CREDIT RISK MODELING

Tomasz R. Bielecki
Department of Applied Mathematics
Illinois Institute of Technology
Chicago, IL 60616, USA

Monique Jeanblanc
Département de Mathématiques
Université d’Évry Val d’Essonne
91025 Évry Cedex, France

Marek Rutkowski
School of Mathematics and Statistics
University of New South Wales
Sydney, NSW 2052, Australia

Center for the Study of Finance and Insurance

Osaka University, Osaka, Japan
2
Contents

1 Structural Approach 9
1.1 Notation and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.1 Defaultable Claims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.2 Risk-Neutral Valuation Formula . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.1.3 Defaultable Zero-Coupon Bond . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Merton’s Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 First Passage Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Distribution of the First Passage Time . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 Joint Distribution of Y and τ . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4 Black and Cox Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 Bond Valuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4.2 Black and Cox Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.3 Corporate Coupon Bond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4.4 Optimal Capital Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.5 Extensions of the Black and Cox Model . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5.1 Stochastic Interest Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.6 Random Barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.6.1 Independent Barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2 Hazard Function Approach 35

2.1 Elementary Market Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.1.1 Hazard Function and Hazard Rate . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.2 Defaultable Bond with Recovery at Maturity . . . . . . . . . . . . . . . . . . 37
2.1.3 Defaultable Bond with Recovery at Default . . . . . . . . . . . . . . . . . . . 40
2.2 Martingale Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.1 Conditional Expectations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.2 Martingales Associated with Default Time . . . . . . . . . . . . . . . . . . . . 42
2.2.3 Predictable Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . 46
2.2.4 Girsanov’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.2.5 Range of Arbitrage Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.2.6 Implied Risk-Neutral Default Intensity . . . . . . . . . . . . . . . . . . . . . . 51
2.2.7 Price Dynamics of Simple Defaultable Claims . . . . . . . . . . . . . . . . . . 52

3
4 CONTENTS

2.3 Pricing of General Defaultable Claims . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.3.1 Buy-and-Hold Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.3.2 Spot Martingale Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3.3 Self-Financing Trading Strategies . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.3.4 Martingale Properties of Arbitrage Prices . . . . . . . . . . . . . . . . . . . . 59
2.4 Single Name Credit Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4.1 Stylized Credit Default Swap . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4.2 Market CDS Spread . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4.3 Price Dynamics of a CDS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4.4 Replication of a Defaultable Claim . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Basket Credit Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.5.1 First-to-Default Intensities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.5.2 First-to-Default Representation Theorem . . . . . . . . . . . . . . . . . . . . 68
2.5.3 Price Dynamics of Credit Default Swaps . . . . . . . . . . . . . . . . . . . . . 70
2.5.4 Valuation of a First-to-Default Claim . . . . . . . . . . . . . . . . . . . . . . 73
2.5.5 Replication of a First-to-Default Claim . . . . . . . . . . . . . . . . . . . . . . 74
2.5.6 Conditional Default Distributions . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5.7 Recursive Valuation of a Basket Claim . . . . . . . . . . . . . . . . . . . . . . 77
2.5.8 Recursive Replication of a Basket Claim . . . . . . . . . . . . . . . . . . . . . 80
2.6 Applications to Copula-Based Models . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.6.1 Independent Default Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.6.2 Archimedean Copulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

3 Hazard Process Approach 87

3.1 Hazard Process and its Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1.1 Conditional Expectations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.1.2 Hazard Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.1.3 Valuation of Defaultable Claims . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.1.4 Defaultable Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.1.5 Martingales Associated with Default Time . . . . . . . . . . . . . . . . . . . . 93
3.1.6 F-Intensity of Default Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.1.7 Reduction of the Reference Filtration . . . . . . . . . . . . . . . . . . . . . . 97
3.1.8 Enlargement of Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2 Hypothesis (H) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2.1 Equivalent Forms of Hypothesis (H) . . . . . . . . . . . . . . . . . . . . . . . 99
3.2.2 Canonical Construction of a Default Time . . . . . . . . . . . . . . . . . . . . 101
3.2.3 Stochastic Barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.3 Predictable Representation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.4 Girsanov’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.5 Invariance of Hypothesis (H) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.5.1 Case of the Brownian Filtration . . . . . . . . . . . . . . . . . . . . . . . . . . 107
CONTENTS 5

3.5.2 Extension to Orthogonal Martingales . . . . . . . . . . . . . . . . . . . . . . . 108

3.6 G-Intensity of Default Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.7 Single Name CDS Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.7.1 Standing Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.7.2 Valuation of a Defaultable Claim . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.7.3 Replication of a Defaultable Claim . . . . . . . . . . . . . . . . . . . . . . . . 119
3.8 Multi-Name CDS Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.8.1 Valuation of a First-to-Default Claim . . . . . . . . . . . . . . . . . . . . . . 122
3.8.2 Replication of a First-to-Default Claim . . . . . . . . . . . . . . . . . . . . . . 128

4 Hedging of Defaultable Claims 131

4.1 Semimartingale Market Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.1.1 Dynamics of Asset Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.2 Trading Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2.1 Unconstrained Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.2.2 Constrained Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.3 Martingale Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.3.1 Defaultable Asset with Zero Recovery . . . . . . . . . . . . . . . . . . . . . . 140
4.3.2 Hedging with a Defaultable Bond . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.3.3 Defaultable Asset with Non-Zero Recovery . . . . . . . . . . . . . . . . . . . 153
4.3.4 Two Defaultable Assets with Zero Recovery . . . . . . . . . . . . . . . . . . . 154
4.4 PDE Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.4.1 Defaultable Asset with Zero Recovery . . . . . . . . . . . . . . . . . . . . . . 158
4.4.2 Defaultable Asset with Non-Zero Recovery . . . . . . . . . . . . . . . . . . . 162
4.4.3 Two Defaultable Assets with Zero Recovery . . . . . . . . . . . . . . . . . . . 165

5 Modeling Dependent Defaults 169

5.1 Basket Credit Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.1.1 The kth-to-Default Contingent Claims . . . . . . . . . . . . . . . . . . . . . . 170
5.1.2 Case of Two Credit Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.2 Conditionally Independent Defaults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
5.2.1 Canonical Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.2.2 Hypothesis (H) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.2.3 Independent Default Times . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.2.4 Signed Intensities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.3 Valuation of FTDC and LTDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
5.4 Copula-Based Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.4.1 Direct Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.4.2 Indirect Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.5 One-factor Gaussian Copula Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
5.6 Jarrow and Yu Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6 CONTENTS

5.6.1 Construction of Default Times . . . . . . . . . . . . . . . . . . . . . . . . . . 178

5.6.2 Case of Two Credit Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
5.7 Kusuoka’s Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.7.1 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.7.2 Bonds with Zero Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
5.8 Basket Credit Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.8.1 Credit Default Index Swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.8.2 Collateralized Debt Obligations . . . . . . . . . . . . . . . . . . . . . . . . . . 185
5.8.3 First-to-Default Swaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
5.8.4 Step-up Corporate Bonds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
5.8.5 Valuation of Basket Credit Derivatives . . . . . . . . . . . . . . . . . . . . . . 188
5.9 Modeling of Credit Ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.9.1 Infinitesimal Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
5.9.2 Transition Intensities for Credit Ratings . . . . . . . . . . . . . . . . . . . . . 191
5.9.3 Conditionally Independent Credit Migrations . . . . . . . . . . . . . . . . . . 192
5.9.4 Examples of Markovian Models . . . . . . . . . . . . . . . . . . . . . . . . . . 192
5.9.5 Forward Credit Default Swap . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.9.6 Credit Default Swaptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.9.7 Spot kth-to-Default Credit Swap . . . . . . . . . . . . . . . . . . . . . . . . . 196
5.9.8 Forward kth-to-Default Credit Swap . . . . . . . . . . . . . . . . . . . . . . . 197
5.9.9 Model Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

A Poisson Processes 203

A.1 Standard Poisson Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
A.2 Inhomogeneous Poisson Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
A.3 Conditional Poisson Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
A.4 The Doléans Exponential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
A.4.1 Exponential of a Process of Finite Variation . . . . . . . . . . . . . . . . . . . 212
A.4.2 Exponential of a Special Semimartingale . . . . . . . . . . . . . . . . . . . . . 212
Introduction

The goal of this text is to give a survey of techniques used in mathematical modeling of credit
risk and to present some recent developments in this area, with the special emphasis on hedging of
defaultable claims. It is largely based on the following papers by T.R. Bielecki, M. Jeanblanc and
M. Rutkowski:

• Modelling and valuation of credit risk. In: Stochastic Methods in Finance, M. Frittelli and W.
Runggaldier, eds., Springer, 2004, 27–126,
• Hedging of defaultable claims. In: Paris-Princeton Lectures on Mathematical Finance 2003,
R. Carmona et al., eds. Springer, 2004, 1–132,
• PDE approach to valuation and hedging of credit derivatives. Quantitative Finance 5 (2005),
257–270,
• Hedging of credit derivatives in models with totally unexpected default. In: Stochastic Processes
and Applications to Mathematical Finance, J. Akahori et al., eds., World Scientific, 2006, 35–
100,
• Hedging of basket credit derivatives in credit default swap market. Journal of Credit Risk 3
(2007), 91–132.
• Pricing and trading credit default swaps in a hazard process model. Forthcoming in Annals
of Applied Probability.

Credit risk embedded in a financial transaction is the risk that at least one of the parties involved
in the transaction will suffer a financial loss due to default or decline in the creditworthiness of the
counter-party to the transaction or, perhaps, of some third party. For example:

• A holder of a corporate bond bears a risk that the market value of the bond will decline due
to decline in credit rating of the issuer.
• A bank may suffer a loss if a bank’s debtor defaults on payment of the interest due and/or the
principal amount of the loan.
• A party involved in a trade of a credit derivative, such as a credit default swap (CDS), may
suffer a loss if a reference credit event occurs.
• The market value of individual tranches constituting a collateralized debt obligation (CDO)
may decline as a result of changes in the correlation between the default times of the underlying
defaultable securities (that is, the collateral assets or the reference credit default swaps).

The most extensively studied form of credit risk is the default risk – that is, the risk that
a counterparty in a financial contract will not fulfil a contractual commitment to meet her/his
obligations stated in the contract. For this reason, the main tool in the area of credit risk modeling
is a judicious specification of the random time of default. A large part of the present text is devoted
to this issue.

7
8 CHAPTER 0. INTRODUCTION

Our main goal is to present a comprehensive introduction to the most important mathematical
tools that are used in arbitrage valuation of defaultable claims, which are also known under the
name of credit derivatives. We also examine in some detail the important issue of hedging these
claims.
This text is organized as follows.

• In Chapter 1, we provide a concise summary of the main developments within the so-called
structural approach to modeling and valuation of credit risk. In particular, we present the
classic structural models, put forward by Merton [124] and Black and Cox [25], and we mention
some variants and extensions of these models. We also study very succinctly the case of a
structural model with a random default triggering barrier.
• Chapter 2 is devoted to the study of an elementary model of credit risk within the hazard
function framework. We focus here on the derivation of pricing formulae for defaultable claims
and the dynamics of their prices. We also deal here with the issue of replication of single-
and multi-name credit derivatives in the stylized credit default swap market. Results of this
chapter should be seen as a first step toward more practical approaches that are presented in
the foregoing chapters.
• Chapter 3 deals with the alternative reduced-form approach in which the main modeling tool is
the hazard process. We examine the pricing formulae for defaultable claims in the reduced-form
setup with stochastic hazard rate and we examine the behavior of the stochastic intensity when
the reference filtration is reduced. Special emphasis is put on the so-called hypothesis (H) and
its invariance with respect to an equivalent change of a probability measure. As an application
of mathematical results, we present here an extension of hedging results established in Chapter
2 for the case of deterministic pre-default intensities to the case of stochastic default intensities.
• Chapter 4 is devoted to a study of hedging strategies for defaultable claims under the assump-
tion that some primary defaultable assets are traded. We first present some theoretical results
on replication of defaultable claims in an abstract semimartingale market model.
Subsequently, we develop the PDE approach to the valuation and hedging of defaultable claims
in a Markovian framework. For the sake of simplicity of presentation, we focus in the present
text on the case of a market model with three traded primary assets and we deal with a single
default time only. However, an extension of the PDE method to the case of any finite number
of traded assets and several default times is readily available.
• Chapter 5 provides an introduction to the area of modeling dependent defaults and, more
generally, to modeling of dependent credit rating migrations for a portfolio of reference credit
names. We present here some applications of these models to the valuation of real-life examples
of actively traded credit derivatives, such as: credit default swaps and swaptions, first-to-
default swaps, credit default index swaps and tranches of collateralized debt obligations.
• For the reader’s convenience, we present in the appendix some well known results regarding the
Poisson process and its generalizations. We also recall there the definition and basic properties
of the Doléans exponential of a semimartingale.

The detailed proofs of most results can be found in papers by Bielecki and Rutkowski [20],
Bielecki et al. [12, 13, 16] and Jeanblanc and Rutkowski [99]. We also quote some of the seminal
papers, but, unfortunately, we were not able to provide here a survey of an extensive research in the
area of credit risk modeling. For more information, the interested reader is thus referred to original
papers by other authors as well as to monographs by Ammann [2], Bluhm, Overbeck and Wagner
[28], Bielecki and Rutkowski [20], Cossin and Pirotte [55], Duffie and Singleton [68], McNeil, Frey
and Embrechts [123], Lando [109], or Schönbucher [138]
Chapter 1

Structural Approach

We start by presenting a rather brief overview of the structural approach to credit risk modeling.
Since it is based on the modeling of the behavior of the total value of the firm’s assets, it is also
known as the value-of-the-firm approach. In order to model credit events (the default event, in
particular), this methodology refers directly to economic fundamentals, such as the capital structure
of a company. As we shall see in what follows, the two major driving concepts in the structural
modeling are: the total value of the firm’s assets and the default triggering barrier. Historically, this
was the first approach used in this area – it can be traced back to the fundamental papers by Black
and Scholes [26] and Merton [124]. The present exposition is largely based on Chapters 2 and 3 in
Bielecki and Rutkowski [20]; the interested reader may thus consult [20] for more details.

1.1 Notation and Definitions

We fix a finite horizon date T ∗ > 0. The underlying probability space (Ω, F, P) is endowed with
some reference filtration F = (Ft )0≤t≤T ∗ , and is sufficiently rich to support the following random
quantities:
• the short-term interest rate process r and thus also a default-free term structure model,
• the value of the firm process V , which is interpreted as a stochastic model for the total value
of the firm’s assets,
• the barrier process v, which is used to specify the default time τ ,
• the promised contingent claim X representing the liabilities to be redeemed to the holder of a
defaultable claim at maturity date T ≤ T ∗ ,
• the process A, which models the promised dividends, that is, the liabilities that are redeemed
continuously or discretely over time to the holder of a defaultable claim,
• the recovery claim Xe representing the recovery payoff received at time T if default occurs prior
to or at the claim’s maturity date T ,
• the recovery process Z, which specifies the recovery payoff at time of default if it occurs prior
to or at the maturity date T.
The probability measure P is aimed to represent the real-world (or statistical ) probability, as opposed
to a martingale measure (also known as a risk-neutral probability). Any martingale measure will be
denoted by Q in what follows.

1.1.1 Defaultable Claims

We postulate that the processes V, Z, A and v are progressively measurable with respect to the
e are FT -measurable. In addition, A is assumed
filtration F, and that the random variables X and X

9
10 CHAPTER 1. STRUCTURAL APPROACH

to be a process of finite variation with A0 = 0. We assume without mentioning that all random
objects introduced above satisfy suitable integrability conditions. Within the structural approach,
the default time τ is typically defined in terms of the firm’s value process V and the barrier process
v. We set
τ = inf { t > 0 : t ∈ T and Vt ≤ vt }
with the usual convention that the infimum over the empty set equals +∞. Typically, the set T is
the interval [0, T ] (or [0, ∞) in the case of perpetual claims). In classic first-passage-time structural
models, the default time τ is given by the formula

τ = inf { t > 0 : t ∈ [0, T ] and Vt ≤ v̄(t)},

where v̄ : [0, T ] → R+ is some deterministic function, termed the barrier.

Remark 1.1.1 In most structural models, the underlying filtration F is generated by a standard
Brownian motion. In that case, the default time τ will be an F-predictable stopping time (as any
stopping time with respect to a Brownian filtration), meaning that there exists a strictly increasing
sequence of F-stopping times announcing the default time.

Provided that default has not occurred before or at time T , the promised claim X is received
in full at the claim’s maturity date T . Otherwise, depending on the market convention regarding
a particular contract, either the amount Xe is received at maturity T , or the amount Zτ is received
at time τ . If default occurs at maturity of the claim, that is, on the event {τ = T }, we adopt the
convention that only the recovery payment X e is received.
It is sometimes convenient to consider simultaneously both kinds of recovery payoff. Therefore,
in this chapter, a generic defaultable claim is formally defined as a quintuplet (X, A, X, e Z, τ ). In
e
other chapters, we set X = 0 and we consider a quadruplet (X, A, Z, τ ), formally identified with a
claim (X, A, 0, Z, τ ). In some cases, we will also set A = 0 so that a defaultable claim will reduce to
a triplet (X, Z, τ ), to be identified with (X, 0, Z, τ ).

1.1.2 Risk-Neutral Valuation Formula

Suppose that our financial market model is arbitrage-free, in the sense that there exists a martingale
measure (risk-neutral probability) Q, meaning that price process of any tradeable security, which
pays no coupons or dividends, becomes an F-martingale under Q, when discounted by the savings
account B, given as
³Z t ´
Bt = exp ru du .
0

We introduce the default process Ht = 1{t≥τ } and we denote by D the process modeling all cash
flows received by the owner of a defaultable claim. Let us write
e {τ ≤T } .
XTd = X1{τ >T } + X1

e Z, τ ) with
Definition 1.1.1 The dividend process D of a defaultable contingent claim (X, A, X,
maturity date T equals, for every t ∈ R+ ,
Z Z
d
Dt = XT 1[T,∞[ (t) + (1 − Hu ) dAu + Zu dHu .
]0,t] ]0,t]

It is apparent that the process D is of finite variation, and

Z Z
(1 − Hu ) dAu = 1{u<τ } dAu = Aτ − 1{t≥τ } + At 1{t<τ } .
]0,t] ]0,t]
1.1. NOTATION AND DEFINITIONS 11

Note that if default occurs at some date t, the promised dividend payment At − At− , which is due
to occur at this date, is not received by the holder of a defaultable claim. Furthermore, if we set
τ ∧ t = min (τ, t) then Z
Zu dHu = Zτ ∧t 1{t≥τ } = Zτ 1{t≥τ } .
]0,t]

Remark 1.1.2 In principle, the promised payoff X could be easily incorporated into the promised
dividends process A. This would not be convenient, however, since in practice the recovery rules
concerning the promised dividends A and the promised claim X are not the same, in general.
For instance, in the case of a defaultable coupon bond, it is frequently postulated that if default
occurs then the future coupons are lost, whereas a strictly positive fraction of the face value is
received by the bondholder.

We are in a position to define the ex-dividend price St of a defaultable claim. At any time t,
the random variable St represents the current value of all future cash flows associated with a given
defaultable claim.

e Z, τ )
Definition 1.1.2 For any date t ∈ [0, T ], the ex-dividend price of a defaultable claim (X, A, X,
is given as
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Ft . (1.1)
]t,T ]

Note that the discounted ex-dividend price St∗ = St Bt−1 , t ∈ [0, T ], satisfies
³Z ¯ ´ Z
¯
St∗ = EQ Bu−1 dDu ¯ Ft − Bu−1 dDu .
]0,T ] ]0,t]

Hence it is a supermartingale (submartingale, respectively) under Q if and only if the dividend

process D is increasing (decreasing, respectively).
The process S c , which is given by the formula
³Z ¯ ´ Z
¯
Stc = Bt EQ Bu−1 dDu ¯ Ft = St + Bt Bu−1 dDu ,
]0,T ] ]0,t]

e Z, τ ).
is called the cumulative price of a defaultable claim (X, A, X,

1.1.3 Defaultable Zero-Coupon Bond

Assume that A = 0, Z = 0 and X = L for some positive constant L > 0. Then the value process
S represents the arbitrage price of a defaultable zero-coupon bond (also referred to as the corporate
discount bond in the sequel) with the face value L and recovery at maturity only. In general, the
price D(t, T ) of such a bond equals
¡ ¯ ¢
e {τ ≤T } ) ¯ Ft .
D(t, T ) = Bt EQ BT−1 (L1{τ >T } + X1

It is convenient to rewrite the last formula as follows

¡ ¯ ¢
D(t, T ) = LBt EQ BT−1 (1{τ >T } + δ(T )1{τ ≤T } ) ¯ Ft ,

e
where the random variable δ(T ) = X/L represents the recovery rate upon default. For a corporate
bond, it is natural to assume that 0 ≤ Xe ≤ L, so that for random variable δ(T ) we obtain the
following bounds 0 ≤ δ(T ) ≤ 1.
12 CHAPTER 1. STRUCTURAL APPROACH

Alternatively, we may re-express the bond price as follows

³ ¡ ¯ ¢´
D(t, T ) = L B(t, T ) − Bt EQ BT−1 w(T )1{τ ≤T } ¯ Ft ,

where
B(t, T ) = Bt EQ (BT−1 | Ft )
is the price of a unit default-free zero-coupon bond and w(T ) = 1 − δ(T ) is the writedown rate
upon default. Generally speaking, the value of a corporate bond depends on the joint probabil-
ity distribution under Q of the three-dimensional random variable (BT , δ(T ), τ ) or, equivalently,
(BT , w(T ), τ ).

Example 1.1.1 According to Merton’s [124] model, the recovery payoff upon default (that is, on
the event {VT < L}) equals X̃ = VT , where the random variable VT is the firm’s value at maturity
date T of a corporate bond. Consequently, the random recovery rate upon default is equal here to
δ(T ) = VT /L and the writedown rate upon default equals w(T ) = 1 − VT /L.

For simplicity, we assume that the savings account B is non-stochastic – that is, the short-
term interest rate r is deterministic. Then the price of a default-free zero-coupon bond equals
B(t, T ) = Bt BT−1 and the price of a zero-coupon corporate bond satisfies

D(t, T ) = Lt (1 − w∗ (t, T )),

where Lt = LB(t, T ) is the present value of future liabilities and w∗ (t, T ) is the conditional expected
writedown rate under Q. It is given by the following equality
¡ ¢
w∗ (t, T ) = EQ w(T )1{τ ≤T } | Ft .

The conditional expected writedown rate upon default equals, under Q,

¡ ¢
EQ w(T )1{τ ≤T } | Ft w∗ (t, T )
wt∗ = = ,
Q(τ ≤ T | Ft ) p∗t
where p∗t = Q(τ ≤ T | Ft ) is the conditional risk-neutral probability of default. Finally, let δt∗ = 1−wt∗
be the conditional expected recovery rate upon default under Q. In terms of p∗t , δt∗ and wt∗ , we obtain

D(t, T ) = Lt (1 − p∗t ) + Lt p∗t δt∗ = Lt (1 − p∗t wt∗ ).

If the random variables w(T ) and τ are conditionally independent with respect to the σ-field Ft
under Q then we have that wt∗ = EQ (w(T ) | Ft ).

Example 1.1.2 In practice, it is common to assume that the recovery rate is non-random. Let
the recovery rate δ(T ) be constant, specifically, δ(T ) = δ for some real number δ. In this case, the
writedown rate w(T ) = w = 1 − δ is non-random as well. Then w∗ (t, T ) = wp∗t and wt∗ = w for
every t ∈ [0, T ]. Furthermore, the price of a defaultable bond has the following representation

D(t, T ) = Lt (1 − p∗t ) + δLt p∗t = Lt (1 − wp∗t ).

We will return to various conventions regarding the recovery values of corporate bonds later on in
this text (see, in particular, Section 2.1).

1.2 Merton’s Model

Classic structural models are based on the assumption that the risk-neutral dynamics of the value
process of the assets of the firm V are given by the following stochastic differential equation (SDE)
¡ ¢
dVt = Vt (r − κ) dt + σV dWt
1.2. MERTON’S MODEL 13

with V0 > 0, where κ is the constant payout ratio (dividend yield) and the process W is a standard
Brownian motion under the martingale measure Q. The positive constant σV represents the volatility.
We first present the classic model put forward by Merton [124], who proposed to base the valuation
of a corporate bond on the following postulates:

• a firm has a single liability with the promised terminal payoff L, interpreted as a zero-coupon
bond with maturity T and face value L > 0,
• the ability of the firm to redeem its debt is determined by the total value VT of firm’s assets
at time T ,
• default may occur at time T only, and the default event corresponds to the event {VT < L}.

Hence the default time τ in Merton’s model equals

τ = T 1{VT <L} + ∞1{VT ≥L} .

Using the present notation, a corporate bond is described by A = 0, Z = 0, and

XTd = VT 1{VT <L} + L1{VT ≥L}

so that X̃ = VT . In other words, the bond’s payoff at maturity date T equals

D(T, T ) = min (VT , L) = L − max (L − VT , 0) = L − (L − VT )+ .

The last equality shows that the valuation of the corporate bond in Merton’s setup is equivalent
to the valuation of a European put option written on the firm’s value with strike equal to the bond
face value.
Let D(t, T ) be the price at time t < T of the corporate bond. It is clear that the value D(Vt ) of
the firm’s debt admits the following representation

D(Vt ) = D(t, T ) = LB(t, T ) − Pt ,

where Pt is the price of a put option with strike L and expiration date T. Hence the value E(Vt ) of
the firm’s equity at time t equals

E(Vt ) = Vt e−κ(T −t) − D(Vt ) = Vt e−κ(T −t) − LB(t, T ) + Pt = Ct ,

where Ct stands for the price at time t of a call option written on the firm’s assets, with strike price
L and exercise date T. To justify the last equality above, we may also observe that at time T we
have
E(VT ) = VT − D(VT ) = VT − min (VT , L) = (VT − L)+ .
We conclude that the firm’s shareholders can be seen as holders of the call option with strike L and
expiry T on the total value of the firm’s assets.
Using the option-like features of a corporate bond, Merton [124] derived a closed-form expression
for its arbitrage price. Let N denote the standard Gaussian cumulative distribution function
Z x
1 2
N (x) = √ e−u /2 du, ∀ x ∈ R.
2π −∞

Proposition 1.2.1 For every t ∈ [0, T [, the value D(t, T ) of a corporate bond equals
¡ ¢ ¡ ¢
D(t, T ) = Vt e−κ(T −t) N − d+ (Vt , T − t) + L B(t, T )N d− (Vt , T − t)

where ¡ ¢
ln(Vt /L) + r − κ ± 21 σV2 (T − t)
d± (Vt , T − t) = √ .
σV T − t
14 CHAPTER 1. STRUCTURAL APPROACH

The unique replicating strategy for a corporate bond involves holding, at any time t ∈ [0, T [, φ1t Vt
units of cash invested in the firm’s value and φ2t B(t, T ) units of cash invested in default-free bonds,
where ¡ ¢
φ1t = e−κ(T −t) N − d+ (Vt , T − t)
and
D(t, T ) − φ1t Vt ¡ ¢
φ2t = = LN d− (Vt , T − t) .
B(t, T )

Let us now examine credit spreads in Merton’s model. For notational simplicity, we set κ = 0.
Then Merton’s formula becomes
¡ √ ¢
D(t, T ) = LB(t, T ) Γt N (−d) + N (d − σV T − t) ,

where we denote Γt = Vt /LB(t, T ) and

ln Γt + 21 σV2 (T − t)
d = d+ (Vt , T − t) = √ .
σV T − t

Since LB(t, T ) represents the current value of the face value of the firm’s debt, the quantity Γt can
be seen as a proxy of the asset-to-debt ratio Vt /D(t, T ). It can be easily verified that the inequality
D(t, T ) < LB(t, T ) is valid. This condition is in turn equivalent to the strict positivity of the
corresponding credit spread, as defined by formula (1.2) below.
Observe that, in the present setup, the continuously compounded yield r(t, T ) at time t on the T -
maturity Treasury zero-coupon bond is constant and equal to the short-term interest rate r. Indeed,
we have
B(t, T ) = e−r(t,T )(T −t) = e−r(T −t) .
Let us denote by rd (t, T ) the continuously compounded yield at time t < T on the corporate bond,
so that
d
D(t, T ) = Le−r (t,T )(T −t) .
From the last equality, it follows that

ln D(t, T ) − ln L
rd (t, T ) = − .
T −t
The credit spread S(t, T ) is defined as the excess return on a defaultable bond, that is, for any t < T ,

1 LB(t, T )
S(t, T ) = rd (t, T ) − r(t, T ) = ln . (1.2)
T −t D(t, T )

In Merton’s model, the credit spread S(t, T ) is given by the following expression
¡ √ ¢
ln N (d − σV T − t) + Γt N (−d)
S(t, T ) = − > 0.
T −t
The property S(t, T ) > 0 is consistent with the real-life feature that corporate bonds have an
expected return in excess of the risk-free interest rate. Indeed, the observed yields on corporate
bonds are systematically higher than yields on Treasury bonds with matching notional amounts and
maturities.
Note, however, that when time t converges to maturity date T then the credit spread in Merton’s
model tends either to infinity or to 0, depending on whether VT < L or VT > L. Formally, if we
define the forward short credit spread at time T as

S(T, T ) := lim S(t, T )

t↑T
1.3. FIRST PASSAGE TIMES 15

then, by straightforward computations, we obtain that

½
0, on the event {VT > L},
S(T, T ) =
∞, on the event {VT < L}.

It is frequently argued in the financial literature that, for realistic values of model’s parameters,
the credit spreads produced by Merton’s model for bonds with short maturities are far below the
spreads observed in the market.

1.3 First Passage Times

Before we present an extension of Merton’s model, put forward by Black and Cox [25], let us present
some well-known mathematical results regarding first passage times, which will prove useful in what
follows.
Let W be a standard one-dimensional Brownian motion under Q with respect to its natural
filtration F. Let us define an auxiliary process Y by setting, for every t ∈ R+ ,

Yt = y0 + νt + σWt , (1.3)

for some constants ν ∈ R and σ > 0. Let us notice that Y inherits from W the strong Markov
property with respect to the filtration F.

1.3.1 Distribution of the First Passage Time

Let τ stand for the first passage time to zero by the process Y , that is,

τ = inf { t ∈ R+ : Yt = 0}. (1.4)

It is known that in an arbitrarily small interval [0, t] the sample path of the Brownian motion started
at 0 passes through origin infinitely many times. Using Girsanov’s theorem and the strong Markov
property of the Brownian motion, it is thus easy to deduce that the first passage time by Y to zero
coincides with the first crossing time by Y of the level 0, that is, with probability 1,

τ = inf { t ∈ R+ : Yt < 0} = inf { t ∈ R+ : Yt ≤ 0}.

In what follows, we will write Xt = νt + σWt for every t ∈ R+ .

Lemma 1.3.1 Let σ > 0 and ν ∈ R. Then for every x > 0 we have
µ ¶ µ ¶
¡ ¢ x − νs 2νσ −2 x −x − νs
Q sup Xu ≤ x = N √ −e N √ (1.5)
0≤u≤s σ s σ s

and for every x < 0

µ ¶ µ ¶
¡ ¢ −x + νs 2νσ −2 x x + νs
Q inf Xu ≥ x = N √ −e N √ . (1.6)
0≤u≤s σ s σ s

Proof. To derive the first equality, we will use Girsanov’s theorem and the reflection principle for a
Brownian motion. Assume first that σ = 1. Let P be the probability measure on (Ω, Fs ) given by
dP ν2
= e−νWs − 2 s , Q-a.s.,
dQ
so that the process Wt∗ := Xt = Wt + νt, t ∈ [0, s], is a standard Brownian motion under P. Also
dQ ∗ ν2
= eνWs − 2 s , P-a.s.
dP
16 CHAPTER 1. STRUCTURAL APPROACH

Moreover, for x > 0,

¡ ¢ ³ ∗ ν2
´
Q sup Xu > x, Xs ≤ x = EP eνWs − 2 s 1{ sup 0≤u≤s Wu∗ >x, Ws∗ ≤x} .
0≤u≤s

ft , t ∈ [0, s]) by setting

We set τx = inf { t ≥ 0 : Wt∗ = x} and we define an auxiliary process (W
ft = Wt∗ 1{τ ≥t} + (2x − Wt∗ )1{τ <t} .
W x x

f is a standard Brownian motion under P. Moreover, we have

By virtue of the reflection principle, W
fu > x, W
{ sup W fs ≤ x} = {W ∗ ≥ x} ⊂ {τx ≤ s}.
s
0≤u≤s

Let ¡ ¢
J := Q sup (Wu + νu) ≤ x .
0≤u≤s

Then we obtain
¡ ¢
J = Q(Xs ≤ x) − Q sup Xu > x, Xs ≤ x
0≤u≤s
∗
³
ν2
´
= Q(Xs ≤ x) − EP eνWs − 2 s 1{ sup 0≤u≤s Wu∗ >x, Ws∗ ≤x}
³ ν2
´
f
= Q(Xs ≤ x) − EP eν Ws − 2 s 1{ sup f f
0≤u≤s Wu >x, Ws ≤x}
³ ∗ ν2
´
= Q(Xs ≤ x) − EP eν(2x−Ws )− 2 s 1{Ws∗ ≥x}
³ ∗ ν2
´
= Q(Xs ≤ x) − e2νx EP eνWs − 2 s 1{Ws∗ ≤−x}
= Q(Ws + νs ≤ x) − e2νx Q(Ws + νs ≤ −x)
µ ¶ µ ¶
x − νs −x − νs
=N √ − e2νx N √ .
s s
This ends the proof of the first equality for σ = 1. For any σ > 0, we have
¡ ¢ ¡ ¢
Q sup (σWu + νu) ≤ x = Q sup (Wu + νσ −1 u) ≤ xσ −1 ,
0≤u≤s 0≤u≤s

and this implies (1.5). Since −W is a standard Brownian motion under Q, we also have that, for
any x < 0, ¡ ¢ ¡ ¢
Q inf (σWu + νu) ≥ x = Q sup (σWu − νu) ≤ −x ,
0≤u≤s 0≤u≤s

and thus (1.6) easily follows from (1.5). ¤

Proposition 1.3.1 The first passage time τ given by (1.4) has the inverse Gaussian probability
distribution under Q. Specifically, for any 0 < s < ∞,
−2
Q(τ ≤ s) = Q(τ < s) = N (h1 (s)) + e−2νσ y0
N (h2 (s)), (1.7)

where N is the standard Gaussian cumulative distribution function and

−y0 − νs −y0 + νs
h1 (s) = √ , h2 (s) = √ .
σ s σ s

Proof. Notice first that

¡ ¢ ¡ ¢
Q(τ ≥ s) = Q inf Yu ≥ 0 = Q inf Xu ≥ −y0 , (1.8)
0≤u≤s 0≤u≤s

where Xu = νu + σWu .
1.3. FIRST PASSAGE TIMES 17

From Lemma 1.3.1, we have that, for every x < 0,

µ ¶ µ ¶
¡ ¢ −x + νs −2 x + νs
Q inf Xu ≥ x = N √ − e2νσ x N √ ,
0≤u≤s σ s σ s
and this yields (1.7), when combined with (1.8). ¤

The following corollary is a consequence of Proposition 1.3.1 and the strong Markov property of
the process Y with respect to the filtration F.

Corollary 1.3.1 For any t < s we have, on the event {t < τ },

µ ¶ µ ¶
−Yt − ν(s − t) −2νσ −2 Yt −Yt + ν(s − t)
Q(τ ≤ s | Ft ) = N √ +e N √ .
σ s−t σ s−t

We are in a position to apply the foregoing results to specific examples of default times. We first
examine the case of a constant lower threshold.

Example 1.3.1 Suppose that the short-term interest rate is constant, that is, rt = r for every
t ∈ R+ . Let the value of the firm process V obey the SDE
¡ ¢
dVt = Vt (r − κ) dt + σV dWt

with constant coefficients κ ∈ R and σV > 0. Let us also assume that the barrier process v is
constant and equal to v̄, where the constant v̄ satisfies v̄ < V0 , so that the default time is given as

τ = inf { t ∈ R+ : Vt ≤ v̄} = inf { t ∈ R+ : Vt < v̄}.

We now set Yt = ln(Vt /v̄). Then it is easy to check that ν = r − κ − 12 σV2 and σ = σV in formula
(1.3). By applying Corollary 1.3.1, we obtain, for every s > t on the event {t < τ },
à v̄ ! à v̄ !
ln Vt − ν(s − t) ³ v̄ ´2a ln Vt + ν(s − t)
Q(τ ≤ s | Ft ) = N √ + N √ ,
σV s − t Vt σV s − t

where we denote
ν r − κ − 12 σV2
a= = .
σV2 σV2
This result was used in Leland and Toft [117].

Example 1.3.2 Let the value process V and the short-term interest rate r be as in Example 1.3.1.
For a strictly positive constant K and an arbitrary γ ∈ R+ , let the barrier function be defined as
v̄(t) = Ke−γ(T −t) for t ∈ R+ , so that the function v̄(t) satisfies

dv̄(t) = γv̄(t) dt, v̄(0) = Ke−γT .

We now set Yt = ln(Vt /v̄(t)) and thus the coefficients in (1.3) are νe = r − κ − γ − 12 σV2 and σ = σV .
We define the default time τ by setting τ = inf { t ≥ 0 : Vt ≤ v̄(t)}. From Corollary 1.3.1, we obtain,
for every t < s on the event {t < τ },
à v̄(t) ! µ ¶2ea à v̄(t) !
ln Vt − νe(s − t) v̄(t) ln Vt + νe(s − t)
Q(τ ≤ s | Ft ) = N √ + N √ ,
σV s − t Vt σV s − t

where
νe r − κ − γ − 12 σV2
e
a= 2 = .
σV σV2
This formula was employed by Black and Cox [25].
18 CHAPTER 1. STRUCTURAL APPROACH

1.3.2 Joint Distribution of Y and τ

We will now find the joint probability distribution, for every y ≥ 0 and s > t,
I := Q(Ys ≥ y, τ ≥ s | Ft ) = Q(Ys ≥ y, τ > s | Ft ),
where τ is given by (1.4). Let us denote by M W and mW the running maximum and minimum of a
one-dimensional standard Brownian motion W , respectively. More explicitly, MsW = sup 0≤u≤s Wu
and mWs = inf 0≤u≤s Wu .
It is well known that for every s > 0 we have
Q(MsW > 0) = 1, Q(mW
s < 0) = 1.

The following classic result – commonly referred to as the reflection principle – is a straightforward
consequence of the strong Markov property of the Brownian motion.

Lemma 1.3.2 We have that, for every s > 0, y ≥ 0 and x ≤ y,

Q(Ws ≤ x, MsW ≥ y) = Q(Ws ≥ 2y − x) = Q(Ws ≤ x − 2y). (1.9)

We need to examine the Brownian motion with non-zero drift. Consider the process X that
equals Xt = νt + σWt . We write MsX = sup0≤u≤s Xu and mX s = inf 0≤u≤s Xu . By virtue of
Girsanov’s theorem, the process X is a Brownian motion, up to an appropriate re-scaling, under an
equivalent probability measure and thus we have, for any s > 0,
Q(MsX > 0) = 1, Q(mX
s < 0) = 1.

Lemma 1.3.3 For every s > 0, the joint distribution of (Xs , MsX ) is given by the expression
−2
Q(Xs ≤ x, MsX ≥ y) = e2νyσ Q(Xs ≥ 2y − x + 2νs)
for every x, y ∈ R such that y ≥ 0 and x ≤ y.

Proof. Since ¡ ¢ ¡ σ ¢
I := Q Xs ≤ x, MsX ≥ y = Q Xsσ ≤ xσ −1 , MsX ≥ yσ −1 ,
where Xtσ = Wt + νtσ −1 , it is clear that we may assume, without loss of generality, that σ = 1. We
will use an equivalent change of probability measure. From Girsanov’s theorem, it follows that X
is a standard Brownian motion under the probability measure P, which is given on (Ω, Fs ) by the
Radon-Nikodým density (recall that σ = 1)
dP ν2
= e−νWs − 2 s , Q-a.s.
dQ
Note also that
dQ ∗ ν2
= eνWs − 2 s , P-a.s.,
dP
where the process (Wt∗ = Xt = Wt + νt, t ∈ [0, s]) is a standard Brownian motion under P. It is
easily seen that
³ ∗ ν2
´ ³ ∗ ν2
´
I = EP eνWs − 2 s 1{Xs ≤x, MsX ≥y} = EP eνWs − 2 s 1{Ws∗ ≤x, MsW ∗ ≥y} .

Since W is a standard Brownian motion under P, an application of the reflection principle (1.9) gives
³ ∗ ν2
´
I = EP eν(2y−Ws )− 2 s 1{2y−Ws∗ ≤x, MsW ∗ ≥y}
³ ∗ ν2
´
= EP eν(2y−Ws )− 2 s 1{Ws∗ ≥2y−x}
³ ∗ ν2
´
= e2νy EP e−νWs − 2 s 1{Ws∗ ≥2y−x} ,

since clearly 2y − x ≥ y.
1.3. FIRST PASSAGE TIMES 19

e say, by setting
Let us define one more equivalent probability measure, P
e
dP ∗ ν2
= e−νWs − 2 s , P-a.s.
dP
Is is clear that
³ ∗ ν2
´
e
I = e2νy EP e−νWs − 2 s 1{Ws∗ ≥2y−x} = e2νy P(W ∗
s ≥ 2y − x).

e and we
ft = Wt∗ + νt, t ∈ [0, s]) is a standard Brownian motion under P
Furthermore, the process (W
have that
e W
I = e2νy P( fs + νs ≥ 2y − x + 2νs).
The last equality easily yields the asserted formula. ¤

It is worthwhile to observe that (a similar remark applies to all formulae below)

Q(Xs ≤ x, MsX ≥ y) = Q(Xs < x, MsX > y).

The following result is a straightforward consequence of Lemma 1.3.3.

Proposition 1.3.2 For any x, y ∈ R satisfying y ≥ 0 and x ≤ y, we have that

µ ¶
¡ X
¢ 2νyσ −2 x − 2y − νs
Q Xs ≤ x, Ms ≥ y = e N √ .
σ s
Hence µ ¶ µ ¶
¡ ¢ x − νs 2νyσ −2 x − 2y − νs
Q Xs ≤ x, MsX ≤ y = N √ −e N √
σ s σ s
for every x, y ∈ R such that x ≤ y and y ≥ 0.

Proof. For the first equality, note that

µ ¶
x − 2y − νs
Q(Xs ≥ 2y − x + 2νs) = Q(−σWs ≤ x − 2y − νs) = N √ ,
σ s

since −σWt has Gaussian law with zero mean and variance σ 2 t. For the second formula, it is enough
to observe that
Q(Xs ≤ x, MsX ≤ y) + Q(Xs ≤ x, MsX ≥ y) = Q(Xs ≤ x)
and to apply the first equality. ¤

It is clear that
Q(MsX ≥ y) = Q(Xs ≥ y) + Q(Xs ≤ y, MsX ≥ y)
for every y ≥ 0, and thus
−2
Q(MsX ≥ y) = Q(Xs ≥ y) + e2νyσ Q(Xs ≥ y + 2νs).

Consequently,
−2
Q(MsX ≤ y) = 1 − Q(MsX ≥ y) = Q(Xs ≤ y) − e2νyσ Q(Xs ≥ y + 2νs).

This leads to the following corollary.

Corollary 1.3.2 The following equality is valid, for every s > 0 and y ≥ 0,
µ ¶ µ ¶
X y − νs 2νyσ −2 −y − νs
Q(Ms ≤ y) = N √ −e N √ .
σ s σ s
20 CHAPTER 1. STRUCTURAL APPROACH

We will now focus on the distribution of the minimal value of X. Observe that we have, for any
y ≤ 0, ¡ ¢ ¡ ¢
Q sup (σWu − νu) ≥ −y = Q inf Xu ≤ y ,
0≤u≤s 0≤u≤s

where we have used the symmetry of the Brownian motion. Consequently, for every y ≤ 0 we have
e e e
Q(mX X
s ≤ y) = Q(Ms ≥ −y), where the process X equals Xt = σWt − νt. It is thus not difficult to
establish the following result.

Proposition 1.3.3 The joint probability distribution of (Xs , mX

s ) satisfies, for every s > 0,
µ ¶ µ ¶
X −x + νs 2νyσ −2 2y − x + νs
Q( Xs ≥ x, ms ≥ y) = N √ −e N √
σ s σ s
for every x, y ∈ R such that y ≤ 0 and y ≤ x.

Corollary 1.3.3 The following equality is valid, for every s > 0 and y ≤ 0,
µ ¶ µ ¶
−y + νs 2νyσ −2 y + νs
Q(mX s ≥ y) = N √ − e N √ .
σ s σ s

Recall that we denote Yt = y0 + Xt , where Xt = νt + σWt . We write

mX
s = inf Xu , mYs = inf Yu .
0≤u≤s 0≤u≤s

Corollary 1.3.4 We have that, for any s > 0 and y ≥ 0,

µ ¶ µ ¶
−y + y0 + νs −2νσ −2 y0 −y − y0 + νs
Q(Ys ≥ y, τ ≥ s) = N √ −e N √ .
σ s σ s

Proof. Since
Q(Ys ≥ y, τ ≥ s) = Q( Ys ≥ y, mYs ≥ 0) = Q( Xs ≥ y − y0 , mX
s ≥ −y0 ),

the asserted formula is rather obvious. ¤

More generally, the Markov property of Y justifies the following result.

Lemma 1.3.4 We have that, for any t < s and y ≥ 0, on the event {t < τ },
µ ¶
−y + Yt + ν(s − t)
Q(Ys ≥ y, τ ≥ s | Ft ) = N √
σ s−t
µ ¶
−2 −y − Yt + ν(s − t)
− e−2νσ Yt N √ .
σ s−t

Example 1.3.3 Assume that the dynamics of the value of the firm process V are
¡ ¢
dVt = Vt (r − κ) dt + σV dWt (1.10)
and set τ = inf { t ≥ 0 : Vt ≤ v̄}, where the constant v̄ satisfies v̄ < V0 . By applying Lemma 1.3.4 to
Yt = ln(Vt /v̄) and y = ln(x/v̄), we obtain the following equality, which holds for x ≥ v̄ on the event
{t < τ },
µ ¶
ln(Vt /x) + ν(s − t)
Q(Vs ≥ x, τ ≥ s | Ft ) = N √
σ s−t
µ ¶2a µ ¶
v̄ ln v̄ 2 − ln(xVt ) + ν(s − t)
− N √ ,
Vt σ s−t
where ν = r − κ − 12 σV2 and a = νσV−2 .
1.4. BLACK AND COX MODEL 21

Example 1.3.4 We consider the setup of Example 1.3.2, so that the value process V satisfies (1.10)
and the barrier function equals v̄(t) = Ke−γ(T −t) for some constants K > 0 and γ ∈ R.
Making use again of Lemma 1.3.4, but this time with Yt = ln(Vt /v̄(t)) and y = ln(x/v̄(s)), we
find that, for every t < s ≤ T and an arbitrary x ≥ v̄(s), the following equality holds on the event
{t < τ }
µ ¶
ln(Vt /v̄(t)) − ln(x/v̄(s)) + νe(s − t)
Q(Vs ≥ x, τ ≥ s | Ft ) = N √
σV s − t
µ ¶2ea µ ¶
v̄(t) − ln(Vt /v̄(t)) − ln(x/v̄(s)) + νe(s − t)
− N √ ,
Vt σV s − t
a = νeσV−2 . Upon simplification, this yields
where νe = r − κ − γ − 12 σV2 and e
µ ¶
ln(Vt /x) + ν(s − t)
Q(Vs ≥ x, τ ≥ s | Ft ) = N √
σV s − t
µ ¶2ea µ 2
¶
v̄(t) ln v̄ (t) − ln(xVt ) + ν(s − t)
− N √ ,
Vt σV s − t
where ν = r − κ − 12 σV2 .

Remark 1.3.1 Note that if we take x = v̄(s) = Ke−γ(T −s) then clearly

1 − Q(Vs ≥ v̄(s), τ ≥ s | Ft ) = Q(τ < s | Ft ) = Q(τ ≤ s | Ft ).

But we also have that

µ ¶ µ ¶
ln(Vt /v̄(s)) + ν(s − t) ln(v̄(t)/Vt ) − νe(s − t)
1−N √ =N √
σV s − t σV s − t
and µ ¶ µ ¶
ln v̄ 2 (t) − ln(v̄(s)Vt ) + ν(s − t) ln(v̄(t)/Vt ) + νe(s − t)
N √ =N √ .
σV s − t σV s − t
By setting x = v̄(s), we rediscover the formula established in Example 1.3.2.

1.4 Black and Cox Model

By construction, Merton’s model does not allow for a premature default, in the sense that the default
may only occur at the maturity of the claim. Several authors have put forward various structural
models for valuation of a corporate debt in which this restrictive and unrealistic feature was relaxed.
In most of these models, the time of default was defined as the first passage time of the value
process V to either deterministic or random barrier. In principle, the bond’s default may thus occur
at any time before or on the maturity date T. The challenge is to appropriately specify the lower
threshold v, the recovery process Z, and to explicitly evaluate the conditional expectation that
appears on the right-hand side of the risk-neutral valuation formula
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Ft ,
]t,T ]

which is valid for t ∈ [0, T [. As one might easily guess, this is a non-trivial mathematical problem,
in general. In addition, the practical problem of the lack of direct observations of the value process
V largely limits the applicability of the first-passage-time models based on the firm value process V .
Black and Cox [25] extend Merton’s [124] research in several directions by taking into account
such specific features of real-life debt contracts as: safety covenants, debt subordination, and re-
strictions on the sale of assets. Following Merton [124], they assume that the firm’s stockholders
22 CHAPTER 1. STRUCTURAL APPROACH

receive continuous dividend payments, which are proportional to the current value of firm’s assets.
Specifically, they postulate that
¡ ¢
dVt = Vt (r − κ) dt + σV dWt , V0 > 0,

where W is a Brownian motion under the risk-neutral probability Q, the constant κ ≥ 0 represents
the payout ratio and σV > 0 is the constant volatility. The short-term interest rate r is assumed to
be constant.
The so-called safety covenants provide the bondholders with the right to force the firm to bank-
ruptcy or reorganization if the firm is doing poorly according to some gauge. The standard for
a poor performance is set by Black and Cox in terms of a time-dependent deterministic barrier
v̄(t) = Ke−γ(T −t) , t ∈ [0, T [, for some constant K > 0. As soon as the total value of firm’s assets
hits this lower threshold, the bondholders take over the firm. Otherwise, default either occurs at
maturity date T or not, depending on whether the inequality VT < L holds or not.
Let us set ½
v̄(t), for t < T ,
vt =
L, for t = T .
The default event occurs at the first time t ∈ [0, T ] at which the firm’s value Vt falls below the
level vt , or the default event does not occur at all. Formally, the default time equals (by convention
inf ∅ = +∞)
τ = inf { t ∈ [0, T ] : Vt ≤ vt }.

The recovery process Z and the recovery payoff Xe are proportional to the value process, specif-
e
ically, Z = β2 V and X = β1 VT for some constants β1 , β2 ∈ [0, 1]. Note that the case examined by
Black and Cox [25] corresponds to β1 = β2 = 1, but, of course, the extension to the case of arbitrary
β1 and β2 is immediate.
To summarize, we consider the following defaultable claim
e = β1 VT , Z = β2 V, τ = τ̄ ∧ τb,
X = L, A = 0, X

where the early default time τ̄ equals

τ̄ = inf { t ∈ [0, T [ : Vt ≤ v̄(t)}

and τb stands for Merton’s default time, that is, τb = T 1{VT <L} + ∞1{VT ≥L} .

1.4.1 Bond Valuation

Similarly as in Merton’s model, it is assumed that the short term interest rate is deterministic and
equal to a positive constant r. We postulate, in addition, that v̄(t) ≤ LB(t, T ) for every t ∈ [0, T ]
or, more explicitly,
Ke−γ(T −t) ≤ Le−r(T −t) ,
so that, in particular, K ≤ L. This additional condition is imposed in order to guarantee that the
payoff to the bondholder at the default time τ will never exceed the face value of the debt, discounted
at a risk-free rate.
Since the dynamics for the value process V are given in terms of a Markovian diffusion, a suitable
partial differential equation can be used to characterize the value process of the corporate bond. Let
us write D(t, T ) = u(Vt , t). Then the pricing function u = u(v, t) of a corporate bond satisfies the
following PDE
1
ut (v, t) + (r − κ)vuv (v, t) + σV2 v 2 uvv (v, t) − ru(v, t) = 0
2
on the domain
{(v, t) ∈ R+ × R+ : 0 < t < T, v > Ke−γ(T −t) }
1.4. BLACK AND COX MODEL 23

with the boundary condition

u(Ke−γ(T −t) , t) = β2 Ke−γ(T −t)
and the terminal condition u(v, T ) = min (β1 v, L).
Alternatively, the price D(t, T ) = u(Vt , t) of a defaultable bond has the following probabilistic
representation, on the event {t < τ } = {t < τ̄ },
³ ¯ ´
¯
D(t, T ) = EQ Le−r(T −t) 1{τ̄ ≥T, VT ≥L} ¯ Ft
³ ¯ ´
¯
+ β1 EQ VT e−r(T −t) 1{τ̄ ≥T, VT <L} ¯ Ft
³ ¯ ´
¯
+ β2 K EQ e−γ(T −τ̄ ) e−r(τ̄ −t) 1{t<τ̄ <T } ¯ Ft .

After default – that is, on the event {t ≥ τ } = {t ≥ τ̄ }, we clearly have

D(t, T ) = β2 v̄(τ )B −1 (τ, T )B(t, T ) = β2 Ke−γ(T −τ ) er(t−τ ) .

We wish find explicit expressions for the conditional expectations arising in the probabilistic repre-
sentation of the price D(t, T ). To this end, we observe that:

• the first two conditional expectations can be computed by using the formula for the conditional
probability Q(Vs ≥ x, τ ≥ s | Ft ),
• to evaluate the third conditional expectation, it suffices to employ the conditional probability
law of the first passage time of the process V to the barrier v̄(t).

1.4.2 Black and Cox Formula

Before we state the bond valuation result due to Black and Cox [25], we find it convenient to
introduce some notation. We denote
1
ν = r − κ − σV2 ,
2
1
m = ν − γ = r − κ − γ − σV2 ,
2
b = mσV−2 .

For the sake of brevity, in the statement of Proposition 1.4.1 we shall write σ instead of σV . As
already mentioned, the probabilistic proof of this result will rely on the knowledge of the probability
law of the first passage time of the geometric (that is, exponential) Brownian motion to an expo-
nential barrier. All relevant results regarding this issue were already established in Section 1.3 (see,
in particular, Examples 1.3.2 and 1.3.4).

Proposition 1.4.1 Assume that m2 +2σ 2 (r−γ) > 0. Prior to default, that is, on the event {t < τ },
the price process D(t, T ) = u(Vt , t) of a defaultable bond equals
¡ ¡ ¢ ¡ ¢¢
D(t, T ) = LB(t, T ) N h1 (Vt , T − t) − Rt2b N h2 (Vt , T − t)
¡ ¡ ¡ ¢¢
+ β1 Vt e−κ(T −t) N h3 (Vt , T − t)) − N h4 (Vt , T − t)
¡ ¡ ¡ ¢¢
+ β1 Vt e−κ(T −t) Rt2b+2 N h5 (Vt , T − t)) − N h6 (Vt , T − t)
¡ ¡ ¢ ¡ ¢¢
+ β2 Vt Rtθ+ζ N h7 (Vt , T − t) + Rtθ−ζ N h8 (Vt , T − t) ,
p
where Rt = v̄(t)/Vt , θ = b + 1, ζ = σ −2 m2 + 2σ 2 (r − γ) and

ln (Vt /L) + ν(T − t)

h1 (Vt , T − t) = √ ,
σ T −t
24 CHAPTER 1. STRUCTURAL APPROACH

ln v̄ 2 (t) − ln(LVt ) + ν(T − t)

h2 (Vt , T − t) = √ ,
σ T −t
ln (L/Vt ) − (ν + σ 2 )(T − t)
h3 (Vt , T − t) = √ ,
σ T −t
ln (K/Vt ) − (ν + σ 2 )(T − t)
h4 (Vt , T − t) = √ ,
σ T −t
ln v̄ 2 (t) − ln(LVt ) + (ν + σ 2 )(T − t)
h5 (Vt , T − t) = √ ,
σ T −t
ln v̄ 2 (t) − ln(KVt ) + (ν + σ 2 )(T − t)
h6 (Vt , T − t) = √ ,
σ T −t
ln (v̄(t)/Vt ) + ζσ 2 (T − t)
h7 (Vt , T − t) = √ ,
σ T −t
ln (v̄(t)/Vt ) − ζσ 2 (T − t)
h8 (Vt , T − t) = √ .
σ T −t

Before proceeding to the proof of Proposition 1.4.1, we will establish an elementary lemma.

Lemma 1.4.1 For any a ∈ R and b > 0 we have, for every y > 0,
Z y µ ¶ µ ¶
ln x + a 1 2 ln y + a − b2
x dN = e 2 b −a N (1.11)
0 b b
and Z y µ ¶ µ ¶
− ln x + a 1 2 − ln y + a + b2
x dN = e2b +a
N . (1.12)
0 b b
Let a, b, c ∈ R satisfy b < 0 and c2 > 2a. Then we have, for every y > 0,
Z y µ ¶
ax b − cx d+c d−c
e dN √ = g(y) + h(y), (1.13)
0 x 2d 2d
√
where d = c2 − 2a and where we denote
µ ¶ µ ¶
b − dy b + dy
g(y) = eb(c−d) N √ , h(y) = eb(c+d) N √ .
y y

Proof. The proof of (1.11)–(1.12) is standard. For (1.13), observe that

Z y µ ¶ Z y µ ¶µ ¶
ax b − cx ax b − cx b c
f (y) := e dN √ = e n √ − 3/2 − √ dx,
0 x 0 x 2x 2 x
where n is the probability density function of the standard Gaussian law. Note also that
à √ !à √ !
√ b − c2 − 2ax b c2 − 2a
0 b(c− c2 −2a)
g (x) = e n √ − 3/2 − √
x 2x 2 x
µ ¶µ ¶
ax b − cx b d
=e n √ − 3/2 − √
x 2x 2 x
and
√ Ã !Ã √ !
√ c2 − 2ax
b+ b c2 − 2a
0 b(c+ c2 −2a)
h (x) = e n √ − 3/2 + √
x 2x 2 x
µ ¶µ ¶
ax b − cx b d
=e n √ − 3/2 + √ .
x 2x 2 x
1.4. BLACK AND COX MODEL 25

Consequently, µ ¶
0 0 ax b b − cx
g (x) + h (x) = −e n √
x3/2 x
and µ ¶
0 0 ax d b − cx
g (x) − h (x) = −e n √ .
x1/2 x
Hence f can be represented as follows
Z
1 y¡ 0 c ¢
f (y) = g (x) + h0 (x) + (g 0 (x) − h0 (x)) dx.
2 0 d

Since limy→0+ g(y) = limy→0+ h(y) = 0, we conclude that we have, for every y > 0,
1 c
f (y) = (g(y) + h(y)) + (g(y) − h(y)).
2 2d
This ends the proof of the lemma. ¤

Proof of Proposition 1.4.1. To establish the asserted formula, it suffices to evaluate the following
conditional expectations:

D1 (t, T ) = LB(t, T ) Q(VT ≥ L, τ̄ ≥ T | Ft ),

¡ ¯ ¢
D2 (t, T ) = β1 B(t, T ) EQ VT 1{VT <L, τ̄ ≥T } ¯ Ft ,
¡ ¯ ¢
D3 (t, T ) = Kβ2 Bt e−γT EQ e(γ−r)τ̄ 1{t<τ̄ <T } ¯ Ft .

For the sake of notational convenience, we will focus on the case t = 0 (of course, the general result
will follow easily).
Let us first evaluate D1 (0, T ), that is, the part of the bond value corresponding to no-default
event. From Example 1.3.4, we know that if L ≥ v̄(T ) = K then
à ! à 2 !
ln VL0 + νT ln v̄LV(0) + νT
Q(VT ≥ L, τ̄ ≥ T ) = N √ − R02ea N √
0

σ T σ T

with R0 = v̄(0)/V0 . It is thus clear that

¡ ¡ ¢ ¡ ¢¢
D1 (0, T ) = LB(0, T ) N h1 (V0 , T ) − R02ea N h2 (V0 , T ) .

Let us now examine D2 (0, T ) – that is, the part of the bond’s value associated with default at time
T . We note that
Z L
D2 (0, T ) ¡ ¢
= EQ VT 1{VT <L, τ̄ ≥T } = x dQ(VT < x, τ̄ ≥ T ).
β1 B(0, T ) K

Using again Example 1.3.4 and the fact that the probability Q(τ̄ ≥ T ) does not depend on x, we
obtain, for every x ≥ K,
µ x ¶ Ã v̄2 (0) !
ln V0 − νT 2e
a
ln xV0 + νT
dQ(VT < x, τ̄ ≥ T ) = dN √ + R0 dN √ .
σ T σ T

Let us denote Z µ ¶
L
ln x − ln V0 − νT
K1 (0) = x dN √
K σ T
and Z µ ¶
L
2 ln v̄(0) − ln x − ln V0 + νT
K2 (0) = x dN √ .
K σ T
26 CHAPTER 1. STRUCTURAL APPROACH

Using (1.11)–(1.12), we obtain

à à ! à !!
(r−κ)T
ln VL0 − νbT ln VK0 − νbT
K1 (0) = V0 e N √ −N √ ,
σ T σ T

where νb = ν + σ 2 = r − κ + 12 σ 2 . Similarly,
 Ã 2 ! Ã v̄2 (0) !
ln v̄LV(0) + ν
b T ln KV0 + ν
b T
K2 (0) = V0 R02 e(r−κ)T N √
0
−N √ .
σ T σ T

Since ¡ ¢
D2 (0, T ) = β1 B(0, T ) K1 (0) + Re0a K2 (0) ,
we conclude that D2 (0, T ) is equal to
¡ ¡ ¡ ¢¢
D2 (0, T ) = β1 V0 e−κT N h3 (V0 , T )) − N h4 (V0 , T )
¡ ¡ ¢ ¡ ¢¢
+ β1 V0 e−κT R02ea+2 N h5 (V0 , T ) − N h6 (V0 , T ) .

It remains to evaluate D3 (0, T ), that is, the part of the bond value associated with the possibility of
the forced bankruptcy before the maturity date T . To this end, it suffices to calculate the following
expected value
Z T
¡ ¢
v̄(0) EQ e(γ−r)τ̄ 1{τ̄ <T } = v̄(0) e(γ−r)s dQ(τ̄ ≤ s),
0

where (see Example 1.3.2)

µ ¶ µ ¶2ea µ ¶
ln(v̄(0)/V0 ) − νes v̄(0) ln(v̄(0)/V0 ) + νes
Q(τ̄ ≤ s) = N √ + N √ .
σ s V0 σ s

Note that v̄(0) < V0 and thus ln(v̄(0)/V0 ) < 0. Using (1.13), we obtain
Z T µ ¶
ln(v̄(0)/V0 ) − νes
v̄(0) e(γ−r)s dN √
0 σ s
a + ζ) θ−ζ ¡
V0 (e ¢ V0 (e
a − ζ) θ+ζ ¡ ¢
= R0 N h8 (V0 , T ) − R0 N h7 (V0 , T )
2ζ 2ζ

and
Z µ ¶
v̄(0)2ea+1 T (γ−r)s ln(v̄(0)/V0 ) + νes
e dN √
V02ea 0 σ s
a + ζ) θ+ζ ¡
V0 (e ¢ V0 (e a − ζ) θ−ζ ¡ ¢
= R0 N h7 (V0 , T ) − R0 N h8 (V0 , T ) .
2ζ 2ζ

Consequently, ¡ ¡ ¢ ¡ ¢¢
D3 (0, T ) = β2 V0 R0θ+ζ N h7 (V0 , T ) + R0θ−ζ N h8 (V0 , T ) .
Upon summation, this completes the proof for t = 0. ¤

Let us consider some special cases of the Black and Cox pricing formula. Assume that β1 = β2 = 1
and the barrier function v̄ is such that K = L. Then necessarily γ ≥ r. It can be checked that for
K = L the pricing formula reduces to D(t, T ) = D1 (t, T ) + D3 (t, T ), where
¡ ¡ ¢ ¡ ¢¢
D1 (t, T ) = LB(t, T ) N h1 (Vt , T − t) − Rt2â N h2 (Vt , T − t) ,
¡ ¡ ¢ ¡ ¢¢
D3 (t, T ) = Vt Rtθ+ζ N h7 (Vt , T − t) + Rtθ−ζ N h8 (Vt , T − t) .
1.4. BLACK AND COX MODEL 27

• Case γ = r. If we also assume that γ = r then ζ = −σ −2 ν̂ and thus

Vt Rtθ+ζ = LB(t, T ), Vt Rtθ−ζ = Vt Rt2â+1 = LB(t, T )Rt2â .
It is also easy to see that in this case
ln(Vt /L) + ν(T − t)
h1 (Vt , T − t) = √ = −h7 (Vt , T − t),
σ T −t
while
ln v̄ 2 (t) − ln(LVt ) + ν(T − t)
h2 (Vt , T − t) = √ = h8 (Vt , T − t).
σ T −t
We conclude that if v̄(t) = Le−r(T −t) = LB(t, T ) then D(t, T ) = LB(t, T ). Note that this result is
quite intuitive. A corporate bond with a safety covenant represented by the barrier function, which
equals the discounted value of the bond’s face value, is equivalent to a default-free bond with the
same face value and maturity.
• Case γ > r. For K = L and γ > r, it is natural to expect that D(t, T ) would be smaller than
LB(t, T ). It is also possible to show that when γ tends to infinity (all other parameters being fixed),
then the Black and Cox price converges to Merton’s price.

1.4.3 Corporate Coupon Bond

We now postulate that the short-term rate r > 0 and that a defaultable bond, of fixed maturity T
and face value L, pays continuously coupons at a constant rate c, so that At = ct for every t ∈ R+ .
The coupon payments are discontinued as soon as the default event occurs. Formally, we consider
here a defaultable claim specified as follows
e = β1 VT , Z = β2 V, τ = inf { t ∈ [0, T ] : Vt < vt }
X = L, At = ct, X
with the Black and Cox barrier v. Let us denote by Dc (t, T ) the value of such a claim at time t < T .
It is clear that Dc (t, T ) = D(t, T ) + A(t, T ), where A(t, T ) stands for the discounted value of future
coupon payments. The value of A(t, T ) can be computed as follows
µZ T ¯ ¶ Z T
−r(s−t) ¯ rt
A(t, T ) = EQ ce 1{τ̄ >s} ds ¯ Ft = ce e−rs Q(τ̄ > s | Ft ) ds.
t t

Setting t = 0, we thus obtain

Z T
Dc (0, T ) = D(0, T ) + c e−rs Q(τ̄ > s) ds = D(0, T ) + A(0, T ),
0

where (recall that we write σ instead of σV )

µ ¶ µ ¶2ea µ ¶
ln(V0 /v̄(0)) + νes v̄(0) ln(v̄(0)/V0 ) + νes
Q(τ̄ > s) = N √ − N √ .
σ s V0 σ s
An integration by parts formula yields
Z T Z T
1³ ´
e−rs Q(τ̄ > s) ds = 1 − e−rT Q(τ̄ > T ) + e−rs dQ(τ̄ > s) .
0 r 0

We assume, as usual, that V0 > v̄(0), so that ln(v̄(0)/V0 ) < 0. Arguing in a similar way as in the
last part of the proof of Proposition 1.4.1 (specifically, using formula (1.13)), we obtain
Z T µ ¶ea+ζe µ e 2T ¶
−rs v̄(0) ln(v̄(0)/V0 ) + ζσ
e dQ(τ̄ > s) = − N √
0 V0 σ T
µ ¶ea−ζe µ e 2T ¶
v̄(0) ln(v̄(0)/V0 ) − ζσ
− N √ ,
V0 σ T
28 CHAPTER 1. STRUCTURAL APPROACH

where
1
νe = r − κ − γ − σ 2 , a = νeσ −2 ,
e
2
and p
ζe = σ −2 νe2 + 2σ 2 r.
Although we have focused on the case when t = 0, it is clear that the derivation of the general
formula for any t < T hinges on basically the same arguments. We thus arrive at the following
result.

Proposition 1.4.2 Consider a defaultable bond with face value L, which pays continuously coupons
at a constant rate c. The price of such a bond equals Dc (t, T ) = D(t, T ) + A(t, T ), where D(t, T )
is the value of a defaultable zero-coupon bond given by Proposition 1.4.1 and A(t, T ) equals, on the
event {t < τ } = {t < τ̄ },
½ ³ ¡
c ¢ ¡ ¢´
A(t, T ) = 1 − B(t, T ) N k1 (Vt , T − t) − Rt2ea N k2 (Vt , T − t)
r
¾
e ¡ ¢ e ¡ ¢
− Reta+ζ N g1 (Vt , T − t) − Reta−ζ N g2 (Vt , T − t) ,

where Rt = v̄(t)/Vt and

ln(Vt /v̄(t)) + νe(T − t)

k1 (Vt , T − t) = √ ,
σ T −t
ln(v̄(t)/Vt ) + νe(T − t)
k2 (Vt , T − t) = √ ,
σ T −t
e 2 (T − t)
ln(v̄(t)/Vt ) + ζσ
g1 (Vt , T − t) = √ ,
σ T −t
e 2 (T − t)
ln(v̄(t)/Vt ) − ζσ
g2 (Vt , T − t) = √ .
σ T −t

Some authors apply the general result to the special case when the default triggering barrier is
assumed to be a constant level. In this special case, the coefficient γ equals zero. Consequently,
νe = ν and p
ζe = σ −2 ν 2 + 2σ 2 r = ζ.
Assume, in addition, that v̄ ≥ L, so that the firm’s insolvency at maturity T is excluded. For the
sake of the reader’s convenience, we state the following immediate corollary to Propositions 1.4.1
and 1.4.2.

Corollary 1.4.1 Let Rt = v̄/Vt . Assume that γ = 0 so that the barrier is constant, v = v̄. If v̄ ≥ L
then the price of a defaultable coupon bond equals, on the event {t < τ } = {t < τ̄ },
c ³ c ´³ ¡ ¢ ¡ ¢´
Dc (t, T ) = + B(t, T ) L − N k1 (Vt , T − t) − Rt2ea N k2 (Vt , T − t)
³ r c ´³ r
¢´
a+ζe ¡ ¢ e ¡
+ β2 v̄ − Rt N g1 (Vt , T − t) + Reta−ζ N g2 (Vt , T − t) .
e
r

Let us mention that the valuation formula of Corollary 1.4.1 coincides with expression (3) in
Leland and Toft [117]. Letting the bond maturity T tend to infinity, we obtain the following
representation of the price of a consol bond (that is, a perpetual coupon bond with infinite maturity)
µ µ ¶ea+ζe ¶ µ ¶ea+ζe
c v̄ v̄
Dc (t) = Dc (t, ∞) = 1− + β2 v̄ .
r Vt Vt
1.4. BLACK AND COX MODEL 29

1.4.4 Optimal Capital Structure

Following Black and Cox [25], we present an analysis of the optimal capital structure of a firm. Let
us consider a firm that has an interest paying bonds outstanding. We assume that it is a consol
bond, which pays continuously coupon rate c. We postulate, in addition, that r > 0 and the payout
rate κ is equal to zero.
The last condition can be given a financial interpretation as the restriction on the sale of assets,
as opposed to issuing of new equity. Equivalently, we may think about a situation in which the
stockholders will make payments to the firm to cover the interest payments. However, they have the
right to stop making payments at any time and either turn the firm over to the bondholders or pay
them a lump payment of c/r per unit of the bond’s notional amount.
Recall that we denote by E(Vt ) (D(Vt ), resp.) the value at time t of the firm equity (debt, resp.),
hence the total value of the firm’s assets satisfies Vt = E(Vt ) + D(Vt ).
Black and Cox [25] argue that there is a critical level of the value of the firm, denoted as v ∗ ,
below which no more equity can be sold. The critical value v ∗ will be chosen by stockholders, whose
aim is to minimize the value of the bonds (equivalently, to maximize the value of the equity). Let us
observe that v ∗ is nothing else than a constant default barrier in the problem under consideration;
the optimal default time τ ∗ thus equals

τ ∗ = inf { t ∈ R+ : Vt ≤ v ∗ }.

To find the value of v ∗ , let us first fix the bankruptcy level v̄. The ODE for the pricing function
u = u∞ (V ) of a consol bond takes the following form (recall that σ = σV )
∞

1 2 2 ∞
V σ uV V + rV u∞
V + c − ru
∞
= 0,
2
subject to the lower boundary condition u∞ (v̄) = min (v̄, c/r) and the upper boundary condition

lim u∞
V (V ) = 0.
V →∞

For the last condition, observe that when the firm’s value grows to infinity, the possibility of default
becomes meaningless, so that the value of the defaultable consol bond tends to the value c/r of the
default-free consol bond. The general solution has the following form:
c
u∞ (V ) = + K1 V + K2 V −α ,
r
where α = 2r/σ 2 and K1 , K2 are some constants, to be determined from boundary conditions. We
find that K1 = 0 and ½ α+1
v̄ − (c/r)v̄ α , if v̄ < c/r,
K2 =
0, if v̄ ≥ c/r,
or, equivalently, µ µ ¶α ¶ µ ¶α
∞ c v̄ v̄
u (Vt ) = 1− + v̄ .
r Vt Vt
It is in the interest of the stockholders to select the bankruptcy level in such a way that the value
of the debt, D(Vt ) = u∞ (Vt ), is minimized and thus the value of firm’s equity
c
E(Vt ) = Vt − D(Vt ) = Vt − (1 − q̄t ) − v̄ q̄t
r
is maximized. It is easy to check that the optimal level of the barrier does not depend on the current
value of the firm, and it equals
c α c
v∗ = = .
rα+1 r + σ 2 /2
30 CHAPTER 1. STRUCTURAL APPROACH

Given the optimal strategy of the stockholders, the price process of the firm’s debt (i.e., of a consol
bond) takes the form, on the event {τ ∗ > t},
µ ¶α+1
c 1 c
D∗ (Vt ) = −
r αVtα r + σ 2 /2

or, equivalently,
c
D∗ (Vt ) = (1 − qt∗ ) + v ∗ qt∗ ,
r
where µ ¶α µ ¶α
v∗ 1 c
qt∗ = = α .
Vt Vt r + σ 2 /2

We end this section by mentioning that other important developments in the area of optimal
capital structure were presented in the papers by Leland [116], Leland and Toft [117], Christensen
et al. [50]. Chen and Kou [46] and Hilberink and Rogers [87] study analogous problems, but they
model the firm’s value as a diffusion process with jumps. This extension was aimed to eliminate an
undesirable feature of other models, in which the spread for corporate bonds converges to zero for
short maturities.

1.5 Extensions of the Black and Cox Model

The Black and Cox first-passage-time approach was later developed by, among others: Brennan and
Schwartz [32, 33] – an analysis of convertible bonds, Nielsen et al. [128] – a random barrier and
random interest rates, Leland [116], Leland and Toft [117] – a study of an optimal capital structure,
bankruptcy costs and tax benefits, Longstaff and Schwartz [119] – a constant barrier combined with
random interest rates, Fouque et al. [74, 75] – a stochastic volatility model and its extension to a
multi-name case.
In general, the default time can be given as

τ = inf { t ∈ R+ : Vt ≤ v(t)},

where v : R+ → R is an arbitrary function and the value of the firm V is modeled as a geometric
Brownian motion.
Moraux [125] proposes to model the default time as a Parisian stopping time. For a continuous
process V and a given t > 0, we first introduce the random variable gtb (V ), representing the last
moment before t when the process V was at a given level b, that is,

gtb (V ) = sup { 0 ≤ s ≤ t : Vs = b}.

The Parisian stopping time is the first time at which the process V is below the level b for a time
period of length greater or equal to a constant α. Formally, the default time τ is given by the formula

τ = inf { t ∈ R+ : (t − gtb (V ))1{Vt <b} ≥ α}.

In the case of the process V governed by the Black-Scholes dynamics, it is possible to find the joint
probability distribution of (τ, Vτ ) by means of the Laplace transform. Another plausible choice for
the default time is the first moment when the process V has spent more than α units of time below
a predetermined level, that is,
τ = inf { t ∈ R+ : AVt > α},
Rt
where we denote AVt = 0 1{Vu <b} du. The probability distribution of this random time is related
to the so-called cumulative options.
1.5. EXTENSIONS OF THE BLACK AND COX MODEL 31

Campi and Sbuelz [43] assume that the default time is given by the first hitting time of 0 by the
CEV process and they study the difficult problem of pricing an equity default swap. More precisely,
they assume that the dynamics under Q of the firm’s value are
³ ´
dVt = Vt− (r − κ) dt + σVtβ dWt − dMt ,

where W is a Brownian motion and M the compensated martingale of a Poisson process (i.e.,
Mt = Nt − λt), and they set τ = inf { t ∈ R+ : Vt ≤ 0}. Put another way, Campi and Sbuelz [43]
define the default time by setting τ = τ β ∧ τ N , where τ N is the first jump of the Poisson process
and τ β is defined as τ β = inf { t ∈ R+ : Xt ≤ 0}, where in turn the process X obeys the following
SDE ³ ´
dXt = Xt− (r − κ + λ) dt + σXtβ dWt .

Using the well-known fact that the CEV process can be expressed in terms of a time-changed Bessel
process and results on the hitting time of 0 for a Bessel process of dimension smaller than 2, they
obtain closed from solutions.
Zhou [146] examines the case where the dynamics under Q of the firm are
³¡ ¢ ´
dVt = Vt− r − λν dt + σ dWt + dXt ,

whereP W is ¡a standard
¢ Brownian motion and X is a compound Poisson process. Specifically, we set
Nt
Xt = i=1 eYi − 1 , where N is a Poisson process with a constant intensity λ, random variables Yi
are independent and have the Gaussian distribution N (a, b2 ). We also set ν = exp(a+b2 /2)−1, since
for this choice of ν the process Vt e−rt is a martingale. Zhou [146] first studies Merton’s problem in
this framework. He also gives an approximation for the first passage time problem when the default
time is given as follows
τ = inf { t ∈ R+ : Vt ≤ L}.

