress tests failed spectacularly when they
were needed most: none of them helped to detect the vulnerabili-
ties in the financial system ahead of the recent financial crisis”. To
improve the performance of macro stress tests, they suggested in-

222

stress_testing.indd 222 20/05/2013 18:45

 Determining the Severity of Macroeconomic Stress Scenarios

creasing the severity of the scenarios. The financial crisis of 2008

can be used as a starting point to gauge the severity of a bank’s
stress scenario. According to the above 2009 BIS paper, “prior to the
(2008) crisis, however, banks generally applied only moderate sce-
narios, either in terms of severity or the degree of interaction across
portfolios or risk types... Scenarios that were considered extreme
or innovative were often regarded as implausible by the board and
senior management.”
Have the banks learned their lessons? Are they designing severe
stress scenarios that will result in estimates of losses that show their
vulnerabilities? In this chapter, we have suggested a simple way to
answer these questions, and discussed alternate methodologies to
answer the same questions. In many countries, banking regulators
are requiring banks to perform stress testing on an annual basis. As
more data is gathered from these exercises, it will enhance further
research on this topic, and hopefully the macro stress tests will be-
come a valuable tool in the banks’ risk-management arsenal.

REFERENCES
BCBS, 2009, “Principles for sound stress testing practices and supervision”, Basel
Committee on Banking Supervision, May, available at: [Link]
[Link].

Borio, Claudio, Mathias Drehmann and Kostas Tsatsaronis, 2012, “Stress-testing

macro stress testing: does it live up to expectations?”, January, Monetary and Eco-
nomic Department, BIS Working Papers No. 369, available at: [Link]
publ/[Link].

Breuer, Thomas, et al, 2009, “How to Find Plausible, Severe and Useful Stress Sce-
narios”, International Journal of Central Banking 5(3), pp. 205–24.

Edge, Rochelle, and Sam Rosen, 2012, “Assessing the Severity of Macro Scenarios
in CCAR”, February 3, Federal Reserve System internal document (availability by
direct request to authors).

Federal Reserve, 2009a, “The Supervisory Capital Assessment Program: Overview

of Results”, Board of Governors of the Federal Reserve System, May 7, available at:
[Link]

Federal Reserve, 2009b, “The Supervisory Capital Assessment Program: Design

and Implementation”, Board of Governors of the Federal Reserve System, April 24,
available at: [Link]

223

stress_testing.indd 223 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

Federal Reserve, 2011a, “Comprehensive Capital Analysis and Review: Objec-

tives and Overview”, Board of Governors of the Federal Reserve System, March
18, available at: [Link]
[Link].

Federal Reserve, 2011b, “Federal Reserve System, Capital Plan Review, Summary
Instructions and Guidance”, Board of Governors of the Federal Reserve System,
November 22, available at: [Link]
bcreg/[Link].

Federal Reserve, 2012, “2013 Supervisory Scenarios for Annual Stress Tests Re-
quired under the Dodd–Frank Act Stress Testing Rules and the Capital Plan Rule”,
Board of Governors of the Federal Reserve System, November 15, available at:
[Link]

Friedman, Ed, “2012, Bank Stress Scenarios Reflect Fed’s Risk Views”, November
26, available at: [Link]

Guerrieri, Luca, and Michelle Welch, 2012, “Can Macro Variables Used in Stress
Testing Forecast the Performance of Banks?”, July 23, Finance and Economics Dis-
cussion Series, Federal Reserve Board, available at: [Link]
gov/pubs/feds/2012/201249/[Link].

Peristian, Stavros, Donald P. Morgan and Vanessa Savino, 2010, “The Information
Value of the Stress Test and Bank Opacity”, July, Federal Reserve Bank of New York,
Staff Report No. 460.

Sarkar, Debashish, 2012, “Severity Scoring Models for Assessment of BHC Stress
Scenarios”, February 10, Federal Reserve Bank of New York internal document
(availability by direct request to author).

Wyman, Oliver, 2012, “Asset Quality Review and Bottom-up Stress Test Exercise”,
September 28, available at: [Link]
informe_ow280912e.pdf.

