Counterparty Credit Risk

Governments and supervisors around the
world are focusing on the interconnectedness

of banks and systemic risk. Banks themselves
recognise that an active approach to this risk is
needed.
Measurement and management

Pricing and hedging
Stress testing
Backtesting and risk capital
Chapters from leading authorities

such as Michael Pykhtin, Evan Picoult
and the editor, Eduardo Canabarro,
explain key issues and provide a
practical toolkit for important areas
including:
Measurement,
Pricing and
Hedging
Calculating counterparty exposure

Collateral management
Valuing collaterised derivatives
Stress testing
Backtesting and calculating
economic capital
Counterparty Credit Risk is the essential
guide for practitioners, regulators,
consultants, accountants, lawmakers,
auditors and researchers.
Counterparty
Credit Risk
Edited by Eduardo Canabarro
Counterparty Credit Risk brings together leading

industry figures, regulators, academics, and
legal exports to provide the complete guide to
managing over-the-counter (OTC) derivative
counterparty credit risk.
The cascade of losses in the aftermath of the

demise of Lehman undermined markets across
the globe. Now, in the aftermath of the worst
financial crisis in a generation, counterparty
credit risk has been widely identified as a blind
spot in risk management and risen to the top of
the regulatory agenda.
The book takes a practical approach to

the essential elements with in-depth
examination and analysis of methods
to measure, price and manage OTC
derivative counterparty risk. Four
sections focus on:
Counterparty Credit Risk
The collapse of Lehman Brothers in September

2008 demonstrated what many in financial
markets had long feared: the failure of a large
financial institution, or even the imminent risk
of such failure, could cause counterparties to
incur severe losses as they scramble to cover
positions with the distressed institution.
PEFC Certified
This book has been
produced entirely from
sustainable papers that
are accredited as PEFC
compliant.
www.pefc.org
CountrPCRiskSbk.indd 1
22/3/10 10:29:11
Whole_CCR_last_amends.indd 1
19/03/2010 17:10
19/03/2010 17:10

Measurement, Pricing and Hedging
19/03/2010 17:10
Published by Risk Books, a Division of Incisive Financial Publishing Ltd

Haymarket House
2829 Haymarket
London SW1Y 4RX
Tel: + 44 (0)207 484 9700
Fax: + 44 (0)207 484 9797
E-mail: books@incisivemedia.com
Sites: www.riskbooks.com

www.incisivemedia.com
2009 Incisive Media.
ISBN 978-1-906348-34-2
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Publisher: Nick Carver

Comissioning Editor: Sarah Hastings
Managing Editor: Jennifer Gibb
Designer: Lisa Ling
Typeset by Tricolour Design, Sudbury, Suffolk
Printed and bound in the UK by PrintonDemand-Worldwide
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 copyright 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
90,Tottenham Court Road, London W1P 0LP.
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
author 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.
19/03/2010 17:10
Contents
About the Editor
vii
About the Authors
ix
Introduction
xvii
PART I: COUNTERPARTY RISK MEASUREMENT AND MANAGEMENT 1

1 Systemic Counterparty Credit Risk
3
Aaron Brown
AQR Capital Management
2 Collateralised Credit Exposure
Michael Pykhtin
Federal Reserve Board
17
3 Efficient Calculation of Counterparty Exposure Conditional on Default 51

David M. Rowe, Philip Koop and Daniel Travers
SunGard
4 Effective, Enterprise-wide Collateral Management
Darren Measures
JP Morgan
63
5 Evolution of the US Legal Framework for Counterparty Risk Mitigation 81

Lauren Teigland-Hunt
Teigland-Hunt LLP
PART II: COUNTERPARTY RISK PRICING AND HEDGING
6 Pricing and Hedging Counterparty Risk: Lessons Re-learned?
Eduardo Canabarro
Morgan Stanley
107
109
7 The Counterparty Risk of Credit Derivative Products

Jon Gregory
Ockham Financial Training and Consulting
137
8 Contingent Credit Default Swaps

Andrew P. Hollings, Shankar Mukherjee and Svein Stokke
Novarum Partners Ltd
165
19/03/2010 17:10
counterparty credit risk
9 Funding Benefit and Funding Cost

Yi Tang and Andrew Williams
Morgan Stanley
185
10 Generalised Valuation of Collateralised Derivatives

Patrick L. Chen, Katsuichiro Uchiyama and Guanghua Cao
Morgan Stanley
199
PART III: STRESS TESTING

11 Stress Testing and Scenario Analysis: Some Second Generation
Approaches
Greg Hopper
Goldman Sachs
219
221
12 Computing and Stress Testing Counterparty Credit Risk Capital

Dan Rosen; David Saunders
The Fields Institute for Research in Mathematical Sciences and R2 Financial
Technologies; University of Waterloo
245
PART IV: ECONOMIC AND REGULATORY CAPITAL

13 Back(testing) to the Future: From Market Risk to Counterparty Credit
Risk Models
Eduardo Epperlein; Sean Paul Hrabak; Wei Zhu, Alan Smillie
Citi; Financial Services Authority; Citi
293
14 Economic Capital for Counterparty Credit Risk from Two Perspectives

Evan Picoult
Citi and Columbia Business School
327
Index
351
295
vi
19/03/2010 17:10
About the Editor
Eduardo Canabarro is the managing director responsible for quantitative risk management at Morgan Stanley. He is responsible
for the development of the methods and models used to measure
market and credit risks as well as for the independent review and
validation of pricing and risk models used by the bank. Prior to
Morgan Stanley, he held a similar position at Lehman Brothers as
managing director and global head of quantitative risk management. Eduardo has also worked for Goldman Sachs and Salomon
Brothers in quantitative modelling and risk management.
Eduardo has published various articles in the Journal of Financial
Engineering, Journal of Fixed Income, The Journal of Risk Financing, The
Journal of Risk and Re-Insurance, and Risk. His articles Counterparty Risk: Measurement and Pricing and Analyzing Counterparty
Risk were cornerstones for the Basel II framework for regulatory
capital on counterparty credit risk. He has spoken at leading risk
management events around the world including the ones sponsored by the Wharton School, BIS, ICBI, Risk, PRMIA and IAFE.
Eduardo holds degrees in electrical engineering and an MBA (finance) from UFRGS Brazil as well as MS and PhD degrees in finance from University of California at Berkeley, US.
vii
19/03/2010 17:10
19/03/2010 17:10
About the Authors
Aaron Brown is a risk manager at AQR Capital Management and

the author of The Poker Face of Wall Street (2006, picked as one of
the ten best books of 2006 by Business Week) and A World of Chance
(2008, with Reuven and Gabrielle Brenner). In his 27-year Wall
Street career, Aaron has been a portfolio manager, trader, head of
mortgage securities and risk manager for firms that include Morgan Stanley and Citigroup; he also did a stint as a finance professor.
Aaron holds degrees in applied mathematics from Harvard and finance from the University of Chicago.
Guanghua Cao is a desk strategist for the Credit Valuation Adjustment (CVA) Strategies Group in the Fixed Income Department at
Morgan Stanley, New York. His responsibilities include the development of quantitative models for CVA valuation, hedging and
management. Prior to joining the CVA Strategies Group, Guanghua
was a quantitative analyst for the Credit Methodology Group at
Morgan Stanley, focusing on the validation of risk models and risk
monitoring. He holds a Masters in financial engineering from the
Haas School of Business, UC Berkeley, a PhD in applied mathematics from the Southern Methodist University and a BA in applied
mathematics from the Shanghai JiaoTong University.
Patrick Chen is currently the global head of the Model Review
Group at Morgan Stanley, where he is responsible for the independent review and validation of pricing models and also for the
risk models used by the bank, as well as model risk measurement
methodology, development and reporting. Prior to joining Morgan
Stanley, Patrick has been the head of Counterparty Credit Risk &
CVA Analytics and head of Operational Risk Analytics at Lehman
Brothers. He has worked for Merrill Lynch, Barclays Capital and
Bank of America in various trading desk and quantitative risk manix
19/03/2010 17:10
agement roles since the start of his career in 1996. Patrick holds a
PhD in theoretical physics from the University of Illinois at UrbanaChampaign and a BSc in physics from the University of Science and
Technology of China.
Eduardo Epperlein is a managing director and the head of Market and Counterparty Risk Analytics and Model Validation at Citi.
He is responsible for defining market and counterparty credit risk
standards, developing methodologies for calculating regulatory
and economic capital and validating all pricing and risk models.
Eduardo is also the main point of contact with regulatory bodies
for issues related to market and counterparty risk. He joined Citi
Toronto in 1994. Eduardo holds a PhD in plasma physics from Imperial College London and spent eight years as a research scientist
at the Laboratory for Laser Energetics before joining Citi.
Jon Gregory is a consultant specialising in counterparty risk and
credit derivatives. Until 2008, he was the global head of credit analytics at Barclays Capital based in London. Jon has worked on many
aspects of credit modelling over the past decade, previously being
with BNP Paribas and Salomon Brothers (now Citi). In addition to
publishing papers on the pricing of credit risk and related topics, he
was in 2001 a co-author of the book Credit: The Complete Guide to Pricing, Hedging and Risk Management, short-listed for the Kulp-Wright
Book Award for the most significant text in the field of risk management and insurance. He has a PhD from Cambridge University.
Andrew Hollings is a co-founder of Novarum. His 20-year career in
the OTC derivatives market has focused on risk management and
trading exotic products. Andrew began his career at JPMorgan Chase
where he held the position of managing director and global head of
the New Financial Products Group. Prior to founding Novarum, he
was a managing director at Citigroup Global Markets where he pioneered the risk management of OTC derivative counterparty credit
risk. Andrew received a BSc (Hon) in economics and finance and an
MA in economics from the University of Wales, Cardiff.
19/03/2010 17:10
about the authors
Greg Hopper is the managing director and head of the Counterparty Credit Risk Analytics Group at Goldman Sachs. The Counterparty Credit Risk Analytics Group is responsible for building global
counterparty exposure models for all products, VaR-based margining models and stress-testing applications and models. He also
oversees counterparty credit risk in the Americas for credit derivatives, equity derivatives, loan trading and prime brokerage. Before
joining Goldman Sachs, Greg was an executive director at Morgan
Stanley. He has a BA in mathematics, an MA in applied mathematics and a PhD in economics, all from the University of Virginia.
Sean Paul Hrabak is a risk specialist in the Prudential Risk Division
at the Financial Services Authority. Previously, he was a director in
the Risk Analytics Group at Citi and headed up the Counterparty
Risk Modelling and Projects Team. Sean holds a PhD in mathematics from Kings College London.
Philip Koop is a quantitative software developer who creates
software models for the pricing and risk-management of financial
derivatives. He is one of the creators of the Adaptiv Analytics solution of SunGards Capital Markets and Investment Banking business segment, where he continues to develop software to support
credit risk management. He has also worked at risk management
software houses TrueRisk and Algorithmics and at the Toronto-Dominion Bank. He has a degree in mathematics (honours computer
science) from the University of Waterloo.
Darren Measures is the executive director for Collateral Management Sales and Relationship Management at JP Morgan in New
York, a position he has held since November 2007. From 2004 until
2007, Darren was the global product manager responsible for the
launch of the Derivatives Collateral Management product and US
Equity TriParty Repo extension. He moved to the US from London in 2000 in order to run the western hemisphere operation after
working on the investment banks OTC derivatives collateral management programme. Darren joined JP Morgan in 1999 after stints
in sales, credit and market risk management at Citibank, SBC and
UBS. These experiences developed his knowledge of OTC Libor,
xi
19/03/2010 17:10
government and credit derivative products. Darren is a working

member of the Asset Managers Forum and International Swaps
and Dealers Associations Collateral and Custodial Committees.
He is a frequent speaker at industry conferences and often contributes to key publications.
Shankar Mukherjee is a co-founder of Novarum. In his 20-year
career in the OTC derivatives market he has focused mainly on
risk management and trading hybrid products. Shankar began
his career with Citibank, Mumbai, before moving to JPMorgan
Chase where he was the managing director of the Credit Portfolio
Group. Prior to founding Novarum, he was a managing director
at Citigroup Global Markets where he helped to set up the OTC
derivative counterparty credit risk management business. Shankar
received a BE(Hon) in electrical and electronics engineering from
BITS, Pilani, India and an MBA from the Indian Institute of Management, Calcutta, India.
Evan Picoult is a managing director within Citis Risk Architecture
Department as well as an adjunct professor in the decision, risk and
operations department of Columbia Universitys Business School.
Over the past few years he has focused on firm-wide projects regarding Basel II, stress testing and the enhancement of the measurement, implementation and use of economic capital. He has led the
development of the methods used at Citi for measuring market risk
and counterparty credit risk. He is a frequent lecturer on risk topics
at professional conferences, regulatory conferences and at universities and has published a number of articles on risk topics. Evan
is on the Advisory Board of the IAFE (International Association of
Financial Engineers) and is co-head of the IAFE Credit Risk Committee. He has been the North American co-chair of ISDAs Risk
Management Committee since the mid-1990s. Evan has a PhD in
experimental particle physics from Columbia University, and after
joining Citibank returned to Columbia part time, to obtain an MBA
in finance from the Executive programme of Columbias Business
School.
xii
19/03/2010 17:10
about the authors
Michael Pykhtin is a vice president in the Counterparty Credit

Risk Analytics Group at Bank of America. His main responsibility
is developing new credit risk methodologies for the banks portfolio of derivatives and other capital market products. Prior to joining
Bank of America in 2005, Michael worked at KeyCorp, where he
was responsible for developing models of portfolio credit risk and
economic capital. Michael edited Counterparty Credit Risk Modelling,
published in 2005. He is also a contributing author to several recent
edited collections. Michael has extensively published in the leading
industry journals. He is an associate editor of The Journal of Credit
Risk. Michael holds a PhD in physics from the University of Pennsylvania.
Dan Rosen is the CEO of R2 Financial Technologies, as well as a
visiting fellow at the Fields Institute for Research in Mathematical
Sciences and an adjunct professor in mathematical finance at the
University of Toronto. Dan acts as an advisor to institutions around
the world and lectures extensively on valuation of derivatives and
structured finance, risk management, economic and regulatory
capital. He has authored numerous risk management and financial
engineering publications, and serves in the editorial board of several industrial and academic journals. Prior to founding R2 in 2006,
Dan had a successful 10-year career at Algorithmics, where he held
senior management roles in strategy, research and financial engineering, business development and product marketing. He holds
an MASc and PhD from the University of Toronto.
David M. Rowe is EVP for risk management at SunGard in London. In this role he provides strategic input to SunGards solutions
for risk management and serves as an external spokesman on risk
management issues. He appears frequently at industry conferences
and has written a monthly column for Risk magazine since 1999.
Prior to joining SunGard, David spent more than 25 years in the
banking and economic forecasting industries. His former positions
include senior vice president of the Risk Management Information
Group at Bank of America in San Francisco and EVP and director of
research at Townsend-Greenspan & Co.
xiii
19/03/2010 17:10
David Saunders is an assistant professor in the department of statistics and actuarial science at the University of Waterloo, and a senior research consultant at R2 Financial Technologies. He is the author of many articles on the subjects of risk management, portfolio
optimisation and derivatives pricing. David holds a PhD in mathematics from the University of Toronto, and is a research fellow of
the HERMES European Center of Excellence on Computational Finance and Economics at the University of Cyprus, and the Waterloo
Research Institute in Insurance, Securities and Quantitative Finance
(WatRISQ) at the University of Waterloo.
Alan Smillie is a senior quantitative analyst in the Risk Analytics
Group at Citi. He is responsible for developing and maintaining
modelling solutions for market risk, counterparty credit risk, stress
testing and economic capital across Citi. Alan received his PhD in
quantitative finance from Imperial College, London in 2006.
Svein Stokke is a co-founder of Novarum. For over 15 years, he has
been in the OTC derivatives markets as a trader, risk manager and
quantitative analyst. Svein began his career with JPMorgan Chase
in Oslo and London in the exotic derivatives business. There, and
later at Citigroup, he developed new risk management techniques
for OTC derivative counterparty credit risk. Before founding Novarum, Svein was head of Counterparty Credit Risk Trading at Citigroup based in New York. He received an MSc in control engineering from the Norwegian University of Science and Technology and
an MBA from Warwick Business School, UK.
Yi Tang is currently with Morgan Stanley as a managing director
and the global head of CVA Strategies Group. He also worked as
the interim global co-head of the Credit Strategies Group. Previously, Yi was a general manager and the head of the Quantitative
Analytics Division of Shinsei Securities responsible for derivatives
modelling in interest rates, foreign exchange, equity, credit, commodities, as well as interest rates/foreign exchange, interest rates/
equity and other hybrids. He also worked at Goldman Sachs as the
head of the CVA Strategies Group in FICC and at Bear Stearns as a
managing director/principal in the FAST Department and the head
xiv
19/03/2010 17:10
about the authors
of a quant group responsible for part of interest rates derivatives

modelling and part of the interest rates/credit hybrid derivatives
modelling. Before switching to the field of quantitative finance,
he worked as an adjunct assistant professor and a postdoctoral
researcher in physics at the University of California, Los Angeles
(UCLA). Yi has been an invited speaker at several conferences/
seminars on quantitative finance and co-authored a book entitled
Quantitative Analysis, Derivatives Modeling, and Trading Strategies in
the Presence of Counterparty Credit Risk for the fixed-income market
(published in 2007). He received his PhD in physics from UCLA.
Lauren Teigland-Hunts practice focuses on derivative, securities
and commodity transactions and US commodities law and regulation. She has extensive experience in representing financial institutions, hedge funds, exchanges and trade associations in a wide
range of trading matters, with a focus on equity, credit, fixed income and commodity derivatives, as well as transactions in physical energy. Lauren also advises on regulatory issues related to
derivative transactions and has acted as counsel to several ISDA
drafting committees. Prior to founding Teigland-Hunt LLP in 2002,
she was an attorney at Sullivan & Cromwell LLP in the firms Commodities, Futures and Derivatives Group (1996-2002) and worked
as a futures trader and banker in New York and Paris (1986-93).
Dan Travers, product manager, Adaptiv Analytics, SunGard Capital Markets & Investment Banking, has a background in pure and
financial mathematics. After spending time at Adelaide University teaching and earning a postgraduate degree, he worked for a
number of years setting up trading and risk management functions
in the energy markets. He moved to work for a software vendor for
the financei, where he has many more years experience in consulting and development of software solutions. He now works as product manager for Adaptiv Analytics, a revolutionary new calculation
engine from SunGard, and Front Arena Middle Office Risk.
Katsuichiro Uchiyama is a vice president for the Credit Methodology Group at Morgan Stanley, New York. His responsibilities
include the development of quantitative methods and models for
xv
19/03/2010 17:10
credit risk and the validation of risk models used by the bank. Prior
to joining Morgan Stanley, Katsuichiro was a quantitative analyst
at Credit Suisse First Boston and Sanwa Bank, focusing on credit
products and fixed income derivatives. He holds a BA and MA in
pure mathematics from Kyoto University, Japan.
Andrew Williams works as a managing director for Morgan Stanley.
Wei Zhu is a director and head of the Market Risk Analytics and
Counterparty Risk Analytics Modeling unit within Citigroups Institutional Client Group (ICG). He has been working on market
risk models to capture both general market risk and issuer specific
risk over the past eight years and has been involved in the quantitative modeling of counterparty credit risk in recent years. Wei
received his PhD in physics from New York University in 2001 and
is a CFA charter holder since 2004.
xvi
19/03/2010 17:10
Introduction
Eduardo Canabarro
Morgan Stanley
We write this book at a grave time, in the wake of a severe financial crisis that caused the demise of many large, well-established
financial institutions, as well as the deepest and broadest economic
downturn since the Great Depression.
At the time of writing (February 2010), governments and financial industry regulators around the world are re-evaluating the economic role of banks, their interconnectedness, pro-cyclical activities and the standards for capital adequacy, liquidity and leverage.
Substantive reforms are on the horizon and, over the course of last
year, the Basel Committee of Banking Supervision (BCBS) issued
important proposals on the risk, liquidity, capital and compensation management of banks. These reforms and the changes that will
come with them will re-shape the global financial industry.
Counterparty credit risk (CCR) was at the centre of the most
pressing moments of the 200708 financial crisis, as it was in other
recent crises; this hardly comes as a surprise, considering the large
scale and scope of inter-bank trading.
Derivatives are flexible tools that allow banks and their clients
to manage and transfer financial risks effectively. They spur financial innovation and promote the efficiency of financial markets. But
they, like other forms of inter-bank trading, create an undesirable
byproduct the connection of the counterparties and the interconnectedness of the banking system.
The failure of a large financial institution, or even the imminent
risk of such failure, may cause counterparties to incur losses as they
are forced to replace their outstanding derivative trades with the
distressed institution. This can create a cascade of losses undermining overall confidence in the financial markets. Inter-bank trading
stalls, impairing the ability of even healthy financial institutions to
participate in the markets and to continue to manage their rapidly
changing risk profiles.
xvii
19/03/2010 17:10
We experienced this scenario during the acute moments of the

200708 financial crisis. The seriousness of the events in the aftermath of Lehman Brothers filing for bankruptcy was a sobering reminder to all of us of the fragility of the equilibrium of trust and
belief that sustains the financial markets.
This book is a collection of cutting-edge analyses and ideas for
enhancing methods and practices to manage over-the-counter
(OTC) derivative CCR.
When I was approached by Risk Books to be the editor of this
book, I immediately recognised it as an extraordinary opportunity
to compile the experiences and insights of a seasoned group of experts who would certainly have much substantive to say in light of
the recent crisis. In fact, I told the publisher that I would only agree
to edit the book if I could find and persuade such a group to participate in the project.
We were fortunate! We were able to line up an outstanding team
of authors whose remarkable expertise in the key pertinent topics
suited our plan for the book extremely well.
The topics covered are technical, but the authors placed great emphasis on making the material as digestible as possible. Our common
objective was to address the substance in a clear and concise form.
My role as the editor was to organise the material and to offer
suggestions to the authors to align their chapters with each other
and with the overall flow of the book. I did not attempt to impose
form on the material submitted, as I thought that preserving the
originality of the contributions was of much value. The reader will
learn not only from the content but also from the way that each
author understands and presents it.
We organised the book in four sections:
o
o
o
o
Counterparty risk measurement and management (five chapters)

Counterparty risk pricing and hedging (five chapters)
Stress testing of counterparty risk (two chapters)
Back-testing and risk capital of counterparty risk (two chapters)
Our choice of sections represents the fundamental and sequential

stages of the risk management processes: understanding and quantification of the risks, identification of risk-reducing actions, pricing
xviii
19/03/2010 17:10
INTRODUCTION
and hedging the risks, performing stress tests and scenario analyses
and calculating and allocating economic and regulatory capital.
The first section covers counterparty risk measurement and management. It focuses on the important current issues related to systemic risk and collateralisation.
In the first chapter, Aaron Brown analyses the effect of counterparty credit risk on the financial system as a whole, concluding that
the total systemic risk may exceed the sum of the individual firm
risks. This conclusion is important to regulators who wish to protect the financial system. It is also important to individual firms.
It is now possible to conceive of scenarios where a firm could be
driven out of the market because of the potential losses that it imposes on other market participants.
In the second chapter, Michael Pykhtin presents the current stateof-the-art in modelling collateralised credit exposure for both unilateral and bilateral margin agreements. He emphasises the critical
role of the margin period of risk in the models. Showing exposure
profiles for several simple examples, he illustrates the main properties of market-driven collateralised credit exposures.
Chapter 3 describes the current state of the computational methods applied to calculate counterparty exposures conditional on
default. Conditional exposures are important to price wrong- and
right-way risks. The authors, David Rowe, Philip Koop and Daniel
Travers, propose an approach to constrain the exposure simulations
to instances that imply default of the counterparty. This reduces the
computational burden for calculating exposure-at-default to little
more than that required to generate a standard unconditional exposure simulation. They also address the opportunities created by the
utilisation of parallel processing and by the expansion of addressable memory in 64-bit technologies.
In Chapter 4, Darren Measures examines the evolution of collateralisation in credit risk mitigation. As collateralisation expanded in
scale and scope, collateral management systems evolved across different product areas. In each area, issues related to eligibility of collateral types, haircuts and optimisation were gradually addressed. The
similarities of functions such as margin calls, settlements and reconciliations across product areas offer valuable opportunities to manxix
19/03/2010 17:10
age these processes in a manner that reduces total enterprise risk.

In the fifth and final chapter of this section, Lauren TeiglandHunt examines how the approaches to managing CCR through
legal documentation have evolved over the last two decades. Beginning with key credit risk mitigation initiatives such as the creation of the International Swaps and Derivatives Association (ISDA)
Master Agreement, she identifies a series of industry undertakings
to enhance legal documentation and standards for managing CCR.
The author then reviews the lessons learned from the financial crisis of 200708, and discusses common approaches to mitigating
CCR via legal documentation today. Finally, she concludes with
an overview of the regulatory reform efforts targeting the financial
sector that are likely to impact the way in which CCR is addressed
in the OTC derivatives markets.
The second section covers the pricing of counterparty risks and of
collateral arrangements. Credit valuation adjustments (CVAs) have
produced about two thirds of the counterparty credit losses incurred by banks during the recent crisis, and have motivated some
of the BCBSs proposals for reforms on the regulatory capital on
counterparty risks.1 Collateral and its relationship to OTC derivative valuation, funding costs and availability of funding are also
current focal issues.
In Chapter 6, Eduardo Canabarro reviews the techniques used to
price and hedge CVAs that have evolved over the last 15 years. He
then discusses how those techniques performed during the 200708
crisis and the lessons that we have (re-)learned for the future.
Chapter 7 examines the wrong-way risk inherent in credit derivatives. Jon Gregory analyses single-name credit default swaps
(CDSs), index tranches and, especially, super-senior tranches. He
then focuses on the monoline insurers acting as sellers of supersenior protection and creating a situation that led to billions of dollars of losses for banks.
In Chapter 8, Andrew Hollings, Shankar Mukherjee and Svein
Stokke consider contingent CDSs (CCDSs) as effective hedges for
market-driven counterparty exposures. They show various features of the structure with respect to counterparty exposure measurement, CVAs and economic capital.
xx
19/03/2010 17:10
INTRODUCTION
Chapter 9 sees Yi Tang and Andrew Williams analysing the funding benefits and costs created by OTC derivatives. At a time when
banks are being required to enhance their liquidity contingency
plans, the pricing and active management of the funding consequences of OTC derivatives is of paramount importance.
In Chapter 10, the authors Patrick Chen, Katsuichiro Uchiyama
and Guanghua Cao provide a detailed analysis of the pricing of
OTC derivatives that are covered by collateral agreements. The exchange of collateral generates cashflows over the life of the trades.
Those cashflows need to be priced as an integral part of the trades.
A portfolio of derivatives covered by a collateral agreement has a
value different from one of the same portfolio without such agreement, and the difference would exist even in the absence of any
credit risks it is caused by the difference between secured and
unsecured financing rates.
The third section covers the stress testing of counterparty risks. The
experience of the financial crisis made it clear that stress-testing
frameworks need to be expanded and enhanced.
In Chapter 11, Greg Hopper presents advanced (second-generation) stress-testing techniques that attempt to cover systematically
the space of risks. He discusses the applicability of algorithmic
stress tests and automatic scenario generation.
Dan Rosen and David Saunders then go on to discuss, in Chapter
12, the implementation of stress-testing techniques to, among others, counterparty risks, regulatory alpha and wrong-way risks.
They address the methodological and computational challenges
faced by the standard implementations of counterparty exposure
measurement systems.
The fourth and final section covers backtesting and the integration of counterparty risks in the economic and regulatory capital
frameworks.
Eduardo Epperlein, Sean Paul Hrabak, Wei Zhu and Alan Smillie
describe in Chapter 14 how the standard VaR backtesting framework, which applies to the entire banks trading portfolio, can be
generalised to multiple portfolios and become a more powerful
diagnostic tool. The framework is then enhanced to test the perxxi
19/03/2010 17:10
formance of counterparty exposure models. Key enhancements are

the ability to account for the path dependence inherent in exposure profiles over long time horizons, the correlations of the various
portfolios, and the direct testing of both specific percentile as well
as expected positive exposure.
Finally, in Chapter 14, Evan Picoult discusses economic capital
from two different perspectives: price risk (trading book) and value
risk (banking book). He then highlights various aspects of the tension
between the two frameworks as they apply to counterparty risks.
I hope that the reader, after studying this book, will have a broad
and clear perspective of the current issues and thinking on counterparty risk management, and of how the authors see the evolution
of the practice from here.
I will close this introduction with a sober statement:
The 200708 financial crisis was not the last crisis, and it may not
even be the most severe one we will ever experience. Other crises
will certainly occur and they will cause losses to financial intermediaries and damage to economies and societies. This is the intrinsic
nature of market systems: the alternating cycles of boom and bust,
optimism and pessimism, greed and risk aversion, expansion and
contraction, under-regulation and over-regulation. As long as human reason and emotions drive economic actions, we are certain to
face busts and crises. Our role as professional risk managers is to
create and promote the continuous building of a solid foundation
of methods and practices to minimise the damage when the crises
occur. Better assessments of risk and uncertainty, more transparency in the description and communication of risks, proper quantitative modelling and valuation of risks seem to be the key disciplines
to pursue. The same economic activities that generate profits and
are conducive to aligned behaviours of profit-maximising firms are
the ones that can and will generate large losses when reasoning and
sentiment change. And those will always change, we just do not
know when.
1
See BCBSs consultative document, Strengthening the Resilience of the Banking Sector,
December 2009.
xxii
19/03/2010 17:10
Part I
Counterparty risk
measurement and
management
19/03/2010 17:10
19/03/2010 17:10
Systemic Counterparty Credit Risk

Aaron Brown
AQR Capital Management
There are two basic ways to protect an interlocking system. First

is to make each component individually unlikely to fail. Second is
to make the system robust to individual failures. Traditional prudential supervision has focused on the first. Each firm computes
a worst-case (in some sense) loss and is required to hold enough
capital to cover that amount. Part of that computation has been to
estimate exposure to counterparty credit risk.
However, the financial crisis that began in 2009 has highlighted the need for the second type of protection as well, limiting not
the counterparty credit risk a firm is exposed to, but the amount
of counterparty credit risk any firm is allowed to inflict upon the
financial system. That is, to focus not only on What is the probability a firm will fail, but also on How much damage will it do if it
fails? This is not only a regulatory issue. Several of the high-profile
failures in 2008 were arguably caused not by actual firm losses exceeding available firm capital, but by regulators and counterparties
losing willingness to bear the systemic risk the firm represented.
Almost every firm, even the best capitalised and most respected,
came closer than managers and shareholders liked to catastrophic
losses of market confidence. This chapter discusses how to define,
measure, monitor and manage this systemic counterparty credit
risk.
DEFINITION OF SYSTEMIC CREDIT COUNTERPARY RISK
1
19/03/2010 17:10
Counterparty credit risk (CCR) is two-sided. If A does an over-thecounter (OTC) swap with B, A has a potential loss if B goes bankrupt while owing net payment to A (most of the chapters in this
book are concerned with estimating and managing that risk). At
the same time, A inflicts a potential loss on B. B may lose money if
A goes bankrupt while owing net payment to B. Concern about this
inverse risk has increased dramatically as a result of the events of
2008, in which it threatened to bring down the financial system and
caused governments to assume massive potential liabilities. For
this reason, we call it systemic CCR (SCCR).
In normal CCR, when A considers its exposure to B, A does not
worry about losses to third parties if B were to go bankrupt. The
problem begins with B and ends with A. But if A inflicts losses on
B by defaulting, A will inflict them on all its other creditors as well,
which may cause some of them to default to others, leading to a systemic problem. Moreover, since SCCR is not currently reported in
standardised ways, nor regulated nor limited, the fear of systemic
problems from As default can cause the actions that lead to As failure and thereby realise the fears even if A was otherwise fully able
to meet its obligations. Normal CCR requires a spark, that is a default, in order to do damage, but SCCR can combust spontaneously.
Although SCCR is not part of Basel capital norms, nor one of the
standard risk metrics reported to senior management, risk managers have computed it for years. The concern was not with systemic
risk, it was to identify counterparties with large exposure to the
firm for general informational reasons. Risk managers considered it
a good idea to know who was exposed to the firms credit, the virtual creditors of the firm, for more or less the same reasons a chief
financial officer (CFO) wants to know the firms large investors. In
future SCCR is likely to be a central concern of regulators, perhaps
even of more concern than normal CCR.
Arguably, events since August 2007 have demonstrated generally skillful management of CCR but deep problems in the management of SCCR. Although many institutions had large counterparty
credit losses, there were few cases of these losses exceeding limits
or causing severe financial stress. Even the most painful liquidation of the crisis, the failure of Lehman Brothers, did not cause a
counterparty to fail (the Reserve Fund and some other funds had
2
19/03/2010 17:10
survival-threatening losses from holding Lehman short-term debt,

but this was investment credit risk, not CCR). All hedge funds that
liquidated did so in an orderly manner, without major counterparty
losses. In fact there have been no major counterparty losses due to
hedge fund blow-ups since the near miss in 1998. The last financial
institution failure to inflict widespread counterparty pain was Refco in 2005, and no unrelated institution failed as a result.
What did threaten the survival of almost all large financial institutions in 2008 was SCCR. Fear of interlocking liabilities froze the
financial markets and caused runs on the bank. There were literal
runs in the case of Northern Rock and E*Trade bank, redemption
surges in many investment funds and free-falling stock prices for
publicly traded investment and commercial banks. The best capitalised and most prestigious companies were not immune, and attempts to reassure the market backfired. Only unprecedented and
massive government intervention prevented meltdown. Self-fulfilling fears of SCCR played a large part, perhaps a decisive part, in the
failures of AIG, Bear Stearns and monoline insurers, and were major contributing factors to the Federal Home Loan Mortgage Corporation (Freddie Mac), the Federal National Mortgage Association
(Fannie Mae), Merrill Lynch and Lehman Brothers.1
The crisis brought calls to increase capital reserves for CCR, but
these seem misdirected. No institution failed or had serious troubles due to inadequate capital to cover counterparty credit losses.
Moreover, it is impractical to reduce the probability of failure of
each individual regulated institution to zero. Even if you did, a systemic problem could arise from an unregulated institution. What
is practical is to reduce the probability of a problem spreading to
harm the entire financial system, whether initiated by a regulated
or unregulated institution, to near zero. That requires controlling
SCCR. In future, we believe no regulated or unregulated institution
will be allowed to inflict unlimited amounts of SCCR on the financial system, regardless of how well capitalised it is against direct
CCR, and certainly regardless of whether it has a AAA credit rating
or quasi-government backing.
THEORY OF SYSTEMIC CREDIT COUNTERPARTY RISK
On first view, it may seem that systemic risk requires no additional
3
19/03/2010 17:10
analysis. If A and B each concern themselves with the CCR to themselves, both sides of the risk are covered. However, As analysis of
the risk it inflicts on B will differ fundamentally from Bs analysis
of the same risk.
The biggest difference is that CCR is both optional (since it can be
hedged) and actual (since it causes direct cash losses). SCCR cannot
be easily hedged and is virtual. It does not cause direct cash losses
to A, but if it is perceived to be too high, A can be forced out of
business. Thus, for CCR, a firm has a choice of hedging or accepting some probability distribution of cash losses, while SCCR must
be managed for the firms survival, and it has to managed in both
perception and reality.
Almost as important is the different direction of feedback. CCR
has negative feedback if it increases, either due to increase in exposure or decline in counterparty credit, a firm will take steps to
reduce exposures or hedge the credit. In the extreme, default of a
counterparty means no further increase in CCR. If SCCR increases
due to trading losses or a decline in credit quality, the steps counterparties and investors take to protect themselves tend to increase
overall SCCR. This will be discussed in more detail later. Also, the
default of a counterparty does not end the build-up of SCCR, it accelerates it. CCR is therefore self-limiting, while SCCR can generate
a self-sustaining dynamic that destroys a firm and inflicts damage
beyond the firm.
The negative feedback can be seen in another way CCR increases when a firm makes money, because those profits are owed to it
by others, while SCCR increases when a firm loses money, because
it owes those losses to others.2 A triggering default is therefore less
likely with CCR than SCCR.
Another difference concerns basic economic theory. If B goes
bankrupt, swap counterparty A may have a net payable or net receivable on the swap. If A has a net receivable, that becomes an
unsecured claim on the bankruptcy estate. If A has a net payable,
it gets no benefit from the bankruptcy, the payment still must be
made.3 Therefore, at swap initiation, A has effectively written a
credit default swap on B for a random notional amount. The value
of that credit default swap is a straightforward loss to A, whether
A pays it in up-front cash by buying protection, or over time in ex4
19/03/2010 17:10
pected hedging losses or as an expected value of loss in default.4

If A goes bankrupt with a net receivable on the swap, the bankruptcy will not affect trade settlement. If A goes bankrupt with
a net payable on the swap, the payable is converted to an unsecured claim, and the cash required to make the payment to B will
be distributed by the bankruptcy court.5 In general, that cash will
aid creditors in order of priority. Unless senior unsecured creditors
make full recovery, the result will be that B has a loss at the expense
of either creditors with higher priority or other senior unsecured
creditors. In no case can creditors with lower priority than senior
unsecured benefit from the SCCR. In particular, equity holders get
no benefit.
There is no reason for A to consider that it gains from Bs loss.
If A acts to maximise shareholder wealth it will be indifferent to a
transfer from one creditor to another. To the extent A has a duty to
either creditors or customers, Bs loss is a gain only if A has a higher
duty to the recipient creditor than to B. In general, theres no reason
to think that is true. In fact, since B may be a customer as well as a
creditor, it might be assumed that A will be more concerned about
Bs interests than the interests of pure creditors or pure customers.
To the extent that A is motivated by regulations, it is likely that Bs
loss is also a loss to A, since counterparty credit losses are usually
considered more damaging to the financial system than bondholder losses.
In theory, SCCR allows A to raise capital at lower cost because
the SCCR increases expected recovery rates to some creditors in
default. So, although A has no benefit from systemic credit risk in
bankruptcy, prior to bankruptcy it can profit by paying lower rates
on some debt. However, this profit is less than Bs loss (since much
of it goes to As existing creditors) and is questionable because the
route is so indirect. In practice, we suspect this is negligible. Most
senior creditors would be more worried about the increased business risk to A from SCCR than the potential effect on their recovery
in default.
DYNAMICS OF SYSTEMIC CREDIT COUNTERPARTY RISK
So far we have considered a static analysis of a single contract. More
asymmetries between CCR and SCCR emerge when we consider an
5
19/03/2010 17:10
entity involved in continuous trading with many counterparties. It

is not difficult to protect a firm from excessive CCR losses due to a
large move of a single market factor or the default of a single counterparty. The large unexpected losses that concern CCR managers
are more likely to result from correlations among market factors
and credit default spread increases of several counterparties, either
because all are exposed to the same risks, or because the counterparties have mutual liabilities and drag each other down.
Default correlation is not a factor for SCCR. If a firm defaults, it
defaults to all counterparties. Since a firm typically has offsetting
trades with different counterparties, it is lack of market correlation
that will maximise systemic counterparty credit losses, as these depend on gross exposures rather than net. So markets in which correlations go to one or negative one (crashes) concern the risk manager of CCR, while the risk manager of SCCR worries about things
that are supposed to be correlated moving independently (spread
blowouts). While 2008 saw both kinds of disruptions, the spread
blowouts were far more anomalous statistically than the crashes, so
it is no surprise that SCCR caused more problems than CCR.
There is a more general point here, about how prudential supervision evolved in response to crashes and panics. While these are
still dangers, as financial sophistication and complexity increase,
crash dangers are mitigated while spread blowouts become a larger
threat. Regulation encourages this in the sense of insisting on protection from crashes but being relatively insensitive to spread risks.
Some of the calls for more simplicity and transparency may be trying to force the financial system to fit existing regulation rather
than adapting regulation to modern financial practice. Enlightened
regulation may be able to encourage an optimal balance between
the two types of risk.
Continuing along this line of analysis reveals a gaping hole in
CCR management from the standpoint of a regulator. CCR depends
on the difference in credit spread between the swap counterparties,
since an OTC contract can result in either payables or receivables.
If two counterparties have the same credit spread, economic CCR
nets out to zero.6 But to a regulator concerned with the overall level
of credit risk, it is the absolute level of spreads that matter. If everyones credit spreads blow out 500 basis points, no one has to revise
6
19/03/2010 17:10
their credit valuation adjustments, but a regulators problems just

got much worse.
CONTAGION AND CONTAINMENT
The last difference is a company can transmit SCCR as well as create it. If A has OTC derivative trades with B, and B also trades with
C, changes in Cs credit affect A, although A has no control over C,
and may not even know of Cs existence. CCR can be managed by
one counterparty at a time, and even by half a counterparty (that
is, A may protect itself from Bs default without B protecting itself
from As default). SCCR is all or nothing it has to be managed
with respect to all counterparties at once, and even beyond that to
counterparties of counterparties. It is also mutual if A protects
itself, it protects its counterparties. In this respect, CCR is like the
health risk of poor eating and lack of exercise, while SCCR is more
like a communicable disease. Clearly these two risks require different types of public health responses. In SCCR, ask not for whom
the bell tolls, it tolls for thee.
Some people have taken this argument even farther and treat all
risks as common risks, leading to some sort of economy-wide risk
management requiring a systemic risk regulator. That is far too
broad there is no freedom without freedom to take risk. SCCR is
special because it nets out to zero and is a risk no one wants. Market risk also nets out to zero, in the sense that profit to one side of
a trade is loss to another. But it is a risk both sides want. Natural
disaster risk is a risk no one wants, but it does not net out to zero.
Restricting overall market risk interferes with the operation of the
financial system, and it is impossible to eliminate natural disaster
risk (although it would be good if we could). However, it is possible
and desirable to restrict overall SCCR, because it is zero for the financial system as a whole, and reducing it improves the operation
of the financial system. It is only for transactional convenience and
capital savings that we permit any SCCR at all.
Two traditional tools for reducing SCCR are clearinghouses and
mark-to-market payments. We have already seen increasing use of
both of these, and that trend seems likely to continue due to the
twin impact of market forces and regulatory encouragement. But
these do not eliminate SCCR. For one thing, clearinghouses can fail.7
7
19/03/2010 17:10
Clearinghouses also require standardisation to be effective, and

there will always be innovation and demand for custom products.
Both central clearing and mark-to-market payments impose costs
that are minor for some applications, such as bilateral trading, but
unacceptably high for others, such as splitting the risk of a pool of
corporate bonds into funding, interest rate and credit risks, among
thousands of individual investors, each with unique requirements.
Some of the risk transfer accomplished in the past by OTC trading
and securitisation may move to open clearinghouses, but there will
still be a lot of SCCR in the financial system.
More important, clearinghouses and mark-to-market both require liquid trading to be effective. In the recent crisis, public trading remained liquid while some OTC markets and trading in securitised products froze. A spectacular illustration of that is how
the markets shrugged off the tremendous dislocations of the quant
equity crash of August 7August 9, 2007. Stock market trading volume smashed previous records as virtually all quant equity funds
delevered rapidly. Yet the entire process was smooth and without
liquidity, credit or operational glitches. This compares very favourably to, say, the interbank lending market, which began experiencing problems around the same time. The quant equity crash was
over by August 10, 2007, with no fatalities, while the interbank
lending crisis has lasted almost two years and has claimed many
victims. This has understandably created support for the increased
use of exchange trading.
Twenty years earlier, however, the situation was reversed. In the
stock market crash of 1987 it was the public stock and futures markets that nearly failed, while OTC trading and securitisation was
operationally unaffected. Public exchange problems are potentially
more damaging than OTC issues because they affect all market
participants at once, including many smaller and less sophisticated
investors. To a risk manager it is obvious that having two parallel
robust trading systems is safer than putting all eggs in one basket.
MEASURING SYSTEMIC CREDIT COUNTERPARTY RISK
The first step to measuring SCCR is the same as the first step to
measuring CCR. You need to project potential future payables and
receivables among all relevant legal entities, both legal entities of
8
19/03/2010 17:10
INTRODUCTION
the firm and trading partners, reflecting netting, margin and other
CCR mitigants. You do this for a large number of simulated futures.
This is a complicated task for a variety of reasons, which will be
addressed in other chapters of this book. However, once you have
completed it, it is a simple matter to add the total payables of the
firm (counting receivables as zero) in various future paths.
CCR is evaluated over the life of contracts (and in some cases for
assumed rollovers and drawdowns, that is for contracts that do not
yet exist). That makes sense, because you cannot count on changes
with a specific counterparty. But for SCCR, it only makes sense to
look at the near term, say 14 days. This is because SCCR is affected
by trading and margin payments throughout the firm, and these
will not stay anything like constant over extended periods. Even if
you knew current positions could cause an unacceptable amount of
SCCR in one year, there would be no reason to do anything about it
now, as it would change in much less than a year and, if it did not,
you could easily fix it later (say by sending a margin payment).
The most popular metric for CCR is expected positive exposure.
This makes sense for a hedgeable risk it is the relevant statistic for
computing the financial cost of the exposure. But since SCCR is not
hedgeable (at least not by the institution that is causing it, and not
by anyone for the financial system as a whole), and because we are
concerned about the danger it presents rather than a financial cost,
it makes more sense to use a value-at-risk (VaR) type metric, say the
99% highest total payables over two weeks in a simulation.
A minor issue is whether to consider right-way versus wrong-way
risk. Right-way risk means you are exposed to a counterparty when
it is likely to have the money to pay, like a sovereign government
buying insurance against its currency weakening. Wrong-way risk
is the opposite, like a company buying call options on its own stock
to protect against dilution from employee stock option grants. In
principle, consideration of right- versus wrong-way risk improves
estimates, both for CCR and SCCR. In practice, it has proven difficult to do for CCR, except in special circumstances. For SCCR, it is
easier. We know we are concerned with times when credit default
swap spreads are high, liquidity is at a premium, risk-flight assets
are high and so forth; also, a financial institution has an excellent
idea of its market exposures. Therefore, it is a good idea to weight
9
19/03/2010 17:10
the simulation toward dangerous paths, in the sense of threatening

confidence in the firm and indicating nervous markets.
MONITORING SYSTEMIC CREDIT COUNTERPARTY RISK
The main output of this process is the potential systemic counterparty exposure (PSCE). There is a 1% chance the firm will owe more
than this gross amount to counterparties sometime over the next
two weeks, assuming normal markets and no trading. The way to
think about this is the firms treasurer should be entirely comfortable with the possibility of making an unsecured 14-day borrowing
of this amount, on top of the firms other needs.
In practice, the actual gross payable will be funded mostly with
gross receivables it will not be borrowed. However, if there is
any suspicion the firm would be unable to borrow it, confidence
in the firm might evaporate. In that case some of the gross receivables might not be paid on time, counterparties might be reluctant to engage in risk-increasing trades and short-term unsecured
lenders might disappear. Think of the firm as running a payments
processing business in connection with its OTC trading, moving
funds from some counterparties to others, and the PSCE as a high
estimate of the total amount of such transfers. Counterparties will
not tolerate significant risk in the payment processing part of OTC
trading, so if they suspect the firm could not easily cover its gross
payables, they will not want to deal with the firm at all.
Since perception of SCCR is as important as its reality, periodic
disclosure of PSCE or the knowledge it is monitored by regulators
will reduce the chance of spontaneous combustion. Of course, like
VaR, the PSCE should be subjected to rigourous backtest and supplemented with stress testing to ensure the firm has contingency
plans for the 1% of the time gross OTC payables exceed PSCE.
MANAGING SYSTEMIC CREDIT COUNTERPARTY RISK
To a regulator, it is natural to think of all reported PSCEs of major
firms as constituting an unfunded clearinghouse for OTC trades.
To be more precise, it is not unfunded, but its funding is held by
individual firms in the form of CCR capital. In a real clearinghouse
the funding capital is combined in a pool available to all members.
The problem with holding it individually is there is no necessary
10
19/03/2010 17:10
relation between a firms PSCE and its CCR capital (you might expect them both to increase roughly with the size and risk of the
firms OTC business, which is probably true for the average levels
of each, but we do not think it is a very close relation at any fixed
point in time). Total CCR capital should be adequate to cover total
PSCE, but the capital may not be in the right places. Moreover, CCR
capital is not segregated from general firm capital, so it might not
be available to cover SCCR.8
A first reaction is this is a bad way to run a clearinghouse and the
regulator should insist on actual funding with pooled capital equal
to each firms PSCE, plus a safety cushion. But remember, there are
reasons these trades are OTC we have already moved everything
we can to real clearinghouses. Marking-to-market and offsetting
are too difficult for these trades, or in other cases revealing trade
details to other exchange members (which is necessary if capital is
to be pooled) exposes competitive business intelligence. Moreover,
requiring firms to post capital to the clearinghouse would result in
a double charge for the same risk, once in capital requirement for
CCR and once in clearinghouse capital for SCCR.9
Some people might not mind double charging,10 figuring anything that encourages OTC trades to move onto a clearinghouse is
good. But it is bad risk management to double count. If you want
more protection, increase CCR capital requirement or lower allowed amounts of PSCE, do not build in overlapping protections
at different steps out of conservatism. That leads to waste, and
each individual protection invariably gets watered down, because
people figure the other protection is there anyway; and people discover innovations that avoid both charges. Worst of all, the lack of
theoretical clarity leads to confusion and error, which stifles a lot of
useful innovation while permitting much bad practice.
There is a way to double charge rationally require third-party
collateral custody with initial margin for both sides. This can be
thought of as setting up one clearinghouse for each pair of counterparties that trade with each other. While this is simple and solves
both the CCR and PCCR problems, it is also extremely expensive.
The capital required to fund it could be put to much better use, that
it could reduce risk much more, if it were pooled with general firm
capital to protect against the diversified portfolio of risks the firm
11
19/03/2010 17:10
faces. The problem with overreacting to the last crisis is you are
likely to divert resources away from the place the next crisis will
occur.
Assuming that regulators are comfortable with single charging
for CCR and SCCR, and allowing the capital to be pooled within
but not among firms, it obviously makes sense for them to limit the
size of the unfunded clearinghouse exposures relative to the credit
strength of its members.
LIMITS FOR SYSTEMIC CREDIT COUNTERPARTY RISK
To set the size limits, a thought experiment is helpful. Suppose the
regulatory plan in the event of crisis is to take over the unfunded
OTC clearinghouse for one day. It cannot be taken over for longer,
because the complexity and customisation of the contracts makes it
too risky. No one entity, and certainly not the regulator, can mark all
the contracts to market. The regulator has no desire or authority to
bail anyone out, just to allocate any credit losses quickly. The point
is to avoid clearing gridlock, to provide liquidity but not credit protection (if credit protection is also required to prevent follow-on
damage, that is a different decision, probably by a different regulator, and is not as time-critical). Remember this is a thought experiment only, as far as we know no legal or organisational basis exists for this to happen, especially since the unfunded clearinghouse
does not exist in the first place. We are only trying to establish the
resources necessary to prevent default of a major OTC counterparty
from bringing the system down, not setting the deployment strategy for those resources.
Assuming that everyone has done their calculations well, and
markets have not moved beyond the levels anticipated in PSCE and
CCR capital computations, the regulator should be able to borrow
(remember, thought experiment) the capital allocated by counterparties to cover the CCR risk of the failed entities, and use that to
cover the gross payables of the failed entities, which should be less
than the combined PSCE of those entities. Once that happens, no
additional counterparties should fail, because all receivables have
been received. Therefore, all payables should be paid, which should
suffice to pay back the loans, minus any actual credit losses.
Of course, even in our thought experiment the regulator cannot
12
19/03/2010 17:10
expect to borrow cash from the non-defaulting counterparties. CCR

is used as part of the capital requirement computation, not the reserve requirement. So the regulator must stand ready to provide
overnight liquidity to OTC counterparties secured by receivables
due from failed counterparties, trusting the CCR capital of counterparties to cover any shortfall.
At this point we need to make some arbitrary assumptions. They
do not matter much, because it is the form of the rule we are interest in, not the parameters. The parameters would be set to allow a
level of OTC trading acceptable to both regulators and market participants. It is sheer guesswork to translate parameters to the probability distribution of liquidity amount the regulator would have to
provide.
A simple model is one where each OTC counterparty has a
known one-day probability of default, with a constant pairwise
correlation of the event of default with other counterparties. The
regulator wants to set a limit on the expected liquidity provision
amount, plus some number of standard deviations. All OTC counterparties are allowed to make equal marginal contribution to this
value. That leads to a rule of the form permitted PSCE equals:
where p is the one-day probability of default and A, B and C are

parameters.
Some reasonable values might be A = 0.0001%, B = 0.1% and C =
0.005%, with the result interpreted in million US dollars. In normal
credit times that would exclude non-investment grade counterparties from any PSCE (that is, any OTC derivative exposure would
require initial margin up to the 99% 14-day loss level) and allow
something like:
o
o
o
o
o
o
US$250 million to Baa3 entities;

US$500 to Baa2;
US$1,000 to Baa1;
US$1,250 to A3;
US$1,500 to A2;
US$1,750 to A1;
13
19/03/2010 17:10
o
o
o
o
US$2,000 to Aa3;
US$3,000 to Aa2;
US$4,000 to Aa3; and
US$8,000 to AAA.
Regulators might prefer higher or lower levels, and possibly a different form of the rule, and possibly base the limit on credit default
swap spreads or regulatory capital rather than credit ratings. But
something like this makes economic sense. Of course, only standalone credit could be used, a holding company could not set up
four A1 subs relying on the parent rating in order to get a total
US$7,000 million PSCE limit. But it could set up bankruptcy remote
special-purpose entities to trade, with independent credit support.
EFFECTS OF LIMITING SYSTEMIC CREDIT COUNTERPARTY RISK
A positive effect of this rule is that if an institutions credit deteriorated, it would be forced to reduce systemic counterparty credit exposure at the same time its counterparties wanted the same thing.
Since both sides are pressured, trading prices are more likely to be
fair and markets more likely to be liquid than in the current system
where only the counterparties care about reducing exposure (in
fact, the troubled firm arguably has an incentive not to reduce exposure). Fair and liquid markets could help cure the financial system, especially as they are likely to be in the types of instruments
of most concern.
Another positive effect is that, if financial firm credit spreads
begin to widen in the market, the firms will be forced to unwind
trades, increase margin or move trades to clearinghouses; or, of
course, they could shore up their credit by issuing equity or other
means. A negative effect, which is really the flip side of the same
thing, is general trading liquidity in unrelated markets will dry up
if problems in one market threaten financial institution credit. This
could even result in institutions being trapped in unwanted exposures, unable to reduce risk. One possible mitigant to this would be
an ISDA trigger allowing either side to terminate a trade if either
party exceeds its regulatory PSCE limit.
A financial institution has a different perspective from a regulator. It is concerned with protecting itself, not the financial system.
14
19/03/2010 17:10
The danger of SCCR is that it will induce a death spiral. Too many
counterparties have too much exposure to the firm, so they start
pulling out funds, novating trades, refusing new credit extensions
and new trading and hedging in the market. This could lead to onesided markets in credit exposure to the firm, a liquidity squeeze
and increasing credit default swap spreads and out-of-the-money
put premiums. These signs could exacerbate worries, leading to
additional actions, spiralling down until the firm is forced out of
business. The regulatory protections could exacerbate this process
as news that a firm is having PSCE troubles could trigger worse
troubles.
Therefore, an institution will strive to limit SCCR based on the
mood of the market, both to credit exposure in general and to the
firms credit in particular. A simple rule that captures this general
effect would be to limit the product of PSCE and one-year credit
default swap spreads on the firm to 10% of its market capitalisation.
Of course, that is just an off-hand example, firms can do analysis to
determine the appropriate levels of PSCE, and will probably come
up with more complex and nuanced rules. Firms already do a version of this, as they know virtual credit extended by counterparties
is an unreliable source of funds, as it will be withdrawn quickly if
the firm is perceived to be in trouble.
SCCR has always been with us, but for the first time we have
the tools to manage it directly. Historically, there has been market
and regulatory pressure to reduce it, but that has not yet been effective enough to avoid periodic disasters. SCCR has led to government guarantees and bailouts, which are very expensive in cash
and public support for regulated markets. There is always pressure
to downplay SCCR, because a run on any firm threatens all firms.
Thus, firms continue trading with a risky counterparty because the
consequences of cutting off trading are unpredictable and potentially severe. This leads to opaqueness and nervous markets. It is
time to manage SCCR cheaply, rationally and openly.
Failure may be too strong in some of these cases; only Lehman filed for bankruptcy and
was liquidated. However, they were all risk management failures in the sense that losses
caused control of the institution to pass beyond management to either courts, government
or government-assisted third-party takeover.
This effect is negligible for a large diversified trading organisation, as that type of entity is
15
19/03/2010 17:10
likely to have offsets for most positions and, in any case, the trading P&L from a few days is
unlikely to be significant relative to capital or equity.
However, the effect can be quite significant for a less diversified organisation, such as a
hedge fund, insurance company or corporate hedger.
3 Actually the situation can be complicated, as other chapters of this book point out, but for
our purposes we can pretend all OTC contracts are terminated upon bankruptcy with immediate settlement required of the non-defaulting party.
4 With bilateral margin terms, or no margin, it is also true that B effectively writes a credit
default swap on A.
5 Again, this can be more complicated, but that does not concern us here.
6 This assumes a perfectly symmetrical contract and margin terms, but that does not affect
the general point.
7 The Paris Sugar Bourse, the Chicago Mercantile Exchange, the London Metal Exchange
(three times) and the NYMEX have all inflicted bankruptcies on trading counterparties, by
changing rules and forcing settlement at non-economic terms, to save their clearinghouses.
This does all the damage of an OTC default, plus has the additional disadvantages of eroding market confidence in trading contracts and allowing the institution that caused the problem to survive and create problems another day.
8 The opposite is equally likely, that more than the CCR capital is available, but regulators
have to plan for the bad case.
9 It is tempting to try to combine these, by making firms contribute actual PSCE capital to the
clearinghouse, but at the same time also counting clearinghouse deposits as equity for capital computations. Unfortunately, if firm A gets in trouble, it draws down on pooled clearinghouse capital, which would then reduce the regulatory capital of other firms, just when they
need it to cover CCR from A. It is only from a single firms point of view that CCR and SCCR
are complementary; you get CCR when you have a receivable under a contract and SCCR
when you have a payable, and never both at the same time. For the system as a whole, one
firms CCR is another firms SCCR, so they always occur at the same time.
10 The current system has the opposite problem of double charging. A major source of liquidity for some prime brokers was the initial and excess margin deposited by their customers.
However, these same funds were counted as equity by the customers. At the same time, they
were used by the prime broker to mitigate counterparty credit exposure, thus reducing capital charge, thus increasing reported capital ratios. In normal times there is a clear distinction
between equity and cash, and we consider only one of the prime broker or customer failing
at one time. In that case the same cash can serve all three purposes simultaneously. In the recent crisis, when only cash mattered, there was only one pool to support three requirements,
all of which were calling for the funds at the same time.
16
19/03/2010 17:10
Collateralised Credit Exposure

Michael Pykhtin
Federal Reserve Board
One of the most popular and effective techniques for mitigating and
controlling counterparty credit risk is collateralisation. The idea of
collateralisation of counterparty risk is very similar to the way collateral is used to mitigate lending risk: collateral is used to reduce
credit exposure. However, because of the uncertainty of counterparty credit exposure and the bilateral nature of counterparty credit
risk, collateral management is much more complex in the case of
counterparty risk. Exposure of one counterparty to another changes
every day, as the markets move. Therefore, to keep the current exposure under control, it is not sufficient to post collateral once instead, collateral must be posted and returned frequently, preferably
on a daily basis. Moreover, because mark-to-market (MTM) value
of many types of over-the-counter (OTC) derivatives can change
sign, credit exposure can change direction. If both counterparties
want to limit their exposure, then either counterparty is required
to post collateral when the others exposure rises. This chapter discusses modelling credit exposure and collateral and collateralised
exposure profiles.
The rules of posting collateral are specified in a legally enforceable margin agreement signed by both counterparties. Due to both
the popularity of collateralisation and its operational and legal
complexity, the standardisation of margin agreements was initiated in the 1990s. Nowadays most margin agreements in developed countries take the form of a Credit Support Annex (CSA) to
17
19/03/2010 17:10
an International Swaps and Derivatives Association (ISDA)1 Master