1.5.1 Stochastic Interest Rates

In this section, we present a generalization of the Black and Cox valuation formula for a corporate
bond to the case of random interest rates. We assume that the underlying probability space (Ω, F, P),
endowed with the filtration F = (Ft )t∈R+ , supports the short-term interest rate process r and the
value process V. The dynamics under the martingale measure Q of the firm’s value and of the price
of a default-free zero-coupon bond B(t, T ) are
¡ ¢
dVt = Vt (rt − κ(t)) dt + σ(t) dWt

and ¡ ¢
dB(t, T ) = B(t, T ) rt dt + b(t, T ) dWt
respectively, where W is a d-dimensional standard Q-Brownian motion. Furthermore, κ : [0, T ] → R,
σ : [0, T ] → Rd and b(·, T ) : [0, T ] → Rd are assumed to be bounded functions. The forward value
FV (t, T ) = Vt /B(t, T ) of the firm satisfies under the forward martingale measure QT
¡ ¢
dFV (t, T ) = −κ(t)FV (t, T ) dt + FV (t, T ) σ(t) − b(t, T ) dWtT ,
Rt
where the process WtT = Wt − 0
b(u, T ) du, t ∈ [0, T ], is a d-dimensional Brownian motion under
QT . We set, for any t ∈ [0, T ],
RT
FVκ (t, T ) = FV (t, T )e− t
κ(u) du
.

Then ¡ ¢
dFVκ (t, T ) = FVκ (t, T ) σ(t) − b(t, T ) dWtT .
32 CHAPTER 1. STRUCTURAL APPROACH

Furthermore, it is apparent that FVκ (T, T ) = FV (T, T ) = VT . We consider the following modification
of the Black and Cox approach

X = L, Zt = β2 Vt , X̃ = β1 VT , τ = inf { t ∈ [0, T ] : Vt < vt },

where β2 , β1 ∈ [0, 1] are constants and the barrier v is given by the formula
½ RT
KB(t, T )e t κ(u) du , for t < T ,
vt =
L, for t = T ,

with the constant K satisfying 0 < K ≤ L. Let us denote, for any t ≤ T,

Z T Z T
κ(t, T ) = κ(u) du, σ 2 (t, T ) = |σ(u) − b(u, T )|2 du,
t t

where | · | is the Euclidean norm in R . For brevity, we write Ft = FVκ (t, T ) and we denote
d

1
η± (t, T ) = κ(t, T ) ± σ 2 (t, T ).
2

Proposition 1.5.1 The forward price FD (t, T ) = D(t, T )/B(t, T ) of the defaultable bond equals,
for every t ∈ [0, T [ on the event {τ > t},
¡ ¡ ¢ ¡ ¢¢
L N b h1 (Ft , t, T ) − (Ft /K)e−κ(t,T ) N b h2 (Ft , t, T )
¡ ¡ ¢ ¡ ¢¢
+ β1 Ft e−κ(t,T ) N b h3 (Ft , t, T ) − N b h4 (Ft , t, T )
¡ ¡ ¢ ¡ ¢¢
+ β1 K N b h5 (Ft , t, T ) − N b h6 (Ft , t, T )
+ β2 KJ+ (Ft , t, T ) + β2 Ft e−κ(t,T ) J− (Ft , t, T ),

where

b ln (Ft /L) − η+ (t, T )

h1 (Ft , t, T ) = ,
σ(t, T )
b 2 ln K − ln(LFt ) + η− (t, T )
h2 (Ft , T, t) = ,
σ(t, T )
b ln (L/Ft ) + η− (t, T )
h3 (Ft , t, T ) = ,
σ(t, T )
b ln (K/Ft ) + η− (t, T )
h4 (Ft , t, T ) = ,
σ(t, T )
b 2 ln K − ln(LFt ) + η+ (t, T )
h5 (Ft , t, T ) = ,
σ(t, T )
b ln(K/Ft ) + η+ (t, T )
h6 (Ft , t, T ) = ,
σ(t, T )

and where we write, for Ft > 0 and t ∈ [0, T [,

Z µ ¶
T
ln(K/Ft ) + κ(t, T ) ± 21 σ 2 (t, u)
J± (Ft , t, T ) = eκ(u,T ) dN .
t σ(t, u)

In the special case when κ = 0, the formula of Proposition 1.5.1 covers as a special case the
valuation result established by Briys and de Varenne [40]. In some other recent studies of first
passage time models, in which the triggering barrier is assumed to be either a constant or an
unspecified stochastic process, typically no closed-form solution for the value of a corporate debt is
available and thus a numerical approach is required (see, for instance, Longstaff and Schwartz [119],
Nielsen et al. [128], or Saá-Requejo and Santa-Clara [136]).
1.6. RANDOM BARRIER 33

1.6 Random Barrier

In the case of the full information and the Brownian filtration, the first hitting time of a deterministic
barrier is a predictable stopping time. This is no longer the case when we deal with an incomplete
information (as, e.g., in Duffie and Lando [64]), or when an additional source of randomness is
present. We present here a formula for credit spreads arising in a special case of a totally inaccessible
time of default. For a more detailed study we refer to Babbs and Bielecki [8] and Giesecke [82]. As
we shall see, the method used here is in fact fairly close to the general method presented in Chapter
3. We now postulate that the barrier which triggers default is represented by a random variable η
defined on the underlying probability space. The default time τ is given as τ = inf { t ∈ R+ : Vt ≤ η},
where V is the value of the firm and, for simplicity, V0 = 1. Note that {t < τ } = { inf u≤t Vu > η}.
We shall denote by mV the running minimum of the continuous process V , that is, mVt = inf u≤t Vu .
With this notation, we have that {τ > t} = {mVt > η}. Note that mV is manifestly a decreasing,
continuous process.

1.6.1 Independent Barrier

We assume that, under the risk-neutral probability Q, a random variable η modeling the barrier is
independent of the value of the firm. We denote by Fη the cumulative distribution function of η,
that is, Fη (z) = Q(η ≤ z). We assume that Fη is a differentiable function and we denote by fη its
derivative (with fη (z) = 0 for z > V0 ).

Lemma 1.6.1 Let us set Ft = Q(τ ≤ t | Ft ) and Γt = − ln(1 − Ft ). Then

Z t
fη (mVu )
Γt = − V
dmVu .
0 Fη (mu )

Proof. If a random variable η is independent of F∞ then

Ft = Q(τ ≤ t | Ft ) = Q(mVt ≤ η | Ft ) = 1 − Fη (mVt ).

The process mV is decreasing and thus Γt = − ln Fη (mVt ). We conclude that

Z t
fη (mVu )
Γt = − V
dmVu ,
0 Fη (mu )

as required. ¤

Example 1.6.1 Let V0 = 1 and let η be a random variable uniformly distributed on the interval
[0, 1]. Then manifestly Γt = − ln mVt . The computation of the expected value EQ (eΓT f (VT )) requires
the knowledge of the joint probability distribution of the pair (VT , mVT ).

We now postulate, in addition, that the value process V is modeled by a geometric Brownian
motion with a drift. Specifically, we set Vt = eXt , where Xt = µt + σWt . It is clear that τ =
inf { t ∈ R+ : mXt ≤ ψ}, where ψ = ln η and m
X
is the running minimum of the process X, that is,
X
mt = inf { Xs : 0 ≤ s ≤ t}. We choose the Brownian filtration as the reference filtration, that is,
we set F = FW . This means that we assume that the value of the firm process V (hence also the
process X) is perfectly observed. The barrier ψ is not observed, however. We only postulate that an
investor can observe the occurrence of the default time. In other words, he can observe the process
Ht = 1{t≥τ } = 1{mX t ≤ψ}
. We denote by H the natural filtration of the process H. The information
available to the investor is thus represented by the joint filtration G = F ∨ H.
We also assume that the default time τ and interest rates are independent under Q. It is then
possible to establish the following result (for the proof, the interested reader is referred to Giesecke
[82] or Babbs and Bielecki [8]).
34 CHAPTER 1. STRUCTURAL APPROACH

Proposition 1.6.1 Under the assumptions stated above, we deal with a unit corporate bond with
zero recovery. Then the credit spread S(t, T ) is given as, for every t ∈ [0, T [,
½ ³ Z T f (mX ) ´¯ ¾
1 ψ u X ¯
S(t, T ) = −1{t<τ } ln EQ exp dmu ¯ Ft .
T −t t Fψ (mX
u )

Note that the process mX is decreasing, so that the stochastic integral with respect to this process
can be interpreted as a pathwise Stieltjes integral. In Chapter 3, we will examine the notion of a
hazard process of a random time with respect to a reference filtration F. It is thus worth mentioning
that for the default time τ defined above, the F-hazard process Γ exists and it is given by the formula
Z t
fψ (mX
u )
Γt = − X
dmXu .
0 Fψ (mu )

Since this process is manifestly continuous, the default time τ is in fact a totally inaccessible stopping
time with respect to the filtration G.
Chapter 2

Hazard Function Approach

In this chapter, we provide a detailed analysis of the relatively simple case of the reduced-form
methodology, when the flow of information available to an agent reduces to the observations of
the random time representing the default event of some credit name. The emphasis is put on the
evaluation of conditional expectations with respect to the filtration generated by a default time with
the use of the hazard function. We also study hedging strategies based on credit default swaps in
a single name setup and in the case of several credit names. We conclude this chapter by dealing
with examples of copula-based credit risk models with several default times.

2.1 Elementary Market Model

We begin with the simple case where risk-free zero-coupon bonds, driven by a deterministic short-
term interest rate (r(t), t ∈ R+ ), are the only traded assets in the default-free market model. Recall
that in that case the price at time t of the risk-free zero-coupon bond with maturity T equals
³ Z T ´ B(t)
B(t, T ) = exp − r(u) du = ,
t B(T )
³Z t ´
where B(t) = exp r(u) du is the value at time t of the savings account.
0

Definition 2.1.1 A default time τ is assumed to be an arbitrary positive random variable defined
on some underlying probability space (Ω, G, Q).

Let F be the cumulative distribution function of a random variable τ so that

Z t
F (t) = Q(τ ≤ t) = f (u) du,
0

where the second equality holds provided that the distribution of the random time τ admits the
probability density function f .
It is assumed throughout that the inequality F (t) < 1 holds for every t ∈ R+ . Otherwise, there
would exist a finite date t0 for which F (t0 ) = 1, so that the default event would occur either before
or at t0 with probability 1.
We emphasize that the random payoff of the form 1{T <τ } cannot be perfectly hedged with
deterministic zero-coupon bonds, which are the only traded primary assets in our elementary market
model. To hedge the default risk, we shall later postulate that some defaultable assets are traded,
e.g., a defaultable zero-coupon bond or a credit default swap. In the first step, we will postulate that

35
36 CHAPTER 2. HAZARD FUNCTION APPROACH

the “fair value” of a defaultable asset is given by the risk-neutral valuation formula with respect to
Q. Let us note in this regard that in practice, the risk-neutral distribution of default time is implied
from market quotes of traded defaultable assets, rather than postulated a priori.

2.1.1 Hazard Function and Hazard Rate

Recall the standing assumption that F (t) < 1 for every t ∈ R+ .

Definition 2.1.2 The hazard function Γ : R+ → R+ of τ is given by the formula, for every t ∈ R+ ,

Γ(t) = − ln(1 − F (t)).

Note that Γ is a non-decreasing function with the initial value Γ(0) = 0 and with the limit
limt→+∞ Γ(t) = +∞. The following elementary result is easy to prove.

Lemma 2.1.1 If the cumulative distribution

Rt function F is absolutely continuous with respect to
the Lebesgue measure, so that F (t) = 0 f (u) du where f is the probability density function of τ ,
Rt
then the hazard function Γ is absolutely continuous as well. Specifically, Γ(t) = 0 γ(u) du where
γ(t) = f (t)(1 − F (t))−1 for every t ∈ R+ .

The function γ is called the hazard rate or the intensity function of default time τ . When τ
admits the hazard rate γ, we have that, for every t ∈ R+ ,
³ Z t ´
Q(τ > t) = 1 − F (t) = e−Γ(t) = exp − γ(u) du .
0

The interpretation of the hazard rate is that it represents the conditional probability of the occurrence
of default in a small time interval [t, t + dt], given that default has not occurred by time t. More
formally, for almost every t ∈ R+ ,

1
γ(t) = lim Q(t < τ ≤ t + h | τ > t).
h→0 h

Remark 2.1.1 Let τ be the moment of the first jump of an inhomogeneous Poisson process with
a deterministic intensity (λ(t), t ∈ R+ ). It is then well known that the probability density function
of τ equals
µ Z t ¶
Q(τ ∈ dt)
f (t) = = λ(t) exp − λ(u) du = λ(t)e−Λ(t) ,
dt 0
Rt
where Λ(t) = 0 λ(u) du and thus F (t) = Q(τ ≤ t) = 1 − e−Λ(t) . The hazard function Γ is thus
equal to the compensator of the Poisson process, that is, Γ(t) = Λ(t) for every t ∈ R+ . In other
words, the compensated Poisson process Nt − Γ(t) = Nt − Λ(t) is a martingale with respect to the
filtration generated by the Poisson process N .
Conversely, if τ is a random time with the probability density function f , setting Λ(t) = − ln(1 −
F (t)) allows us to interpret τ as the moment of the first jump of an inhomogeneous Poisson process
with the intensity function equal to the derivative of Λ.

Remark 2.1.2 It is not difficult to generalize the study presented in what follows to the case where
τ does not admit a density, by dealing with the right-continuous version of the cumulative function.
The case where τ is bounded can also be studied along the same method.
2.1. ELEMENTARY MARKET MODEL 37

2.1.2 Defaultable Bond with Recovery at Maturity

We denote by H = (Ht , t ∈ R+ ) the right-continuous increasing process Ht = 1{t≥τ } , referred to
as the default indicator process. Let H stand for the natural filtration of the process H. It is clear
that the filtration H is the smallest filtration which makes τ a stopping time. More explicitly, for
any t ∈ R+ , the σ-field Ht is generated by the events {s ≥ τ } for s ≤ t. The key observation is that
any Ht -measurable random variable X has the form

X = h(τ )1{t≥τ } + c1{t<τ } ,

where h : R+ → R is a Borel measurable function and c is a constant.

Remark 2.1.3 It is worth mentioning that if the cumulative distribution function F is continuous
then τ is known to be a totally inaccessible stopping time with respect to H (see, e.g., Dellacherie
[57] or Dellacherie and Meyer [60], Page 107). We will not use this important property explicitly,
however.

Our next goal is to derive some useful valuation formulae for defaultable bonds with differing
recovery schemes.
For the sake of simplicity, we will first assume that a bond is represented by a single payoff at its
maturity T . Therefore, it is possible to value a bond as a European contingent claim X maturing
at T , by applying the standard risk-neutral valuation formula
µ ¶
X ¯¯
πt (X) = B(t) EQ ¯ Ht = B(t, T ) EQ (X | Ht ).
B(T )

For the ease of notation, we will consider, without loss of generality, a defaultable bond with the
face value L = 1.

Constant Recovery at Maturity

A defaultable (or corporate) zero-coupon bond (a DZC) with maturity T , unit par value, and recovery
value δ paid at maturity, consists of:

• the payment of one monetary unit at time T if default has not occurred before T , i.e., if τ > T ,
• the payment of δ monetary units, made at maturity, if τ ≤ T , where δ ∈ [0, 1] is a constant.

The price at time 0 of the defaultable zero-coupon bond is formally defined as the expectation
under Q of the discounted payoff, so that
¡ ¢
Dδ (0, T ) = B(0, T ) EQ 1{T <τ } + δ1{τ ≤T } .

Consequently,
Dδ (0, T ) = B(0, T ) − (1 − δ)B(0, T )F (T ).

The value of the defaultable zero-coupon bond is thus equal to the value of the default-free
zero-coupon bond minus the discounted value of the expected loss computed under the risk-neutral
probability. Of course, for δ = 1 we recover, as expected, the price of a default-free zero-coupon
bond.
Obviously, the price defined above is not a hedging price, since the payoff at maturity of the
defaultable bond cannot be replicated by trading in primary assets; recall that only default-free
zero-coupon bonds are traded in the present setup. Therefore, we deal with an incomplete market
model and the risk-neutral pricing formula for the defaultable zero-coupon bond is thus postulated,
rather than derived from replication.
38 CHAPTER 2. HAZARD FUNCTION APPROACH

The value of the bond at any date t ∈ [0, T ] depends whether or not default has happened before
this time.
On the one hand, if default has occurred before or at time t, the constant payment of δ will be
made at maturity date T and thus the price of the DZC is obviously δB(t, T ).
On the other hand, if default has not yet occurred before or at time t, the date of its occurrence
is uncertain. It is thus natural in this situation to define the ex-dividend price Dδ (t, T ) at time
t ∈ [0, T [ of the DZC maturing at T as the conditional expectation under Q of the discounted payoff
¡ ¢
B(t, T ) 1{T <τ } + δ1{τ ≤T } (2.1)

given the information, which is available at time t, that is, given the no-default event {τ > t}.
In view of specification (2.1) of the bond’s payoff, we thus obtain
e δ (t, T ),
Dδ (t, T ) = 1{t≥τ } δB(t, T ) + 1{t<τ } D

where the pre-default value D e δ (t, T ), t ∈ [0, T ], is defined as

¡ ¯ ¢
De δ (t, T ) = EQ B(t, T ) (1{T <τ } + δ1{τ ≤T } ) ¯ t < τ .

e δ (t, T ), we observe that

To compute D
³ ¯ ´
e δ (t, T ) = B(t, T ) 1 − (1 − δ)Q(τ ≤ T ¯ t < τ )
D
µ ¶
Q(t < τ ≤ T )
= B(t, T ) 1 − (1 − δ)
Q(t < τ )
µ ¶
G(t) − G(T )
= B(t, T ) 1 − (1 − δ) , (2.2)
G(t)
where we denote G(t) = 1 − F (t). Let us define, for every t ∈ [0, T ],

G(T ) ³ Z T ´
B γ (t, T ) = B(t, T ) = exp − (r(u) + γ(u)) du .
G(t) t

Then pre-default value of the bond can be represented as follows

¡ ¢
De δ (t, T ) = B γ (t, T ) + δ B(t, T ) − B γ (t, T ) .

In particular, for δ = 0, that is, for the case of the bond with zero recovery, we obtain the equality
De 0 (t, T ) = B γ (t, T ), and thus the price D0 (t, T ) satisfies

e 0 (t, T ) = 1{t<τ } B γ (t, T ).

D0 (t, T ) = 1{t<τ } D

It is worth noting that the value of the DZC is discontinuous at default time τ , since we have,
on the event {τ ≤ T },
e δ (τ, T ) = (δ − 1)B γ (t, T ) < 0,
Dδ (τ, T ) − Dδ (τ −, T ) = δB(τ, T ) − D

where the last inequality holds for any δ < 1. Recall that for δ = 1 the DZC is simply a default-free
zero-coupon bond.
For practical purposes, equality (2.2) can be rewritten as follows
e δ (t, T ) = B(t, T )(1 − LGD × DP),
D

where the loss given default (LGD) is defined as 1 − δ and the conditional default probability (DP)
is given by the formula
Q(t < τ ≤ T )
DP = = Q(τ ≤ T | t < τ ).
Q(t < τ )
2.1. ELEMENTARY MARKET MODEL 39

If the hazard rate γ ≥ 0 is constant then the pre-default credit spread equals
³ ´
e T ) = 1 ln B(t, T ) = γ − 1 ln 1 + δ(eγ(T −t) − 1) .
S(t,
T −t D e δ (t, T ) T −t
It is thus easily seen that the pre-default credit spread converges to the constant γ(1 − δ) when time
to maturity T − t tends to zero. It is thus strictly positive when γ > 0 and 0 ≤ δ < 1.
Recall that for δ = 0, the equality De 0 (t, T ) = B γ (t, T ) is valid. Hence the short-term interest rate
has simply to be adjusted by adding the credit spread (equal here to γ) in order to price DZCs with
zero recovery using the formula for default-free bonds. The default-risk-adjusted interest rate equals
rb = r + γ and thus it is higher than the risk-free interest rate r if γ is positive. This corresponds to
the real-life feature that the value of a DZC with zero recovery is strictly smaller than the value of a
default-free zero-coupon with the same par value and maturity provided, of course, that the real-life
probability of default event during the bond’s lifetime is positive.

General Recovery at Maturity

Let us now assume that the payment is a deterministic function of the default time, denoted as
δ : R+ → R. Then the value at time 0 of this defaultable zero-coupon is
¡ ¢
Dδ (0, T ) = B(0, T ) EQ 1{T <τ } + δ(τ )1{τ ≤T }
or, more explicitly,
³ Z T ´
Dδ (0, T ) = B(0, T ) G(T ) + δ(s)f (s) ds ,
0
where, as before, G(t) = 1 − F (t) stands for the survival probability. More generally, the ex-dividend
price is given by the formula, for every t ∈ [0, T [,
¡ ¯ ¢
Dδ (t, T ) = B(t, T ) EQ 1{T <τ } + δ(τ )1{τ ≤T } ¯ Ht .
The following result furnishes an explicit representation for the bond’s price in the present setup.

Lemma 2.1.2 The price of the bond satisfies, for every t ∈ [0, T [,
e δ (t, T ) + 1{t≥τ } δ(τ )B(t, T ),
Dδ (t, T ) = 1{t<τ } D (2.3)
e δ (t, T ) equals
where the pre-default value D
¡ ¯ ¢
e δ (t, T ) = B(t, T ) EQ 1{T <τ } + δ(τ )1{τ ≤T } ¯ t < τ
D
Z
G(T ) B(t, T ) T
= B(t, T ) + δ(u)f (u) du
G(t) G(t) t
Z
B γ (t, T ) T
= B γ (t, T ) + δ(u)f (u) du
G(T ) t
Z T RT
= B γ (t, T ) + B γ (t, T ) δ(u)γ(u)e u γ(v) dv du.
t

e δ (t, T ), t ∈ [0, T ]) are

The dynamics of the process (D
e δ (t, T ) = (r(t) + γ(t))D
dD e δ (t, T ) dt − B(t, T )γ(t)δ(t) dt. (2.4)

The proof of the lemma is based on straightforward computations. To derive the dynamics of
e δ (t, T ), it is useful to observe, in particular, that
D
dB γ (t, T ) = (r(t) + γ(t))B γ (t, T ) dt.
The risk-neutral dynamics of the discontinuous process Dδ (t, T ) involve also the H-martingale M
introduced in Section 2.2 below (see Example 2.2.2).
40 CHAPTER 2. HAZARD FUNCTION APPROACH

2.1.3 Defaultable Bond with Recovery at Default

Let us now consider a corporate bond with recovery at default. A holder of a defaultable zero-coupon
bond with maturity T is now entitled to:

• the payment of one monetary unit at time T if default has not yet occurred,
• the payment of δ(τ ) monetary units, where δ is a deterministic function; note that this payment
is made at time τ if τ ≤ T .

The price at time 0 of this defaultable zero-coupon bond is

¡ ¢
Dδ (0, T ) = EQ B(0, T ) 1{T <τ } + B(0, τ )δ(τ )1{τ ≤T }
Z T
= Q(T < τ )B(0, T ) + B(0, u)δ(u) dF (u)
0
Z T
= G(T )B(0, T ) + B(0, u)δ(u)f (u) du.
0

Obviously, if the default has occurred before time t, the value of the DZC is null (this was not the
case for the recovery payment made at the bond maturity), since, unless explicitly stated otherwise,
we adopt throughout the ex-dividend price convention for all assets.

Lemma 2.1.3 The price of the bond satisfies, for every t ∈ [0, T [,
e δ (t, T ),
Dδ (t, T ) = 1{t<τ } D (2.5)

where the pre-default value D e δ (t, T ) equals

¡ ¯ ¢
De δ (t, T ) = EQ B(t, T ) 1{T <τ } + B(t, τ )δ(τ )1{τ ≤T } ¯ t < τ
Z T
G(T ) 1
= B(t, T ) + B(t, u)δ(u) dF (u)
G(t) G(t) t
Z T
1
= B γ (t, T ) + B(t, u)δ(u)f (u) du
G(t) t
Z T
= B γ (t, T ) + B γ (u, T )δ(u)γ(u) du.
t

e δ (t, T ), t ∈ [0, T ]) are

The dynamics the process (D
e δ (t, T ) = (r(t) + γ(t))D
dD e δ (t, T ) dt − δ(t)γ(t) dt. (2.6)

As expected, the dynamics of the price process Dδ (t, T ) will also include a jump with a negative
value occurring at time τ (see Proposition 2.2.2).

Fractional Recovery of Par Value

Assume that a DZC pays a constant recovery δ at default. The pre-default value of the bond is
here the same for the recovery at maturity scheme with the function δB −1 (t, T ). This follows from
a simple reasoning, but it can also be deduced from the formulae established in Lemmas 2.1.2 and
2.1.3.

Fractional Recovery of Treasury Value

We now consider the recovery δ(t) = δB(t, T ) at the moment of default. The pre-default value is in
this case the same as for a defaultable bond with a constant recovery δ at maturity date T . Once
2.2. MARTINGALE APPROACH 41

again, this property is also a consequence of Lemmas 2.1.2 and 2.1.3. Under this convention, we
obtain the following expressions for the pre-default value of the bond
Z
e δ (t, T ) = e−
RT δB(t, T ) T
(r(u)+γ(u)) du
D t + γ(u)G(u) du
G(t) t
Z T Ru
e 0 (t, T ) + δB(t, T )
=D γ(u)e− t γ(v) dv du.
t

Fractional Recovery of Market Value

Let us finally assume that the recovery at the moment of default equals δ(t)D e δ (t, T ), where δ is a
δ
deterministic function. Equivalently, the recovery payoff is given as δ(τ )D (τ −, T ). The dynamics
e δ (t, T ) are now given by (see Duffie and Singleton [66])
of the pre-default value D
¡ ¢ δ
dD e δ (t, T ) = r(t) + γ(t)(1 − δ(t)) D
e (t, T ) dt,

e δ (t, T ) = 1, so that, for every t ∈ [0, T ],

with the terminal condition D
³ Z T Z T ´
e δ
D (t, T ) = exp − r(u) du − γ(u)(1 − δ(u)) du .
t t

2.2 Martingale Approach

We shall work under the standing assumption that F (t) = Q(τ ≤ t) < 1 for every t ∈ R+ , but we
do not impose any further restrictions on the cumulative distribution function F of a default time τ
under Q at this stage. In particular, we do not postulate, in general, that F is a continuous function.

2.2.1 Conditional Expectations

We first give an elementary formula for the computation of the conditional expectation with respect
to the σ-field Ht , as presented, for instance, in Brémaud [30], Dellacherie [57, 58], or Elliott [69].

Lemma 2.2.1 For any Q-integrable and G-measurable random variable X we have that

EQ (X1{t<τ } )
1{t<τ } EQ (X | Ht ) = 1{t<τ } . (2.7)
Q(t < τ )

Proof. The conditional expectation EQ (X | Ht ) is, obviously, Ht -measurable. Therefore, it can be

represented as follows
EQ (X | Ht ) = h(τ )1{t≥τ } + c1{t<τ } (2.8)
for some Borel measurable function h : R+ → R and some constant c. By multiplying both members
by 1{t<τ } and taking the expectation, we obtain

EQ (1{t<τ } EQ (X | Ht )) = EQ (X1{t<τ } ) = cQ(t < τ ),

so that c = (Q(t < τ ))−1 EQ (X1{t<τ } ). By combining this equality with (2.8), we get the desired
result. ¤

Let us recall the notion of the hazard function (cf. Definition 2.1.2).

Definition 2.2.1 The hazard function Γ of a default time τ is defined by the formula Γ(t) =
− ln(1 − F (t)) for every t ∈ R+ .
42 CHAPTER 2. HAZARD FUNCTION APPROACH

Corollary 2.2.1 Assume that X is an H∞ -measurable and Q-integrable random variable, so that
X = h(τ ) for some Borel measurable function h : R+ → R such that EQ |h(τ )| < +∞. If the hazard
function Γ of τ is continuous then
Z ∞
EQ (X | Ht ) = 1{t≥τ } h(τ ) + 1{t<τ } h(u)eΓ(t)−Γ(u) dΓ(u). (2.9)
t

If, in addition, τ admits the intensity function γ then

Z ∞ Ru
EQ (X | Ht ) = 1{t≥τ } h(τ ) + 1{t<τ } h(u)γ(u)e− t
γ(v) dv
du.
t

In particular, we have, for any t ≤ s,

Rs
Q(s < τ | Ht ) = 1{t<τ } e− t
γ(v) dv

and ³ Rs ´
Q(t < τ < s | Ht ) = 1{t<τ } 1 − e− t γ(v) dv .

2.2.2 Martingales Associated with Default Time

We first consider the general case of a possibly discontinuous cumulative distribution function F of
the default time τ .

Proposition 2.2.1 The process (Mt , t ∈ R+ ) defined as

Z
dF (u)
Mt = Ht − (2.10)
]0,t∧τ ] 1 − F (u−)

is an H-martingale.

Proof. Let t < s. Then, on the one hand, we obtain

F (s) − F (t)
EQ (Hs − Ht | Ht ) = 1{t<τ } EQ (1{t<τ ≤s} | Ht ) = 1{t<τ } , (2.11)
1 − F (t)

where the second equality follows from equality (2.7) with X = 1{s≥τ } .
On the other hand, by applying once again formula (2.7), we obtain
ÃZ ! ÃZ !
dF (u) ¯¯ dF (u) ¯¯
EQ ¯ Ht = 1{t<τ } EQ ¯ Ht
]t∧τ,s∧τ ] 1 − F (u−) ]t∧τ,s∧τ ] 1 − F (u−)
ÃZ !
1 dF (u)
= 1{t<τ } EQ 1{u≤τ }
Q(t < τ ) ]t,s] 1 − F (u−)
Z
1 dF (u)
= 1{t<τ } Q(u ≤ τ )
Q(t < τ ) ]t,s] 1 − F (u−)
Z
1 dF (u)
= 1{t<τ } (1 − F (u−))
Q(t < τ ) ]t,s] 1 − F (u−)
Z
1
= 1{t<τ } dF (u)
1 − F (t) ]t,s]
F (s) − F (t)
= 1{t<τ } .
1 − F (t)

In view of (2.11), this proves the result. ¤

2.2. MARTINGALE APPROACH 43

Assume now that the cumulative distribution function F is continuous. Then the process (Mt , t ∈
R+ ), defined as
Z τ ∧t
dF (u)
Mt = Ht − ,
0 1 − F (u)
is an H-martingale.
Moreover, we have that
Z t
dF (u)
= − ln(1 − F (t)) = Γ(t).
0 1 − F (u)
These observations yield the following corollary to Proposition 2.2.1.

Corollary 2.2.2 Assume that F (and thus also Γ) is a continuous function. Then the process
Mt = Ht − Γ(t ∧ τ ), t ∈ R+ , is an H-martingale.

In particular, if F is an absolutely continuous function then the process

Z τ ∧t Z t
Mt = Ht − γ(u) du = Ht − γ(u)(1 − Hu ) du (2.12)
0 0

is an H-martingale.

Remark 2.2.1 From Corollary 2.2.2, we obtain the Doob-Meyer decomposition of the submartin-
gale H as Ht = Mt + Γ(t ∧ τ ). The predictable increasing process At = Γ(t ∧ τ ) is called the
compensator (or the dual predictable projection) of increasing and H-adapted process H.

Example 2.2.1 In the case where N is an inhomogeneous Poisson process with deterministic R t inten-
sity λ and τ is the moment of the first jump of N , let Ht = Nt∧τ . It is well known that Nt − 0 λ(u) du
is a martingale with respect to Rthe natural filtration of N . Therefore, the process stopped at time τ
t∧τ
is also a martingale, i.e., Ht − 0 λ(u) du is a martingale. Furthermore, we have seen in Remark
2.1.1 that we can reduce our attention to this case, since any random time in the present setup can
be viewed as the moment of the first jump of an inhomogeneous Poisson process.

We are in a position to derive the dynamics of a defaultable zero-coupon bond with recovery
δ(τ ) paid at default. We will use the property that the process M is an H-martingale under the
risk-neutral probability Q. For convenience, we shall work under the assumption that τ admits the
hazard rate γ. We emphasize that we are working here under a risk-neutral probability. In the
sequel, we shall see how to compute the risk-neutral default intensity from the historical one, using
a suitable Radon-Nikodým density process.

Proposition 2.2.2 Assume that τ admits the hazard rate γ. Then the risk-neutral dynamics of a
DZC with recovery δ(τ ) paid at default, where δ : R+ → R is a Borel measurable function, are
¡ ¢
e δ (t, T ) dMt
dDδ (t, T ) = r(t)Dδ (t, T ) − δ(t)γ(t)(1 − Ht ) dt − D
where the H-martingale M is given by (2.12).

Proof. Combining the equality (cf. (2.5))

e δ (t, T ) = (1 − Ht )D
Dδ (t, T ) = 1{t<τ } D e δ (t, T )

e δ (t, T ), we obtain
with dynamics (2.6) of the pre-default value D
dDδ (t, T ) = (1 − Ht ) dDe δ (t, T ) − D
e δ (t, T ) dHt
³ ´
= (1 − Ht ) (r(t) + γ(t))D e δ (t, T ) − δ(t)γ(t) dt − D e δ (t, T ) dHt
¡ ¢
= r(t)Dδ (t, T ) − δ(t)γ(t)(1 − Ht ) dt − D e δ (t, T ) dMt ,

as required. ¤
44 CHAPTER 2. HAZARD FUNCTION APPROACH

Example 2.2.2 Assume that τ admits the hazard rate γ. By combining the pricing formula (2.3)
with the pre-default dynamics (2.4), it is possible to show that the risk-neutral dynamics of the price
Dδ (t, T ) of a DZC with recovery δ(τ ) paid at maturity are
¡ ¢
e δ (t, T ) dMt .
dDδ (t, T ) = r(t)Dδ (t, T ) dt + δ(t)B(t, T ) − D (2.13)

From the last formula, one may also derive the integral representation of the H-martingale B(0, t)Dδ (t, T )
for t ∈ [0, T [, which gives the discounted price of the bond, in terms of the H-martingale M associated
with τ (see also Proposition 2.2.6 in this regard).

The foregoing results furnish a few more examples of H-martingales.

Proposition 2.2.3 The process (Lt , t ∈ R+ ), given by the formula

Lt = 1{t<τ } eΓ(t) = (1 − Ht )eΓ(t) , (2.14)

is an H-martingale. If the hazard function Γ is continuous then the process L satisfies

Z
Lt = 1 − Lu− dMu , (2.15)
]0,t]

where the H-martingale M is given by the formula Mt = Ht − Γ(t ∧ τ ).

Proof. We will first show that L is an H-martingale. We have that, for any t > s,

EQ (Lt | Hs ) = eΓ(t) EQ (1{t<τ } | Hs ).

From (2.7), we obtain

1 − F (t)
EQ (1{t<τ } | Hs ) = 1{s<τ } = 1{s<τ } eΓ(s)−Γ(t) .
1 − F (s)
Hence
EQ (Lt | Hs ) = 1{s<τ } eΓ(s) = Ls .

To establish (2.15), it suffices to note that L0 = 1 and to apply the integration by parts formula
for the product of two functions of finite variation. Recall also that Γ is continuous. We thus obtain

dLt = −eΓ(t) dHt + (1 − Ht )eΓ(t) dΓ(t) = −eΓ(t) dMt

= −1{t≤τ } eΓ(t) dMt = −Lt− dMt .

Alternatively, it is possible to show directly that the process L given by (2.14) is the Doléans
exponential of M , that is, that L is the unique solution of the SDE

dLt = −Lt− dMt , L0 = 1.

Note that this SDE can be solved pathwise, since M is manifestly a process of finite variation (see
Section A.4 for more details). ¤

Proposition 2.2.4 Assume that the hazard function Γ is continuous. Let h : R+ → R be a Borel
measurable function such that the random variable h(τ ) is Q-integrable. Then the process (M̄th , t ∈
R+ ), given by the formula
Z t∧τ
h
M̄t = 1{t≥τ } h(τ ) − h(u) dΓ(u), (2.16)
0
is an H-martingale. Moreover, for every t ∈ R+ ,
Z Z
M̄th = h(u) dMu = − e−Γ(u) h(u) dLu . (2.17)
]0,t] ]0,t]
2.2. MARTINGALE APPROACH 45

Proof. The proof given below provides an alternative proof of Corollary 2.2.2. We wish to establish,
through direct calculations, the martingale property of the process M̄ h given by formula (2.16). On
the one hand, formula (2.9) in Corollary 2.2.1 yields
Z s
¡ ¢ Γ(t)
EQ h(τ )1{t<τ ≤s} | Ht = 1{t<τ } e h(u)e−Γ(u) dΓ(u).
t

On the other hand, we note that

³ Z s∧τ ´ ¡ ¢
J := EQ h(u) dΓ(u) = EQ h̃(τ )1{t<τ ≤s} + h̃(s)1{s<τ } | Ht ,
t∧τ
Rs
where we write h̃(s) = t
h(u) dΓ(u). Consequently, using again (2.9), we obtain
³Z s ´
J = 1{t<τ } eΓ(t) h̃(u)e−Γ(u) dΓ(u) + e−Γ(s) h̃(s) .
t

To conclude the proof, it is enough to observe that the Fubini theorem yields
Z s Z u
−Γ(u)
e h(v) dΓ(v) dΓ(u) + e−Γ(s) h̃(s)
t t
Z s Z s Z s
−Γ(v) −Γ(s)
= h(u) e dΓ(v) dΓ(u) + e h(u) dΓ(u)
Zt s u t

= h(u)e−Γ(u) dΓ(u),
t

as required. The proof of formula (2.17) is left to the reader. ¤

Corollary 2.2.3 Assume that the hazard function Γ of τ is continuous. Let h : R+ → R be a

Borel measurable function such that the random variable eh(τ ) is Q-integrable. Then the process
fth , t ∈ R+ ), given by the formula
(M
Z t∧τ
¡ ¢
fth = exp 1{t≥τ } h(τ ) −
M (eh(u) − 1) dΓ(u),
0

is an H-martingale.

Proof. It suffices to observe that

¡ ¢
exp 1{t≥τ } h(τ ) = 1{t≥τ } eh(τ ) + 1{t<τ } = 1{t≥τ } (eh(τ ) − 1) + 1,

and to apply Proposition 2.2.4 to the function eh − 1. ¤

Proposition 2.2.5 Assume that the hazard function Γ of τ is continuous. Let h : R+ → R be a

Borel measurable function such that h ≥ −1 and, for every t ∈ R+ ,
Z t
h(u) dΓ(u) < +∞.
0

cth , t ∈ R+ ), given by the formula

Then the process (M
³ Z t∧τ ´
¡ ¢
cth = 1 + 1{t≥τ } h(τ ) exp −
M h(u) dΓ(u) ,
0

is a non-negative H-martingale.
46 CHAPTER 2. HAZARD FUNCTION APPROACH

Proof. We start by noting that

³ Z t ´
cth
M = exp − (1 − Hu )h(u) dΓ(u)
0
³ Z τ ´
+ 1{t≥τ } h(τ ) exp − (1 − Hu )h(u) dΓ(u)
0
³ Z t ´
= exp − (1 − Hu )h(u) dΓ(u)
0
Z ³ Z u ´
+ h(u) exp − (1 − Hs )h(s) dΓ(s) dHu .
]0,t] 0

Using Itô’s formula, we thus obtain

³ Z t ´¡ ¢
ch
dMt = exp − (1 − Hu )h(u) dΓ(u) h(t) dHt − (1 − Ht )h(t) dΓ(t)
0
³ Z t ´
= h(t) exp − (1 − Hu )h(u) dΓ(u) dMt .
0

This shows that Mch is a non-negative local H-martingale and thus a supermartingale. It can be
cth ) = 1 for every t ∈ R+ . Hence the process M
checked directly that EQ (M ch is indeed an H-
martingale. ¤

2.2.3 Predictable Representation Theorem

In this subsection, we assume that the hazard function Γ is continuous, so that the process Mt =
Ht − Γ(t ∧ τ ) is an H-martingale. The next result shows that the martingale M has the predictable
representation property for the filtration H generated by the default process. Let us observe that
this filtration is also generated by the martingale M .

Proposition 2.2.6 Let h : R+ → R be a Borel measurable function such that the random variable
h(τ ) is integrable under Q. Then the martingale Mth = EQ (h(τ ) | Ht ) admits the representation
Z
Mth = M0h + (h(u) − g(u)) dMu , (2.18)
]0,t]

where Z ∞
1
g(t) = h(u) dF (u) = eΓ(t) EQ (h(τ )1{t<τ } ) = EQ (h(τ ) | t < τ ).
G(t) t

Moreover, g is a continuous function and g(t) = Mth on {t < τ }, so that

Z
h h h
Mt = M0 + (h(u) − Mu− ) dMu .
]0,t]

Proof. From Lemma 2.2.1, we obtain

EQ (h(τ )1{t<τ } )
Mth = h(τ )1{t≥τ } + 1{t<τ }
Q(t < τ )
= h(τ )1{t≥τ } + 1{t<τ } eΓ(t) EQ (h(τ )1{t<τ } ).

We first consider the event {t < τ }. On this event, we clearly have that Mth = g(t). The integration
by parts formula yields
Z ∞
¡ ¢
Mth = g(t) = eΓ(t) EQ h(τ )1{t<τ } = eΓ(t) h(u) dF (u)
t
2.2. MARTINGALE APPROACH 47
Z ∞ Z t Z t
Γ(u)
= h(u) dF (u) − e h(u) dF (u) + e−Γ(u) g(u) deΓ(u)
0 0 0
Z ∞ Z t Z t
= h(u) dF (u) − eΓ(u) h(u) dF (u) + g(u) dΓ(u).
0 0 0

On the other hand, the right-hand side of (2.18) yields, on the event {t < τ },
Z t
EQ (h(τ )) − (h(u) − g(u)) dΓ(u)
0
Z ∞ Z t Z t
Γ(u)
= h(u) dF (u) − e h(u) dF (u) + g(u) dΓ(u),
0 0 0
Γ(u)
where we used the equality dΓ(u) = e dF (u). Hence equality (2.18) is established on the event
{t < τ }.
To prove that (2.18) holds on the event {t ≥ τ } as well, it suffices to note that the process M h
and the process given by the right-hand side of (2.18) are constant on this event (that is, they are
stopped at τ ) and the jump at time τ of both processes are identical.
Specifically, on the event {t ≥ τ } we have that
∆Mτh = Mτh − Mτh− = h(τ ) − g(τ ).
This completes the proof. ¤
Assume that the default time τ admits the intensity function γ. Then an alternative derivation
of (2.18) consists in computing the conditional expectation
Z ∞
h Γ(t)
Mt = EQ (h(τ ) | Ht ) = h(τ )1{t≥τ } + 1{t<τ } e h(u) dF (u)
t
Z Z ∞
= h(u) dHu + (1 − Ht )eΓ(t) h(u) dF (u)
]0,t] t
Z
= h(u) dHu + (1 − Ht )g(t).
]0,t]

Noting that
dF (t) = e−Γ(t) dΓ(t) = e−Γ(t) γ(t) dt,
we obtain
dg(t) = EQ (h(τ )1{t<τ } ) deΓ(t) − eΓ(t) h(t)e−Γ(t) γ(t) dt = (g(t) − h(t))γ(t) dt.
Consequently, the Itô formula yields
dMth = (h(t) − g(t)) dHt + (1 − Ht )(g(t) − h(t))γ(t) dt = (h(t) − g(t)) dMt ,
since, obviously,
dMt = dHt − γ(t)(1 − Ht ) dt.

The following corollary to Proposition 2.2.6 emphasizes the role of the basic martingale M .
R
Corollary 2.2.4 Any H-martingale (Xt , t ∈ R+ ) can be represented as Xt = X0 + ]0,t] ζs dMs ,
where (ζt , t ∈ R+ ) is an H-predictable process.

Remark 2.2.2 Assume that the hazard function Γ is right-continuous. One may establish the
following formula
Z
EQ (h(τ ) | Ht ) = EQ (h(τ )) − e∆Γ(u) (g(u) − h(u)) dMu ,
]0,t∧τ ]

where ∆Γ(u) = Γ(u) − Γ(u−) and g is defined in Proposition 2.2.6.

48 CHAPTER 2. HAZARD FUNCTION APPROACH

2.2.4 Girsanov’s Theorem

Let τ be a non-negative random variable on a probability space (Ω, G, Q). We denote by F the
cumulative distribution function of τ under Q. It is assumed throughout that F (t) < 1 for every
t ∈ R+ , so that the hazard function Γ of τ under Q is well defined.
Let P be an arbitrary probability measure on (Ω, H∞ ), which is absolutely continuous with respect
to Q. Let η stand for the H∞ -measurable Radon-Nikodým density of P with respect to Q
dP
η := = h(τ ) ≥ 0, Q-a.s., (2.19)
dQ
where h : R+ → R+ is a Borel measurable function satisfying
Z
EQ (h(τ )) = h(u) dF (u) = 1. (2.20)
[0,∞[

The probability measure P is equivalent to Q if and only if the inequality in formula (2.19) is strict,
Q-a.s.
Let Fb be the cumulative distribution function of τ under P, that is,
Z
b
F (t) := P(τ ≤ t) = h(u) dF (u).
[0,t]

We assume Fb(t) < 1 for any t ∈ R+ or, equivalently, that

Z
P(τ > t) = 1 − Fb(t) = h(u) dF (u) > 0. (2.21)
]t,∞[

b of τ under P is well defined (of course, this property always holds

Therefore, the hazard function Γ
if P is equivalent to Q).
Put another way, we assume that
Z
¡ ¢
g(t) = eΓ(t) EQ 1{t<τ } h(τ ) = eΓ(t) h(u) dF (u) = eΓ(t) P(τ > t) > 0.
]t,∞[

b = − ln(1 − Fb (t))
Our first goal is to examine the relationship between the hazard functions Γ(t)
and Γ(t) = − ln(1 − F (t)). The first result is an immediate consequence of the definition of the
hazard function.

Lemma 2.2.2 We have, for every t ∈ R+ ,

³R ´
b
Γ(t) ln ]t,∞[
h(u) dF (u)
= .
Γ(t) ln(1 − F (t))

From now on, we assume, in addition, that F is a continuous function. The following result can
be seen as an elementary counterpart of the celebrated Girsanov theorem for a Brownian motion.

Lemma 2.2.3 Assume that the cumulative distribution function F of τ under Q is continuous.
Then the cumulative distribution function Fb of τ under P is continuous and we have that, for every
t ∈ R+ ,
Z t
b =
Γ(t) b
h(u) dΓ(u),
0

where the function b h : R+ → R+ is given by the formula b h(t) = h(t)/g(t). Hence the process
ct , t ∈ R+ ), which is given by the formula
(M
Z t∧τ Z t∧τ
ct := Ht −
M b
h(u) dΓ(u) = Mt − (b
h(u) − 1) dΓ(u),
0 0
is an H-martingale under P.
2.2. MARTINGALE APPROACH 49

Proof. Indeed, if F (and thus Fb) is continuous, we obtain

b = dFb(t) d(1 − e−Γ(t) g(t)) g(t) dΓ(t) − dg(t) b

dΓ(t) = = = h(t) dΓ(t),
b
1 − F (t) e −Γ(t) g(t) g(t)

where we used the following, easy to check, equalities 1 − Fb(t) = e−Γ(t) g(t) and dg(t) = (g(t) −
h(t)) dΓ(t). ¤

b is the hazard function of τ under P, we necessarily have

Remark 2.2.3 Since Γ
Z ∞
b =
lim Γ(t) b
h(t) dΓ(t) = +∞. (2.22)
t→+∞ 0

Conversely, if a continuous function Γ is the hazard function of τ under Q and b

h : R+ → R+ is a
Borel measurable function such that, for every t ∈ R+ ,
Z t
b :=
Γ(t) b
h(u) dΓ(u) < +∞ (2.23)
0

and (2.22) holds, then it is possible to find a probability measure P absolutely continuous with
b is the hazard function of τ under P (see Remark 2.2.4).
respect to Q such that Γ

In the special case when F is an absolutely continuous function, so that the intensity function γ
of τ under Q is well defined, the cumulative distribution function Fb of τ under P equals
Z t
b
F (t) = h(u)f (u) du,
0

so that Fb is an absolutely continuous function as well. Therefore, the intensity function γ

b of τ under
P exists and it is given by the formula

h(t)f (t) h(t)f (t)

γ
b(t) = = Rt .
b
1 − F (t) 1 − 0 h(u)f (u) du

b(t) = b
From Lemma 2.2.3, it follows that γ h(t)γ(t). To re-derive this result, observe that

h(t)f (t) h(t)f (t) h(t)f (t) h(t)f (t)

γ
b(t) = = Rt = R∞ = −Γ(t)
b
1 − F (t) 1 − 0 h(u)f (u) du t
h(u)f (u) du e g(t)
f (t)
=b
h(t) =b
h(t)γ(t).
1 − F (t)

Let us now examine the Radon-Nikodým density process (ηt , t ∈ R+ ), which is given by the formula

dP ¯¯
ηt := Ht = EQ (η | Ht ).
dQ

Proposition 2.2.7 Assume that F is a continuous function and let E stand for the Doléans expo-
nential (see Section A.4). Then
Z
ηt = 1 + ηu− (b
h(u) − 1) dMu (2.24)
]0,t]

or, equivalently,
³Z ´
η t = Et (b
h(u) − 1) dMu . (2.25)
]0, · ]
50 CHAPTER 2. HAZARD FUNCTION APPROACH

Proof. Note that ηt = Mth where Mth = EQ (h(τ ) | Ht ). Using Proposition 2.2.6 and noting that
η0 = M0h = 1, we thus obtain (cf. (2.18))
Z Z
ηt = η0 + (h(u) − g(u)) dMu = 1 + (h(u) − ηu− ) dMu
]0,t] ]0,t]
Z
= 1+ ηu− (bh(u) − 1) dMu .
]0,t]

Formula (2.25) follows from the definition of the Doléans exponential. ¤

It is worth noting that
Z ∞
ηt = 1{t≥τ } h(τ ) + 1{t<τ } h(u)eΓ(t)−Γ(u) dΓ(u),
t
but also (this can be deduced from (2.24))
³ Z t∧τ ´
¡ ¢
ηt = 1 + 1{t≥τ } κ(τ ) exp − κ(u) dΓ(u) ,
0

where we write κ = b
h − 1. Since b
h is a non-negative function, it is clear that the inequality κ ≥ −1
holds.

Remark 2.2.4 Let κ be any Borel measurable function κ ≥ −1 (κ > −1, respectively) such that
Rt
the inequality 0 κ(u) dΓ(u) < +∞ holds for every t ∈ R+ . Then, by virtue of Proposition 2.2.5, the
process
³Z ´
ηtκ := Et κ(u) dMu
]0, · ]
follows a non-negative (positive, respectively) H-martingale under Q. If, in addition, we have that
Z ∞
(1 + κ(u)) dΓ(u) = +∞
0
then ηtκ = EQ (η∞κ κ
| Ht ), where η∞ = limt→∞ ηtκ . In that case, we may define a probability measure
P on (Ω, H∞ ) by setting dP = η∞ κ
dQ. The hazard function Γ b of τ under P satisfies dΓ(t)
b =
(1 + κ(t)) dΓ(t). Note also that in terms of κ, we have (cf. Theorem 3.4.1)
Z t∧τ Z t∧τ
Mct := Mt − κ(u) dΓ(u) = Ht − (1 + κ(u)) dΓ(u).
0 0

2.2.5 Range of Arbitrage Prices

In order to study the model completeness, we first need to specify the class of primary traded
assets. In our elementary model, the primary traded assets are risk-free zero-coupon bonds with
deterministic prices and thus there exists infinitely many equivalent martingale measures (EMMs).
Indeed, the discounted asset prices are constant and thus the class Q of all EMMs coincide with the
set of all probability measures equivalent to the historical probability. Let us assume that under
the historical probability the default time is an unbounded random variable with a strictly positive
probability density function. For any Q ∈ Q, we denote by FQ the cumulative distribution function
of τ under Q, that is,
Z T
FQ (t) = Q(τ ≤ t) = fQ (u) du.
0
The range of prices is defined as the set of all potential prices that do not induce arbitrage oppor-
tunities. For instance, in the case of a DZC with a constant recovery δ ∈ [0, 1[ paid at maturity, the
range of arbitrage prices is equal to the set
{B(0, T ) EQ (1{T <τ } + δ1{τ ≤T } ), Q ∈ Q}.
It is easy to check that this set is exactly the open interval ]δB(0, T ), B(0, T )[. Let us note that this
range of arbitrage (or viable) prices is manifestly too wide to be useful for practical purposes.
2.2. MARTINGALE APPROACH 51

2.2.6 Implied Risk-Neutral Default Intensity

The absence of arbitrage opportunities in a financial model is commonly interpreted in terms of the
existence of an EMM. If defaultable zero-coupon bonds (DZCs) issued by a given firm are traded,
their prices are observed in the bond market. Therefore, the equivalent martingale measure Q, to
be used for pricing purposes for other credit derivatives with the same reference credit name, is in
some sense selected by the market, rather than arbitrarily postulated. To support this claim, we
will show that it is possible to derive the cumulative distribution function of τ under an implied
martingale measure Q from market quotes for default-free and defaultable zero-coupon bonds, that
is, from observed Treasury and corporate yield curves.
It is important to stress that, in the present setup, no specific relationship between the risk-
neutral default intensity and the historical one is expected to hold, in general. In particular, the
risk-neutral default intensity can be either higher or lower than the historical one. The historical
default intensity can be deduced from observation of default times for a cohort of credit names,
whereas the risk-neutral one is obtained from prices of traded defaultable claims for a given credit
name.

Zero Recovery

If a defaultable zero-coupon bond with zero recovery and maturity T is traded at some price D0 (t, T )
belonging to the interval ]0, B(t, T )[ then the process B(0, t)D0 (t, T ) is a martingale under a risk-
neutral probability Q. We do not postulate that the market model is complete, so we do not claim
that an equivalent martingale measure is unique. The following equalities are thus valid under some
martingale measure Q ∈ Q

B(0, t)D0 (t, T ) = EQ (B(0, T )1{T <τ } | Ht )

³ Z T ´
= B(0, T )1{t<τ } exp − γ Q (u) du ,
t

where γ Q (u) = fQ (u)(1 − FQ (u))−1 . Note that the knowledge of the intensity γ Q is manifestly
sufficient for a computation of the implied cumulative distribution function FQ .
Let us now consider t = 0. It is easily seen that if for any maturity date T the price D0 (0, T )
belongs to the range of viable prices ]0, B(0, T )[ then the function γ Q is strictly positive and the
converse implication holds as well. Assuming that prices D0 (0, T ), T > 0, are observed, the function
γ Q that satisfies, for every T > 0,

³ Z T ´
0
D (0, T ) = B(0, T ) exp − γ Q (u) du
0

is the implied risk-neutral default intensity, that is, the unique Q-intensity of τ that is consistent
RT
with the market data for DZCs. More precisely, the value of the integral 0 γ Q (u) du is known for
any T > 0 as soon as defaultable zero-coupon bonds with all maturities are traded at time 0.
The unique risk-neutral intensity can be formally obtained from the market quotes for DZCs by
differentiation with respect to maturity date T , specifically,

r(t) + γ Q (t) = −∂T ln D0 (0, T ) | T =t .

Of course, the last formula is valid provided that the partial derivative in the right-hand side of this
formula is well defined.
52 CHAPTER 2. HAZARD FUNCTION APPROACH

Recovery at Maturity

Assume that the prices of DZCs with different maturities and fixed recovery δ at maturity, are
known. Then we deduce from (2.1.2)

B(0, T ) − Dδ (0, T )
FQ (T ) = .
B(0, T )(1 − δ)

Hence the probability distribution of τ under the EMM implied by the market quotes of DZCs is
uniquely determined. However, as observed by Hull and White [89], extracting risk-neutral default
probabilities from bond prices is in practice more complicated, since most corporate bonds are
coupon-bearing bonds, rather than zero-coupons.

Recovery at Default

In this case, the cumulative distribution function can also be obtained by differentiation of the
defaultable zero-coupon curve with respect to the maturity. Indeed, denoting by ∂T Dδ (0, T ) the
derivative of the value of the DZC at time 0 with respect to the maturity and assuming that
G = 1 − F is differentiable, we obtain from (2.5)

∂T Dδ (0, T ) = g(T )B(0, T ) − G(T )B(0, T )r(T ) − δ(T )g(T )B(0, T ),

where we write g(t) = G0 (t). By solving this equation, we obtain

à Z !
t
δ (K(u))−1
Q(τ > t) = G(t) = K(t) 1 + ∂T D (0, u) du ,
0 B(0, u)(1 − δ(u))

³Z t
r(u) ´
where we denote K(t) = exp du .
0 1 − δ(u)

2.2.7 Price Dynamics of Simple Defaultable Claims

This section examines the dynamics of prices of some simple defaultable claims. For the sake of
simplicity, we postulate here that the interest rate r is constant and we assume that the default
intensity γ is well defined.

Recovery at Maturity

Let S be the price of an asset that only delivers a recovery Z(τ ) at time T for some function Z.
e = Z(τ ). We know
Formally, this corresponds to the defaultable claim (0, 0, Z(τ ), 0, τ ), that is, X
already that the process
Z t
Mt = Ht − (1 − Hu )γ(u) du
0

is an H-martingale. Recall that γ(t) = f (t)/G(t), where f is the probability density function of τ .
Observe that

e−rt St = EQ (Z(τ )e−rT | Ht )

EQ (Z(τ )1{t<τ ≤T } )
= 1{t≥τ } e−rT Z(τ ) + 1{t<τ } e−rT
G(t)
Z
= e−rT e
Z(u) dHu + 1{t<τ } e−rT Z(t),
]0,t]
2.2. MARTINGALE APPROACH 53

where the function Ze : [0, T ] → R is given by the formula

RT
e = EQ (Z(τ )1{t<τ ≤T } ) =
Z(t) t
Z(u)f (u) du
.
G(t) G(t)
It is easily seen that
RT
e = f (t) t
Z(u)f (u) du Z(t)f (t) e f (t) dt − Z(t)f (t) dt
dZ(t) dt − dt = Z(t)
G2 (t) G(t) G(t) G(t)
and thus
µ ¶
f (t) ¡ e ¢
e
d(e−rt St ) = e−rT Z(t) dHt + (1 − Ht ) Z(t) − Z(t) dt − Z(t−) dHt
G(t)
¡ −rT −rt
¢¡ ¢
= e Z(t) − e St− dHt − (1 − Ht )γ(t) dt
¡ ¢
= e−rt e−r(T −t) Z(t) − St− dMt .

The discounted price is here an H-martingale under the risk-neutral probability Q and the price S
does not vanish (unless Z equals zero).

Recovery at Default

Assume now that the recovery payoff is received at default time. Hence we deal here with the
defaultable claim (0, 0, 0, Z, τ ) and thus the price of this claim is obviously equal to zero after τ . In
general, we have

EQ (e−rτ Z(τ )1{t<τ ≤T } )

e−rt St = EQ (e−rτ Z(τ )1{t<τ ≤T } | Ht ) = 1{t<τ } ,
G(t)
b
so that e−rt St = 1{t<τ } Z(t), where the function Zb : [0, T ] → R equals
Z T
b = 1
Z(t) Z(u)e−ru f (u) du.
G(t) t

Note that
RT
b = −Z(t)e f
−rt (t) Z(u)e−ru f (u)du
dZ(t) dt + f (t) t dt
G(t) G2 (t)
f (t) b f (t) dt
= −Z(t)e−rt dt + Z(t)
G(t) G(t)
¡ ¢
b
= γ(t) Z(t) − Z(t)e −rt
dt.

Consequently,
¡ ¢
b − Z(t)e−rt dt − Z(t)
d(e−rt St ) = (1 − Ht )γ(t) Z(t) b dHt
¡ ¢
b
= Z(t)e−rt − Z(t) dMt − Z(t)e−rt (1 − Ht )γ(t) dt
= e−rt (Z(t) − St− ) dMt − Z(t)e−rt (1 − Ht )γ(t) dt.

In that case, the discounted price is not an H-martingale under the risk-neutral probability. By
contrast, the process Z t∧τ
St e−rt + e−ru Z(u)γ(u) du
0
is an H-martingale. It is also worth noting that the recovery can be formally interpreted as a
dividend stream paid at the rate Zγ up to time τ ∧ T .
54 CHAPTER 2. HAZARD FUNCTION APPROACH

2.3 Pricing of General Defaultable Claims

We will now examine the behaviour of the arbitrage price of a general defaultable claim. Let us
first recall the standing notation. A strictly positive random variable τ , defined on the probability
space (Ω, G, P), is called the random time. In view of its interpretation, it will be later referred to as
the default time. We introduce the default indicator process Ht = 1{t≥τ } associated with τ and we
denote by H the filtration generated by this process. We assume from now on that we are given, in
addition, some auxiliary filtration F and we write G = H ∨ F, meaning that we have Gt = σ(Ht , Ft )
for every t ∈ R+ . Note that P is aimed to represent the real-life probability measure.
We simplify slightly the definition of a defaultable claim of Section 1.1.1 by setting X e = 0, so
that a generic defaultable claim is now formally reduced to a quadruplet (X, A, Z, τ ).

Definition 2.3.1 By a defaultable claim maturing at time T we mean a quadruplet (X, A, Z, τ ),

where X is an FT -measurable random variable, A is an F-adapted process of finite variation, Z is
an F-predictable process, and τ is a random time.

As in Section 1.1.1, the role of each component of a defaultable claim will become clear from the
definition of the dividend process D (cf. Definition 1.1.1), which describes all cash flows associated
with a defaultable claim over the lifespan ]0, T ], that is, after the contract was initiated at time 0.
Of course, the choice of 0 as the date of inception is merely a convention.

Definition 2.3.2 The dividend process D of a defaultable claim maturing at T equals, for every
t ∈ [0, T ], Z Z
Dt = X1{τ >T } 1[T,∞[ (t) + (1 − Hu ) dAu + Zu dHu .
]0,t] ]0,t]

The financial interpretation of the process D justifies the following terminology (cf. Section 1.1):
• X is the promised payoff at maturity T ,
• A represents the process of promised dividends,
• the recovery process Z specifies the recovery payoff at default.
It is worth stressing that we maintain here the convention that the cash payment (premium) at time
0 is not included in the dividend process D associated with a defaultable claim.

Example 2.3.1 When dealing with a credit default swap (CDS), it is natural to assume that the
premium paid at time 0 is equal to zero and the process A represents the fee (annuity), which is paid
in instalments up to maturity date or default, whichever comes first. For instance, if At = −κt for
some constant κ > 0, then the market quote of a stylized credit default swap is formally represented
by this constant, referred to as the continuously paid CDS spread or premium (see Section 2.4.1 for
more details).
If the other covenants of the contract are known (i.e., the payoff X and recovery Z are given),
the valuation of a credit default swap is equivalent to finding the level of the rate κ that makes the
swap valueless at inception. Typically, in a credit default swap we have X = 0, whereas the default
protection process Z is specified in reference to recovery rate of an underlying credit name. In a more
realistic approach, the process A is discontinuous, with jumps occurring at the premium payment
dates. In this text, we will only deal with a stylized CDS with a continuously paid premium. For a
discussion of market conventions for CDSs, see, for instance, Brigo [35].

Let us return to the general setup. It is clear that the dividend process D follows a process of
finite variation on [0, T ]. Since
Z Z
(1 − Hu ) dAu = 1{u<τ } dAu = Aτ − 1{t≤τ } + At 1{t<τ } ,
]0,t] ]0,t]
2.3. PRICING OF GENERAL DEFAULTABLE CLAIMS 55

it is also apparent that if default occurs at some date t, the promised dividend At − At− that is due
to be received or paid at this date is canceled. If we denote τ ∧ t = min (τ, t) then we have
Z
Zu dHu = Zτ ∧t 1{t≥τ } = Zτ 1{t≥τ } .
]0,t]

Let us stress that the process Du − Dt , u ∈ [t, T ], represents all cash flows from a defaultable claim
to be received by an investor who has purchased it at time t. Of course, the process Du − Dt may
depend on the past behavior of the claim (e.g., through some intrinsic parameters, such as credit
spreads) as well as on the history of the market prior to t. The past cash flows from a claim are not
valued by the market, however, so that the current market value at time t of a claim (that is, the
price at which it is traded at time t) depends only on future cash flows to be either paid or received
over the time interval ]t, T ].
We will work under the standing assumption that our underlying financial market model is
arbitrage-free, in the sense that there exists a spot martingale measure Q (also referred to as a risk-
neutral probability), meaning that Q is equivalent to the real-life probability P on (Ω, GT ) and the
price process of any traded security, paying no coupons or dividends, follows a G-martingale under
Q, when discounted by the savings account B, which is, as usual, given by the formula
µZ t ¶
Bt = exp ru du .
0

2.3.1 Buy-and-Hold Strategy

We write S i , i = 1, 2, . . . , k to denote the price processes of k primary securities in an arbitrage-free
financial model. We make the standard assumption that the processes S i , i = 1, 2, . . . , k − 1 follow
semimartingales. In addition, we set S k = B so that S k represents the value process of the savings
account.
The last assumption is not necessary, however. One may assume, for instance, that S k is the
price of a T -maturity risk-free zero-coupon bond, or choose any other strictly positive price process
as a numéraire.
For the sake of convenience, we assume that S i , i = 1, 2, . . . , k − 1 are non-dividend-paying assets
and we introduce the discounted price processes S i∗ by setting S i∗ = B −1 S i . All processes are
assumed to be given on a filtered probability space (Ω, G, P), where P is the real-life (i.e., statistical)
probability measure.
Let us now assume that we have an additional traded security that pays dividends during its
lifespan, assumed to be the time interval [0, T ], according to a process of finite variation D, with
D0 = 0. Let S denote a (yet unspecified) price process of this security. In particular, we do not
postulate a priori that S follows a semimartingale. It is not necessary to interpret S as a price
process of a defaultable claim, though we have here this particular interpretation in mind.
Let a G-predictable, Rk+1 -valued process φ = (φ0 , φ1 , . . . , φk ) represent a generic trading strat-
egy, where φjt represents the number of shares of the jth asset held at time t. We identify here S 0
with S, so that S is the 0th asset. In order to derive a pricing formula for this asset, it suffices to
examine a simple trading strategy involving S, namely, the buy-and-hold strategy.
Suppose that one unit of the 0th asset was purchased at time 0, at the initial price S0 , and it
was held until time T . We assume all dividends are immediately reinvested in the savings account
B. Formally, we consider a buy-and-hold strategy ψ = (1, 0, . . . , 0, ψ k ), where ψ k is a G-predictable
process. The wealth process V (ψ) of ψ equals, for every t ∈ [0, T ],
Vt (ψ) = St + ψtk Bt . (2.26)

Definition 2.3.3 We say that a strategy ψ = (1, 0, . . . , 0, ψ k ) is self-financing if

dVt (ψ) = dSt + dDt + ψtk dBt ,
56 CHAPTER 2. HAZARD FUNCTION APPROACH

or more explicitly, for every t ∈ [0, T ],

Z t
Vt (ψ) − V0 (ψ) = St − S0 + Dt + ψuk dBu . (2.27)
0

We assume from now on that the process ψ k is chosen in such a way (with respect to S, D and
B) that a buy-and-hold strategy ψ is self-financing. In view of (2.26)–(2.27) this means that, for
every t ∈ [0, T ],
Z t
ψtk Bt = V0 (ψ) − S0 + Dt + ψuk dBu .
0

InR addition, we make the standing assumption that the random variable Y defined by the equality
Y = ]0,T ] Bu−1 dDu is Q-integrable, where Q is a martingale measure.

Lemma 2.3.1 The discounted wealth process V ∗ (ψ) = B −1 V (ψ) of any self-financing buy-and-hold
trading strategy ψ satisfies, for every t ∈ [0, T ],
Z
Vt∗ (ψ) = V0∗ (ψ) + St∗ − S0∗ + Bu−1 dDu . (2.28)
]0,t]

Therefore, we have, for every t ∈ [0, T ],

Z
VT∗ (ψ) − Vt∗ (ψ) = ST∗ − St∗ + Bu−1 dDu . (2.29)
]t,T ]

Proof. We define an auxiliary process Vb (ψ) by setting Vbt (ψ) = Vt (ψ) − St = ψtk Bt for t ∈ [0, T ]. In
view of (2.27), we have
Z t
b b
Vt (ψ) = V0 (ψ) + Dt + ψuk dBu ,
0

and thus the process Vb (ψ) follows a semimartingale. An application of Itô’s product rule yields
¡ ¢
d Bt−1 Vbt (ψ) = Bt−1 dVbt (ψ) + Vbt (ψ) dBt−1
= Bt−1 dDt + ψtk Bt−1 dBt + ψtk Bt dBt−1
= Bt−1 dDt ,

where we have used the obvious identity Bt−1 dBt + Bt dBt−1 = 0. By integrating the last equality,
we obtain Z
¡ ¢ ¡ ¢
Bt−1 Vt (ψ) − St = B0−1 V0 (ψ) − S0 + Bu−1 dDu ,
]0,t]

and this immediately yields (2.28). ¤

It is worth noting that Lemma 2.3.1 remains valid if the assumption that S k represents the
savings account B is relaxed. It suffices to assume that the price process S k is a numéraire, that is,
a strictly positive continuous semimartingale. For the sake of brevity, let us write S k = β. We say
that ψ = (1, 0, . . . , 0, ψ k ) is self-financing if the wealth process V (ψ), defined as

Vt (ψ) = St + ψtk βt ,

satisfies, for every t ∈ [0, T ],

Z t
Vt (ψ) − V0 (ψ) = St − S0 + Dt + ψuk dβu .
0
2.3. PRICING OF GENERAL DEFAULTABLE CLAIMS 57

Lemma 2.3.2 The relative wealth Vt∗ (ψ) = βt−1 Vt (ψ) of a self-financing trading strategy ψ satisfies,
for every t ∈ [0, T ], Z
Vt∗ (ψ) = V0∗ (ψ) + St∗ − S0∗ + βu−1 dDu ,
]0,t]

where S ∗ = βt−1 St .

Proof. The proof proceeds along the same lines as the proof of Lemma 2.3.1. It suffices to note that
the equality βt−1 dβt + βt dβt−1 + dhβ, β −1 it = 0 holds for every t ∈ [0, T ]. ¤

2.3.2 Spot Martingale Measure

Our next goal is to derive the risk-neutral valuation formula for the ex-dividend price process S.
Recall that we have assumed that our market model is arbitrage-free, meaning that it admits a (not
necessarily unique) martingale measure Q, equivalent to P, which is associated with the choice of B
as a numéraire. Let us recall the definition of a spot martingale measure.

Definition 2.3.4 We say that Q is a spot martingale measure if the discounted price S i∗ = S i B −1
of any non-dividend paying traded security S i follows a Q-martingale with respect to the filtration
G.

It is well known that the discounted wealth process V ∗ (φ) = V (φ)B −1 of any self-financing
trading strategy φ = (0, φ1 , φ2 , . . . , φk ) is a local martingale under any martingale measure Q.
In what follows, we only consider admissible trading strategies, that is, strategies for which the
discounted wealth process V ∗ (φ) is a martingale under some martingale measure Q.
A market model in which only admissible trading strategies are allowed is arbitrage-free, that is,
there are no arbitrage opportunities in this model.
Following this line of arguments, we postulate that the trading strategy ψ introduced in Section
2.3.1 is also admissible, so that its discounted wealth process V ∗ (ψ) is a martingale under Q with
respect to G. This assumption is quite natural if we wish to prevent arbitrage opportunities to
appear in the extended model of the financial market. Indeed, since we postulate that S is traded,
the wealth process V (ψ) can be formally seen as an additional non-dividend paying traded security.
To derive a pricing formula for a defaultable claim, we make a natural assumption that the
market value at time t of the 0th security comes exclusively from the future dividends stream, that
is, from the cash flows occurring in the open interval ]t, T [. Since the lifespan of S is [0, T ], this
amounts to postulate that ST = ST∗ = 0. To emphasize this property, we shall refer to S as the
ex-dividend price of the 0th asset.

Definition 2.3.5 A process S with ST = 0 is the ex-dividend price of the 0th asset if the dis-
counted wealth V ∗ (ψ) of any self-financing buy-and-hold strategy ψ follows a G-martingale under a
martingale measure Q.

As a special case, we obtain the ex-dividend price a defaultable claim with maturity T .

Proposition 2.3.1 The ex-dividend price process S associated with the dividend process D satisfies,
for every t ∈ [0, T ],
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Gt . (2.30)
]t,T ]

Proof. The postulated martingale property of the discounted wealth process V ∗ (ψ) yields, for every
t ∈ [0, T ], ¡ ¯ ¢
EQ VT∗ (ψ) − Vt∗ (ψ) ¯ Gt = 0.
58 CHAPTER 2. HAZARD FUNCTION APPROACH

Taking into account (2.29), we thus obtain

³ Z ¯ ´
¯
St∗ = EQ ST∗ + Bu−1 dDu ¯ Gt .
]t,T ]

Since, by virtue of the definition of the ex-dividend price, the equalities ST = ST∗ = 0 are valid, the
last formula yields (2.30). ¤

It is not difficult to show that the ex-dividend price S satisfies the equality St = 1{t<τ } Set for
t ∈ [0, T ], where the process Se represents the ex-dividend pre-default price of a defaultable claim.
The cumulative price process S c associated with the dividend process D is given by the formula, for
every t ∈ [0, T ],
³Z ¯ ´
¯
Stc = Bt EQ Bu−1 dDu ¯ Gt . (2.31)
]0,T ]

The corresponding discounted cumulative price process, S c∗ := B −1 S c , is a G-martingale under Q.

The savings account B can be replaced by an arbitrary numéraire β. The corresponding valuation
formula becomes, for every t ∈ [0, T ],
³Z ¯ ´
¯
St = βt EQβ βu−1 dDu ¯ Gt ,
]t,T ]

where Qβ is a martingale measure on (Ω, GT ) associated with a numéraire β, that is, a probability
measure on (Ω, GT ) given by the formula

dQβ B0 βT
= , Q-a.s.
dQ β0 BT

2.3.3 Self-Financing Trading Strategies

Let us now consider a general trading strategy φ = (φ0 , φ1 , . . . , φk ) with G-predictable components.
Pk
The associated wealth process V (φ) is given by the equality Vt (φ) = i=0 φit Sti , where, as before
S 0 = S. A strategy φ is said to be self-financing if Vt (φ) = V0 (φ) + Gt (φ) for every t ∈ [0, T ], where
the gains process G(φ) is defined as follows, for every t ∈ [0, T ],
Z k Z
X
Gt (φ) = φ0u dDu + φiu dSui .
]0,t] i=0 ]0,t]

Corollary 2.3.1 Let S k = B. Then for any self-financing trading strategy φ, the discounted wealth
process V ∗ (φ) = B −1 V (φ) is a martingale under Q.

Proof. Since B is a continuous process of finite variation, the Itô product rule yields dSti∗ =
Sti dBt−1 + Bt−1 dSti for i = 0, 1, . . . , k. Consequently,

dVt∗ (φ) = Vt (φ) dBt−1 + Bt−1 dVt (φ)

³X k ´
−1 −1
= Vt (φ) dBt + Bt φit dSti + φ0t dDt
i=0
k
X ¡ ¢
= φit Sti dBt−1 + Bt−1 dSti + φ0t Bt−1 dDt
i=0
k−1
X ¡ ¢ k−1
X
= φit dSti∗ + φ0t dSt∗ + Bt−1 dDt = φit dSti∗ + φ0t dStc∗ ,
i=1 i=1
2.3. PRICING OF GENERAL DEFAULTABLE CLAIMS 59

where the auxiliary process S c∗ is given by the following expression

Z
Stc∗ = St∗ + Bu−1 dDu .
]0,t]

To conclude, it suffices to observe that in view of (2.30) the process S c∗ satisfies

³Z ¯ ´
¯
Stc∗ = EQ Bu−1 dDu ¯ Gt , (2.32)
]0,T ]

and thus it is a martingale under Q. ¤

It is worth noting that Stc∗ , as given by formula (2.32), represents the discounted cumulative
price at time t of the 0th asset, that is, the arbitrage price at time t of all past and future dividends
associated with the 0th asset over its lifespan. To check this, let us consider a buy-and-hold strategy
such that ψ0k = 0. Then, in view of (2.29), the terminal wealth at time T of this strategy equals
Z
VT (ψ) = BT Bu−1 dDu .
]0,T ]

It is clear that VT (ψ) represents all dividends from S in the form of a single payoff at time T . The
arbitrage price πt (Yb ) at time t ∈ [0, T [ of the claim Yb = VT (ψ) equals (under the assumption that
this claim is attainable)
³Z ¯ ´
¯
πt (Yb ) = Bt EQ Bu−1 dDu ¯ Gt
]0,T ]

and thus Stc∗

= Bt−1 πt (Yb ).
It is clear that discounted cumulative price follows a martingale under
Q (under the standard integrability assumption).

Remarks 2.3.1 (i) Under the assumption of uniqueness of a spot martingale measure Q, any Q-
integrable contingent claim is attainable, and the valuation formula established above can be justified
by means of replication.
(ii) Otherwise – that is, when a martingale probability measure Q is not uniquely determined by
the model (S 1 , S 2 , . . . , S k ) – the right-hand side of (2.30) may depend on the choice of a particular
martingale probability, in general. In this case, a process defined by (2.30) for an arbitrarily chosen
spot martingale measure Q can be taken as the no-arbitrage price process of a defaultable claim. In
some cases, a market model can be completed by postulating that S is also a traded asset.

2.3.4 Martingale Properties of Arbitrage Prices

In the next result, we summarize the martingale properties of arbitrage prices of a generic defaultable
claim.

Corollary 2.3.2 The discounted cumulative price process (Stc∗ , t ∈ [0, T ]) of a defaultable claim is
a Q-martingale with respect to G. The discounted ex-dividend price (St∗ , t ∈ [0, T ]) satisfies, for
every t ∈ [0, T ], Z
St∗ = Stc∗ − Bu−1 dDu
]0,t]

and thus it follows a supermartingale under Q if and only if the dividend process D is increasing.

In an application considered in Section 2.4, the finite variation process A is interpreted as the
positive premium paid in instalments by the claim-holder to the counterparty in exchange for a pos-
itive recovery. It is thus natural to assume that A is a decreasing process, whereas other components
of the dividend process are increasing processes (that is, X ≥ 0 and Z ≥ 0). It is rather clear that,
60 CHAPTER 2. HAZARD FUNCTION APPROACH

under these assumptions, the discounted ex-dividend price S ∗ is neither a super- nor submartingale
under Q, in general.
Assume now that A = 0, so that the premium for a defaultable claim is paid upfront at time
0 and it is not accounted for in the dividend process D. We postulate, as before, that X ≥ 0 and
Z ≥ 0. In this case, the dividend process D is manifestly increasing and thus the discounted ex-
dividend price S ∗ is a supermartingale under Q. This feature is quite natural since the discounted
expected value of future dividends decreases when time elapses.

2.4 Single Name Credit Derivatives

Following Bielecki et al. [18], we will now apply the general theory to a widely particular class
of credit derivatives, namely, to credit default swaps. We do not need to specify explicitly the
underlying market model at this stage, but we make the following standing assumption.

