Rating Based Modeling of Credit Risk: Theory and Application of Migration Matrices

Rating Based Modeling of Credit Risk

Theory and Application of Migration Matrices

Trueck Stefan

Rachev Svetlozar T.

Brief Table of Contents

Copyright

Dedication

Preface

Chapter 1. Introduction

Chapter 2. Rating and Scoring Techniques

Chapter 3. The New Basel Capital Accord

Chapter 4. Rating Based Modeling

Chapter 5. Migration Matrices and the Markov Chain Approach

Chapter 6. Stability of Credit Migrations

Chapter 7. Measures for Comparison of Transition Matrices

Chapter 8. Real-World and Risk-Neutral Transition Matrices

Chapter 9. Conditional Credit Migrations

Chapter 10. Dependence Modeling and Credit Migrations

Chapter 11. Credit Derivatives

Table of Contents

Copyright

Dedication

Preface

Chapter 1. Introduction

1.1. Motivation

1.2. Structural and Reduced Form Models

1.3. Basel II, Scoring Techniques, and Internal Rating Systems

1.4. Rating Based Modeling and the Pricing of Bonds

1.5. Stability of Transition Matrices, Conditional Migrations, and Dependence

1.6. Credit Derivative Pricing

1.7. Chapter Outline

Chapter 2. Rating and Scoring Techniques

2.1. Ratings Agencies, Rating Processes, and Factors

2.1.1. The Rating Process

2.1.2. Credit Rating Factors

2.1.3. Types of Rating Systems

2.2. Scoring Systems

2.3. Discriminant Analysis

2.4. Logit and Probit Models

2.4.1. Logit Models

2.4.2. Probit Models

2.5. Model Evaluation: Methods and Difficulties

2.5.1. Model Performance and Benchmarking

2.5.2. Model Accuracy, Type I and II Errors

Chapter 3. The New Basel Capital Accord

3.1. Overview

3.1.1. The First Pillar—Minimum Capital Requirement

3.1.2. The Second Pillar—Supervisory Review Process

3.1.3. The Third Pillar—Market Discipline

3.2. The Standardized Approach

3.2.1. Risk Weights for Sovereigns and for Banks

3.2.2. Risk Weights for Corporates

3.2.3. Maturity

3.2.4. Credit Risk Mitigation

3.3. The Internal Ratings Based Approach

3.3.1. Key Elements and Risk Components

3.3.2. Derivation of the Benchmark Risk Weight Function

3.3.3. Asset Correlation

3.3.4. The Maturity Adjustment

3.3.5. Expected, Unexpected Losses and the Required Capital

3.4. Summary

Chapter 4. Rating Based Modeling

4.1. Introduction

4.2. Reduced Form and Intensity Models

4.2.1. The Model by Jarrow and Turnbull (1995)

4.2.2. The Model Suggested by Madan and Ünal (1998)

4.2.3. The Model Suggested by Lando (1998)

4.2.4. The Model of Duffie and Singleton (1999)

4.3. The CreditMetrics Model

4.4. The CreditRisk+ Model

4.4.1. The First Modeling Approach

4.4.2. Modeling Severities

4.4.3. Shortcomings of the First Modeling Approach

4.4.4. Extensions in the CR+ Model

4.4.5. Allocating Obligors to One of Several Factors

4.4.6. The pgf for the Number of Defaults

4.4.7. The pgf for the Default Loss Distribution

4.4.8. Generalization of Obligor Allocation

4.4.9. The Default Loss Distribution

Chapter 5. Migration Matrices and the Markov Chain Approach

5.1. The Markov Chain Approach

5.1.1. Generator Matrices

5.2. Discrete Versus Continuous-Time Modeling

5.2.1. Some Conditions for the Existence of a Valid Generator

5.3. Approximation of Generator Matrices

5.3.1. The Method Proposed by Jarrow, Lando, and Turnbull (1997)

5.3.2. Methods Suggested by Israel, Rosenthal, and Wei (2000)

5.4. Simulating Credit Migrations

5.4.1. Time-Discrete Case

5.4.2. Time-Continuous Case