Agreement. All the trades covered by a Master Agreement are subject to close-out netting: in the event of default of one of the counterparties, they are treated as single transaction whose value equals
the sum of the individual trades values. A CSA may cover all the
trades under a Master Agreement, or a subset of trades.
Margin agreements have several important contractual features:
o Coverage. The set of trades covered by a margin agreement has
to be specified. From the pure credit exposure reduction point
of view, it is optimal to include all the trades between the counterparties in a margin agreement. However, inclusion of more
exotic trades increases the risk of valuation disputes and, as a
consequence, also increases delays in posting the full amount of
required collateral. Thus, it may be optimal to exclude difficult to
value trades from the margin agreement.
o Type of collateral. Admissible types of collateral have to be specified. Cash collateral has become the most common, but other
types of collateral are possible. Usually, non-cash collateral includes high-quality securities such as government bonds or
highly rated corporate bonds.
o Haircuts for non-cash collateral. Non-cash collateral carries the risk
of depreciation. To account for this risk, haircuts are specified for
each type of non-cash admissible collateral. Margin calls are usually made in terms of cash-equivalent collateral, and the haircuts
are applied to the market values of the assets that serve as collateral to obtain the cash equivalents. It should be kept in mind
that haircuts protect from the depreciation risk only to a certain
extent. For example, if a corporate bond is posted as collateral,
default of the issuer would bring the value of the bond down by
a far greater extent than what is covered by the haircut. This risk
of rapid depreciation can be mitigated by the requirement that
the posted security be replaced should its rating fall below a prespecified level (although this would not protect from a sudden
default).
o Interest rate for cash collateral. If cash collateral is posted, it earns an
interest rate, which needs to be specified in the margin agreement.
o Currency of collateral. An admissible currency of collateral has to
be specified. When two counterparties do not have the same
18
19/03/2010 17:10
local currency, one of them will have to take foreign exchange

risk linked to the posted collateral, even when it is in the form
of cash. Securities in different currencies specified as admissible
collateral may attract larger haircuts due to the additional foreign exchange risk.
Unilateral/bilateral. Margin agreements can be either unilateral
(in one of the counterparties favour) or bilateral. Under a unilateral (bilateral) agreement, one (either) of the counterparties is
required to post collateral when the unsecured exposure (ie, exposure calculated without taking into account any collateral) of
the other counterparty rises above a threshold. If the credit quality of the counterparties is comparable, the agreement is usually
bilateral. If the credit quality of one of the counterparties is significantly worse than that of the other, the stronger counterparty
may insist on a unilateral agreement in its favour.
Threshold(s). Under a unilateral margin agreement, a threshold
is specified for one of the counterparties. This counterparty has
to post collateral if the unsecured exposure of the other counterparty exceeds the threshold. The amount of posted collateral
must be sufficient to cover the excess. Under a bilateral margin
agreement, two distinct thresholds are specified one for either
counterparty. Either counterparty has to post collateral when
the others unsecured exposure exceeds the threshold. Threshold is essentially the maximum level of credit exposure that the
other counterparty will tolerate. Threshold usually depends on
the counterpartys credit quality: the better the credit quality, the
higher the threshold. Non-investment-grade counterparties (including hedge funds) are often assigned the threshold of zero.
High-credit-quality counterparties may receive a high value of
the threshold. However, margin agreement may require that the
threshold drop if the counterparty is downgraded.
Minimum transfer amount (MTA). If the amount of collateral that
needs to be posted or returned is less than the minimum transfer
amount, no transfer of collateral occurs. The purpose of MTA is
to reduce the frequency of collateral exchange.
Margin call frequency. This is the time period which specifies how
often revaluation of the portfolio has to be performed to determine the amount of collateral (if any) that needs to be posted or
returned. Daily margin call frequency has become standard.
19
19/03/2010 17:10
Let us consider the mechanics of margin agreements. Suppose

that two counterparties, A and B, have a portfolio of nettable trades
supported by a unilateral margin agreement in As favour with a
daily margin call frequency. Counterparty A keeps to the following
procedure on the daily basis. It evaluates the entire portfolio and
compares the portfolio MTM value to counterparty Bs threshold
and the amount of cash-equivalent collateral counterparty B has
already posted to counterparty A. If the portfolio value is below the
threshold, counterparty A should return any collateral that it holds
to counterparty B, subject to the MTA. If the portfolio value is above
the threshold, the collateral amount required by counterparty A is
equal to the difference between the portfolio value and the threshold. If the required collateral exceeds the value of the posted collateral and the excess is greater than the MTA, a margin call is made. If
the required collateral is less than the value of the posted collateral
by more than the MTA, counterparty A returns this difference to
counterparty B. In a bilateral agreement, counterparty B follows the
same procedure as counterparty A.
It is worth remembering that, although margin agreements are a
very powerful counterparty credit risk mitigation technique, they
are operationally very complex. Information on all contractual features should be maintained accurately for each counterparty. Accurate valuation of all trades with collateralised counterparties must
be performed on a daily basis. All calls for collateral must be followed up until the required collateral is received. Failure to deliver
collateral is a dangerous signal that may indicate that the counterparty is about to default.
MODELLING CREDIT EXPOSURE
Exposure simulation framework2
Let us now consider a portfolio of trades between two counterparties. To make a distinction between them, we will call one of the
counterparties a bank. Unless mentioned otherwise, in this chapter
we will do all the analysis from the banks perspective in other
words, we will be interested in the banks exposure to the counterpartys default.
Because of the complex nature of banks portfolios, exposure distribution at future time points is usually obtained via the Monte
Carlo simulation process. This process typically consists of three
20
19/03/2010 17:10
major steps:
o Scenario generation. The dynamics of market risk factors (eg, interest rates, foreign exchange rates, etc) is specified via relatively
simple stochastic processes (such as geometric Brownian motion). These processes are calibrated either to historical data or to
market implied data. Future values of the market risk factors are
simulated for a fixed set of future time points.
o Instrument valuation. For each simulation time point and for each
realisation of the underlying market risk factors, valuation is performed for each trade in the counterparty portfolio.
o Aggregation. For each simulation time point and for each realisation of the underlying market risk factors, counterparty-level
exposure is obtained by applying the necessary netting and collateral rules.
The outcome of this process is a set of realisations of the counterparty-level exposure (each realisation corresponds to one market
scenario) at each simulation time point.
Because of the computational intensity required to calculate
counterparty exposures especially for a bank with a large portfolio certain compromises between the accuracy and the speed of
the calculation are usually made: relatively small number of market scenarios (typically, a few thousand) and simulation time points
(typically, in the 50200 range), simplified valuation methods, etc.
Non-collateralised exposure
The banks exposure to the counterparty at a given future time is
given by the banks economic loss in the event of the counterpartys default at that time. If the counterparty defaults, the bank must
close out all of its positions with the counterparty. To determine the
loss arising from the counterpartys default, it is convenient to assume that the bank enters into exactly the same portfolio of trades
with another counterparty in order to maintain its market position.
Since the banks market position is unchanged after replacing the
portfolio, the loss is determined by the portfolios replacement cost
at the time of default.
If the portfolio MTM value at the time of default is negative for
the bank, the bank receives this amount when it replaces the portfolio, but has to forward the money to the defaulting counterparty, so
21
19/03/2010 17:10
that the net loss is zero. If the portfolio value at the time of default
is positive for the bank, the bank pays this amount when replacing
the portfolio, but receives nothing (assuming no recovery) from the
defaulting counterparty, so that the net loss is equal to the portfolio
value.
Thus, future scenarios with positive portfolio MTM value result
in exposure equal to the portfolio value, while scenarios with negative portfolio value result in zero exposure. In mathematical terms,
the non-collateralised exposure E(t) at time t is related to portfolio
MTM value V(t) via
(2.1)
Equation 2.1 is valid for any scenario at any future time point t.
Collateralised exposure
Suppose now that the netting agreement between the bank and the
counterparty is covered by a margin agreement. In the presence of
a margin agreement, exposure at time t is reduced by the amount
of collateral C(t) available to the bank at time t. Still, exposure is
floored at zero because, in the event of the counterpartys default,
any collateral in excess of the banks replacement cost must be returned to the counterparty. We can easily formalise this and define
the collateralised exposure EC(t) via
(2.2)
The reader should keep in mind that both V(t) and C(t) are not
known today and are, therefore, random variables from the modelling point of view. In an exposure simulation framework, Equation
2.2 is applied on the scenario-by-scenario basis: for each realisation
of portfolio value and collateral at simulation time point t, a realisation of collateralised exposure is calculated.
We have written Equation 2.2 assuming that that bank can only
receive collateral. It is less obvious that the banks posting collateral
can also result in the banks exposure to the counterpartys default,
and that this credit exposure can still be described by Equation 2.2.
For Equation 2.2 to be valid regardless of whether the bank posts
or receives collateral, we have to assume the following convention:
C(t) is positive if the banks holds collateral at time t and negative if
the bank has posted collateral at time t.
22
19/03/2010 17:10
Let us verify that, if we use this convention, Equation 2.2 will

describe the banks credit exposure for the scenarios when the bank
posts collateral. Suppose that at time t the bank has posted collateral in the amount |C(t)|, so that the received collateral under our
convention is C(t) = |C(t)| < 0. Suppose also that the counterparty
defaults at time t. To determine the banks loss, let us consider three
possible types of scenarios for the portfolio MTM value V(t):
o V(t) C(t). Since the portfolio value is negative, the bank receives
the amount |V(t)| from another counterparty when it replaces
the portfolio. This amount is greater than the collateral amount
|C(t)| that the bank has posted. To settle the portfolio with
the defaulting counterparty, the bank needs to pay the amount
|V(t)|, but it withholds the amount |C(t)| to recover collateral
that it has posted. There is no loss to the bank under this scenario.
o C(t) < V(t) 0. The portfolio value is still negative, so the bank
receives the amount |V(t)| from another counterparty when it
replaces the portfolio. However, this amount is not sufficient to
recover the collateral amount |C(t)| that the bank has posted.
Even if the bank does not pay anything to the defaulting counterparty, it still loses the amount |C(t)||V(t)|= V(t) C(t).
o V(t) > 0. Under this scenario the bank pays the amount V(t) to
replace the portfolio. The posted collateral |C(t)| is completely
lost. The net loss is V(t) +|C(t)|= V(t) C(t).
Combining these three scenario types into a single expression
produces the banks loss equal to max {V(t) C(t),0}, which is precisely the right-hand side of Equation 2.2.
Thus, if we know both the portfolio value and collateral received
or posted by the bank at a given point in time, we can easily calculate the collateralised exposure at that time. We know how to simulate portfolio value at future time points, but do not know yet how
to calculate the collateral for a given portfolio value scenario. In the
next section we will develop approaches to modelling collateral.
MODELLING COLLATERAL
Minimum transfer amount approximation
As we have mentioned above, most margin agreements have contractual MTA. No collateral call or return is made unless the collat23
19/03/2010 17:10
eral amount that needs to be transferred exceeds the MTA. Modelling this feature accurately requires daily simulation time points
for the portfolio MTM value, which is usually not feasible due to
the long-term nature of counterparty risk and the typical size of the
trading book of large banks. Instead, a simplified approach is often
used: MTA is added to the threshold. In other words, the margin
agreement is treated as the one that has (i) an effective threshold
equal to the sum of the contractual threshold and the contractual
MTA; and (ii) an effective MTA equal to zero. We will refer to this
approach as the MTA approximation.
The reader should note that including MTA into the threshold is
only a crude approximation that captures some features of MTA,
but is by no means equivalent to full modelling of the effect of
MTA. To illustrate this point, suppose that a bank has a counterparty with the threshold equal to US$5, the MTA equal to US$1,
the portfolio MTM value equal to zero and no collateral posted. If
the portfolio value starts rising, the first margin call by the bank is
made when the portfolio value exceeds US$6. The MTA approximation captures this behaviour, but mis-states the amount of this
margin call. If, for example, the first margin call is made when the
portfolio value equals US$7, the actual amount called for would be
US$2, while according to the MTA approximation it would be only
US$1.
In spite of its crudeness, the MTA approximation has become
standard in modelling exposure under margin agreements. We will
incorporate the MTA approximation in all our models of collateral.
Nave collateral model
Unilateral margin agreement
Suppose that there is a unilateral margin agreement between a
bank and a counterparty in the banks favour. The counterparty
must post collateral when the banks unsecured exposure exceeds
the counterparty effective threshold Hcpt (which is equal to the sum
of the contractual threshold and the MTA). The amount of collateral
that the bank receives must be sufficient to cover this excess. We
could describe this relation mathematically by specifying collateral
C(t) that is available to the bank at time t as:
(2.3)
24
19/03/2010 17:10
Collateralised exposure is obtained by substituting Equation 2.3

into the Equation 2.2:
(2.4)
where we have used Equation (2.1) that relates the portfolio MTM
value V(t) and the non-collateralised exposure E(t).
The interpretation of Equation 2.4 is very straightforward. If the
portfolio value is below the threshold, collateral is zero and there
is no difference between collateralised and non-collateralised exposures. If the portfolio value exceeds the threshold, the collateral
covers the excess, so that the collateralised exposure is equal to the
threshold. Figure 2.1 illustrates this behaviour for a sample portfolio value scenario, assuming the effective threshold at the level of
US$10.
Figure 2.1 Relation between portfolio value and collateralised
exposure under the nave collateral model, assuming US$10
effective threshold
Bilateral margin agreement

Under the bilateral agreement, both the bank and the counterparty
25
19/03/2010 17:10
post collateral. The counterparty posts collateral in the same manner as described, ie, when V(t) > Hcpt. The bank has to post collateral
when the counterpartys unsecured exposure to the banks default
rises above the banks effective threshold, equal to the sum of the
banks contractual threshold and the MTA. Since we are doing the
analysis from the banks perspective, it is convenient to specify
the banks effective threshold as a negative quantity: Hbnk < 0. The
bank posts collateral whenever the portfolio MTM value (from the
banks perspective) falls below this threshold, ie, when V(t) < Hbnk.
The amount that the bank has to post is equal to the difference between the threshold and the portfolio value. We can quantify this
by setting the collateral C(t) available to the bank by time t equal to:
(2.5)
The first term in the right-hand side of Equation 2.5 describes the
scenarios when the bank receives collateral (ie, C(t) > 0), while the
second term describes the scenarios when the bank posts collateral
(ie, C(t) < 0). This is consistent with our convention that collateral
available to the bank is negative when the bank posts collateral.
Note that both terms cannot be non-zero simultaneously.
Let us now calculate collateralised exposure using Equation 2.2.
We can consider three possible cases separately:
o V(t) > Hcpt. Then, from Equation (2.5), collateral is equal to C(t) =
V(t) Hcpt. This quantity is positive, so the bank receives collateral. Calculating collateralised exposure we obtain EC(t) = max{V(t)
[V(t) Hcpt], 0} = max{Hcpt, 0} = Hcpt
o Hbnk V(t) Hcpt. Then, from Equation (2.5), collateral is equal to
zero, and the collateralised exposure equals the uncollateralised
exposure.
o V(t) < Hbnk. Then, from Equation (2.5), collateral is equal to C(t) =
V(t) Hbnk. This quantity is negative, so the bank posts collateral.
Calculating collateralised exposure, we obtain EC(t) = max{V(t)
[V(t) Hbnk], 0} = max{Hbnk, 0} = 0 because the banks effective
threshold is negative.
Thus, the second term in Equation 2.5 is irrelevant from the exposure point of view: the banks posting collateral does not have any
26
19/03/2010 17:10
effect on the banks credit exposure to the counterparty. Therefore,

the collateralised exposure is still given by Equation 2.4 that we
obtained for a unilateral agreement:
Thus, under the nave model set-up, the distinction between
unilateral agreements and bilateral agreements can be ignored: all
bilateral agreements can be treated as unilateral in the banks favour.
Margin period of risk
The model of collateral we have just discussed is very attractive
because of its simplicity. However, there is good reason why we
call this model nave, as it makes an unrealistic oversimplifying
assumption that collateral is delivered immediately after a margin
call is made.
In reality, collateral is not delivered immediately even when the
contract states daily margin call frequency. Even when the counterparty is willing to post collateral, it may take several days between a margin call and the time when the bank actually receives
collateral. However, we have to assume a longer delay: since we are
interested in exposure at default, we have to assume that the counterparty has financial difficulties and is not willing to post collateral. The counterparty may pretend to dispute the margin call, and
it may take the bank several days to realise that the counterparty is
actually defaulting rather than disputing the call. When the bank
is certain of the counterpartys default, it issues a notice of default.
Usually a notice of default has a grace period of several days. During this grace period the counterparty can still post collateral and
avoid default. Only if the counterparty does not post collateral by
the expiration of the grace period, the counterparty is officially in
default and the bank can start liquidating collateral (if it is noncash) and settling the trades with the counterparty. Settling the
trades requires their valuation, which is often done by requesting
third-party quotes on equivalent trades. The quotes can be obtained
quickly for simple trades (such as vanilla interest rate swaps), but it
may take longer to obtain quotes for more exotic trades. Only when
the portfolio is closed out does the magnitude of the banks loss due
to the counterpartys default become certain.
27
19/03/2010 17:10
The delay between the margin call that the counterparty does not
respond to and the time by which the bank is able to close out and/
or replace the counterparty portfolio is known as the margin period
of risk (MPR). Generally, it is impossible to predict how long this
delay is going to be, as it involves many uncertain factors such as
those mentioned above. In spite of this, MPR is usually specified as
a deterministic quantity in a collateral model. It is prudent to allow
different values of MPR for different margin agreements to account
for different contractual margin call frequencies and for different
levels of trade liquidity in the portfolios. For daily contractual margin call frequency and portfolios of liquid trades, a two-week MPR
is usually assumed. For less frequent margin calls, and/or more
exotic trades, longer periods should be used.
Including the margin period of risk in the collateral model
The risk that was completely ignored in the nave model is that
portfolio MTM value changes between the last successful margin
call and the time of default. Indeed, suppose that the bank closes
out the defaulting counterpartys portfolio by time t. For modelling
purposes, time t is the counterpartys default time because this is
when the banks loss on the counterparty materialises. Then, any
margin call that happened at time t is irrelevant because the defaulting counterparty is not going to respond to it. Moreover, if the
counterparty defaults at time t, we should assume that the latest
margin call which the counterparty would respond to is at time t
dt, where dt denotes the MPR appropriate for a given margin agreement. Thus, any change in portfolio MTM value between t dt and
t will not be covered by collateral.
Unilateral margin agreement
The rules for posting collateral are the same as the ones we used
in the nave model: the counterparty posts collateral when portfolio MTM value exceeds the counterpartys effective threshold Hcpt,
and the amount of collateral posted by the counterparty should be
sufficient to cover the excess. The main departure from the nave
model is that, to determine collateral available to the bank at time t,
we need to use portfolio MTM value at the earlier time point t dt.
Suppose that we know both the portfolio value V(t dt) and the
28
19/03/2010 17:10
collateral available to the bank C(t dt) at time point t dt. To obtain collateral C(t) at time t, we need to calculate the required extra collateral DC(t dt) that needs to be posted by (or returned to)
the counterparty as of time t dt and add it to C(t dt). This extra
amount is given by the difference between the collateralised portfolio value (ie, the portfolio value minus the available collateral) at
the time point t-t and the counterpartys effective threshold: DC(t
dt) = V(t dt) C(t dt) Hcpt. If the collateralised portfolio value is
above the threshold, DC(t dt) is positive, so that the counterparty
has to post this amount. If the collateralised portfolio value is below
the threshold, DC(t dt) is negative, so that the bank must return
this amount to the counterparty. However, the amount of collateral
the bank will return to the counterparty is limited by the collateral
the counterparty has already posted, ie, C(t dt). Taking this into
account, we can specify DC(t dt) as:
(2.6)
By adding the required collateral given by Equation 2.6 to the already available collateral C(t-t), we obtain the collateral available
at time t:
(2.7)
There are two important observations we can make about Equation 2.7:
o Collateral C(t) is path-independent in the sense that it does not
depend on collateral amounts at earlier time points. This is a
very useful property of the model because it allows arbitrarily
large time steps in exposure/collateral simulation to be used.
If, for instance, C(t) depended on C(t dt), then we would have
a recursive definition and would have to simulate collateral at
equally spaced time points, separated by dt. Given that exposure
is simulated over a long term (eg, 30 years), while typical values
of dt are rather small (eg, two weeks), the total number of simulation time points would be prohibitively large.
o Equation (2.7) looks strikingly similar to the nave collateral expression of Equation (2.3), which states C(t) = max{V(t) Hcpt, 0}.
29
19/03/2010 17:10
The dependence of collateral on the portfolio MTM value is indeed exactly the same as in the nave model. The only difference
between the expressions is that portfolio value now is measured
at an earlier time point. If we let the MPR dt go to zero in Equation (2.7), we will obtain Equation (2.3).

Collateralised exposure is obtained by substituting Equation 2.7
into Equation 2.2 that defines collateralised exposure:
(2.8)
where we have used the notation x+ = max{x, 0}. Collateralised exposure given by Equation 2.8 differs from the nave model result
(Equation 2.4) by an extra term added to the threshold. This extra
term, V(t) V(t dt), is the increment of the portfolio value over the
MPR. As expected, in the limit dt = 0 this term disappears and we
recover the nave model result: EC(t) = min{E(t), Hcpt}.
Let us consider a single scenario of portfolio MTM value such
that the bank receives collateral at time t. Under the nave model set-up, collateralised exposure would be exactly equal to the
threshold. In contrast, the collateralised exposure given by Equation 2.8 can be above or below the threshold, depending on the sign
of the portfolio value increment; if the path of portfolio value is
rising (falling) from t dt to t, collateralised exposure will be above
(below) the threshold.
As we discussed earlier in this chapter, exposure distribution
is usually obtained via the Monte Carlo simulation process. Trade
MTM values are simulated on a fixed set of time points {tk} (k =
1, 2, ...), which we will call the primary time points. Suppose that
we want to simulate collateralised exposure at time point tk. To be
able to do this we need to know the portfolio MTM value at two
time points: the primary point tk and the look-back point tk dt. The
short-term time points can be densely spaced, and we might have
already simulated trade values at tk dt as at one of the previous
primary time points. However, the interval between two adjacent
30
19/03/2010 17:10
time points typically increases with the simulation time. Very soon
this interval becomes larger than the MPR dt, so that the inequality
tk-1 < tk dt holds for all k -s above a certain value. Thus, to simulate collateralised exposure at a primary time point tk, we need to
include the look-back time point tk dt in the simulation set, unless the look-back point coincides with one of the previous primary
points. Then, trade values are simulated in the path-by-path manner at the extended time point set, which includes all the primary
and look-back time points. Collateralised exposure is calculated
only at the primary time points.3
Figure 2.2 illustrates the process of simulating collateralised exposure. The horizontal dashed line shows the counterpartys effective
threshold. The dotted curve shows one of the continuous portfolio
value scenarios that happens to be above the effective counterparty
threshold Hcpt. Two primary simulation time points are shown: tk-1
and tk. To obtain collateralised exposure at these time points, portfolio value needs to be simulated at four time points: both primary
points and the look-back points tk-1 dt and tk dt. These simulated
portfolio values are shown as small circles on the scenario curve.
Collateral available to the bank at the primary point is calculated
as the difference between the portfolio value at the look-back point
and the effective threshold, subject to a floor of zero. Then collateralised exposure is obtained by subtracting this collateral amount
from the portfolio value at the primary time point, subject to a floor
of zero. After following this procedure in Figure 2.2, we can observe
that the collateralised exposure is above the threshold for tk-1 and
is below the threshold for tk. This is consistent with our analysis of
Equation 2.8 above: the portfolio value is rising from tk-1 dt to tk-1
and is falling from tk dt to tk.
The reader should always remember that Equation 2.8 relates
portfolio value and collateralised exposure only at the level of a
single scenario path, and should not be applied to quantities derived from portfolio value distributions, such as expected portfolio
value, expected exposure, potential future exposure, etc. We will
discuss these derived quantities in the next section.
Bilateral margin agreement
From our discussion of unilateral agreements we have seen that
31
19/03/2010 17:10
Figure 2.2 Simulating collateralised exposure

V(tkdt)
V(tk-1)
V(tk)
V(tk-1dt)
C(tk)
C(tk-1)
EC(tk-1)
Hcpt
EC(tk)
dt
tk-1
dt
tk
collateral is determined by exactly the same rules as in the nave

model with only one difference: portfolio value that determines collateral available at the primary time point t should be measured
at the look-back time point t dt. Therefore, by replacing the primary time point with the corresponding look-back time point in the
right-hand side of Equation 2.5, we obtain collateral available to the
bank under the MPR model:
C(t) = max{V(t dt) Hcpt, 0} + min{V(t dt) Hbnk, 0}
(2.9)
As before, the first term in Equation 2.9 describes the scenarios

when the bank receives collateral (ie, C(t) > 0), while the second
term describes the scenarios when the bank posts collateral (ie, C(t)
< 0).
Let us use the scenario analysis, similar to the one we applied to
the nave model, to determine the collateralised exposure.
o V(t dt) > Hcpt. Then, from Equation (2.9), collateral is equal to
C(t) = V(t dt) Hcpt. This quantity is positive, so the bank receives collateral. Calculating collateralised exposure we obtain:
EC(t) = max{V(t) [V(t dt) Hcpt], 0} =max{Hcpt+V(t)-V(t-t), 0}
32
19/03/2010 17:10
o Hbnk V(t dt) Hcpt. Then, from Equation (2.9), collateral is equal
to zero, and the collateralised exposure equals the uncollateralised exposure.
o V(t dt) < Hbnk. Then, from Equation (2.9), collateral is equal to
C(t) = V(t dt) Hbnk. This quantity is negative, so the bank posts
collateral. Calculating collateralised exposure we obtain EC(t) =
max{V(t) [V(t dt) Hbnk], 0} = max{Hbnk + V(t) V(t dt), 0}.
Summarising these three types of scenarios in a single expression,
we can write collateralised exposure as:
(2.10)
The distribution of collateralised exposure is obtained by the

same Monte Carlo approach as for the case of the unilateral agreement. Portfolio MTM value is simulated for all the primary time
points {tk} and the corresponding look-back time points {tk dt} in a
path-by-path manner. Then, at each primary time point tk, Equation
2.10 is applied to each scenario to obtain the realisation of collateralised exposure for that scenario. After the simulation is finished,
the array of these realisations at each primary time point describes
the distribution of collateralised exposure. Derived quantities like
collateralised potential future exposure or collateralised expecting
exposure can be obtained directly from this distribution.
Earlier in the chapter, when we introduced the concept of collateralised exposure, we discussed the possibility that collateral posted by the bank to the counterparty can generate credit exposure for
the bank. However, we did not have any collateral model at that
point. Under the first collateral model that we have developed, the
nave model, posting collateral did not result in any credit exposure. However, in the more realistic MPR model, posting collateral
can result in credit exposure. This fact immediately follows from
the third row of the right-hand side of Equation 2.10: if the bank
posts collateral (ie, V(t dt) < Hbnk), credit exposure is positive when
Hbnk + V(t) V(t dt) > 0, ie, when portfolio value increase from the
look-back time point t dt to the primary time point t is greater than
33
19/03/2010 17:10
the magnitude of the banks effective threshold.

To understand the condition under which the bank can experience credit loss resulting from posting collateral, let us examine
Figure 2.3. The horizontal dashed line shows the banks effective
threshold (a negative quantity). Suppose that portfolio MTM value
at the look-back time point V(t dt) is below the threshold (shown
by the small circle on the plot). Then, collateral C(t) available to the
bank at time t is given by the difference V(t dt) Hbnk, which is
negative because the bank actually posts collateral. Conditionally
on V(t dt), portfolio value V(t) at the primary time point t can
take a range of values, shown by the vertical stripe. As we have discussed earlier in the chapter, collateral level C(t) (shown on the plot
by a short horizontal line that crosses the vertical stripe) divides the
range of possible values of V(t) into two regions: the region of no
credit exposure V(t) C(t) (the stripe is dashed) and the region of
positive credit exposure V(t) > C(t) (the stripe is solid). For portfolio
MTM value to be in the positive exposure region at time t, it must
experience an upward move from time t dt to time t such that it is
larger than the magnitude of the banks effective threshold |Hbnk|.
Figure 2.3 Credit exposure arising from posted collateral
V(t)
dt
Hbnk
C(t) = V(tdt) Hbnk
V(tdt)
34
19/03/2010 17:10
The nave model can be obtained by setting the MPR dt to zero.

In that case the entire range of possible values of V(t) degenerates
into a single point V(t dt). Since this point is below the level of
collateral C(t), no credit exposure is created by posting collateral.
COLLATERALISED EXPOSURE PROFILES
The most complete characterisation of future credit exposure
is given by its probability distribution at each future time point.
In practice, this probability distribution is usually estimated via
Monte Carlo simulation of exposure at multiple future time points.
However, for many risk management applications, a single deterministic quantity characterising exposure at a given time point is
needed. For example, to determine whether credit exposure for a
given counterparty is above its credit limit, a single number characterising the exposure would be useful. A collection of such numbers
obtained by applying the same procedure to exposure distributions
at all simulation time points is known as exposure profile. Two
types of exposure profiles are widely used in practice: potential future exposure (PFE) and expected exposure (EE). Potential future
exposure profile is obtained by calculating a high-confidence-level
percentile (eg, 95%) of exposure at each simulation time point. Expected exposure profile is obtained by calculating the sample mean
of the simulated exposure realisations at each simulation time point.
In this section we will apply the collateralised exposure framework that we have developed to calculate collateralised PFE and
EE profiles for several simple examples. We will see how the collateralised exposure profiles differ from the non-collateralised ones
and examine how they depend on the MPR and on the level of the
thresholds.
Examples
We will build exposure profiles for two sample trades over a fixed
time period of five years.
Forward-starting swap
The first trade is a forward starting five-year interest rate payer
swap that starts in five years. A distinctive feature of this trade is
that no cashflows occur during our five-year period of interest. Fig35
19/03/2010 17:10
ure 2.4 shows profiles of non-collateralised PFE at 95% confidence

level and non-collateralised EE for this trade in the units of swap
notional. These profiles start at zero (the swap is at-the-money at
inception) and are monotonically increasing with time, indicating
that the exposure distribution is getting wider as it moves further
into the future. This behaviour is explained by the simple fact that
more uncertainty is associated with a more distant future. Such
profiles are characteristic of any trade/portfolio that does not have
cashflows in the given time interval.
Figure 2.4 Non-collateralised PFE (at 95%) and EE profiles for
forward starting swap
Immediately-starting swap
The second trade is an immediately starting five-year interest rate
payer swap. Figure 2.5 shows profiles of non-collateralised PFE at
95% and non-collateralised EE for this trade. The profiles now look
very different: they start at zero, rise to a maximum, and then fall
to zero again. While the uncertainty effect responsible for the rising of the exposure is still present, there is another effect, known as
amortisation, that acts in the opposite direction. As we consider
36
19/03/2010 17:10
more distant future time points, fewer cashflows remain between

those time points and the trade maturity. Fewer future cashflows
means less uncertainty, so that the width of the trade value distribution is reduced, and exposure profiles come down.
Figure 2.5 Non-collateralised PFE (at 95%) and EE profiles for
an immediately starting swap
Margin agreements change exposure profiles dramatically. We will

consider margin agreements with three levels of effective threshold: 0.5%, 1.0% and 2.0% of the swap notional. To illustrate how
these thresholds relate to the levels of uncollateralised exposure,
the thresholds are shown by dashed horizontal lines in Figures 2.5
and 2.6.
Collateralised potential future exposure
Figure 2.6 shows collateralised PFE at the confidence level of 95%
for the forward starting swap under three unilateral margin agreements that differ by the value of the effective threshold. To make
the visual analysis easier, we used the same scale for all three pan37
19/03/2010 17:10
els that represent different threshold values. Three curves on each

panel correspond to three different values for the MPR: zero, two
weeks (2W) and one month (1M). The PFE profiles for bilateral
agreements are not shown because they coincide with the unilateral
profiles for this example. In most cases this will be the case because
losses arising from posting collateral are relatively small, so that
they do not contribute to the tail of the exposure distribution. However, as we will see below, these losses will affect the collateralised
EE profiles.
Figure 2.6 Collateralised PFE at 95% confidence level for the forward
starting swap
The PFE profiles in Figure 2.6 start rising quickly, but come to
a plateau soon after they cross the threshold. The profile for zero
MPR stays exactly at the threshold, as we know that under the nave model collateralised exposure cannot go above the threshold.
The amount by which the other two curves exceed the threshold is
nearly the same for all three values of the threshold. The one-month
curve is always above the two-week curve.
38
19/03/2010 17:10
Figure 2.6 (Continued)
39
19/03/2010 17:10
We can understand this behaviour if we examine Equation 2.8:
EC(t) = min E(t),[Hcpt+V(t t,t)]
where we have introduced the notation

for the increment of the portfolio value over the MPR. The tail of the
distribution of EC(t) is determined by the scenarios with large E(t)
and large dV(t dt,t). For the cases of low and moderate thresholds,
large E(t) is usually larger than large Hcpt + dV(t dt,t). Since the
lesser quantity defines collateralised exposure, the tail of EC(t) is
usually determined by the tail of dV(t dt,t).
Now we can make the following conclusions:
o The time interval dt for the portfolio value increment is the same
for all t. Therefore, in the absence of cashflows, the tail of dV(t
dt,t) does not change much with time, and the collateralised PFE
profile is relatively flat.
o Since the distribution of dV(t dt,t) does not depend on the level
of threshold, the amount by which the collateralised PFE profile
exceeds the threshold does not depend on the threshold either.
Because of this, the performance of the nave model does not
improve as the threshold is getting larger.
o The distribution of dV(t dt,t) is always wider for larger values
of dt because of more uncertainty associated with larger times.
Therefore, the larger is the assumed MPR, the higher above the
threshold the collateralised PFE profile will be.
Figure 2.7 shows collateralised PFE at the confidence level of 95%
for the immediately starting swap under margin agreements with
the same three values of the effective threshold. As before, the
curves on each panel correspond to the MPR of zero, two weeks
(2W) and one month (1M) and the same scale is used for all three
plots.
All the effects that we have discussed for the case of the forward
starting swap are still present in Figure 2.7. The key difference between Figures 2.6 and 2.7 is that the PFE curves in Figure 2.7 quickly reach a maximum above the threshold and then start declining
towards the threshold instead of staying on a plateau.
40
19/03/2010 17:10
Figure 2.7 Collateralised PFE at 95% confidence level for the

immediately starting swap
41
19/03/2010 17:10
It is not difficult to understand why this happens. As we have discovered for the case of the forward starting swap, the tail of the
collateralised exposure distribution EC(t) is usually determined by
the tail of the portfolio value increment dV(t dt,t). However, for
the immediately starting swap, the distribution of dV(t dt,t) is
getting narrower as the simulation time progresses because fewer
cashflows remain until the swap maturity. This is the same amortisation effect that is responsible for the shape of the uncollateralised
PFE profile of the immediately starting swap. However, for the collateralised PFE, the amortisation effect starts dominating the uncertainty effect much sooner because the uncertainty time does not
increase with t in this case instead, it remains fixed at the MPR dt.
Collateralised expected exposure
Figure 2.8 shows collateralised EE for the forward starting swap
under margin agreements with the same three values of the counterpartys effective threshold as above. As before, the curves on
42
19/03/2010 17:10
each panel correspond to the MPR of zero, two weeks (2W) and
one month (1M) and the same scale is used for all three plots. However, there are now five curves on each plot this time. The extra
two curves show the collateralised EE profiles for bilateral margin
agreements (BL on the plots), which are in general different from
the unilateral EE profiles (UL on the plots). Apart from the counterpartys threshold, bilateral agreements also specify the banks
threshold. In our examples of bilateral agreements we have set the
banks threshold equal to the counterpartys threshold (eg, the bilateral profiles on panel (a) are for a margin agreement with both
the counterpartys and the banks effective thresholds equal to 0.5%
of the swap notional).
Similarly to the collateralised PFE profiles, the collateralised EE
profiles in Figure 2.8 start rising quickly, but come to a plateau. Another common feature is that the one-month curve is always above
the two-week curve.
Figure 2.8 Collateralised EE for the forward starting swap
43
19/03/2010 17:10
44
19/03/2010 17:10
However, there are several important features of collateralised

EE profiles that are different from the collateralised PFE profiles of
Figure 2.6:
o For almost all examples, the plateau is below the counterpartys
effective threshold. In fact, for zero MPR, this is always the case.
Indeed, collateralised EE is the simple average of all simulated
collateralised exposure scenarios. When the MPR is zero, any
collateralised exposure scenario is bounded by zero from below
and by the threshold from above, so that the average is also between zero and the threshold. For non-zero MPR, collateralised
exposure can go above the threshold, but the excess over the
threshold is determined by the portfolio value increment over
the MPR, which cannot be very large due to the small magnitude
of the MPR. Unless the threshold is very small or the time zero
portfolio MTM value is very large, the contribution from the scenarios above the threshold is not sufficient to bring the average
above the threshold.
o Collateralised EE profiles for bilateral agreements are always
above the ones for unilateral agreements that have the same
counterpartys threshold. Scenarios where the bank receives collateral are exactly the same for a unilateral agreement and a bilateral agreement that have the same counterpartys threshold.
Therefore, these scenarios contribute exactly the same amount to
the average. However, scenarios where the bank posts collateral
may result in non-zero credit exposure for the bilateral agreement, while they produce no credit exposure for the unilateral
agreement. These scenarios produce extra contribution to the average for the bilateral margin agreement, thus resulting in higher
collateralised EE.
o The excess of collateralised EE for the bilateral agreement over
the one for the unilateral agreement is always larger for larger
MPR. This excess is due to portfolio value scenarios that first go
below the banks threshold (the bank posts collateral) and then
rise by more than the threshold magnitude over the MPR (positive exposure for the bank is created). For larger MPR, there is
more chance for the portfolio value to rise above the same fixed
amount. Thus, the number of scenarios where the bank posts
collateral is the same for any MPR, but the percentage of these
45
19/03/2010 17:10
scenarios that result in non-zero exposure is larger for larger MPR.

These extra scenarios produce higher averages for larger MPR.
o The excess of collateralised EE for the bilateral agreement over
the one for the unilateral agreement quickly decreases as the
magnitude of the banks threshold becomes larger. As the magnitude of the banks threshold becomes larger, fewer scenarios
result in the banks posting collateral because the portfolio value
must fall lower to cross the banks threshold. Moreover, a smaller
percentage of these scenarios result in non-zero credit exposure
because this requires the portfolio value to rise by more than the
banks threshold magnitude over the same short time period,
equal to the MPR.
o Including non-zero MPR in the model becomes less and less
important as the magnitude of the counterpartys threshold increases. Indeed, the difference between the collateralised EE profiles for margin agreements with non-zero MPR and zero MPR
in Figure 2.8 becomes smaller as the counterpartys threshold
becomes larger. This behaviour is in striking contrast with the
collateralised PFE profiles, where this difference did not change
much as the threshold increased. The primary effect that contributes to this behaviour can be understood from Equation 2.8 (for
unilateral agreement) or Equation 2.10 (for bilateral agreement).
If we consider only the scenarios where the bank receives colassocilateral, both equations produce the same exposure
ated with those scenarios:
When the counterpartys threshold Hcpt is small in comparison to

the range of potential changes of the portfolio MTM value over
the MPR, the zero floor skews the average over all collateral-receiving scenarios upward. As the threshold becomes larger, the
effect of the zero floor on the average decreases. As the threshold
becomes significantly larger than the standard deviation of dV(t
dt,t), the effect of the zero floor can be practically neglected, and
the average of the collateralised exposure over the collateral-receiving scenarios becomes approximately equal to the threshold
plus the average increment of portfolio value. The latter term is
often quite small, so that the contribution of the collateral-receiv46
19/03/2010 17:10
ing scenarios to the collateralised EE is approximately equal to

Hcpt for large values of the threshold.
Now the reader should recall that under the nave model of zero
MPR, any scenario where the bank receives collateral results in exposure equal to the counterpartys threshold. Thus, for large counterpartys thresholds, collateralised EE profile does not depend significantly on the MPR, and the nave model can produce rather
accurate collateralised EE.
Figure 2.9 shows collateralised EE for the immediately starting
swap under unilateral and bilateral margin agreements with the
same values of the effective threshold as for the case of forward
starting swap. As before, the curves on each panel correspond to
the MPR of zero, two weeks (2W) and one month (1M) and the
same scale is used for all three plots.
Figure 2.9 Collateralised EE for the immediately starting swap
47
19/03/2010 17:11
48
19/03/2010 17:11
All the effects that we have discussed for the case of the forward
starting swap are still present in Figure 2.9. The key difference between Figures 2.8 and 2.9 is the shape of the curves, which is entirely due to the amortisation effect we have discussed for the collateralised PFE profiles. For example, we can easily explain why
the two-week and one-month MPR profiles approach the zero MPR
curve as the simulation time increases. Since fewer cashflows remain until maturity for larger simulation times, the standard deviation of the portfolio value increment dV(t dt,t) declines with t. As
we have discussed for the case of forward starting swap, smaller
standard deviation of dV(t dt,t) in relation to the threshold diminishes the dependence of the collateralised EE on the MPR. Thus,
as the simulation time passes, the collateralised EE profile for any
fixed MPR is getting closer to the zero MPR profile.
SUMMARY
Collateralisation is a very important risk management tool. It allows a financial institution to greatly reduce its credit exposure to
a counterparty and keep the exposure under control. However, because of the uncertainty inherent in counterparty credit exposure,
collateralisation in the context of OTC derivatives is much more
complex than it is in the context of lending risk both operationally and from the modelling perspective. While we have briefly addressed the operational challenges, the main focus of this chapter
has been on modelling. We have shown the current state of the art
in modelling collateralised credit exposure for both unilateral and
bilateral margin agreements. We have discussed the critical role
that the margin period of risk plays in these models. Using exposure profiles for several simple examples, we have illustrated the
main properties of collateralised credit exposure.
The opinions expressed here are those of the author, and do not necessarily reflect the views or policies of the Federal Reserve Board or its staff.
1
2
ISDA is a global trade association that represents leading participants in the OTC derivatives industry.
Our description of exposure simulation process is very brief. For more details, the reader
is encouraged to go to Canabarro and Duffie (2003), De Prisco and Rosen (2005) or Pykhtin
and Zhu (2007).
While the full Monte Carlo approach for collateralised exposure is very flexible, its clear
disadvantage is that for collateralised counterparties the simulation time is doubled due
49
19/03/2010 17:11
to inclusion of the look-back time points. The only way to avoid this doubling is to use an
for calculating collateralised expected exposure (EE) that does not require portfolio value
simulation at the look-back time points. Gibson (2005) has calculated collateralised EE analytically, without any exposure simulation at all, but the price he had to pay was the assumption that portfolio MTM value at any time point is normally distributed with a known mean
and standard deviation.
REFERENCES
Canabarro, E. and D. Duffie, 2003, Measuring and Marking Counterparty Risk,

in L. Tilman, (ed.), Asset/Liability Management for Financial Institutions, (London: Institutional Investor Books).
De Prisco, B. and D. Rosen, 2005, Modelling Stochastic Counterparty Credit Exposures for Derivatives Portfolios, in M. Pykhtin, (ed.), Counterparty Credit Risk
Modelling, (London: Risk Books).
Gibson, M., 2005, Measuring Counterparty Credit Exposure to a Margined Counterparty, in M. Pykhtin, (ed.), Counterparty Credit Risk Modelling, (London: Risk
Books).
Pykhtin, M., 2009, Modeling Credit Exposures for Collateralized Counterparties,
The Journal of Credit Risk, 5(4) pp. 3-27.
Pykhtin M. and S. Zhu, 2007, A Guide to Modeling Counterparty Credit Risk,
GARP Risk Review, July/August, pp. 1622.
50
19/03/2010 17:11
Efficient Calculation of Counterparty

Exposure Conditional on Default
David M. Rowe, Philip Koop and Daniel Travers
SunGard
Discussions of counterparty credit have long recognised that it is a

dual contingency risk. An institution experiences exposure to potential credit loss when a counterparty loses money on its bilateral
trades with the institution. This creates positive mark-to-market
values on the institutions balance sheet reflecting the present value
of money owed by the counterparty. Actual realised losses (credit
risk rather than just credit exposure) require that the counterparty
default after having experienced losses on its bilateral portfolio of
trades with the institution.
If the likelihood of default is statistically independent of the market events giving rise to exposure, it is valid to calculate expected
credit losses by simulating expected exposure separately and then
multiplying by the likelihood of default. Unfortunately, this is not a
valid approach if default and exposure are correlated (either positively or negatively) with each other. This chapter describes a means
of simulating exposure conditional on default of a counterparty. The
expected value of such conditional exposure can then be multiplied
by the probability of default to derive a statistically valid estimate of
expected loss even when exposure and default are correlated. This
makes derivation of expected loss no more computationally burdensome than a standard unconditional exposure simulation.
INTRODUCTION
Simulation of credit exposure to derivative counterparties dates
back to the mid-1980s. It was initially prompted by the need to in51
19/03/2010 17:11
clude a provision in the Basel I Capital Accord for potential credit

losses on banks rapidly growing derivative market-making activities. Early analysis focused on individual interest rate swaps and
foreign exchange transactions, and was designed to provide an empirical basis for the parameters in what came to be known as the
mark-to-market plus add-on approach. This type of calculation was
originally designed as a simple way for banks to derive their total
potential credit exposure to all derivative counterparties. An unfortunate effect of implementing this approach in the Basel I rules was
that nearly all banks proceeded to implement a similar (although
often more conservative) calculation for measuring and controlling
potential credit exposure to individual counterparties. Arguably,
this delayed serious efforts to develop more realistic exposure measurement techniques for many years.
In the early days, application of netting was not legally certain,
even in major G7 economies. Thus, the simple addition of assessments for single transactions did not introduce a serious distortion.
As netting became more widely recognised in legislation and case
law precedents, some means of reflecting this risk-reducing phenomenon in credit assessments became more urgent. Even then,
however, this often took the form of some easing of the parameters
in the mark-to-market plus add-on approach, rather than developing a full-blown simulation process.
A few banks began work on exposure simulation systems in the
early 1990s.1 Such efforts were necessarily hampered by limitations
on the computing capacity that was commercially practical to devote to such systems. The focus was primarily on making significant
improvements to the mark-to-market plus add-on approach, which
was quite a modest goal given the well-known shortcoming of that
technique. In general, the philosophy was to err on the side of conservatism so that simulations would not understate exposure versus
approved limits. Achieving significantly greater sophistication was
deemed commercially impractical and of marginal importance.
Much has changed in the past 20 years, both in the derivative
market itself and in the technology available to support related
risk systems. The biggest change in the derivative market has been
the explosion of transactions for which some form of credit risk is
the primary variable driving the value. Single-name credit default
52
19/03/2010 17:11
Efficient Calculation of Counterparty Exposure Conditional on Default
swaps and a vast array of structured credit securities, both primary

and synthetic, have introduced a new dimension of complexity to
valuation and to counterparty exposure simulation. More recently,
the crisis in financial markets that erupted in 2008 sharpened the
recognition of counterparty risk as a very real threat that must be
taken seriously. This has generated increased demand for more sophisticated exposure measures and a desire to price for counterparty credit risk in the normal course of daily trading activity.
Fortunately, some advances in technology since the turn of the
century are well suited to supporting these aspirations for counterparty credit risk analysis that is both timelier and more sophisticated. The two advances with the most to offer are:
o grid computing that enables massive parallel processing; and
o 64-bit architecture that supports a massive expansion in addressable memory capacity.
While these are not extremely recent innovations, they have only
become proven, commercially attractive, mainstream technologies
in recent years. Furthermore, as with all major advances in hardware architecture, pre-existing software has had to undergo significant revisions to take full advantage of the new possibilities. The
combination of these hardware advances and progressive software
adaptation has created dramatic new opportunities for improved
management of counterparty credit risk. For the first time it is commercially feasible to deploy incremental Monte Carlo simulation as
a means of assessing the exposure impact of a proposed new deal
sufficiently rapidly to provide trading decision support. This is essential if traders are going to be charged for the incremental expected credit default losses implied by their trades credit valuation
adjustment (CVA) and for the capital to provide a cushion against
potential unexpected credit losses. This chapter focuses on an efficient derivation of the exposure at default (EAD) appropriate for
input into the economic capital calculation which is sensitive to nonzero correlation between exposure and the probability of default.
WRONG-WAY AND RIGHT-WAY RISK
Wrong-way credit risk burst into market consciousness in the midst
of the Asian currency crisis of 199798. Many Asian corporations
53
19/03/2010 17:11
had borrowed in G7 currencies, most often US dollars or yen. If

their revenues were primarily in domestic currency, this presented a
major foreign exchange risk. Depreciation of these borrowers local
currency would significantly increase their effective debt burden. To
alleviate this contingency, many such corporations hedged their foreign exchange risk in the currency swap market. Furthermore, they
often executed these transactions with Western banks. Having executed these trades with an Asian corporation, the Western bank now
held a sizable open foreign exchange position that they needed to
hedge. In looking for suitable professional counterparties, the most
active players often would be money centre banks in the original
borrowers home country. In executing such a hedge, however, the
Western banks took on classic wrong-way risk. The Western bank
was paying the Asian currency and receiving its own currency from
an Asian bank. When several Asian currencies depreciated dramatically, the value of these trades moved decisively in favour of
the Western banks just as the local economies of their counterparty
banks were falling into crisis. The local turmoil resulted in several
Asian bank failures and consequent credit losses for the Western
bank counterparties to these significantly in-the-money trades.
As always, experience was a harsh but effective teacher. Since the
derivative credit losses stemming from the largely inadvertent buildup of wrong-way risk prior to the Asian currency crisis, the correlation between exposure and default probability has remained a lively
topic for debate. Since most end-users are primarily attempting to
hedge fundamental business risks most of the time, there should be
a strong presumption that most derivative positions represent rightway risk. An example is an airline entering into a swap where they
pay fixed and receive a floating payment based on the future price
of jet fuel. They lose money on the swap when jet fuel prices fall,
but this price decline tends to increase the profitability of their core
business, making them a better credit risk. Even though most swaps
and other derivatives are used to hedge fundamental business risks,
the Asian crisis and its associated losses drove home the realisation
that this cannot be taken for granted in any given bilateral portfolio.
IMPLICATIONS FOR CALCULATION OF ECONOMIC CAPITAL
In general, we think of expected credit losses as the product of the
probability of default times expected exposure (adjusted for expect54
19/03/2010 17:11
ed recoveries). This is quite correct if exposure is static and predetermined. It also is correct if we can assume that random exposure
is uncorrelated with the probability of counterparty default. As noted in the previous section, however, such statistical independence
between exposure and default will tend to be rare. In most cases it
is to be expected that exposure is negatively (ie, favourably) correlated with the likelihood of default. If a counterparty is hedging
a fundamental business risk, the bank will tend to make money
on their bilateral trades when the fundamental economic factors
are favourable to the counterpartys business, thereby reducing the
probability of default. In other, probably less likely but potentially
important, situations the toxic combination of rising exposure and
rising default probability needs to be recognised.
In calculating unexpected losses used for deriving economic
capital, it is similarly important not to assume the independence
of exposure and probability of default, as these unlikely but toxic
cases of wrong-way risk may have an even greater impact on the
tail losses driving trading book economic capital than on the level
of expected loss. In the discussion that follows, we first consider
credit defaults as the only driver of economic capital. At the end we
briefly describe how the method we propose could be extended to
reflect the impact of credit rating transitions short of default.
The usual process for dealing with the correlation of exposure
and default is to include a credit quality variable in the simulation,
and to correlate this with all the market variables that drive the
value of trades in the bilateral portfolio. The credit quality variable
is then evaluated against a default threshold. For every time point
in every simulation, it is then possible to derive the value of the
exposure and a binary default/no default indicator for the status
of the counterparty. By generating a sufficient number of paths, it
is possible to determine what proportion of the time a default occurs and the distribution of exposure across those instances when it
does. The obvious problem with this brute force approach is that
it is massively wasteful of computing resources. In the overwhelming majority of instances, for all but the shakiest of counterparties,
there will be no default. In these cases the exposure amount is irrelevant to the credit calculation since no credit loss has occurred.
Of course, it is possible to simulate the full set of market drivers,
55
19/03/2010 17:11
including the credit quality variable, and only calculate the implied
exposure if a default has occurred. Even so, given the rarity of a
default it would often be necessary to generate a million or more
full market scenarios to obtain as few as a thousand instances of default. Our proposal is to take this thought process a step further and
simulate the drivers of the portfolio value conditional on the credit
quality variable being in default status at a specified time horizon.
This approach encounters problems if the desire is to generate a full
life of the portfolio analysis since, in such an approach, the requirement is to capture those situations where default occurs at a given
time while not having occurred prior to that time. Nevertheless,
for a single period analysis (say at a one-year horizon, commonly
used for calculating economic capital) simulating exposure conditional on default can lead to a significant reduction in the necessary
amount of computation.
THE SIMULATION FRAMEWORK
The analytics behind this approach are as follows. Let the scalar
V(t) be the future value of a portfolio of securities as a function of
time. For portfolios of practical interest, V(t) is unknown and a fundamental risk management problem is to estimate its distribution.
A common solution to this problem is a Monte Carlo simulation in
which V(t) is calculated as a function v of a vector of realisations
of a set of stochastic market price processes p(t). These processes
are generally modelled mathematically as continuous functions but
in the simulation can be realised only at a set of n discrete times
t1, t2, tn. For simplicity, we assume that this set of times is the
same for each process. Each Monte Carlo scenario s consists of an
equally probable vector of discrete process realisations ps(t) and a
corresponding evaluation of Vs(t) = v[ps(t)]. To lighten the notation,
the remaining discussion is in the context of a single Monte Carlo
scenario. Also, we do not assume any particular form of information filtration, though most implementations operate in the natural filtration.
In the framework we are considering, an individual price process
P(t) may be of a form which could comprise diffusive, jump and
complicated drift components, but all randomness is derived from
a vector z of correlated standard normal variables generated at
56
19/03/2010 17:11
each simulation realisation time. In other words, the value of each

process realisation at the kth simulation time P(tk) can be written as
a function of zk and the realisation of p(t) at previous simulation
times. (The size m of z may be either smaller or larger than the size
of p. On the one hand, some processes may have more than one
stochastic driver; on the other, some stochastic drivers may be systemic and be shared among several processes.) We assume that the
correlation matrix of z is the same at each simulation time.
In this framework, the form of the individual price processes is
quite general, but their possible joint distribution is constrained by
the underlying multivariate Gaussian distribution. Despite this limitation, the framework is justly popular because of its relative tractability with respect to calibration and implementation. It also is amenable to various extensions, one of which we describe in this chapter.
CONDITIONAL SIMULATION
Suppose that at a given simulation time, the m normal variables of z
are partitioned into subsets z1 and z2 of sizes q and m - q respectively:
where z2 is fixed in advance. The distribution of z1 conditioned on z2

is the conditional multivariate normal distribution. In general terms
(if the variables of z do not have identical, standard normal distributions), when z is characterised by the mx1 mean vector and the
mxm covariance matrix then, conditional on fixed values Z2 of the
stochastic vector z2, z1 is a multivariate normal distribution characterised by the qx1 mean vector q and the qxq covariance matrix q.
If we partition the parameters of the unconstrained stochastic
process such that
with dimensions
[qx1],
[qxq], [qx(m-q)]
[(m-q)x1], [(m-q)xq], [(m-q)x(m-q)]
then the parameters of the constrained stochastic process for z1
57
19/03/2010 17:11
conditional on the realisation Z2 of the stochastic vector z2 can be

derived as
(Note that, in general, neither nor 22 is positive definite; consequently, the inverse of 22 must be represented by some approximation such as the Moore-Penrose pseudo-inverse.) The significance
of the conditional multivariate normal distribution is that it is not
necessary to generate complete scenario realisations of all market
processes if we want to restrict our attention to scenarios for which
some subset of the processes p(t) (lets refer to this subset as r(t))
exhibits a particular property. It is sufficient to generate candidates
of r(t) alone because we can fill-in the complete scenario by conditioning the remaining normal draws driving the simulation from
the subset of z2 that generates acceptable values for r(t).
CONDITIONING APPLIED TO DEFAULT
In the setting we have described, the processes r(t) represent the
credit status process of the counterparty, and we are interested in
realisations of r(t) that imply counterparty default. We have stated
earlier that in this chapter we restrict our attention to short simulation horizons for which a single period default model is suitable.
From this perspective, the credit status process can have only two
outcomes, namely default and survival. Suppose, then, that we
choose a credit status process in which only the terminal value is
relevant in deciding default. We will not have a default time to
guide us in choosing a particular exposure value in the simulation
scenario to use as the exposure at default. For regulatory capital
this would be consistent with the use of so-called effective exposure, in which exposure is floored at its previous maximum value,
as the measure of exposure at default. For economic capital purposes, another measure of exposure, such as a simple or a weighted
average over the one-year time horizon, may be considered.
Why are we interested in a complete realisation of the credit status process if its terminal value is the only thing that determines
default? The reason is that exposure may be path dependent if it
arises from some path-dependent valuation feature such as a bar58
19/03/2010 17:11
rier event or an exercise decision, as occurs in a physically settled

swaption. We would therefore like to condition complete paths of
market realisations on counterparty default.
The attraction of a terminal model of default is that it can be
used to generate market scenarios that incorporate counterparty
default very efficiently. Let the credit status process be a standard
Wiener process W(t) with W(0) = 0. The conditioning vector z2 is
therefore of length one, consisting of just one normal draw at each
simulation time. The probability of default is an exogenous input
to the model; the appropriate probability to use is the probability
PD(T) that the time of default D is less than or equal to the simulation horizon T. The interpretation of the default process is that default occurs if the terminal value W(T) is below the threshold value
implied by the given default probability. Since the Wiener process
has a normal marginal distribution, this means that default is associated with the following terminal values of W
For our purposes, the Wiener process has two important properties. The first is that a Wiener process that is pinned at its endpoints
can be randomly filled in using the Brownian bridge. The second
is that the sequence of normal draws that would produce a discrete
realisation of a Wiener process can be deduced from the realisation
itself. We use these properties as follows. We pin W(T) randomly in
the default region by applying a standard uniform variable u to the
formula for the terminal distribution
Having pinned W(0) and W(T), we construct a complete discrete
realisation of W(t) from a sequence of independent standard normal variables X1, X2, , Xn by applying the Brownian bridge
Note that the sequence of random variables x is not the same as

the sequence z2 we require to condition the complete market scenario. That is because the increments of a Brownian bridge process
59
19/03/2010 17:11
are different from the increments of a Wiener process when considered as a function of the random normal drivers. To condition the
scenario, we must interpret a given sequence of Brownian bridge
increments as Wiener process increments. The required sequence
of z2 is the one that would have produced the bridged realisation
of W(t) had it been generated by an unconditioned simulation. But
this can be deduced from the increments of W
In summary, we can efficiently generate a default scenario by applying the inverse normal distribution function to a standard uniform variable to generate a random W(T), applying the Brownian
bridge to this value to construct a complete default process realisation W(t), and calculating the sequence of conditioning normal variables z2 that is implied by this realisation. The simulated realisation
of p(t) that is conditioned on z2 using the conditional multivariate
normal distribution will be consistent with counterparty default to
the extent allowed by the correlation framework of the simulation.
ECONOMIC CAPITAL CALCULATIONS
Outside the trading book context, unexpected credit losses used in
deriving economic capital are driven by variations in default rates.
Exposure is treated as a pre-determined input to this stochastic
process. One common procedure for treating market-driven counterparty exposure is to take expected exposure from unconditional
simulations as the static input to the economic capital model. As
described above, however, this fails to account for the consistent
presence of correlation (positive of negative) between the level of
exposure and the probability of default. In this sense, expected exposure conditional on default is a much more defensible static input to the simulation of default losses across the full credit portfolio
than is unconditional expected exposure.
Of course, this still leaves an open question. We have derived
simulations for multiple paths of exposure for discrete points over
the one-year time horizon conditional on the modelled value of the
credit status variable pinned in the default range at one year. From
60
19/03/2010 17:11
this we can extract an expected exposure at each simulation point.

We still need to decide how to transform this vector of conditional
expected exposures into a fixed input to the economic capital model. Remember that we have constrained the credit status indicator
in a way that guarantees default status sometime in the following
year, rather than at a specific point in time. This argues for taking
some type of average over the year as our input rather than the terminal value at one year, even though one-year forward is the only
point where default status of the credit variable is guaranteed. Using the terminal value also would have the unfortunate property of
showing zero expected exposure at default when all trades run off
during the year, even though default could occur before year end.
The spirit of Basel II would argue for using the average over time
of effective expected exposure, where each periods value is floored
at the previous maximum. This approach is based on the assumption of continuing replacement business as existing transactions
mature. For economic capital, this violates the spirit of most counterparty exposure calculations, which are premised on analysing
a static portfolio with existing trades ageing and running off over
time. A seemingly attractive option would be to evaluate the values
of the credit status variable based on the Brownian bridge against
the default threshold at each point over the year. It then would be
possible to take the exposure at the point the credit indicator first
crossed the default threshold for each simulation and use the average of these exposures as input to the economic capital model. One
argument against this is that tractable diffusion processes like the
one used here are notoriously poor at reaching the default barrier at
early times as frequently as we observe in the real world. With the
process we propose, the simulated default times would be strongly
biased toward the simulation horizon. While there is no obviously right answer for how to derive a single conditional expected
exposure value from our results, a simple average of the expected
exposures across all simulation points out to one year seems like a
reasonable approach.
EXTENSION TO INCLUDE CREDIT TRANSITIONS
Clearly not all credit losses are driven by default, and transition
losses feature significantly in recent thinking by the regulators. To
61
19/03/2010 17:11
include such losses in economic capital calculations, the technique

in this chapter can be extended in a straightforward manner to include transitions. Given the probability of transition to a given state
over a one-year horizon, we can determine the range within which
the credit status variable at one year must lie to be consistent with
this transition. Pinning the terminal value of the credit status variable in that range and following the logic outlined will give exposure paths conditional on this transition. One drawback is that this
process requires repeating the exposure simulation for each terminal transition state we wish to consider. This still must be massively
more efficient, however, than a brute force approach.
1
Notable among these banks were Citibank, Bankers Trust, Morgan Guarantee Trust and
Bank of America.
62
19/03/2010 17:11
Effective, Enterprise-wide
Collateral Management
Darren Measures
JP Morgan
Credit risk exists whenever one counterparty has a future, or potential future, cashflow or securities obligation to another. The primary
risk, of course, is that the party that owes that payment will not be
able to meet their obligation. Collateral has historically proven to
be an effective mechanism to mitigate that risk. This is not because
collateralisation changes the likelihood of a default, nor impacts the
value of a defaulted transaction. Instead, collateralisation works to
reduce credit risk by offsetting some, or all, of the loss should a
default occur.
In essence, collateralisation involves the transfer of credit risk
to operational, legal, settlement, liquidity, correlation and reputation risk. An institutions ability to successfully mitigate credit risk
through the use of collateral, in other words, is dependant upon its
ability to effectively handle the operational requirements of managing that collateral.
Long utilised in repo and securities loan transactions, collateralisation has grown exponentially with rapid rise of over-the-counter
(OTC) derivative trading. Both the number of collateralised OTC
trades and the total volume of collateral in circulation increased
steadily between 2003 and 2008. According to a survey by the International Swaps and Derivatives Association (ISDA), the percentage
of trades collateralised rose from 30% to 59%, while the collateral
volume grew by 60%. This trend rapidly accelerated in 2009, when
market events increased the focus on counterparty risk. Estimated
63
19/03/2010 17:11
collateral in circulation will increase by 86% this year to US$4 trillion (greater than the GDP of Germany). While this may well represent a near-term peak, an upward drift in the use of collateral is
expected.
Figure 4.1 Growth of value of total reported and estimated collateral 200009 (US$m) (2009 reported number is approximate)
Source: ISDA
Beyond a simple growth in volume, the use of collateral has grown

more complex as a wider range of trading instruments was introduced, and the need to monitor credit risk concentration was
brought into sharp relief by the recent banking crisis. Traditionally
the province of sell-side institutions, which used collateral management to mitigate credit exposure to less creditworthy counterparties, collateral management has become a primary concern of the
buy side. Needing to understand the full credit exposure to any
given counterparty across all traded instruments, buy-side institutions have joined their sell-side counterparts in the search for an
efficient and effective enterprise-wide collateral management solution. At that point, collateral management truly becomes a competitive trading advantage.
In this chapter, we will explore some of the difficulties of developing bilateral collateral management systems that span fixed income, credit and equity markets, and discuss some of the methods
and resources that are being developed to meet this need. First, it
is important to understand how the use of collateral has evolved
to mitigate risk in three main trade components: OTC derivative,
repo and securities lending. We will look at how collateral is used
to reduce risk in each of these markets in isolation, after which we
64
19/03/2010 17:11
Effective, Enterprise-wide Collateral Management
will discuss why looking at these markets in isolation is the very

last thing the financial industry should be doing.
COLLATERAL USE IN OTC DERIVATIVES TRADING
OTC derivatives are highly flexible financial instruments used in
a broad range of investing and hedging strategies. One of the fastest-growing sectors of the financial markets throughout the 1990s
and 2000s, OTC derivative trading reached US$415 trillion by 2006.
Huge increases in volume drove a corresponding increase in counterparty credit risk.
By using collateral, OTC derivative traders are able to reduce
their exposure to credit losses should their counterparty default on
their obligations. This allows them to expand their OTC derivative
trading activities while adhering to their capital requirements. As
collateralisation reduces credit risk and credit losses, it stands to
reason that credit reserves can be reduced as collateral is deployed.
This applies to both economic capital requirements set by management to act as a buffer against credit losses and regulatory capital
thresholds set by regulators.
The use of collateral:
o liberates credit lines to engage in more transactions with highly
creditworthy counterparties (ie, short-term credit rate swaps between leading dealers);
o reduces credit charges (spreads) by mitigating the risk of default;
o enables parties to enter into agreements where future exposures
are less certain (ie, very long-dated interest rate swaps); and
o allows institutions to enter into agreements with leveraged counterparties and unrated institutions (such as hedge funds).
THE EVOLUTION OF OTC COLLATERAL MANAGEMENT PROCESSES
In the early stages of the OTC derivative market, dealers applied a
technique called close-out netting to mitigate their overall credit
exposure to counterparties. (The provisions for this were defined
by the ISDA Master Agreement of 1992.) In the event of a default,
the non-defaulting counterparty was allowed to accelerate and terminate all transactions, and net their market value. This resulted in
a single sum that was owed by one counterparty to another. These
65
19/03/2010 17:11
practices provided some measure of credit risk mitigation, and the

general principles are still in use today. Close-out netting, however,
was an incomplete solution, and even carried questions concerning
the legal enforceability of such clauses in different jurisdictions in
the case of bankruptcies.
Looking for more certain methods of reducing credit exposures,
dealers turned to collateralisation. While early efforts were hampered by a lack of standardised documentation, the publication of
an ISDA standardised credit support document in 1994, and the
publication of ISDA Guidelines for Collateral Practioners in 1995,
brought standard practices to the industry. These standards were
further refined in the wake of the 199798 banking crisis, when
ISDA published a series of recommendations covering:
o
o
o
o
the methods used to determine haircuts;

coordination between collateral, payments and settlement functions;
the examination of the type of collateral used; and
the importance of reducing operational risk.
Despite such recommendations, most OTC derivative trades were

still documented with phone calls, faxes and scraps of paper as late
as 2003, when regulators stepped in to direct more efficient and orderly back-office processes. The resulting move towards standard,
automated methods of confirming, affirming and documenting
trades helped simplify the movement of collateral.
With better standards being developed, collateralisation grew at
an exponential pace. Collateral in use grew from US$175200 billion in 1999 to more than US$4 trillion in 2008. (Of course, further
refinement of collateralisation standards is being considered in the
wake of recent market events. These new recommendations are discussed later in this chapter.)
THE MECHANICS OF BILATERAL OTC DERIVATIVE COLLATERALISATION
As a rule, OTC derivative trades begin as net neutral, with neither
party owing nor being owed. With time and market volatility, however, the valuation of the trade will change, causing one party to
become indebted to the other. This creates a credit risk that can be
mitigated through the use of collateral.
66
19/03/2010 17:11
To determine how much collateral needs to be posted, derivative

transactions are marked to market on a periodic basis. Once a collateral transfer limit has been passed, collateral is transferred from
the counterparty that is out of the money to the counterparty in the
money. A number of terms, negotiated in a bilateral OTC derivative
collateral agreement, determine how much of what type of collateral needs to be posted under what conditions.
o Collateral thresholds define the dollar limit at which collateral
must first be posted. Thresholds are often asymmetrical, reflecting
the relative creditworthiness of the counterparties. Less creditworthy parties may even be required to post collateral from the outset.
o Minimum transfer amounts define the dollar amount at which it is
necessary to transfer collateral. These provisions, which may also
be asymmetrical, are designed to reduce overall transfer costs.
o Acceptable collateral defines what securities can be posted. Most
agreements call for G7 or OECD sovereign debt, or cash.
o Haircuts reflect the percentage discount that is invariably applied to the market value of securities taken as collateral, based
on the expected volatility of the security over the duration of the
agreement.
o Portfolio margining looks at a group of transactions between two
counterparties to determine initial margin requirements. This
may reduce collateral requirements for a specific transaction, as
other transactions in the portfolio may offset the risk.
o Collateral calls are set at a predetermined frequency. While collateral calls once took place on a weekly, or even monthly, basis,
most major traders are moving towards daily reconciliations and
collateral calls.
o Confirmations are sent to record the amount of collateral delivered from one party to another. Agreements determine how and
when such confirmations must be sent.
The use of collateral in OTC derivatives is almost exclusively documented using the Credit Support Annex (CSA) from the main ISDA
Master Agreement. These provide for the collateralisation of the
net current exposure on the portfolio of transactions covered by the
master agreement. The CSA specifies:
67
19/03/2010 17:11
o
o
o
o
o
o
o
o
o
o
how collateral calls and collateral returns are to be calculated;

the mechanics and timing of transfers;
the method and timing of valuations made by the valuation agent;
collateral substitutions or exchanges;
resolution of disputes regarding valuation and/or exposures;
enforcement on default;
representations and warranties
allocation of expenses related to collateral arrangements;
default interest; and
re-hypothecation (or re-use) of securities.
COLLATERAL USE IN REPO

