Time-inhomogeneous Lévy processes

in interest rate and credit risk

models

Dissertation zur Erlangung des Doktorgrades

der Fakultät für Mathematik und Physik
der Albert-Ludwigs-Universität Freiburg im Breisgau

vorgelegt von
Wolfgang Kluge
Juli 2005
Dekan: Prof. Dr. Josef Honerkamp
Referenten: Prof. Dr. Ernst Eberlein
Prof. Dilip B. Madan, Ph.D., Ph.D.

Datum der Promotion: 11. Oktober 2005

Abteilung für Mathematische Stochastik

Albert-Ludwigs-Universität Freiburg
Eckerstr. 1
D-79104 Freiburg im Breisgau
 Abstract

In this thesis, we present interest rate models and a credit risk model, all driven
by time-inhomogeneous Lévy processes, i.e. stochastic processes whose incre-
ments are independent but in general not stationary.
In the interest rate part, we discuss a Heath–Jarrow–Morton forward rate
model (the Lévy term structure model), a model for forward bond prices (the
Lévy forward price model) and a Libor model (the Lévy Libor model) which
generalizes the Libor market model. In all of these models, explicit valuation
formulae are established for the most liquid interest rate derivatives, namely
caps, floors, and swaptions. The formulae can numerically be evaluated fast and
thus allow to calibrate the models to market data. In the Lévy term structure
model, we also price floating range notes. Their payoffs are path-dependent.
In the credit risk part, the Lévy Libor model (and therewith, as a special
case, the Libor market model) is extended to defaultable forward Libor rates.
We present a rigorous construction of the model and price some of the most
heavily traded credit derivatives, namely credit default swaps, total rate of
return swaps, credit spread options and credit default swaptions.
 Acknowledgements

I am indebted to many people for contributing to my development in mathe-

matics and finance or to my personal development or to both. Any list would
definitely be incomplete, so let me just name those that, in my opinion, have
the closest link to this thesis:

First and foremost, I would like to thank my academic teacher and supervi-
sor Prof. Dr. Ernst Eberlein. He introduced me to the theory of Lévy pro-
cesses as well as to mathematical finance and gave me valuable advice and
support. I would also like to thank all my colleagues from the Department
of Mathematical Stochastics at the University of Freiburg, and especially Jan
Bergenthum and Antonis Papapantoleon, for discussing mathematical and non-
mathematical ideas with me. Further thanks go to Monika Hattenbach, Anto-
nis Papapantoleon, and Albrecht Zimmermann for valuable comments on the
manuscript.

Very special thanks go to my family and to my friends for their support. In

particular, I would like to thank Marie-Luise Kluge, Wieland Kluge, Andreas
Kluge, and Vera Henrike Meyer-Borgstädt.

Financial support by the Deutsche Forschungsgemeinschaft (DFG) is gratefully

acknowledged. I also acknowledge the permission from Blackwell Publishing to
use the content of Eberlein and Kluge (2005) in this thesis.
Contents

1 Introduction 1
1.1 Interest rate models . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Modelling credit risk . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Non-homogeneous Lévy processes . . . . . . . . . . . . . . . . . . 6

2 The Lévy term structure model 15

2.1 Presentation of the model . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Tools for derivative valuation . . . . . . . . . . . . . . . . . . . . 19
2.3 Valuation of caps and floors . . . . . . . . . . . . . . . . . . . . . 25
2.4 Swaption pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.5 Valuation of floating range notes . . . . . . . . . . . . . . . . . . 36
2.5.1 Digital Options . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.2 Range Notes . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 An example of calibration to market data . . . . . . . . . . . . . 49
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3 Lévy models for effective rates 59

3.1 The Lévy forward price model . . . . . . . . . . . . . . . . . . . . 61
3.1.1 The model as a special case of the LTSM . . . . . . . . . 64
3.2 The Lévy Libor model . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.1 Valuation of caps and floors . . . . . . . . . . . . . . . . . 71
3.2.2 Swaption pricing . . . . . . . . . . . . . . . . . . . . . . . 77

4 The Lévy Libor model with default risk 85

4.1 Presentation of the model . . . . . . . . . . . . . . . . . . . . . . 86
4.2 Construction of the time of default . . . . . . . . . . . . . . . . . 89
4.3 Specification of the drift . . . . . . . . . . . . . . . . . . . . . . . 93
4.4 Defaultable forward measures . . . . . . . . . . . . . . . . . . . . 102
4.5 Recovery rules and bond prices . . . . . . . . . . . . . . . . . . . 106
4.6 Credit-sensitive swap contracts . . . . . . . . . . . . . . . . . . . 110
4.6.1 Credit default swaps . . . . . . . . . . . . . . . . . . . . . 111
4.6.2 Total rate of return swaps . . . . . . . . . . . . . . . . . . 112
4.6.3 Asset swaps . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.7 Credit options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.7.1 Options on defaultable bonds . . . . . . . . . . . . . . . . 114
4.7.2 Credit spread options . . . . . . . . . . . . . . . . . . . . 119
ii CONTENTS

4.7.3 Credit default swaptions . . . . . . . . . . . . . . . . . . . 120

4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

A Proof of Proposition 4.8 123

Chapter 1

Introduction

A trader in a bank asked for an offer by a client, who wants to purchase a

financial product, faces two questions:

• What price shall he offer the product for?

• Assuming the client accepts the offer, how can the trader hedge himself?

The answers to these questions are easy if there is a liquid market for this
product. The trader can use the market price, add a margin and offer it to the
client. Suppose his offer is accepted, then he buys the product at the market,
sells it to the client and keeps the margin.
In case there is no liquid market for the product, the questions are much
harder to answer. For a specific financial product, namely a call (or put) option
on a stock price, these questions have been addressed in a famous article by
Black and Scholes (1973). Under the assumption of existence of a liquid market
for the underlying stock and a risk-free asset, they duplicate the option by a
permanently re-balanced portfolio consisting of stocks and a certain amount of
the risk-free asset. In this way, they derive a unique price for the option and a
perfect hedging strategy at the same time.
Unfortunately, the price as well as the hedging strategy depend heavily on
the way Black and Scholes model the stock price. They model it – first suggested
by Samuelson (1965) – as a geometric Brownian motion. Since their article
has been published, a large variety of different approaches for modelling stock
prices has been proposed by several authors. These approaches usually lead to
different option prices and hedging strategies. Often, they do not even produce
unique prices for options or perfect hedging strategies. Similar observations can
be made for other derivatives, as e.g. interest rate, foreign exchange or credit
derivatives: a reasonable price for a derivative will usually depend on the way
the underlying is modelled, a fact that imposes the next question: How should
it be modelled?
Of course, there is no objective answer to this question since every model has
its advantages and drawbacks. Two desirable model features are generality and
tractability. However, these two properties do not, usually, come together. Very
general models, e.g. models driven by general semimartingales, are often not
tractable. On the other hand, models driven by Brownian motions usually can
2 Introduction

be handled nicely but cannot explain statistical behaviour of real-life financial

data (e.g. fat-tailed log return distributions or volatility smiles).
Lévy processes are excellent tools for modelling price processes of financial
securities. The resulting models are very flexible and allow for tractable pric-
ing formulae for derivatives in many cases. In this thesis, we go a step further
towards generality and consider various models driven by time-inhomogeneous
(also called non-homogeneous) Lévy processes. They are usually as easy to han-
dle as models driven by (homogeneous) Lévy processes but allow for additional
flexibility. Our main focus lies on the construction of the models as well as on
the derivation of pricing formulae for options and other derivatives. Hedging
strategies are not addressed.
Among the derivatives that are traded over-the-counter (OTC) interest rate
products form the largest part. According to the Bank for International Settle-
ments (BIS), the notional amount of outstanding OTC interest rate derivative
contracts added up to 187 trillions of US dollars in December 2004, compared
to 248 trillions for all OTC derivative contracts.1 We devote chapters 2 and
3 to modelling of interest rates and to pricing of interest rate derivatives. In
chapter 2, a Heath–Jarrow–Morton forward rate model introduced by Eber-
lein and Raible (1999), the Lévy term structure model, is presented. Explicit
valuation formulae for caps, floors, swaptions and for a derivative with a path-
dependent payoff, namely a floating range note, are derived. All formulae can
numerically be evaluated fast, a fact that allows us to calibrate the model to
market data. The main results of chapter 2 can also be found in Eberlein and
Kluge (2004, 2005). Two models for effective rates due to Eberlein and Özkan
(2005), the Lévy forward price model and the Lévy Libor model, are presented
in chapter 3. We show that the first one can be seen as a special case of the
Lévy term structure model. The second approach generalizes the Libor market
model to non-homogeneous Lévy processes. Eberlein and Özkan (2005) deduce
approximate pricing solutions for caps and floors but do not consider swaption
pricing. We suggest an alternative approximation to price caps and floors and
compare the results. Moreover, a valuation formula for swaptions (also based
on an approximation) is derived. Symmetry results relating the prices of caps
and floors in all of these interest rate models are established in Eberlein, Kluge,
and Papapantoleon (2005). However, they are not part of this thesis.
A market that is growing rapidly in size is the market for credit derivatives.
“According to the data presented in the Triennial Central Bank Survey of For-
eign Exchange and Derivatives Market Activity, the growth of credit related
derivatives in the three years ending June 2004 amounted to 568 %, against
121 % for all OTC products”.2 In chapter 4 we develop a credit risk model that
extends the Lévy Libor model (and therewith the Libor market model). Pricing
formulae for some of the most popular credit derivatives, as e.g. credit default
swaps, total rate of return swaps, credit spread options and credit default swap-
tions, are also deduced.

1
Source: BIS Quarterly Review, June 2005, p. A99.
2
Source: BIS Quarterly Review, June 2005, p. 50.
1.1 Interest rate models 3

In the remaining sections of this chapter, we give very brief introductions

to interest rate and credit risk modelling. Moreover, non-homogeneous Lévy
processes are defined and some of their properties are established.

1.1 Interest rate models

In this section, we introduce notations and give definitions for some very com-
mon financial products and mathematical concepts related to interest rate the-
ory which are used in the subsequent chapters. Moreover, a brief survey of the
most important types of interest rate models is presented. However, this sec-
tion cannot, and is not supposed to, serve as an introduction to interest rate
theory. There is a large amount of literature discussing the subject in detail. To
name just a few references, let us mention the books of Musiela and Rutkowski
(1998), Brigo and Mercurio (2001), and Björk (2004).
A zero coupon bond is a financial security that pays an amount of one cur-
rency unit to its owner at maturity of the contract. The price at time t of a zero
coupon bond with maturity T (t ≤ T ) is denoted by B(t, T ) in what follows.
Obviously, we have B(T, T ) = 1 for any maturity date T . Since there are no
intermediate payments, one could expect the value of a zero coupon bond to
be always less than or equal to one. However, this property does not hold in
models that admit negative interest rates. To put it differently, not all models
allow to carry cash at no costs.
Instantaneous (continuously compounded) forward rates are mathematical
concepts rather than observable rates in the markets. However, they are very
convenient for modelling purposes. By f (t, T ) we denote the forward rate as
seen at time t for borrowing or lending money over an infinitesimal time period
starting at T . Formally, instantaneous forward rates are defined by f (t, T ) :=
∂
− ∂T log B(t, T ) (assuming the derivative exists). Zero
 Rcoupon bond  prices can
T
be recovered from forward rates via B(t, T ) = exp − t f (t, u) du .
The instantaneous interest rate that prevails at time t for immediate lending
or borrowing over an infinitesimal time interval is called spot rate or short rate
and denoted by rt . Clearly, we have the relationship rt = f (t, t). An amount
of one unit of cash R t at time 0 that is continuously reinvested at the short rate
yields Bt := exp 0 rs ds at time t. B is usually referred to as savings account,
money market account or discount factor.
Libor is an abbreviation for London Inter-Bank Offer Rate and refers to
the interest rate that is paid between banks. In contrast to instantaneous rates,
Libor rates are effective or simply compounded, i.e. interest accrues according to
a discrete grid. Although banks can default, many interest rate models neglect
this risk and assume that Libor rates are default-free. Under this assumption,
the Libor rate for a period of length δ starting at T can be related to zero
B(T,T )
coupon bond prices via L(T, T ) := 1δ B(T,T +δ) − 1 . The forward Libor rate
L(t, T ) is theLibor rate atT as it is seen by the market at time t. More precisely,
B(t,T )
L(t, T ) := 1δ B(t,T +δ) − 1 .
A (plain vanilla) interest rate swap is an agreement between two parties
4 Introduction

to exchange fixed against floating interest rate payments. More precisely, one
party agrees to pay a fixed interest rate on a notional principal in return for a
floating interest rate (usually the Libor) on the same notional and for the same
period of time. The fixed rate that makes the initial value of this contract equal
to zero is called swap rate.
A cap (resp. floor) consists of a series of call (resp. put) options on sub-
sequent Libor rates. These single options are called caplets (resp. floorlets). A
caplet that is settled in arrears with a notional of 1, maturity T and a strike rate
of K on the Libor rate L(T, T ) pays off δ(L(T, T ) − K)+ at T + δ. The payoff of
the respective floorlet equals δ(K − L(T, T ))+ . Note that a caplet can be seen
as a put option on a zero coupon bond, since a payoff of δ(L(T, T ) − K)+ at
T + δ equals a payoff of B(T, T + δ)δ(L(T, T ) − K)+ at T and

B(T, T + δ)δ(L(T, T ) − K)+ = (1 − (1 + δK)B(T, T + δ))+

= (1 + δK)((1 + δK)−1 − B(T, T + δ))+ .

Similarly, a floorlet can be regarded as a call option on a zero coupon bond. Caps
and floors are commonly used as insurances against rising or falling interest
rates.
A swaption is an option on a forward swap, i.e. on an interest rate swap
which starts in the future. At maturity of the option, its holder has the right to
enter into the swap at a pre-specified fixed rate. There are payer and receiver
swaptions giving their owners the right to enter into the swap as fixed rate payer
or receiver respectively. The holder of a payer (receiver) swaption will exercise
the option if the swap rate at option maturity is higher (lower) than the strike
rate of the swaption.

The most classical approaches to modelling fixed income markets are short
rate models. They exogenously specify the dynamics of the short rate r. Deriva-
tives in these models are not only caps, floors, and swaptions but also zero
coupon bonds. In other words, initial zero coupon bond prices are an output
of and not an input to the model. Brigo and Mercurio (2001) give a good
overview of various models, also commenting on their particular advantages
and drawbacks. Short rate models describe the evolution of the whole fixed in-
come market by one explanatory variable (the short rate) only, a feature that
is often criticized. Another common handicap of all (time-homogeneous) short
rate models is their inability to exactly reproduce a given initial term struc-
ture, i.e. the bond prices B(0, ·) that are observed in the market (see e.g. Björk
(2004)). Nevertheless, short rate models are still widely used.
The idea to exogenously specify the evolution of the whole term structure of
interest rates was pioneered by Heath, Jarrow, and Morton (1992). Subject to
modelling in a Heath–Jarrow–Morton (henceforth HJM) framework are either
zero coupon bond prices or instantaneous forward rates. Initial bond prices enter
as a model input, i.e. any given initial term structure is perfectly reproduced.
There are various HJM-type models differing mainly in the specification of the
process that drives the forward rates or bond prices. At the high end as far as
generality is concerned let us mention the semimartingale approach of Björk,
1.2 Modelling credit risk 5

Di Masi, Kabanov, and Runggaldier (1997). These authors use a finite number
of Wiener processes plus an integer-valued random measure as drivers.
In a series of papers by Sandmann, Sondermann, and Miltersen (1995),
Miltersen, Sandmann, and Sondermann (1997), Brace, Gatarek, and Musiela
(1997), Jamshidian (1997), and Musiela and Rutkowski (1997) the forward Libor
model and the forward swap model were developed. Subject to modelling in these
so-called market models are the dynamics of forward Libor or swap rates. Among
practitioners, the models are very popular since they reproduce well-established
market formulae for caps/floors and swaptions respectively. More precisely, the
forward Libor model can be calibrated perfectly to (at-the-money) quotes of
caps and floors whereas the forward swap model is able to reproduce market
prices of swaptions exactly. For an extensive survey of the market models we
refer to Brigo and Mercurio (2001, Section 6).

1.2 Modelling credit risk

Most credit risk models may be classified into two categories: structural models
and reduced form models. We comment very briefly on the two approaches and
refer to Schönbucher (2000) and Bielecki and Rutkowski (2002) for more details.
In structural models or firm’s value models the value process of a firm’s assets
is exogenously specified. Default is defined as the first time at which the value
process hits or falls below a default triggering barrier. The barrier is a second
exogenously specified process that is usually related to the debt of the firm.
An advantage of this approach is the direct link between default and the firm’s
capital structure. The structural approach is well-suited whenever products
that depend on more than one security issued by a firm (e.g. a callable bond
which can be converted into shares when called) have to be priced. A question
that arises naturally in this setting is wherefrom to observe the value process?
Another drawback is the fact that defaultable bonds are no fundamentals but
an output of the model. Moreover, in structural models that are driven by
continuous processes the time of default is a predictable stopping time. As a
consequence, credit spreads tend to zero as the time to maturity of a debt
tends to zero. This contradicts what can be observed in real markets. From the
large amount of papers that have contributed to the structural approach let us
mention a few: Merton (1974), Black and Cox (1976), Geske (1977), Longstaff
and Schwartz (1995), and Zhou (1997).
Reduced form models, also called intensity based models or hazard rate mod-
els, specify default by a totally inaccessible stopping time. In many cases, exoge-
nous specifications of hazard rates or stochastic intensities are used to construct
the time of default. Some authors link the hazard rate to the value of the firm’s
assets or stocks, as e.g. Madan and Unal (1998, 2000). The resulting models are
often referred to as hybrid models since they combine elements from the struc-
tural and the intensity based approach. However, most reduced form models
do not rely on the value of the firm’s assets at all. Instead, they often extend
default-free interest rate models. The Lévy Libor model with default risk which
we present in chapter 4 is one example of the intensity based approach. From
6 Introduction

the long list of other examples, let us mention a very small sample: Artzner
and Delbaen (1992), Jarrow and Turnbull (1995), Lando (1997), and Duffie and
Singleton (1999).

1.3 Non-homogeneous Lévy processes

In financial mathematics Lévy processes are very popular since they allow for
models that are much more flexible than models driven by Brownian motions.
At the same time, these models are often still tractable. Important special
cases of Lévy processes include Brownian motions, Poisson and compound Pois-
son processes as well as Lévy processes generated by variance gamma (intro-
duced by Madan and Seneta (1987, 1990)), CGMY (Carr, Geman, Madan,
and Yor (2002)), normal inverse Gaussian or generalized hyperbolic distribu-
tions (Barndorff-Nielsen (1977, 1998) and Eberlein (2001)). We are not going to
review the definition and properties of a Lévy process. Instead, we refer to ex-
isting literature. There are many books examining these processes in detail, e.g.
Bertoin (1996), Sato (1999), and Applebaum (2004). Schoutens (2003) discusses
applications of Lévy processes to mathematical finance.
Non-homogeneous Lévy processes are more general than (homogeneous)
Lévy processes. In contrast to their homogeneous counterparts, non-homo-
geneous Lévy processes do not, generally, possess stationary increments. As we
will see in the subsequent chapters, relaxation of the stationarity assumption
provides us with additional flexibility in the models. Fortunately, this flexibil-
ity does not come at a high pice, i.e. the models usually do not become more
complicated. A book that examines non-homogeneous Lévy processes in some
detail is Cont and Tankov (2003).
We assume a stochastic basis (Ω, F, F, P) to be given, i.e. a probability
space (Ω, F, P) equipped with a filtration F := (Ft )t≥0 . By a filtration we mean
an increasing and right continuous family of sub-σ-fields of F (compare Jacod
and Shiryaev (2003, Definition I.1.2)). Note that, unless explicitly stated, we
do not assume completeness of the stochastic basis, i.e. we do not require that
the usual conditions hold. Since all models that are considered in what follows
have a finite time horizon T ∗ , we work with the following definition for a non-
homogenous Lévy process:

Definition 1.1 An adapted stochastic process L = (Lt )0≤t≤T ∗ with values in

Rd is a non-homogeneous Lévy process, sometimes also called time-inhomo-
geneous Lévy process or process with independent increments and absolutely
continuous characteristics, henceforth abbreviated by PIIAC, if the following
conditions hold:

1. L has independent increments,

i.e. Lt − Ls is independent of Fs (0 ≤ s < t ≤ T ∗ ).

2. For every t ∈ [0, T ∗ ], the law of Lt is characterized by the characteristic

1.3 Non-homogeneous Lévy processes 7

function
Zt 
h
ihu,Lt i
i 1
IE e = exp ihu, bs i − hu, cs ui (1.1)
2
0
Z 
ihu,xi
+ (e − 1 − ihu, xi1l{|x|≤1} )Fs (dx) ds.
Rd

Here, bs ∈ Rd , cs is a symmetric nonnegative-definite d×d matrix, and Fs

is a measure on Rd that integrates (|x|2 ∧1) and satisfies Fs ({0}) = 0. The
Euclidian scalar product on Rd is denoted by h·, ·i, the respective norm by
| · |. It is assumed that

ZT ∗ Z 
2
|bs | + ||cs || + (|x| ∧ 1)Fs (dx) ds < ∞, (1.2)
0 Rd

where || · || denotes any norm on the set of d × d matrices.

We call (b, c, F ) := (bs , cs , Fs )0≤s≤T ∗ the characteristics of L.

Remark: It is also possible to define a non-homogeneous Lévy process L on a

probability space without assuming a filtration to be given in advance. Instead,
the probability space can then be equipped with the filtration generated by L.

Note that definition 1.1 can also be used to define a homogeneous Lévy
process if bs , cs , and Fs are assumed not to depend on s. Let us point out some
properties of a non-homogeneous Lévy process L:

Lemma 1.2 Fix t ∈ [0, T ∗ ]. The distribution of Lt is infinitely divisible with

Lévy–Khintchine triplet (b, c, F ), where

Zt Zt Zt
b := bs ds, c := cs ds, F (dx) := Fs (dx) ds.
0 0 0

(The integrals are to be understood componentwise.)

Proof: Clearly, b ∈ Rd and c is a symmetric nonnegative-definite d × d matrix.