5.4.3. Nonparametric Approach

Chapter 6. Stability of Credit Migrations

6.1. Credit Migrations and the Business Cycle

6.2. The Markov Assumptions and Rating Drifts

6.2.1. Likelihood Ratio Tests

6.2.2. Rating Drift

6.2.3. An Empirical Study

6.3. Time Homogeneity of Migration Matrices

6.3.1. Tests Using the Chi-Square Distance

6.3.2. Eigenvalues and Eigenvectors

6.4. Migration Behavior and Effects on Credit VaR

6.5. Stability of Probability of Default Estimates

Chapter 7. Measures for Comparison of Transition Matrices

7.1. Classical Matrix Norms

7.2. Indices Based on Eigenvalues and Eigenvectors

7.3. Risk-Adjusted Difference Indices

7.3.1. The Direction of the Transition (DIR)

7.3.2. Transition to a Default or Nondefault State (TD)

7.3.3. The Probability Mass of the Cell (PM)

7.3.4. Migration Distance (MD)

7.3.5. Devising a Distance Measure

7.3.6. Difference Indices for the Exemplary Matrices

7.4. Summary

Chapter 8. Real-World and Risk-Neutral Transition Matrices

8.1. The JLT Model

8.2. Adjustments Based on the Discrete-Time Transition Matrix

8.3. Adjustments Based on the Generator Matrix

8.3.1. Modifying Default Intensities

8.3.2. Modifying the Rows of the Generator Matrix

8.3.3. Modifying Eigenvalues of the Transition Probability Matrix

8.4. An Adjustment Technique Based on Economic Theory

8.5. Risk-Neutral Migration Matrices and Pricing

Chapter 9. Conditional Credit Migrations

9.1. Overview

9.2. The CreditPortfolioView Approach

9.3. Adjustment Based on Factor Model Representations

9.3.1. Deriving an Index for the Credit Cycle

9.3.2. Conditioning of the Migration Matrix

9.3.3. A Multifactor Model Extension

9.4. Other Methods

9.5. An Empirical Study on Different Forecasting Methods

9.5.1. Forecasts Using the Factor Model Approach

9.5.2. Forecasts Using Numerical Adjustment Methods

9.5.3. Regression Models

9.5.4. In-Sample Results

9.5.5. Out-of-Sample Forecasts

Chapter 10. Dependence Modeling and Credit Migrations

10.1. Introduction

10.1.1. Independence

10.1.2. Dependence

10.2. Capturing the Structure of Dependence

10.2.1. Under General Multivariate Distributions

10.3. Copulas

10.3.1. Examples of Copulas

10.3.2. Properties of Copulas

10.3.3. Constructing Multivariate Distributions with Copulas

10.4. Modeling Dependent Defaults

10.5. Modeling Dependent Migrations

10.5.1. Dependence Based on a Credit Cycle Index

10.5.2. Dependence Based on Individual Transitions

10.5.3. Approaches Using Copulas

10.6. An Empirical Study on Dependent Migrations

10.6.1. Distribution of Defaults

10.6.2. The Distribution of Rating Changes

Chapter 11. Credit Derivatives

11.1. Introduction

11.1.1. Types of Credit Derivatives

11.1.2. Collateralized Debt Obligations (CDO)

11.2. Pricing Single-Named Credit Derivatives

11.3. Modeling and Pricing of Collateralized Debt Obligations and Basket Credit Derivatives

11.3.1. Estimation of Macroeconomic Risk Factors

11.3.2. Modeling of Conditional Migrations and Recovery Rates

11.3.3. Some Empirical Results

11.4. Pricing Step-Up Bonds

11.4.1. Step-Up Bonds

11.4.2. Pricing of Step-Up Bonds

Copyright

Academic Press is an imprint of Elsevier

30 Corporate Drive, Suite 400, Burlington, MA 01803, USA

525 B Street, Suite 1900, San Diego, California 92101-4495, USA

84 Theobald’s Road, London WC1X 8RR, UK

Copyright © 2009 by Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher.

Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone: (+44) 1865 843830, fax: (+44) 1865 853333, E-mail: permissions@elsevier.com. You may also complete your request online via the Elsevier homepage (http://www.elsevier.com), by selecting Support & Contact then Copyright and Permission and then Obtaining Permissions.

Library of Congress Cataloging-in-Publication Data

Application submitted

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

ISBN: 978-0-12-373683-3

For information on all Academic Press publications visit our Web site at: http://www.elsevierdirect.com

Printed in the United States of America

08 09 10 9 8 7 6 5 4 3 2 1

Dedication

To my parents and Prasheela (S.T.)

To Svetlozar Todorov Iotov (S.T.R)

Preface

Credit risk has become one of the most intensely studied topics in quantitative finance in the last decade. A large number of books on the topic have been published in recent years, while on the excellent home-page that is maintained by Greg Gupton there are more than 1200 downloadable working papers related to credit risk. The increased interest in modeling and management of credit risk in academia seems only to have started in the mid-1990s. However, due to the various issues involved, including the ability to effectively apply quantitative modeling tools and techniques and the dramatic rise of credit derivatives, it has become one of the major fields research in finance literature.

As a consequence of an increasingly complex and competitive financial environment, adequate risk management strategies quantitative modeling know-how and the ability to effectively apply this expertise and techniques. Also, with the revision of the Basel Capital Accord, various credit risk models have been analyzed with respect to their feasibility, and a significant focus has been on good risk-management practices with respect to credit risk. Another consequence of Basel II is that most financial institutions will have to develop internal models to adequately determine the risk arising from their credit exposures. It can therefore be expected that in particular the use and application of rating based models for credit risk will be increasing further.

On the other hand, it has to be acknowledged that rating agencies are the center of the subprime mortgage crisis, as they failed to provide adequate ratings for many diverse products in the credit and credit derivative markets like mortgage bonds, asset backed securities, commercial papers, collateralized debt obligations, and derivative products for companies and also for financial institutions. Despite some deficiencies of the current credit rating structure—recommendations for their improvements are thoroughly analyzed in Crouhy et al. (2008) but are beyond the scope of this book—overall, rating based models have evolved as an industry standard. Therefore, credit ratings will remain one of the most important variables when it comes to measurement and management of credit risk.

The literature on modeling and managing credit risk and credit derivatives has been widely extended in recent years; other books in the area include the excellent treatments by Ammann (2002), Arvanitis and Gregory (2001), Bielecki and Rutkowski (2002), Bluhm et al. (2003), Bluhm and Overbeck (2007b), Cossin and Pirotte (2001), Duffie and Singleton (2003), Fabozzi (2006a,b), Lando (2004), Saunders and Allen (2002), and Schönbucher (2003), just to mention a few. However, in our opinion, so far there has been no book on credit risk management mainly focusing on the use of transition matrices, which, while popular in academia, is even more widely used in industry. We hope that this book provides a helpful survey on the theory and application of transition matrices for credit risk management, including most of the central issues like estimation techniques, stability and comparison of rating transitions, VaR simulation, adjustment and forecasting migration matrices, corporate-yield curve dynamics, dependent migrations, and the modeling and pricing of credit derivatives. While the aim is mainly to provide a review of the existing literature and techniques, a variety of very recent results and new work also has been incorporated into the book. We tried to keep the presentation thorough but also accessible, such that most of the chapters do not require a very technical background and should be useful for academics, regulators, risk managers, practitioners, and even students who require an introduction or a more extensive and advanced overview of the topic. The large number of applications and numerical examples should also help the reader to better identify and follow the important implementation issues of the described models.