While collateral has been used to mitigate credit risk in repurchase
agreements (repo) for many years, current bilateral repo practices
have remained fairly consistent since the late 1990s. That is when
trading practices were considerably tightened following the Barings, Russian and Long Term Capital Management defaults.
While initially employed as a mechanism to reduce the cost of
financing, repo has grown in application and sophistication to support a broad variety of additional strategies. It is now used to fund
long positions and cover short ones, enhance liquidity, improve
upon returns, extract tax efficiencies and support any number of
other objectives.
MECHANICS OF REPO TRANSACTIONS
In a repo transaction, one party exchanges securities for cash with
an agreement to repurchase the securities at a specified future date.
The securities serve as collateral for a de facto cash loan, while the
cash serves as collateral for a de facto securities loan. The repo market, therefore, is accessed both by those seeking to borrow cash and
those needing to borrow securities in order to make delivery on
futures contracts, leverage funding of long positions, fund short
positions to hedge interest rate risks, or accommodate a variety of
other needs.
There are primarily two types of repo transactions: classic repo
and buy/sellback. Used for the same purposes, they have very similar structures. The primary difference is that classic repo transactions are effected with bilateral documentation and mark-to-market
68
19/03/2010 17:11
arrangements, where buy/sellback transactions are often not. This

leads to a number of legal and administrative distinctions.
Classic repo
A repo is a sale of securities (originally bonds, but now also securities) for cash with a corresponding legal agreement to repurchase
similar securities from the original provider for cash plus an agreed
upon return. Depending on the underlying jurisdiction, legal title
to the securities can pass between parties, but outright market risk
and economic benefit (such as coupons or dividends) do not. Both
parties have the right to call for margin.
In bilateral transactions, the repoed securities can be re-deployed
(or rehypothecated) by the purchaser for other market purposes.
Figure 4.2 Typical bilateral repurchase agreement transaction
Buy/sellback
A buy/sellback transaction is an outright sale of a bond or security for value with an outright purchase of that security for value
on a forward date. As with classic repo, legal title to the securities
passes, but economic benefit does not. As opposed to classic repo,
these transactions generally are not evidenced by a legal agreement.
Therefore, neither party has the right to call for margin. While there
is no explicit repo rate in these agreements, the setting of the forward price serves the same purpose.
69
19/03/2010 17:11
COLLATERAL USE IN SECURITIES LENDING

Similar to repo, securities lending can be used to affect the exchange
of cash and securities between two parties. In repo, that exchange
is structured as a buying and selling transaction. In securities lending, it is structured as a collateralised loan. That leads to a number
of important distinctions:
o the basic form of the exchange is securities against collateral
rather than cash against collateral;
o repo always uses cash on one side of the trade; securities lending
can employ a broader range of collateral, cash, bonds, letters of
credit or guarantees to collateralise the loan of specific securities;
and
o return is paid to the providers of non-collateral securities, rather
than the provider of cash, and is paid in the form of a fee (while
interest is paid on cash collateral).
Securities lending is used by four types of users:
o borrowers of securities tend to be securities dealers, banks and
hedge funds deploying trading, hedging and arbitrage strategies
that require short positions;
o lenders of securities are generally long-term security holders like
insurance companies, pension funds and endowments, that are
looking for incremental returns or seeking to offset custody fees;
banks and brokerdealers may also lend securities as a customer
service or to capitalise on market opportunities;
o borrowers of cash are primarily securities dealers, banks and
hedge funds using securities they own as collateral to borrow
cash to finance their long positions; and
o cash investors use securities lending as a flexible, short-term
cash management tool, with trade maturities tailored to their investment needs.
THE EVOLUTION OF SECURITIES LENDING
In use since the 19th Century, securities lending came into its own in
North America in the 1960s, when US custodian banks began lending
specific stocks to brokerdealers on behalf of insurance companies,
endowment funds and other large long-term investors. The practice
70
19/03/2010 17:11
steadily spread to other domestic markets through the 1970s and

1980s. Securities lending grew dramatically throughout the 1990s
with the expansion of the derivatives market, development of new
arbitrage activities and greater use of active short-selling strategies.
Impediments to cross-border securities lending relaxed during this
period, and the market became steadily more global.
More recently, the growth of the hedge fund industry and the
increasing use of prime brokerage arrangements further expanded
securities lending. This expansion has led some larger hedge funds
to use several prime broker arrangements to diversify counterparty
credit risk.
THE MECHANICS OF SECURITIES LENDING
A securities loan involves one party borrowing specific securities
from another and providing collateral of comparable value in return. As a general rule:
o the lender lends securities to borrower for a fixed period, or on
demand;
o depending on the underlying jurisdiction, legal title can pass,
but the benefit of any coupon or dividend reverts to the lender;
o rights on the securities, such as voting rights, are exercised as
instructed by the lender;
o the borrower pays a fee to the lender (called a rebate in the
US); and
o the lender takes collateral in the form of securities or cash (in
which case, the lender pays interest to borrower) and margining
is taken in the same way as for repo.
THE LIMITATIONS OF BILATERAL COLLATERAL MANAGEMENT IN
REPO AND SECURITIES LENDING
So far, we have discussed collateral management for repo and securities lending on a strictly bilateral basis. This is the traditional
form of collateral management, where the two counterparties in a
transaction rely upon their internal systems to implement all the
functions of a collateralised trade, including: the collection and return of cash and collateral; recall and substitution; asset serving;
and marking collateral to market. Most collateral in use is still man71
19/03/2010 17:11
aged internally on a bilateral basis, and virtually all prominent sellside institutions have traditionally maintained extensive collateral
management operations for that purpose.
Figure 4.3 Typical bilateral securities lending transaction
As the uses of collateral have grown in scale and complexity, however, the core tasks of effective management on a bilateral basis have
placed an increasing strain on internal resources. Effective collateral management requires substantial infrastructure investments,
specialised expertise and experience across the full spectrum of collateralised trading. For many institutions, particularly on the buy
side, the development of these capabilities on an internal basis can
be daunting. This will only be exacerbated as the industry evolves
toward enterprise-wide collateral management systems.
In addition to the administrative burden, bilateral collateral
management poses other limitations. Typically, buy-side institutions maintain bilateral agreements with a relatively small number
of institutions. This creates credit risk concentration concerns, and
may limit their flexibility to use the broadest range of collateral in
the most efficient manner. It may also be difficult to aggregate and
collateralise smaller positions.
CONCERNS SPECIFIC TO MANAGING OTC DERIVATIVES COLLATERAL INTERNALLY
The collateralisation of OTC derivatives can be particularly complex, and the risk of performing these operations without adequate
72
19/03/2010 17:11
resources is substantial. At the operational level, having a poorly

run derivatives collateral programme is worse than not securitising
trades at all, as it creates a false impression of security. This has led
to more active trading than if the firm were capped out of an available credit line. As the recent market crisis exposed, the concerns
are far more than hypothetical.
Common operational errors resulting from user-defined spreadsheets and databases create operational risk within collateral
management systems. Siloed, spreadsheet-based systems are ill
equipped to cope with changes and large amounts of dynamic
data, and are still highly prone to human error. Replacing such systems, however, requires a substantial investment from already constrained IT budgets.
Of course, even advanced systems would not perform well without experienced people to implement them. Operational risk, therefore, is further increased by the difficulties of securing and retaining
staff experienced in derivatives collateral management. A lack of
collateral knowledge may lead to errors if the intent of the original
legal agreement is not carried forth into day-to-day practice. Inexperience and turnover can also lead to major mistakes in such areas
as novations, upfronts, cash drops and reconciliations. Some of the
more nuanced features of the ISDA/CSA collateral structure call
for close attention to detail by an experienced hand. For example,
the legal jurisdiction, types of collateral and haircuts chosen can all
affect the ease with which a derivative portfolio can be unwound.
ADVANTAGES OF TRI-PARTY COLLATERAL MANAGEMENT FOR
REPO AND SECURITIES LENDING
For many institutions, tri-party collateral management offers a less
expensive and more flexible alternative to internal bilateral systems
for repo and securities lending.
A tri-party trade begins with a master agreement between the
two counterparties specifying the terms of the transaction. Unlike a
bilateral trade, however, the agreement does not specify details of
securities to be pledged. Instead, the agreement is sent to an independent third-party agent. This agent manages the collateral and
ensures that the lenders exposure is adequately covered by automatically picking eligible securities from the borrowers collateral
73
19/03/2010 17:11
pools. Borrowers are able to optimise collateral use by ranking collateral in terms of ideal delivery order within the eligibility criteria.
Figure 4.4a How a tri-party stock loan trade works
Figure 4.4b How a tri-party repo trade works
The core functions of a tri-party service include: the set-up and

maintenance of eligibility criteria, settlement of securities, acceptance and rejection of collateral in adherence to eligibility criteria,
daily mark to market reporting provided intra day and overnight,
74
19/03/2010 17:11
margin notification to borrowers, the calculation and payment of

coupons if collateral is held over record date, and a comprehensive corporate action and tax service. Additional functions include
record date substitution, optimised collateral allocation, automatic
search and replacement of collateral on an intraday basis plus complete reporting solutions.
Tri-party collateral management covers both securities lending
and borrowing as well as repo activities. This allows clients active on
both sides to centralise their collateral management in one location.
COLLATERAL AGENCY STRUCTURES OFFER SIMILAR ADVANTAGES FOR OTC DERIVATIVES
Similar to tri-party collateral management in repo and securities
lending, collateral agency structures offer a less expensive and
more flexible alternative to internal systems for OTC derivative
trading. In addition to the operational efficiencies and funding optimisation, bilateral agency platforms offer faster, more visible processes for reconciling portfolios, resolving disputes and marking to
market even hard to value transactions.
In a best-case scenario, a third-party agency would work closely
with other service providers or utilities to offer value-added services to trading desks. For example, this may include the opportunity to compare valuations with counterparties and identify discrepancies far faster than traditional internal mechanisms. Through
platforms that provide counterparties with common views of each
others data, operations can reconcile trade differences and resolve
valuation issues. Such services can also serve as an early warning
system for systemic patterns of trade population and valuation
discrepancies. With shared data, desks can clean up discrepancies
stemming from timing and data source issues, clearing the way to
identify areas of genuine dispute. Standardised margin messaging
would increase automation options and enhance speed.
The peer-to-peer reconciliation platforms may also play a role in
resolving discrepancies. In addition to providing an early-warning
system, the platforms ability to aggregate data from over 70% of all
non-cleared OTC trades provides precedent to assist with difficult
valuation issues.
75
19/03/2010 17:11
THE NEED FOR ENTERPRISE-WIDE COLLATERAL MANAGEMENT

The move toward third-party solutions adds considerable efficiency
and flexibility to the collateral management of repo, securities loans
and OTC derivative trading. Looking at each of these markets in
isolation, however, still leaves considerable room for undue levels
of counterparty credit risk. In times of financial crisis, it is clear that
institutions must be able to quickly understand their net exposure
to any given counterparty across all trading platforms to know
what their exposure is and where the collateral offsetting any exposure resides. Many organisations, including banks, are reassessing
their historical judgments on where credit risk lies. The full gamut
of financial instruments is now being parsed, with a close look at
the assumptions and processes that surround them. This includes
OTC derivatives, exchange-traded derivatives, repo, stock loan, to
be announced (TBA) bond forwards, deposits, reserves, clearance
activity, loans and structures, among others.
This process, which began with the need to have holistic credit
risk management on the sell side, has now migrated across the full
field of financial instrument users. It is often referred to as enterprise risk management or enterprise-wide collateral management (EWC) when discussing securitisation. Being unable to
assess credit risk exposures and associated collateral at a holistic
level, in a timely and accurate manner, is no longer an option. The
need is clear: risk managers must assess and move their concentrations as the market dynamics and perceptions change. How to
implement is the challenge for all parties.
The evolving answer seems to lie in the aggregation of collateral management functions, across all products, into a third-party
agency. The ideal such agency would have a strong presence in triparty repo and stock loan, sound credentials in the OTC derivatives space, an industrial-strength infrastructure and established
networks for holding almost all conceivable collateral asset types
on a global basis. Combined with a specialist staff dedicated purely
to collateral management, such an agency would offer a wealth of
options and safeguards to its clients.
In addition, an established collateral agent would eliminate the
need to reinvent the wheel, and the attendant risks of doing so. It
eradicates the requirement of maintaining both infrastructure and
76
19/03/2010 17:11
technology thereafter, which would otherwise pull focus and important resources from the clients natural core competency. The
development and widespread use of EWC solutions also benefits
the industry as a whole, since it reduces the need for massive infrastructure projects across a thousand different firms and allows
those resources to be aligned against other initiatives.
FUNDING EFFICIENCIES ARE PARAMOUNT
While the prime motivation for the EWC undertaking has been the
need to understand where the risks lie and what is held against
those risks, funding efficiencies are increasingly emerging as another important driver. Although this initially stemmed from the
banks, buy-side firms are increasingly seeking funding efficiencies
that result from having all collateral tracked, held or pledged from
a central service provider. This will allow full rehypothecation (the
reuse of security assets) to ensure that assets are fully employed;
cash is optimised across different programmes; the cheapest-to-deliver assets are always used; and economies of scale are realised at
the programme level, rather than at a product silo level.
It is at this point that collateral management goes beyond its traditional role as a vital component of credit risk mitigation and holistic credit risk management. Effective collateral management truly becomes a competitive trading advantage for those firms that can
achieve a fully holistic global collateral programme. Finally, from
a cost-saving perspective, the EWC programme also makes sense.
For example, rather than having 15, 25 or even up to a 100 different
collateral management programmes on an enterprise-wide basis,
managing one programme achieves operational synergies through
coordinated settlement, data maintenance, pricing, asset servicing, reconciliations and reporting. Cutting out duplication reduces
costs, while dramatically reducing operational risk.
THE EVOLUTION TOWARD ENTERPRISE SOLUTIONS
The market evolution of traded products has pushed buy-side/
sell-side entities to set up their credit risk management processes in
very different ways. These variances were based on a historic underlying market assumption that credit risk management is more a
preserve of the sell side, because the risk of default from the strong
77
19/03/2010 17:11
credit sell side was minimal. In light of recent market events, those
assumptions have been turned on their head.
Today, processes that were once seen as sidebars and ancillary
to trading with the sell side are front and centre for buy-side users.
Bulking up the product line by purchasing collateral management
tools and services such as OTC derivatives collateral management
makes tactical sense as the industry deals with product nuances.
The real challenge for the industry will be how it copes with collateral management at the holistic level, without having to pursue
grand infrastructure projects at a time of tightening investment dollars. Enterprise-wide collateral management, through a centralised
collateral agent, shines a torch on the road ahead. But moving forward will require a real commitment at the most senior levels, so
that a full end-to-end review of current risk profiles may take place
and the effort to consolidate across all sectors of the industry
gathers sustained momentum.
PROPOSED MEASURES TO IMPROVE COLLATERAL MANAGEMENT
In the wake of the recent market crisis, the focus on mitigating all
credit exposures through securitisation is abundantly clear. The largest task will be to bolster any and all derivatives collateral management processes, including the need for firm and clear collateral segregation from operating balances. A number of initiatives, prompted
by both regulators and the industry itself, have been proposed.
The perception of systemic risk in the financial market, along with
the interwoven nature of derivative transactions across all market
participants, has sparked a discussion about whether some or all
bilateral credit risks can be moved from an OTC basis to a clearer
or an exchange. Measures that would mandate such movement are
among the most commonly discussed regulatory proposals.
Where possible, some sort of central clearing approach would
help take some of that interrelatedness out of the market. The ability to reduce and change all OTC derivatives credit risk in this way
is questionable, however, due largely to the costs of insuring everyone against everything, and the general lack of liquidity of all
instruments across all tenors. As the vast majority of OTC positions
are not standardised, such measures would be unlikely to seriously
reduce collateral use.
78
19/03/2010 17:11
Ahead of such regulatory change, the financial industry has taken

a proactive stance towards mitigating counterparty risk. In its June,
2009, letter to the Federal Reserve, ISDA defined three key pillars
for collateral management:
o To rapidly put in place robust portfolio reconciliation practices
to detect significant trade population and valuation differences
that could give rise to disputed collateral calls. The Fed 16 dealers have already made strides towards reconciling portfolios on
a daily basis.
o To set out a roadmap for collateral management, focusing on independent amount risk issues; electronic communications that
will standardise margin calls; portfolio reconciliation; CSA review; and the development of best practices for collateral management. Many of these recommendations are on track for implementation by year-end.
o To develop a new collateral dispute resolution process for the
industry.
These three pillars are closely aligned to the recommendations
set forth by the Credit Risk Market Policy Group in August, 2008,
which advocated a back to basics approach focused on enhancing corporate governance, monitoring risk, estimating risk appetite, focusing on contagion and enhancing oversight.
CONCLUSION
A practice that is nearly as old as banking itself, the collateralisation of credit risk is driving the most exciting financial innovations
of recent decades. From exchange-traded derivatives, through the
repo and securities lending markets, to OTC derivatives, the transformative power of collateral is opening up opportunities for innovative investment strategies, and lowering the costs and risks in
the securities and settlement infrastructure.
However, while collateralised financing mitigates credit exposures, it also creates new forms of operational risk. The tasks of
optimising collateral, eligibility testing, recalling, substituting and
re-using collateral, marking collateral to market and meeting the
demand to finance new types of collateral are complicated ones.
They require substantial infrastructure investments, specialised
79
19/03/2010 17:11
expertise and experience across the full spectrum of collateralised

trading.
As collateralisation grows in scale and complexity, systems must
become more efficient and better integrated to offer a true, enterprise-wide understanding of where counterparty risk lies and how
it can be most effectively minimised. The ultimate solution may lie
in the creation of a single seamless platform where virtually any
transaction could be collateralised with any counterparty using the
most efficient collateral. Until such a day, it is vital for all institutions to continually examine collateral management processes, both
internal and outsourced, seeking to minimise operational risks and
maximise funding efficiencies.
80
19/03/2010 17:11
Evolution of the US Legal Framework

for Counterparty Risk Mitigation
Lauren Teigland-Hunt
Teigland-Hunt LLP
The over-the-counter (OTC) derivatives markets have seen their

fair share of market crises and counterparty failures since their
advent in the early 1980s. With each crisis and failure, however,
various lessons emerged, and one of the fundamental ways market
participants have sought to address these lessons is through improving the legal framework and documentation that governs OTC
derivatives transactions.
The legal nature of an OTC derivative contract in the US1
The typical OTC derivative transaction whether it is a swap, forward or option is merely a bilateral contract between two parties. Subject to the terms and conditions specified in the relevant
contract, the parties agree to make certain payments and deliveries
to each other in the future based on changes in the value or performance of the underlying reference asset, rate or price.
In contrast to exchange-traded or listed derivative contracts
(such as futures contracts), typical OTC derivative contracts are not
cleared through a third-party clearinghouse or central counterparty
mechanism.2 In addition, they are private contracts that are not protected by any governmental insurance programme or customer asset protection regime. As a result, each party to an OTC derivative
contract directly bears the risk of default or performance failure of
its counterparty ie, counterparty credit risk (CCR). It is therefore
incumbent upon each party to negotiate for itself the contractual
81
19/03/2010 17:11
protections and remedies it requires in order to manage and mitigate the CCR it faces with a particular counterparty. The creation
of a comprehensive documentation architecture that adapts to new
and evolving products and changing needs has been essential to
derivative trading relationships3 and, as such, to the management
of CCR.
Examining the evolution of the contractual legal framework
The OTC derivatives markets have grown dramatically over the last
25 years, and the amount of CCR being borne in the marketplace
has necessarily grown along with it. By the end of 2008, gross credit
exposure for OTC derivative transactions among major banks and
dealers in the G104 countries was estimated by the Bank for International Settlements at US$5 trillion, compared to US$937 billion in
2000.5 The size6 and growth of this market have, perhaps not surprisingly, led to a number of initiatives to improve the way in which
CCR is managed and mitigated in industry documentation.
This chapter will examine how industry approaches to managing
CCR through legal documentation have evolved over the past two
decades. We begin with the evolution of the industrys recognition
and treatment of CCR, focusing on major initiatives intended to establish and improve industry standard approaches to OTC derivatives documentation. Next, we summarise some common approaches for mitigating CCR via legal documentation. We will then address
some of the lessons learned from the financial crisis of 200809, concluding with an overview of the regulatory reform efforts targeting
the financial sector that are likely to impact the way in which CCR is
addressed in the OTC derivatives markets going forward.
EVOLUTION OF INDUSTRY EFFORTS TO ADDRESS CCR IN LEGAL
DOCUMENTATION
In the early years of the OTC derivatives industry, efforts to address
CCR through legal documentation focused primarily on providing
for two forms of netting: the ability to net transaction values upon
default of a counterparty (close-out netting), and the ability to net
routine payments under two parties derivatives transactions when
they fall due on the same day and in the same currency (payment
netting). As we will discuss, the need for close-out netting was typi82
19/03/2010 17:11
Evolution of the US Legal Framework for Counterparty Risk Mitigation
cally met through the use of enforceable master agreements that

provided a clear mechanism for the netting of transaction values
when terminating (or closing out) transactions under the master
agreement. In addition, master agreements typically allowed parties to elect to apply some level of payment netting, either in respect
of individual transactions (at a minimum) or across multiple transactions (where operationally feasible).
Over time, a handful of prominent counterparty failures raised
awareness among market participants of the need for measures
beyond netting to address CCR. In particular, the industry began
to focus greater attention on implementing effective collateral arrangements and enhancing collateral management practices. In
addition, legal efforts to address CCR began to focus on including more comprehensive event of default and set-off provisions, as
well as providing greater flexibility to terminating parties during
the close-out process.
The following discussion highlights some of the defining industry developments that led to the modern CCR mitigation techniques
contained in industry documentation today. As will be seen, a number of the improved practices and approaches used today were recommended in various noteworthy studies and reports published
over the last two decades in response to market developments and
crises.
1987: The creation of the Master Agreement and the basis for netting
Shortly after the International Swap Dealers Association (ISDA)
was formed in 1985, one of its primary mandates became the development of standardised documentation for the derivatives industry. At the time, the OTC derivatives industry was only a few
years old, and swaps were generally being documented as standalone transactions under lengthy, bespoke documentation that took
a great deal of time to draft and negotiate.
ISDA produced its first standardised Master Agreement for interest rate swaps and currency transactions in 1987. A master agreement for OTC derivatives transactions spells out general terms and
conditions governing the two parties derivatives trading relationship, including termination provisions that will allow a party to
unwind derivative transactions early upon the occurrence of cer83
19/03/2010 17:11
tain events of default or termination events. The terms of the Master Agreement, as well as the terms of the individual transactions
governed by it, may be negotiated by the two parties as they see
fit, allowing for customisation to the parties particular needs and
desires. The 1987 Master Agreement effectively offered market participants an industry-standard contractual basis for close-out netting for the first time.
Five years later, ISDA published new versions of its master agreement for derivatives, but without limiting their scope to particular
products or asset classes. In ISDAs Users Guide to the 1992 Master
Agreement, ISDA noted that one of the primary objectives of the
new forms was to promote the benefits of cross-product netting.
The most recent version of the ISDA Master Agreement was published in 2002 and similarly is not product-specific, thereby permitting netting across asset classes.
The close-out netting provisions of the ISDA Master Agreement
are fundamental to mitigating CCR, as they allow a party that is terminating its derivative transactions early (eg, due to a bankruptcy
or other default by a counterparty) to net the different values owed
between the parties in respect of the various trades into a single
net payment owed by one party to the other. One of the key advantages of an effective close-out netting mechanism is that it prevents a bankrupt counterparty (or its trustee/administrator) from
cherry picking among terminated transactions to the detriment
of the non-defaulting party ie, it prevents a defaulting party from
demanding full payment with respect to transactions that are favourable to it while paying only cents on the dollar in respect of the
transactions in which it owes money to the non-defaulting party.
Furthermore, without the ability to engage in close-out netting,
parties would be required to calculate their credit exposures and,
notably, their capital adequacy requirements, where applicable on
a gross basis instead of a net basis.7 Industry estimates indicate that
the ability to determine credit exposure on a net basis can reduce
overall credit risk by over 80%.8
As a legal matter, this reduction in credit risk can best be achieved
when each transaction is entered into as part of a single agreement, documented under an enforceable master agreement, such
as the ISDA Master Agreement, in a legal jurisdiction that recognises the enforceability of close-out netting. As noted by ISDA,
84
19/03/2010 17:11
[t]he widespread use of master agreements providing for closeout nettinghas [an] all important beneficial effect on systemic
risk. Market participants face a considerable exposure to risks
stemming from the failure of a major market participant causing
cascading insolvencies through counterparties. The substantial
credit risk reduction achieved by close-out netting considerably
lessens this consequential effect by reducing counterparty exposure at each node in the network of relationships between market participants and by encouraging best practices by properly
documenting trades at the earliest possible time.9
Now re-named the International Swaps and Derivatives Association, Inc, ISDA10 continues to pursue its mission of producing sound
legal documentation and promoting the benefits and enforceability
of netting. This role is critical because netting is not legally enforceable in every jurisdiction against every type of counterparty.
In fact, in the late 1980s the enforceability of close-out netting
was generally not considered to have been sufficiently tested in the
courts. Consequently, when the Basel Committee on Banking Supervision issued the Basel Capital Accord in 1988 (setting out the
minimum capital standards that the G10 central banks would apply to their banking industries), it did not recognise the ability of
banks to net exposures on the basis of agreements providing for
close-out netting for purposes of determining capital adequacy requirements. Rather, it recommended that banks continue to assess
the enforceability of close-out netting.11
Following the issuance of the November 1990 Lamfalussy report
on interbank netting schemes, the Basel Committee reconsidered
the treatment of netting in the 1988 Capital Accord. In the revised
Basel Capital Accord of July 1994, the committee recognised banks
ability to determine exposures on the basis of close-out netting provisions for capital adequacy purposes. Specifically, the Basel Committee concluded that a bank could net OTC derivatives exposures
under a netting agreement for capital adequacy purposes, so long
as it had written legal opinions stating that netting was enforceable
under the laws of the jurisdictions relevant to the trading relationship. Therefore, in support of its mission, ISDA commissions annual legal opinions from over 50 jurisdictions that address the legal
enforceability of the ISDA Master Agreements netting provisions.12
85
19/03/2010 17:11
1993: Industry response to concerns over OTC derivatives markets

growth the G30 Study
By the early 1990s, it became clear that derivatives had fundamentally changed modern approaches to risk management by providing significant new tools to manage risk. The rapid development
and perceived complexity of derivatives markets, however, troubled a number of industry observers, and concerns about the prospect of misuse of derivatives leading to the failure of individual
firms or systemic risk began to emerge.
In response to these concerns, the Group of Thirty (G30), a private international body comprised of senior representatives of the
private and public sectors and academia, formed the Global Derivatives Study Group (GDSG). The GDSG issued a study in 1993
entitled Derivatives: Practices and Principles (G30 Study), which
contained 20 recommendations to help dealers and end users manage their derivatives activity and continue to benefit from the use
of derivatives. The G30 Study remains a significant report on the issues to consider when managing derivatives trading activities, and
many of the recommendations in subsequent reports and studies
echo principles highlighted in this study.
The G30 Study also included a working paper of the GDSGs
Credit Risk Measurement and Management Subcommittee. This
working paper sought to address concerns about derivatives by
explaining their uses and by formulating and disseminating recommendations about their management. The paper also discussed
documentation provisions that serve to reduce and control CCR in
three categories:
o up-front risk assessment;
o ongoing monitoring and risk reduction; and
o early termination to limit exposure.
The GDSG recommended using industry standard documentation
containing provisions addressing common concerns in these categories, and allowing parties to further tailor provisions to address
additional credit concerns. In particular, the GDSG encouraged
dealers and end users to use a single master agreement that incorporates a full two-way payment approach, meaning that a termination payment may be payable to either party based on the results
86
19/03/2010 17:11
of the close-out netting process, regardless of which party is in default.13 In addition, the GDSG working paper advocated that parties
use a single master agreement with an effective close-out netting
mechanism to avoid potential cherry-picking among transactions
following a counterpartys insolvency, as well as robust payment
netting provisions given that the ability to net payment obligations
under a large number of transactions serves to minimise settlement
risk by reducing the amount and number of payments required to
be made between the parties on a given day.
199495: ISDA publishes standardised credit support documentation
In 1994, ISDA published its first standard form collateral documents, including the 1994 ISDA Credit Support Annex for use with
ISDA master agreements governed by New York law (NY CSA).
The NY CSA was published as a bilateral form, ie, it provides that
either party may be required to post collateral to the other under its
terms. As ISDAs Users Guide to the NY CSA explains, its principal
purpose is to provide a means to document security arrangements
involving the use of cash or readily marketable securities to secure
the risk, or exposure, that either or both parties may have under the
ISDA Master Agreement to which the Annex relates. In 1995, ISDA
went on to publish standardised credit support documentation for
use with master agreements governed by English law, as well as
documentation for use when assets located in Japan are to be used
as credit support.
1998: The near collapse of Long-Term Capital Management
The enforceability of netting remained one of the primary focuses of
industry standard practices for mitigating CCR until a major counterparty default drew attention to the challenges of closing out a
major counterparty in a stressed market, the risks presented by unsecured mark-to-market exposure and the prospect of systemic risk.
During the mid-1990s a prominent hedge fund, managed by
Long-Term Capital Management (LTCM), enjoyed several years of
spectacular investment returns. Then, in 1998, LTCM unexpectedly
lost US$4.6 billion in less than four months as a result of market
movements stemming from the Russian financial crisis that occurred when the Russian government defaulted on its government
87
19/03/2010 17:11
bonds. The Federal Reserve Bank of New York helped to organise

a bailout by the major creditors (including most major derivatives
dealers) of the fund in order to allow for an orderly liquidation of
the funds assets and avoid potential further disruption in the financial markets.
In the aftermath of the near failure of LTCM, a number of guidance documents and recommendations were issued by various
groups, several of which are summarised below.
The G10 report
Two committees of the G10 jointly organised a study group to provide a comprehensive survey and analysis of the practices and procedures that participants in OTC derivatives markets used to manage CCR in practice. Like the G30 Study, the resulting report, issued
in September 1998 and entitled OTC Derivatives: Settlement Procedures and Counterparty Risk Management, recommended using industry standard documentation such as the ISDA Master
Agreement in order to reduce counterparty exposure on outstanding transactions through the use of close-out netting provisions.
The report concluded that close-out netting was a key risk management tool used by most dealers, generally much more extensively than payment netting. The report also described how the failure to complete a master agreement negotiation in a timely manner
prior to trading can exacerbate CCR by jeopardising a dealers ability to close out and net obligations in the event of a counterpartys
default. For this reason, many dealers began using long-form
confirmations that included terms providing that, until a master
agreement has been finalised between the parties, the standard
terms of the published master agreement will apply. In this way the
trade in question would be deemed to be governed by the events
of default and close-out netting provisions of the industry standard
master agreement, even if one had not yet been executed.
The PWG report
In April 1999, the Presidents Working Group on Financial Markets14
issued a report entitled Hedge Funds, Leverage, and the Lessons
of Long-Term Capital Management. The report recommended
that financial institutions draft and publish standards to enhance
88
19/03/2010 17:11
private sector practices for CCR management in a number of areas

relevant to legal documentation. Recommendations included the
following:
o documentation should provide for the timely receipt of counterparty information for use in making credit decisions and to ensure early identification of potential credit problems;
o policies should address the use of collateral to mitigate CCR,
such as providing for more frequent collateral calls with a shorter time to post collateral;
o well-documented procedures should be developed for close out
and liquidation of contracts and collateral; and
o termination rights and triggers should be re-examined to
ensure their effectiveness (noting that many triggers based on
declines in a hedge funds net asset value (NAV) look only at
year-end NAV or a 12-month rolling average NAV, which may
be late in signalling the deterioration of a hedge funds financial
condition).
ISDA 1999 collateral review
In March 1999, ISDA issued a review of collateral management
practices that drew upon lessons from collateral managers experiences during the LTCM collapse. The review set out 22 recommendations for enhancing collateral management practices, and an
action plan for facilitating their implementation. The ISDA report
suggested amending credit support documentation to shorten the
time cycles applicable to valuation, delivery and liquidation of collateral in order to minimise CCR.
However, the report also noted that compliance with shorter
time cycles can present the most operational and financial challenges at times of extreme volatility, when they are needed the most.
Furthermore, the report recommended that counterparties provide
that cash will be accepted in lieu of agreed forms of non-cash collateral, both in the context of a delivery to cover a collateral call and
a return of collateral. The delivery of cash as a fallback was thought
to reduce time cycles and preserve the liquidity of the party receiving the collateral until the agreed upon form of non-cash collateral
could be properly delivered.
The report further recommended that parties ensure that collat89
19/03/2010 17:11
eral agreements address the possibility of credit changes that alter

the relative risk between the two parties and establish appropriate collateral thresholds that vary in response to variables such as
credit ratings or NAV levels.
Finally, the report encouraged collateral managers to assess the
possible liquidity implications of unilateral credit support documentation, which was used by many institutions at the time.
CRMPG report
In January 1999, the formation of the Counterparty Risk Management Policy Group (CRMPG), comprised of representatives from 12
major commercial and investment banks, was announced. The stated
objective of CRMPG was to promote enhanced practices in counterparty credit and market risk management. The first CRMPG report,
Improving Counterparty Risk Management Practices, was published in June 1999 and recommended a variety of modifications to
documentation as a means of reducing CCR, including the following:
o Valuations upon termination: Documentation should provide a
non-defaulting party with the flexibility to value transactions in
a good faith and commercially reasonable manner.
o Payment netting: Documentation should be revised as necessary
to provide for netting of all amounts payable on the same day (in
the same currency) across different types of transactions. Documentation should also provide for payment netting across multiple products appropriately linked under a master agreement.
o Cross-product netting and cross-collateralisation: Parties should
use multi-product master agreements to facilitate the netting of
obligations and collateral across product lines. Where the parties do not have the ability to net collateral across product lines,
documentation should allow the secured party to retain excess
collateral to secure other obligations of the pledgor to that party.
o Set off: The non-defaulting party should be permitted to exercise
broad rights of set off, including the right to set off:
o obligations between the non-defaulting party and the defaulting party under other transactions or other documentation;
o collateral or property of the defaulting party held by the nondefaulting party in connection with other transactions or under other documentation;
90
19/03/2010 17:11
o obligations between affiliates of the non-defaulting party and

the defaulting party and obligations between the non-defaulting party and affiliates of the defaulting party under other
transactions or under other documentation; and
o collateral or property of the defaulting party held by the
non-defaulting party or its affiliates in connection with other
transactions or under other documentation; in addition, the
non-defaulting party should be permitted to transfer excess
collateral to an affiliate to secure obligations of the defaulting
party to such affiliate.
o Events of default: Cross-default provisions should capture any
default by the counterparty under any other transaction or
agreement with the non-defaulting party or the non-defaulting
partys affiliates. Parties should consider the need for broader
cross-default provisions in individual cases.
2002: ISDA re-publishes the Master Agreement
Hundreds of individuals representing firms from a variety of market sectors and regions worked with ISDA for over two years to
produce a new version of ISDAs Master Agreement that would
strengthen the ability of market participants to more effectively
manage risk. Significant changes included:
o The reduction of the standard grace periods associated with
several events of default, including failure to pay, default under specified transaction and bankruptcy events of default. This
change was made in response to market participants experiences with major counterparty defaults (such as LTCM) during periods of market stress, where firms found that the standard grace
periods prevented them from terminating their OTC derivatives
transactions in a timely manner.
o The introduction of Close-out Amount as a more flexible standard for measuring damages upon early termination (or close out).
This change, discussed further on, was intended to reflect the
increase in the volume and complexity of transactions in prior
years and the perceived need to improve the valuation methodology that applies during the close-out process.
91
19/03/2010 17:11
o The addition of a set off provision allowing a non-defaulting

party to set off the termination payment that may be owed under
an ISDA Master Agreement against other obligations owed between it and a defaulting party. Where enforceable, this type of a
provision can provide significant CCR mitigation if the parties
have other trading or business relationships with each other in
addition to their derivatives trading relationship.
2005: Addressing the credit derivative, leverage and hedge fund
tsunami15
After the original CRMPG report was issued in 1999, much progress was made in implementing its principles: firms increased their
use of collateral as a risk mitigant, and documentation standards
were updated to address lessons learned and achieve greater consistency in legal and credit terms across products. By 2005, however, new challenges appeared on the horizon. Some industry observers expressed concern not only about the continuing growth of
the derivatives markets (and credit derivatives in particular), but
also the growth in hedge funds, which were estimated to manage
US$1 trillion in assets globally. In addition, the markets had, in the
intervening years, experienced new stresses and disturbances (such
as the dot.com bubble, the failure of Enron and the 9/11 terrorist attacks). Moreover, the low interest rate environment and high availability of credit were perceived to present new potential risks and
challenges.16
As a result, the Counterparty Risk Management Policy Group was
re-formed in January 2005 (CRMPG II) in order to produce a second
report, entitled Toward Greater Financial Stability: A Private Sector Perspective. The primary purpose of CRMPG II was to update
the groups original recommendations and examine what additional
steps should be taken by the private sector to promote the efficiency,
effectiveness and stability of the global financial system.
The CRMPG II report encouraged market participants to increasingly elect for payment netting through industry standard documentation. Trade associations and market participants were also
encouraged to adopt as best practice cross-affiliate and cross-product netting and cross-default provisions in master agreements, to
achieve greater efficiency and thus reduce CCR across product lines.
92
19/03/2010 17:11
March 2008: The near failure of Bear Stearns

In March 2008, the Federal Reserve Bank of New York helped to orchestrate a rescue takeover of Bear Stearns by JP Morgan to avoid a
collapse of what was one of the largest global investment banks and
securities trading and brokerage firms in the world. One month
later the formation of the Counterparty Risk Management Policy
Group III (CRMPG III) was announced. The CRMPG III published
a new report entitled Containing Systemic Risk: The Road to Reform in early August 2008, which focused its primary attention on
the steps that should be taken by the private sector to reduce the
frequency and severity of future financial shocks.
Calling for financial statesmanship ahead of individual interests, CRMPG III recommended that all large integrated financial intermediaries promptly adopt the more flexible Close-out Amount
approach for early termination valuations, in order to facilitate the
close-out process in the event of a default in their counterparty relationships with each other. ISDA facilitated the implementation
of this recommendation by drafting a Close-out Amount Master
Amendment in August 2009, which was signed by virtually all of
the major derivatives dealers, including Lehman Brothers. As a result, when Lehman collapsed the following month, most of the major dealers were able to use the Close-out Amount approach when
terminating their OTC derivatives positions with Lehman.
In response to subsequent requests from members, ISDA then
released its Close-out Amount Protocol, which offers market participants a way to amend their 1992 ISDA Master Agreements on a
multilateral basis to replace other termination valuation methods
(eg, Market Quotation or Loss) with Close-out Amount. The
ISDA Close-out Amount Protocol is open to ISDA members and
non-members and, as of February 2010, had 75 adherents.
In addition, CRMPG III recommended that all market participants promptly and periodically review their existing documentation and, where appropriate, take immediate steps to update
trading agreements to ensure that they have appropriate events
of default and termination events. CRMPG III also strongly recommended that the industry develop central counterparty (CCP)
clearing arrangements for the credit derivatives market as soon as
possible. Multiple CCPs were ultimately launched for this purpose,
93
19/03/2010 17:11
as noted in a later section of this chapter.

While approaches for mitigating CCR through legal documentation have improved dramatically in recent decades, market events
continue to shed light on areas where additional improvements
could be made. The CRMPG III report properly anticipated the
prospect of future financial shocks on the horizon, but several of
the lessons of Black September 2008 were not fully anticipated
and resulted in new initiatives to improve derivatives documentation in order to address CCR.
CCR LESSONS FROM 2008S BLACK SEPTEMBER
The collapse of Lehman Brothers in September 2008, combined with
the stressed trading environment that existed for most of the autumn of that year, resulted in a series of learning experiences for the
derivatives marketplace. In particular, the events of 2008 focused
market participants attention on the need for robust collateral arrangements that function well in liquid and illiquid markets, and
the need to ensure that collateral is properly protected from the potential failure of a party holding that collateral. In addition, the first
failure of a major dealer counterparty resulted in increased scrutiny
of the CCR presented by certain large financial institutions and the
need to have appropriate CCR mitigants in place with dealers as
well as non-dealers. These, and some of the other fundamental lessons that emerged from the Lehman experience, are addressed in
more detail below.
Ensuring the right to call for collateral
Since 1998 there has been an increasing trend toward the use of bilateral17 collateral agreements: respondents to ISDAs 2009 Margin
Survey reported that approximately 80% of their ISDA credit support agreements, and 75% of all agreements, are bilateral.18 Nonetheless, some parties still seek to amend ISDAs bilateral Credit
Support Annexes so that they are one way, ie, so that only one
party will be required to post collateral to the other. During the
volatile markets of 200708, some market participants, faced with
sizeable unsecured exposures to potentially vulnerable firms, were
surprised to discover that they had one-way collateral arrangements in place when they sought to obtain collateral from their
94
19/03/2010 17:11
dealer counterparties. This very fundamental lesson should encourage market participants to be more vigilant about their credit
support documentation going forward, and ensure that they have
negotiated the appropriate rights to call for collateral.
Implementing effective means to resolve collateral call disputes
The Credit Support Annexes published by ISDA contain standard
provisions providing that disputes over collateral calls will be resolved using a dealer poll mechanism (eg, if the parties cannot
agree on a valuation or mark for a trade, one party will seek
quotes from four dealers in the particular product, and the trade will
be valued at the average of the quotes obtained). While this mechanism in principle may be workable in liquid markets, it has been
viewed by many as cumbersome in practice and ineffective for illiquid products or in markets that are stressed. These weaknesses
became very apparent not only to market participants but also to
regulators during September 2008, and consequently ISDA promptly committed to establish a working group to (i) review the margin
dispute resolution language and practices in common use across the
derivative market currently, and (ii) recommend alternative and improved approaches. This working group presented the results of its
work to regulators on September 30, 2009 in the form of the ISDA
2009 Collateral Dispute Resolution Procedure, a 20-plus page document that provides a far more detailed approach to collateral call
dispute resolution than previously existed. The new procedure is to
be implemented on an experimental pilot basis until the end of 2009,
followed by a trial period in the first half of 2010. Formal market
adoption is then expected to occur starting in summer 2010.
Tightening collateral delivery requirements/shortening transfer timing
During the periods of high market volatility experienced in 2007
08, many firms found that industry standard credit support documentation did not provide for deliveries or returns of collateral in
the timeframes desired. As a result, many firms are revisiting their
existing collateral arrangements in an effort to adjust transfer timing requirements and notification times to ensure timely transfers
of collateral.
95
19/03/2010 17:11
Protecting collateral from the insolvency of a secured party

Many collateral arrangements in the OTC derivatives markets
provide that a party may be required to post initial margin, or independent amounts, in respect of certain transactions. Typically,
independent amounts are posted only by non-dealer parties that
trade actively, such as hedge funds, and constitute an additional
cushion to protect their dealer counterparties against residual
CCR that may persist, even where the trading relationship is fully collateralised. Consequently, independent amounts posted to a
dealer often represent excess collateral relative to the dealers actual exposure to the posting counterparty. In addition, certain market
participants have been known to voluntarily leave excess collateral
with their derivatives counterparties rather than call for the return
of excess collateral when entitled.
The collapse of Lehman Brothers highlighted the risk presented
by these instances of over-collateralisation. Many counterparties
that effectively over-collateralised Lehman through the provision
of required independent amounts, or by willingly leaving excess
collateral with Lehman, were disappointed to discover that their
claims for the return of such excess collateral would rank merely as
general unsecured claims in the Lehman insolvency proceedings.
As a consequence of having general unsecured claims ranking in
priority behind secured creditors and other higher-ranking claims,
these counterparties will recover only a percentage of the value of
the excess collateral they posted (based on the pro rata distribution
that will ultimately be made to general unsecured creditors).
One of the primary reasons for the reduced rate of recovery is
that collateral was generally delivered directly to Lehman with the
right of rehypothecation consistent with market practice. Therefore, collateral was permitted to be freely commingled and used by
Lehman and was not segregated or afforded any other client asset
protections. As a result, many counterparties to derivative trades
with Lehman entities were not able to directly recover any excess
collateral they had posted once the Lehman entities had filed for
bankruptcy, and instead had to resort to filing unsecured creditor
claims as part of the relevant Lehman insolvency proceedings.
Some market participants have sought to protect themselves from
the risks associated with over-collateralisation by establishing third96
19/03/2010 17:11
party custodial arrangements for independent amounts, or through

other methods (eg, posting letters of credit instead of cash or agreeing that independent amounts will no longer be required to be posted upon a certain level of credit deterioration of a holding party).
In an effort to address industry concerns in this area, a white
paper examining the risks associated with posting independent
amounts was jointly published in October 2009 by ISDA, Managed
Funds Association and the Securities Industry and Financial Markets Association. In addition, an ISDA working group has been established to explore and develop alternatives that could be used to
protect market participants from these risks. These concerns have
also caught the attention of regulators and Capitol Hill. Commodity Futures Trading Commission (CFTC) Chairman Gary Gensler
raised the over-collateralisation issue in a letter to Congressional
leaders dated August 17, 2009, proposing that mandatory setaside requirements be imposed with respect to collateral received
by swap dealers on all OTC derivatives. A few months later the US
House of Representatives passed legislation that included provisions that would, if enacted, require OTC derivatives collateral to
be segregated with an independent third-party custodian upon request of a posting counterparty.19
Negotiating effective early warning triggers/termination rights, other remedies
Despite the fact that there were clear signs that Lehmans creditworthiness was deteriorating in early September 2008, few of
Lehmans derivative counterparties had the contractual right to
terminate their transactions early or demand increased collateral or
other forms of security. Prior to filing for insolvency, most Lehman
entities continued to fully perform under their trades, and therefore
few, if any, events of default or other termination events were triggered in advance of their filings. Even where parties had negotiated
termination rights based upon credit rating downgrades in respect
of Lehman, most of these provisions were not triggered because
the Lehman parent company, Lehman Brothers Holdings Inc, maintained an A/A2 credit rating from S&P and Moodys (respectively)
up until its bankruptcy filing on September 15, 2008.
Since Lehmans collapse, many market participants have sought
97
19/03/2010 17:11
to negotiate more effective termination triggers and other remedies

based on measures of credit deterioration other than credit rating
downgrades. For example, some firms have been seeking early termination rights based upon the widening of credit default swap or
commercial paper spreads in respect of a counterpartys debt beyond a certain level. Reserve levels for capital adequacy purposes
have also been considered a possible trigger. However, certain parties have been reluctant to agree to such terms out of concern that a
widening in these spreads can be driven by market forces that are
not specifically related to a deterioration in their own credit quality (eg, spread widening may occur due to more general concerns
about their industry or sector). By way of compromise on these issues, firms have also explored implementing alternative remedies
(instead of termination) to address such indicators of increased
credit risk, such as greater collateral posting obligations (or supercollateralisation), the reset of collateral thresholds or independent
amounts to zero, or a requirement that collateral owed to the deteriorating party be held by a third-party custodian.
PRINCIPAL METHODS FOR MITIGATING CCR THROUGH DOCUMENTATION TODAY
Since a party to a typical OTC derivative contract directly bears
the credit risk of its counterparty, the ability to manage and mitigate CCR is a fundamentally important consideration for parties
entering into OTC derivative transactions. It affects not only their
willingness to do business with each other, but also the types of
provisions that they seek in the legal documentation governing
their trades and their trading relationship. As demonstrated in prior sections, various market events and experiences have led to an
increasing array of methods used to address CCR in industry documentation. Below we summarise the principal methods being used
in the marketplace today, as well as some emerging trends.
Broader enforceability and scope of netting and set off
Close-out netting remains one of the most important approaches
to mitigating CCR through the use of enforceable trading documentation. As of June 30, 2009, close-out netting was estimated to
reduce counterparty credit exposure at US banks by 88%. In con98
19/03/2010 17:11
trast, netting reduced credit exposure by only about 45% during

1996.20 Similarly, the Bank for International Settlements estimated
that close-out netting reduced credit exposure for all OTC derivatives by 85% as of December 31, 2008 from US$34 trillion in gross
mark-to-market value to US$5 trillion in net exposure (a substantial
change relative to the reduction rate of 53% estimated as of June
1998).21
As noted, ensuring the enforceability of the netting provisions of
the ISDA Master Agreement remains an essential undertaking by
ISDA due to its importance in reducing the credit risk arising from
the business. In 2009, ISDA obtained netting opinions with respect
to over 50 different countries, and additional jurisdictions will be
added based upon ISDA member demand. ISDA arranges to have
these opinions updated annually by leading local counsel in the
relevant jurisdictions in order to comply with the requirements of
relevant central banks and regulators.
ISDAs advocacy efforts also have resulted in the adoption of legislation ensuring the enforceability of netting in countries around
the world, and ISDA continues to work with relevant legislative
and regulatory representatives to promote the enforceability of netting provisions in various emerging markets jurisdictions where
netting legislation has yet to be adopted. For example, as of this
writing, netting legislation is under consideration in Argentina,
Pakistan and Russia.
The collapse of Lehman, as a large, diversified financial complex
operating out of multiple legal entities across the globe, has resulted in renewed desire to see enforceable master netting and set-off
agreements that can effect netting of exposures across products as
well as affiliates. However, netting across affiliates presents material
legal enforceability challenges in many jurisdictions (including the
US), and therefore careful legal analysis and drafting, as well as legislative amendments, may be needed to overcome such challenges.
Greater use of collateral
There continues to be a significant trend toward increased use of
collateral in OTC derivative contracts, in terms of both number of
trades and amount of credit exposure covered. According to ISDAs
2009 Margin Survey, the reported number of collateral agreements
99
19/03/2010 17:11
in place has grown from about 12,000 in 2000 to almost 151,000 in

2009, while the estimated amount of collateral in circulation has
grown from about US$200 billion to almost US$4 trillion over the
same period. Respondents to the survey also reported that 65% of
OTC derivatives trades are subject to collateral agreements, compared with only 30% in 2003. Collateral agreement coverage of
credit default swaps at major dealers is even higher, at over 90%.
Further, 66% of OTC derivative credit exposure is now covered
by collateral, compared with 29% in 2003; a recent report from the
OCC identified similar levels of coverage. Notably, coverage of exposure to hedge funds at US bank holding companies is substantially higher, averaging over 200% as of June 30, 2009.22
In support of this trend, ISDA also solicits legal opinions on the
enforceability of the ISDA Credit Support Annexes in over 40 jurisdictions. The existence of this collateral opinion library is one of the
reasons that ISDA credit support documentation is chosen most frequently by practitioners, with 87% of the over 151,000 existing collateral agreements being based on ISDA documents. Respondents
to the ISDA 2009 Margin Survey predicted a further 26% growth in
the number of collateral agreements outstanding during 2009.
More conservative early termination provisions
Since the failure of LTCM in 1998, major market participants have
sought to make the standard events of default in the ISDA Master Agreement more conservative in order to better protect parties facing troubled counterparties. For example, consistent with
the approach adopted in ISDAs 2002 Master Agreement, parties
often shorten grace periods for certain default events so that a nondefaulting party can terminate transactions earlier than originally
contemplated in the 1992 ISDA Master Agreement (which is still
widely used today). In addition, market participants have been creating additional termination events specifically tailored to the risk
profiles of their counterparties in an attempt to secure termination
rights as soon as red flags arise, without having to wait for payment
or performance failures.
More flexible close-out mechanics
As demonstrated by the changes made in the 2002 ISDA Master
100
19/03/2010 17:11
Agreement and the recommendations advanced by CRMPG III,

there continues to be a trend toward attempting to provide terminating parties with more flexibility in the process of closing out a
defaulting party. The Lehman close-out experience may provide
additional bases for considering enhancements to the close-out
mechanics under the ISDA Master Agreement that will facilitate
prompt and enforceable termination and valuation so as to minimise CCR.
Although the private sector has already made significant steps
toward addressing lessons that emerged from the financial crisis
of 200809, lawmakers and regulators remain intent on reshaping
the legal and regulatory framework applicable to OTC derivative
transactions. The next section provides an overview of likely financial regulatory reforms and the potential implications for the way
in which CCR is managed through legal documentation.
THE FUTURE: HOW PROPOSED FINANCIAL MARKET REFORMS
MAY IMPACT LEGAL APPROACHES TO MANAGING CCR IN OTC
DERIVATIVES MARKETS
The severe financial crisis experienced in the US during 200809
resulted in numerous calls for financial regulatory reform. While
the crisis is believed to have had several causes, US government officials have asserted that a lack of transparency in the OTC derivatives markets and the failure of risk management systems to keep
pace with the complexity of new financial products contributed to
the need for substantive reforms in the US regulatory framework,
which currently exempts or excludes most OTC derivatives from
regulation. As a result, the major reform proposals that have been
advanced contemplate establishing a new regulatory framework
for OTC derivatives.
Although the US House of Representatives passed the Wall Street
Reform and Consumer Protections Act (Wall Street Reform Act) on
December 11, 2009, definitive legislation has yet to be enacted as of
this writing. The act passed by the House includes, however, many
of the key reform proposals made in respect of OTC derivatives
since 2008. Generally, they seek to create a more comprehensive and
coordinated regulatory framework that will produce more market
transparency and improve market discipline. Below we discuss
101
19/03/2010 17:11
those elements that are likely to impact the way in which market
participants manage and mitigate CCR in OTC derivatives markets.
OTC clearing
Several reform proposals seek to promote market transparency and
reduce systemic risk by requiring certain OTC derivative transactions to be cleared through regulated CCPs. CCP structures vary,
but a CCP is typically an independent, regulated legal entity that
places itself between the two parties to a derivative transaction.
When trading through a CCP, the single contract between two initial counterparties that is the hallmark of an OTC trade may still
be executed on a bilateral basis at the outset, but it is then replaced
by two new contracts between the CCP and each of the two contracting parties. At that point the original parties to the trade are no
longer counterparties to each other instead, each faces the CCP
as its counterparty. While the practice of clearing OTC derivatives
trades is not entirely new, it is still a relatively recent phenomenon
that has been gradually gathering steam.
In 1998, the only significant OTC clearing was done by OM Stockholm in Sweden for both standardised and tailor-made contracts.
Since then LCH Clearnet Ltd (LCH) launched clearing services for
interest rate swap contracts through its SwapClear entity, and, as of
the end of May 2009, was clearing approximately 60% of the eligible
inter-dealer market in interest rate swaps (or approximately US$85
trillion in trades).23 In addition, other CCPs have emerged over the
years that offer clearing of equity derivative and commodity derivative trades, such as LCHs BClear and IntercontinentalExchange,
Inc, respectively. Most recently, industry regulators have strongly
encouraged market participants to increase OTC derivatives clearing in order to enhance the resilience of the financial system. As a
result, several CCPs in the US and Europe launched clearing services for credit derivatives, and 15 major derivatives dealers committed to centrally clear more than 90% of their credit and interest
rate derivatives by the end of 2009.24
One of the key benefits associated with OTC clearing is the potential to reduce CCP clearing members credit exposures through
multilateral netting (relative to exposures that would exist in bilateral relationships). The potential reduction in counterparty credit
102
19/03/2010 17:11
exposures may result in a reduction in economic or regulatory capital beyond that which can be achieved through bilateral netting and
collateralisation. In addition, clearing houses in most jurisdictions
are subject to insolvency regimes that protect their actions from
challenge in a default and provide explicit support for the application of their default rules. Consequently, there typically is little difference in legal risk, from a netting perspective, between clearing a
trade through a CCP and transacting on a bilateral basis.
Nonetheless, the risk mitigation potential of a CCP necessarily depends on the effectiveness of the CCPs risk management standards
and procedures eg, membership standards, margin requirements,
financial resources and default procedures. In particular, the liquidity and valuation challenges that may be presented by OTC derivatives require CCPs to develop default management procedures that
recognise these challenges, and it is therefore important that market
participants assess the implications of a particular CCPs default
procedures in managing their credit exposure to the CCP and its
participants. Many of the procedures and documentation relevant
to the newer CCPs are still in development (eg, relative to buy-side
access), so these assessments remain a work in progress.
Mandatory margin/collateral and capital/prudential requirements
Many of the reform proposals contemplate imposing mandatory
margin requirements for OTC derivatives either market-wide (subject to certain exemptions) or, at a minimum on systemically significant market participants. The proposed requirements typically
envision minimum initial and variation margin requirements with
respect to bilateral (non-cleared) credit exposures (recognising that
CCPs will establish minimum margin requirements for cleared
trades). For example, the Wall Street Reform Act directs regulators to set margin levels for counterparties in transactions that are
not cleared where one of the counterparties is a dealer or major
swap participant. In addition, most proposals call for minimum
capital requirements applicable to banks, swap dealers and major
swap participants. Non-cleared transactions are often proposed to
be subject to higher margin and capital requirements than cleared
trades. In addition, some proposals call for greater protections for
collateral posted in respect of both cleared and non-cleared OTC
103
19/03/2010 17:11
derivatives trades (eg, through segregation of collateral with a

third-party custodian).
The extent to which the above elements will be incorporated into
final legislation remains to be seen. However, it is likely that any
new US regulatory regime for OTC derivatives will have material
implications for the way in which CCR is managed in these markets
going forward.