A monotone convergence argument yields that F is a measure on the Borel sets
of Rd and we get
Z Zt Z
f (x)F (dx) = f (x)Fs (dx) ds (1.3)
Rd 0 Rd

for any integrable function f . Thus, (|x|2 ∧1)F (dx) < ∞ by (1.2) and F ({0}) =
R

0. The claim now follows from (1.1) and the Lévy–Khintchine formula. 
8 Introduction

Lemma 1.3 L is an additive process in law, i.e. a stochastically continuous

process with independent increments and L0 = 0 a.s.

Proof: The last property follows directly from the characteristic function of L0 .
To verify stochastic continuity let us have a look at the characteristic function
of Lt − Lv for v < t. By the independence of the increments

IE eihu,Lt i
h i  
ihu,Lt −Lv i
IE e =  ihu,L i 
IE e v

Zt 
1
= exp ihu, bs i − hu, cs ui
2
v
Z 
ihu,xi
+ (e − 1 − ihu, xi1l{|x|≤1} )Fs (dx) ds.
Rd

As v approaches t, the characteristic function of Lt − Lv converges pointwise

to 1. Thus, Lt − Lv converges to 0 in distribution and, since that is a constant,
also stochastically. 

Every additive process in law has a modification that is càdlàg, which means
that all paths are right-continuous and admit left-hand limits (see e.g. Sato
(1999, Theorem 11.5)). We will always work with this modification of L. Al-
though not every càdlàg adapted process with independent increments is a
semimartingale (Jacod and Shiryaev (2003, Chapter II, §4c) give a counter-
example), non-homogeneous Lévy processes are semimartingales:

Lemma 1.4 The process L is a semimartingale with respect to the stochastic

basis (Ω, F, F, P).

Proof: This property can be established by looking at the characteristic func-

tion of Lt . More precisely, (1.2) implies that for every u ∈ Rd the function
t 7→ f (u)t where
h i
f (u)t := log IE eihu,Lt i
Zt 
1
= ihu, bs i − hu, cs ui
2
0
Z 
+ (eihu,xi − 1 − ihu, xi1l{|x|≤1} )Fs (dx) ds
Rd

has finite variation over finite intervals. Hence, t 7→ exp f (u)t has finite varia-
tion over finite intervals and Jacod and Shiryaev (2003, Chapter II, Theorem
4.14) yields that L is a semimartingale. 

We describe L by its semimartingale characteristics in the sense of Jacod

and Shiryaev (2003, Chapter II, Definition 2.6):
1.3 Non-homogeneous Lévy processes 9

Lemma 1.5 The semimartingale characteristics of L associated with the trun-

cation function h(x) := 1l{|x|≤1} are given by

Zt Zt Zt Z
Bt = bs ds, Ct = cs ds, ν([0, t] × A) = Fs (dx) ds (A ∈ B(Rd )).
0 0 0 A

Proof: We use Jacod and Shiryaev (2003, Chapter II, Corollary 2.48) and look
at
Z
1
A(u)t := ihu, Bt i − hu, Ct ui + (eihu,xi − 1 − ihu, xi1l{|x|≤1} )ν([0, t] × dx)
2
Rd

for u ∈ Rd . The function t 7→ A(u)t is continuous and has finite variation over
finite intervals. Moreover, A(u)t equals the characteristic exponent of Lt . Thus,
h i
E[A(u)] = exp A(u) = IE eihu,L· i ,

where E denotes the stochastic exponential. By independence of the increments

of L

IE[eihu,Lt i |Fs ] = IE[eihu,Lt −Ls i eihu,Ls i |Fs ]

IE[eihu,Lt i ] ihu,Ls i
= e .
IE[eihu,Ls i ]

Hence, eihu,L· i /E[A(u)] is a martingale and the cited corollary yields that B, C
and ν are indeed the characteristics of L. 

These characteristics allow us to write L in its canonical representation (see

Jacod and Shiryaev (2003, II.2.34))

Zt Zt Z X
Lt = bs ds + Lct + x1l{|x|≤1} (µ − ν)(ds, dx) + ∆Ls 1l{|∆Ls |>1} . (1.4)
0 0 Rd s≤t

Here, Lc denotes the continuous martingale part of L and µ is the random

measure associated Rwith the jumps of L. From the characteristic C we can
t√
conclude that Lct = 0 cs dWs , where W is a standard d-dimensional Brownian
√
motion and cs is a measurable version of the square root of cs .
In many applications we will require the existence of exponential moments
and put the following integrability condition on the measures Fs :
Assumption (EM). There are constants M, ε > 0 such that for every u ∈
[−(1 + ε)M, (1 + ε)M ]d

ZT ∗ Z
exphu, xiFs (dx) ds < ∞. (1.5)
0 {|x|>1}
10 Introduction

R
Without loss of generality, {|x|>1} exphu, xiFs (dx) is assumed to be finite for
all s.

An equivalent condition to (1.5) in terms of the existence of exponential

moments of L is given by the following lemma:

Lemma 1.6 Assumption (EM) holds if and only if there are constants M, ε > 0
such that IE[exphu, Lt i] < ∞ for all t ∈ [0, T ∗ ] and u ∈ [−(1 + ε)M, (1 + ε)M ]d .

Proof: Assume that (1.5) holds and fix u ∈ [−(1 + ε)M, (1 + ε)M ]d and
t ∈ [0, T ∗ ]. Let L e 1 ∼ Lt . Then, its generating triplet
e be a Lévy process with L
(b, c, F ) is given by lemma 1.2. By (1.3) we have

Z Zt Z
exphu, xiF (dx) = exphu, xiFs (dx) ds < ∞.
{|x|>1} 0 {|x|>1}

Sato (1999, Theorem 25.3) yields that IE[exphu, L e 1 i] < ∞. Since L e 1 ∼ Lt , we

also get IE[exphu, Lt i] < ∞.
Conversely, let us assume that for u ∈ [−(1 + ε)M, (1 + ε)M ]d we have
IE[exphu, LT ∗ i] < ∞. Let Le be a Lévy process with L e 1 ∼ LT ∗ and (b, c, F ) its
generating triplet. Then IE[exphu, L1 i] = IE[exphu, LT ∗ i] < ∞. Again (1.3) and
e
Sato (1999, Theorem 25.3) imply

ZT ∗ Z Z
exphu, xiFs (dx) ds = exphu, xiF (dx) < ∞. 
0 {|x|>1} {|x|>1}

From the preceding lemma we can conclude that under assumption (EM)
the expected value of Lt is finite. Hence, the characteristic function of Lt can
be written as
Zt 
h
ihu,Lt i
i 1
IE e = exp ihu, bs i − hu, cs ui
2
0
Z 
ihu,xi
+ (e − 1 − ihu, xi)Fs (dx) ds. (1.6)
Rd

Of course, the bs in this representation differ from those in equation (1.1) since
we changed the truncation function.R In fact, it follows from Jacod and Shiryaev
(2003, II.2.30) that they differ by Rd x1l{|x|>1} Fs (dx). Henceforth, whenever we
work under assumption (EM), we will use the characteristics that correspond
to equation (1.6). Also, in this setting L is not only a semimartingale but a
special semimartingale:

Lemma 1.7 L is a special semimartingale.

1.3 Non-homogeneous Lévy processes 11

Proof: We use Jacod and Shiryaev (2003, II.2.29) and show that (|x|2 ∧ |x|) ∗ ν
is an adapted process with locally integrable variation. Since (|x|2 ∧ |x|) ∗ ν is
increasing and deterministic, we only need to show finiteness of

ZT ∗Z
(|x|2 ∧ |x|) ∗ νT ∗ = (|x|2 ∧ |x|)Fs (dx) ds
0 Rd
ZT ∗ Z ZT ∗ Z
2
= |x| Fs (dx) ds + |x|Fs (dx) ds.
0 {|x|≤1} 0 {|x|>1}

The finiteness of the first term is guaranteed by (1.2), while (1.5) implies that
the second summand is finite. 

Consequently, the canonical representation of L simplifies to

Zt Zt Zt Z
√
Lt = bs ds + cs dWs + x(µ − ν)(ds, dx). (1.7)
0 0 0 Rd

In one of the models that follow we will need an assumption which is slightly
stronger than assumption (EM) from a mathematical point of view. In appli-
cations, both assumptions are practically equal and not very restrictive:
Assumption (SUP). It holds that
 Z 
2
sup |bs | + ||cs || + (|x| ∧ |x|)Fs (dx) < ∞ (1.8)
0≤s≤T ∗
Rd

(where || · || denotes any norm on the set of d × d matrices) and there are
constants M, ε > 0 such that for every u ∈ [−(1 + ε)M, (1 + ε)M ]d
 Z 
sup exphu, xiFs (dx) < ∞. (1.9)
0≤s≤T ∗
{|x|>1}

In the remaining part of this section we assume that (EM) is in force and
present a proposition that proves to be very useful for the derivation of drift
conditions in term structure models as well as for option pricing. To simplify
notation, let us denote by θs the cumulant associated with the infinitely divisible
distribution characterized by the Lévy–Khintchine triplet (bs , cs , Fs ), i.e. for
z ∈ [−(1 + ε)M, (1 + ε)M ]d where M is the constant from assumption (EM) we
have
Z
1
θs (z) := hz, bs i + hz, cs zi + (ehz,xi − 1 − hz, xi)Fs (dx). (1.10)
2
Rd
12 Introduction

According to Sato (1999, Theorem 25.17) we can extend θs to complex numbers

z ∈ Cd with <(zj ) ∈ [−(1 + ε)M, (1 + ε)M ] for j ∈ {1, . . . , d} and write the
characteristic function of Lt as

h i Zt
ihu,Lt i
IE e = exp θs (iu) ds (1.11)
0

Note that iu := (iuj )1≤j≤d and the scalar product on Rd is extended to com-
plex numbers, that is hw, zi := dj=1 wj zj for w, z ∈ Cd . Hence, h·, ·i is not
P
the Hermitian scalar product. If L is a (homogeneous) Lévy process, i.e. the
increments of L are stationary, bs , cs , and Fs and thus also θs do not depend
on s. In this case we write θ for short. θ then equals the cumulant (also called
log moment generating function) of L1 .
The characteristic function of Lt can also be extended to a strip in the
complex plane, as the following lemma shows:

Lemma 1.8 Fix t ∈ [0, T ∗ ]. For z ∈ Cd with <(z) ∈ [−(1 + ε)M, (1 + ε)M ]d
we have IE[|ehz,Lt i |] < ∞ and

Zt
hz,Lt i
IE[e ] = exp θs (z) ds. (1.12)
0

Proof: We get IE[|ehz,Lt i |] = IE[eh<(z),Lt i ] < ∞ from lemma 1.6. Let L e be a

e 1 ∼ Lt . Clearly, IE[eh<(z),L1 i ] = IE[eh<(z),Lt i ] < ∞. The
Lévy process with L
e

Lévy–Khintchine triplet (b, c, F ) of L

e is given by lemma 1.2 and Sato (1999,
Theorem 25.17) yields that
Z
1
Ψ(z) := hz, bi + hz, czi + (ehz,xi − 1 − hz, xi)F (dx)
2
Rd

hz,L1 i ] = eΨ(z) . By using (1.3) we see that Ψ(z) equals

Rist definable and IE[e
e
hz,Lt i ] = exp t θ (z) ds.
R
0 θs (z) ds and therefore IE[e 0 s 

The following proposition will frequently be used for option pricing:

Proposition 1.9 Suppose that f : R+ → Cd is a continuous function such

that |<(f i (x))| ≤ M for all i ∈ {1, . . . , d} and x ∈ R+ , then

  ZT  ZT
IE exp f (s) dLs = exp θs (f (s)) ds.
t t

(The integrals are to be understood componentwise for real and imaginary part.)

Proof: This proof uses the idea of the proof of lemma 3.1 in Eberlein and Raible
(1999). By independence of the increments of L it is sufficient to consider the
1.3 Non-homogeneous Lévy processes 13

case t = 0. Since f is continuous and deterministic, it is locally bounded (take

Tn := n as localizing sequence) and we have

ZT d Z
X
T

f (s) dLs = f i (s) dLis .

0 i=1 0

For any partition 0 = t0 < t1 < . . . < tN +1 = T

N N
" !#
X Y  
IE exp hf (tk ), Ltk+1 − Ltk i = IE exphf (tk ), Ltk+1 − Ltk i
k=0 k=0
N  
Y IE exphf (tk ), Ltk+1 i
=
IE [exphf (tk ), Ltk i]
k=0
t
N Zk+1
X 
= exp θs (f (tk )) ds .
k=0 t
k

We used the independence of the increments of L for the first two equalities
and lemma 1.8 for the third. NowR let the mesh of the partition go to zero. The
T
right-hand side converges to exp 0 θs (f (s)) ds.
Let us have a look at the left-hand side. According to Jacod and Shiryaev
(2003, Proposition I.4.44)

N
X ZT
i
f (tk )(Litk+1 − Litk ) −→ f i (s) dLis in measure for each i.
k=0 0

Continuous transformations preserve convergence in measure. Consequently,

N d X
N
! !
X X
exp hf (tk ), Ltk+1 − Ltk i = exp f i (tk )(Litk+1 − Litk )
k=0 i=1 k=0
d ZT
!
X
i
−→ exp f (s) dLis
i=1 0

ZT
= exp f (s) dLs in measure.
0

If we can show that the approximating sequence is uniformly integrable,

convergence in measure will imply convergence in L1 and the claim is proved.
To show uniform integrability, we use Dellacherie and Meyer (1978, Theorem
14 Introduction

II.22) and check that the sequence is bounded in L1+ε :


N
! 1+ε 
X
IE  exp hf (tk ), Ltk+1 − Ltk i 
k=0
N
" !#
X
= IE exp h(1 + ε)<(f (tk )), Ltk+1 − Ltk i
k=0
tZk+1
N
!
X
= exp θs ((1 + ε)<(f (tk ))) ds ,
k=0 t
k

where the last equality follows as in the chain of equations above. The term
on the right-hand side converges in R if the mesh of the partition goes to zero.
Hence, we can find an upper bound for this term independent of N . 
Chapter 2

The Lévy term structure

model

In designing a model for fixed income markets that is interesting for both, the
academic world as well as the financial industry, one has to have two aspects
in mind: the model should allow for analytical expressions at least for the most
important interest rate-sensitive instruments such as bonds, swaps, caps, floors
and swaptions. At the same time, it should be possible to calibrate it fast and
accurately to market data. In particular, models should be able to reproduce
a given term structure and prices of the most liquid interest rate derivatives,
namely caps, floors and swaptions, with a sufficient degree of accuracy. We try
to fulfill both needs by presenting a generalization of the Lévy term structure
model introduced in Eberlein and Raible (1999) to non-homogeneous Lévy pro-
cesses (see also Eberlein, Jacod, and Raible (2005)). Within this framework,
we derive explicit formulae for the prices of caps, floors and swaptions. These
formulae can numerically be evaluated fast and allow for a calibration of the
model to market data. Moreover, we provide a valuation formula for a derivative
with a path-dependent payoff, namely a floating range note.
Among the variety of different interest rate models, the most popular ap-
proach is probably the Libor market model. Its popularity results from the fact
that it is consistent with the market practice of pricing caps and floors. In
other words, the model allows for a perfect calibration to cap and floor quotes.
Unfortunately, the model prices fit only the at-the-money quotes well. Away-
from-the-money there may be substantial misvaluations. Our goal is not only to
reproduce the market prices of at-the-money caps. We intend to get an accurate
calibration to cap prices across different strike rates and across all maturities
with a reasonable number of parameters.
It is well known that exponential Lévy models for stock prices allow for
an excellent calibration to implied volatility patterns for single maturities and
also for a certain range of maturities, but fail to reproduce option prices with
the same accuracy over the full range of different maturities. We made a simi-
lar observation in the Lévy term structure model. It performs very well when
calibrating prices of caps with different strikes for a certain range of maturi-
ties. However, results become worse when the calibration is done across strikes
16 The Lévy term structure model

and across the full range of maturities. This is due to the restrictive assump-
tion of stationary increments. We drop this assumption and allow for a non-
homogeneous Lévy process as driving process.
The outline of the chapter is as follows. Section 2.1 presents the details of the
model. Some mathematical tools that are needed for derivative pricing in the
subsequent sections are established in section 2.2. In section 2.3 we discuss the
pricing of caps and floors. The main techniques used are change-of-numeraire
and Laplace transformation methods. Analytical formulae that can numerically
be evaluated fast are derived. The same tools together with an idea of Jamshid-
ian (1989) can be applied to price swaptions under an additional assumption on
the volatility structure. This is the content of section 2.4. Change-of-numeraire
and Laplace transformation techniques can also be employed in a path depen-
dent context. Concretely, we use a non-standard numeraire plus Laplace trans-
formation methods in section 2.5 to determine the value of floating range notes.
As a necessary tool and nice side result, digital options are priced. In section
2.6 we give an example of a model calibration to real market prices of caps as
well as to swaption prices. Results for driving homogeneous Lévy processes are
compared to those that are obtained when a non-homogeneous Lévy process is
used. Section 2.7 concludes.

2.1 Presentation of the model

Let us briefly recall the HJM framework for modelling the term structure of
interest rates. Suppose that T ∗ > 0 is a fixed time horizon and assume that for
every T ∈ [0, T ∗ ] there is a zero coupon bond maturing at T traded on the mar-
ket. Subject to modelling are either zero coupon bond prices or instantaneous,
continuously compounded forward rates. Since forward rates can be deduced
from bond prices and vice versa (see section 1.1), the term structure can be
modelled by specifying either of them. Here, forward rates are specified and
zero coupon bond prices are deduced.
The model is driven by a d-dimensional non-homogeneous Lévy process L
with characteristics (b, c, F ) on a probability space (Ω, F, P) equipped with
the filtration (Fs )0≤s≤T ∗ which is generated by L. More precisely, we assume
F = FT ∗ and (Fs )0≤s≤T ∗ is the smallest right continuous filtration to which L
is adapted. The dynamics of the instantaneous forward rates for T ∈ [0, T ∗ ] are
postulated to be given by

Zt Zt
f (t, T ) = f (0, T ) + α(s, T ) ds − σ(s, T ) dLs (0 ≤ t ≤ T ). (2.1)
0 0

The initial values f (0, T ) are deterministic, and bounded and measurable in T .
Moreover, α and σ are stochastic processes with values in R and Rd respectively
defined on Ω × [0, T ∗ ] × [0, T ∗ ] that satisfy the following conditions:

1. (ω, s, T ) 7→ α(ω, s, T ) and (ω, s, T ) 7→ σ(ω, s, T ) are measurable with

respect to P ⊗ B([0, T ∗ ]).
2.1 Presentation of the model 17

2. For s > T we have α(ω, s, T ) = 0 and σ(ω, s, T ) = 0.

3. sups,T ≤T ∗ (|α(ω, s, T )| + |σ(ω, s, T )|) < ∞.

These conditions ensure that we can find a “joint version” of all f (t, T ) such that
(ω, t, T ) 7→ f (t, T )(ω)1l{t≤T } is O ⊗ B([0, T ∗ ])-measurable (see Eberlein, Jacod,
and Raible (2005)). Here, P and O denote the predictable and the optional
σ-field on Ω × [0, T ∗ ].
From the forward rates we can deduce explicit expressions for zero coupon
bond prices and the risk free savings account:

Lemma 2.1 The bond price B(t, T ) is given by

Zt Zt !
B(t, T ) = B(0, T ) exp (r(s) − A(s, T )) ds + Σ(s, T ) dLs , (2.2)
0 0

where
ZT ZT
A(s, T ) := α(s, u) du and Σ(s, T ) := σ(s, u) du. (2.3)
s∧T s∧T

Proof: The claim can be proved in the same way as proposition 3.1 in Özkan
(2002). Although the statement there is for a Lévy process and special semi-
martingale L, the proof neither uses the stationarity of the increments nor the
fact that L is a special semimartingale. Thus, it also applies in the present set-
ting. 
Rt
Setting T = t in lemma 2.1, the risk free savings account Bt := exp 0 r(s) ds
can be written as
Zt Zt !
1
Bt = exp A(s, t) ds − Σ(s, t) dLs . (2.4)
B(0, t)
0 0

This leads to the following representation for the bond price which will be useful
later:
Zt Zt !
B(0, T )
B(t, T ) = exp − A(s, t, T ) ds + Σ(s, t, T ) dLs , (2.5)
B(0, t)
0 0

where we used the abbreviations

A(s, t, T ) := A(s, T ) − A(s, t)

and
Σ(s, t, T ) := Σ(s, T ) − Σ(s, t). (2.6)
In the remaining part of this chapter, we consider only deterministic volatil-
ity structures. More precisely, we require that the driving process L satisfies
assumption (EM) (see chapter 1) as well as the following condition:
18 The Lévy term structure model

Assumption (DET). The volatility structure σ is deterministic and bounded.

For 0 ≤ s, T ≤ T ∗ we have

0 ≤ Σi (s, T ) ≤ M (i ∈ {1, . . . , d}) (2.7)

where Σ is given by (2.3) and M is the constant from assumption (EM). Note
that Σ(s, s) = 0.

Our goal in the remaining part of this section is to derive a condition on

the drift term that ensures the martingale property for discounted bond prices.
In other words, we are looking for a condition on the drift term that makes the
model work directly under a martingale measure. This will allow us to price
integrable contingent claims by taking the P-expectation of their discounted
payoffs.
Let us use proposition 1.9 with f (s) := Σ(s, T ) for a fixed T ∈ [0, T ∗ ]. The
proposition yields
" Zt !# Zt
IE exp Σ(s, T ) dLs = exp θs (Σ(s, T )) ds.
0 0
R•
If we set A(s, T ) := θs (Σ(s, T )) and X := 0 Σ(s, T ) dLs , this reads as

Zt
IE[exp Xt ] = exp A(s, T ) ds.
0

Lemma 2.1 implies the following expression for the discounted bond price:
1 exp Xt
Z(t, T ) := B(t, T ) = B(0, T ) .
Bt IE[exp Xt ]
X is a process with independent increments. Therefore
IE[exp Xt ]
IE[exp Xt | Fs ] = IE[exp(Xt − Xs )| Fs ] exp Xs = exp Xs .
IE[exp Xs ]

Hence, Z(·, T ) is a martingale. Summing up, we get the following proposition:

Proposition 2.2 Let

A(s, T ) := θs (Σ(s, T )), (2.8)
then for all T ∈ [0, T ∗ ] the discounted bond price process Z(t, T ) := 1
Bt B(t, T )
is a martingale.