In the process of writing this book, we received a lot of help from various people in both academia and industry. First of all, we highly appreciated feedback and comments on the manuscript by many colleagues and friends. We would also like to thank various master, research, and PhD students who supplied corrections or contributed their work to several of the chapters. In particular, we are grateful to Arne Benzin, Alexander Breusch, Jens Deidersen, Stefan Harpaintner, Jan Henneke, Matthias Laub, Nicole Lehnert, Andreas Lorenz, Christian Menn, Jingyuan Meng, Emrah Özturkmen, Peter Niebling, Jochen Peppel, Christian Schmieder, Robert Soukup, Martin Sttzel, Stoyan Stoyanov, and Wenju Tian for their contributions. Finally, we would like to thank Roxana Boboc and Stacey Walker at Elsevier for their remarkable help and patience throughout the process of manuscript delivery.

Stefan Trueck and Svetlozar T. Rachev

Sydney and Karlsruhe, August 2008

Chapter 1. Introduction - Credit Risk Modeling, Ratings, and Migration Matrices

1.1. Motivation

The aim of this book is to provide a review on theory and application of migration matrices in rating based credit risk models. In the last decade, rating based models in credit risk management have become very popular. These systems use the rating of a company as the decisive variable and not—like the formerly used structural models the value of the firm—when it comes to evaluate the default risk of a bond or loan. The popularity is due to the straightforwardness of the approach but also to the new Capital Accord (Basel II) of the Basel Committee on Banking Supervision (2001), a regulatory body under the Bank of International Settlements (BIS). Basel II allows banks to base their capital requirements on internal as well as external rating systems. Thus, sophisticated credit risk models are being developed or demanded by banks to assess the risk of their credit portfolio better by recognizing the different underlying sources of risk. As a consequence, default probabilities for certain rating categories but also the probabilities for moving from one rating state to another are important issues in such models for risk management and pricing. Systematic changes in migration matrices have substantial effects on credit Value-at-Risk (VaR) of a portfolio but also on prices of credit derivatives like Collaterized Debt Obligations (CDOs). Therefore, rating transition matrices are of particular interest for determining the economic capital or figures like expected loss and VaR for credit portfolios, but can also be helpful as it comes to the pricing of more complex products in the credit industry.

This book is in our opinion the first manuscript with a main focus in particular on issues arising from the use of transition matrices in modeling of credit risk. It aims to provide an up-to-date reference to the central problems of the field like rating based modeling, estimation techniques, stability and comparison of rating transitions, VaR simulation, adjustment and forecasting migration matrices, corporate-yield curve dynamics, dependent defaults and migrations, and finally credit derivatives modeling and pricing. Hereby, most of the techniques and issues discussed will be illustrated by simplified numerical examples that we hope will be helpful to the reader. The following sections provide a quick overview of most of the issues, problems, and applications that will be outlined in more detail in the individual chapters.

1.2. Structural and Reduced Form Models

This book is mainly concerned with the use of rating based models for credit migrations. These models have seen a significant rise in popularity only since the 1990s. In earlier approaches like the classical structural models introduced by Merton (1974), usually a stochastic process is used to describe the asset value V of the issuing firm

where μ and σ are the drift rate and volatility of the assets, and W(t) is a standard Wiener process. The firm value models then price the bond as contingent claims on the asset. Literature describes the event of default when the asset value drops below a certain barrier. There are several model extensions, e.g., by Longstaff and Schwartz (1995) or Zhou (1997), including stochastic interest rates or jump diffusion processes. However, one feature of all models of this class is that they Model credit risk based on assuming a stochastic process for the value of the firm and the term structure of interest rates. Clearly the problem is to determine the value and volatility of the firm’s assets and to model the stochastic process driving the value of the firm adequately. Unfortunately using structural models, especially short-term credit spreads, are generally underestimated due to default probabilities close to zero estimated by the models. The fact that both drift rate and volatility of the firm’s assets may also be dependent on the future situation of the whole economy is not considered.