I would like to acknowledge Sara Hayes, Josh Riezman and Amanda
Silverman for their valuable research assistance in producing this chapter.
I am also grateful to Robert Pickel, David Mengle and GuyLaine Charles
for their insightful comments and suggestions in respect of this important
topic. Any errors in, or omissions from, this chapter are solely my own.
This chapter has been written based on New York law and the legal and regulatory regime
that applies to OTC derivatives in the US. Different analysis or conclusions may apply in
other jurisdictions or under other bodies of law.
2 In the last section of this chapter, we discuss recent legislative and regulatory initiatives to
increase clearing of OTC derivatives.
3 Robert Pickel, Evolution and Intelligent Design: Developing the Derivatives Infrastructure, Capital Markets Law Journal, 2006.
4 Belgium, Canada, France, Germany, Italy, Japan, the Netherlands, Sweden, Switzerland, the
UK and the US.
5 See Bank for International Settlements (BIS), Semiannual Over-the-Counter (OTC) Derivatives Markets Statistics, tbl. March 19, 2009: www.bis.org/statistics/otcder/dt1920a.pdf;
BIS, Press release: The Global OTC Derivatives Market Continues to Grow, tbl. 1, Nov. 13,
2000: www.bis.org/publ/otc_hy0011.pdf?noframes=1. It is worth noting that credit exposure has remained at or below 1% of notional amount outstanding during this period.
6 In 2009, the US Comptroller of the Currency (OCC) estimated the notional value of derivatives held by US commercial banks alone at US$202 trillion, compared with less than US$33
trillion in 1998. OCCs Quarterly Report on Bank Trading and Derivatives Activities First
Quarter 2009: www.occ.treas.gov/ftp/release/2009-72a.pdf; OCC Bank Derivatives Report
Third Quarter 1998: www.occ.treas.gov/ftp/deriv/dq398.pdf.
7 If two parties have entered into several different transactions with one another, from the
standpoint of either party at any given time some transactions may constitute assets (where
that party is in-the-money), while others are liabilities (where that party is out-of-themoney). When both parties are able to net their transactions that are liabilities against those
that are assets, the overall credit risk is considerably reduced.
8 BIS, OTC Derivatives Market Activity in the Second Half of 2008, May 2009: www.bis.
org/publ/otc_hy0905.pdf?noframes=1.
9 Robert Pickel, ISDA Memorandum Benefits of Close-out Netting, August 10, 2001: www.
isda.org/c_and_a/pdf/Argentina_netting_legis.pdf.
10 ISDA is among the worlds largest global financial trade associations (in terms of number of
member firms), with over 830 member institutions from 58 countries. ISDA News Release,
September 17, 2009: www.isda.org/press/press091709.html.
1
104
19/03/2010 17:11
11 Netting of exposures was permitted in the case of agreements providing for netting by
novation, which refers to the replacement of existing contracts between two parties eg, for
the delivery of a specified currency on the same date by a single contract for that date, such
that the original contracts are satisfied and discharged. Basel Committee on Banking Supervision, Prudential Supervision of Banks Derivatives Activities, December 1994: www.bis.
org/publ/bcbs14.htm.
12 ISDA has collected legal opinions on enforceability of close-out netting and credit support
arrangements in many countries around the world. And these opinions on enforceability
are not subject to customary bankruptcy or insolvency exceptions. They provide a full and
reasoned assessment of the prospect for enforcing those rights without the intervention of
an administrator for the insolvent counterparty. Robert Pickel, Evolution and Intelligent
Design: Developing the Derivatives Infrastructure, Capital Markets Law Journal, 2006. As
noted, the 1994 Capital Accord provided that banks need to have legal opinions supporting the enforceability of close-out netting in order to be able to net their OTC derivatives
exposures for capital adequacy purposes. The opinions obtained by ISDA serve to meet this
requirement.
13 This concept contrasts with provisions calling for limited two-way payments, whereby a
non-defaulting party is not obligated to make a termination payment to a defaulting party.
This approach is often characterised as a walkaway clause and is now rarely used given
its negative implications for capital adequacy requirements.
14 The Presidents Working Group on Financial Markets was formed in March 1988 by President Reagan to give recommendations as to legislative and private sector solutions for
enhancing the integrity, efficiency, orderliness, and competitiveness of [United States]
financial markets and maintaining investor confidence... See Executive Order No. 12631, 3
C.F.R. 559 (1988).
15 The tsunami reference was used by certain CRMPGII members to describe need for report.
16 Zdenka Seiner Griswold, Counterparty Risk Management Policy Group II: OTC Documentation Practices in a Changing Risk Environment, Futures & Derivatives L. Rep., Issue 25#6.
17 As noted above, bilateral collateral agreements provide that either party may be required to
post collateral depending on the direction of total net exposure.
18 In contrast, ISDAs 2005 Margin Survey reported that only 65% of ISDA CSAs were bilateral.
19 Wall Street Reform and Consumer Protection Act, H.R. 4173, 111th Cong. 3108 (as passed
by the House, December 11, 2009).
20 OCC, Second Quarter 2009: www.occ.treas.gov/ftp/release/2009-114a.pdf.
21 BIS, OTC Derivatives Market Activity in the Second Half of 2008, May 2009, www.bis.
org/publ/otc_hy0905.pdf?noframes=1.
22 Consolidated Financial Statements for Bank Holding Companies, June 30, 2009.
23 Serena Ng, Clearnet to Expand its Service on Swaps, Wall Street Journal, May 28, 2009.
24 In September 2009, the president of the Federal Reserve Bank of New York, William C. Dudley, stated, These [clearing] targets will push major dealers to accelerate their progress.
We also expect them to work with central counterparties to rapidly expand the universe of
eligible products and to continue to increase clearing levels beyond these initial targets.
Press release, Federal Reserve Bank of New York, Market Participants Commit to Expand
Central Clearing for OTC Derivatives, September 8, 2009: www.newyorkfed.org/newsevents/news/markets/2009/ma090908.html.
105
19/03/2010 17:11
19/03/2010 17:11
Part II
Counterparty risk
pricing and hedging
19/03/2010 17:11
19/03/2010 17:11
Pricing and Hedging Counterparty Risk:

Lessons Re-learned?
Eduardo Canabarro
Morgan Stanley
In the financial crisis that began in 2007, over-the-counter (OTC)

derivative counterparty risk was again an important source of systemic financial risk. This has been a recurring feature of the recent
financial crises, most noticeably in the Russia/LTCM crisis of 1998.
The paradox is that, over the past 15 years, banks and bank regulators have spent enormous resources to build and enhance their
capabilities to measure, price and manage counterparty credit
risks.1 Also, and in parallel, a broader and more liquid credit derivative market has expanded opportunities for the risk management
of counterparty credit risk as a trading book operation with active
hedging.
The most sophisticated commercial and investment banks have
built risk management systems and have established trading desks
dedicated to the management of their counterparty risks. And still,
despite so much progress on so many important fronts, counterparty risks inflicted large losses on banks during the latest crisis to
the tune of US$10 billion for one bank and multiples of that for the
entire banking industry.
In this chapter we examine the key features and the evolution
of counterparty risk pricing and risk management in the context of
our experience over the past fifteen years.
The pricing of counterparty risk is usually referred to as credit
valuation adjustment (CVA). We use this terminology and acronym
throughout this chapter. For many years there was confusion and
109
19/03/2010 17:11
disagreement about the concept of CVA: should it be a unilateral

calculation (ie, pricing only the risks faced by the bank) or should
it be a bilateral calculation (ie, pricing the credit risks faced by both
the bank and its counterparty). Only gradually has the issue been
settled, in practice and in the accounting standards, and the bilateral calculation is now widely recognised and adopted.2
DEFINITION
Throughout this chapter we use the following setup: we, bank A,
have a portfolio of OTC derivative trades with our client, counterparty B. All valuations are from our (ie, bank As) perspective.
Definition: CVA is the adjustment to be applied to the credit-riskfree value of the portfolio of OTC derivatives between counterparties A and B to reflect the market value of the credit risks faced by
both counterparties with respect to each other.
We refer to the credit-risk-free value of the derivatives as their
mid-market value.
If bank A faces more credit risk than counterparty B, the CVA is
negative (ie, it reduces the value of the OTC derivatives from the
perspective of A). If bank A faces less credit risk than counterparty
B, the CVA is positive (ie, it increases the value of the derivatives
from the perspective of A). If the present values of the credit risks
are the same, the CVA is equal to zero.
CVA CALCULATION: A SIMPLE MODEL
Lets represent the calculation of the CVA as
CVA = EAsA EBsB
where:
o EA is the present-valued expected exposure faced by B with respect to A;
o sA is the loss rate (ie, the product of risk-neutral PD and riskneutral LGD) of A, where PD is the probability of default and
LGD is the loss-given default;
o EB is the present-valued expected exposure faced by A with respect to B; and
o sB is the loss rate of B.
110
19/03/2010 17:11
PRICING AND HEDGING COUNTERPARTY RISK
The scalar notation is simplistic but useful for examining the fundamental elements of pricing and risking CVA.
A numerical example:
EA = US$200
sA = 2%
EB = US$100
sB = 5%
CVA = 200 x 0.02 100 x 0.05 = 4 5 = US$1
The CVA is a negative adjustment to the mid-market value of the
portfolio of trades as seen by bank A, because bank A faces more
credit risk than counterparty B.
Assuming that the mid-market value of the portfolio is US$50,
the portfolio would be worth US$51 for bank A and +US$51 for
counterparty B. Both counterparties would agree on this value.
Fair value: adding counterparty C
Suppose that counterparty B is to assign the portfolio of trades to
counterparty C, sC = 2%. In order for the assignment to happen, all
three counterparties need to agree on the values of the portfolio of
trades and on the amounts that they will have to exchange to effect
the assignment.
Current status:
o A sees a value of US$51 for the portfolio of OTC derivatives
o B sees a value of +US$51
After the assignment of the portfolio to C:
CVA = 200 x 0.02 100 x 0.02 = 4 2 = US$2
o A sees a value of US$48
o C sees a value of +US$48
So, in order to execute the assignment, the following payments
need to take place:
111
19/03/2010 17:11
o A pays US$3 to B, that is, 48 (51) = 3.

o C pays US$48 to B.
o B receives US$51 = 3 + 48.
Notice that A is willing to pay US$3 to B for the assignment of the
trades because the credit risk that it faces is reduced due to the
better credit quality (ie, lower credit spread) of C. Also, C is only
willing to pay US$48 to B because that is the value of the portfolio
of trades inclusive of the CVA against A. The three counterparties
agree on the fair values of the portfolio of OTC derivatives and on
the amounts to be exchanged.3
Fair value: a recent example
During the latest financial crisis a large component of the CVA
losses incurred by banks were attributable to their CVAs related to
monoline insurers (also called simply monolines).4
Monoline insurers sold credit protection on senior and supersenior tranches of ABS/CMBS CDOs to banks. When the implied
credit losses of the underlying pools of ABS/CMBS bonds skyrocketed, the banks found themselves in large creditor positions
with respect to the monolines.5 At the same time, and partly because of the monolines large debtor positions, the credit qualities
of the monolines deteriorated massively and the monolines credit
spreads widened to thousands of basis points per year.
Some banks unwound the trades with monolines at deep discounts from mid-market values, implicitly reflecting the actual
CVAs.6
Example of unwinding:
Monoline B owes US$2.5 billion (at mid-market) to bank A.
Unwinds trades paying US$500 million to bank A.
Based on the unwinding value:
o CVA to bank A is US$2.0 billion;
o CVA to monoline B is +US$2.0 billion.
112
19/03/2010 17:11
CVA calculation: the full model

The fully fledged model of CVA calculation is specified by the Monte Carlo simulation described below:
1. The default events of both counterparties (as well as the evolutions of all market prices) are simulated over the lives of the
derivative trades and according to the risk-neutral dynamics
(hazard rates, recovery rates, volatilities, correlations, drifts, etc).
2. If and when the first default event (of either A or B) occurs on
a simulated market evolution (path), the loss or gain relative to
the mid-market valuation of the trades to the non-defaulting
counterparty is calculated and discounted to time zero using the
appropriate stochastic discount factor (which is simulated along
the market path).
3. As default causes a loss or a gain (relative to mid-market) to
B when the trades are closed out. Bs default causes a loss or
a gain (relative to mid-market) to A. As losses/gains are negative/positive, respectively. Bs losses/gains are positive/negative, respectively. The CVA is the average of all losses and gains
over all simulated paths.
This Monte Carlo model includes the key elements of the CVA calculation:
o Risk-neutral market dynamics for market factors and events,
including the implied correlations between default events and
other market variables (wrong- or right-way risks)
o Full assessment of loss to the non-default counterparty at the
time of default. The loss calculation should be comprehensive
and include all costs that would be incurred due to the default of
the counterparty, including estimated transaction costs related to
the replacement of illiquid trades.
o We assume that, at the time of the default of a counterparty, the
OTC derivative trades are closed out. Even though the non-defaulting party may have the option to not close out the trades, we
believe that in most cases the counterparties (especially bank A, a
derivatives dealer) will not exercise this option.7
113
19/03/2010 17:11
CVA calculation: the simplified model used by banks

Most banks do not implement the fully fledged Monte Carlo simulator. Banks prefer to implement simplified models where EA and
EB are profiles (ie, curves or vectors) of the present-valued, conditional-on-default expected exposures over the life of the portfolio
of trades between A and B. Those profiles are calculated analytically or via Monte Carlo simulations, and they factor in credit risk
mitigants such as netting agreements, margin agreements and early
termination provisions.
The loss rates sA and sB are profiles of forward loss rates as implied from the prices of credit default swaps (CDSs) on A and B,
respectively. The CVA calculation is the inner product of the vectors of present-valued expected exposures and forward loss rates,
and it reflects the net present values of the credit risks faced by the
counterparties.
Figure 6.1 Graphical example of the present-valued, conditional-on-default
expected exposure EB and the forward loss rate sB profiles
The credit risks of many counterparties do not trade in the marketplace. For those, the loss rates used to mark the CVA must be
derived from generic credit spread curves.
Usually those generic curves are constructed for each credit rating category used by the bank. The generic credit curves should be
based on traded credit spreads (eg, traded credit indexes) so that
those traded prices can be used as hedges.
If the generic credit curves are not based on traded prices, the
114
19/03/2010 17:11
bank will not be able to hedge the CVAs. It is important to mark all
CVAs (traded and non-traded counterparty risks) to traded prices,
directly or indirectly.
The zero-CVA case
Lets assume a perfect and frictionless market environment:
o continuous price processes no price gaps that would prevent
continuous hedging;
o frictionless markets no transaction costs, infinite liquidity, etc;
o perfect margin agreements with continuous marking to market
and continuous exchange of collateral at zero collateral thresholds; and
o perfect and absolutely certain close-out procedures, legally enforceable netting and collateral rights.
In this perfect market environment, the CVA is zero since both
counterparties A and B face zero credit exposures with respect to
each other, as calculated by the fully fledged model.
A zero CVA means that the fair value of the portfolio of trades
between A and B is its mid-market value. That is, its value at the
level of quoted prices for example, in OTC derivative brokers
published screens.
The author refers to this idealised case as the zero CVA case. It
is a useful reference case against which all other cases can be compared to determine the need for and value of a CVA.
Funding benefit or credit risk pricing?
Practitioners often refer to EA.sA as a funding benefit to bank A. That
is, bank A benefits from its expected borrowing from its counterparty B. In other words, if counterparty B is exposed to bank A, it
means that it is lending to A the value of that exposure.
That is not correct though. EA.sA is the value of the credit risk
faced by B. It can be associated with funding provided to A by B,
but not always.
Consider the following example:
Suppose that A and B have two identical and symmetric (long and
short) derivative trades with each other. Assume that there are no
115
19/03/2010 17:11
netting or collateral agreements between A and B. The mid-market

value of the portfolio of the two trades is identically equal to zero
over the life of the trades. The funding costs/benefits produced by
this portfolio of identical and symmetric trades are thus equal to
zero. But, because there is no netting agreement in place, both counterparties face credit exposures with respect to each other, and the
CVA is not zero (except in very special cases). In particular, EA.sA is
not zero and yet there is no funding benefit at all to A.
Funding costs and benefits need to be analysed at the level of the
full portfolio of derivatives of the bank, including all trades and all
margin arrangements with all counterparties. They are bank specific and affect market prices only to the extent that preferences and
risk-aversions determine marginal prices.
Price signals and competitiveness
Sometimes derivative traders and their derivative sales colleagues
argue that if they were to include the CVA in the prices that they
offer to their clients, they would be off market (charging their clients too much) and would lose business to competitors that do not
price in the CVA. This is frequently the situation in which the credit
quality of the counterparty is low as in the case, for example, of
many commodity derivatives counterparties. But it is also the case
when out-of-the-money and wrong-way risks are not clearly and
correctly identified ex ante (such as monoline insurers before 2007).
The point here is obvious: market values are market values and
a trader may decide to trade against them or despite of them for
many good reasons, but they should mark their trades to market
based on the prevailing market prices. A bank may trade with a
client even though it incurs a CVA-induced loss because of the benefits of the client relationship, its desire to participate in a particular
market or transaction or other long-term and profitable benefits.
A bank should not underestimate the benefits of the mark to
market discipline (including CVA). And it should not underestimate the costs of not having that discipline either.
Price signals create economic incentives that lead banks and their
traders to engage in sound economic actions. If the price signals are
wrong, they will induce poor economic behaviour and consequent
costs. Wrong price signals (eg, caused by flawed pricing, erroneous
116
19/03/2010 17:11
capital charges, etc) can have serious and sometimes fatal consequences for banks.
PANEL 6.1 Mark-to-market discipline
E. Gerald Corrigan, a former president of the New York Federal Reserve
Bank and later managing director and co-chair of Goldman Sachs
Global Risk Management Committee, made the authoritative statement
below (Corrigan 2002):
While on the subject of accounting practice, I want to acknowledge
that there is one critical area in which my own thinking has changed
materially since my days at the Fed. That change relates to the need to
substantially accelerate the move to the universal adoption of fair-value
accounting for all financial institutions.
Having learned the advantages of historical cost accounting and the
perils of fair-value accounting for banks from the master, Paul Volcker,
I am more than mindful that this subject is controversial and that there
are practical, policy and philosophical issues to be addressed in managing this transition. Without in any way diminishing these issues, the
critical factor that has changed my thinking on this subject is the discipline associated with the need to mark-to-market all positions on and
off balance sheet on a daily basis.
From my vantage point as co-chairman of the Global Risk Management Committee at Goldman Sachs, I have witnessed at first hand
the ways in which fair value and mark-to-market accounting impose
prompt discipline on business practices and risk appetites. Recognizing
losses (and gains) as they occur constructively influences behavior in
ways that I simply did not comprehend from the distance of my former
lofty perch on Liberty Street.
CVA desks
Banks have chosen different structures for their CVA risk management operations. Some banks opted for a centralised model where
a single, central CVA desk acquires and consolidates all CVA risk,
charging the derivatives trading desks a transfer price. At the inception of each trade, the CVA desk sets the price at which it will acquire the counterparty risk. Sometimes the price is positive, sometimes it is negative, depending on how the new trade affects that
counterpartys existing exposures. Trades that reduce the existing
exposures by offsetting existing trades generate a payment from the
CVA desk to the derivatives trading desk. This payment creates an
incentive and provides a competitive advantage to the derivatives
117
19/03/2010 17:11
trading desk to seek those risk-reducing trades.

Other banks have opted for a decentralised model and have established more than one CVA desk, each within one of its major
trading units (eg, in rates, commodities, equities business units,
etc). Those desks perform the same function as the central CVA
desk but take and risk manage the counterparty risk of the derivatives desks within their business unit only.
The fact that both models co-exist today suggests that each has
its advantages and disadvantages. The centralised model allows
for more aggregation and consolidation of risks and offers the opportunity for higher operational leverage. But it inevitably creates
organisational tensions related to transfer prices and the different
philosophies of risk management of different business units.
The decentralised model, on the other hand, creates some redundancy of operating structures (with their attached operating costs)
but reduces the transfer price tension and keeps the ownership of
the risk closer to the points where the risk is originated. The latter
can be an important aspect of the overall risk management framework, especially when the exposures are generated by new and
complex products where the specific trading desks have developed
more expertise than CVA desks.
During the recent financial crisis some of the largest CVA losses
came from highly structured credit products. Some people claim
that banks underestimated and under-priced the credit risks at the
inception of those trades. Those wrong-way counterparty risks became large during the crisis, hedging was difficult because of the
illiquidity of the stressed markets and banks incurred large CVA
losses. Some argue that had those risks been owned by the trading desks originating the structured trades, the banks would not
have accumulated the large amounts of those risks that they did. Of
course, it is quite debatable if this assessment is correct given that,
ex ante, the risks were seen as just a very small fraction of the sizes
that eventually materialised (Rowe 2006).
A CVA desk should be subject to risk limits like any other trading
desk. But since the function of the CVA desk is primarily risk management (as opposed to revenue generation) it should be subject to
limits that are designed consistently with its objectives. The limits
will typically include limits on sensitivities (greeks), jumps-to-de118
19/03/2010 17:11
fault, value-at-risk (VaR), stress tests, etc. The CVA desk will be given some ability to manoeuver around and manage risks proactively
but the limits will preclude it from warehousing concentrated and
outsized risk positions.
The goals of the CVA desk should be to:
o mark the CVAs to market;
o eliminate the concentrations of CVA risks via hedging; and
o optimise the CVA portfolios risk-return profile while operating
under its risk limits.
HEDGING CVA
One of the most important benefits of a strong marking to market
discipline is that it induces banks and their traders to hedge. Hedging is the only way to comply with risk limits and avoid economic
capital charges.
A bank that marks counterparty risk to market via CVA should
hedge its CVA. However, this is likely to be one of the most complex and difficult risk management activities that the bank will ever
pursue.
CVA is a complex, multi-dimensional, hybrid contingent claim
whose value depends on:
o dynamics of the market risk factors driving the underlying counterparty exposures;
o dynamics of the price of the credit of the counterparties;
o correlations and co-dependencies between the above;
o uncertainty about the enforceability of netting rights;
o uncertainty about recovery rates; and
o uncertainty about trade replacement costs at default.
All the above factors need to be considered carefully in the CVA
hedging strategy. Usually the risk factors driving the exposures are
liquid and hedgeable. They are comprised of interest rates, foreign
exchange rates, equity and commodity prices and credit spreads.
But some factors can become illiquid at times and hedge re-balancing can become costly.
Also, and importantly, some counterparty exposures have the
119
19/03/2010 17:11
characteristics of crowded trades because the banks client may

have engaged in the same type of derivatives with many different
banks. The crowded-trade situation (many banks having to re-balance hedges in the same direction and simultaneously) can exacerbate the transaction costs and the market impact of the dynamic
re-balancing of CVA hedges, especially in cases where wrong-way
risk is present.
The counterparties credit risk may not be liquid. There may be
no CDSs or bonds of the counterparty available in the marketplace.
In this case the bank may have to use proxies for hedging.
Systematic components of counterparty credit risks may be
hedgeable, but the name-specific (idiosyncratic) ones may not be.
Exposures to illiquid counterparties need to be limited at the inception of the trades. After the trades are on, the market-driven exposures will fluctuate and they can become large if not limited appropriately at inception. When credit protection on a counterparty
is not available, the bank will have to pursue sub-optimal hedging
strategies.
What are the risk factors that drive CVAs?
Lets focus on the credit risk faced by bank A. This is the term -EB.
sB of the CVA expression. Bank A needs to hedge the changes of
the exposure EB and the changes in the loss rate sB. This requires
hedges on the market factors driving the exposures (eg, interest
and foreign exchange rates, credit spreads, equity and commodity prices) and hedges on the credit spread of counterparty B. The
hedges have to be continuously rebalanced as the exposure and the
spreads change. The larger the credit spreads, the larger the hedges
on the exposures. The larger the exposures, the larger the hedges
on the spreads.
Formally, the local sensitivities of the CVA with respect to its
components are:
CVA / EB = - sB
CVA / sB = - EB
2CVA / (EB sB ) = - 1
120
19/03/2010 17:11
The expressions show that the hedges of the CVA need to incorporate hedges of the market risk factors driving the exposures and
hedges of the credit spread of the counterparty. Moreover, there
is negative cross-gamma term that can be substantial when the
changes in counterparty Bs spreads and exposures are large.
During the recent credit crisis, due to the high volatility of markets (ie, large EBs and sBs), the negative cross-gamma term created difficulties for CVA desks hedging the CVAs, and it caused
losses in cases where the realised correlation between EB and sB
was positive (wrong-way risk).
Should a bank hedge the CVAs?
If the bank marks to market its CVAs and the bank does not hedge
them, it will experience P&L (and earnings) volatility. Importantly, in a trending and deteriorating credit market environment, the
bank could suffer substantial cumulative losses. In the recent credit
crisis, some banks lost many billions of US dollars in CVAs. This
was particularly the case for banks that did not actively hedge their
CVAs (despite the fact that they were marking them to market).
The recurring story of not hedging CVA and, in general, other
credit risks, goes as follows:
o the bank does not hedge the CVA because it is eager (and greedy)
to earn the often quite compressed credit spread;
o then the credit spreads widen, counterparty risk grows and the
bank incurs losses;
o then the banks appetite for additional losses is reduced and the
bank decides to hedge the CVA; and
o at that point the bank locks in the realised losses and, even if the
CVAs were to decline in the future, the bank will not realise any
gains because of the offsetting effect of hedges now in place.
This is an unfortunate risk management story it is quite amazing how it repeats itself over time and contexts. The lesson to be
learned from the latest playing of this story is that banks should
hedge their CVAs outright, otherwise they are likely to start hedging it too late, at the worst possible time and could then be subject
to large transaction and market impact costs.
121
19/03/2010 17:11
Should a bank hedge its own credit risk?

So far we have focused on the component -EB.sB, which is usually
referred to as the asset or receivables side of the CVA. This is the
market value of the credit risk faced by bank A. There is another
component in the CVA, EA.sA, which reflects the market price of the
credit risk faced by counterparty B. The risk of this component also
affects the P&L of bank A. It is usually referred to as the liability or
payables side of bank As CVA.
The interesting feature of this component of the CVA is that it
will produce P&L gains for bank A when its credit spread widens.
During the recent financial crisis, the credit spreads of many banks
widened quite a lot, causing large CVA gains in the liability component of the CVA. Those gains reversed after the critical part of
the crisis was over, and the gains left were not enough to offset the
substantial losses in the asset side of the CVAs. The net CVA results
of the banks, including both asset and liability CVAs, were highly
negative.
The key questions here are:
o should bank A hedge the liability side CVA?
o more specifically, should bank A hedge its own credit spread?
o and, if so, how could bank A hedge its own credit spread?
Those questions have been asked many times over the last 15 years.
As in the case of the asset component of the CVA, the relevant
risk sensitivities to consider are:
o CVA / EA = sA
o CVA / sA = EA
o 2CVA / (EA sA ) = 1
Changes in the exposure EA can be relatively easily hedged by taking positions on the market risk factors that drive the exposure (eg,
interest and foreign exchange rates, credit spreads, equity and commodity prices).
Changes in bank As own loss rate sA are more challenging to
hedge. The systematic risk component of bank As credit spread
can possibly be hedged by positions on suitable credit indexes.
The bank-specific, idiosyncratic risk component is more difficult to
122
19/03/2010 17:11
hedge. It would require that bank A sell credit protection on itself.

This is a very difficult proposition given the strong wrong-way nature of the trade. Thus, many banks have chosen to retain the idiosyncratic component of their credit spread risk. Sometimes they
choose to retain the entire liability CVA spread risk, systematic and
idiosyncratic components.
When banks attempt to hedge the systematic components of
their own spreads, they do it in two different ways:
o using the systematic risk components of their counterparties
spreads; and
o by engaging in trades based on credit indexes or credit baskets.
The first of these alternatives is a natural way of using the asset side
CVA as a hedge for the liability side CVA. It makes sense because
some (perhaps a large part) of the credit spread part of the asset
CVA can only be hedged systematically anyway ie, the idiosyncratic, name-specific risks do not trade in the credit markets. Thus,
that component can be hedged with the systematic part of bank As
spread.
The second alternative has caused problems because some credit
baskets are exposed to large idiosyncratic risks. For example, when
Lehman Brothers filed for bankruptcy in September 2008, banks
that were selling credit protection on baskets of financials that included Lehman incurred sizeable losses.
The credit spread risks of the asset and liability CVAs can be large
and of the order of US$510 million per basis point on each side
at the largest OTC derivatives dealers. This is equivalent to a total
notional of 5-year credit default swaps of approximately US$1224
billion. Those are quite large credit spread positions.
Friction costs
Similarly to other complex contingent claims, the risk management
of CVAs requires dynamic re-balancing of the hedges. When the
counterparty exposures and the credit spreads of the counterparties
are large and volatile, the rebalancing requirements may be intense
and costly, due to illiquidity, wide bid-ask spreads and market impact of the hedging programme, especially when in crowded situations.
The costs of the frictions mentioned above are usually not ex123
19/03/2010 17:11
plicitly captured in the CVA pricing models, but they can be a most
relevant cost component of large, concentrated CVA risks.
Banks have used simulation models to assess the size of those
potential costs of replication over the life of the portfolio of trades
with a counterparty. The simulation models incorporate the frictions and imperfections mentioned above and provide a more realistically description of the probability distribution of potential CVA
hedging costs. Those costs should be considered in the risk/return
analyses of new and large OTC derivative transactions. During the
recent crisis, the realised dynamic hedging costs of CVA hedging
proved to be quite material in some cases.
Wrong-way risks
We refer to wrong-way risk as the adverse (negative) correlation
between the exposure to the counterparty and its credit quality. Alternatively, it can be stated as the positive correlation between exposure and credit spread. Those are situations where bank A would
expect to have large exposure to a counterparty in market scenarios
where the credit quality of that counterparty would be expected to
deteriorate.
Examples are:
o the emerging market countrys bank selling hard currencies
against local currency;
o the company selling put options on its stock;
o the company selling CDS on its own name;
o banks selling of credit protection on their obligors; and
o the producer of commodities over-selling forward its future production.
The last of the examples above highlights the difficulty evaluating
wrong- and right-way risks. If counterparty B has similar trades/
hedges with many banks, it is difficult to assess the full impact of
its hedges in the determination of the correlation between bank As
exposure to counterparty B and the credit quality of counterparty
B. Also, and importantly, the fact that counterparty B may have
margin agreements with bank A and/or with various other banks
may impose liquidity pressure on B in the form of margin requirements when banks exposures to B increase. The case of the mining
124
19/03/2010 17:11
company Ashanti in the late 1990s is an example of this.

Despite the many efforts of banks and their regulators, the ex ante
identification, monitoring and pricing of wrong-way risks continue
to pose great challenges in practice. Sophisticated banks have created systematic identification and screening frameworks for wrongway risks.
In addition to the above, there are wrong-risks that are specific
to CVA hedging. Those wrong-way risks relate to crowded counterparty risks. When a counterparty has entered in similar and large
OTC derivatives trades with many banks, the hedging programmes
of the banks with respect to their exposures will create wrong-way
risk: all banks will need to buy or sell credit protection on the name
as their exposures to the counterparty increase or decrease, respectively. Usually, those wrong-way risks do not show up until credit
spreads and/or exposure have grown to a large level. But from that
point on, they could be present. During the recent credit crisis this
occurred with respect to monolines, as well as other concentrated
counterparty exposures.
The reverse side of wrong-way risk is what is called right-way
risk. Examples are:
o the company selling call options on its own stock; and

o the producer of commodities selling forward a small portion of
its future production.
Whenever material, both wrong- and right-way counterparty risks
should be priced in the CVA. This would affect not only the pricing
but also the hedge ratios of the CVA. To incorporate those effects,
CVA models need to explicitly recognise the joint dynamics of credit
spreads and the market factors that drive counterparty exposures.
In the simplified model used by banks, this requires the computation of expected exposures conditional on default. The fully fledged
CVA model accounts for right- and wrong-way risks automatically
by explicitly modelling the dependencies between credit hazard
rates and other market prices.
Out-of-the-money risks
These are counterparty risks that are very small at the inception of the
OTC derivatives trades, but which can become very large because of
sizeable market moves and inability to hedge along the move.
125
19/03/2010 17:11
Examples of OTC derivatives trades that fit in this category are:8

o bank A buying credit protection on senior and super-senior ABS/
CMBS CDO tranches from monoline insurers when the running
premium of those tranches were 10 basis points per year then
the implied cum losses of the tranches increased and some were
priced at 8090% total loss;
o bank A selling oil at-the-money-forward to airlines when the
spot price of oil was US$140/bbl it then dropped massively to
US$40/bbl later on; and
o bank A buying long-dated out-of-the-money equity index puts
from insurance company when the S&P 500 was at 1400 the
S&P 500 then dropped massively to 700.
The problem with those exposures is that the potential future exposure (PFE) models used for CVA calculation are not good predictors
of massive market dislocations. PFE models often turn out to be
inadequate for capturing large and abnormal market movements.
CVA desks need to be cautious in the pricing and hedging of outthe-money counterparty exposures. The ability to hedge those exposures in the future, as they grow, needs to be assessed prudently,
considering not only the positions that bank A has but the overall
crowdedness of the market. The pricing of CVA should incorporate a (potentially large) premium to account for future hedge costs
and for the risk of large hedging slippages. The profitability of such
trades needs to be evaluated considering the potential CVA risks
and costs embedded in the trades.
The author suspects that the majority of bank As deep-out-ofthe-money OTC derivatives trades have substantially inferior riskadjusted returns when all risks and costs are properly accounted
for. That is, bank A should evaluate them very carefully.
Stress tests
Stress tests are a fundamental component of a sound CVA risk management programme. CVAs are some of the most complex and black box
risks that are run by banks. A comprehensive and well-designed stress
test framework around the CVA model is of paramount importance.
The fundamental goals of the stress test framework should be the:
126
19/03/2010 17:11
o identification of concentrations of market and credit risks (those

could be related to single counterparties or groups of counterparties);
o identification of out-of-the-money exposures suggesting that
some stresses should incorporate very large and outsized shocks;
o identification of wrong-way risks; and
o identification of potentially large dynamic hedging costs.
Since the calculation of the CVA can be so computationally intense
and expensive, it is important to engineer the CVA systems and
models efficiently. Various analytical and computational techniques
have evolved to allow for fast calculations. One of the most effective, and which has been used since the very early days of CVA,
are price/value tables or grids (as the ones used in finite difference solvers). Those tables can be generated once and then used to
calculate CVAs as required to obtain the various hedge ratios and
stress tests. The simulation and pricing models used within a CVA
framework need to be designed to meet the requirements of computational efficiency and speed.
Netting and financial collateral
Netting and collateral agreements are powerful credit risk mitigants. They are designed to reduce counterparty exposures by
providing for the offsetting of positive- and negative-value trades
when a counterparty defaults, and by providing for the exchange
of collateral as one counterparty gets exposed to the other. But they
are by no means perfect and, consequently, do not eliminate completely the counterparty risks (see Gibson 2005).
Netting enforceability is a legal matter subject to much uncertainty in some cases. Some jurisdictions have a solid history of cases
where netting has been ultimately enforced. Others have no history
at all and the netting enforceability is quite uncertain. The CVA calculation has to price netting enforceability or the lack of it. Without
netting, counterparty exposures are higher.
CVA is a measure that is attributable to a netting node, that is,
to the portfolio of trades that can be netted at close out. A netting
node consists of one trade (when there is no netting across trades)
or many trades (when there is legally enforceable netting).
127
19/03/2010 17:11
It is important to be aware that CVA models attempt to value netting rights at present and in the future. Netting rights can change
over time, the overall trend being more netting once jurisdictions
evolve their legal frameworks and the history of netting cases
builds up. Netting rights can also change as the domicile of the
counterparty changes from one jurisdiction to another as trades are
assigned to different affiliated legal entities. CVA models have to
price future netting and future netting possibilities. Netting rights,
as priced in CVAs, may have option-like characteristics. Bank A has
to think about the present value of the cost of obtaining netting
rights in the future if those do not exist today, and consider reflecting them in the CVA. This is difficult and can be subjective, but
sometimes it can be material and needs to be considered.
Example:
Suppose that netting is deemed unenforceable today. Bank A calculates the CVA as US$100. If bank A could net, the CVA would
be US$10. Bank A assesses that it would cost US$20 to persuade
counterparty B to assign the OTC derivatives trades to its subsidiary C, which is domiciled in a netting-friendly jurisdiction. What
should be the CVA today? According to the discussion above, it
should be US$30 (as opposed to US$100) to include the present
value of bank As option to obtain netting. Of course the subjectivity of this assessment is the fact that the option is not in the form of
a contract. Yet, the economic value of the option is so large that it
cannot be ignored in the decision-making process of, for example,
determining the amount of hedges to be executed against the CVA.
Financial collateral is another powerful counterparty risk mitigant. It is extensively used in the market, especially between financial industry counterparties. A typical margin agreement stipulates
that trades are marked to market daily and, if the exposure is above
certain threshold levels, the counterparty that owes money to the
other has to post collateral. Cash and high-quality securities are the
usual collateral. Non-cash collateral is haircut to account for the uncertainty of its price due to market illiquidity and volatility.
PFE models account for collateral in the calculation of counterparty exposures. The models measure residual exposures that exist because of the thresholds that limit the continuous exchange
128
19/03/2010 17:11
of collateral and the market risks that exist during the closeout
of a defaulted counterparty. Market movements during the closeout period and the cost of replacing the trades with the defaulted
counterparty need to be accounted for properly. Bank A should not
underestimate the all-in costs of replacing trades with a defaulted
counterparty, especially when that counterparty is a large market
participant (eg, financial institutions like Lehman Brothers, other
OTC derivative dealers or a large hedge fund) and its default can
impair the liquidity and increase the volatility of the markets where
the derivatives trade.
The above has implications for counterparty risk measurement
and CVAs because it suggests that there is a (perhaps strong) component of wrong-way risk embedded in the residual, after-collateral credit exposures to large and important counterparties: their
default events are associated with market volatility and illiquidity
that can increase residual exposures. Those risks need to be modelled adequately by PFE models and priced in CVAs.
Other credit risk mitigants
Beside netting and financial collateral, there are other credit risk
mitigants to counterparty risks: non-financial collateral, letters of
credit, guarantees, early termination arrangements, like liquidity
puts and credit rating triggers.
Non-financial collateral (eg, physical assets like power plants,
airplanes, etc) are sometimes pledged as collateral against OTC
derivatives trades associated with special lending or investment
banking transactions. They can be reflected in the CVA calculation
by reducing sB appropriately and/or by explicitly modelling the
counterparty exposures below and above the estimated value of the
physical asset. The correlation between exposures, credit quality of
the counterparty and collateral are important to consider.
Example:
Suppose that an airline buys jet fuel forward from bank A and
pledges airplanes as collateral. Then the economy goes in recession,
jet fuel prices drop, the airlines revenues drop (due to less travelling) and, consequently, the market value of airplanes drop.
Letters of credit are usually obtained by a low credit quality
129
19/03/2010 17:11
counterparty from banks, and they guarantee the OTC derivatives

exposures up to a certain amount. The guarantees can be incorporated in the CVA calculation by reducing sB appropriately and/
or by explicitly modelling the counterparty exposures below and
above the amounts guaranteed.
Early termination arrangements like liquidity puts and credit
rating triggers provide for the early termination and settlement
of OTC derivatives trades prior to their final maturities. Liquidity
puts give the counterparties the option to terminate trades at certain future dates. Credit triggers establish that, if the public credit
rating of any one of the counterparties falls below a certain level,
the trades automatically terminate and settle. Those provisions can
and should be priced in CVAs but caution has to be exercised: bank
A will often face challenging client relationship issues to exercise
those options.
Currently, financial institutions are working very intensely to establish central counterparties (CCP). Those would provide broader
and more effective netting, as well as daily marking-to-market and
margining. Banks would face the CCP as opposed to each other,
and the CCP would be sufficiently capitalised and would impose
sufficiently strict margin requirements to eliminate counterparty
risks to a large extent. CCPs will lead to a greater standardisation of
derivatives over time, which is a positive trend.
REGULATORY CAPITAL: TRADING BOOK OR BANKING BOOK?
The Basel I and Basel II frameworks treat counterparty risk as an
item of the banking book. That is, the capital is the product of 8%
x EAD x RW. The exposure at default (EAD) is obtained from the
profile of expected exposures to the counterparty. In the early 2000s,
banks and their regulators made substantive progress in evolving
the counterparty risk regulatory capital treatment towards a more
risk-sensitive framework as part of the Basel II development process
(see Pykhtin and Zhu 2006, Rowe 2003 and Canabarro and Duffie
2003). But, despite much progress, the banking book capital treatment of counterparty risk in Basel II is greatly disconnected from
the risk management practices of the large and sophisticated banks.
CVA is a trading book item: banks mark it to market and the
most sophisticated banks hedge it actively. The recent financial cri130
19/03/2010 17:11
sis showed clearly that the banks that hedged CVA actively had
CVA losses that were substantially smaller than the banks that did
not hedge it.9
The regulatory capital treatment of counterparty risk (either in
Basel I or Basel II) creates important difficulties and disincentives
for active CVA hedging: the hedges of CVAs are traded financial instruments in interest rates, foreign exchange rates, equity and commodity prices and credit spreads. Thus, they are part of the banks
trading book while CVA is not. This creates a split hedge problem
for banks that actively hedge CVA and, perversely, the more hedges
a bank executes to neutralise the CVA volatility, the more it may get
charged regulatory capital on the naked hedges in its trading book.
This effect is highly undesirable from an economic point of view.
Economic capital
CVA risks are included in the economic capital models used by
banks. It is important to include both the regular mark-to-market
risks of the CVAs as well as the jumps-to-default (JTDs), especially
the JTDs of low credit quality counterparties.
An integrated economic capital model aggregates CVA risks
with other risks of the trading book and deals with all of them
consistently. The liquidities of the risks and their implications for
risk horizons are a crucial component of a sound economic capital
model.
Illiquid risks (eg, name-specific credit spreads that do not trade
in credit markets) should be associated with long risk horizons and
get treatment equivalent to buy-and-hold positions. However, it
is important to recognise that only the illiquid components of the
risks should be evaluated over long risk horizons. Components of
risks that are liquid can be hedged and should be associated with
short risk horizons.
Counterparty risk systems
Over the last 15 years banks have spent large amounts of resources
building information technology (IT) systems to enable them to
measure, price and risk manage counterparty risks. Those systems
are some of the most complex systems built by banks IT groups.
Size and complexity exist in almost all levels, including data sourc131
19/03/2010 17:11
ing, risk analytics, computation engines, risk reporting and downstream feeds.
The banks that implemented the most successful systems were
the ones that were able to modularise their systems; they built parallel processing capabilities and were conscious that extreme accuracy (which the author often calls false precision) needs to be sacrificed in favour of modularity and speed of calculations.
Figure 6.2 Main modules of a CVA calculation system
The data sourcing is broad and large. Data feeds include: legal entity definitions, netting and collateral agreements, legal opinions
on the agreements, guarantees, letters of credit, early termination
arrangements, trade descriptions and market data to calibrate the
simulation models and to price the trades. Those need to be captured across the entire bank and need to be aggregated appropriately. Usually the collection and preparation of the data (of size
measured in terabytes) accounts for half or more of the processing
time of CVAs.
The analytics require sound and pragmatic financial engineering.
Accuracy often has to be traded off against computational speed.
The design of the analytics as simple as possible (but no simpler,
as Einstein pointed out) and as fast as possible has to be the fundamental principle. It is typical that a CVA model will contain about
six broad market simulators and perhaps 300500 trade pricing calculators. It will use parallel processing and will store data in forms
that are adequate for future multiple uses.
The planning of what data to store and in what form is a key element for the success of the implementation. Storing tables (grids)
of trade values and exposures (many simulation paths, each with
many time horizons) per netting node seem to be quite useful storage units. Those structures can be re-used many times for multiple
132
19/03/2010 17:11
purposes. Typically, and daily, a large sophisticated bank will process millions of derivatives trades over 10002000 market paths,
each containing 50100 future dates from 050 years in the future.
The CVAs will also be calculated multiple times in order to construct the hedge ratios and risk metrics used by the bank.
The central processing unit (CPU) farms underlying CVA systems are large: a structure with thousands of CPUs is typical in the
largest sophisticated banks. Managing the workflow through such
a large farm is not trivial. Banks have invested many resources to
enhance their capabilities to dispatch tasks to the farm of CPUs
(sometimes contracted from external providers) and to collect back
the processed results. As technology continues to evolve, the capability of using faster and a greater number of CPUs is very important. This requires that the design of the CVA system be founded
on modularity and scalability. The author cannot emphasise those
requirements enough.
CONCLUSION
Counterparty risk management continues to be a challenging and
evolving field. We have achieved much over the last 15 years but
have still much to progress. The 20078 crisis provides clear evidence of the need for further work.
As credit markets grow more complete and liquid, and as financial technology evolves and is absorbed by the banks, the opportunities to enhance counterparty risk management practices will
continue to increase.
Regulatory capital frameworks need to recognise the market risk
nature of liquid counterparty risks since banks and other derivative users hedge actively counterparty risks as part of their trading
books.
The benefit of sound mark-to-market discipline as applied to
counterparty and other types of risk cannot be underestimated,
given its positive effect in shaping the behaviour and risk appetites
of banks and their traders.
The author expects to see continuing and much-needed progress
on all fronts of counterparty risk management over the next coming
years, particularly in:
133
19/03/2010 17:11
o
o
o
o
pricing and risk management models;

computational capabilities;
institutional and legal frameworks; and
regulatory capital rules.
Progress will require concerted efforts by market participants, lawmakers and regulators. Ultimately, as we have seen so evidently
in the 200708 crisis, we all share the benefits of and are responsible for preserving the integrity and sound functioning of the
financial markets. When counterparty risks increase during crises,
markets seize up and even healthy financial institutions can collapse because of their inability to find counterparties to trade and
manage their risks.
Panel 6.2Direction for the future
In the conclusion of an article in 2003 (Canabarro and Duffie),we said:
The traditional potential-exposure framework discussed[here] takes
a static buy-and-hold view of counterparty risks, without incorporating
the possibility of dynamic hedging of liquid credit risks. As the liquidity
and completeness of the credit derivatives market continues to grow,
more and more counterparty risks will become hedgeable. Static potential exposure measures will be superseded by Value-at-Risk (VaR)
and stress limits similar to those currently used for other liquid market
risks (such as those inherent in interest rates and currencies). Some
firms have adopted internal credit risk hedging via desks whose purpose is to provide credit protection, at a transfer price, to other desks
within the firm and then to choose how much of the firms net exposure
to each counterparty is to be traded to outside counterparties. More
decentralised firms have desk-by-desk hedging of credit risks.
These developments will have a fundamental impact in the asset
and liability management of financial institutions that deal in derivatives. A large part of the counterparty risk of these institutions will be
transferred via dynamic hedging, hopefully improving the risk sharing
among many economic agents and increasing market efficiency.
Much has evolved in the direction that we indicated (Panel 2.6).

But much still has to be accomplished. The direction for the future
remains the same.
The ideas and interpretations contained in this chapter are the authors own and do not intend to represent those of the authors
134
19/03/2010 17:11
employer, Morgan Stanley. The author would like to thank Aaron

Brown, Darrell Duffie, Eduardo Epperlein, Jon Gregory and David
Lamb for comments. The author is deeply grateful to Darrell Duffie,
Robert Litzenberger, Charles Monet, Evan Picoult and Tom Wilde for
fruitful exchanges on this subject for many years. The author is solely
responsible for all errors and omissions herein.
4
5
6
7
For a historical perspective of the banks and regulators work on counterparty risk, see the
reports of Gerald Corrigans Counterparty Risk Management Policy Groups (CRMPG 1999,
2005 and 2008) and the new Basel II framework as laid out in the Basel Committee on Banking Supervisions paper in 2005.
Bilateral CVAs have long been recognised as the proper valuation by academics, industry
practitioners and the US tax courts. See Sorensen and Bollier (1994), Duffie and Huang
(1996), United States Tax Court (2003), Duffies expert report (2001), Canabarro and Duffie
(2003), Duffie and Singleton (2003). However it was not until the implementation of FAS 157
in 2007 that the bilateral valuation became widely accepted and practiced by most US banks.
See Gregory (2009) for issues that are still discussed by practitioners today.
The implicit assumption used in the calculations above is that C does not have any outstanding trades with A prior to the assignment. If C had trades with A prior to the assignment, and
if those trades were nettable against the ones to be assigned, C could see a different value
for the portfolio trades being assigned. This is an important feature of counterparty risk: the
ability to net trades (at close-out) eliminates credit risk and banks should take advantage of
the opportunities to increase netting in their CVA trading.
Cases similar to our example can be found in the banks financial releases and statements
in 2008.
The ABS/CMBS CDO senior and super-senior tranches were in fact tranches of re-securitisations of ABS/CMBS bonds that, in their turn, were tranches of securitisations of ABS/
CMBS pools of loans. The bonds tended to be quite thin tranches (eg, 68% attachment-detachment) creating a sharply steep loss profile for the bonds and for their re-securitisations.
When the implied cum losses of the loan pools increased in 20078, the prices of the bonds
and the CDOs fell over the cliff.
See Merrill Lynchs press release July 28, 2008.
After Lehman Brothers filed for bankruptcy many counterparties chose not to close out their
trades with Lehman. This was particularly the case of counterparties that owed them large
sums of money on their out-of-the-money derivative trades.
We assume that there are no margin agreements between bank A and counterparty B. Margin agreements would eliminate much of the counterparty credit risks mentioned in those
examples.
As an illustration, compare the CVA losses disclosed in the 2008 Annual Reports of JPMorgan Chase and Citibank, and how the former benefited from hedging its CVA.
REFERENCES
Basel Committee on Banking Supervision, 2005, The Application of Basel II to

Trading Activities and Treatment of Double Default Effects, Bank of International
Settlements, July.
135
19/03/2010 17:11

in L. Tilman (ed) in Asset/Liability Management of Financial Institutions, (London: Euromoney Books).
Canabarro, E., E. Picoult and T. Wilde, 2003, Analysing Counterparty Risk, Risk,
16(9), pp. 117122.
Corrigan, E. G., 2002, The Boom-Bust Capital Spending Cycle in the US: Lessons
Learned, Per Jacobsson Lecture at the Annual Meeting of the World Bank and International Monetary Fund, September.
Counterparty Risk Management Policy Group I, 1999, Improving Counterparty
Risk Management Practices, June.
Counterparty Risk Management Policy Group II, 2005, Toward Greater Financial
Stability: A Private Sector Perspective, July.
Counterparty Risk Management Policy Group III, 2008, Containing Systemic
Risk: The Road to Reform, August.
Duffie, D., 2001, Expert Report of Darrell Duffie, in United States Tax Court, Bank
One Corporation, Petitioner v. Commissioner of Internal Revenue, Respondent.
Duffie, D. and M. Huang, 1996, Swap Rates and Credit Quality, Journal of
Finance, 51, pp. 921949.
Duffie, D. and K. Singleton, 2003, Credit Risk Modeling for Financial Institutions,
(Princeton University Press).
Financial Accounting Standards Board (FASB), 2006, Statement of Financial Accounting Standards No. 157 - Fair Value Measurements, September.
Gibson, M., 2005, Measuring Counterparty Credit Exposure to a Margined Counterparty, Finance and Economics Discussion Series, Federal Reserve Board, Washington, DC.
Gregory, J., 2009, Being Two-Faced Over Counterparty Credit Risk?, Risk, 22(2),
pp. 8690.
Litzenberger, R., 1992, Swaps: Plain and Fanciful, Journal of Finance, 47, pp. 831
850.
Pykhtin, M. and S. Zhu, 2006, Measuring Counterparty Risk for Traded Products
under Basel II, in M. Pykhtin, (ed.), Counterparty Credit Risk Modelling, (London:
Risk Books).
Rowe, D., 2003, Reason for Hope, Risk, September.
Rowe, D., 2006, The Arrogance of Hindsight, Risk, 19(10).
Sorensen, E. and T. Bollier, 1994, Pricing Swap Default Risk, Financial
Analysts Journal, 50, pp. 2333, MayJune.
United States Tax Court, 2003, Bank One Corporation, Petitioner v. Commissioner of
Internal Revenue, Respondent, May 2.
136
19/03/2010 17:11
The Counterparty Risk of Credit

Derivative Products
Jon Gregory
Ockham Financial Training and Consulting
The quantification of counterparty risk is often made under a key

simplifying assumption of no wrong-way risk. Wrong-way risk is
the term generally used to indicate an unfavourable dependence
between exposure and counterparty credit quality ie, the exposure is high when the counterparty is more likely to default. The
majority of derivatives transactions can be considered to have little
or no wrong-way risk, but credit derivatives are a crucial exception. In this chapter, we will look at the counterparty risk of credit
derivatives, in particular considering credit default swaps (CDSs),
tranches of credit portfolios and finally super-senior tranches. We
will see that wrong-way risk in credit derivatives transactions can
be devastating if ignored. The market is learning by experience
about the counterparty risks inherent in credit derivatives products, which must be controlled in order for recovery and growth
within this area to be possible.
The standard product in the credit derivatives market is a CDS,
which allows the transfer of default risk of one or more corporations or sovereign entities from one party to another. In a singlename CDS, the protection buyer pays a periodic fee (the premium) to the protection seller for a certain notional amount of debt
of a specified reference entity. Credit risk is transferred from the
protection buyer to the protection seller since, if the reference entity
undergoes a credit event,1 then the latter must compensate the
former for the associated loss. This loss is defined by the notional
137
19/03/2010 17:11
value of the contract multiplied by one minus the percentage recovery rate of the senior debt, and is achieved by means of a prespecified settlement procedure. The premium is paid until either
the maturity date or the credit event time, whichever comes first.
The reference entity is not a party to the contract, and it is not necessary for the buyer or seller to obtain the reference entitys consent to
enter into a CDS.
It is quite easy to understand the wrong-way risk inherent in
credit derivatives products. When buying protection in a CDS contract, exposure arises from a gain giving a positive mark-to-market (MtM), which is the result of the reference entities credit quality worsening. However, we would rather that the counterpartys
credit quality is not deteriorating also! In the case of a strong relationship between the credit quality of the reference entity and counterparty then clearly there is extreme wrong-way risk. On the other
hand, with such a strong relationship, then selling CDS protection
should be a right-way trade with little or no counterparty risk.
QUANTIFYING THE RISK OF JOINT DEFAULTS
The market standard model for pricing credit products sensitive to
joint defaults is the so-called Gaussian copula model, attributed
to Li (2000). Other more sophisticated approaches for CDS counterparty risk have been considered see, for example, Brigo and
Chourdakis (2008) and Lipton and Sepp (2009). However, due to
the uncertainty inherent in parametrising such models, we will focus on a simple and parsimonious copula approach in which the
only key parameter is the correlation between the default times of
the counterparty and reference entity.
The so-called copula approach to modelling default events allows a joint default probability to be defined by a two-dimensional
or bivariate Gaussian distribution function. For two names A and
B, with default probabilities pA and pB, the joint default probability,
pAB, is given by:
where F-1 represents the inverse of a cumulative Gaussian distribution function, F2 is the cumulative bivariate Gaussian distribution
function and rAB is the correlation between A and B, often referred
to as the asset correlation.
138
19/03/2010 17:11
The Counterparty Risk of Credit Derivative Products
We now have the means to calculate joint default probabilities

as a function of correlation. This is shown in Figure 7.1 for default
probabilities of pA = 1% and pB = 2%. For negative correlation, the
joint default probability is extremely small, tending towards zero at
maximum negative correlation (mutually exclusive events). With
zero correlation (independence), the joint default probability is
0.02% (1% x 2%) while at maximum correlation it increases to 1%.
Figure 7.1 Joint default probability as a function of asset correlation for two
counterparties with individual default probabilities of 2% and 1%
We can see that the magnitude of correlation is clearly crucial in

determining the impact of counterparty risk for two or more names.
COUNTERPARTY RISK IN CDS
Wrong-way risk in CDS and credit derivative products in general is
a direct consequence of the nature of the products and can lead to
serious counterparty risk issues. We start with a discussion of counterparty risk in the basic CDS product before extending the analysis to consider more complex credit derivative structures. The first
important point to note about a CDS contract is the highly asymmetric payoff profile as illustrated in Figure 7.2. When buying CDS
protection, there is a potentially large gain to be made if the underlying credit event is triggered. Indeed, the maximum exposure for
a protection buyer is 100% of the contract notional (zero recovery
139
19/03/2010 17:11
assumption). On the other hand, a potential gain when selling CDS

protection, while more likely, is significantly smaller.
Figure 7.2 Illustration of the asymmetry of counterparty risk for a five-year CDS
When selling protection, gains arise only from credit spread

tightening of the reference entity CDS premium (and additionally
time decay). For better rated names, the possible spread tightening
is more limited; for example, the assumption for an Aa credit is a
spread tightening of 50bp which, with a duration of 4.33, gives a
maximum exposure of 50 x 4.33 / 10,000 = 2.17%.
In addition to the asymmetry of the payoff in a CDS contract is
a potentially significant correlation effect. A protection buyer in a
CDS contract has a payoff with respect to a reference entitys default, but is at risk in case the counterparty in the contract suffers a
similar fate. If these default events are correlated then there is a serious wrong-way risk problem. This can be understood by considering the four possible cases of relevance when buying protection in
a single-name CDS transaction as illustrated in Figure 7.3.
Considering the possible sequence of events depicted in Figure
7.3, we can see that if the reference entity default is likely to be preceded by default of the counterparty (which is more likely if there
is a correlation between the default events), then the counterparty
risk is significantly increased (second case in Figure 7.3). Further140
19/03/2010 17:11
more, even if the reference entity does not default, then there could
be a significant loss of the MtM of the CDS protection at the counterparty default time (last case). If the counterparty default implies
a significantly positive MtM for the CDS protection (since the correlated reference entity is expected to have a worsening credit quality), then the loss would be expected to be significant.
Figure 7.3 Illustration of the payoff of a CDS contract subject to counterparty risk
The credit valuation adjustment (CVA) for a CDS is best considered as a special case compared to the general formula normally
used. This is because we have to consider the default time of both
the counterparty and the reference entity and, more specifically,
the order in which they occur. In particular, the computation of the
MtM of the remaining protection at the counterparty default time
represents an American Monte Carlo problem. An elegant solution
is provided by Mashal and Naldi (2005), who show that there are
upper and lower bounds for the value of protection that can be
computed reasonably easily, in addition Gregory (2010) gives an
analytical solution for the upper bound. Similar calculations have
been shown by Turnbull (2005). We will take the same approach
here and use a simple Monte Carlo simulation to value a CDS with
141
19/03/2010 17:11
counterparty risk with all results shown representing an average of

the upper and lower bounds (in most cases they are quite close).
We will ignore the impact of any collateral in the following analysis. This will be conservative since the use of collateral may be
considered to reduce significantly CDS counterparty risk. However, due to the highly contagious and systemic nature of CDS risks,
the impact of collateral may be hard to assess and indeed may be
quite limited, especially in cases of high correlation. We note also
that many protection sellers in the CDS market (such as monolines
and CDPCs, which are discussed later in this chapter) have not traditionally entered into collateral arrangements anyway.
We are interested in the risky value of buying CDS protection as
a function of correlation between the reference entity and counterparty (the counterparty is selling protection). We assume initially
that the fair premium of the CDS contract is 120bp2 and the counterparty is twice as risky (ie, their own CDS premium is 240bp). All
results assume recovery rates of 40% for both reference entity and
counterparty, and a maturity date of five years.
We start by assessing the size of the CVA for counterparty risk by
considering the fair premium (ie, reduced to account for counterparty risk) that should be paid in order to buy protection from the
risky counterparty as illustrated in Figure 7.4. We can also observe
the very strong impact of correlation: we should be willing only
to pay 100bp at 60% correlation to buy protection, compared with
paying 120bp with a risk-free counterparty. The CVA impact is
20bp or one sixth of the risk-free CDS premium. At extremely high
correlations, the impact is even more severe and the CVA can be
seen to be huge. At a maximum correlation of 100%, the CDS premium is just 48bp, which relates to the assumed recovery value.3
We now consider a reverse situation where counterparty risk is
less severe. We assume that the fair premium of the CDS contract is
240bp and the counterparty is half as risky (ie, their own CDS premium is 120bp). The fair CDS premium is shown in Figure 7.5. We
can notice that the contract does not contain as much counterparty
risk since the protection seller has a better credit quality than the
reference entity. We also notice that the counterparty risk vanishes
as the correlation goes to 100%. This is due to the fact that, with
perfect correlation, the riskier reference entity will always default
142
19/03/2010 17:11
first. This facet might be considered slightly unnatural. An obvious

way to correct for it would be to have some concept of joint default
of the reference entity and counterparty, or build in a settlement
period to the analysis. These points are discussed respectively by
Gregory (2009b) and Turnbull (2005).
(bp)
Figure 7.4 Fair CDS premium when buying protection subject to counterparty risk
compared to the standard (risk-free) premium
(%)
The counterparty CDS premium is assumed to be 240bp.
We finally repeat the analysis for the former case when selling
CDS protection, where the risky premium is higher in order for
the protection seller to be compensated for counterparty risk (as
it is assumed that there is no counterparty risk on the protection
buyers side). There is only a small CVA, which is diminished at
high correlation values (see Figure 7.6). This is intuitive since, at the
counterparty default time, it is unlikely that the MtM of the shortprotection CDS contract is positive (this would require a somewhat
unlikely improvement in the credit quality of the reference entity).
Results showing the minimal impact of counterparty risk on
short CDS protection positions are shown also by Turnbull (2005)
and Gregory (2009b). As noted by Brigo and Chourdakis (2009),
credit spread volatility, a missing component from the modelling
approach used here, may have the impact of increasing counterparty risk for such positions.
143
19/03/2010 17:11
(bp)
Figure 7.5 Fair CDS premium when buying protection subject to counterparty risk
(%)
The counterparty and reference entity credit quality are exchanged compared to Figure
7.4, and so the counterparty CDS premium is assumed to be 120bp.
(bp)
Figure 7.6 Fair CDS premium when selling protection subject to counterparty risk
(%)
However, the overall conclusion that long-protection CDS positions have severe CVA adjustments due to wrong-way risk, and
144
19/03/2010 17:11
short-protection CDS positions have only small adjustments (rightway risk), is generally true.
Note also that the calculations consider one-way or unilateral
counterparty risk rather than the increasingly common bilateral
CVA calculations that consider default of both parties. Indeed, in
the previous example, the long-protection counterparty risk was an
order of magnitude larger than that for short-protection positions.
Since the wrong-way risk tends to skew CVA towards one counterparty, a consideration of bilateral counterparty risk is not particularly important in the case of credit derivatives.
COUNTERPARTY RISK IN TRANCHES OF CREDIT DERIVATIVES
While CDS counterparty risk represents a challenge to quantify due
to the wrong-way risk and uncertainty of the correlation between
reference entity and protection seller (or buyer), structured credit
has given rise to even more complexities. A significant proportion
of the credit derivatives market is dominated by correlation products, including index tranches and synthetic collateralised debt obligations (synthetic CDOs). Compared to the default of just a single
name, index tranches and synthetic CDOs can be traded in CDS
form4 but cover pre-specified losses on some underlying portfolio.
For example, a X, Y% tranche will reference losses between X% and
Y% on the underlying index. The subordination or attachment
point of the tranche is X%, while Y% is referred to as the detachment point. The size of the tranche is Y-X%. The standard index
tranches for the most common DJ (Dow Jones) iTraxx Europe and
DJ CDX NA indexes5 are illustrated in Figure 7.7.
The index tranche that takes the first loss, 03%, is referred to as
the equity tranche, with the very high up tranches referred to as
senior or super-senior and the intermediate tranches referred to as
mezzanine (there is no definitive naming convention for individual
tranches). Index tranches vary substantially in the risk they constitute: equity tranches carry large amount of risk but promise potentially large returns, while tranches that are more senior have less
risk but pay only moderate coupons. At the far end, super-senior
tranches might be considered to have no risk whatsoever (in terms
of experiencing losses), a point we will analyse in more depth later
in the chapter. Tranching creates a leverage effect since the more junior tranches carry more risk than the index, while the most senior
tranches6 have less risk.
145
19/03/2010 17:11
Figure 7.7 Illustration of the index tranches corresponding to the DJ

iTraxx Europe and DJ CDX North American credit indexes
All tranches are shown to scale except the 22100% and 30100%.
Equity tranches 03% have always traded with an upfront premium and fixed running spread of 500bp to avoid the annuity risk
that exists for such a relatively high-risk tranche. For iTraxx, more
recently the 36% and 69% tranches have changed to trade in the
same way. The remaining tranches trade on a running basis. CDX
tranches trade at 500bp running for 03%, 37% and 710%, and
100bp running for 1015%, 1530%, and 30100%.
In addition to index tranches, there also exist many kinds of CDO
structure, which are all broadly characterised by their exposure to a
146
19/03/2010 17:11
certain range of losses on a portfolio. However, irrespective of trading convention and type of CDO, the important aspect for counterparty risk is that these structures reference only a certain range of
the losses on the portfolio. The analysis we make below will be for
index tranches (the simplest and most standard CDO product), but
the conclusions will hold for any unfunded CDO structure.
The counterparty risk problem is more difficult since we need to
assess where the counterparty might default compared to the names
in the underlying portfolio. We now extend the analysis of the previous section to discuss the impact of counterparty risk on the value
of credit derivatives index tranches. More details on the calculations can be found in Gregory (2009b) (as well as Turnbull (2005)
and Pugachevsky (2005)). We assume the index is trading at a level
of 120bp.7 Again, all recovery rates are assumed to be 40%, and we
use a five-year maturity. We assume initially that the counterparty is
more risky than the index on average with a CDS premium of 240bp
similar to the single-name case shown in Figure 7.4.
We first compute the fair CDS premium when buying protection
on a CDS index (ie, the 0100% tranche) that is shown in Figure 7.8
compared to the risk-free value. There is almost exactly the same
result as seen previously for a single-name CDS with equivalent
parameters (Figure 7.4). Hence, we can conclude that a credit index behaves in a very similar way to a similar single-name CDS in
terms of counterparty risk.
While a credit index behaves rather like a single-name in terms
of the counterparty risk, index tranches exhibit very different behaviour based on the losses they reference. The key point is that
tranches have very different exposures to underlying losses. It is
important to understand how the impact of counterparty risk can
change across the capital structure. We choose tranches according
to the standard iTraxx Europe portfolio that are defined by the attachment and detachment points (0%, 3%, 6%, 9%, 12%, 22%, 100%).
Since we are interested only in understanding the qualitative impact of counterparty risk for different tranches, we use the market
standard Gaussian copula model with a fixed correlation parameter of 50%, the premiums for which are shown in Table 7.1.8 Due
to constraints on the correlation matrix, this means we consider the
correlation between the counterparty default and the other names
in the portfolio in the range 0, 70%.9
147
19/03/2010 17:11
(bp)
Figure 7.8 Fair CDS premium when buying protection on a CDS index subject to
counterparty risk compared to the standard (risk-free) premium
(%)
Table 7.1 Index tranche premiums

calculated using a Gaussian copula
model with 50% correlation
Running premium (bp)
Index
120
03%
1750
36%
830
69%
545
912%
386
1222%
210
22100%
16.4
The premium of tranches varies substantially across the capital

structure from the most risky equity tranches to the least risky super-senior tranches. In order to compare all tranches on the same
scale, we plot the ratio of fair risky premium10 to the risk-free premium: this value will have a maximum at unity and decrease to148
19/03/2010 17:11
wards the recovery value11 as counterparty risk becomes more significant. These results are shown in Figure 7.9. We note that this
analysis does not consider the upfront premiums that occur as a
result of common trading conventions of many of these tranches.
Such impacts are considered in Gregory (2009b) but do not change
any of the results substantially.
Figure 7.9 Impact of counterparty risk across the capital structure
Fair risky tranche premium divided by the risk-free premium for all tranches in the capital structure and compared to the index (0100% tranche).
The results in Figure 7.9 show some important features. First,

as in the single-name case, an increase in correlation creates more
counterparty risk and reduces the fair risky premium. However,
the distribution of counterparty risk across tranches is very inhomogeneous. For the most risky 03% equity tranche, we can see
that the counterparty risk impact is actually quite small, even at
high correlation values. At the 40% recovery rate assumed, the equity tranche covers the first 6.25 defaults12 in the portfolio. Even
though the counterparty is riskier, the chance that it defaults at
149
19/03/2010 17:11
some point before the equity tranche has completely defaulted is

relatively small.13 Whereas the equity tranche has less risk (trading
on a running basis) than the index, all other more senior tranches
have more risk (except the 36% tranche at high correlations). Indeed, from a counterparty risk perspective, we can view tranching
as segregating the counterparty risk the more senior a tranche, the
more risk it contains on a relative basis. It is a very important point
that wrong-way risk increases for more senior tranches. Indeed, at
high correlation the two most senior tranches trade at a ratio tending towards the assumed recovery value (40%). The implication of
this is that by the time the tranche has taken losses, the more risky
counterparty is extremely likely to be in default, and hence only the
recovery fraction is received.
Figure 7.10 As Figure 7.9 but for a less risky counterparty with a CDS premium
of 90bp
The analysis concerned a situation where the counterparty is

more risky than the average of the names in the index. We briefly
150
19/03/2010 17:11
summarise results for a less risky counterparty with a CDS premium of 90bp in Figure 7.10. While the overall impact is, as expected,
not so significant we still see that there is still considerable counterparty risk, especially for the most senior tranches.
We can also note from Figure 7.10 that the extreme counterparty risk of the 22100% tranche is not significantly decreased from
trading with the counterparty that is over two and a half times less
risky. In this case, the high seniority of the tranche is more important than the actual default probability of the counterparty.
SUPER-SENIOR TRANCHES
The last section has illustrated the most severe counterparty arises
from the most senior tranches in the capital structure, referred to as
super-senior tranches. Consider, for example, the super-senior 22
100% tranche. Assuming an average 40% recovery, there need to be
45.8 defaults14 before this tranche takes any loss, and so the chance
that the counterparty is still around to honour these payments is
expected to be much smaller than for other tranches.
Figure 7.11 Fair premium when buying protection on the 22100% super-senior
tranche as a function of correlation with the parameters given in the text
The fair premium based on a recovery value is shown this assumes the counterparty
will default before any losses are settled.
151
19/03/2010 17:11
In Figure 7.11, we show the fair risky premium for this tranche
compared to the risk-free premium and the recovery value. Not
surprisingly, the counterparty risk impact is dramatic with the fair
premium tending towards the recovery value (40% of the risk-free
premium) at high correlation. In such a case there is virtually no
chance to settle any losses on the protection before the counterparty
has defaulted.
Super-senior tranches are interesting due to their high subordination and therefore relatively minimal default risk. Before the credit
crisis, it was not easy for banks to buy super-senior protection in
order to hedge their books.15 Conventional sellers of super-senior
protection were in short supply since, while the underlying default
risk was considered minimal, the premiums were not attractive to
most investors. Furthermore, the inherent counterparty risk created
another challenge for those needing to buy super-senior protection.
One way to attempt to get around the problems is to buy supersenior protection via a leveraged structure, a so-called leveraged
super-senior (LSS) transaction. The protection seller (investor) will
be able to get an enhanced return (due to the leverage) for taking
the risk on a super senior, almost default risk free, tranche while
the issuer will attempt to mitigate their counterparty risk by having the right to terminate the transaction. The leverage in an LSS
transaction reflects the fact that the investors cash participation is
less than the notional of the tranche. For example, a US$100 million
investment may be leveraged 10 times into a tranche with a notional of US$1 billion. The investor has sold protection on US$1 billion of protection but posted only US$100 million initial collateral
(the magnitude of these figures are representative of actual trades).
Generally, for a leverage of x times, the investor will initially commit 1/x units of collateral as illustrated in Figure 7.12.
There needs to be a mechanism to mitigate the risk that the LSS
protection buyer retains via the uncollateralised protection. This is
achieved using a trigger event, where the investor might have
the option to de-leverage by posting more collateral but will otherwise face the structure being unwound by the issuer at prevailing
market rates. The trigger has the effect of converting counterparty
risk into so-called gap risk. Counterparty risk is removed from
the structure since the protection buyer has an initial amount of col152
19/03/2010 17:11
lateral (often called independent amount or upfront margin), and

has also the right to unwind the transaction (hopefully) before the
value of the underlying tranche exhausts this level. While there is
no counterparty risk, there is the risk that the tranche value gaps
significantly in the protection buyers favour.16 Note that the protection seller may have the right, but will not be obliged, to make
payments above the initial level of collateral. Hence, if the tranche
MtM (and unwind costs) exceeds 1/x before the issuer is able to terminate the structure, they will suffer a loss even though the counterparty may not be in default.
Figure 7.12 Illustration of the LSS concept
A tranche is effectively multiplied by a leverage factor (x) and the resulting

tranche is collateralised as shown.