Remark: The drift condition from proposition 2.2 ensures that P is a risk-
neutral measure. If the dimension of the driving process L is d = 1, P is the
unique martingale measure. For d ≥ 2 this property does not hold in general. A
discussion on the uniqueness of martingale measures in this model framework
can be found in Eberlein, Jacod, and Raible (2005). If there is more than one
martingale measure the problem of which one to choose arises. In this case, we
2.2 Tools for derivative valuation 19

assume that P is the risk-neutral measure chosen by the market and price inte-
grable contingent claims by taking the P-expectation of the discounted payoffs.

Note that in the special case of a driving standard Brownian motion, i.e.
θs (z) = hz,zi d
2 for z ∈ C , equation (2.8) is the well-known Heath–Jarrow–Morton
drift condition for the multifactor Gaussian HJM model. In all sections that
follow, the drift condition from proposition 2.2 is assumed to be in force. Ex-
pression (2.5) for the bond price can then be expressed as

 Zt  Zt 
B(0, T ) 
B(t, T ) = exp θs (Σ(s, t)) − θs (Σ(s, T )) ds + Σ(s, t, T ) dLs .
B(0, t)
0 0
(2.9)

2.2 Tools for derivative valuation

The aim of this section is to present the main mathematical tools that are
needed for derivative pricing in the subsequent sections. One method that will be
applied is the change-of-numeraire technique developed by Geman, El Karoui,
and Rochet (1995). It will prevent us from having to evaluate joint probabil-
ity laws and, therefore, save us time in the computation of derivative prices.
Standard and non-standard numeraires will be used, i.e. we employ forward
martingale measures as well as a measure that we call adjusted forward mea-
sure. The other pillar on which the derivation of pricing formulae for derivatives
will rest is an integral transform method. Integral transform methods are very
useful whenever the characteristic function or bilateral Laplace transform of the
underlying is known in closed form. They go back to Carr and Madan (1999)
who use Fourier transforms and to Raible (2000) whose approach is based on
bilateral Laplace transforms. In the context of deriving hedging strategies sim-
ilar methods have been used by Hubalek and Krawczyk (1998). The idea to use
characteristic functions for option pricing has already been applied in Heston
(1993).
Remember that the forward martingale measure for the settlement day T ,
denoted by PT , is defined by the Radon–Nikodym derivative

dPT 1
:= . (2.10)
dP BT B(0, T )

Usually, this measure is defined on (Ω, FT ) only, but we can and do define it on
(Ω, FT ∗ ). P and PT are equivalent and from (2.4) we get the explicit expression

 ZT ZT 
dPT
= exp − A(s, T ) ds + Σ(s, T ) dLs . (2.11)
dP
0 0
20 The Lévy term structure model

Restricted to the σ-field Ft for t ≤ T this becomes

 
dPT 1 B(t, T )
= IEP Ft =
dP Ft BT B(0, T ) Bt B(0, T )
Zt Zt !
= exp − A(s, T ) ds + Σ(s, T ) dLs . (2.12)
0 0

Let us derive some properties of L under PT . The next proposition provides

the semimartingale characteristics of L under the forward measure. Since we
assume (EM), L is a special semimartingale with respect to P. However, it might
not be a PT -special semimartingale (at least it is not clear at this point that it
actually is). Hence, we work with a general truncation function h:

Proposition 2.3 The semimartingale characteristics of L with respect to PT

associated with the truncation function h are given by
Zs Zs Z
(BsT )i (h) = Bsi (h) + ciu Σ(u, T ) du + hi (x)(ehΣ(u,T ),xi − 1)ν(du, dx),
0 0 Rd
CsT = Cs ,
ν T (ds, dx) = ehΣ(s,T ),xi ν(ds, dx),
where ciu denotes the i-th row of the matrix cu .

Proof: To derive the semimartingale characteristics we use Girsanov’s Theo-

rem for semimartingales as presented in Jacod and Shiryaev (2003, III.3.24).
That is, we look for two predictable processes β and Y describing the change of
measure. Denote by Z the density process as given in (2.12). Then, combining
proposition 2.2, equation (1.10), and the canonical decomposition of L, we get
Zs Zs
1
Zs = exp − hΣ(u, T ), bu i du − hΣ(u, T ), cu Σ(u, T )i du
2
0 0
Zs Z  
hΣ(u,T ),xi
− e − 1 − hΣ(u, T ), xi ν(du, dx)
0 Rd
Zs Zs
√
+ hΣ(u, T ), bu i du + cu Σ(u, T ) dWu
0 0
Zs Z !
L
+ hΣ(u, T ), xi(µ − ν)(du, dx)
0 Rd
•
√
Z
= Es cu β(u) dWu
0
Z• Z !
+ (Y (u, x) − 1) (µL − ν)(du, dx)
0 Rd
2.2 Tools for derivative valuation 21

where β(u) := Σ(u, T ) and Y (u, x) := exphΣ(u, T ), xi are the candidates for
Girsanov’s Theorem. The last equality follows from Kallsen and Shiryaev (2002,
Lemma 2.6). The following lemma shows that these candidates meet the pre-
requisites of the theorem (compare also Shiryaev (1999, Chapter VII, Section
3g, Theorem 1)). We obtain the characteristics as given above. 

Lemma 2.4 Define β, Y , and Z as above. Then

Zt
c i,c
hZ , L it = Zu− ciu β(u) du
0

 
Z
and Y is a nonnegative version of MµPL Z− 1l{Z− >0} P
e .

Remark: Here, h·, ·i· denotes the angle bracket relative to P and MµPL is a
positive measure on (Ω × [0, T ∗ ] × Rd , F ⊗ B([0, T ∗ ]) ⊗ B(Rd )) defined by
MµPL (W ) := IE[W ∗ µT ∗ ] for a measurable nonnegative function W . MµPL (·|P)
e
P
denotes the “conditional expectation” relative to MµL with respect to the σ-
e := P ⊗ B(Rd ). For more details we refer to Jacod and Shiryaev (2003,
field P
Section III.3c).

Proof: To show the first claim, define a process N by

Zs Zs Z
√
Ns := cu β(u) dWu + (Y (u, x) − 1)(µL − ν)(du, dx).
0 0 Rd

N is well defined since (Y − 1)2 ∗ νT ∗ < ∞ and thus, by Jacod and Shiryaev
R s (Y − 1) ∈ Gloc . Moreover, Z = E(N ) or, written
(2003, Theorem II.1.33),
differently, Zs = 1 + 0 Zu− dNu . Thus, for 1 ≤ i ≤ d
*Z•  Z• i +
c i,c √
hZ , L it = Zs− dNsc , cs dWs
0 0 t
Zt *  Z• i +
√
= Zu− d N c , cs dWs
0 0 u
Zt
= Zu− ciu β(u) du.
0

To prove the second claim, we have to show that for any P-measurable
e
P P Z
nonnegative U the equation MµL (Y U ) = MµL ( Z− 1l{Z− >0} U ) holds. Since
22 The Lévy term structure model

( ZZ− 1l{Z− >0} )s = exphΣ(s, T ), ∆Ls i, we get

" ZT ∗Z #
IE Y (s, x)U (s, x)µL (ds, dx)
0 Rd
" #
X
hΣ(s,T ),∆Ls i
= IE e U (s, ∆Ls )1l{∆Ls 6=0}
0≤s≤T ∗
" ZT ∗Z #
Zs L
= IE 1l U (s, x)µ (ds, dx)
Zs− {Zs− >0}
0 Rd

and the assertion is proved. 

Lemma 2.5 L is a special semimartingale and a non-homogeneous Lévy pro-

cess with respect to PT .

Proof: To verify that L is a special semimartingale, note that

ZT ∗Z
(|x|2 ∧ |x|) ∗ νTT ∗ = (|x|2 ∧ |x|)ehΣ(s,T ),xi Fs (dx) ds
0 Rd
ZT ∗ Z
≤ C1 |x|2 Fs (dx) ds
0 {|x|≤1}

ZT ∗ Z
+ |x|ehΣ(s,T ),xi Fs (dx) ds
0 {|x|>1}

for some constant C1 and use the arguments of the proof of lemma 1.7.
L is a process with independent increments since there is a deterministic
version of its semimartingale characteristics given in proposition 2.3 (see Jacod
and Shiryaev (2003, II.4.15)). The same theorem (or alternatively proposition
1.9) can be used to calculate the characteristic function of L under PT :

h
ihu,Ls i
i 1
IEPT e = exp ihu, BsT (h)i − hu, Cs ui
2
s
Z Z 
ihu,xi T
+ (e − 1 − ihu, h(x)i)ν (dt, dx) .
0 Rd

We can write this characteristic function in the same form as in (1.1) with
Z  
T
bs = bs + cs Σ(s, T ) + ehΣ(s,T ),xi − 1 x1l{|x|≤1} Fs (dx),
Rd
cTs = cs ,
FsT (dx) = ehΣ(s,T ),xi Fs (dx).
2.2 Tools for derivative valuation 23

It can easily be checked that bT, cT , and F T satisfy an equation analogous to

(1.2). Consequently, L is a non-homogeneous Lévy process with respect to PT .
Of course, path properties are preserved under the equivalent change of mea-
sure. 

Another measure of which we will make use in the context of pricing floating
range notes is the following:
For T 0 < T we define the adjusted forward measure PT 0 ,T on (Ω, FT ∗ ) via

dPT 0 ,T F (T 0 , T 0 , T ) B(0, T )
:= 0
= , (2.13)
dPT F (0, T , T ) B(0, T 0 )B(T 0 , T )
0
where F (·, T 0 , T ) := B(·,T )
B(·,T ) denotes the forward price process. Restricting this
density to Ft for t ≤ T 0 we get
dPT 0 ,T F (t, T 0 , T ) B(0, T )B(t, T 0 )
= = (2.14)
dPT Ft F (0, T 0 , T ) B(0, T 0 )B(t, T )

since (F (t, T 0 , T ))0≤t≤T 0 is a PT -martingale. Thus we have

dPT 0 ,T B(0, T )B(t, T 0 ) B(t, T ) B(t, T 0 )

= = ,
dP Ft B(0, T 0 )B(t, T ) Bt B(0, T ) Bt B(0, T 0 )

i.e. the forward measure PT 0 and the adjusted forward measure PT 0 ,T are equal
once restricted to (Ω, Ft ) for t ≤ T 0 . However, on (Ω, Ft ) for t > T 0 they are
usually different. Choose for example T 0 < t < T , then in general
dPT 0 ,T B(t, T )
=
dP Ft B(T 0 , T )Bt B(0, T 0 )
Zt Zt !
(2.2) 1
= exp − A(s, T ) ds + Σ(s, T ) dLs
BT 0 B(0, T 0 )
T0 T0
1
6 =
BT 0 B(0, T 0 )
dPT 0
= .
dP Ft

Using (2.2) and (2.4) we can write the density process Z 0 of PT 0 ,T with
respect to P as
Zt Zt !
dPT 0 ,T
Zt0 = = exp − AT 0 ,T (s) ds + ΣT 0 ,T (s) dLs , (2.15)
dP Ft
0 0

where
AT 0 ,T (s) := A(s, T 0 )1l{s≤T 0 } + A(s, T )1l{s>T 0 } (2.16)
and
ΣT 0 ,T (s) := Σ(s, T 0 )1l{s≤T 0 } + Σ(s, T )1l{s>T 0 } . (2.17)
24 The Lévy term structure model

Proceeding as in the proof of proposition 2.3, the two predictable processes in

Girsanov’s Theorem can be identified as
β(u) = ΣT 0 ,T (u) and Y (u, x) = exphΣT 0 ,T (u), xi.
Similar arguments as in the proof of lemma 2.5 lead to the conclusion that
L is also a non-homogeneous Lévy process and a special semimartingale with
respect to the adjusted forward measure PT 0 ,T .
The integral transform method we are going to apply traces back to Raible
(2000). The key idea is to express derivative prices as convolutions and perform
Laplace transformations followed by inverse Laplace transformations. This tech-
nique exploits the fact that the Laplace transform of a convolution equals the
product of the Laplace transforms of the convolution factors. It is especially
useful whenever these Laplace transforms of the convolution factors can easily
be calculated or are even known in closed form. We cite the following theo-
rems from Raible (2000, Theorems B.2 and B.3) since they will be used very
frequently in the subsequent sections and chapters:
Theorem 2.6 Let F1 and F2 be measurable complex-valued functions on the
real line. Let z ∈ C and R := <z. If
Z Z
−Rx
e |F1 (x)| dx < ∞ and e−Rx |F2 (x)| dx < ∞
R R

and if x 7→ e−Rx |F1 (x)| is bounded, then the convolution F (x) := F1 ∗ F2 (x)
exists and is continuous for all x ∈ R, and we have
Z
e−Rx |F (x)| dx < ∞
R

and Z Z Z
e−zx F (x) dx = e−zx F1 (x) dx e−zx F2 (x) dx.
R R R

Theorem 2.7 Let F be a measurable complex-valued function on the real line.

Let R ∈ R be such that
Z
f (z) = e−zx F (x) dx (z ∈ C, <z = R),
R

with the integral converging absolutely for z = R. Let x ∈ R be such that the
integral
R+i∞
Z
ezx f (z) dz
R−i∞
exists as a Cauchy principal value. Assume that F is continuous at the point x.
Then
R+i∞
Z
1
F (x) = ezx f (z) dz,
2πi
R−i∞
2.3 Valuation of caps and floors 25

where the integral is to be understood as the Cauchy principal value if the inte-
grand is not absolutely integrable.

2.3 Valuation of caps and floors

Our goal in this section is to derive explicit formulae for the prices of caps and
floors that can numerically be evaluated fast. This is crucial in order to be able
to calibrate the model to market prices of caps and floors in reasonable time.
Remember that a cap (resp. floor) is a series of call (resp. put) options on
subsequent variable rates. These single options are called caplets (resp. floorlets).
Each caplet is equivalent to a put option on a zero coupon bond, each floorlet
can be considered as a call option (see section 1.1 or James and Webber (2000,
3.1.5)). Thus, if we derive suitable formulae for calls and puts on zero coupon
bonds, we immediately have formulae for caps and floors.
As described in a previous section, the discounted bond price process Z(·, T )
is a martingale with respect to the measure P and the given filtration for each
T ∈ [0, T ∗ ]. Consequently, we can price an integrable contingent claim by taking
the conditional expectation of the discounted payoff. The time-s value of a call
with strike K and maturity t on a bond which matures at T is then given by
 
1 +
Cs (t, T, K) := IE (B(t, T ) − K) |Fs (s ≤ t).
Bt

To determine today’s value of the call we need to evaluate IE Bt−1(B(t, T )−K)+ .

 

A straightforward approach is to derive the joint (conditional) distribution of

the random variables Bt and B(t, T ). Although this can easily be done ana-
lytically (compare Eberlein and Raible (1999)), the numerical evaluation of the
resulting expression is extremely time consuming. Instead, we use the change-of-
numeraire technique to circumvent the calculation of the joint probability law,
i.e. we switch from the spot martingale measure P to the forward martingale
measure for the settlement day t, denoted by Pt . Equation (2.10) for the density
and expression (2.9) for the bond price lead to the following price for the call:

C0 (t, T, K) = B(0, t)IEPt [(B(t, T ) − K)+ ]

= B(0, t)IEPt [(D exp(X) − K)+ ],

where
Zt  !
B(0, T ) 
D := exp θs (Σ(s, t)) − θs (Σ(s, T )) ds
B(0, t)
0

is deterministic and
Zt
X := Σ(s, t, T ) dLs
0

is Ft -measurable. To calculate the option price we only need to know the distri-
bution of X under the measure Pt , which we denote by PX t . Suppose that this
26 The Lévy term structure model

distribution possesses a Lebesgue-density ϕ in R, then

Z
C0 (t, T, K) = B(0, t) (Dex − K)+ ϕ(x) dx. (2.18)
R

Before deriving a formula for the option price, let us shortly discuss the
assumption that PX t possesses a Lebesgue-density. This distribution possesses a
Lebesgue-density if and only if it is absolutely continuous (with respect to the
Lebesgue-measure on R). Since P and Pt are equivalent, this is the case if and
only if the distribution of X with respect to P, denoted by PX , is absolutely
continuous. Whether or not PX is absolutely continuous cannot be answered in
general. The answer depends on the choice of the volatility structure and the
driving process, as the following examples show:

1. Let Σ(s, t) = Σ(s, T ) for s ∈ [0, t] (i.e. Σ(s, t, T ) = 0), then X = 0 and
PX cannot be absolutely continuous.

2. Choose the Ho–Lee volatility structure, i.e. Σ(s, T ) = σb(T − s), and let
L be a Poisson process, then X = σ b(T − t)Lt , whose distribution is not
absolutely continuous since the distribution of Lt is not.

The following proposition gives sufficient conditions for absolute continuity:

Proposition 2.8 Assume that Σ(s, t, T ) 6= 0 for s ∈ [0, t]. Then each of the
following conditions implies that PX is absolutely continuous with respect to the
Lebesgue-measure λ\:

1. There is a Borel set S ⊂ [0, t] with λ\(S) > 0 such that cs is positive
definite for s ∈ S.

2. Denote by Φs the characteristic function associated with the Lévy–Khint-

chine triplet (bs , cs , Fs ), i.e. for u ∈ Rd
 Z 
1
Φs (u) = exp ihu, bs i − hu, cs ui + (eihu,xi − 1 − ihu, xi)Fs (dx)
2
Rd
= exp(θs (iu)).

Then
|Φs (u)| ≤ C exp (−γ|u|η ) (s ∈ [0, t])

for real constants C, γ > 0, η > 0 that do not depend on s.

Proof: We show that ΦX , i.e. the characteristic function of X, is integrable.

Using proposition 1.9 we get

Zt
X
Φ (u) = exp θs (iuΣ(s, t, T )) ds (u ∈ R). (2.19)
0
2.3 Valuation of caps and floors 27

Let us first suppose that condition 1 is satisfied and define L1 and L2 by

Zt Zt
√
L1t := bs ds + cs dWs ,
0 0
Zt Z
L2t := x(µ − ν)(ds, dx).
0 Rd

Both processes are PIIACs and L = L1 + L2 . By (2.19) and using proposition

1.9 on L1 and L2 , we have
1 2
ΦX (u) = ΦX (u)ΦX (u),
h  R i
j t
where ΦX (u) := IE exp iu 0 Σ(s, t, T ) dLjs for j ∈ {1, 2}. Since both fac-
tors are bounded above by 1, it is enough to show that one of them is integrable.
But
 Zt 
X1 1
Φ (u) = exp − u2 hΣ(s, t, T ), cs Σ(s, t, T )i ds
2
0
1  
=: exp − u2 C ,
2
where C > 0 due to the positive definiteness of cs for s ∈ S and the fact that
1
λ\(S) > 0. Hence ΦX is integrable.
Now suppose condition 2 is satisfied. Then by (2.19)

Zt  
X
|Φ (u)| = exp < θs (iuΣ(s, t, T )) ds
0
Zt
= exp log |Φs (uΣ(s, t, T ))| ds
0
Zt  
≤ exp log C exp(−γ|uΣ(s, t, T )|η ) ds
0
 Zt 
t η η
= C exp − γ|u| |Σ(s, t, T )| ds
0
=: C t exp (−e
γ |u|η ) ,

e > 0 since Σ(s, t, T ) 6= 0 for s ∈ [0, t]. Consequently, ΦX is integrable. 

where γ

Let us come back to option pricing and denote by MtX the moment gene-
rating function of the random variable X with respect to the measure Pt . The
next theorem gives an analytic expression for the price of the call:
28 The Lévy term structure model

Theorem 2.9 Suppose that the distribution of X possesses a Lebesgue-density.

Choose an R < −1 such that MtX (−R) < ∞. Then we have

Z∞
1 1
C0 (t, T, K) = KB(0, t)eRξ eiuξ M X (−R − iu) du
2π (R + iu)(R + 1 + iu) t
−∞
(2.20)
with
Zt 
B(0, t) 
ξ := log − θs (Σ(s, t)) − θs (Σ(s, T )) ds + log K.
B(0, T )
0

Before proving the theorem let us point out that it is always possible to find an
R which satisfies the prerequisites of the theorem. This is part of the following
lemma which also gives an explicit expression for MtX (−R − iu):

Lemma 2.10 Choose M and ε in assumption (EM) such that Σ(s, T ) ≤ M 0

componentwise for an M 0 < M and for all s, T ∈ [0, T ∗ ]. Then, for each R ∈
−M 0
[−1 − MM 0 , −1) we have MtX (−R) < ∞. Moreover, for z ∈ C with <z = −R

Zt  
MtX (z) = exp θs (zΣ(s, T ) + (1 − z)Σ(s, t)) − θs (Σ(s, t)) ds. (2.21)
0

M −M 0
Proof: Fix R ∈ [−1 − M 0 , −1). For z ∈ C with <z = −R we have

< zΣi (s, T ) + (1 − z)Σi (s, t) = < Σi (s, T ) + (1 − z)(Σi (s, t) − Σi (s, T ))

≤ Σi (s, T ) + |1 + R||Σi (s, t) − Σi (s, T )|

M − M0 0
≤ M0 + M = M.
M0
Hence,
" Zt !#
MtX (z) = IEPt exp z (Σ(s, T ) − Σ(s, t)) dLs
0
Zt !
(2.11)
= exp − A(s, t) ds
0
" Zt !#
× IEP exp (zΣ(s, T ) + (1 − z)Σ(s, t)) dLs
0
Zt  
= exp θs (zΣ(s, T ) + (1 − z)Σ(s, t)) − θs (Σ(s, t)) ds,
0

where the last equality follows from (2.8) and proposition 1.9. In particular,
MtX (−R) is finite. 
2.3 Valuation of caps and floors 29

Proof of theorem 2.9: The arguments are similar to the proof of theorem
3.2 in Raible (2000).
Using representation (2.18) for the option price and defining ξ := − log D +
log K and v(x) := (e−x − 1)+ yields
Z
C0 (t, T, K) = B(0, t) (Dex − K)+ ϕ(x) dx
R
Z
= KB(0, t) (DK −1 ex − 1)+ ϕ(x) dx
R
Z
= KB(0, t) v(ξ − x)ϕ(x) dx
R
= KB(0, t)(v ∗ ϕ)(ξ) =: V (ξ).