The second major class of models—the reduced form models—does not condition default explicitly on the value of the firm. They are more general than structural models and assume that an exogenous random variable drives default and that the probability of default (PD) over any time interval is non-zero. An important input to determine the default probability and the price of a bond is the rating of the company. Thus, to determine the risk of a credit portfolio of rated issuers one generally has to consider historical average defaults and transition probabilities for current rating classes. The reduced form approach was first introduced by Fons (1994) and then extended by several authors, including Jarrow et al. (1997) and Duffie and Singleton (1999). Quite often in reduced form approaches the migration from one rating state to another is modeled using a Markov chain model with a migration matrix governing the changes from one rating state to another. An exemplary transition matrix is given in Table 1.1.

TABLE 1.1. Average One-Year Transition Matrix of Moody’s Corporate Bond Ratings for the Period 1982–2001

Besides the fact that they allow for realistic short-term credit spreads, reduced form models also give great flexibility in specifying the source of default. We will now give a brief outlook on several issues that arise when migration matrices are applied in rating based credit modeling.

1.3. Basel II, Scoring Techniques, and Internal Rating Systems

As mentioned before, due to the new Basel Capital Accord (Basel II) most of the international operating banks may determine their regulatory capital based on an internal rating system (Basel Committee on Banking Supervision, 2001). As a consequence, a high fraction of these banks will have ratings and default probabilities for all loans and bonds in their credit portfolio. Therefore, Chapters 2 and 3 of this book will be dedicated to the new Basel Capital Accord, rating agencies, and their methods and a review on scoring techniques to derive a rating. Regarding Basel II, the focus will be set on the internal ratings based (IRB) approach where the banks are allowed to use the results of their own internal rating systems. Consequently, it is of importance to provide a summary on the rating process of a bank or the major rating agencies. As will be illustrated in Chapter 6, internal and external rating systems may show quite a different behavior in terms of stability of ratings, rating drifts, and time homogeneity.

While Weber et al. (1998) were the first to provide a comparative study on the rating and migration behavior of four major German banks, recently more focus has been set on analysing rating and transition behavior also in internal rating systems (Bank of Japan, 2005; European Central Bank, 2004). Recent publications include, for example, Engelmann et al. (2003), Araten et al. (2004), Basel Committee on Banking Supervision (2005), and Jacobson et al (2006). Hereby, Engelmann et al. (2003) and the Basel Committee on Banking Supervision (2005) are more concerned with the validation, respectively, classification of internal rating systems. Araten et al. (2004) discuss issues in evaluating banks’ internal ratings of borrowers comparing the ex-post discrimination power of an internal and external rating system. Jacobson et al. (2006) investigate internal rating systems and differences between the implied loss distributions of banks with equal regulatory risk profiles. We provide different technologies to compare rating systems and estimated migration matrices in Chapters 2 and 7.

Another problem for internal rating systems arises when a continuous-time approach is chosen for modeling credit migrations. Since for bank loans, balance sheet data or rating changes are reported only once a year, there is no information on the exact time of rating changes available. While discrete migration matrices can be transformed into a continuous-time approach, Israel et al. (2000) show that for several cases of discrete transition matrices there is no true or valid generator. In this case, only an approximation of the continuous-time transition matrix can be chosen. Possible approximation techniques can be found in Jarrow et al. (1997), Kreinin and Sidelnikova (2001), or Israel et al. (2000) and will be discussed in Chapter 5.

1.4. Rating Based Modeling and the Pricing of Bonds

A quite important application of migration matrices is also their use for determining the term structure of credit risk. In 1994, Fons (1994) developed a reduced form model to derive credit spreads using historical default rates and a recovery rate estimate. He illustrated that the term structure of credit risk, i.e., the behavior of credit spreads as maturity varies, depends on the issuer’s credit quality, i.e., its rating. For bonds rated investment grade, the term structures of credit risk have an upward sloping structure. The spread between the promised yield-to-maturity of a defaultable bond and a default-free bond of the same maturity widens as the maturity increases. On the other hand, speculative grade rated bonds behave in the opposite way: the term structures of the credit risk have a downward-sloping structure. Fons (1994) was able to provide a link between the rating of a company and observed credit spreads in the market.