In defining a trigger, an issuer is trying to ensure that the percentage value of the tranche will always be below 1/a, with the likely
incorporation of some cushion that will be appropriate given the
153
19/03/2010 17:11
risk in unwinding the trade. The trigger definition represents a balance between a simple definition that may ease documentation,
understanding of the product and the related ratings process and
a more complex definition that leaves less unwind risk for the issuer. In practice, three types of trigger have been used: loss, spread
and MtM based. While different triggers have particular merits, the
general conclusions below hold for all trigger types.
An LSS might seem like a clever way to convert extreme uncontrollable counterparty risk into less extreme and controllable gap
risk.
LSS structures were initially valued as a full protection position less some gap risk component (analogous to a CVA) that was
argued to be small and sometimes even zero. Gregory (2008a)
makes a more thorough quantitative analysis and derives the valuation formulas for protection purchased in LSS, which is argued to
be substantially less valuable than the equivalent full protection.
Essentially, the gap risk created in an LSS structure is potentially
just as severe as the counterparty risk it replaced. The problem arises since, rather than having protection on an [A, B] tranche (say)
less some gap risk, the LSS value can be shown to be equivalent to
a much smaller tranche [A, A + (B - A)/x] (where x is the leverage
ratio as defined above) plus a complex trigger option due to the
issuers ability to unwind the transaction early. An illustration of
this pricing result is shown in Figure 7.13. The difference between
the incorrect value based on flawed gap risk assumptions and the
actual LSS value is substantial.
The pricing of an LSS represented in Figure 7.13 suggests that
there is a huge amount of gap risk in an LSS structure. Issuers
should have realised that there was a potential problem with LSS
structures in that the underlying default risk of the super-senior
tranches is so small. The tranches will only ever take losses in an
end of the world scenario with the structure being unwound half
way to the end of the world. Explained in such a way, it might
suggest that the ability to unwind the transaction would be of extremely limited practical use.
The extreme gap risk suggested by Figure 7.13 has been empirically proven. In 2007, a sudden wave of volatility in the credit
market meant that the triggers of LSS trades were severely tested.
154
19/03/2010 17:11
Super-senior tranches suffered from both credit spread widening

and increases in market-traded correlations, meaning the long-protection positions increased in value substantially. By August 2007,
many LSS structures were in significant danger of hitting unwind
triggers17 and the market was forced still wider (especially in terms
of implied correlations) due to these fears. It was practically impossible to buy the protection on super-senior tranches that would be
needed in order to unwind LSS transactions. Hence, the gap risk
was clearly severe and the chance of unwinding an LSS transaction without suffering massive losses was practically zero, despite
the original optimism from issuers and rating agencies about the
safety of the structures. An LSS is a fatally flawed structure because
the underlying counterparty risk in the super-senior tranche is converted into gap risk that is at least as complicated and toxic.
Figure 7.13 Schematic illustration of leveraged super-senior valuation as
described by Gregory (2008a)
MONOLINES
The most significant portion of super-senior protection in the credit
derivative markets has involved monoline insurers (technically
they enter into insurance contracts rather than selling protection
via credit derivative contracts, but this has no material difference).
Monoline insurers are financial guarantee companies that are tri155
19/03/2010 17:11
ple-A rated and provide super-senior insurance for a variety of

different credit portfolio structures. The triple-A rating is clearly
important due to the significant counterparty risk inherent in the
structures they trade, as illustrated in the last section.
A monoline will have equity capital against which it will invest
(sell protection) on a (much larger) notional of contracts, with the
ratio between the latter and the former defining an effective leverage of the company. Hence, a monoline is like an LSS with multiple
underlying transactions. Monolines typically achieve triple-A ratings based on conservative default risk stresses of their portfolio or
a capital adequacy model agreed with the relevant rating agencies
which must assess the possibility of default losses to be extremely
small. Most importantly, monolines do not typically post collateral;
this point both allows and probably also requires them to achieve a
triple-A rating. Due to the high leverage of a monoline, they have a
problem with negative mark-to-market changes on positions since
losses will be magnified by the leverage factor. Monolines do not
post collateral against positions (at least in normal times) to avoid
exposure to short-term market volatility and liquidity issues. Since
they carry triple-A ratings then surely it is unnecessary for them
to post collateral since they are extremely unlikely ever to fail. By
not posting collateral, a monoline can essentially try and ride the
wave of short-term volatility and illiquidity that may imply large
MtM losses on positions. In the end, this can be just considered simply noise as long as actual default losses never occur.
We note the posting of collateral means the crystallisation of losses for the monoline. Without the need to post collateral, a monoline
can always hope that any MtM losses will be regained at some point
in the future (if default losses do not occur). However, in the common scenario that ratings worsening of a portfolio are preceded by
MtM losses, then a monoline will be forced to realise significant
losses. The monoline may be unable to post collateral or unwind/
restructure trades to reduce the capital accordingly. The monoline
would then enter a death spiral from which it is unable to regain
its triple-A rating, and hence would be eventually be forced into a
so-called run-off. Run-off involves a monoline becoming static
and settling losses on transactions as and when they occur from
whatever equity capital is left.
156
19/03/2010 17:11
In December 2007, the market was around three or fours months

into a credit crisis that was to prove longer and more painful than
most market participants thought possible. Concerns started to
arise over the triple-A ratings of monolines and that they had insufficient capital to justify their ratings. However, this placed the ratings agencies in a subtle situation since the downgrading of monolines would potentially trigger a chain reaction. Investors would be
required to mark down assets due to the loss of the triple-A rating
and monolines would be forced to raise more capital due to being
required to post collateral against positions with worried counterparties. Swift ratings action could have therefore triggered an immediate crisis with so much resting on the viability of the monolines triple-A ratings. On the other hand, by leaving the monoline
rating alone, as long as the crisis did not worsen then some MtM
losses would eventually be recovered (due to spreads tightening
again on assets with minimal default risk) and the triple-A ratings
would again look firm.
Of course, the rating agencies should never have been placed (or
allowed themselves to be placed) in the difficult moral situation
where removal of a triple-A rating, while the correct action, would
cause insolvency of the monoline and probable immediate systemic
failure of all monolines. This situation should have been envisaged
at the time of first awarding ratings and should have been a large
clue that the whole concept of rating monolines was fundamentally
flawed. Rating agencies chose the wait-and-see approach to downgrading monolines at the start of the credit crisis, implicitly making
a bet that the crisis would be a short one. For example, in December
2007, Standard & Poors reaffirmed the rating of XL Financial Assurance Ltd with negative outlook. In late December, Fitch placed
their triple-A rating under review, saying that US$2 billion of new
capital needed to be raised (based on revised loss estimates that
were still overly aggressive in retrospect). By mid-2008, XL Financial Assurance Ltd had been downgraded below investment grade
by at least one rating agency. On May 27, 2009, an auction determined a final settlement value of 15 cents on the dollar for credit
default swaps referencing Bermuda-based Syncora Guarantee (the
monoline formerly known as XL Capital Assurance).
Many of the problems suffered by monolines in 2008 and 2009
157
19/03/2010 17:11
were caused by high leverage coupled with the unprecedented

increase in value of super-senior protection, leading to death spiral effects as discussed. The credit spreads of monolines widened
from 510 to several hundred basis points or more. Banks that had
bought super-senior insurance from monolines had to realise substantial losses due to the increased counterparty risk. For example,
as of June 2008, UBS was estimated to have US$6.4 billion at risk to
monoline insurers, while the equivalent figures for Citigroup and
Merrill Lynch were US$4.8 billion and US$3 billion respectively (Financial Times, 2008). We will now illustrate that the monoline venture into structured finance via taking super-senior risk was fatally
flawed from the outset.
We can very simply illustrate the basic flaw in the monoline business model with a simple calculation. Consider an institution has
bought protection from a counterparty (monoline) on a contract
with a payoff defined by a binary event B (B is zero if the event
has not occurred and 1 otherwise).18 The counterparty will not post
collateral against the position. Denote the current time by t and the
maturity date of the contract as T. Assuming zero interest rates, the
value of this contract is just the expected payoff V(t) = E[B] = q.
Denoting the counterparty default time by t and assuming zero recovery, the risky value is
where I(.) represents an indicator function which is unity if the
statement is true and zero otherwise and
is the default probability of the monoline in the period of interest. Now
assume a simple Gaussian relationship between the counterparty
default and payoff. The last term in the above equation, which is
identified as a CVA, can be written as
where F2d(.) is a cumulative Gaussian distribution function and r
is a correlation parameter. In Figure 7.14, we show, as a function of
correlation, the risky value for a digital contract with a risk-free value of 0.05% (akin to a super-senior tranche type probability) traded
with a counterparty (monoline) with default probabilities of 0.05%
and 0.l%.
158
19/03/2010 17:11
Figure 7.14 Simple model showing the value of a digital contract purchased with
no collateral posting as a function of correlation
Two different monoline default probabilities are shown.
It is clear from the example that wrong-way risk increases catastrophically with correlation (between the monoliness default
and the value of the underlying contract). Indeed, we can see that
counterparty risk depends less on the absolute credit quality of the
monoline and more on correlation. For example, at high correlation values the doubling of credit quality has an impact similar to
that of a decrease in correlation of only about 5%. The above point
suggests that, when trading with a monoline, we should be less
concerned about our credit quality and more concerned with the
correlation between credit quality and the value of the underlying
contract. This implies that a counterparty to a monoline, for example, should be highly concerned if the monoline is specialising in
one area (credit derivatives for example!).
Monolines represent a perfect storm in terms of counterparty
risk. First, they specialise in the rather senior (out of the money)
tranches. Second, their involvement in this market is so extreme
that the correlation between the value of such assets and their own
default probability is by construction extremely high. It could even
be argued that they are so highly exposed to an asset class that
159
19/03/2010 17:11
the correlation is close to unity. Third, they have focused strongly

on their triple-A ratings as a justification for having only minimal
counterparty risk. Even if this triple-A rating is correct then there
can be significant counterparty risk.
Figure 7.15 Illustration of the value of tranched protection purchased from a
credit insurer in a run-off or termination state
The quantitative treatment of counterparty risk presented by

monolines is explored in more detail by Gregory (2008b). The conclusion of this analysis is that an institution buying protection from
a monoline ends up with a structure that is even more complex than
the leveraged super-senior trade discussed in the last section. In addition to the problems inherent in the LSS structure, there are additional issues due to the uncertainty over when and if the protection
would be unwound19 and the changing leverage of the monoline. In
analogy with the LSS valuation, an institution buying protection on
an [A, B] tranche from a monoline effectively has protection on only
160
19/03/2010 17:11
a smaller A, A +(B -A)/x tranche (see Figure 7.15) and, additionally,

some complex optionality related to the possibility that the tranche
may be unwound by the monoline.20 The value of the protection is
inversely proportional to the leverage of the monoline, and so the
higher the leverage, the smaller the value of protection. As an example of the leverage of monolines, in late 2007 it was revealed that
ACA Financial Guaranty sold protection on US$59 billion against
capital of US$425 million (Das, 2008), which gives a leverage of 138
times (ACA was being downgraded to triple-C at this point).
The quantitative results of Gregory (2008b) summarised in Figure 7.16 were empirically tested almost immediately. For example,
in 2008, Merrill Lynch reported a net credit valuation loss of US$10.4
billion,21 largely because of counterparty risk related losses with
monoline insurance counterparties. Essentially, this stems from the
realisation that insurance purchased had little or no value due to a
massive CVA adjustment .
Figure 7.16 Illustration of valuation approaches for protection purchased
from a monoline insurer
The incorrect valuation arises from simply assuming that there is a small amount of
counterparty risk that can be assessed via the simple computation of a CVA; the actual value, that is calculated from a proper quantification of the relevant cashflows,
is significantly less and very complex to assess (Gregory, 2008b).
161
19/03/2010 17:11
SUMMARY
In this chapter, we have discussed the wrong-way counterparty risk
inherent in credit derivatives transactions. We have analysed the
impact of the dependence between exposure (the value of a credit
derivatives contract) and default probability of the counterparty,
so-called wrong-way risk. Wrong-way risk for credit derivatives
is significant and potentially devastating, increasing counterparty
risk to substantial, and sometimes even comical, levels. We have
discussed specifically single-name CDS contracts, index tranches
with a significant consideration given to the most dangerous supersenior risks that have been shown to create massive counterparty
risk problems. LSS tranches have been described as an example
where the attempted conversion of counterparty risk to gap risk
is fundamentally flawed. Finally, following on from the LSS ideas,
we have discussed the issues arising from monoline insurers acting
as super-senior protection sellers in the credit derivatives market,
which represents a perfect storm for counterparty risk and led to
billions of US dollars of losses for banks.
Controlling the counterparty risk in credit derivatives is not easy.
As the credit derivatives market recovers and develops, it is important to make proper use of robust risk mitigation techniques such
as collateral agreements and not rely on spurious triple-A ratings
and complex transactions such as LSSs. The credit derivative market has also provided some important lessons for the quantification
and control of wrong-way counterparty risk. Counterparty risk will
be a significant challenge for global financial markets in the coming
years and credit derivatives products are a major component of this
new dimension of financial risk.
1
2
3
4
Credit events included bankruptcy, failure to pay and, in some cases, debt restructuring.
All calculations assume flat credit curves, ie, equal CDS premiums for all maturities.
The premium based only on recovery value, ie, there is no chance of receiving any default
payment, is 120 x 40% = 48bp.
They can also be traded in funded form via a credit-linked note (CLN), which removes the
counterparty risk for both parties by utilising a special-purpose vehicle (SPV), although the
Lehman Brothers bankruptcy showed the basis of such SPVs to be potentially flawed.
These are the benchmark investment grade credit indexes. DJ iTraxx Europe contains 125
equally weighted European corporate investment grade reference entities, while DJ CDX
NA IG is similar, but references North American entities.
Due to their size, potentially only the super-senior tranches 22100% and 30100% will have
a leverage of less than one, and all other tranches could be more highly leveraged than
162
19/03/2010 17:11
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
the index.
This means approximately that the average CDS premium of the 125 names in the index is
120bp.
This does not produce prices close to the market, but the standard approach of base correlation used to reproduce market prices does not have an obvious associated way in which
to correctly price counterparty risk. We have checked that the qualitative conclusions of
these results hold at different correlations levels.
The upper limit for this correlation due to constraints of positive semi-definitiveness on the
.
correlation matrix is approximately
As before, we use an average of the upper and lower bounds.
This case corresponds to the counterparty always defaulting before the tranche takes losses,
and therefore the payment received is the recovery times the MtM of the tranche at the
counterparty default time. Hence, the fair premium is the risk-free premium times the recovery rate.
3% x 125 / (140%).
The counterparty must be one of the first seven defaults for there to be any counterparty
risk, since after this point the tranche is completely wiped out.
22% x 125 / (1-40%).
Required to either directly to offload risk in full capital structure trades or indirectly for
convexity hedging in CDO books.
Note that an LSS position will likely be hedged via selling the equivalent amount of unleveraged super-senior protection, and therefore an unwind will involve buying back this
protection.
Amazingly, such triggers were considered just months previously to be almost impossible to
hit, and triple-A ratings were given to LSS structures on this basis.
This is relevant for a single-name CDS and hence is only an approximation for a tranche, but
the general conclusion is not changed.
It is possible (but not certain) that a tranche might be unwound, for example if a monoline
was trying to reduce capital requirements.
Unlike the LSS case, the lower bound for the protection value is zero since the earlier losses
of other protection buyers may be settled first.
Merrill Lynch form 10-K.
REFERENCES
Brigo, D. and K. Chourdakis, 2008, Counterparty Risk for Credit Default Swaps,
Impact of Spread Volatility and Default Correlation, working paper.
Das, S., 2008, The Credit Default Swap (CDS) Market Will it Unravel?, February,
URL: www.eurointelligence.com/Article3.1018+M583ca062a10.0.html.
Gregory, J., 2008a, A Trick of the Credit Tail, Risk, March, pp. 8892.
Gregory, J., 2008b, A Free Lunch and the Credit Crunch, Risk, August, pp. 7477.
Gregory, J., 2009a, Being Two-faced Over Counterparty Credit Risk, Risk, 22(2),
pp. 8690.
Gregory, J, 2009b, Counterparty Credit Risk: The New Challenge for Global Financial
Markets, (Chichester: John Wiley).
163
19/03/2010 17:11
Gregory, J., 2010, Quantifying CVA in Credit Default Swaps in A. Lipton and A.
Rennie (eds), The Oxford Handbook of Credit Derivatives, (Oxford University Press)
Li, D. X., 2000, On Default Correlation: A Copula Function Approach, Journal of
Fixed Income, 9(4), March, pp. 4354.
Lipton, A. and A. Sepp, 2009, Credit Value Adjustment for Credit Default Swaps
Via the Structural Default Model, The Journal of Credit Risk, 5(2), pp. 123146.
Mashal, R. and M. Naldi, 2005, Pricing Multi-name Default Swaps with Counterparty Risk, Journal of Fixed Income, 14(4), pp. 316.
OKane, D., 2008, Pricing Single-name and Portfolio Credit Derivatives (Chichester:
Wiley).
Pugachevsky, D., 2005, Pricing Counterparty Risk in Unfunded Synthetic CDO
Tranches, in M. Pykhtin, (ed.), Counterparty Credit Risk Modelling, (London: Risk
Books).
Turnbull, S., 2005, The Pricing Implications of Counterparty Risk for Non-linear
Credit Products, in M. Pykhtin, (ed.), Counterparty Credit Risk Modelling, (London:
Risk Books).
164
19/03/2010 17:11
Contingent Credit Default Swaps

Andrew P. Hollings, Shankar Mukherjee and
Svein Stokke
Novarum Partners Ltd
The days when counterparty credit risk (CCR) could be regarded as

a benign artifact of the over-the-counter (OTC) derivatives market
are now over. The failure of Lehman Brothers and the subsequent
unprecedented disruption in financial markets has convinced policy
makers that the magnitude of CCR retained by market participants
represents an unacceptable systemic risk to the stability of financial
markets. While the proposed establishment of a central counterparty (CCP) to clear OTC derivative transactions will reduce the
operational risk between professional parties willing and able to
post collateral, it will do little to alleviate the unsecured CCR arising from OTC derivatives transacted with end-users unwilling or
unable to post collateral. In response, market participants have created the contingent credit default swap (CCDS) to hedge unsecured
CCR. The pay-off profile of the CCDS is designed to replicate the
unsecured CCR to a specified counterparty (reference credit) arising from a specified OTC derivative transaction (reference transaction). Executed under standard ISDA documentation, the CCDS
can be fully integrated into the calculation of CCR. This provides
market participants with a single, standardised hedge instrument
that can be employed to construct a robust, transparent and scalable risk practice that alleviates CCR constraints, mitigates credit
valuation adjustment (CVA) volatility and reduces both economic
capital and risk-weighted assets. This chapter considers contingent
CCDSs as effective hedges for market-driven counterparty expo165
19/03/2010 17:11
sures. We show various features of the structure with respect to

counterparty exposure measurement, CVAs and economic capital.
MEASURING COUNTERPARTY CREDIT RISK
The mere mention of OTC derivative CCR is enough to intimidate
even the most seasoned risk manager. Many market practitioners
mistakenly believe that this esoteric artifact of the OTC derivatives
market is too complex to accurately measure, let alone effectively
hedge. However, while the models and methods employed to measure and report CCR are indeed complex, the conceptual framework underlying these measurements is straightforward. Unlike
traditional debt instruments, for which the exposure at default
(EAD) is contractually known at inception, the EAD of an OTC derivative contract is determined by referencing the prevailing price
of one or more market indexes. While the prices of these market indexes are known today, their future prices are uncertain. Therefore,
in order to determine the EAD over the life of an OTC derivative
contract, a risk manager needs to forecast the future prices of all
market indexes used in the calculation of its closeout value.
Exposure at default: the basic building block
Calculating the EAD of an OTC derivative contract requires a methodology for forecasting the future prices of all relevant market indexes. The most popular solution to this forecasting problem is to
use Monte Carlo methods to simulate distributions of prices for
every market index used in the calculation of the closeout value of
all outstanding OTC derivative contracts for future dates up to and
including the longest maturity date.1 These market data tables, each
consisting of a discrete set of time steps on the x-axis and simulated
market indexes on the y-axis, are then used to generate an EAD distribution table for each OTC derivative contract, consisting of a distribution of future closeout values at each discrete time step. Next,
the risk manager must determine if the OTC derivative contract is
governed under a legally enforceable ISDA Master Netting Agreement (referred to collectively as a netting set). Where a netting set is
comprised of a number of OTC derivative contracts, the economic
benefit of closeout netting in default is captured by simply adding
together the tables of future closeout values of each OTC deriva166
19/03/2010 17:11
CONTINGENT CREDIT DEFAULT SWAPS
tive contract.2 Finally, where the bankruptcy code requires that a

closeout liability be paid by the non-defaulting party to the defaulting party (in the absence of an enforceable recision or walk-away
clause), all negative future closeout values are set to zero.3 These
netting set EAD distribution tables are the basic building blocks
from which all CCR metrics are derived.
In summary, to construct a netting set EAD distribution table, a
risk manager must:
o generate market data tables consisting of simulated future prices
for all relative market indexes in the portfolio;
o for each OTC derivative contract, construct a distribution of future closeout values at each of the time steps in the market data
tables;
o if a legally enforceable netting set exists, sum together the future
closeout values for all OTC derivative contracts in the netting
set; and
o isolate exposures by setting all negative future closeout values
to zero.
CCR metrics
The netting set EAD distribution tables are used by risk managers,
in combination with additional inputs and assumptions, to produce
a suite of CCR metrics to measure and report specific characteristics
of the CCR they retain. It is important to remember that, given the
high degree of abstraction and reduction employed in the derivation of these metrics, each metric alone is not meant to provide a
complete description of counterparty EAD.
Expected and peak positive exposure profiles
Given these netting set EAD distribution tables, the risk manager
can now analyse the CCR to a specific legal entity or a family of
legal entities by simply adding together the pertinent netting set
EAD distribution tables to create a single counterparty EAD distribution table. From this counterparty EAD table, the risk manager
can derive a rich set of risk metrics to measure and report CCR. A
popular metric used by many risk managers is the expected positive exposure, defined as the mean (average) of the EAD distribution for a stipulated future date. Typically, the expected positive
167
19/03/2010 17:11
exposure is calculated for a series of stipulated future dates, up to

and including the longest contract maturity, to create an expected
positive exposure profile. In addition, many risk managers measure
the potential variance from the expected positive exposure profile
by sampling the EAD distribution table at higher percentiles (typically 95% or 99%) to create a peak positive exposure profile. The
expected and peak positive exposure profiles are the primary risk
metrics used by risk managers to approve, allocate, monitor and
report retained CCR.
Credit valuation adjustment4
US GAAP and IAS accounting rules require companies to make a
CVA to the mid-market value of outstanding OTC derivative transactions, so that they reflect the fair market value of the CCR inherent in these contracts. A simplified CVA model can be derived from
the counterparty EAD distribution table and the credit spread of
the counterparty.5 If the creditworthiness of the counterparty is assumed to be independent of the market prices used in the determination of EAD, CVA can be computed by applying the counterpartys forward credit spread to the expected positive exposure profile
to construct an expected CVA profile. The present value of the expected CVA profile is the CVA. However, if the creditworthiness of
the counterparty is correlated with the market prices used in the
determination of EAD,6 the calculation of CVA requires additional
steps. First, a table of credit spreads incorporating their correlation
with market prices is derived from the market data tables. These
correlated credit spreads are applied to the counterparty EAD distribution table to create a counterparty CVA distribution table. The
expected CVA profile is defined as the mean (average) of each CVA
distribution for future dates up to and including the longest contract maturity date. Once again, the present value of the expected
CVA profile is the CVA.
Economic capital
CVA can be viewed as a measure of expected losses associated with
the CCR arising from the OTC derivative transactions outstanding
with a specific counterparty. However, risk managers may also wish
to measure the economic capital required over and above the CVA
168
19/03/2010 17:11
to support unexpected losses at a selected confidence level. This can

be achieved by constructing new tables of stressed credit spreads,
which can then be applied to the counterparty EAD distribution
table to create stressed counterparty CVA distribution tables. The
stressed counterparty CVA distribution tables can be sampled at
any desired percentile to create a stressed CVA profile. The present
value of the stressed CVA profile provides the risk manager with a
stressed CVA to compare with the current CVA.
Figure 8.1 Measuring counterparty credit risk
169
19/03/2010 17:11
Risk-weighted assets
Regulated entities are required to translate and report CCR in units
of risk-weighted assets (RWA) using the methodology prescribed
in the Basel II Capital Accords.7 This translation begins with the reduction of the expected positive exposure profile over the first year8
into a single exposure metric, the effective expected positive exposure (EEPE). EEPE is defined as the average of the expected positive
exposures9 over the first year, constrained to be non-decreasing and
weighted in proportion to the time interval each expected positive
exposure value represents of the first year. The translation of CCR
into RWA is completed by multiplying the EEPE by the following
factors:
o Alpha (). As noted in the calculation of CVA, it may be inappropriate to assume that EAD is independent of market prices
or that the portfolio of outstanding OTC derivative contracts is
fully diversified by counterparty. To address this, EEPE is multiplied by a factor . Pursuant to the Basel II Capital Accords, is
set at 1.4, with regulatory discretion to impose a higher or lower
multiplier where appropriate, subject to a floor of 1.2.
o Counterparty risk weighting (RW). The RW captures the creditworthiness of the asset through the counterpartys probability
of default (PD) and loss given default (LGD), and the duration
of the expected positive exposure profile through the effective
maturity (EM) adjustment.10
COUNTERPARTY CREDIT RISK MITIGATION
As mentioned above, CCR can no longer be seen as a benign artifact of the OTC derivatives market. The downfall of Lehman Brothers and the disruption of the financial markets has highlighted the
magnitude of CCR retained by market participants and the lack of
effective market mechanisms to mitigate CCR.
Exponential growth
The OTC derivatives market is a victim of its own success. According to the Bank for International Settlements, the total notional of
outstanding OTC derivative contracts grew from US$1 trillion in
1987 to US$592 trillion by the end of 2008. Inevitably, such dramatic
growth has come at a cost. At the end of 2008, OTC derivative deal170
19/03/2010 17:11
ers reported that the current EAD, defined as the gross value of
contracts that have a positive market value after taking into account
legally enforceable bilateral netting agreements but before collateral, stood at US$4.736 trillion.
Figure 8.2 OTC derivatives outstanding (notional amount, US$ trillions)*
*According to ISDA and Bank for International Settlements.
Market enhancements
Policy makers have decided that the current architecture of the
OTC derivatives market represents an unacceptable systemic risk
to the stability of financial markets. In response, the International
Swaps and Derivatives Association (ISDA) has sponsored a wide
range of initiatives to enhance market standardisation, price transparency and effective CCR mitigation. These initiatives fall into two
broad categories:
o the establishment of a central counterparty (CCP) to clear OTC
derivative transactions to mitigate the operational risk between
professional parties willing and able to post collateral; and
o the establishment of a simple market mechanism to enable and
encourage market participants to hedge the unsecured CCR arising from OTC derivatives transacted with end users unwilling or
unable to post collateral.
Caveat utilitor
The early attempts of risk managers to mitigate CCR were, at best,
171
19/03/2010 17:11
an attempt to manage CCR within broad acceptable limits through

the management of aggregate CVA or expected positive exposure.
However, as previously noted, given the degree of abstraction and
reduction employed in the derivation of these metrics, even if a risk
manager were successful in replicating these metrics at any single
point in time, the risk manager could not assume that the underlying
source of CCR, counterparty EAD, had been effectively mitigated.
For example, some risk managers chose to manage aggregate CVA
as a proxy for CCR, by constructing a replicating hedge portfolio of
traded products to offset changes in its economic value. However,
even if the market is sufficiently complete to establish this hedge
portfolio, individual risk positions would need to be constantly rebalanced through time to ensure that the hedge portfolio remained
effective. The cost of this rebalancing is uncertain. Therefore, while
this hedging strategy may be successful in reducing CVA volatility,
it is difficult to claim an effective transfer of CCR when the cost of
this hedging strategy is only known ex post. Furthermore, in practice given the incomplete nature of markets the replication of
aggregate CVA has proven to be very difficult, extremely complex
and computationally onerous.
Similarly, those risk managers who chose to focus on the management of expected positive exposure as a proxy for CCR faced similar problems. Since expected positive exposure is usually approved,
monitored and reported within a conventional loan framework,11
some risk managers assumed that expected positive exposure
could be hedged using conventional credit default swaps (CDS).
However, since expected positive exposure is constantly changing,
these CDS hedges need to be constantly rebalanced. Once again,
while this hedging strategy may be successful in reducing aggregate credit exposure, it is difficult to claim an effective transfer of
CCR when the cost and notional amount of CDS required to hedge
CCR is only known ex post.
Contingent credit default swap
In light of the difficulties described above, risk managers began to
reevaluate their approach to the mitigation of CCR. They soon concluded that what they needed was a product that would precisely
replicate the EAD of an OTC derivative contract, that could be fully
172
19/03/2010 17:11
integrated into the derivation of counterparty EAD and that would

automatically be reflected in the downstream calculation of expected/peak positive exposure, CVA, economic capital and RWA. To
this end, market participants created the contingent credit default
swap (CCDS). Governed under standard ISDA documentation12
and executed with an independent third party,13 a CCDS is a simple
variation of a standard CDS where the notional amount is defined
as the closeout value of a specified OTC derivative transaction (reference transaction) determined upon an event of default of a specified counterparty (reference credit), subject to a minimum value of
zero. Since an OTC derivative transaction is not transferable without the mutual consent of both parties, the settlement of the CCDS
is determined by reference to a publicly traded security of the reference credit (reference obligation). To effectively replicate the EAD
of an OTC derivative contract, the CCDS reference obligation must
have a level of payment priority equal or junior to the OTC derivative claim in the event of default. Furthermore, the events of default
stipulated under the OTC derivative contract must incorporate the
events of default of the CCDS reference obligation either implicitly
or explicitly through the inclusion of a general cross-default clause.14
CCDS RT EAD distribution table
The CCDS reference transaction (CCDS RT) is included in the determination of counterparty EAD as a simple contra EAD distribution table. Deriving the CCDS RT distribution table is identical to
the derivation of the EAD distribution tables for each underlying
OTC derivative contract. First, using the same market data tables,
an EAD distribution table for the CCDS RT is generated, consisting
of distributions of future closeout values of the CCDS RT at each
discrete time step. Next, since the notional amount of a CCDS contract can only be positive, all negative future closeout values in the
CCDS RT EAD distribution table are set to zero. The CCDS RT EAD
distribution table is then integrated with the counterparty EAD distribution table by simply subtracting the table of future closeout
values of the CCDS RT. The final step is to eliminate any net negative future closeout values from the counterparty EAD distribution
table by setting them to zero. This ensures that only those market
scenarios where the CCDS RT is an effective EAD mitigant are in173
19/03/2010 17:11
cluded in the final counterparty EAD distribution table.

In summary:
o for the CCDS RT, construct a distribution of future closeout values at each of the time steps in the market data tables;
o set all negative future closeout values to zero;
o subtract the future closeout values of the CCDS RT from the
counterparty EAD distribution table; and
o isolate the net exposures by setting any net negative future closeout values to zero.
Figure 8.3 Hedging counterparty credit risk
174
19/03/2010 17:11
EAD substitution
The EAD of a CCDS cannot be smaller than the EAD of the CCDS
RT, given that the CCDS notional amount is contractually defined
as the closeout value of the CCDS RT, subject to a minimum value of
zero. Therefore, a conservation of aggregate CCR is ensured, since
any reduction in EAD to the counterparty arising from the inclusion of the CCDS RT will be fully offset by an increase in EAD to the
CCDS provider, before the inclusion of collateral.15 The implementation of the CCDS and CCDS RT requires only a minor extension
of existing processes and models. Furthermore, where the CCDS
RT is a plain vanilla instrument, existing models can be reused to
derive the CCDS RT distribution tables.
CCDS effectiveness
Whether or not the CCDS gives rise to a reduction in aggregate
CVA, economic capital or RWA will depend upon both the efficacy
of the CCDS RT and the creditworthiness of the CCDS provider.
The efficacy of the CCDS RT is self evident when integrating the
CCDS RT EAD distribution table with the counterparty EAD distribution table. CCDS RT effectiveness is maximised by minimising the absolute value (both positive and negative) of all net future
closeout values in the counterparty EAD distribution table. For example, where a single OTC derivative transaction is hedged with
an identical CCDS RT, all future closeout values in the counterparty EAD distribution table will net to zero.16 Finally, the higher the
credit quality of the CCDS provider relative to the CCDS reference
credit, the greater the aggregate reduction in CVA, economic capital
and RWA, ceteris paribus.17
CCDS RISK PRACTICE
The CCDS provides market participants with a single standardised
hedge instrument which can be employed to construct a robust,
transparent and scalable risk practice to alleviate CCR constraints,
mitigate CVA volatility and reduce economic capital and RWA. Presented below are four examples to illustrate how risk managers can
optimise their risk preferences using CCDS. To highlight the transfer and mitigation of CCR, the following simplifying assumptions
have been made:
175
19/03/2010 17:11
o there is only one OTC derivative transaction outstanding with

the counterparty;
o the specification of the CCDS RT is identical to that of the outstanding OTC derivative transaction;
o the CCDS is executed with a highly creditworthy counterparty
and is fully cash collateralised under a legally enforceable collateral agreement; and
o the collateral agreement is assumed to be fully effective in defeasing the EAD to the CCDS provider18
Figure 8.4 Standard CCDS
Standard CCDS
In Figure 8.4, the risk manager completely mitigates the counter176
19/03/2010 17:11
party expected positive exposure profile through the use of a standard CCDS contract. The original counterparty expected positive
exposure profile is replaced by the expected positive exposure profile to the CCDS provider, which is subsequently defeased by the
collateral agreement.
Figure 8.5 Forward starting CCDS
Forward starting CCDS

In Figure 8.5, the risk manager retains the original counterparty expected positive exposure profile out to a specific future date. The
expected positive exposure profile beyond this date is mitigated
through the use of a forward starting CCDS, which is a variation of
177
19/03/2010 17:11
a standard CCDS in which the start date of the CCDS is a future date
that is contractually specified at the CCDSs inception. The original
counterparty expected positive exposure profile is completely mitigated and replaced by an expected positive exposure profile to the
CCDS provider beyond the start date of the CCDS. The expected
positive exposure profile to the CCDS provider is subsequently defeased by the collateral agreement.
Figure 8.6 Short maturity CCDS
Short maturity CCDS

In Figure 8.6, the risk manager retains the original counterparty expected positive exposure profile only beyond a specific future date.
178
19/03/2010 17:11
The expected positive exposure profile before this date is mitigated

through the use of a short maturity CCDS, which is a variation of a
standard CCDS in which the maturity date of the CCDS is shorter
than the maturity date of the CCDS reference transaction. The original counterparty expected positive exposure profile is completely
mitigated and replaced by an expected positive exposure profile
to the CCDS provider up until the maturity date of the CCDS. The
expected positive exposure profile to the CCDS provider is subsequently defeased by the collateral agreement.
Figure 8.7 Threshold CCDS
179
19/03/2010 17:11
Threshold CCDS
In Figure 8.7, the risk manager caps the original counterparty
expected positive exposure profile at a predetermined amount by
using a threshold CCDS, which is a variation of a standard CCDS
in which the CCDS notional amount is contractually defined as
the closeout value of the CCDS reference transaction less a predetermined threshold amount, subject to a minimum value of zero.
The original counterparty expected positive exposure profile is
completely mitigated above the threshold amount and replaced by
an expected positive exposure profile to the CCDS provider. The
expected positive exposure profile to the CCDS provider is subsequently defeased by the collateral agreement.
CCDS toolkit
By using simple combinations of CCDS, risk managers can precisely define the amount of CCR they wish to retain, both by magnitude and maturity. Once these CCDS hedges are in place, there is no
additional action that risk managers need to take, unless their risk
preference changes, in which case they can simply amend the specification of the CCDS hedges accordingly. Where counterparty EAD
arises from a portfolio of outstanding OTC derivative transactions,
Figure 8.8 CCDS toolkit
180
19/03/2010 17:11
a portfolio of CCDS hedges can be constructed using standard optimisation techniques to mitigate the CCR arising from the most
complex counterparty EAD distribution table.
Figure 8.9 Optimising CCR metrics
Optimising downstream metrics

The effective transfer of counterparty EAD to the CCDS provider
and any subsequent defeasance through the action of collateral,
181
19/03/2010 17:11
is automatically reflected in all downstream CCR metrics. This allows the risk manager to select which CCR output metric against
which to optimise. Where the amount of CCR retained by a risk
manager is not constrained by underlying credit risk appetite, but
rather by the magnitude and volatility of CVA, a portfolio of CCDS
hedges can be constructed to specifically mitigate the high beta
contributors to aggregate CVA. Similarly, where the risk manager
is required to operate within overall economic capital or regulatory
RWA constraints, a portfolio of CCDS hedges can be constructed
to specifically target the counterparty EADs that have the largest
marginal impact on economic capital or RWA.
ECONOMICS OF HEDGING
Effective CCR mitigation will result in a reduction in the magnitude
of CVA, economic capital and RWA. This reduction may arise from
either an increase in the overall aggregate credit quality of retained
CCR and/or a reduction in aggregate counterparty EAD, ceteris paribus. The marginal economic benefit of CCR mitigation will vary
between firms given differences in models and methods used in
the derivation of CCR and in the marginal cost of capital. Where
the marginal economic benefit of CCR mitigation is greater than
the cost of CCDS protection, the risk manager has a clear economic incentive to hedge. There is no mandated single set of models,
methods and market inputs with which to derive and report CCR.
Therefore, even where the underlying portfolio of OTC derivative
transactions are identical, the magnitude of CCR arising from these
transactions will vary between firms. Furthermore, notwithstanding differences in the derivation of CCR, the economic benefit of a
reduction in economic capital and RWA is driven by a firms marginal cost of capital. Since the cost of capital can vary significantly
between firms, the marginal economic benefit of CCR mitigation
will also vary between firms, being greater for those firms with a
higher marginal cost of capital.
CCDS PRICING
A number of market vendors support a general form of the CCDS
standard valuation model (CCDS-SVM). The CCDS-SVM is defined as a CDS with a time-varying notional determined as a se182
19/03/2010 17:11
ries of options on the CCDS reference transaction where changes

in the market value of the CCDS reference transaction are assumed
to be independent of changes in the market credit spread of the
CCDS reference credit. The CCDS-SVM provides market participants with a simple valuation model with which to benchmark the
price of a CCDS. Many market participants subsequently augment
the CCDS-SVM price to reflect qualitative assessments of market
imperfections, expected hedge costs and potential correlations between the CCDS reference transaction and reference credit.
CONCLUSION
The unprecedented disruption witnessed in the financial markets
has highlighted the need for an effective and transparent market
mechanism to mitigate CCR. The simplicity and ease of implementation of the CCDS enables CCR mitigation to be fully captured and
integrated within the CCR metrics of the firm, and provides risk
managers with a simple and scalable toolkit with which to recycle
scarce credit and capital resources.
1
2
3
4
6
7
8
9
10
11
12
13
While underlying Monte Carlo methods are generally similar, there are large differences in
how these methods are parameterised between firms, especially with respect to non-observable variables such as correlations and long-term volatilities.
Where closeout netting is not legally enforceable, each individual OTC derivative contract is
treated as a separate, stand-alone netting set.
Where closeout liabilities are represented as negative values and closeout assets as positive
values.
The CVA is not to be confused with the debt valuation adjustment (DVA), which is made to
the mid-market value of OTC derivative transactions to reflect the market value of expected
closeout liabilities.
The CVA can be implied from estimates of the counterpartys probability of default (PD)
and loss given default (LGD), or derived from the market prices of securities issued by the
counterparty and/or credit default swaps referencing such securities.
Commonly termed wrong-way or right-way risk.
Basel Committee on Banking Supervision: International Convergence of Capital Measurement and Capital Standards, June 2006.
Or the longest contract maturity date, if all contracts mature before one year.
Please note that the definition of expected positive exposure used in this chapter is referred
to as expected exposure in the Basel II Capital Accords.
The calculation of the effective maturity adjustment is prescribed in the Basel II Capital Accords and is subject to a cap of five years.
Expected positive exposure is still referred to as loan equivalent exposure in many firms.
ISDA Confirmation for Contingent Credit Default Swap Transaction (February 6, 2007).
To demonstrate an effective transfer of CCR, the CCDS provider must be independent of
the CCDS purchaser.
183
19/03/2010 17:11
14 It is common for an ISDA Master Agreement to include additional events of default not
present in the publicly traded debt of the counterparty. This asymmetry creates a timing
advantage over creditors of equivalent or more senior status by enabling the non-defaulting
party to pursue a claim against the defaulting party prior to the instigation of formal insolvency proceedings.
15 If the CCDS is executed under a legally enforceable collateral agreement, the risk manager
can reduce the EAD to the CCDS provider to take account of the risk mitigating effect of
the collateral.
16 Many risk managers use simple optimisation techniques to determine the portfolio of CCDS
that optimise their risk preference.
17 Ignoring any benefit from the application of double default treatment.
18 In practice, a residual EAD is retained to the CCDS provider, representing the potential time
period during which a collateral call can be disputed by either party. This time interval is
commonly referred to as the cure period. Since the cure period is generally only five to
seven business days, for simplicity of representation, the resulting residual EAD is assumed
to be de minimus.
19 A total convergence in CCR models and methods is desirable but impractical given the large
number of asset classes and bespoke pay-off profiles that are transacted in the OTC derivatives market.
184
19/03/2010 17:11
Funding Benefit and Funding Cost

Yi Tang and Andrew Williams
Morgan Stanley1
OVERVIEW
Business significance
Under 2010 market conditions, credit valuation adjustment (CVA)
has grown to be one of the most significant risks for a bank or securities firm. A closely related topic of increasing criticality is the funding
of over-the-counter (OTC) derivatives, as it significantly affects an
entitys liquidity and cost of running an OTC derivatives business.
Failure to obtain adequate funding is a leading factor that drives an
entity to default or to file for Chapter 11 (bankruptcy protection).
Thus, the risks involved in funding need to be properly priced
and actively managed. The net funding cost of the OTC derivatives
business for a major bank or a securities firm can easily reach hundreds of millions of US dollars per annum.
While there are many ways for an entity to obtain funding, such
as secured or unsecured funding, derivatives payables and equity
funding (such as regulatory capital incurred by the OTC derivatives
business), we will focus on the pari passu senior unsecured funding
of OTC derivatives through senior unsecured debt issuance, while
touching upon some aspects of secured funding. This chapter analyses the funding benefits and costs created by OTC derivatives.
Joint market and credit states
In order to intuitively understand the economics of the funding
cost and benefit, and their relationships with asset CVA and liabil185
19/03/2010 17:11
ity CVA, it is beneficial to examine the joint market and credit states
for two trading parties, A and B, with a single OTC derivatives
trade (or a nettable portfolio) without any collateral agreements.
The market states (M) can be partitioned based on the sign of the
MTM (mark to the market) of the portfolio to one of the parties, say
party A, which can be positive (+), or in the money, or negative (),
or out of the money, to party A. The credit states (C) can be partitioned into four states, ie, none of the parties defaults (ND), party
A defaults first (AD), party B defaults first (BD) and both parties
default (D) at the same time (or within the same time interval).
Furthermore, we use VA(M,C) to denote the MTM value of the
portfolio to party A at a joint market and credit state (M,C), and
RA and RB to denote the default recovery rates for party A and B,
which can also be state dependent. Here we assume the recovery as
a fraction of the MTM value or the mid-market replacement value
VA(M,ND) of the corresponding non-defaultable or Libor-quality
trade. Other recovery assumptions can be handled in a similar
manner (eg, with a state dependent effective recovery).
Table 9.1 The joint market and credit states for two trading parties
A and B
Market states (M) or MTM to A
Credit states (C)
ND
AD
BD
ND
AD
BD
The MTM for the trade (or the nettable portfolio) without any collateral agreements (or the market states) can be positive (+) or negative () to party A.
Table 9.2 Loss given default (LGD)2 (on mid market basis excluding
the bidoffer re-hedging cost)3 from the point view of party A under
the prevailing asymmetric default treatment
ND
AD
BD
(1-RB)VA(+,ND)
(1-RB)VA(+,ND)
(1-RA)VA(-,ND)
(1-RA)VA(-,ND)
The asymmetry comes from the fact that party A has default gain in state (,AD), but no default
loss in state (+,AD). Similarly, party A has default loss in state (+,BD), but no default gain in
state (,BD).
186
19/03/2010 17:11
Figure 9.1 The MTM to party A at various states
ASSET CVA, LIABILITY CVA, FUNDING COST AND FUNDING BENEFIT

Both asset and liability CVA are a measure of the economic value
of default risk. In contrast, funding cost and funding benefit are
a measure of the economic value of funding an OTC derivatives
trade or a portfolio of derivatives. Generally, CVA is included as
part of the fair valuation of a portfolio of OTC derivatives, as required under international accounting standards, whereas funding
cost and funding benefit are usually accrued over time. We discuss
later in this chapter how these measures are related and whether
they should be viewed together or in isolation when considering
the true economics of an OTC derivatives trade or a portfolio of
derivatives.
Default risk
The asset CVA for party A stems from the risk of party Bs default
when the OTC derivatives trade (or a nettable portfolio) is in the
money, ie, party A has a net receivable from party B. Its value is the
same as the default option party A (implicitly) sold to party B, which
is priced by reference to party Bs credit default swap (CDS) spreads
187
19/03/2010 17:11
(if available). Asset CVA sometimes is also termed credit cost.

The liability CVA for party A stems from party As own default
when the OTC derivatives trade (or a nettable portfolio) is out of
the money, ie, party A has a payable to party B. Its value is the same
as the default option party A (implicitly) bought from party B,
which is priced by reference to party As CDS spreads (if available).
Liability CVA sometimes is also termed credit benefit.
The net CVA is the sum of the asset CVA and the liability CVA.
Where both asset CVA and liability CVA are included in a derivatives price, it is referred to as the bilateral CVA. If just the asset CVA
is included, it is referred to as the unilateral CVA. International accounting standards recognise and require the bilateral CVA to be
incorporated into a derivatives valuation to achieve fair market
value.
Funding
Party A incurs funding costs when the OTC derivatives trade (or a
nettable portfolio) is in the money, ie, party A has a net receivable
from party B. It is like party A extending a loan to party B. Since
OTC derivatives receivables typically cannot be financed via repo,4
pricing the funding cost is similar to pricing the asset CVA except
that, instead of using party Bs CDS spreads, it uses the relevant
asset funding spread, ie, the spread of party As senior unsecured
bonds or some blended rate if multiple funding sources are available. Economically, party As funding cost is to compensate party
As senior unsecured bond holders and other funding providers.
It is worthwhile to mention the following points regarding the
funding cost.
o A dealer will generally include some or all of this funding cost
in the bidoffer portion of the final execution price, since it represents the true expected cost of providing the OTC derivatives
trade to the customer.
o The fair value (reflected in party As books and records) of the
OTC derivatives trade is not dependent on party As funding
cost per se. However, if the actual cost of carrying the derivatives
position is significantly more than the discount rate assumed in
the derivatives valuation, a negative carry cost will be experienced by Party A over time. If preferred, this negative carry can
188
19/03/2010 17:11
be largely offset by establishing a funding reserve equal to the

funding cost at inception of the trade and amortising this reserve
over time.
Party A enjoys a funding benefit when the OTC derivatives trade
(or a nettable portfolio) is out of the money, ie, party A has a net
payable to party B. It is like party A receiving a loan from party
B, which can be used to fund other receivables or to retire pari passu
debt. Pricing the funding benefit is similar to pricing the liability
CVA except that, instead of using party As CDS spreads, it uses
the relevant liability funding spread, ie, the spread of party As pari
passu bonds.
It is worthwhile to mention the following points regarding the
funding benefit.
o Except for the potential difference in party As CDS and senior
unsecured bond spreads, the liability CVA and the funding benefit are largely the same in value and in economic origin.
o If the fair value of the OTC derivatives trade already includes
the liability CVA (a measure of default risk), it would generally
be double counting if the value of the funding benefit is also included in the valuation or trade pricing.
o The funding benefit can be viewed as something that accrues
over time, as party A is not in default (and has a net payable to
party B) and largely offsets the decay in value of the liability CVA
(to the first order).
Table 9.3 shows that, when funding is included, both bilateral and
unilateral CVA approaches give broadly similar aggregated result
(to the first order) from an economic perspective. By EPE, we mean
the expected positive exposure (PVd to today), and by ENE, we
mean the expected negative exposure (PVd to today).
EXAMPLES OF CALCULATIONS OF ASSET CVA, LIABILITY CVA,
FUNDING COST AND FUNDING BENEFIT
The following examples are intended to show the pricing considerations from the perspective of party A when attempting to address
the actual economics borne by party A (the dealer) in providing the
OTC derivatives trade to party B (the client). For simplicity, the
189
19/03/2010 17:11
Table 9.3 A summary of the pricing considerations when considering both

default risk and funding costs (to the first order) from party As perspective
Economic measure
Type
Bilateral CVA
Unilateral CVA
Credit cost or asset Default risk

CVA
EPE priced at party EPE priced at party

B credit spread
B credit spread
Credit benefit or li- Default risk

ability CVA
ENE priced at party Zero

A credit spread
Funding cost
Funding/carry cost
EPE priced at the EPE priced at the

relevant party A as- relevant
party
set funding spread A asset funding
spread
Funding benefit
Funding/carry cost
Zero
ENE priced at the

relevant party A
liability funding
spread
examples here do not fully incorporate the bidoffer spread that

a dealer may charge. Also, in this chapter we do not consider the
set-off benefit as well as structured products, such as extinguishers,
which warrant a separate discussion.
As highlighted, the fair value of an OTC derivatives trade from
an accounting perspective should include consideration of the
credit quality of both parties to the trade it does not generally
include any adjustment for the respective funding costs of party A
or party B. However, when considering the actual economics of executing and carrying out an OTC trade, the dealer should consider
the expected funding costs, and this should/may be included in the
final price charged to the client.
o If bilateral CVA pricing is considered the starting point, consideration should be given to the additional asset side funding costs
not included from an economic perspective.
o If unilateral CVA pricing is the starting point, consideration
should be given to both the additional asset side funding cost
and liability side funding benefit not included from an economic
perspective.
190
19/03/2010 17:11
Example 1
Figure 9.2 Party A makes a loan to party B
Party A is lending money to party B. Contrary to some lending

activities of a commercial bank, a derivatives dealer will typically
seek to cover the actual cost of funding the position and be compensated for the default risk of trading with party B. The trade is likely
to bear both a credit and a funding cost.
Table 9.4 Illustrative formulas assuming constant EPE and ENE with T
being the time to maturity of the counterparty portfolio
Bilateral CVA
Unilateral CVA
Credit cost
EPE * party B probabil- EPE * party B probability of default *

ity of default *
(1 party B recovery)
(1 party B recovery)
Credit benefit
Zero
Funding cost
EPE * party A asset EPE * party A asset

funding spread *T
funding spread *T
Funding benefit
Zero
Zero
Zero
Example 2
Party A is borrowing money from party B. A derivatives dealer may
pay some of the credit benefit or some of the funding benefit to
party B, but not both.
191
19/03/2010 17:11
Figure 9.3 Party B makes a loan to party A
Table 9.5 Illustrative formulas assuming constant EPE and ENE with T
being the time to maturity of the counterparty portfolio
Bilateral CVA
Credit cost
Credit benefit
Funding cost
Funding benefit
Zero
Unilateral CVA
Zero
ENE * party A prob- Zero

ability of default * (1
party A recovery)
Zero
Zero
Zero
ENE * party A liability

funding spread
Example 3
Figure 9.4 Party A and party B enter in an interest rate swap with two-way
payments and a symmetrical exposure profile
192
19/03/2010 17:11
Party A is neither net borrowing nor net lending from party B. A derivatives dealer will typically seek to cover the credit cost (default
risk) of trading with party B but will not charge anything for the
funding, assuming the asset funding spread approximately equals
the liability funding spread, ie, the credit benefit offsets the funding
cost in a bilateral CVA approach or the funding benefit offsets the
funding cost in a unilateral CVA approach.
Table 9.6 Illustrative formulas assuming time averaged EPE and ENE
with T being the time until maturity of the counterparty portfolio
Bilateral CVA
Credit cost
Credit benefit
Funding cost
Funding benefit
Unilateral CVA
EPE * party B probabili- EPE * party B probability of default * (1 party ty of default * (1 party
B recovery)
B recovery)
ENE * party A probability of default * (1 Zero
party A recovery)
EPE * party A asset EPE * party A asset
funding spread
funding spread *T
Zero
ENE * party A liability

funding spread *T
Note that, in cases where the derivatives receivables can be financed

via repo and the asset funding spread is less than the liability funding spread, the above trade could represent a net funding gain for
party A.
EFFECT OF NETTING
Netting of the portfolio of OTC derivatives trades with the same
counterparty can significantly reduce the counterparty credit risk,
and thus reduce the (absolute values of) asset and liability CVA.
There are cases whereby the netting is absent, such as where the
netting agreement is not enforceable (depending on the relevant jurisdiction of the domicile of the counterparty), and cases whereby
no netting is desirable due to specific trading strategies, eg, convertible bond tax optimisation. OTC derivatives trades with different legal entities sharing the same parent are normally not netted.
On the other hand, it is interesting to be aware that netting or not
has no effect on the net funding cost in the unilateral CVA case.
193
19/03/2010 17:11
A simple example is where party A and party B have two exactly

offsetting trades, but with no netting. In this case, both parties have
non-zero asset CVA and non-zero liability CVA and, most likely,
non-zero net CVA, but they both have zero net funding cost. In
other words, the receivables are funded by the payables. The same
applies when one intermediates a trade (with no collateral).
However, in the case of bilateral CVA, the funding cost needs to
be computed following the same netting rules of the CVA calculation and the funding benefit is always set to zero, as the funding
benefit is captured in the liability CVA to the first order.
EFFECT OF COLLATERAL
Collateralisation of OTC derivatives trades has significant, but possibly different, impact on both CVA and funding.
For instance, if party A has a net receivable from party B and party B posts collateral to party A, party As asset CVA is reduced (in
absolute value). However, whether or not party As funding cost is
reduced depends on whether or not the collateral can be re-hypothecated5 or re-used. Securities collateral is typically re-hypothecated
in the US market, whereas cash collateral is generally re-used.
If securities collateral is received and can be re-hypothecated,
then party A can post the same collateral to another party to obtain
secured funding at a much lower funding rate (such as the Federal
funds rate) than an unsecured funding rate. Likewise, if cash collateral is received, this cash can be used directly and party A will
pay the cash collateral rate, which, again, is usually from Federal
funds rate. Thus, in either case, receiving collateral that can be used
reduces party As funding cost.
If the collateral cannot be re-hypothecated or re-used, such as
when the collateral is posted to a third party, then the collateral cannot be posted out to obtain secured funding or used directly, and so
does not reduce party As funding cost.
From party Bs perspective, if it has posted collateral, its liability
CVA or funding benefit have also been reduced.
OVERNIGHT INDEX SWAP DISCOUNTING FOR
DERIVATIVES PRICING
OTC derivatives that are fully collateralised, with collateral that
194
19/03/2010 17:11
may be re-hypothecated or re-used, are effectively self-financed

at the funding rate of the collateral. For typical, high-grade OTC
collateral, this rate is usually Federal funds rate for US Treasuries
or US dollar cash, Euro OverNight Index Average (Eonia) for euro
denominated government bonds or euro cash, and Sterling OverNight Index Average (Sonia) for UK gilts or sterling cash.
Naturally, for pricing these derivatives, the payoffs should be
discounted at the funding rate or, for simplicity, the corresponding
overnight index swap (OIS) rate. The OIS rate is relevant because
it provides a link between Libor rates quoted in the market and
the overnight rates, such as the Federal funds rate. Discounting at
the OIS rate helps to capture correctly the economics of OTC derivatives that are fully collateralised and hence funded in a secured
way. Securities firms tend to have a profit and loss (P&L) change
when switching to OIS discounting. This turns out to be one of
most challenging projects for banks and securities firms, as it not
only changes the derivatives valuation and models, it also significantly changes the hedge ratios.
Other challenges include:
o the collateral may not be funded exactly at the OIS rate;
o some of the OTC derivatives may be partially collateralised due
to thresholds;
o some of the OTC derivatives may be collateralised differently depending upon a counterpartys credit rating; and
o often, there are switch or cheapest-to-deliver options embedded
in the collateral agreements.
It is natural for the underlying derivatives model, trading system,
and traders to handle simplified collateral terms (eg, fully collateralised with a predetermined collateral currency) and for the CVA
model, trading system and traders to handle the rest.
TRADING AND MODELLING OF FUNDING COST AND FUNDING BENEFIT
As compared to OIS discounting (which handles the secured funding), in this case, when discussing funding we mainly mean senior
unsecured funding.
The modelling of funding cost and funding benefit is virtually
195
19/03/2010 17:11
the same as the modelling of CVA, except with different model inputs. Therefore, it is natural for the CVA model, trading system and
traders to handle the funding cost and funding benefit.
This is the same with CVA, where the modelling of funding includes the simulations of underlying market factors together with
the counterpartys default, pricing the future exposure properly,
handling netting and collateral, including ratings-based collateral
or termination, mutual break and/or re-couponing. The firms
treasury also needs additional information such as expected future
funding needs and funding needs under various stressed scenarios.
The derivatives funding pricing and modelling is a critical part
of a firms OTC derivatives business, as it is not only required to
properly price the cost of doing business, but it also helps manage
the firms liquidity position by properly incentivising the underlying derivatives trading desks to function in line with the firm. For
instance, an underlying derivatives desk can enter a trade with 30year receivables without collateral (and is paid in day 1 or year 1),
and the firm is left with potentially 30 years of senior unsecured
funding cost. Such funding costs needs to be properly priced to better position the firms long-term funding profile.
STRATEGIES FOR FUNDING COST REDUCTION
Important functions of the derivatives funding trading desk include helping the firm source funding, reducing the funding cost
and reducing the balance sheet consumptions.
Such strategies include portfolio optimisation via trading operations, such as mutual break, trade assignment, re-couponing and
via enhancing the credit terms, such as netting and collateral. In
addition to standard collateral, alternative collateral, such as stocks
and a counterpartys own bonds, can also be used (which requires
proper modelling). The underlying derivatives desks can also be
encouraged to create payables which can then be used to finance
and offset receivable balances. OTC derivatives repo can also be
used if and when available.
The repackaging of OTC derivatives receivables into creditlinked notes (CLNs) or collateralised derivatives obligations can
also possibly reduce the balance sheet consumption. It is difficult to
execute this strategy under current market conditions, however.
196
19/03/2010 17:11
DEBIT VALUATION ADJUSTMENT

It is worthwhile to mention debt valuation adjustment (DVA),
which is related to funding and is under spotlight for the losses
it has incurred. DVA is very similar to liability CVA, but it results
from the election of fair valuation of debt under FAS 159 (or FAS
157) (though it can also be used as a synonym of liability CVA).
The main rationale of FAS 159 (or FAS 157) is to try to resolve the
inconsistency of the handling of asset and liabilities of a financial
firm. The asset is typically PVd or fair valued but the liability is
typically not (and carried on an accrued basis). Under FAS 159 (or
FAS 157), a firm can elect to receive fair value selected debt.
It is an interesting question to ask, although some firms have incurred significant loss in DVA, economically, whether or not the
DVA P&L should affect the shareholders equity. Imagine a situation where a firm is moving toward default and its credit spread
widens significantly; its DVA will show significant P&L gain which,
in turn (and on the surface), will result in a gain of the shareholders
equity. Theoretically, the firm can capture the gain by buying back
its debt, but, practically, it is difficult to execute, as this is the time
the firm needs cash or liquid assets the most. Should the firm go
bankrupt, the gain of the shareholders equity would disappear.
CONCLUSION
Under 2010 conditions, pricing and risk managing OTC derivatives
funding are now more important than ever, as it significantly affects an entitys liquidity and cost of running an OTC derivatives
business. Failure to obtain adequate funding is a leading factor in
driving an entity to default or to file Chapter 11 (bankruptcy protection).
In this chapter, we have presented simple examples for illustrating the methodology and the thought processes behind the pricing
the OTC derivatives funding (consistent with CVA pricing).
The authors acknowledge useful discussions with the following people (in alphabetical order of the last names): Larbi Beldjoudi, Eduardo Canabarro, Patrick Chen, Ryan Comerford, Jay Dweck, Alex
Eydeland, Gerard Frewen, Seb Korbei, David Lamb, Andrew Ross,
Aseem Mehta, Patrick Morris-Suzuki, Stephan OConnor, Vladimir
Rodionov, Michael Sternberg and Adrian Thilo.
197
19/03/2010 17:11
1
2
3
4
5
The views expressed here are the authors own and do not necessarily reflect the views of
the entities that the authors are affiliated to. The authors are solely responsible for the errors.
LGD can be positive, which means a default gain, and can occur when defaulting payables
(which means a default loss to the counterparty).
This is a simplified assumption. In practice, the bidoffer re-hedging cost can be very significant and needs to be handled properly.
There is a small repo market for selected OTC derivatives, which became less active during
the recent market crisis.
If collateral can be re-hypothecated, it means that the receiver of the collateral can post it
out to a third party.
REFERENCES
in L. Tilman, (ed), Asset/Liability Management for Financial Institutions (London: Institutional Investor Books).
Johannes, M. S. and S. M. Sundaresan, 2003, Pricing Collateralized Swaps, available at SSRN, URL: http://ssrn.com/abstract=412342.
Tang, Y. and Bin L., 2008, Quantitative Analysis, Derivatives Modeling, and Trading
Strategies in the Presence of Counterparty Credit Risk for the Fixed-Income Market, (Singapore: World Scientific Publishing).
Zhu, S. H. and M. A. Pykhtin, 2007, Guide to Modeling Counterparty Credit
Risk. GARP Risk Review, July/August, available at SSRN, URL: http://ssrn.com/
abstract=1032522.
198
19/03/2010 17:11
10
Generalised Valuation of
Collateralised Derivatives
Patrick L. Chen, Katsuichiro Uchiyama
and Guanghua Cao
Morgan Stanley
Having a collateral agreement to mitigate counterparty credit

risk arising from over-the-counter (OTC) derivatives transactions
has been a common practice in the markets over the past 10 years.
Based on the agreement, counterparties make margin calls, typically on a daily basis. The recent credit/liquidity crisis exemplifies
the need for the financial industry to pursue further cash or cashequivalent liquid assets as collateral. Within the agreement, the
counterparty posting collateral also earns interest on the collateral
based on a predetermined collateral index rate, which is often the
overnight indexed swap (OIS) rate.
Hence, collateral posting creates additional intermediate cashflows that affect the pricing and risk management of OTC derivative transactions. In other words, if counterparty A receives cash
collateral from counterparty B, it means that A is borrowing the
amount of current exposure from B at the OIS rate, which is typically lower than the Libor rate. Thus, A benefits from the spread
between the OIS and Libor rates and vice versa. This chapter will
determine the benefits or costs by pricing a collar option on the
future present value of the trade as an underlying, thereby allowing us to compute the benefits or costs with general threshold conditions. Given the significant widening of the OISLibor spreads
during the recent crisis, this will lead to potentially large valuation
differences relative to marking-to-market. This chapter provides a
detailed analysis of the pricing of OTC derivatives that are covered
by collateral agreements.
199
19/03/2010 17:11
METHODOLOGY
Assumptions
In this chapter, we examine the valuation of non-dividend-paying
derivative claims with terminal payoff XT at time T > 0 (no callable
or knockout features), collateralised to the mark-to-market (MTM)
or present value. We assume that:
o there is no dispute in MTM value between the bank (bank) and
counterparty (cpty);
o collateral posting is continuous (as proxy for at least daily margin
calls);
o collateral is cash (or cash equivalent) that generates the same
short-term risk-free money market account rate rt ;
o the counterparty posting collateral receives collateral index rate
rtc , which is typically different from the risk-free rate rt ;
When the risk-free rate is different from the collateral index rate,
there is economic benefit (if rt rt c > 0) or cost (if rt rtc < 0) by
holding collaterals.
Valuation of default-free collateralised derivatives
Figure 10.1 Net collateral balance amount as function of derivative value
Assuming unsecured thresholds (q) and minimum transfer amounts

(MTA) conditions apply as shown in Figure 10.1, then, given the future/forward market value Vt at future horizon t, we can express
200
19/03/2010 17:11
Generalised Valuation of Collateralised Derivatives
the net collateral balance amount (Ct) at future horizon t as
Ct = (Vt cpty )(Vt cpty MTAcpty ) (Vt bank )(Vt bank MTAbank )
{
{(V
= (Vt cpty MTAcpty )+ + MTAcpty((Vt cpty MTAcpty )

t
bank
MTAbank ) + MTAbank (Vt bank MTAbank )

+
where Q(.) is the Heaviside step function.
0, x < 0
( x ) =
1, x 0
Intuitively, Ct is equivalent to a combination of long a vanilla call,

short a vanilla put, and long a digital call, short a digital put, at different strike prices.
For simplicity, we further assume MTAcpty = MTAbank. Therefore
Ct = (Vt cpty )+ (Vt bank )+.
(10.1)
This can be explained as a collar long a call option and short a put
option on trade value Vt , at strike price qcpty and qbank , respectively.
In risk-neutral probability measure Q, we have the expression of
the present value of the derivatives as
T

Xtc = EtQ e t
rs ds
T ru du
XTc + e t
ys Cs ds
t
(10.2)
where ys = rs rsc , the differential or spread between the risk-free

rate and the collateral index rate (ie, the OIS rate),1 reflecting the
funding rate benefit or cost of holding the cash collateral earning
the risk-free interest rate on the collateral balance held on hand and
paying the OIS rate to the client.
This expression can be simplified under four different conditions.
Fully collateralised bilateral with zero threshold
If both thresholds are set to zero (qcpty = qbank = 0) then the collateral
balance is always equal to the trade value. The delta of the collar
becomes 1.
201
19/03/2010 17:11
Ct = (Vt )+ (Vt )+ = Vt
From Appendix 1, we have
T

X = E e t
c
rs = rs ys
c
t
Q
t
rsc ds
XTc
where rsc is the collateral index rate or the OIS rate as the discounting rate.
Unilateral, counterparty posts collateral only
If the counterparty posts collateral only, but the bank does not post
collateral
+
( cpty > 0, bank = + ), then Ct = (Vt cpty ) and2
T

Xtc = EtQ e t
rs ds
T ru du
XTc + e t
ys ( X s cpty )+ ds
t
Unilateral, the bank posts collateral only

If the counterparty does not post collateral ( cpty = +, bank > 0 ),
then3
T

Xtc = EtQ e t
rs ds
T ru du
XTc e t
ys ( X s bank )+ ds
t
Fully uncollateralised
If neither the bank nor the counterparty posts collateral
( cpty = +, bank = + ), then obviously the discounting rate remains as the Libor rate
T

X = E e t
c
t
Q
t
rs ds
XTc
General model
In general, we may express the present value of the derivatives
with an effective discounting rate rs as
T

Xtc = EtQ e t
rs ds
XTc
202
19/03/2010 17:11
where rsc rs rs , if the net effect is counterparty posting collateral;

rs > rs otherwise. Here we assume that the terminal payoff XT is net
receivable to the bank.
Alternatively, we can also rewrite 10.2 in a form that is similar to the
expression of credit valuation adjustment (CVA)
(10.3)
where
) (
Pt = Vt + Vt cpty Vt cpty MTAcpty
N t = ( Vt ) ( Vt bank ) ( Vt bank MTAbank )

+
In the above expression, Pt (or Nt ) represents the positive (or negative) exposure after applying the collateral respectively. Therefore,
the second and third adjustment term in 10.3 can be calculated in a
similar way as most firms calculate CVA. Also, note that in general
the benefit or cost of holding collateral depends not only on the
exposure profile of each trade but also on the exposure profile of
the portfolio.
Special cases
If we assume Xt follows a normal Gaussian process and interest
rates rs is constant4
Xt ~ N ( X 0 ert , t ),
Xt = X0 ert + t x
(10.4)
Then, further assuming y to be constant, we can write (see Appendix 2)

203
19/03/2010 17:11
y
X0c E0Q XT e ( r y )T + bank cpty T
2
Where
y=
1
T
2
2
bank + cpty
Dbank
+ Dcpty
ds y 1
1
+
12
2 s
(10.5)
In particular, when both thresholds are zero ( qcpty= qbank=0), we have
X0c = E0Q XT e ( r y )T .
NUMERICAL RESULTS (GAUSSIAN PROCESS)
In this section, we show numerical results assuming the exposure
profile Xt follows a normal Gaussian process. Additionally, we assume the following conditions throughout this section:
rt = 150 bp
rtc = 50 bp
s = US$3,000,000
T = 3.0
This assumption corresponds to the OIS spread of 100bp.