We apply theorem 2.6 to the functions F1 (x) := v(x) and F2 (x) := ϕ(x), that is
we express the bilateral Laplace transform of their convolution as the product
of the bilateral Laplace transforms of the convolution factors. The prerequisites
of the theorem are satisfied since x 7→ e−Rx v(x) is bounded,
Z
1
e−Rx |v(x)| dx = < ∞,
R(R + 1)
R

and Z Z
−Rx
e |ϕ(x)| dx = e−Rx ϕ(x) dx = MtX (−R) < ∞
R R
by assumption. The cited theorem yields

L[V ](R + iu) = B(0, t)KL[v](R + iu)L[ϕ](R + iu) (u ∈ R)

where L[V ] denotes the bilateral Laplace transformRof V (analogously for v and
ϕ). It also yields that ξ 7→ V (ξ) is continuous and R e−Rξ V (ξ) dξ is absolutely
convergent. Therefore, we may apply theorem 2.7 and get
R+iY
Z
1
V (ξ) = lim ezξ L[V ](z) dz
2πi Y →∞
R−iY
ZY
1
= lim e(R+iu)ξ L[V ](R + iu) du
2π Y →∞
−Y
ZY
1
= B(0, t)KeRξ lim eiuξ L[v](R + iu)L[ϕ](R + iu) du,
2π Y →∞
−Y

if this limit exists. We have

L[ϕ](R + iu) = MtX (−R − iu)

30 The Lévy term structure model

and, according to Raible (2000, p. 66),

1
L[v](R + iu) = .
(R + iu)(R + 1 + iu)

The above limit exists (moreover, the integral converges absolutely) due to the
fact that |eiuξ | = 1, |MtX (−R − iu)| ≤ MtX (−R) < ∞ independent of u and
1
R
R (R+iu)(R+1+iu) du < ∞. This proves our assertion. 

In a similar manner we can derive the price P0 (t, T, K) for a put with strike
K and maturity t on a bond which matures at T :

Corollary 2.11 Suppose the distribution of X possesses a Lebesgue-density.

Choose an R > 0 such that MtX (−R) < ∞. Then we have
Z∞
1 1
P0 (t, T, K) = KB(0, t)eRξ eiuξ M X (−R − iu) du
2π (R + iu)(R + 1 + iu) t
−∞
(2.22)
with
Zt 
B(0, t) 
ξ := log − θs (Σ(s, t)) − θs (Σ(s, T )) ds + log K,
B(0, T )
0

Remark: Note that the formulae for the call and the corresponding put coin-
cide. The difference is in the permitted values for R.

2.4 Swaption pricing

In this section, an explicit formula for pricing swaptions is derived under an
additional assumption on the volatility structure. The numerical evaluation of
the pricing formula can be done fast. Once again, this is crucial for calibration
purposes.
Remember that a swaption is an option on a forward swap, i.e. on a swap
which starts in the future. At maturity of the option the holder has the right to
enter into the swap at a pre-specified fixed rate. There are payer and receiver
swaptions giving their owners the right to enter into the swap as fixed rate
payer or receiver respectively. We interpret the swaption as a right to exchange
a coupon bond having the fixed rate of the swap as its coupon against a floater,
whose value is always equal to 1. Thus, a receiver (resp. payer) swaption can be
seen as a call (resp. put) on a coupon bond with an exercise price of 1 (compare
Musiela and Rutkowski (1998, Section 16.2.3)).
We price options on coupon bonds in our model framework with the follow-
ing restriction on the volatility structure:
Assumption (VOL). For all T ∈ [0, T ∗ ] we have σ(·, T ) 6≡ (0, . . . , 0) and

σ(s, T ) = σ2 (T )σ1 (s) (0 ≤ s ≤ T )

2.4 Swaption pricing 31

where σ1 : [0, T ∗ ] → Rd and σ2 : [0, T ∗ ] → R+ are continuously differentiable.

Remarks:
1. Under assumption (VOL), the short rate process r is Markovian. This can
be proved using exactly the same arguments as in Eberlein and Raible
(1999, Theorems 4.2 and 4.3) even though L does not possess stationary
increments.

2. In the one-dimensional case (d = 1) and for a homogeneous Lévy process

L which is not identically zero, the converse is also true. That is, a Marko-
vian short rate implies that the volatility structure factorizes as above.
This has been proved by Eberlein and Raible (1999) under an additional
assumption on L and by Küchler and Naumann (2003) in the general case.

3. In this setting, P is the unique martingale measure. This is an immediate

consequence of Eberlein, Jacod, and Raible (2005, Theorem 6.4) since the
dimension dt of the vector space Et := span(Σ(t.T ) : T ∈ [0, T ∗ ]) satisfies
dt ≤ 1 for all t and we can thus Rreduce the dimensionality of the model
•
to one (e.g. by using the PIIAC 0 σ1 (s) dLs as driving motion).

Example 2.12 The following volatility structures satisfy the condition above:
1. σ(s, T ) = σ
b (Ho–Lee volatility structure),

be−a(T −s)
2. σ(s, T ) = σ (Vasiček volatility structure),
1+γT
3. σ(s, T ) = σ
b 1+γs e−a(T −s) (Moraleda–Vorst volatility structure)
b, γ > 0 and a 6= 0.
for real constants σ

To price options on coupon bonds we use an idea of Jamshidian (1989) as

well as change of numeraire and Laplace transformation techniques. Denote by
BC (t, T1 , . . . , Tn ) the time t price of a coupon bond with maturity Tn paying to
its owner an amount of C1 , . . . , Cn at the dates T1 , . . . , Tn . Then, for 0 ≤ t < T1

BC (t, T1 , . . . , Tn ) = C1 B(t, T1 ) + C2 B(t, T2 ) + . . . + Cn B(t, Tn ).

The time-0 price of a call with strike price 1 and maturity t on that bond is
obtained by taking the expectation of the discounted payoff, i.e.

C0 := C0 (t, T1 , . . . , Tn , C1 , . . . , Cn )
n
" #
1 X +
:= IE Ci B(t, Ti ) − 1
Bt
i=1
" n #
X +
= B(0, t)IEPt Ci B(t, Ti ) − 1
i=1

 Zt
 
n
X  +
= B(0, t)IEPt  Di exp Σ(s, t, Ti ) dLs − 1  .
i=1 0
32 The Lévy term structure model

For the last equation we used (2.9) and defined the constants

Zt  !
B(0, Ti ) 
Di := Ci exp θs (Σ(s, t)) − θs (Σ(s, Ti )) ds .
B(0, t)
0

Since for 0 ≤ s ≤ t ≤ T ≤ T ∗

ZT ZT
Σ(s, t, T ) = σ(s, u) du = σ2 (u) du σ1 (s),
t t

we get
R Ti
σ2 (u) du
Σ(s, t, Ti ) = R tTn Σ(s, t, Tn ).
t σ2 (u) du
Hence,
n
" + #
X
Bi X
C0 = B(0, t)IEPt Di e −1 , (2.23)
i=1

where R Ti
σ2 (u) du
0 < Bi := R tTn ≤1 (i ∈ {1, . . . , n})
t σ2 (u) du
and
Zt
X := Σ(s, t, Tn ) dLs .
0

We only need to know the distribution of X with respect to the measure Pt

to calculate the option price. Suppose that this distribution PX
t possesses a
Lebesgue-density in R, then
n
Z X +
Bi x
C0 = B(0, t) Di e − 1 ϕ(x) dx, (2.24)
R i=1

dPX
where ϕ := dλt\ .
We proceed as in the previous section by performing a Laplace transforma-
tion followed by an inverse Laplace transformation. As before, denote by MtX
the moment generating function of the random variable X with respect to the
measure Pt .

Theorem 2.13 Suppose the distribution of X possesses a Lebesgue-density.

Choose an R < −1 such that MtX (−R) < ∞ and let Z be the unique zero
of the strictly increasing and continuous function
n
X
g(x) := Di eBi x − 1.
i=1
2.4 Swaption pricing 33

Then we have
ZY
1
C0 = B(0, t) lim L[v](R + iu)MtX (−R − iu) du, (2.25)
2π Y →∞
−Y

where L[v] denotes the bilateral Laplace transform of v : R → R defined by

v(x) := (g(−x))+ . Moreover,
X n   
−1 1
L[v](R + iu) = e(R+iu)Z Di eBi Z + (u ∈ R)
Bi + R + iu R + iu
i=1

and for z ∈ C with <z = −R

Zt  
MtX (z) = exp θs (zΣ(s, Tn ) + (1 − z)Σ(s, t)) − θs (Σ(s, t)) ds.
0

Proof: Since g is a strictly increasing and continuous function taking negative

as well as positive values, it has a unique zero. Consequently, there is a real
number Z such that
(g(x))+ = 1l[Z,∞) (x)g(x).
Define v(x) := (g(−x))+ , then (2.24) implies
C0 = B(0, t)(v ∗ ϕ)(0).
First, we calculate the bilateral Laplace transform of v:
Z n
!+
X
L[v](z) := e−zx Di e−Bi x − 1 dx
R i=1

Z−Z n
X
!
−zx −Bi x
= e Di e − 1 dx
−∞ i=1

n
X Z−Z ! Z−Z
= Di e−zx e−Bi x dx − e−zx dx.
i=1 −∞ −∞

For any constant 0 ≤ C ≤ 1 and any z ∈ C with <z < −C

Z−Z Z0
−(C+z)x
e dx = e−(C+z)(x−Z) dx
−∞ −∞
Z1
1
= e(C+z)Z t−(C+z) dt
t
0
(C+z)Z
=e B(−C − z, 1)
Γ(−C − z)Γ(1)
= e(C+z)Z
Γ(−C − z + 1)
−1
= e(C+z)Z .
C +z
34 The Lévy term structure model

B(·, ·) and Γ(·) denote the Euler Beta and Gamma function respectively. For
some comments on the chain of equalities we refer to Raible (2000, p. 66)
where the bilateral Laplace transform of a similar function is derived. It can be
concluded that the Laplace transform of v exists for z ∈ C with <z < −1 and
that it is given by

n   !
X −1 1
L[v](z) = ezZ Di eBi Z + .
Bi + z z
i=1

Again, we apply theorem 2.6 to the functions F1 (x) := v(x) and F2 (x) := ϕ(x)
and, proceeding similar to the proof of theorem 2.9, we obtain

ZY
1
C0 = B(0, t) lim L[v](R + iu)MtX (−R − iu) du,
2π Y →∞
−Y

if this limit exists. The next lemma shows that it exists, although the integral
does not converge absolutely as it does in formulae (2.20) and (2.22). The ex-
pression for MtX can be derived as in lemma 2.10. 

Lemma 2.14 For any C < 0 and under the assumptions of theorem 2.13 the
limit
ZY
1
lim eiuZ M X (−R − iu) du
Y →∞ C + iu t
−Y

exists.

Proof: As before, let ϕ denote the Lebesgue-density of PX

t . For Y ∈ R+
Fubini’s theorem yields

ZY
1
I(Y ) := eiuZ M X (−R − iu) du
C + iu t
−Y
Z Z !
1
= 1l{|u|≤Y } eiuZ e−(R+iu)x ϕ(x) dx du
C + iu
R R
!
−
Z Z
C iu
= 1l{|u|≤Y } eiu(Z−x) 2 du e−Rx ϕ(x) dx
C + u2
R R
Z
= 2J(x, Y )e−Rx ϕ(x) dx,
R
2.4 Swaption pricing 35

with
ZY  
iu(Z−x) C − iu
J(x, Y ) := < e du
C 2 + u2
0
ZY
C
= cos(u(Z − x)) du
C2 + u2
0
ZY ZY
1 C2
+ sin(u(Z − x)) du − sin(u(Z − x)) du
u uC 2 + u3
0 0
1 2 3
=: J (x, Y ) + J (x, Y ) − J (x, Y ).

If we can show that J(x, Y ) is bounded by a constant that does not depend on
x or Y and that limY →∞ J(x, Y ) exists for allR x, this will imply the existence
of limY →∞ I(Y ) (remember that MtX (−R) = R e−Rx ϕ(x) dx < ∞ by assump-
tion). It is clear that J 1 and J 3 have the two desired properties. Now let us
have a look at J 2 :
 
 0 for x = Z 

 |Z−x|Y 

2
J (x, Y ) = = Si(|Z − x|Y ),
Z
1
 sin t dt for x 6= Z 

 t 

0

where Si denotes the sine integral. From the properties of the sine integral we
can conclude that J 2 (x, Y ) is bounded by a constant that does not depend on
x or Y and that limY →∞ J 2 (x, Y ) exists. 

In a similar manner as for the call we can derive the price P0 of a put with
strike 1 and maturity t on the coupon bond:
Corollary 2.15 Suppose the distribution of X possesses a Lebesgue-density.
Choose an R > 0 such that MtX (−R) < ∞ and let Z be the unique zero of the
strictly decreasing and continuous function
n
X
g(x) := 1 − Di eBi x .
i=1

Then we have
ZY
1
P0 = B(0, t) lim L[v](R + iu)MtX (−R − iu) du, (2.26)
2π Y →∞
−Y

where L[v] denotes the bilateral Laplace transform of v : R → R defined by

v(x) := (g(−x))+ . Moreover,
n   !
X −1 1
L[v](R + iu) = e(R+iu)Z Di eBi Z + (u ∈ R)
Bi + R + iu R + iu
i=1
36 The Lévy term structure model

and for z ∈ C with <z = −R

Zt  
MtX (z) = exp θs (zΣ(s, Tn ) + (1 − z)Σ(s, t)) − θs (Σ(s, t)) ds.
0

Remark: We can observe a similarity to the pricing formulae for calls and puts
on zero coupon bonds. The formulae for the call and the respective put on a
coupon bearing bond coincide. Different are again only the permitted values for
R.

2.5 Valuation of floating range notes

Turnbull (1995) as well as Navatte and Quittard-Pinon (1999) derived ex-
plicit pricing formulae for floating range notes in a one-factor Gaussian Heath–
Jarrow–Morton (HJM) model. Nunes (2004) extended their results to a multi-
factor Gaussian HJM framework. The main aim of this section is to generalize
their results by providing explicit valuation formulae for floating range notes in
the Lévy term structure model.
Range notes are structured products, convenient for investors with a strong
belief that interest rates will stay within a certain corridor. They provide in-
terest payments which are proportional to the time in which a reference index
rate (most commonly the Libor rate) lies inside that range. In return for the
drawback that no interest will be paid for the time the corridor is left, they offer
higher rates than comparable standard products, like e.g. floating rate notes.
Floating range notes pay coupon rates which are linked to some reference index
rate (e.g. 3-month Libor plus 100 basis points) whereas the coupon rates of
fixed range notes are specified in advance. Let us stress that coupon payments
of both products depend on the path of the reference index rate.
Turnbull (1995) provided an explicit valuation formula for floating range
notes in the one-factor Gaussian HJM framework. Using the same model and
the change-of-numeraire technique developed by Geman, El Karoui, and Rochet
(1995), Navatte and Quittard-Pinon (1999) derived a pricing solution in a more
intuitive way. For this purpose, they introduced double delayed digital options.
The value of each floating range note coupon is shown to be equal to the value
of a portfolio of those options plus some additional term. This extra term only
involves the cumulative density function of a standard normal distribution.
Nunes (2004) managed to generalize the former results to a multifactor Gaussian
HJM model. His valuation formula for floating range notes looks very similar,
i.e. each coupon is written as a portfolio of delayed digital options plus some
extra term. This extra term, although given in closed form, is quite complicated
and stems from evaluating the joint probability law of two random variables.
One purpose of this section is to show that the calculation of the joint prob-
ability distribution can be circumvented by changing the probability measure
in a suitable way. Concretely, we will use the adjusted forward measure that
has been introduced in section 2.2. Proceeding this way, a much simpler pricing
2.5 Valuation of floating range notes 37

formula can be obtained in the multifactor Gaussian HJM model (see theo-
rem 2.21). However, our main goal is to price range notes in the more general
framework of the Lévy term structure model. Once again, we use the change-
of-numeraire technique as well as a Laplace transform method. As a necessary
tool for pricing range notes (and a nice side result), we begin by deriving a
valuation formula for digital options.

2.5.1 Digital Options

In this section, we discuss the pricing of interest rate digital options. For con-
venience we adopt the notation of Nunes (2004).
A standard European interest rate digital call (put) with strike rate rk is a
financial security that pays an amount of one currency unit to its owner if and
only if the simply compounded interest rate for the period [T, T + δ] lies above
(below) rk at maturity T of the option. More precisely, the time-T value of this
option is given by

SD(Θ)T [rn (T, T + δ); rk ; T ] := 1l{Θrn (T,T +δ)>Θrk }

with  
1 1
rn (T, T + δ) := −1 , (2.27)
δ B(T, T + δ)
where Θ = 1 for a digital call and Θ = −1 for a digital put.
In accordance with Navatte and Quittard-Pinon (1999) and Nunes (2004)
we call an interest rate digital option delayed if option maturity T and payment
date T1 differ (T1 > T ). The time-T1 price of a delayed digital option is given
by
DD(Θ)T1 [rn (T, T + δ); rk ; T1 ] := 1l{Θrn (T,T +δ)>Θrk } ,
where again Θ = 1 for a delayed digital call and Θ = −1 for a delayed digital
put. Since a standard digital option is a special case of a delayed digital option
(T1 = T ), we will only consider the latter in the following.
Delayed range digital options provide a terminal payoff equal to 1 paid at
T1 if and only if at option maturity T (T ≤ T1 ) the underlying interest rate
lies inside a prespecified corridor. Consequently, the time-T1 price of a delayed
range digital option is

DRDT1 [rn (T, T + δ); rl ; ru ; T1 ] := 1l{rn (T,T +δ)∈[rl ,ru ]} .

By arbitrage arguments, the time-t prices (t ∈ [0, T1 ]) of delayed digital calls,

puts, and range options are related via

DRDt [rn (T, T + δ); rl ; ru ; T1 ] = B(t, T1 ) − DD(1)t [rn (T, T + δ); ru ; T1 ]

−DD(−1)t [rn (T, T + δ); rl ; T1 ].

Unfortunately, a call-put parity like

DD(1)t [rn (T, T + δ); rk ; T1 ] = B(t, T1 ) − DD(−1)t [rn (T, T + δ); rk ; T1 ] (2.28)
38 The Lévy term structure model

does not hold for all t ∈ [0, T1 ] (note that in case rn (T, T + δ) = rk equality fails
for t ∈ [T, T1 ]). However, equation (2.28) holds for t < T in models in which the
distribution of B(T, T + δ) does not have point masses (like e.g. in the Gaussian
HJM model with a reasonable volatility structure). If L is a Poisson process,
equation (2.28) might fail for t < T though. The technique that we are going
to present for option valuation only works for model specifications that do not
produce point masses in the distribution of B(T, T + δ) (see proposition 2.8 and
the discussion preceding it). In these cases, we have the call-put parity (2.28)
for t < T and can thus price any of the mentioned digital options if we are able
to price a delayed digital call.
We calculate the value of the call by taking the P-conditional expectation
of its discounted payoff, i.e.

Dt := DD(1)t [rn (T, T + δ); rk ; T1 ]

 
1
= Bt IE 1l Ft
BT1 {rn (T,T +δ)>rk }
 
= B(t, T1 )IEPT1 1l{rn (T,T +δ)>rk } Ft
 
(2.27)
= B(t, T1 )IEPT1 1l B(T,T +δ)< 1
n o Ft
δrk +1

(2.2)
= B(t, T1 )
(2.5)
 
× IEPT1 1l B(t,T +δ) exp − T A(s,T,T +δ) ds+ T Σ(s,T,T +δ) dL < 1
n o Ft .
[ t s ] δr +1
R R
B(t,T ) t k

For the change of numeraire we used equations (2.11)–(2.12) and the abstract
Bayes formula. By independence of the increments of L and since B(t,T +δ)
B(t,T ) is Ft -
measurable, we get (compare Musiela and Rutkowski (1998, lemma A.0.1.(v)))
 
B(t, T + δ)
Dt = B(t, T1 )h (2.29)
B(t, T )
with h : R → [0, 1] given by
 
h(y) := IEPT1 1l y exp − R T A(s,T,T +δ) ds+R T Σ(s,T,T +δ) dL < 1
n o .
[ t t s ] δr +1
k

To calculate h(y) for y > 0, observe that

Z Z
h(y) = 1l eX < K dPT1 = 1lnex < K o dPX
n o
T1 (x), (2.30)
y y
Ω R

where
ZT
X := Σ(s, T, T + δ) dLs ,
t
ZT
1
K := exp A(s, T, T + δ) ds, (2.31)
δrk + 1
t
2.5 Valuation of floating range notes 39

and PX
T1 denotes the distribution of X under PT1 . If this distribution possesses
a Lebesgue-density ϕ in R then
Z
h(y) = 1lnex < K o ϕ(x) dx
y
R
Z
= fy (−x)ϕ(x) dx
R
= (fy ∗ ϕ)(0) = V (0) (2.32)

with fy (x) := 1lne−x < K o (x) and V (ξ) := (fy ∗ϕ)(ξ). Denote by MTX1 the moment
y
generating function of the random variable X with respect to the measure PT1 .
The next theorem gives an analytic expression for the price of the call that can
numerically be evaluated fast:
Theorem 2.16 Suppose the distribution of X possesses a Lebesgue-density.
Choose an R > 0 such that MTX1 (−R) < ∞. Then

Z∞  R+iu !
1 B(t, T ) 1
Dt = B(t, T1 ) < K M X (−R − iu) du
π B(t, T + δ) R + iu T1
0

with
ZT 
1 
K := exp θs (Σ(s, T + δ)) − θs (Σ(s, T )) ds.
δrk + 1
t

Remark: It is always possible to choose an R that satisfies the prerequisites of

the theorem (compare lemma 2.10). The particular choice of R does not – of
course – have an impact on the option price, but it has influence on the speed
at which the integral can be evaluated numerically (see Raible (2000, Section
3.7)).