Assumption 2.4.1 We assume throughout that:

• the underlying probability measure Q represents a spot martingale measure on (Ω, HT ),
• the short-term interest rate r = 0, so that Bt = 1 for every t ∈ R+ .

2.4.1 Stylized Credit Default Swap

A stylized T -maturity credit default swap is formally introduced through the following definition.

Definition 2.4.1 A credit default swap (CDS) with a constant rate κ and protection at default is
a defaultable claim (0, A, Z, τ ) where Z(t) = δ(t) and A(t) = −κt for every t ∈ [0, T ]. A function
δ : [0, T ] → R represents the default protection whereas κ is the CDS spread (also termed the rate,
premium or annuity of a CDS).

As usual, we denote by F the cumulative distribution function of default time τ under Q and we
assume that F is a continuous function, with F (0) = 0 and F (T ) < 1. Also, we write G = 1 − F
to denote the survival probability function of τ , so that the inequality G(t) > 0 is valid for every
t ∈ [0, T ].
Since we start with only one traded asset in our model (the savings account), it is clear that
any probability measure Q e on (Ω, HT ) equivalent to Q can be chosen as a spot martingale measure.
The choice of Q is reflected in the cumulative distribution function F (in particular, in the default
intensity if F admits a probability density function). Note that in practical applications of reduced-
form models, the choice of F is done by calibration.
Since the ex-dividend price of a CDS is the price at which the contract is actually traded, we
shall refer to the ex-dividend price as the price in what follows. Recall that we have also introduced
the cumulative price, which encompasses also all past payoffs from a CDS, assumed to be reinvested
in the savings account.
Let s ∈ [0, T ] be a fixed date. We consider a stylized T -maturity credit default swap with a
constant spread κ and default protection function δ, initiated at time s and maturing at T .
The dividend process of a CDS equals
Z Z
Dt = δ(u) dHu − κ (1 − Hu ) du (2.33)
]s,t] ]s,t]

and thus, in view of (2.30), the ex-dividend price of this contract equals, for t ∈ [s, T ],
³ ¯ ´ ³ ¡ ¢ ¯¯ ´
¯
St (κ, δ, T ) = EQ 1{t<τ ≤T } δ(τ ) ¯ Ht − EQ 1{t<τ } κ (τ ∧ T ) − t ¯ Ht ,
2.4. SINGLE NAME CREDIT DERIVATIVES 61

where the first conditional expectation represents the current value of the default protection stream
(or simply the protection leg) and the second expectation is the value of the survival annuity stream
(or the fee leg). To alleviate notation, we shall write St (κ) instead of St (κ, δ, T ) in what follows.

Lemma 2.4.1 The ex-dividend price at time t ∈ [s, T ] of a credit default swap started at s, with
spread κ and default protection δ, equals
Z Z
1 ³ T T ´
St (κ) = 1{t<τ } − δ(u) dG(u) − κ G(u) du . (2.34)
G(t) t t

Proof. We have, on the event {t < τ },

RT Ã RT !
δ(u) dG(u)
t
− t u dG(u) + T G(T )
St (κ) = − −κ −t
G(t) G(t)
µ Z T ³ Z T ´¶
1
= − δ(u) dG(u) − κ T G(T ) − tG(t) − u dG(u) .
G(t) t t

Since Z Z
T T
G(u) du = T G(T ) − tG(t) − u dG(u),
t t

we conclude that (2.34) holds. ¤

The pre-default price is defined as the unique function S(κ)e such that we have, for every t ∈ [0, T ]
(see Lemma 2.5.1 with n = 1)
St (κ) = 1{t<τ } Set (κ). (2.35)
Combining (2.34) with (2.35), we find that the pre-default price of the CDS equals, for t ∈ [s, T ],
Z Z
1 ³ T T ´
Set (κ) = − δ(u) dG(u) − κ G(u) du (2.36)
G(t) t t

so that Set (κ) = Pe(t, T ) − κA(t,

e T ), where
Z T
1
Pe(t, T ) = − δ(u) dG(u)
G(t) t

is the pre-default price at time t of the protection leg, and

Z T
e T) = 1
A(t, G(u) du
G(t) t

represents the pre-default price at time t of the fee leg for the period [t, T ] per one unit of the CDS
e T ) as the CDS annuity (it is also known as the present
spread κ. We shall refer henceforth to A(t,
value of one basis point of a CDS). Note that, under our standing assumption that the survival
e
function G is continuous, the pre-default price S(κ) is a continuous function.

2.4.2 Market CDS Spread

A CDS that has null value at its inception plays an important role as a benchmark CDS and thus
we introduce a formal definition, in which it is implicitly assumed that a protection function δ of a
CDS is given and that we are on the event {s < τ }, that is, the default of the reference name has
not yet occurred prior to or at time s.
62 CHAPTER 2. HAZARD FUNCTION APPROACH

Definition 2.4.2 A market CDS started at s is the CDS initiated at time s whose initial value is
equal to zero. The T -maturity market CDS spread (also known as the fair CDS spread) at time s is
the fixed level of the spread κ = κ(s, T ) that makes the T -maturity CDS started at s valueless at
its inception. The market CDS spread at time s is thus determined by the equation Ses (κ(s, T )) = 0
where Ses (κ) is given by the formula (2.36).

Under the present assumptions, by virtue of (2.36), the T -maturity market CDS spread κ(s, T )
equals, for every s ∈ [0, T ],
RT
Pe(s, T ) δ(u) dG(u)
κ(s, T ) = = − sR T . (2.37)
e
A(s, T ) G(u) du
s

Example 2.4.1 Assume that δ(t) = δ is constant, and F (t) = 1 − e−γt for some constant default
intensity γ > 0 under Q. In that case, the valuation formulae for a CDS can be further simplified. In
view of Lemma 2.4.1, the ex-dividend price of a (spot) CDS with spread κ equals, for every t ∈ [0, T ],
³ ´
St (κ) = 1{t<τ } (δγ − κ)γ −1 1 − e−γ(T −t) .

The last formula (or the general formula (2.37)) yields κ(s, T ) = δγ for every s < T , so that the
market spread κ(s, T ) is here independent of s. As a consequence, the ex-dividend price of a market
CDS started at s equals zero not only at the inception date s, but indeed at any time t ∈ [s, T ], both
prior to and after default. Hence this process is trivially a martingale under Q. As we shall see in
what follows, this martingale property of the ex-dividend price of a market CDS is an exception, in
the sense so that it fails to hold if the default intensity varies over time.

In what follows, we fix a maturity date T and we assume that credit default swaps with different
inception dates have a common default protection δ. We shall write briefly κ(s) instead of κ(s, T ).
Then we have the following result, in which the quantity ν(t, s) = κ(t) − κ(s) represents the calendar
CDS market spread for a given maturity T .

Proposition 2.4.1 The price of a market CDS started at s with protection δ at default and maturity
T equals, for every t ∈ [s, T ],

e T ) = 1{t<τ } ν(t, s)A(t,

St (κ(s)) = 1{t<τ } (κ(t) − κ(s)) A(t, e T ). (2.38)

Proof. It suffices to observe that St (κ(s)) = St (κ(s)) − St (κ(t)), since St (κ(t)) = 0, and to use (2.36)
with κ = κ(t) and κ = κ(s). ¤

Note that formula (2.38) can be extended to any value of κ, specifically,

e T ),
St (κ) = 1{t<τ } (κ(t) − κ)A(t,

assuming that the CDS with spread κ was initiated at some date s ∈ [0, t]. The last representation
shows that the price of a CDS can take negative values. The negative value occurs whenever the
current market spread is lower than the contracted spread.

2.4.3 Price Dynamics of a CDS

In the remainder of Section 2.4, we assume that the hazard function satisfies, for every t ∈ [0, T ],
µ Z t ¶
G(t) = Q(τ > t) = exp − γ(u) du ,
0
2.4. SINGLE NAME CREDIT DERIVATIVES 63

where the default intensity γ(t) under Q is a strictly positive deterministic function. Recall that the
process M , given by the formula, for every t ∈ [0, T ],
Z t
Mt = Ht − (1 − Hu )γ(u) du, (2.39)
0

is an H-martingale under Q.
We first focus on the dynamics of the price of a CDS, with spread κ, which was initiated at some
date s < T .

Lemma 2.4.2 (i) The dynamics of the price St (κ), t ∈ [s, T ], are

dSt (κ) = −St− (κ) dMt + (1 − Ht )(κ − δ(t)γ(t)) dt. (2.40)

(ii) The cumulative price process Stc (κ), t ∈ [s, T ], is an H-martingale under Q, specifically,
¡ ¢
dStc (κ) = δ(t) − St− (κ) dMt . (2.41)

Proof. To prove (i), it suffices to recall that

St (κ) = 1{t<τ } Set (κ) = (1 − Ht )Set (κ)

so that the integration by parts formula yields

dSt (κ) = (1 − Ht ) dSet (κ) − Set− (κ) dHt .

Using formula (2.34), we find easily that

dSet (κ) = γ(t)Set (κ) dt + (κ − δ(t)γ(t)) dt.

In view of (2.39) and the fact that Sτ − (κ) = Seτ − (κ) and St (κ) = 0 for t ≥ τ , the proof of (2.40) is
complete.
To prove part (ii), we note that (2.30) and (2.31) yield

Stc (κ) − Ssc (κ) = St (κ) − Ss (κ) + Dt − Ds .

Consequently,
Z Z t
Stc (κ) − Ssc (κ) = St (κ) − Ss (κ) + δ(u) dHu − κ (1 − Hu ) du
]s,t] s
Z Z t
= St (κ) − Ss (κ) + δ(u) dMu − (1 − Hu )(κ − δ(u)γ(u)) du
]s,t] s
Z
¡ ¢
= δ(u) − Su− (κ) dMu
]s,t]

where the last equality follows from (2.40). ¤

Equality (2.40) emphasizes the fact that a single cash flow of δ(τ ) occurring at time τ can be
formally treated as a dividend stream at the rate δ(t)γ(t) paid continuously prior to default. It is
clear that we also have

dSt (κ) = −Set− (κ) dMt + (1 − Ht )(κ − δ(t)γ(t)) dt.

64 CHAPTER 2. HAZARD FUNCTION APPROACH

2.4.4 Replication of a Defaultable Claim

Our goal is to show that, in order to replicate a general defaultable claim, it suffices to trade
dynamically in two assets: a CDS maturing at T and the savings account B, assumed here to be
constant. Since one may always work with discounted values, the last assumption is not restrictive.
Moreover, it is also possible to take a CDS with any maturity U ≥ T .
Let φ0 , φ1 be H-predictable processes and let C : [0, T ] → R be a function of finite variation with
C(0) = 0. We say that (φ, C) = (φ0 , φ1 , C) is a self-financing trading strategy with dividend stream
C if the wealth process V (φ, C), defined as

Vt (φ, C) = φ0t + φ1t St (κ),

where St (κ) is the price of a CDS at time t, satisfies

¡ ¢
dVt (φ, C) = φ1t dSt (κ) + dDt − dC(t) = φ1t dStc (κ) − dC(t),

where the dividend process D of a CDS is in turn given by (2.33). Note that C represents both
outflows and infusions of funds. It will be used to cover the running cash flows associated with a
claim we wish to replicate.
Consider a defaultable claim (X, A, Z, τ ) where X is a constant, A is a continuous function of
finite variation, and Z is some recovery function. In order to define replication of a defaultable claim
(X, A, Z, τ ), it suffices to consider trading strategies on the random interval [0, τ ∧ T ].

Definition 2.4.3 We say that a trading strategy (φ, C) replicates a defaultable claim (X, A, Z, τ )
if:
(i) the processes φ = (φ0 , φ1 ) and V (φ, C) are stopped at τ ∧ T ,
(ii) C(τ ∧ t) = A(τ ∧ t) for every t ∈ [0, T ],
(iii) the equality Vτ ∧T (φ, C) = Y holds, where the random variable Y equals

Y = X1{τ >T } + Z(τ )1{τ ≤T } . (2.42)

Remark 2.4.1 Alternatively, one may say that a self-financing trading strategy φ = (φ, 0) (i.e., a
trading strategy with C = 0) replicates a defaultable claim (X, A, Z, τ ) if and only if Vτ ∧T (φ) = Yb ,
where we set
Yb = X1{τ >T } + A(τ ∧ T ) + Z(τ )1{τ ≤T } . (2.43)
However, in the case of non-zero (possibly random) interest rates, it is more convenient to define
replication of a defaultable claim via Definition 2.4.3, since the running payoffs specified by A are
distributed over time and thus, in principle, they need to be discounted accordingly (this does not
show in (2.43), since it is assumed here that r = 0).

Let us denote, for every t ∈ [0, T ],

µ Z T ¶
e = 1
Z(t) XG(T ) − Z(u) dG(u)
G(t) t

and
Z T
e = 1
A(t) G(u) dA(u).
G(t) t

Let π and π e be the risk-neutral value and the pre-default risk-neutral value of a defaultable claim
under Q, so that πt = 1{t<τ } πe(t) for every t ∈ [0, T ]. Also, let π
b stand for its risk-neutral cumulative
price. It is clear that the equalities π
e(0) = π(0) = π b(0) = EQ (Yb ) are valid.
2.4. SINGLE NAME CREDIT DERIVATIVES 65

Proposition 2.4.2 The pre-default risk-neutral value of a defaultable claim (X, A, Z, τ ) equals
π e + A(t)
e(t) = Z(t) e for every t ∈ [0, T [ (clearly, π
e(T ) = 0). Therefore, for every t ∈ [0, T [,

de
π (t) = γ(t)(e
π (t) − Z(t)) dt − dA(t). (2.44)

Moreover
dπt = −e
π (t−) dMt − γ(t)(1 − Ht )Z(t) dt − dA(t ∧ τ ) (2.45)
and
db
πt = (Z(t) − π
e(t−)) dMt .

Proof. The proof of equality π e + A(t)

e(t) = Z(t) e is similar to the derivation of formula (2.36). We
have, for t ∈ [0, T [,
³ ¯ ´
¯
πt = EQ 1{t<τ } Y + A(τ ∧ T ) − A(τ ∧ t) ¯ Ht
Z T
1 ³ ´
= 1{t<τ } XG(T ) − Z(u) dG(u)
G(t) t
Z T
1
+ 1{t<τ } G(u) dA(u)
G(t) t
e + A(t))
= 1{t<τ } (Z(t) e = 1{t<τ } π
e(t).

By elementary computations, we obtain the following equalities

e = γ(t)(Z(t)
dZ(t) e − Z(t)) dt

and
e = γ(t)A(t)
dA(t) e dt − dA(t),

so that (2.44) holds. Formula (2.45) follows easily from (2.44) and the integration by parts formula
applied to the equality πt = (1 − Ht )e
π (t) (see the proof of Lemma 2.4.2 for similar computations).
The last formula is also easy to check. ¤

The next proposition shows that the risk-neutral value of a defaultable claim is also its replication
price, that is, a defaultable claim derives its value from the price of the traded CDS.

Theorem 2.4.1 Assume that the inequality Set (κ) 6= δ(t) holds for every t ∈ [0, T ]. Let φ1t =
φe1 (τ ∧ t), where the function φe1 : [0, T ] → R is given by the formula

Z(t) − π
e(t−)
φe1 (t) = (2.46)
δ(t) − Set (κ)

and let φ0t = Vt (φ, A) − φ1t St (κ), where the process V (φ, A) is given by the formula
Z
Vt (φ, A) = πe(0) + φe1 (u) dSuc (κ) − A(t ∧ τ ). (2.47)
]0,τ ∧t]

Then the strategy (φ0 , φ1 , A) replicates the defaultable claim (X, A, Z, τ ).

Proof. Assume first that a trading strategy φ = (φ0 , φ1 , C) is a replicating strategy for (X, A, Z, τ ).
By virtue of condition (i) in Definition 2.4.3 we have C = A and thus, by combining (2.47) with
(2.41), we obtain
dVt (φ, A) = φ1t (δ(t) − Set (κ)) dMt − dA(τ ∧ t)
For φ1 given by (2.46), we thus obtain

dVt (φ, A) = (Z(t) − π

e(t−)) dMt − dA(τ ∧ t).
66 CHAPTER 2. HAZARD FUNCTION APPROACH

It is thus clear that if we take φ1t = φe1 (τ ∧ t) with φe1 given by (2.46), and the initial condition
V0 (φ, A) = π
e(0) = π0 , then we have that Vt (φ, A) = πe(t) for every t ∈ [0, T [ on the event {t < τ }.
By examining, in particular, the jump of the wealth process V (φ, A) at the moment of default, one
may check that all conditions of Definition 2.4.3 are indeed satisfied. ¤

Remark 2.4.2 Of course, if we take as (X, A, Z, τ ) a CDS with spread κ and protection function
δ, then we have Z(t) = δ(t) and π e(t) = Set (κ), so that φ1t = 1 for every t ∈ [0, T ].
e(t−) = π

2.5 Basket Credit Derivatives

In this section, we shall examine hedging of first-to-default basket claims with single name credit
default swaps on the underlying n credit names, denoted as 1, 2, . . . , n. The standing Assumption
2.4.1 is maintained throughout this section.
Let the random times τ1 , τ2 , . . . , τn , given on a common probability space (Ω, G, Q), represent
the default times of n reference credit names. We denote by

τ(1) = τ1 ∧ τ2 ∧ . . . ∧ τn = min (τ1 , τ2 , . . . , τn )

the moment of the first default, so that no defaults are observed on the event {t < τ(1) }. Let

F (t1 , t2 , . . . , tn ) = Q(τ1 ≤ t1 , τ2 ≤ t2 , . . . , τn ≤ tn )

be the joint probability distribution function of default times. We assume that the probability
distribution of default times is jointly continuous, and we write f (t1 , t2 , . . . , tn ) to denote the joint
probability density function. Also, let

G(t1 , t2 , . . . , tn ) = Q(τ1 > t1 , τ2 > t2 , . . . , τn > tn )

stand for the joint probability that the names 1, 2, . . . , n have survived up to times t1 , t2 , . . . , tn . In
particular, the joint survival function is given by the formula, for every t ∈ R+ ,

G(t, . . . , t) = Q(τ1 > t, τ2 > t, . . . , τn > t) = Q(τ(1) > t) = G(1) (t).

For i = 1, 2, . . . , n, we define the default indicator process Hti = 1{t≥τi } and the corresponding
filtration Hi = (Hti )t∈R+ where Hti = σ(Hui : u ≤ t). We denote by G the joint filtration generated
by default indicator processes H 1 , H 2 , . . . , H n , so that G = H1 ∨ H2 ∨ . . . ∨ Hn . It is clear that τ(1)
is a G-stopping time as the infimum of G-stopping times.
(1) (1)
Finally, we define the process Ht = 1{t≥τ(1) } and the associated filtration H(1) = (Ht )t∈R+
(1) (1)
where Ht = σ(Hu : u ≤ t).
Since we postulate that Q(τi = τj ) = 0 for any i 6= j, i, j = 1, 2, . . . , n, we also have that
n
X
(1) (1) i
Ht = Ht∧τ(1) = Ht∧τ(1)
.
i=1

We now fix a finite horizon date T > 0, and we make the standing assumption that

G(1) (T ) = Q(τ(1) > T ) > 0.

For any t ∈ [0, T ], the event {t < τ(1) } is an atom of the σ-field Gt . Hence the following simple,
but useful, result.

Lemma 2.5.1 Let X be a Q-integrable stochastic process on (Ω, G, Q). Then

e
1{t<τ(1) } EQ (Xt | Gt ) = 1{t<τ(1) } X(t),
2.5. BASKET CREDIT DERIVATIVES 67

e : [0, T ] → R is given by the formula

where the function X
¡ ¢
EQ 1{t<τ(1) } Xt
e
X(t) = .
G(1) (t)

If X is a G-adapted, Q-integrable stochastic process then, for every t ∈ [0, T ],

e + 1{t≥τ } Xt .
Xt = 1{t<τ(1) } X(t) (1)

e : [0, T ] → R is called the pre-default value of the process X.

By convention, the function X

2.5.1 First-to-Default Intensities

In this section, we introduce the notion of the first-to-default intensity. This natural concept will
prove useful in the valuation and hedging of a first-to-default claim.

ei : R+ → R+ , given by the formula

Definition 2.5.1 The function λ

ei (t) = lim 1 Q(t < τi ≤ t + h | τ(1) > t),

λ
h↓0 h

e : R+ → R+ , given by
is called the ith first-to-default intensity. The function λ

e = lim 1 Q(t < τ(1) ≤ t + h | τ(1) > t),

λ(t) (2.48)
h↓0 h

is called the first-to-default intensity.

Let us denote
∂G(t1 , t2 , . . . , tn )
∂i G(t, . . . , t) = ¯ .
∂ti ¯t1 =t2 =...=tn =t
Then we have the following elementary lemma summarizing the properties of first-to-default inten-
ei and λ.
sities λ e

Lemma 2.5.2 The ith first-to-default intensity λ ei satisfies

R∞ R∞
e t
. . . t f (u1 , . . . , ui−1 , t, ui+1 , . . . , un ) du1 . . . dui−1 dui+1 . . . dun
λi (t) =
G(t, . . . , t)
R∞ R∞
. . . t F (du1 , . . . , dui−1 , t, dui+1 , . . . , dun ) ∂i G(t, . . . , t)
= t =− .
G(1) (t) G(1) (t)

e satisfies
The first-to-default intensity λ

e =− 1 dG(1) (t) f(1) (t)

λ(t) = ,
G(1) (t) dt G(1) (t)

e
where f(1) (t) is the probability density function of the random time τ(1) . The equality λ(t) =
Pn e
i=1 λi (t) holds for every t ∈ R+ .

Proof. Clearly
R∞ R t+h R∞
ei (t) = lim 1 t
... t
... t
f (u1 , . . . , ui , . . . , un ) du1 . . . dui . . . dun
λ
h↓0 h G(t, . . . , t)
68 CHAPTER 2. HAZARD FUNCTION APPROACH

and thus the first asserted formula follows. The second equality follows directly from (2.48) and the
e = Pn λ
definition of the joint survival function G(1) . Finally, equality λ(t) e
i=1 i (t) is equivalent to the
equality
n
1 X 1
lim Q(t < τi ≤ t + h | τ(1) > t) = lim Q(t < τ(1) ≤ t + h | τ(1) > t),
h↓0 h h↓0 h
i=1
which in turn is easy to establish. ¤

Remarks 2.5.1 The ith first-to-default intensity λei should not be confused with the marginal
intensity function λi of τi , which is defined as
fi (t)
λi (t) = , ∀ t ∈ R+ ,
Gi (t)
where fi is the marginal probability density function of τi , that is,
Z ∞ Z ∞
fi (t) = ... f (u1 , . . . , ui−1 , t, ui+1 , . . . , un ) du1 . . . dui−1 dui+1 . . . dun
0 0
R∞
and where Gi (t) = 1 − Fi (t) = t fi (u) du. Indeed, we have that λ ei 6= λi , in general. However, if
ei = λi , that is, the first-to-default and marginal
τ1 , . . . , τn are mutually independent under Q then λ
default intensities coincide.

It is also rather clear that the first-to-default e is not equal to the sum of marginal
intensity λ
e 6= Pn λi (t), in general.
default intensities, that is, we have that λ(t) i=1

2.5.2 First-to-Default Representation Theorem

We will now prove an integral representation theorem for any G-martingale stopped at τ(1) with
respect to some finite collection of G-martingales stopped at τ(1) . To this end, we define, for every
i = 1, 2, . . . n, Z t∧τ(1)
cti = Ht∧τ
M i
− ei (u) du,
λ ∀ t ∈ R+ . (2.49)
(1)
0
Then we have the following result, referred to as the first-to-default predictable representation theo-
rem.

ct = EQ (Y | Gt ), t ∈ [0, T ], where Y is a Q-integrable

Proposition 2.5.1 Consider the G-martingale M
random variable given by the expression
n
X
Y = Zi (τi )1{τi ≤T, τi =τ(1) } + X1{τ(1) >T } (2.50)
i=1

c admits the
for some functions Zi : [0, T ] → R, i = 1, 2, . . . , n and some constant X. Then M
following representation
X n Z
ct = EQ (Y ) +
M cui
hi (u) dM (2.51)
i=1 ]0,t]

where the functions hi , i = 1, 2, . . . , n are given by

ct− = Zi (t) − M
hi (t) = Zi (t) − M f(t−), ∀ t ∈ [0, T ], (2.52)

where Mf is the unique function such that M

ct 1{t<τ } = M
f(t)1{t<τ } for every t ∈ [0, T ]. The
(1) (1)
f satisfies M
function M f0 = EQ (Y ) and
n
X ¡ ¢
f(t) =
dM ei (t) M
λ f(t) − Zi (t) dt. (2.53)
i=1
2.5. BASKET CREDIT DERIVATIVES 69

More explicitly,
³Z t ´ Z tX
n ³Z t ´
f(t) = EQ (Y ) exp
M e ds −
λ(s) ei (s)Zi (s) exp
λ e
λ(u) du ds.
0 0 i=1 s

Proof. To alleviate notation, we provide the proof of this result in a bivariate setting only, so that
τ(1) = τ1 ∧ τ2 and Gt = Ht1 ∨ Ht2 . We start by noting that

ct = EQ (Z1 (τ1 )1{τ ≤T, τ >τ } | Gt ) + EQ (Z2 (τ2 )1{τ ≤T, τ >τ } | Gt )
M 1 2 1 2 1 2

+ EQ (X1{τ(1) >T } | Gt ),

and thus (see Lemma 2.5.1)

3
X
ct = 1{t<τ } M
1{t<τ(1) } M f(t) = 1{t<τ } Ye i (t)
(1) (1)
i=1

where the auxiliary functions Ye i : [0, T ] → R, i = 1, 2, 3, are given by

RT R∞
e 1 t
duZ1 (u) u dvf (u, v)
Y (t) = ,
G(1) (t)
RT R∞
e 2 t
dvZ2 (v) v duf (u, v)
Y (t) = ,
G(1) (t)
XG(1) (T )
Ye 3 (t) = .
G(1) (t)
By elementary calculations and using Lemma 2.5.2, we obtain
R∞ RT R∞
dYe 1 (t) Z1 (t) t dvf (t, v) t
duZ1 (u) u dvf (u, v) dG(1) (t)
= − −
dt G(1) (t) G2(1) (t) dt
R∞
dvf (t, v) Ye (t) dG(1) (t)
1
= −Z1 (t) t −
G(1) (t) G(1) (t) dt
e e 1 e e
= −Z1 (t)λ1 (t) + Y (t)(λ1 (t) + λ2 (t)), (2.54)

and thus, by symmetry,

dYe 2 (t) e2 (t) + Ye 2 (t)(λ

e1 (t) + λ
e2 (t)).
= −Z2 (t)λ (2.55)
dt
Moreover,
dYe 3 (t) XG(1) (T ) dG(1) (t) e1 (t) + λ
e2 (t)).
=− = Ye 3 (t)(λ (2.56)
dt G2(1) (t) dt

Therefore, recalling that Mf(t) = P3 Ye i (t), we obtain from (2.54)–(2.56)

i=1
¡ ¢ ¡ ¢
f(t) = −λ
dM e1 (t) Z1 (t) − M
f(t) dt − λe2 (t) Z2 (t) − M f(t) dt. (2.57)

Consequently, since the function M f is continuous, we have, on the event {τ(1) > t},
¡ ¢ ¡ ¢
dMct = −λ e1 (t) Z1 (t) − M e2 (t) Z2 (t) − M
ct− dt − λ ct− dt.

We shall now check that both sides of equality (2.51) coincide at time τ(1) on the event {τ(1) ≤ T }.
To this end, we note that, on the event {τ(1) ≤ T },

cτ = Z1 (τ1 )1{τ =τ } + Z2 (τ2 )1{τ =τ } ,

M (1) (1) 1 (1) 2
70 CHAPTER 2. HAZARD FUNCTION APPROACH

whereas the right-hand side in (2.51) is equal to

Z Z
c
M0 + c 1
h1 (u) dMu + cu2
h2 (u) dM
]0,τ(1) [ ]0,τ(1) [
Z Z
+ 1{τ(1) =τ1 } h1 (u) dHu1 + 1{τ(1) =τ2 } h2 (u) dHu2
[τ(1) ] [τ(1) ]
¡ ¢
=M f(τ(1) −) + Z1 (τ1 ) − M f(τ(1) −) 1{τ =τ }
(1) 1
¡ ¢
+ Z2 (τ2 ) − Mf(τ(1) −) 1{τ =τ }
(1) 2

= Z1 (τ1 )1{τ(1) =τ1 } + Z2 (τ2 )1{τ(1) =τ2 }

f(τ(1) −) = M
as M cτ − . Since the processes on both sides of equality (2.51) are stopped at time τ(1) ,
(1)
we conclude that equality (2.51) is valid for every t ∈ [0, T ]. Let us finally observe that formula
(2.53) was also established in the proof (see formula (2.57)). ¤
ci are in fact G-martingales. They will be referred to
The next result shows that the processes M
as the basic first-to-default martingales.

ci given by the formula (2.49) is a G-

Corollary 2.5.1 For each i = 1, 2, . . . , n, the process M
martingale stopped at time τ(1) .

Proof. Let us fix k ∈ {1, 2, . . . , n}. It is clear that the process Mck is stopped at τ(1) . Note that
Z t
Mfk (t) = − ei (u) du
λ
0

is the unique function such that, for every t ∈ [0, T ],

cti = 1{t<τ } M
1{t<τ(1) } M fk (t).
(1)

Let us take hk (t) = 1 and hi (t) = 0 for any i 6= k in formula (2.51) or, equivalently, let us set
fk (t),
Zk (t) = 1 + M fk (t), i 6= k,
Zi (t) = M

in the definition (2.50) of the random variable Y . Finally, let the constant X in (2.50) be chosen
in such a way that the random variable Y satisfies EQ (Y ) = Mck . Then we may deduce from (2.51)
0
ck = M
that M c and thus we conclude that M ck is a G-martingale. ¤

2.5.3 Price Dynamics of Credit Default Swaps

As primary traded assets in the market model under consideration, we take the constant savings ac-
count and a family of single-name CDSs with default protections δi and spreads κi for i = 1, 2, . . . , n.
For convenience, we assume that the CDSs have the same maturity T , but this assumption can
be easily relaxed. The ith traded CDS is formally defined by its dividend process (Dti , t ∈ [0, T ]),
which is given by the formula
Z
Dti = δi (u) dHui − κi (t ∧ τi ).
]0,t]

Consequently, the price at time t of the ith CDS equals

¡ ¡ ¢¯ ¢
Sti (κi ) = EQ (1{t<τi ≤T } δi (τi ) | Gt ) − κi EQ 1{t<τi } (τi ∧ T ) − t ¯ Gt .

To replicate a first-to-default claim, we only need to examine the dynamics of each CDS on the
interval [0, τ(1) ∧ T ]. The following lemma will prove useful in this regard.
2.5. BASKET CREDIT DERIVATIVES 71

Lemma 2.5.3 We have, on the event {t < τ(1) },

³ X ¯ ´
¯
Sti (κi ) = EQ 1{t<τ(1) =τi ≤T } δi (τ(1) ) + 1{t<τ(1) =τj ≤T } Sτi(1) (κi ) ¯ Gt
j6=i
³ ¯ ´
¯
− EQ κi 1{t<τ(1) } (τ(1) ∧ T − t) ¯ Gt .

Proof. We first note that the price Sti (κi ) can be represented as follows, on the event {t < τ(1) },
³ ¯ ´
¯
Sti (κi ) = EQ 1{t<τ(1) =τi ≤T } δi (τ(1) ) ¯ Gt
X ³ ¯ ´
¯
+ EQ 1{t<τ(1) =τj ≤T } 1{τ(1) <τi ≤T } δi (τi ∧ T ) ¯ Gt
j6=i
X ³ ¯ ´
¯
− κi EQ 1{t<τ(1) =τj ≤T } 1{τ(1) <τi } (τi − τ(1) ) ¯ Gt
j6=i
¡ ¯ ¢
− κi EQ 1{t<τ(1) } (τ(1) ∧ T − t) ¯ Gt .
By conditioning first on the σ-field Gτ(1) , we obtain the claimed formula. ¤

The representation established in Lemma 2.5.3 is by no means surprising; it merely shows that
in order to compute the price of a CDS prior to the first default, we can either do the computations
in a single step, by considering the cash flows occurring on ]t, τi ∧ T ] or, alternatively, we can first
compute the price of the contract at time τ(1) ∧ T and subsequently value all cash flows occurring
between t and τ(1) ∧ T .
In view of Lemma 2.5.3, we can argue that in what follows, instead of considering the original ith
CDS maturing at T , we can deal with the corresponding synthetic CDS contract with the random
maturity τ(1) ∧ T .
Similarly as in Section 2.4.1, we will write Sti (κi ) = 1{t<τ(1) } Seti (κi ), where the pre-default price
Seti (κi ) satisfies
Seti (κi ) = Pei (t, T ) − κi A
ei (t, T ),
where Pei (t, T ) and κi A
ei (t, T ) stand for the pre-default values of the protection leg and the fee leg,
respectively.
i
For any j 6= i, we define a function St|j (κi ) : [0, T ] → R, which represents the price of the ith
CDS at time t on the event {τ(1) = τj = t}. Formally, this quantity is defined as the unique function
satisfying
1{τ(1) =τj ≤T } Sτi(1) (κi ) = 1{τ(1) =τj ≤T } Sτi(1) |j (κi ),
so that X
1{τ(1) ≤T } Sτi(1) (κi ) = 1{τ(1) =τj ≤T } Sτi(1) |j (κi ).
j6=i

1
Let us examine, for instance, the case of two credit names. Then the function St|2 (κ1 ), t ∈ [0, T ],
represents the price of the first CDS at time t on the event {τ(1) = τ2 = t}.
1
Lemma 2.5.4 The function Sv|2 (κ1 ), v ∈ [0, T ], equals
RT RT R∞
1 δ (u)f (u, v)du
vR 1
du u dzf (z, v)
v R
Sv|2 (κ1 ) = ∞ − κ1 ∞ . (2.58)
v
f (u, v) du v
f (u, v) du

Proof. Note that the conditional cumulative distribution function of τ1 given that τ1 > τ2 = v
equals, for u ∈ [v, ∞],
Ru
f (z, v) dz
Q(τ1 ≤ u | τ1 > τ2 = v) = Fτ1 |τ1 >τ2 =v (u) = v∞
R ,
v
f (z, v) dz
72 CHAPTER 2. HAZARD FUNCTION APPROACH

so that the conditional tail equals, for u ∈ [v, ∞],

R∞
f (z, v) dz
Gτ1 |τ1 >τ2 =v (u) = 1 − Fτ1 |τ1 >τ2 =v (u) = Ru∞ .
v
f (z, v) dz
Let J be the right-hand side of (2.58). It is clear that
Z T Z T
J =− δ1 (u) dGτ1 |τ1 >τ2 =v (u) − κ1 Gτ1 |τ1 >τ2 =v (u) du.
v v

Combining Lemma 2.4.1 with the fact that Sτ1(1) (κi ) is equal to the conditional expectation with
respect to σ-field Gτ(1) of the cash flows of the ith CDS on ]τ(1) ∨ τi , τi ∧ T ], we conclude that J
1
coincides with Sv|2 (κ1 ), the price of the first CDS on the event {τ(1) = τ2 = v}. ¤

The following result extends Lemma 2.4.2.

Lemma 2.5.5 The dynamics of the pre-default price Seti (κi ) are
³ n
X ´
e Sei (κi ) dt + κi − δi (t)λ
dSeti (κi ) = λ(t) ei (t) − S i
(κ i )λei (t) dt (2.59)
t t|j
j6=i

e = Pn λ
where λ(t) e
i=1 i (t) or, equivalently,
¡ ¢
ei (t) Sei (κi ) − δi (t) dt
dSeti (κi ) = λ (2.60)
t
X ¡ ¢
+ ej (t) Sei (κi ) − S i (κi ) dt + κi dt.
λ t t|j
j6=i

The cumulative price of the ith CDS stopped at τ(1) satisfies

Z
Stc,i (κi ) = Sti (κi ) + i
δi (u) dHu∧τ (1)
(2.61)
]0,t]
X Z
i j
+ Su|j (κi ) dHu∧τ (1)
− κi (τ(1) ∧ t),
j6=i ]0,t]

and thus ¡ ¢ X¡ ¢
dStc,i (κi ) = δi (t) − Set−
i cti +
(κi ) dM i
St|j (κi ) − Set−
i ctj .
(κi ) dM (2.62)
j6=i

Proof. We shall consider the case n = 2. Using the formula derived in Lemma 2.5.3, we obtain
RT R∞ RT 1
R∞
du δ1 (u) u dvf (u, v) t
dv Sv|2 (κ1 ) v duf (u, v)
e 1
P (t, T ) = t
+ .
G(1) (t) G(1) (t)
By adapting equality (2.54), we get
³ ´
e1 (t) + λ
dPe1 (t, T ) = (λ e2 (t))e e1 (t)δ1 (t) − λ
g1 (t) − λ e2 (t)S 1 (κ1 ) dt.
t|2

To establish (2.59)–(2.60), we need also to examine the fee leg. Its price equals
³ ¡ ¢ ¯¯ ´
ei (t, T ),
EQ 1{t<τ(1) } κ1 (τ(1) ∧ T ) − t ¯ Gt = 1{t<τ(1) } κ1 A
To evaluate the conditional expectation above, it suffices to use the cumulative distribution function
F(1) of the random time τ(1) . As in Section 2.4.1 (see the proof of Lemma 2.4.1), we obtain
Z T
ei (t, T ) = 1
A G(1) (u) du, (2.63)
G(1) (t) t
and thus ¡ ¢
dAei (t, T ) = 1 + (λe1 (t) + λ
e2 (t))A
ei (t, T ) dt.
Since Set1 (κ1 ) = Pei (t, T ) − κi A
ei (t, T ), the formulae (2.59) and (2.60) follow. Formula (2.61) is rather
clear. Finally, dynamics (2.62) can be deduced easily from (2.60) and (2.61). ¤
2.5. BASKET CREDIT DERIVATIVES 73

2.5.4 Valuation of a First-to-Default Claim

In this section, we shall analyze the risk-neutral valuation of first-to-default claims on a basket of n
credit names.

Definition 2.5.2 A first-to-default claim (FTDC) with maturity T is a defaultable claim (X, A, Z, τ(1) )
where X is a constant amount payable at maturity if no default occurs, A : [0, T ] → R with A0 = 0
is a continuous function of bounded variation representing the dividend stream up to τ(1) , and
Z = (Z1 , Z2 , . . . , Zn ) is the vector of functions Zi : [0, T ] → R where Zi (τ(1) ) specifies the recovery
received at time τ(1) if the ith name is the first defaulted name, that is, on the event {τi = τ(1) ≤ T }.

We define the risk-neutral value π of an FTDC by setting

Xn ³ Z T ¯ ´
¯
πt = EQ Zi (τi )1{t<τ(1) =τi ≤T } + 1{t<τ(1) } (1 − Hu(1) ) dA(u) ¯ Gt
i=1 t
³ ¯ ´
¯
+ EQ X1{τ(1) >T } ¯ Gt

and the risk-neutral cumulative value π b of an FTDC by the formula

Xn ³ Z T ¯ ´
¯
π
bt = EQ Zi (τi )1{t<τ(1) =τi ≤T } + 1{t<τ(1) } (1 − Hu(1) ) dA(u) ¯ Gt
i=1 t
n Z
X Z t
i
+ EQ (X1{τ(1) >T } |Gt ) + Zi (u) dHu∧τ (1)
+ (1 − Hu(1) ) dA(u)
i=1 ]0,t] 0

where the last two terms represent the past dividends. Let us stress that the risk-neutral valuation of
an FTDC will be later supported by replication arguments (see Theorem 2.5.1) and thus risk-neutral
value π of an FTDC will be shown to be its replication price.
By the pre-default risk-neutral value associated with a G-adapted process π, we mean the function
π
e such that πt 1{t<τ(1) } = π
e(t)1{t<τ(1) } for every t ∈ [0, T ]. Direct calculations lead to the following
result, which can also be deduced from Proposition 2.5.1.

Lemma 2.5.6 The pre-default risk-neutral value of an FTDC equals

Xn Z T
Ψi (t) 1 G(1) (T )
π
e(t) = + G(1) (u) dA(u) + X (2.64)
i=1
G(1) (t) G(1) (t) t G(1) (t)

where
Z T Z ∞ Z ∞ Z ∞ Z ∞
Ψi (t) = ... ... Zi (ui )
ui =t u1 =ui ui−1 =ui ui+1 =ui un =ui
F (du1 , . . . , dui−1 , dui , dui+1 , . . . , dun ).

The next result extends Proposition 2.4.2 to the multi-name setup. Its proof is similar to the
proof of Lemma 2.5.5 and thus it is omitted.

Proposition 2.5.2 The pre-default risk-neutral value of an FTDC satisfies

X ¡ ¢
de
π (t) = ei (t) π
λ e(t) − Zi (t) dt − dA(t).
i=1

Moreover, the risk-neutral value of an FTDC satisfies

n
X
dπt = − π cui − dA(τ(1) ∧ t)
e(t−) dM (2.65)
i=1
74 CHAPTER 2. HAZARD FUNCTION APPROACH

and the risk-neutral cumulative value π

b of an FTDC satisfies
n
X
db
πt = (Zi (t) − π cui .
e(t−)) dM
i=1

2.5.5 Replication of a First-to-Default Claim

Let the savings account with the price B = 1 and single-name credit default swaps with prices
S 1 (κ1 ), . . . , S n (κn ) be primary traded assets. We say that a G-predictable process φ = (φ0 , φ1 , . . . , φn )
and a function C of finite variation with C(0) = 0 define a self-financing strategy with dividend stream
C if the wealth process V (φ, C), defined as
n
X
Vt (φ, C) = φ0t + φit Sti (κi ),
i=1

satisfies
n
X n
X
¡ ¢
dVt (φ, C) = φit dSti (κi ) + dDti − dC(t) = φit dStc,i (κi ) − dC(t) (2.66)
i=1 i=1

where S i (κi ) (S c,i (κi ), respectively) is the price (cumulative price, respectively) of the ith traded
CDS.

Definition 2.5.3 We say that a trading strategy (φ, C) replicates an FTDC (X, A, Z, τ(1) ) whenever
the following conditions are satisfied:
(i) the processes φ = (φ0 , φ1 , . . . , φn ) and V (φ, C) are stopped at τ(1) ∧ T ,
(ii) C(τ(1) ∧ t) = A(τ(1) ∧ t) for every t ∈ [0, T ],
(iii) the equality Vτ(1) ∧T (φ, C) = Y holds, where the random variable Y equals
n
X
Y = X1{τ(1) >T } + Zi (τ(1) )1{τi =τ(1) ≤T } .
i=1

We are now in a position to extend Theorem 2.4.1 to the case of a first-to-default claim written
on a basket of n reference credit names.

Theorem 2.5.1 Assume that det N (t) 6= 0 for every t ∈ [0, T ], where
 
δ1 (t) − Set1 (κ1 ) 2
St|1 (κ2 ) − Set2 (κ2 ) . . . St|1
n
(κn ) − Setn (κn )
 1 
 St|2 (κ1 ) − Set1 (κ1 ) δ2 (t) − Set2 (κ2 ) n
. . . St|2 (κn ) − Setn (κn ) 
N (t) = 
 .. .. .. ..


 . . . . 
1
St|n (κ1 ) − Set1 (κ1 ) St|n
2
(κ1 ) − Set2 (κ1 ) . . . δn (t) − Setn (κn )

For every t ∈ [0, T ], let φ(t) e = (φe1 (t), φe2 (t), . . . , φen (t)) be the unique solution to the linear equation
e
N (t)φ(t) = h(t) where h(t) = (h1 (t), h2 (t), . . . , hn (t)) with hi (t) = Zi (t) − π e(t−) and where πe
e e e
is given by Lemma 2.5.6. More explicitly, the functions φ1 , φ2 , . . . , φn satisfy, for t ∈ [0, T ] and
i = 1, 2, . . . , n,
¡ ¢ X ¡ j ¢
φei (t) δi (t) − Seti (κi ) + φej (t) St|i (κj ) − Setj (κj ) = Zi (t) − π
e(t−). (2.67)
j6=i

Let us set φit = φei (τ(1) ∧ t) for i = 1, 2, . . . , n and let, for every t ∈ [0, T ],
n
X
φ0t = Vt (φ, A) − φit Sti (κi ), (2.68)
i=1
2.5. BASKET CREDIT DERIVATIVES 75

where the process V (φ, A) is given by the formula

X n Z
Vt (φ, A) = π
e(0) + φei (u) dSuc,i (κi ) − A(τ(1) ∧ t). (2.69)
i=1 ]0,τ(1) ∧t]

Then the trading strategy (φ, A) replicates the FTDC (X, A, Z, τ(1) ).

Proof. The proof is based on similar arguments as the proof of Theorem 2.4.1. It suffices to check
that under the assumption of the theorem, for a trading strategy (φ, A) stopped at τ(1) , we obtain
from (2.62) and (2.66) that
n
X ³¡ ¢ X¡ ¢ ´
dVt (φ, A) = φit δi (t) − Set−
i ci +
(κi ) dM t S i
t|j (κ i ) − ei (κi ) dM
S t−
ctj
i=1 j6=i
− dA(τ(1) ∧ t).

For φit = φei (τ(1) ∧ t), where the functions φe1 , φe2 , . . . , φen solve (2.67), we thus obtain
n
X
dVt (φ, A) = (Zi (t) − π cti − dA(τ(1) ∧ t).
e(t−)) dM
i=1

By comparing the last formula with (2.65), we conclude that if, in addition, V0 (φ, A) = π0 = π
e0 and
φ0 is given by (2.68), then the strategy (φ, A) replicates an FTDC (X, A, Z, τ(1) ). ¤

2.5.6 Conditional Default Distributions

In the case of first-to-default claims, it was enough to consider the unconditional distribution of
default times. As expected, in order to deal with a general basket defaultable claim, we need to
analyze conditional distributions of default times. It is possible to extend the approach presented
in the preceding sections and to explicitly derive the dynamics of all processes of interest on the
time interval [0, T ]. However, since we deal here with a simple model of joint defaults, it suffices to
make a non-restrictive assumption that we work on the canonical space Ω = Rn and to use simple
arguments based on the conditioning with respect to past defaults.
Suppose that k names out of a total of n names have already defaulted. To introduce a convenient
notation, we adopt the convention that the n − k non-defaulted names are in their original order
j1 < . . . < jn−k , whereas the k defaulted names i1 , . . . , ik are ordered in such a way that u1 <
. . . < uk . For the sake of brevity, we write Dk = {τi1 = u1 , . . . , τik = uk } to denote the information
structure of the past k defaults.

Definition 2.5.4 The joint conditional distribution function of default times τj1 , . . . , τjn−k equals,
for every t1 , . . . , tn−k > uk ,

F (t1 , . . . , tn−k | τi1 = u1 , . . . , τik = uk )

¡ ¢
= Q τj1 ≤ t1 , . . . , τjn−k ≤ tn−k | τi1 u1 , . . . , τik = uk .

The joint conditional survival function of default times τj1 , . . . , τjn−k is given by the expression

G(t1 , . . . , tn−k | τi1 = u1 , . . . , τik = uk )

¡ ¢
= Q τj1 > t1 , . . . , τjn−k > tn−k | τi1 = u1 , . . . , τik = uk

for every t1 , . . . , tn−k > uk .

As expected, the conditional first-to-default intensities are defined using the joint conditional
distributions, instead of the joint (unconditional) distribution of default times. We will denote
G(1) (t |Dk ) = G(t, . . . , t |Dk ).
76 CHAPTER 2. HAZARD FUNCTION APPROACH

Definition 2.5.5 Given the event Dk , for any jl ∈ {j1 , . . . , jn−k } the conditional first-to-default
intensity of a surviving name jl is denoted by λej (t |Dk ) = λ ej (t | τi = u1 , . . . , τi = uk ). It is given
l l 1 k
by the formula
R∞R∞ R∞
e t t
. . . t dF (t1 , . . . , tl−1 , t, tl+1 , . . . , tn−k |Dk )
λjl (t |Dk ) =
G(1) (t |Dk )

for every t ∈ [uk , T ].

i
In Section 2.5.3, we introduced the processes St|j (κj ) representing the value of the ith CDS at
time t on the event {τ(1) = τj = t}. According to the notation introduced above, we thus dealt with
the conditional value of the ith CDS with respect to the event D1 = {τj = t}. It is clear that to
value a CDS for each surviving name, one can proceed as prior to the first default, except that one
should now use the conditional distribution

F (t1 , . . . , tn−1 | D1 ) = F (t1 , . . . , tn−1 | τj = j), ∀ t1 , . . . , tn−1 ∈ [t, T ],

rather than the unconditional distribution F (t1 , . . . , tn ), which was employed in Proposition 2.5.6.
The same argument can be applied to any default event Dk . The corresponding conditional version
of Proposition 2.5.6 is rather easy to formulate and prove and thus we decided not to provide an
explicit conditional pricing formula here.
The conditional first-to-default intensities introduced in Definition 2.5.5 will allow us to construct
the conditional first-to-default martingales in a similar way as we defined the first-to-default mar-
ei . However, since any name can default
tingales M i associated with the first-to-default intensities λ
at any time, we need to introduce an entire family of conditional martingales, whose compensators
are based on intensities conditioned on the information about the past defaults.

Definition 2.5.6 Given the default event Dk = {τi1 = u1 , . . . , τik = uk }, for each surviving name
cjl by setting, for
jl ∈ {j1 , . . . , jn−k }, we define the basic conditional first-to-default martingale Mt|Dk
t ∈ [uk , T ],
Z t
Mcjl = Ht∧τ jl
− ej (u |Dk ) du.
1{u<τ(k+1) } λ (2.70)
t|Dk (k+1) l
uk

The process M cjl , t ∈ [uk , T ], is a martingale under the conditional probability measure Q|Dk ,
t|Dk
that is, the probability measure Q conditioned on the event Dk and with respect to the filtration
j
generated by default processes of the surviving names, that is, the filtration GtDk := Htj1 ∨. . .∨Ht n−k
for t ∈ [uk , T ].

Conditionally on the event Dk , we have τ(k+1) = τj1 ∧ τj2 ∧ . . . ∧ τjn−k , so that τ(k+1) is the first
default for all surviving names. Formula (2.70) is thus a rather straightforward generalization of
ci
formula (2.49). In particular, for k = 0 we obtain M ci
t|D0 = Mt , t ∈ [0, T ], for any i = 1, 2, . . . , n.

The martingale property of the process M cjl , as stated in Definition 2.5.6, follows from Propo-
t|Dk
sition 2.5.3; this property can also be seen as a conditional version of Corollary 2.5.1.
We are in a position to state the conditional version of the first-to-default predictable repre-
sentation theorem of Section 2.5.2. Formally, this result is nothing else than a restatement of the
martingale representation formula of Proposition 2.5.1 in terms of conditional first-to-default inten-
sities and conditional first-to-default martingales.
Let us fix an event Dk and let us write GDk = Hj1 ∨ . . . ∨ Hjn−k .

Proposition 2.5.3 Let Y be a random variable given by the formula

n−k
X
Y = Zjl |Dk (τjl )1{τjl ≤T, τjl =τ(k+1) } + X1{τ(k+1) >T }
l=1
2.5. BASKET CREDIT DERIVATIVES 77

for some functions Zjl |Dk : [uk , T ] → R, l = 1, 2, . . . , n − k and some constant X (possibly dependent
on Dk ). Let us define, for t ∈ [uk , T ],

ct|D = EQ|D (Y | GtDk ).

M k k

ct|D , t ∈ [uk , T ], is a GDk -martingale with respect to the conditional probability

Then the process M k
measure Q|Dk .
ct|D admits the following representation, for t ∈ [uk , T ],
Furthermore, M k

XZ
n−k
ct|D = M
M c0|D + cjl ,
hjl (u|Dk ) dM
k k u|Dk
l=1 ]uk ,t]

where the processes hjl are given by

ct−|D ,
hjl (t |Dk ) = Zjl |Dk (t) − M ∀ t ∈ [uk , T ].
k

Proof. The proof relies on a rather straightforward extension of arguments used in the proof of
Proposition 2.5.1 to the context of conditional default distributions. Therefore, we leave the details
to the reader. ¤

2.5.7 Recursive Valuation of a Basket Claim

We are ready to extend the results developed in the context of first-to-default claims to value and
hedge general basket claims. A generic basket claim is any contingent claim that pays a specified
amount on each default from a basket of n credit names and a constant amount at maturity T if no
defaults have occurred prior to or at T .

Definition 2.5.7 A basket claim associated with a family of n credit names is given as (X, A, Z̄, τ̄ )
where X is a constant amount payable at maturity only if no default occurs prior to or at T , the
vector τ̄ = (τ1 , . . . , τn ) represents default times and the time-dependent matrix Z̄ represents the
recovery payoffs at defaults, specifically,
 
Z1 (t |D0 ) Z2 (t |D0 ) ... Zn (t |D0 )
 Z1 (t |D1 ) Z2 (t |D1 ) ... Zn (t |D1 ) 
 
Z̄ =  .. .. .. .. .
 . . . . 
Z1 (t |Dn−1 ) Z2 (t |Dn−1 ) ... Zn (t |Dn−1 )

Note that the above matrix Z̄ is presented in the shorthand notation. In fact, in each row one
needs to specify, for an arbitrary choice of the event Dk = {τi1 = u1 , . . . , τik = uk } and any name
jl ∈
/ {i1 , . . . , ik }, the conditional payoff function at the moment of the (k + 1)th default, that is,

Zjl (t |Dk ) = Zjl (t | τi1 = u1 , . . . , τik = uk ), ∀ t ∈ [uk , T ].

In the financial interpretation, the function Zjl (t |Dk ) specifies the recovery payment at the
default of the name jl , conditional on the event Dk and on the event {τjl = τ(k+1) = t}, that is,
assuming that the name jl is the first defaulted name among all surviving names.
In particular, Zi (t |D0 ) := Zi (t) represents the recovery payment at the default of the ith name
at time t ∈ [0, T ], given that no defaults have occurred prior to t, that is, at the moment of the first
default. We will use the symbol D0 to denote the situation where no defaults have occurred prior
to time t.
78 CHAPTER 2. HAZARD FUNCTION APPROACH

Example 2.5.1 Let us consider the kth-to-default claim for some fixed k ∈ {1, 2, . . . , n}. Assume
that the payoff at the kth default depends only on the moment of the kth default and the identity
of the kth defaulted name. Then all rows of the matrix Z̄ are equal to zero, except for the kth row,
which equals, for every t ∈ [0, T ],

[Z1 (t | k − 1), Z2 (t | k − 1), . . . , Zn (t | k − 1)].

We write here k − 1, rather than Dk−1 , in order to emphasize that the knowledge of timings and
identities of the k defaulted names is not relevant under the present assumptions.
More generally, for a generic basket claim in which the payoff at the ith default depends on the
time of the ith default and identity of the ith defaulted name only, the recovery matrix Z̄ reads
 
Z1 (t) Z2 (t) ... Zn (t)
 Z1 (t |1) Z2 (t |1) ... Zn (t |1) 
 
Z̄ =  .. .. .. .. 
 . . . . 
Z1 (t |n − 1) Z2 (t |n − 1) . . . Zn (t |n − 1)

where Zj (t |k − 1) represents the payoff at the moment τ(k) = t of the kth default if j is the kth
defaulting name, that is, on the event {τj = τ(k) = t}. This shows that in several practically
important examples of basket credit derivatives, the matrix Z̄ of recoveries will have a relatively
simple structure.

It is clear that any basket claim can be represented as a static portfolio of kth-to-default claims
for k = 1, 2, . . . , n. However, this decomposition does not seem to be advantageous for our purposes.
In what follows, we prefer to represent a basket claim as a sequence of conditional first-to-default
claims, with the same value between any two defaults as a basket claim under consideration. Using
this approach, we will be able to directly apply previously developed results for the case of first-to-
default claims and thus to produce a rather straightforward recursive algorithm for the valuation
and hedging of a basket claim.
Instead of stating a formal result, which would require heavy notation, we prefer to focus first
on the computational algorithm for valuation and hedging of a basket claim. An important concept
in this algorithm is the conditional pre-default price
e |Dk ) = Z(t
Z(t e | τi = u1 , . . . , τi = uk ), ∀ t ∈ [uk , T ],
1 k

e |Dk ), t ∈ [uk , T ], is defined as the risk-

of a conditional first-to-default claim. The function Z(t
neutral value of a conditional FTDC on n − k surviving names, with the following recovery payoffs
upon the first default at any date t ∈ [uk , T ]

Zbjl (t | Dk ) = Zjl (t | Dk ) + Z(t

e | Dk , τj = t).
l
(2.71)

Assume for the moment that for any name jm ∈ / {i1 , . . . , ik , jl } the conditional recovery payoff
Zbjm (t | τi1 = u1 , . . . , τik = uk , τjl = uk+1 ) upon the first default after date uk+1 is known. Then we
can compute the function
e | τi = u1 , . . . , τi = uk , τj = uk+1 ),
Z(t ∀ t ∈ [uk+1 , T ],
1 k l

exactly as in Lemma 2.5.6, but using the conditional default distribution. The assumption that the
conditional payoffs are known is not restrictive, since the functions appearing in right-hand side of
(2.71) are known from the previous step in the following recursive pricing algorithm.

• First step. We first derive the value of a basket claim assuming that all but one defaults have
already occurred. Let
Dn−1 = {τi1 = u1 , . . . , τin−1 = un−1 }.
2.5. BASKET CREDIT DERIVATIVES 79

For any t ∈ [un−1 , T ], we deal with the payoffs

Zbj1 (t |Dn−1 ) = Zj1 (t |Dn−1 ) = Zj1 (t | τi1 = u1 , . . . , τin−1 = un−1 ),
for j1 ∈/ {i1 , . . . , in−1 } where the recovery payment Zj1 (t |Dn−1 ) for t ∈ [un−1 , T ] is given by
the specification of the basket claim. Hence we can evaluate the pre-default value Z(t e |Dn−1 )
at any time t ∈ [un−1 , T ], as a value of a conditional first-to-default claim with the said payoff,
using the conditional distribution under Q|Dn−1 of the random time τj1 = τin on the interval
[un−1 , T ].
• Second step. In this step, we assume that all but two names have already defaulted. Let
Dn−2 = {τi1 = u1 , . . . , τin−2 = un−2 }.
For each surviving name j1 , j2 ∈/ {i1 , . . . , in−2 }, the payoff Zbjl (t |Dn−2 ) for t ∈ [un−2 , T ], of a
basket claim at the moment of the next default formally comprises the recovery payoff from the
defaulted name jl which is Zjl (t |Dn−2 ) as well as the pre-default value Z(t e | Dn−2 , τj = t), t ∈
l
[un−2 , T ], which was computed in the first step. Therefore, we have, for every t ∈ [un−2 , T ],
Zbjl (t | Dn−2 ) = Zjl (t | Dn−2 ) + Z(t
e | Dn−2 , τj = t).
l

To find the value of a basket claim between the moments of the (n − 2)th and the (n − 1)th
default, it suffices to compute the pre-default value of the conditional FTDC associated with
the two surviving names, j1 , j2 ∈/ {i1 , . . . , in−2 }. Since the conditional payoffs Zbj1 (t |Dn−2 )
b
and Zj2 (t |Dn−2 ) are already known at this stage, it is sufficient to compute the expectation
under the conditional probability measure Q|Dn−2 in order to find the pre-default value of this
conditional FTDC for any t ∈ [un−2 , T ].
• General induction step. We now assume that exactly k defaults have occurred, that is, we
assume that we are working on the event
Dk = {τi1 = u1 , . . . , τik = uk }.
From the preceding step, we know the function Z(t e |Dk+1 ) where the event Dk+1 is given as
Dk+1 = {τi1 = u1 , . . . , τik = uk , τjl = uk+1 }.
In order to evaluate Z(te |Dk ), we set, for t ∈ [uk , T ],

Zbjl (t | Dk ) = Zjl (t | Dk ) + Z(t

e | Dk , τj = t),
l
(2.72)
for any j1 , . . . , jn−k ∈ e |Dk ) for every t ∈ [uk , T ] as the risk-
/ {i1 , . . . , ik } and we compute Z(t
neutral value under the conditional probability Q|Dk of the conditional FTDC with payoffs
given by (2.72).

We are in a position to state the valuation result for a basket claim, which can be formally
established using the reasoning outlined above.

Proposition 2.5.4 The risk-neutral value at time t ∈ [0, T ] of a basket claim (X, A, Z̄, τ̄ ) equals,
for t ∈ [0, T ],
n−1
X
πt = e |Dk )1[τ ∧T,τ
Z(t ∧T [ (t),
(k) (k+1)
k=0
where Dk = Dk (ω) = {τi1 (ω) = u1 , . . . , τik (ω) = uk } for k = 1, 2, . . . , n and D0 means that no
defaults have yet occurred.

Proof. Assume that we are at some date t ∈ [0, T ] and suppose that exactly k names (for some
k = 1, 2, . . . , n) have already defaulted, hence the set Dk is known to us (so that t ≥ uk ). Form the
point of view of valuation, the basket claim can be seen this point of time as a conditional FTDC
with the conditional payoff Z(t b |Dk ) = Z(t |Dk ) + Z(t
e |Dk+1 ). We can now use the pricing formula
of Proposition 2.5.6 (using conditional distribution) for an FTDC in order to derive the value of
e |Dk ) for every t ∈ [uk , T ].
Z(t ¤
80 CHAPTER 2. HAZARD FUNCTION APPROACH

2.5.8 Recursive Replication of a Basket Claim

From our discussion, it is clear that a basket claim can be conveniently interpreted as a specific
sequence of conditional first-to-default claims and thus the replication of a basket claim relies on
hedging of a sequence of conditional first-to-default claims. In the next result, we denote τ(0) = 0.

Theorem 2.5.2 For any k = 0, 1, . . . , n, the replicating strategy φ for a basket claim (X, A, Z̄, τ̄ )
on the time interval [τk ∧ T, τk+1 ∧ T ] coincides with the replicating strategy for the conditional
FTDC with payoffs Z(tb |Dk ) given by (2.72). The replicating strategy φ = (φ0 , φj1 , . . . , φjn−k , A),
corresponding to the units of savings account and units of CDS on each surviving name at time t,
has the wealth process
n−k
X j j
Vt (φ, A) = φ0t + φtl St l (κjl )
l=1
jl
where the processes φ , l = 1, 2, . . . , n − k can be computed by the conditional version of Theorem
2.5.1.

Proof. We know that the basket claim can be decomposed into a series of conditional first-to-
default claims. So, at any given moment of time t ∈ [0, T ], assuming that k defaults have already
occurred, our basket claim is equivalent to the conditional FTDC with payoffs Z(t b | Dk ) and the
e |Dk ). This conditional FTDC is alive up to the next default τ(k+1) or maturity
pre-default value Z(t
T , whichever comes first. Hence it is clear that the replicating strategy of a basket claim over the
random interval [τk ∧ T, τk+1 ∧ T ] need to coincide with the replicating strategy for this conditional
first-to-default claim and thus it can be found along the same lines as in Theorem 2.5.1, using the
conditional distribution under Q|Dk of defaults for surviving names. ¤

2.6 Applications to Copula-Based Models

We will now apply the general results to simple models, in which some copula functions (cf. Section
5.4) are used to describe the dependence of default times. For various applications of copula functions
to credit risk modeling and to valuation of credit derivatives, the interested reader is referred to,
e.g., Andersen and Sidenius [3], Burtschell et al. [41, 42], Cherubini and Luciano [48], Cherubini et
al. [49], Embrechts et al. [72], Frey et al. [79], Gennheimer [80], Giesecke [81], Laurent and Gregory
[115], Li [118], McNeil et al. [123], and Schönbucher and Schubert [137].
For simplicity of exposition, we only consider the bivariate situation and we work under the
following standing assumptions.

Assumption 2.6.1 We assume that:

• we are given a first-to-default claim (X, A, Z, τ(1) ) where Z = (Z1 , Z2 ) for some constants
Z1 , Z2 and X,
• the default times τ1 and τ2 have exponential marginal distributions with parameters λ1 and
λ2 ,
• the protection δi of the ith credit default swap is constant and κi = λi δi for i = 1, 2.

2.6.1 Independent Default Times

Let us first consider the case where the default times τ1 and τ2 are independent (of course, this
corresponds to the product copula C(u, v) = uv). In view of independence, the marginal intensities
and the first-to-default intensities can be easily shown to coincide. We have, for i = 1, 2,
Gi (u) = Q(τi > u) = e−λi u
2.6. APPLICATIONS TO COPULA-BASED MODELS 81

and thus the joint survival probability equals, for every (u, v) ∈ R2+ ,

G(u, v) = G1 (u)G2 (v) = e−λ1 u e−λ2 v .

Consequently, we obtain

F (du, dv) = G(du, dv) = λ1 λ2 e−λ1 u e−λ2 v dudv = f (u, v) dudv

and
G(du, u) = −λ1 e−(λ1 +λ2 )u du.

Proposition 2.6.1 Assume that the default times τ1 and τ2 are independent. Then the replicating
strategy for an FTDC (X, 0, Z, τ(1) ) is given as

Z1 − π
e(t) Z2 − π
e(t)
φe1 (t) = , φe2 (t) = ,
δ1 δ2
where
(Z1 λ1 + Z2 λ2 )
π
e(t) = (1 − e−(λ1 +λ2 )(T −t) ) + Xe−(λ1 +λ2 )(T −t) .
λ1 + λ2

Proof. From the previous remarks, we obtain

RT R∞ RT R∞
Z1 t u dF (u, v) Z2 t v dF (u, v) G(T, T )
π
e(t) = + +X
G(t, t) G(t, t) G(t, t)
R T −(λ +λ )u R T −(λ +λ )v
Z 1 λ1 t e 1 2
du Z2 λ2 t e 1 2
dv G(T, T )
= + +X
e−(λ1 +λ2 )t e−(λ1 +λ2 )t G(t, t)
Z 1 λ1 Z 2 λ2
= (1 − e−(λ1 +λ2 )(T −t) ) + (1 − e−(λ1 +λ2 )(T −t) )
(λ1 + λ2 ) (λ1 + λ2 )
G(T, T )
+X ,
G(t, t)
and thus
(Z1 λ1 + Z2 λ2 )
π
e(t) = (1 − e−(λ1 +λ2 )(T −t) ) + Xe−(λ1 +λ2 )(T −t) .
λ1 + λ2
Under the assumption of independence of default times, we also have that St|j i
(κi ) = Seti (κi ) for
i, j = 1, 2 and i 6= j. Furthermore, from Example 2.4.1, we have that Seti (κi ) = 0 for t ∈ [0, T ] and
thus the matrix N (t) in Theorem 2.5.1 reduces to
· ¸
δ1 0
N (t) = .
0 δ2

e = h(t) where
The replicating strategy can be found easily by solving the linear equation N (t)φ(t)
h(t) = (h1 (t), h2 (t)) with the function hi given by the formula

hi (t) = Zi − π
e(t−) = Zi − π
e(t)

for i = 1, 2. ¤

As an important example of a first-to-default claim, we will now consider the case of a first-to-
default swap (FTDS). A stylized FTDS is formally defined by setting X = 0, A(t) = −κ(1) t where
κ(1) is the swap spread and Zi (t) = δi ∈ [0, 1) for some constants δi , i = 1, 2. Hence an FTDS can be
equivalently seen as the FTDC (0, −κ(1) t, (δ1 , δ2 ), τ(1) ). Under the present assumptions, we obtain

1 − eλT ³ ´
π0 = π
e(0) = (λ1 δ1 + λ2 δ2 ) − κ(1)
λ
82 CHAPTER 2. HAZARD FUNCTION APPROACH

where we denote λ = λ1 + λ2 . The FTDS market spread is the level of κ(1) that makes the FTDS
valueless at initiation. Hence in this elementary example this spread equals λ1 δ1 + λ2 δ2 . In addition,
it can be shown that under the present assumptions we have that π e(t) = 0 for every t ∈ [0, T ].
Suppose that we wish to hedge the short position in the FTDS using two CDSs, say CDSi ,
i = 1, 2, with respective default times τi , protection payments δi and spreads κi = λi δi . Recall that
in the present setup we have that, for every t ∈ [0, T ],
i
St|j (κi ) = Seti (κi ) = 0, i, j = 1, 2, i 6= j. (2.73)

Consequently, we have here that hi (t) = −Zi (t) = −δi for every t ∈ [0, T ]. It then follows from
e
equation N (t)φ(t) = h(t) that φe1 (t) = φe2 (t) = 1 for every t ∈ [0, T ] and thus φ0t = 0 for every
t ∈ [0, T ]. This result is by no means surprising; we hedge a short position in the FTDS by holding
a static portfolio of two single-name CDSs since, under the present assumptions, the FTDS is
equivalent to such a portfolio of corresponding single name CDSs. Of course, one would not expect
that this feature will still hold in a general case of dependent default times.
The first equality in (2.73) is due to the standing assumption of independence of default times
τ1 and τ2 and thus it will no longer be true for other copulae. The second equality follows from our
simplifying postulate that the risk-neutral marginal distributions of default times are exponential.
In practice, the risk-neutral marginal distributions of default times are obtained by fitting a model to
market data (i.e., market prices of single name CDSs) and thus, typically, they are not exponential.

2.6.2 Archimedean Copulae

We now proceed to the case of exponentially distributed, but dependent, default times. The mutual
dependence will be specified by a choice of some Archimedean copula. Recall that a bivariate
Archimedean copula is defined as C(u, v) = ϕ−1 (ϕ(u), ϕ(v)), where ϕ is called the generator of
a copula.

Clayton Copula

Recall that the generator of the Clayton copula is given as ϕ(s) = s−θ − 1 for every s ∈ R+ , for some
strictly positive parameter θ. Hence the bivariate Clayton copula can be represented as follows
1
C(u, v) = (u−θ + v −θ − 1)− θ .

Under the present assumptions, the corresponding joint survival function G(u, v) equals
1
G(u, v) = C(G1 (u), G2 (v)) = (eλ1 uθ + eλ2 vθ − 1)− θ ,

so that
G(u, dv) 1
= −λ2 eλ2 vθ (eλ1 uθ + eλ2 vθ − 1)− θ −1
dv
and
G(du, dv) 1
f (u, v) = = (θ + 1)λ1 λ2 eλ1 uθ+λ2 vθ (eλ1 uθ + eλ2 vθ − 1)− θ −2 .
dudv
We only provide explicit formulae for φe1 and Sv|2
1
(κ1 ), since the quantities φe2 and Su|1
2
(κ2 ) are given
by symmetric expressions.

Proposition 2.6.2 Let the joint distribution of (τ1 , τ2 ) be given by the Clayton copula with some
θ > 0. Then the replicating strategy for an FTDC (X, 0, Z, τ(1) ) is given by the expression
2
δ2 (Z1 − π
e(t)) + St|1 (κ2 )(Z2 − π
e(t))
φe1 (t) = 1 (κ )S 2 (κ ) , (2.74)
δ1 δ2 − St|2 1 t|1 2
2.6. APPLICATIONS TO COPULA-BASED MODELS 83

where
R eλ1 θT λ2 1 R eλ2 θT λ1 1
eλ1 θt
(s + s λ1 − 1)− θ −1 ds eλ2 θt
(s + s λ2 − 1)− θ −1 ds
π
e(t) = Z1 1 + Z2 1
θ(eλ1 θt + eλ2 θt − 1)− θ θ(eλ1 θt + eλ2 θt − 1)− θ
1
(eλ1 θT + eλ2 θT − 1)− θ
+X 1
(eλ1 θt + eλ2 θt − 1)− θ
and
1 1
1 [(eλ1 θT + eλ2 θT − 1)− θ −1 − (eλ1 θv + eλ2 θv − 1)− θ −1 ]
Sv|2 (κ1 ) = δ1 1
(eλ1 θv + eλ2 θv − 1)− θ −1
RT 1
v
(eλ1 θu + eλ2 θv − 1)− θ −1 du
− κ1 1 .
(eλ1 θv + eλ2 θv − 1)− θ −1

Proof. Observe that

Z T Z ∞ Z T
1
du f (u, v)dv = λ1 eλ1 uθ (eλ1 uθ + eλ2 uθ − 1)− θ −1 du
t u t
Z eλ1 θT
1 λ2 1
= (s + s λ1 − 1)− θ −1 ds
θ eλ1 θt

and thus, by symmetry,

Z T Z ∞ Z eλ2 θT
1 λ1 1
dv f (u, v)du = (s + s λ2 − 1)− θ −1 ds.
t v θ eλ2 θt

We thus obtain
RT R∞ RT R∞
Z1 t
dG(u, v) Z2
u
dG(u, v) G(T, T )
t v
π
e(t) = + +X
G(t, t) G(t, t) G(t, t)
R eλ1 θT λ2
− 1
−1
R eλ2 θT λ1 1
eλ1 θt
(s + s λ1 − 1) θ ds eλ2 θt
(s + s λ2 − 1)− θ −1 ds
= Z1 1 + Z 2 1
θ(eλ1 θt + eλ2 θt − 1)− θ θ(eλ1 θt + eλ2 θt − 1)− θ
1
(eλ1 θT + eλ2 θT − 1)− θ
+X − θ1
.
(eλ1 θt + eλ2 θt − 1)
We are in a position to determine the replicating strategy. Under the standing assumption that
κi = λi δi for i = 1, 2 we still have that Seti (κi ) = 0 for i = 1, 2 and for t ∈ [0, T ]. Hence the matrix
N (t) reduces to " #
2
δ1 −St|1 (κ2 )
N (t) = 1
−St|2 (κ1 ) δ2
where
RT RT R∞
1 v
f (u, v) du f (z, v) dzdu
Sv|2 (κ1 ) = δ1 R ∞ − κ1 v R ∞
u

v
f (u, v) du v
f (u, v) du
1 1
[(eλ1 θT + eλ2 θT − 1)− θ −1 − (eλ1 θv + eλ2 θv − 1)− θ −1 ]
= δ1 − θ1 −1
(eλ1 θv + eλ2 θv − 1)
RT 1
v
(eλ1 θu + eλ2 θv − 1)− θ −1 du
− κ1 1 .
(eλ1 θv + eλ2 θv − 1)− θ −1
2
The expression for Su|1 (κ2 ) can be found by analogous computations. By solving the equation
e
N (t)φ(t) = h(t), we obtain the required expression (2.74) for the replicating strategy. ¤
84 CHAPTER 2. HAZARD FUNCTION APPROACH

Gumbel Copula

As another example of an Archimedean copula, we consider the Gumbel copula with the generator
ϕ(s) = (− ln s)θ for every s ∈ R+ where the parameter θ satisfies θ ≥ 1. The bivariate Gumbel
copula can thus be written as
θ 1
+(− ln v)θ ] θ
C(u, v) = e−[(− ln u) .

Under our standing assumptions, the corresponding joint survival function G(u, v) equals
θ θ 1
+λθ2 v θ ) θ
G(u, v) = C(G1 (u), G2 (v)) = e−(λ1 u .

Consequently, the partial derivatives of our interest satisfy

dG(u, v) 1
= −G(u, v)λθ2 v θ−1 (λθ1 uθ + λθ2 v θ ) θ −1
dv
and
dG(u, v) 1 ¡ 1 ¢
= G(u, v)(λ1 λ2 )θ (uv)θ−1 (λθ1 uθ + λθ2 v θ ) θ −2 (λθ1 uθ + λθ2 v θ ) θ + θ − 1 .
dudv
As in the case of the Clayton copula, it is enough to derive the formulae for φe1 and Sv|2 1
(κ1 ), since
e 2
φ2 and S (κ2 ) are given by symmetric expressions.
u|1

Proposition 2.6.3 Assume that the joint distribution of (τ1 , τ2 ) is given by the Gumbel copula with
θ ≥ 1. Then the replicating strategy for an FTDC (X, 0, Z, τ(1) ) is given by
2
δ2 (Z1 − π
e(t)) + St|1 (κ2 )(Z2 − π
e(t))
φe1 (t) = 1 (κ )S 2 (κ ) ,
δ1 δ2 − St|2 1 t|1 2

where
e(t) = (Z1 λθ1 + Z2 λθ2 )λ−θ (e−λt − e−λT ) + Xe−λ(T −t)
π
1
with λ = (λθ1 + λθ2 ) θ and
θ θ 1
+λθ2 v θ ) θ 1
1 e−(λ1 T (λθ1 T θ + λθ2 v θ ) θ −1 − e−λv λ1−θ v 1−θ
Sv|2 (κ1 ) = δ1
e−λv λ1−θ v 1−θ
RT θ θ 1
+λθ2 v θ ) θ 1
v
e−(λ1 T (λθ1 uθ + λθ2 v θ ) θ −1 du
− κ1 .
e−λv λ1−θ v 1−θ

Proof. We have that

Z TZ ∞ Z T 1
1 θ θ
dG(u, v) = λθ1 (λθ1 + λθ2 ) θ −1 e−(λ1 +λ2 ) θ u du
t u t
= (−λθ1 λ−θ e−λu )|u=T θ −θ −λt
u=t = λ1 λ (e − e−λT ),
1
where λ = (λθ1 + λθ2 ) θ . Similarly, we also obtain
Z TZ ∞
dG(u, v) = λθ2 λ−θ (e−λt − e−λT ).
t v

Furthermore, G(T, T ) = e−λT and G(t, t) = e−λt . Hence

RT R∞ RT R∞
dG(u, v) dG(u, v) G(T, T )
πe(t) = Z1 t u + Z2 t v +X
G(t, t) G(t, t) G(t, t)
= Z1 λθ1 λ−θ (e−λt − e−λT ) + Z2 λθ2 λ−θ (e−λt − e−λT ) + Xe−λ(T −t) ,
2.6. APPLICATIONS TO COPULA-BASED MODELS 85

and thus
e(t) = (Z1 λθ1 + δ2 Z2θ )λ−θ (e−λt − e−λT ) + Xe−λ(T −t) .
π
In order to find the replicating strategy, we proceed as in the proof of Proposition 2.6.2. Under the
1
present assumptions, we obtain the following expression for Sv|2 (κ1 )
RT RT R∞
1 f (u, v)du f (z, v)dzdu
Sv|2 (κ1 ) = δ1 R v∞ − κ1 v Ru
∞
v
f (u, v)du v
f (u, v)du
1
−(λθ1 T θ +λθ2 v θ ) θ 1
e (λθ1 T θ + λθ2 v θ ) θ −1 − e−λv λ1−θ v 1−θ
= δ1
e−λv λ1−θ v 1−θ
RT θ θ 1
+λθ2 v θ ) θ 1
v
e−(λ1 T (λθ1 uθ + λθ2 v θ ) θ −1 du
− κ1 −λv
.
e λ v 1−θ
1−θ

2
By the symmetry of the model, a similar expression is valid for the value Su|1 (κ2 ). This completes
the proof of the proposition. ¤

Though the copula-based models are widely used for modeling of dependent defaults, they suffer
a major shortcoming of being inherently static models. Therefore, their use is limited to risk-neutral
valuation of credit derivatives, as opposed to the arbitrage pricing of defaultable claims, which hinges
on the dynamic replication technique.
86 CHAPTER 2. HAZARD FUNCTION APPROACH
Chapter 3

Hazard Process Approach

In the general reduced-form (or hazard process) approach, we deal with two kinds of information: the
information conveyed by assets prices and other economic factors, denoted as F = (Ft )0≤t≤T ∗ , and
the information about the occurrence of the default time, that is, the knowledge of the time where
the default occurred in the past, if the default has indeed already happen. As we already know, the
latter information is modeled by the filtration H generated by the default process H.
At the intuitive level, the reference filtration F is generated by prices of some assets, or by other
economic factors (such as, e.g., interest rates). This filtration can also be a subfiltration of the
filtration generated by the asset prices. The case where F is the trivial filtration is exactly what we
have studied in the previous chapter. Though in a typical example F is chosen to be the Brownian
filtration, most theoretical results do not rely on a particular choice of the reference filtration F. We
denote by Gt = Ft ∨ Ht the full filtration (sometimes referred to as the enlarged filtration).
Special attention will be paid in what follows to the so-called hypothesis (H). In the present
context, it postulates the preservation of the martingale property with respect to the enlargement of
F by the observations of default time. It is important to note that this hypothesis is not preserved
under an equivalent change of a probability measure, in general.
In order to examine the precise meaning of market completeness in a defaultable security market
model and to derive the hedging strategies for credit derivatives, we shall also establish a suitable
version of the predictable representation theorem.
Most results presented in Sections 3.1–3.6 can be found, for instance, in survey papers by Jean-
blanc and Rutkowski [99, 100]; see also the papers by Artzner and Delbaen [5], Bélanger et al. [10],
Jarrow and Turnbull [94], Lando [107], and Wong [142].
Sections 3.7–3.8 are based on the paper by Bielecki at al. [19].