Bilateral case
If the counterparty threshold qcpty equals to the bank threshold qbank,
then the effective discount rate used in the valuation formula differs by y from the risk-free rate r as in Equation 10.5. The funding
rate benefit y generally depends on i) threshold amount; ii) volatility of future present value; and iii) initial mark to market value of X0.
Figure 10.2 shows that the funding rate benefit exponentially
decays towards zero with increasing threshold amount. The gradient depends on the volatility of the exposure, but it also heavily
depends on X0. Figure 10.3 indicates that if the absolute value of
X0 exceeds the threshold amount by a large margin, then the relationship between the funding benefit and the threshold becomes
almost linear. This property is explained as the collar option (i) gets
deep in-the-money in such cases.
In case that the counterparty threshold differs from the bank
threshold, the funding rate benefit also exponentially decays towards zero with increasing threshold amounts. Note that in this
case, however, the second term in Equation 10.5 becomes non-zero.
We will see the effect of the second term in detail in unilateral cases.
204
19/03/2010 17:11
Figure 10.2 Plot of funding rate benefit
y v. threshold amount (X0 = 0)
y v. threshold amount and X0
205
19/03/2010 17:11
y v. threshold amounts (X0 = 0)
Unilateral case
In the unilateral case, the second term in Equation 10.5 represents
the most part of the collateral benefit (or cost). Therefore, here we
define a spread y that also factors in the second term as follows
E0Q XT e (r y)T = X0c
(10.6)
Note that, since the second term is not proportional to the initial
mark-to-market value of X0 , the above defined spread could be
higher than the initial LiborOIS spread when X0 is close to zero.
Unilateral, counterparty posts collateral only
In this case, the bank can benefit from receiving but not posting any
collateral. Figure 10.5 shows a similar shape for the funding rate
benefit as in Figure 10.2; however, it is worth emphasising that the
maximum benefit of this case exceeds the LiborOIS spread of 1%.
Unilateral, the bank posts collateral only
In this case, the bank loses from posting but not receiving any collateral. Figure 10.7 shows that our funding rate exceeds the Libor rate
by about 1% at maximum if we assume X0 equals to US$1 million.
206
19/03/2010 17:11
Figure 10.5 Plot of funding rate benefit y v. counterparty threshold

amount (X0 = 1mm)
Figure 10.6 Plot of funding rate benefit y v. counterparty threshold amount

and X0
207
19/03/2010 17:11
Figure 10.7 Plot of funding rate benefit y v. bank threshold amount

(X0 = 1mm)

and X0
y v. bank threshold amount
208
19/03/2010 17:11
NUMERICAL RESULTS (INTEREST RATE SWAP)

In this section we show numerical results for the interest rate swaps
listed in Table 10.1. The future present values of these trades are
simulated using the multi-factor HeathJarrowMorton (HJM)
model calibrated to the US market as of July 17, 2009. Additionally,
we assume the OIS spread is flat and equal to 30bp.
To assess the funding benefit (or cost), we compute a flat spread
y to define an effective discounting rate rt = rt y to discount back
the swap cashflows (see Equation 10.6).5
Table 10.1 List of sample interest rate swaps
Name
Notional
Maturity
Pay
Receive
US$100mm
10 years
3.7%
US$3m Libor
YS$100mm
5 years
3.0%
US$3m Libor
US$100mm
3 years
1.8%
US$3m Libor
Bilateral case
Figure 10.9 indicates the funding benefit for each swap transactions,
which shows a similar pattern to that seen in Figure 10.2. Since the
volatility of the future present value is proportional to the duration,
the figure shows that the high volatility trade obtains a large funding benefit and vice versa.
To compare these exact results to the Gaussian approximation
methodology described in the previous section, we calibrate the
volatility term structure and discount factors in Equation 10.4 to
the simulated results obtained by the multi-factor HJM model.
Figure 10.10 shows the comparison of the funding benefits computed by the full simulation model and the Gaussian model. While
the approximated result shows a similar shape and pattern as the
full simulated results, the benefit differs by about 1bp to 6bp. Considering that the above comparison is done for a single trade, we
expect this difference will become much smaller if the approximation is applied to the diversified portfolio. Also note that, as expected, the figure shows the high volatility trade suffers more difference and vice versa.
209
19/03/2010 17:11
Figure 10.9 Plot of funding rate benefit v. threshold amount
Figure 10.10 Plot of funding rate benefit computed by the full simulation and
Gaussian model
210
19/03/2010 17:11
Figure 10.11 shows how the funding rate benefit changes along
with the threshold for the counterparty and the bank. As expected,
if the net effect is counterparty posting collateral, then the bank obtains a large funding benefit and vice versa.
Figure 10.11 Plot of funding rate benefit v. counterparty and bank threshold
(Swap A)
Unilateral case
In the unilateral case, the results also show a similar pattern that we
have seen in the Gaussian case (see Figures 10.5 and 10.7). Note that
when the counterparty posts collateral only, the maximum benefit
exceeds the LiborOIS spread of 0.30%.
CONCLUSION
In this chapter, we showed that the spread between the risk-free
money market account rate and the effective discount rate used to
value collateralised derivatives, which reflects the benefit of collateral agreements, can be determined by pricing a collar option on
the future exposure of the trade as an underlying.
211
19/03/2010 17:11
Figure 10.13 Plot of funding rate cost v. bank threshold (bank posts collateral only)
Figure 10.12 Plot of funding rate benefit v. counterparty threshold (counterparty

posts collateral only)
212
19/03/2010 17:11
As a special case, we derived a formula to estimate the spread

for Gaussian-type exposure profile that corresponds to most of
the profile for FX and equity trades that have one time cashflow at
its maturity.
In this case, if we have bilateral Credit Support Annex (CSA)
with a counterparty that has the same credit rating as the bank
does, then the collateral benefit decreases exponentially towards
zero along with the increase in creditworthiness. Note that, however, when the trade is deep in/out-the-money, the relationship
between the collateral benefit and the credit rating becoming almost linear.
If we have unilateral CSA that requires the counterparty to post
collateral, then the collateral benefit has a similar shape. However,
in case the bank needs to post collateral but not receive collateral,
the collateral benefit becomes negative, meaning that the effective
discount rate exceeds the risk-free money market account rate.
We also showed that similar results hold for interest rate swap
transactions using the multi-factor HJM model to simulate the future exposure profile.
Note that the same approach can be applied to analyse the effect of threshold on the discounting rate for a defaultable swap (see
Duffie and Huang 1996, for the un-collateralised case).
APPENDIX 1
Suppose we have a non-dividend paying derivative claim with terminal payoff XT( i ) in currency i at time T > 0, which is fully collateralised to the MTM or present value. We further assume that:
o there is no dispute in MTM value between the bank (bank) and
counterparty (cpty);
o collateral posting is continuous (as proxy for at least daily margin calls);
o collateral is cash in currency j (or cash equivalent) that generates
the same short-term, risk-free money market account rate rt( j )
for currency j6;
o the counterparty posting collateral receives collateral index rate
rtc ( j ), which is typically different from the risk-free rate rt( j ) for
collateral currency j.
213
19/03/2010 17:11
In risk-neutral probability measure Q denominated X in currency i,

the price for the collateralised derivative is
c(i )
t

= E e t
Q
t
rs( i ) ds
c(i )
T
(i)
ru
e t
y X sc (i ) ds
du ( j )
s
where ys( j ) = rs( j ) rsc ( j ), the differential or spread between the riskfree rate and the collateral index rate in currency j, and XTc ( i ) = XT( i ).
We show that the final valuation formula is
T
(rs( i ) ys( j ) ) ds c ( i ) .
Xtc (i ) = EtQ e t
XT
Define

rs( i ) ds
Bt = e 0
, Dt = e 0
(r
(i )
s
ys( j ) ds
(10.7)
, Yt (i ) = Dt1 Xtc (i ) .
To prove Equation 10.7, it is equivalent to show Yt ( i ) is a martingale

in risk-neutral measure, which is
Yt (i ) = EtQ Ys(i ) for every t < s

or equivalently to prove that
EtQ dYt (i ) = 0 .
By definition, we get
t
(i )
rs
Bt1 DtYt (i ) = Bt1 Xtc (i ) = e 0
ds

EtQ e t
rs( i ) ds
(i )
T ru
XTc (i ) + e t
t
y X sc (i ) ds
du ( j )
s
T
rs( i ) ds c ( i )
ru( i ) du ( j ) c ( i )
= EtQ e 0
XT + EtQ e 0
ys X s ds
Note that the first term and the integrand of the second term in the
above equation is a martingale in risk-neutral measure by definition.
Hence
t
ru( i ) du ( j ) c (i )
EtQ d Bt1 DtYt (i ) = EtQ e 0
yt Xt dt
(i )
ru
= e 0
du
y X
( j)
t
c (i )
t
(10.8)
dt = y B X
( j)
t
1
t
c (i )
t
dt
214
19/03/2010 17:11
On the other hand, by definition
EtQ d Bt1 DtYt ( i ) = EtQ yt( j ) Bt1 DtYt (i )dt + Bt1 Dt dYt (i )
= yt( j ) Bt1 Xtc (i )dt + Bt1 Dt EtQ dYt (i )
(10.9)
Combining above Equations 10.8 and 10.9, we get
EtQ dYt (i ) = 0.
We have thus proved that Equation 10.7 holds.
APPENDIX 2
If we assume the interest rate rs is constant and the exposure profile Xt follows a normal Gaussian process in risk-neutral probability
measure Q
Xt ~ N ( X0 ert , t ),
Then
Xt = X0 ert + t x
+
+ )+ =
E Q ( X+ t cpty
( X ert + t x cpty )+ ( x )dx

E ( Xt 0cpty
) = ( X0 ert+ t0 x cpty )+ ( x )dx
x 2
+
x 2 1
+
= rtX0 ert ( X0 ert + t x 1 cpty )2
e 2 dx

= cpty X0 ert (Xcpty
e
+
)
e
dx
2
cpty
0 t
2
t
2
Dcpty
2
=1 t Dcpty 1 e 2 D (
D )
cpty
cpty
= t
e 2 2Dcpty ( Dcpty
)
Q
0
where (..) is the standard normal distribution function, and
Dcpty =
cpty X0 ert
t
Similarly, we can also get

2
1 Dbank
E0Q ( Xt bank )+ = t
e 2 + Dbank ( Dbank ) ,
2
where
Dbank =
bank X0 ert
.
t
Let us simplify further by assuming ys to be constant, and then we

215
19/03/2010 17:11
can write by expanding

T
T ru du
rs ds
X = E e 0 XT + e 0 ys Cs ds
0
c
0
Q
0
= X0 +
2
1 Dcpty
1 D2bank
2
ye s
e
Dcpty ( Dcpty )
e
Dbank ( Dbank ) ds
2
2
rs
= X0 + ds ye
rs
2
Dbank
bank cpty bank + cpty
+ Dc2pty
rs
+
X0 e +
1 2 s 1
2
12
bank cpty
bank cpty
yT
yT = X0 (1 + yT ) +
X0 + X0 +
2
2
y
X0 e yT + bank cpty T
2
y
= E0Q XT e ( r y )T + bank cpty T
2
where
2
2
bank + cpty

Dbank
+ Dcpty
ds
y
.
0
12
2
s

In particular, when both thresholds are zero ( q cpty = q bank = 0 ),
y=
1
T
we have
Therefore
Dcpty = Dbank =
X 0 ert
.
t
X 0c = X 0 + yX 0 e rs ers ds = (1 + yT ) X 0
e X0
yT
= E0Q XT e (rr y )T
The comments and views expressed here are the authors own and do
not intend to represent that of the entities the authors are affiliated to.
The authors are solely responsible for all errors and omissions herein. The authors acknowledge useful discussions with the following
people (in alphabetical order of the last names): Eduardo Canabarro,
Zhengyun Hu, Yi Tang.
216
19/03/2010 17:11
REFERENCES
Duffie, D. and M. Huang, 1996, Swap Rates and Credit Quality, Journal of Finance
51(3), pp. 921949.
1
2
3
4
5
6
The spread ys can be stochastic.

Since it gives only second-order difference, we use the future present value Xs computed
without the additional cashflow to calculate the collateral amount.
Same as Endnote 2.
Note that it is easy to extend the model having a time dependent volatility structure and
discount rates.
Here we chose swaps to be net receivable, meaning that pv01 to the discounting rate is positive, thereby positive funding benefit y~ corresponds to a positive impact to the swap value.
Note that it is easy to extend the valuation formula for the case having multiple collateral
currencies.
217
19/03/2010 17:11
19/03/2010 17:11
Part III
stress testing
19/03/2010 17:11
19/03/2010 17:11
11
Stress Testing and Scenario Analysis:

Some Second Generation Approaches
Greg Hopper
Goldman Sachs
The market turmoil that began in 2007 highlighted the need for alternative methodologies to measure market and credit risks. Given
the sometimes very large and unusual changes we have observed
in financial prices, changes that have sometimes proven difficult for
models to predict in advance, there is a renewed interest in stress
testing and other analytical methodologies as a means of capturing these unusual risks. Traditionally, stress testing has consisted of
running a series of standardised risk factor shocks on market and
credit portfolios. While this sort of stress testing is enormously useful in general for quantifying potential market and credit risks that
may not be well captured by models, it does not go far enough in
mining the full potential of stress testing for risk management. In
situations in which the risks of the portfolio are more complex, traditional stress testing should be supplemented by more advanced
methodologies.
In this chapter, we describe more advanced stress-testing methodologies, algorithmic stress tests and automatic scenario generation, discussing particular examples. Both methodologies are
designed to address limitations of traditional stress tests. An algorithmic stress test is designed to capture the risks of portfolios that
contain particular strategies or particular trades which cannot be
captured well by risk models or standard stress tests. An automatic
scenario generation model is designed to address the problem of
how to define a complete set of stress tests.
221
19/03/2010 17:11
ALGORITHMIC STRESS TESTING

Suppose we want to assess the risk in a hedge funds multi-strategy credit derivatives portfolio. In such a portfolio, we may have a
longshort portfolio of plain vanilla credit default swaps (CDS) on
paired names, a longshort portfolio of CDS on the same names but
at different tenors, a CDS versus bond basis portfolio and a long
short portfolio of index tranches at various attachment points. We
may be the funds risk manager interested in quantifying the risk of
the portfolio, or we may be the risk manager of the funds trading
counterparty who is interested in setting a credit exposure limit or
calculating a margin requirement. How should we measure the risk
of such a portfolio?
One possibility would be to estimate a value-at-risk (VaR) model
using data on CDS, bond spreads and index tranche spreads. While
a VaR methodology is certainly one useful way to characterise the
risk, it does suffer from an important limitation. Unlike a standard trading portfolio, which arises from flow business and does
not necessarily express a particular point of view on the market,
a hedge fund portfolio is designed to take a stand on the direction
of market risk factors. Hedge fund portfolios may reflect a particular strategy or set of strategies, which, if formulated by examining
empirical data, implicitly contain expectations about which events
are rare in the data and which are common. The VaR methodology,
of course, will be likely estimated on the same data that was used
to formulate the strategies. The strategies may be, by their very nature, based on relatively limited data as well. The VaR model will
thus contain the same view on the relative risks in the portfolio as
are contained in the strategies and will not be able to provide an
independent view of the risk. In such a situation, the risk manager
might well be surprised to see an outcome that could not have been
reasonably expected as a result of using the VaR model.
The more complex and strategy specific the risks in the portfolio,
the more important it is to obtain an independent view of the risk.
The VaR model fails to give this independent view since its simulations are not strategy specific, ie, designed to stress the particular strategies expressed in the portfolio. But traditional stress tests
cannot provide an effective independent view either, since they are
similarly not strategy specific. Consider, for example, the risk of the
222
19/03/2010 17:11
Stress Testing and Scenario Analysis
longshort credit portfolio. An analysis of the data would suggest

that there is a substantial common component to the movements
of credit spreads across names. The VaR model would put a lot of
weight on simulating this common movement. The standard stress
test might bump up and down all spreads by predefined amounts,
such as 20%, +20%, 50%, +50%, etc. However, the risk to a long
short portfolio is not the common movement of spreads, but rather
the idiosyncratic movement of particular names. We thus need to
stress this specific strategy to appropriately quantify the risk.
A longshort CDS strategy stress test might be designed to capture
a market dislocation caused, perhaps, by the unwind of a crowded
trading strategy. To perform such a stress test, we would like to have
a methodology that stresses up all spreads in which protection has
been sold and stresses down all spreads in which protection has
been bought. We might also want to base the amount of the stress on
the level of the spread, the industry or geographical location of the
counterparty or some combination of these or other factors.
To implement such a stress test, we would need to write an algorithm that bases the particular stresses on the details of the trades
in the portfolio. An algorithmic stress test is designed to compute
a set of stress tests that are functions of the strategies as well as the
trades in the portfolio. As a consequence, an algorithmic stress test
will, in general, generate stresses that vary with the portfolio, while
a VaR model or traditional stress test will subject each portfolio to
the same stresses.
By their nature, algorithmic stress tests can be quite complicated.
For example, our hypothetical credit derivatives portfolio would
require algorithmic scenarios that stress idiosyncratic movements
of spreads, moves along the term structure of spreads, stresses that
capture the liquidity premium between bond and CDS spreads and
movements of the base correlation curve across a particular index
or between indices. Because the stresses are not predictable a priori
but depend on the portfolio, it will be important to build into the
algorithm mechanisms to correct impossible stresses, such as a base
correlation curve that permits arbitrage (unless, of course, an arbitrageable base correlation curve is one of the stresses). It will also
be necessary to combine the stresses into empirically plausible scenarios. For example, we might want to combine a general spread
223
19/03/2010 17:11
rise with an increase of the bondCDS basis and idiosyncratic dislocations of CDS spreads in particular industries.
The credit derivative portfolio we have discussed is an example
of a set of static strategies. However, it is possible for a portfolio
to incorporate a dynamic strategy as well. Consider, for example,
a hedge fund that invests in G7 currency strategies. The fund is a
fundamental investor that estimates the equilibrium value of the
exchange rate and buys when the currency is undervalued and sells
when it is overvalued. To implement the strategy, the fund borrows
US dollars and invests in sterling, paying the US dollar interest rate
when borrowing and receiving the sterling interest rate on sterling
purchases. When the fund wants to buy more sterling, it borrows
more US dollars and converts them to sterling. When it sells sterling, it converts back to US dollars and pays off its US dollar loans.
As risk manager, we want to get a sense of what the strategy is
as well as what the risks are. The portfolio manager shows us an
example trading record with a positive P&L of US$2.4 million on
sterling, as shown in Table 11.1.
Since the risk looks relatively simple, it would seem reasonable
to implement a VaR model, impose a suite of daily stress tests and
then define VaR and stress-test limits. However, despite the fact
that over the next 24 weeks the VaR and stress-test limits were not
breached, we would be surprised to see the following trading outcome, shown in Table 11.2.
The strategy lost US$14 million despite the VaR and stress-test
limits. How could this happen? In reality, the trading records show
the sale of a put option on sterling with a strike price of US$1.5/
pound and a 24-week maturity that is created by dynamically
hedging the underlying currency. The point of this example is not
to suggest that our hypothetical portfolio manager is really selling
puts. Rather, the point is that traditional investment strategies such
as buy low/sell high can look very much like a dynamic option
hedging strategy. To synthetically create the put, it is necessary to
buy more of the currency as the price drops, and to sell it as the
price rises. In this particular case, as long as the US dollar is not
appreciating, the strategy will synthetically monetise the premium
inherent in selling the option. Even if the portfolio manager is not
literally selling an option, the strategy of buying low and selling
224
19/03/2010 17:11
high could yield profits in a fortuitous exchange rate regime.

Because the strategy is dynamic, the short-term risks will be well
captured by a VaR model or stress tests but the longer-term cumulative risks will be ignored. However, a well-defined algorithmic
stress test could capture this risk. The algorithm would have to generate dynamic paths of exchange and interest rates and then simulate the buy and sell decision. Such a stress test would be difficult
to formulate, however, since the portfolio manager is not literally
selling an option but rather is just behaving as if they are doing so.
The portfolio manager may not be able to articulate exactly how
buy and sell decisions are made in a way that is precise enough to
encode in an algorithm.
Whether static or dynamic, it is crucial for the risk manager to
understand the underlying strategies in order to formulate the appropriate algorithmic stress tests. The greatest challenge in implementing an algorithmic stress test is in identifying the strategies
that may be expressed in the portfolio. Such understanding may
be hampered by the proprietary nature of the strategies or by the
simple fact that a portfolio manager cannot explain the strategy in a
precise way. In such a case, the risk manager may have to resort to
statistical analysis of trading records to understand the strategy.
In our example, we focused on portfolios that expressed static
or dynamic strategies. To measure the risk of these portfolios, the
stress tests needed to be formulated as a function of the particular
trades in the portfolio. But algorithmic stress tests may also be usefully employed in portfolios that do not express a strategy. As an
example, we could generate an algorithmic liquidity stress test. The
algorithm would vary the stress on each trade or set of trades according to estimated liquidity, which could depend on the type of
trade or the number or volume of the trades in the portfolio.
AUTOMATIC SCENARIO GENERATION
Standard stress tests are adequate for most portfolios since they do
not contain complex, strategy-specific risks. Nonetheless, standard
stress tests do suffer from the problem of completeness. Even if we
have a long list of predefined stresses that we run on portfolios,
how can we be sure that we have not missed a relevant stress test
when we formulated our suite?
225
19/03/2010 17:11
Cum cost w/
int
19/03/2010 17:11
US$/
Spot purchased
US$1.600
26,767,909
42,828,654
42,828,654
65,710
-90,351
42,804,013
-42,804,013
US$1.580
4,078,595
6,442,204
49,270,858
75,594
-103,941
49,242,510
-6,438,498
US$1.569
2,196,543
3,445,479
52,716,337
80,880
-111,210
52,686,007
-3,443,497
US$1.578
-2,783,998
-4,393,872
48,322,466
74,139
-101,941
48,294,664
4,391,344
US$1.585
-2,033,408
-3,222,227
45,100,238
69,195
-95,143
45,074,290
3,220,373
US$1.558
6,259,009
9,752,709
54,852,948
84,158
-115,717
54,821,388
-9,747,098
US$1.548
2,536,285
3,925,179
58,778,126
90,180
-123,998
58,744,309
-3,922,920
US$1.563
-4,897,648
-7,656,232
51,121,894
78,434
-107,846
51,092,482
7,651,827
US$1.580
-5,229,157
-8,263,249
42,858,645
65,756
-90,414
42,833,986
8,258,495
US$1.560
4,953,445
7,728,047
50,586,692
77,612
-106,717
50,557,587
-7,723,600
10
US$1.568
-2,961,330
-4,642,832
45,943,860
70,489
-96,923
45,917,426
4,640,161
11
US$1.585
-5,799,471
-9,194,223
36,749,636
56,383
-77,527
36,728,493
9,188,934
12
US$1.574
2,227,396
3,506,114
40,255,750
61,762
-84,923
40,232,589
-3,504,096
13
US$1.540
10,432,917
16,068,567
56,324,317
86,415
-118,821
56,291,912
-16,059,322
14
US$1.540
-693,949
-1,068,674
55,255,644
84,776
-116,567
55,223,853
1,068,059
15
US$1.570
-12,058,790
-18,937,940
36,317,703
55,720
-76,615
36,296,808
18,927,045
16
US$1.532
13,636,192
20,889,088
57,206,791
87,769
-120,683
57,173,878
-20,877,070
Week
Cost
Cum cost
US$ int cost
int gain
P&L
226
Table 11.1 Weekly purchases and sales of pound sterling
Table 11.1 (continued)

17
US$1.520
4,818,378
7,322,407
64,529,198
99,004
-136,130
64,492,072
-7,318,194
18
US$1.494
13,339,793
19,924,020
84,453,218
129,572
-178,162
84,404,629
-19,912,557
19
US$1.489
2,744,844
4,088,333
88,541,551
135,845
-186,786
88,490,610
-4,085,981
20
US$1.475
10,033,876
14,798,901
103,340,452
158,550
-218,006
103,280,996
-14,790,386
21
US$1.460
11,862,651
17,315,969
120,656,421
185,117
-254,535
120,587,002
-17,306,006
22
US$1.403
19,433,407
27,265,750
147,922,171
226,949
-312,055
147,837,065
-27,250,063
23
US$1.439
-622,455
-895,524
147,026,647
225,575
-310,166
146,942,056
895,009
24
US$1.399
1,758,965
2,460,721
149,487,369
229,350
-315,357
149,401,362
-2,459,306
Total
spot
100,000,000
Cum P&L -149,401,362

227
19/03/2010 17:11
US
US
US
US
US
US
US
US
US
US
US
US
US
US
US
US
19/03/2010 17:11
US
228
Table 11.2 Weekly purchases of pound sterling
US
US
US
US
US
US
US
US
229
19/03/2010 17:11
The advantage of a risk simulation model is that it can generate

thousands of scenarios of empirically consistent risk factors. But
the disadvantage of a model is that those scenarios are essentially
repetitions of history, with each scenarios probability of occurrence
consistent with its relative frequency in the data set used to estimate the model. It would be useful if we could generate thousands
of empirically reasonable stress tests, but with frequencies that are
not necessarily consistent with the historical sample.
An automatic scenario generation (ASG) model is a hybrid model that combines empirical calibration with stress testing. An ASG
model is capable of generating thousands of scenarios whose magnitudes are empirically plausible but whose frequencies are defined
by judgement as well as historical data. Developing an ASG model
is as much art as science. Below, we sketch how an ASG model
could be developed for short-term interest rate risks as well as longterm credit risks.
ASG model for interest rate risks
To measure the risk of an interest rate portfolio we can develop an
ASG model based on a principal components analysis of the yield
curve. Table 11.3 reports the first four principal components of the
daily changes in various tenors of the euro swap curve.
The principal components method is a statistical technique that
provides a description of the correlation between daily changes
of interest rates for various tenors on the yield curve. Each principal component explains some of the variance in the data, with the
first principal component explaining the most variance, the second
principal component explaining the next most variance, etc. In general, almost all of the variance of the data can be explained by the
first three or four principal components. We have chosen for this
application to use the first four principal components and discard
the rest.
The principal components have a natural interpretation in terms
of changes in shape of the yield curve. For example, the first principal component gives rise to a roughly parallel shift in the yield
curve. The second principal component describes the steepening
and flattening of the curve, while the third component describes
changes in curvature.
230
19/03/2010 17:11
Table 11.3 First four principal components of the euro swap curve
We can generate empirically reasonable stress moves in the yield

curve by using the principal components. A one standard deviation
one-day move for each tenor in basis points can be constructed by
multiplying the estimated standard deviation by the weight of the
tenor. If we want a more customised stress, we can scale to N days
by multiplying by the square root of N, and we can choose different
numbers of standard deviations to size the stress. We can also scale
the implicit normal distribution into a fatter tailed distribution such
as a t-distribution by multiplying by the appropriate number of
standard deviations.
For example, suppose we wanted to generate a twist in the yield
curve over a five-day period that happens at the 99.9% confidence
level assuming t-5 distribution scaling. Table 11.4 and Figure 11.1
report the results.
To obtain the stress of 2.1 basis points for the three-month point,
we multiply 5.262 (standard deviation) x 3.02% x number of standard
231
19/03/2010 17:11
Table 11.4 Stressed euro curve

Basis point moves
implied by principal
components analysis
Base and stressed euro

swap curve in percentage
points
Figure 11.1 Twist applied to base euro curve
deviations of t-5 distribution x sqrt(5) (to scale to a five-day move).

Although the change in the yield curve looks small, the data suggest that this is a very rare move indeed and could be well suited to
a stress test. To verify that small but unusual interest rate stresses
might be relevant, consider the following hypothetical portfolio of
232
19/03/2010 17:11
euro interest rates swaps that we assume we have credit risk to.
First, we do a six standard deviation upward move stress on the
euro yield curve, based on the principal components model.
Figure 11.2 Six standard deviation upward shift in euro yield curve
Figure 11.3 Six standard deviation steepening combined with flattening

of euro yield curve
The increase in credit exposure is 23.5 million. Next, we do a six

standard deviation steepening and flattening of the yield curve
stress test.
Although the second stress seems benign, it is much more significant, accounting for an exposure increase of US$55 million. But
in designing a suite of stress tests a priori for this simple looking
233
19/03/2010 17:11
Table 11.5 Changes in value of trades in portfolio resulting from stress test
Maturity
Fixed rate
(%)
Notional ()
Base (US$)
Stress (US$)
Receive
15/05/2005
2.55
fixed
1,750,103,216
4,337,757
-8,369,712
24/07/2005
2.65
fixed
2,178,242,302
10,764,706
-18,207,617
09/10/2005
2.75
fixed
539,923,141
2,544,884
-5,148,514
01/08/2006
2.85
fixed
309,596,299
2,321,874
-3,921,841
08/04/2006
2.95
fixed
38,324,956
242,876
-581,744
24/07/2006
3.05
float
554,740,582
-5,054,423
10,313,470
28/09/2006
3.15
float
370,746,956
-3,306,555
7,275,276
15/03/2007
3.3
float
32,652,419
-353,227
798,961
30/07/2007
3.4
fixed
29,371,281
388,393
-825,367
07/10/2007
3.45
fixed
13,444,756
147,238
-415,775
02/01/2008
3.55
fixed
14,148,595
245,781
-422,741
14/05/2008
3.6
float
40,735,463
-417,238
1,533,900
18/09/2008
3.7
float
94,299,879
-1,346,925
3,732,378
10/09/2009
3.875
fixed
536,863,490
7,668,879
-27,261,184
01/04/2010
3.975
fixed
447,129,633
6,246,626
-25,235,264
31/08/2010
4.05
float
238,985,385
-3,930,440
14,134,109
21/02/2011
4.125
float
734,185,844
-12,486,748
46,633,485
11/09/2011
5.2
fixed
230,408,409
21,965,510
1,787,298
24/04/2012
5.37
fixed
626,426,458
66,577,449
10,064,734
30/10/2012
fixed
428,927,165
31,472,118
-8,252,016
24/04/2013
4.35
float
93,509,156
-1,154,141
7,623,010
28/08/2013
4.4
float
529,878,764
-9,170,861
42,925,511
10/06/2014
4.45
fixed
45,897,777
889,061
-3,965,743
10/09/2015
4.55
fixed
167,297,334
2,395,003
-16,448,669
12/06/2016
4.6
float
173,664,690
-3,752,842
17,032,293
10/09/2017
4.7
fixed
602,181,995
10,298,061
-65,157,500
02/07/2018
4.75
float
385,477,064
-8,934,783
41,731,565
234
19/03/2010 17:11

11/09/2022
4.9
float
105,743,048
-1,592,715
13,645,432
03/04/2028
float
58,926,603
-804,362
8,549,505
13/02/2033
5.05
float
221,439,561
-4,295,520
33,606,323
08/02/2037
5.05
fixed
20,451,445
336,082
-3,261,644
28/08/2013
4.32125
float
402,915,171
-3,934,618
35,528,742
18/09/2008
3.63375
fixed
502,754,237
5,503,432
-21,533,881
09/10/2005
2.67
fixed
502,754,237
1,660,118
-5,497,647
27/10/2012
4.15
float
100,550,847
29,286
8,996,736
27/01/2013
4.3625
float
251,377,119
-5,562,264
18,199,899
01/02/2014
4.37
float
100,550,847
-1,138,111
9,093,884
23/02/2013
4.2775
float
502,754,237
-5,849,625
41,460,276
19/02/2012
4.05
float
170,936,441
-401,886
14,380,397
19/02/2029
4.9475
fixed
80,440,678
545,072
-12,418,399
16/03/2014
4.27
float
193,057,627
600,378
20,097,884
06/04/2006
2.3475
fixed
904,957,627
-6,925,255
-26,236,264
06/04/2009
3.20125
fixed
452,478,814
-9,154,313
-35,664,015
04/04/2014
4.0675
float
392,148,305
9,867,210
48,961,522
06/04/2013
3.915
float
1,093,993,22
028,019,878
128,855,221
06/04/2014
4.02875
fixed
1,005,508,475
-29,368,225
-129,377,102
15/04/2006
2.3025
fixed
1,005,508,475
-9,114,491
-30,628,628
20/04/2009
3.24375
fixed
1,078,910,593
-20,502,185
-83,722,781
20/04/2013
3.985
float
653,580,508
13,430,899
73,752,472
06/05/2014
4.2825
fixed
290,591,949
-2,066,834
-31,214,947
09/05/2014
4.2925
fixed
387,321,864
-2,491,132
-41,341,345
06/05/2012
4.07125
float
502,754,237
1,170,372
43,974,420
09/05/2006
2.5975
fixed
427,341,102
-1,355,358
-10,595,760
09/05/2014
4.3
float
100,550,847
572,186
10,661,910
27/05/2009
3.497
float
180,991,525
1,454,935
12,087,577
30/05/2006
2.641
fixed
502,754,237
-1,532,57
-12,448,748
235
19/03/2010 17:11
Table 11.6 Changes in value of trades in portfolio

Maturity
Fixed rate
(%)
Notional ()
Base (US$)
Stress (US$)
Receive
15/05/2005
2.55
fixed
1,750,103,216
4,337,757
8,002,826
24/07/2005
2.65
fixed
2,178,242,302
10,764,706
18,267,692
09/10/2005
2.75
fixed
539,923,141
2,544,884
4,280,311
08/01/2006
2.85
fixed
309,596,299
2,321,874
3,548,925
08/04/2006
2.95
fixed
38,324,956
242,876
371,737
24/07/2006
3.05
float
554,740,582
-5,054,423
-7,110,930
28/09/2006
3.15
float
370,746,956
-3,306,555
-4,482,914
15/03/2007
3.3
float
32,652,419
-353,227
-440,807
30/07/2007
3.4
fixed
29,371,281
388,393
447,372
07/10/2007
3.45
fixed
13,444,756
147,238
158,499
02/01/2008
3.55
fixed
14,148,595
245,781
251,815
14/05/2008
3.6
float
40,735,463
-417,238
-322,384
18/09/2008
3.7
float
94,299,879
-1,346,925
-1,077,076
10/09/2009
3.875
fixed
536,863,490
7,668,879
5,049,979
01/04/2010
3.975
fixed
447,129,633
6,246,626
3,801,262
31/08/2010
4.05
float
238,985,385
-3,930,440
-2,598,888
21/02/2011
4.125
float
734,185,844
-12,486,748
-8,173,852
11/09/2011
5.2
fixed
230,408,409
21,965,510
20,013,915
24/04/2012
5.37
fixed
626,426,458
66,577,449
57,597,367
30/10/2012
fixed
428,927,165
31,472,118
24,557,511
24/04/2013
4.35
float
93,509,156
-1,154,141
410,653
28/08/2013
4.4
float
529,878,764
-9,170,861
-217,874
10/06/2014
4.45
fixed
45,897,777
889,061
77,218
10/09/2015
4.55
fixed
167,297,334
2,395,003
-968,453
12/06/2016
4.6
float
173,664,690
-3,752,842
-169,300
10/09/2017
4.7
fixed
602,181,995
10,298,061
-3,452,514
02/07/2018
4.75
float
385,477,064
-8,934,783
90,084
11/09/2022
4.9
float
105,743,048
-1,592,715
1,657,501
236
19/03/2010 17:11

03/04/2028
float
58,926,603
-804,362
1,388,136
13/02/2033
5.05
float
221,439,561
-4,295,520
4,603,083
08/02/2037
5.05
fixed
20,451,445
336,082
-536,831
28/08/2013
4.32125
float
402,915,171
-3,934,618
2,862,275
18/09/2008
3.63375
fixed
502,754,237
5,503,432
4,062,125
09/10/2005
2.67
fixed
502,754,237
1,660,118
3,274,598
27/10/2012
4.15
float
100,550,847
29,286
1,627,358
27/01/2013
4.3625
float
251,377,119
-5,562,264
-1,605,862
01/02/2014
4.37
float
100,550,847
-1,138,111
613,470
23/02/2013
4.2775
float
502,754,237
-5,849,625
2,194,676
19/02/2012
4.05
float
170,936,441
-401,886
1,659,559
19/02/2029
4.9475
fixed
80,440,678
545,072
-2,461,413
16/03/2014
4.27
float
193,057,627
600,378
4,041,984
06/04/2006
2.3475
fixed
904,957,627
-6,925,255
-3,896,839
06/04/2009
3.20125
fixed
452,478,814
-9,154,313
-11,159,537
04/04/2014
4.0675
float
392,148,305
9,867,210
16,911,027
06/04/2013
3.915
float
1,093,993,220
28,019,878
45,943,009
06/04/2014
4.02875
fixed
1,005,508,475
-29,368,225
-47,436,286
15/04/2006
2.3025
fixed
1,005,508,475
-9,114,491
-5,835,720
20/04/2009
3.24375
fixed
1,078,910,593
-20,502,185
-25,514,664
20/04/2013
3.985
float
653,580,508
13,430,899
24,265,352
06/05/2014
4.2825
fixed
290,591,949
-2,066,834
-7,428,880
09/05/2014
4.2925
fixed
387,321,864
-2,491,132
-9,654,688
06/05/2012
4.07125
float
502,754,237
1,170,372
8,482,447
09/05/2006
2.5975
fixed
427,341,102
-1,355,358
-56,710
09/05/2014
4.3
float
100,550,847
572,186
2,432,224
27/05/2009
3.497
float
180,991,525
1,454,935
2,402,047
30/05/2006
2.641
fixed
502,754,237
-1,532,537
-119,051
75,698,453
130,626,563
237
19/03/2010 17:11
portfolio, it is much more likely that we would have included the

first stress test in the suite as opposed to the second. The advantage of an ASG model is that we can automatically generate thousands of hybrid stress tests, many of which will look like the twist
scenario. We could do that straightforwardly by just generating
many combinations from the principal components. Or we could
put more structure on the model. For example, we could have a
model that starts with a number of driving events, such as market
surprised by CPI report, attach some judgement-based probability
to this event, and then generate hundreds of stress scenarios consistent with that event. In this way, we impose interpretable structure
on the hybrid model.
Imposing an interpretable structure on an ASG model adds another important advantage. Risk managers are much more likely
to pay attention to and act on a stress if it can be shown to be empirically plausible and if it can be rationalised in a story line that
reflects current economic conditions.
DEVELOPMENT OF AN ASG MODEL TO HELP ANALYSE THE RISKS
OF A LOAN PORTFOLIO
To show another example of how an ASG model might be developed, consider the following example. Suppose we have a loan
portfolio consisting of 100 high-yield commitments, all of which
are 0% funded and have a notional value of US$1 million. We assume that 54 of the loan names are present in a high-yield index but
that 46 names are not present in the index. We want to calculate a
static hedge over the life of the portfolio that will best optimise the
riskreturn trade-off. A static hedge is a hedge that is put on once
and is maintained over the period analysed. A risk manager might
be interested in a static hedge because they are sceptical that the
portfolio can be dynamically hedged during times of market stress.
Or they may be interested in how to alter the riskreturn properties
of the portfolio. We assume for the purposes of this example that
the potential hedging instruments are a US$300 million 1015%
mezzanine tranche on the high-yield index, or a CDS on the index
whose notional is scaled so that the running cost of the two hedges
is similar. What are the trade-offs in terms of risk and return to various hedging strategies?
238
19/03/2010 17:11
To understand how the risks of the portfolio are affected by possible hedges, it is necessary to simulate many spread scenarios. However, standard simulation models are not very useful in this context.
Modelling correlation of spreads is problematic in standard models,
but the basis risk between the idiosyncratic and index spreads is
central to the question of how best to hedge with credit indexes.
Moreover, typical econometrically based models do not easily allow
incorporation of credit judgement that we may want to impose in
the stress tests. An ASG model seems an ideal solution: given some
specific credit judgements coupled with empirical analysis, an ASG
model could generate thousands of self-consistent scenarios.
To develop such a model, it is very helpful to start by generating
basic scenarios to get a sense of what to include in the model. The
most obvious scenario to consider is a basis risk stress in which
the index hedge spread moves more or less than the spreads of the
names in the loan portfolio. In Table 11.7, we perform a stress in
which we shock index spreads up 40% and loan portfolio spreads
up 25%, and compare the index tranche hedge to the CDS hedge.
The annual running cost of the index tranche versus the index CDS
is similar. However, as expected, the portfolio hedged with the
tranche shows a smaller loss, since it benefits from the inherent leverage of the tranche.
Table 11.7 40% increase in index spread and 25% increase in loan
spreads in a hypothetical portfolio
CDO price
1,934,640
Annual CDO cost

Net CDO
-900,000
1,034,640
Loan marks
-2,101,202
Loan revenue
Portfolio value
520,000
-1,066,562
Def swap value

Annual cost
-903,099
Net default swap
388,244
Loan marks
-2,101,202
Loan revenue
520,000
Portfolio value
Average idiosyncratic movement

Average index movements
1,291,343
-1,712,958
1.25
1.4
We can reverse the basis risk stress test to get a sense of the sensitivities.
239
19/03/2010 17:11
Table 11.8 40% increase in loan spreads and 25% increase in index
spread in a hypothetical portfolio
CDO price
1,228,806
Annual CDO cost

Net CDO
-900,000
328,806
Loan marks
-2,073,535
Loan revenue
Portfolio value
520,000
-1,744,728
Def swap value
1,213,198
Annual cost
-903,099
Net default swap
310,100
Loan marks
-2,073,535
Loan revenue
520,000
Portfolio value

-1,763,435
1.4
1.25
In this stress test, index name spreads move up 25% while idiosyncratic name spreads move up 40%. The tranche on the index provides comparable protection to default swaps because the tranche
is leveraged. If we put high probability or weight on a spread blowout in the near term, these results suggest that the tranche may be
superior in a wide range of correlation circumstances.
We can also perform a spread-tightening stress test.
Table 11.9 20% decrease in loan spreads and 30% decrease in index
spread in a hypothetical portfolio
CDO price
-1,497,530
Annual CDO cost

Net CDO
-900,000
-2,397,530
Loan marks
-327,722
Loan revenue
Portfolio value
520,000
-2,725,251
Def swap value
-1,051,639
Annual cost
-903,099
Net default swap
-1,954,737
Loan marks
-327,722
Loan revenue
520,000
Portfolio value
-2,282,459
0.8
0.7
Index name spreads are stressed down 30% while idiosyncratic

name spreads are stressed down 20%. In this case, the leverage of
240
19/03/2010 17:11
the tranche works against us, accentuating the loss.

We might also consider time decay as a possible scenario. How
sensitive is the portfolio to ageing, even if spreads do not change?
Table 11.10 Spreads constant but portfolio ages one year
CDO price
-1,017,945
Annual CDO cost
-900,000
Def swap value
-337,259
Annual cost
-903,099
Net CDO
-1,917,945
Net default swap
-1,240,358
Loan marks
-1,125,252
Loan marks
-1,125,252
Loan revenue
Portfolio value
520,000
-3,043,197
Loan revenue
520,000
Portfolio value
-2,365,610
In this stress test, spreads do not change but one year passes. The
tranche hedge loses much more because the effect of time decay is
greater in the tranche than it is in the CDS. Thus, it will be important to age the trades appropriately in scenarios.
Table 11.11 Effect of implied correlation changes
36% implied correlation
CDO price
285,617
25% implied correlation

CDO price
100,640
Annual CDO cost
-900,000
Annual CDO cost
-900,000
Net CDO
-617,383
Net CDO
-799,360
Loan marks
Loan revenue
Portfolio value
-2,101,202
520,000
-2,715,585
Loan marks
Loan revenue
Portfolio value
-2,101,202
520,000
-2,900,562
As Table 11.11 reports, we also do a joint stress test in which index

name spreads are up 40% while idiosyncratic name spreads are up
25%. At the same time, one year has passed. We run the stress for
two values of implied correlation. We can see that changes in im241
19/03/2010 17:11
plied correlation can have significant effects on the portfolio.

The results of this analysis suggest we should include basis risk
and correlation shocks in the set of scenarios and age the trades
appropriately in the dynamic stress tests. To complete the model,
we also need to develop a mechanism to generate the scenarios automatically. Although there are many ways to generate these scenarios, we suggest one method.
Since large changes in spreads are generally produced by changes in business cycles, we could attempt to simulate the state of the
business cycle and then tie the scenarios to the simulated state of
the economy. One way to simulate business cycle states is to use a
regime-switching model, first suggested by Hamilton (1989).
In a regime-switching model, the economy follows a Markov
chain process in which we assume that there are two states: expansion and contraction. Each quarter, if the current state of the economy is expansion, there is a probability p that the economy will be
in a state of expansion next quarter and probability 1p that the
next quarter will be a recession state. On the other hand, if the current state this quarter is recession, there is a probability q that next
quarter the economy will be in recession and probability 1q that
the economy will transition to an expansion.
We can estimate p and q from historical data and thus simulate patterns of expansions and contractions that are consistent with the data.
Or, since this is a stress-testing framework, we might judgementally
set p and q to create a customised business cycle. For example, we
might estimate from data that q = 0.75. In that case, the expected
length of a simulated recession would be 1/(1.75) or four quarters.
But if we wanted to simulate severe recessions, we might set q = 7/8,
which would make recessions on average eight quarters long.
The next step would be to tie the scenarios to the current state of
the economy and the current length of being in a particular state.
For example, if in the simulation the economy has been in an expansion for several years, we might want to simulate low spreads.
Conversely, if the economy has been in recession for over a year, we
want to simulate very high spreads. Or we might decide to tie the
scenarios to the state of the simulated economy in a different way.
There are many legitimate ways to generate the scenarios; the ultimate choices will of course depend on the application and interests
of the risk manager.
242
19/03/2010 17:11
CONCLUSION
In this chapter, we have discussed two more advanced stress-testing methodologies: algorithmic stress tests and automatic scenario
generation. Algorithmic stress tests are very useful when the risks
of a portfolio depend on the trades in the portfolio or the strategy
expressed in the portfolio. Algorithmic stress tests can be usefully
employed whenever the risks in a portfolio are particularly complex. ASG models are hybrid alternatives to risk simulation models
that combine empirical calibration and judgement. ASG models,
since they allow for thousands of scenarios to be automatically
generated, help solve the problem that a relevant stress test might
be ignored when formulating the stress-testing suite. For particular
applications, both methodologies may be very useful supplements
to traditional risk models such as VaR, potential exposure models
and standard stress-testing regimes.
REFERENCES
Hamilton, J. D., 1989, A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle, Econometrica 57(2), pp. 35784.
243
19/03/2010 17:11
19/03/2010 17:11
12
Computing and Stress Testing

Counterparty Credit Risk Capital
Dan Rosen; David Saunders
The Fields Institute for Research in Mathematical Sciences
and R2 Financial Technologies; University of Waterloo
The financial crisis has highlighted the need for financial institutions to better understand the complex portfolio problem underlying counterparty credit risk (CCR) capital computations, and to
develop systematic stress testing approaches for identifying the
embedded model risks. Given the complexity and interconnectedness of the web of financial transactions that constitutes the overthe-counter (OTC) markets, accurately measuring and managing a
portfolios CCR is very challenging. In this context, it is difficult to
capture accurately in a credit portfolio model the stochastic nature
of counterparty exposures driven by market factors, as well as the
dependence structure between exposures, between counterparty
defaults and between defaults and exposures (wrong- or right-way
risk). Thus, in addition to the intricate modelling and computational
complexity of computing CCR capital, challenges remain in estimating these market-credit correlations in a dependable way, as well as
validating them. These correlations can also vary considerably with
the portfolio composition and market conditions. Their impact on
the final capital estimate is significant and difficult to assess.
From a regulatory perspective, the Basel II Accord (BCBS 2006)
allows banks to use an internal ratings-based approach (IRB) to
compute capital requirements for the CCR of derivatives portfolios.
Counterparty capital charges are determined through a risk-weight
formula, which uses four quantitative inputs:
o the probability of default (PD);
245
19/03/2010 17:11
o exposure at default (EAD);

o loss given default (LGD); and
o maturity (M).
The Accord allows the use of internal ratings systems to compute
probabilities of default and losses given default. Furthermore,
banks are allowed to use internal models for EADs and M, based
on the concepts of expected positive exposure (EPE), effective EPE,
effective maturity and the alpha multiplier.
A portfolios alpha is given by the ratio of CCR economic capital
from a joint simulation of market and credit risk factors and the
economic capital when counterparty exposures are deterministic
and equal to EPE. The computation of the numerator thus requires
a credit portfolio model that incorporates stochastic exposures and
their correlations to credit events as well. From a modelling perspective, alpha can be seen as a means to condition internal EPE
estimates on a bad state of the economy, and thus adjusts internal
EPEs for:
o the uncertainty of, and correlation between, counterparty exposures (and LGDs);
o the correlation between counterparty exposures and defaults;
and
o the lack of granularity across the portfolio.
In addition, regulators also view alpha as a means to offset model
and estimation errors. The regulatory base value for alpha is 1.4,
but banks can seek to estimate their own internal alpha, with a
regulatory floor set at 1.2. Supervisors require conservative alpha
estimates, with the ability to capture wrong-way risk and stress test
key parameters such as exposures and correlations.
This chapter presents a practical example of computing and
stress testing CCR capital and alpha in a realistic trading book. We
discuss the methodological and computational challenges faced
when modelling CCR capital, and when implementing a systematic approach for stress testing. Throughout the example we use an
efficient approach for computing alpha that was introduced in Garcia Cespedes et al (2010). The methodology fully leverages existing
counterparty exposure simulations used for risk management and
246
19/03/2010 17:11
Computing and Stress Testing Counterparty Credit Risk Capital
credit limits, and preserves the joint distribution of counterparty

exposures. It further makes the nature of the dependence between
market and credit risk more transparent and amenable for stress
testing, an essential requirement both for internal capital management and regulatory purposes. This allows us to better understand
the nature of CCR in the portfolio and wrong-way risk, by showing
the quantitative impact on several economic capital measures from
various model parameters, including:
o
o
o
o
the magnitude and direction of market-credit correlations;

systematic market factors;
stressed exposures; and
the timing of defaults.
The rest of the chapter is organised as follows. The second section

introduces the basic concepts and definitions on CCR capital and
the Basel II regulation. The third section presents the basic credit
portfolio model and methodology for computing and stress testing
alpha, while the fourth covers the case study, which includes the
computation and stress testing of alpha for a derivatives portfolio
in a realistic trading book. The fifth section provides some concluding remarks. In addition, the two appendices cover some mathematical modelling details and algorithms, as well as an example of
estimating correlations between market and credit factors.
BACKGROUND: COUNTERPARTY CREDIT RISK, BASEL II AND ALPHA
In this section, we review the basic concepts behind CCR and the
treatment of CCR in the Basel II accord (BCBS 2006), with a particular focus on the modelling of counterparty credit exposures.
Counterparty exposures and CCR
We define counterparty exposure as the economic loss incurred on
all outstanding transactions if the counterparty defaults, accounting for netting and collateral but unadjusted by possible recoveries. It is essentially the cost of replacing the set of contracts at the
time of default. Thus, for a single position or a set of transactions
within a netting agreement, Exposure = max (0, mark-to-market).
More generally, exposure can account for the complexity of netting
hierarchies within a given counterparty, as well as the details of
247
19/03/2010 17:11
mitigation techniques such as collateral and margin calls (see De

Prisco and Rosen (2005) and Fleck and Schmidt (2005) for a detailed
treatment of exposure measures).
We are interested in both the current exposures and the potential
future exposures (PFE), ie, the future changes in exposures during
the contracts lives. This is important for derivatives since their values can change substantially over time according to the state of the
market. PFEs take into account the ageing of the portfolio and market factor movements, which affect contracts future values.
PFEs are typically computed using simulation, where the future
is described by discrete sets of times {t0,t1,,tN=T}, and scenarios.
A scenario is a path containing all the market information up to
T. Denote the sth scenario by s and its probability by ps, s = 1,...,S.
For counterparty j, the PFE at time tk in scenario s is denoted by
PFEj(s,tk), j=1,, M. We refer to the matrix of PFE values for every
scenario s and time tk as the counterparty PFE profile. Typically,
counterparty PFE profiles are computed over 1,0005,000 scenarios
and 1260 time steps (depending also on the simulation horizon).
Once this PFE profile is calculated, various statistical measures can
be defined, such as average exposures (over time and scenarios),
peak exposures, etc. In particular, Table 12.1 presents the PFE statistics in which are more relevant for the calculation of CCR capital
and alpha.
Table 12.1 Potential future exposure (PFE) statistics

Name
Definition
Expected exposure
(over all scenarios at tk)

Time-averaged
exposure
(for scenario s up to tk)

Effective expected exposure

Expected positive exposure (EPE)
Effective EPE
248
19/03/2010 17:11
The expected positive exposure (EPE) is the average of the PFE over
time and scenarios. In contrast, the effective EPE first computes the
maximum of the expected exposures at a given date or any prior
date, and then averages this over the entire horizon. As we explain
below, EPE is generally considered as the relevant exposure measure for economic capital, and the Basel II regulation uses a more
conservative EAD measure in terms of effective EPE.
Basel II internal ratings-based (IRB) approach for CCR
Based on the Basel II formula, the minimum capital requirements
associated to a given obligor are given by:
(10.1)
where PDj , EADj and LGDj are the banks internal estimates of
the counterpartys one-year default probability, exposure at default
and the loss given default, respectively; the asset correlation j and
maturity adjustment bj are parameters defined in the accord;1 and
Mj is the maturity. This Basel II minimum capital formula measures
capital at the 99.9% level over a one-year horizon, based on a singlefactor credit model, and on the assumption of an asymptotic finegrained portfolio.
For derivatives portfolios, Basel II originally outlined the same
treatment of EADs as in the Basel I Accord, through the mark-tomarket plus add-on method (BCBS 1995). The final Basel II accord
responded to industry comments and allows for internal models
for EADs and M, based on the internal computation of PFEs. In this
case, the minimum capital requirements are given by Equation 12.1,
with internal estimates of the effective maturity for the jth counterparty portfolio, given by

(12.2)
249
19/03/2010 17:11
and the exposure at default

(12.3)
The definition of a portfolios alpha is

(12.4)
where ECTotal is the economic capital for CCR from a joint simulation of market and credit risk factors, and ECEPE is the economic
capital when counterparty exposures are deterministic and equal
to expected positive exposures.
Since the denominator of alpha in Equation 10.4 assumes that
EADs are deterministic (and given by the EPEs for T=1 year), it can
be computed by standard single-step credit portfolio models using
analytical or simulation methods. However, the calculation of the
numerator further requires a joint simulation of market and credit
risk factors, which incorporates:
o the uncertainty of, and correlation between, counterparty exposures; and
o the correlation between exposures and defaults.
The use of EPE is justified as follows. For very large, infinitely granular portfolios, and where PFEs are independent of each other and
of default events, the portfolios economic capital can be computed
by a model that assumes deterministic exposures given by the EPEs
( =1). Alpha essentially represents the deviation from this ideal
case, driven by the stochastic nature of the exposures and their correlations, the correlations of exposures and credit events and portfolio granularity.
COMPUTING COUNTERPARTY CREDIT RISK CAPITAL AND ALPHA
In this section, we describe an efficient approach for computing
CCR capital and alpha, introduced in Garcia Cespedes et al (2010).
As explained below, the methodology leverages existing counterparty exposure simulations used for risk management and preserves the joint distribution of counterparty exposures. It further
makes the assumptions regarding market-credit correlation easy to
understand and stress test.
250
19/03/2010 17:12
Credit portfolio model

Consistent with the credit risk treatment in the Basel II accord, we
restrict the presentation to a single-factor Gaussian copula model.
Extensions to multi-factor models are discussed in Garcia Cespedes
et al (2010). We consider first the single-step model and then its natural extension to random default times.
Consider a derivatives portfolio with M counterparties indexed
by j=1,..., M. For each counterparty j we define its creditworthiness
index
(12.5)
where Z,, j are independent standard normal random variables

and j is the factor loading giving the sensitivity of CWIj to the systematic factor Z. Counterparty j has default probability PDj and
defaults if CWIj -1(PDj), where is the standard cumulative normal distribution function and -1 is its inverse. We focus on singleperiod losses due to counterparty default over a fixed horizon (eg,
one year). The portfolios credit loss can be written as
(12.6)
where wj is the realised loss at default, wj=EADjLGDj, LGDj is the

loss given default for the jth counterparty, and EADj is its (generally
random) exposure at default. The superscript TS refers to total stochastic losses, the qualifier total implying that both systematic and
idiosyncratic losses are considered, while stochastic refers to the
fact that exposures at default are allowed to be random. We are also
interested in the case when realised losses at default are replaced by
their mean values, which we refer to as total deterministic losses.
(12.7)
where mj is the mean of the random variable wj.

Remaining within a single-step credit portfolio model, we can use
the time-averaged exposures (Table 12.1) as the single-step stochastic exposure equivalent in the loss variable to compute the numerator. This is consistent with the use of EPE in the denominator of
251
19/03/2010 17:12
alpha. The case of a multi-step default portfolio model is discussed

later in this section. Thus, if we denote by EC(L) the economic capital for a loss variable L at the confidence level q, then we can write
alpha in Equation 12.4
(12.8)
In order to understand the influence of name concentration and idiosyncratic risk, it is also useful to consider the analogous systematic credit loss variables, defined as follows

(12.9)
(12.10)
General methodology
Assume that the PFE profiles are pre-computed from a simulation of
market factors affecting the counterparty credit exposures. Denote by
A the counterparty exposure matrix with entries given by the timeaveraged exposure of counterparty j in scenario s,
. The
matrix A represents a non-parametric joint distribution of counterparty exposures.2 We assume that the co-dependence between exposures and default events is modelled through a Gaussian copula3
of the single systematic credit factor Z and a single factor driving
the exposures.
The simulation algorithm proceeds as follows:
1. Construct the matrix A of counterparty EADs. Sort the columns
of the matrix (ie, the exposure scenarios) in order of increasing
total portfolio exposure.
2. Define a standard normal random variable Y and select the exposure scenario corresponding to column s of the sorted matrix if
(12.11)
where
, and the thresholds
252
19/03/2010 17:12
Cs, s = 1, ..., S 1 are calibrated to match the exposure scenario

probabilities. For equally probable MC scenarios, Cs = F-1(s/S), s
=1,...,S 1.
3. Simulate from the joint distribution of the systematic credit factor Z and the exposure factor Y according to a standard bivariate
Gaussian distribution (idiosyncratic credit risk is assumed to
be independent of both Y and Z).
A correlation parameter specifies the joint distribution of the
systematic credit factor Z and the exposure Gaussian factor Y.
4. Compute VaR, EC and any other required risk statistics for the
level of correlation and then compute alpha.
5. Repeat alpha calculations for various levels of market-credit correlation.
Alternative ordering methods
In the above algorithm, the columns of the matrix were sorted by
the value of total portfolio exposure, and then the joint distribution
of the exposure scenarios and the systematic credit factor was defined through a Gaussian copula. Essentially, this implies that with
the parameter r we specify the normal rank correlation between total portfolio exposure and the systematic credit factor. Market-credit risk is correlated in the direction of the total portfolio exposure.
Clearly, total portfolio exposure is not the only way to order the
exposure scenarios, ie, not the only direction of market-credit correlation that may be considered. Indeed, a significant advantage of
the algorithm is its flexibility, in that we can stress test alpha calculations to the direction of market-credit correlation simply by using
different criteria for sorting the columns of the exposure scenario
matrix A. In this section, we examine some alternatives to sorting
the scenarios by total portfolio exposure.
One strategy is to define a vector v of counterparty weights, and
sort the scenarios based on the weighted sum of exposures, that is
by the value of the ordering factor
(12.12)
Once the scenarios (columns of the matrix A) have been sorted by

253
19/03/2010 17:12
the value of c under each scenario, the alpha simulation algorithm

proceeds as above. Some possible choices for weighting vectors v
include the following.
o Total portfolio exposure. This method has already been discussed. It corresponds to the choice
(12.13)
It is conservative in the sense that when Y and Z are strongly

(negatively) correlated, large defaults will tend to occur when
exposures are large. However, it may not be the most risk-sensitive approach, as it ignores default probabilities (instruments
with large exposures might have small default probabilities and
contribute relatively little to overall portfolio risk).
o Expected loss. In this case, the counterparty exposure scenarios
are sorted by the portfolios expected loss under each scenario;
ie, the weight vector is given by the individual counterparty PDs
(12.14)
Here high correlation between Y and Z implies that more defaults

tend to occur when the exposures are large for those counterparties that have relatively large default probabilities (ie, when total
portfolio expected loss conditional on the exposure scenario is
higher). Similar to an expected loss factor, we can define a conditional Basel II capital factor, where instead of using the PDs as
weights, we use the Basel II weights4
(12.15)
o Principal components of the exposure matrix. We may specify

the ordering factor based on the principal components of the matrix A.5 For example, choosing the first principal component (ie,
the one corresponding to the largest eigenvalue of the exposures
covariance matrix), is equivalent to selecting v to be the direction
of greatest variation in the exposures. This maximises the joint
variation in the market and credit components of the model.
To the extent that maximising variance implies maximising risk,
254
19/03/2010 17:12
this assumption can also be seen as a conservative choice for the

ordering methodology. Note that any PC or combination of them
may be potentially used as an ordering factor.
Ordering scenarios using market factors
An alternative to defining an exposure ordering factor based on a
weighted combination of counterparty exposures is to use a weighted combination of market factors (interest rates, foreign exchange
rates, equity prices, market indexes, etc) driving these exposures.
Suppose that there are R market factors, and denote by MFr(s,tk)
the value of market factor r, under scenario s at time tk, r=1,...,R.
In the preceding sections, the time-compressed values of the exposures were denoted by
(12.16)
The method for generating time-compressed values of the market

factors depends on the financial definitions of the factors themselves. For example, for equity factors the relevant value is the total
return, defined by
(12.17)
For interest rates and exchange rates, time averages may be more
appropriate, although returns can also be considered (eg, for a given
interest rate, we can consider the total return that would accrue from
an investment in the corresponding zero coupon bond). When considering a joint simulation of market factors MF and potential future
exposures PFE, we arrive at the augmented simulation matrix
(12.18)
The columns of this matrix (market scenarios) can be ordered by

the value of any of the exposure scenarios, as described above, or
255
19/03/2010 17:12
by the value of a market sorting factor b defined by a weight vector u in a way completely analogous to that for exposure sorting
factors above
(12.19)
In this chapter, we typically consider only a single market factor,

such as a global equity index.
Multi-step default model (random default times)
The basic CCR model introduced so far assumes a single-step default model. Essentially, default occurs at the end of the horizon.
For derivatives portfolios, exposures at default can vary over time
across the capital horizon. We now briefly outline a simple multistep extension of the single-step CCR methodology. For simplicity, we remain in the single-factor Gaussian copula framework, although extensions are simple and well known. The main difficulty
is how to correlate market and credit risk, which is accomplished
by sorting scenarios according to the time-averaged exposure (or
market factor) values.
The basic ingredients for the multi-step credit model are the
same as in the single-step case. Each counterparty has a creditworthiness indicator CWIj. We assume that the marginal distribution of
the default time of each counterparty is known and given by Fj
(12.20)
The default time of counterparty j is then defined to be

(12.21)
Standard results from statistics imply that the default time as defined above has the correct marginal distribution. In practice, the
complete marginal distribution Fj is not observable. At best, it may
be possible to observe default probabilities at a finite set of times
Fj(tk), and some interpolation assumptions must then be employed
in performing the multi-step default simulation. In the simplest
case, only the one-year default probability estimates PDj employed
in the single-step simulation are available. The simplest alternative
256
19/03/2010 17:12
is then to assume that default times are exponentially distributed,

with rates calibrated to match these one-year default probabilities.
Explicitly

(12.22)
In running the simulation, there are a finite number of times at

which counterparty credit exposures are available, given by the potential future exposure profiles PFE(s,tk) and we must select one of
the simulation times tk as default time. With t0 = 0, the current time,
the default times are specified simply as
(12.23)
and no default otherwise.