Proof: We use the convolution representation (2.32) and apply theorem 2.6
to the functions F1 (x) := fy (x) and F2 (x) := ϕ(x), that is we express the
bilateral Laplace transform of their convolution as the product of the bilateral
Laplace transforms of the convolution factors. The prerequisites of the theorem
are satisfied since x 7→ e−Rx fy (x) is bounded,
Z  R
−Rx 1 K
e |fy (x)| dx = < ∞,
R y
R

and Z
e−Rx |ϕ(x)| dx = MTX1 (−R) < ∞
R

by assumption. The cited theorem together with (2.32) yields

L[V ](R + iu) = L[fy ](R + iu)L[ϕ](R + iu) (u ∈ R)

40 The Lévy term structure model

where L[V ] denotes the bilateral Laplace transform of V (analogously

R −Rξfor fy and
ϕ). The theorem also yields that ξ 7→ V (ξ) is continuous and that R e V (ξ) dξ
is absolutely convergent. Therefore, applying theorem 2.7 and proceeding as in
the proof of theorem 2.9 we obtain

ZY
1
V (0) = lim L[fy ](R + iu)L[ϕ](R + iu) du,
2π Y →∞
−Y

if this limit exists. Note that the integrand evaluated at u equals the complex
conjugate of the integrand evaluated at −u. Therefore, using the relationship
z + z̄ = 2 <(z) one arrives at

ZY
1  
V (0) = lim < L[fy ](R + iu)L[ϕ](R + iu) du.
π Y →∞
0

We have
L[ϕ](R + iu) = MTX1 (−R − iu)

and, since R > 0, one obtains

 R+iu
1 K
L[fy ](R + iu) =
(R + iu) y

and (after some calculations) concludes that the above limit exists. Plugging in
the expressions from (2.29), (2.31), and (2.32) as well as remembering the drift
condition, i.e. A(s, T, T + δ) = θs (Σ(s, T + δ)) − θs (Σ(s, T )), yields the claim.


Theorem 2.17 Under the assumptions of theorem 2.16 we have an explicit

expression for MTX1 , namely for u ∈ R

ZT h i
MTX1 (−R − iu) = exp θs (gs (−R − iu)) − θs (gs (0)) ds (2.33)
t

with gs (z) := zΣ(s, T, T + δ) + Σ(s, T1 ).

Proof: To obtain the expression for the moment generating function of X we

use equation (2.11), the independence of the increments of L, the fact that we
RT RT
have IE[exp t Σ(s, T ) dLs ] = exp t A(s, T ) ds, equation (2.8), and proposition
2.5 Valuation of floating range notes 41

1.9 (in this order) and get for z ∈ C with <(z) = −R

" ZT !#
MTX1 (z) = IEPT1 exp z Σ(s, T, T + δ) dLs
t
ZT1 !
= exp − A(s, T1 ) ds
0
" ZT ZT1 !#
× IE exp z Σ(s, T, T + δ) dLs + Σ(s, T1 ) dLs
t 0
ZT !
= exp − A(s, T1 ) ds
t
" ZT  
!#
× IE exp zΣ(s, T, T + δ) + Σ(s, T1 ) dLs
t
ZT ! ZT !
= exp − θs (gs (0)) ds exp θs (gs (z)) ds
t t

with gs (z) := zΣ(s, T, T + δ) + Σ(s, T1 ). Now (2.33) follows. 

Let us consider the multifactor Gaussian HJM model as a special case, i.e.
L is a d-dimensional standard Brownian motion under P. Then θ(x) = hx,xi
2 for
d
x ∈ C . From (2.33) and using (2.6) we get for z ∈ C

ZT 
hzΣ(s, T, T + δ) + Σ(s, T1 ), zΣ(s, T, T + δ) + Σ(s, T1 )i
MTX1 (z) = exp
2
t

hΣ(s, T1 ), Σ(s, T1 )i
− ds
2
 2 ZT
z
= exp ||Σ(s, T + δ) − Σ(s, T )||2 ds
2
t
ZT 
+z hΣ(s, T + δ) − Σ(s, T ), Σ(s, T1 )i ds .
t

Consequently, X is normally distributed under PT1 with mean

ZT
m(t, T, T + δ, T1 ) := hΣ(s, T + δ) − Σ(s, T ), Σ(s, T1 )i ds
t
42 The Lévy term structure model

and variance
ZT
g(t, T, T + δ) := ||Σ(s, T + δ) − Σ(s, T )||2 ds.
t

From (2.30) we get

K
log y
log K
!
y − m(t, T, T + δ, T1 )
Z  
K
h(y) = dPX
T1 (x) = PT1 X ≤ log =Φ p ,
y g(t, T, T + δ)
−∞

where Φ denotes the cumulative density function of a standard normal distri-

bution. Plugging in the expression for K from (2.31) and using (2.29) we arrive
at
 
log B(t,TB(t,T )
+δ)(δrk +1) + 1
2 g(t, T, T + δ) − l(t, T, T + δ, T 1 )
Dt = B(t, T1 )Φ  p ,
g(t, T, T + δ)

where
ZT
l(t, T, T + δ, T1 ) := hΣ(s, T + δ) − Σ(s, T ), Σ(s, T1 ) − Σ(s, T )i ds.
t

This formula coincides with the one derived in Nunes (2004, Proposition 3.3).
Note that for a standard digital call (T1 = T ) one gets l(t, T, T + δ, T1 ) = 0.

2.5.2 Range Notes

The purpose of this section is to derive a formula for pricing range notes in
the Lévy term structure model. As a special case, we will also consider the
multifactor Gaussian HJM model and obtain a pricing formula that is simpler
than the one provided by Nunes (2004). Once again, his notation is adopted.
In the following, we put ourselves at time t, the valuation date of the range
note. Consider a bond with bullet redemption having had its previous coupon
payment date at T0 (≤ t) and having its N future coupons paid at times Tj+1
(j = 0, . . . , N − 1). Based on some day count convention, let nj (δj ) denote the
number of days (years) between the times Tj and Tj+1 . For the current period
we split up n0 into the sum of n− +
0 and n0 , representing the number of days
between T0 and t and between t and T1 respectively. Furthermore, denote by
Tj,i the date that corresponds to i days after Tj and by δj,i the length (in years)
of the compounding period starting at time Tj,i .
For the multifactor Gaussian HJM model, Nunes (2004) shows that the
value of a fixed range note equals the value of a portfolio of delayed range
digital options. Although the Lévy term structure model is more general, the
same arguments apply since they do not depend on the driving process. We
refer the reader to Nunes (2004, Proposition 4.1) and concentrate on floating
range notes in what follows.
2.5 Valuation of floating range notes 43

To value floating range notes we will first switch from the spot measure to
a suitable forward measure. Afterwards, another change of measure from the
forward measure to an adjusted forward measure will be performed. Proceeding
this way, we will not have to deal with a joint probability distribution of two
random variables.
We cite the following definition from Nunes (2004, Definition 4.2):

Definition 2.18 For a floating range note, the value of the (j + 1)th coupon,
at time Tj+1 , is equal to

rn (Tj , Tj + δj ) + sj
νj+1 (Tj+1 ) := H(Tj , Tj+1 ),
Dj

where sj represents the spread over the reference interest rate paid by the bond
during the (j + 1)th compounding period, Dj is the number of days in a year for
the (j + 1)th compounding period, and
nj
X
H(Tj , Tj+1 ) := 1l{rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )}
i=1

denotes the number of days, in the (j + 1)th compounding period, that the refer-
ence interest rate lies inside a prespecified range, which is equal to [rl (Tj,i ), ru (Tj,i )]
for the ith day of the (j + 1)th compounding period.

Consequently, the time-t value of the floating range note is given by

N
X −1
F lRN (t) := B(t, TN ) + νj+1 (t)
j=0

where B(t, TN ) corresponds to the discounted value of the final payment of 1.

For the valuation of the first coupon we follow Nunes (2004) and get, since
rn (T0 , T0 + δ0 ) is already known at time t or, mathematically speaking, mea-
surable with respect to Ft ,
 
1 rn (T0 , T0 + δ0 ) + s0
ν1 (t) = Bt IE H(T0 , T1 ) Ft
BT1 D0
rn (T0 , T0 + δ0 ) + s0 h i
= B(t, T1 )IEPT1 H(T0 , T1 ) Ft
D0

rn (T0 , T0 + δ0 ) + s0
= B(t, T1 )H(T0 , t)
D0
Xn0 h i
+ B(t, T1 )IEPT1 1l{rl (T0,i )≤rn (T0,i ,T0,i +δ0,i )≤ru (T0,i )} Ft
i=n−
0 +1

rn (T0 , T0 + δ0 ) + s0
= B(t, T1 )H(T0 , t)
D0
Xn0 
+ DRDt [rn (T0,i , T0,i + δ0,i ); rl (T0,i ); ru (T0,i ); T1 ]
i=n−
0 +1
44 The Lévy term structure model

with
−
n0
X
H(T0 , t) := 1l{rl (T0,i )≤rn (T0,i ,T0,i +δ0,i )≤ru (T0,i )} .
i=1
For the subsequent coupons, we get
νj+1 (t)
 
rn (Tj , Tj+1 ) + sj
1
= Bt IE H(Tj , Tj+1 ) Ft
BTj+1 Dj
 
rn (Tj , Tj+1 ) + sj
= B(t, Tj+1 )IEPTj+1 H(Tj , Tj+1 ) Ft
Dj
  nj
(2.27) sj 1 X h i
= − B(t, Tj+1 ) IEPTj+1 1l{rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )} Ft
Dj δj Dj
i=1
nj  
B(t, Tj+1 ) X 1
+ IEPTj+1 1l{rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )} Ft
δj Dj B(Tj , Tj+1 )
i=1
1 2
=: νj+1 (t) + νj+1 (t).
1 (t) we proceed as before and get
To evaluate νj+1
 nj
X
1 sj 1
νj+1 (t) = − DRDt [rn (Tj,i , Tj,i + δj,i ); rl (Tj,i ); ru (Tj,i ); Tj+1 ].
Dj δj Dj
i=1

For the evaluation of 2 (t)

νj+1
we switch from the forward measure PTj+1 to the
adjusted forward measure PTj ,Tj+1 . This procedure has the advantage that we
do not have to deal with the joint distribution of the two random variables
B(Tj , Tj + δj ) and rn (Tj,i , Tj,i + δj,i ). Using the abstract Bayes formula together
with (2.13)–(2.14) and denoting by IETj ,Tj+1 the expectation with respect to
PTj ,Tj+1 yields
nj
2
X B(t, Tj+1 )
νj+1 (t) =
δj Dj
i=1
 
1
× IEPTj+1 1l Ft
B(Tj , Tj+1 ) {rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )}
nj
X B(t, Tj+1 ) B(0, Tj )
=
δj Dj B(0, Tj+1 )
i=1
 
F (Tj , Tj , Tj+1 )
× IEPTj+1 1l Ft
F (0, Tj , Tj+1 ) {rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )}
nj
X B(t, Tj+1 ) B(0, Tj ) B(0, Tj+1 )B(t, Tj )
=
δj Dj B(0, Tj+1 ) B(0, Tj )B(t, Tj+1 )
i=1
h i
× IETj ,Tj+1 1l{rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )} Ft
nj
X B(t, Tj ) h i
= IETj ,Tj+1 1l{rl (Tj,i )≤rn (Tj,i ,Tj,i +δj,i )≤ru (Tj,i )} Ft .
δj Dj
i=1
2.5 Valuation of floating range notes 45

The summands on the right hand side look (except for a multiplicative constant)
very similar to the time-t value of a range digital option, the only difference
being that the expectation is taken under the adjusted forward measure. We can
proceed in the same way as we did for digital options and use the independence
of the increments of L to obtain
nj
2 B(t, Tj ) X j,i
νj+1 (t) = Dt .
δj Dj
i=1

Here,
" #
(2.27)
Dtj,i = IETj ,Tj+1 1l 1 1
ff Ft
δj,i ru (Tj,i )+1
≤B(Tj,i ,Tj,i +δj,i )≤ δ r (T
j,i l j,i )+1
" #
(2.9)
= IETj ,Tj+1 1l B(t,Tj,i +δj,i ) j,i
ff Ft
K j,i ≤ B(t,Tj,i )
exp(X j,i )≤K
 
j,i B(t, Tj,i + δj,i )
= h (2.34)
B(t, Tj,i )

with hj,i : R+ → [0, 1] given by

Z
j,i j,i
h (y) = 1ln 1 K j,i ≤ex ≤ 1 K j,i o dPX
Tj ,Tj+1 (x), (2.35)
y y
R

where

ZTj,i
X j,i := Σ(s, Tj,i , Tj,i + δj,i ) dLs ,
t
ZTj,i
j,i 1 
K := exp θs (Σ(s, Tj,i + δj,i )) − θs (Σ(s, Tj,i )) ds,
δj,i rl (Tj,i ) + 1
t
ZTj,i
1  
K j,i := exp θs (Σ(s, Tj,i + δj,i )) − θs (Σ(s, Tj,i )) ds,
δj,i ru (Tj,i ) + 1
t

j,i
and PXTj ,Tj+1 denotes the distribution of X
j,i with respect to P
Tj ,Tj+1 .
To improve readability, let us simplify notation and fix j and i. In what
follows, we omit the sub– and superscripts j, i and write T , δ, Dt , h, X, K and
K for short. Denote by MTXj ,Tj+1 the moment generating function of the random
variable X with respect to PTj ,Tj+1 . Then we have the following pricing formula
for Dt , which immediately gives us a formula for the value of the floating range
note:
46 The Lévy term structure model

Theorem 2.19 Suppose the distribution of X possesses a Lebesgue-density.

Choose an R > 0 such that MTXj ,Tj+1 (−R) < ∞. Then

Z∞  R+iu !
1 B(t, T ) 1
Dt = < K MX (−R − iu) du
π B(t, T + δ) R + iu Tj ,Tj+1
0
Z∞  R+iu !
1 B(t, T ) 1
− < K MX (−R − iu) du
π B(t, T + δ) R + iu Tj ,Tj+1
0

with
ZT 
1 
K := exp θs (Σ(s, T + δ)) − θs (Σ(s, T )) ds, (2.36)
δrl (T ) + 1
t
ZT 
1 
K := exp θs (Σ(s, T + δ)) − θs (Σ(s, T )) ds. (2.37)
δru (T ) + 1
t

Proof: Observe that

Z Z
h(y) = 1lnex ≤ K o dPX
Tj ,Tj+1 (x) − 1lnex < K o dPX
Tj ,Tj+1 (x).
y y

Applying exactly the same arguments as in the proof of theorem 2.16 yields the
claim. The only difference is that we consider the moment generating function
of X with respect to an adjusted forward measure and not with respect to a
forward measure. 

The next theorem gives an expression for MTXj ,Tj+1 :

Theorem 2.20 Under the assumptions of theorem 2.19 we have for u ∈ R

ZT h i
MTXj ,Tj+1 (−R − iu) = exp θs (gs (−R − iu)) − θs (gs (0)) ds, (2.38)
t

with gs (z) := zΣ(s, T, T + δ) + Σ(s, Tj )1l{s≤Tj } + Σ(s, Tj+1 )1l{Tj <s} .

Proof: Equations (2.15)–(2.17) yield for z ∈ C with <(z) = −R

" ZT #
MTXj ,Tj+1 (z) = IETj ,Tj+1 z Σ(s, T, T + δ) dLs
t
TZj+1 !
= exp − ATj ,Tj+1 (s) ds
0
" ZT TZj+1 !#
× IE exp z Σ(s, T, T + δ) dLs + ΣTj ,Tj+1 (s) dLs .
t 0
2.5 Valuation of floating range notes 47

Using the independence of the increments of L plus the fact that for s1 < s2

" Zs2 # Zs2

IE exp ΣTj ,Tj+1 (s) dLs = exp θs (ΣTj ,Tj+1 (s)) ds
s1 s1
Zs2
= exp ATj ,Tj+1 (s) ds (2.39)
s1

(which is a consequence of proposition 1.9 and the drift condition (2.8)) yields

ZT !
MTXj ,Tj+1 (z) = exp − ATj ,Tj+1 (s) ds
t
" ZT #


× IE exp zΣ(s, T, T + δ) + ΣTj ,Tj+1 (s) dLs .
t

Making use of proposition 1.9 again and keeping in mind (2.39) as well as the
definition of ΣTj ,Tj+1 in (2.17), one arrives at equation (2.38). 

Once again, let us consider the special case of a multifactor Gaussian HJM
model. We have θ(x) = hx,xi d
2 for x ∈ C and from (2.38) we get for z ∈ C

ZT
z2

MTXj ,Tj+1 (z) = exp ||Σ(s, T, T + δ)||2 ds
2
t
ZT 
+z hΣ(s, T, T + δ), Σ(s, Tj )1l{s≤Tj } + Σ(s, Tj+1 )1l{Tj <s} i ds .
t

Consequently, X is normally distributed under PTj ,Tj+1 with mean

ZT
m(t, T, T +δ, Tj , Tj+1 ) := hΣ(s, T, T +δ), Σ(s, Tj )1l{s≤Tj } +Σ(s, Tj+1 )1l{Tj <s} i ds
t

and variance
ZT
g(t, T, T + δ) := ||Σ(s, T, T + δ)||2 ds. (2.40)
t
48 The Lévy term structure model

From (2.35) we get

K
log y
Z
h(y) = dPX
Tj ,Tj+1 (x)
K
log y
 
K K
= PTj ,Tj+1 log ≤ X ≤ log
y y
 
log K
y − m(t, T, T + δ, T j , T j+1 )
= Φ p 
g(t, T, T + δ)

log K
!
y − m(t, T, T + δ, Tj , Tj+1 )
−Φ p .
g(t, T, T + δ)

Plugging in the expression for K and K from (2.36)–(2.37) and using (2.34) we
arrive at
 
B(t,T ) 1
log B(t,T +δ)(δr l (T )+1)
+ 2 g(t, T, T + δ) − l(t, T, T + δ, T j , T j + 1)
Dt = Φ  p 
g(t, T, T + δ)
 
B(t,T ) 1
log B(t,T +δ)(δr u (T )+1)
+ 2 g(t, T, T + δ) − l(t, T, T + δ, T j , T j + 1)
−Φ  p ,
g(t, T, T + δ)

where

l(t, T, T + δ, Tj , Tj+1 ) := (2.41)

ZT
hΣ(s, T, T + δ), Σ(s, Tj )1l{s≤Tj } + Σ(s, Tj+1 )1l{Tj <s} − Σ(s, T )i ds.
t

Putting pieces together, we obtain the following result:

Theorem 2.21 Using the notation introduced above, the time-t price of a float-
ing range note in the multifactor Gaussian HJM model is equal to

N
X −1
F lRN (t) = B(t, TN ) + νj+1 (t)
j=0

with

rn (T0 , T0 + δ0 ) + s0
ν1 (t) = B(t, T1 )H(T0 , t)
D0
Xn0 
+ DRDt [rn (T0,i , T0,i + δ0,i ); rl (T0,i ); ru (T0,i ); T1 ]
i=n−
0 +1
2.6 An example of calibration to market data 49

and
 nj
X
sj 1
νj+1 (t) = − DRDt [rn (Tj,i , Tj,i + δj,i ); rl (Tj,i ); ru (Tj,i ); Tj+1 ]
Dj δj Dj
i=1
nj
B(t, Tj ) X  
+ Φ(ηj,i (rl (Tj,i ))) − Φ(ηj,i (ru (Tj,i )))
δj Dj
i=1

where
B(t,T ) 1
log B(t,Tj,i +δj,ij,i
)(δj,i r+1) + 2 g(t, Tj,i , Tj,i +δj,i )−l(t, Tj,i , Tj,i +δj,i , Tj , Tj+1 )
ηj,i (r) := p
g(t, Tj,i , Tj,i +δj,i )

and g and l are defined as in (2.40) and (2.41).

2.6 An example of calibration to market data

To test the ability of the Lévy term structure model to reproduce observed
option prices, we calibrate it to market data. Our data set consists of caplet
and swaption prices (quoted via their implied volatilities) of February 19, 2002
as well as of the zero coupon bond yield curve observed at the same day. The
calibration is done for a driving homogeneous as well as for a non-homogeneous
Lévy process.

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

2.0
0 5 10 15 20
Time to maturity (years)

Figure 2.1: Euro yield curve on February 19, 2002

Before being able to start a calibration, we first have to specify two ingre-
dients to the model, namely the volatility structure σ and the driving process
L. We choose the Vasiček volatility structure, that is σ(s, T ) = e−a(T −s) for a
real a. Note that we set σ
b = 1 in example 2.12 since this multiplicative constant
can be included in the process L and is therefore redundant in the volatility
structure. Two cases are considered for the driving process:
50 The Lévy term structure model

T 0.5 Y 1Y 1.5 Y 2Y 2.5 Y

B(0,T) 0.9833630 0.9647388 0.9435826 0.9228903 0.9006922
T 3Y 3.5 Y 4Y 4.5 Y 5Y
B(0,T) 0.8790279 0.8568412 0.8352144 0.8133497 0.7920573
T 5.5 Y 6Y 6.5 Y 7Y 7.5 Y
B(0,T) 0.7706813 0.7498822 0.7292111 0.7091098 0.6893282
T 8Y 8.5 Y 9Y 9.5 Y 10 Y
B(0,T) 0.6700983 0.6514969 0.6334118 0.6159873 0.5990420
T 10.5 Y 11 Y 11.5 Y 12 Y 12.5 Y
B(0,T) 0.5822022 0.5658357 0.5497905 0.5342002 0.5189684
T 13 Y 13.5 Y 14 Y 14.5 Y 15 Y
B(0,T) 0.5041710 0.4898851 0.4760039 0.4626807 0.4497303
T 15.5 Y 16 Y 16.5 Y 17 Y 17.5 Y
B(0,T) 0.4369867 0.4246043 0.4126901 0.4011102 0.3897380
T 18 Y 18.5 Y 19 Y 19.5 Y 20 Y
B(0,T) 0.3786882 0.3678357 0.3572942 0.3472375 0.3374637

Table 2.1: Euro zero coupon bond prices on February 19, 2002

• First, the model is driven by a (homogeneous) Lévy process. Concretely,

we model increments of length 1 to be NIG-distributed (see e.g. Barndorff-
Nielsen (1998)) with parameters µ, α, β, δ. It can be shown that para-
meter µ does not have an impact on option prices. Thus, µ can be set
equal to zero. This leads to a model with a total number of 4 parameters.