3.1 Hazard Process and its Applications

The concepts introduced in the Chapter 2 will now be extended to a more general setup, in which
an additional flow of information – formally represented hereafter by some filtration F – is available.
We denote by τ a non-negative random variable on a probability space (Ω, G, Q), satisfying
Q(τ = 0) = 0 and Q(τ > t) > 0 for any t ∈ R+ . We introduce the right-continuous default indicator
process H by setting Ht = 1{t≥τ } for t ∈ R+ and we write H to denote the filtration generated by
the process H, so that Ht = σ(Hu : u ≤ t) for every t ∈ R+ .
We assume that we are given an auxiliary reference filtration F such that G = H ∨ F, that is,
Gt = Ht ∨ Ft for any t ∈ R+ . For each t ∈ R+ , the total information available at time t is captured
by the σ-field Gt .

87
88 CHAPTER 3. HAZARD PROCESS APPROACH

All filtrations considered in what follows are implicitly assumed to satisfy the ‘usual conditions’
of right-continuity and completeness. For the sake of simplicity, we assume that the σ-field F0 is
trivial (so that G0 is the trivial σ-field as well).
The process H is obviously G-adapted, but it is not necessarily F-adapted. In other words, the
random time τ is a G-stopping time, but it may fail to be an F-stopping time.

Lemma 3.1.1 Assume that the filtration G satisfies G = H ∨ F. Then G ⊆ G∗ where the filtration
G∗ = (Gt∗ ) t∈R+ is defined as follows
© ª
Gt∗ := A ∈ G : A ∩ {τ > t} = B ∩ {τ > t} for some B ∈ Ft .

Proof. It is rather clear that the class Gt∗ is a sub-σ-field of G. Therefore, it is enough to check that
Ht ⊆ Gt∗ and Ft ⊆ Gt∗ for every t ∈ R+ . Put another way, we need to verify that if either A = {τ ≤ u}
for some u ≤ t or A ∈ Ft then there exists an event B ∈ Ft such that A ∩ {τ > t} = B ∩ {τ > t}.
In the former case, we may take B = ∅ and in the latter B = A. ¤

For any t ∈ R+ , we write Ft = Q(τ ≤ t | Ft ) and we denote by G the F-survival process of τ with
respect to the filtration F, given as
Gt := 1 − Ft = Q(τ > t | Ft ).
For any 0 ≤ t ≤ s the inclusion {τ ≤ t} ⊆ {τ ≤ s} holds, and thus
¡ ¯ ¢
EQ (Fs | Ft ) = EQ Q(τ ≤ s | Fs ) ¯ Ft = Q(τ ≤ s | Ft ) ≥ Q(τ ≤ t | Ft ) = Ft .
This shows that the process F (G, respectively) follows a bounded and non-negative F-submartingale
(F-supermartingale, respectively) under Q and thus we may deal with the right-continuous modifi-
cations of F and G with finite left-hand limits. It is worth noting that F0 = 0 and limt→∞ Ft = 1.
The next definition introduces a straightforward generalization of the concept of the hazard
function (see Definition 2.2.1).

Definition 3.1.1 Assume that Ft < 1 for t ∈ R+ . The F-hazard process of τ under Q, denoted by
Γ, is defined through the formula 1 − Ft = e−Γt . Equivalently, Γt = − ln Gt = − ln (1 − Ft ) for
every t ∈ R+ .

Since G0 = 1, it is clear that Γ0 = 0. Moreover, limt→∞ Γt = ∞ since limt→∞ Gt = 0. For the

sake of conciseness, we shall refer briefly to Γ as the F-hazard process, rather than the F-hazard
process under Q, unless there is a danger of misunderstanding.
Throughout this chapter, we will work under the standing assumption that the inequality Ft < 1
holds for every t ∈ R+ , so that the F-hazard process Γ is well defined. Therefore, the case when τ
is an F-stopping time (that is, the case when F = G) is not dealt with here.

3.1.1 Conditional Expectations

We will first focus on the conditional expectation EQ (1{t<τ } X | Gt ), where X is a Q-integrable
random variable. We start by extending the formula established in Lemma 2.2.1.

Lemma 3.1.2 For any G-measurable and Q-integrable random variable X we have, for any t ∈ R+ ,
EQ (1{t<τ } X | Ft )
EQ (1{t<τ } X | Gt ) = 1{t<τ } EQ (X | Gt ) = 1{t<τ } . (3.1)
Q(t < τ | Ft )
In particular, for any t ≤ s
Q(t < τ ≤ s | Ft )
Q(t < τ ≤ s | Gt ) = 1{t<τ } = 1{t<τ } EQ (1 − eΓt −Γs | Ft ).
Q(t < τ | Ft )
3.1. HAZARD PROCESS AND ITS APPLICATIONS 89

Proof. Since Ft ⊆ Gt , it suffices to check that

¡ ¯ ¢ ¡ ¯ ¢
EQ 1C XQ(C | Ft ) ¯ Gt = EQ 1C EQ (1C X | Ft ) ¯ Gt ,

where we denote C = {t < τ }. Put another way, we need to show that for any A ∈ Gt we have
Z Z
1C XQ(C | Ft ) dQ = 1C EQ (1C X | Ft ) dQ. (3.2)
A A

In view of Lemma 3.1.1, for any A ∈ Gt we have A ∩ C = B ∩ C for some event B ∈ Ft , and so
Z Z Z
1C XQ(C | Ft ) dQ = XQ(C | Ft ) dQ = XQ(C | Ft ) dQ
A A∩C B∩C
Z Z
= 1C XQ(C | Ft ) dQ = EQ (1C X | Ft )Q(C | Ft ) dQ
ZB B
Z
= EQ (1C EQ (1C X | Ft ) | Ft ) dQ = EQ (1C X | Ft ) dQ
B B∩C
Z Z
= EQ (1C X | Ft ) dQ = 1C EQ (1C X | Ft ) dQ.
A∩C A

We thus conclude that (3.2) holds. ¤

The following corollary to Lemma 3.1.2 is rather straightforward.

Corollary 3.1.1 Let X be a FT -measurable and Q-integrable random variable. Then, for every
t ≤ T,
EQ (X1{T <τ } | Ft )
EQ (X1{T <τ } | Gt ) = 1{t<τ } = 1{t<τ } EQ (XeΓt −ΓT | Ft ).
EQ (1{t<τ } | Ft )

The following result will be used in valuation of a recovery payoff that occurs at default.

Lemma 3.1.3 Assume that Z is an F-predictable process such that the random variable Zτ 1{τ ≤T }
is Q-integrable. Then we have, for every t ≤ T ,
³Z ¯ ´
¯
1{t<τ } EQ (Zτ 1{τ ≤T } | Gt ) = 1{t<τ } eΓt EQ Zu dFu ¯ Ft . (3.3)
]t,T ]

Let F = N + C be the Doob-Meyer decomposition of F , where N is an F-martingale, and C is an

F-predictable increasing process. Then, for every t ≤ T ,
³Z ¯ ´
¯
1{t<τ } EQ (Zτ 1{τ ≤T } | Gt ) = 1{t<τ } eΓt EQ Zu dCu ¯ Ft . (3.4)
]t,T ]

If F is a continuous, increasing process then F = C = e−Γt so that the equality dFt = e−Γt dΓt is
valid. Consequently,
³Z T ¯ ´
¯
1{t<τ } EQ (Zτ 1{τ ≤T } | Gt ) = 1{t<τ } EQ Zu eΓt −Γu dΓu ¯ Ft .
t

Proof. We start by noting that (3.3) implies that

1{t<τ } EQ (Zτ 1{τ ≤T } | Gt ) = 1{t<τ } eΓt EQ (Zτ 1{t<τ ≤T } | Ft ).

Pn
Let us first assume that Z is a stepwise F-predictable process; specifically, Zu = i=0 Zti 1{ti <u≤ti+1 }
for t < u ≤ T , where t0 = t < · · · < tn+1 = T , and Zti is an Fti -measurable random variable for
90 CHAPTER 3. HAZARD PROCESS APPROACH

i = 0, . . . , n. Then we obtain
EQ (Zτ 1{t<τ ≤T } | Ft ) = EQ (Zτ 1{t<τ ≤T } | Ft )
³Xn ¯ ´
¯
= EQ 1{ti <τ ≤ti+1 } Zti ¯ Ft
i=0
³X
n ¯ ´
¯
= EQ Zti (Fti+1 − Fti ) ¯ Ft .
i=0

Hence for any stepwise, bounded, F-predictable process Z we have

¡ ¢ ³Z ¯ ´
¯
EQ 1{t<τ ≤T } Zτ | Ft = EQ Zu dFu ¯ Ft . (3.5)
]t,T ]

In the next step, Z is approximated by a suitable sequence of bounded, stepwise, F-predictable

processes. The sum under the sign of the conditional expectation converges to the Itô integral (or to
the Lebesque-Stieltjes integral if F is of finite variation). The boundedness of Z and F is a sufficient
condition for the convergence of sequence of conditional expectations. ¤

The next auxiliary result will prove useful in valuation of defaultable securities that pay dividends
prior to the default time.

Proposition 3.1.1 Assume that A is a bounded, F-predictable process of finite variation. Then, for
every t ≤ T ,
³Z ¯ ´ ³Z ¯ ´
¯ ¯
EQ (1 − Hu ) dAu ¯ Gt = 1{t<τ } eΓt EQ (1 − Fu ) dAu ¯ Ft
]t,T ] ]t,T ]

or, equivalently,
³Z ¯ ´ ³Z ¯ ´
¯ ¯
EQ (1 − Hu ) dAu ¯ Gt = 1{t<τ } EQ eΓt −Γu dAu ¯ Ft .
]t,T ] ]t,T ]

Proof. For a fixed, but arbitrary, t ≤ T, we introduce an auxiliary process A b by setting A

bu = Au −At
b
for u ∈ [t, T ]. It is clear that A is a bounded, F-predictable process of finite variation; the same remark
applies to its left-continuous version A bt− .
Therefore,
³Z ¯ ´
¯
Jt = EQ (1 − Hu ) dAu ¯ Gt
]t,T ]
³Z ¯ ´
= EQ bu ¯¯ Gt
1{τ >u} dA
]t,T ]
³ ¯ ´
= EQ A bT 1{τ >T } ¯¯ Gt
bτ − 1{t<τ ≤T } + A
³Z ¯ ´
= 1{t<τ } eΓt EQ bu− dFu + A
A bT (1 − FT ) ¯¯ Ft ,
]t,T ]

where the last equality follows from formulae (3.1) and (3.3). Using an obvious equality Gt = 1 − Ft ,
we obtain
³Z ¯ ´ ³ Z ¯ ´
EQ Abu− dFu + A bT (1 − FT ) ¯¯ Ft = EQ − Abu− dGu + AbT GT ¯¯ Ft .
]t,T ] ]t,T ]

b is a process of finite variation (so that its continuous martingale part vanishes), the following
Since A
version of Itô’s product rule is valid
Z Z
b b
AT GT = At Gt + b
Au− dGu + Gu dAbu .
]t,T ] ]t,T ]
3.1. HAZARD PROCESS AND ITS APPLICATIONS 91

bt = 0, and thus
But A
³Z ¯ ´ ³Z ¯ ´
b b ¯ ¯
EQ Au− dFu + AT (1 − FT ) ¯ Ft = EQ (1 − Fu ) dAu ¯ Ft .
]t,T ] ]t,T ]

This proves the first formula. The second equality is merely a restatement of the first one. ¤

3.1.2 Hazard Rate

Rt
Let the process F be absolutely continuous, that is, Ft = 0 fu du for some F-progressively measur-
able, non-negative process f . Then necessarily F is an increasing process and thus Γ is an absolutely
continuous and increasing
Rt process. Specifically, it is easy to check that Γ admits the F-hazard rate
γ, that is, Γt = 0 γu du where in turn the F-progressively measurable, non-negative process γ is
given by the formula γt = (1 − Ft )−1 ft . We will sometimes refer to γ as the F-intensity (or simply
stochastic intensity) of default time τ (see Section 3.1.6).

3.1.3 Valuation of Defaultable Claims

Our next goal is to establish a convenient representation for the pre-default value of a default-
able claim in terms of the hazard process Γ of the default time. We postulate that Q represents
a martingale measure associated with the choice of the savings account B as a discount factor
(or a numéraire). Therefore, in the present setup, the risk-neutral valuation formula reads (for a
justification of this formula, see Section 2.3)
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Gt , (3.6)
]t,T ]

where S is the ex-dividend price process, B is the savings account and D is the dividend process
associated with a defaultable claim (see Section 1.1.2), that is,
Z Z
Dt = XTd 1[T,∞[ (t) + (1 − Hu ) dAu + Zu dHu . (3.7)
]0,t] ]0,t]

For the sake of conciseness, we will write

³Z ¯ ´
¯
It = Bt EQ Bu−1 (1 − Hu ) dAu ¯ Gt
]t,T ]
¡ ¯ ¢
Jt = Bt EQ 1{t<τ ≤T } Bτ−1 Zτ ¯ Gt ,
and ¡ ¯ ¢
Kt = Bt EQ BT−1 X1{T <τ } ¯ Gt .
In view of (3.6)–(3.7), it is clear that the ex-dividend price of a generic defaultable claim (X, A, Z, τ )
(cf. Definition 2.3.1) can be represented as follows St = It + Jt + Kt . It is noteworthy that the
default time τ does not appear explicitly in the conditional expectation in the right-hand side of
pricing formulae of Proposition 3.1.2.

Proposition 3.1.2 For every t ∈ [0, T ], the ex-dividend price of a defaultable claim (X, A, Z, τ )
admits the following representation
³Z ¯ ´
¯
St = 1{t<τ } G−1
t B E
t Q Bu−1 (Gu dAu − Zu dGu ) + GT BT−1 X ¯ Ft .
]t,T ]

If F (and thus also Γ) is an increasing, continuous process then

³Z ¯ ´
¯
St = 1{t<τ } Bt EQ Bu−1 eΓt −Γu (dAu + Zu dΓu ) + BT−1 XeΓt −ΓT ¯ Ft .
]t,T ]
92 CHAPTER 3. HAZARD PROCESS APPROACH

R
Proof. By applying Proposition 3.1.1 to the process of finite variation ]0,t]
Bu−1 dAu , we obtain
³Z ¯ ´
¯
It = 1{t<τ } G−1
t Bt EQ Bu−1 Gu dAu ¯ Ft
]t,T ]

or, equivalently,
³Z ¯ ´
¯
It = 1{t<τ } Bt EQ Bu−1 eΓt −Γu dAu ¯ Ft .
]t,T ]

Furthermore, Lemma 3.1.3 yields

³Z ¯ ´
¯
Jt = 1{t<τ } eΓt Bt EQ Bu−1 Zu dFu ¯ Ft .
]t,T ]

If, in addition, the hazard process Γ is an increasing continuous process then

³Z T ¯ ´
¯
Jt = 1{t<τ } Bt EQ Bu−1 eΓt −Γu Zu dΓu ¯ Ft .
t

Finally, it follows from (3.1) that

Kt = 1{t<τ } eΓt Bt EQ (1{τ >T } BT−1 X | Ft ).

Since the random variables X and BT are FT -measurable, we also have

¡ ¢
Kt = 1{t<τ } eΓt Bt EQ (GT BT−1 X | Ft ) = 1{t<τ } Bt EQ BT−1 XeΓt −ΓT | Ft .

Both formulae of the proposition are obtained upon summation. ¤

Let us note that St = 1{t<τ } Set , where the F-adapted process Se represents the pre-default value
of a defaultable claim (X, A, Z, τ ). The next result is a straightforward consequence of Proposition
3.1.2.

Corollary 3.1.2 Assume that F (and thus also Γ) is an increasing, continuous process. Then
the pre-default value of a defaultable claim (X, A, Z, τ ) coincides with the pre-default value of a
R
b 0, τ ), where the process A
defaultable claim (X, A, bt = At + t Zu dΓu for
b is given by the formula A
0
t ∈ [0, T ].

Let us consider the case of a default time τ that admits the F-intensity process γ. The second
formula in Proposition 3.1.2 now becomes
³Z Ru ¯ ´
¯
St = 1{t<τ } EQ e− t (rv +γv )dv (dAu + γu Zu du) ¯ Ft
]t,T ]
³ RT ¯ ´
¯
+ 1{t<τ } EQ e− t (rv +γv )dv X ¯ Ft .

To obtain a more intuitive representation for the last expression, we introduce the default-risk-
b which is
adjusted interest rate rb = r + γ and the associated default-risk-adjusted savings account B,
given by the formula
³Z t ´
Bbt = exp rbu du . (3.8)
0

Although the process B b does not represent the price of a tradeable security, it enjoys the features
of the savings account B. Specifically, B b is an F-adapted, continuous process of finite variation
(typically, though not necessarily, an increasing process). In terms of the process B, b we have
³Z Z T ¯ ´
bt EQ
St = 1{t<τ } B bu−1 dAu +
B B b −1 X ¯¯ Ft .
bu−1 Zu γu du + B (3.9)
T
]t,T ] t
3.1. HAZARD PROCESS AND ITS APPLICATIONS 93

3.1.4 Defaultable Bonds

Consider a defaultable zero-coupon bond with the par (face) value L and maturity date T . We will
re-examine the following recovery schemes: the fractional recovery of par value and the fractional
recovery of Treasury value; recall that these schemes were already studied in Section 2.1 in the case
of deterministic intensity. The fractional recovery of market value scheme is more difficult to deal
with, though it is still tractable (cf. Duffie et al. [65] and Duffie and Singleton [66]).
We assume in this subsection that τ admits the F-hazard rate γ.

Fractional Recovery of Par Value

Under this scheme, a fixed fraction of the bond face value is paid to the bondholders at the time of
default. Formally, we deal here with a defaultable claim (X, 0, Z, τ ), which settles at time T , with
the promised payoff X = L, where L stands for the bond’s face value and with the constant recovery
process Z = δL for some δ ∈ [0, 1]. The ex-dividend price at time t ∈ [0, T ] of the bond is thus given
by the following expression
¡ ¯ ¢
Dδ (t, T ) = LBt EQ δBτ−1 1{t<τ ≤T } + BT−1 1{T <τ } ¯ Gt .

If τ admits the F-intensity process γ then the pre-default value of the bond equals
³ Z T ¯ ´
e δ b
D (t, T ) = LBt EQ δ B b −1 ¯¯ Ft .
bu−1 γu du + B (3.10)
T
t

Fractional Recovery of Treasury Value

According to this convention, the fixed fraction of the face value is paid to bondholders at maturity
date T . A corporate zero-coupon bond is now given by a defaultable claim (X, 0, Z, τ ) with the
promised payoff X = L and the recovery process Zt = δLB(t, T ) where, as usual, B(t, T ) stands for
the price at time t of a unit zero-coupon Treasury bond with maturity T . The defaultable bond is
here equivalent to a single contingent claim Y , which settles at time T and equals
¡ ¢
Y = L 1{τ >T } + δ1{τ ≤T } .

The ex-dividend price Dδ (t, T ) of this claim at time t < T thus equals
¡ ¯ ¢
Dδ (t, T ) = LBt EQ BT−1 (δ1{T ≥τ } + 1{T <τ } ) ¯ Gt

or, equivalently, ¡ ¯ ¢
St = LBt EQ δBτ−1 B(τ, T )1{t<τ ≤T } + BT−1 1{T <τ } ¯ Gt .
e δ (t, T ) of a defaultable bond that is subject to the fractional recovery of
The pre-default value D
Treasury value scheme is given by the expression
³ Z T ¯ ´
e δ (t, T ) = LB
D bt EQ δ B b −1 ¯¯ Ft .
bu−1 B(u, T )γu du + B
T
t

3.1.5 Martingales Associated with Default Time

We will now examine some important martingales associated with τ .

Proposition 3.1.3 (i) The process Lt = (1 − Ht )eΓt is a G-martingale.

(ii) If X is an F-martingale and the process XL is integrable then it is a G-martingale.
(iii) If the process F (or, equivalently, Γ) is increasing and continuous then the process Mt =
Ht − Γ(t ∧ τ ) is a G-martingale.
94 CHAPTER 3. HAZARD PROCESS APPROACH

Proof. (i) From Lemma 3.1.2, we obtain, for any t ≤ s,

EQ (Ls | Gt ) = 1{t<τ } eΓt EQ (1{s<τ } eΓs | Ft ) = 1{t<τ } eΓt = Lt ,

since the tower rule yields (obviously, Ft ⊂ Fs )

EQ (1{s<τ } eΓs |Ft ) = EQ (Q(τ > s | Fs )eΓs |Ft ) = 1.

(ii) Using again Lemma 3.1.2, we get, for any t ≤ s,

EQ (Ls Xs | Gt ) = EQ (1{s<τ } Ls Xs | Gt )
¡ ¯ ¢
= 1{t<τ } eΓt EQ 1{s<τ } eΓs Xs ¯ Ft
¡ ¯ ¢
= 1{t<τ } eΓt EQ EQ (1{s<τ } | Fs )eΓs Xs ¯ Ft
= Lt X t .

(iii) Note that H is a process of finite variation and Γ is an increasing, continuous process. Hence
from the integration by parts formula, we obtain

dLt = (1 − Ht )eΓt dΓt − eΓt dHt .

Moreover, the process Mt = Ht − Γ(t ∧ τ ) can be represented as follows

Z Z t Z
Mt = dHu − (1 − Hu ) dΓu = − e−Γu dLu ,
]0,t] 0 ]0,t]

and thus it is a G-martingale, since L is G-martingale and e−Γt is a bounded process. It should
be noted that if the hazard process Γ is not assumed to be increasing then the Itô differential deΓt
becomes more complicated. ¤

Note that the process F (or, equivalently, Γ) is not necessarily of finite variation. Hence part
(iii) in Proposition 3.1.3 does not yield the general form of the Doob-Meyer decomposition of the
submartingale H. For simplicity, in the next result we shall assume that F is a continuous process.
It is worth noting that part (iii) in Proposition 3.1.3 is a consequence of Proposition 3.1.4, since for
a continuous and increasing F we have that F = C = 1 − e−Γt .

Proposition 3.1.4 Assume that F is a continuous process with the Doob-Meyer decomposition
F = N + C. Then the process M = (Mt , t ∈ R+ ), which is given by the formula
Z t∧τ
dCu
Mt = Ht − , (3.11)
0 1 − Fu
is a G-martingale.

Proof. We split the proof into two steps.

First step. We shall prove that, for any t ≤ s,

EQ (Hs | Gt ) = Ht + 1{t<τ } eΓt EQ (Cs − Ct | Ft ). (3.12)

Indeed, we have that

EQ (Hs | Gt ) = 1 − Q(s < τ | Gt ) = 1 − 1{t<τ } eΓt EQ (1 − Fs | Ft )

= 1 − 1{t<τ } eΓt EQ (1 − Ns − Cs | Ft )
¡ ¢
= 1 − 1{t<τ } eΓt 1 − Nt − Ct − EQ (Cs − Ct | Ft )
¡ ¢
= 1 − 1{t<τ } eΓt 1 − Ft − EQ (Cs − Ct | Ft )
= 1{t≥τ } + 1{t<τ } eΓt EQ (Cs − Ct | Ft ).
3.1. HAZARD PROCESS AND ITS APPLICATIONS 95

Second step. Let us denote Z Z

t t
dCu
Ut = = eΓu dCu .
0 1 − Fu 0
We shall prove that, for any t ≤ s,
EQ (Us∧τ | Gt ) = Ut∧τ + 1{t<τ } eΓt EQ (Cs − Ct | Ft ).
From Lemma 3.1.3, we obtain
µZ ∞ ¯ ¶
¯
EQ (Us∧τ | Gt ) = Ut∧τ 1{t≥τ } + 1{t<τ } eΓt EQ Us∧u dFu ¯ Ft
t
µZ s Z ∞ ¯ ¶
Γt ¯
= Ut∧τ 1{t≥τ } + 1{t<τ } e EQ Uu dFu + Us dFu ¯ Ft
µZt s s
¯ ¶
Γt ¯
= Ut∧τ 1{t≥τ } + 1{t<τ } e EQ Uu dFu + Us (1 − Fs ) ¯ Ft .
t

Using the integration by parts formula and the fact that U is a continuous process of finite variation,
we obtain
d(Ut (1 − Ft )) = −Ut dFt + (1 − Ft ) dUt = −Ut dFt + dCt .
Consequently,
Z s
Uu dFu + Us (1 − Fs ) = −Us (1 − Fs ) + Ut (1 − Ft ) + Cs − Ct
t
+ Us (1 − Fs ) = Ut (1 − Ft ) + Cs − Ct .
It follows that, for any t ≤ s,
¡ ¢
EQ (Us∧τ | Gt ) = 1{t≥τ } Ut∧τ + 1{t<τ } eΓt EQ Ut (1 − Ft ) + Cs − Ct | Ft
= Ut∧τ + 1{t<τ } eΓt EQ (Cs − Ct | Ft ).
By combining the formula above with (3.12), we conclude that the process M given by (3.11) is a
G-martingale. ¤

Proposition 3.1.5 Assume that the bounded submartingale F admits the Doob-Meyer decomposi-
tion F = N + C. Then the process M = (Mt , t ∈ R+ ), which is given by the formula
Z t∧τ
dCu
Mt = Ht − , (3.13)
0 1 − Fu−
is a G-martingale.

Proof. In the first part of the proof, we proceed along the same lines as in the proof of Proposition
2.2.1. In view of Lemma 3.1.2, we find that, in the present case, it is enough to show that the
following equalities hold, for every t ≤ s,
³Z dCu ¯¯ ´
I := EQ ¯ Ft = EQ (Fs − Ft | Ft ) = EQ (Cs − Ct | Ft ),
]t,s∧τ ] 1 − Fu−

where the second equality is simply a consequence of the definition of C. We have

Z Z
³ dCu dCu ¯¯ ´
I = EQ 1{s<τ } + 1{t<τ ≤s} ¯ Ft
]t,s] 1 − Fu− ]t,s∧τ ] 1 − Fu−
Z Z
³ ³ dCu ¯¯ ´ dCu ¯¯ ´
= EQ EQ 1{s<τ } ¯ Fs + 1{t<τ ≤s} ¯ Ft
]t,s] 1 − Fu− ]t,s∧τ ] 1 − Fu−
³ Z Z Z ¯ ´
dCu dCv ¯
= EQ (1 − Fs ) + dCu ¯ Ft
]t,s] 1 − F u− ]t,s] ]t,u] 1 − F v−
³ Z ¯ ´
¯
= EQ (Λs − Λt )(1 − Fs ) + (Λu − Λt ) dCu ¯ Ft ,
]t,s]
96 CHAPTER 3. HAZARD PROCESS APPROACH

where the third equality follows from formula (3.5) and where we denote, for every r ∈ R+ ,
Z
dCu
Λt = . (3.14)
]0,t] 1 − Fu−
Since Λ is an F-predictable process and N is an F-martingale, we obtain
³Z ¯ ´
¯
EQ (Λu − Λt ) dNu ¯ Ft = 0,
]t,s]

and this in turn yields

³ Z ¯ ´
¯
I = EQ (Λs − Λt )(1 − Fs ) + (Λu − Λt ) dCu ¯ Ft
]t,s]
³ Z ¯ ´
¯
= EQ (Λs − Λt )(1 − Fs ) + (Λu − Λt ) dFu ¯ Ft .
]t,s]

Recall that our goal is to show that I = EQ (Cs − Ct | Ft ). To this end, we observe that
Z Z
(Λu − Λt ) dFu = −Λt (Fs − Ft ) + Λu dFu .
]t,s] ]t,s]

Since Λ is a process of finite variation, Itô’s product rule yields

Z Z
Λu dFu = Λs Fs − Λt Ft − Fu− dΛu . (3.15)
]t,s] ]t,s]

Finally, it follows from (3.14) that

Z
Fu− dΛu = Λs − Λt − Cs + Ct .
]t,s]

Combining the above formulae, we conclude that

Z
(Λs − Λt )(1 − Fs ) + (Λu − Λt ) dFu = Cs − Ct . (3.16)
]t,s]

This completes the proof. ¤

3.1.6 F-Intensity of Default Time

Assume that F admits the Doob-Meyer decomposition F = N +C, Rwhere the process C is absolutely
t
continuous with respect to the Lebesgue measure, so that Ct = 0 cu du for some F-progressively
measurable process c.

Definition 3.1.2 The F-intensity of default time τ is a non-negative and F-progressively measurable
process λ such that the process Z t∧τ
Mt = Ht − λu du
0
is a G-martingale.

Note that under the present assumptions the F-intensity is given by the formula λt = ct (1−Ft )−1
for every t ∈ R+ . If we assume that the process F is absolutely continuous, then we recover the
definition of the hazard rate of Section 3.1.2, since manifestly the equality λ = γ holds in that case.
The proof of the next lemma is left to the reader.

Lemma 3.1.4 The F-intensity of default time satisfies, for almost every t ∈ R+ ,
1 Q(t < τ < t + h | Ft )
λt = lim .
h→0 h Q(t < τ | Ft )
3.1. HAZARD PROCESS AND ITS APPLICATIONS 97

3.1.7 Reduction of the Reference Filtration

In this section, we follow Jeanblanc and LeCam [97]. Suppose that F e is a sub-filtration of F, so that
e by setting Get = Fet ∨ Ht for every t ∈ R+ .
Fet ⊂ Ft for every t ∈ R+ . We define the full filtration G
The hazard process of τ with respect to Fe is given by Γ e t = − ln G
e t with

e t = Q(t < τ | Fet ) = EQ (Gt | Fet ).

G

For any Q-integrable random variable Y , Lemma 3.1.2 implies that

e
EQ (1{t<τ } Y | Get ) = 1{t<τ } eΓt EQ (1{t<τ } Y | Fet ).

In particular, if Y is a Fes -measurable random variable then, for every t ≤ s,

e
EQ (1{s<τ } Y | Get ) = 1{t<τ } eΓt EQ (G
e s Y | Fet ).

From the obvious equality

EQ (1{s<τ } Y | Get ) = EQ (EQ (1{s<τ } Y | Gt ) | Get ),

we also obtain
³ ´
EQ (1{s<τ } Y | Get ) = EQ 1{t<τ } eΓt EQ (Gs Y | Ft ) | Get
³ ¯ ´
e
= 1{t<τ } eΓt EQ 1{t<τ } eΓt EQ (Gs Y | Ft ) ¯ Fet .

From the uniqueness of the pre-default F-adapted process, it can now be deduced that the following
result is true.

Lemma 3.1.5 For any Q-integrable and Fes -measurable random variable Y we have, for every t ≤ s,
³ ¯ ´
e s Y | Fet ) = EQ 1{t<τ } eΓt EQ (Gs Y | Ft ) ¯ Fet .
EQ (G

Proof. We provide a direct proof of the asserted formula. We have

³ ¯ ´
EQ 1{t<τ } EQ (Gs Y | Ft )eΓt ¯ Fet
³ ¯ ´
= EQ EQ (1{t<τ } | Ft )eΓt EQ (Gs Y | Ft ) ¯ Fet
³ ¯ ´
= EQ EQ (Gs Y | Ft ) ¯ Fet = EQ (Gs Y | Fet )
³ ¯ ´
= EQ EQ (Gs | Fes )Y ¯ Fet = EQ (G e s Y | Fet ),

since we assumed that Y is Fes -measurable. ¤

Let F = N + C be the Doob-Meyer decomposition of the submartingale F with R t respect to F
and let us assume that C is absolutely continuous with respect to t, that is, Ct = 0 cu du. Since C
is an increasing process, it is easily seen that the process Cet = EQ (Ct | Fet ) is a submartingale with
e Let us denote by C
respect to F. e = ze + α
e its Doob-Meyer decomposition with respect to F e and let
us set Net = EQ (Nt | Fet ). Since N e
e is an F-martingale, we see that the submartingale

Fet = Q(t ≥ τ | Fet ) = EQ (Ft | Fet )

admits the Doob-Meyer decomposition Fe = m e +α e

e, where the F-martingale part equals m
e =N e + ze.
The next lemma furnishes an explicit relationship between the increasing processes C and α
e.
98 CHAPTER 3. HAZARD PROCESS APPROACH

Rt
Lemma 3.1.6 Let Ct = 0 cu du be the F-predictable increasing process in the Doob-Meyer decom-
e
position of the F-submartingale F . Then the F-predictable increasing process in the Doob-Meyer
e
decomposition F = m e
e of the F-submartingale Fe equals, for every t ∈ R+ ,
e +α
Z t
α
et = EQ (cu | Feu ) du. (3.17)
0

Proof. To establish (3.17), we will show that the process

Z t
MtF = EQ (Ft | Fet ) − EQ (cu | Feu ) du
0

e
is an F-martingale. e
Clearly, the process M F is integrable and F-adapted. Moreover, for every t ≤ s,
³ Z s ¯ ´
¯
EQ (MsF | Fet ) = EQ EQ (Fs | Fes ) − EQ (cu | Feu ) du ¯ Fet
0
³Z t ¯ ´
¯
= EQ (Fs | Fet ) − EQ EQ (cu | Feu ) du ¯ Fet
0
³Z s ¯ ´
¯
− EQ EQ (cu | Feu ) du ¯ Fet
t
³Z t ¯ ´ ³Z s ¯ ´
et + EQ ¯ ¯
=N cu du ¯ Fet + EQ cu du ¯ Fet
0 t
Z t ³Z s ¯ ´
¯
− EQ (cu | Feu ) du − EQ EQ (cu | Feu ) du ¯ Fet
0 t

and thus
³Z s ¯ ´ ³Z s ¯ ´
¯ ¯
EQ (MsF | Fet ) = MtF + EQ cu du ¯ Fet − EQ EQ (cu | Feu ) du ¯ Fet
t t
Z s Z s ³ ¯ ´
¯
= MtF + EQ (cu | Fet ) du − EQ EQ (cu | Feu ) ¯ Fet du
t t
Z s Z s
F
= Mt + e
EQ (cu | Ft ) du − EQ (cu | Fet ) du = MtF .
t t

e
We have thus shown that the process M F is an F-martingale. e
Moreover, the F-adapted process αe,
e
given by (3.17) is manifestly continuous, and thus it is F-predictable. By virtue of uniqueness of the
Doob-Meyer decomposition, we conclude that M F = m e and formula (3.17) is valid. ¤

ct = EQ (ct | Fet ). The process

Corollary 3.1.3 Let us denote e
Z t∧τ
ft = Ht − e
cu
M du
0 1 − Feu

e
is a G-martingale e
and the F-intensity et = e
of τ is equal to λ e −1
ct G t .

Remark 3.1.1 It is worth noting that, typically, the inequality

EQ (λt | Fet ) = EQ (ct G−1 e e e −1 e

t | Ft ) 6= EQ (ct | Ft )Gt = λt

e
holds. This means that the F-intensity of τ is not given by the optional projection of the F-intensity
e
on the reduced filtration F, in general.
3.2. HYPOTHESIS (H) 99

3.1.8 Enlargement of Filtration

Assume that G is any enlarged filtration. Then we may work directly with the filtration G, provided
that the decomposition of any F-martingale in this filtration is known up to time τ . For example, if W
is an F-Brownian motion, then it is not necessarily a G-martingale and its Doob-Meyer decomposition
in the G filtration up to time τ is
Z t∧τ
dhW, Giu
Wt∧τ = βt∧τ + ,
0 Gu−

where (βt∧τ , t ∈ R+ ) is a continuous G-martingale with the increasing process t ∧ τ . Suppose, for
instance, that the dynamics of an asset S are given by
¡ ¢
dSt = St rt dt + σt dWt

in the default-free framework, that is, with respect to the filtration F. Then its dynamics with
respect to the enlarged filtration G are
³ dhW, Git ´
dSt = St rt dt + σt + σt dβt
Gt−

provided that we restrict our attention to the behavior of S prior to default. We conclude that the
possibility of default changes the drift term in the price dynamics. The interested reader is referred
to Mansuy and Yor [121] for more information.

3.2 Hypothesis (H)

As already mentioned above, an arbitrary F-martingale does not remain a G-martingale, in general.
We shall now study a particular case in which this martingale invariance property (also known as
the immersion property between F and G) actually holds.

3.2.1 Equivalent Forms of Hypothesis (H)

Once again we consider a general situation where G = H ∨ F for some reference filtration F. We
shall now examine the so-called hypothesis (H) which reads as follows.

Hypothesis (H) Every F-local martingale is a G-local martingale.

This hypothesis implies, for instance, that any F-Brownian motion remains a Brownian motion
with respect to the enlarged filtration G. It was studied, among others by, Brémaud and Yor [31],
Jeanblanc and Le Cam [96], Mazziotto and Szpirglas [122], Kusuoka [106], and Nikeghbali and Yor
[129].
Let us first examine some equivalent forms of hypothesis (H) (see, e.g., Dellacherie and Meyer
[59]).

Lemma 3.2.1 Assume that G = F ∨ H, where F is an arbitrary filtration and H is generated by the
process Ht = 1{t≥τ } . Then the following conditions are equivalent to hypothesis (H) .
(i) For any t, h ∈ R+ , we have

Q(τ ≤ t | Ft ) = Q(τ ≤ t | Ft+h ). (3.18)

(i0 ) For any t ∈ R+ , we have

Q(τ ≤ t | Ft ) = Q(τ ≤ t | F∞ ). (3.19)
100 CHAPTER 3. HAZARD PROCESS APPROACH

(ii) For any t ∈ R+ , the σ-fields F∞ and Gt are conditionally independent given Ft under Q, that
is,
EQ (ξ η | Ft ) = EQ (ξ | Ft ) EQ (η | Ft )
for arbitrary bounded, F∞ -measurable random variable ξ and any bounded, Gt -measurable random
variable η.
(iii) For any t ∈ R+ and any u ≥ t, the σ-fields Fu and Gt are conditionally independent given Ft .
(iv) For any t ∈ R+ and any bounded, F∞ -measurable random variable ξ: EQ (ξ | Gt ) = EQ (ξ | Ft ).
(v) For any t ∈ R+ , and any bounded, Gt -measurable random variable η: EQ (η | Ft ) = EQ (η | F∞ ).

Proof. If hypothesis (H) holds then (3.19) is valid as well. If (3.19) holds then the fact that Ht
is generated by the events {τ ≤ s}, s ≤ t, proves that the σ-fields F∞ and Ht are conditionally
independent given Ft . The desired property now follows. The equivalence between (3.19) and (3.18)
is left to the reader.
Using the monotone class theorem, it can be shown that conditions (i) and (i0 ) are equivalent.
The proof of equivalence of conditions (i0 )–(v) can be found, for instance, in Section 6.1.1 of Bielecki
and Rutkowski [20] (for related results, see Elliott et al. [70]).
Let us show, for instance, that condition (iv) and hypothesis (H) are equivalent.
Assume first that hypothesis (H) holds. Consider any bounded, F∞ -measurable random variable
ξ. Let Mt = EQ (ξ | Ft ) be the martingale associated with ξ. Of course, M is a martingale with
respect to F. Then hypothesis (H) implies that M is a local martingale with respect to G and thus
a G-martingale, since M is bounded (any bounded local martingale is a martingale). We conclude
that Mt = EQ (ξ | Gt ) and thus (iv) holds.
Suppose now that (iv) holds. First, we note that the standard truncation argument shows that
the boundedness of a random variable ξ in condition (iv) can be replaced by the assumption that ξ
is Q-integrable. Hence any F-martingale M is an G-martingale, since M is clearly G-adapted and
we have, for every t ≤ s,
Mt = EQ (Ms | Ft ) = EQ (Ms | Gt ),
where the second equality follows from (iv).
Suppose now that M is an F-local martingale. Then there exists an increasing sequence of F-
stopping times τn such that limn→∞ τn = ∞, for any n the stopped process M τn follows a uniformly
integrable F-martingale. Hence M τn is also a uniformly integrable G-martingale and this means
that M is a G-local martingale. ¤

Remarks 3.2.1 (i) Equality (3.19) appears in numerous papers on default risk, typically without
any reference to hypothesis (H). For example, in Madan and Unal [120], the main theorem follows
from the fact that (3.19) holds (see the proof of B9 in the appendix of [120]). This is also the case
for the model studied by Wong [142].
(ii) If τ is F∞ -measurable and (3.19) holds then τ is an F-stopping time. If τ is an F-stopping time
then equality (3.18) holds.
(iii) Though hypothesis (H) is not necessarily valid, in general, it is satisfied when τ is constructed
through the so-called canonical approach (or for Cox processes). It also holds when τ is independent
of F∞ (see Greenfield [84]).
(iv) If hypothesis (H) holds then from the condition

Q(τ ≤ t | Ft ) = Q(τ ≤ t | F∞ ), ∀ t ∈ R+ ,

we deduce easily that F is an increasing process. The property that F is increasing is equivalent
to the fact that any F-martingale stopped at time τ is a G-martingale. Nikeghbali and Yor [129]
proved that this is equivalent to EQ (Mτ ) = M0 for any bounded F-martingale M .
(v) The hypothesis (H) was also studied by Florens and Fougère [73], who coined the term non-
causality. For more comments on hypothesis (H), we refer to Elliott et al. [70].
3.2. HYPOTHESIS (H) 101

Proposition 3.2.1 Assume that hypothesis (H) holds. If a process X is an F-martingale then the
processes XL and [L, X] are G-local martingales.

Proof. From Proposition 3.1.3(ii), the process XL is a G-martingale. Since

Z Z
[L, X]t = Lt Xt − Lu− dXu − Xu− dLu ,
]0,t] ]0,t]

and the process X is an F-martingale (and thus also a G-martingale), we conclude that the process
[L, X] is a G-local martingale, as the sum of three G-local martingales. ¤

3.2.2 Canonical Construction of a Default Time

We shall now briefly describe the most commonly used construction of a default time associated
with a given hazard process Γ. It should be stressed that the random time obtained through this
particular method – which will be called the canonical construction in what follows – has certain
specific features that are not necessarily shared by all random times with a given F-hazard process
Γ. We assume that we are given an F-adapted, right-continuous, increasing process Γ defined on
a filtered probability space (Ω, G, Q). As usual, we assume that Γ0 = 0 and Γ∞ = +∞. In many
instances, Γ is given by the equality, for every t ∈ R+ ,
Z t
Γt = γu du
0

for some non-negative, F-progressively measurable intensity process γ.

To construct a random time τ , we postulate that the underlying probability space (Ω, G, Q)
is sufficiently rich to support a random variable ξ, which is uniformly distributed on the interval
[0, 1] and independent of the filtration F under Q. In this version of the canonical construction, Γ
represents the F-hazard process of τ under Q.
We define the random time τ : Ω → R+ by setting

τ = inf { t ∈ R+ : e−Γt ≤ ξ } = inf { t ∈ R+ : Γt ≥ η },

where the random variable η = − ln ξ has a unit exponential law under Q. It is not difficult to find
the process Ft = Q(τ ≤ t | Ft ). Indeed, since clearly {τ > t} = {ξ < e−Γt } and the random variable
Γt is F∞ -measurable, we obtain

Q(τ > t | F∞ ) = Q(ξ < e−Γt | F∞ ) = Q(ξ < e−x )x=Γt = e−Γt .

Consequently, we have
¡ ¯ ¢
1 − Ft = Q(τ > t | Ft ) = EQ Q(τ > t | F∞ ) ¯ Ft = e−Γt ,

and so F is an F-adapted, right-continuous, increasing process. It is also clear that the process Γ
represents the F-hazard process of τ under Q. As an immediate consequence of the last two formulae,
we obtain the following property of the canonical construction of the default time (cf. (3.19))

Q(τ ≤ t | F∞ ) = Q(τ ≤ t | Ft ), ∀ t ∈ R+ . (3.20)

To summarize, we have that

Q(τ ≤ t | F∞ ) = Q(τ ≤ t | Fu ) = Q(τ ≤ t | Ft ) = 1 − e−Γt

for any two dates 0 ≤ t ≤ u.

102 CHAPTER 3. HAZARD PROCESS APPROACH

3.2.3 Stochastic Barrier

Suppose that
Q(τ ≤ t | Ft ) = Q(τ ≤ t | F∞ ) = 1 − e−Γt ,
where Γ is a continuous, strictly increasing, F-adapted process. Our goal is to show that there exists
a random variable Θ, independent of F∞ , with exponential distribution of parameter 1, such that
τ = inf {t ≥ 0 : Γt > Θ}. Let us set Θ := Γτ . Then

{t < Θ} = {t < Γτ } = {Ct < τ },

where C is the right inverse of Γ, so that ΓCt = t. Therefore

Q(Θ > u | F∞ ) = e−ΓCu = e−u .

We have thus established the required properties, namely, the probability distribution of Θ and its
independence of the σ-field F∞ . Furthermore,

τ = inf{ t ∈ R+ : Γt > Γτ } = inf{ t ∈ R+ : Γt > Θ}.

3.3 Predictable Representation Theorem

Kusuoka [106] established the following representation theorem in which the reference filtration F is
generated by a Brownian motion.

Theorem 3.3.1 Assume that hypothesis (H) is satisfied under Q. Then any square-integrable mar-
tingale with respect to G admits a representation as the sum of a stochastic integral with respect
to the Brownian motion and a stochastic integral with respect to the discontinuous martingale M
associated with τ .

To derive a version of the predictable representation theorem, we will assume, for simplicity, that
F is continuous and Ft < 1 for every t ∈ R+ . Since hypothesis (H) holds, F is also an increasing
process and thus
dFt = e−Γt dΓt , deΓt = eΓt dΓt . (3.21)
The following result extends Proposition 2.2.6 to the case of the reference filtration F that only
supports continuous martingale; in particular, this result covers the case when F is the Brownian
filtration.

Theorem 3.3.2 Suppose that hypothesis (H) holds under Q and that any F-martingale is con-
tinuous. Then the martingale Mth = EQ (hτ | Gt ), where h is an F-predictable process such that
EQ |hτ | < ∞, admits the following decomposition in the sum of a continuous martingale and a
discontinuous martingale
Z t∧τ Z
Mth = mh0 + eΓu dmhu + h
(hu − Mu− ) dMu , (3.22)
0 ]0,t∧τ ]

where mh is the continuous F-martingale given by

³Z ∞ ¯ ´
¯
mht = EQ hu dFu ¯ Ft
0

and M is the discontinuous G-martingale defined as Mt = Ht − Γt∧τ .

3.4. GIRSANOV’S THEOREM 103

Proof. We start by noting that

³Z ∞ ¯ ´
¯
Mth = EQ (hτ | Gt ) = 1{t≥τ } hτ + 1{t<τ } eΓt EQ hu dFu ¯ Ft
t
³ Z t ´
= 1{t≥τ } hτ + 1{t<τ } eΓt mht − hu dFu . (3.23)
0

We will sketch two slightly different derivations of (3.22).

First derivation. Let the process J be given by the formula, for t ∈ R+ ,

³ Z t ´
Jt = eΓt mht − hu dFu .
0

Noting that Γ is a continuous increasing process and mh is a continuous martingale, we deduce from
the Itô integration by parts formula that
³ Z t ´
Γt
dJt = e dmht Γt
− e ht dFt + mht − hu dFu eΓt dΓt
0
= eΓt dmht − eΓt ht dFt + Jt dΓt .

Therefore, from (3.21),

dJt = eΓt dmht + (Jt − ht ) dΓt
or, in the integrated form,
Z t Z t
Jt = M0h + eΓu dmhu + (Ju − hu ) dΓu .
0 0

h
Note that Jt = Mth = Mt− on the event {t < τ }. Therefore, on the event {t < τ },
Z t∧τ Z t∧τ
Mth = M0h + eΓu dmhu + h
(Mu− − hu ) dΓu .
0 0

From (3.23), the jump of M h at time τ equals

hτ − Jτ = hτ − Mτh− = Mτh − Mτh− .

Equality (3.22) now easily follows.

Second derivation. Equality (3.23) can be re-written as follows

Z t ³ Z t ´
Mth = hu dHu + (1 − Ht )e Γt
mht − hu dFu .
0 0

Hence formula (3.22) can be obtained directly by the Itô integration by parts formula. ¤

3.4 Girsanov’s Theorem

We now start by defining a random time τ on a probability space (Ω, G, Q) and we postulate that
it admits the continuous F-hazard process Γ under Q. Hence, from Proposition 3.1.4, we know that
the process Mt = Ht − Γt∧τ is a G-martingale. We postulate that hypothesis (H) holds under Q.
Finally, we postulate that the reference filtration F is generated by an F- (hence also G-) Brownian
motion under Q.
104 CHAPTER 3. HAZARD PROCESS APPROACH

Let us fix T > 0. For a probability measure P equivalent to Q on (Ω, GT ) we introduce the
G-martingale (ηt , t ∈ [0, T ]) by setting
dP ¯¯
ηt := ¯ Gt = EQ (X | Gt ), Q-a.s. (3.24)
dQ
Note that X = ηT is here some GT -measurable random variable such that Q(X > 0) = 1 and
EQ X = 1.
Using Theorem 3.3.1, we deduce that the Radon-Nikodým density process η admits the following
representation, for every t ∈ [0, T ],
Z t Z
ηt = 1 + ξu dWu + ζu dMu ,
0 ]0,t]

where ξ and ζ are G-predictable stochastic processes. Since η is a strictly positive martingale, by
−1 −1
setting θt = ξt ηt− and κt = ζt ηt− , we obtain
Z
¡ ¢
ηt = 1 + ηu− θu dWu + κu dMu (3.25)
]0,t]

where θ and κ are G-predictable processes, with κ > −1. This means the process η is the Doléans
exponential, more explicitly
µZ · ¶ µZ ¶
(1) (2)
ηt = Et θu dWu Et κu dMu = ηt ηt , (3.26)
0 ]0, · ]

where we write µZ ¶ µZ Z ¶
· t t
(1) 1
ηt = Et θu dWu = exp θu dWu − θu2 du ,
0 0 2 0
and
µZ ¶
(2)
ηt = Et κu dMu (3.27)
]0, · ]
µZ Z t∧τ ¶
= exp ln(1 + κu ) dHu − κu γu du .
]0,t] 0

Then we have the following extension of Girsanov’s theorem.

Theorem 3.4.1 Let P be a probability measure on (Ω, GT ) equivalent to Q. If the Radon-Nikodým

ct , t ∈
density of P with respect to Q is given by (3.24) with η satisfying (3.25) then the process (W
[0, T ]), given by
Z t
Wct = Wt − θu du,
0
ct , t ∈ [0, T ]), given
is a Brownian motion with respect to the filtration G under P and the process (M
by Z t∧τ Z t∧τ
c
Mt := Mt − κu dΓu = Ht − (1 + κu ) dΓu ,
0 0
c under P.
is a G-martingale orthogonal to W

Proof. Note first that, for every t ∈ [0, T ], we have

ct ) = W
d(ηt W ct dηt + ηt− dW
ct + d[W
c , η]t
=Wct dηt + ηt− dWt − ηt− θt dt + ηt− θt d[W, W ]t
ct dηt + ηt− dWt .
=W
3.4. GIRSANOV’S THEOREM 105

This shows that W c is a G-local martingale under P. Since the quadratic variation of W c under P
c c c
equals [W , W ]t = t and W is continuous, using the Lévy characterization theorem, we conclude that
c follows a Brownian motion under P. Similarly, for every t ∈ [0, T ],
W

ct ) = M
d(ηt M ct dηt + ηt− dMct + d[M c, η]t
=Mct dηt + ηt− dMt − ηt− κt dΓt∧τ + ηt− κt dHt
=Mct dηt + ηt− (1 + κt ) dMt .

c is a G-martingale under P. To complete the proof of the proposition, it suffices

We conclude that M
c c is a process of finite variation; hence W
to observe that W is a continuous process and M c and M c
are orthogonal G-martingales under P. ¤

Corollary 3.4.1 Let Y be a G-martingale under P, where the probability measure P is defined in
Theorem 3.4.1. Then Y admits the following integral representation
Z t Z
Yt = Y0 + cu +
ξu∗ dW cu ,
ζu∗ dM (3.28)
0 ]0,t]

where ξ ∗ and ζ ∗ are G-predictable stochastic processes.

Proof. Consider the process Ye given by the formula

Z Z
e
Yt = −1
ηu− d(ηu Yu ) − −1
ηu− Yu− dηu .
]0,t] ]0,t]

It is clear that Ye is a G-local martingale under Q. Notice also that Itô’s formula yields
−1 −1 −1
ηu− d(ηu Yu ) = dYu + ηu− Yu− dηu + ηu− d[Y, η]u ,

and thus Z
Yt = Y0 + Yet − −1
ηu− d[Y, η]u . (3.29)
]0,t]

From the predictable representation theorem, we know that the process Ye admits the following
integral representation
Z t Z
Yet = Y0 + ξeu dWu + ζeu dMu (3.30)
0 ]0,t]

for some G-predictable processes ξe and ζ.

e Therefore,

dYt = ξet dWt + ζet dMt − ηt−

−1
d[Y, η]t = ξet dW
ct + ζet (1 + κt )−1 dM
ct

since (3.25) combined with (3.29)–(3.30) yield

−1
ηt− d[Y, η]t = ξet θt dt + ζet κt (1 + κt )−1 dHt .

To derive the last equality we observe, in particular, that in view of (3.29) we have (we take into
account continuity of Γ)
∆[Y, η]t = ηt− ζet κt dHt − κt ∆[Y, η]t .

We conclude that Y satisfies (3.28) with the processes ξ ∗ = ξe and ζ ∗ = ζ(1

e + κ)−1 , where in turn
e e
the processes ξ and ζ are given by (3.30). ¤
106 CHAPTER 3. HAZARD PROCESS APPROACH

3.5 Invariance of Hypothesis (H)

Kusuoka [106] shows by means of a counter-example (see Example 3.5.1) that hypothesis (H) is not
invariant with respect to an equivalent change of the underlying probability measure, in general. It
is worth noting that his counter-example is based on two filtrations, H1 and H2 , generated by the
two random times τ 1 and τ 2 and H1 is chosen to play the role of the reference filtration F. We shall
argue that in the case where F is generated by a Brownian motion, the above-mentioned invariance
property is valid under mild technical assumptions.
Let us first examine a general setup in which G = F ∨ H, where F is an arbitrary filtration and
H is generated by the default process H. We say that Q is locally equivalent to P if Q is equivalent
to P on (Ω, Gt ) for every t ∈ R+ . Then there exists the Radon-Nikodým density process η such that,
for every t ∈ R+ ,
dQ | Gt = ηt dP | Gt . (3.31)
For part (i) in Lemma 3.5.1, we refer to Blanchet-Scalliet and Jeanblanc [27] or Proposition 2.2 in
Jamshidian [92]. For part (ii), see Jeulin and Yor [101].
In this section, we will work under the standing assumption that hypothesis (H) is valid under
P.

Lemma 3.5.1 (i) Let Q be a probability measure equivalent to P on (Ω, Gt ) for every t ∈ R+ , with
the associated Radon-Nikodým density process η. If the density process η is F-adapted then we have
that, for every t ∈ R+ ,
Q(τ ≤ t | Ft ) = P(τ ≤ t | Ft ).
Hence hypothesis (H) is also valid under Q and the F-intensities of τ under Q and under P
coincide.
(ii) Assume that Q is equivalent to P on (Ω, G) and dQ = η∞ dP, so that ηt = EP (η∞ | Gt ). Then
hypothesis (H) is valid under Q whenever we have, for every t ∈ R+ ,
EP (η∞ Ht | F∞ ) EP (ηt Ht | F∞ )
= . (3.32)
EP (η∞ | F∞ ) EP (ηt | F∞ )

Proof. To prove (i), assume that the density process η is F-adapted. We have for each t ≤ s ∈ R+
EP (ηt 1{t≥τ } | Ft )
Q(τ ≤ t | Ft ) = = P(τ ≤ t | Ft )
EP (ηt | Ft )
= P(τ ≤ t | Fs ) = Q(τ ≤ t | Fs ),
where the last equality follows by another application of the Bayes formula. The assertion now
follows from part (i) in Lemma 3.2.1.
To prove part (ii), it suffices to establish the equality
Fbt := Q(τ ≤ t | Ft ) = Q(τ ≤ t | F∞ ), ∀ t ∈ R+ .

Note that since the random variables ηt 1{t≥τ } and ηt are P-integrable and Gt -measurable, using
the Bayes formula, part (v) in Lemma 3.2.1, and assumed equality (3.32), we obtain the following
chain of equalities
EP (ηt 1{t≥τ } | Ft ) EP (ηt 1{t≥τ } | F∞ )
Q(τ ≤ t | Ft ) = =
EP (ηt | Ft ) EP (ηt | F∞ )
EP (η∞ 1{t≥τ } | F∞ )
= = Q(τ ≤ t | F∞ ).
EP (η∞ | F∞ )
We conclude that hypothesis (H) holds under Q if and only if the equality (3.32) is valid. ¤

Unfortunately, a straightforward verification of condition (3.32) is rather cumbersome. For this

reason, we shall provide alternative sufficient conditions for the preservation of the hypothesis (H)
under a locally equivalent probability measure.
3.5. INVARIANCE OF HYPOTHESIS (H) 107

3.5.1 Case of the Brownian Filtration

Let W be a Brownian motion under P and let F be its natural filtration. Since we work under the
standing assumption that hypothesis (H) is satisfied under P, the process W is also a G-martingale,
where G = F ∨ H. Hence W is a Brownian motion with respect to G under P. Our goal is to show
that hypothesis (H) is still valid under Q ∈ Q for a large class Q of (locally) equivalent probability
measures. We postulate that τ admits the hazard rate γ with respect to F under P.
Let Q be an arbitrary probability measure locally equivalent to P on (Ω, G). The predictable
representation theorem implies that there exist G-predictable processes θ and κ > −1 such that the
Radon-Nikodým density η of Q with respect to P satisfies the following SDE
¡ ¢
dηt = ηt− θt dWt + κt dMt

with the initial value η0 = 1, so that η is given by (3.26). By virtue of a suitable version of the
Girsanov theorem, the following processes Wc and M c are G-martingales under Q
Z t Z t∧τ
ct = Wt −
W θu du, M ct = Mt − γu κu du.
0 0

Proposition 3.5.1 Assume that hypothesis (H) holds under P. Let Q be a probability measure
locally equivalent to P with the associated Radon-Nikodým density process η given by formula (3.26).
If the process θ is F-adapted then the hypothesis (H) is valid under Q and the F-intensity of τ
under Q equals γ bt = (1 + κet )γt , where κ
e is the unique F-predictable process such that the equality
κ
et 1{t≤τ } = κt 1{t≤τ } holds for every t ∈ R+ .

e be the probability measure locally equivalent to P on (Ω, G), given by

Proof. Let P
µZ ¶
e (2)
dP | Gt = Et κu dMu dP | Gt = ηt dP | Gt . (3.33)
]0, · ]

We claim that hypothesis (H) holds under P. e From Girsanov’s theorem, the process W follows a
e
Brownian motion under P with respect to both F and G. Moreover, from the predictable represen-
tation property of W under P,e we deduce that any F-local martingale L under P e can be written as a
stochastic integral with respect to W . Specifically, there exists an F-predictable process ξ such that
Z t
Lt = L0 + ξu dWu .
0

e Since
This shows that L is also a G-local martingale, and thus hypothesis (H) holds under P.
µZ · ¶
dQ | Gt = Et θu dWu dPe|G ,
t
0

by virtue of part (i) in Lemma 3.5.1, hypothesis (H) is valid under Q as well. The last claim in the
statement of the lemma can be deduced from the fact that the hypothesis (H) holds under Q and,
by Girsanov’s theorem, the process
Z t Z t
ct = Mt −
M 1{u<τ } γu κu du = Ht − 1{u<τ } (1 + κ
eu )γu du
0 0

is a Q-martingale. ¤

We claim that the equality P e = P holds on the filtration F. Indeed, we have dP

e | F = ηet dP | F ,
t t
(2)
where we write ηet = EP (ηt | Ft ). Furthermore, for every t ∈ R+ ,
µ µZ ¶¯ ¶
(2) ¯
EP (ηt | Ft ) = EP Et κu dMu ¯ F∞ = 1, (3.34)
]0, · ]
108 CHAPTER 3. HAZARD PROCESS APPROACH

where the first equality follows from part (v) in Lemma 3.2.1. To establish the second equality
in (3.34), we first note that since the process M is stopped at τ , we may assume, without loss
of generality, that κ = κe where the process κe is F-predictable. Moreover, the conditional cumu-
lative distribution function of τ given F∞ has the form 1 − exp(−Γt (ω)). Hence, for arbitrarily
selected sample paths of processes κ and Γ, the claimed equality can be seen as a consequence of
the martingale property of the Doléans exponential.

3.5.2 Extension to Orthogonal Martingales

Equality (3.34) suggests that Proposition 3.5.1 can be extended to the case of arbitrary orthogonal
local martingales. Such a generalization is convenient, if we wish to cover the situation considered
in Kusuoka’s counterexample. Let N be a local martingale under P with respect to the filtration F.
It is also a G-local martingale, since we maintain the assumption that hypothesis (H) holds under
P. Let Q be an arbitrary probability measure locally equivalent to P on (Ω, G). We assume that the
Radon-Nikodým density process η of Q with respect to P equals
¡ ¢
dηt = ηt− θt dNt + κt dMt

for some G-predictable processes θ and κ > −1 (the properties of the process θ depend, of course, on
the choice of a local martingale N ). The next result covers the case where N and M are orthogonal
G-local martingales under P, so that the product M N is a G-local martingale.

Proposition 3.5.2 Assume that the following conditions hold:

(a) N and M are orthogonal G-local martingales under P,
(b) N has the predictable representation property under P with respect to F, in the sense that any
F-local martingale L under P there exists an F-predictable process ξ such that, for every t ∈ R+ ,
Z
Lt = L0 + ξu dNu ,
]0,t]

(c) Pe is a probability measure on (Ω, G) such that (3.33) holds.

Then we have:
e
(i) hypothesis (H) is valid under P,
(ii) if the process θ is F-adapted then hypothesis (H) is valid under Q.

Lemma 3.5.2 Under the assumptions of Proposition 3.5.2, we have:

e
(i) N is a G-local martingale under P,
e
(ii) N has the predictable representation property for F-local martingales under P.

e | G = η (2) dP | G , where the density process η (2) is given by (3.27),

Proof. In view of (c), we have dP t t t
(2) (2)
so that dηt = ηt− κt dMt . From the assumed orthogonality of N and M , it follows that N and η (2)
are orthogonal G-local martingales under P and thus N η (2) is a G-local martingale under P as well.
This means that N is a G-local martingale under P, e so that (i) holds.
To establish part (ii) in the lemma, we first define the auxiliary process ηe by setting ηet =
(2) e | F = ηet dP | F , and thus in order to show that any F-local mar-
EP (ηt | Ft ). Then manifestly dP t t
e
tingale under P is an F-local martingale under P, it suffices to check that ηet = 1 for every t ∈ R+ ,
so that Pe = P on F. To this end, we note that, for every t ∈ R+ ,
µ µZ ¶¯ ¶
(2) ¯
EP (ηt | Ft ) = EP Et κu dMu ¯ F∞ = 1,
]0, · ]

where the first equality follows from part (v) in Lemma 3.2.1 and the second one can established
similarly as the second equality in (3.34).
3.6. G-INTENSITY OF DEFAULT TIME 109

e Then it follows also

We are in a position to prove (ii). Let L be an F-local martingale under P.
an F-local martingale under P and thus, by virtue of (b), it admits an integral representation with
respect to N under P and P.e This shows that N has the predictable representation property with
e
respect to F under P. ¤

Proof of Proposition 3.5.2. We shall argue along the similar lines as in the proof of Proposition
3.5.1. To prove (i), note that by part (ii) in Lemma 3.5.2 we know that any F-local martingale
under Pe admits the integral representation with respect to N . But, by part (i) in Lemma 3.5.2,
e We conclude that L is a G-local martingale under P
N is a G-local martingale under P. e and thus
e
hypothesis (H) is valid under P. Assertion (ii) now follows from part (i) in Lemma 3.5.1. ¤

Example 3.5.1 Kusuoka [106] presents a counter-example based on the two independent random
R t∧τ
times τ1 and τ2 given on some probability space (Ω, G, P). We write Mti = Hti − 0 i γi (u) du, where
Hti = 1{t≥τi } and γi is the deterministic intensity function of τi under P. Let us set dQ | Gt = ηt dP | Gt ,
(1) (2)
where ηt = ηt ηt and, for i = 1, 2 and every t ∈ R+ ,
Z t µZ ¶
(i) (i)
ηt = 1 + ηu− κ(i)
u dM i
u = Et κ (i)
u dM i
u
0 ]0, · ]

for some G-predictable processes κ(i) , i = 1, 2, where G = H1 ∨ H2 . We set F = H1 and H = H2 .

Manifestly, hypothesis (H) holds under P.
Moreover, in view of Proposition 3.5.2, it is still valid under the equivalent probability measure
P e | G = η (2) dP | G . It is clear that P
e given by dP e = P on F, since we have that, for every t ∈ R+ ,
t t t

µ µZ ¶¯ ¶
(2) 2 ¯ 1
EP (ηt | Ft ) = EP Et κ(2)
u dM u ¯ Ht = 1.
]0, · ]

However, hypothesis (H) is not necessarily valid under Q if the process κ(1) fails to be F-adapted. In
Kusuoka’s counter-example, the process κ(1) was chosen to be explicitly dependent on both random
times and it was shown that hypothesis (H) fails to hold under Q.
For an alternative approach to Kusuoka’s example, through an absolutely continuous change of
a probability measure, the interested reader may consult Collin-Dufresne et al. [51].

3.6 G-Intensity of Default Time

In an alternative approach to modeling of default time, we start by assuming that we are given a
default time τ and some filtration G such that τ is a G-stopping time. In this setup, the default
intensity is defined as follows.

Definition 3.6.1 A G-intensity of default time τ is any non-negative and G-predictable process
(λt , t ∈ R+ ) such that the process (Mt , t ∈ R+ ), which is given as
Z t∧τ
Mt = Ht − λu du,
0

is a G-martingale.

The existence of a G-intensity of τ hinges on the fact that H is a bounded increasing process,
therefore a bounded sub-martingale, and thus, by the Doob-Meyer decomposition, it can be written
as a sum of a martingale M and a G-predictable, increasing process A, which is stopped at τ . In
the case where τ is a predictable stopping time, obviously A = H. In fact, it is known that the
G-intensity exists only if τ is a totally inaccessible stopping time with respect to G. In the present
110 CHAPTER 3. HAZARD PROCESS APPROACH

setup, the default intensity is not well defined after time τ . Specifically, if λ is a G-intensity then
for any non-negative, G-predictable process g the process λ,e given by the expression

et = λt 1{t≤τ } + gt 1{t>τ } ,
λ
Rt
is also a version of a G-intensity. Let us write Λt = 0
λu du. The following result is a counterpart
of Lemma 3.1.3(i).

Lemma 3.6.1 The process Lt = 1{t<τ } eΛt for t ∈ R+ is a G-martingale.

Proof. From the integration by parts formula, we get

¡ ¢
dLt = eΛt (1 − Ht )λt dt − dHt = −eΛt dMt .

This shows that L is a G-martingale. ¤

The following result is due to Duffie et al. [65].

Proposition 3.6.1 For any GT -measurable and Q-integrable random variable X we have

EQ (X1{T <τ } | Gt ) = 1{t<τ } eΛt EQ (Xe−ΛT | Gt ) − EQ (1{t<τ ≤T } ∆Yτ eΛτ | Gt ),

where the process Y is defined by setting, for every t ∈ R+ ,

¡ ¯ ¢
Yt = EQ Xe−ΛT ¯ Gt .

Proof. Let us denote U = LY . The Itô integration by parts formula yields

dUt = Lt− dYt + Yt− dLt + d[L, Y ]t = Lt− dYt + Yt− dLt + ∆Lt ∆Yt .

Since L and Y are G-martingales, we obtain

¡ ¢ ¡ ¢
EQ (UT | Gt ) = EQ X1{T <τ } | Gt = Ut − EQ 1{t<τ ≤T } ∆Yτ eΛτ | Gt .

Consequently,
¡ ¢ ¡ ¢
EQ (X1{T <τ } | Gt ) = 1{t<τ } eΛt EQ Xe−ΛT | Gt − EQ 1{t<τ ≤T } eΛτ ∆Yτ | Gt

as required. ¤

It is worthwhile to compare the next result with the formula established in Corollary 3.1.1.
¡ ¯ ¢
Corollary 3.6.1 Assume that the process Yt = EQ Xe−ΛT ¯ Gt is continuous at time τ , that is,
∆Yτ = 0. Then for any GT -measurable, Q-integrable random variable X
¡ ¯ ¢
EQ (X1{T <τ } | Gt ) = 1{t<τ } EQ XeΛt −ΛT ¯ Gt .

It should be stressed that the continuity of the process Y at time τ depends on the choice of λ
after time τ and that this condition is rather difficult to verify, in general. Furthermore, the jump
size ∆Yτ is usually quite hard to compute explicitly. It is thus worth noting that Collin-Dufresne
et al. [51] apply an absolutely continuous change of a probability measure that leads to an essential
simplification of the formula of Proposition 3.6.1. In a recent paper by Jeanblanc and Le Cam [98],
the authors provide a detailed comparison of the two alternative approaches to the modeling of
default time.
3.7. SINGLE NAME CDS MARKET 111

3.7 Single Name CDS Market

A strictly positive random variable τ defined on a probability space (Ω, G, Q) is termed a random
time. In view of its financial interpretation, we will refer to it as a default time. We define the
default indicator process Ht = 1{t≥τ } and we denote by H the filtration generated by this process.
We assume that we are given, in addition, some auxiliary filtration F and we write G = H ∨ F,
meaning that we have Gt = σ(Ht , Ft ) for every t ∈ R+ . The filtration G is referred to as to the full
filtration. It is clear that τ is an H-stopping time, as well as a G-stopping time (but not necessarily
an F-stopping time).
All processes are defined on the space (Ω, G, Q), where Q is to be interpreted as the real-life (i.e.,
statistical) probability measure. Unless otherwise stated, all processes considered in what follows
are assumed to be G-adapted and with càdlàg sample paths.

3.7.1 Standing Assumptions

We assume that the underlying market model is arbitrage-free, meaning that it admits a spot mar-
tingale measure Q (not necessarily unique) equivalent to Q. A spot martingale measure is associated
with the choice of the savings account B as a numéraire, in the sense that the price process of any
traded security, which pays no coupons or dividends, is a G-martingale under Q when discounted
by the savings account B. As usual, B is given by
³Z t ´
Bt = exp ru du , ∀ t ∈ R+ ,
0

where the short-term r is assumed to follow an F-progressively measurable stochastic process. The
choice of a suitable term structure model is arbitrary and it is not discussed in the present work.
Recall that Gt = Q(τ > t | Ft ) is the survival process of τ with respect to a filtration F. We
postulate that G0 = 1 and Gt > 0 for every t ∈ R+ (hence the case where τ is an F-stopping time is
excluded) so that the hazard process Γ = − ln G of τ with respect to the filtration F is well defined.
Clearly, the process G is a bounded F-supermartingale and thus it admits the unique Doob-
Meyer decomposition G = µ − ν, where µ is an F-martingale with µ0 = 1 and ν is an F-predictable,
increasing process. If F = N + C is the Doob-Meyer decomposition of F then, of course, µ = 1 − N
and C = ν. We shall work throughout under the following standing assumption.

Assumption 3.7.1 We postulate that G is a continuous process and the increasing process C in
its Doob-Meyer decomposition is absolutely continuous with respect to the Lebesgue measure, so
that dCt = ct dt for some F-progressively measurable, non-negative process c.

We denote by λ the F-progressively measurable process defined as λt = G−1 t ct . Let us note for
a further reference that, under Assumption 3.7.1, we have that dGt = dµt − λt Gt dt, where the
F-martingale µ is continuous. Moreover, in view of the Lebesgue dominated convergence theorem,
the continuity of G implies that the expected value EQ (Gt ) = Q(τ > t) is a continuous function,
and thus Q(τ = t) = 0 for any fixed t ∈ R+ . Finally, we already know that under Assumption 3.7.1
the process M , given by
Z t∧τ
Mt = Ht − Λt∧τ = Ht − λu du, (3.35)
0

is a G-martingale, where the increasing, absolutely continuous, F-adapted process Λ is given by

Z t Z t
Λt = G−1
u dCu = λu du.
0 0

Recall that the F-progressively measurable process λ is called the F-intensity of default.
112 CHAPTER 3. HAZARD PROCESS APPROACH

3.7.2 Valuation of a Defaultable Claim

Let us first recall the concept of a generic defaultable claim (cf. Section 1.1.1 and Definition 2.3.1).
In this section, we work within a single-name framework, so that τ is the moment of default of a
reference credit name. A generic defaultable claim is now specified by the following extension of
e = 0).
Definition 2.3.1 (note that, similarly as in Definition 2.3.1, we set X

Definition 3.7.1 By a defaultable claim with maturity date T we mean a quadruplet (X, A, Z, τ )
where X is an FT -measurable random variable, (At , t ∈ [0, T ]) is an F-adapted, continuous process
of finite variation with A0 = 0, (Zt , t ∈ [0, T ]) is an F-predictable process and τ is a random time.

As usual, the financial interpretation of components of a defaultable claim can be inferred from
the specification of the dividend process D describing all cash flows associated with a defaultable
claim over its lifespan ]0, T ], that is, excluding the initial premium, if any. We follow here our
standard convention that the date 0 is the inception date of a defaultable contract.

Definition 3.7.2 The dividend process (Dt , t ∈ R+ ) of a defaultable claim (X, A, Z, τ ) maturing at
T equals, for every t ∈ R+ ,
Z t∧T Z
Dt = X1{T <τ } 1[T,∞[ (t) + (1 − Hu ) dAu + Zu dHu .
0 ]0,t∧T ]

It is clear that the dividend process D is an F-adapted process of finite variation on [0, T ].
Let us recall the financial interpretation of D is as follows: X is the promised payoff, the process
A represents the promised dividends and the process Z, termed the recovery process, specifies the
recovery payoff at default. As already mentioned above, according to our convention, a possible cash
payment (premium) at time 0 is not included in the dividend process D associated with a defaultable
claim.

Price Dynamics of a Defaultable Claim

For any fixed t ∈ [0, T ], the process Du − Dt , u ∈ [t, T ], represents all cash flows from a defaultable
claim received by an investor who purchased it at time t. In general, the process Du − Dt may
depend on the past prices of underlying assets and on the history of the market prior to t. The past
dividends are not valued by the market, however, so that the current market value at time t ∈ [0, T ]
of a defaultable claim – that is, the price at which it is traded at time t – will only reflect future
cash flows over the time interval ]t, T ]. This leads to the following definition of the ex-dividend price
of a defaultable claim (cf. formula (3.6))

Definition 3.7.3 The ex-dividend price process S of a defaultable claim (X, A, Z, τ ) equals, for
every t ∈ [0, T ],
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Gt . (3.36)
]t,T ]

Obviously, ST = 0 for any dividend process D. We work throughout under the natural integra-
bility assumptions: EQ |BT−1 X| < ∞,
¯Z T ¯
¯ ¯
EQ ¯ Bu−1 (1 − Hu ) dAu ¯ < ∞
0

and
EQ |Bτ−1
∧T Zτ ∧T | < ∞,
3.7. SINGLE NAME CDS MARKET 113

which ensure that the ex-dividend price St is well defined for any t ∈ [0, T ]. We will later need the
following technical assumption
³Z T ´
EQ (Bu−1 Zu )2 dhµiu < ∞. (3.37)
0

We first derive a convenient representation for the ex-dividend price S of a defaultable claim.

Proposition 3.7.1 The ex-dividend price of a defaultable claim (X, A, Z, τ ) equals, for t ∈ [0, T [,
Z
Bt ³ T ¡ ¢ ¯¯ ´
St = 1{t<τ } EQ BT−1 GT X + Bu−1 Gu Zu λu du + dAu ¯ Ft .
Gt t

Proof. For any t ∈ [0, T [, the ex-dividend price is given by the conditional expectation
³ Z T ∧τ ¯ ´
¯
St = Bt EQ BT−1 X1{T <τ } + Bu−1 dAu + Bτ−1 Zτ 1{t<τ ≤T } ¯ Gt .
t∧τ

Let us fix t and let us introduce two auxiliary processes Y = (Yu )u∈[t,T ] and R = (Ru )u∈[t,T ] by
setting Z u Z u
−1 −1
Yu = Bv dAv , Ru = Bu Zu + Bv−1 dAv = Bu−1 Zu + Yu .
t t
Then St can be represented as follows
³ ¯ ´
¯
St = Bt EQ BT−1 X1{T <τ } + 1{T <τ } YT + Rτ 1{t<τ ≤T } ¯ Gt .