1 In 2013, six US BHCs were subject to estimate trading losses: Bank of America Corp, Citi-
group, Goldman Sachs, JPMorgan Chase, Morgan Stanley and Wells Fargo & Co.
2 Historical Data: 1976 through Second Quarter 2012--October 9, 2012 (Excel) - available for
download at [Link]
3 See [Link]
4 The nine quarters from Q1 2008 to Q1 2010 are converted to Q3 2010 as the starting values.

224

stress_testing.indd 224 20/05/2013 18:45

 Index
(page numbers in italic type refer to figures and tables)

A calculation of risk-weighted
Advanced Internal Ratings Based assets and capital
(AIRB) approach 89, 96–7 requirements for credit risk
Akaike Information Criterion (AIC) 158 122–4
Asian consumer portfolios 152–4 and French large banks’
Korea’s credit-card crisis and corporate credit portfolio and
structural changes in the evolution over time of
consumer market 154 exposures at default 122
Basel Final Rule 58–9, 60, 64, 67
B board of directors 2–3
Banco de España stress scenario 217 Brazil 150
bank-loan portfolios as diagnostic tool banking system 165–70, 168, 169
71–87 discussed 165–70
model estimation 82–4 financial crisis’s effects on 170–80
scenario design 79–82 and Fundo Garantidor de
and stress-test goals, defining and Créditos (FGC) 177
achieving 73–8 and liquidity risk 161–91, 171,
practical objectives 77–8 172, 173, 174, 175, 176, 178,
and stress-test limitations 74–7 179, 180, 184, 185, 187, 188
bank models for loan losses and and liquidity stress tests 180–8
reserves 91 business disruption, and operational
see also loan losses and reserves, risk 60
stress-test modelling for
banks’ corporate credit portfolios, C
stress-testing 109–25, 111, 114, 115, challenges to modelling operational
116, 117, 123 risk 61–3
framework 113–21 see also operational risk: modelling
and aggregate default rate, charge-off rates across loan types 101–2,
linking GDP to 118–20 102
“conversion” function: mapping Committee of European Banking
aggregate default rate into Supervisors (CEBS) 127–8
latent credit index to generate common pitfalls in stress-testing
stressed transition matrix counterparty credit risk 52–3
120–1 see also counterparty credit-risk
data: Prudential Common exposures, stress-testing of
Reporting 113–14 Comprehensive Capital Analysis and
data: S&P Transition Matrices Review (CCAR) xxxiii, 30, 72, 75–6,
114–17 79, 85, 90, 93, 193–6 passim, 218
forecasting default rate 119–20 Consolidated Reports of Condition and
model specification 110–12, 111 Income 221
correlation factor, estimation of “conversion” function: mapping
112 aggregate default rate into latent
general 111–12 credit index to generate stressed
numerical application 122–4 transition matrix 120–1

225

stress_testing.indd 225 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

corporate credit portfolios, stress- capital exercise launched by 141

testing of 109–25, 111, 114, 115, 116, and EU-wide stress test 127–42,
117, 123 138–40
framework 113–21 key characteristics of 128–33
and aggregate default rate, lessons learned from 135–7
linking GDP to 118–20 results and disclosure 133–4, 134
“conversion” function: mapping European Systemic Risk Board (ESRB)
aggregate default rate into 127
latent credit index to generate external fraud, and operational risk 59
stressed transition matrix
120–1 F
data: Prudential Common Federal Deposit Insurance Corporation
Reporting 113–14 to develop their forecasting models.
data: S&P Transition Matrices By applying the historical
114–17 information 221
forecasting default rate 119–20 Final Rule 59, 60, 64, 67
model specification 110–12, 111 Final Supervisory Guidance 64
correlation factor, estimation of financial crisis (2007–9) 15, 23, 25, 26,
112 71, 89, 129, 165, 180
general 111–12 and Brazilian banking system 170–
numerical application 122–4 80, 177, 189; see also Brazil
calculation of risk-weighted and dependence structure 67
assets and capital and European Banking Authority,
requirements for credit risk establishment of 127
122–4 to gauge severity of stress scenario
and French large banks’ corporate 223
credit portfolio and evolution impacts of extreme illiquidity
over time of exposures at default highlighted by 28
122 and liquidity, importance of 182
counterparty credit-risk exposures, Fundo Garantidor de Créditos (FGC)
stress-testing of 37–54, 47, 48 177
common pitfalls 52–3
and credit valuation adjustment G
(CVA) 49–52 governance:
current exposure 41–4, 42 over stress testing 1–14
evolution of management of 37–40 capital and liquidity 13
implications of 40–1 coverage 11
loan equivalent 44–9 internal audit 10–11
management of 37–40 other key aspects of 11–13
evolution 37–40 policies, procedures and
Counterparty Risk Management Policy documentation 5–7
Group 38 types and approaches 12
credit risk, stress testing for 27–9 validation and independent
review 7–10
D structure of 2–5
data diagnostics and model board of directors 2–3
specification 157–8 senior management 3–5
Dodd–Frank Wall Street Reform and Greece 144
Consumer Protection Act (2010) xxi,
xxx, 17, 30, 72, 80, 90, 95, 194 I
Iceland 144
E independent review and validation 7–10
Edge, Rochelle 220 see also stress testing: governance over
European Banking Authority 30, 109 internal audit, and stress testing 10–11;