To this point, the default model is the basic Gaussian copula default model of Li (2001). We employ the same basic technique to
introduce correlation between exposures and systematic credit risk
for the multi-step case as we did in the single-step case. In particular, a market factor Y is defined to have a standard normal distribution such that the correlation between Y and Z is r. Exposure
scenarios are sorted according to one of the many criteria discussed
above. Once Y is simulated, the exposure scenario is determined by
(12.24)
exactly as in the single-step exposure model. The only differences,

then, between the single-step and the multi-step simulation for the
exposures are:
o exposure simulations in the multi-step mode result in an entire
sample path PFEj(s,tk) for the exposure of each counterparty
rather than a single number; and
o the method for ordering multi-step simulations is more difficult
to define than the ordering method for single-step simulations.
As with single-step simulations, there are four possible modes for
loss simulation in the multi-step model. These are distinguished by
whether exposures are deterministic or stochastic, and by whether
257
19/03/2010 17:12
total risk (idiosyncratic and systematic) or only systematic risk is

considered. The loss variables are specified in Table 12.2. With stochastic exposures, the potential future exposure scenarios are generated as described. With deterministic exposures, the exposure at
each time is replaced by its average over all scenarios
(12.25)
With total risk, the specification of the default time j is as above.

With systematic risk, once we have simulated Z, rather than simulating j by simulating j and using the defining equation for j, we
replace the default indicator over each interval by the conditional
probability of the counterparty defaulting over that interval given
the value of the systematic factor Z.
Table 12.2 Multi-step loss variables
Loss type
Random variable
Total stochastic
Total deterministic
Systematic stochastic
Systematic deterministic
Multi-step exposure scenario sets are ordered through their timecompressed values. That is, to determine an ordering of the scenario set corresponding to the potential future exposure profiles
PFE(s,tk), we order the corresponding scenario set in which the
path for a given counterparty (for each scenario) is replaced by the
time-averaged potential future exposure over that path,
(see
Table 12.1). Market-factor scenarios are similarly replaced by their
time averages when used for sorting. Once the time averages have
been computed, the scenarios are sorted using any of the criteria
described above.
258
19/03/2010 17:12
CASE STUDY
In this section, we analyse the counterparty credit risk of a derivatives portfolio representative of a financial institutions trading
book, using the methodology described. We begin with a brief description of the portfolio, including a name concentration analysis
and a discussion of the exposure distribution. This is followed by a
presentation of base case economic capital and counterparty credit
risk numbers, including a discussion of the correlation between
systematic credit risk and instrument exposures. We then proceed
to a series of stress tests, studying the impact of modifying various
base case assumptions. These tests include an analysis of the impact
on counterparty credit risk of:
o
o
o
o
o
the strength of market-credit correlation;

the choice of risk measure;
default timing (single-step v. multi-step);
stressed exposures (stress testing market factors); and
choice of market and exposure factors (the nature of marketcredit correlation).
In particular, we note that we analyse wrong-way risk by stress testing both the magnitude of the market-credit correlations, as well as
the direction, ie, the actual market factor that is correlated to the
underlying credit factor.
Finally, we demonstrate how the counterparty credit risk multiplier alpha can be decomposed to show the effects of different
model assumptions and sources of counterparty credit risk. The
calculation of a portfolios alpha is only a first step in the risk management process. Once alpha has been calculated, it is important
for risk managers to understand where it comes from by decomposing it into its constituent sources. We propose a decomposition,
where a stressed alpha can be expressed as the product of several
multiplicative factors, each representing separate stress test dimensions.
Portfolio description and exposures summary
We analyse a large portfolio of over-the-counter derivatives, including significant positions in interest rate swaps and credit default
swaps, with over 1,100 counterparties. The analysis was performed
at the end of June 2008.
259
19/03/2010 17:12
Counterparty exposures are simulated with 1,000 scenarios over

59 time steps, using a mean reverting model for the market factors
(interest rates, exchange rates, etc). We consider three cases of exposures in the credit portfolio model: a single step credit simulation,
a full multi-step simulation and a simulation with 12 time steps.
Exposures for the single-step credit portfolio model are given by
the time-averaged values from the full multi-step simulation (over
one year). Similarly, in the 12 time step credit simulation, exposures
are bucketed over each interval, and their time-averaged values
over that interval are used. Throughout this section, exposures are
normalised and expressed already adjusted for loss given default.
More than 60 of the over 1,100 counterparties have current exposures that are zero. Figures 12.112.3 present a basic concentration
analysis of the portfolio exposures (single-step). Figure 12.1 presents the effective number of counterparties (defined as the inverse
of the Herfindahl index)6 among the largest N exposures, as N increases. The effective number of counterparties for the entire portfolio is 67. The largest 500 counterparties account for nearly all of
the portfolio exposure.
Figure 12.1 Effective number of counterparties in the portfolio (based on
single-step exposures over one year)
Figure 12.2 and Table 12.3 present, respectively, the histogram of

the exposure distribution and the summary statistics. For quantities measured in US dollars (means, medians, standard deviations,
260
19/03/2010 17:12
percentiles), the figures are expressed as a percentage of the mean

exposure in the base case exposure dataset. Dimensionless quantities (skewness, kurtosis) are given in terms of their values.
Figure 12.2 Histogram of base case exposures
Table 12.3 Summary statistics of base case exposure dataset

Base case exposures
Mean (% of base case mean)
100%
Median (% of base case mean)
10.15%
Std dev (% of base case mean)
399.13%
Skewness
10.85
Kurtosis
151.71
Maximum (% of base case mean)
7,388.80%
10th Largest (% of base case mean)
3,514.38%
The portfolio exposure distribution is highly skewed, with very fat

tails (extreme leptokurtosis). It is also extremely volatile, with the
standard deviation being approximately four times the mean. For
individual counterparty exposures, the average coefficient of variation is 43.6%.
Figure 12.3 plots the range of individual counterparty exposures.
Counterparties are arranged in the order of decreasing mean ex261
19/03/2010 17:12
posure. The 95th and 5th percentiles of the exposure distribution for
each counterparty are expressed as percentages of its mean exposure. Larger counterparties with higher mean exposures tend to
have less volatility (as a percentage of mean exposure) than smaller
counterparties (with lower mean exposures).
Figure 12.3 Percentiles of the distribution of exposures for individual counterparties (counterparties are arranged in order of decreasing mean exposures)
Economic capital: base case

The base case credit capital analysis is performed using the exposures described above. Internal bank credit ratings, PDs and correlations in the single-factor credit model are used. On average, internal credit correlations are slightly lower than Basel II correlations.
As a market (sorting) factor we use the first principal component
of the exposures. We further set a base case market-credit correlation of 25%.
The market factor (first principal component) explains 46% of total exposure variation. Since any single factor explains only a moderate percentage, it is clearly important to stress test the results with
respect to other assumptions regarding the nature of market-credit
correlation, to ensure the appropriate conservatism in the alpha calculations.
The histogram of the distribution of total losses based on a Monte
Carlo simulation with one million scenarios is given in Figure 12.4.
Losses are expressed as a percentage of expected portfolio loss; Ta262
19/03/2010 17:12
ble 12.4 presents the statistics of the portfolio loss distribution. It includes losses for all four loss types (stochastic total and systematic
losses, as well as deterministic total and systematic losses).
Figure 12.4 Histogram of total portfolio losses base case (losses
are expressed as a percentage of the expected loss)
Table 12.4 Statistics of loss distribution base case (statistics expressed as

fractions of the mean total deterministic loss)
Loss type
Mean
Standard
deviation
Value-at-risk
(99.9%)
Economic capital
(99.9%)
Total deterministic
1.00
2.9
32.5
31.5
Systematic
deterministic

1.00
2.1
24.9
23.9
1.03
3.1
36.5
35.5
1.03
2.2
26.8
25.8

Total stochastic

Systematic stochastic
The alpha multiplier for the base case can be calculated using valueat-risk or economic capital (defined as VaR EL) at the 99.9% confidence level (and for both total and systematic losses). In particular,
for total losses (ie, covering both idiosyncratic and systematic risk),
alpha is approximately 1.12 for both value-at-risk (36.5/32.5) and
economic capital (35.5/31.5). Alpha for systematic losses is lower, at
263
19/03/2010 17:12
approximately 1.08. The base case alpha in all cases is thus significantly below the regulatory floor of 1.2.
We will now stress test several assumptions in the base case
model and assess the degree of sensitivity of the alpha multiplier to
variations in these assumptions.
Stress testing CCR capital and alpha
The CCR capital calculations for the base case result in alpha estimates below the Basel II regulatory floor of 1.2. Since these calculations employ a number of significant assumptions, it is important
for both effective risk management and regulatory compliance to
determine the sensitivity of the alpha estimates to variations in
those assumptions. In this section, we perform several stress tests
on the model assumptions:
o wrong-way risk (as reflected by both the strength of the marketcredit correlation as well as the nature of the correlation between
market risk and systematic credit risk);
o the choice of risk measure and confidence level
o default timing
o stressed exposures; and
o credit correlations.
Alpha as a function of market-credit correlations
Recall that in the base case economic capital calculations presented
earlier, exposure scenarios are sorted according to the value of the
first principal component of portfolio exposures, and the correlation between the exposure factor and the systematic credit factor
in a Gaussian copula model was set at 25%. The value of this parameter may be extremely difficult to estimate based on the limited
data available. Given this parameter uncertainty, it is evidently important to stress test alpha estimates to the assumed value of the
market-credit correlation parameter r.
Figure 12.5 presents alpha across all levels of market-credit correlation for different risk measures and loss types. Alphas for both
total and systematic losses are presented, as well as using value-atrisk or economic capital at the 99.9% confidence level.9 The curve
plotting alpha as a function of the market-credit correlation is generally increasing, although it shows a slight smile effect for extreme negative correlations.
264
19/03/2010 17:12
Alpha remains at or below the regulatory default value of 1.4 for

all levels of market-credit correlation. Indeed, it is easy to solve the
inverse problem of finding the maximum market-credit correlation
that produces an alpha below the Basel Accord floor of 1.2. In this
case, alpha remains below the floor for all correlations less than or
equal to 0.5.
Figure 12.5 Alpha as a function of the level of market-credit
correlations
265
19/03/2010 17:12
Choice of risk measure and confidence level

As discussed, the definition of alpha may be constructed in general from various risk measures, such as VaR, economic capital
or expected shortfall. Furthermore, alpha may be considered as a
function of the confidence level for a given measure such as VaR.
Depicting alpha at different confidence levels and for different risk
measures provides a better understanding of the nature of the loss
distributions, and the impact the market-credit correlation has on
the additional counterparty credit risk owing to stochastic (and correlated) instrument exposures.
Figure 12.6 presents alpha for both total and systematic VaR, at
different confidence levels, and across market-credit correlations.
As in the base case, the exposure sorting factor is the value of the
first principal component of portfolio exposures. Expected shortfall (also known as conditional value-at-risk) is also given at the
99.9% confidence level. Generally, alpha is an increasing function of
market-credit correlation (systematic wrong-way risk). However,
for the extreme tail risk measures (VaR and expected shortfall at
the 99.9% confidence level), the alpha curve has a slight u-shape,
bending upwards once market-credit correlation becomes strongly
negative. Both the shape of the alpha curves, as well as the magnitudes of the alphas, remain broadly similar for total and systematic
risk. Note that, as expected, the correlation has a higher impact on
alpha as the measure captures the risk further in the tail.
Alpha for all risk measures remains below the regulatory floor
of 1.2 until market-credit correlation reaches the level 0.5. At extreme values of market-credit correlation (very near 1), VaR at the
99.9% confidence level produces an alpha almost exactly equal to
the regulatory default value of 1.4, while expected shortfall at the
same confidence level produces an alpha that exceeds this value by
a small amount.
Default timing (single step v. multi-step exposure scenarios)
The examples considered so far have assumed a single-step credit
model, with time-averaged exposures from the market-factor simulation. We now consider the effect of a multi-step simulation, as
described in the third section. To understand better the impact of
default timing on CCR capital, we consider two multi-step cases:
266
19/03/2010 17:12
o a full multi-step simulation using the 59 time-steps of the exposure simulation; and
o a 12 time step simulation, where the one-year time horizon is
divided into 12 time buckets, and simulated exposure values are
replaced by their time-averaged values over each time bucket.
Figure 12.6 Alphas for different risk measures
267
19/03/2010 17:12
Figure 12.7 presents the results of the simulations for VaR at the
99.9% confidence level across all values of market-credit correlation, for both systematic and total alpha. We express the ratios for:
o single-step simulation (for both the numerator and denominator);
o multi-step simulation in the numerator and the denominator
(MS/MS); and
o multi-step simulation in the numerator and single-step in the denominator (MS/SS).
The MS/MS ratio isolates the impact of stochastic exposures in a
multi-step setting, while the MS/SS ratio allows us to determine
the total combined impact from stochastic exposures and a multistep setting (when compared to a single-step deterministic setting).
All the alpha curves in Figure 12.7 have very similar shapes. The
curves are generally upward sloping (wrong-way systematic risk),
although alpha may also increase for extreme negative correlations.
The additional volatility associated with multiple time steps generally increases the VaR. For the MS/MS ratios this happens both
in the numerator and denominator, thus resulting in only a slight
difference in alphas compared to the single-step case. However,
the MS/SS ratios are greater than the MS/MS corresponding ones.
There appear to be few discernible differences between the 12-step
and the 59-step simulation results. Thus, any additional conservatism gained through employing the multi-step simulation appears
to be sufficiently captured with the (more efficient) 12-step simulation. In summary, for this example, the multi-step simulations can
be seen to add in the order 3-4% to the alpha multiplier.
Credit correlations
As a simple stress test, we assess the impact on alpha from the
credit correlations. As mentioned earlier, the correlations used in
the single-factor credit model come from a banks representative internal model. In this case, we would like to understand the impact
on alpha from using instead the Basel II correlations. Figure 12.8
compares alpha for the base case and the Basel correlations for all
levels of market-credit correlation. At a correlation level of 25%, alpha increases from 1.08 to 1.12 (however, overall regulatory capital
using the Basel II formula is more than 20% lower).
268
19/03/2010 17:12
Figure 12.7 Single-step and multi-step alphas
While alpha increases only slightly, in this example the internal

model is more conservative and, on average, the Basel correlations
are slightly lower. This results in lower systematic risk and between
15-25% lower capital. The net effect on alpha is much smaller, since
the effect on capital is reflected in both the numerator and denominator (from 1.08 to 1.12).
269
19/03/2010 17:12
Figure 12.8 Effect on alpha of model using Basel credit correlations
Stressed exposures
In its discussion of the guidelines for computing own estimates
of alpha, the Basel Accord states: Where appropriate, volatilities
and correlations of market risk factors used in the joint simulation
of market and credit risk should be conditioned on the credit risk
factor to reflect potential increases in volatility or correlation in an
economic downturn.10 We compare the results of computing alpha
using the base case exposures (simulated using the current best estimates of market risk factor volatilities and correlations), and using
a stressed exposure dataset, where both market-factor volatilities
270
19/03/2010 17:12
and correlations have been stressed.

Figures 12.9 and 12.10 and Table 12.5 present the concentration
analysis (similar to the previous section) for the stressed exposures
dataset, and provide the base case results for comparison. The
stressed exposures have a larger number of effective counterparties
(over 72), as well as a larger mean and thicker tail corresponding to
large exposures.
Figure 12.9 Effective numbers of counterparties (base case and stressed
exposures)
Figure 12.10 Histograms of exposures (base case and stressed exposures)
271
19/03/2010 17:12
Table 12.5 Summary statistics (base case and stressed exposures)

Base case exposures
Stressed exposures
100%
108.81%
Median (% of base case mean)
10.15%
13.55%
Std. dev. (% of base case mean)
399.13%
409.71%
Skewness
10.85
10.35
Kurtosis
151.71
139.78
Maximum (% of base case mean)
7,388.80%
7,407.23%
10th largest (% of base case mean)
3,514.38%
3,592.43%
Mean (% of base case mean)
As in the base case, we use as the market scenario (for sorting the
exposure scenarios) the value of the first principal component of the
exposures. The percentages of total exposure variation explained
by each of the three first principal components for both the base
case and the stressed exposures are presented in Table 12.6. The
results are very similar, with the first principal component explaining approximately 45% of total variation, the first two components
explaining approximately 65% of total variation and the first three
components explaining about 75% of total variation. Table 12.6
shows the base case exposures and the stressed exposures presented in the sorted order (ie, sorted by the value of the first principal
component of exposures in each case). In both cases, we see that
exposures tend to be an increasing function of the value of the first
principal component (we note that, in other examples, a smile behaviour has also been observed, see Garcia Cespedes et al (2010)).
Table 12.6 Explained variations by the first three
principal components of the exposures (base case
and stressed exposures)
PC
Base case %
Stressed exposures %
45.6
44.5
21.70
21.6
11.0
11.1
Figure 12.11 shows alpha for VaR at the 99.9% level across all levels
272
19/03/2010 17:12
of market credit correlation for both total and systematic losses. As

expected, alpha is higher for the stressed exposure dataset across all
correlation levels. Although for low correlation levels (between zero
and 25%) they are very close, alpha for stressed exposures becomes
much higher for extreme positive values of market credit correlation. Base case alphas remain below the regulatory floor of 1.2 for
correlations below 0.5, but the stressed exposure alphas reach this
level at about 40%, with the difference being more pronounced for
systematic risk. The stressed exposure alphas hit the regulatory default value of 1.4 near a market-credit correlation of 0.75, and then
exceed it for high values of market-credit correlation, hitting 1.6 for
total risk and 1.7 for systematic risk in the case of a perfect correlation between the exposure factor Y and the systematic credit factor.
We note in this example that the stressed exposures data set is very
conservative, and in this case, the regulatory floor is still not hit
with market-credit correlations below 40%.
Market and exposure factors
A key advantage of the methodology presented in this chapter is
its flexible modelling of market-credit dependence, which enables
comprehensive stress testing of wrong-way risk, while leveraging
existing exposure simulations. In this section, we examine the impact of assumptions on the direction of market-credit correlation,
by choosing different exposure factors (which are used to sort the
base case exposure scenarios). In particular, we consider the following exposure factors:
o
o
o
o
o
first principal component (this is the base case);

second and third principal components;
average of the first three principal components;
expected loss; and
total portfolio exposure.
Figure 12.12 presents the alpha curves (for VaR at the 99.9% confidence level) corresponding to the different exposures factors. The
most conservative approaches result from using total exposure or
expected loss, and produce very similar results. Using the second
and third principal components generally result in the lowest alpha
multipliers. Somewhat surprisingly, in this case, sorting by the av273
19/03/2010 17:12
erage of the first three principal components produces results that

are almost as conservative as those arising from sorting by total
exposure or expected loss, and are more conservative than sorting
by any of the components individually.
Figure 12.11 Alpha at the 99.9% confidence level (base case and
stressed exposures)
274
19/03/2010 17:12
Figure 12.12 Alpha for 99.9% VaR for different exposure factors
For the conservative sorting methods (sorting by total exposure

and sorting by expected loss), the alpha multiplier is already at
approximately 1.25 for a market-credit correlation of 0.5. This is
above the regulatory floor of 1.2, which is hit by these sorting
methods at a market-credit correlation just below 0.4. The default
alpha value of 1.4 specified in the Basel Accord is hit at a marketcredit correlation of 0.75, and correlations above this level produce
alphas in excess of 1.4.
275
19/03/2010 17:12
Figure 12.13 sheds further light on these results, by showing the

total portfolio exposure at each scenario, when they are sorted by
each of the factors. We observe that the sorting methods that produce the most conservative capital numbers tend to exhibit a significant positive co-dependence with the total portfolio exposure, with
the corresponding curve in Figure 12.13 trending upwards. Sorting
methods that produce less conservative results tend to exhibit more
noise and a weaker relationship between the sorted scenarios and
total portfolio exposure. Similar to Figure 12.13, there are other visual diagnostics that could be employed to obtain more information
about the factors and the sorted scenarios. For example, the value
of any of the factors can be plotted against the scenarios sorted by
another factor (eg, expected loss against scenarios sorted by total
exposure). We can also examine quantitative measures, such as correlations and rank correlations, which provide information about
the relationships between the various sorting factors (see below).
After having examined various methods for correlating market
and credit risk based on the simulated values of the portfolio exposures themselves, we now proceed to consider correlating market
and credit risk with a general market factor (which drives exposures). Although the analysis can be performed for any given market factor, for illustrative purposes we use a global equity index.
Figure 12.13 Total exposure by scenario scenarios sorted according

to different exposure factors
276
19/03/2010 17:12
Figure 12.13 (continued)
277
19/03/2010 17:12
Figure 12.14 compares the base case alphas (using as a factor the
first principal component of portfolio exposures) with those produced by a model based on the market factor.11 The results are generally quite similar. Alphas cross the regulatory floor of 1.2 at about
the level of 50% market-credit correlation.
Figure 12.15 further compares the (single-step) base case alphas
against single-step and multi-step alphas with stressed exposures,
all based on using the global equity index as a market factor. Results are very consistent with those of the earlier section.
278
19/03/2010 17:12
Figure 12.14 Alpha (99.9% VaR) base case and using a global equity index as a market factor
The additional effect of multi-step credit losses is small compared

to the effect of stress exposures, and is reduced with market-credit
correlation level (as the timing of the loss becomes less important
once the market is stressed).
We can further characterise the relationship within the model between the market and exposure factors, as well as their relationship
with the credit factor.
279
19/03/2010 17:12
Figure 12.15 Alpha (99.9% VaR) with scenarios sorted by market

factor single-step base case exposures, and stressed exposures
(single and multi-step)
280
19/03/2010 17:12
Note that both market and exposure factors arise from linear combinations of variables in the (augmented) simulation matrix. Thus,
their codependence is entirely defined within this matrix. In particular, the correlation between any two factors can be computed in
a straightforward way as the sample correlation over the scenarios.
We can generally express it as either a linear correlation or a normal
rank correlation. Linear and normal rank correlations for the various sorting factors are given in Tables 12.7 and 12.8 respectively.
The sorting factors in the table are denoted by market (a market
index), PCi (value of the i-th principal component of the exposure
matrix), PC AVG (average of the first three principal components),
TE (total exposure) and EL (expected loss).
Table 12.7 Correlation matrix of the exposure and the market factors
Market
PC1
PC2
PC3
PC AVG
TE
EL
Market
1.00
-0.52
-0.21
-0.35
-0.64
-0.60
-0.58
PC1
-0.52
1.00
0.00
0.00
0.76
0.84
0.71
PC2
-0.21
0.00
1.00
0.00
0.53
0.31
0.48
PC3
-0.35
0.00
0.00
1.00
0.38
0.28
0.32
PC AVG
-0.64
0.76
0.53
0.38
1.00
0.91
0.92
TE
-0.60
0.84
0.31
0.28
0.91
1.00
0.95
EL
-0.58
0.71
0.48
0.32
0.92
0.95
1.00
Table 12.8 Normal rank correlation matrix of the exposure sorting

factors and the market factor
Market
PC1
PC2
PC3
PC AVG TE
EL
Market
1.00
-0.52
-0.20
-0.33
-0.65
-0.61
-0.59
PC1
-0.52
1.00
-0.01
0.00
0.75
0.84
0.70
PC2
-0.20
-0.01
1.00
-0.01
0.49
0.24
0.43
PC3
-0.33
0.00
-0.01
1.00
0.36
0.27
0.31
PC AVG
-0.65
0.75
0.49
0.36
1.00
0.89
0.90
TE
-0.61
0.84
0.24
0.27
0.89
1.00
0.93
EL
-0.59
0.70
0.43
0.31
0.90
0.93
1.00
281
19/03/2010 17:12
First, note the very strong correlation between TE, EL and PC AVG.
Also, the largest absolute correlation coefficients between the market factor and the exposure factors occur for TE, EL and PC AVG,
the factors with the potential to generate more wrong-way risk produce the most conservative alphas. Conversely, the lowest absolute
correlations occur for the factors that potentially generate the least
wrong-way risk (PC2, PC3).
In addition to the sample market and exposure factor correlations, we can also determine how other market or exposures factors
may relate to a systematic credit factor, given an assumed marketcredit correlation r. It is possible to derive analytical expressions for
these correlations. Table 12.9 presents the implied correlations between the systematic credit factor Z and each of the exposure sorting factors, assuming that exposure scenarios have been ordered
according to the value of the market factor, across different levels of
assumed market-credit correlation. The correlations of the various
exposure factors move in the opposite direction to the market exposure factor, with conservative exposure methods having implied
correlations that are larger and closer in magnitude to the correlation between the market factor and the systematic credit factor.
Empirical market-credit correlations
The base case analysis assumes a 25% correlation between the first
exposure PC and the credit factor. In this case, alpha is 1.08. We
have then performed various stress tests on the level of this correlation, as well as by changing the market factor. Also, the previous
section shows the correlations between the various market factors.
We would also like to understand better what are reasonable empirical levels for these correlations. For example, Fleck and Schmidt
(2005) estimate market-credit correlations based on interest rates
and Moodys default rates for the period 1971-2003, and find an
r-squared correlation coefficient of 0.19.
In this section, we briefly consider the estimation of the strength
of the dependence between market factors and the systematic credit factor, based on observed time series for the market factor returns and credit default data. Before proceeding to the analysis, two
relevant observations should be considered. First, given the short
history and low frequency of available default data, we should be
282
19/03/2010 17:12
Table 12.9 Implied exposure credit correlations

-1.0
-0.75
-0.50
-0.25
0.0
0.25
0.50
0.75
1.0
Market (sample)
-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
PC1
0.53
0.39
0.26
0.13
0.00
-0.13
-0.26
-0.39
-0.53
PC2
0.21
0.16
0.11
0.05
0.00
-0.05
-0.11
-0.16
-0.21
PC3
0.35
0.26
0.18
0.09
0.00
-0.09
-0.18
-0.26
-0.35
PC AVG
0.65
0.48
0.32
0.16
0.00
-0.16
-0.32
-0.48
-0.65
TE
0.61
0.45
0.30
0.15
0.00
-0.15
-0.30
-0.45
-0.61
EL
0.58
0.44
0.29
0.15
0.00
-0.15
-0.29
-0.44
-0.58
283
Market-credit corr
19/03/2010 17:12
aware that such market-credit dependence is very difficult to assess

accurately, and estimates are subject to significant error. Second,
given the non-normal nature of the loss distribution, there may be
some model differences, depending on the market factor chosen for
a particular analysis, for specific portfolios.
Appendix 1 presents a brief empirical study estimating this level
of correlation using default rates from 1981 to 2007 (Standard and
Poors, 2008) and assuming an asymptotic single-factor Gaussian copula model. We obtain estimated market-credit correlations
ranging between 17-37% (with most estimates around 25-30%).
Based on these empirical levels, we see from Figure 12.14 that alpha would be in the 1.10-1.18 range for base case exposures, and
1.18-1.25 for the much more conservative stressed exposures.
Stressed alpha decomposition
A stressed alpha multiplier allows a simple intuitive decomposition
into the different sources of risk. This allows the risk manager to
identify the various sources of model risk.
Alpha can be written as a product, where each term includes one
additional dimension of risk in the counterparty credit risk calculation. This is best illustrated with an example. Consider the alpha
multiplier for 99.9% VaR defined as follows:
o denominator: VaR based on base case, single-step, deterministic
exposures (BC, SS, DE)
o numerator: VaR based on stressed, multi-step, stochastic exposures (SC,MS, SE)12
Then we have
where SE denotes stochastic exposures, DE denotes deterministic

exposures, SS denotes single-step losses, MS denotes multi-step
losses, BC denotes base case exposures, while SC denotes stressed
case exposures.
284
19/03/2010 17:12
The factors in the above decomposition, from right to left, correspond to:
: basic alpha multiplier capturing the incremental capital arising from stochastic exposures (when compared to deterministic exposures given by EPE).
o
: incremental risk from using stressed exposures
rather than base case exposures.
o
: incremental risk from using multi-step losses
rather than single-step losses.
o
Table 12.10 gives the alpha decomposition for the sample portfolio,
for both total and systematic alpha. In this case, stressed exposures
contribute an additional 7% beyond the base case alpha and multistep losses contribute an additional 3% after stressed exposures
have been included.
Table 12.10 Alpha decomposition
Alpha
SC,MS-SS
SC-BC,SS
BC
Total
1.235
1.029
1.072
1.119
Systematic
1.201
1.029
1.076
1.085
Finally, note that the decomposition above is not unique.13 In the

previous decomposition, we first consider the impact of stochastic exposures, then of stressed exposures and finally of multi-step
losses. Instead, we may consider first the impact of stochastic exposures, then the impact of multi-step losses and finally the impact of
stressed exposures
285
19/03/2010 17:12
SUMMARY AND CONCLUSION

The credit crisis has glaringly reminded us of the significant assumptions that CCR and capital calculations require. It is thus
important for both effective risk management and regulatory compliance to determine the sensitivity of the CCR capital and alpha
estimates to variations in those assumptions. In this chapter, we
have presented a practical example of computing and stress testing CCR capital and alpha in a realistic trading book. We use an
efficient approach for computing alpha, which leverages standard
counterparty exposure simulations and makes the dependence between market and credit risk transparent and amenable for stress
testing. This allows us to better understand the nature of CCR in
the portfolio and show the quantitative impact on several economic
capital measures from various model parameters.
We perform stress tests in a systematic way on the model assumptions for the example portfolio. In particular, we show how
one can analyse wrong-way risk by stress testing both the magnitude of the market-credit correlations, as well as the direction, ie,
the actual market factor that is correlated to the underlying credit
factor. We also demonstrate through a simple empirical analysis the
possible levels of market-credit correlations (which may be used
also with different factors). Other stress tests of the model include
analysing the impact on counterparty credit risk of the choice of
risk measure, the default timing (single-step v. multi-step credit
models), the credit correlations and stressed exposures (stress testing market factors).
APPENDIX 1. EMPIRICAL CORRELATIONS BETWEEN MARKET
AND CREDIT FACTORS
This appendix contains a brief empirical analysis of market-credit
correlation, based on available default data. Default rates from 1981
to 2007 are taken from S&P (Standard and Poors, 2008). Assuming
an asymptotic single-factor model Gaussian copula model (using
the notation from the third section of this chapter), with homogeneous default probabilities and correlations, we obtain that the conditional default rate given a realisation Zt of the systematic credit
factor for a given year t becomes
286
19/03/2010 17:12
(12.26)
In practice, we observe the default rate DRt (ie, the percentage of

companies that defaulted in the year under consideration). In order to estimate the correlation between the (observable) market factor and the systematic credit factor Z, we need to invert the above
equation in order to determine the systematic factor value from the
observed default rate
(12.27)
The above equation depends on two unknown parameters: PD and

, which are estimated from the observed default rates using the
method of moments (as mentioned above, with only 26 observations of annual default rates, even if we were to make the unwarranted assumption that the default model is correct, these estimates
would be subject to significant estimation error).
The time series for default rates (both for all companies, as well
as for only investment grade companies), and a set of different market factors (including the MSCI Global Equity index, another global
equity index used as the market-sorting factor in the fourth section (market factor 1), and a similar global equity index constructed
from data over a shorter time horizon (market factor 2)) are given
in Figures 12.16 to 12.19. The time series of the inferred value of
the systematic credit factor was then regressed against each of the
market-factor time series in order to estimate the strength of the
market-credit linear correlation. The results are provided in Table
12.11. We see that the estimated market-credit correlations lie in the
range of 17.1-37.2%, with about half of the estimated correlations
being very close to 30%.
287
19/03/2010 17:12
Figure 12.16 Data for market-credit correlation analysis
Table 12.11 Correlation matrix for factors in the regression analysis

Syst. credit
ALL
Syst. credit
INV
EQ index
Market
factor 1
Market
factor 2
Syst. credit ALL
1.0
0.701
0.235
0.372
0.296
Syst. credit INV
0.701
1.0
0.295
0.171
0.293
EQ index
0.235
0.295
1.0
0.446
0.599
Market factor 1
0.372
0.171
0.446
1.0
0.644
Market factor 2
0.296
0.293
0.599
0.644
1.0
288
19/03/2010 17:12
Figure 12.17 Data for regressions with the equity index
Figure 12.18 Data for regressions with market factor 1
289
19/03/2010 17:12
Figure 12.19 Data for regressions with market factor 2
290
19/03/2010 17:12
The authors gratefully acknowledge the fruitful collaboration and

many helpful discussions with Juan Antonio de Juan Herrero, Juan
Carlos Garcia Cespedes, Valentin Sanchez, Nelson Weisman, Brian
Horgan, Alexis Dalmeida, Philippe Rouanet and Benoit Fleury.
Thanks also for many helpful comments and suggestions to Gary
Dunn, Nathanael Benjamin, the OSFI Risk Analytics Working Group
and participants of the December 2008 ICBI RiskMinds Conference.
The authors gratefully acknowledge financial support from R2 Financial Technologies (DR, DS), The Fields Institute (DR) and NSERC
(DS).
1
For corporate exposures they are given by

,
2
3
4
5
For simplicity, we assume throughout that exposures are already adjusted by LGD, and thus
represent the realised losses should the counterparties default.
While we use a Gaussian copula, the methodology is general and can be used with other
copulas.
Note that they can be interpreted as the exposure-weighted expected default probabilities,
conditional on a bad credit scenario.
The principal components of A are the eigenvalues of the covariance matrix of the exposure
scenarios, S=(A-m)(A-m)T where m is a vector containing the average exposure of each counterparty, ie, mj = EPEj(T).
Exposures are ranked in decreasing order. With the nth largest exposure denoted by wn, the
Herfindahl index of the N largest exposures is defined as
The effective number of counterparties among the N largest exposures is (HN)-1.

The coefficient of variation is defined to be the standard deviation of the counterpartys
exposures over the scenario set divided by the mean of the counterpartys exposures.
8 Note that the loss histogram uses a logarithmic scale.
9 Generally, it can be shown that alpha for economic capital is higher than alpha for VaR. As
in this example, for most credit portfolios the mean is generally quite small compared to the
99.9% percentile, and alpha for VaR and for economic capital are virtually indistinguishable.
10 BCBS (2006), Annex 4, Paragraph 37.
11 The observant reader may notice, when looking at Tables 12.7 and 12.8, the negative correlations between the market factor and the first PC, and be surprised to find that the alpha
curves have similar shapes. For the base case, the figure actually shows the results for scenarios sorted by the negative of the first PC. More generally, all figures depicting sorting methods
place more conservative results on the right (extreme positive correlation). This is intended
to resolve the ambiguity arising from the fact that if v is an eigenvector corresponding to the
exposure matrix A, then so is v.
7
291
19/03/2010 17:12
12 For this example, exposure scenarios are sorted by the first principal component of portfolio exposures, and the assumed correlation between the exposure factor and the systematic
credit factor is 25%.
13 This is similar to the case of risk and P&L attributions.
REFERENCES
Basel Committee on Banking Supervision, 1995, Basel Capital Accord: Treatment
of Potential Exposure for Off-Balance-Sheet Items, URL: www.bis.org.
Basel Committee on Banking Supervision (BCBS), 2006, International Convergence of Capital Measurement and Capital Standards: A Revised Framework (Comprehensive Version), URL: www.bis.org.
De Prisco, B., I. Iscoe, Y. Jiang and H. Mausser, 2009, Compound Scenarios: An
Efficient Framework for Integrated Market-Credit Risk, The Journal of Risk, 11(2).
De Prisco, B. and D. Rosen, 2005, Modelling Stochastic Counterparty Credit Exposures for Derivatives Portfolios, in M. Pykhtin (ed.), Counterparty Credit Risk Modelling, (London: Risk Books).
Fleck, M. and A. Schmidt, 2005, Analysis of Basel II Treatment of Counterparty
Credit Risk, in M. Pykhtin (ed.), Counterparty Credit Risk Modelling, (London: Risk
Books).
Garcia Cespedes, J. C., J. A. de Juan Herrero, D. Rosen and D. Saunders, 2010, Effective Modelling of CCR Capital and Alpha in Basel II, The Journal of Risk Model
Validation, 4(1).
Iscoe I., A. Kreinin and D. Rosen, 1999, An Integrated Market and Credit Risk
Portfolio Model, Algo Research Quarterly, 2 (3), pp. 2138.
Li, D., 2001, On Default Correlation: A Copula Function Approach, Journal of
Fixed Income, 9, pp. 4354.
Rosen D. and D. Saunders, 2009, Analytic Methods for Hedging Systematic Credit
Risk with Linear Factor Portfolios, Journal of Economic Dynamics and Control, 33(1),
pp. 3752.
Standard & Poors, 2008, Default, Transition, and Recovery: 2007 Annual Global
Corporate Default Study and Rating Transitions, URL: www.standardandpoors.com.
292
19/03/2010 17:12
Part IV
economic and
regulatory capital
19/03/2010 17:12
19/03/2010 17:12
13
Back(testing) to the Future: From Market

Risk to Counterparty Credit Risk Models
Eduardo Epperlein; Sean Paul Hrabak;
Wei Zhu, Alan Smillie
Citi; Financial Services Authority; Citi1
The standard procedures for backtesting market risk models have

been well established for more than a decade, and are used by both
risk managers, to confirm a models ability to forecast risk and reliably inform risk management decisions, and by regulators, as a
necessary prerequisite for a model that is used to calculate regulatory capital. In addition to regulatory capital, market risk models
are also often used in the calculation of portfolio initial margin for
prime brokerages serving hedge funds, as well as central counterparties. Given the wide-ranging importance of these models, it has
always been critical to ensure that the backtesting analysis is as informative as possible.
In July 2005, the Bank for International Settlements (2005), introduced for the first time the use of firms internal counterparty
credit risk models for the calculation of capital. It was natural,
therefore, that successful backtesting of a counterparty credit risk
model would be seen as a prerequisite for its use in the calculation
of capital.
However, at that time, there were numerous challenges to backtesting counterparty credit risk models (eg, ISDA, BBA and LIBA
2008). Counterparty credit exposure is typically managed by means
of an exposure profile, calculated at many future time points out
to the longest maturity of any trade2 within the portfolio, as opposed to the single 10-day horizon typically used for market risk.
This difference has fundamental consequences, both from a risk
295
19/03/2010 17:12
management and a capital perspective. For risk management, the

exposure is limited by risk appetite at these future time horizons,
which generally decreases as the time horizon increases, reflecting
the increased uncertainty of the future market environment and the
future creditworthiness of the counterparty. Capital calculations
also depend on the entire exposure profile (see Annex 4 in the Basel
II Accord).3
There are many challenges that must be faced when backtesting
counterparty credit risk models, some of which are outlined below.
o Small sample size. As compared to market risk, it is much harder
to build independent samples upon which to assess in a statistical sense the performance of the model. This is because counterparty exposure is calculated over time horizons that are far
greater than market risk.
In market risk it is usual to backtest the model based upon a
one-day prediction of P&L to a 99% confidence level over one
day, and then compare this to the realised P&L. For each portfolio, each year of backtesting yields approximately 250 independent trials of the model (assuming breaks are independent).
In contrast, when working with counterparty credit risk, the
exposure profiles typically predict exposure for many years and
the estimated risk must be compared with the entire realised exposure path. For example, a one-year FX forward has exposure
up to and including the day of maturity. The realised exposure of
the FX forward would then be compared to the models exposure
profile. One year of historical data, used to compute the realised
exposure, plus any data required to calibrate the model, constitutes one sample in the backtest. To achieve the same number of
samples as compared to market risk, backtesting would naively
require more than 250 years of data.
o Independence of portfolios. Market risk is typically tested across
many market factors simultaneously. For example, we may test
a portfolio containing all the stocks and indexes that a firm is
currently trading. Because of this multiplicity of market factors
per asset class, and therefore a large number of degrees of freedom, it is possible to construct portfolios per asset class which
contain transactions sensitive to subsets of these market factors,
and which can justifiably be treated as independent.
296
19/03/2010 17:12
Back(testing) to the Future: From Market Risk to Counterparty Credit Risk Models
However, because counterparty portfolios span multiple asset

classes but with far fewer market factors per portfolio, we need
to use a framework for counterparty credit risk which works
even for portfolios that may be dependent.
o Generalised risk measures. Counterparty credit risk profiles are
not always based upon a percentile measure, such as value-atrisk (VaR), but rather are often based upon non-percentile measures such as expected positive exposure (EPE) or expected
shortfall (ESF).4 It is therefore necessary to develop techniques
that can also test these more general measures.
We close this introduction with some remarks concerning credit
value adjustments (CVAs). CVAs use much of the same conceptual framework as the calculation of counterparty credit exposure.
However, CVAs are harder to calculate because they involve the
coherent simulation of: (i) the (possibly conditional) probability of
default of each counterparty (possibly relative to the other counterparty) in the bilateral relationship; (ii) the exposure each counterparty has to the other; and (iii) the loss given default (LGD).
Furthermore, since it is a valuation adjustment, which will affect
profits, it must be calculated as accurately as possible so that it can
be efficiently hedged. Traditionally, this level of accuracy has not
been a concern to counterparty credit exposure models as limits
are often set at some high percentile of the exposure, and various
approximations can be derived on the basis that the tails of the exposures may be relatively unaffected by the details of the pricing.
The accuracy and completeness of CVA sensitivities can be verified by performing an analysis comparing: (i) the actual daily P&L;
and (ii) the hypothetical P&L, as derived from the prior day sensitivities and the actual daily market factor movements. However,
this only verifies the local accuracy of the CVA model, and may be
insensitive to shortcomings in its global behaviour. As an alternative, where the CVA is calculated based upon long-term probability
models of the market (risk neutral models), we can verify the accuracy of the CVA by testing the underlying probability distributions
produced by the modes, using the same backtesting techniques as
would be used for the more traditional real-world credit risk models. This is not, however, a direct test of the CVA, since we may be
297
19/03/2010 17:12
willing to accept a calculation based upon a poor probability model

if its output closely reproduces market prices.
It is also important to note that the introduction of CVAs has affected risk management practices. In some firms it is commonplace
to find desks that hedge exposure to certain counterparties. These
desks calculate and hedge the estimated value of the bilateral counterparty credit risk between the firm and its counterparty (CVA).
The result of this hedging is that the tenor of the risk is effectively reduced from the full tenor of the portfolio, to a much shorter
time horizon, determined largely by the potential illiquidity of the
hedges. This time horizon is still longer than the typical market risk
horizon, but is much shorter than that of counterparty credit risk.
However, the CVA calculation itself is dependent upon long-term
simulations of exposure, as indeed are economic risk capital calculations. Therefore, even if risk management practices see radical
change in the future, perhaps turning all counterparty credit risk
into market risk via CVA, the techniques described in this chapter
will still be of use.5 This chapter describes how the standard VaR
backtesting framework, which applies to the entire banks trading
portfolio, can be generalised to multiple portfolios and become a
more powerful diagnostic tool. The framework is then enhanced to
test the performance of counterparty exposure models.
BASIC PRINCIPALS OF BACKTESTING
Market risk and counterparty credit risk measurement require statistical models which use a probability measure to predict a range
of possible future states of observable market prices. A diagrammatic representation of such a model is shown in Figure 13.1.
As these are statistical models, the logic of statistical inference
must be applied to form a judgement about their performance.
Here we consider backtesting as an assessment of the predictive
ability of a model in the estimation of risk measures.
In practice it is these summary risk measures, rather than the
probability distribution itself, that play the major role in risk management and the calculation of regulatory capital. Therefore, it is
more appropriate to directly test the predictive ability of our estimates of certain risk measures, for example, the 99th percentile or
the expected positive exposure. It is possible to construct tests that
298
19/03/2010 17:12
attempt to verify the entire distribution, but a drawback of such an

approach is that the performance of the distribution is only indirectly related to the performance of the risk measure. Such indirect
linkages can reduce the power of the test with respect to the risk
measure itself. It is also possible to perform backtesting on various
percentiles of the distribution (see the exposition Campbell, 2005).
Backtesting is a specific form of statistical test most commonly
applied in counterparty and market risk; however, it is also important to note that other forms of statistical and econometric tests can
show whether or not a model captures certain important features of
the observed data, and should also be used.
Figure 13.1 A diagrammatic representation of a statistical model based on historical calibration for a real-world simulation methodology
For risk neutral, the approach is to use only the initial conditions for calibration, where the calibration
is determined by matching market prices.
HYPOTHESIS TESTING
The standard backtesting methods were designed to work with percentile risk measures, such as value-at-risk (BIS 1996). These methods concentrate on the time series of exceptions, points at which the
P&L exceeds the risk measure, and use various hypothesis tests to
verify that the models predictions are consistent with the empirical data. These tests determine critical thresholds for the number of
exceptions.
299
19/03/2010 17:12
In what follows we shall describe more general backtesting

methods. These generalised tests do not result in simple threshold
values for the number of breaks, but are based upon a comparison
between the complementary cumulative probability distribution
function (CCDF) of the breaks as predicted by the model with the
observed frequency of breaks for a set of portfolios over a period of
time. This comparison will be made using the classical hypothesistesting framework and so, for pedagogical reasons, we now describe the key structure of the framework. (For a more exhaustive
account the reader is referred to Berry and Lindgren 1990.)
The basic procedure for hypothesis testing is:
o Identify the null hypothesis H. We describe below the null hypothesis by the statement that the CCDF of the model is equal to
the empirical CCDF. Here the model CCDF is derived from the
model itself (either analytically or numerically) and the empirical CCDF is determined by the observation of the test results.
o Identify ~H, the alternative to H. This is an important task in that
it is the alternative hypothesis, which enables us to measure the
power of the hypothesis test. Unfortunately, it is also quite a difficult (but not impossible) task in that an alternative model must
be proposed.
o Specify or construct a test statistic K, one that discriminates between H and ~H. There are many test statistics that can be chosen, and they must be evaluated based upon their utility in terms
of assisting key criteria of the model. For example, one statistic
might measure how close to perfection your model is (such as
the KolmogorovSmirnov statistic), while another might simply
determine if your model is conservative.
o Specify values of K that are extreme under H and in the direction ~H. These are the critical values mentioned above, and are
such that they would indicate that ~H provides a better explanation of the data than H.
o Obtain data and calculate the corresponding P-value. The P-value is the probability, under the assumption that the hypothesis
H is true, of seeing a value of K equal to, or more extreme than,
the observed value. The smaller the P-value, the stronger the evidence against H.
300
19/03/2010 17:12
To assess the results there are generally accepted interpretation of

P-values in the statistical literature, see Berry and Lindgren, 1990.
o P<0.01: Strong evidence against H highly statistically significant
o 0.01<=P<0.05: Moderate evidence against H statistically significant
o 0.05<=P<0.1: Little or no evidence against H
o 0.1<=P: No evidence against H
The 1996 Basel accord makes use of the same threshold with the
exception of P<0.01, which is lowered to P<0.001 (see the next section below).
The P-value is sometimes called the observed level of significance and, as such, irrespective of the above classification, gives a
useful indicator of the performance of the model. So, for example, if
your aim was to improve your model, rather than simply test if you
can reject your model given the data statistically, then the P-value
becomes the model discriminant, with higher P-values indicating
an improved model.
MAKING A DECISION, STATISTICAL POWER AND SAMPLE SIZE
Type I and type II errors
Given the results of a statistical analysis of some data, we would
like to be able to drive a decision process. For example, if the results
of the hypothesis testing of our statistical model indicate a poor
result for our model, then we would like to invest effort in improving the model. However, in order for this decision to be sound, it is
necessary to verify that we have sufficiently large sample to make
the inference reliable.
The method for making the assessment as to the reliability of the
result of the hypothesis test is to calculate the statistical power of
the test.
To be more specific, we may have the following types of error:
o type I error the test rejects a true null hypothesis; and
o type II error the test does not reject a false null hypothesis.
The most reliable test is the one which minimises the type I and
type II errors (Berry and Lindgren 1990).
301
19/03/2010 17:12
The likelihood of making a type I error is controlled by specifying

the level of P at which we are willing to reject the hypothesis the
empirical data matches the model. If, as described in the previous section, we choose to reject the model when P < 0.05, then the
likelihood of making a type I error (incorrectly rejecting a true
model) is 5%.
Figure 13.2 This diagram illustrates type I and type II errors
The distribution on the left corresponds to the null distribution and the distribution on the right corresponds to the alternative distribution.
The likelihood of making a type II error is determined by the power

of the test to reject the model when it does not match the empirical
data. The power of a test is defined to be the likelihood that a false
null hypothesis is rejected, ie, Power = 1 , where = Prob (type
II error). The power of a statistical test is determined by the size of
the region in which the null hypothesis is not rejected. If the nonrejection region is large (the middle left region in Figure 13.2), we
expect that the test will lack the ability to distinguish between true
and false hypotheses.
When applied to backtesting, the size of the non-rejection region
is affected by the width in the test statistic if the null distribution
is very wide, it is relatively difficult to distinguish the hypothesis
from the alternative hypothesis, and the power of the test will be
low. The width of the null distribution is, in turn, determined by
302
19/03/2010 17:12
the effective number of market factors/portfolios being tested and

the degree of correlation between them using a small number of
independent factors/portfolios may be as effective at shrinking the
non-rejection region as using many highly correlated factors.
There are no definitive values in the literature as to the acceptable thresholds for the probability of type I and type II errors, and
they are regarded as fundamentally arbitrary.
BACKTESTING A SINGLE PORTFOLIO VAR THE 1996 BASEL ACCORD
The 1996 Market Risk Amendment to the Basel Accord introduced
a simple framework for backtesting the corporate level VaR of an
institution (BIS 1996). The VaR represents the maximum6 loss a
portfolio can generate at a specific confidence level, over a certain
time horizon. For regulatory capital, the confidence level is 99%
and the time horizon is 10 days, although it is common to calculate
a one-day VaR, which is then scaled up to a 10-day VaR.
The test is designed to be applied to the one-day VaR of a single,
high-level portfolio. A backtest is conducted by comparing the ex-ante
VaR, as calculated at the close of business day t to the ex-post P&L, as
calculated between the close of business t and t + 1. As a test statistic,
the method uses the number of VaR exceptions that occur over a 250day (one year) period. This is illustrated in Figure 13.3. If the model
is correct, VaR exceptions should occur independently, and therefore
the number of breaks should follow a binomial distribution. For a
perfect one-day VaR model at a 99% confidence level, we expect the
realised loss to exceed the VaR only on one out of 100 days.
We can test our hypothesis - that the VaR breaks are distributed
binomially with a given chance of a break - by calculating the probability, P, of the observed break count under the assumption that
the hypothesis is true. If the probability that a correct VaR model would produce at least the observed number of breaks is low,
then we can reject the model. The required confidence level of the
test determines a threshold number of breaks. If we observe more
breaks than this threshold level, we reject the hypothesis and assume the model is incorrect.
The Basel Committee rules divide the number of VaR exceptions
during the year into three zones:
o green zone: 14. This indicates a sufficiently accurate (or, more
303
19/03/2010 17:12
precisely, conservative) model, so no action is required;

o yellow zone: 59. The banks are required to increase their capital
multiplier to compensate possible model understatement; and
o red zone: 10 or more. The banks are required to use a fixed multiplier of 4.
Figure 13.3 A sample market risk backtesting that shows four VaR exceptions during the year
Relating back to the thresholds for critical P-values in the section

above:
o 0.05<=P<0.1 corresponds to four exceptions and defines the end
of the green zone;
o 0.001<= P <0.05 to define the beginning of the yellow zone; and
o P < 0.001 at which point the red zone begins.
As a consequence of the choice of using a standard of 250 days of
data, only one exception separates the 0.05 < = P < 0.1 region and
the 0.01 < = P < 0.05 region. This makes the introduction of zone
separating these confidence levels of little value. It is not known by
the authors why the red zone is set at such a high confidence level,
viz 0.001 rather than 0.01, which is more commonly used in the
statistics literature.
If the market risk model assumes a normal distribution for the
profit and loss of a portfolio, then it is possible to derive a scaling
factor that can be applied to the VaR to compensate for an unreliable model. The confidence level for a model that produces VaR
304
19/03/2010 17:12
breaks can be inferred using the maximum likelihood estimator for

a binomially distributed random variable. This gives an effective
confidence level of 1 - x/250. The scaling factor required to bring
the confidence level back to 99% is
where
is the inverse of the standard cumulative normal distribution function. This is the basis of the increasing multiplier in the
Basel Framework. Table 13.1 shows that the increase in multiplier
from the Basel Accord is in agreement with the above equation, for
models in the yellow zone. Note that, as we shall see, the probability distributions for counterparty credit risk are typically very
specific to each backtest, therefore it is hard to generalise the multiplication formalism from market risk to counterparty risk.
In the red zone, it is likely that further increases of the multiplier
will not correct the model as expected. Therefore, the Basel standards require that the supervisor should also begin investigating
the reasons why the banks model produced such a large number
of misses, and should require the bank to begin work on improving
its model immediately.
Along with the above three-zone approach, the Basel Committee
does acknowledge that the results in the yellow zone are plausible
for both accurate and inaccurate models. For example, there is approximately a 10.8% chance of an accurate model producing five or
more exceptions. However, the Basel Committee further states, the
burden of proof in these situations should not be on the supervisor
to prove that a problem exists, but rather should be on the bank to
prove that their model is fundamentally sound.
APPROACH FOR MANY PORTFOLIOS WITHIN MARKET RISK
BACKTESTING
One of the weaknesses of the Basel backtesting framework is that
it only applies to the single consolidated total VaR (the firms entire portfolio), and does not apply in the context of multiple portfolios. It is usual in a large trading institution to conduct the Basel
backtesting across hundreds of fairly independent portfolios and
across different product classes over any given time interval. This
305
19/03/2010 17:12
type of hypothetical portfolio backtesting aids in the resolution of

poor model performance. If we blindly apply Table 13.1 to these 100
independent tests we would find (on average) that approximately
11% of the 100 portfolios will fall outside the green zone, which
might suggest larger capital multipliers, even for a perfect VaR
model. Conversely, it is also possible (though less likely) for all 100
portfolios to be in the green zone, which implies that VaR model is
good when, in fact, it may be overly conservative.

Table 13.1 Increase in multiplier in the yellow and red zone

Zone
Green
Yellow
Red
Breaks
Effective
confidence
level
Basel Accord
Calculated
(x)
Multiplier
Ratio
99.0%
3.00
1.00
1.00
98.0%
3.40
1.13
1.13
97.6%
3.50
1.17
1.18
97.2%
3.65
1.22
1.22
96.8%
3.75
1.25
1.26
96.4%
3.85
1.28
1.29
10
96.0%
4.00
1.33
1.33
There is one more factor to consider. Irrespective of where the excessions per portfolio lie in the Basel zones, looking at each portfolio and applying the Basel testing framework does not allow bias
to be detected in the backtesting results that could indicate that the
model is not a good out-of-sample predictor. For example, it is possible that the number of exceptions for each portfolio is within the
green zone, but this does not mean that there is no significant evidence, when you look across portfolios, to reject the model. Figure
13.4 shows such a hypothetical example.
This latter example is very important, for if a firm relies on the
1996 framework to prove the accuracy of the use of a market risk
model for counterparty credit risk (such as prime broker portfolio
margining or CCPs), they may be unable to detect serious flaws.
The answer to this apparent inconsistency was alluded to above,
namely that the simple test should be replaced, based on a critical
306
19/03/2010 17:12
threshold for the total number of exceptions, with a test that compares the complementary cumulative distribution function (CCDF)
corresponding to the empirical and theoretical frequency distributions for the number of breaks across portfolios. The choice of the
CCDF, rather than just the cumulative distribution function (CDF),
is based on a desire to make the analysis as intuitive as possible.
Figure 13.4Illustration of the weakness of the Basel backtesting framework
In this representation, the number of exceptions x is the abscissa

and the probability of having x or more exceptions is the ordinate.
Therefore if we consider two models whose CCDFs do not intersect, the one that is more conservative, ie, has a smaller probability
of x or more exceptions, has a CCDF that always lies below the
other. So, in using a CCDF, above is less conservative and below is
more conservative. Typical methods for performing this comparison include the KolmogorovSmirnov or CramrVon Mises tests.
In what follows we assume that the portfolios that we are backtesting are independent. When we turn to counterparty credit risk
backtesting, we will show how the more general framework there
incorporates the dependence between portfolios into the test distribution. In the case of independent portfolios in market risk VaR,
where the exceptions are independent, the probability distribution
of the excessions is binomial. To illustrate the new approach let us
consider a stylised example. Here we take the binomial CCDF and
compare it to some sample distributions that were generated by
taking a 1,000 random draws from a truncated student-t distribu307
19/03/2010 17:12
tion with standard deviation 1.6 and mean 4. We take the maximum of the random numbers and zero and then convert to integer
values. This forms sample 1 in Figure 13.5. The two remaining
samples, sample 2 and sample 3, are formed by progressively
shifting the mean of the distribution to the left, such that the means
are 2 and 1 respectively.
Figure 13.5 Histogram of sample 1 constructed by a student t with standard deviation 1.6, mean 4 and degrees of freedom 3, together with the implied breaks given
the assumption of a binomial distribution
Samples 2 and 3 are built from sample 1 by progressively moving the mean to the left.
308
19/03/2010 17:12
We then form the CCDFs shown in Figure 13.6.