We use the formula of Corollary 3.1.1, to evaluate the conditional expectations

³ ¯ ´ Bt ³ ¯ ´
¯ ¯
Bt EQ 1{T <τ } BT−1 X ¯ Gt = 1{t<τ } EQ BT−1 GT X ¯ Ft ,
Gt
and ³ ¯ ´ ³ ¯ ´
¯ Bt ¯
Bt EQ 1{T <τ } YT ¯ Gt = 1{t<τ } EQ GT YT ¯ Ft .
Gt
In addition, we will use of the following formula
³ Z T ¯ ´
1 ¯
EQ (1{t<τ ≤T } Rτ | Gt ) = −1{t<τ } EQ Ru dGu ¯ Ft ,
Gt t

which is known to hold for any F-predictable process R such that EQ |Rτ | < ∞. We thus obtain, for
any t ∈ [0, T [,
³ Z T ¯ ´
Bt ¯
St = 1{t<τ } EQ BT−1 GT X + GT YT − (Bu−1 Zu + Yu ) dGu ¯ Ft ,
Gt t

Moreover, since dGt = dµt − λt Gt dt, where µ is an F-martingale, we also obtain

³ Z T ¯ ´ ³Z T ¯ ´
¯ ¯
EQ − Bu−1 Zu dGu ¯ Ft = EQ Bu−1 Gu Zu λu du ¯ Ft ,
t t

where we have used the assumed inequality (3.37).

To complete the proof, it remains to observe that G is a continuous semimartingale and Y is a
continuous process of finite variation with Yt = 0, so that the Itô integration by parts formula yields
Z T Z T Z T
GT YT − Yu dGu = Gu dYu = Bu−1 Gu dAu ,
t t t
114 CHAPTER 3. HAZARD PROCESS APPROACH

where the second equality follows from the definition of Y . We conclude that the asserted formula
holds for any t ∈ [0, T [, as required. ¤

Proposition 3.7.1 implies that the ex-dividend price S satisfies, for every t ∈ [0, T ],

St = 1{t<τ } Set

e which is termed the ex-dividend pre-default price of a defaultable

for some F-adapted process S,
claim. Note that S may not be continuous at time T , in which case ST − 6= ST = 0.

Definition 3.7.4 The cumulative price process S c associated with the dividend process D is defined
by setting, for every t ∈ [0, T ],
³Z ¯ ´ Z
c −1 ¯
St = Bt EQ Bu dDu ¯ Gt = St + Bt Bu−1 dDu . (3.38)
]0,T ] ]0,t]

Note that the discounted cumulative price process S c∗ = B −1 S c follows a G-martingale under
Q. We deduce immediately from Proposition 3.7.1 and Definition 3.7.4 that the following corollary
is valid.

Corollary 3.7.1 The cumulative price of a defaultable claim (X, A, Z, τ ) equals, for t ∈ [0, T ],
Z T
Bt ³ ¡ ¢ ¯¯ ´
Stc = 1{t<τ } EQ BT−1 GT X1{t<T } + Bu−1 Gu Zu λu du + dAu ¯ Ft
G
Z t t

+ Bt Bu−1 dDu .
]0,t]

The pre-default cumulative price is the unique F-adapted process Sec that satisfies, for every
t ∈ [0, T ],
1{t<τ } Stc = 1{t<τ } Setc . (3.39)

Our next goal is to derive the dynamics under Q for the (pre-default) price of a defaultable claim
in terms of some G-martingales and F-martingales. To simplify the presentation, we shall work from
now on under the following standing assumption.

Assumption 3.7.2 Any F-martingale is a continuous process.

The following auxiliary result is well known (see, for instance, Lemma 5.1.6 in [20]). Recall that
µ is the F-martingale appearing in the Doob-Meyer decomposition of G.

Lemma 3.7.1 Let n be any F-martingale. Then the process ň given by

Z t∧τ
ňt = nt∧τ − G−1
u dhn, µiu (3.40)
0

is a continuous G-martingale.

In particular, the process µ̌ given by

Z t∧τ
µ̌t = µt∧τ − G−1
u dhµ, µiu (3.41)
0

is a continuous G-martingale.
In the next result, we deal with the dynamics of the ex-dividend price process S. Recall that the
G-martingale M is given by formula (3.35).
3.7. SINGLE NAME CDS MARKET 115

Proposition 3.7.2 The dynamics of the ex-dividend price S on [0, T ] are

¡ ¢
dSt = −St− dMt + (1 − Ht ) (rt St − λt Zt ) dt − dAt (3.42)
¡ ¢ ¡ ¢
+ (1 − Ht )G−1
t Bt dmt − St dµt + (1 − Ht )G−2
t St dhµit − Bt dhµ, mit
where the continuous F-martingale m is given by the formula
Z T
³
−1
¡ ¢ ¯¯ ´
mt = EQ BT GT X + Bu−1 Gu Zu λu du + dAu ¯ Ft . (3.43)
0

e In view of Propo-
Proof. We shall first derive the dynamics of the pre-default ex-dividend price S.
sition 3.7.1, the price S can be represented as follows, for t ∈ [0, T [,

St = 1{t<τ } Set = 1{t<τ } Bt G−1

t Ut ,

where the auxiliary process U equals

Z t Z t
Ut = mt − Bu−1 Gu Zu λu du − Bu−1 Gu dAu ,
0 0

where in turn the continuous F-martingale m is given by (3.43). It is thus obvious that Se = BG−1 U
for t ∈ [0, T [ (of course, SeT = 0). Since G = µ − C, an application of Itô’s formula leads to
d(G−1 −1 −1 −1
t Ut ) = Gt dmt − Bt Zt λt dt − Bt dAt
³ ´
+ Ut G−3 −2 −2
t dhµit − Gt (dµt − dCt ) − Gt dhµ, mit .

Therefore, since under the present assumptions dCt = λt Gt dt, using again Itô’s formula, we obtain
¡ ¢ ¡ ¢
dSet = (λt + rt )Set − λt Zt dt − dAt + G−1
t Bt dmt − Set dµt (3.44)
¡
−2 e
¢
+ Gt St dhµit − Bt dhµ, mit .

Note that, under the present assumptions, the pre-default ex-dividend price Se follows on [0, T [ a
continuous process with dynamics given by (3.44). This means that St− = Set on {t ≤ τ } for any
t ∈ [0, T [. Moreover, since G is continuous, we clearly have that Q(τ = T ) = 0. Hence for the
process St = (1 − Ht )Set we obtain, for every t ∈ [0, T ],
¡ ¢
dSt = −St− dMt + (1 − Ht ) (rt St − λt Zt ) dt − dAt
¡ ¢ ¡ ¢
+ (1 − Ht )G−1
t Bt dmt − St dµt + (1 − Ht )G−2
t St dhµit − Bt dhµ, mit
as expected. ¤

Let us now examine the dynamics of the cumulative price. As expected, the discounted cumula-
tive price S c∗ = B −1 S c is a G-martingale under Q (see formula (3.46) below).

Corollary 3.7.2 The dynamics of the cumulative price S c on [0, T ] are

¡ ¢
dStc = rt Stc dt + (Zt − St− ) dMt + (1 − Ht )G−1 t Bt dmt − St dµt (3.45)
¡ ¢
+ (1 − Ht )G−2 t St dhµit − Bt dhµ, mit
with the F-martingale m given by (3.43). Equivalently,
dStc = rt Stc dt + (Zt − St− ) dMt + G−1
t (Bt dm̌t − St dµ̌t ), (3.46)
where the G-martingales m̌ and µ̌ are given by (3.40) and (3.41) respectively. The pre-default
cumulative price Sec satisfies, for t ∈ [0, T ],
¡ ¢
dSetc = rt Setc dt + λt (Set − Zt ) dt + G−1
t Bt dmt − Set dµt (3.47)
−2 e
¡ ¢
+ Gt St dhµit − Bt dhµ, mit .
116 CHAPTER 3. HAZARD PROCESS APPROACH

Proof. Formula (3.38) yields

³ Z ´
dStc = dSt + d Bt Bu−1 dDu = dSt + rt (Stc − St ) dt + dDt
]0,t]
= dSt + rt (Stc − St ) dt + (1 − Ht ) dAt + Zt dHt . (3.48)

By combining (3.48) with (3.42), we obtain (3.45). Formulae (3.46) and (3.47) are immediate
consequences of (3.40), (3.41) and (3.45). ¤

Dynamics under hypothesis (H). Let us now consider the special case where hypothesis (H) is
satisfied under Q between the filtrations F and G = H ∨ F. This means that the immersion property
holds for the filtrations F and G, in the sense that any F-martingale under Q is also a G-martingale
under Q. In that case, the survival process G of τ with respect to F is known to be non-increasing,
so that G = −C. In other words, the continuous martingale µ in the Doob-Meyer decomposition of
G vanishes. Consequently, formula (3.42) becomes
¡ ¢
dSt = −St− dMt + (1 − Ht ) (rt St − λt Zt ) dt − dAt + (1 − Ht )Bt G−1
t dmt .

Similarly, (3.45) reduces to

dStc = rt Stc dt + (Zt − Set ) dMt + (1 − Ht )G−1

t Bt dmt

and (3.47) becomes

dSetc = rt Setc dt + λt (Set − Zt ) dt + G−1
t Bt dmt .

Remark 3.7.1 Hypothesis (H) is a rather natural assumption in the present context. Indeed, it
can be shown that it is necessarily satisfied under the postulate that the underlying F-market model
is complete and arbitrage-free, and the extended G-market model is arbitrage-free (see Blanchet-
Scalliet and Jeanblanc [27]).

Price Dynamics of a CDS

In Definition 3.7.5 of a stylized T -maturity credit default swap, we follow the convention adopted
in Section 2.4. Unlike in Section 2.4, the default protection stream is now represented by an F-
predictable process δ. We assume that the default protection payment is received at the time of
default and it equals δt if default occurs at time t prior to or at maturity date T . Note that δt
represents the protection payment, so that according to our notational convention the recovery rate
equals 1 − δt rather than δt . The notional amount of the CDS is equal to one monetary unit.

Definition 3.7.5 The stylized T -maturity credit default swap (CDS) with a constant spread κ and
protection at default is a defaultable claim (0, A, Z, τ ) in which we set Zt = δt and At = −κt for
every t ∈ [0, T ]. An F-predictable process δ : [0, T ] → R represents the default protection and a
constant κ is the fixed CDS spread (also termed the rate or premium of the CDS).

A credit default swap is thus a particular defaultable claim in which the promised payoff X is null
and the recovery process Z is determined in reference to the estimated recovery rate of the reference
credit name. We shall use the notation D(κ, δ, T, τ ) to denote the dividend process of a CDS. It
follows immediately from Definition 3.7.2 that the dividend process D(κ, δ, T, τ ) of a stylized CDS
equals, for every t ∈ R+ ,
Dt (κ, δ, T, τ ) = δτ 1{t≥τ } − κ(t ∧ T ∧ τ ). (3.49)
In a more realistic approach, the process A is discontinuous, with jumps occurring at the premium
payment dates. In this section, we shall only deal with a stylized CDS with a continuously paid
premium.
3.7. SINGLE NAME CDS MARKET 117

Let us first examine the valuation formula for a stylized T -maturity CDS. Since we now have
X = 0, Z = δ and At = −κt, we deduce easily from (3.36) that the ex-dividend price of such CDS
contract equals, for every t ∈ [0, T ],
¡ ¢
St (κ, δ, T, τ ) = 1{t<τ } Pe(t, T ) − κA(t,
e T) , (3.50)

where we denote, for any t ∈ [0, T ],

Bt ³ ¯ ´
¯
Pe(t, T ) = EQ 1{t<τ ≤T } Bτ−1 δτ ¯ Ft
Gt

and Z
³ T ∧τ ¯ ´
e T ) = Bt EQ
A(t,
¯
Bu−1 du ¯ Ft .
Gt t

The quantity Pe(t, T ) is the pre-default value at time t of the protection leg, whereas A(t,
e T ) represents
the pre-default present value at time t of one risky basis point paid up to the maturity T or the default
time τ , whichever comes first. For ease of notation, we shall write St (κ) in place of St (κ, δ, T, τ ) in
what follows. Note that the quantities Pe(t, T ) and A(t, e T ) are well defined at any date t ∈ [0, T ],
and not only prior to default as the terminology ‘pre-default values’ might suggest.
We are in a position to state the following immediate corollary to Proposition 3.7.1.

Corollary 3.7.3 The ex-dividend price of a CDS equals, for every t ∈ [0, T ],
³ Z T ¯ ´
Bt ¯
St (κ) = 1{t<τ } EQ Bu−1 Gu (δu λu − κ) du ¯ Ft (3.51)
Gt t

and thus the cumulative price of a CDS equals, for every t ∈ [0, T ],
³ Z T ¯ ´ Z
Bt ¯
Stc (κ) = 1{t<τ } EQ Bu−1 Gu (δu λu − κ) du ¯ Ft + Bt Bu−1 dDu .
Gt t ]0,t]

The next result is a direct consequence of Proposition 3.7.2 and Corollary 3.7.2.

Corollary 3.7.4 The dynamics of the ex-dividend price S(κ) are

dSt (κ) = −St− (κ) dMt + (1 − Ht )(rt St + κ − λt δt ) dt (3.52)

¡ ¢
+ (1 − Ht )G−1
t (Bt dnt − St dµt ) + (1 − Ht )Gt
−2
St dhµit − Bt dhµ, nit

with the F-martingale n given by the formula

³Z T ¯ ´
¯
nt = EQ Bu−1 Gu (δu λu − κ) du ¯ Ft . (3.53)
0

The cumulative price S c (κ) satisfies, for every t ∈ [0, T ],

¡ ¢ ¡ ¢
dStc (κ) = rt Stc (κ) dt + δt − St− (κ) dMt + (1 − Ht )G−1
t Bt dnt − St (κ) dµt
¡ ¢
+ (1 − Ht )G−2 t St (κ) dhµit − Bt dhµ, nit

or, equivalently,
¡ ¢ ¡ ¢
dStc (κ) = rt Stc (κ) dt + δt − St− (κ) dMt + G−1
t Bt dňt − St (κ) dµ̌t , (3.54)

where the G-martingales ň and µ̌ are given by (3.40) and (3.41) respectively.
118 CHAPTER 3. HAZARD PROCESS APPROACH

Dynamics under hypothesis (H). If the immersion property of F and G holds, the martingale
µ is null and thus (3.52) reduces to
¡ ¢
dSt (κ) = −Set (κ) dMt + (1 − Ht ) rt St (κ) + κ − λt δt dt + (1 − Ht )Bt G−1
t dnt

since the process Set (κ), t ∈ [0, T ], is continuous and satisfies (cf. (3.44))
¡ ¢
dSet (κ) = (λt + rt )Set (κ) + κ − λt δt dt + Bt G−1
t dnt .

Let us note that the quantity κ − λt δt has the intuitive interpretation as the pre-default dividend
rate of a CDS.
Similarly, we obtain from (3.54)
¡ ¢
dStc (κ) = rt Stc (κ) dt + δt − Set (κ) dMt + (1 − Ht )Bt G−1
t dnt (3.55)

and
¡ ¢
dSetc (κ) = rt Setc (κ) dt + λt Set (κ) − δt dt + Bt G−1
t dnt .

Dynamics of the Market CDS Spread

Let us now introduce the notion of the market CDS spread. It reflects the real-world feature that
for any date s the CDS issued at this time has the fixed spread chosen in such a way that the CDS
is worthless at its inception. Note that the protection process (δt , t ∈ [0, T ]) is fixed throughout.
We fix the maturity date T and we assume that credit default swaps with different inception dates
have a common protection process δ.

Definition 3.7.6 The T -maturity market CDS spread κ(s, T ) at any date s ∈ [0, T ] is the level of
the CDS spread that makes the values of the two legs of a CDS equal to each other at time s.

It should be noted that CDSs are quoted in terms of spreads. At any date t, one can take at no
cost a long or short position in the CDS issued at this date with the fixed spread equal to the actual
value of the market CDS spread for a given maturity and a given reference credit name.
Let us stress that the market CDS spread κ(s, T ) is not defined neither at the moment of default
nor after this date, so that we shall deal in fact with the pre-default value of the market CDS
spread. Observe that κ(s, T ) is represented by an Fs -measurable random variable. In fact, it follows
immediately from (3.51) that κ(s, T ) admits the following representation, for every s ∈ [0, T ],
³R ¯ ´
T −1 ¯ Fs
e
P (s, T ) EQ s
B u G δ λ
u u u du K1
κ(s, T ) = = ³R ¯ ´ = s2 ,
e T)
A(s, EQ
T
Bu−1 Gu du ¯ Fs Ks
s

where we denote
³Z T ¯ ´
Ks1 = EQ Bu−1 Gu δu λu du ¯ Fs
s

and
³Z T ¯ ´
Ks2 = EQ Bu−1 Gu du ¯ Fs .
s

In what follows, we shall write briefly κs instead of κ(s, T ). The next result furnishes a convenient
representation for the price at time t of a CDS issued at some date s ≤ t, that is, the marked-to-
market value of a CDS that exists already for some time (recall that the market value of the just
issued CDS is null).
3.7. SINGLE NAME CDS MARKET 119

Proposition 3.7.3 The ex-dividend price S(κs ) of a T -maturity market CDS initiated at time s
equals, for every t ∈ [s, T ],
e T ) = 1{t<τ } Set (κs ),
St (κs ) = 1{t<τ } (κt − κs ) A(t, (3.56)

where Set (κs ) is the pre-default ex-dividend price at time t.

Proof. To establish (3.56), it suffices to observe that St (κs ) = St (κs ) − St (κt ) since St (κt ) = 0.
Therefore, in order to conclude it suffices to use (3.50) with κ = κt and κ = κs . ¤

Let us now derive the dynamics of the market CDS spread. We define the F-martingales
³Z T ¯ ´ Z s
1 −1 ¯ 1
ms = EQ Bu Gu δu λu du ¯ Fs = Ks + Bu−1 Gu δu λu du
0 0

and
³Z T ¯ ´ Z s
¯
m2s = EQ Bu−1 Gu 2
du ¯ Fs = Ks + Bu−1 Gu du.
0 0
Under Assumption 3.7.2, the F-martingales m1 and m2 are continuous. Therefore, using the Itô
formula, we find easily that the semimartingale decomposition of the market spread process reads
1 ³ κs 1 ´
dκs = 2 Bs−1 Gs (κs − δs λs ) ds + 2 dhm2 is − 2 dhm1 , m2 is
Ks Ks Ks
1 ¡ ¢
+ 2 dm1s − κs dm2s .
Ks

3.7.3 Replication of a Defaultable Claim

We now assume that k credit default swaps with certain maturities Ti ≥ T , spreads κi and protection
payments δ i for i = 1, 2, . . . , k are traded over the time interval [0, T ]. All these contracts are
supposed to refer to the same underlying credit name and thus they have a common default time τ .
Formally, this family of CDSs is represented by the associated dividend processes Di = D(κi , δ i , Ti , τ )
given by formula (3.49). For brevity, the corresponding ex-dividend price will be denoted as S i (κi )
rather than S(κi , δ i , Ti , τ ). Similarly, S c,i (κi ) will stand for the cumulative price process of the ith
traded CDS. The 0th traded asset is the savings account B.

Trading Strategies in the CDS Market

Our goal is to examine hedging strategies for a defaultable claim (X, A, Z, τ ). As expected, we
will trade in k credit default swaps and the savings account. To this end, we will consider trading
strategies φ = (φ0 , . . . , φk ) where φ0 is a G-adapted process and the processes φ1 , . . . , φk are G-
predictable.
In the present setup, we consider trading strategies that are self-financing in the standard sense,
as recalled in the following definition.

Definition 3.7.7 The wealth process V (φ) of a strategy φ = (φ0 , . . . , φk ) in the savings account B
and ex-dividend CDS prices S i (κi ), i = 1, 2, . . . , k equals, for any t ∈ [0, T ],
k
X
Vt (φ) = φ0t Bt + φit Sti (κi ).
i=1

A strategy φ is said to be self-financing if Vt (φ) = V0 (φ) + Gt (φ) for every t ∈ [0, T ], where the gains
process G(φ) is defined as follows
Z Xk Z
Gt (φ) = φ0u dBu + φiu d(Sui (κi ) + Dui ),
]0,t] i=1 ]0,t]
120 CHAPTER 3. HAZARD PROCESS APPROACH

where Di = D(κi , δ i , Ti , τ ) is the dividend process of the ith CDS (see formula (3.49)).

The following lemma is fairly general; in particular, it is independent of the choice of the under-
lying model. Indeed, in the proof of this result we only use the obvious relationships dBt = rt Bt dt
and the relationship (cf. (3.38))
Z
c,i i
St (κi ) = St (κi ) + Bt Bu−1 dDui . (3.57)
]0,t]

Let V ∗ (φ) = B −1 V (φ) stand for the discounted wealth process and let S c,i,∗ (κi ) = B −1 S c,i (κi ) be
the discounted cumulative price.

Lemma 3.7.2 Let φ = (φ0 , . . . , φk ) be a self-financing trading strategy in the savings account B
and ex-dividend prices S i (κi ), i = 1, 2, . . . , k. Then the discounted wealth process V ∗ = B −1 V (φ)
satisfies, for t ∈ [0, T ]
Xk
dVt∗ (φ) = φit dStc,i,∗ (κi ). (3.58)
i=1

Proof. We have
¡ ¢
dVt∗ (φ) = Bt−1 dVt (φ) − rt Bt−1 Vt (φ) dt = Bt−1 dVt (φ) − rt Vt (φ) dt
h Xk
¡ ¢ i
= Bt−1 φ0t rt Bt dt + φit dSti (κi ) + dDti − rt Vt (φ) dt
i=1
h³ k
X ´ k
X ¡ ¢i
= Bt−1 Vt (φ) − φit Sti (κi ) rt dt + φit dSti (κi ) + dDti − rt Bt−1 Vt (φ) dt
i=1 i=1
k
X ³ ´
= Bt−1 φit dSti (κi ) − rt Sti (κi ) dt + dDti
i=1
k
X ³ ´
= φit d(Bt−1 Sti (κi )) + Bt−1 dDti .
i=1

By comparing the last formula with (3.57), we see that (3.58) holds. ¤

Replication with Ex-Dividend Prices of CDSs

Recall that the cumulative price of a defaultable claim (X, A, Z, τ ) is denoted as S c . We adopt the
following, quite natural, definition of replication of a defaultable claim. Note that the set of traded
assets is not explicitly specified in this definition. Hence this definition can be used for any choice
of primary traded assets.

Definition 3.7.8 We say that a self-financing strategy φ = (φ0 , . . . , φk ) replicates a defaultable

claim (X, A, Z, τ ) if its wealth process V (φ) satisfies Vt (φ) = Stc for every t ∈ [0, T ]. In particular,
c
the equality Vt∧τ (φ) = St∧τ holds for every t ∈ [0, T ].

In the remaining part of this section we assume that hypothesis (H) holds. Hence the hazard
process Γ of default time is increasing and thus, by Assumption 3.7.1, we have that, for any t ∈ [0, T ],
Z t
Γt = Λt = λu du.
0
c,i,∗
The discounted cumulative price S (κi ) of the ith CDS is governed by (cf. (3.55))
¡ ¢
dStc,i,∗ (κi ) = Bt−1 δti − Seti (κi ) dMt + (1 − Ht )G−1 i
t dnt , (3.59)
3.7. SINGLE NAME CDS MARKET 121

where (cf. (3.53))

³Z Ti ¯ ´
¯
nit = EQ Bu−1 Gu (δui λu − κi ) du ¯ Ft . (3.60)
0

The next lemma yields the dynamics of the wealth process V (φ) for a self-financing strategy φ. To
prove (3.61), it suffices to combine (3.58) with (3.59).

Lemma 3.7.3 The discounted wealth process V ∗ (φ) = B −1 V (φ) of any self-financing trading strat-
egy φ satisfies, for any t ∈ [0, T ],
k
X ³ ¡ ¢ ´
dVt∗ (φ) = φit Bt−1 δti − Seti (κi ) dMt + (1 − Ht )G−1 i
t dnt . (3.61)
i=1

It is clear from the lemma that it is enough to search for the components φ1 , . . . , φk of a strategy
φ. The same remark applies to self-financing strategies introduced in Definition 3.7.7. It is worth
stressing that in what follows, we shall only consider admissible trading strategies, that is, strategies
for which the discounted wealth process V ∗ (φ) = B −1 V (φ) is a G-martingale under Q. The market
model in which only admissible trading strategies are allowed is arbitrage-free, that is, arbitrage
opportunities are ruled out. Admissibility of a replicating strategy will be ensured by the equality
V (φ) = S c and the fact that the discounted cumulative price S c∗ = B −1 S c of a defaultable claim is
a G-martingale under Q.
We work throughout under the standing Assumptions 3.7.1 and 3.7.2 and the following postulate.

Assumption 3.7.3 The filtration F is generated by a d-dimensional Brownian motion W under Q.

Since hypothesis (H) is assumed to hold, the process W is also a Brownian motion with respect
to the enlarged filtration G = H ∨ F. Recall that any (local) martingales with respect to a Brownian
filtration is necessarily continuous. Hence Assumption 3.7.2 is obviously satisfied under Assumption
3.7.3.
The crucial observation is that, by the predictable representation property of a Brownian motion,
there exist F-predictable, Rd -valued processes ξ and ζ i , i = 1, 2, . . . , k such that dmt = ξt dWt and
dnit = ζti dWt , where m and ni are given by (3.43) and (3.60), respectively.
The crucial observation is that, by the predictable representation property of a Brownian motion,
there exist F-predictable, Rd -valued processes ξ and ζ i , i = 1, 2, . . . , k such that dmt = ξt dWt and
dnit = ζti dWt , where the F-martingales m and ni are given by (3.43) and (3.60) respectively.
We are now in a position to state the hedging result for a defaultable claim in the single-name
setup. We consider a defaultable claim (X, A, Z, τ ) satisfying the natural integrability conditions
under Q, which ensure that the cumulative price process S c for this claim is well defined.

Theorem 3.7.1 Assume that there exist F-predictable processes φ1 , . . . , φk satisfying the following
conditions, for any t ∈ [0, T ],
k
X k
X
¡ ¢
φit δti − Seti (κi ) = Zt − Set , φit ζti = ξt . (3.62)
i=1 i=1

Let the process V (φ) be given by (3.61) with the initial condition V0 (φ) = S0c and let φ0 be given by,
for t ∈ [0, T ],
³ X k ´
φ0t = Bt−1 Vt (φ) − φit Sti (κi ) .
i=1

Then the self-financing trading strategy φ = (φ , . . . , φk ) in the savings account B and the assets
0

S i (κi ), i = 1, 2, . . . , k replicates the defaultable claim (X, A, Z, τ ).

122 CHAPTER 3. HAZARD PROCESS APPROACH

Proof. From Lemma 3.7.3, we know that the discounted wealth process satisfies
k
X ³ ´
dVt∗ (φ) = φit Bt−1 (δti − Seti (κi )) dMt + (1 − Ht )G−1 i
t dnt . (3.63)
i=1

Recall also that the discounted cumulative price S c∗ of a defaultable claim is governed by

dStc∗ = Bt−1 (Zt − Set ) dMt + (1 − Ht )G−1

t dmt . (3.64)

We will show that if the two conditions in (3.62) are satisfied for any t ∈ [0, T ], then the equality
Vt (φ) = Stc holds for any t ∈ [0, T ].
Let Ve ∗ (φ) = B −1 Ve (φ) stand for the discounted pre-default wealth, where Ve (φ) is the unique
F-adapted process such that 1{t<τ } Vt (φ) = 1{t<τ } Vet (φ) for every t ∈ [0, T ]. On the one hand, using
(3.62), we obtain

k
X ³ ´
dVet∗ (φ) = φit λt Bt−1 (Seti (κi ) − δti ) dt + G−1 i
t ζt dWt
i=1

= λt Bt−1 (Set − Zt ) dt + G−1

t ξt dWt .

On the other hand, the discounted pre-default cumulative price Sec∗ satisfies

dSetc∗ = λt Bt−1 (Set − Zt ) dt + G−1

t ξt dWt .

Since by assumption Ve0∗ (φ) = V0 (φ) = S0c = Se0c∗ , it is clear that Vet∗ (φ) = Setc∗ for every t ∈ [0, T ].
We thus conclude that the pre-default wealth Ve (φ) of φ and the pre-default cumulative price Sec of
the claim coincide. Note that the first equality in (3.62) is in fact only essential for those values of
t ∈ [0, T ] for which λt 6= 0.
To complete the proof, we need to check what happens when default occurs prior to or at maturity
T . To this end, it suffices to compare the jumps of S c and V (φ) at time τ . In view of (3.62), (3.63)
and (3.64), we obtain
∆Vτ (φ) = Zτ − Seτ = ∆Sτc
c
and thus Vt∧τ (φ) = St∧τ for any t ∈ [0, T ]. After default, we have dVt (φ) = rt Vt (φ) dt and dStc =
rt St dt, so that we conclude that the desired equality Vt (φ) = Stc holds for any t ∈ [0, T ].
c
¤

3.8 Multi-Name CDS Market

In this section, we shall deal with a market model driven by a Brownian filtration in which a finite
family of CDSs with different underlying names is traded.

3.8.1 Valuation of a First-to-Default Claim

Our first goal is to extend the pricing results of Section 3.7.1 to the case of a multi-name credit risk
model with stochastic default intensities.

Joint Survival Process

We assume that we are given n strictly positive random times τ1 , . . . , τn , defined on a common
probability space (Ω, G, Q), and referred to as default times of n credit names. We postulate that
this space is endowed with a reference filtration F, which satisfies Assumption 3.7.2.
3.8. MULTI-NAME CDS MARKET 123

In order to describe dynamic joint behavior of default times, we introduce the conditional joint
survival process G(u1 , . . . , un ; t) by setting, for every u1 , . . . , un , t ∈ R+ ,

G(u1 , . . . , un ; t) = Q(τ1 > u1 , . . . , τn > un | Ft ).

Let us set τ(1) = τ1 ∧ . . . ∧ τn and let us define the process G(1) (t; t), t ∈ R+ by setting

G(1) (t; t) = G(t, . . . , t; t) = Q(τ1 > t, . . . , τn > t | Ft ) = Q(τ(1) > t | Ft ).

It is easy to check that G(1) is a bounded supermartingale. It thus admits the unique Doob-Meyer
decomposition G(1) = µ−C. We shall work throughout under the following extension of Assumption
3.7.1.

Assumption 3.8.1 We assume that the process G(1) is continuous and the increasing process C
is absolutely continuous with respect to the Lebesgue measure, so that dCt = ct dt for some F-
progressively measurable, non-negative process c. We denote by λ e the F-progressively measurable
et = G−1 (t; t)ct . The process λ
process defined as λ e is called the first-to-default intensity.
(1)

We denote Hti = 1{t≥τi } and we introduce the following filtrations Hi , H and G

Hti = σ(Hsi ; s ∈ [0, t]), Ht = Ht1 ∨ . . . ∨ Htn , Gt = Ft ∨ Ht ,

We assume that the usual conditions of completeness and right-continuity are satisfied by these
filtrations. Arguing as in Section 3.7.1, we see that the process
Z t∧τ(1) Z t
ct = Ht(1) − Λ
M e t∧τ = Ht(1) − eu du = H (1) −
λ t
eu du,
(1 − Hu(1) )λ
(1)
0 0

(1) Rt
is a G-martingale, where we denote Ht et =
= 1{t≥τ(1) } and Λ eu du. Note that the first-to-default
λ
0
e satisfies
intensity λ

et = lim 1 Q(t < τ(1) ≤ t + h | Ft ) =

λ
1 1
lim (Ct+h − Ct ).
h↓0 h Q(τ(1) > t | Ft ) G(1) (t; t) h↓0 h

ei and
We make an additional assumption, in which we introduce the first-to-default intensity λ
ci
the associated martingale M for each credit name i = 1, . . . , n.

ei given by
Assumption 3.8.2 For any i = 1, 2, . . . , n, the process λ

ei = lim 1 Q(t < τi ≤ t + h, τ(1) > t | Ft )

λ t
h↓0 h Q(τ(1) > t | Ft )

ci , given by the formula

is well defined and the process M
Z t∧τ(1)
cti = Ht∧τ
M i
− ei du,
λ (3.65)
(1) u
0

is a G-martingale.

Pn ei = λ c = Pn M
e and M ci are valid.
It is worth noting that the equalities i=1 λ i=1
124 CHAPTER 3. HAZARD PROCESS APPROACH

Special Case

Let Γb i , i = 1, 2, . . . , n be a given family of F-adapted, increasing, continuous processes, defined on

a filtered probability space (Ω, e F, P). We postulate that Γ b i = 0 and limt→∞ Γ b it = ∞. For the
0
construction of default times satisfying Assumptions 3.8.1 and 3.8.2, we postulate that (Ω, b F, b is
b P)
an auxiliary probability space endowed with a family ξi , i = 1, 2, . . . , n of random variables uniformly
distributed on [0, 1] and such that their joint probability distribution is given by an n-dimensional
copula function C (see Section 5.4). We then define, for every i = 1, 2, . . . , n,

τi (e
ω, ω b it (e
b ) = inf { t ∈ R+ : Γ ω ) ≥ − ln ξi (b
ω )}.

We endow the space (Ω, G, Q) with the filtration G = F ∨ H1 ∨ · · · ∨ Hn , where the filtration Hi is
generated by the process Hti = 1{t≥τi } for every i = 1, 2, . . . , n.
We have that, for any T > 0 and arbitrary t1 , . . . , tn ≤ T ,

Q(τ1 > t1 , . . . , τn > tn | FT ) = C(Kt11 , . . . , Ktnn ),

bi
where we denote Kti = e−Γt .
Schönbucher and Schubert [137] show that the following equality holds, for arbitrary s ≤ t,
µ ¶
C(Ks1 , . . . , Kti , . . . , Ksn ) ¯¯
Q(τi > t | Gs ) = 1{s<τ(1) } EQ ¯ Fs .
C(Ks1 , . . . , Ksn )
R i
Consequently, assuming that Γb it = t γ
b du, the ith survival intensity equals, on the event {t < τ(1) },
0 u
∂
C(Kt1 , . . . , Ktn ) ∂
ei = γ
λ bti Kti ∂vi
bti Kti
=γ ln C(Kt1 , . . . , Ktn ).
t
C(Kt1 , . . . , Ktn ) ∂vi

One can now easily show that the process Mci , which is given by formula (3.65), is a G-martingale.
This indeed follows from Aven’s lemma [7].

Price Dynamics of a First-to-Default Claim

We will now analyze the risk-neutral valuation of first-to-default claims on a basket of n credit
names. As before, τ1 , . . . , τn are respective default times and τ(1) = τ1 ∧ . . . ∧ τn stands for the
moment of the first default.

Definition 3.8.1 A first-to-default claim with maturity T associated with τ1 , . . . , τn is a defaultable

claim (X, A, Z, τ(1) ), where X is an FT -measurable amount payable at maturity T if no default occurs
prior to or at T and an F-adapted, continuous process of finite variation A : [0, T ] → R with A0 = 0
represents the dividend stream up to τ(1) . Finally, Z = (Z 1 , . . . , Z n ) is the vector of F-predictable,
real-valued processes, where Zτi(1) specifies the recovery received at time τ(1) if default occurs prior
to or at T and the ith name is the first defaulted name, that is, on the event {τi = τ(1) ≤ T }.

The next definition extends Definition 3.7.2 to the case of a first-to-default claim. Recall that
(1)
we denote Ht = 1{t≥τ(1) } for every t ∈ [0, T ].

Definition 3.8.2 The dividend process (Dt , t ∈ R+ ) of a first-to-default claim maturing at T equals,
for every t ∈ R+ ,
Z t∧T
Dt = X1{T <τ(1) } 1[T,∞[ (t) + (1 − Hu(1) ) dAu
0
Z n
X
+ 1{τ(1) =τi } Zui dHu(1) .
]0,t∧T ] i=1
3.8. MULTI-NAME CDS MARKET 125

We are in a position to examine the prices of the first-to-default claim. Note that

1{t<τ(1) } Stc = 1{t<τ(1) } Setc , 1{t<τ(1) } St = 1{t<τ(1) } Set ,

where Sec and Se are pre-default values of S c and S, where the price processes S c and S are given by
Definitions 3.7.3 and 3.7.4, respectively. We postulate that EQ |BT−1 X| < ∞,
¯Z T ¯
¯ ¯
EQ ¯ Bu−1 (1 − Hu(1) ) dAu ¯ < ∞,
0

and for i = 1, 2, . . . , n
EQ |Bτ−1
(1) ∧T
Zτi(1) ∧T | < ∞,

so that the ex-dividend price St (and thus also cumulative price S c ) is well defined for any t ∈ [0, T ].
In the next auxiliary result, we denote Y i = B −1 Z i . Hence Y i is a real-valued, F-predictable process
such that EQ |Yτi(1) ∧T | < ∞.

Lemma 3.8.1 We have that

³X
n ¯ ´
¯
Bt EQ 1{t<τ(1) =τi ≤T } Yτi(1) ¯ Gt
i=1
³ Z T n
X ¯ ´
Bt ei G(1) (u; u) du ¯¯ Ft .
= 1{t<τ(1) } EQ Yui λ u
G(1) (t; t) t i=1

Proof. Let us fix i and let us consider the process Yui = 1A 1]s,v] (u) for some fixed date t ≤ s < v ≤ T
and some event A ∈ Fs . We note that
Z v∧τ(1)
i
1{s<τ(1) =τi ≤v} = Hv∧τ − H i
= cvi − M
M csi + ei du.
λ
(1) s∧τ(1) u
s∧τ(1)

Using Assumption 3.8.2, we thus obtain

³ ¯ ´ ³ ¯ ´
¯ ¯
EQ 1{t<τ(1) =τi ≤T } Yτi(1) ¯ Gt = EQ 1A 1{s<τ(1) =τi ≤v} ¯ Gt
³ ³ Z v∧τ(1) ´¯ ´
= EQ 1A M cvi − M csi + ei du ¯¯ Gt
λ u
s∧τ(1)
³ ³ Z v∧τ(1) ¯ ´¯ ´
cvi − M
= EQ 1A EQ M csi + ei du ¯¯ Gs ¯¯ Gt
λ u
s∧τ(1)
³Z T ∧τ(1) ¯ ´
= EQ ei du ¯¯ Gt
Yui λ u
t∧τ(1)
³ Z T ¯ ´
1 ei G(1) (u; u) du ¯¯ Ft ,
= 1{t<τ(1) } EQ Yui λ u
G(1) (t; t) t

where the last equality follows from the formula

³Z T ∧τ(1) ¯ ´ 1 ³ Z T ¯ ´
¯ ¯
EQ Ru du ¯ Gt = 1{t<τ(1) } EQ Ru G(1) (u; u) du ¯ Ft ,
t∧τ(1) G(1) (t; t) t

which holds for any F-predictable process R such that the right-hand side is well defined. ¤

Given Lemma 3.8.1, the proof of the next result is very much similar to that of Proposition 3.7.1
and thus is omitted.
126 CHAPTER 3. HAZARD PROCESS APPROACH

Proposition 3.8.1 The pre-default ex-dividend price Se of a first-to-default claim (X, A, Z, τ(1) )
satisfies
Bt ³Z T ³Xn ´¯ ´
Set = EQ Bu−1 G(1) (u; u) ei du + dAu ¯¯ Ft .
Zui λ u
G(1) (t; t) t i=1
Bt ³ ¯ ´
¯
+ EQ BT−1 G(1) (T ; T )X1{t<T } ¯ Ft .
G(1) (t; t)

By proceeding as in the proof of Proposition 3.7.2, one can also establish the following result,
which gives dynamics of price processes Se and S c of a first-to-default claim.
Recall that µ is the continuous martingale arising in the Doob-Meyer decomposition of the process
G(1) (see Assumption 3.8.1).

Proposition 3.8.2 The dynamics of the pre-default ex-dividend price Se of a first-to-default claim
(X, A, Z, τ(1) ) on [0, τ(1) ∧ T ] are
n
X ¡ ¢
et )Set dt −
dSet = (rt + λ ei Z i dt − dAt + G−1 (t; t) Bt dmt − Set dµt
λ t t (1)
i=1
¡ ¢
+ G−2 e
(1) (t; t) St dhµit − Bt dhµ, mit ,

where the continuous F-martingale m is given by the formula

³Z T ³Xn ´¯ ´
mt = EQ −1
Bu G(1) (u; u) ei du + dAu ¯¯ Ft
Zui λ u
0 i=1
³ ¯ ´
¯
+ EQ BT−1 G(1) (T ; T )X ¯ Ft .

The dynamics of the cumulative price S c on [0, τ(1) ∧ T ] are

n
X ³ n
X ´
dStc = (Zti − Set− ) dMti + rt Set − ei Z i dt − dAt
λ t t
i=1 i=1
¡ ¢ ¡ ¢
+ G−1 e −2 e
(1) (t; t) Bt dmt − St dµt + G(1) (t; t) St dhµit − Bt dhµ, mit .

Hypothesis (H)

As in the single-name case, the most explicit results can be derived under an additional assumption
of the immersion property of filtrations F and G.

Assumption 3.8.3 Any F-martingale under Q is a G-martingale under Q. This also implies that
hypothesis (H) holds between F and G. In particular, any F-martingale is also a Gi -martingale for
i = 1, 2, . . . , n, that is, hypothesis (H) holds between F and Gi for i = 1, 2, . . . , n.

It is worth stressing that, in general, there is no reason to expect that any Gi -martingale is
necessarily a G-martingale. We shall argue that even when the reference filtration F is trivial this
is not the case, in general (except for some special cases, for instance, under the independence
assumption).
1|2
Example 3.8.1 Let us take n = 2 and let us denote Gt = Q(τ1 > t | Ht2 ) and G(u, v) = Q(τ1 >
u, τ2 > v). It is then easy to prove that
µ ¶
1|2 ∂2 G(t, t) G(t, t)
dGt = − dMt2
∂2 G(0, t) G(0, t)
µ ¶
2 2 ∂1 G(t, t)
+ Ht ∂1 h(t, τ2 ) + (1 − Ht ) dt,
G(0, t)
3.8. MULTI-NAME CDS MARKET 127

∂2 G(t,u)
where h(t, u) = ∂2 G(0,u) and M 2 is the H2 -martingale given by
Z t∧τ2
∂2 G(0, u)
Mt2 = Ht2 + du.
0 G(0, u)

If hypothesis (H) holds between H2 and H1 ∨ H2 then the martingale part in the Doob-Meyer
decomposition of G1|2 vanishes. We thus see that hypothesis (H) is not always valid, since clearly
∂2 G(t, t) G(t, t)
−
∂2 G(0, t) G(0, t)
does not vanish, in general. One can note that in the special case when τ2 < τ1 , the martingale part
in the above-mentioned decomposition disappears and thus hypothesis (H) holds between H2 and
H1 ∨ H2 .

From now on, we shall work under Assumption 3.8.3. In that case, the dynamics of price processes
obtained in Proposition 3.8.1 simplify, as the following result shows.

Corollary 3.8.1 The pre-default ex-dividend price Se of a first-to-default claim (X, A, Z, τ(1) ) sat-
isfies
n
X
et )Set dt −
dSet = (rt + λ ei Z i dt − dAt + Bt G−1 (t; t) dmt
λ t t (1)
i=1

with the continuous F-martingale m defined in Proposition 3.8.2. The cumulative price S c of a
first-to-default claim (X, A, Z, τ(1) ) is given by the expression, for t ∈ [0, T ∧ τ(1) ],
n
X
dStc = rt Stc dt + (Zti − Set ) dM
cti + Bt G−1 (t; t) dmt .
(1)
i=1

Equivalently, for t ∈ [0, T ∧ τ(1) ],

n
X
dStc = rt Stc dt + (Zti − Set ) dM
cti + Bt G−1 (t; t) dm̌t ,
(1)
i=1

where m̌ is a G-martingale given by m̌t = mt∧τ(1) for every t ∈ [0, T ].

In what follows, we assume that F is generated by a Brownian motion. Then there exists an
F-predictable process ξ for which dmt = ξt dWt so that the last formula in Corollary 3.8.1 yields the
following result.

Corollary 3.8.2 The discounted cumulative price of a first-to-default claim (X, A, Z, τ(1) ) satisfies,
for t ∈ [0, T ∧ τ(1) ],
Xn
c∗
dSt = Bt−1 (Zti − Set ) dM
cti + G−1 (t; t)ξt dWt .
(1)
i=1

Price Dynamics of a CDS

By the ith CDS we mean the credit default swap written on the ith reference name, with the maturity
date Ti , the constant spread κi and the protection process δ i , as specified by Definition 3.7.5. Let
i
St|j (κi ) stand for the ex-dividend price at time t of the ith CDS on the event τ(1) = τj = t for some
j 6= i. This value can be represented using a suitable extension of Proposition 3.8.1, but we decided
to omit the derivation of this pricing formula.
i
Assume that we have already computed the value of St|j (κi ) for every t ∈ [0, Ti ]. Then the ith
CDS can be seen, on the random interval [0, Ti ∧ τ(1) ], as a first-to-default claim (X, A, Z, τ(1) ) with
128 CHAPTER 3. HAZARD PROCESS APPROACH

i
X = 0, Z = (St|1 (κi ), . . . , δ i , . . . , St|n
i
(κi )) and At = −κi t. The last observation applies also to the
random interval [0, T ∧τ(1) ] for any fixed date T ≤ Ti . Let us denote by ni the following F-martingale
n Z
³X Ti ³ n
X ´ ¯ ´
nit = EQ Bu−1 G(1) (u; u) ei
δui λ + i
Su|j ej − κi du ¯¯ Ft .
(κi )λ
u u
i=1 0 j=1 ,j6=i

The following result can be easily deduced from Proposition 3.8.1.

Corollary 3.8.3 The cumulative price of the ith CDS satisfies, for every t ∈ [0, Ti ∧ τ(1) ],

dStc,i (κi ) = rt Stc,i (κi ) dt + (δti − Seti (κi )) dMci

t
Xn
+ i
(St|j (κi ) − Seti (κi )) dMctj + Bt G−1 (t; t) dni .
(1) t
j=1, j6=i

Consequently, the discounted cumulative price of the ith CDS satisfies, for every t ∈ [0, Ti ∧ τ(1) ],
n
X
dStc,i,∗ (κi ) = Bt−1 (δti − Seti (κi )) dM
cti + Bt−1 i
(St|j (κi ) − Seti (κi )) dM
ctj
j=1, j6=i

+ G−1 i
(1) (t; t)ζt dWt ,

where ζ i is an F-predictable process such that dnit = ζti dWt .

3.8.2 Replication of a First-to-Default Claim

Our final goal is to extend Theorem 3.7.1 of Section 2.2.7 to the case of several credit names in a
hazard process model in which credit spreads are driven by a multi-dimensional Brownian motion.
We consider a self-financing trading strategy φ = (φ0 , . . . , φk ) with G-predictable components, as
defined in Section 3.7.3. The 0th traded asset is thus the savings account; the remaining k primary
assets are single-name CDSs with different underlying credit names and/or maturities. As before, for
any l = 1, 2, . . . , k we will use the short-hand notation S l (κl ) and S c,l (κl ) to denote the ex-dividend
and cumulative prices of CDSs with respective dividend processes D(κl , δ l , Tl , τ̃l ) given by formula
(3.49). Note that here τ̃l = τj for some j = 1, 2, . . . , n. We will thus write τ̃l = τjl in what follows.

Remark 3.8.1 Note that, typically, we will have k = n + d so that the number of traded assets will
be equal to n + d + 1.

Recall that we denote by S c the cumulative price of a first-to-default claim (X, A, Z, τ(1) ), where
the recovery process Z is n-dimensional, specifically, Z = (Z 1 , . . . , Z n ). We already know that if
hypothesis (H) is satisfied by the filtrations F and G then the dynamics of S c under the risk-neutral
measure Q are (see Corollary 3.8.1)
n
X
dStc = rt Stc dt + (Zti − Set ) dM
cti + Bt G−1 (t; t) dmt ,
(1)
i=1

where the continuous F-martingale m under Q is given by the formula (see Proposition 3.8.2)
³Z T ³X
n ´¯ ´
mt = EQ Bu−1 G(1) (u; u) ei du + dAu ¯¯ Ft
Zui λ u
0 i=1
³ ¯ ´
¯
+ EQ BT−1 G(1) (T ; T )X ¯ Ft .

We adopt the following natural definition of replication of a first-to-default claim.

3.8. MULTI-NAME CDS MARKET 129

Definition 3.8.3 We say that a self-financing strategy φ = (φ0 , . . . , φk ) replicates a first-to-default

c
claim (X, A, Z, τ(1) ) if its wealth process V (φ) satisfies the equality Vt∧τ(1) (φ) = St∧τ (1)
for any
t ∈ [0, T ].

When dealing with replicating strategies in the sense of the definition above, we may and do
assume, without loss of generality, that the components of the process φ are F-predictable processes.
This is rather obvious from the mathematical point of view, since it is well known that prior to default
any G-predictable process is equal to the unique F-predictable process. Also this property supports
the common intuition that the knowledge of default time should not be used in the construction of
the replicating strategy for a first-to-default claim.
The following auxiliary result is a direct counterpart of Lemma 3.7.3.

Lemma 3.8.2 We have, for any t ∈ [0, T ∧ τ(1) ],

k
X ¡ ¢
dVt∗ (φ) = φlt Bt−1 δtl − Setl (κl ) dM
ctjl
l=1
k ³
X n
X ´
¡ l ¢
+ Bt−1 St|j (κl ) − Setl (κl ) dM
ctj + G−1 (t; t) dnlt ,
(1)
l=1 j=1 ,j6=jl

where
³Z Tl ³ n
X ´ ¯ ´
nlt = EQ e jl +
Bu−1 G(1) (u; u) δul λ l
Su|j ej − κl du ¯¯ Ft .
(κl )λ
u u
0 j=1 ,j6=jl

Proof. The proof of the lemma easily follows from Lemma 3.7.2 combined with Corollary 3.8.3. The
details are left to the reader. ¤

We are now in a position to extend Theorem 3.7.1 to the case of a first-to-default claim on a
basket of n credit names. At the same time, Theorem 3.8.1 is also a generalization of Theorem 2.5.1
to the case of non-trivial reference filtration F.
Let ξ and ζ l , l = 1, 2, . . . , k be F-predictable, Rd -valued processes such that the following repre-
sentations are valid
dmt = ξt dWt
and
dnlt = ζtl dWt .
The existence of processes ξ and ζ l for l = 1, 2, . . . , k is an immediate consequence of the classic
predictable representation theorem for the Brownian filtration (of course, these processes are rarely
known explicitly).

Theorem 3.8.1 Assume that the processes φe1 , . . . , φen satisfy, for t ∈ [0, T ] and i = 1, 2, . . . , n
k
X k
X
¡ ¢ ¡ l ¢
φelt δtl − Setl (κl ) + φelt St|i (κl ) − Setl (κl ) = Zti − Set
l=1, jl =i l=1, jl 6=i

and
k
X
φelt ζtl = ξt .
l=1

Let us set = φei (t ∧ τ(1) ) for i = 1, 2, . . . , k and t ∈ [0, T ]. Let the process V (φ) be given by Lemma
φit
3.8.2 with the initial condition V0 (φ) = S0c and let φ0 be given by
k
X
Vt (φ) = φ0t Bt + φlt Stl (κl ).
l=1
130 CHAPTER 3. HAZARD PROCESS APPROACH

Then the self-financing strategy φ = (φ0 , . . . , φk ) replicates the first-to-default claim (X, A, Z, τ(1) ).

Proof. The proof goes along the similar lines as the proof of Theorem 3.7.1. It suffices to examine
replicating strategy on the random interval [0, T ∧ τ(1) ]. On the one hand, in view of Lemma 3.8.2,
the wealth process of a self-financing strategy φ satisfies on [0, T ∧ τ(1) ]

k
X ¡ ¢
dVt∗ (φ) = φelt Bt−1 δtl − Setl (κl ) dM
ctjl
l=1
k ³
X n
X ´
¡ l ¢
+ Bt−1 St|j (κl ) − Setl (κl ) dM
ctj + G−1 (t; t)ζ l dWt .
(1) t
l=1 j=1 ,j6=jl

On the other hand, the discounted cumulative price of a first-to-default claim (X, A, Z, τ(1) ) satisfies
on the interval [0, T ∧ τ(1) ]
n
X
dStc∗ = ci + (1 − Ht(1) )G−1 (t; t)ξt dWt .
Bt−1 (Zti − St− ) dMt (1)
i=1

A comparison of the last two formulae leads directly to the stated conditions. To complete the proof,
it suffices to verify that the strategy φ = (φ0 , . . . , φk ) introduced in the statement of the theorem
replicates a first-to-default claim, in the sense of Definition 3.8.3. Since this verification is rather
standard, we leave the details to the reader. ¤
Chapter 4

Hedging of Defaultable Claims

In this chapter, we study hedging strategies for credit derivatives under the assumption that certain
primary assets are traded. We follow here Bielecki et al. [14, 16] and we put special emphasis on
the PDE approach in a Markovian setup. For related methods and results, we refer to Arvanitis and
Laurent [6], Blanchet-Scalliet and Jeanblanc [27], Collin-Dufresne and Hugonnier [52], Greenfield
[84], Laurent et al. [111], Laurent [112], Petrelli et al. [131], Rutkowski and Yousiph [135], and
Vaillant [141].

4.1 Semimartingale Market Model

We assume that we are given a probability space (Ω, G, P) endowed with a (one- or multi-dimensional)
standard Brownian motion W and a random time τ , which admits the F-intensity γ under P, where
F is the filtration generated by the process W . Since the default time is assumed to admit the F-
intensity, it is not an F-stopping time. Indeed, it is well known that any stopping time with respect
to a Brownian filtration is predictable, and thus does not admit an F-intensity.

4.1.1 Dynamics of Asset Prices

We interpret τ as the common default time for all defaultable assets in our model. In what follows,
we fix a finite horizon date T > 0. For simplicity, we assume that only three primary assets are
traded in the market and the dynamics under the historical probability P of their prices are, for
i = 1, 2, 3 and every t ∈ [0, T ],
¡ ¢
dYti = Yt−
i
µi,t dt + σi,t dWt + κi,t dMt , (4.1)
R t∧τ
where Mt = Ht − 0
γu du is a martingale or, equivalently,
¡ ¢
dYti = Yt−
i
(µi,t − κi,t γt 1{t<τ } ) dt + σi,t dWt + κi,t dHt . (4.2)

The processes (µi , σi , κi ) = (µi,t , σi,t , κi,t , t ∈ R+ ), i = 1, 2, 3, are assumed to be G-adapted, where
G = F ∨ H. In addition, we assume that Y0i > 0 and κi ≥ −1 for any i = 1, 2, 3, so that Y i are
non-negative processes and they are strictly positive prior to τ . In the case of constant coefficients,
we have
2
Yti = Y0i eµi t eσi Wt −σi t/2 e−κi γi (t∧τ ) (1 + κi )Ht .

According to Definition 4.1.2 below, replication refers to the behavior of the wealth process V (φ)
on the random interval [[0, τ ∧ T ]] only. Therefore, for the purpose of replication of defaultable claims
of the form (X, Z, τ ), it is sufficient to consider prices of primary assets stopped at τ ∧ T . This

131
132 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

implies that instead of dealing with G-adapted coefficients in (4.1), it suffices to focus on F-adapted
coefficients for the price processes stopped at τ ∧ T . However, for the sake of completeness, we will
also deal with a T -maturity claim of the form Y = G(YT1 , YT2 , YT3 , HT ) (see Section 4.4 below).

Pre-Default Values

As will become clear in what follows, when dealing with defaultable claims of the form (X, Z, τ ), we
will be mainly concerned with the pre-default prices. The pre-default price Ye i of the ith asset is an
F-adapted, continuous process, given by the equation, for i = 1, 2, 3 and t ∈ [0, T ],
¡ ¢
dYeti = Yeti (µi,t − κi,t γt ) dt + σi,t dWt (4.3)

with Ye0i = Y0i . Put another way, Ye i is the unique F-predictable process such that the equality

Yeti 1{t≤τ } = Yti 1{t≤τ }

holds for every t ∈ R+ . When dealing with the pre-default prices, we may and do assume, without
loss of generality, that the processes µi , σi and κi are F-predictable.
Let us stress that the historically observed drift coefficient is µi,t − κi,t γt , which appears in (4.2),
rather than the drift µi,t , which appears (4.1). The drift coefficient µi,t is already credit-risk adjusted
in the sense of our model and it is not directly observed. This convention was chosen here for the
sake of simplicity of notation. It also lends itself to the following intuitive interpretation: if φi is the
number of units of the ith asset held in our portfolio at time t then the gains/losses from trades in
this asset, prior to default time, can be represented by the differential
¡ ¢
φit dYeti = φit Yeti µi,t dt + σi,t dWt − φit Yeti κi,t γt dt.

The last term in the formula above may be formally treated as an effect of dividends that are paid
continuously at the dividend rate κi,t γt . This nice interpretation is not necessarily useful in practice,
since the quantity κi,t γt cannot be observed directly and, as is well known, a reliable estimation of
the drift coefficient in dynamics (4.3) is extremely difficult anyway. Moreover, it is a delicate issue
how to disentangle in practice the two components of the drift coefficient in (4.3). Still, if this formal
interpretation is adopted, it is sometimes possible to make use of the standard results concerning
the valuation of derivatives of dividend-paying assets.
We shall argue below that, although there is formally nothing wrong with the dividend-based
approach, a more pertinent and simpler approach to hedging of defaultable claims hinges on the
assumption that only the effective drift, which is given by the expression

µ
bi,t = µi,t − κi,t γt ,

is observable. Moreover, in practical approach to hedging, the values of drift coefficients in dynamics
of asset prices will play no essential role, so that we will not postulate that they are among market
observables.

Market Observables

To summarize, we assume throughout that the market observables are: the pre-default market prices
of primary assets, their volatilities and correlations, as well as the jump coefficients κi,t (the financial
interpretation of jump coefficients is examined in the next subsection). To summarize, we postulate
that under the statistical probability P the processes Y i , i = 1, 2, 3 satisfy
¡ ¢
dYti = Yt−i
µ
ei,t dt + σi,t dWt + κi,t dHt

where the drift terms µ ei,t are not observed, but we can observe the volatilities σi,t (and thus the
asset correlations) and we have an a priori assessment of jump coefficients κi,t . In this general setup,
4.1. SEMIMARTINGALE MARKET MODEL 133

the most natural assumption is that the dimension of a driving Brownian motion W coincide with
the number of tradable assets. However, for the sake of simplicity of presentation, we will frequently
assume that the process W is one-dimensional.
One of our goals will be to establish closed-form expressions for replicating strategies for derivative
securities in terms of market observables only (whenever replication of a given claim is actually
feasible). To achieve this goal, we shall combine a general theory of hedging defaultable claims
within a continuous semimartingale setup, with a judicious specification of particular models with
deterministic volatilities and correlations.

Recovery Schemes

It is clear that the sample paths of price processes Y i are continuous, except for a possible discon-
tinuity at time τ . Specifically, we have that

∆Yτi := Yτi − Yτi− = κi,τ Yτi− ,

so that the value of Y i at τ is given by

Yτi = Yτi− (1 + κi,τ ) = Yeτi− (1 + κi,τ ).

A primary asset Y i is termed a default-free asset (defaultable asset, respectively) if κi = 0 (κi 6= 0,

respectively). In the special case when κi = −1, we say that a defaultable asset Y i is subject to
the zero recovery scheme, since its price drops to zero at time τ and remains null at any later date.
Such an asset ceases to exist after default, in the sense that it is no longer traded after default. This
feature makes the case of a zero recovery essentially different from other cases, as we shall see in the
sequel.
In the market practice, it is much more common for a credit derivative to deliver a positive
recovery if default event occurs during the contract’s lifetime (for instance, a protection payment of
a credit default swap).
Formally, the value of recovery at default is given as the value of some predetermined stochastic
process, that is, it is equal to the value at time τ of some F-adapted recovery process Z.
For instance, the recovery process Z can be equal to δ, where δ is a constant, or to g(t, δYt )
where g is a deterministic function and (Yt , t ∈ R+ ) is the price process of some default-free asset.
Typically, the recovery is paid at default time, but it is sometimes postponed to the maturity date.
Let us observe that the case where a defaultable asset Y i pays a pre-determined recovery at
default is covered by our setup defined in (4.1). For example, the case of a constant recovery payoff
i −1
δi ≥ 0 at default time τ corresponds to the process κi,t = δi (Yt− ) − 1. Under this convention, the
i
price Y is governed under P by the SDE
¡ ¢
dYti = Yt−
i i −1
µi,t dt + σi,t dWt + (δi (Yt− ) − 1) dMt .

If the recovery is proportional to the pre-default value Yτi− and is paid at default time τ (this scheme
is known as the fractional recovery of market value), we set κi,t = δi − 1 and thus
¡ ¢
dYti = Yt−
i
µi,t dt + σi,t dWt + (δi − 1) dMt .

Defaultable Claims

For the purpose of this chapter, it will be enough to define a generic defaultable claim as follows
(note that, formally, it suffices to set A = 0 in Definition 3.7.1).

Definition 4.1.1 A defaultable claim with maturity date T is represented by a triplet (X, Z, τ ),
where:
134 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

(i) the default time τ specifies the random time of default, and thus also the default events {t ≥ τ }
for every t ∈ [0, T ],
(ii) the promised payoff X ∈ FT represents the random payoff received by the owner of the claim
at time T , provided that there was no default prior to or at time T ; the actual payoff at time T
associated with X thus equals X1{T <τ } ,
(iii) the F-adapted recovery process (Zt , t ∈ [0, T ]) specifies the recovery payoff Zτ received by the
owner of a claim at time of default (or at maturity), provided that the default occurred prior to or
at maturity date T .

In practice, hedging of a credit derivative after default time is usually of minor interest. Also, in
a model with a single default time, hedging after default reduces to replication of a non-defaultable
claim. It is thus natural to define the replication of a defaultable claim in the following way.

Definition 4.1.2 We say that a self-financing strategy φ replicates a defaultable claim (X, Z, τ ) if its
wealth process (Vt (φ), t ∈ [0, T ]) satisfies VT (φ)1{T <τ } = X1{T <τ } and Vτ (φ)1{T ≥τ } = Zτ 1{T ≥τ } .

When dealing with replicating strategies, in the sense of Definition 4.1.2, we will always assume,
without loss of generality, that the components of the process φ are F-predictable processes, rather
than G-predictable.

4.2 Trading Strategies

In this section, we consider a fairly general setup. In particular, processes (Yti , t ∈ [0, T ]) for
i = 1, 2, 3 are assumed to be non-negative semimartingales on a probability space (Ω, G, P) endowed
with some filtration G. We assume that they represent spot prices of traded assets in our model of
the financial market. Neither the existence of a savings account nor the market completeness are
postulated, in general. We restrict our attention to the case where only three primary assets are
traded. The general case of k traded assets was examined by Bielecki et al. [15, 17].
Our goal is to characterize contingent claims which are hedgeable, in the sense that they can be
replicated by continuously rebalanced portfolios consisting of primary assets. Here, by a contingent
claim we mean an arbitrary GT -measurable random variable. We will work throughout under the
standard assumptions of a frictionless market (no transaction costs or taxes, no restrictions on the
short sale of assets, perfect liquidity, etc.)

4.2.1 Unconstrained Strategies

Let φ = (φ1 , φ2 , φ3 ) be a trading strategy; in particular, each process φi is predictable with respect
to the filtration G. The corresponding wealth process (Vt (φ), t ∈ [0, T ]) is defined by the formula,
for every t ∈ [0, T ],
X3
Vt (φ) = φit Yti .
i=1
A trading strategy φ is said to be self-financing if the wealth process satisfies, for every t ∈ [0, T ],
3 Z
X
Vt (φ) = V0 (φ) + φiu dYui .
i=1 ]0,t]

Let Φ stand for the class of all self-financing trading strategies. We shall first prove that a self-
financing strategy is determined by its initial wealth and the two components φ2 , φ3 . To this end,
we postulate that the price of Y 1 follows a strictly positive process and we choose Y 1 as a numéraire
asset. We shall now analyze the relative values, V 1 and Y i,1 , which are given by
Vt1 (φ) = Vt (φ)(Yt1 )−1 , Yti,1 = Yti (Yt1 )−1 .
4.2. TRADING STRATEGIES 135

Lemma 4.2.1 (i) For any φ ∈ Φ, we have, for every t ∈ [0, T ],

3 Z
X
Vt1 (φ) = V01 (φ) + φiu dYui,1 .
i=2 ]0,t]

(ii) Conversely, let X be a GT -measurable random variable, and let us assume that there exists x ∈ R
and G-predictable processes φi , i = 2, 3 such that

³ 3 Z
X ´
X= YT1 x+ φiu dYui,1 . (4.4)
i=2 ]0,T ]

Then there exists a G-predictable process φ1 such that the trading strategy φ = (φ1 , φ2 , φ3 ) is self-
financing and replicates X. Moreover, the wealth process of φ (that is, the price of X at time t)
satisfies Vt (φ) = Vt1 Yt1 , where, for every t ∈ [0, T ],
3 Z
X
Vt1 =x+ φiu dYui,1 . (4.5)
i=2 ]0,t]

Proof. In the case of continuous semimartingales, the result is well known; the demonstration for
discontinuous semimartingales is not much different. Nevertheless, for the reader’s convenience, we
provide a detailed proof.
Let us first introduce some notation. As usual, [X, Y ] stands for the quadratic covariation (the
bracket) of the two semimartingales X and Y , as formally defined by the Itô integration by parts
formula Z Z
Xt Yt = X0 Y0 + Xu− dYu + Yu− dXu + [X, Y ]t .
]0,t] ]0,t]

For any càdlàg process Y , we denote by ∆Yt = Yt −Yt− the size of the jump at time t. Let V = V (φ)
be the value of a self-financing strategy and let V 1 = V 1 (φ) = V (φ)(Y 1 )−1 be its value relative to
the numéraire Y 1 . The integration by parts formula yields

dVt1 = Vt− d(Yt1 )−1 + (Yt−

1 −1
) dVt + d[(Y 1 )−1 , V ]t .
P3
From the self-financing condition, we have dVt = i=1 φit dYti . Hence, using elementary rules to
compute the quadratic covariation [X, Y ] of the two semimartingales X, Y , we obtain

dVt1 = φ1t Yt−

1
d(Yt1 )−1 + φ2t Yt−
2
d(Yt1 )−1 + φ3t Yt−
3
d(Yt1 )−1
1 −1 1
+ (Yt− ) φt dYt1 + (Yt− 1 −1 2
) φt dYt1 + (Yt− 1 −1 3
) φt dYt1
+ φ1t d[(Y 1 )−1 , Y 1 ]t + φ2t d[(Y 1 )−1 , Y 2 ]t + φ3t d[(Y 1 )−1 , Y 1 ]t
¡ 1 ¢
= φ1t Yt− d(Yt1 )−1 + (Yt− 1 −1
) dYt1 + d[(Y 1 )−1 , Y 1 ]t
¡ 2 ¢
+ φ2t Yt− d(Yt1 )−1 + (Yt− 1 −1
) dYt− 1
+ d[(Y 1 )−1 , Y 2 ]t
¡ 3 ¢
+ φ3t Yt− d(Yt1 )−1 + (Yt− 1 −1
) dYt− 1
+ d[(Y 1 )−1 , Y 3 ]t .

We now observe that

1
Yt− d(Yt1 )−1 + (Yt−
1 −1
) dYt1 + d[(Y 1 )−1 , Y 1 ]t = d(Yt1 (Yt1 )−1 ) = 0

and
i
Yt− d(Yt1 )−1 + (Yt−
1 −1
) dYti + d[(Y 1 )−1 , Y i ]t = d((Yt1 )−1 Yti ).
Consequently,
dVt1 = φ2t dYt2,1 + φ3t dYt3,1 ,
136 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

as was claimed in part (i). We now proceed to the proof of part (ii). We assume that (4.4) holds for
some constant x and processes φ2 , φ3 and we define the process V 1 by setting, for every t ∈ [0, T ]
(cf. (4.5)),
X3 Z
1
Vt = x + φiu dYui,1 .
i=2 ]0,t]

Next, we define the process φ1 as follows

3
X ³ 3
X ´
φ1t = Vt1 − φit Yti,1 = (Yt1 )−1 Vt − φit Yti ,
i=2 i=2

where we set Vt = Vt1 Yt1 for t ∈ [0, T ]. Since

3
X
dVt1 = φit dYti,1 ,
i=2

for the process V we obtain

dVt = d(Vt1 Yt1 ) = Vt−

1
dYt1 + Yt−
1
dVt1 + d[Y 1 , V 1 ]t
3
X ¡ 1 ¢
= 1
Vt− dYt1 + φit Yt− dYti,1 + d[Y 1 , Y i,1 ]t .
i=2

From the Itô integration by parts formula, we obtain

dYti = d(Yti,1 Yt1 ) = Yt−

i,1
dYt1 + Yt−
1
dYti,1 + d[Y 1 , Y i,1 ]t ,

and thus
3
X
1
¡ i,1 ¢
dVt = Vt− dYt1 + φit dYti − Yt− dYt1
i=2
³ 3
X ´ 3
X
1 i,1
= Vt− − φit Yt− dYt1 + φit dYti .
i=2 i=2
P3
Our aim was to prove that dVt = i=1 φit dYti . The last equality is indeed satisfied if
3
X 3
X
φ1t = Vt1 − φit Yti,1 = Vt−
1
− i,1
φit Yt− , (4.6)
i=2 i=2

that is, provided that

3
X
∆Vt1 = φit ∆Yti,1 ,
i=2

which is satisfied, in view of definition (4.5) of V 1 . Note also that, from the second equality in
(4.6), we deduce that the process φ1 is G-predictable. Finally, the wealth process of φ satisfies
Vt (φ) = Vt1 Yt1 for every t ∈ [0, T ] and thus VT (φ) = X. ¤

We say that a self-financing strategy φ replicates a claim X ∈ GT if

3
X
X= φiT YTi = VT (φ)
i=1

or, equivalently,
3 Z
X
X = V0 (φ) + φit dYti .
i=1 ]0,T ]
4.2. TRADING STRATEGIES 137

Suppose that there exists an EMM for some choice of a numéraire asset, and let us restrict our
attention to the class of all admissible trading strategies, so that our model is arbitrage-free.
Assume that a claim X can be replicated by some admissible trading strategy, so that it is
attainable (or hedgeable). Then, by definition, the arbitrage price at time t of X, denoted as πt (X),
equals Vt (φ) for any admissible trading strategy φ that replicates X.
In the context of Lemma 4.2.1, it is natural to choose as an EMM a probability measure Q1
equivalent to P on (Ω, GT ) and such that the prices Y i,1 , i = 2, 3, are G-martingales under Q1 . If a
contingent claim X is hedgeable, then its arbitrage price satisfies

πt (X) = Yt1 EQ1 (X(YT1 )−1 | Gt ). (4.7)

We emphasize that even when an EMM Q1 is not unique, the price of any hedgeable claim X is
given by the conditional expectation above. Put another way, in the case of a hedgeable claim, the
conditional expectations (4.7) under various equivalent martingale measures coincide.

4.2.2 Constrained Strategies

In this section, we make an additional assumption that the price process Y 3 is strictly positive. Let
φ = (φ1 , φ2 , φ3 ) be a self-financing trading strategy satisfying the following constraint
2
X
φit Yt−
i
= Zt , ∀ t ∈ [0, T ], (4.8)
i=1

for a predetermined, G-predictable process Z. In the financial interpretation, equality (4.8) means
that a portfolio φ is rebalanced in such a way that the total wealth invested in assets Y 1 , Y 2 matches
a predetermined stochastic process Z. For this reason, the constraint given by (4.8) is referred to as
the balance condition.
Our first goal is to extend part (i) in Lemma 4.2.1 to the case of constrained strategies. Let
Φ(Z) stand for the class of all (admissible) self-financing trading strategies that satisfy the balance
condition (4.8). They will be sometimes referred to as constrained strategies. Since any strategy
P3
φ ∈ Φ(Z) is self-financing, from dVt (φ) = i=1 φit dYti , we obtain

3
X 3
X
∆Vt (φ) = φit ∆Yti = Vt (φ) − φit Yt−
i
.
i=1 i=1

By combining this equality with (4.8), we deduce that

3
X
Vt− (φ) = φit Yt−
i
= Zt + φ3t Yt−
i
.
i=1

Let us write
Yti,3 = Yti (Yt3 )−1 , Zt3 = Zt (Yt3 )−1 .
The following result extends Lemma 1.7 in Bielecki et al. [12] from the case of continuous semi-
martingales to the general case (see also [15, 17]). It is apparent from Proposition 4.2.1 that the
wealth process V (φ) of a strategy φ ∈ Φ(Z) depends only on a single component of φ, namely, φ2 .

Proposition 4.2.1 The relative wealth Vt3 (φ) = Vt (φ)(Yt3 )−1 of any trading strategy φ ∈ Φ(Z)
satisfies à ! Z
Z 2,3
3 3 2 2,3 Yu− 1,3 Zu3 1,3
Vt (φ) = V0 (φ) + φu dYu − 1,3 dYu + 1,3 dYu . (4.9)
]0,t] Yu− ]0,t] Yu−
138 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

Proof. Let us consider discounted values of price processes Y 1 , Y 2 , Y 3 , with Y 3 taken as a numéraire
asset. By virtue of part (i) in Lemma 4.2.1, we thus have
2 Z
X
Vt3 (φ) = V03 (φ) + φiu dYui,3 . (4.10)
i=1 ]0,t]

The balance condition (4.8) implies that

2
X i,3
φit Yt− = Zt3 ,
i=1

and thus ³ ´
1,3 −1 2,3
φ1t = (Yt− ) Zt3 − φ2t Yt− . (4.11)
By inserting (4.11) into (4.10), we arrive at the asserted formula (4.9). ¤

The next result will prove particularly useful for deriving replicating strategies for defaultable
claims.

Proposition 4.2.2 Let a GT -measurable random variable X represent a contingent claim that settles
at time T . We set
Y 2,3
dYt∗ = dYt2,3 − t−1,3 dYt
1,3
= dYt2,3 − Yt−
2,1
dYt1,3 , (4.12)
Yt−
where, by convention, the initial value is Y0∗ = 0. Assume that there exists a G-predictable process
φ2 , such that à !
Z Z
3 2 ∗ Zt3 1,3
X = YT x + φt dYt + 1,3 dYt . (4.13)
]0,T ] ]0,T ] Yt−

Then there exist G-predictable processes φ1 and φ3 such that the strategy φ = (φ1 , φ2 , φ3 ) belongs to
Φ(Z) and replicates X. The wealth process of φ equals, for every t ∈ [0, T ],
à Z Z !
3 2 ∗ Zu3 1,3
Vt (φ) = Yt x + φu dYu + 1,3 dYu .
]0,t] ]0,t] Yu−

Proof. As expected, we first set (note that the component φ1 follows a G-predictable process)
1 ³ ´
φ1t = 1 Zt − φ2t Yt−
2
(4.14)
Yt−
and Z Z
Zu3
Vt3 = x + φ2u dYu∗ + 1,3
1,3 dYu .
]0,t] ]0,t] Yu−
Arguing along the same lines as in the proof of Proposition 4.2.1, we obtain
2 Z
X
Vt3 = V03 + φiu dYui,3 .
i=1 ]0,t]

Now, we define
2
X ³ 2
X ´
φ3t = Vt3 − φit Yti,3 = (Yt3 )−1 Vt − φit Yti ,
i=1 i=1

where Vt = Vt3 Yt3 . As in the proof of Lemma 4.2.1, we check that

2
X i,3
φ3t = Vt−
3
− φit Yt− ,
i=1
4.2. TRADING STRATEGIES 139

and thus the process φ3 is G-predictable. It is clear that the strategy φ = (φ1 , φ2 , φ3 ) is self-financing
and its wealth process satisfies Vt (φ) = Vt for every t ∈ [0, T ]. In particular, VT (φ) = X, so that φ
replicates X. Finally, equality (4.14) implies (4.8) and thus φ belongs to the class Φ(Z). ¤

Note that equality (4.13) is a necessary (by Lemma 4.2.1) and sufficient (by Proposition 4.2.2)
condition for the existence of a constrained strategy that replicates a given contingent claim X.

Synthetic Asset
P2
Let us take Z = 0 so that φ ∈ Φ(0). Then the balance condition becomes i=1 φit Yt−
i
= 0 and
formula (4.9) reduces to
à !
2,3
Y
dVt3 (φ) = φ2t dYt2,3 − t−
1,3 dYt
1,3
. (4.15)
Yt−

The process Ȳ 2 = Y 3 Y ∗ , where Y ∗ is defined in (4.12) is called a synthetic asset. It corresponds

2,1
to a particular self-financing portfolio, with the long position in Y 2 , the short position of Yt−
1
number of shares of Y , and suitably re-balanced positions in the third asset, so that the portfolio
is self-financing, as in Lemma 4.2.1.
It is not difficult to show (see Bielecki et al. [15, 17]) that trading in primary assets Y 1 , Y 2 , Y 3
is formally equivalent to trading in assets Y 1 , Ȳ 2 , Y 3 . This observation supports the name synthetic
asset attributed to the process Ȳ 2 . It is worth noting, however, that the synthetic asset process may
take negative values, so that it is unsuitable as a numéraire, in general.

Case of Continuous Asset Prices

In the case of continuous asset prices, the relative price Y ∗ = Ȳ 2 (Y 3 )−1 of the synthetic asset can be
given an alternative representation, as the following result shows. Recall that the predictable bracket
of the two continuous semimartingales X and Y , denoted as hX, Y i, coincides with their quadratic
covariation [X, Y ].

Proposition 4.2.3 Assume that the price processes Y 1 and Y 2 are continuous. Then the relative
price of the synthetic asset satisfies
Z t
Yt∗ = (Yu3,1 )−1 eαu dYbu ,
0

where we denote Ybt = Yt2,1 e−αt and

Z t
αt = hln Y 2,1 , ln Y 3,1 it = (Yu2,1 )−1 (Yu3,1 )−1 dhY 2,1 , Y 3,1 iu . (4.16)
0

In terms of the auxiliary process Yb , formula (4.9) becomes

Z t Z t
Zu3
Vt3 (φ) = V03 (φ) + φbu dYbu + 1,3
1,3 dYu ,
0 0 Yu−

where φbt = φ2t (Yt3,1 )−1 eαt .

Proof. It suffices to give the proof for Z = 0. The proof relies on the integration by parts formula
stating that we have, for any two continuous semimartingales, say X and Y ,
¡ ¢
Yt−1 dXt − Yt−1 dhX, Y it = d(Xt Yt−1 ) − Xt dYt−1 ,
140 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

provided that Y is strictly positive. By applying this formula to processes X = Y 2,1 and Y = Y 3,1 ,
we obtain
(Yt3,1 )−1 (dYt2,1 − (Yt3,1 )−1 dhY 2,1 , Y 3,1 it ) = d(Yt2,1 (Yt3,1 )−1 ) − Yt2,1 d(Y 3,1 )−1
t .