• For the second calibration a non-homogeneous Lévy process is used. Mo-

tivated by the observation that traders act differently on the short term
bond market (maturities up to one year), the middle (one to five years),
and the long term market (greater than five years), we use a “piecewise
Lévy process” as driving motion, i.e. a non-homogeneous Lévy process
whose increments on [0, 1] are stationary as well as the increments on
[1, 5] and [5, T ∗ ]. Again, increments of length 1 are modelled to be NIG-
distributed. We end up with a model consisting of a total number of 10
parameters, one for the volatility structure and 3 × 3 parameters for the
process L.

Note that in both cases the martingale measure is unique and the short rate
follows a Markov process. Of course, these particular choices are just two out
of many possibilities in our modelling framework.
Let us consider the calibration to prices of caplets first. We use market
prices for caplets with maturities ranging from one to ten years and 12 different
strike rates from 2.5 % to 10 %. They are given via their implied volatilities (in
%) as shown in figure 2.2 and table 2.2. Note that the shape of the implied
volatility surface is quite typical, i.e. for the short maturities there is a smile in
the implied volatilities whereas for the long maturities a skewed shape can be
observed. All caplets are linked to the 6-month EURIBOR. Let us stress that
all except for the one year caplet are in fact portfolios of two caplets. The two
2.6 An example of calibration to market data 51

30.0

28.0

26.0

24.0

22.0

20.0

18.0

16.0

14.0

12.0

10.0 0
2.5 2
4.0 4
6.0 6
8.0 8
10.0 10 Maturity (in years)
Strike rate (in %)

Figure 2.2: Euro caplet implied volatility surface on February 19, 2002

2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 7.0 8.0 9.0 10.0
1Y 27.2 23.6 20.1 19.9 20.6 21.0 21.8 21.7 22.8 22.9 22.0 24.3
2Y 27.4 24.3 21.3 18.6 18.3 18.1 18.6 19.2 20.0 21.7 23.5 26.4
3Y 26.9 22.9 20.0 18.7 16.1 15.6 15.5 15.7 17.0 18.9 21.3 23.6
4Y 26.3 22.0 19.4 17.4 15.6 14.7 14.6 14.3 14.7 16.4 17.0 18.4
5Y 25.4 21.4 19.8 16.8 15.6 14.5 13.9 13.4 13.5 12.8 14.8 15.8
6Y 25.2 21.7 19.6 17.5 15.9 14.2 13.2 13.3 13.1 13.8 14.4 15.4
7Y 23.6 20.9 18.4 16.2 15.2 14.1 13.2 12.2 12.1 12.2 13.1 13.8
8Y 23.5 20.4 18.5 16.3 14.8 13.7 13.1 12.3 12.3 13.5 13.5 13.6
9Y 22.9 21.0 17.5 16.6 15.1 13.3 12.1 12.2 12.2 12.9 12.7 13.9
10 Y 22.2 19.0 17.7 15.7 14.1 13.0 12.2 11.8 11.8 12.5 13.4 13.8

Table 2.2: Euro caplet implied volatilities on February 19, 2002

year “caplet” for example consists of a caplet with option maturity date in one
year as well as of a caplet that matures in one and a half years.
The calibration is done by minimizing the sum of the squared errors between
theoretical and market price relative to the at-the-money caplet market price
for the respective maturity. That is, we minimize the sum of
 2
model price - market price
.
ATM market price for the respective maturity

The differences (resp. absolute differences) between implied volatility of

the model price and implied volatility of the market price are shown in ta-
ble 2.3 (figure 2.3) for the homogeneous and in table 2.4 (figure 2.4) for the
non-homogeneous Lévy process. To improve the readability of figure 2.3 and
to make the two pictures comparable, we truncated differences in the implied
52 The Lévy term structure model

volatilities that exceed 10 %. In the tables, the two strike rates that are closest
to the at-the-money strike rate are highlighted in red.

10.0

9.0

8.0

7.0

6.0

5.0

4.0

3.0

2.0

1.0

0
3.0 1
4.0
5.0 5
6.0
8.0
10.0 10
Strike rate (in %) Maturity (years)

Figure 2.3: Absolute errors in caplet calibration for the driving homogeneous
Lévy process

2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 7.0 8.0 9.0 10.0
1Y 10.4 4.4 -0.5 -3.0 1.3 5.3 8.1 11.1 14.5 17.8 21.2 20.9
2Y 4.4 2.0 0.4 -0.7 -2.1 -1.1 0.1 1.0 2.7 2.8 2.4 0.5
3Y 1.3 1.2 0.8 -0.5 0.2 -0.2 -0.1 0.3 0.4 -0.3 -1.8 -3.4
4Y -0.4 0.6 0.5 0.4 0.5 0.4 0.0 0.3 0.5 -0.5 -0.5 -1.4
5Y -1.0 0.1 -0.6 0.5 0.3 0.3 0.3 0.5 0.5 1.5 0.0 -0.7
6Y -2.1 -1.1 -1.1 -0.7 -0.4 0.3 0.6 0.1 0.0 -0.5 -0.9 -1.7
7Y -1.4 -1.2 -0.6 0.1 0.0 0.1 0.3 0.8 0.5 0.3 -0.5 -1.1
8Y -2.2 -1.3 -1.1 -0.3 0.0 0.2 0.1 0.5 -0.1 -1.5 -1.5 -1.5
9Y -2.2 -2.4 -0.5 -1.0 -0.5 0.4 1.0 0.3 -0.3 -1.1 -1.1 -2.3
10 Y -2.0 -0.7 -1.1 -0.3 0.2 0.6 0.7 0.6 0.0 -1.0 -2.1 -2.6

Table 2.3: Errors in caplet calibration for the driving homogeneous Lévy
process

The model parameters that lead to these results are

a = 0.0504489

and
α = 48.9992, β = −5.47554, δ = 0.00417802
for the driving homogeneous Lévy process.
2.6 An example of calibration to market data 53

10.0

9.0

8.0

7.0

6.0

5.0

4.0

3.0

2.0

1.0

0
3.0 1
4.0
5.0 5
6.0
8.0
10.0 10
Strike rate (in %) Maturity (in years)

Figure 2.4: Absolute errors in caplet calibration for the driving

non-homogeneous Lévy process

2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 7.0 8.0 9.0 10.0
1Y 2.2 0.8 0.9 0.4 1.0 2.3 3.0 4.4 5.3 6.6 8.5 6.9
2Y 0.1 -1.0 -1.0 -0.2 -0.7 -0.6 -0.8 -0.9 -0.3 -0.4 -0.7 -2.2
3Y 0.8 0.6 0.3 -0.7 0.3 0.0 -0.1 -0.1 -0.4 -1.0 -2.3 -3.7
4Y -0.1 0.6 0.4 0.3 0.4 0.2 -0.2 0.0 0.1 -0.9 -0.7 -1.5
5Y -0.3 0.4 -0.5 0.4 0.0 0.0 -0.1 0.2 0.3 1.5 0.2 -0.1
6Y -0.5 -0.2 -0.6 -0.6 -0.6 -0.1 0.2 -0.3 0.1 0.0 0.2 0.0
7Y 0.6 0.2 0.2 0.4 -0.1 -0.3 -0.1 0.4 0.5 1.0 0.8 0.8
8Y 0.1 0.3 -0.2 0.0 0.0 -0.2 -0.4 0.1 -0.1 -0.7 0.0 0.5
9Y 0.2 -0.8 0.4 -0.6 -0.6 0.0 0.5 -0.1 -0.2 -0.4 0.3 -0.3
10 Y 0.4 0.8 -0.2 0.0 0.1 0.2 0.2 0.1 0.0 -0.2 -0.7 -0.6

Table 2.4: Errors in caplet calibration for the driving non-homogeneous Lévy
process

In case of the piecewise Lévy process we get

a = 0.0864322
and
α = 179.818, β = −34.6837, δ = 0.01282933 on [0, 1],
α = 37.531, β = − 2.1500, δ = 0.00386291 on [1, 5],
α = 14.132, β = − 3.0837, δ = 0.00220217 on [5, 10].

From these results we can draw some conclusions:

54 The Lévy term structure model

1. Except for the short maturities (i.e. the one and two year caplets), the
model driven by a homogeneous Lévy process produces very good results.
In a large interval around the at-the-money strike rates differences in the
implied volatilities are well within a 1 % range. Only for far out-of-the-
money options errors exceed the 1 % range. Remember that the model is
driven by four parameters only.

2. The model driven by a piecewise Lévy process produces excellent results

across all maturities and strike rates. For the one year caplet this may not
be obvious from the picture since we observe large differences in implied
volatilities away-from-the-money. Nevertheless, the respective differences
between model and market price are very small. For the one year caplet
with a strike rate of 9 % for example, we notice the largest difference
(8.5 %) in implied volatilities, but model and market price differ only by
6 × 10−8 (less than 1/1000 of a basis point).

3. It is not surprising that a model driven by 10 parameters outperforms a

model that depends on 4 parameters only. However, not only the smaller
number of parameters makes the model driven by a homogeneous Lévy
process inferior to its non-homogeneous counterpart. It is well known that
exponential Lévy models for stock prices allow for an excellent calibration
to implied volatility patterns for single maturities and also for a certain
range of maturities, but fail to reproduce option prices with the same
accuracy over the full range of different maturities because the smile in
the implied volatilities flattens as time to maturity increases. In other
words, a model that produces the correct smile for short maturities will
usually not give rise to enough smile for the long maturities. Conversely,
a model that gets the long maturities right will produce a smile that
is too strong in the short end. The same observation can be made in
the Lévy term structure model as figure 2.5 shows. The model driven by
a homogeneous Lévy process produces an accurate smile for the 5 year
caplet but too much of a smile in the short end and too few for the 10 year
caplet. Note that the model driven by a non-homogeneous Lévy process
is flexible enough to reproduce the smiles observed in the market across
all maturities with high accuracy.

4. The calibration results for the piecewise Lévy model can be improved by
increasing the number of parameters in the model. For example, when a
driving process is used whose increments are stationary and NIG-distri-
buted on [0, 1], [1, 2], . . . , [9, T ∗ ] (that makes 31 parameters in total) the
average error (model minus market price divided by ATM market price)
reduces by 24 %. We leave the question to the reader whether or not this
higher calibration accuracy justifies the larger number of parameters.

The second model calibration is done to market prices of swaptions. For this
purpose, we use prices of swaptions with maturities of one, two, three, four, five,
seven, and ten years. The tenors of the underlying swaps reach from one to ten
years. The implied volatilities of these market prices are shown in table 2.5.
2.6 An example of calibration to market data 55

Unfortunately, our data set consists of at-the-money options only. Therefore,

we will not be able to discuss any effects on the smile and we are not going to
give any pictures.

1Y 2Y 3Y 4Y 5Y 6Y 7Y 8Y 9Y 10 Y
1Y Opt 18.0 16.6 15.4 14.4 13.7 13.3 12.8 12.5 12.2 11.8
2Y Opt 15.1 14.3 13.4 13.0 12.6 12.3 12.0 11.7 11.5 11.2
3Y Opt 14.5 13.3 12.7 12.3 12.0 11.8 11.5 11.3 11.1 10.9
4Y Opt 13.7 12.5 12.1 11.8 11.6 11.4 11.2 10.9 10.7 10.5
5Y Opt 13.0 12.0 11.6 11.4 11.3 11.1 10.9 10.7 10.5 10.3
7Y Opt 12.3 11.2 10.9 10.7 10.6 10.4 10.3 10.1 10.0 9.9
10 Y Opt 11.4 10.3 10.0 9.8 9.6 9.5 9.5 9.4 9.3 9.3

Table 2.5: Euro at-the-money swaptions: Implied volatilities on 19 February

2002 (in %)

The calibration is done by minimizing the sum of the squared relative errors,
that is the sum of the squared differences between theoretical and market price
relative to the market price. The differences in the implied volatilities of model
and market prices are shown in table 2.6 for the homogeneous and in table 2.7
for the non-homogeneous Lévy process. Both models reproduce market prices
with high accuracy.

1Y 2Y 3Y 4Y 5Y 6Y 7Y 8Y 9Y 10 Y
1Y Opt -1.02 -0.78 -0.42 -0.11 -0.01 -0.13 -0.08 -0.15 -0.16 -0.07
2Y Opt 0.05 0.14 0.43 0.28 0.21 0.09 0.05 0.06 -0.03 0.00
3Y Opt -0.38 0.24 0.32 0.26 0.16 0.04 0.07 -0.01 -0.08 -0.13
4Y Opt -0.39 0.31 0.27 0.19 0.08 0.02 -0.06 -0.02 -0.06 -0.08
5Y Opt -0.36 0.21 0.25 0.16 0.01 -0.06 -0.12 -0.17 -0.18 -0.18
7Y Opt -0.59 0.27 0.35 0.26 0.09 0.03 -0.10 -0.09 -0.19 -0.27
10 Y Opt -0.48 0.32 0.35 0.32 0.33 0.21 0.03 -0.06 -0.15 -0.31

Table 2.6: Error of swaption calibration (in %) for the homogeneous Lévy
process

1Y 2Y 3Y 4Y 5Y 6Y 7Y 8Y 9Y 10 Y
1Y Opt -0.14 -0.17 0.00 0.18 0.19 0.00 0.01 -0.08 -0.11 -0.03
2Y Opt 0.00 0.05 0.31 0.15 0.08 -0.03 -0.05 -0.02 -0.09 -0.04
3Y Opt -0.54 0.08 0.18 0.14 0.07 -0.02 0.03 -0.02 -0.05 -0.07
4Y Opt -0.52 0.20 0.19 0.14 0.07 0.04 0.00 0.08 0.07 0.08
5Y Opt -0.52 0.09 0.15 0.09 -0.02 -0.06 -0.09 -0.11 -0.10 -0.06
7Y Opt -0.73 0.14 0.25 0.19 0.04 0.01 -0.09 -0.06 -0.13 -0.18
10 Y Opt -0.60 0.22 0.27 0.26 0.29 0.20 0.04 -0.03 -0.10 -0.24

Table 2.7: Error of swaption calibration (in %) for the non-homogeneous Lévy
process
56 The Lévy term structure model

The model parameters that lead to these results are

a = 0.0486698

and
α = 2730.651, β = −230.663, δ = 0.161142
for the (homogeneous) Lévy process. In case of the piecewise Lévy process we
get
a = 0.0413190
and

α = 12.0637, β = − 6.9660, δ = 0.0024161 on [0, 1],

α = 41.4422, β = − 0.1376, δ = 0.0025918 on [1, 5],
α = 64.9670, β = −28.4413, δ = 0.0027980 on [5, 20].

2.7 Conclusion
As a generalization of the Lévy term structure model introduced in Eberlein and
Raible (1999) we discussed a term structure model driven by non-homogeneous
Lévy processes. For deterministic volatility structures, pricing formulae have
been derived for caps and floors as well as for digital options and for a deriva-
tive with a path-dependent payoff, namely a floating range note. As a side
result, a valuation formula derived by Nunes (2004) for floating range notes
in the multifactor Gaussian HJM model has been simplified. A formula for
swaption valuation has also been established under an additional restriction
on the volatility structure which still allows for some well known examples as
the Ho–Lee or Vasiček volatility function. An advantage of all of these pricing
formulae is the speed at which they can be evaluated numerically. This gave
us the opportunity to calibrate the model to market data of the most liquid
interest rate derivatives, namely caps, floors, and swaptions. Calibrations were
done for a driving homogeneous Lévy process as well as for a driving process
with independent and piecewise stationary increments. This led to models with
4 and 10 parameters respectively. Both models proved to be flexible enough to
reproduce the given derivatives’ prices with high accuracy, although the lat-
ter clearly outperformed the other (which is not too surprising since it has
more parameters). The model driven by a homogeneous Lévy process revealed
a weakness which also occurs when modelling stock prices with Lévy processes,
namely that smiles in the implied volatilities of option prices flatten too much
as option maturity increases. This drawback can be removed by working with
non-homogeneous Lévy processes.
2.7 Conclusion 57

32
Market price
Model price (homogeneous)
30 Model price (non−homogeneous)

28
Implied volatility (in %)
26

24

22

20

18

16
3 4 5 6 7 8 9 10
Strike rate (in %)

26
Market price
Model price (homogeneous)
24 Model price (non−homogeneous)
Implied volatility (in %)

22

20

18

16

14

12
3 4 5 6 7 8 9 10
Strike rate (in %)

24
Market price
Model price (homogeneous)
22 Model price (non−homogeneous)
Implied volatility (in %)

20

18

16

14

12

10
3 4 5 6 7 8 9 10
Strike rate (in %)

Figure 2.5: Implied volatility curves for the

2 year, 5 year, and 10 year caplet
Chapter 3

Lévy models for effective rates

Short rate models as well as term structure (Heath–Jarrow–Morton) models,

like the Lévy term structure model discussed in the previous chapter, specify
the dynamics of continuously compounded interest rates. From a mathematical
point of view, these rates that apply for an infinitesimal time interval are very
convenient for modelling purposes. However, interest rates quoted in real mar-
kets are usually effective (simply compounded). Sandmann, Sondermann, and
Miltersen (1995), Miltersen, Sandmann, and Sondermann (1997), and Brace,
Gatarek, and Musiela (1997) managed to incorporate a model for effective
rates into an HJM-framework. Their Forward Libor Model which is often re-
ferred to as the Libor market model became a very popular approach among
practitioners since it is consistent with the market practice of pricing caps and
floors. Jamshidian (1999) generalized this model by considering semimartin-
gales as driving processes, but pricing of caps and floors was not discussed in
this setup. A model that lies in between Jamshidians’s approach and the Libor
market model as far as generality is concerned has recently been developed by
Eberlein and Özkan (2005). Their Lévy Libor model is a lot more flexible than
the Libor market model since it uses general (time-inhomogeneous) Lévy pro-
cesses as drivers instead of the special case of a Brownian motion. Moreover,
explicit pricing formulae for caps and floors can be obtained.
Eberlein and Özkan (2005) take two different approaches to model forward
Libor rates in a discrete tenor setting. Both approaches have in common that
they do not specify zero coupon bond prices directly (it is only assumed that
the processes describing the evolution of bond prices are special semimartingales
whose values as well as all left hand limits are strictly positive; moreover, the
terminal value of each bond equals one). Instead, ratios of bond prices are
specified. Eberlein and Özkan consider a fixed time horizon T ∗ as well as a
discrete tenor structure 0 = T0 < T1 < . . . < Tn = T ∗ and build up the model
in one of the two following ways:

• The first approach uses the (ordinary) exponential of a non-homogeneous

Lévy process to model forward Libor rates directly, which are defined by
 
1 B(t, Tk )
L(t, Tk ) := −1 (k ∈ {1, . . . , n − 1}),
δk B(t, Tk+1 )
60 Lévy models for effective rates

where δk := Tk+1 − Tk . This specification generalizes the Libor market

model and we will refer to it as the Lévy Libor model.

• In the second approach forward price processes, i.e.

B(t, Tk )
F (t, Tk , Tk+1 ) := (k ∈ {1, . . . , n − 1}),
B(t, Tk+1 )

are specified as starting value times the exponential of a non-homogeneous

Lévy process. Of course, this immediately provides a model for forward
Libor rates since L(t, Tk ) = δ1k (F (t, Tk , Tk+1 ) − 1). In order to be able
to distinguish this ansatz from the other one we call it the Lévy forward
price model.

Although L(t, Tk ) and F (t, Tk , Tk+1 ) differ only by an additive and a multiplica-
tive constant, the two specifications lead to models that are quite different. If in
a small time interval from t1 to t2 the driving process in the exponent changes
its value by a small amount ∆, then in the Lévy Libor model we have

L(t2 , Tk ) = L(t1 , Tk ) exp ∆ ≈ L(t1 , Tk ) + ∆L(t1 , Tk )

whereas in the Lévy forward price model

1 ∆
L(t2 , Tk ) = ((1 + δk L(t1 , Tk )) exp ∆ − 1) ≈ L(t1 , Tk ) +
δk δk

for reasonable values of L(t1 , Tk ). In other words, (small) changes in the driving
process have different impact on the forward Libor rates. In the Lévy Libor
model, forward Libor rates change by an amount that is relative to their level
while the change in the Lévy forward price model does not depend on the actual
level.
The two models also differ in tractability. The Lévy forward price model is
very pleasant from an analytical point of view. The driving process remains a
non-homogeneous Lévy process under all forward measures, a fact that simpli-
fies the valuation of derivatives considerably. It might be seen as a drawback
that this model allows for negative Libor rates. In the Lévy Libor model, Libor
rates are always positive. The driving process is usually only a Lévy process with
respect to one forward measure. This makes option pricing more complicated
and forces us to work with approximations.
The aim of this chapter is to push further the derivation of option prices
within the Lévy Libor model as well as within the Lévy forward price model.
Our focus lies on the most common interest rate derivatives, i.e. caps, floors,
and swaptions.
In the Lévy forward price model, exact pricing formulae can be obtained. We
prove that this model can be regarded as a special case of the Lévy term struc-
ture model. Consequently, the pricing formulae for caps, floors, and swaptions
from chapter 2 can be used.
Eberlein and Özkan (2005) provide an approximate pricing formula for caps
and floors in the Lévy Libor model. Valuation of swaptions is not considered.
 3.1 The Lévy forward price model 61

Our goal is twofold: first, we derive an alternative valuation formula for caps
and floors based on a different approximation. The advantage of our formula is
the much higher speed at which cap and floor prices can be computed. Second,
we present a method to price swaptions.
The chapter is organized as follows: Section 3.1 reviews the construction of
the Lévy forward price model which is done only very briefly in Eberlein and
Özkan (2005). In section 3.1.1 we prove that this model can be embedded into
the Lévy term structure model. A brief presentation of the Lévy Libor model is
given in section 3.2. The remaining sections are devoted to the pricing of caps,
floors, and swaptions within this model.