Figure 13.6 Some sample empirical CCDFs as compared with a binomial CCDF
To compare the empirical CCDF (built from the observed breaks),

, to the theoretical CCDFs, F(x), we calculate a test statistic that
is designed to determine if there is evidence to reject the hypothesis
that the sample of observed breaks comes from the binomial distribution. To do this we can use the following test statistics:
o KolmogorovSmirnov two-sided statistic
o KolmogorovSmirnov one-sided statistic
}
The critical values of the two-sided and one-sided test can be approximated analytically, and have the following asymptotic values
309
19/03/2010 17:12
(Berry and Lindgren 1990). In Table 13.2 we give the critical values
for a range of significance levels. The critical values are calculated
for a sample size of 1,000 in line with the stylised examples underlying Figure 13.6. The asymptotic formulas are quoted for the
two-sided test only. However, the confidence levels for the one-sided test are given by the asymptotic values corresponding to confidence levels, which are approximately half those of the two-sided
test. To obtain the critical values for the one-sided test, the asymptotic formulas were approximated for the one-sided test by linearly
interpolating the values of the numerator.
Table 13.2 Critical values for the KolmogorovSmirnov test statistic
Significance
level
Asymptotic
formula
(two-sided)
Critical value
(two sided,
N=1000)
Critical value
(one sided,
N=1000)
0.10
1.22
N
0.039
0.033
0.05
1.36
N
0.043
0.039
0.01
1.63
N
0.052
0.048
Note that the significance levels for the one-sided test are half those of the two-sided test
for the same critical value. The critical values for the one-sided test are obtained from the
two-sided test values using linear interpolation.
If the KS statistic chosen is greater than these values then the hypothesis can be rejected and the model fails to be a good out-of-sample
predictor. Table 12.3 shows the results for the hypothetical example.
Table 12.3Results for the hypothetical example
KS for
significance
level 0.01
Sample 1
Sample 2
Sample 3
Model with
bias
One-sided
(0.377 |FAIL )
(0.016 | PASS)
( 0.01| PASS)
(0.343|FAIL)
Two-sided
( 0.377|FAIL )
( 0.127| FAIL)
( 0.319|FAIL )
(0.343|FAIL)
310
19/03/2010 17:12
Note that sample 2 would have failed under the 1996 Basel Framework but is considered plausible under the new framework, although it is clearly not the perfect model as, by the design of this hypothetical example, there are two different underlying probability
distributions. Also, the model with bias, which would have passed
if the portfolios were tested under the Basel framework, clearly fails
as we may have expected by inspecting the CCDFs.
These comparative results show that you should be careful in
designing your test statistic. The two-sided KS statistic is useful if
you want to build the most accurate model. However, from a prudent risk management perspective, one may be focused on building a conservative model but less concerned if it is too conservative,
in which case the one-sided statistic may be more appropriate.
The KS statistic was used here because of the ease of application
and the availability of asymptotic formulas for the critical values.
The KS statistic is easy to apply because the maximum distance
is measured between the CCDFs and, given the assumption of the
number of independent portfolios, then the appropriate critical
values are looked up using the asymptotic formulas or tabularised
values. However, the KS statistic, given that it depends on only a
point-by-point comparison of the CCDFs, could be viewed from
the perspective of making maximum use of the available information as inferior to the CramrVon Mises statistic, which is sensitive to the entire area between the CCDFs.
As we shall see in the remaining sections of this chapter, the use of
Monte Carlo techniques permits a more general formulation to enable us to calculate the critical values for any test statistic that the
model developer finds useful. Moreover, the use of Monte Carlo techniques also enables us to drop the assumption of independence of
portfolios and incorporate the correlation between portfolios directly.
COUNTERPARTY RISK BACKTESTING
In this final section we shall describe the generalisation of the market risk framework described above to counterparty credit risk. The
central feature of the generalisation is the use of Monte Carlo simulations, in order to determine the probability distribution of breaks
implied by the model. The use of Monte Carlo methods for hypothesis testing and for the determination of critical values for distribu311
19/03/2010 17:12
tions is well founded in the statistics literature (for example, Berry

and Lindgren 1990, and Greenwood, Landaw and Brown 1999).
Backtesting then becomes a comparison between the probability
distribution of the observed exceptions and the probability distribution derived from the models simulation outputs, as opposed to
an a priori analytical probability distribution used in market risk. If
sufficient statistical evidence is found to reject the hypothesis that
the model fits the observed data, then we are prompted to improve
the model. In fact, even if this is not the case, we may seek to improve the model based on the actual size of the P-value, higher Pvalues reflecting more accurate7 models. This is especially important if these models are used for CVA.
The main elements of the approach to be outlined here were designed to ensure we are able to meet the following criteria:
o To account for the term structure of the probability of exceptions
for coherent measures. Unlike percentile measures, coherent
measures (such as EPE and ESF) do not correspond, a priori, to a
given percentile, as they are, by definition, an average over some
part of the distribution. However, for the purposes of backtesting, we need to ask what the probability of exceeding the value
of the coherent measure is at each point in time. To do this we
must take recourse in the Monte Carlo simulations that have
determined the value of the measure. The value of the coherent
measure will correspond to a certain percentile on the distribution, and thus we determine the probability of exceeding its value. An example of such values is shown in Figure 13.7.
o To capture path dependence. Since counterparty credit risk measures the exposure of the portfolio at multiple points out to the
maturity of the longest trade in the portfolio, it is clear that if the
realised value of the portfolio moves towards the profile then the
probability of exceeding the profile will increase, and likewise
once an exception has occurred a subsequent exception is more
likely to occur. Therefore, this path dependence means that exceptions are not independent and we must therefore incorporate
this into the models CCDF (see Figure 13.7 for an example of this).
o To determine the complementary cumulative distribution function for a given model. The path dependency and other properties of the model, such as drift and skew, all have an impact on
312
19/03/2010 17:12
the probability distribution of exceptions.

o Aggregate samples drawn from different CCDFs. As there are no
analytical formulas for the probability distribution of exceptions,
we must find a way to aggregate observed exceptions while preserving their probability of occurrence.
o To construct appropriate test statistics. Since we have access to
the underlying Monte Carlo distributions we are able to explicitly build the null distributions corresponding to the hypothesis.
o To determine if we have sufficient sample size for a good test.
If we want to be able to make a commitment to improve a failing
or inaccurate model, we must make sure we have a large enough
sample to make the decision robust.
Figure 13.7 An example of a 10-year cross-currency swap (N=100 MM US$)
On this page we have plotted the PFE (PSE equivalent) and the EPE (PSLE equivalent) in US$ as
a function of time; on the next page we have plotted the probability of exceeding each profile as a
function of time. It is clear that the probability of exceeding the PFE profile is constant, while that of
exceeding the EPE is an increasing function of time (for this example).
313
19/03/2010 17:12
We now pause to remind the reader of the definitions of potential

future exposure (PFE) and expected positive exposure (EPE):
o PFE(t) is defined to be the X-th percentile of the forecast distributions at each observation time, t. The probability of exceeding the
PSE at time t, determined at the prediction time t=0, is always
1-X%.
o EPE(t) is defined to be the average of the values in the distribution once any negative values have been set to zero. The probability of exceeding the EPE at time t determined at the prediction time t=0 can be any set of values in [0,1],8 usually different
for each t.
How to account for the term structure of probability of exceptions
for EPE and ESF9
The following method allows us to count exceptions along a path
where the probability of exceptions is not homogeneous. This is the
case for coherent measures. Let us suppose that we are comparing
314
19/03/2010 17:12
Figure 13.8 Two backtests
The one at the top shows the margined exposure backtest, and the one at the bottom
shows the unmargined exposure backtest. The grey lines are the PFE barriers. The
white lines are the EPE barriers, and the black lines are the historical exposures. In
the top diagram left we see the effect of path dependence, where a large move at the
beginning of the profiles leaves the historical path above the expected for almost the
entire one-year backtest.
315
19/03/2010 17:12
the exposure profile produced by our model with the realised portfolio value at discrete points in time ti, i=1,,. We denote the
increments between these times by d(ti). At each point in time we
know the probability of exceeding the value of the exposure profile (as determined from the Monte Carlo simulations described
above). We denote this probability by p(ti). We then calculate the
time-weighted average probability and denote it f. If a break occurs when the probability of a break is high we want it to count
for less than a break that occurs when the probability of a break is
low. Therefore, we introduce the ratio of the average probability to
the probability at that point in time as an adjustment to the time
weighting of that period. The reader should note that this adjustment is not unique, but does define a natural adjustment or penalisation function.
Let 1(ti) denote the indicator function that takes a value of one when
an exception occurs and a value of zero otherwise. Let T equal the
sum of the time periods or total time of the backtest. Then we may
write the following expression that counts the exceptions along a
path
The reader should note that when p(ti) = f for any i, as is the case for
percentile backtesting, then this formula reduces to a simple count
of the exceptions.
Calculating the model CCDF incorporating path dependency
Let us suppose that we have the following dimensions (see Figure
13.9) to the sample space for the backtest:
o Backtesting window. We have conducted a number of backtests
at several starting points.
o Portfolio index. We have tested a number of portfolios over each
backtesting window.
o Simulation index. Each Monte Carlo has run the same number
of simulations over the same number of time steps (a simplifica316
19/03/2010 17:12
tion). Note that within each backtesting window the simulations

are coherent, while between backtesting windows a different random draw is used. Therefore, the assumption of zero correlations
is acceptable so long as the backtesting windows do not overlap.
Within the same window, maintain coherence between portfolio
backtesting, which captures the correlation between the portfolios
and encodes this into the null distribution (see below).
According to these dimensions, we capture the following pieces
of information:
o the number of observed exceptions, given by (empirical), as
per the formula, for each portfolio and each backtesting window;
and
o the number of model exceptions (model) implied by the model
for each portfolio at each simulation path for each backtesting
window.
Figure 13.9 Aggregate cube of model breaks
This is important in the construction of the hypothesis test, as certain slices provide key tools in the
construction. In particular, note that the slices defined by constant alpha are such that all underlying
simulated markets are coherent.
317
19/03/2010 17:12
Before we take another step, we are required to normalise the observed and model exceptions. This is because the exceptions themselves are not comparable, but rather only their probabilities can be
aggregated across models since the models CCDF is defined only
at the level of granularity given by backtest and by portfolio.
In order to affect this, we introduce a map that translates the
exceptions onto a standardised probability distribution such that
their probabilities are preserved.
The choice of normalisation distribution
The choice of normalisation distribution is arbitrary so long as it
leads to a monotonically decreasing CCDF. One natural choice
of normalising distribution is given by calculating the empirical
CCDF across all model exceptions as if there were a single sample,
Figure 13.10 A plot of typical CCDFs for a multiple portfolio backtest
In this instance the underlying portfolios are UK natural gas swaps
318
19/03/2010 17:12
denoted E(total). Another natural choice would be the CCDF corresponding to the uniform distribution. The choice of normalisation
distribution then leads to a transformation designed to preserve the
probability of exceptions.
The probability preserving map
To construct this map we first calculate the empirical CCDF for each
portfolio and backtest from the model exceptions that correspond
to it, denoted E(factor). A set of E(factor) CCDFs corresponding to a
multiple portfolio backtest is shown in Figure 13.10.
The probability preserving map Y, is then defined by the following function
Note that this function acts to the right. So that for each exception
Here E(factor) maps W onto the probability of W, and E(total)-1 maps

the probability of W onto an E(total) equivalent number of exceptions.
See Figure 13.11 for a diagrammatic representation of the action
of this map, and an example showing the actual changes in value
corresponding to the example in Figure 13.10.
This operation is completed for all model and observed exceptions before computing the null distributions of the chosen test statistic and the empirical CCDFs.
It is important to note that this normalisation is essential for the aggregation of any results.
The hypothesis-testing framework for counterparty credit risk
As explained in the prior section, we are now able to construct the
null distributions and the empirical CCDF given the normalised
exceptions.
The empirical CCDF is straightforward to compute, denoted
E(empirical). To derive the null distributions we must compute the
model implied CCDFs per simulation dimension, denoted E(sim).
This results in 1,000 CCDFs for 1,000 simulations. An example is
shown in Figure 13.12. Note that the width of the distribution is
sensitive to the number of independent portfolios, as the depen319
19/03/2010 17:12
dence in the portfolios rise the E(sim) disperse and fill the space of
possible CCDFs. This has the effect of reducing the power of your
test, irrespective of your choice of alternative model.
Figure 13.11 The diagram on the top
illustrates the action of the probability
preserving map, while the diagram on
the bottom shows an example of the
actual transformations of the observed
breaks (the backtest was based on
portfolios of UK natural gas swaps)
320
19/03/2010 17:12
The null distributions are built from the E(sim) using the test statistic. Just as we used the KolmogorovSmirnov statistic in two
variants in the last section, we can now use the CramerVon Mises
statistic here in two variants. The reason for preferring this statistic is the fact that it measures the goodness of fit for the model to
realise across the entire CCDF, and therefore has more information
available.
Figure 13.12 A plot of typical CCDFs defined for simulation for a multiple
portfolio backtest
This distribution of CCDF is then used to construct the null distribution by applying the
test statistic (in this instance, the underlying portfolios are UK natural gas swaps).
We the compute a family of CCDFs, one for each simulation path,

as follows. Take all normalised model exceptions corresponding
to a given simulation path s, and calculate the empirical CCDF,
denotedE(sim). The mean of the E(sim) is equal to the CCDF calculated from the entire sample; E(total) above. Note that E(total) can be
arbitrary, which is another nice and natural feature.
The two test statistics are defined as follows:
321
19/03/2010 17:12
o CramrVon Mises two-sided statistic

This statistic tells us how close the theoretical distribution is
to the empirical distribution. Note we explicitly expressed the
CCDF E(.) as a function of the number of breaks x.
o CramrVon Mises one-sided statistic
This statistic tells us how conservative our model is by only measuring values where the empirical lies above the theoretical.
The attentive reader will notice that we do not use the exact form of
the CramrVon Mises statistic here, but prefer a linear alternative
(Anderson 1962). We can use this because we measure the distance
without using the root mean square, which we can work effectively as we are using Monte Carlo calculations rather than deriving
analytical approximation formulas. It has the desirable feature of
equally weighting distance from E(total) rather than a quadratic
weight.
To construct the null distribution we compute the value of the
test statistic corresponding to each simulation path, indexed by
sim (sim = 1, 2, 3, )
here sim replaces sample in the above equations for the test statistic.
We thereby generate the null distribution of the test statistic with
the same sample size as the simulations. The critical values for rejection given the significance levels can then be directly computed.
The value of the test statistic is then computed for the realised
data of E(empirical); here empirical replaces sample in the above
322
19/03/2010 17:13
equations for the test statistic (see Figures 13.13 and 13.14 for an example). The value of the empirical test statistic is then compared to
the values derived from the null distribution. If the value is less than
the 0.01 significance level, then the model is determined to have
failed the hypothesis test and requires remediation. In our example,
the value of the test statistic corresponds to a P-Value greater than
50, in which case the model is performing well and is a long way
from the rejection region.
To compute the null distribution corresponding to the alternative
hypothesis, you need to construct a family of CCDFs that are from
a model other than the one being tested. Since you have no knowledge of what the new model may be, you are free to construct the
alternative model either by running a different model/calibration or
by appropriately parameterising the space of CCDFs that are more
conservative than your model. Using these approaches you can estimate if you have sufficient statistics.
A very important consequence of calculating the null distribution of the hypothesis from the Monte Carlo simulations is that we
need not assume that the portfolios are independent, so long as the
backtests are conducted such that the path-by-path exposures for
all portfolios have been simulated coherently. When this is done
correctly, the width of the null distribution then carries information
about the effective sample size, the width increases as sample size
decreases. A small sample size leads to a very wide null distribution.
This, in turn, affects the statistical power of the test, as a wider null
distribution leads to a lower statistical power. For stylised test cases
with 1050 independent (zero correlation) portfolios, the power of
the test ranges from 80% (beta of 0.2) to 90% (beta of 0.1), respectively.
SUMMARY AND CONCLUDING REMARKS
The chapter provides an overview of the counterparty credit risk
backtesting framework. We started from a review of the Basel market
risk backtesting framework. The recasting of the market risk framework into a comparison of CCDFs leads on to a counterparty credit
risk framework involving a simulation-based statistical inference
method. While we have found these method useful in testing and
developing counterparty credit risk models, we remind the reader
323
19/03/2010 17:13
that the methods described here are only a subset of the methods
that could presumably be developed, each with a certain utility.
Figure 13.13 This shows an example of the null distribution corresponding to the CramrVon Mises two-sided statistic
This distribution is formed by taken the distribution of CCDFs in Figure 13.12 and applying the test statistic. Below the graph is shown the value of the statistic (distance), the P-value and the power of the test Beta.
324
19/03/2010 17:13
Figure 13.14 This shows E(total) and E(empirical) corresponding to the example of the
UK natural gas
Although it appears that the two distributions are close, it is the computation of the P-value that confirms this
as a statistical fact.
2
3
5
6
7
8
9
The analysis and conclusions set forth are those of the authors. Citi is not responsible for any statement or conclusion herein, and opinions or theories presented herein do not necessarily reflect the
position of Citi. This chapter is based on work previously undertaken by Sean Paul Hrabak in his
former position as head of counterparty credit modelling at Citi and does not reflect in any way the
views or opinions of the Financial Services Authority.
The exposure is calculated out for the full maturity even if the portfolio is covered by a CSA as
margining reduces the size of the exposure but not the tenor of the exposure.
The capital calculations depend on the lifetime exposure profile in that in the IRB framework EAD
is constructed from the first year of the profile, and the maturity adjustment M is constructed from
the entire exposure profile.
EPE(t) is defined to be the average of the values in the distribution once any negative values have
been set to zero. ESF is defined to be the average of the values in the distribution beyond some
percentile X. It therefore represents the expected value of the exposure that the model predicts will
occur with a probability less than or equal to 1-X.
We should note that economic risk capital is also calculated over a long-term horizon, viz one year.
99% VaR is sometimes also referred as the minimum loss that can occur 1% of the time.
Please note that when we refer to accuracy of the model, it must be understood that this is predicated by the statistical nature of the model. That is we mean accurate in a statistical sense.
Although this probability is unlikely in principle ever to be exactly one or zero, the use of Monte
Carlo simulation to approximate the theoretical distribution makes it likely in practice.
The following holds for all coherent measures, EPE and ESf being examples thereof.
325
19/03/2010 17:13
REFERENCES
Anderson, T. W., 1962, On the Distribution of the Two Sample Cramer-von Mises
Criterion, Annals Math. Stat., 33(3), pp. 11481159.
Bank for International Settlements, 1996, Supervisory Framework for the Use
of Backtesting in Conjunction with the Internal Models Approach to Market Risk
Capital Requirements, URL: www.bis.org/publ/bcbs22.htm.
Bank for International Settlements, 2005, The Application of Basel II to Trading
Activities and the Treatment of Double Default Effects.
Bank for International Settlements, 2006, International Convergence of Capital
Measurement and Capital Standards, URL: www.bis.org/publ/bcbs128.pdf.
Berry, D. A. and B. W. Lindgren, 1990, Statistics, Theory and Methods, Brookes/
Cole.
Campbell, S. D., 2005, A Review of Backtesting and Backtesting Frameworks,
Finance and Economics Discussion Series, 21, Board of Governors of the Federal
Reserve System.
Greenwood, A., E. M. Landaw and T. H. Brown, 1999, Testing the Fit of a Quantal
Model of Neurotransmission, Biophysical Journal, 76(4), pp. 18471855.
ISDA, BBA and LIBA, 2008, Observations: Back Testing in IMM Firms, URL: www.
bba.org.uk/content/1/c6/01/49/87/Joint_association_response_to_UK_FSA_observations_on_back_testing_in_IMM_firm.pdf.
326
19/03/2010 17:13
14
Economic Capital for Counterparty

Credit Risk from Two Perspectives
Evan Picoult
Citi and Columbia Business School1
Economic capital for counterparty credit risk (CCR) can be quantified from two very different risk perspectives, which for the purpose of this chapter we will initially refer to as trading book and
banking book risk perspectives. We will later refine these concepts and designate the two perspectives as the price risk and the
value risk perspective.2 The measurement of risk in all portfolios
that have credit-sensitive transactions can be calculated from either
of these two risk perspectives. The magnitude of virtually every
type of risk measurement, including stress tests and economic capital, will differ depending on which risk perspective is employed.
This chapter is organised as follows:
o an illustration of the two perspectives a comparison of the difference in the risk measurements of a loan in a banking book
versus a bond in a trading book;
o a comparison of a) the banking book risk perspective, by which
CCR traditionally has been managed, accounted for and assigned economic and regulatory capital, and b) the trading book
risk perspective, in which the credit value adjustment (CVA)
for counterparty credit risk3 is accounted for and dynamically
hedged as a component of market risk;
o the definition of economic in terms of its primary use;
o the more precise definition of price risk and value risk, the
conditions under which it is appropriate to apply each perspective to the measurement of risk and the measurement of econom
327
19/03/2010 17:13
ic capital from each perspective;

o the measurement of economic capital for CCR from each perspective; and
o conflicts between marking-to-market the CVA and calculating
economic capital and regulatory capital from a value risk perspective.
The distinction between the two perspectives is meaningful, corresponds to different ways of managing, funding and accounting
for a credit-sensitive portfolio and has very real consequences for
the measurement of risk. The economic capital per unit exposure
for a credit-sensitive portfolio tends to be higher when measured in
a robust way from a trading book rather than a banking book risk
perspective, for reasons that will be explained below.
Under the current Basel II rules,4 the minimum regulatory capital
required for CCR is calculated under the implicit assumption that
CCR is accounted for and managed as traditional banking book risk.
However, with the adoption of the fair value accounting option, under FASB 157, many large banks now account for and actively hedge
their CCR (or, more specifically, they hedge their CVA) as a form of
trading market risk. The breakage between the assumptions underlying the calculation of minimum regulatory capital for CCR and
the reality of how the risk is accounted for, hedged and managed
has led to several problems, which will also be described below.
ILLUSTRATION OF THE TWO PERSPECTIVES LOAN V. BOND
To put the treatment of CCR into context, let us contrast the accounting and risk measurement of a fixed-rate loan portfolio,
viewed from a banking book perspective with a fixed-rate bond
portfolio, viewed from a trading book perspective. For the purpose
of this thought experiment, let us assume the default risk characteristics of the two portfolios are identical ie, they have the same
distribution of probability of default (PD), loss given default (LGD)
and effective duration (M). From either perspective, over any future time period there is a risk of losses from some assets defaulting
and a risk of losses from the non-defaulted assets.
Loan portfolio in a banking book
In a banking book perspective, the risk of potential losses of a loan
328
19/03/2010 17:13
Economic Capital for Counterparty Credit Risk from Two Perspectives
portfolio over a future time horizon can be categorised into three

components.
o The interest rate risk between the assets and liabilities (ALM
risk) of the loan portfolio. This is the risk of potential changes, in
both: a) the net interest income within the period and b) the present value, as of the end of the period, of the net interest income
over the remaining life of the portfolio. .
o The risk of an increase in the loan loss reserve (LLR) at the end of
the period. The accounting value of a loan portfolio is the sum of
the par value of the loans minus the provision for loan loss, also
known as a loan loss reserve . The LLR is the expected default
loss over some future time horizon (eg, the remaining life of the
portfolio).
o The risk of default losses in the future period.
As will be explained in more detail below, from a banking book perspective, annualised stress tests and economic capital for the credit
risk of a loan portfolio should be based on the risk of unexpected
default losses within one year and unexpected changes in the LLR
at the end of the year.
Bond portfolio in a trading book
The mark-to-market value of a bond in a trading book is equal to:
PVbond = PV risk-free Bond credit risk premium
(14.1)
Where PVrisk-free = The present value of all future cashflows discounted at a risk-free yield. In this context, risk-free yield means a
base yield curve, like a sovereign yield or Libor.
Bond credit risk premium PVrisk-free * DurationBond * SpreadBond
(14.2)
The risk of a fixed-rate bond portfolio, in a trading book, over any

future time period, has three components.
o The risk of changes in the risk-free yield curve.
o The risk of changes in the spread to the risk-free yield (ie, changes
329
19/03/2010 17:13
in bonds credit risk premium) for bonds that do not default in

the future period.
o The risk of default losses in the future period.
Let us now contrast the treatment of the risk of a portfolio of fixed
bonds with a portfolio of fixed-rate loans, as summarised in Table 14.1.
Table 14.1 Risk measurement: loan v. bonds
Risk measure for assets simulated to have:
No default in period
Default in
period
Risk component
Risk-free yield
curve
Issuer/borrower
credit risk
Loans in banking book
ALM risk
Change in LLR
Default
loss
Bonds in trading book
Interest rate
risk
Credit spread risk
Default
loss
Both portfolios are sensitive to changes in the level and shape of

the risk-free yield curve. The primary difference between the two
perspectives lies in the timing of the accounting recognition of
gains and losses caused by changes in the risk-free yield curve. In
a trading book changes in the risk-free yield curve immediately affects the present value of all expected future cashflows. In a banking book, changes in the risk-free yield curve affect the net interest
income in the current and in future periods, but, in contrast to the
trading book, the net interest income drips in over time.
In a severe systemic stress event, material differences can occur
between the changes in the credit risk premium of a bond portfolio
(due to changes in market spreads) and a banks objective assessment of changes in the LLR of an equivalent portfolio of loans. This
difference is not primarily caused by a difference in the timing of
the recognition of cashflows, as is the case above. The difference can
be understood by decomposing the credit spread of a bond into its
four components:
o the expected default loss, which is the product of PD and LGD;
o a credit risk premium for uncertainty in the expected default loss;
330
19/03/2010 17:13
o a trading liquidity risk premium, for uncertainty about the price

the security could be sold for, after it has been bought; and
o a funding liquidity risk premium, for uncertainty about the cost
of funding the security, after it has been bought.
During times of severe systemic risk, such as occurred during 2008,
each of the four components will increase. However, the third and
fourth components will tend to have a larger relative increase in
magnitude because liquidity risk tends to manifest itself disproportionately in a severe systemic event. As a consequence, under a severe stress event, the bond credit risk premium will tend to increase
much more than an objective assessment of the change in lifetime
default loss.5 The ultimate source of this difference lies in how each
type of portfolio is managed, as will be described below.
TWO PERSPECTIVES ON CCR
Table 14.2 Risk measurement: OTC derivative portfolio traditional v.
trading book
Risk measure for assets simulated to have:
No default in period
Default in
period
Value component
Counterparty
risk free market
value*
Counterparty credit
risk premium
Traditional
method
Market risk
(ex-CVA)*
Change in EDL**
Default loss
Full trading book

method
Market risk
(ex-CVA)*
Market risk of CVA
Default loss
* Market risk of OTC derivatives discounted at Libor, independent of credit spread of counterparty.
** EDL is the expected default loss for CCR. This can be accounted for as a contra-asset account
eg, either as a provision for CCR default loss or an adjustment to the gross positive market value
of a portfolio of OTC derivatives.
Traditional treatment of CCR

A portfolio of OTC derivatives in a trading book traditionally has
331
19/03/2010 17:13
been bifurcated into two very different, but somewhat overlapping

realms, somewhat like the treatment of a loan in a banking book
described above. In analogy to Equation 14.1
PVOTC_Derivatives = PVrisk-free Counterparty credit risk premium (14.3)
Where PVrisk-free = Marked-to-market value of derivative portfolio,
in which all future cashflows are discounted to present value at Libor.
It is a risk-free valuation with respect to counterparty credit risk.
The counterparty credit risk premium of an OTC derivative portfolio traditionally has been calculated in different ways, depending
in part on whether the bank was a commercial bank (which had a
loan portfolio) or an investment bank (which essentially marked all
of its assets to market). The common denominator of either approach
was the creation of a contra-asset account to reduce the gross positive
marked-to-market value of a portfolio of OTC derivatives, in recognition of potential future credit losses. Firms differed in the sophistication of how this calculation was done. In essence, in a traditional
banking book approach a bank would create a provision for potential
counterparty default loss. In an investment banking approach, the
bank would create an adjustment to the market value of its OTC
derivatives, to reflect its CCR. The latter practice evolved into todays
CVA calculation.
In the more traditional banking approach, the counterparty credit
risk premium for OTC derivatives was calculated as provision for
potential future CCR default losses. The method of calculating the
provision changed over time as individual banks, and the industry
as a whole, gained more knowledge. In a traditional approach at a sophisticated bank, the provision, for each netting set or counterparty,
could have been estimated as the product of a) an expected exposure
profile (eg, expected positive exposure, EPE), and b) an expected default loss per unit exposure (ie, the product of the historic expected PD * LGD). The only recognition of the stochastic nature of the
underlying exposure would have been in the use, per netting set or
counterparty, of the EPE profile as the loan equivalent for calculating the provision. Equivalently, for each netting set or counterparty,
the loan equivalent exposure could have been based on the average
EPE profile over the first year and the effective duration, M, of the
EPE profile (Picoult 2005).
332
19/03/2010 17:13
It should be noted that Basel II implicitly assumes that the credit

risk premium for counterparty credit risk is a component of the provision for credit loss, since Basel II treats CCR as a form of wholesale
credit risk.6
At some investment banks the counterparty credit risk premium
was calculated as an adjustment (ie, a reduction) of the marked-tomarket value of a portfolio of OTC, per netting set or counterparty.
At a sophisticated investment bank this adjustment could have
been calculated as the product of a) the expected positive exposure
(EPE) profile, and b) market spreads (ie, essentially the CVA_asset).
It was obvious well before the 2008 financial crisis that the expected historic default loss percentage tended to be much less than
credit spreads for investment grade ratings and very short tenors.
This illustrated the relative importance of the credit risk premium,
trading liquidity risk premium and funding liquidity risk premium
components of credit spreads.
In summary, in the tradition banking book bifurcated perspective, CCR was accounted for and measured as a provision for default loss. Although the provision would have been sensitive to the
credit rating of the counterparty (which would have determined
the PD of the counterparty), it would not have been directly sensitive to changes in the counterpartys credit spread. The stochastic
nature of potential credit exposure would have been taken into account in the calculation of the provision by using the EPE profile as
the loan equivalent exposure of each netting set or counterparty.
As I have framed the issue, the essential difference in the calculation of the CCR risk premium between a traditional banking
book approach and an investment bank approach is in the use of
expected default loss as a percent of exposure (eg, PD*LGD) by the
former versus market spreads by the latter. The specific accounting
category (ie, provision for default loss versus adjustment of market value) is non-essential in this context. They are each a type of
contra-asset account.
Full trading book treatment of CCR
In the full trading book treatment of CCR, the counterparty credit
risk premium for a portfolio of OTC derivatives is the market value
of its CVA. The CVA is the analogue of the market-based issuer credit risk premium of a bond, described in Equation 14.2. An important
333
19/03/2010 17:13
difference between the two is that the CVA must be simulated on a

portfolio basis, taking netting and margin agreements into account,
in recognition of the stochastic uncertain future exposure.
A second difference is that under GAAP fair value accounting
rules, the CVA must be calculated from a bilateral perspective. This
means that both the CVA_asset and the CVA_liability are part of the
mark-to-market value of an OTC derivative portfolio. The CVA_asset is the analogue of the issuer risk premium of a purchased bond.
The analogue of the CVA_liability is the issuer risk premium of a
banks own debt security. The inclusion of the CVA_liability (or a
banks own debt securities) in a mark-to-market valuation leads
to the counterintuitive consequence that a bank will record a P&L
gain, all else held constant, when its credit spread widens, and a
P&L loss when its credit spread narrows.
We will temporarily put aside the question of the appropriateness of using a unilateral or bilateral CVA until later in the chapter.
When CCR is evaluated from a full trading perspective, the risk
of a portfolio of OTC derivatives over any future period has three
broad components.
o Market risk arising from potential changes in the counterparty
credit risk free value of the portfolio. This is the traditional market risk of a portfolio of OTC derivatives and consists of the risk
of changes in all the market factors that affect the non-CVA market value of the portfolio.
o Potential changes in the market value of the CVA for the counterparties that do not default in the future period. The drivers of potential changes in the CVA of each netting set (or counterparty)
are changes in:
i) the market-based exposure profile7 of each netting set or
counterparty;
ii) the counterpartys CDS spread; and
iii) the implied correlation between i) and ii).
o Risk of default losses in the future period.

DEFINITION OF ECONOMIC CAPITAL
Economic capital is a particular type of risk measurement, primar334
19/03/2010 17:13
ily used to ascertain the capital adequacy of a firm with respect to

the risk of insolvency. The economic capital of a firm is its potential
unexpected loss, over a one-year horizon, calculated at a very high
confidence level (eg, 99.97%). Unexpected loss (UL) is the difference between the loss at the specified high confidence level and
the expected loss (EL). This relationship is illustrated in Figure 14.1
below.
Figure 14.1 Economic capital illustrated for a loan portfolio
The essential framework underlying the concept of economic capital is that to avoid insolvency over the following 12 months, a bank
needs two types of financial buffers to absorb potential losses.
o A bank should have provisions (eg, loan loss reserves for potential credit losses in the banking book) that, at a minimum, are
sufficient to cover its expected loss, for all forms of risk,8 over
that time period.
o It should have available financial resources9 (AFR) that, at a minimum, are sufficient to absorb potential unexpected losses at the
high confidence level used to calculate economic capital, to enable the bank to avoid insolvency and to continue functioning as
a going concern. Thus, economic capital sets a minimum level of
required AFR.
The confidence level used to measure economic capital should cor335
19/03/2010 17:13
respond to a target credit rating. To see the relationship between a

target credit rating and a confidence level, consider the following
example. Let us assume that the average, through-the-cycle probability of default (PD) of firms with the credit rating of AA is 3bp. It
follows that a minimum condition for having a AA rating would be
that a banks AFR (its buffer to absorb unexpected losses to avoid
insolvency) is larger than its economic capital, when the latter is
calculated at the 99.97% CL (ie, 1 the average PD of the target rating). The prior sentence can be more directly stated as: A minimum
condition for having an AA rating would be that the probability of a
banks unexpected loss exceeding its AFR should be less than 3bp.
One way of ensuring that a bank is adequately capitalised for
a severe economic downturn is to calculate economic capital under severe systemic stress conditions, rather than conditions corresponding to the then current state of the economy.
Although the primary use of economic capital is the assessment
of the capital adequacy of a firm, it has several important subsidiary uses as a component in the analysis of pricing, performance
evaluation and strategic planning.
DEFINITION OF PRICE RISK PERSPECTIVE
In a price risk perspective, all forms of risk measurement are calculated in terms of potential changes in the market value of a portfolio
and the potential incremental credit losses due to obligor or counterparty defaults.10 A robust methodology would avoid doublecounting these two types of losses.
The price risk perspective should be used to analyse the market and
credit risk of all portfolios, except those that have the specific attributes needed to qualify for a value risk perspective, as defined below.
Defining stress scenarios in a price risk perspective
The potential change in the market value of a portfolio is simulated
in terms of changes in the underlying market factors that determine its value, such as the level and shape of base (ie, risk-free)
yield curves, credit spreads, equity and equity indexes, commodity
prices and the implied volatilities and correlations of these factors.
Consequently, from a price risk perspective, stress scenarios over
some time horizon are specified in terms of a) changes in market
336
19/03/2010 17:13
factors, needed to measure the potential loss of market value for all
transactions, except those for which issuers and counterparties are
simulated to default during the time horizon, and b) macroeconomic factors, needed to specify potential default losses of issuers and
counterparties over that time horizon. Systemic stress-test scenarios (SSTSs) can be defined by very broad changes in similar types
of market factors, eg, all equity indexes and prices fall by 40%, under the assumption that the correlations of changes in similar types
of market factors will approach 1.0 during a severe systemic stress
event.11 In contrast, business-specific stress-test scenarios (BSSTSs)
focus on the basis risk of particular desks and need to be specified
in the context of the positions each trading desk takes.
From a price risk perspective, economic capital can be defined
in terms of an annualised potential loss of market value and an annualised potential incremental default loss (IDR), relative to the expected default loss:
The annualised loss of market value should be based on some
combination of:
o the annualised scaled VaR;12
o one or more13 annualised systemic stress tests {SST}; and
o one or more annualised business specific stress tests {BSST}.
IDR is an incremental measure of default loss for two reasons:
o it is incremental relative to the expected default loss;14 and
o it is incremental relative to the economic capital attributable to
the loss of market value, to ensure that there is no double counting ie, for any counterparty, there is a probability of default
(PD) and a probability (1 PD) of a change in the market value
of transactions without default.
The simplest way of specifying economic capital for trading market is
ECPrice_Risk = Max(scaled VaR, {SST}, {BSST}) + IDR
(14.4)
A more sophisticated approach would entail integrating VaR, stress

testing and default risk into a coherent simulation.15
337
19/03/2010 17:13
DEFINITION OF VALUE RISK PERSPECTIVE

A value risk perspective, in essence, is a hold-to-maturity and
funded-to-maturity default risk perspective. As explained above,
the price risk perspective should be used to analyse the risk of all
credit-sensitive portfolios, except those that have the specific attributes needed to qualify for a value risk perspective. The criteria for
applying a value risk perspective are:
o there is no intention of selling, hedging or securitising the transactions in a portfolio;
o the transactions are not subject to mark-to-market accounting;
o changes in the mark-to-market value of the assets do not affect
book equity through the OCI process; and
o the assets in the portfolio are funded until maturity ie, there is
no accrual funding liquidity risk.
If these conditions are met, default risk is the only source of uncertainty about the cashflows that will be realised over the life of the
portfolio. The cashflows of such a portfolio would have no sensitivity to funding or trading liquidity risk.
As a consequence, the primary measure of risk for such a portfolio over its lifetime would be the probability distribution of the
cumulative default losses over that time horizon.
A lifetime unexpected loss (UL) for such a portfolio would be the
difference between the potential default loss (DL) measured at a high
confidence level (eg, 99.97%) and the expected default loss (EDL)
UL(;T0,TMat) = DL(; T0,TMat) EDL(T0,TMat)
(14.5)
Where T0 = current date

TMat = highest maturity date of transactions in portfolio
= confidence level of measurement
This can be re-written as the sum of two terms
UL(;T0,TMat) = UL(;T0,T1) + UL(;T1,TMat)
(14.6)
Let us examine each of these terms. To put them in context, let us

apply this analysis to a loan portfolio, assuming that the LLR for
338
19/03/2010 17:13
this portfolio was equal to its expected lifetime default loss.

The first term, UL(;T0,T1), is simply the unexpected default loss
over the next year ie, the economic capital for default loss occurring
over the next year, when is set to the appropriate confidence level.
The second term, UL(;T1,TMat), is the unexpected default loss
from the end of the first year to the maturity of the portfolio. More
specifically, this is the unexpected change in the remaining provision for default loss at the end of the first year, calculated for an
amortising portfolio with no roll-overs or additional transactions.
The unexpected change in the provision for default loss at the end
of the first year can be attributed to a) the unexpected deterioration in the credit quality of the portfolio (ie, net migration to lower
credit ratings), combined with b) the deterioration in the forecasted
state of the economy over the remaining life of the portfolio.17
In summary, from a banking book perspective, economic capital
for a credit-sensitive portfolio can be expressed as
ECValue_Risk = UL(99.97%; T0,TMat)

= UL(99.97%; T0,T1) + UL(99.97%; T1,TMat)
The right-hand side of the first equation is simply the unexpected
default loss over the lifetime of the portfolio. The second equation
decomposes this unexpected lifetime default loss into the sum of
two terms: the unexpected default loss over the first year, plus the
unexpected change in the remaining provision for default loss at
the end of the first year due to a severe deterioration of the credit
quality of the portfolio, and a deterioration in the expected state of
the economy over the remaining life of the portfolio.
COMPARISON OF ECPrice_Risk V. ECValue_Risk
In summary, we have two general definitions of economic capital
for a credit-sensitive portfolio
Price risk perspective: ECPrice_Risk = Max(scaled VaR, {SST}, {BSST}) + IDR
Value risk perspective: ECValue_Risk = UL(99.97%; T0,T1) + UL(99.97%; T1,TMat)
To illuminate the difference in economic capital between the two

perspectives, let us conduct another thought experiment. Assume:
339
19/03/2010 17:13
o the assets in the portfolio have a market value of US$500 million

and consist of a highly diverse set of A rated, 10-year floatingrate corporate bonds;
o for the value risk analysis, the portfolio is fully funded to maturity by the issuance of US$500 million, 10-year floating-rate debt;
o for the price risk analysis, the portfolio is funded in the shortterm repo market;
o ECPrice_Risk will be driven by the systemic stress-test scenario that
has the largest increase in corporate spreads; and
o ECValue_Risk will equal the stressed incremental cumulative default loss over the life of the portfolio relative to the expected
cumulative default loss.
Under these conditions ECValue_Risk < ECPrice_Risk. This will occur because the loss of market value due to the widening of credit spreads
in a severe systemic stress scenario, such as occurred during 2008,
will exceed the increase in unexpected lifetime default loss of the
assets, conditional on a severe systemic stress scenario. As explained above, this occurs because of the very material widening of
the trading and funding liquidity risk premium components of the
spread in a 2008-type stress scenario.
APPLICATION OF ABOVE FRAMEWORK TO ECONOMIC CAPITAL
FOR COUNTERPARTY CREDIT RISK18
There are two fundamentally different approaches to measuring,
monitoring and managing counterparty credit risk: the price risk
(ie, CVA) approach, and the traditional value risk (ie, hold-to-maturity banking book) approach. The introduction of fair value accounting, and its requirement that a bank mark-to-market both its
CVA_asset and its CVA_liability, has been a big incentive for banks
to adopt a price risk perspective in measuring and dynamically
hedging their counterparty credit risk.
ECPrice_Risk for CCR
ECPrice_Risk for CCR is based on treating the CVA and its hedges like
any other credit-sensitive component of a trading portfolio. ECPrice_
would equal the potential large loss arising from changes in the
Risk
market value of:
340
19/03/2010 17:13
o CVA (ie, the market view of the counterparty credit risk premium); and
o hedges of the CVA, which would include hedges on:
o EPE per netting set, or counterparty (ie, the market vector
component of the CVA)
o general spread risk component of CVA (eg, the use of a CDS or
a CCDS on an index).
o specific counterparty spread risk (including default risk) component of CVA (eg, the use of a CDS or CCDS on individual
names).
In general, economic capital should not include the CVA_liability,
although the affect of the CVA_liability on reducing the funding
costs of the CVA_asset should be taken into account as a higherlevel adjustment.19 However, given that it is a requirement of fair
value accounting under GAAP, a bank may need to calculate two
measures:
o an economic capital, which only includes the CVA_asset and its
hedges, as a measure of economic risk; and
o a shadow accounting economic capital, which would also include the CVA_liability and its hedges, as a measure of the potential unexpected P&L loss under fair value accounting.
The method of calculating ECPrice_Risk for the CVA_asset and its
hedges is not trivial, as it needs to take into account each component of Equation 14.4
ECPrice_Risk = Max(scaled VaR, {SST}, {BSST} + IDR
This includes the capture and use of the following:
o The factor sensitivities of each component of the CVA_asset per
netting set or counterparty and the factor sensitivities of the corresponding hedges. The sensitivities need to be measured both
for a) the short-term market factor shocks used in the VaR calculation, and b) long-term market factor shocks that are used in
the systemic and business-specific stress scenarios (eg, the stress
changes in market factors over a period of one year for potentially illiquid markets).
341
19/03/2010 17:13
o The implied market correlations of each combination of four

factors: a) the market factors that materially affect EPE per netting set, b) general spreads by ratings, c) counterparty specific
spreads, and d) counterparty PD. The capture and robust modelling of these correlations is the means by which general and
specific wrong-way and right-way CCR could be captured.
o The appropriate liquidity horizon applicable for stress changes in
market factors used in the CVA calculation. This would depend
on assumptions about the feasibility of dynamically hedging
each component of the CVA, even in adverse market conditions.
ECValue_Risk for CCR
Under the banking book risk perspective
ECValue_Risk = UL(99.97%; T0,T1) + UL(99.97%; T1,TMat)
The first term is the economic capital for unexpected default loss
over the first year. As shown in Canabarro, Picoult and Wilde
(2003), this can be derived from the unexpected default loss of an
equivalent loan portfolio that has an exposure at default per obligor
equal to alpha*EPE.
The second term is the unexpected change in the provision for
CCR default loss at the end of the first year. In contrast to a loan
portfolio, there will be three drivers of UL(99.97%; T1,TMat) for CCR:
a) Unexpected deterioration in both the credit quality of the counterparties at the end of the first year and in the expected state of
the economy over the remaining life of the portfolio.
b) Overall unexpected increase in the EPE profile, per netting set or
counterparty, from the end of the first year to the maturity of the
portfolio, conditional on the stressed state of the economy in a).
c) Unexpected changes in the correlation of a) and b).
As explained above, unexpected is measured with respect to the
expected values of the quantities today, eg, the values, for the period of time from T1 to TMat, that were used in the calculation of the
current provision for CCR default loss.
In the traditional value risk perspective, both internal economic
capital and regulatory risk-weighted assets (RWA) for CCR are calculated as a form of wholesale credit risk.
342
19/03/2010 17:13
For example, the current Basel II rules continue to treat counterparty credit risk as a form of wholesale credit risk. RWA for counterparty risk is calculated by the same method used for wholesale
loans: RWAs are specified as the product of EAD (exposure at default) and a risk weight that is a function of the standard risk parameters of a netting set: a) the PD of the counterparty, b) the LGD and
the M of the netting set or counterparty. The unique characteristic of
CCR (ie, the stochastic future exposure) is captured in the methods
used to calculate EAD and M. Implicit in the Basel II method is an
assumption that there is a provision for CCR default loss.
The traditional calculation of ECValue_Risk for CCR should include
all three components (a, b and c) described. However, the current
Basel II method does not include component b) and c) above. That
is, Basel II currently only takes component a) into account but does
not take into account the risk that at the end of the year there also
could be unexpected changes in provisions because of unexpected
changes in EPE over the remaining life.
CONFLICT BETWEEN CURRENT PRACTICE AND TRADITIONAL
MEASUREMENT OF EC AND REGULATORY CAPITAL FOR CCR
The difference between the price and value risk perspectives for
CCR is summarised in Table 14.2, above. The essential difference
in the two approaches lies in the method used to account for and
manage the counterparty credit risk premium.
Let us examine the different forms in which a conflict can arise
between the value risk perspective for economic capital and RWA,
and the practice of marking-to-market the CVA.
In this section, let us assume that the challenges of building a
rigorous method for calculating ECPrice-Risk for CVA have been overcome. As explained above, a primary hurdle is the inclusion of a
validated model and means of simulating, for each netting set,
changes in the implied correlations between each combination of:
a) the market factors affecting counterparty exposure, b) general
spreads by currency, rating and asset type, c) the counterparty specific component of spread, and d) the counterparty PD.
Mark-to-market CVA_asset only and use value risk perspective for
EC and RWA
If the CVA_asset is not hedged, then ECValue_Risk < ECPrice-Risk, as ex343
19/03/2010 17:13
plained above. This means the loss of value in the CVA_asset due
to a widening of spreads in a 2008-type systemic stress event will
tend to exceed ECValue_Risk for CCR. The fundamental reason, as explained, is the material increase in the funding risk premium and
liquidity risk premium components of credit spreads will tend to
result in a loss of market value that is greater than the unexpected
increase in lifetime default loss, even when the latter is calculated
for the same systemic stress event. Since the current version of B-II
does not include components b) and c) for ECValue_Risk, described, we
can also say that if CVA_asset is not hedged then ECBasel_II < ECValue_
< ECPrice-Risk This relation would be true if all three measures were
Risk
calculated at the same confidence level.
If the CVA_asset is hedged, other problems arise. The hedges
of CVA_asset reduce economic risk by offsetting the market factor
sensitivities and default risk of the CVA_asset.
In principle, it should be possible to include the effect of all the
CVA hedges on the calculation of ECValue_Risk for CCR. Such an analysis is not trivial, but possible in principle.
The primary problem is that a literal interpretation of the current
Basel II rules would imply that the hedges of the CVA_asset should
be included in the VaR calculation of trading book risk, without in
any way reducing the RWA for CCR calculated as wholesale loan
risk.21 This would create two errors:
o Although the Basel II calculation of RWA for CCR underestimates
the potential P&L loss of an unhedged CVA in a 2008-type systemic
stress scenario, it would tend to overstate the potential P&L loss for
a maximally hedged CVA under the same systemic stress scenario.
Including the market factor sensitivities of the hedges in VaR, but
o not the market factor sensitivities of the CVA they are hedging,
will tend to overstate the magnitude of VaR.
Consequently, a bank that consistently hedges its CVA_asset to the
maximum degree possible will reduce its P&L volatility, but would
be penalised with RWA that are materially larger than appropriate.
Mark-to-market both CVA_asset and CVA_liability and use a value
risk perspective for EC and RWA
In a bilateral CVA, the total CVA is the difference between the CVA_
344
19/03/2010 17:13
asset and the CVA_liability

CVA_total = CVA_asset CVA_liability
Consequently Equation 14.3 can be re-written as

PVOTC_Derivatives = PVrisk-free - (CVA_asset CVA_liability )
= PVrisk-free - CVA_asset + CVA_liability
(14.7)
Although we do not think it makes sense to include the CVA_liability in the calculation of a banks economic capital, a bank that must
adhere to current GAAP fair value accounting may want to generate a shadow accounting-type of EC to estimate the full impact of
the bilateral CVA on its unexpected P&L loss.
The CVA_asset, per netting set or counterparty, is sensitive to
the counterpartys CDS spread, whereas the CVA_liability is sensitive to a banks own CDS spread. A robust calculation of ECPrice_Risk
for the bilateral approach would include simulating all the components described above for the CVA_asset, as well as changes in a)
the ENE profile, per netting set, b) a banks own spread, c) the implied correlation of a banks own spread with all the market factors
that affect the ENE profile, d) the implied correlation of a banks
own spread and general market spread, etc.
Because of bilateral margin agreements and netting agreements
in the interbank market, the CVA_asset of a typical large bank will
tend to be dominated by its exposure to non-financial end users (ie,
corporates and/or sovereigns). During the 2008 crisis, particularly
after the collapse of Lehman Brothers, spreads on financial institutions widened considerably more than spreads on non-financial
firms. As a consequence of this, a bank that calculated an unhedged
bilateral CVA would have had large gains on its CVA_liability,
which would have offset some or all of the losses on its CVA_asset.
However, when spreads on financial institutions later began to narrow more than spreads on non-financial institutions, the relatively
large decrease in the magnitude of the (unhedged) CVA_liability
would have generated large P&L losses. In summary, an unhedged
bilateral CVA would introduce additional sources of P&L volatility.
Whereas the hedges on the CVA_asset can, in principle, be integrated into the calculation of ECValue_Risk for CCR, it is not possible to
345
19/03/2010 17:13
integrate the risk of changes in the CVA_liability into either ECValue_

or RWA for CCR. The reason is that ECValue_Risk and RWA for CCR
Risk
are only focused on counterparty default risk and have no means
of taking into account the risk of changes in P&L due to changes in
a banks own credit spread.
POTENTIAL SOLUTIONS TO REDUCE BREAKAGE BETWEEN
MARKING-TO-MARKET CVA AND CURRENT BASEL II
In conclusion here are two potential solutions to reduce the breakage between the practice of marking-to-market a unilateral (or bilateral) CVA and the current treatment of CCR under Basel II:
o One solution would be to allow banks that mark-to-market their
CVA to a) no longer calculate RWA for CCR as a form of wholesale credit risk, and b) incorporate the CVA and its hedges into
the calculation of RWA for the trading book. As of this writing,
this would require the calculation of VaR, stressed VaR and IRC
(incremental risk charge for downgrade and default risk) for the
CVA (unilateral or bilateral) and its hedges.
o A less complete solution would require banks to continue to calculate RWA for CCR as a form of wholesale credit risk, while
allowing banks the option of excluding from VaR both the CVA
and all hedges of the CVA. This solution would have the benefit
of avoiding the increase in RWA for trading that would occur
if the hedges, but not the underlying CVA being hedged, were
included in VaR. However, this solution still has the shortcoming of not incorporating into the calculation of RWA the benefit
of CVA hedges that were entered into. To fully implement this
solution, a bank would need to use an expanded version of an
internal model approach that would capture components b) and
c) described above for ECValue_Risk.
The author has benefited, over many years, from discussions about
counterparty credit risk with colleagues at Citi and other firms, including David E. Lawrence, David Lamb, Byron Nicas, Henry Wayne,
Eduardo Canabarro, Tom Wilde and Charles Monet. The distinction
between and the implications of the two risk perspectives, which is
discussed in this chapter, has benefited very materially from the almost
daily discussions on this topic that the author had for the last year with
346
19/03/2010 17:13
an exceptional working group at Citi, that has included Charles Monet,

Eduardo Epperlein, Alan Smillie and Jorge Sobehart, among others.
The ideas and conclusions described in this chapter, as well as any errors, are the authors
own. The author does not speak for his employer Citi.
2 The distinction between a banking book and trading book accounting perspective is well
known. The difference between a banking book and trading book risk perspective for a
loan portfolio became dramatically clear during 2008, as a consequence of firm-wide stress
tests we began running. Charles Monet made the same observations but developed a more
fundamental analysis of the difference between the two perspectives, which he labelled
price risk v. value risk (Epperlein et al 2009). A small working group at Citi consisting of
Charles, Eduardo Epperlein, Alan Smillie, Jorge Sobehart, Nic Fries and the author has had
extensive discussions on this topic over the last year.
3 See below for further explanation, as well as Canabarro in this book, and Picoult (2005).
4 This is true as of October, 2009. However, the Basel Risk Measurement and Modeling Group
(RMMG) is revisiting these rules, partly for the issue under discussion.
5 This is not just the result of a thought experiment, but also of the actual analysis of the risk
measurement of such portfolios by me and other colleagues at Citi.
6 In fact, in Basel II, for all forms of wholesale and retail credit risk, a bank needs to compare
its expected default loss over the next 12 months to its provisions for default loss. If the provisions are less than the expected default loss, then additional regulatory capital is required.
7 One component of the CVA of each netting set or counterparty is the lifetime EPE profile. In
a full market perspective, the EPE profile should be simulated using market implied volatilities and correlations of each market factor, when available, instead of their historic volatilities and correlations.
8 Credit risk and operational risk have a daily expected loss greater than zero. The expected
daily loss of a trading portfolio due to market risk is zero under the assumption that market
prices are efficiently set. A bank might not normally set aside provisions for expected general operational risk losses, and instead account for such expected losses in budget forecasting and pricing. Banks do set aside provisions for the expected loss of specific operational
risk events (eg, provision for a specific legal process).
9 A prudent measure of AFR is tangible common equity (TCE), which is equal to book equity
minus goodwill, other intangibles and preferred shares. Other broader measures of AFR can
be defined, which would include items such as preferred shares, etc. The current Basel II
definitions of Tier 1 capital and Total capital (the sum of Tier 1 and Tier 2 capital) are examples of AFR defined more broadly than TCE.
10 The author is using obligor default to mean the default by the borrower of a loan or the
issuer of a debt or equity security, and counterparty default to mean the default of a counterparty on a forward or derivative transaction.
11 To avoid pro-cyclicality, stress changes in credit spreads need to be specified in terms of
very large changes in the magnitude of spreads, independent of the current level of spreads.
12 Scaled VaR is an appropriate measure of economic capital for a portfolio in a liquid market
that has a relatively constant level of daily VaR, and for which the sign and magnitude of individual factor sensitivities frequently change. For such a portfolio, the distribution of P&L
over a one-year horizon will be specified by N independent, identical draws from the same
underlying daily P&L distribution, where N is the number of trading days in a year. Given
the assumptions, the annualised distribution of P&L will tend to be normal and economic
capital over one year can be derived by scaling the daily VaR. However, under systemic
stress conditions trading liquidity dries up, the serial correlation of daily changes in some
market rates
1
347
19/03/2010 17:13
13
14
15
16
17
18
19
20
21
becomes high, the volatilities and correlations that characterise markets in ordinary times
change and, consequently, the assumptions underlying the calculation of economic capital
from scaled VaR break down.
Realistically, there should be at least two large systemic stress scenarios, corresponding to an
inflationary and deflationary macro-environment.
Note that, in contrast, the expected change in market value is zero. Thus the incremental unexpected change in market value at any confidence level is identical to the change in market
value at that confidence level.
Epperlein, Smillie and Monet (2009). iVAST is a Citi acronym for integrated VaR and stress
testing. It is a process for integrating VaR, systemic stress tests and business-specific stress
tests into the measurement of economic capital for all forms of price risk, including the market risk of trading, the market risk of the CVA and other risks.
Under GAAP, the mark-to-market changes in the assets in an available for sale (AFS) portfolio, other than that due to permanent impairment, are not included in net income but do
affect the value of book equity through the other comprehensive income (OCI) account. This
definition of the price risk perspective implicitly assumes that the available financial resources (AFR) of a bank should be equated to its TCE. Under that definition of AFR, an AFS
portfolio would need to have its risk evaluated from a price risk perspective because changes in the market value of the AFS portfolio would affect TCE (and thus the banks AFR).
economic capital, as a reminder, is defined with respect to potential loss of value of AFR.
This analysis assumes that credit ratings are defined with respect to through-the cycle PDs.
At any point in time, the PD for the next year for a given credit rating will be conditional on
the expected state of the economy.
Some of the material in this section is identical to material written by the author for an
industry association response to a question from the Basel Risk Management and Modeling
Group (RMMG) on the relation between CVA and the Basel II treatment of RWA for counterparty credit risk. The author wrote that section of the industry association response, which
has not been published.
As explained in the prior footnote, a banks CVA_liability has a gain/loss in value when the
banks own credit spread widens/narrows. Unless a bank can monetise this gain or loss,
it has no economic reality. However, an increase in the banks expected negative exposure
(ENE) can reduce the funding cost of the banks expected positive exposure (EPE). This
should be taken into account in pricing.
General/specific wrong-way CCR occurs when there is a positive correlation between an
increase in general/counterparty specific credit spreads and exposure to that counterparty.
Right-way CCR occurs when there is a negative correlation between an increase in such
spreads and an increase in the exposure to the counterparty. The assessment of the degree
of wrong-way risk (or right-way risk) requires some knowledge of counterpartys total
business. For example, assume a bank has bought CDS credit protection on some subprime
mortgage CDO tranches from a monoline insurer. Assume the monoline had sold large volumes of such transactions without hedging its exposure with offsetting transactions. The
monoline would be very vulnerable to a systemic stress event that included a material fall in
residential housing prices and a widening of spreads. To model this wrong-way risk prior to
its actual occurrence would require detailed knowledge of the type of business the monoline
was running. In light of 2008, this is obviously not a hypothetical example.
This interpretation is arrived at because the hedges of the CVA_asset are marked-to-market,
are in the trading book and are not entered into for the purpose of hedging the banking book.
348
19/03/2010 17:13
REFERENCES
Canabarro, E., E. Picoult and T. Wilde, 2003, Analysing Counterparty Risk, Risk,
September, pp. 11722.
Epperlein, E., A. Smille and C. Monet, 2009, internal non-public Citigroup documents on iVAST.
Picoult, E., 2005, Calculating and Hedging Exposure, Credit Value Adjustment
and Economic Capital for Counterparty Credit Risk, in M. Pykhtin (ed), Counterparty Credit Risk Modelling (London: Risk Books).
Picoult, E, E. Epperlein and C. Monet, 2009, Stress Testing, Price Risk vs. Value
Risk and iVAST, three integrated, sequential presentations at the Ri$kMinds conference, Geneva.
349
19/03/2010 17:13
19/03/2010 17:13
Index
(page numbers in italic type denote tables and figures)
B
backtesting 295325, 299, 302, 303, 306,
307, 308, 309, 310, 31314, 315, 317, 318,
320, 324, 325
basic principles of 2989
hypothesis use and 299301
many portfolios, approach for
30511
single-portfolio VaR 3035
statistical power, sample size and,
3013
Bank for International Settlements 82,
99, 170, 295
Barings 68
Bear Sterns 3, 93
Black September, lessons learned from,
for counterparty credit risk 948
C
calculation of economic capital 601
implications for 546
collateral management 6380, 64
derivatives, over-the-counter
(OTC), mechanics of 668, 69
concerns over managing 723
enterprise solutions and, evolution
towards 778
enterprise-wide, need for 767
funding efficiencies and 77
over-the-counter (OTC) derivatives
and, agency structures advan-
tages 75
trading and 65
over-the-counter (OTC), evolution
of 656
proposed measures to improve 789

in repo 689
buy/sellback 69
classic 69
limitations of bilateral collateral
management in securities
lending and 712, 72
securities lending and 702
tri-party, for repo and securities
lending 735, 74
collateralised derivatives 199216
default-free, valuation of 2004
generalised valuation of 199216,
200, 205, 206, 207, 208, 209, 210,
211, 212
methodology 200
assumptions 200
default-free 2004
numerical results (Gaussian process)
204
bilateral case 204
unilateral case 206
numerical results (interest rate swap)
20910
bilateral case 209
unilateral case 211
conservative early-termination provi sions 100
counterparty credit risk (CCR):
backtesting and 295325, 299, 302,
303, 306, 307, 308, 309, 310,
31314, 315, 317, 318, 320, 324,
325
basic principles of 2989
hypothesis use and 299301
many portfolios, approach for
351
19/03/2010 17:13
30511
single-portfolio VaR 3035
statistical power, sample size
and, 3013
Basel II internal ratings-based (IRB)
approach for 24950
Black September and, lessons from
948
collateral and, effective means to
resolve call disputes concern ing 95
collateral and, protection of, from
insolvency 967
collateral and, right call for 945
collateral delivery requirements
and 95
early-warning triggers and 97
flexible close-out mechanics and
1001
capital and alpha, computing 2508
alternative ordering methods
2535
case study concerning 25985
credit portfolio model 2512
general methodology 2523
market factors and, ordering
scenario using 2556
collateralisation to mitigate and
control, see credit exposure: col lateralised
computing and stress-testing capital
concerning 24590; see also stress
testing
counterparty exposures and 2479,
248
credit default swaps (CDS) and:
contingent (CCDS) 16583
counterparty risk in 13945, 140,
141, 143, 144
derivative products and, see credit
derivative products
economic capital for, from two
perspectives 32746, 331
breakage between marking-to-
market CVA and current Basel II

and, potential solutions to reduce
346
comparison 33940
current versus traditional mea surement of 3436
loan versus bond 32831
full-trading-book treatment of 3334
hedging, see counterparty credit
risk (CCR): pricing and hedging
in legal documentation, industry
efforts to address 8294
Bear Stearns near-failure and
934
CRMPG report 901
derivative, leverage and hedge
fund tsunami and 92
G10 study 88
G30 study 867
ISDAs collateral review 8990
ISDAs standardised credit sup port documentation 87
Long Term Capital Management
near-collapse and 878
Master Agreement and, creation
of 835
Master Agreement and, ISDA
republishes 912
netting, basis for 835
PWG report 889
managing, in over-the-counter
(OTC) derivatives market 1014
clearing 1023
mandatory margin/collateral
and capital/prudential require ments 1034
measuring 16670, 169
credit valuation adjustment 168
economic capital and 1689
expected and peak positive expo sure profiles 1678
exposure at default 1667
metrics 167, 181, 1812
risk-weighted assets and 16970
352
19/03/2010 17:13
INDEX
mitigation 1705, 171

caveat utilitor and 1712
EAD substitution and 1745
exponential growth and 170
market enhancements 171
reference translation EAD distri bution table and 1734
mitigation of, through documenta tion today 98101
broader enforceability and 989
conservative early-termination
provisions and 100
greater use of collateral and
99100
pricing and hedging 10934, 174
by banks 1223
counterparty risk systems and
1312
credit valuation adjustments
(CVAs) and 10933 passim, 114; see also
main entry
definition of 110
economic capital and 131
fair value and 11113
friction costs and 1234
funding benefit versus credit risk
pricing 11516
hedging: economics of 182
mark-to-market discipline and
117
netting, financial collateral and
1279
other mitigants and 12930
out-of-the-money risks and 1256
price signals, competitiveness
and 11617
regulatory capital and 1301
stress tests for 1267
wrong-way risks and 1245
stress testing and, see main entry
systematic (SCCR) 115
contagion and containment 78
definition of 13
dynamics of 57
effects of limiting 1415

limits for 1214
managing 1012
measuring 810
monitoring 10
theory of 35
traditional treatment of 3323
counterparty exposure conditional on
default:
calculation of 5162
calculations 601
conditional simulation and 578
conditioning 5860
credit transitions and, extension
to include 612
simulation framework and 567
wrong-way, right-way risk and
534
Counterparty Risk Management Policy
Group (CRMPG) 90, 924 passim
counterparty risk mitigation, US legal
framework for 81104
contractual, examination of 82
credit default swaps (CDS):
contingent (CCDS) 16583
effectiveness of 175
forward-starting 177, 1778
pricing 1823
risk practice concerning 17582
short-maturity 178, 1789
standard 176, 1767
threshold 179, 17980
toolkit for 180, 180
counterparty risk in 13945, 140,
141, 143, 144
credit derivative products:
counterparty risk of 13762
in credit default swaps (CDS)
13945, 140, 141, 143, 144
impact of, across capital structure
149, 150
joint defaults, quantifying risk of
1389, 139
monolines 15561, 159, 161
353
19/03/2010 17:13
super senior tranches of 1515, 151,

153, 155
tranches of, counterparty risk in
14551, 146, 148
credit exposure:
collateralised 1749, 25; see also col lateral management
expected exposure: forward starting swap 427
expected exposure: immediately
starting swap 479
minimum transfer amount ap proximation 234
modelling 234
nave model 247, 25
potential future exposure:
forward-starting swap 3740
potential future exposure: im mediately starting swap 40
profile examples: forward-start ing swap 356
profile examples: immediately
starting swap 36
profiles 3549, 36, 37, 389, 412,
434, 478
margin period of risk and 279, 49
bilateral agreement 315
unilateral agreement 2831
modelling 203
non-collateralised 212
simulation framework 201
Credit Support Annex (CSA) 1718,
678, 213
credit valuation adjustments (CVAs) 53,
10933 passim, 114
calculation, full model 113
calculation, main modules of 132
calculation, simple model 11011
calculation, simplified model used
by banks 11415
desks 11719
hedging 11920, 1213
marking-to-market CVA and cur rent Basel II and, potential solu
tions to reduce
breakage between 346
risk factors that drive 1201
stress tests for 1267
zero case 115
D
derivatives, over-the-counter (OTC)
756
bilateral, mechanics of 667
collateral agency structures offer
similar advantages for 75
collateral use in 65; see also collater alised derivatives
collateralised, see collateralised
derivatives
evolution of 656
funding and, see funding
G30 study and 867
industry response to concerns over
86
legal approaches to managing,
reforms impact on 1014
clearing 1023
and capital/prudential
requirements 1034
legal nature of, in US 812
in US, legal nature of 812
derivatives trading, collateral use in 65
E
early-warning triggers 978
economic capital:
calculation of 601
for counterparty credit risk, from
two perspectives 32746, 331
breakage between marking-to market CVA and current
Basel II and, potential
solutions to reduce 346
comparison 33940
current versus traditional mea-
354
19/03/2010 17:13
INDEX
surement of 3436
loan versus bond 32831
definition of 3356, 335
implications for calculation of 546
measuring counterparty credit risk
and 1689
price-risk perspective and 3367
pricing and hedging counterparty
risk and 131
value-risk perspective and 3389
enterprise solutions, evolution towards
778
ETrade bank 3
F
Fannie Mae 3
flexible close-out mechanics 1001
Freddie Mac 3
funding, benefit and cost of 18597,
186, 187
asset CVA, liability CVA and 18793
calculations of 18993
default risk and 1878
funding and 1889
asset CVA, liability CVA and, ex amples of 18993, 191, 192, 193
business significance 185
collateral and 194
cost reduction and 196
debit valuation adjustment and 197
derivatives pricing and 1945
joint market and credit states 1856
modelling and trading of 1956
netting and 1934
overnight index swap discounting
and 1945
trading and modelling of 1956
G
Global Derivatives Study Group
(GDSG) 867
I
implications for calculation of economic
capital 546
International Swaps and Derivatives
Association (ISDA), 17, 63, 79, 835
passim, 912, 93, 97, 99, 1001, 166,
171
collateral review of 8990
credit-support documentation pub lished by 87, 95
Guidelines for Collateral Practitio ners issued by 66
J
joint defaults, quantifying risk of 138
JP Morgan 93
L
legal documentation, industry efforts
to address counterparty credit risk
in 8294
Lehman Brothers 23, 93, 94, 96, 97, 99,
101, 123, 165, 170, 345
Long Term Capital Management 68,
878, 89, 100, 109
M
margin period of risk 279, 49
included in collateral model 28
minimum transfer amount approxima tion 234
monolines 112, 15561, 159, 161
N
nave collateral model 247, 25
bilateral margin agreement and
256
unilateral margin agreement and
245
Northern Rock 3
O
over-the-counter (OTC) collateral man agement, evolution of 656
756
355
19/03/2010 17:13
bilateral, mechanics of 667

collateral agency structures offer
similar advantages for 75
collateral use in 65; see also collater alised derivatives
collateralised, see collateralised
derivatives
evolution of 656
funding and, see funding
G30 study and 867
industry response to concerns over
86
legal approaches to managing,
reforms impact on 1014
clearing 1023
and capital/prudential
requirements 1034
in US, legal nature of 812
R
regulatory capital, trading book versus
banking book 1301
right-way/wrong-way credit risk 534
S
securities lending 702, 72
collateral use in 70, 72
evolution of 701
limitations of bilateral collateral
management in repo and 71
mechanics of 71
tri-party collateral management for
repo and 735, 74
stress testing:
algorithmic 2225
automatic scenario generation
(ASG) and 22542
model, to analyse risks of loan
portfolio 23842
model, for interest rate risks
2308
computing and 24590
counterparty credit risk capital and

24590, 258
alpha and 2508
alternative ordering methods
2535
background to 24750
Basel II internal ratings-based
(IRB) approach and 24950
case study concerning 25985,
260, 261, 262, 263, 265, 267,
269, 270, 271, 272, 274, 275,
2768, 279, 280, 281, 283, 285
credit portfolio model 2512
exposure 2479, 248
general methodology 2523
market and credit factors con cerning, empirical correla tions between 2868, 288,
28990
multi-step default model 256
ordering scenario using market
factors 2556
scenario analysis and 22143, 2267,
2289, 231, 232, 233, 2345, 2367,
239, 240, 241
systematic counterparty credit risk
(SCCR) 115
contagion and containment 78
definition of 13
dynamics of 57
effects of limiting 1415
limits for 1214
managing 1012
measuring 810
monitoring 10
theory of 35
U
US, counterparty risk mitigation in
81104
W
wrong-way/right-way credit risk 534
356
19/03/2010 17:13

Counterparty Credit Risk

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Counterparty Credit Risk

Uploaded by

Copyright:

Available Formats

Governments and supervisors around the

world are focusing on the interconnectedness

Measurement and management

Chapters from leading authorities

Calculating counterparty exposure

Edited by Eduardo Canabarro