The relative wealth Vt3 (φ) = Vt (φ)(Yt3 )−1 of a strategy φ ∈ Φ(0) satisfies
Z t
Vt3 (φ) = V03 (φ) + φ2u dYu∗
0
Z t
3
= V0 (φ) + φ2u (Yu3,1 )−1 eαu dYbu ,
0
Z t
= V03 (φ) + φbu dYbu
0

where we denote φbt = φ2t (Yt3,1 )−1 eαt . ¤

Remark 4.2.1 The financial interpretation of the auxiliary process Yb will be studied below. Let
us only observe here that if Y ∗ is a local martingale under some probability Q then Yb is a Q-local
b
martingale (and vice versa, if Yb is a Q-local martingale under some probability Qb then Y ∗ is a
b
Q-local martingale). Nevertheless, for the reader’s convenience, we shall use two symbols Q and Q,b
since this equivalence holds for continuous processes only.

Remark 4.2.2 It is thus worth stressing that we will apply Proposition 4.2.3 to pre-default values of
assets, rather than directly to asset prices, within the setup of a semimartingale model with a common
default, as described in Section 4.1.1. In this model, the asset prices may have discontinuities, but
their pre-default values follow continuous processes.

4.3 Martingale Approach

Our goal is to derive quasi-explicit conditions for replicating strategies for a defaultable claim in a
fairly general setup introduced in Section 4.1.1. In this section, we only deal with trading strategies
based on the reference filtration F and the underlying price processes (that is, prices of default-free
assets and pre-default values of defaultable assets) are assumed to be continuous. Therefore, our
arguments will hinge on Proposition 4.2.3, rather than on a more general Proposition 4.2.1. We
shall also adapt Proposition 4.2.2 to our current purposes.
To simplify the presentation, we make the standing assumption that all coefficient processes are
such that the SDEs, which appear in what follows, admit unique strong solutions and all Doléans
exponentials (the Radon-Nikodým derivatives) are true martingales under respective probabilities.

4.3.1 Defaultable Asset with Zero Recovery

In this section, we shall examine in some detail a particular model where the two assets, Y 1 and Y 2 ,
are default-free and satisfy, for i = 1, 2,
¡ ¢
dYti = Yti µi,t dt + σi,t dWt ,
where W is a one-dimensional Brownian motion. The third asset is a defaultable asset with zero
recovery, so that ¡ ¢
dYt3 = Yt−
3
µ3,t dt + σ3,t dWt − dMt .
Since we will be interested in replicating strategies in the sense of Definition 4.1.2, we may and do
assume, without loss of generality, that the coefficients µi,t , σi,t , i = 1, 2, are F-predictable, rather
than G-predictable. Recall that, in general, there exist F-predictable processes µ e3 and σ
e3 such that
µ
e3,t 1{t≤τ } = µ3,t 1{t≤τ } , σ
e3,t 1{t≤τ } = σ3,t 1{t≤τ } .
4.3. MARTINGALE APPROACH 141

We assume throughout that Y0i > 0 for every i, so that the price processes Y 1 , Y 2 are strictly
positive and the process Y 3 is non-negative and has strictly positive pre-default value.

Default-Free Market

It is natural to postulate that the default-free market with two traded assets, Y 1 and Y 2 , is arbitrage-
free. To be more specific, we choose Y 1 as a numéraire and we require that there exists a probability
measure P1 , equivalent to P on (Ω, FT ), and such that the process Y 2,1 is a P1 -martingale. The
dynamics of processes (Y 1 )−1 and Y 2,1 are
¡ 2 ¢
d(Yt1 )−1 = (Yt1 )−1 (σ1,t − µ1,t ) dt − σ1,t dWt , (4.17)

and ¡ ¢
dYt2,1 = Yt2,1 (µ2,t − µ1,t + σ1,t (σ1,t − σ2,t )) dt + (σ2,t − σ1,t ) dWt ,
respectively. Hence the necessary condition for the existence of an EMM P1 is the inclusion A ⊆ B,
where A = {(t, ω) ∈ [0, T ] × Ω : σ1,t (ω) = σ2,t (ω)} and B = {(t, ω) ∈ [0, T ] × Ω : µ1,t (ω) = µ2,t (ω)}.
The necessary and sufficient condition for the existence and uniqueness of an EMM P1 reads
½ µZ · ¶¾
EP ET θu dWu =1 (4.18)
0

where the process θ is given by the formula, for every t ∈ [0, T ],

µ1,t − µ2,t
θt = σ1,t − , (4.19)
σ1,t − σ2,t

where, by convention, 0/0 = 0. Note that in the case of constant coefficients, if σ1 = σ2 then the
considered model is arbitrage-free only in the trivial case when µ2 = µ1 .

Remark 4.3.1 Since the martingale measure P1 is unique, the default-free model (Y 1 , Y 2 ) is com-
plete. However, this assumption is not necessary and thus it can be relaxed. As we shall see in
what follows, it is typically more natural to assume that the driving Brownian motion W is multi-
dimensional.

Arbitrage-Free Property

Let us now consider also a defaultable asset Y 3 . Our goal is now to find a martingale measure Q1 (if
it exists) for relative prices Y 2,1 and Y 3,1 . Recall that we postulate that the hypothesis (H) holds
under P for filtrations F and G = F ∨ H. The dynamics of Y 3,1 under P are
n o
3,1 ¡ ¢
dYt3,1 = Yt− µ3,t − µ1,t + σ1,t (σ1,t − σ3,t ) dt + (σ3,t − σ1,t ) dWt − dMt .

Let Q1 be any probability measure equivalent to P on (Ω, GT ) and let η be the associated Radon-
Nikodým density process, so that
dQ1 | Gt = ηt dP | Gt , (4.20)
where the process η is a G-martingale under P and satisfies

dηt = ηt− (θt dWt + ζt dMt ) (4.21)

for some G-predictable processes θ and ζ.

c and M
From Girsanov’s theorem (cf. Theorem 3.4.1), the processes W c, which are given by the
expressions Z t
ct = Wt −
W θu du (4.22)
0
142 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

and Z t∧τ
ct = Mt −
M γu ζu du, (4.23)
0

are G-martingales under Q1 .

To ensure that Y 2,1 is a Q1 -martingale, we postulate that conditions (4.18) and (4.19) are
satisfied. Consequently, for the process Y 3,1 to be a Q1 -martingale, it is necessary and sufficient
that a process ζ satisfies
µ1,t − µ2,t
γt ζt = µ3,t − µ1,t − (σ3,t − σ1,t ).
σ1,t − σ2,t

To ensure that Q1 is a probability measure equivalent to P, we require that the inequality ζt > −1
is valid. Then the unique martingale measure Q1 is given by formula (4.20) where η solves (4.21),
so that µZ ¶ ÃZ !
·
η t = Et θu dWu Et ζu dMu .
0 ]0, · ]

We are in a position to formulate the following result.

Proposition 4.3.1 Assume that the process θ given by (4.19) satisfies (4.18) and
µ ¶
1 µ1,t − µ2,t
ζt = µ3,t − µ1,t − (σ3,t − σ1,t ) > −1. (4.24)
γt σ1,t − σ2,t

Then the market model M = (Y 1 , Y 2 , Y 3 ; Φ) is arbitrage-free and complete. The dynamics of relative
prices under the unique martingale measure Q1 are
ct ,
dYt2,1 = Yt2,1 (σ2,t − σ1,t ) dW
¡ ¢
dYt3,1 = Y 3,1 (σ3,t − σ1,t ) dWct − dM
ct .
t−

Since the coefficients µi,t , σi,t , i = 1, 2, are F-adapted, the process Wc is an F-martingale (hence,
1
a Brownian motion) under Q . Therefore, by virtue of Proposition 3.5.1, hypothesis (H) holds under
Q1 , and the F-intensity of default under Q1 equals
µ ¶
µ1,t − µ2,t
γ
bt = γt (1 + ζt ) = γt + µ3,t − µ1,t − (σ3,t − σ1,t ) .
σ1,t − σ2,t

Example 4.3.1 We present an example where the condition (4.24) does not hold and thus arbitrage
opportunities arise. Assume that the coefficients are constant and satisfy µ1 = µ2 = σ1 = 0, µ3 < −γ
for a constant default intensity γ > 0. Then
µ ¶
1
Yt3 = 1{t<τ } Y03 exp σ3 Wt − σ32 t + (µ3 + γ)t
2
µ ¶
1
≤ Y03 exp σ3 Wt − σ32 t = Vt (φ),
2
σ3
where V (φ) represents the wealth of a self-financing strategy (φ1 , φ2 , 0) with φ2 = σ2 . Hence the
arbitrage strategy would be to sell the asset Y 3 and to follow the strategy φ.

Remark 4.3.2 Let us stress once again, that the existence of an EMM is a necessary condition for
the model viability, but the uniqueness of an EMM is not always a natural condition to be imposed.
In fact, when constructing a model, we should be mostly concerned with its flexibility and ability
to reflect the pertinent risk factors, rather than with its mathematical completeness. In the present
context, it would be natural to postulate that the dimension of the underlying Brownian motion
coincides with the number of traded risky assets.
4.3. MARTINGALE APPROACH 143

Hedging a Survival Claim

We first focus on replication of a survival claim (X, 0, τ ), that is, a defaultable claim represented by
the terminal payoff X1{T <τ } , where X is an FT -measurable random variable. For the moment, we
maintain the simplifying assumption that W is one-dimensional. As we shall see in what follows,
it may lead to certain pathological features of a model. If, on the contrary, the driving noise is
multi-dimensional, most of the analysis remains valid, except that the model completeness is no
longer ensured, in general.
Recall that Ye 3 stands for the pre-default price of Y 3 , defined as follows (see (4.3))
¡ ¢
dYet3 = Yet3 (e
µ3,t + γt ) dt + σ
e3,t dWt

with Ye03 = Y03 . This strictly positive, continuous, F-adapted process enjoys the property that Yt3 =
1{t<τ } Yet3 . Let us denote the pre-default values relative to the numéraire Ye 3 by Yeti,3 = Yti (Yet3 )−1 for
i = 1, 2 and let us introduce the pre-default relative price Ye ∗ of the synthetic asset Ȳ 2 by setting

Ye 2,3
dYet∗ = dYet2,3 − t1,3 dYet1,3
Ye
³¡ t ¢ ´
= Yet2,3 µ2,t − µ1,t + σ e3,t (σ1,t − σ2,t ) dt + (σ2,t − σ1,t ) dWt .

We postulate that σ1,t − σ2,t 6= 0. It is useful to note that the process Yb defined in Proposition 4.2.3
satisfies ³¡ ´
¢
dYbt = Ybt µ2,t − µ1,t + σ e3,t (σ1,t − σ2,t ) dt + (σ2,t − σ1,t ) dWt .

We will show that when α given by (4.16) is deterministic, the process Yb has the financial interpre-
tation as the credit-risk adjusted forward price of Y 2 relative to Y 1 . Therefore, it is more convenient
to work with the process Ye ∗ when dealing with the general case, but to use instead the process Yb
when analyzing a model with deterministic volatilities.
Consider an F-predictable self-financing strategy φ satisfying the balance condition φ1t Yt1 +
φ2t Yt2
= 0, and the corresponding wealth process

3
X
Vt (φ) := φit Yti = φ3t Yt3 .
i=1

Let us set Vet (φ) := φ3t Yet3 . Since the process Ve (φ) is F-adapted, it is rather clear that it represents the
pre-default price process of the portfolio φ, in the sense that the equality 1{t<τ } Vt (φ) = 1{t<τ } Vet (φ)
is valid for every t ∈ [0, T ]. We shall call the process Vet (φ) the pre-default wealth of φ. Consequently,
the process Vet3 (φ) := Vet (φ)(Yet3 )−1 = φ3t is termed the relative pre-default wealth.
Using Proposition 4.2.1, with suitably modified notation, we find that the F-adapted process
Ve 3 (φ) satisfies, for every t ∈ [0, T ],
Z t
Vet3 (φ) = Ve03 (φ) + φ2u dYeu∗ .
0

Define a new probability Q∗ on (Ω, FT ) by setting

dQ∗ = ηT∗ dP,

where dηt∗ = ηt∗ θt∗ dWt and

µ2,t − µ1,t + σe3,t (σ1,t − σ2,t )
θt∗ = . (4.25)
σ1,t − σ2,t
144 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

The process (Yet∗ , t ∈ [0, T ]) is a (local) martingale under Q∗ driven by a Brownian motion. We shall
require that this process is in fact a true martingale; a sufficient condition for this is that
Z T ³ ´2
EQ∗ Yet2,3 (σ2,t − σ1,t ) dt < ∞.
0

From the predictable representation theorem for the Brownian filtration, it follows that for any
random variable X ∈ FT , such that the random variable X(YeT3 )−1 is square-integrable under Q∗ ,
there exists a constant x and an F-predictable process φ2 such that
µ Z ¶
e 3
X = YT x + 2 e∗
φu dYu . (4.26)
]0,T ]

We now deduce from Proposition 4.2.2 that there exists a self-financing strategy φ with the pre-
default wealth Vet (φ) = Yet3 Vet3 for every t ∈ [0, T ], where we set
Z t
e 3
Vt = x + φ2u dYeu∗ . (4.27)
0

Moreover, the balance condition φ1t Yt1 + φ2t Yt2 = 0 is satisfied for every t ∈ [0, T ]. Since, clearly,
VeT (φ) = X, we have that

VT (φ) = φ3T YT3 = 1{T <τ } φ3T YeT3 = 1{T <τ } VeT (φ) = 1{T <τ } X.
We conclude that the strategy φ replicates the survival claim (X, 0, τ ). In particular, we have that
Vt (φ) = 0 on the random interval [[τ, T ∧ τ ]].

Definition 4.3.1 We say that a survival claim (X, 0, τ ) is attainable if the process Ve 3 given by
(4.27) is a martingale under Q∗ .

The following result is an immediate consequence of (4.26) and (4.27).

Corollary 4.3.1 Let X ∈ FT be such that X(YeT3 )−1 is square-integrable under Q∗ . Then the
survival claim (X, 0, τ ) is attainable. Moreover, the pre-default price π et (X, 0, τ ) of the survival
claim (X, 0, τ ) is given by the following conditional expectation, for every t ∈ [0, T ],

et (X, 0, τ ) = Yet3 EQ∗ (X(YeT3 )−1 | Ft ).

π (4.28)

e(X, 0, τ )(Ye 3 )−1 is an F-martingale under Q∗ .

The process π

Proof. Since X(YeT3 )−1 is square-integrable under Q∗ , we know from the predictable representation
theorem for the Brownian filtration that the process φ2 in formula (4.26) is such that
µZ T ¶
EQ∗ (φ2t )2 dhYe ∗ it < ∞.
0

Therefore, the process Ve 3 given by (4.27) is a true martingale under Q∗ . We conclude that the
survival claim (X, 0, τ ) is attainable.
Now, let us denote by πt (X, 0, τ ) the price at time t of the survival claim (X, 0, τ ). Since φ is
a replicating strategy for the claim (X, 0, τ ), we have that Vt (φ) = πt (X, 0, τ ) for every t ∈ [0, T ].
Consequently, for every t ∈ [0, T ],

et (X, 0, τ ) = 1{t<τ } Vet (φ) = 1{t<τ } Yet3 EQ∗ (VeT3 | Ft )

1{t<τ } π
= 1{t<τ } Ye 3 EQ∗ (X(Ye 3 )−1 | Ft ).
t T

This proves equality (4.28). ¤

In view of the last result, it is justified to refer to Q∗ as the pricing measure relative to Y 3 for
attainable survival claims.
4.3. MARTINGALE APPROACH 145

Remark 4.3.3 It can be proved that there exists a unique absolutely continuous probability mea-
sure Q̄ on (Ω, GT ) such that we have
µ ¶ µ ¶
1{T <τ } X ¯¯ e X ¯¯
Yt3 EQ̄ ¯ G t = 1 Y
{t<τ } t
3
EQ∗ ¯ F t .
YT3 YeT3

However, this probability measure is manifestly not equivalent to Q∗ , since its Radon-Nikodým
density process vanishes after τ (for a related result, see the paper by Collin-Dufresne et al. [51]).

Example 4.3.2 We provide here an explicit calculation of the pre-default price of a survival claim.
For simplicity, we assume that X = 1, so that the claim represents a defaultable zero-coupon bond.
Also, we set γt = γ = const, µi,t = 0, and σi,t = σi , i = 1, 2, 3. Straightforward calculations yield
the following pricing formula
1 2
e0 (1, 0, τ ) = Y03 e−(γ+ 2 σ3 )T .
π

We see that here the pre-default price π e0 (1, 0, τ ) depends explicitly on the intensity γ, or rather
on the drift term in dynamics of the pre-default value of a defaultable asset. Indeed, from the
practical viewpoint, the interpretation of the drift coefficient in dynamics of Y 2 as the real-world
default intensity is questionable, since, within the present setup, the default intensity never appears
as an independent variable; indeed, it is merely one component of the drift term in dynamics of the
pre-default value of Y 3 .
Note also that we deal here with a model in which three traded assets are driven by a common
one-dimensional Brownian motion. No wonder that this model enjoys the nice property of market
completeness, but, at the same time, it also exhibits an undesirable property that the pre-default
values of all three assets are perfectly correlated.
As we shall see later, if traded primary assets are judiciously chosen then, typically, the pre-
default price (and hence the price) of a survival claim will not depend explicitly on the default
intensity process.

Remark 4.3.4 From the practical perspective, it seems natural to consider a given market model
as an acceptable model if its implementation does not require estimation of drift parameters of
pre-default prices, at least for the purpose of hedging and valuation of a sufficiently large class of
defaultable contingent claims of interest. It is worth recalling that we do not postulate that the
drift coefficients are market observables. Since the default intensity can formally be interpreted as a
component of the drift term in dynamics of pre-default prices, in an acceptable model there should
be no need to estimate this quantity. From this perspective, the model considered in Example 4.3.2
may serve as an example of an ‘unacceptable’ model, since its implementation requires the knowledge
of the drift parameter in dynamics of Y 3 .
Let us stress that we do not claim that it is always possible to hedge derivative assets without
using the drift coefficients in dynamics of traded assets; we merely argue that one should strive to
develop market models in which this knowledge is not essential.

Hedging a Recovery Process

Let us now briefly study the situation where the promised payoff equals zero and the recovery payoff
is paid at time τ and equals Zτ for some F-adapted process Z. Put another way, we consider a
defaultable claim of the form (0, Z, τ ). Once again, we make use of Propositions 4.2.1 and 4.2.2. In
view of (4.13), we need to find a constant x and an F-predictable process φ2 such that
Z T Z T
Zt e 1,3
ψT := − dY =x+ φ2t dYet∗ .
0 Yt1 t 0
146 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

Similarly as before, we conclude that, under suitable integrability conditions on ψT , there exists φ2
such that dψt = φ2t dYt∗ , where ψt = EQ∗ (ψT | Ft ). We now set
Z t Z T e3
e 3 2 ∗ Zu e 1,3
Vt = x + φu dYu + dYu ,
0 e
0 Yu
1,3

so that, in particular, VeT3 = 0. Then it is possible to find processes φ1 and φ3 such that the strategy
φ is self-financing and it satisfies: Vet (φ) = Vet3 Yet3 and Vt (φ) = Zt + φ3t Yt3 for every t ∈ [0, T ]. It is
thus clear that Vτ (φ) = Zτ on the event {T ≥ τ } and VT (φ) = 0 on the event {T < τ }.

4.3.2 Hedging with a Defaultable Bond

Of course, an abstract semimartingale model considered until now furnishes only a generic framework
for a construction of acceptable models for hedging of default risk. A choice of traded assets and
specification of their dynamics need to be examined on a case-by-case basis, rather than in an abstract
semimartingale setup. We shall address these important issues by examining a few practically
appealing examples of defaultable claims and the corresponding models.
For the sake of concreteness, we postulate throughout this section that Yt1 = B(t, T ) is the price
of a default-free ZCB with maturity T , whereas Yt3 = D0 (t, T ) is the price of a defaultable ZCB
with zero recovery, that is, a defaultable asset with the terminal payoff YT3 = 1{T <τ } at maturity T .
We postulate that the dynamics under P of the default-free ZCB are
¡ ¢
dB(t, T ) = B(t, T ) µ(t, T ) dt + b(t, T ) dWt

for some F-predictable processes µ(t, T ) and b(t, T ) and we select the process Yt1 = B(t, T ) as
a numéraire. Since the prices of the other two assets are not given a priori, we may take any
probability measure Q equivalent to P on (Ω, GT ) to play the role of Q1 .
In such a case, the probability measure Q1 is commonly referred to as the forward martingale
measure for the date T and is denoted by QT . Hence the Radon-Nikodým density of QT with respect
to P is given by (4.21) for some F-predictable processes θ and ζ, and the process
Z t
WtT = Wt − θu du, ∀ t ∈ [0, T ],
0

is a Brownian motion under QT . Under QT the default-free ZCB is governed by

¡ ¢
dB(t, T ) = B(t, T ) µb(t, T ) dt + b(t, T ) dWtT

where µ
b(t, T ) = µ(t, T ) + θt b(t, T ).
Let now Γ b stand for the F-hazard process of default time τ under QT , so that Γ b t = − ln(1 − Fbt ),
where Fbt = QT (τ ≤ t | Ft ). Assume that hypothesis (H) is valid under QT so that, in particular,
b is increasing. We define the price process of the defaultable ZCB with zero recovery
the process Γ
by the formula
¡ b b ¯ ¢
D0 (t, T ) := B(t, T ) EQT (1{T <τ } | Gt ) = 1{t<τ } B(t, T ) EQT eΓt −ΓT ¯ Ft .

It is then easily seen that Yt3,1 = D0 (t, T )(B(t, T ))−1 is a QT -martingale and the pre-default
price De 0 (t, T ) equals
¡ ¯ ¢
De 0 (t, T ) = B(t, T ) EQ eΓbt −ΓbT ¯ Ft .
T

The next result examines the basic properties of the auxiliary process Γ(t, b T ), which is given as,
for every t ∈ [0, T ],
¡ ¯ ¢
e 0 (t, T )(B(t, T ))−1 = EQ eΓbt −ΓbT ¯ Ft .
b T ) = Yet3,1 = D
Γ(t, T
4.3. MARTINGALE APPROACH 147

b T ) can be interpreted as the conditional probability under QT that default will

The quantity Γ(t,
not occur prior to the maturity date T , given that we observe Ft and we know that the default has
not yet happened. We will be interested in its volatility process β(t, T ), which is implicitly defined
by the following result.

Lemma 4.3.1 Assume that the F-hazard process Γ b of τ under QT is continuous. Then the process
b
Γ(t, T ), t ∈ [0, T ], is a continuous F-submartingale and
¡ ¢
b T ) = Γ(t,
dΓ(t, b T ) dΓ
b t + β(t, T ) dW T (4.29)
t

b T ) is of finite variation if and only if the

for some F-predictable process β(t, T ). The process Γ(t,
b T ) = eΓbt −ΓbT .
b is deterministic. In this case, we have Γ(t,
hazard process Γ

Proof. We have
¡ ¢
b T ) = EQ eΓbt −ΓbT | Ft = eΓbt Lt ,
Γ(t, T

¡ b ¢
where we set Lt = EQT e−ΓT | Ft . Hence Γ(t, b T ) is equal to the product of a strictly positive,
b
increasing, right-continuous, F-adapted process eΓt and a strictly positive, continuous F-martingale
b T ) such that L satisfies
L. Furthermore, there exists an F-predictable process β(t,

b T ) dW T
dLt = Lt β(t, t

¡ b ¢
with the initial condition L0 = EQT e−ΓT . Formula (4.29) now follows by an application of Itô’s
b b
formula and by setting β(t, T ) = e−Γt β(t, T ). To complete the proof, it suffices to recall that a
continuous martingale is never a process of finite variation, unless it is a constant process. ¤

Remark 4.3.5 It can be checked that β(t, T ) is also the volatility of the process
¡ ¯ ¢
Γ(t, T ) = EP eΓt −ΓT ¯ Ft .

R
Assume that Γbt = t γ
b du for some F-predictable, non-negative default intensity process γ b under
0 u
QT . Then we have the following auxiliary result, which yields, in particular, the volatility process
of the defaultable ZCB.

e 0 (t, T ) are
Corollary 4.3.2 The dynamics under QT of the pre-default price D
¡ ¢
e 0 (t, T ) = D
dD e 0 (t, T ) µb(t, T ) + b(t, T )β(t, T ) + γ
bt dt
¡ ¢
+D e T ) dW T .
e 0 (t, T ) b(t, T ) + β(t, T ) d(t, t

Equivalently, the price D0 (t, T ) of the defaultable ZCB satisfies under QT

³¡ ¢ ´
dD0 (t, T ) = D0 (t, T ) µ e T ) dW T − dMt .
b(t, T ) + b(t, T )β(t, T ) dt + d(t, t

e T ) = b(t, T ) + β(t, T ).
where we denote d(t,

It is worth noting that the process β(t, T ) can be expressed in terms of market observables, in
e T ) and b(t, T ) of pre-default
the sense, that it can be represented as the difference of volatilities d(t,
prices of traded assets.
148 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

Credit-Risk-Adjusted Forward Price

Assume that the price Y 2 satisfies under the statistical probability P

¡ ¢
dYt2 = Yt2 µ2,t dt + σt dWt (4.30)

with F-predictable coefficients µ and σ. Let FY 2 (t, T ) = Yt2 (B(t, T ))−1 be the forward price of YT2 .
For an appropriate choice of θ (see 4.25), we shall have that
¡ ¢
dFY 2 (t, T ) = FY 2 (t, T ) σt − b(t, T ) dWtT .

Therefore, the dynamics of the pre-default synthetic asset Yet∗ under QT are
¡ ¢¡ ¢
dYet∗ = Yet2,3 σt − b(t, T ) dWtT − β(t, T ) dt ,

and the process Ybt = Yt2,1 e−αt (see Proposition 4.2.3 for the definition of α) satisfies
¡ ¢¡ ¢
dYbt = Ybt σt − b(t, T ) dWtT − β(t, T ) dt .

b be an equivalent probability measure on (Ω, GT ) such that Yb (or, equivalently, Ye ∗ ) is a

Let Q
b
Q-martingale. By virtue of Girsanov’s theorem, the process W c given by the formula, for t ∈ [0, T ],
Z t
c T
Wt = Wt − β(u, T ) du,
0

is a Brownian motion under Q.b Thus, the forward price FY 2 (t, T ) satisfies under Qb
¡ ¢¡ ¢
dFY 2 (t, T ) = FY 2 (t, T ) σt − b(t, T ) dWct + β(t, T ) dt . (4.31)

It appears that the valuation results are easier to interpret when they are expressed in terms
of forward prices associated with vulnerable forward contracts, rather than in terms of spot prices
of primary assets. For this reason, we shall now examine credit-risk-adjusted forward prices of
default-free and defaultable assets.

Definition 4.3.2 Let Y be a GT -measurable claim. An Ft -measurable random variable K is called

the credit-risk-adjusted forward price of Y if the pre-default value at time t of the vulnerable forward
contract represented by the claim 1{T <τ } (Y − K) equals 0.

Lemma 4.3.2 The credit-risk-adjusted forward price FbY (t, T ) of an attainable survival claim (X, 0, τ ),
which is represented by a GT -measurable claim Y = X1{T <τ } , equals π et (X, 0, τ )(De 0 (t, T ))−1 , where
et (X, 0, τ ) is the pre-default price of (X, 0, τ ). The process FbY (t, T ), t ∈ [0, T ], is an F-martingale
π
under Q.b

Proof. The forward price is defined as an Ft -measurable random variable K such that the claim

1{T <τ } (X1{T <τ } − K) = X1{T <τ } − KD0 (T, T )

is worthless at time t on the event {t < τ }. It is clear that the pre-default value at time t of this
claim equals π e 0 (t, T ). Consequently, we obtain FeY (t, T ) = π
et (X, 0, τ ) − K D e 0 (t, T ))−1 .
et (X, 0, τ )(D
¤

Let us now focus on default-free assets. It is clear that the credit-risk-adjusted forward price of
the bond B(t, T ) equals 1. To find the credit-risk-adjusted forward price of Y 2 , let us write

FbY 2 (t, T ) := FY 2 (t, T ) eαT −αt = Yt2,1 eαT −αt , (4.32)

4.3. MARTINGALE APPROACH 149

where α is given by (see (4.16))

Z t ¡ ¢
αt = σu − b(u, T ) β(u, T ) du (4.33)
0
Z t ¡ ¢¡ ¢
= e T ) − b(u, T ) du.
σu − b(u, T ) d(u,
0

Lemma 4.3.3 Assume that α given by (4.33) is a deterministic function. Then the credit-risk-
adjusted forward price of Y 2 , denoted as FbY 2 (t, T ), is given by (4.32) for every t ∈ [0, T ].

Proof. According to Definition 4.3.2, the price FbY 2 (t, T ) is an Ft -measurable random variable K,
which makes the forward contract represented by the claim D0 (T, T )(YT2 − K) worthless on the set
{t < τ }. Assume that the claim YT2 − K is attainable. Since D e 0 (T, T ) = 1, from equation (4.28) it
follows that the pre-default value of this claim is given by the conditional expectation
¡ ¯ ¢
De 0 (t, T ) E b Y 2 − K ¯ Ft .
Q T

Consequently,
¡ ¯ ¢ ¡ ¯ ¢
FbY 2 (t, T ) = EQb YT2 ¯ Ft = EQb FY 2 (T, T ) ¯ Ft = FY 2 (t, T ) eαT −αt ,

as was claimed. ¤

It is worth noting that the process FbY 2 (t, T ) is a (local) martingale under the pricing measure
b
Q, since it satisfies
dFbY 2 (t, T ) = FbY 2 (t, T )(σt − b(t, T )) dW
ct . (4.34)
Under the present assumptions, the auxiliary process Yb introduced in Proposition 4.2.3 and the
credit-risk-adjusted forward price FbY 2 (t, T ) are closely related to each other. Indeed, we have
FbY 2 (t, T ) = Ybt eαT , so that the two processes are proportional.

Vulnerable Option on a Default-Free Asset

We shall now analyze a vulnerable call option with the payoff

CTd = 1{T <τ } (YT2 − K)+

for a constant strike K. Our goal is to find a replicating strategy for this claim, which is interpreted
as a survival claim (X, 0, τ ) with the promised payoff X = CT = (YT2 − K)+ , where CT is the payoff
of an equivalent non-vulnerable option. The method presented below is quite general, however, so
that it can be applied to any survival claim with the promised payoff X = G(YT2 ) for some function
G : R → R satisfying mild integrability assumptions.
We assume that Yt1 = B(t, T ), Yt3 = D0 (t, T ) and the price of a default-free asset Y 2 is governed
by (4.30). Then
CTd = 1{T <τ } (YT2 − K)+ = 1{T <τ } (YT2 − KYT1 )+ .

We are going to apply Proposition 4.2.3. In the present setup, we have Yt2,1 = FY 2 (t, T ) and
Ybt = FY 2 (t, T )e−αt . Since a vulnerable option is an example of a survival claim, in view of Lemma
4.3.2, its credit-risk-adjusted forward price satisfies FbC d (t, T ) = C
etd (D
e 0 (t, T ))−1 .

Proposition 4.3.2 Suppose that the volatilities σ, b and β are deterministic functions. Then the
credit-risk-adjusted forward price of a vulnerable call option written on a default-free asset Y 2 equals

FbC d (t, T ) = FbY 2 (t, T )N (d+ (FbY 2 (t, T ), t, T )) − KN (d− (FbY 2 (t, T ), t, T ))
150 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

where
ln z − ln K ± 12 v 2 (t, T )
d± (z, t, T ) =
v(t, T )
and Z T
2
v (t, T ) = (σu − b(u, T ))2 du.
t

The replicating strategy φ in the spot market satisfies,for every t ∈ [0, T ] on the event {t < τ },

φ1t B(t, T ) = −φ2t Yt2 ,

φ2t = De 0 (t, T )(B(t, T ))−1 N (d+ (t, T ))eαT −αt ,
e 0 (t, T ) = C
φ3t D etd ,

where d+ (t, T ) = d+ (FbY 2 (t, T ), t, T ).

Proof. In the first step, we establish the valuation formula. Assume for the moment that the option
is attainable. Then the pre-default value of the option equals, for every t ∈ [0, T ],
¡ ¯ ¢
etd = D
C e 0 (t, T ) E b (FY 2 (T, T ) − K)+ ¯ Ft
Q
¡ ¯ ¢
=D e 0 (t, T ) E b (FbY 2 (T, T ) − K)+ ¯ Ft .
Q

In view of (4.34), the conditional expectation above can be computed explicitly, yielding the claimed
valuation formula.
To find the replicating strategy and establish attainability of the option, we consider the Itô
differential dFbC d (t, T ) and we identify terms in (4.27). It appears that

dFbC d (t, T ) = N (d+ (t, T )) dFbY 2 (t, T ) = N (d+ (t, T ))eαT dYbt (4.35)
= N (d+ (t, T ))Yet3,1 eαT −αt dYet∗ ,

so that the process φ2 in (4.26) equals

φ2t = Yet3,1 N (d+ (t, T ))eαT −αt .

etd (D
Moreover, φ1 is such that φ1t B(t, T ) + φ2t Yt2 = 0 and φ3t = C e 0 (t, T ))−1 . It is easily seen that this
proves also the attainability of the option. ¤

Let us examine the financial interpretation of the last result.

First, equality (4.35) shows that it is easy to replicate the option using vulnerable forward
contracts. Indeed, we have
ed Z T
C
FbC d (T, T ) = X = 0
+ N (d+ (t, T )) dFbY 2 (t, T )
e 0 (0, T )
D 0

so that it is enough to invest the premium C e d = C d in defaultable ZCBs of maturity T and take, at
0 0
any instant t prior to default, N (d+ (t, T )) positions in vulnerable forward contracts. It is apparent
that if default occurs prior to T , all outstanding vulnerable forward contracts become void.
Second, it is worth stressing that neither the arbitrage price, nor the replicating strategy for a
vulnerable option, depend explicitly on the default intensity. This remarkable feature is due to the
fact that the default risk of the writer of the option can be completely eliminated by trading in
defaultable zero-coupon bond with the same exposure to credit risk as a vulnerable option.
In fact, since the volatility β is invariant with respect to an equivalent change of a probability
measure, and so are the volatilities σ and b(t, T ), the formulae of Proposition 4.3.2 are valid for any
choice of a forward measure QT equivalent to P (and, of course, they are valid under P as well).
4.3. MARTINGALE APPROACH 151

The only way in which the choice of a forward measure QT impacts these results is through the
pre-default value of a defaultable ZCB.
We conclude that we deal here with the volatility based relative pricing a defaultable claim. This
should be contrasted with more popular intensity-based risk-neutral pricing, which is commonly
used to produce an arbitrage-free model of traded defaultable assets. Recall, however, that if traded
assets are not chosen carefully for a given class of survival claims, then both hedging strategy and
pre-default price may depend explicitly on values of drift parameters that appear in our market
model and which, in turn, can be linked to the default intensity (see Example 4.3.2).

Remark 4.3.6 Assume that the promised payoff X = G(YT2 ) for some function G : R → R. The
pricing formula of Proposition 4.3.2 leads to the conjecture that the credit-risk-adjusted forward
price FbY (t, T ) of the survival claim Y = 1{T <τ } G(YT2 ) satisfies the equality

FbY (t, T ) = w(t, FbY 2 (t, T )),

where the pricing function w solves the PDE

∂t w(t, z) + 12 (σt − b(t, T ))2 z 2 ∂zz w(t, z) = 0

with the terminal condition w(T, z) = G(z). Let us mention that the PDE approach is studied in
some detail in Section 4.4 below.

Remark 4.3.7 Proposition 4.3.2 is still valid if the driving Brownian motion is two-dimensional,
rather than one-dimensional. In an extended model, the volatilities σt , b(t, T ) and β(t, T ) take values
in R2 and the respective products are interpreted as inner products in R3 . Equivalently, one may
prefer to deal with real-valued volatilities, but with correlated one-dimensional Brownian motions.

Abstract Vulnerable Swaption

In this section, we relax the assumption that Y 1 is the price of a default-free bond. We now let Y 1
and Y 2 to be arbitrary default-free assets, with dynamics
¡ ¢
dYti = Yti µi,t dt + σi,t dWt , i = 1, 2. (4.36)

We still take the defaultable zero-coupon bond with zero recovery and the price process Yt3 =
D0 (t, T ) to be the third traded asset.
We maintain the assumption that the model is arbitrage-free, but we no longer postulate that
it is complete. In other words, we postulate the existence an EMM Q1 , as defined in subsection on
the arbitrage-free property, but not the uniqueness of Q1 .
We take the first asset as the numéraire, so that all prices are expressed in units of Y 1 . In
particular, Yt1,1 = 1 for every t ∈ R+ , and the relative prices Y 2,1 and Y 3,1 satisfy under Q1 (cf.
Proposition 4.3.1)

ct ,
dYt2,1 = Yt2,1 (σ2,t − σ1,t ) dW
¡ ¢
dYt3,1 = Y 3,1 (σ3,t − σ1,t ) dWct − dM
ct .
t−

It is natural to postulate that the driving Brownian noise is two-dimensional. In such a case, we
may represent the joint dynamics of relative prices Y 2,1 and Y 3,1 under Q1 as follows

dYt2,1 = Yt2,1 (σ2,t − σ1,t ) dWt1 ,

3,1 ¡ ¢
ct ,
dYt3,1 = Yt− (σ3,t − σ1,t ) dWt2 − dM

where W 1 , W 2 are one-dimensional Brownian motions under Q1 , such that dhW 1 , W 2 it = ρt dt for
a deterministic instantaneous correlation coefficient ρ taking values in [−1, 1].
152 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

We assume from now on that the volatilities σi , i = 1, 2, 3 are deterministic. Let us set
Z t
αt = hln Ye 2,1 , ln Ye 3,1 it = ρu (σ2,u − σ1,u )(σ3,u − σ1,u ) du, (4.37)
0

and let Qb be an equivalent probability measure on (Ω, GT ) such that the process Ybt = Y 2,1 e−αt
t
b
is a Q-martingale. To clarify the financial interpretation of the auxiliary process Yb in the present
context, we introduce the concept of credit-risk-adjusted forward price relative to the numéraire Y 1 .

Definition 4.3.3 Let Y be a GT -measurable claim. An Ft -measurable random variable K is called

the time-t credit-risk-adjusted Y 1 -forward price of Y if the pre-default value at time t of a vulnerable
forward contract, represented by the claim

1{T <τ } (YT1 )−1 (Y − KYT1 ) = 1{T <τ } (Y (YT1 )−1 − K),

equals 0.

The credit-risk-adjusted Y 1 -forward price of Y is denoted by FbY |Y 1 (t, T ) and it is also interpreted
as an abstract defaultable swap rate. The following auxiliary results are easy to establish, by arguing
along the same lines as in Lemmas 4.3.2 and 4.3.3.

Lemma 4.3.4 The credit-risk-adjusted Y 1 -forward price of a survival claim Y = (X, 0, τ ) equals

FbY |Y 1 (t, T ) = π e 0 (t, T ))−1 ,

et (X 1 , 0, τ )(D

where X 1 = X(YT1 )−1 is the price of X in the numéraire Y 1 and π

et (X 1 , 0, τ ) is the pre-default value
1
of a survival claim with the promised payoff X .

Proof. It suffices to note that for Y = 1{T <τ } X we have

1{T <τ } (Y (YT1 )−1 − K) = 1{T <τ } X 1 − KD0 (T, T ),

where X 1 = X(YT1 )−1 , and to consider the pre-default values. ¤

Lemma 4.3.5 The credit-risk-adjusted Y 1 -forward price of the asset Y 2 equals

FbY 2 |Y 1 (t, T ) = Yt2,1 eαT −αt = Ybt eαT ,

where α, assumed here to be deterministic, is given by formula (4.37).

Proof. It suffices to find an Ft -measurable random variable K for which

¡ ¯ ¢
De 0 (t, T ) E b YT2 (YT1 )−1 − K ¯ Ft = 0.
Q

From the last equality, we obtain K = FbY 2 |Y 1 (t, T ), where

¡ ¯ ¢
FbY 2 |Y 1 (t, T ) = EQb YT2,1 ¯ Ft = Yt2,1 eαT −αt = Ybt eαT .

b
We have used here the facts that Ybt = Yt2,1 e−αt is a Q-martingale and α is deterministic. ¤

We are in a position to examine a vulnerable option to exchange default-free assets with the
payoff
CTd = 1{T <τ } (YT1 )−1 (YT2 − KYT1 )+ = 1{T <τ } (YT2,1 − K)+ . (4.38)
The last expression shows that the option can be interpreted as a vulnerable swaption associated
with the assets Y 1 and Y 2 . It is useful to observe that
µ ¶+
CTd 1{T <τ } YT2
= −K ,
YT1 YT1 YT1
4.3. MARTINGALE APPROACH 153

so that, when expressed in units of the numéraire Y 1 , the payoff becomes

CTd,1 = D0,1 (T, T )(YT2,1 − K)+ ,
where Ctd,1 = Ctd (Yt1 )−1 and D0,1 (t, T ) = D0 (t, T )(Yt1 )−1 stand for the prices relative to the
numéraire Y 1 .
It is clear that we deal here with a model analogous to the model examined in previous subsections
in which, however, all prices are expressed in units of the numéraire asset Y 1 . This observation allows
us to directly deduce the valuation formula from Proposition 4.3.2.

Proposition 4.3.3 Let us consider the market model (4.36) with a two-dimensional Brownian mo-
tion W and deterministic volatilities σi , i = 1, 2, 3. The credit-risk-adjusted Y 1 -forward price of a
vulnerable call option, with the terminal payoff given by (4.38), equals
¡ ¢ ¡ ¢
FbC d |Y 1 (t, T ) = Fbt N d+ (Fbt , t, T ) − KN d− (Fbt , t, T ) ,

where we write Fbt = FbY 2 |Y 1 (t, T ) and

ln z − ln K ± 12 v 2 (t, T )
d± (z, t, T ) =
v(t, T )
with Z T
2
v (t, T ) = (σ2,u − σ1,u )2 du.
t
The replicating strategy φ in the spot market satisfies, on the event {t < τ },
φ1t Yt1 = −φ2t Yt2 , φ2t = D e 0 (t, T ) = C
e 0 (t, T )(Yt1 )−1 N (d+ (t, T ))eαT −αt , φ3t D etd ,
¡ ¢
where d+ (t, T ) = d+ FbY 2 |Y 1 (t, T ), t, T .

Proof. The proof is analogous to that of Proposition 4.3.2 and thus it is omitted. ¤

It is worth noting that the payoff (4.38) was judiciously chosen. Suppose instead that the option
payoff is not defined by (4.38), but it is given by an apparently simpler expression
CTd = 1{T <τ } (YT2 − KYT1 )+ .
Since the payoff CTd can be represented as follows
b T1 , YT2 , YT3 ) = YT3 (YT2 − KYT1 )+ ,
CTd = G(Y
b 1 , y2 , y3 ) = y3 (y2 − Ky1 )+ , we deal with an option to exchange the second asset for K
where G(y
units of the first asset, but with the payoff expressed in units of the defaultable asset Y 3 . When
expressed in relative prices, the payoff becomes
CTd,1 = 1{T <τ } (YT2,1 − K)+ .
where 1{T <τ } = D0,1 (T, T )YT1 . It is thus rather clear that it is not longer possible to apply the same
method as in the proof of Proposition 4.3.2.

4.3.3 Defaultable Asset with Non-Zero Recovery

In this section, we still postulate that Y 1 and Y 2 are default-free assets with price processes
¡ ¢
dYti = Yti µi,t dt + σi,t dWt ,
where W is a one-dimensional Brownian motion. We now assume, however, that
dYt3 = Yt−
3
(µ3 dt + σ3 dWt + κ3 dMt )
with κ3 > −1 and κ3 6= 0. We assume that Y03 > 0, so that Yt3 > 0 for every t ∈ R+ . We shall
briefly describe the same steps as in the case of a defaultable asset with zero recovery.
154 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

Arbitrage-Free Property

As usual, we need first to impose specific constraints on model coefficients, so that the model is
arbitrage-free. In general, an EMM Q1 exists if there exists a pair (θ, ζ) such that, for i = 2, 3,
κi − κ1 κ1
θt (σi − σ1 ) + ζt ξt = µ1 − µi + σ1 (σi − σ1 ) + ξt (κi − κ1 ) .
1 + κ1 1 + κ1

To ensure the existence of a solution (θ, ζ) on the event {τ < t} under the present assumptions, we
impose the condition
µ1 − µ2 µ1 − µ3
σ1 − = σ1 − ,
σ1 − σ2 σ1 − σ3
that is,
µ1 (σ3 − σ2 ) + µ2 (σ1 − σ3 ) + µ3 (σ2 − σ1 ) = 0.
Since κ1 = κ2 = 0, on the event {τ ≥ t}, we have to solve the following equations

θt (σ2 − σ1 ) = µ1 − µ2 + σ1 (σ2 − σ1 ),
θt (σ3 − σ1 ) + ζt γκ3 = µ1 − µ3 + σ1 (σ3 − σ1 ).

If, in addition, (σ2 − σ1 )κ3 6= 0, we obtain the unique solution

µ1 − µ2 µ1 − µ3
θ = σ1 − = σ1 − ,
σ1 − σ2 σ1 − σ3
ζ = 0 > −1,

so that the martingale measure Q1 exists and is unique.

Observe that, since ζ = 0, the default intensity under Q1 coincides here with the default intensity
under the real-life probability Q. It is interesting to note that, in a more general situation when all
three assets are defaultable with non-zero recovery, the default intensity under Q1 coincides with
the default intensity under the real-life probability Q if and only if the process Y 1 is continuous.
For more details, the interested reader is referred to Bielecki at al. [14] where the general case is
studied.

4.3.4 Two Defaultable Assets with Zero Recovery

We shall now assume that we have only two assets and both are defaultable assets with zero recovery.
This case was recently examined by Carr [44], who studied an imperfect hedging of digital options.
Note that here we present results for replication, that is, perfect hedging.
We shall briefly outline the analysis of hedging of a survival claim. Under the present assumptions,
we have, for i = 1, 2, ¡ ¢
dYti = Yt−i
µi,t dt + σi,t dWt − dMt , (4.39)
where W is a one-dimensional Brownian motion, so that

Yt1 = 1{t<τ } Yet1 , Yt2 = 1{t<τ } Yet2 ,

with the pre-default prices governed by the SDEs

¡ ¢
dYeti = Yeti (µi,t + γt ) dt + σi,t dWt . (4.40)

The wealth process V associated with the self-financing trading strategy (φ1 , φ2 ) satisfies, for every
t ∈ [0, T ],
µ Z t ¶
1 1 2 e 2,1
Vt = Yt V0 + φu dYu ,
0
4.3. MARTINGALE APPROACH 155

where Yet2,1 = Yet2 /Yet1 . Since both primary traded assets are subject to zero recovery, it is clear that
the present model is incomplete, in the sense, that not all defaultable claims can be replicated.
We shall check in what follows that, under the assumption that the driving Brownian motion W
is one-dimensional, all survival claims satisfying mild technical conditions are hedgeable, however.
In the more realistic case of a two-dimensional noise, we will still be able to hedge a large class of
survival claims, including options on a defaultable asset and options to exchange defaultable assets.

Hedging a Survival Claim

For the sake of expositional simplicity, we assume in this subsection that the driving Brownian
motion W is one-dimensional. Arguably, this is not the right choice, since we deal here with two
risky assets, so that they will be perfectly correlated. However, this assumption is convenient for
the expositional purposes, since it ensures the model completeness with respect to survival claims.
We will later relax this temporary assumption so it is fair to say that this assumption is not crucial.
We shall now argue that in a market model with two defaultable assets that are subject to
zero recovery, the replication of a survival claim (X, 0, τ ) is in fact equivalent to replication of an
associated promised payoff X using the pre-default price processes.

Lemma 4.3.6 If a trading strategy φi , i = 1, 2, based on pre-default values Ye i , i = 1, 2, is a repli-

cating strategy for an FT -measurable claim X, that is, if φ is such that the process

Vet (φ) = φ1t Yet1 + φ2t Yet2

satisfies, for every t ∈ [0, T ],

dVet (φ) = φ1t dYet1 + φ2t dYet2 ,

VeT (φ) = X,

then for the process Vt (φ) = φ1t Yt1 + φ2t Yt2 we have, for every t ∈ [0, T ],

dVt (φ) = φ1t dYt1 + φ2t dYt2 ,

VT (φ) = 1{T <τ } X.

This means that the strategy φ replicates the survival claim (X, 0, τ ).

Proof. It is clear that Vt (φ) = 1{t<τ } Vt (φ) = 1{t<τ } Vet (φ). From the equality

φ1t dYt1 + φ2t dYt2 = −(φ1t Yet1 + φ2t Yet2 ) dHt + (1 − Ht− )(φ1t dYet1 + φ2t dYet2 ),

it follows that
φ1t dYt1 + φ2t dYt2 = −Vet (φ) dHt + (1 − Ht− )dVet (φ),
that is,
φ1t dYt1 + φ2t dYt2 = d(1{t<τ } Vet (φ)) = dVt (φ).
It is also easily seen that the equality VT (φ) = X1{T <τ } holds. ¤
1 2
Combining the last result with Lemma 4.2.1, we see that a strategy (φ , φ ) replicates a survival
claim (X, 0, τ ) whenever we have
³ Z T ´
e 1
YT x + φ2t dYet2,1 = X
0

for some constant x and some F-predictable process φ2 , where, in view of (4.40),
³¡ ¢ ´
dYet2,1 = Yet2,1 µ2,t − µ1,t + σ1,t (σ1,t − σ2,t ) dt + (σ2,t − σ1,t ) dWt .
156 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

We introduce a probability measure Q, e equivalent to P on (Ω, GT ), and such that Ye 2,1 is an F-

e
martingale under Q. It is easily seen that the Radon-Nikodým density η satisfies, for t ∈ [0, T ],
µZ · ¶
e
dQ | Gt = ηt dP | Gt = Et θs dWs dP | Gt
0

with
µ2,t − µ1,t + σ1,t (σ1,t − σ2,t )
θt = ,
σ1,t − σ2,t
provided, of course, that the process θ is well defined and satisfies suitable integrability conditions.
e
We shall show that a survival claim is attainable if the random variable X(YeT1 )−1 is Q-integrable.
Indeed, the pre-default value Vet at time t of a survival claim equals
¡ ¢
Vet = Yet1 EQe X(YeT1 )−1 | Ft

and, from the predictable representation theorem, we deduce that there exists a process φ2 such
that Z t
¡ ¢ ¡ ¢
e 1 −1 e
E e X(YT ) | Ft = E e X(YT ) 1 −1
+ φ2u dYeu2,1 .
Q Q
0

The component φ of the self-financing trading strategy φ = (φ1 , φ2 ) is then chosen in such a way
1

that, for every t ∈ [0, T ],

φ1t Yet1 + φ2t Yet2 = Vet .
To conclude, by focusing on pre-default values, we have shown that the replication of survival claims
can be reduced here to classic results on replication of (non-defaultable) contingent claims in a
default-free market model.

Option on a Defaultable Asset

In order to get a complete model with respect to survival claims, we postulated in the preceding
subsection that the driving Brownian motion in dynamics (4.39) is one-dimensional. This assumption
is questionable, since it clearly implies the perfect correlation between risky assets. However, we may
relax this restriction and work instead with the two correlated one-dimensional Brownian motions.
The model will no longer be complete, but options on a defaultable asset will still be attainable.
The payoff of a (non-vulnerable) call option written on the defaultable asset Y 2 equals

CT = (YT2 − K)+ = 1{T <τ } (YeT2 − K)+ ,

so that it is natural to interpret this contract as a survival claim with the promised payoff X =
(YeT2 − K)+ .
To deal with this option in an efficient way, we consider a model in which
¡ ¢
dYti = Yt−i
µi,t dt + σi,t dWti − dMt ,

where W 1 and W 2 are two one-dimensional correlated Brownian motions with the instantaneous
e 0 (t, T ) repre-
correlation coefficient ρt . More specifically, we assume that Yt1 = D0 (t, T ) = 1{t<τ } D
sents a defaultable ZCB with zero recovery, and Yt2 = 1{t<τ } Yet2 is a generic defaultable asset with
zero recovery. Within the present setup, the payoff can also be represented as follows

CT = (YT2 − KYT1 )+ = g(YT1 , YT2 ),

where g(y1 , y2 ) = (y2 − Ky1 )+ , and thus it can also be seen as an option to exchange the second
asset for K units of the first asset.
4.4. PDE APPROACH 157

e implies that
The requirement that the process Yet2,1 = Yet2 (Yet1 )−1 is an F-martingale under Q
³¡ ¢ q ´
dYet2,1 = Yet2,1 σ2,t ρt − σ1,t dW ft1 + σ2,t 1 − ρ2t dW
ft2 ,

where Wf = (W f1, W
f 2 ) follows a two-dimensional Brownian motion under Q. e Since Ye 1 = 1, a
T
replication of the option reduces to finding a constant x and an F-predictable process φ2 satisfying
Z T
x+ φ2t dYet2,1 = (YeT2 − K)+ .
0
To obtain closed-form expressions for the option price and replicating strategy, we postulate that
the volatilities σ1 , σ2 and the correlation coefficient ρ are deterministic. Let
FbY 2 (t, T ) = Yet2 (D
e 0 (t, T ))−1
and
FbC (t, T ) = C
et (D
e 0 (t, T ))−1
stand for the credit-risk-adjusted forward price of the second asset and of the option, respectively.
The proof of the following valuation result is fairly standard and thus it is omitted.

Proposition 4.3.4 Assume that σ1 , σ2 and ρ are deterministic. Let Y 1 be a defaultable zero-coupon
bond with zero recovery. Then the credit-risk-adjusted forward price of the option written on a
defaultable asset Y 2 equals
¡ ¢ ¡ ¢
FbC (t, T ) = FbY 2 (t, T )N d+ (FbY 2 (t, T ), t, T ) − KN d− (FbY 2 (t, T ), t, T ) .
Equivalently, the pre-default price of the option equals
¡ ¢ ¡ ¢
Cet = Yet2 N d+ (FbY 2 (t, T ), t, T ) − K D
e 0 (t, T )N d− (FbY 2 (t, T ), t, T ) ,
where
ln z − ln K ± 12 v 2 (t, T )
d± (z, t, T ) =
v(t, T )
and Z T
2 2 2
v (t, T ) = (σ1,u + σ2,u − 2ρu σ1,u σ2,u ) du.
t
Moreover the replicating strategy φ in the spot market satisfies, for every t ∈ [0, T ] on the event
{t < τ },
¡ ¢ ¡ ¢
φ1t = −KN d− (FbY 2 (t, T ), t, T ) , φ2t = N d+ (FbY 2 (t, T ), t, T ) .

4.4 PDE Approach

In the remaining part of this chapter, in which we follow Bielecki et al. [14] (see also Rutkowski and
Yousiph [135]), we shall take a different perspective. We assume that trading occurs on the time
interval [0, T ] and our goal is to replicate a contingent claim, which settles at time T , and has the
form
Y = G(YT1 , YT2 , YT3 , HT ) = 1{T ≥τ } g1 (YT1 , YT2 , YT3 ) + 1{T <τ } g0 (YT1 , YT2 , YT3 ).
We do not need to assume here that the coefficients in the dynamics of primary assets are F-
predictable. Since our goal is to develop the PDE approach, it will be essential to postulate a
Markovian character of a model. For the sake of simplicity, we use the notation with constant
coefficients, so that we write, for i = 1, 2, 3,
¡ ¢
dYti = Yt−i
µi dt + σi dWt + κi dMt .
The assumption of constant coefficients is rarely, if ever, satisfied in practically relevant models of
credit risk. It is thus important to stress that it is postulated here mainly for the sake of notational
convenience and the results established in this section cover also the non-homogeneous Markov case
1 2 3 1 2 3
in which µi,t = µi (t, Yt− , Yt− , Yt− , Ht− ), σi,t = σi (t, Yt− , Yt− , Yt− , Ht− ), etc.
158 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

4.4.1 Defaultable Asset with Zero Recovery

We first assume that Y 1 and Y 2 are default-free, so that κ1 = κ2 = 0, and the third asset is subject
to total default, that is, κ3 = −1 and thus
¡ ¢
dYt3 = Yt−
3
µ3 dt + σ3 dWt − dMt .
We work throughout under the assumptions of Proposition 4.3.1. This means that any Q1 -integrable
contingent claim Y = G(YT1 , YT2 , YT3 ; HT ) is attainable and its arbitrage price equals, for every
t ∈ [0, T ],
πt (Y ) = Yt1 EQ1 (Y (YT1 )−1 | Gt ). (4.41)
The following auxiliary result is thus rather obvious.

Lemma 4.4.1 The process (Y 1 , Y 2 , Y 3 , H) has the Markov property with respect to the filtration G
under the martingale measure Q1 . Consequently, for any attainable claim Y = G(YT1 , YT2 , YT3 ; HT )
there exists a pricing function v : [0, T ] × R3 × {0, 1} → R such that πt (Y ) = v(t, Yt1 , Yt2 , Yt3 ; Ht ).

We introduce the pre-default pricing function v(· ; 0) = v(t, y1 , y2 , y3 ; 0) and the post-default
pricing function v(· ; 1) = v(t, y1 , y2 , y3 ; 1).
In fact, since we manifestly have that Yt3 = 0 if Ht = 1, it suffices to study the post-default
function v(t, y1 , y2 ; 1) = v(t, y1 , y2 , 0; 1). We denote
µ1 − µ2
αi = µi − σi , b = (µ3 − µ1 )(σ1 − σ2 ) − (µ1 − µ3 )(σ1 − σ3 ).
σ1 − σ2
Let γ > 0 be the default intensity under P and let ζ > −1 be given by (4.24).

Proposition 4.4.1 Assume that the functions v(· ; 0) and v(· ; 1) belong to the class C1,2 ([0, T ] ×
R3+ , R). Then v(t, y1 , y2 , y3 ; 0) satisfies the PDE
2
X 3
1 X
∂t v(· ; 0) + αi yi ∂i v(· ; 0) + (α3 + ζ)y3 ∂3 v(· ; 0) + σi σj yi yj ∂ij v(· ; 0)
i=1
2 i,j=1
³ b ´£ ¤
− α1 v(· ; 0) + γ − v(t, y1 , y2 ; 1) − v(t, y1 , y2 , y3 ; 0) = 0
σ1 − σ2
with the terminal condition v(T, y1 , y2 , y3 ; 0) = G(y1 , y2 , y3 ; 0). Furthermore, the function v(t, y1 , y2 ; 1)
satisfies the PDE
2
X 2
1 X
∂t v(· ; 1) + αi yi ∂i v(· ; 1) + σi σj yi yj ∂ij v(· ; 1) − α1 v(· ; 1) = 0
i=1
2 i,j=1

with the terminal condition v(T, y1 , y2 ; 1) = G(y1 , y2 , 0; 1).

Proof. For simplicity, we write Ct = πt (Y ). Let us define

∆v(t, y1 , y2 , y3 ) = v(t, y1 , y2 ; 1) − v(t, y1 , y2 , y3 ; 0).
Then the jump ∆Ct = Ct − Ct− can also be represented as follows
¡ ¢
1{τ =t} v(t, Yt1 , Yt2 ; 1) − v(t, Yt1 , Yt2 , Yt−
3
; 0) = 1{τ =t} ∆v(t, Yt1 , Yt2 , Yt−
3
).
We write ∂i to denote the partial derivative with respect to the variable yi and we typically omit
1 2 3
the variables (t, Yt− , Yt− , Yt− , Ht− ) in expressions ∂t v, ∂i v, ∆v, etc. We shall also make use of the
fact that for any Borel measurable function g we have
Z t Z t
g(u, Yu2 , Yu−
3
) du = g(u, Yu2 , Yu3 ) du
0 0
4.4. PDE APPROACH 159

since Yu3 and Yu−3

differ only for at most one value of u (for each ω). Let ξt = 1{t<τ } γ. An application
of Itô’s formula yields

3
X 3
1 X j
dCt = ∂t v dt + ∂i v dYti + i
σi σj Yt− Yt− ∂ij v dt
i=1
2 i,j=1
³ ´
3
+ ∆v + Yt− ∂3 v dHt
3
X 3
1 X j
= ∂t v dt + ∂i v dYti + i
σi σj Yt− Yt− ∂ij v dt
i=1
2 i,j=1
³ ´¡ ¢
3
+ ∆v + Yt− ∂3 v dMt + ξt dt ,

and this in turn implies that

3
X 3
i
¡ ¢ 1 X i j
dCt = ∂t v dt + Yt− ∂i v µi dt + σi dWt + σi σj Yt− Yt− ∂ij v dt
i=1
2 i,j=1
³ ´
3
+ ∆v dMt + ∆v + Yt− ∂3 v ξt dt
n X3
1 X
3 o
i i j
= ∂t v + µi Yt− ∂i v + σi σj Yt− Yt− ∂ij v dt
i=1
2 i,j=1
³ ´ ³X
3 ´
3 i
∆v + Yt− ∂3 v ξt dt + σi Yt− ∂i v dWt + ∆v dMt .
i=1

bt = Ct (Yt1 )−1
The Itô integration by parts formula and (4.17) yield for C

³ ´ ³ 3
X ´
bt = C
dC bt− (−µ1 + σ12 ) dt − σ1 dWt + (Yt−
1 −1
) ∂t v + i
µi Yt− ∂i v dt
i=1
½ 3 ³ ´ ¾
1 −1 1 X i j 3
+ (Yt− ) σi σj Yt− Yt− ∂ij v + ∆v + Yt− ∂3 v ξt dt
2 i,j=1
3
X
1 −1 i 1 −1
+ (Yt− ) σi Yt− ∂i v dWt + (Yt− ) ∆v dMt
i=1
3
X
1 −1 i
− (Yt− ) σ1 σi Yt− ∂i v dt.
i=1

Using (4.22)–(4.23), we obtain

³¡ ¢ ´ 3
X
bt = C
dC bt− ct + (Y 1 )−1
− µ1 + σ12 − σ1 θ dt − σ1 dW i
µi Yt− ∂i v dt
t−
i=1
½ 3 ³ ´ ¾
1 −1 1 X i j 3
+ (Yt− ) ∂t v + σi σj Yt− Yt− ∂ij v + ∆v + Yt− ∂3 v ξt dt
2 i,j=1
3
X 3
X
1 −1
+ (Yt− ) i
σi Yt− ct + (Yt−
∂i v dW 1 −1
) i
σi Yt− θ∂i v dt
i=1 i=1
3
X
1 −1
+ (Yt− ct + (Yt−
) ∆v dM 1 −1 1 −1
) ζξt ∆v dt − (Yt− ) σ1 σ i Yt−
i
∂i v dt.
i=1
160 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

b admits the following decomposition under Q1

This means that the process C
3
X
¡ ¢
bt = C
dC bt− − µ1 + σ12 − σ1 θ dt + (Yt−
1 −1
) i
µi Yt− ∂i v dt
i=1
½ 3 ³ ´ ¾
1 −1 1 X i j 3
+ (Yt− ) ∂t v + σi σj Yt− Yt− ∂ij v + ∆v + Yt− ∂3 v ξt dt
2 i,j=1
3
X
1 −1 i 1 −1
+ (Yt− ) σi Yt− θ∂i v dt + (Yt− ) ζξt ∆v dt
i=1
3
X
1 −1 i
− (Yt− ) σ1 σi Yt− ∂i v dt + a Q1 -martingale.
i=1

b is a martingale under Q1 . Therefore, the continuous finite

From (4.41), it follows that the process C
variation part in the above decomposition necessarily vanishes, and thus we get
3
X
1 −1
¡ ¢
0= Ct− (Yt− ) − µ1 + σ12 − σ1 θ + 1 −1
(Yt− ) µi Yti ∂i v
i=1
½ 3 ¾
1 X ¡ ¢
1 −1
+ (Yt− ) ∂t v + σi σj Yti Ytj ∂ij v + ∆v + Yt3 ∂3 v ξt
2 i,j=1
3
X 3
X
1 −1 i 1 −1 1 −1 i
+ (Yt− ) σi Yt− θ∂i v + (Yt− ) ζξt ∆v − (Yt− ) σ1 σi Yt− ∂i v.
i=1 i=1

Consequently, we have that

¡ ¢
0 = Ct− − µ1 + σ12 − σ1 θ
3
X 3
1 X ¡ ¢
+ ∂t v + µi Yti ∂i v + σi σj Yti Ytj ∂ij v + ∆v + Yt3 ∂3 v ξt
i=1
2 i,j=1
3
X 3
X
+ σi Yti θ∂i v + ζξt ∆v − σ1 σi Yti ∂i v.
i=1 i=1

Finally, we conclude that

2
X 3
1 X
∂t v + αi Yti ∂i v + (α3 + ξt ) Yt3 ∂3 v + σi σj Yti Ytj ∂ij v
i=1
2 i,j=1
− α1 Ct− + (1 + ζ)ξt ∆v = 0.

Recall that ξt = 1{t<τ } γ. It is thus clear that the pricing functions v(·, 0) and v(·; 1) satisfy the
PDEs given in the statement of the proposition. ¤

It should be stressed that in what follows we only examine the form of a replicating strategy
prior to default time.

Proposition 4.4.2 The replicating strategy φ for the claim Y is given by formulae

φ3t Yt−
3
= −∆v(t, Yt1 , Yt2 , Yt−
3
) = v(t, Yt1 , Yt2 , Yt−
3
; 0) − v(t, Yt1 , Yt2 ; 1),
3
X
φ2t Yt2 (σ2 − σ1 ) = −(σ1 − σ3 )∆v − σ1 v + i
Yt− σi ∂i v,
i=1
φ1t Yt1 = v − φ2t Yt2 − φ3t Yt3 .
4.4. PDE APPROACH 161

Proof. Let us sketch the proof. As a by-product of our computations, we obtain

3
X
bt = −(Yt1 )−1 σ1 v dW
dC ct + (Yt1 )−1 i
σi Yt− ct + (Yt1 )−1 ∆v dM
∂i v d W ct .
i=1

The self-financing strategy that replicates Y is determined by two components φ2 , φ3 and the fol-
lowing relationship
bt = φ2 dYt2,1 + φ3 dYt3,1
dC t t
³ ´
ct + φ3 Y 3,1 (σ3 − σ1 ) dW
= φ2t Yt2,1 (σ2 − σ1 ) dW ct − dM
ct .
t t−

3,1
By identification, we thus obtain φ3t Yt− = (Yt1 )−1 ∆v and
3
X
φ2t Yt2 (σ2 − σ1 ) − (σ3 − σ1 )∆v = −σ1 Ct + i
Yt− σi ∂i v.
i=1

This yields the required formulae. ¤

Corollary 4.4.1 In the case of a defaultable claim with zero recovery, the hedging strategy satisfies
the balance condition φ1t Yt1 + φ2t Yt2 = 0 for every t ∈ [0, T ].

Proof. A zero recovery corresponds to the equality G(y1 , y2 , y3 , 1) = 0. We now have v(t, y1 , y2 ; 1) =
0 and thus necessarily
3
φ3t Yt− 3
= v(t, Yt1 , Yt2 , Yt− ; 0)
for every t ∈ [0, T ]. Hence the equality φ1t Yt1 + φ2t Yt2 = 0 holds for every t ∈ [0, T ]. The last equality
is the balance condition for Z = 0; it ensures that the wealth of a replicating portfolio jumps to zero
at default time. ¤

Hedging with the Savings Account

Let us now study the particular case where Y 1 is the savings account, i.e.,

dYt1 = rYt1 dt, Y01 = 1.

Of course, this corresponds to µ1 = r and σ1 = 0. Let rb = r + γ

b, where γ
b, which equals
σ3
γ
b = γ(1 + ζ) = γ + µ3 − r + (r − µ2 ),
σ2
represents the default intensity under the martingale measure Q1 . The quantity rb defined above
has a rather natural interpretation as the risk-neutral credit-risk adjusted short-term interest rate.
Straightforward calculations yield the following corollary to Proposition 4.4.1.

Corollary 4.4.2 Assume that σ2 6= 0 and

dYt1 = rYt1 dt,

¡ ¢
dYt2 = Yt2 µ2 dt + σ2 dWt ,
¡ ¢
dYt3 = Yt−
3
µ3 dt + σ3 dWt − dMt .

Then the function v(· ; 0) satisfies

∂t v(t, y2 , y3 ; 0) + ry2 ∂2 v(t, y2 , y3 ; 0) + rby3 ∂3 v(t, y2 , y3 ; 0) − rbv(t, y2 , y3 ; 0)

3
1 X
+ σi σj yi yj ∂ij v(t, y2 , y3 ; 0) + γ
bv(t, y2 ; 1) = 0
2 i,j=2
162 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

with v(T, y2 , y3 ; 0) = G(y2 , y3 ; 0) and the function v(· ; 1) satisfies

∂t v(t, y2 ; 1) + ry2 ∂2 v(t, y2 ; 1) + 12 σ22 y22 ∂22 v(t, y2 ; 1) − rv(t, y2 ; 1) = 0

with v(T, y2 ; 1) = G(y2 , 0; 1).

In the special case of a survival claim, the function v(· ; 1) vanishes identically since the value of
the claim after default is obviously zero, and thus the following result can be established.

Corollary 4.4.3 The pre-default pricing function v(· ; 0) of a survival claim Y = 1{T <τ } G(YT2 , YT3 )
is a solution of the following PDE

∂t v(t, y2 , y3 ; 0) + ry2 ∂2 v(t, y2 , y3 ; 0) + rby3 ∂3 v(t, y2 , y3 ; 0)

3
1 X
+ σi σj yi yj ∂ij v(t, y2 , y3 ; 0) − rbv(t, y2 , y3 ; 0) = 0
2 i,j=2

with the terminal condition v(T, y2 , y3 ; 0) = G(y2 , y3 ).

The replicating strategy φ satisfies, on the event {t < τ },

3
1 X
φ2t = i
σi Yt− ∂i v(t, Yt2 , Yt−
3
; 0) + σ3 v(t, Yt2 , Yt−
3
; 0),
σ2 Yt2 i=2
3 −1
φ3t = (Yt− 3
) v(t, Yt2 , Yt− ; 0),
1 −rt
¡ ¢
φt = e Ct − φt Yt + φ3t Yt3 ,
2 2

where C is the price of Y , that is, Ct = e−r(T −t) EQ1 (Y | Gt ).

Example 4.4.1 Consider a survival claim Y = 1{T <τ } g(YT2 ), that is, a vulnerable claim with a
default-free underlying asset. Its pre-default pricing function v(· ; 0) does not depend on y3 and
satisfies the following PDE

∂t v(t, y2 ; 0) + ry2 ∂2 v(t, y2 ; 0) + 12 σ22 y22 ∂22 v(t, y2 ; 0) − rbv(t, y2 ; 0) = 0

with the terminal condition v(T, y2 ; 0) = g(y2 ). One can check that the solution to this PDE can be
represented as follows

v(t, y) = e−(br−r)(T −t) vgr,σ2 (t, y) = e−bγ (T −t) vgr,σ2 (t, y),

where the function vgr,σ2 is the price of the default-free claim g(YT2 ) when the dynamics of price
processes (Y 1 , Y 2 ) are given by the Black-Scholes model with the interest rate r and the volatility
parameter σ2 .

4.4.2 Defaultable Asset with Non-Zero Recovery

We now assume that the price of a defaultable asset is governed by the SDE

dYt3 = Yt−
3
(µ3 dt + σ3 dWt + κ3 dMt )

with κ3 > −1 and κ3 6= 0. We assume that Y03 > 0, so that the inequality Yt3 > 0 is valid for every
t ∈ R+ . We shall briefly describe the same steps as in the case of a defaultable asset with zero
recovery.
4.4. PDE APPROACH 163

Arbitrage-Free Property

Assume that the prices Y 1 , Y 2 , Y 3 of traded assets are governed by the following equations
dYt1 = rYt1 dt,
¡ ¢
dYt2 = Yt2 µ2 dt + σ2 dWt ,
¡ ¢
dYt3 = Yt−
3
µ3 dt + σ3 dWt + κ3 dMt ,
where we postulate that σ2 6= 0 and σ3 6= 0.
The existence of an EMM for this model was examined in Section 4.3.3. Recall that in order to
ensure the existence of an EMM, on the event {t > τ }, we need to impose the following condition
r − µ2 r − µ3
= ,
σ2 σ3
that is,
r(σ3 − σ2 ) − µ2 σ3 + µ3 σ2 = 0.
Furthermore, on the event {t ≤ τ }, we obtain the following equations
θt σ2 = r − µ2 ,
θt σ3 + ζt γκ3 = r − µ3 + σ1 .
If, in addition, (σ2 − σ1 )κ3 6= 0, we obtain the unique solution
r − µ2 r − µ3
θ= = ,
σ2 σ3
ζ = 0 > −1,
so that the martingale measure Q1 for Y 2,1 and Y 3,1 exists and is unique.

Pricing PDE and Replicating Strategy

We are in a position to derive the pricing PDEs. For the sake of simplicity, we assume that Y 1 is
the savings account, so that the foregoing result is a counterpart of Corollary 4.4.2. For the proof
of Proposition 4.4.3, the interested reader is referred to Bielecki et al. [14].

Proposition 4.4.3 Let σ2 6= 0 and let the price processes Y 1 , Y 2 , Y 3 satisfy

dYt1 = rYt1 dt,
¡ ¢
dYt2 = Yt2 µ2 dt + σ2 dWt ,
¡ ¢
dYt3 = Yt−
3
µ3 dt + σ3 dWt + κ3 dMt .
Assume, in addition, that σ2 (r − µ3 ) = σ3 (r − µ2 ) and κ3 6= 0, κ3 > −1. Then the price of a
contingent claim Y = G(YT2 , YT3 , HT ) can be represented as πt (Y ) = v(t, Yt2 , Yt3 , Ht ), where the
pricing functions v(· ; 0) and v(· ; 1) satisfy the following PDEs
∂t v(t, y2 , y3 ; 0) + ry2 ∂2 v(t, y2 , y3 ; 0) + y3 (r − κ3 γ) ∂3 v(t, y2 , y3 ; 0)
3
1 X
− rv(t, y2 , y3 ; 0) + σi σj yi yj ∂ij v(t, y2 , y3 ; 0)
2 i,j=2
¡ ¢
+ γ v(t, y2 , y3 (1 + κ3 ); 1) − v(t, y2 , y3 ; 0) = 0
and
∂t v(t, y2 , y3 ; 1) + ry2 ∂2 v(t, y2 , y3 ; 1) + ry3 ∂3 v(t, y2 , y3 ; 1) − rv(t, y2 , y3 ; 1)
3
1 X
+ σi σj yi yj ∂ij v(t, y2 , y3 ; 1) = 0
2 i,j=2
164 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

subject to the terminal conditions

v(T, y2 , y3 ; 0) = G(y2 , y3 ; 0), v(T, y2 , y3 ; 1) = G(y2 , y3 ; 1).

The replicating strategy φ satisfies, on the event {t < τ },

3
1 X
φ2t = σi yi ∂i v(t, Yt2 , Yt−
3
, Ht− )
σ2 Yt2 i=2
σ3 ¡ ¢
− 2 v(t, Yt2 , Yt−3
(1 + κ3 ); 1) − v(t, Yt2 , Yt−
3
; 0) ,
σ2 κ3 Yt
1 ¡ ¢
φ3t = 2 3 2 3
3 v(t, Yt , Yt− (1 + κ3 ); 1) − v(t, Yt , Yt− ; 0) ,
κ3 Yt−
¡ ¢
φ1t = e−rt Ct − φ2t Yt2 + φ3t Yt3 ,

where C is the price of Y , that is, Ct = e−r(T −t) EQ1 (Y | Gt ).

Hedging of a Survival Claim

We shall now illustrate Proposition 4.4.3 by means of examples. As a first example, we will examine
hedging of a survival claim Y of the form

Y = G(YT2 , YT3 , HT ) = 1{T <τ } g(YT3 ).

Then the post-default pricing function v(· ; 1) vanishes identically and the pre-default pricing function
v(· ; 0) solves the PDE

∂t v(t, y2 , y3 ; 0) + ry2 ∂2 v(t, y2 , y3 ; 0) + y3 (r − κ3 γ) ∂3 v(t, y2 , y3 ; 0)

3
1 X
+ σi σj yi yj ∂ij v(t, y2 , y3 ; 0) − (r + γ)v(t, y2 , y3 ; 0) = 0
2 i,j=2

with the terminal condition v(T, y2 , y3 ; 0) = g(y3 ). Let us denote α = r − κ3 γ and β = γ(1 + κ3 ). It
is not difficult to check that the function

v(t, y2 , y3 ; 0) = eβ(T −t) vgα,σ3 (t, y3 )

is a solution of the above equation, where the function w(t, y3 ) = vgα,σ3 (t, y3 ) is the solution of the
following version of the Black-Scholes PDE

∂t w + αy3 ∂y3 w + 21 σ32 y32 ∂y3 y3 w − αw = 0

with the terminal condition vgα,σ3 (T, y3 ) = g(y3 ), that is, the price of the default-free claim g(YT3 )
when the dynamics of (Y 1 , Y 3 ) are given by the Black-Scholes model with the interest rate r = α
and the volatility σ3 .
Let Ct be the current value of the contingent claim Y , so that

Ct = 1{t<τ } eβ(T −t) vgα,σ3 (t, Yt3 ).

The hedging strategy for this survival claim satisfies, on the event {t < τ },
1 −β(T −t) α,σ3 Ct
φ3t Yt3 = − e vg (t, Yt3 ) = − ,
κ3 κ3
σ3 ³ 3 −β(T −t) ´
φ2t Yt2 = Yt e ∂y vg (t, Yt ) − φ3t Yt3 ,
α,σ3 3
σ2
φ1t Yt1 = Ct − φ2t Yt2 + φ3t Yt3 .
4.4. PDE APPROACH 165

Hedging of a Recovery Payoff

As another illustration of Proposition 4.4.3, we shall now consider the claim G(YT2 , YT3 , HT ) =
1{T ≥τ } g(YT2 ), that is, we assume that recovery is paid at maturity and equals g(YT2 ). We argue that
the post-default pricing function v(· ; 1) is independent of y3 . Indeed, the post-default pricing PDE
∂t v(t, y2 , y3 ; 1) + ry∂2 v(t, y2 , y3 ; 1) + 12 σ22 y 2 ∂22 v(t, y2 , y3 ; 1) − rv(t, y2 , y3 ; 1) = 0
with the terminal condition v(T, y2 , y3 ; 1) = g(y), admits a unique solution vgr,σ2 (t, y2 ), which is the
price of g(YT2 ) in the Black-Scholes model with the interest rate r and the volatility σ2 . Prior to
default, the price of the claim can be found by solving the following PDE
∂t v(t, y2 , y3 ; 0) + ry2 ∂2 v(t, y2 , y3 ; 0) + y3 (r − κ3 γ)∂3 v(t, y2 , y3 ; 0)
3
1 X
+ σi σj yi yj ∂ij v(t, y2 , y3 ; 0) − (r + γ)v(t, y2 , y3 ; 0) = −γvgr,σ2 (t, y2 )
2 i,j=2

with the terminal condition v(T, y2 , y3 ; 0) = 0. It is not difficult to check that

v(t, y2 , y3 ; 0) = (1 − e−γ(T −t) )vgr,σ2 (t, y2 ).
It could be instructive to compare this result with Example 4.4.1.