3.1 The Lévy forward price model

The Lévy forward price model is constructed by backward induction. It is
∗
driven by a non-homogeneous Lévy process LT on a complete stochastic basis
(Ω, FT ∗ , F = (Fs )0≤s≤T ∗ , PT ∗ ). The measure PT ∗ plays the role of the forward
∗
measure associated with the settlement day T ∗ . The process LT is supposed
to have exponential moments in the sense of assumption (EM) from chapter
∗ ∗ ∗
1. Two of the characteristics (bT , c, F T ) of LT can be chosen freely, namely
∗ ∗
c and F T , whereas the drift characteristic bT will be derived later. Since we
proceed by backward induction, let us denote Ti∗ := Tn−i and δi∗ := δn−i for
i ∈ {0, . . . , n}. The following assumptions are made:

(FP.1): For any maturity Ti there is a bounded deterministic volatility function

λ(·, Ti ) : [0, T ∗ ] → Rd which represents the volatility of the forward price
process F (·, Ti , Ti+1 ). In particular, for all k ∈ {1, . . . , n − 1}

k
X
λj (s, Ti ) ≤ M (s ∈ [0, T ∗ ], j ∈ {1, . . . , d}), (3.1)
i=1

where M is the constant from assumption (EM) and λ(s, Ti ) = 0 for

s > Ti .

(FP.2): The initial term structure of zero coupon bond prices B(0, Ti ) is strictly
positive (i ∈ {1, · · · , n}).

We begin by constructing the forward price with the longest maturity and
postulate that
 t 
Z
∗
F (t, T1∗ , T ∗ ) = F (0, T1∗ , T ∗ ) exp  λ(s, T1∗ ) dLTs  (3.2)
0

subject to the initial condition

B(0, T1∗ )
F (0, T1∗ , T ∗ ) = .
B(0, T ∗ )
62 Lévy models for effective rates

Another way to write (3.2) is in terms of the forward Libor rate

Zt
 
T∗
1 + δ1∗ L(t, T1∗ ) = (1 + δ1∗ L(0, T1∗ )) exp  λ(s, T1∗ ) dLs  .
0

∗
Our goal is to specify the drift characteristic bT in such a way that the forward
price process F (·, T1∗ , T ∗ ) (or equivalently the forward Libor rate L(·, T1∗ )) is a
∗
martingale with respect to PT ∗ . For this purpose, we choose bT such that

Zt Zt
∗ 1
hλ(s, T1∗ ), bTs i ds =− hλ(s, T1∗ ), cs λ(s, T1∗ )i ds
2
0 0
Zt Z   ∗
∗
− ehλ(s,T1 ),xi − 1 − hλ(s, T1∗ ), xi ν T (ds, dx),
0 Rd

∗ ∗
where ν T (ds, dx) := FsT (dx) ds is the compensator of the random measure µL
∗
that is associated with the jumps of LT . Lemma 2.6 in Kallsen and Shiryaev
(2002) yields that the forward price F (·, T1∗ , T ∗ ) can then be expressed as the
stochastic exponential of a local martingale, namely

F (t, T1∗ , T ∗ ) = F (0, T1∗ , T ∗ )Et (H(·, T1∗ ))

with
Zt
√ ∗
H(t, T1∗ ) = cs λ(s, T1∗ ) dWsT
0
Zt Z  
∗ ∗
+ ehλ(s,T1 ),xi − 1 (µL − ν T )(ds, dx). (3.3)
0 Rd

Note that H(·, T1∗ ) is also a non-homogeneous Lévy process. The stochastic
exponential of a process that is a local martingale as well as a non-homogeneous
Lévy process is not only a local martingale, but in fact a martingale (see e.g.
Eberlein, Jacod, and Raible (2005) for a proof). Hence, F (·, T1∗ , T ∗ ) and thus
also L(·, T1∗ ) are martingales.
We define the forward martingale measure associated with the date T1∗ by
setting
dPT1∗ F (T1∗ , T1∗ , T ∗ )
:= = ET1∗ (H(·, T1∗ )).
dPT ∗ F (0, T1∗ , T ∗ )
From equation (3.3) we can immediately identify the two predictable processes
β and Y in Girsanov’s Theorem for semimartingales (see Jacod and Shiryaev
(2003, Theorem III.3.24)) that describe the change of measure, namely

β(s) = λ(s, T1∗ ) and Y (s, x) = exp hλ(s, T1∗ ), xi.

3.1 The Lévy forward price model 63

T∗ ∗ Rt√
In particular, Wt 1 := WtT − 0 cs λ(s, T1∗ ) ds is a standard Brownian motion
∗ ∗
with respect to PT1∗ and ν T1 (dt, dx) := exphλ(s, T1∗ ), xiν T (dt, dx) is the PT1∗ -
∗
compensator of µL . We have the following PT1∗ -canonical representation of LT :

Zt Zt Zt Z
∗ √ T∗ ∗
LTt = bbs ds + cs dWs 1 + x(µL − ν T1 )(ds, dx)
0 0 0 Rd

with a deterministic drift coefficient bb which can be calculated using Girsanov’s

Theorem.
Now we are ready to construct the forward price F (·, T2∗ , T1∗ ) by postulating
that  t 
Z
T ∗
F (t, T2∗ , T1∗ ) = F (0, T2∗ , T1∗ ) exp  λ(s, T2∗ ) dLs 1  ,
0
where
Zt Zt Zt Z
T∗ T∗ √ T∗ ∗
Lt 1 = bs 1 ds + cs dWs 1 + x(µL − ν T1 )(ds, dx).
0 0 0 Rd

In order to ensure that F (·, T2∗ , T1∗ ) is a PT1∗ -martingale, we choose the drift
∗
characteristic bT1 appropriately, namely such that
Zt Zt
T∗ 1
hλ(s, T2∗ ), bs 1 i ds =− hλ(s, T2∗ ), cs λ(s, T2∗ )i ds
2
0 0
Zt Z   ∗
∗
− ehλ(s,T2 ),xi − 1 − hλ(s, T2∗ ), xi ν T1 (ds, dx).
0 Rd
∗ ∗
Note that LT1 differs from LT only by a deterministic drift term. In partic-
ular, both processes are non-homogeneous Lévy processes under PT ∗ and PT1∗ .
Again, we can express the forward price process F (·, T2∗ , T1∗ ) as the stochastic
exponential of a non-homogeneous Lévy process and local martingale H(·, T2∗ )
F (t,T2∗ ,T1∗ )
 
and use the martingale F (0,T2∗ ,T1∗ ) 0≤t≤T ∗ to define the forward martingale
2
measure associated with the date T2∗ by setting
dPT2∗ F (T2∗ , T2∗ , T1∗ )
:= .
dPT1∗ F (0, T1∗ , T1∗ )

Proceeding as before, forward prices F (·, Ti∗ , Ti−1 ∗ ) for i = 3, . . . , n − 1 and

forward measures PTi∗ for i = 3, . . . , n − 2 are defined inductively. We obtain a

model where the forward price F (·, Ti∗ , Ti−1 ∗ ) is given by

 t 
Z ∗
T
F (t, Ti∗ , Ti−1
∗
) = F (0, Ti∗ , Ti−1
∗
) exp  λ(s, Ti∗ ) dLs i−1  (3.4)
0
64 Lévy models for effective rates

with

Zt Zt Zt Z
T∗ T∗ √ T∗ ∗
Lt i−1 = bs i−1 ds + cs dWs i−1 + x(µL − ν Ti−1 )(ds, dx). (3.5)
0 0 0 Rd

∗ ∗
W Ti−1 is a PTi−1 T
∗ -standard Brownian motion and ν i−1 is the PT ∗ -compensator
i−1
of µL . It is given by

i−1
X 
∗ ∗
ν Ti−1
(dt, dx) = exp hλ(t, Tj∗ ), xi FtT (dx) dt. (3.6)
j=1

∗
The characteristic bTi−1 satisfies

Zt Zt
T∗ 1
hλ(s, Ti∗ ), bs i−1 i ds =− hλ(s, Ti∗ ), cs λ(s, Ti∗ )i ds (3.7)
2
0 0
Zt Z   ∗
∗
− ehλ(s,Ti ),xi − 1 − hλ(s, Ti∗ ), xi ν Ti−1 (ds, dx).
0 Rd

∗
Observe that the driving processes LTi differ only by deterministic drift terms.
Hence, all of them are non-homogeneous Lévy processes with respect to each
forward measure.

3.1.1 The Lévy forward price model as a special case of the

Lévy term structure model
The aim of this section is to show that the Lévy forward price model can be
seen as a special case of the Lévy term structure model. By a special case we
mean that the model parameters in the term structure model can be chosen in
such a way that we end up with forward price processes as given in (3.4)–(3.7).
Let us briefly recall those properties of the Lévy term structure model that will
be needed to embed the forward price model.
In the Lévy term structure model, the price B(·, T ) of a zero coupon bond
with maturity T is given by

 Zt  Zt 
B(0, T ) 
B(t, T ) = exp θs (Σ(s, t)) − θs (Σ(s, T )) ds + Σ(s, t, T ) dLs ,
e e e
B(0, t)
0 0
(3.8)
where
Σ(s, t, T ) := Σ(s, T ) − Σ(s, t)

and Le is a non-homogeneous Lévy process with characteristics (eb, e

c, Fe) under
the spot martingale measure P. L satisfies assumption (EM) and θs denotes the
e e
3.1 The Lévy forward price model 65

cumulant associated with the infinitely divisible distribution characterized by

the Lévy–Khintchine-triplet (ebs , e
cs , Fes ). The volatility structure is given by

ZT
Σ(s, T ) := σ(s, u) du
s∧T

for a measurable, bounded and deterministic function σ. The forward martin-

gale measure PTi∗ for the settlement day Ti∗ is related to P via the Radon-
Nikodym derivative

∗ ∗
ZTi ZTi
dPTi∗
 
= exp − θes (Σ(s, Ti∗ )) ds + Σ(s, Ti∗ ) dL
es .
dP
0 0

e is also a non-homogeneous Lévy process with respect to PT ∗ and its PT ∗ -

L i i
∗ ∗ ∗
characteristics (ebTi , e
cTi , FeTi ) are given by

Z 
∗ 
ebTs i = ebs + e
cs Σ(s, Ti∗ ) +
∗
ehΣ(s,Ti ),xi − 1 x Fes (dx), (3.9)
T∗ Rd
cs i = e
e cs ,
T∗ ∗
Fes i (dx) = ehΣ(s,Ti ),xi Fes (dx). (3.10)

e is also a PT ∗ -special semimartingale, it can be written in its PT ∗ -

Since L i i
canonical representation as

Zt Zt p Zt Z
∗ Ti∗
L
et = ebTs i ds + cs dWs +
e
e ∗
x(µL − νeTi )(ds, dx),
0 0 0 Rd

∗ ∗ T ∗
where W Ti is a PTi∗ -standard Brownian motion and νeTi (ds, dx) := Fes i (dx) ds
is the PTi∗ -compensator of µL , the random measure associated with the jumps
e

of the process L.
e
66 Lévy models for effective rates

From (3.8) we can deduce the forward price process

∗
F (t, Ti+1 , Ti∗ )
∗ )
B(t, Ti+1
=
B(t, Ti∗ )
∗ )  Zt 
B(0, Ti+1 
= exp θes (Σ(s, Ti∗ )) − θes (Σ(s, Ti+1
∗
)) ds
B(0, Ti∗ )
0
Zt 
+ Σ(s, Ti∗ , Ti+1
∗
) dL
es
0
 Zt p
∗ T∗
= F (0, Ti+1 , Ti∗ ) exp It1 + It2 + ecs Σ(s, Ti∗ , Ti+1
∗
) dWs i
0
Zt Z   
∗
+ hΣ(s, Ti∗ , Ti+1
∗
), xi µL − νeTi (ds, dx) ,
e

0 Rd

with

Zt  
It1 := θes (Σ(s, Ti∗ )) − θes (Σ(s, Ti+1
∗
)) ds
0
Zt 
= − hΣ(s, Ti∗ , Ti+1
∗
), ebs i
0
1 1
+ hΣ(s, Ti∗ ), e cs Σ(s, Ti∗ )i − hΣ(s, Ti+1
∗
), e ∗
cs Σ(s, Ti+1 )i
2Z 2 
 ∗ ),xi

hΣ(s,Ti∗ ),xi hΣ(s,Ti+1 ∗ ∗
+ e −e + hΣ(s, Ti , Ti+1 ), xi Fs (dx) ds
e
Rd

and

Zt
T∗
It2 := hΣ(s, Ti∗ , Ti+1
∗
), ebs i i ds
0
Zt 
(3.9)
= hΣ(s, Ti∗ , Ti+1
∗
), ebs i + hΣ(s, Ti∗ , Ti+1
∗
cs Σ(s, Ti∗ )i
), e
0
Z   
hΣ(s,Ti∗ ),xi
+ hΣ(s, Ti∗ , Ti+1
∗
), xi e − 1 Fs (dx) ds.
e
Rd
3.1 The Lévy forward price model 67

Summing up I 1 and I 2 yields

Zt
1
It1 + It2 =− hΣ(s, Ti∗ , Ti+1
∗
cs Σ(s, Ti∗ , Ti+1
), e ∗
)i ds
2
0
Zt Z 
T∗
∗ ∗

− ehΣ(s,Ti ,Ti+1 ),xi − 1 − hΣ(s, Ti∗ , Ti+1
∗
), xi Fes i (dx) ds.
0 Rd

Hence, the forward price process in the Lévy term structure model is given by
∗
F (t, Ti+1 , Ti∗ )
 Zt
∗ 1
= F (0, Ti+1 , Ti∗ ) × exp − hΣ(s, Ti∗ , Ti+1
∗
cs Σ(s, Ti∗ , Ti+1
), e ∗
)i ds
2
0
Zt Z 
T∗
∗ ∗

− ehΣ(s,Ti ,Ti+1 ),xi − 1 − hΣ(s, Ti∗ , Ti+1
∗
), xi Fes i (dx) ds
0 Rd
Zt p T∗
+ ecs Σ(s, Ti∗ , Ti+1
∗
) dWs i
0
Zt Z   
Ti∗
+ hΣ(s, Ti∗ , Ti+1
∗
), xi L
µ − νe (ds, dx) .
e

0 Rd

The next step is to specify the model parameters, that is the volatility
structure σ and the characteristics (eb, e
c, Fe) of L,
e in such a way that these forward
price dynamics match the dynamics given in (3.4)-(3.7). First, we choose the
volatility function such that

Σ(s, Ti∗ , Ti+1

∗ ∗
) = λ(s, Ti+1 ).

This can be reached by setting

n
X 1 ∗
σ(s, u) := − ∗ λ(s, Ti+1 )1l[Ti+1 ,Ti ) (u)
∗ ∗
δi+1
i=0

since
∗
ZTi
Σ(s, Ti∗ , Ti+1
∗ ∗
) = Σ(s, Ti+1 ) − Σ(s, Ti∗ ) = − ∗
σ(s, u) du = λ(s, Ti+1 ).
∗
Ti+1

Of course there are many other possibilities to specify σ. It is also possible

to choose a volatility structure σ that is continuous or smooth in the second
variable.
68 Lévy models for effective rates

Next, we specify the characteristics (eb, e

c, Fe) of the driving process L.
e For
∗
s ∈ [0, T ] let bs be arbitrary, e
e cs = cs and Fs such that
e
∗
Fes (dx) = exph−Σ(s, T ∗ ), xiFsT (dx). (3.11)
∗ ∗
Remember that F T is the third characteristic of the driving process LT in the
Lévy forward price model. Then
T ∗ (3.10) ∗
Fes i (dx) = exphΣ(s, Ti∗ ) − Σ(s, T ∗ ), xiFsT (dx)
i
!
∗
X
= exp hΣ(s, Tj∗ ) − Σ(s, Tj−1∗
), xi FsT (dx)
j=1
i
!
∗
X
= exp hλ(s, Tj∗ ), xi FsT (dx)
j=1

and we arrive at the forward price process

Zt !
∗
∗
F (t, Ti+1 , Ti∗ ) = F (0, Ti+1
∗
, Ti∗ ) exp ∗
λ(s, Ti+1 e Ts i
) dL ,
0

where
Zt Zt Zt Z
e Ti
∗ T∗ √ T∗ ∗
L = bs i ds + cs dWs i + x(µL − νeTi )(ds, dx).
e
t
0 0 0 Rd

∗
The PTi∗ -compensator νeTi of µL is given by
e

i
X 
Ti∗ ∗
νe (dt, dx) = exp hλ(t, Tj∗ ), xi FtT (dx) dt
j=1

T∗
and (bs i ) satisfies

Zt Zt
∗ T∗ 1 ∗ ∗
hλ(s, Ti+1 ), bs i i ds =− hλ(s, Ti+1 ), e
cs λ(s, Ti+1 )i ds
2
0 0
t
Z Z  ∗
 ∗
− ehλ(s,Ti+1 ),xi − 1 − hλ(s, Ti+1
∗
), xi νeTi (ds, dx).
0 Rd

Hence, we obtain forward price dynamics in the Lévy term structure model as
given in (3.4)–(3.7) for the forward price model. Consequently, we can regard
the Lévy forward price model as a special case of the Lévy term structure model.
In particular, the option pricing formulae developed in chapter 2 can be used.
Remark: This embedding only works for driving processes that are non-homo-
genous Lévy processes. If both models are driven by a process with stationary
 3.2 The Lévy Libor model 69

∗
increments, that is FsT and Fes do not depend on s, we usually cannot embed
the forward price model into the term structure model. Due to equation (3.11)
∗
this only works if FsT = Fes = 0 or in the pathetic case that Σ(s, T ∗ ) does not
depend on s (which implies that Σ(·, T ∗ ) is equal to zero).

3.2 The Lévy Libor model

The goal of this section is to give a short overview over the Lévy Libor model.
We are not going to present a construction of the model since this is done
in Eberlein and Özkan (2005) in detail. Instead, we list some of the model
properties that will be needed for option pricing in the subsequent sections as
well as for the Lévy Libor model with default risk which will be discussed in
chapter 4.
The model is constructed by backward induction and driven by a non-homo-
∗
geneous Lévy process LT on a complete stochastic basis (Ω, F = FT ∗ , F =
(Fs )0≤s≤T ∗ , PT ∗ ). As in the Lévy forward price model, PT ∗ should be regarded
∗
as the forward measure associated with the settlement day T ∗ . LT is required
to satisfy assumption (EM) and can be written in its canonical decomposition
as
Zt Zt Z
T∗ √ T∗ ∗
Lt = cs dWs + x(µ − ν T )(ds, dx).
0 0 Rd
∗
Here, W T denotes a standard Brownian motion, µ is the random measure asso-
∗ ∗ ∗
ciated with the jumps of LT , and ν T (dt, dx) = FsT (dx) dt is the compensator
∗ ∗
of µ. The characteristics of LT are given by (0, c, F T ). Note that without
∗
loss of generality we assume LT to be driftless. The following assumptions are
made:

(LR.1): For any maturity Ti there is a deterministic function λ(·, Ti ) : [0, T ∗ ] →

Rd , which represents the volatility of the forward Libor rate process
L(·, Ti ). In addition,
n−1
X
|λj (s, Ti )| ≤ M for all s ∈ [0, T ∗ ] and j ∈ {1, . . . , d}, (3.12)
i=1

where M is the constant from assumption (EM) and λ(s, Ti ) = 0 for

s > Ti .

(LR.2): The initial term structure B(0, Ti ) (i ∈ {1, . . . , n}) is strictly positive and
strictly decreasing (in i).

The dynamics of the forward Libor rates are specified as

 t
Zt

Z
T
L(t, Tk ) = L(0, Tk ) exp  bL (s, Tk , Tk+1 ) ds + λ(s, Tk ) dLs k+1  (3.13)
0 0
70 Lévy models for effective rates

with initial condition

 
1 B(0, Tk )
L(0, Tk ) = −1 .
δk B(0, Tk+1 )
∗
LTk+1 equals LT plus some – in general non-deterministic – drift term which is
chosen in such a way that LTk+1 is driftless under the forward measure associated
with the settlement day Tk+1 , henceforth denoted by PTk+1 . More precisely,

Zt Zt Z
T √ T
Lt k+1 = cs dWs k+1 + x(µ − ν Tk+1 )(ds, dx), (3.14)
0 0 Rd

where W Tk+1 is a standard Brownian motion with respect to PTk+1 and ν Tk+1 is
the PTk+1 -compensator of µ. The drift term bL (s, Tk , Tk+1 ) is specified in such
a way that L(·, Tk ) becomes a PTk+1 -martingale, i.e.
1
bL (s, Tk , Tk+1 ) = − hλ(s, Tk ), cs λ(s, Tk )i (3.15)
2 Z
 
T
− ehλ(s,Tk ),xi − 1 − hλ(s, Tk ), xi Fs k+1 (dx).
Rd

The connection between different forward measures is given by

n−1 n−1
dPTk+1 Y 1 + δl L(Tk+1 , Tl ) B(0, T ∗ ) Y
= = (1 + δl L(Tk+1 , Tl )).
dPT ∗ 1 + δl L(0, Tl ) B(0, Tk+1 )
l=k+1 l=k+1
(3.16)
Once restricted to the σ-field Ft this becomes
n−1
dPTk+1 B(0, T ∗ ) Y
= (1 + δl L(t, Tl )) (t ∈ [0, Tk+1 ]). (3.17)
dPT ∗ Ft B(0, Tk+1 )
l=k+1

The Brownian motions and compensators with respect to the different measures
are connected via
Zt n−1
!
Tk+1 ∗ √ X
Wt = WtT − cs α(s, Tl , Tl+1 ) ds (3.18)
0 l=k+1

with
δl L(s−, Tl )
α(s, Tl , Tl+1 ) := λ(s, Tl ) (3.19)
1 + δl L(s−, Tl )
and
n−1
!
∗ T
Y
Tk+1
ν (dt, dx) = β(s, x, Tl , Tl+1 ) ν T (dt, dx) =: Fs k+1 (dx) ds, (3.20)
l=k+1

where
δl L(s−, Tl )  hλ(s,Tl ),xi 
β(s, x, Tl , Tl+1 ) := e − 1 + 1. (3.21)
1 + δl L(s−, Tl )
3.2 The Lévy Libor model 71

Note that LTk+1 is usually not a (non-homogeneous) Lévy process under any
∗
of the measures PTi (except for k = n − 1, since LT is by definition a PIIAC
B(·,T )
under PT ∗ ). The construction by backward induction guarantees that B(·,Tkj ) is
a PTk -martingale for all j, k ∈ {1, . . . , n}.