However, obviously not only the worst case event of default has influence on the price of a bond, but also a change in the rating of a company can affect prices of the issued bond. Therefore, with CreditMetrics JP Morgan provides a framework for quantifying credit risk in portfolios using historical transition matrices (Gupton et al., 1997). Further, refining the Fons model, Jarrow et al. (1997) introduced a discrete-time Markovian model to estimate changes in the price of loans and bonds. Both approaches incorporate possible rating upgrades, stable ratings, and rating downgrades in the reduced form approach. Hereby, for determining the price of credit risk, both historical default rates and transition matrices are used. The model of Jarrow et al. (1997) is still considered one of the most important approaches as it comes to the pricing of bonds or credit derivatives and will be described in more detail in Chapter 8.

Both the CreditMetrics framework and Markov chain approach heavily rely on the use of adequate credit migration matrices it will be illustrated in Chapters 4 and 5. Further, the application of migration matrices for deriving cumulative default probabilities and the pricing of credit derivatives will be illustrated in Chapter 11.

1.5. Stability of Transition Matrices, Conditional Migrations, and Dependence

As mentioned before, historical transition matrices can be used as an input for estimating portfolio loss distributions and credit VaR figures. Unfortunately, transition matrices cannot be considered to be constant over a longer time period; see e.g., Allen and Saunders (2003) for an extensive review on cyclical effects in modeling credit risk measurement. Further, migrations of loans in internal bank portfolios may behave differently than the transition matrices provided by major rating agencies like Moody’s or Standard & Poor’s would suggest (Krüger et al., 2005; Weber et al., 1998). Nickell et al. (2000) show that there is quite a big difference between transition matrices during an expansion of the economy and a recession. The results are confirmed by Bangia et al. (2002) who suggest that for risk management purposes it might be interesting not only to simulate the term structure of defaults but to design stress test scenarios by the observed behavior of default and transition matrices through the cycle. Jafry and Schuermann (2004) investigate the mobility in migration behavior using 20 years of Standard & Poor’s transition matrices and find large deviations through time. Kadam and Lenk (2008) report significant heterogeneity in default intensity, migration volatility, and transition probabilities depending on country and industry effects. Finally, Trueck and Rachev (2005) show that the effect of different migration behavior on exemplary credit portfolios may lead to substantial changes in expected losses, credit VaR, or confidence sets for probabilities of default (PDs). During a recession period of the economy the VaR for one and the same credit portfolio can be up to eight times higher than during an expansion of the economy.

As a consequence, following Bangia et al. (2002), it seems necessary to extend transition matrix application to a conditional perspective using additional information on the economy or even forecast transition matrices using revealed dependencies on macroeconomic indices and interest rates. Based on the cyclical behavior of migration, the literature provides some approaches to adjust, re-estimate, or change migration matrices according to some model for macroeconomic variables or observed empirical prices. Different approaches suggest conditioning the matrix based on macroeconomic variables or forecasts that will affect future credit migrations. The first model developed to explicitly link business cycles to rating transitions was in the 1997 CreditPortfolioView (CPV) by Wilson (1997a, b). Kim (1999) develops a univariate model whereby ratings respond to business cycle shifts. The model is extended to a multifactor credit migration model by Wei (2003) while Cowell et al. (2007) extend the model by replacing the normal with an α-stable distribution for modeling the risk factors. Nickell et al. (2000) propose an ordered probit model which permits migration matrices to be conditioned on the industry, the country domicile, and the business cycle. Finally, Bangia et al. (2002) provide a Markov switching model, separating the economy into two regimes. For each state of the economy—expansion and contraction—a transition matrix is estimated such that conditional future migrations can be simulated based on the state of the economy.

To