4.4.3 Two Defaultable Assets with Zero Recovery

We shall now assume that only two primary assets are traded, and they are defaultable assets with
zero recovery. We postulate that, for i = 1, 2,
¡ ¢
dYti = Yt−i
µi dt + σi dWt − dMt .

This means that Yti = 1{t<τ } Yeti , i = 1, 2, with the pre-default prices governed by the SDEs, for
i = 1, 2, ¡ ¢
dYeti = Yeti (µi + γ) dt + σi dWt .
In the case where the promised payoff X of a survival claim Y = X1{T <τ } is path-independent, so
that
Y = X1{T <τ } = G(YT1 , YT2 )1{T <τ } = G(YeT1 , YeT2 )1{T <τ }
for some function G, it is possible to use the PDE approach in order to value and replicate a survival
claim prior to default. Under the present assumptions, we need not to examine the balance condition,
since, if default event occurs prior to the maturity date of the claim, the wealth of the portfolio will
fall to zero, as it should in view of the equality Z = 0.
From the martingale approach presented in Section 4.3.4, we already know that hedging of a
survival claim Y = X1{T <τ } is formally equivalent in the present framework to replication of the
promised payoff X = G(YeT1 , YeT2 ) using the pre-default values Ye 1 and Ye 2 of traded assets.
We shall find the pre-default pricing function v(t, y1 , y2 ; 0), which is required to satisfy the ter-
minal condition
v(T, y1 , y2 ; 0) = G(y1 , y2 ),
as well as the replicating strategy (φ1 , φ2 ) for a survival claim. The replicating strategy φ is such
that for the pre-default value Ce of the considered claim Y we have
et := v(t, Yet1 , Yet2 ; 0) = φ1t Yet1 + φ2t Yet2 ,
C
and
et = φ1t dYet1 + φ2t dYet2 .
dC (4.42)
The following result furnishes the pre-default pricing PDE and an explicit formulae for the replication
strategy for a survival claim.
166 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS

Proposition 4.4.4 Assume that σ1 6= σ2 . Then the pre-default pricing function v = v(t, y1 , y2 ; 0)
satisfies the PDE
³ µ2 − µ1 ´ ³ µ2 − µ1 ´
∂t v + y1 µ1 + γ − σ1 ∂1 v + y2 µ2 + γ − σ2 ∂2 v
σ2 − σ1 σ2 − σ1
1³ ´ ³ µ2 − µ1 ´
+ y12 σ12 ∂11 v + y22 σ22 ∂22 v + 2y1 y2 σ1 σ2 ∂12 v = µ1 + γ − σ1 v
2 σ2 − σ1

with the terminal condition v(T, y1 , y2 ) = G(y1 , y2 ). The replicating strategy satisfies

φ1t Yet1 + φ2t Yet2 = v(t, Yet1 , Yet2 )

and
Yet1 σ1 ∂1 v(t, Yet1 , Yet2 ) + Yet2 σ2 ∂2 v(t, Yet1 , Yet2 ) − σ1 v(t, Yet1 , Yet2 )
φ2t Yet2 = .
σ2 − σ1

Proof. Let us sketch the derivation of the pricing PDE and the replicating strategy. By applying
the Itô formula to v(t, Yet1 , Yet2 ) and comparing the diffusion terms in (4.42) and in the Itô differential
dv(t, Yet1 , Yet2 ), we find that

y1 σ1 ∂1 v + y2 σ2 ∂2 v = φ1 y1 σ1 + φ2 y2 σ2 , (4.43)

where φi = φi (t, y1 , y2 ), i = 1, 2 is a replicating strategy. Since we have

φ1 y1 = v(t, y1 , y2 ) − φ2 y2 , (4.44)

we deduce from (4.43) that

y1 σ1 ∂1 v + y2 σ2 ∂2 v = vσ1 + φ2 y2 (σ2 − σ1 ),

and thus the function φ2 equals

y1 σ1 ∂1 v + y2 σ2 ∂2 v − vσ1
φ2 y2 = . (4.45)
σ2 − σ1

Furthermore, by identification of drift terms in (4.43), we obtain

∂t v + y1 (µ1 + γ)∂1 v + y2 (µ2 + γ)∂2 v

1³ 2 2 ´
+ y1 σ1 ∂11 v + y22 σ22 ∂22 v + 2y1 y2 σ1 σ2 ∂12 v
2
= φ1 y1 (µ1 + γ) + φ2 y2 (µ2 + γ).

Upon elimination of φ1 and φ2 , we arrive at the stated PDE. Formulae (4.44) and (4.45) yield the
claimed equalities for the replicating strategy. ¤
Recall that the historically observed drift terms in dynamics of traded assets are µ bi = µi + γ,
rather than µi . The pre-default pricing PDE derived in Proposition 4.4.4 can thus be represented
as follows
µ ¶ µ ¶
µ
b2 − µb1 µ
b2 − µ
b1
∂t v + y1 µ b1 − σ1 ∂1 v + y 2 µb2 − σ2 ∂2 v
σ2 − σ1 σ2 − σ1
µ ¶
1³ 2 2 ´ µ
b2 − µ
b1
+ y1 σ1 ∂11 v + y22 σ22 ∂22 v + 2y1 y2 σ1 σ2 ∂12 v = v µ b1 − σ1 .
2 σ2 − σ1

It is worth noting that the pre-default pricing function v does not depend on default intensity. In
order to further simplify the pre-default pricing PDE for a survival claim, we will make an additional
assumption about the corresponding payoff function G.
4.4. PDE APPROACH 167

Specifically, we suppose, in addition, that the payoff function G of our claim is such that
G(y1 , y2 ) = y1 g(y2 /y1 ) for a certain function g : R+ → R or, equivalently, that the equality
G(y1 , y2 ) = y2 h(y1 /y2 ) holds for some function h : R+ → R. In that case, it is enough to focus on
the relative pre-default prices defined as follows
bt = C
C et (Yet1 )−1 , Yet2,1 = Yet2 (Yet1 )−1 .

The corresponding pre-default pricing function vb(t, z), which is defined as the function such that
bt = vb(t, Yet2,1 ) holds for every t ∈ [0, T ], satisfies the following PDE
the equality C
1
∂t vb + (σ2 − σ1 )2 z 2 ∂zz vb = 0
2
with the terminal condition vb(T, z) = g(z).
We thus conclude that the pre-default price C et = Yet1 vb(t, Yet2,1 ) does not depend directly on the
drift coefficients µ
b1 and µb2 . Therefore, in principle, one should be able to derive an expression for
the price of the claim in terms of market observables, that is, the prices of the underlying assets,
their volatilities and the correlation coefficient. Put another way, neither the default intensity nor
the drift coefficients of the underlying assets appear as independent parameters in the pre-default
pricing function.
Let us conclude this chapter by mentioning that we have presented here only some special cases
of market models and pricing PDEs considered in papers by Bielecki et al. [14] and Rutkowski and
Yousiph [135].
168 CHAPTER 4. HEDGING OF DEFAULTABLE CLAIMS
Chapter 5

Modeling Dependent Defaults

Modeling of dependent defaults is the most important and challenging research area with regard to
credit risk and credit derivatives. We describe the case of conditionally independent default time,
the industry standard copula-based approach, as well as the Jarrow and Yu [95] approach to the
modeling of default times with dependent stochastic intensities. We conclude by summarizing one
of the approaches that were recently developed for the purpose of modeling joint credit ratings
migrations for several firms. It should be acknowledged that several other methods of modeling
dependent defaults, proposed in the literature, are not examined in this text.
Valuation of basket credit derivatives covers, in particular:

• the first-to-default swaps (e.g., Duffie [62], Kijima and Muromachi [105]) – they provide a
protection against the first default in a basket of defaultable claims,
• the kth-to-default claims (e.g., Bielecki and Rutkowski [22]) – a protection against the first k
defaults in a basket of defaultable claims.

Modeling issues arising in the context of dependent defaults include:

• conditionally independent default times (Kijima and Muromachi [105]),

• simulation of correlated defaults (Duffie and Singleton [67]),
• modeling of infectious defaults (Davis and Lo [56]),
• asymmetric default intensities (Jarrow and Yu [95]),
• copulae (Laurent and Gregory [114], Schönbucher and Schubert [137]),
• dependent credit ratings (Lando [108], Bielecki and Rutkowski [21]),
• dependent credit migrations (Kijima et al. [104]),
• modeling defaults via the Marshall-Olkin copula (Elouerkhaoui [71]),
• modeling of losses for a large portfolio (Frey and McNeil [78]).

5.1 Basket Credit Derivatives

Basket credit derivatives are credit derivatives deriving their cash flows (and thus their values) from
credit risks of several reference entities (or prespecified credit events).
Standing assumptions. We assume that:

169
170 CHAPTER 5. DEPENDENT DEFAULTS

• we are given a collection of default times τ1 , τ2 , . . . , τn defined on a common probability space

(Ω, G, Q),
• Q(τi = 0) = 0 and Q(τi > t) > 0 for every i and t,
• Q(τi = τj ) = 0 for arbitrary i 6= j (in a continuous time setup).

For a given collection τ1 , τ2 , . . . , τn of default times, we define the ordered sequence τ(1) < τ(2) <
· · · < τ(n) , where τ(k) stands for the random time of the kth default. Formally, we set
τ(1) = min {τ1 , τ2 , . . . , τn }
and, recursively, for k = 2, 3, . . . , n
© ª
τ(k) = min τi : i = 1, 2, . . . , n, τi > τ(k−1) .
In particular, τ(n) represents the moment of the last default, that is,
τ(n) = max {τ1 , τ2 , . . . , τn }.

5.1.1 The kth-to-Default Contingent Claims

We set Hti = 1{t≥τi } and we denote by Hi the filtration generated by the process H i , that is, by the
observations of the default time τi . In addition, we are given a reference filtration F on the space
(Ω, G, Q). The filtration F is related to some other market risks, for instance, to the interest rate
risk. Finally, we introduce the enlarged filtration G by setting
G = F ∨ H1 ∨ H2 ∨ . . . ∨ Hn .
Note that the σ-field Gt models the total information available to market participants at time t.
A general kth-to-default contingent claim, which matures at time T , is formally specified by the
following covenants:
• if τ(k) = τi ≤ T for some i = 1, 2, . . . , n then the claim pays at time τ(k) the amount Zτi(k) ,
where Z i is an F-predictable recovery process,
• if τ(k) > T then the claim pays at time T an FT -measurable promised amount X.

5.1.2 Case of Two Credit Names

For the sake of notational simplicity, we shall frequently consider the case of two reference credit
names. In that case, the cash flows of considered contracts can be described as follows.

Cash flows of a first-to-default claim (FTDC):

• if τ(1) = min {τ1 , τ2 } = τi ≤ T for i = 1, 2, the claim pays at time τi the amount Zτii ,
• if min {τ1 , τ2 } > T , it pays at time T the amount X.

Cash flows of a last-to-default claim (LTDC):

• if τ(2) = max {τ1 , τ2 } = τi ≤ T for i = 1, 2, the claim pays at time τi the amount Zτii ,
• if max {τ1 , τ2 } > T , it pays at time T the amount X.

We recall that the savings account B equals

³Z t ´
Bt = exp ru du ,
0
and the probability measure Q is interpreted as a martingale measure for our model of the financial
market, which is assumed to include defaultable securities. Consequently, the price B(t, T ) of a
zero-coupon default-free bond maturing at T equals, for every t ∈ [0, T ],
¡ ¢
B(t, T ) = Bt EQ BT−1 | Gt .
5.2. CONDITIONALLY INDEPENDENT DEFAULTS 171

Pricing of FTDC and LTDC

In general, the ex-dividend price at time t of a defaultable claim (X, Z, τ ) is given by the risk-neutral
valuation formula
³Z ¯ ´
¯
St = Bt EQ Bu−1 dDu ¯ Gt
]t,T ]

where D is the dividend process, which describes all cash flows associated with a given defaultable
claim. Consequently, the ex-dividend price at any date t ∈ [0, T ] of an FTDC is given by the
expression
³ ¯ ´
(1) ¯
St = Bt EQ Bτ−1 1
Z 1
τ1 {τ1 <τ2 , t<τ1 ≤T } ¯ Gt
1
³ ¯ ´
¯
+ Bt EQ Bτ−1 2
Z 2
τ2
1 {τ2 <τ 1 , t<τ 2 ≤T } ¯ Gt
³ ¯ ´
¯
+ Bt EQ BT−1 X1{T <τ(1) } ¯ Gt .

Similarly, the ex-dividend price of an LTDC equals, for every t ∈ [0, T ],

³ ¯ ´
(2) ¯
St = Bt EQ Bτ−1 1
Z 1
1
τ1 {τ2 <τ1 , t<τ1 ≤T } ¯ Gt
³ ¯ ´
¯
+ Bt EQ Bτ−12
Zτ22 1{τ1 <τ2 , t<τ2 ≤T } ¯ Gt
³ ¯ ´
¯
+ Bt EQ BT−1 X1{T <τ(2) } ¯ Gt .

Both expressions above are merely special cases of a general formula. The goal is to either derive
more explicit representations under various assumptions about τ1 and τ2 or to provide ways of
efficient calculation of involved expected values by means of Monte Carlo simulation (using perhaps
an equivalent probability measure).

5.2 Conditionally Independent Defaults

The concept of conditional independence of default times with respect to a reference filtration F is
defined as follows.

Definition 5.2.1 The random times τi , i = 1, 2, . . . , n are said to be conditionally independent with
respect to F under Q if we have, for any T > 0 and any t1 , . . . , tn ∈ [0, T ],
n
Y
Q(τ1 > t1 , . . . , τn > tn | FT ) = Q(τi > ti | FT ).
i=1

Let us comment briefly on Definition 5.2.1.

• Conditional independence has the following intuitive interpretation: the reference credits
(credit names) are subject to common risk factors that may trigger credit (default) events.
In addition, each credit name is subject to idiosyncratic risks that are specific for this name.
• Conditional independence of default times means that once the common risk factors are fixed
then the idiosyncratic risk factors are independent of each other. This means that most
computations can be done similarly as in the case of independent default times.
• It is worth stressing that the property of conditional independence is not invariant with respect
to an equivalent change of a probability measure (for a suitable counterexample, see Section
5.7).
172 CHAPTER 5. DEPENDENT DEFAULTS

5.2.1 Canonical Construction

Let Γi , i = 1, 2, . . . , n be a given family of F-adapted, increasing, continuous processes, defined on
a probability space (Ω̃, F, P̃). We make the standard assumptions that Γi0 = 0 and Γi∞ = ∞. Let
b F,
(Ω, b be an auxiliary probability space with a sequence ξi , i = 1, 2, . . . , n of independent random
b P)
variables uniformly distributed on [0, 1]. We define the random times τ1 , . . . , τn by setting

b ) = inf { t ∈ R+ : Γit (ω̃) ≥ − ln ξi (b

τi (ω̃, ω ω )}

for every elementary event (ω̃, ω b F∞ ⊗

b ) belonging to the product probability space (Ω, G, Q) = (Ω̃×Ω,
F, b We endow the space (Ω, G, Q) with the filtration G = F ∨ H ∨ · · · ∨ H .
b P̃ ⊗ P). 1 n

Proposition 5.2.1 Let the random variables ξ1 , . . . , ξn be independent and uniformly distributed on
[0, 1]. Then the process Γi is the F-hazard process of τi and thus, for any s ≥ t,
¡ i i ¢
Q(τi > s | Ft ∨ Hti ) = 1{t<τi } EQ eΓt −Γs | Ft .

We have that Q(τi = τj ) = 0 for every i 6= j and the default times τ1 , . . . , τn are conditionally
independent with respect to F under Q.

Proof. It suffices to note that, for ti < T ,

Q(τ1 > t1 , . . . , τn > tn | FT ) = Q(Γ1t1 ≥ − ln ξ1 , . . . , Γntn ≥ − ln ξn | FT )

Yn
i
= e−Γti .
i=1

The details are left to the reader. ¤

Rt
Recall that if Γit = 0 γui du then γ i is the F-intensity of τi . Intuitively,

Q(τi ∈ [t, t + dt] | Ft ∨ Hti ) ≈ 1{t<τi } γti dt.

5.2.2 Hypothesis (H)

If hypothesis (H) holds between the filtrations F and G then it also holds between the filtrations F
and F ∨ Hi1 ∨ · · · ∨ Hik for any i1 , . . . , ik . However, there is no reason for hypothesis (H) to hold
between F ∨ Hi1 and G. Note that, if hypothesis (H) holds then one has, for every t1 ≤ . . . ≤ tn ≤ T ,

Q(τ1 > t1 , . . . , τn > tn | FT ) = Q(τ1 > t1 , . . . , τn > tn | F∞ ).

It is not difficult to check that hypothesis (H) holds when the random times τ1 , . . . , τn are given by
the canonical construction of Section 5.2.1.

5.2.3 Independent Default Times

We shall first examine the case of default times τ1 , . . . , τn that are independent under Q. Suppose
that for every i = 1, 2, . . . , n we know the cumulative distribution function Fi (t) = Q(τi ≤ t) of the
default time of the ith reference entity. The cumulative distribution functions of τ(1) and τ(n) are
n
Y
F(1) (t) = Q(τ(1) ≤ t) = 1 − (1 − Fi (t))
i=1

and
n
Y
F(n) (t) = Q(τ(n) ≤ t) = Fi (t).
i=1
5.2. CONDITIONALLY INDEPENDENT DEFAULTS 173

More generally, for any i = 1, 2, . . . , n we have

n
X X Y Y
F(i) (t) = Q(τ(i) ≤ t) = Fkj (t) (1 − Fkl (t)),
m=i π∈Πm j∈π l6∈π

where Πm denote the family of all subsets of {1, 2, . . . , n} consisting of m elements. Suppose, in
addition, that the default times τ1 , . . . , τn admit intensity functions γ1 (t), . . . , γn (t), so that the
processes Z t∧τi
Hti − γi (u) du
0
Rt
are Hi -martingales. Recall that Q(τi > t) = e− 0
γi (u) du
. It is easily seen that, for every t ∈ R+ ,
n
Y Rt
Q(τ(1) > t) = Q(τi > t) = e− 0
γ(1) (u) du
,
i=1

where γ(1) (t) = γ1 (t) + . . . + γn (t) for every t ∈ R+ . Therefore, the process
Z t∧τ(1)
(1)
Ht − γ(1) (u) du
0
(1)
follows an H -martingale, where the filtration H(1) is generated by the process Ht = σ(τ(1) ∧ t).
(1)

By similar computations, it is also possible to find the intensity function γ(k) of the kth default time
τ(k) for every k = 2, 3, . . . , n.
Let us consider, for instance, a digital default put of basket type. To be more specific, we postulate
that a contract pays a fixed amount (e.g., one unit of cash) at the moment τ(k) of the kth default
provided that τ(k) ≤ T . If the interest rate is non-random then its value at time 0 equals
Z
¡ −1 ¢
S0 = EQ Bτ 1{τ(k) ≤T } = Bu−1 dF(k) (u).
]0,T ]

If default times τ1 , . . . , τn admit intensity functions γ1 , . . . , γn then

Z T Z T Ru
−1
S0 = Bu dF(k) (u) = Bu−1 γ(k) (u)e− 0 γ(k) (v) dv du.
0 0

5.2.4 Signed Intensities

Some authors (see, e.g., Kijima and Muromachi [105]) examine credit risk models in which the
negative values of “default intensities” are allowed. In that case, the process chosen to model the
“default intensity” does not play the role of the actual default intensity, in particular, the process
Z t∧τ
Mt = Ht − γt dt
0

is not necessarily a martingale. Negative values of the “default intensity” process clearly contradict
the usual interpretation of the intensity as the conditional default probability over an infinitesimal
time interval.
Nevertheless, for a given collection Γi , i = 1, 2, . . . , n of F-adapted, continuous processes, with
Γi0= 0, which are defined on (Ω̃, F, P̃), one can construct random times τi , i = 1, 2, . . . , n on the
enlarged probability space (Ω, G, Q) by setting
τi = inf { t ∈ R+ : Γit ≥ − ln ξi }. (5.1)

Let us denote Γ b it = max u≤t Γiu . Observe that if the process Γi is absolutely continuous then so is
b i b i with respect
the process Γ . In that case, the actual intensity of τi is obtained as the derivative of Γ
to the time variable. The following result examines the case of signed intensities.
174 CHAPTER 5. DEPENDENT DEFAULTS

Lemma 5.2.1 Random times τi , i = 1, 2, . . . , n given by (5.1) are conditionally independent with
respect to F under Q. In particular, for any T > 0 and every t1 , . . . , tn ≤ T ,
n
Y Pn
bi bi
Q(τ1 > t1 , . . . , τn > tn | FT ) = e−Γti = e− i=1 Γ t
i .
i=1

5.3 Valuation of FTDC and LTDC

Pricing of a first-to-default claim or a last-to-default claim is straightforward under the assumption
of conditional independence of default times as manifested by the following result in which, for
notational simplicity, we consider only the case of two credit names. As usual, we do not state
explicitly integrability conditions that should be imposed on a recovery process Z and a terminal
payoff X.

Proposition 5.3.1 Let the default times τj , j = 1, 2 be F-conditionally independent. Assume that
the recovery Z = Z 1 = Z 2 is an F-predictable process and the terminal payoff X is FT -measurable.
(i) If hypothesis (H) holds between F and G and processes F i , i = 1, 2 are continuous then the price
at time t = 0 of the first-to-default claim with Z 1 = Z 2 = Z equals
³Z T
1 2
´
(1) (1)
S0 = EQ Bu−1 Zu e−(Γu +Γu ) d(Γ1u + Γ2u ) + BT−1 XGT , (5.2)
0

where we denote
(1) 1 2
GT = Q(τ(1) > T | FT ) = Q(τ1 > T | FT ) Q(τ2 > T | FT ) = e−(ΓT +ΓT ) .

(ii) In the general case, let

Z t
Fti = Q(τi ≤ t | Ft ) = Nti + Cti = Nti + ciu du,
0

where N i is a continuous F-martingale. Then we have

³Z T ¡ 1 2 ¢ ´
(1) (1)
S0 = EQ Bu−1 Zu e−(Γu +Γu ) (λ1u + λ2u ) du + dhN 1 , N 2 iu + BT−1 XGT
0

where λiu = ciu (1 − Fui )−1 .

Proof. To simplify the notation, we will only consider the case where B = 1. A computation of the
expectation EQ (X1{τ(1) >T } ) is straightforward. Thus, let us focus on the evaluation of the expected
value
EQ (Zτ 1{τ ≤T } ),
where, for brevity, we denote τ = τ(1) = τ1 ∧ τ2 .
From Lemma 3.1.3, we know that if Z is F-predictable then
¡ ¢ ³Z ´
EQ Zτ 1{τ ≤T } = EQ Zu dFu ,
]0,T ]

where Fu = Q(τ ≤ u | Fu ). For τ = τ1 ∧ τ2 , the conditional independence assumption yields

1 − Fu = Q(τ1 > u, τ2 > u | Fu ) = Q(τ1 > u | Fu ) Q(τ2 > u | Fu )

= (1 − Fu1 )(1 − Fu2 ).
5.4. COPULA-BASED APPROACHES 175

Case (i). Under the assumption that hypothesis (H) holds between filtrations F and Gi for i = 1, 2,
the processes F i are continuous and increasing. Consequently,
1 2 1 2
dFu = e−Γu dFu2 + e−Γu dFu1 = e−(Γu +Γu ) d(Γ1u + Γ2u ),
and this in turn yields
¡ ¢ ³Z T
1 2
´
EQ Zτ1 ∧τ2 1{τ1 ∧τ2 <T } = EQ Zu e−(Γu +Γu ) d(Γ1u + Γ2u ) .
0

Case (ii). In the general case, the Doob-Meyer decomposition of the process F i is F i = N i + C i
and, under our assumptions, the process
Z t∧τi
Hti − λiu du
0
i
is a G -martingale, where we write λiu = ciu (1 − Fui )−1 . We now have
1 2
dFu = e−Γu dFu2 + e−Γu dFu1 + dhN 1 , N 2 iu .
Since N 1 and N 2 are martingales, it follows that
¡ ¢ ³Z T 1 2
´
EQ Zτ1 ∧τ2 1{τ1 ∧τ2 <T } = EQ Zu (e−Γu dCu2 + e−Γu dCu1 + dhN 1 , N 2 iu )
0
³Z T ¡ ¢´
−(Γ1u +Γ2u )
= EQ Zu e (λ1u + λ2u ) du + dhN 1 , N 2 iu ,
0

as required. ¤

The valuation formula (5.2) can be easily extended to the case of an arbitrary date t ∈ [0, T ].
This is left as an exercise for the reader.

5.4 Copula-Based Approaches

As already mentioned in Section 2.6, the classic concept of a copula function provides a conve-
nient tool for producing multidimensional probability distributions with predetermined univariate
marginal distributions.

Definition 5.4.1 A function C : [0, 1]n → [0, 1] is called a copula function if the following conditions
are satisfied:
(i) C(1, . . . , 1, vi , 1, . . . , 1) = vi for any i = 1, 2, . . . , n and any vi ∈ [0, 1],
(ii) C is an n-dimensional cumulative distribution function.

The following well known theorem, due to Sklar, underpins the theory of copula functions. For
the proof of this result and further properties of copula functions, see Nelsen [127].

Theorem 5.4.1 For any cumulative distribution function F on Rn there exists a copula function
C such that F (x1 , . . . , xn ) = C(F1 (x1 ), . . . , Fn (xn )) for every x1 , . . . , xn ∈ R, where Fi is the ith
marginal cumulative distribution function. If, in addition, the function F is continuous then C is
unique.

Let us first give a few examples of copula functions:

(i) the product copula C(v1 , . . . , vn ) = Πni=1 vi , which corresponds to the independence,
(ii) the Gumbel copula, which is given by the formula, for θ ∈ [1, ∞),
µ ·X n ¸1/θ ¶
C(v1 , . . . , vn ) = exp − (− ln vi )θ ,
i=1
176 CHAPTER 5. DEPENDENT DEFAULTS

(iii) the Gaussian copula, which is given by the expression

¡ ¢
C(v1 , . . . , vn ) = NΣn N −1 (v1 ), . . . , N −1 (vn ) ,
where NΣn is the cumulative distribution function for the n-variate central Gaussian distribution
with the linear correlation matrix Σ and N −1 is the inverse of the cumulative distribution function
for the univariate standard Gaussian distribution.
(iv) the Student t-copula, defined as
¡ ¢
C(v1 , . . . , vn ) = Θnν,Σ t−1 −1
ν (v1 ), . . . , tν (vn ) ,

where Θnν,Σ stands for the cumulative distribution function of the n-variate t-distribution with ν
degrees of freedom and with the linear correlation matrix Σ and t−1
ν is the inverse of the cumulative
distribution function of the univariate Student t-distribution with ν degrees of freedom.

5.4.1 Direct Approach

In a direct application, we first postulate that a (univariate marginal) cumulative distribution func-
tion Fi for each random variable τi , i = 1, 2, . . . , n is given. A particular copula function C is then
chosen in order to introduce an appropriate dependence structure of the random vector (τ1 , . . . , τn ).
The joint probability distribution of the random vector (τ1 , . . . , τn ) is given as
¡ ¢
Q(τ1 ≤ t1 , . . . , τn ≤ tn ) = C F1 (t1 ), . . . , Fn (tn ) .
The direct copula-based approach has an apparent shortcoming of being essentially a static approach,
in the sense that it makes no account of the flow of market information, which can be represented
by some reference filtration.

5.4.2 Indirect Approach

A less straightforward application of copula functions relies on an extension of the canonical construc-
tion of conditionally independent default times. In the approach described below, the dependence
between the default times is enforced both through the dependence between the marginal hazard
processes Γb i , i = 1, 2, . . . , n and through the choice of a copula function C. For this reason, it is
sometimes referred to as the double correlation case.
Assume that the joint probability distribution of (ξ1 , . . . , ξn ) in the canonical construction is
given by an n-dimensional copula function C. Similarly as in Section 5.2.1, we postulate that the
random vector (ξ1 , . . . , ξn ) is independent of F and we set

τi (e
ω, ω b it (e
b ) = inf { t ∈ R+ : Γ ω ) ≥ − ln ξi (b
ω )}.
We have that, for any T > 0 and arbitrary t1 , . . . , tn ≤ T ,
Q(τ1 > t1 , . . . , τn > tn | FT ) = C(Kt11 , . . . , Ktnn ),
bi
where we denote Kti = e−Γt . Schönbucher and Schubert [137] show that the following equality holds,
for arbitrary s ≤ t on the event {τ1 > s, . . . , τn > s},
µ ¶
C(Ks1 , . . . , Kti , . . . , Ksn ) ¯¯
Q(τi > t | Gs ) = EQ ¯ Fs .
C(Ks1 , . . . , Ksn )
bi
dΓ
bti =
Consequently, assuming that the derivatives γ dt
t
exist, the ith survival intensity equals, on the
event {τ1 > t, . . . , τn > t},
∂
∂vi C(Kt1 , . . . , Ktn ) ∂
λit = γ
bti Kti bti Kti
=γ ln C(Kt1 , . . . , Ktn ),
C(Kt1 , . . . , Ktn ) ∂vi
5.5. ONE-FACTOR GAUSSIAN COPULA MODEL 177

where λit is understood as the following limit

λit = lim h−1 Q(t < τi ≤ t + h | Ft , τ1 > t, . . . , τn > t).

h↓0

It appears that, in general, the survival intensity of the ith name jumps at time t if the jth name
defaults at time t for some j 6= i. In fact, it can be shown that
∂2
C(Kt1 , . . . , Ktn )
i ∂vi ∂vj
λi,j
t = bti
γ Kt ∂ 1 n
,
∂vj C(Kt , . . . , Kt )

where λi,j
t is defined as follows

λi,j
t = lim h
−1
Q(t < τi ≤ t + h | Ft , τk > t, k 6= j, τj = t).
h↓0

Schönbucher and Schubert [137] examine the behavior of survival intensities after defaults of some
names. Let us fix s, and let ti ≤ s for i = 1, 2, . . . , k < n and ti ≥ s for i = k + 1, k + 2, . . . , n. They
show that
¡ ¢
Q τi > ti , i = k + 1, . . . , n | Fs , τj = tj , j = 1, . . . , k, τi > s, i = k + 1, . . . , n
³ ¯ ´
k ¯
EQ ∂v1∂...∂vk C(Kt11 , . . . , Ktkk , Ktk+1
k+1
, . . . , Ktnn ) ¯ Fs
= ∂k 1 k k+1
.
∂v1 ...∂vk C(Kt1 , . . . , Ktk , Ks , . . . , Ksn )

Unfortunately, in this approach it is difficult to control the jumps of intensities, otherwise than
by a judicious choice of the copula function C.

5.5 One-factor Gaussian Copula Model

Laurent and Gregory [114] examine a simplified version of Schönbucher and Schubert [137] approach,
corresponding to the trivial reference filtration F (we thus deal here with the direct approach). The
bi are deterministic functions and the marginal distributions of defaults
marginal default intensities γ
are given by the expression
Rt
bi (u) du
Q(τi > t) = 1 − Fi (t) = e− 0
γ
.

They derive closed-form expressions for certain conditional default intensities by making specific
assumptions regarding the choice of a copula C.
Let us describe the one-factor Gaussian copula model, proposed by Li [118], which was adopted by
the financial industry as a benchmark model for valuing traded and bespoke tranches of collateralized
debt obligations (see Section 5.8.2). Let us set
p
Xi = ρV + 1 − ρ2 Vi ,

where V and Vi , i = 1, 2, . . . , n are independent, standard Gaussian variables under Q and the corre-
lation parameter ρ belongs to (−1, 1). Let C be the copula function corresponding to the distribution
of the vector (X1 , . . . , Xn ), that is, let C be given by the expression, for every v1 , . . . , vn ∈ [0, 1],
¡ ¢
C(v1 , . . . , vn ) = Q X1 < N −1 (v1 ), . . . , Xn < N −1 (vn ) .

We define the default times τi , i = 1, 2, . . . , n by the formula

n Z t o
τi = inf t ∈ R+ : bi (u) du > − ln ξi
γ
0
178 CHAPTER 5. DEPENDENT DEFAULTS

or, equivalently,
τi = inf { t ∈ R+ : 1 − Fi (t) < ξi },
where the uniformly distributed random barriers are defined by the equality ξi = 1 − N (Xi ). It is
worth noting that the random vectors (X1 , . . . , Xn ), (ξ1 , . . . , ξn ) and (τ1 , . . . , τn ) share a common
Gaussian copula function C; this follows from the monotonicity of the transformations involved.
Moreover, the following equality is valid, for every i = 1, 2, . . . , n and every t ∈ R+ ,
½ ¾
N −1 (Fi (t)) − ρV
{τi ≤ t} = {ξi ≥ 1 − Fi (t)} = Vi ≤ p .
1 − ρ2
By the conditional independence of X1 , . . . , Xn with respect to the common factor V , which repre-
sents the market-wide (or systematic) credit risk, we thus obtain, for every t1 , . . . , tn ∈ R+ ,
Z Y n
à !
N −1 (Fi (ti )) − ρv
Q(τ1 ≤ t1 , . . . , τn ≤ tn ) = N p n(v) dv,
R i=1 1 − ρ2

where n is the probability density function of V . It is worth noting that the components Vi are
aimed to represent the firm-specific (or idiosyncratic) part of the credit risk for individual names in
a credit portfolio. For numerical issues arising in implementations of the Li model, see Joshi and
Kainth [102] and Chen and Glasserman [47].

5.6 Jarrow and Yu Model

Jarrow and Yu [95] (see also Yu [144]) approach can be considered as another attempt to develop a
dynamic approach to dependence between default times by modeling directly the contagion effect.
For a given finite family of reference credit names, Jarrow and Yu [95] propose to make a distinction
between primary and secondary firms. At the intuitive level, the rationale for their approach can be
summarized as follows:
• the class of primary firms encompasses these entities whose probabilities of default are influ-
enced by macroeconomic conditions, but not by the credit risk of counterparties; the pricing of
bonds and other defaultable securities issued by primary firms is feasible through the standard
intensity-based methodology,
• it is thus sufficient to focus on defaultable securities issued by a secondary firm, that is, a firm
for which the intensity of default depends explicitly on the status of some other firms.

Let {1, 2, . . . , n} represent the set of all firms in our model and let F stand for some reference
filtration. Jarrow and Yu [95] formally postulate that:
• for any firm from the set {1, 2, . . . , k} of primary firms, the ‘default intensity’ depends only on
a reference filtration F,
• the ‘default intensity’ for any credit name that belongs to the class {k + 1, k + 2, . . . , n} of
secondary firms may depend not only on the filtration F, but also on the status (default or
no-default) of the primary firms.

5.6.1 Construction of Default Times

First step. We first construct default times for all primary firms. To this end, we assume that we
are given a family of F-adapted ‘intensity processes’ λ1 , . . . , λk and we produce a collection τ1 , . . . , τk
of F-conditionally independent random times through the canonical method, that is, we set
n Z t o
τi = inf t ∈ R+ : λiu du ≥ − ln ξi
0
5.6. JARROW AND YU MODEL 179

where ξi , i = 1, 2, . . . , k are mutually independent and identically distributed random variables with
the uniform distribution on [0, 1] under the martingale measure Q.

Second step. In the second step, we are going to construct default times corresponding to secondary
firms. To this end, we assume that:
• the probability space (Ω, G, Q) is large enough to support a family ξi , i = k + 1, k + 2, . . . , n of
mutually independent random variables, with uniform distribution on [0, 1],
• these random variables are independent not only of the filtration F, but also of the already
constructed in the first step default times τ1 , . . . , τk of primary firms.

The default times τi , i = k + 1, k + 2, . . . , n are also defined by means of the standard formula
n Z t o
τi = inf t ∈ R+ : λiu du ≥ − ln ξi .
0

i
However, the ‘intensity processes’ λ for i = k + 1, k + 2, . . . , n are now given by the following
expression
k
X
λit = µit + νti,l 1{t≥τl } ,
l=1
i i,l
where µ and ν are F-adapted stochastic processes. In case where the default of the jth primary
firm does not affect the ‘default intensity’ of the ith secondary firm, we set ν i,j = 0.
b = F ∨ Hk+1 ∨ . . . ∨ Hn be
Let G = F ∨ H1 ∨ . . . ∨ Hn stand for the enlarged filtration and let F
the filtration generated by the reference filtration F and the observations of defaults of secondary
firms. Then:
• the default times τ1 , . . . , τk of primary firms are conditionally independent with respect to F,
• the default times τ1 , . . . , τk of primary firms are no longer conditionally independent when we
replace the filtration F by F, b
• in general, the default intensity of a primary firm with respect to the filtration F b differs from
the intensity λi with respect to F.

5.6.2 Case of Two Credit Names

To illustrate the credit contagion effect, we will now consider the case of only two credit names, A
and B say, and we postulate that A is a primary firm, whereas B is a secondary firm.
Let the constant process λ1t = λ1 represent the F-intensity of default time for firm A, so that
n Z t o
τ1 = inf t ∈ R+ : λ1u du = λ1 t ≥ − ln ξ1 ,
0

where ξ1 is a random variable independent of F with the uniform distribution on [0, 1]. For the
second firm, the ‘default intensity’ is assumed to satisfy

λ2t = λ2 1{t<τ1 } + α2 1{t≥τ1 }

for some positive constants λ2 and α2 . We set

n Z t o
τ2 = inf t ∈ R+ : λ2u du ≥ − ln ξ2 ,
0

where ξ2 is a random variable with the uniform probability distribution, independent of F, and such
that ξ1 and ξ2 are mutually independent. The following result summarizes properties of processes
Λ1 and Λ2 .
180 CHAPTER 5. DEPENDENT DEFAULTS

Lemma 5.6.1 The following properties hold:

(i) the process Λ1 is the hazard process of τ1 with respect to F,
(ii) the process Λ2 is the hazard process of τ2 with respect to F ∨ H1 ,
(iii) the process Λ1 is not the hazard process of τ1 with respect to F ∨ H2 if the inequality λ2 6= α2
holds.

Assume for simplicity that r = 0. We wish to price a defaultable zero-coupon bond with the
default time τi and with constant recovery payoff δi . We thus need to compute the following
conditional expectation, for i = 1, 2,
Diδi (t, T ) = EQ (1{τi >T } + δi 1{τi ≤T } | Gt ), (5.3)
where Gt = Ht1 ∨Ht2 .To this end, we will first find the joint probability distribution of the pair
(τ1 , τ2 ). Let us denote G(s, t) = Q(τ1 > s, τ2 > t). We write ∆ = λ1 + λ2 − α2 and we assume that
∆ 6= 0.

Lemma 5.6.2 The joint distribution of (τ1 , τ2 ) under Q is given by, for every 0 ≤ t ≤ s,
Q(τ1 > s, τ2 > t) = e−λ1 s−λ2 t
and, for every 0 ≤ s < t,
1 ¡ ¢
Q(τ1 > s, τ2 > t) = λ1 e−α2 t e−s∆ − e−t∆ + e−(λ1 +λ2 )t .
∆
Proof. Let ψi = − ln ξi . For t < s, we have λ2t = λ2 t on the set {s < τ1 }. The equalities
{τ1 > s} ∩ {τ2 > t} = {Λ1s < ψ1 } ∩ {Λ2t < ψ2 } = {λ1 s < ψ1 } ∩ {λ2 t < ψ2 }
and the independence of ψ1 and ψ2 lead to
Q(τ1 > s, τ2 > t) = e−λ1 s−λ2 t .
In particular, by setting t = 0, we obtain the equality Q(τ1 > s) = e−λ1 s for every s ∈ R+ .
For t > s, we have that
{τ1 > s} ∩ {τ2 > t} = {{t > τ1 > s} ∩ {τ2 > t}} ∪ {{τ1 > t} ∩ {τ2 > t}}
and
{t > τ1 > s} ∩ {τ2 > t} = {t > τ1 > s} ∩ {Λ2t < ψ2 }
= {t > τ1 > s} ∩ {λ2 τ1 + α2 (t − τ1 ) < ψ2 }.
The independence between ψ1 and ψ2 implies that the random variable τ1 is independent of ψ2 (note
that τ1 = (λ1 )−1 ψ1 ). Consequently,
³ ´
Q(t > τ1 > s, τ2 > t) = EQ 1{t>τ1 >s} e−(λ2 τ1 +α2 (t−τ1 ))
Z t
= e−(λ2 u+α2 (t−u)) λ1 e−λ1 u du
s
1 ³ ´
= λ1 e−α2 t e−(λ1 +λ2 −α2 )s − e−(λ1 +λ2 −α2 )t .
λ1 + λ2 − α2
Denoting ∆ = λ1 + λ2 − α2 , it follows that
1 ¡ ¢
Q(τ1 > s, τ2 > t) = λ1 e−α2 t e−∆s − e−∆t + e−(λ1 +λ2 )t .
∆
In particular, for s = 0, we obtain
1 ³ ³ −α2 t ´ ´
Q(τ2 > t) = λ1 e − e−(λ1 +λ2 )t + ∆e−(λ1 +λ2 )t .
∆
This completes the proof. ¤
5.6. JARROW AND YU MODEL 181

Bonds with Non-Zero Recovery

In view of (5.3), to find the price D1δ1 (t, T ), it suffices to compute

Q(τ1 > T | Ht2 )

Q(τ1 > T | Gt ) = Q(τ1 > T | Ht1 ∨ Ht2 ) = 1{t<τ1 } .
Q(τ1 > t | Ht2 )

Observe that
µ ¶
∂2 G(T, τ2 ) G(T, t)
Q(τ1 > T | Gt ) = 1{t<τ1 } 1{t≥τ2 } + 1{t<τ2 } .
∂2 G(t, τ2 ) G(t, t)

Similarly, valuation of D2δ2 (t, T ) follows from the computation of

Q(τ2 > T | Ht1 )

Q(τ2 > T | Gt ) = 1{t<τ2 } ,
Q(τ2 > t | Ht1 )

where, by symmetry, we have that

µ ¶
∂1 G(τ1 , T ) G(t, T )
Q(τ2 > T | Gt ) = 1{t<τ2 } 1{t≥τ1 } + 1{t<τ1 } .
∂1 G(τ1 , t) G(t, t)

By straightforward computations, we obtain the following pricing result for defaultable bonds
with non-zero recovery.

Corollary 5.6.1 The prices of defaultable bonds equal, for every t ∈ [0, T ]
¡ ¢
D1δ1 (t, T ) = 1{t≥τ1 } δ1 + 1{t<τ1 } e−λ1 (T −t) + δ1 (1 − e−λ1 (T −t) )

and
½
D2δ2 (t, T ) = δ2 + (1 − δ2 )1{t<τ2 } 1{t≥τ1 } e−α2 (T −t)

1 ³ ´¾
−α2 (T −t) −(λ1 +λ2 )(T −t)
+ 1{t<τ1 } λ1 e + (λ2 − α2 )e .
λ1 + λ2 − α2

Bonds with Zero Recovery

Assume that λ1 +λ2 −α2 6= 0 and that the bond is subject to the zero recovery scheme. We maintain
the assumption that r = 0 so that B(t, T ) = 1 for t ≤ T . Therefore, we have D20 (t, T ) = Q(τ2 >
T | Ht1 ∨ Ht2 ) and thus the general formula yields

Q(τ2 > T | Ht1 )

D20 (t, T ) = 1{t<τ2 } .
Q(τ2 > t | Ht1 )

The following pricing result is an immediate consequence of Corollary 5.6.1.

Corollary 5.6.2 Assume that the recovery δ2 = 0. Then D2 (t, T ) = 0 on the event {t ≥ τ2 }. On
the event {t < τ2 } we have

1 ³ ´
D20 (t, T ) = 1{t<τ1 } λ1 e−α2 (T −t) + (λ2 − α2 )e−(λ1 +λ2 )(T −t)
λ1 + λ2 − α2
−α2 (T −t)
+ 1{t≥τ1 } e .
182 CHAPTER 5. DEPENDENT DEFAULTS

5.7 Kusuoka’s Model

Following Kusuoka [106] (see also Bielecki and Rutkowski [21]), we will argue that the assumption
that some firms are classified as primary, while some other are considered to be secondary, is of
no relevance from the point of view of modeling. For simplicity, we make the following standing
assumptions:
• we set n = 2, that is, we consider the case of two credit names,
• the interest rate r equals zero, so that B(t, T ) = 1 for every t ≤ T ,
• the reference filtration F is trivial,
• all corporate bonds are subject to the zero recovery scheme.

In view of the model symmetry, it suffices to analyze a bond issued by the first firm. By definition,
the price of this bond at time t ∈ [0, T ] equals

D10 (t, T ) = Q(τ1 > T | Ht1 ∨ Ht2 ).

Of course, this value is based on the complete information, as modeled by the full filtration G =
H1 ∨ H2 . For the sake of comparison, we will also evaluate the corresponding values, which are based
on the assumption that only a partial observation is available; specifically, we will compute
b 10 (t, T ) = Q(τ1 > T | Ht1 ),
D D̄10 (t, T ) = Q(τ1 > T | Ht2 ).

5.7.1 Model Specification

We follow here Kusuoka [106]. Under the original probability measure P the random times τi , i = 1, 2
are assumed to be mutually independent random variables with exponential laws with parameters
λ1 and λ2 , respectively.
For a fixed T > 0, we define a probability measure Q equivalent to P on (Ω, G) by setting
dQ
= ηT , P-a.s.,
dP
where the Radon-Nikodým density process (ηt , t ∈ [0, T ]) satisfies
2 Z
X
ηt = 1 + ηu− κiu dMui ,
i=1 ]0,t]

where in turn the processes M 1 and M 2 are given by

Z t∧τi
Mti = Hti − λi du = Hti − (t ∧ τi )λi ,
0

where we write, as usual, Hti = 1{t≥τi } , and the G-predictable processes κ1 and κ2 are given by the
following expressions ³α ´
1
κ1t = 1{t>τ2 } −1
λ1
and ³α ´
2
κ2t = 1{t>τ1 } −1
λ2
for some constants αi > 0 for i = 1, 2. Note that the inequality κi > −1 holds for i = 1, 2. It is
not difficult to check, using Girsanov’s theorem, that the G-intensities (cf. Section 3.6) of τ1 and τ2
under Q are given by the expressions

λ1t = λ1 1{t<τ2 } + α1 1{t≥τ2 }

5.7. KUSUOKA’S MODEL 183

and
λ2t = λ2 1{t<τ1 } + α2 1{t≥τ1 } .
Rt
We focus on τ1 and we denote Λ1t = 0 λ1u du. Let us make few observations. First, we note that
the process λ1 is H2 -predictable and the process
Z t∧τ1
1 1
Mt = Ht − λ1u du = Ht1 − Λ1t∧τ1
0

is a G-martingale under Q, so that the process λ1 is a version of G-intensity of τ under Q. In general,

the process λ1 is not the H2 -intensity of τ1 under Q, since we have
¡ 1 1 ¢
Q(τ1 > s | Ht1 ∨ Ht2 ) 6= 1{t<τ1 } EQ eΛt −Λs | Ht2 .
It is also interesting to observe that the process λ1 is the H2 -intensity of τ1 under a probability
measure Q,e which is equivalent to P and is given by

e
dQ
= ηeT , P-a.s.,
dP
where the Radon-Nikodým density process (eηt , t ∈ [0, T ]) satisfies
Z
ηet = 1 + ηeu− κ2u dMu2 .
]0,t]

It can be checked that the following equality is satisfied, for every s > t,
¡ ¢
e 1 > s | H1 ∨ H2 ) = 1{t<τ } E e eΛ1t −Λ1s | H2 .
Q(τ t t 1 Q t

Recall that the processes λ1 and λ2 have jumps provided that αi 6= λi .

The next result shows that the processes λ1 and λ2 specify the transition intensities, so that the
model can be dealt with as a two-dimensional Markov chain (for related results and applications,
see Lando [108], Herbertsson [85], and Shaked and Shanthikumar [139]).

Proposition 5.7.1 For i = 1, 2 and every t ∈ R+ we have

λi = lim h−1 Q(t < τi ≤ t + h | τ1 > t, τ2 > t),
h↓0

α1 = lim h−1 Q(t < τ1 ≤ t + h | τ1 > t, τ2 ≤ t),

h↓0

α2 = lim h−1 Q(t < τ2 ≤ t + h | τ2 > t, τ1 ≤ t).

h↓0

5.7.2 Bonds with Zero Recovery

We now present the bond valuation result in Kusuoka’s [106] model. We focus on the bond price
D10 (t, T ); an analogous formula is valid for D20 (t, T ) as well. Recall that we consider corporate bonds
with zero recovery.

Proposition 5.7.2 The price D10 (t, T ) equals, on the event {t < τ1 },
1 ³ −α1 (T −t) ´
D10 (t, T ) = 1{t<τ2 } λ2 e + (λ1 − α1 )e−λ(T −t)
λ − α1
+ 1{t≥τ2 } e−α1 (T −t) ,
where λ = λ1 + λ2 . Furthermore,
−α1 T
b 10 (t, T ) = 1{t<τ } λ2 e + (λ1 − α1 )e−λT
D 1
λ2 e−α1 t + (λ1 − α1 )e−λt
184 CHAPTER 5. DEPENDENT DEFAULTS

and
λ − α2 (λ1 − α1 )e−λ(T −t) + λ2 e−α1 (T −t)
D̄10 (t, T ) = 1{t<τ2 }
λ − α1 λ1 e−(λ−α2 )t + λ2 − α2
−α1 (T −τ2 )
(λ − α2 )λ2 e
+ 1{t≥τ2 } .
λ1 α2 e(λ−α2 )τ2 + λ(λ2 − α2 )

It is worth noting that:

• the prices D10 (t, T ) and D20 (t, T ) correspond to the Jarrow and Yu price of the bond issued by
a secondary firm (cf. Corollary 5.6.2),
• the processes D10 (t, T ) and D b 0 (t, T ) represent ex-dividend prices of the bond issued by the first
1
firm and thus they vanish after the default time τ1 ,
• the second remark does not apply to the process D̄10 (t, T ).

5.8 Basket Credit Derivatives

We will now describe the mainstream basket credit derivatives, focusing in particular on the recently
developed standardized instruments, the credit default swap indices, and related contracts, such
as collateralized debt obligations and first-to-default swaps. For various methods of valuation and
hedging of basket credit products, we refer to Andersen and Sidenius [4], Cont and Minca [54], Cont
and Kan [53], Brasch [29], Brigo [34], Brigo and Alfonsi [36], Brigo and El-Bachir [37], Brigo and
Morini [38, 126], Brigo et al. [39], Burtschell et al. [41, 42], Di Graziano and Rogers [61], Duffie and
Gârleanu [63], Frey and Backhaus [76, 77], Giesecke and Goldberg [83], Herbertsson [86], Ho and
Wu [88], Hull and White [89, 90, 91], Laurent et al. [113], Laurent and Gregory [114], Pedersen [130],
Rutkowski and Armstrong [134], Sidenius et al. [140], Wu [143] and Zheng [145].

5.8.1 Credit Default Index Swaps

A credit default index swap (CDIS), also known as a CDS index, is a static portfolio of n equally
weighted credit default swaps with standard maturities, typically five or ten years. Standard ex-
amples of a CDIS are iTraxx and CDX. A credit default index swap usually matures few months
before the underlying CDSs. The CDSs in the pool are selected from among those with highest
trading volume in the respective industry sector. Credit default index swaps are issued by a pool of
licensed financial institutions, which are called the market makers. At time of issuance of a CDIS,
the market makers determine an annual rate, known as the index spread, to be paid out to investors
on a periodic basis. The index spread, denoted by κ0 , is constant over the lifetime of a CDIS. Let
us summarize the main provisions of a CDIS.
• We assume that the face value of each reference entity is one. Thus the total notional of a
CDIS equals n. The notional on which the market maker pays the spread, henceforth referred
to as residual protection, is reduced by 1 after each default. For instance, after the first default,
the residual protection is revised from the original value n to n − 1.
• By purchasing a CDIS, an investor assumes the role of a protection seller, so that the market
makers play the role of protection buyers. As the protection seller, an investor agrees to absorb
all losses due to defaults in the reference portfolio, occurring between the time of inception 0
and the maturity T . In case of default of a reference entity, an investor makes the protection
payment to a market maker in the amount of 1 − δ, where δ ∈ [0, 1] is a constant recovery rate,
which is pre-determined in a given CDIS (typically, it equals to 40%).
• In exchange, the protection seller receives from a market maker a periodic fixed premium on
the residual protection at the annual rate of κ0 , which equals the fair CDIS spread at the
inception date.
5.8. BASKET CREDIT DERIVATIVES 185

• A CDIS is also traded after its issuance date. Recall that whenever one of reference entities
defaults, its weight in the index is set to zero. Therefore, by purchasing one unit of an index
at time t, an investor owes protection only on those names that have not yet defaulted prior
to time t. If the quotation of the market CDIS spread at time t differs from the index spread
fixed at issuance, i.e., κt 6= κ0 , the credit-risky present value of the difference is settled through
an upfront payment.

The provisions of a single-name CDS correspond to the CDIS with n = 1, except for the fact
that, by the market convention, a buyer of a single-name CDS is the protection buyer, rather than
the protection seller.
We denote by τi the default time of the ith name in the index portfolio and by H i the process
defined as Hti = 1{t≥τi } for every i = 1, 2, . . . , n. Also, we set N0 = n and we write
n
X
Nt = N0 − Hti (5.4)
i=1

to denote the residual protection (or the reduced nominal) at time t ∈ [0, T ].
Let tj , j = 0, 1, . . . , J with t0 = 0 and tJ = T denote the tenor of the premium leg payments
dates. The discounted cumulative cash flows associated with a CDIS are as follows

B0 ³ ´
XJ Xn XJ
B0
Premium leg = κ0 N0 − Htij = κ0 Ntj
j=1
B tj i=1 j=1
Btj

and
Xn
B0 i
Protection leg = (1 − δ) H .
i=1
Bτi T

5.8.2 Collateralized Debt Obligations

Collateralized debt obligations (CDO) are credit derivatives backed by portfolios of assets. If the
underlying portfolio is made up of bonds, loans or other securitised receivables, the collateralized
debt obligation is known as the cash CDO. Alternatively, the underlying portfolio may consist of
credit derivatives referencing a pool of debt obligations. In the latter case, a CDO is said to be
synthetic.
Because of their recently acquired popularity, we focus our discussion on standardized synthetic
CDO contracts backed by CDS indices. We begin with an overview of the covenants of a typical
synthetic collateralized debt obligation.

• The time of issuance of the contract is 0 and its maturity is T . The notional of the CDO
contract at any date t after issuance is equal to the residual protection Nt of the reference
CDS index (cf. formula (5.4))
• The credit risk (that is, the potential loss due to credit events) borne by the reference pool is
layered into various standardized risk levels, with the range in between two adjacent risk levels
called a CDO tranche. The lower bound of a tranche is usually referred to as attachment point
and the upper bound as detachment point. The credit risk is originally sold in these tranches
to protection sellers. For instance, in a typical CDO contract on iTraxx, the credit risk is split
into the equity tranche (0 − 3% of the total losses), four mezzanine tranches (corresponding to
3 − 6%, 6 − 9%, 9 − 12% and 12 − 22% of the total losses respectively), and the senior tranche
(over 22% of the total losses). At issuance, the notional value of each tranche is equal to the
CDO notional weighted by the respective tranche width.
• The tranche buyer sells partial protection to the pool owner, by agreeing to absorb the pool’s
losses comprised in between the tranche attachment and detachment point. This is better
186 CHAPTER 5. DEPENDENT DEFAULTS

understood by an example. Assume, for instance, that at time 0 the protection seller purchases
the 6−9% tranche with a given notional value. One year later, consequently to a default event,
the cumulative loss breaks through the attachment point, reaching 8%. The protection seller
then fulfills his obligation by disbursing two thirds (= 8%−6%
9%−6% ) of a currency unit. The tranche
notional is then reduced to one third of its pre-default event value. We refer to the remaining
tranche notional as residual tranche protection.
• In exchange, as of time t and up to time T , the CDO issuer (protection buyer) makes periodic
payments to the tranche buyer according to a predetermined rate – termed tranche spread – on
the residual tranche protection. Returning to our example, after the loss reaches 8%, premium
payments are made on 13 (= 9%−8%
9%−6% ) of the tranche notional, until the next credit event occurs
or the contract matures.

We denote by Ll and Ul the lower and upper attachment points for the lth tranche and by κl0
the corresponding spread. It is convenient to introduce the percentage loss process
n
1−δ X i N0 − Nt
Qt = Ht = (1 − δ) ,
n i=1 N0

where N0 = n is the number of credit names in the reference portfolio and the residual protection
Nt is given by (5.4). Finally, denote by Cl = Ul − Ll the width of the lth tranche; in particular, for
the first (i.e., equity) tranche we have C1 = U1 since L1 = 0.
Purchasing one unit of the lth tranche at time 0 generates the following discounted cash flows

XJ
B0 l
Premium leg = κl0 Ntj ,
j=1
Btj

where Ntl is the residual tranche protection at time t, that is,

³ ¡ ¢´
Ntl = N0 Cl − min Cl , max (Qt − Ll , 0) .

The discounted cash flows of the protection leg are

Xn
B0 i
Protection leg = (1 − δ) H 1{Ll <Qτi ≤Ul } .
i=1
Bτi T

The equity tranche of the CDO on iTraxx or CDX is quoted differently; specifically, it is quoted in
terms if an upfront rate, say κ10 , on the total tranche notional, in addition to 500 basis points (5%
rate) paid annually on the residual tranche nominal. The discounted premium leg cash flows of the
equity tranche are thus given by the expression

XJ
B0 0
κ10 N0 C0 + .05 Ntj
j=1
B tj

or, more explicitly,

XJ
B0 ¡ ¢
κ10 nC0 + .05 n C0 − min (C0 , Qtj ) .
j=1
B tj

Additionally to standard traded tranches of a CDO, some non-standard tranches – commonly referred
to as bespoke tranches – are traded over-the-counter. Typically, a credit risk model is first calibrated
to market quotes for standard tranches, and subsequently it is used to value bespoke tranches.
5.8. BASKET CREDIT DERIVATIVES 187

5.8.3 First-to-Default Swaps

A kth-to-default swap is a basket credit instrument backed by a portfolio of single name CDSs. Due
to the rapid growth in popularity of credit default swap indices and the associated derivatives, the
kth-to-default swaps have become rather illiquid. Currently, such products are typically customized
contracts between a bank and its customer, and hence they are relatively bespoke to the customer’s
credit portfolio.
For this reason, in the sequel we focus our attention on first-to-default swaps issued on the iTraxx
index, which are the only ones with a certain degree of liquidity. Standardized first-to-default swaps
are now issued on each of the iTraxx sector sub-indices. Each first-to-default swap is backed by an
equally weighted portfolio of five single name CDSs in the relative sub-index, chosen according to
some liquidity criteria. Let us describe the main provisions of a first-to-default swap (FTDS).

• The time of issuance of the contract is 0 and the maturity is T .

• By investing in a first-to-default swap, the protection seller agrees to absorb the loss produced
by the first default in the reference credit portfolio.
• In exchange, the protection seller is paid a periodic premium, known as FTDS spread, up to
maturity T or the moment of the first default, whichever comes first. We denote the FTDS
spread at time 0 by κ0 .

Recall that by tj , j = 0, 1, . . . , J with t0 = 0 and tJ = T we denote the tenor of the premium leg
payments dates. As usual, we denote by τ(1) the random time of the first default in the pool. The
discounted cumulative cash flows associated with a first-to-default swap are as follows

XJ
B0
Premium leg = κ0 1{tj ≤τ(1) }
j=1
Btj

and
B0
Protection leg = (1 − δ) 1{τ(1) ≤T } .
Bτ(1)
It is worth stressing that the market convention stipulates that the notional corresponding to each
credit name in the reference credit portfolio is equal. Moreover, the recovery rate is assumed to be
constant, that is, the recovery rate does not depend on a particular credit name.

5.8.4 Step-up Corporate Bonds

As of now, step-up corporate bonds are not traded in baskets; however, they are of our interest since
they offer protection against credit events other than defaults, for instance, the downgrade of the
rating of the reference name.
Step-up corporate bonds are coupon-bearing bond issues for which the amounts of coupon pay-
ments depend on the credit quality of the bond’s issuer. As the name of the bond suggests, the
coupon payment increases when the credit quality of the issuer declines.
In practice, the above-mentioned credit quality is reflected by a credit rating assigned to an issuer
by at least one specialized ratings agency (such as: Moody’s KMV, Fitch, or Standard & Poor’s).
The provisions linking the cash flows of the step-up bonds to the credit rating of an issuer have
different step amounts and different rating event triggers. In some cases, a step-up of the coupon
requires a downgrade to the trigger level by both rating agencies. In other cases, there are step-up
triggers for actions of each rating agency. Under this specification, a downgrade by one of agencies
will trigger an increase in the coupon regardless of the rating from the other agency.
Provisions also vary with respect to step-down features which, as the name suggests, trigger a
lowering of the coupon if the company regains its original rating after a downgrade. In general, there
is no step-down below the initial coupon for ratings exceeding the initial rating.
188 CHAPTER 5. DEPENDENT DEFAULTS

Let Xt stand for some indicator of the issuer’s credit quality at time t. Assume that tj , j =
1, 2, . . . , J are coupon payment dates and denote by cj = c(Xtj−1 ) the coupon amount at time tj .
The time t discounted cumulative cash flows of a step-up bond are given by the expression
Z
Bt Bt
(1 − HT ) + (1 − Hu ) dCu + recovery payment
BT ]t,T ] Bu
P
where we denote by C the process given by the expression Ct = tj ≤t cj .

5.8.5 Valuation of Basket Credit Derivatives

Computation of the fair spread at time t for a basket credit derivative involves evaluating the
conditional expectation under the martingale measure Q of the associated discounted cash flows.
In the case of CDS indices, CDOs and FTDSs, the fair spread at time t is such that the value of
the contract at time t is exactly zero, i.e., the risk-neutral conditional expectations of discounted
cumulative cash flows of the premium and protection legs are identical.
The following expressions for fair spreads or values at time t ∈ [0, T ] can be easily derived from
PJ
the discounted cumulative cash flows given in the preceding subsections (note, however, that j=1
PJ
now stands for j=1, tj ≥t and we assume that the CDO tranches were issued at time 0):
• the fair spread of a single name CDS on the ith credit name
³ ¯ ´
¯
(1 − δi ) EQ BBτt (HTi − Hti ) ¯ Gt
i ³ i
¯ ´ ,
κt = PJ Bt
EQ (1 − H i )¯G
j=1 Bt tj ¯ t
j

• the fair spread of a CDIS

³P ¯ ´
n Bt i i ¯
(1 − δ) EQ i=1 Bτi (H T − H )
t ¯ Gt
κt = ³P ³ Pn ´ ¯ ´ ,
J Bt ¯
EQ j=1 Bt n − i=1 Htij ¯ Gt
j

• the fair spread of the lth tranche of a CDO

³P ¯ ´
n Bt i i ¯
(1 − δ) EQ i=1 Bτi (H T − H t )1{L l ≤Q τi
≤U l } ¯ Gt
l ³P ³ ´ ¯ ´,
κt = ¡ ¢ ¯
J Bt
EQ j=1 Bt n C l − min C l , max (Q t j
− L l , 0) ¯ Gt
j

• the fair upfront rate of the equity tranche of a CDO

1 ³ Xn
Bt ¯ ´
¯
κ1t = EQ (1 − δ) (HTi − Hti )1{Qτi ≤U0 } ¯ Gt
nC0 i=1
B τi

.05 ³XJ
Bt ¡ ¢ ¯¯ ´
− EQ n C0 − min (C0 , Qtj ) ¯ Gt ,
nC0 j=1
Btj
• the fair spread of a first-to-default swap
³ ¯ ´
¯
(1 − δ) EQ BBτ t 1{τ(1) ≤T } ¯ Gt
κt = ³P (1)
¯ ´ ,
J Bt ¯
EQ j=1 Bt 1{tj ≤τ(1) } ¯ Gt
j

• the fair value of a step-up corporate bond

³ Z ¯ ´
Bt Bt ¯
EQ (1 − HT ) + (1 − Hu ) dCu + recovery payment ¯ Gt .
BT ]t,T ] Bu

Depending on the dimensionality of the problem, the above conditional expectations will be
evaluated either by means of Monte Carlo simulation or through some other numerical method.
5.9. MODELING OF CREDIT RATINGS 189

5.9 Modeling of Credit Ratings

We will now give a brief description of a generic Markovian market model that can be efficiently
used for valuation and hedging basket credit instruments. The model presented below is a special
case of a general approach examined in Bielecki et al. [11]. Some preliminary empirical studies of
this model and its extensions are reported in Bielecki et al. [23, 24].
For related methods and models, the interested reader is referred to, e.g., Albanese and Chen [1],
Chen and Filipović [45], Frey and Backhaus [76, 77], Jarrow et al. [93], Kijima and Komoribayashi
[103], and Kijima et al. [104].
Let the underlying probability space be denoted by (Ω, G, G, Q), where Q is a risk-neutral measure
inferred from the market via calibration and G = H ∨ F is a filtration containing all information
available to market agents. The filtration H carries information about evolution of credit events,
such as changes in credit ratings or defaults of respective credit names. An additional filtration F
is called a reference filtration; it is meant to contain the information pertaining to the evolution of
relevant macroeconomic variables.
We consider n credit names and we assume that the credit quality of each reference entity falls to
the set K = {1, 2, . . . , K} of K rating categories, where, by convention, the category K corresponds
to default.
Let X i , i = 1, 2, . . . , n be some stochastic processes defined on (Ω, G, Q) and taking values in
the finite state space K, where the process X i represents the evolution of credit ratings of the ith
underlying entity. Then we define the default time τi of the ith credit name by setting

τi = inf { t ∈ R+ : Xti = K}.

We postulate that the default state K is absorbing, so that for each credit name the default event
can only occur once.
We denote by X = (X 1 , X 2 , . . . , X n ) the joint credit ratings process for a given portfolio of n
credit names. The state space of X is thus X = Kn and the elements of X will be denoted by x.
We postulate that the filtration H is the natural filtration of the process X, whereas the reference
filtration F is generated by an Rd -valued factor process Y , which represents the evolution of other
relevant economic variables, like interest rates or equity prices.

5.9.1 Infinitesimal Generator

Under the standing assumption that the factor process Y is Rd -valued, the state space for the process
M = (X, Y ) equals X × Rd . At the intuitive level, we wish to model the process M = (X, Y ) as a
combination of a Markov chain X modulated by a Lévy-type process Y and a Lévy-type process Y
modulated by a Markov chain X.
For this purpose, we start by making a general postulate that the infinitesimal generator A of
M is given by the expression
d d
1 X X
Af (x, y) = alm (x, y)∂l ∂m f (x, y) + bl (x, y)∂l f (x, y)
2
l,m=1 l=1
Z
¡ ¢
+ γ(x, y) f (x, y + g(x, y, y 0 )) − f (x, y) Π(x, y; dy 0 )
R d
X
+ λ(x, x0 ; y)f (x0 , y),
x0 ∈X

where λ(x, x0 ; y) ≥ 0 for every x = (x1 , x2 , . . . , xn ) 6= (x01 , x02 , . . . , x0n ) = x0 and

X
λ(x, x; y) = − λ(x, x0 ; y).
x0 ∈X , x0 6=x
190 CHAPTER 5. DEPENDENT DEFAULTS

Here ∂l denotes the partial derivative with respect to the variable yl . The existence and uniqueness
of a Markov process M with the generator A will follow (under appropriate technical conditions)
from the classic results regarding solutions to martingale problems.
We find it convenient to refer to X (Y , respectively) as the Markov chain component of M (the
jump-diffusion component of M , respectively). At any time t, the intensity matrix of the Markov
chain component is given as Λt = [λ(x, x0 ; Yt )]x,x0 ∈X . The jump-diffusion component satisfies the
SDE Z
dYt = b(Xt , Yt ) dt + σ(Xt , Yt ) dWt + g(Xt− , Yt− , y 0 ) π(Xt− , Yt− ; dy 0 , dt),
Rd
d 0
where, for any fixed (x, y) ∈ X × R , π(x, y; dy , dt) is a Poisson measure with the intensity measure
γ(x, y)Π(x, y; dy 0 )dt and σ(x, y) satisfies the equality σ(x, y)σ(x, y)T = a(x, y).

Remarks 5.9.1 If we take g(x, y, y 0 ) = y 0 and we suppose that the coefficients σ = [σij ], b = [bi ], γ
and the measure Π do not depend on x and y then the factor process Y is a Poisson-Lévy process
with the characteristic triplet (a, b, ν), where the diffusion matrix is a(x, y) = σ(x, y)σ(x, y)T , the
drift vector equals b(x, y) and the Lévy measure ν satisfies ν(dy) = γΠ(dy).

In order to proceed further with the analysis of the model, we need to provide with more structure
the Markov chain component of the infinitesimal generator A. To this end, we make the following
standing assumption.

Assumption (M). The infinitesimal generator of the process M = (X, Y ) has the following form
d d
1 X X
Af (x, y) = alm (x, y)∂l ∂m f (x, y) + bl (x, y)∂l f (x, y)
2
l,m=1 l=1
Z
¡ ¢
+ γ(x, y) f (x, y + g(x, y, y 0 )) − f (x, y) Π(x, y; dy 0 ) (5.5)
Rd
n X
X
+ λi (x, x0i ; y)f (x0i , y),
i=1 xi0 ∈K

where we use the shorthand notation x0i = (x1 , x2 , . . . , xi−1 , x0i , xi+1 , . . . , xn ). Note that x0i is simply
the vector x = (x1 , x2 , . . . , xn ) with the ith coordinate xi replaced by x0i .

In the case of two reference credit entities (that is, when n = 2), the infinitesimal generator A
becomes
d d
1 X X
Af (x, y) = alm (x, y)∂l ∂m f (x, y) + bl (x, y)∂l f (x, y)
2
l,m=1 l=1
Z
¡ ¢
+ γ(x, y) f (x, y + g(x, y, y 0 )) − f (x, y) Π(x, y; dy 0 )
Rd
X X
+ λ (x, x01 ; y)f (x01 , y) +
1
λ2 (x, x02 ; y)f (x02 , y),
x01 ∈K x02 ∈K

where x = (x1 , x2 ), x01 = (x01 , x2 ) and x02 = (x1 , x02 ). Returning to the general form, we have that,
for x = (x1 , x2 ) and x0 = (x01 , x02 ),
 1
 λ (x, x01 ; y), if x2 = x02 ,
0
λ(x, x ; y) = λ2 (x, x02 ; y), if x1 = x01 ,

0, otherwise.

Similar expressions can be derived for the case of an arbitrary number of underlying credit names.
Note that the model specified by (5.5) does not allow for simultaneous jumps of credit ratings X i
5.9. MODELING OF CREDIT RATINGS 191

0
and X i for i 6= i0 . This is not a serious lack of generality, however, since the ratings of both credit
names may still change in an arbitrarily small time interval. The advantage is that, for the purpose
of simulation of paths of process X, rather than dealing with K n × K n intensity matrix [λ(x, x0 ; y)],
it will be sufficient to deal with n intensity matrices [λi (x, x0i ; y)] of dimension K × K (for any fixed
y). Within the present setup, the current credit rating of the credit name i has a direct influence on
the level of the transition intensity for the current rating of the credit name i0 , and vice versa. This
property, known as frailty, is likely to contribute to the default contagion effect.

Remarks 5.9.2 (i) It is clear that we can incorporate in the model the case when at least some
components of the factor process Y follow Markov chains themselves. This feature is important, as
factors such as economic cycles may be modeled as Markov chains. It is known that default rates
are strongly related to business cycles.
(ii) Some of the factors Y 1 , Y 2 , . . . , Y d may represent cumulative duration of visits of processes X i
in particular rating states. For example, we may set
Z t
Yt1 = 1{Xu1 =1} du.
0

so that b1 (x, y) = 1{x1 =1} (x) and the corresponding components of coefficients σ and g equal zero.
(iii) In the area of structural arbitrage, the so-called credit–to–equity models and/or equity–to–credit
models are studied. The market model presented in this section nests both types of interactions.
For example, if one of the factors is the price process of the equity issued by a credit name, and
if credit migration intensities depend on this factor (either implicitly or explicitly), then we have a
equity–to–credit type interaction. If the credit rating of a given name impacts the equity dynamics
for this name (and/or some other names), then we deal with a credit–to–equity type interaction.
Pn
Let Hti = 1{t≥τi } for every i = 1, 2, . . . , n and let the process H be defined as Ht = i=1 Hti . It
can be observed that the process S = (H, X, Y ) is a Markov process on the state space {0, 1, . . . , n}×
X × Rd with respect to its natural filtration. Given the form of the infinitesimal generator of the
process (X, Y ), we can easily describe the infinitesimal generator of the process (H, X, Y ). To this
end, it is enough to observe that the transition intensity at time t of the component H from the
Pn (i)
state Ht to the state Ht + 1 is equal to i=1 λi (Xt , K; Xt , Yt ), provided that Ht < n (otherwise,
(i)
the transition intensity equals zero), where we write Xt = (Xt1 , . . . , Xti−1 , Xti+1 , . . . , Xtn ) and we
i 0 (i) i 0i
set λ (xi , xi ; x , y) = λ (x, x ; y).

5.9.2 Transition Intensities for Credit Ratings

One should always strive to find a right balance between the realistic features of a financial model
and its complexity. This issue frequently nests the issues of functional representation of a model, as
well as its parameterization. In what follows, we present an example of a particular model for credit
ratings, which is rather arbitrary, but is nevertheless relatively simple, and thus it should be easy
to estimate and/or calibrate.
Let X̄t be the average credit rating at time t, so that
n
1X i
X̄t = X .
n i=1 t

Let L = {i1 , i2 , . . . , inb } be a subset of the set of all credit names, where n
b < n. We consider L
to be a collection of “major players” whose economic situation, reflected by their credit ratings,
effectively impacts all other credit names in the pool. The following exponential-linear regression
model appears to be a plausible model for the ratings transitions intensities
d
X
ln λi (x, x0i ; y) = αi,0 (xi , x0i ) + αi,l (xi , x0i )yl + βi,0 (xi , x0i )h
l=1
192 CHAPTER 5. DEPENDENT DEFAULTS

n
b
X
+ βi,k (xi , x0i )xk + β̃i (xi , x0i )x̄ + βbi (xi , x0i )(xi − x0i ), (5.6)
k=1
Pn
where h represents a generic value of Ht , so that h = 1{K} (xi ). Similarly, x̄ stands for a
Pn i=1
generic value of X̄t , that is, x̄ = n1 i=1 xi .
The number of parameters involved in (5.6) can easily be controlled by the number of model
variables, in particular, the number of factors and the number of credit ratings, as well as structure
of the transition matrix (see Section 5.9.9 below). In addition, the reduction of the number of
parameters can be obtained if the pool of all n credit names is partitioned into a (small) number of
homogeneous sub-pools. All of this is a matter of a practical implementation of a specific Markovian
model of credit ratings.
Assume, for instance, that there are n e << n homogeneous sub-pools of credit names, and the
parameters α, β, βe and βb in (5.6) do not depend on xi , x0i . Then the migration intensities (5.6) are
parameterized by n e(d + n
b + 4) parameters.

5.9.3 Conditionally Independent Credit Migrations

Suppose that the transition intensities λi (x, x0i ; y) do not depend on the vector

x(i) = (x1 , x2 , . . . , xi−1 , xi+1 , . . . , xn )

for every i = 1, 2, . . . , n. In addition, assume that the dynamics of the factor process Y do not depend
on the migration process X. It turns out that in this case, given the structure of our generator as
in (5.5), the credit ratings processes X i , i = 1, 2, . . . , n, are conditionally independent given the
sample path of the factor process Y .
We shall illustrate this point in the case of only two credit names in the pool (i.e., for n = 2) and
assuming that there is no factor process, so that conditional independence really means independence
between migration processes X 1 and X 2 . For this, suppose that X 1 and X 2 are independent Markov
chains, each taking values in the state space K and with the infinitesimal generator matrices Λ1 and
Λ2 , respectively. It is clear that the joint process X = (X 1 , X 2 ) is a Markov chain on K × K. An
easy calculation reveals that the infinitesimal generator of the process X is given as

Λ = Λ1 ⊗ IdK + IdK ⊗ Λ2 ,

where IdK is the identity matrix of size K and ⊗ denotes the matrix tensor product. This result is
consistent with the structure (5.5) in the present case.

5.9.4 Examples of Markovian Models

We will now present three pertinent examples of Markovian market models.

Markov Chain Credit Ratings Process

In the first example, we assume that there is no factor process Y and thus we only deal with a ratings
migration process X. In this situation, an attractive and efficient way to model credit ratings is to
postulate that X is a birth-and-death process with absorption at state K. The intensity matrix Λ is
here tri-diagonal. Let us write pt (k, k 0 ) = Q(Xs+t = k 0 | Xs = k).
The transition probabilities pt (k, k 0 ) are known to satisfy the following system of ordinary dif-
ferential equations, for t ∈ R+ and k 0 = 1, 2, . . . , K,

dpt (1, k 0 )
= −λ(1, 2)pt (1, k 0 ) + λ(1, 2)pt (2, k 0 ),
dt
5.9. MODELING OF CREDIT RATINGS 193

dpt (k, k 0 )
= λ(k, k − 1)pt (k − 1, k 0 ) − (λ(k, k − 1) + λ(k, k + 1))pt (k, k 0 )
dt
+ λ(k, k + 1)pt (k + 1, k 0 )
for k = 2, 3, . . . , K − 1, whereas for k = K we simply have that
dpt (K, k 0 )
= 0,
dt
with the initial conditions p0 (k, k 0 ) = 1{k=k0 } . Once the transition intensities λ(k, k 0 ) are specified,
the above system can be easily solved. Note, in particular, that pt (K, k 0 ) = 0 for every t if k 0 6= K.
The advantage of this representation is that the number of parameters can be kept relatively small.
A more flexible credit ratings model is obtained if we allow for jumps to the default state K
from any other state. In that case, the intensity matrix is no longer tri-diagonal and the ordinary
differential equations for transition probabilities take the following form, for t ∈ R+ and k 0 =
1, 2, . . . , K,
dpt (1, k 0 )
= −(λ(1, 2)+λ(1, K))pt (1, k 0 )+λ(1, 2)pt (2, k 0 )+λ(1, K)pt (K, k 0 )
dt
dpt (k, k 0 )
= λ(k, k − 1)pt (k−1, k 0 ) − (λ(k, k−1) + λ(k, k+1))pt (k, k 0 )
dt
+ λ(k, K)pt (k, k 0 )+λ(k, k+1)pt (k+1, k 0 )+λ(k, K)pt (K, k 0 )
for k = 2, 3, . . . , K − 1 and for k = K
dpt (K, k 0 )
= 0,
dt
with initial conditions p0 (k, k 0 ) = 1{k=k0 } .