226

stress_testing.indd 226 20/05/2013 18:45

 bibliography

see also stress testing: governance over credit migration analyses 97–8
internal fraud, and operational risk 59 migration matrices, variant on
international exposures and activities: use of 101
stress testing across 143–60, 145, 150, stylised retail transition matrix 98
153 pool-level loss models 93–5
Asian consumer portfolios 152–4 Long-Term Capital Management 38
data challenges concerning 146–8 loss-distribution approach 58–61, 63, 67
estimation results 158–9 stress-testing frequency distribution
and foreign regulatory actions, within 63–4
factoring in 151–2 stress-testing severity distribution
and idiosyncratic world 143–6 within 64
and Korean credit cards and
unsecured loans, estimating M
losses on 155–6 macroeconomic stress scenarios:
and Korean credit cards and determining severity of 193–223,
unsecured loans, independent 197–8, 200–2, 204, 205, 206, 207,
variables 156 208, 209, 210, 211, 212, 214, 216,
and model specification, data 219
diagnostics may alter 157–8 historical trend of 203–8
retail portfolios 148–9, 153 nine-quarters projections 208–13
structural breaks and dummy other methodologies for 218–22
variables 155 overall assessment 213–17
structural and cyclical issues US supervisory 194–203, 197–8
149–51 aligning 196–203
Ireland 144 market risk, stress testing for 25–36
aggregation of results 28–9
K choice of scenarios 29–30
Korea 154, 155–6 and credit risk, distinction
between 27–9
L and stressed scenarios,
Liquidity Coverage Ratio (LCR) 163, 181 revaluations and computation
liquidity risk: of P&L under 32–4
and Brazilian banking system 161– mark-to-market versus
91, 171, 172, 173, 174, 175, 176, market-to-model
178, 179, 180, 184, 185, 187, 188 valuations 32–3
stress tests: central bank’s model failures, cross-effects,
approach to 180–2 approximations, specific risk
stress tests: individual banks’ and use of proxies 34
approaches to 182–8 revaluations: in practice 33–4
loan losses and reserves, stress-test revaluations: sensitivity-based,
modelling for 89–108, 92 grid-based or full 33
allowance for loan and lease loss time horizon in 30–2
models 91–3, 92 see also stress testing
bank models 91 migration matrices, variant on use of 101
economic modelling issues 103–5 model risk 68, 75
estimation 104–5 model specification, data diagnostics
specification 103–4 may alter 157–8
loan-level loss models 96–102
charge-off rates across loan types N
101–2, 102 Net Stable Funding Ratio 163
credit and behavioural scoring
models with macroeconomic O
factors 97 operational risk:

227

stress_testing.indd 227 20/05/2013 18:45

 Stress Testing: Approaches, Methods and Applications

and business disruption and system lessons learned from 135–7

failures 60 results and disclosure 133–4, 134
and clients, products and business see also European Banking Authority
practices 60 stress testing:
and damage to physical assets 60 across international exposures and
and employment practices and activities 143–60, 145, 150, 153
workplace safety 59 Asian consumer portfolios 152–4
and execution, delivery and process data challenges concerning 146–8
management 60 estimation results 158–9
and external fraud 59 and foreign regulatory actions,
and internal fraud 59 factoring in 151–2
modelling 58–63 and idiosyncratic world 143–6
challenges to 61–3 and Korean credit cards and
loss-distribution approach 59–61 unsecured loans, estimating
stress-testing approaches 63–8 losses on 155–6
1-in-N-year event, converting and Korean credit cards and
macroeconomic scenario into unsecured loans, independent
66–7 variables 156
dependence structure 67–8 and model specification, data
frequency distribution within diagnostics may alter 157–8
LDA 63–4 retail portfolios 148–9, 153
model risk 68 structural breaks and dummy
severity distribution within 64 variables 155
using scenarios 64–6 structural and cyclical issues
stress-testing methodologies 149–51
concerning 57–70 aggregation of results from 28–9
approaches to 63–8 of bank-loan portfolios, as
modelling 58–63 diagnostic tool 71–87
and workplace safety and model estimation 82–4
employment practices 59 scenario design 79–82
and stress-test goals, defining
P and achieving 73–8
physical assets, damage to, and of banks’ corporate credit portfolios
operational risk 60 109–25, 111, 114, 115, 116, 117, 123
policies, procedures and framework 113–21
documentation 5–7 model specification 110–12
see also stress testing: governance over numerical application 122–4
as class of tools xi
R of counterparty exposures 37–54, 47,
risk-management tools 15–24, 17, 18, 48, 53
19, 22 common pitfalls 52–3
enterprise-wide stress testing 16–20 and credit valuation adjustment
simple example 17–20 (CVA) 49–52
Rosen, Sam 220 and current exposure 41–4, 42
Russia 147 evolution of 37–40
implications 40–1
S and loan equivalent 44–9
S&P CreditPro 114–17 management of 37–40
scenarios, stress testing using 64–6 for credit risk 27–9
see also stress testing credit valuation adjustment (CVA)
senior management 3–5 49–52
Spain, and stress scenario: case study 217 goals of, defining and achieving 73–8
stress test, EU-wide (2011) 127–42, 138–40 practical objectives 77–8
key characteristics 128–33 and stress-test limitations 74–7

228

stress_testing.indd 228 20/05/2013 18:45

 bibliography

governance over 1–14; see also Supervisory Capital Assessment

governance Program (SCAP) 71, 79, 90, 193, 195,
capital and liquidity 13 218
coverage 11 “Supervisory Guidance on Stress
internal audit 10–11 Testing for Banking Organizations
other key aspects of 11–13 with More Than $10 Billion in Total
policies, procedures and Consolidated Assets” 72
documentation 5–7 system failures, and operational risk 60
structure 2–5
types and approaches 12 U
validation and independent US supervisory stress scenarios 194–203
review 7–10 aligning 196–203
liquidity, Brazil 180–8 see also macroeconomic stress scenarios
central bank’s approach 180–2
individual banks’ approaches 182–8 V
many approaches to xxvi validation and independent review 7–10
for market risk 25–36 see also stress testing: governance over
aggregation of results 28–9 value-at-risk:
choice of scenarios 29–30 measures, stressed calibration of
and credit risk, distinction 20–2, 22
between 27–9 models, in stress tests 20
and stressed scenarios,
revaluations and computation
of P&L under 32–4
time horizon in 30–2
methodologies, for operational risk
57–70
approaches to 63–8
modelling 58–63
modelling for loan losses and
reserves 89–108, 92
allowance for loan and lease loss
models 91–3, 92
bank models 91
economic modelling issues 103–5
loan-level loss models 96–102
pool-level loss models 93–5
need for effectiveness in xxvi
and other risk-management tools
15–24, 17, 18, 19, 22
enterprise-wide stress testing
16–20
practical objectives for 77–8
scenarios, macroeconomic 193–223,
197–8, 200–2, 204, 205, 206, 207,
208, 209, 210, 211, 212, 214, 216, 219
historical trend of 203–8
nine-quarters projections 208–13
severity, other methodologies for
measuring 218–22
severity, overall assessment of
213–17
US supervisory 194–203, 197–8

229

stress_testing.indd 229 20/05/2013 18:45

stress_testing.indd 230 20/05/2013 18:45