Remark 5.9.1 Some authors model migrations of credit ratings using a proxy diffusion, possibly
with a jump to default. The birth-and-death process with jumps to default furnishes a Markov chain
counterpart of such proxy diffusion models. The nice feature of the Markov chain model is that,
at least in principle, the credit ratings are here observable state variables, whereas in the case of a
proxy diffusion model they are not directly observable.

Diffusion-type Factor Process

We will now extend the Markov chain process by adding a factor process Y . We may postulate, for
instance, that the factor process follows a diffusion process and that the generator of the Markov
process M = (X, Y ) takes the following form
d d
1 X X
Af (x, y) = alm (x, y)∂l ∂m f (x, y) + bl (x, y)∂l f (x, y)
2
l,m=1 l=1
X
+ λ(x, x0 ; y)(f (x0 , y) − f (x, y)).
x0 ∈K, x0 6=x

Let φ(t, x, y, x0 , y 0 ) be the transition probability of M , specifically,

φ(t, x, y, x0 , y 0 ) dy 0 = Q(Xs+t = x0 , Ys+t ∈ dy 0 | Xs = x, Ys = y).
In order to determine the function φ, one needs to examine the Kolmogorov equation of the form
dv(s, x, y)
+ Av(s, x, y) = 0. (5.7)
ds
For the generator A of the present form, the corresponding equation (5.7) is commonly known as
the reaction-diffusion equation (see, for instance, Becherer and Schweizer [9]). Let us mention that
a reaction-diffusion equation is a special case of a more general integro-partial-differential equation,
which was extensively studied in the mathematical literature.
194 CHAPTER 5. DEPENDENT DEFAULTS

Forward CDS Spread Model

Suppose now that the factor process Yt = κ(t, TS , TM ) is the forward CDS spread (for the definition
of κ(t, TS , TM ), see Section 5.9.6 below). We now postulate that the generator of M = (X, Y ) is
1 d2 f (x, y) X
Af (x, y) = y 2 a(x) 2
+ λ(x, x0 )(f (x0 , y) − f (x, y)),
2 dy 0 0
x ∈K, x 6=x

so that the forward CDS spread process satisfies the following SDE
dκ(t, TS , TM ) = κ(t, TS , TM )σ(Xt ) dWt
p
for some Brownian motion process W , where σ(x) = a(x). Note that in this example κ(t, TS , TM )
is a conditionally log-Gaussian process given a particular sample path of the credit ratings process
X. Therefore, we are in a position to make use of Proposition 5.9.1 below to value a credit default
swaption.

5.9.5 Forward Credit Default Swap

Let us first examine two examples of a single-name credit derivative. We assume that the reference
asset is a corporate bond maturing at time U and we consider a forward CDS with the maturity
date TM < U and the start date TS < TM . If default occurs prior to or at time TS the contract
is terminated with no exchange of payments. Therefore, the two legs of this CDS are manifestly
TS -survival claims and thus the valuation of a forward CDS is not much different from valuation of
a spot CDS.

Protection Leg

Assume that the notional amount of the bond equals 1 and denote by δ a deterministic recovery rate
in case of default. Under the assumption that the recovery is paid at default time τ of the reference
credit name, the value at time t of the protection leg of a forward CDS is equal to, for every t ≤ TS ,
¡ ¢
Pt = P (t, TS , TM ) = (1 − δ)Bt EQ 1{TS <τ ≤TM } Bτ−1 | Mt .
The valuation of the protection leg relies on computation of this conditional expectation for a given
term structure model. In particular, if the savings account B is a deterministic function of time
then the computation reduces to the following integration
Z TM
¡ ¢
Bt EQ 1{TS <τ ≤TM } Bτ−1 | Mt = Bt Bu−1 Q(τ ∈ du | Mt ).
TS

Premium Leg

Let us denote by t1 < t2 < . . . < tJ the tenor of premium payments, where TS < t1 < · · · < tJ ≤ TM .
We assume that the premium accrual covenant is in force, so that the cash flows associated with the
premium leg are
³X J XJ
t − tj−1 ´
κ 1{tj <τ } 1tj (t) + 1{tj−1 <τ ≤tj } 1τ (t) .
j=1 j=1
tj − tj−1
where κ is the fixed CDS spread. Consequently, the value at time t ∈ [0, TS ] of the premium leg
equals κAt , where At = A(t, TS , TM ) equals
³ XJ h
Bt Bt τ − tj−1 i¯¯ ´
At = EQ 1{TS <τ } 1{tj <τ } + 1{tj−1 <τ ≤tj } ¯ Mt .
j=1
Btj Bτ tj − tj−1

Under the assumption that B is deterministic and the conditional distribution Q(τ ≤ s | Mt ) is
known, this conditional expectation can be evaluated.
5.9. MODELING OF CREDIT RATINGS 195

5.9.6 Credit Default Swaptions

We consider a forward CDS swap starting at TS and maturing at TM > TS , as described in the
previous section. Our next goal is to examine valuation of the corresponding credit default swaption
with expiry date T < TS and the strike CDS spread K. The swaption’s payoff at its expiry date T
equals
¡ ¢+
PT − KAT ,
and thus the swaption’s price equals, for every t ∈ [0, T ],
³ ¡ ¢+ ¯¯ ´
Bt EQ BT−1 PT − KAT ¯ Mt
³ ¡ ¢+ ¯¯ ´
= Bt EQ BT−1 AT κ(T, TS , TM ) − K ¯ Mt ,

where the process κ(t, TS , TM ) = Pt /At , t ∈ [0, TS ], represents the forward CDS spread.
Note that the random variables Pt and At are strictly positive on the event {τ > T } for t ≤ T <
TS and thus the forward CDS spread κ(t, TS , TM ) enjoys this property as well.

Conditionally Gaussian Case

In order to provide a more explicit representation for the value of a CDS swaption, we assume that
B is deterministic and the forward CDS spread is conditionally log-Gaussian under Q. It is worth
recalling that an example of such a model was presented in Section 5.9.4.

Proposition 5.9.1 Suppose that, on the event {τ > T } and for arbitrary t < t1 < · · · < tk ≤ T ,
the conditional distribution
³ ¯ ´
¯
Q κ(tm , TS , TM ) ≤ km , m = 1, 2, . . . , k ¯ σ(Mt ) ∨ FTX

is lognormal, Q-a.s. Let us denote by σ(u, TS , TM ), u ∈ [t, T ], the conditional volatility of the process
κ(u, TS , TM ), u ∈ [t, T ], with respect to the σ-field σ(Mt ) ∨ FTX . Then the price of a CDS swaption
equals, for every t ∈ [0, T ],
³ ¡ ¢+ ¯¯ ´
Bt EQ BT−1 PT − KAT ¯ Mt
³ h i¯ ´
¯
= Bt EQ 1{τ >T } AT BT−1 κt N (d+ (t, T )) − KN (d− (t, T )) ¯ Mt ,

where we denote κt = κ(t, TS , TM ),

ln κKt υt,T
d± (t, T ) = ± ,
υt,T 2

and Z T
2
υt,T = υ(t, T, TS , TM )2 = σ(u, TS , TM )2 du.
t

Proof. We start be noting that

³ ¡ ¢+ ¯¯ ´
Bt EQ BT−1 PT − KAT ¯ Mt
³ ¡ ¢+ ¯¯ ´
= Bt EQ 1{τ >T } BT−1 PT − KAT ¯ Mt
³ ³¡ ¢+ ´¯ ´
¯
= Bt EQ 1{τ >T } BT−1 EQ PT − KAT | σ(Mt ) ∨ FTX ¯ Mt
³ ³¡ ¢+ ´¯ ´
¯
= Bt EQ 1{τ >T } AT BT−1 EQ κT − K | σ(Mt ) ∨ FTX ¯ Mt .
196 CHAPTER 5. DEPENDENT DEFAULTS

In view of the present assumptions, we also have that

³¡ ¢+ ¯¯ ´
EQ κT − K ¯ σ(Mt ) ∨ FTX
µ κt ¶ µ κt ¶
ln K υt,T ln K υt,T
= κt N + − KN − .
υt,T 2 υt,T 2

By combining the above equalities, we arrive at the stated formula. ¤

5.9.7 Spot kth-to-Default Credit Swap

Let us now examine the valuation of credit derivatives with several underlying credit names within
the present framework. Feasibility of closed-form computations of relevant conditional expectations
depends to a large extent on the type and amount of information one wishes to utilize. Typically, in
order to efficiently deal with exact calculations of conditional expectations, one will need to amend
specifications of the underlying model so that information used in calculations is given by a coarser
filtration, or perhaps by some proxy filtration.
In this subsection, we will discuss the valuation of a generic kth-to-default credit swap relative to
a portfolio of n reference corporate bonds. The deterministic notional value of the ith constituent
bond is denoted by Ni and the corresponding deterministic recovery rate equals δi .
The maturities of the bonds in the portfolio are T1 , T2 , . . . , Tn , whereas the maturity of the swap
is TM < min {T1 , T2 , . . . , Tn }. Let us consider, for instance, a plain-vanilla basket CDS written on
such a portfolio of corporate bonds under the convention of the fractional recovery of par value.
This means that, on the event {τ(k) < TM }, the protection buyer receives at time τ(k) the
cumulative compensation X
(1 − δi )Ni ,
i∈Lk

where Lk is the (random) set of all constituent credit names that defaulted in the time interval
]0, τ(k) ]. This means that the protection buyer is protected against the cumulative effect of the first
k defaults. Recall that, in view of the model assumptions, the possibility of simultaneous defaults
is excluded.

Protection Leg

The cash flows associated with the protection leg are given by the expression
X
(1 − δi )Ni 1{τ(k) ≤TM } 1τ(k) (t),
i∈Lk

so that the value at time t of the protection payment leg is equal to

³ X ¯ ´
(k) ¯
Pt = P (k) (t, TM ) = Bt EQ 1{t<τ(k) ≤TM } Bτ−1
(k)
(1 − δi )Ni ¯ Mt .
i∈Lk

In general, this conditional expectation will need to be evaluated numerically by means of Monte
Carlo simulations.
A special case of a kth-to-default credit swap is when the protection buyer is protected against
losses associated with the kth default only. In that case, the cash flow associated with the default
protection leg is given by the expression
n
X
(1 − δι(k) )Nι(k) 1{τ(k) ≤TM } 1τ(k) (t) = (1 − δi )Ni 1{Hτi =k} 1{τi ≤TM } 1τi (t),
i=1
5.9. MODELING OF CREDIT RATINGS 197

where ι(k) stands for the identity of the kth defaulting credit name. Under the assumption that the
numéraire process B is deterministic, we can represent the value at time t of the protection leg as
the following conditional expectation
n
X ³ ¯ ´
(k) ¯
Pt = Bt EQ 1{t<τi ≤TM } 1{Hτi =k} Bτ−1
i
(1 − δi )N i ¯ Mt
i=1
n
X Z TM
= Bt (1 − δi )Ni Bu−1 Q(Hu = k | τi = u, Mt ) Q(τi ∈ du | Mt ).
i=1 t

Note also that the conditional probability Q(Hu = k | τi = u, Mt ) can be approximated by the
following expression
i
Q(Hu = k, Xu−² 6= K, Xui = K | Mt )
Q(Hu = k | τi = u, Mt ) ≈ i
.
Q(Xu−² 6= K, Xui = K | Mt )
Therefore, if the number n of credit names is small, so that the Kolmogorov equations for the
(k)
conditional distribution of the process (H, X, Y ) can be solved, the value of Pt can be approximated
analytically.

Premium Leg

Let t1 < t2 < . . . < tJ denote the tenor of premium payments, where 0 = t0 < t1 < · · · < tJ < TM .
Under the assumption that the premium accrual covenant is in force, the cash flows associated with
the premium leg of the kth-to-default CDS admit the following representation
µXJ XJ ¶
t − tj−1
κ(k) 1{tj <τ(k) } 1tj (t) + 1{tj−1 <τ(k) ≤tj } 1τ(k) (t) ,
j=1 j=1
tj − tj−1

where κ(k) is the fixed spread of the kth-to-default CDS. Consequently, the value at time t of the
(k)
premium leg equals κ(k) At , where
µ J
X ¯ ¶
(k) (k) Bt ¯
At =A (t, TM ) = EQ 1{t<τ(k) } 1{tj <τ(k) } ¯ Mt
B tj
j=j(t)
µ ¶
τ(k) − tj−1 ¯¯
J
X Bt
+ EQ 1{t<τ(k) } 1{tj−1 <τ(k)} ≤tj } ¯ Mt ,
Bτ(k) tj − tj−1
j=j(t)

where j(t) is the smallest integer such that tj(t) > t. Again, in general, the above conditional
expectation will need to be approximated by simulation. And again, for a small portfolio size n,
if either exact or a numerical solution of relevant Kolmogorov equations can be derived, then an
analytical computation of the expectation can be done, at least in principle.

5.9.8 Forward kth-to-Default Credit Swap

A forward kth-to-default credit swap has an analogous structure to a forward CDS. The notation
used here is consistent with the notation that was introduced in Sections 5.9.5 and 5.9.7.

Protection Leg

The cash flow associated with the protection leg of a forward kth-to-default credit swap can be
expressed as follows X
(1 − δi )Ni 1{TS <τ(k) ≤TM } 1τ(k) (t).
i∈Lk
198 CHAPTER 5. DEPENDENT DEFAULTS

Consequently, the value of the protection leg equals, for every t ∈ [0, TS ],
³ X ¯ ´
(k) ¯
Pt = P (k) (t, TS , TM ) = Bt EQ 1{TS <τ(k) ≤TM } Bτ−1
(k)
(1 − δi )Ni ¯ Mt .
i∈Lk

Premium Leg

As before, let t1 < t2 < . . . < tJ be the tenor of premium payments, where TS < t1 < · · · < tJ < TM .
Under the premium accrual covenant, the cash flows associated with the premium leg are
µX
J J
X ¶
(k) t − tj−1
κ 1{tj <τ(k) } 1tj (t) + 1{tj−1 <τ(k) ≤tj } 1τ(k) (t) ,
j=1 j=1
tj − tj−1

(k)
where κ(k) is the fixed spread. Thus, the value at time t of the premium leg is κ(k) At , where the
(k)
random variable At = A(k) (t, TS , TM ) is given by the expression
µ ¶
hX τ − tj−1 i ¯¯
J XJ
Bt Bt
EQ 1{t<τ(k) } 1{tj <τ } + 1{tj−1 <τ(k) ≤tj } ¯ Mt .
j=1
B tj j=1
Bτ tj − tj−1

We have only presented here two examples of credit derivatives with several reference credit
names. Computations of arbitrage prices and fair spreads for other examples of basket credit deriv-
ative involve evaluating the conditional expectations presented in Section 5.8.5. The choice of a
particular model for the valuation of a given class of basket credit derivatives should be motivated
by arguments regarding its practical relevance as well as its mathematical tractability. In the next
section, we will examine some issues arising in this context.

5.9.9 Model Implementation

Let us now briefly discuss some issues related to the model implementation. As already mentioned,
when one deals with basket products involving several reference credit names, direct computations
may not be feasible, since the cardinality of the state space K for the migration process X is
equal to K n . Thus, for example, in case of K = 18 rating categories, as in Moody’s ratings,1
and in case of a portfolio of n = 100 credit names, the cardinality of the state space K equals
18100 . If one aims at closed-form expressions for conditional expectations, but K is large, then,
typically, it will be infeasible to work directly with information provided by the state vector (X, Y ) =
(X 1 , X 2 , . . . , X n , Y ) and with the corresponding infinitesimal generator A. An essential reduction in
the amount of information that can be effectively used for analytical computations will be required.
This goal can be achieved by reducing the number of rating categories; this is typically done by
considering only two categories: pre-default and default. However, this reduction may still not be
sufficient enough in some circumstances and thus further simplifying structural modifications to the
model may need to be called for. Some types of additional modifications – such as: homogeneous
grouping of credit names and mean-field interactions between credit names – were proposed in the
financial literature to address this important issue.

Recursive Simulation Procedure

When closed-form computations are not feasible, but one does not want to give up on potentially
available information, an alternative may be to carry approximate calculations by means of either
approximating some involved formulae and/or by simulating sample paths of underlying random
processes. We will briefly examine the Monte Carlo simulations approach.
1 We refer here to the following rating categories attributed by Moody’s: Aaa, Aa1, Aa2, Aa3, A1, A2, A3, Baa1,

Baa2, Baa3, Ba1, Ba2, Ba3, B1, B2, B3, Caa, D(efault).
5.9. MODELING OF CREDIT RATINGS 199

In general, a simulation of the evolution of the process X will be infeasible, due to the curse
of dimensionality. However, by virtue of the postulated structure of the infinitesimal generator A
(see (5.5)), a simulation of the evolution of the process X reduces to a recursive simulation of the
evolution of processes X i , whose state spaces are only of size K each. To facilitate simulations even
further, we also postulate that each migration process X i behaves like a birth-and-death process
with absorption at default and with possible jumps to default from every intermediate state (see
Section 5.9.4).
(i)
Recall that we denote Xt = (Xt1 , . . . , Xti−1 , Xti+1 , . . . , Xtn ).
Given the state (x(i) , y) of the process (X (i) , Y ), the intensity matrix of the ith migration process
is sub-stochastic and is given as

 1 2 ··· K −1 K 
1 λi (1, 1) λi (1, 2) ··· 0 λi (1, K)
2  λi (2, 1) λi (2, 2) ··· 0 λi (2, K)
 
3  0 λi (3, 2) ··· 0 λi (3, K)
 
..  .. .. .. .. , ..
.  . . . .  .
 
K −1 0 0 ··· λi (K − 1, K − 1) λi (K − 1, K) 
K 0 0 ··· 0 0

where we use the shorthand notation λi (xi , x0i ) = λi (x, x0i ; y).
Also, we find it convenient to write λi (xi , x0i ; x(i) , y) = λi (x, x0i ; y) in what follows. Then the
diagonal elements are given as follows, for xi 6= K,

λi (x, x; y) = −λi (xi , xi − 1; x(i) , y) − λi (xi , xi + 1; x(i) , y) − λi (xi , K; x(i) , y)

X³ ´
− λl (xl , xl − 1; x(l) , y) + λl (xl , xl + 1; x(l) , y) + λl (xl , K; x(l) , y)
l6=i

with the convention that λi (1, 0; x(i) , y) = 0 for every i = 1, 2, . . . , n.

It is implicit in the above description that λi (K, xi ; x(i) , y) = 0 for any i = 1, 2, . . . , n and
xi = 1, 2, . . . , K. Suppose now that the current state of the process (X, Y ) is (x, y). Then the
intensity of a jump of the process X equals
n
X
λ(x, y) := − λi (x, x; y).
i=1

Conditional on the occurrence of a jump of X, the probability distribution of a jump for the com-
ponent X i , i = 1, 2, . . . , n, is given as follows:

• the probability of a jump from xi to xi − 1 equals

λi (xi , xi − 1; x(i) , y)
pi (xi , xi − 1; x(i) , y) = ,
λ(x, y)

• the probability of a jump from xi to xi + 1 equals

λi (xi , xi + 1; x(i) , y)
pi (xi , xi + 1; x(i) , y) = ,
λ(x, y)

• the probability of a jump from xi to K equals

λi (xi , K; x(i) , y)
pi (xi , K; x(i) , y) = .
λ(x, y)
200 CHAPTER 5. DEPENDENT DEFAULTS

As expected, the following equality is valid

n
X ¡ ¢
pi (xi , xi − 1; x(i) , y) + pi (xi , xi + 1; x(i) , y) + pi (xi , K; x(i) , y) = 1.
i=1

For a generic state x = (x1 , x2 , . . . , xn ) of the migration process X, we define the jump space
n
[
J (x) = {(xi − 1, i), (xi + 1, i), (K, i)}
i=1

with the convention that (K + 1, i) = (K, i), where the shorthand notation (a, i) refers to the
ith component of X. Given that the process (X, Y ) is in the state (x, y) and conditional on the
occurrence of a jump of X, the process X jumps to a point in the space J (x) according to the
probability distribution denoted by p(x, y) and determined by the probabilities pi described above.
Thus, if a random variable ζ has the distribution given by p(x, y) then we have that, for any
(x0i , i) ∈ J (x),
Q(ζ = (x0i , i)) = pi (xi , x0i ; x(i) , y).

Simulation Algorithm

We conclude this section by presenting in some detail the simulation algorithm for the case when
the dynamics of the factor process Y do not depend on the credit ratings process X. The general
case appears to be much harder.
Under the assumption that the dynamics of the factor process Y do not depend on the process
X, the simulation procedure splits into two steps. In Step 1, a sample path of the process Y is
simulated; then, in Step 2, for a given sample path Y , a sample path of the process X is simulated.
We consider here simulations of sample paths over some generic time interval, say [t1 , t2 ], where
0 ≤ t1 < t2 . We assume that the number of defaulted names at time t1 is less than k, that is
Ht1 < k. We conduct the simulation either until the kth default occurs or until time t2 , depending
on whichever occurs first.

Step 1: The dynamics of the factor process are now given by the SDE
Z
dYt = b(Yt ) dt + σ(Yt ) dWt + g(Yt− , y)π(Yt− ; dy, dt), t ∈ [t1 , t2 ].
Rd

Any standard procedure can be used to simulate a sample path of Y . Let us denote by Yb the
simulated sample path of Y .

Step 2: Once a sample path of Y has been simulated, simulate a sample path of X on the interval
[t1 , t2 ] until the kth default time.

We exploit the fact that, according to our assumptions about the infinitesimal generator A,
the components of the credit ratings process X do not have simultaneous jumps. Therefore, the
following algorithm for simulating the evolution of X appears to be feasible:

Step 2.1: Set the counter m = 1 and simulate the first jump time of the process X in the time
interval [t1 , t2 ]. Towards this end, simulate first a value, say ηb1 , of a unit exponential random
variable η1 . The simulated value of the first jump time, τ1X , is then given as
n Z t o
τb1X = inf t ∈ [t1 , t2 ] : λ(Xt1 , Ybu ) du ≥ ηb1 ,
t1

where by convention the infimum over an empty set is +∞. If τb1X = +∞, set the simulated
value of the kth default time to be τb(k) = +∞, stop the current run of the simulation procedure
and go to Step 3. Otherwise, go to Step 2.2.
5.9. MODELING OF CREDIT RATINGS 201

Step 2.2: Simulate the jump of X at time τb1X by drawing from the distribution p(Xt1 , Ybτb1X − ) (see
bτbX , as well as the
the discussion in Section 5.9.9). In this way, one obtains a simulated value X 1

simulated value of the number of defaults H b τbX < k then let m := m + 1 and go to
b τbX . If H
1 1
X
Step 2.3; otherwise, set τb(k) = τb1 and go to Step 3.
Step 2.3: Simulate the mth jump of process X. Towards this end, simulate a value, say ηbm , of
X
a unit exponential random variable ηm . The simulated value of the mth jump time τm is
obtained from the formula
n Z t o
X
τbm = inf t ∈ [bX
τm−1 , t2 ] : λ(Xτbm−1
X , Ybu ) du ≥ ηbm .
X
τbm−1

X
In case τbm = +∞, let the simulated value of the kth default time to be τb(k) = +∞; stop the
current run of the simulation procedure and go to Step 3. Otherwise, go to Step 2.4.

Step 2.4: Simulate the jump of X at time τbm X

by drawing from the distribution p(Xτbm−1
X , Ybτbm
X − ).

b
In this way, produce a simulated value Xτbm X , as well as the simulated value of the number of

b
defaults Hτbm b
X . If Hτ X < k, let m := m + 1 and go to Step 2.3; otherwise, set τ
bm
X
b(k) = τbm and go
to Step 3.

Step 3: Calculate a simulated value of a relevant functional. For example, in case of the kth-to-
default CDS, compute X
(k)
Pbt1 = 1{t1 <bτ(k) ≤T } B
bt B
1
b −1
τb(k) (1 − δi )Ni
bk
i∈L

and
XJ bt XJ bt
b(k) B 1 B 1
τb(k) − tj−1
A t1 = 1{tj <bτ(k) } + 1{tj−1 <bτ(k) ≤tj } ,
b
Btj b
B tj − tj−1
j=j(t1 ) j=j(t1 ) τb(k)

where, as before, the ‘hat’ indicates that we deal with simulated values.

Concluding Remarks

The issue of evaluating functionals associated with multiple credit migrations is prominent with
regard to measuring and managing of portfolio credit risk. In some segments of the credit derivatives
market, only the deterioration of the value of a portfolio of debts (bonds or loans) due to defaults
is essential. For instance, such is the situation regarding the tranches of both cash and synthetic
collateralized debt obligations, as well as the tranches of traded credit default swap indices, such as:
CDX and iTraxx.
It is rather apparent, however, that a valuation model reflecting the possibility of intermediate
credit migrations through other ratings classes, and not only defaults, is called for in order to better
account for changes in creditworthiness of the reference credit entities. Likewise, for the purpose of
managing risks of a debt portfolio, it is necessary to account for changes in value of the portfolio due
to intertemporal variations in credit ratings of constituent credit names. Needless to say that these
issues are currently intensively studied by academics and practitioners alike, and new approaches
are proposed and analyzed in financial literature.
202 CHAPTER 5. DEPENDENT DEFAULTS
Appendix A

Poisson Processes

In some credit risk models, we need to model a sequence of successive random times. This can be
done by making use of the F-conditional Poisson process, which is also known as the doubly stochastic
Poisson process. The general idea is quite similar to the canonical construction of a single random
time. We start by assuming that we are given a stochastic process Λ, to be interpreted as the hazard
process, and we construct a jump process, with unit jump size, such that the probabilistic features
of jump times are governed by the hazard process Λ.

A.1 Standard Poisson Process

Let us first recall the definition and the basic properties of the Poisson process N with a constant
intensity λ > 0.

Definition A.1.1 A process N defined on a probability space (Ω, G, P) is called the (standard)
Poisson process with intensity λ with respect to the filtration G if N0 = 0 and for any 0 ≤ s < t the
following two conditions are satisfied:
(i) the increment Nt − Ns is independent of the σ-field Gs ,
(ii) the increment Nt −Ns has the Poisson law with parameter λ(t−s); specifically, for any k = 0, 1, . . .
we have
λk (t − s)k −λ(t−s)
P(Nt − Ns = k | Gs ) = P(Nt − Ns = k) = e .
k!

The Poisson process of Definition A.1.1 is termed time-homogeneous, since the probability law of
the increment Nt+h − Ns+h is invariant with respect to the shift h ≥ −s. In particular, for arbitrary
s < t the probability law of the increment Nt − Ns coincides with the law of the random variable
Nt−s . Let us finally observe that, for every 0 ≤ s < t,

EP (Nt − Ns | Gs ) = EP (Nt − Ns ) = λ(t − s). (A.1)

It is standard to take a version of the Poisson process whose sample paths are, with probability
1, right-continuous stepwise functions with all jumps of size 1. Let us set τ0 = 0 and let us denote
by τ1 , τ2 , . . . the G-stopping times given as the random moments of the successive jumps of N . For
any k = 0, 1, . . .

τk+1 = inf { t > τk : Nt 6= Nτk } = inf { t > τk : Nt − Nτk = 1}.

One shows without difficulties that P( limk→∞ τk = ∞) = 1. It is convenient to introduce the

sequence (ξk , k ∈ N) of non-negative random variables, where ξk = τk − τk−1 for every k ∈ N. Let
us quote the following well known result.

203
204 APPENDIX A. POISSON PROCESSES

Proposition A.1.1 The random variables ξk , k ∈ N are mutually independent and identically dis-
tributed, with the exponential law with parameter λ, that is, for any k ∈ N we have, for every
t ∈ R+ ,
P(ξk ≤ t) = P(τk − τk ≤ t) = 1 − e−λt .

Proposition A.1.1 suggests a simple construction of a process N , which follows a time-homogeneous

Poisson process with respect to its natural filtration FN . Suppose that the probability space (Ω, G, P)
is large enough to support a family of mutually independent random variables ξk , k ∈ N with the
common exponential law with parameter λ > 0. We define the process N on (Ω, G, P) by setting
Nt = 0 if {t < ξ1 } and, for any natural k,
k
X k+1
X
Nt = k if and only if ξi ≤ t < ξi .
i=1 i=1

It can checked that the process N defined in this way is indeed a Poisson process with parameter
N
λ, with
Pkrespect to its natural filtration F . The jump times of N are, of course, the random times
τk = i=1 ξi , k ∈ N.
Let us recall some useful equalities that are not hard to establish through elementary calculations
involving the Poisson law. For any a ∈ R and every 0 ≤ s < t we have
¡ ¯ ¢ ¡ ¢ ia
EP eia(Nt −Ns ) ¯ Gs = EP eia(Nt −Ns ) = eλ(t−s)(e −1)

and ¡ ¯ ¢ ¡ ¢ a
EP ea(Nt −Ns ) ¯ Gs = EP ea(Nt −Ns ) = eλ(t−s)(e −1) .
The next result is an easy consequence of (A.1) and the above formulae. The proof of the proposition
is thus left to the reader.

Proposition A.1.2 The following stochastic processes are G-martingales.

b defined as
(i) The compensated Poisson process N

bt = Nt − λt.
N

(ii) For any k ∈ N, the compensated Poisson process stopped at τk

ctk = Nt∧τ − λ(t ∧ τk ).

M k

(iii) For any a ∈ R, the exponential martingale M a given by the formula

a b a
Mta = eaNt −λt(e −1)
= eaNt −λt(e −a−1)
.

(iv) For any fixed a ∈ R, the exponential martingale K a given by the formula
ia b ia
Kta = eiaNt −λt(e −1)
= eiaNt −λt(e −ia−1)
.

Remark A.1.1 (i) For any G-martingale M , defined on some filtered probability space (Ω, G, P),
and an arbitrary G-stopping time τ , the stopped process Mtτ = Mt∧τ is necessarily a G-martingale.
Thus, the second statement of the proposition is an immediate consequence of the first, combined
with the simple observation that each jump time τk is a G-stopping time.
(ii) Consider the random time τ = τ1 , where τ1 is the time of the first jump of the Poisson process
N . Then Nt∧τ = Nt∧τ1 = Ht , so that the process M c1 introduced in part (ii) of the proposition
c
coincides with the martingale M associated with τ .
(iii) The property described in part (iii) of Proposition A.1.2 characterizes the Poisson process in
the following sense: if N0 = 0 and for every a ∈ R the process M a is a G-martingale, then N
A.1. STANDARD POISSON PROCESS 205

is the Poisson process with parameter λ. Indeed, the martingale property of M a yields, for every
0 ≤ s < t, ¡ ¯ ¢ a
EP ea(Nt −Ns ) ¯ Gs = eλ(t−s)(e −1) .
By standard arguments, this implies that the random variable Nt − Ns is independent of the σ-field
Gs and has the Poisson law with parameter λ(t − s). A similar remark applies to property (iv) in
Proposition A.1.2.

Let us consider the case of a Brownian motion W and a Poisson process N that are defined
on a common filtered probability space (Ω, G, P). In particular, for every 0 ≤ s < t, the increment
Wt − Ws is independent of the σ-field Gs and has the Gaussian law N (0, t − s).
It might be useful to recall that for any real number b the following processes follow martingales
with respect to G:
ct = Wt − t, mb = ebWt − 21 b2 t , ktb = eibWt + 12 b2 t .
W t

Proposition A.1.3 Let a Brownian motion W with respect to G and a Poisson process N with
respect to G be defined on a common probability space (Ω, G, P). Then the processes W and N are
mutually independent.

Proof. Let us sketch the proof. For a fixed a ∈ R and any t > 0, we have
X Z
eiaNt = 1 + (eiaNt − eiaNt− ) = 1 + (eia − 1)eiaNu− dNu ,
0<u≤t ]0,t]
Z Z t
= 1+ bu + λ
(eia − 1)eiaNu− dN (eia − 1)eiaNu− du.
]0,t] 0

On the other hand, for any b ∈ R, the Itô formula yields

Z t Z
ibWt ibWu 1 2 t ibWu
e = 1 + ib e dWu − b e du.
0 2 0

The continuous martingale part of the compensated Poisson process N b is identically equal to 0 (since
b is a process of finite variation), and obviously the processes N
N b and W have no common jumps.
Thus, using the Itô product rule for semimartingales, we obtain
Z t Z t
1
ei(aNt +bWt ) = 1 + ib ei(aNu +bWu ) dWu − b2 ei(aNu +bWu ) du
0 2 0
Z Z t
ia i(aNu− +bWu ) b
+ (e − 1)e dNu + λ (e − 1)ei(aNu +bWu ) du.
ia
]0,t] 0

Let us denote fa,b (t) = EP (ei(aNt +bWt ) ). By taking the expectations of both sides of the last equality,
we get Z t Z
ia 1 2 t
fa,b (t) = 1 + λ (e − 1)fa,b (u) du − b fa,b (u) du.
0 2 0
By solving the last equation, we obtain, for arbitrary a, b ∈ R,
¡ ¢ ia 1 2 ¡ ¢ ¡ ¢
EP ei(aNt +bWt ) = fa,b (t) = eλt(e −1) e− 2 b t = EP eiaNt EP eibWt .

Thus, for any t ∈ R+ the random variables Wt and Nt are mutually independent under P.
In the second step, we fix 0 < t < s and we consider the following expectation, for arbitrary real
numbers a1 , a2 , b1 and b2 ,
¡ ¢
f (t, s) := EP ei(a1 Nt +a2 Ns +b1 Wt +b2 Ws ) .
206 APPENDIX A. POISSON PROCESSES

Let us denote ã1 = a1 + a2 and b̃1 = b1 + b2 . Then

¡ ¢
f (t, s) = EP ei(a1 Nt +a2 Ns +b1 Wt +b2 Ws )
¡ ¡ ¯ ¢¢
= EP EP ei(ã1 Nt +a2 (Ns −Nt )+b̃1 Wt +b2 (Ws −Wt )) ¯ Gt
¡ ¡ ¯ ¢¢
= EP ei(ã1 Nt +b̃1 Wt ) EP ei(a2 (Ns −Nt )+b2 (Ws −Wt )) ¯ Gt
¡ ¡ ¢¢
= EP ei(ã1 Nt +b̃1 Wt ) EP ei(a2 Nt−s +b2 Wt−s )
¡ ¢
= fa1 ,b1 (t − s) EP ei(ã1 Nt +b̃1 Wt )
= fa1 ,b1 (t − s)fã1 ,b̃1 (t),

where we have used, in particular, the independence of the increment Nt − Ns (and Wt − Ws ) of the
σ-field Gt , and the time-homogeneity of N and W . By setting b1 = b2 = 0 in the last formula, we
obtain ¡ ¢
EP ei(a1 Nt +a2 Ns ) = fa1 ,0 (t − s)fã1 ,0 (t),
while the choice of a1 = a2 = 0 yields
¡ ¢
EP ei(b1 Wt +b2 Ws ) = f0,b1 (t − s)f0,b̃1 (t).

It is not difficult to check that

fa1 ,b1 (t − s)fã1 ,b̃1 (t) = fa1 ,0 (t − s)fã1 ,0 (t)f0,b1 (t − s)f0,b̃1 (t).

We conclude that for any 0 ≤ t < s and arbitrary a1 , a2 , b1 , b2 ∈ R:

¡ ¢ ¡ ¢ ¡ ¢
EP ei(a1 Nt +a2 Ns +b1 Wt +b2 Ws ) = EP ei(a1 Nt +a2 Ns ) EP ei(b1 Wt +b2 Ws ) .

This means that the random variables (Nt , Ns ) and (Wt , Ws ) are mutually independent. By proceed-
ing along the same lines, one may check that the random variables (Nt1 , . . . , Ntn ) and (Wt1 , . . . , Wtn )
are mutually independent for any n ∈ N and for any choice of 0 ≤ t1 < · · · < tn . ¤
Let us now examine the behavior of the Poisson process under a specific equivalent change of
the underlying probability measure. For a fixed T > 0, we introduce a probability measure Q on
(Ω, GT ) by setting
dQ ¯¯
¯ = ηT , P-a.s., (A.2)
dP GT
where the Radon-Nikodým density process (ηt , t ∈ [0, T ]) satisfies

bt ,
dηt = ηt− κ dN η0 = 1, (A.3)

for some constant κ > −1. Since Y := κN b is a process of finite variation, (A.3) admits a unique
b
solution, denoted as Et (Y ) or Et (κN ). Clearly, this solution can be seen as a special case of the
Doléans (or stochastic) exponential. By solving (A.3) in the path-by-path manner, we obtain
Y c Y
b ) = eYt
ηt = Et (κN (1 + ∆Yu )e−∆Yu = eYt (1 + ∆Yu ),
0<u≤t 0<u≤t
P
where Ytc := Yt − 0<u≤t ∆Yu is the path-by-path continuous part of Y . Direct calculations show
that Y
ηt = e−κλt (1 + κ∆Nu ) = e−κλt (1 + κ)Nt = eNt ln(1+κ)−κλt ,
0<u≤t

where the last equality is valid provided that κ > −1. Upon setting a = ln(1 + κ) in part (iii) of
Proposition A.1.2, we obtain η = M a ; this confirms that the process η is a G-martingale under P.
We have thus proved the following result.
A.1. STANDARD POISSON PROCESS 207

Lemma A.1.1 Assume that κ > −1. The unique solution η to the SDE (A.3) is an exponential
G-martingale under P. Specifically,
b
ηt = eNt ln(1+κ)−κλt = eNt ln(1+κ)−λt(κ−ln(1+κ)) = Mta , (A.4)

where a = ln(1+κ). In particular, the random variable ηT is strictly positive, P-a.s. and EP (ηT ) = 1.
Furthermore, the process M a solves the following SDE

dMta = Mt−
a bt ,
(ea − 1) dN M0a = 1.

We are in a position to establish the well-known result, which states that the process (Nt , t ∈
[0, T ]) is a Poisson process with the constant intensity λ∗ = (1 + κ)λ under Q.

Proposition A.1.4 Assume that under P a process N is a Poisson process with intensity λ with
respect to the filtration G. Suppose that the probability measure Q is defined on (Ω, GT ) through
(A.2) and (A.3) for some κ > −1.
(i) The process (Nt , t ∈ [0, T ]) is a Poisson process under Q with respect to G with the constant
intensity λ∗ = (1 + κ)λ.
(ii) The compensated process (Nt∗ , t ∈ [0, T ]) defined as

bt − κλt,
Nt∗ = Nt − λ∗ t = Nt − (1 + κ)λt = N

is a G-martingale under Q.

Proof. From Remark A.1.1(iii), we know that it suffices to find λ∗ such that, for any fixed b ∈ R,
fb , given as
the process M
ftb := ebNt −λ∗ t(eb −1) , ∀ t ∈ [0, T ],
M (A.5)
is a G-martingale under Q. By standard arguments, the process M fb is a Q-martingale if and only
f b
if the product M η is a martingale under the original probability measure P. But in view of (A.4),
we have ³ ¡ ¢ ¡ ¢´
ftb ηt = exp Nt b + ln(1 + κ) − t κλ + λ∗ (eb − 1) .
M

Let us write a = b + ln(1 + κ). Since b is an arbitrary real number, so is a. Then, by virtue of part
(iii) in Proposition A.1.2, we necessarily have

κλ + λ∗ (eb − 1) = λ(ea − 1).

fb defined by (A.5)
After simplifications, we conclude that, for any fixed real number b, the process M
is a G-martingale under Q if and only if λ = (1 + κ)λ. In other words, the intensity λ∗ of N under
∗

Q satisfies λ∗ = (1 + κ)λ. Also the second statement is clear. ¤

Assume that W is a Brownian motion and N is a Poisson process under P with respect to G.
Let η satisfy ¡ ¢
bt , η0 = 1,
dηt = ηt− θt dWt + κ dN (A.6)
for some G-predictable stochastic process θ and some constant κ > −1. A simple application of
the Itô’s product rule shows that if processes η 1 and η 2 satisfy the SDEs dηt1 = ηt−
1
θt dWt and
dηt2 = ηt−
2 bt then their product ηt = ηt1 ηt2 satisfies (A.6).
κ dN
Taking the uniqueness of solutions to the linear SDE (A.6) for granted, we conclude that the
unique solution to this SDE is given by the expression:
³Z t Z
1 t 2 ´ ¡ ¢
ηt = exp θu dWu − θu du exp Nt ln(1 + κ) − κλt . (A.7)
0 2 0
We leave the proof of the next result as an exercise for the reader.
208 APPENDIX A. POISSON PROCESSES

Proposition A.1.5 Let the probability Q be given by (A.2) and (A.7) for some constant κ > −1
and a G-predictable process θ such that EP (ηT ) = 1.
¡ Rt ¢
(i) The process Wt∗ = Wt − 0 θu du, t ∈ [0, T ] is a Brownian motion under Q with respect to the
filtration G.
(ii) The process (Nt , t ∈ [0, T ]) is a Poisson process with the constant intensity λ∗ = (1 + κ)λ under
Q with respect to the filtration G.
(iii) Processes W ∗ and N are mutually independent under Q.

A.2 Inhomogeneous Poisson Process

R∞
Let λ : R+ → R+ be any non-negative, locally integrable function such that 0 λ(u) du = ∞. By
definition, the process N (with N0 = 0) is the Poisson process with intensity function λ if for every
0 ≤ s < t the increment Nt − Ns is independent of the σ-field GsR and has the Poisson law with
t
parameter Λ(t) − Λ(s), where the hazard function Λ equals Λ(t) = 0 λ(u) du.
More generally, let Λ : R+ → R+ be a right-continuous, increasing function with Λ(0) = 0 and
Λ(∞) = ∞. The Poisson process with the hazard function Λ satisfies, for every 0 ≤ s < t and every
k = 0, 1, . . .
(Λ(t) − Λ(s))k −(Λ(t)−Λ(s))
P(Nt − Ns = k | Gs ) = P(Nt − Ns = k) = e .
k!

Example A.2.1 The most convenient, and thus widely used, method of constructing a Poisson
process with a hazard function Λ runs as follows: we take a Poisson process N e with the constant
e eΛ(t) .
intensity λ = 1 with respect to some filtration G and we define the time-changed process Nt := N
The process N is easily seen to follow a Poisson process with the hazard function Λ with respect to
the time-changed filtration G, where Gt = GeΛ(t) for every t ∈ R+ .

Since for arbitrary 0 ≤ s < t

EP (Nt − Ns | Gs ) = EP (Nt − Ns ) = Λ(t) − Λ(s),

it is clear that the compensated Poisson process N bt = Nt − Λ(t) is a G-martingale under P. A

suitable generalization of Proposition A.1.3 shows that a Poisson process with the hazard function
Λ and a Brownian motion with respect to G follow mutually independent processes under P. The
proof of the next lemma relies on a direct application of the Itô formula and so it is omitted.

Lemma A.2.1 Let Z be an arbitrary bounded, G-predictable process. Then the process M Z , given
by the formula
³Z Z t ´
Z
Mt = exp Zu dNu − (eZu − 1) dΛ(u) ,
]0,t] 0
Z
is a G-martingale under P. Moreover, M is the unique solution to the SDE

dMtZ = Mt−
Z bt ,
(eZt − 1) dN M0Z = 1.

In the case of a Poisson process with intensity function λ, it can be easily deduced from Lemma
A.2.1 that for any Borel measurable function κ : R+ →] − 1, ∞[ the process
³Z Z t ´
ζt = exp ln(1 + κ(u)) dNu − κ(u)λ(u) du
]0,t] 0

bt with η0 = 1. Using similar arguments as in the

is the unique solution to the SDE dζt = ζt− κ(t) dN
case of a constant κ, one can show that the unique solution to the SDE
¡ ¢
bt , η0 = 1,
dηt = ηt− θt dWt + κ(t) dN
A.3. CONDITIONAL POISSON PROCESS 209

is given by the following expression

³Z t
1
Z t ´
ηt = ζt exp θu dWu − θu2 du . (A.8)
0 2 0

The next result generalizes Proposition A.1.5. Again, the proof is left to the reader.

Proposition A.2.1 Let Q be a probability measure, equivalent to P on (Ω, GT ), such that the density
process η in (A.2) is given by (A.8). Then under Q and with respect to G we have that:
¡ Rt ¢
(i) the process Wt∗ = Wt − 0 θu du, t ∈ [0, T ] is a Brownian motion,
(ii) the process (Nt , t ∈ [0, T ]) is a Poisson process with the intensity function λ∗ given by λ∗ (t) =
1 + κ(t)λ(t),
(iii) the processes W ∗ and N are mutually independent under Q.

A.3 Conditional Poisson Process

We start by assuming that we are given a filtered probability space (Ω, G, P) and a certain sub-
filtration F of G. Let Λ be an F-adapted, right-continuous, increasing process Rwith Λ0 = 0 and
t
Λ∞ = ∞. We refer to Λ as the hazard process. In some cases, we have Λt = 0 λu du for some
F-progressively measurable process λ with locally integrable sample paths. Then the process λ is
called the F-intensity process.
We are in a position to state the definition of the F-conditional Poisson process, which is also
known as the doubly stochastic Poisson process. A slightly different, but essentially equivalent,
definition of a conditional Poisson process can be found in Brémaud [30] and Last and Brandt [110].

Definition A.3.1 A process N defined on a probability space (Ω, G, P) is called the F-conditional
Poisson process with respect to G, associated with the hazard process Λ, if for any 0 ≤ s < t and
every k = 0, 1, . . .
(Λt − Λs )k −(Λt −Λs )
P(Nt − Ns = k | Gs ∨ F∞ ) = e , (A.9)
k!
where F∞ = σ(Fu : u ∈ R+ ).

At the intuitive level, if a particular sample path Λ· (ω) of the hazard process is known, the
process N has exactly the same probabilistic properties as the Poisson process with respect to G
with the hazard function Λ· (ω). In particular, it follows from (A.9) that

P(Nt − Ns = k | Gs ∨ F∞ ) = P(Nt − Ns = k | F∞ ),

i.e., conditionally on the σ-field F∞ the increment Nt − Ns is independent of the σ-field Gs for any
0 ≤ s < t.
Similarly, for any 0 ≤ s < t ≤ u and every k = 0, 1, . . ., we have

(Λt − Λs )k −(Λt −Λs )

P(Nt − Ns = k | Gs ∨ Fu ) = e . (A.10)
k!
In other words, conditionally on the σ-field Fu , the process (Nt , t ∈ [0, u]) behaves like a Poisson
process with the hazard function Λ· (ω).
Consequently, for any n ∈ N, any non-negative integers k1 , k2 , . . . , kn , and arbitrary non-negative
real numbers s1 < t1 ≤ s2 < t2 ≤ . . . ≤ sn < tn , we have that

³\n ´ ³Yn ¡ ¢ki ´

Λ ti − Λ si
P {Nti − Nsi = ki } = EP e−(Λti −Λsi ) .
i=1 i=1
ki !
210 APPENDIX A. POISSON PROCESSES

Let us observe that in all conditional expectations above, the reference filtration F can be replaced
by the filtration FΛ generated by the hazard process. In fact, the F-conditional Poisson process
with respect to G is also the conditional Poisson process with respect to the filtrations FN ∨ F and
FN ∨ FΛ with the same hazard process.
We shall henceforth postulate that EP (Λt ) < ∞ for every t ∈ R+ .

bt = Nt − Λt is a martingale with respect to G.

Lemma A.3.1 The compensated process N

Proof. It is enough to notice that, for arbitrary 0 ≤ s < t,

bt | Gs ) = EP (EP (Nt − Λt | Gs ∨ F∞ ) | Gs ) = EP (Ns − Λs | Gs ) = N
EP (N bs ,

where, in the second equality, we have used the property of a Poisson process with a deterministic
hazard function. ¤

Given the two filtrations F and G and the hazard process Λ, it is not obvious whether we
may find a process N satisfying Definition A.3.1. To provide a simple construction of a conditional
Poisson process, we assume that the underlying probability space (Ω, G, P), endowed with a reference
filtration F, is sufficiently large to accommodate for the following stochastic processes: a Poisson
process N e with the constant intensity equal to 1 and an F-adapted hazard process Λ. In addition,
we postulate that the Poisson process N e is independent of the filtration F

Remark A.3.1 Given a filtered probability space (Ω, F, P), it is always possible to enlarge this
space in such a way that there exists a Poisson process Ne , which is defined on the enlarged space,
has the constant intensity equal to 1 and is independent of the filtration F,

Under the present assumptions, we have that, for every 0 ≤ s < t and u ∈ R+ , and any non-
negative integer k,
et − N
P(N es = k | F∞ ) = P(N
et − N
es = k | Fu ) = P(N
et − N
es = k)

and
k
et − N
P(N es = k | FsÑ ∨ Fs ) = P(N es = k) = (t − s) e−(t−s) .
et − N
k!
The next result describes an explicit construction of a conditional Poisson process. This con-
struction is based on a random time change associated with the increasing process Λ.

Proposition A.3.1 Let N e be a Poisson process with the constant intensity equal to 1 such that N e
is independent of a reference filtration F. Let Λ be an F-adapted, right-continuous, increasing process
with Λ0 = 0 and Λ∞ = ∞. Then the process Nt = N eΛ , t ∈ R+ , is the F-conditional Poisson process
t
with the hazard process Λ with respect to the filtration G = FN ∨ F.

Proof. Since Gs ∨ F∞ = FsN ∨ F∞ , it suffices to check that

(Λt − Λs )k −(Λt −Λs )

P(Nt − Ns = k | FsN ∨ F∞ ) = e
k!
or, equivalently,
k
P(NeΛ − NeΛ = k | FΛÑ ∨ F∞ ) = (Λt − Λs ) e−(Λt −Λs ) .
t s s
k!
The last equality follows from the assumed independence of Ne and F. ¤

Remark A.3.2 Within the setup of Proposition A.3.1, any F-martingale is also a G-martingale, so
that hypothesis (H) is satisfied.
A.3. CONDITIONAL POISSON PROCESS 211

Example A.3.1 Cox process. In some applications, it is natural to consider a special case of an
F-conditional Poisson process, with the filtration F generated by a certain stochastic process, repre-
senting the state variables. To be more specific, on considers a conditional Poisson process with the
intensity process λ given as λt = g(t, Yt ), where Y is an Rd -valued stochastic process independent
of the Poisson process N e and g : R+ × Rd → R+ is some function. The reference filtration F is
typically chosen to be the natural filtration of the process Y ; that is, we set F = FY . In that case,
the resulting F-conditional Poisson process is referred to as the Cox process associated with the
state-variables process Y and the intensity map g.

Our last goal is to examine the behavior of an F-conditional Poisson process N under an equivalent
change of a probability measure. For the sake of simplicity, we assume that the hazard process Λ is
continuous, and the reference filtration F is generated by a process W , which is a Brownian motion
with respect to G. For a fixed T > 0, we define the probability measure Q on (Ω, GT ) by setting
dQ ¯¯
¯ GT = ηT , P-a.s., (A.11)
dP
where the Radon-Nikodým density process (ηt , t ∈ [0, T ]) solves the SDE
¡ ¢
dηt = ηt− θt dWt + κt dN bt , η0 = 1, (A.12)
for some G-predictable processes θ and κ such that κ > −1 and EP (ηT ) = 1. An application of the
Itô product rule shows that the unique solution to (A.12) is equal to the product νζ, where ν and
ζ are solutions to SDEs
dνt = νt θt dWt
and
bt
dζt = ζt− κt dN
with the initial values ν0 = ζ0 = 1. The unique solutions to these SDEs are given by the expressions
³Z t Z
1 t 2 ´
νt = exp θu dWu − θ du
0 2 0 u
and Y
ζt = exp(Ut ) (1 + ∆Uu ) exp (−∆Uu ),
0<u≤t
R
respectively, where we denote Ut = bu . Observe that ζ admits also the following equivalent
κu dN
]0,t]
representations Z
³ t ´ Y
ζt = exp − κu dΛu (1 + κu ∆Nu )
0 0<u≤t
and
³Z Z t ´
ζt = exp ln(1 + κu ) dNu − κu dΛu .
]0,t] 0

Proposition A.3.2 Let the Radon-Nikodým density of Q with respect to P be given by (A.11)–
(A.12). Then the process W ∗ defined by, for t ∈ [0, T ],
Z t
Wt∗ = Wt − θu du,
0
is a Brownian motion with respect to G under Q and the process N ∗ given by, for t ∈ [0, T ],
Z t Z t
Nt∗ = Nbt − κu dΛu = Nt − (1 + κu ) dΛu ,
0 0
is a G-martingale under Q. If, in addition, the process κ is F-adapted then the process N is under
Q the F-conditional Poisson process with respect to G and the hazard process of N under Q equals
Z t
∗
Λt = (1 + κu ) dΛu .
0
212 APPENDIX A. POISSON PROCESSES

A.4 The Doléans Exponential

In this section, we recall some well-known results from stochastic analysis, regarding the Doléans
exponential (also known as the stochastic exponential). For more details on stochastic integration
and stochastic differential equations, the interested reader is referred to, Elliott [69], Protter [132],
or Revuz and Yor [133].

A.4.1 Exponential of a Process of Finite Variation

Let us first examine a special case of the Doléans exponential for a process of finite variation. Let A
be a real-valued, càdlàg process of finite variation defined on a probability space (Ω, F, P). Consider
the following linear stochastic differential equation

dZt = Zt− dAt ,

with the initial condition Z0 = 1 or, equivalently,

Z
Zt = 1 + Zu− dAu , (A.13)
]0,t]

where the integral is the pathwise Stieltjes integral.

Proposition A.4.1 The unique solution Zt = Et (A) to (A.13), referred to as the Doléans exponen-
tial of A, is given by the expression
Y c Y
Et (A) = eAt (1 + ∆Au )e−∆Au = eAt (1 + ∆Au ), (A.14)
0<u≤t 0<u≤t

where Ac is the path-by-path continuous

P part of A, that is, the continuous process of finite variation
given by the formula Act = At − 0<u≤t ∆Au for every t ∈ R+ .

A.4.2 Exponential of a Special Semimartingale

Let Y be a real-valued, càdlàg, special semimartingale defined on (Ω, F, P).

Definition A.4.1 We denote by E(Y ) the Doléans exponential of Y , that is, the unique solution Z
of the linear stochastic differential equation

dZt = Zt− dYt , (A.15)

with the initial condition Z0 = 1.

Note that (A.15) is a shorthand notation for the integral equation

Z
Zt = 1 + Zu− dYu , (A.16)
]0,t]

where the integral is the Itô stochastic integral.

Recall that the process of quadratic variation of an arbitrary semimartingale Y is defined by the
formula, for every t ∈ R+ , Z
[Y ]t = Yt2 − Y02 − 2 Yu− dYu .
]0,t]

The next result furnishes an extension of Proposition A.4.1 to the case of a process Y that follows
a special semimartingale.
A.4. THE DOLÉANS EXPONENTIAL 213

Proposition A.4.2 Assume that Y is a special semimartingale. Then the unique solution to linear
stochastic differential equation (A.15) is given by the formula
³ 1 ´ Y
Et (Y ) = exp Yt − Y0 − [Y ]ct (1 + ∆Yu ) exp(−∆Yu ),
2
0<u≤t

where ∆Yu = Yu − Yu− and [Y ]c is the path-by-path continuous part of [Y ].

It is well known that [Y ]c = hM c i, where M c is the continuous martingale part of a special

semimartingale Y . Recall that any special semimartingale Y admits the unique decomposition

Yt = Y0 + Mtc + Mtd + At , ∀ t ∈ R+ ,

where M c is a continuous local martingale, M d is a purely discontinuous local martingale, and

A is a predictable process of finite variation, with the initial values M0c = M0d = A0 = 0. This
decomposition is commonly known as the canonical decomposition of a special semimartingale Y .
The following result summarizes the properties of the Doléans exponential that are useful in the
context of Girsanov’s theorem.

Proposition A.4.3 (i) The Doléans exponential of Y is a strictly positive process if and only if the
jumps of Y satisfy ∆Yt > −1 for every t ∈ R+ .
(ii) If Y is a local martingale such that ∆Yt > −1 for every t ∈ R+ then E(Y ) is a strictly positive
local martingale and thus a supermartingale. In that case, it is a martingale whenever EP (Et (Y )) = 1
for every t ∈ R+ .
(iii) The Doléans exponential of a local martingale Y satisfying ∆Yt > −1 for every t ∈ R+ is a
uniformly integrable martingale whenever EP (E∞ (Y )) = 1, where E∞ (Y ) = lim t→∞ Et (Y ) and the
limit is known to exist.
214 APPENDIX A. POISSON PROCESSES
Bibliography

[1] C. Albanese and O.X. Chen. Discrete credit barrier models. Quantitative Finance, 5:247–256,
2005.

[2] M. Ammann. Credit Risk Valuation: Methods, Models and Applications. Springer-Verlag,
Berlin Heidelberg New York, 2nd edition, 2001.

[3] L. Andersen and J. Sidenius. Extensions to the Gaussian copula: Random recovery and random
factor loadings. Journal of Credit Risk, 1:29–70, 2004/2005.

[4] L. Andersen and J. Sidenius. CDO pricing with factor models: Survey and comments. Journal
of Credit Risk, 1:71–88, 2005.

[5] P. Artzner and F. Delbaen. Default risk insurance and incomplete markets. Mathematical
Finance, 5:187–195, 1995.

[6] A. Arvanitis and J.-P. Laurent. On the edge of completeness. Risk, 10:61–62, 1999.

[7] T. Aven. A theorem for determining the compensator of a counting process. Scandinavian
Journal of Statistics, 12:69–72, 1985.

[8] S. Babbs and T.R. Bielecki. A note on short spreads. Working paper, 2003.

[9] D. Becherer and M. Schweizer. Classical solutions to reaction-diffusion systems for hedging
problems with interacting Itô and point processes. Annals of Applied Probability, 15:1111–1144,
2005.

[10] A. Bélanger, S.E. Shreve, and D. Wong. A general framework for pricing credit risk. Mathe-
matical Finance, 14:317–350, 2004.

[11] T.R. Bielecki, S. Crépey, M. Jeanblanc, and M. Rutkowski. Valuation of basket credit deriva-
tives in the credit migrations environment. In J.R. Birge and V. Linetsky, editors, Financial
Engineering, Handbooks in Operations Research and Management Science, Vol. 15. Elsevier,
2008.

[12] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Hedging of defaultable claims. In R.A.
Carmona et al., editors, Paris-Princeton Lectures on Mathematical Finance 2003, Lecture
Notes in Mathematics 1847, pages 1–132. Springer-Verlag, Berlin Heidelberg New York, 2004.

[13] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Modelling and valuation of credit risk. In
M. Frittelli and W. Runggaldier, editors, CIME-EMS Summer School on Stochastic Methods
in Finance, Bressanone, Lecture Notes in Mathematics 1856, pages 27–126. Springer-Verlag,
Berlin Heidelberg New York, 2004.

[14] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. PDE approach to valuation and hedging of
credit derivatives. Quantitative Finance, 5:257–270, 2005.

215
216 BIBLIOGRAPHY

[15] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Completeness of a general semimartingale

market under constrained trading. In M. do Rosário Grossinho et al., editors, Stochastic
Finance, pages 83–106. Springer-Verlag, Berlin Heidelberg New York, 2006.

[16] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Hedging of credit derivatives in models with
totally unexpected default. In J. Akahori et al., editors, Stochastic Processes and Applications
to Mathematical Finance, pages 35–100. World Scientific, Singapore, 2006.

[17] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Replication of contingent claims in a reduced-
form credit risk model with discontinuous asset prices. Stochastic Models, 22:661–687, 2006.

[18] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Hedging of basket credit derivatives in credit
default swap market. Journal of Credit Risk, 3:91–132, 2007.

[19] T.R. Bielecki, M. Jeanblanc, and M. Rutkowski. Pricing and trading credit default swaps in
a hazard process model. Annals of Applied Probability, forthcoming.

[20] T.R. Bielecki and M. Rutkowski. Credit Risk: Modelling, Valuation and Hedging. Springer-
Verlag, Berlin Heidelberg New York, 2002.

[21] T.R. Bielecki and M. Rutkowski. Dependent defaults and credit migrations. Applicationes
Mathematicae, 30:121–145, 2003.

[22] T.R. Bielecki and M. Rutkowski. Modelling of the defaultable term structure: Conditionally
Markov approach. IEEE Transactions on Automatic Control, 49:361–373, 2004.

[23] T.R. Bielecki, A. Vidozzi, and L. Vidozzi. An efficient approach to valuation of credit basket
products and ratings triggered step-up bonds. Working paper, 2006.

[24] T.R. Bielecki, A. Vidozzi, and L. Vidozzi. A Markov copulae approach to pricing and hedging
of credit index derivatives and ratings triggered step-up bonds. Journal of Credit Risk, 4:47–76,
2008.

[25] F. Black and J.C. Cox. Valuing corporate securities: Some effects of bond indenture provisions.
Journal of Finance, 31:351–367, 1976.

[26] F. Black and M. Scholes. The pricing of options and corporate liabilities. Journal of Political
Economy, 81:637–654, 1973.

[27] C. Blanchet-Scalliet and M. Jeanblanc. Hazard rate for credit risk and hedging defaultable
contingent claims. Finance and Stochastics, 8:145–159, 2004.

[28] C. Bluhm, L. Overbeck, and C. Wagner. An Introduction to Credit Risk Modeling. Chapman
& Hall/CRC, Boca Raton, 2004.

[29] H.-J. Brasch. Exact replication of k-th-to-default swaps with first-to-default swaps. Working
paper, 2006.

[30] P. Brémaud. Point Processes and Queues. Martingale Dynamics. Springer-Verlag, Berlin
Heidelberg New York, 1981.

[31] P. Brémaud and M. Yor. Changes of filtration and of probability measures. Z. Wahrschein-
lichkeitsheorie verw. Gebiete, 45:269–295, 1978.

[32] M.J. Brennan and E.S. Schwartz. Convertible bonds: Valuation and optimal strategies for call
and conversion. Journal of Finance, 32:1699–1715, 1977.

[33] M.J. Brennan and E.S. Schwartz. Analyzing convertible bonds. Journal of Financial and
Quantitative Analysis, 15:907–929, 1980.
BIBLIOGRAPHY 217

[34] D. Brigo. Constant maturity credit default swap pricing with market models. Working paper,
2004.
[35] D. Brigo. Candidate market models and the calibrated CIR++ stochastic intensity model for
credit default swap options and callable floaters. Working paper, 2005.
[36] D. Brigo and A. Alfonsi. Credit default swaps calibration and option pricing with the SSRD
stochastic intensity and interest-rate model. Finance and Stochastics, 9:29–42, 2005.
[37] D. Brigo and N. El-Bachir. An exact formula for default swaptions’ pricing in the SSRJD
stochastic intensity model. Working paper, 2008.
[38] D. Brigo and M. Morini. CDS market formulas and models. Working paper, 2005.
[39] D. Brigo, A. Pallavicini, and R. Torresetti. Calibration of CDO tranches with the dynamical
generalized-Poisson loss model. Working paper, 2006.
[40] E. Briys and F. de Varenne. Valuing risky fixed rate debt: An extension. Journal of Financial
and Quantitative Analysis, 32:239–248, 1997.
[41] X. Burtschell, J. Gregory, and J.-P. Laurent. A comparative analysis of CDO pricing models.
Working paper, 2005.
[42] X. Burtschell, J. Gregory, and J.-P. Laurent. Beyond the Gaussian copula: Stochastic and
local correlation. Journal of Credit Risk, 3:31–62, 2007.
[43] L. Campi and A. Sbuelz. Closed form pricing of benchmark equity default swaps under the
CEV assumption. Working paper, 2005.
[44] P. Carr. Dynamic replication of a digital default claim. Working paper, 2005.
[45] L. Chen and D. Filipović. A simple model for credit migration and spread curves. Finance
and Stochastics, 9:211–231, 2005.
[46] N. Chen and S. Kou. Credit spreads, optimal capital structure, and implied volatility with
endogenous default and jump risk. Working paper, 2005.
[47] Z. Chen and P. Glasserman. Fast pricing of basket default swaps. Working paper, 2006.
[48] U. Cherubini and E. Luciano. Pricing and hedging credit derivatives with copulas. Economic
Notes, 32:219–242, 2003.
[49] U. Cherubini, E. Luciano, and W. Vecchiato. Copula Methods in Finance. J. Wiley, Chichester,
2004.
[50] P.O. Christensen, C.R. Flor, D. Lando, and K.R. Miltersen. Dynamic capital structure with
callable debt and debt renegotiations. Working paper, 2002.
[51] P. Collin-Dufresne, R.S. Goldstein, and J.-N. Hugonnier. A general formula for valuing de-
faultable securities. Econometrica, 72:1377–1407, 2004.
[52] P. Collin-Dufresne and J.-N. Hugonnier. On the pricing and hedging of contingent claims in
the presence of extraneous risks. Working paper, 1998.
[53] R. Cont and Y.H. Kan. Dynamic hedging of portfolio credit derivatives. Working paper, 2008.
[54] R. Cont and A. Minca. Recovering portfolio default intensities implied by CDO quotes. Work-
ing paper, 2008.
[55] D. Cossin and H. Pirotte. Advanced Credit Risk Analysis. J. Wiley, Chichester, 2001.
[56] M. Davis and V. Lo. Infectious defaults. Quantitative Finance, 1:382–386, 2001.
218 BIBLIOGRAPHY

[57] C. Dellacherie. Un exemple de la théorie générale des processus. In P.A. Meyer, editor,
Séminaire de Probabilités IV, Lecture Notes in Mathematics 124, pages 60–70. Springer-Verlag,
Berlin Heidelberg New York, 1970.
[58] C. Dellacherie. Capacités et processus stochastiques. Springer-Verlag, Berlin Heidelberg New
York, 1972.
[59] C. Dellacherie and P.A. Meyer. Probabilités et potentiel, chapitres I-IV. Hermann, Paris, 1975.
English translation: Probabilities and Potential, Chapters I-IV, North-Holland, 1978.
[60] C. Dellacherie and P.A. Meyer. Probabilités et potentiel, chapitres V-VIII. Hermann, Paris,
1980. English translation: Probabilities and Potential, Chapters V-VIII, North-Holland, 1982.
[61] G. Di Graziano and L.C.G. Rogers. A dynamic approach to the modelling of correlation credit
derivatives using Markov chains. Working paper, 2006.
[62] D. Duffie. First-to-default valuation. Working paper, 1998.
[63] D. Duffie and N. Gârleanu. Risk and the valuation of collateralized debt obligations. Financial
Analyst’s Journal, 57:41–59, 2001.
[64] D. Duffie and D. Lando. Term structures of credit spreads with incomplete accounting infor-
mation. Econometrica, 69:633–664, 2001.
[65] D. Duffie, M. Schroder, and C. Skiadas. Recursive valuation of defaultable securities and the
timing of resolution of uncertainty. Annals of Applied Probability, 6:1075–1090, 1996.
[66] D. Duffie and K. Singleton. Modeling term structure of defaultable bonds. Review of Financial
Studies, 12:687–720, 1999.
[67] D. Duffie and K. Singleton. Simulating correlated defaults. Working paper, 1999.
[68] D. Duffie and K. Singleton. Credit Risk: Pricing, Measurement and Management. Princeton
University Press, Princeton, 2003.
[69] R.J. Elliott. Stochastic Calculus and Applications. Springer-Verlag, Berlin Heidelberg New
York, 1982.
[70] R.J. Elliott, M. Jeanblanc, and M. Yor. On models of default risk. Mathematical Finance,
10:179–196, 2000.
[71] Y. Elouerkhaoui. Etude des problèmes de corrélation et d’incomplétude dans les marchés de
crédit. Doctoral dissertation, 2006.
[72] P. Embrechts, F. Lindskog, and A.J. McNeil. Modelling dependence with copulas and appli-
cations to risk management. In S. Rachev, editor, Handbook of Heavy Tailed Distributions in
Finance, pages 329–384. Elsevier North Holland, 2003.
[73] J.-P. Florens and D. Fougère. Noncausality in continuous time. Econometrica, 64:1195–1212,
1996.
[74] J.-P. Fouque, R. Sircar, and K. Solna. Stochastic volatility effects on defaultable bonds. Applied
Mathematical Finance, 13:215–244, 2006.
[75] J.-P. Fouque, B.C. Wignall, and X. Zhou. Modeling correlated defaults: First passage model
under stochastic volatility. Journal of Computational Finance, 11:43–78, 2008.
[76] R. Frey and J. Backhaus. Credit derivatives in models with interacting default intensities: A
Markovian approach. Working paper, 2006.
[77] R. Frey and J. Backhaus. Dynamic hedging of synthetic CDO tranches with spread risk and
default contagion. Working paper, 2007.
BIBLIOGRAPHY 219

[78] R. Frey and A.J. McNeil. Dependent defaults in models of portfolio credit risk. Journal of
Risk, 6:59–92, 2003.

[79] R. Frey, A.J. McNeil, and A. Nyfeler. Copulas and credit models. Risk, 10:111–114, 2001.

[80] H. Gennheimer. Model risk in copula based default pricing models. Working paper, 2002.

[81] K. Giesecke. Correlated default with incomplete information. Journal of Banking and Finance,
28:1521–1545, 2004.

[82] K. Giesecke. Default and information. Journal of Economic Dynamics and Control, 2006.

[83] K. Giesecke and L. Goldberg. A top-down approach to multi-name credit. Working paper,
2005.

[84] Y.M. Greenfield. Hedging of the credit risk embedded in derivative transactions. Doctoral
dissertation, 2000.

[85] A. Herbertsson. Default contagion in large homogeneous portfolios. Working paper, 2007.

[86] A. Herbertsson. Pricing synthetic CDO tranches in a model with default contagion using the
matrix-analytic approach. Working paper, 2007.

[87] B. Hilberink and L.C.G. Rogers. Optimal capital structure and endogenous default. Finance
and Stochastics, 6:237–263, 2002.

[88] S.L. Ho and L. Wu. Arbitrage pricing of credit derivatives. Working paper, 2007.

[89] J. Hull and A. White. Valuing credit default swaps (I): No counterparty default risk. Journal
of Derivatives, 8:29–40, 2000.

[90] J. Hull and A. White. Valuing credit default swaps (II): Modeling default correlations. Journal
of Derivatives, 8:12–22, 2000.

[91] J. Hull and A. White. The valuation of credit default swap options. Journal of Derivatives,
10:40–50, 2003.

[92] F. Jamshidian. Valuation of credit default swap and swaptions. Finance and Stochastics,
8:343–371, 2004.

[93] R.A. Jarrow, D. Lando, and S.M. Turnbull. A Markov model for the term structure of credit
risk spreads. Review of Financial Studies, 10:481–523, 1997.

[94] R.A. Jarrow and S.M. Turnbull. Pricing options on derivative securities subject to credit risk.
Journal of Finance, 50:53–85, 1995.

[95] R.A. Jarrow and F. Yu. Counterparty risk and the pricing of defaultable securities. Journal
of Finance, 56:1756–1799, 2001.

[96] M. Jeanblanc and Y. Le Cam. Immersion property and credit risk modelling. Working paper,
2007.

[97] M. Jeanblanc and Y. Le Cam. Intensity versus hazard process approaches. Working paper,
2007.

[98] M. Jeanblanc and Y. Le Cam. Reduced form modelling for credit risk. Working paper, 2007.

[99] M. Jeanblanc and M. Rutkowski. Modeling default risk: An overview. In Y. Jiongmin and
R. Cont, editors, Mathematical Finance: Theory and Practice, pages 171–269. Higher Educa-
tion Press, Beijing, 2000.
220 BIBLIOGRAPHY

[100] M. Jeanblanc and M. Rutkowski. Modeling default risk: Mathematical tools. Working paper,
2000.

[101] T. Jeulin and M. Yor. Nouveaux résultats sur le grossissement des tribus. Ann. Scient. ENS,
4e série, 11:429–443, 1978.

[102] M. Joshi and D. Kainth. Rapid and accurate development of prices and Greeks for nth to
default credit swaps in the Li model. Quantitative Finance, 4:266–275, 2004.

[103] M. Kijima and K. Komoribayashi. A Markov chain model for valuing credit risk derivatives.
Journal of Derivatives, 6:97–108, 1998.

[104] M. Kijima, K. Komoribayashi, and E. Suzuki. A multivariate Markov model for simulating
correlated defaults. Working paper, 2002.

[105] M. Kijima and Y. Muromachi. Credit events and the valuation of credit derivatives of basket
type. Review of Derivatives Research, 4:55–79, 2000.

[106] S. Kusuoka. A remark on default risk models. Advances in Mathematical Economics, 1:69–82,
1999.

[107] D. Lando. On Cox processes and credit risky securities. Review of Derivatives Research,
2:99–120, 1998.

[108] D. Lando. On rating transition analysis and correlation. In Credit Derivatives. Applications for
Risk Management, Investment and Portfolio Optimisation, pages 147–155. Risk Publications,
London, 1998.

[109] D. Lando. Credit Risk Modeling. Princeton University Press, Princeton, 2004.

[110] G. Last and A. Brandt. Marked Point Processes on the Real Line. The Dynamic Approach.
Springer-Verlag, Berlin Heidelberg New York, 1995.

[111] J.-P. Laurent. Applying hedging techniques to credit derivatives. Credit Risk Conference,
London, 2001.

[112] J.-P. Laurent. A note of risk management of CDOs. Working paper, 2006.

[113] J.-P. Laurent, A. Cousin, and J.D. Fermanian. Hedging default risks of CDOs in Markovian
contagion models. Working paper, 2007.

[114] J.-P. Laurent and J. Gregory. Basket defaults swaps, CDOs and factor copulas. Working
paper, 2002.

[115] J.-P. Laurent and J. Gregory. Correlation and dependence in risk management. Working
paper, 2003.

[116] H. Leland. Corporate debt value, bond covenants, and optimal capital structure. Journal of
Finance, 49:1213–1252, 1994.

[117] H. Leland and K. Toft. Optimal capital structure, endogenous bankruptcy, and the term
structure of credit spreads. Journal of Finance, 51:987–1019, 1996.

[118] D.-X. Li. On default correlation: A copula approach. Journal on Fixed Income, 9:43–54, 2000.

[119] F.A. Longstaff and E.S. Schwartz. A simple approach to valuing risky fixed and floating rate
debt. Journal of Finance, 50:789–819, 1995.

[120] D. Madan and H. Unal. Pricing the risks of default. Review of Derivatives Research, 2:121–160,
1998.
BIBLIOGRAPHY 221

[121] R. Mansuy and M. Yor. Random Times and Enlargements of Filtrations in a Brownian Setting.
Springer-Verlag, Berlin Heidelberg New York, 2006.

[122] G. Mazziotto and J. Szpirglas. Modèle général de filtrage non linéaire et équations différentielles
stochastiques associées. Ann. Inst. Henri Poincaré, 15:147–173, 1979.

[123] A.J. McNeil, R. Frey, and P. Embrechts. Quantitative Risk Management: Concepts, Tech-
niques, and Tools. Princeton University Press, Princeton, 2005.

[124] R. Merton. On the pricing of corporate debt: The risk structure of interest rates. Journal of
Finance, 3:449–470, 1974.

[125] F. Moraux. Valuing corporate liabilities when the default threshold is not an absorbing barrier.
Working paper, 2002.

[126] M. Morini and D. Brigo. No-armageddon arbitrage-free equivalent measure for index options
in a credit crisis. Working paper, 2007.

[127] R.B. Nelsen. An Introduction to Copulas. Lecture Notes in Statistics 139. Springer-Verlag,
Heidelberg Berlin New York, 1999.

[128] T.N. Nielsen, J. Saá-Requejo, and P. Santa-Clara. Default risk and interest rate risk: The
term structure of default spreads. Working paper, 1993.

[129] A. Nikeghbali and M. Yor. A definition and some properties of pseudo-stopping times. Annals
of Probability, 33:1804–1824, 2005.

[130] C. Pedersen. Valuation of portfolio credit default swaptions. Technical report, Lehman Broth-
ers, 2003.

[131] A. Petrelli, O. Siu, J. Zhang, and V. Kapoor. Optimal static hedging of defaults in CDOs.
Working paper, 2006.

[132] P. Protter. Stochastic Integration and Differential Equations. Springer-Verlag, Berlin Heidel-
berg New York, 2nd edition, 2005.

[133] D. Revuz and M. Yor. Continuous Martingales and Brownian Motion. Springer-Verlag, Berlin
Heidelberg New York, 3rd edition, 1999.

[134] M. Rutkowski and A. Armstrong. Valuation of credit default swaptions and credit default
index swaptions. Working paper, 2007.

[135] M. Rutkowski and K. Yousiph. PDE approach to valuation and hedging of basket credit
derivatives. International Journal of Theoretical and Applied Finance, 10:1261–1285, 2007.

[136] J. Saá-Requejo and P. Santa-Clara. Bond pricing with default risk. Working paper, 1999.

[137] P. Schönbucher and D. Schubert. Copula-dependent default risk in intensity models. Working
paper, 2001.

[138] P.J. Schönbucher. Credit Derivatives Pricing Models. Wiley Finance, New York, 2003.

[139] M. Shaked and J.G. Shanthikumar. The multivariate hazard construction. Stochastic Processes
and their Applications, 24:241–258, 1987.

[140] J. Sidenius, V. Piterbarg, and L. Andersen. A new framework for dynamic credit portfolio loss
modelling. International Journal of Theoretical and Applied Finance, 11:163–197, 2007.

[141] N. Vaillant. A beginner’s guide to credit derivatives. Technical report, Nomura International,
2001.
222 BIBLIOGRAPHY

[142] D. Wong. A unifying credit model. Technical report, Capital Markets Group, 1998.

[143] L. Wu. Arbitrage pricing of single-name credit derivatives. Working paper, 2005.
[144] F. Yu. Correlated defaults in intensity-based models. Mathematical Finance, 17:155–173, 2007.
[145] H. Zheng. Efficient hybrid methods for portfolio credit derivatives. Quantitative Finance,
6:349–357, 2006.
[146] C. Zhou. The term structure of credit spreads with jumps risk. Journal of Banking and
Finance, 25:2015–2040, 2001.