3.2.1 Valuation of caps and floors

Eberlein and Özkan (2005) present a valuation formula for caps (and, there-
with, via the cap-floor-parity, also for floors) based on an approximation and
Laplace transform methods. In this section, we derive an alternative pricing for-
mula by making use of the same methods but a different approximation. The
advantage of our formula is the much higher speed at which it can be evaluated
numerically. Both approximations yield exact prices if the driving process is
continuous. At the end of this section, we consider an example to compare the
two approximations.
Remember that a cap (resp. floor) is a series of call (resp. put) options on
subsequent Libor rates. These single options are called caplets (resp. floorlets).
The time-t price of a caplet with strike K and maturity Ti is given by

Ct (K, Ti ) := δi B(t, Ti+1 )IEPTi+1 [(L(Ti , Ti ) − K)+ | Ft ].

A problem in evaluating the expression on the right-hand side arises from the
fact that L(·, Ti ) is not – generally – driven by a non-homogeneous Lévy process
under PTi+1 . To put it differently, the random measure associated with the
jumps of the driving process does not possess a deterministic PTi+1 -compensator
(except for the case i = n − 1, i.e. PTi+1 = PT ∗ ). Eberlein and Özkan (2005)
solve this problem by approximating the non-deterministic compensator with
δl L(s−,Tl )
a deterministic one. Concretely, they replace the stochastic term 1+δ l L(s−,Tl )
δl L(0,Tl )
in (3.21) by its deterministic initial value 1+δ l L(0,Tl )
. Combined with Laplace
transformation techniques, this leads to the following approximation for the
price of the caplet (compare Eberlein and Özkan (2005, Theorem 5.1)):

Proposition 3.1 The price of a caplet is approximately given by

1
C0 (K, Ti ) = δi B(0, Ti+1 )K (3.22)
π
Z∞  R+iu !
K 1
× < χ(iR − u) du,
L(0, Ti ) (R + iu)(R + 1 + iu)
0

where
 ZTi
1
χ(z) = exp − (z 2 + iz)hλ(s, Ti ), cs λ(s, Ti )i ds
2
0
ZTi Z   
izhλ(s,Ti ),xi hλ(s,Ti ),xi Ti+1
+ e − 1 − ize + iz νe (ds, dx)
0 Rd
72 Lévy models for effective rates

and νeTi+1 is an approximation for ν Ti+1 given by

n−1
Y  δk L(0, Tk )   
Ti+1 hλ(s,Tk ),xi ∗
νe (ds, dx) = e − 1 + 1 ν T (ds, dx).
1 + δk L(0, Tk )
k=i+1

Here,R R < −1 has to be chosen in such a way that χ(iR) < ∞ and it is assumed
∞
that −∞ |χ(u)| du < ∞.
To evaluate the expression in (3.22) one usually has to deal with a triple
integral, whose numerical evaluation is time consuming. Using a different ap-
proximation, we can get a pricing formula that only involves a double integral,
provided that the characteristic exponent θs of the infinitely divisible distribu-
∗
tion associated with the Lévy triplet (0, cs , FsT ) is known in closed form. Note
that
C0 (K, Ti ) = δi B(0, Ti+1 )IEPTi+1 [(L(Ti , Ti ) − K)+ ]
(3.16) B(0, T ∗ )
= δi B(0, Ti+1 )
B(0, Ti+1 )
" n−1 #
Y
+
IEPT ∗ (1 + δk L(Ti , Tk ))(L(Ti , Ti ) − K)
k=i+1
∗
= δi B(0, T )K IEPT ∗ [(MT1i − MT2i )+ ],
where the PT ∗ -martingales (Mt1 )0≤t≤Ti and (Mt2 )0≤t≤Ti+1 are given by
n−1
Y L(t, Ti )
Mt1 := (1 + δk L(t, Tk ))
K
k=i+1
n−1 Zt Zt !!
T
Y
L
= 1 + δk L(0, Tk ) exp b (s, Tk , Tk+1 ) ds + λ(s, Tk ) dLs k+1
k=i+1 0 0
Zt Zt !
L(0, Ti ) T
× exp bL (s, Ti , Ti+1 ) ds + λ(s, Ti ) dLs i+1
K
0 0
n−1 Zt !!
T∗
Y
= 1 + δk L(0, Tk ) exp λ(s, Tk ) dLs + drift
k=i+1 0
Zt !
L(0, Ti ) T∗
× exp λ(s, Ti ) dLs + drift
K
0

since, for all i, LTi and L T∗ differ only by a (non-deterministic) drift; similarly,
n−1
Y
Mt2 := (1 + δk L(t, Tk ))
k=i+1
n−1 Zt !!
T∗
Y
= 1 + δk L(0, Tk ) exp λ(s, Tk ) dLs + drift .
k=i+1 0
3.2 The Lévy Libor model 73

In the following, we approximate M 1 and M 2 by two processes for which we

require that they remain martingales with respect to PT ∗ . Let us exploit the
ε
fact that 1 + ε exp(x) ≈ (1 + ε) exp 1+ε x for small absolute values of x and
approximate

Zt !
T∗
1 + δk L(0, Tk ) exp λ(s, Tk ) dLs + drift
0

by

Zt !
δk L(0, Tk ) T∗
(1 + δk L(0, Tk )) exp λ(s, Tk ) dLs + new drift .
1 + δk L(0, Tk )
0

We obtain the approximations

Zt !
ft1 := L(0, Ti ) B(0, Ti+1 ) exp
M T∗
(f i (s) + λ(s, Ti )) dLs + drift
K B(0, T ∗ )
0

and
Zt !
f2 := B(0, Ti+1 ) exp
M T∗
f i (s) dLs + drift ,
t
B(0, T ∗ )
0

where
n−1
X δk L(0, Tk )
f i (s) := λ(s, Tk ). (3.23)
1 + δk L(0, Tk )
k=i+1

f1 and M
We derive the drift terms from our requirement that M f2 have to be
PT ∗ -martingales and get

Zt !
ft1 = L(0, Ti ) B(0, Ti+1 ) exp
M T∗
(f i (s) + λ(s, Ti )) dLs + Dt1 ,
K B(0, T ∗ )
0
Zt !
f2 B(0, Ti+1 ) i ∗
M t = exp f (s) dLTs + Dt2 (3.24)
B(0, T ∗ )
0

with
  Zt −1 
∗
Dt1 := log IEPT ∗ exp (f (s) +i
λ(s, Ti )) dLTs ,
0
  Zt −1 
∗
Dt2 := log IEPT ∗ exp f i
(s) dLTs .
0
74 Lévy models for effective rates

Hence,
C0 (K, Ti ) ≈ δi B(0, T ∗ )K IEPT ∗ [(M
f1 − M
Ti
f2 )+ ]
Ti
 f1 + 
MTi

∗ 2
= δi B(0, T )K IEPT ∗ MTi f −1 .
Mf2
Ti

Next, we make use of the change of numeraire technique and define a new
measure P i+1 on (Ω, FTi+1 ) by
eT
T
Zi+1 !
dP
eT f2
M Ti+1 T∗
i+1
:= = exp f i (s) dLs + DT2 i+1 . (3.25)
dPT ∗ f2
M 0 0

Then, denoting
t
f1
M L(0, Ti )
Z
∗
t
Xt := log = log + λ(s, Ti ) dLTs + Dt1 − Dt2
M
f 2
t
K
0

and assuming that the distribution of XTi with respect to P eT

i+1 possesses a
Lebesgue-density ϕ, we get
h
+ i
C0 (K, Ti ) ≈ δi B(0, Ti+1 )KIEPeT eXTi − 1 (3.26)
i+1

= δi B(0, Ti+1 )K(g ∗ ϕ)(0),

+
where g(x) := (e−x − 1) . We obtain the following approximation for the caplet
price:

Proposition 3.2 Suppose that the distribution of XTi possesses a Lebesgue-

fXTi the P
density. Denote by M eT -moment generating function of XT . Choose
Ti+1 i+1 i
XTi
an R < −1 such that M f (−R) < ∞. Then approximately
Ti+1

Z∞   R+iu
1 K 1
C0 (K, Ti ) = δi B(0, Ti+1 )K <
π L(0, Ti ) (R + iu)(R + 1 + iu)
0
ZTi 
× exp θs (f i (s) − (R + iu)λ(s, Ti ))
0
 
i i
+(R + iu) θs (f (s) + λ(s, Ti )) − (R + 1 + iu) θs (f (s)) ds du

with f i given by equation (3.23).

Proof: Proceeding in the same way as in the proof of theorem 2.16 we obtain
1
C0 (K, Ti ) = δi B(0, Ti+1 )K
π
Z∞  
1 XTi
× < M
f (−R − iu) du.
(R + iu)(R + 1 + iu) Ti+1
0
3.2 The Lévy Libor model 75

The claim now follows from the fact that for z ∈ C with < z = −R we have
" z ZTi !#
fXTi (z) L(0, Ti ) ∗
M Ti+1 = IEPeT exp zλ(s, Ti ) dLTs + z(DT1 i − DT2 i )
i+1 K
0
 z " ZTi #−1
L(0, Ti ) i ∗
= IEPT ∗ exp f (s) dLTs
K
0
" ZTi #
∗
× IEPT ∗ exp (f i (s) + zλ(s, Ti )) dLTs
0
h RT ∗
i !z
IEPT ∗ exp 0 i f i (s) dLTs
× h RT i
IEPT ∗ exp 0 i (f i (s) + λ(s, Ti )) dLTs ∗
 z ZTi 
L(0, Ti )
= exp θs (f i (s) + zλ(s, Ti ))
K
0

i i
−z θs (f (s) + λ(s, Ti )) + (z − 1) θs (f (s)) ds,

where the second equality follows from (3.25) and the last equality results from
proposition 1.9. 

The pricing formula of Eberlein and Özkan (2005) is exact if the model is
driven by a Brownian motion since in this case there is no compensator which
has to be approximated. The above approximation is also exact for a driving
Brownian motion as the following proposition shows:
Proposition 3.3 The above approximation yields the exact price for the caplet
if the driving process does not have jumps.
Proof: Equation (3.26) yields the approximate caplet price
" ZTi ! !+ #
√ fsTi+1
δi B(0, Ti+1 )IEPeT L(0, Ti ) exp cs λ(s, Ti ) dW + drift −K ,
i+1
0

where Wf Ti+1 denotes a PeT -standard Brownian motion. From the construc-
i+1

tion above it is clear that (exp Xt )0≤t≤Ti is a P

eT -martingale. We can thus
i+1
determine the drift and get
  ZTi
√ fsTi+1
C0 (K, Ti ) ≈ δi B(0, Ti+1 )IEPeT L(0, Ti ) exp cs λ(s, Ti ) dW
i+1
0
ZTi  + 
1
− hλ(s, Ti ), cs λ(s, Ti )i ds − K .
2
0
76 Lévy models for effective rates

On the other hand, the exact price for the caplet is given by

C0 (K, Ti ) = δi B(0, Ti+1 )IEPTi+1 [(L(Ti , Ti ) − K)+ ]

  ZTi
√ T
= δi B(0, Ti+1 )IEPTi+1 L(0, Ti ) exp cs λ(s, Ti ) dWs i+1
0
ZTi  + 
1
− hλ(s, Ti ), cs λ(s, Ti )i ds − K .
2
0

Hence, the approximate caplet price is exact. 

In the remaining part of this section, let us consider an example to compare

the caplet price approximation suggested by Eberlein and Özkan (2005) with
ours. Suppose that

T0 = 0, T1 = 0.5, T2 = 1, T3 = 1.5, ..., T10 = 5 = T ∗ .

We take discount factors (zero coupon bond prices) as quoted on February 19,
2002 (see table 2.1) and constant volatilities

λ(s, T1 ) = 0.20 λ(s, T2 ) = 0.19 λ(s, T3 ) = 0.18

λ(s, T4 ) = 0.17 λ(s, T5 ) = 0.16 λ(s, T6 ) = 0.15
λ(s, T7 ) = 0.14 λ(s, T8 ) = 0.13 λ(s, T9 ) = 0.12.

For the driving process, three different cases are considered:

(d1): The driving process is a standard Brownian motion. In this case we are
in the market model of Brace, Gatarek, and Musiela (1997) (henceforth
BGM model).

(d2): The driving process is a Lévy process generated by the NIG-distribution

with parameters α = δ = 100, µ = β = 0. This distribution has zero mean,
variance one, and it is very close to the standard normal distribution (see
Eberlein and v. Hammerstein (2004) for a survey of limiting cases of
generalized hyperbolic distributions).

(d3): The driving process is a Lévy process generated by the NIG-distribution

with parameters α = δ = 1.5, µ = β = 0. This distribution has zero mean
and variance one, but it has much fatter tails than the standard normal
distribution. Keep in mind that the parameters have to be chosen in such
a way that they satisfy (3.12).
P For an NIG-distribution with β = 0 this
means we have to have α > 9i=1 |λ(s, Ti )| = 1.44.

We price caplets with maturities ranging from T1 to T9 and strike rates ranging
from 2.5 % to 7 %. To calculate the option prices for (d1) the market formula
can be employed. In cases (d2) and (d3) we use the approximation of Eberlein
and Özkan (2005) and the approximation developed above.
3.2 The Lévy Libor model 77

The results can be found at the end of this chapter. Table 3.1 gives the caplet
prices for scenario (d1). Using distribution (d2) and any of the two approxima-
tions leads to exactly the same caplet prices (up to truncation of 1/1000 of a
basis point) as in table 3.1 and hence to the BGM-implied volatilities as given
in table 3.2. Tables 3.3 and 3.5 show caplet prices for scenario (d3) using the
approximation of Eberlein and Özkan (2005) and the approximation developed
above. The corresponding BGM-implied volatilities are given in tables 3.4 and
3.6. From these results we can draw some conclusions:

1. Not only are both approximations exact in the case of a driving Brownian
motion, they also (at least in this example) produce caplet prices for a
(purely discontinuous) driving Lévy process whose underlying distribution
is close to standard normal (case (d2)) that virtually perfectly fit the
BGM-prices.

2. In case (d3) a smile in the implied volatilities can be observed. This is

not surprising since the underlying distribution has fat tails. Note that
the two approximations produce almost the same caplet prices.

3. In case (d3) a phenomenon can be observed that already occurred in the

Lévy term structure model. For this driving homogeneous Lévy process
the smile in the implied volatilities flattens as time to maturity increases.

Let us shortly comment on the time that is needed to calculate the prices for
this set of 90 caplets. Naturally, the time depends on many factors as e.g. on the
choice of R, the upper limit of integration in the infinite integral as well as on the
numerical integration algorithm or the error bound in the numerical integration.
For choices that we consider to be reasonable, the following amounts of time
were needed on a personal computer: using the approximation of Eberlein and
Özkan (2005), the computation of the caplet prices lasted 385,04 and 436.31
seconds in the scenarios (d2) and (d3) respectively. The calculation using our
approximation lasted 0.12 and 0.09 seconds respectively. Besides the additional
integral, the fact that the Lebesgue-density of the Lévy measure of an NIG-
distribution contains a Bessel function made the first approximation slower.

3.2.2 Swaption pricing

The aim of this section is to provide a swaption pricing formula whose numerical
evaluation can be done fast. We use an approximation that has already been
employed by Brace, Gatarek, and Musiela (1997) in the Libor market model.
Moreover, we put the following restriction on the volatility structure which is
similar to assumption (VOL) in the Lévy term structure model:
Assumption ([Link]). The volatility structure factorizes, i.e.

λ(s, Ti ) = λi λ(s) (0 ≤ s ≤ Ti , i ∈ {1, . . . , n − 1})

where λi is a positive constant and λ : [0, T ∗ ] → Rd does not depend on i.

Examples of volatility structures for Libor models that satisfy this assumption
78 Lévy models for effective rates

can be found in the book of Brigo and Mercurio (2001, Section 6.3.1).
Remember that a payer (resp. receiver) swaption can be seen as a put (resp.
call) option on a coupon bond with an exercise price of 1 (compare Musiela
and Rutkowski (1998, Section 16.3.2)). Consider a payer swaption with strike
rate K where the underlying is a swap that starts at option maturity Ti and
matures at Tm (i < m ≤ n). It’s time-Ti value is given by
m
!+
X
πTi (K, Ti , Tm ) := 1 − ck B(Ti , Tk )
k=i+1
m k−1
! !+
X Y
= 1− ck (1 + δl L(Ti , Tl ))−1
k=i+1 l=i

with ck := K for i + 1 ≤ k ≤ m − 1 and cm := 1 + K.

To calculate todays value we proceed as before, that is, in order to be able
to apply Laplace transform methods, we derive a convolution representation
for the option price first. The value of the swaption is obtained by taking the
PTi -expectation of its time-Ti value. More precisely,
π0 := π0 (K, Ti , Tm )

m k−1
!!+ 
X Y
= B(0, Ti )IEPTi  1 − ck (1 + δl L(Ti , Tl ))−1 
k=i+1 l=i
(3.16)
= B(0, T ∗ )

n−1 m k−1
!!+ 
Y X Y
× IEPT ∗  (1 + δl L(Ti , Tl )) 1 − ck (1 + δl L(Ti , Tl ))−1 
l=i k=i+1 l=i
 !!+ 
m
X n−1
Y
= B(0, T ∗ )IEPT ∗  − ck (1 + δl L(Ti , Tl )) 
k=i l=k

with ci := −1. Combining (3.13), (3.14), and (3.15) with (3.18)–(3.21) yields
Zt Zt
√ ∗
L(t, Tl ) = L(0, Tl ) exp bL (s, Tl , T ∗ ) ds + cs λ(s, Tl ) dWsT
0 0
Zt Z !
T∗
+ hλ(s, Tl ), xi(µ − ν )(ds, dx)
0 Rd

where
bL (s, Tl , T ∗ ) :=
n−1
1 X δj L(s−, Tj )
− hλ(s, Tl ), cs λ(s, Tl )i − hλ(s, Tj ), cs λ(s, Tl )i
2 1 + δj L(s−, Tj )
j=l+1
 n−1
Z !
 Y ∗
− ehλ(s,Tl ),xi − 1 β(s, x, Tj , Tj+1 ) − hλ(s, Tl ), xi FsT (dx).
j=l+1
Rd
3.2 The Lévy Libor model 79

δj L(s−,Tj )
We approximate the stochastic term 1+δj L(s−,Tj ) in the drift by its starting value
δj L(0,Tj ) δj L(s−,Tj )
1+δj L(0,Tj ) (remember that 1+δj L(s−,Tj ) is also contained in β(s, x, Tj , Tj+1 ))
and call the resulting approximation for the drift term bL ∗
0 (s, Tl , T ). Similar
approximations have already been employed by Brace, Gatarek, and Musiela
(1997), Rebonato (1998), and Schlögl (2002). Using the assumption ([Link])
on the volatility structure yields
π0 ≈ B(0, T ∗ )
 
m n−1
Y  !!+
X λl
× IEPT ∗  − ck 1 + δl L(0, Tl ) exp XT + Bl 
λsum i
k=i l=k

with
n−1
X
λsum := λl ,
l=i
Zt n−1 Zt
∗ ∗
X
Xt := λ(s, Tl ) dLTs = λsum λ(s) dLTs ,
0 l=i 0
ZTi
∗
Bl := bL
0 (s, Tl , T ) ds.
0

Note that, due to the assumption on the volatility structure, we have derived a
representation for the price of the swaption that depends only on the distribu-
tion of one random variable, namely on the distribution of XTi with respect to
PT ∗ . Assume that this distribution possesses a Lebesgue-density ϕ, then
π0 ≈ B(0, T ∗ )
Z m n−1
Y  !!+
X λl
× − ck 1 + δl L(0, Tl ) exp x + Bl ϕ(x) dx
λsum
R k=i l=k
∗
= B(0, T )(g ∗ ϕ)(0) (3.27)
with g(x) := (v(x))+ and
m n−1
Y  !
X λl
v(x) := − ck 1 + δl L(0, Tl ) exp − x + Bl , (3.28)
λsum
k=i l=k

i.e. we have derived a convolution representation for the price of the swaption.
The next step is to determine the bilateral Laplace transform of g. Observe
that v has a unique zero; let us write v in a more complicated form as
n−1
Y  
λl
v(x) = 1 + δl L(0, Tl ) exp − x + Bl
λsum
l=i
m k−1
Y  −1 !!
X λl
× 1− ck 1 + δl L(0, Tl ) exp − x + Bl .
λsum
k=i+1 l=i
80 Lévy models for effective rates

Since the first (n − i) factors on the right hand side are strictly positive and
the last factor is continuous, strictly decreasing, and takes positive as well as
negative values, v has a unique zero Z. Consequently,

g(x) = v(x)1l(−∞,Z] (x).

Note that v(x) can be written (compare (3.28)) as a finite sum of expressions of
the type “c1 exp(−c2 x)” with c1 ∈ R and c2 ∈ [0, 1]. For z ∈ C with < z < −1
we get
−c1 −Z(z+c2 )
Z
e−zx c1 e−c2 x 1l(−∞,Z] (x) dx =


e .
z + c2
R

Hence, the Laplace transform of g exists for all z ∈ C with < z < −1 and a
closed form expression (depending on Z) can be derived. However, since the
number of summands of the above form in v increases exponentially as (n − i)
increases, a numerical evaluation of the Laplace transform is (at least for large
values of (n − i)) more appropriate. Note that we can save computational time
by applying the following multiplication scheme to v(x):
m
X n−1
Y   n−1
Y
ck dl = cm +dm−1 (cm−1 +dm−2 (cm−2 +dm−3 (. . . (ci+1 +di ci )))) dl .
k=i l=k l=m

Putting pieces together, we obtain the following formula for the swaption price:

Proposition 3.4 Suppose that the distribution of XTi possesses a Lebesgue-

XT
density. Denote by MT ∗ i the PT ∗ -moment generating function of XTi . Choose
XT
an R < −1 such that MT ∗ i (−R) < ∞. Then approximately

Z∞  ZTi 
1
π0 (K, Ti , Tm ) = B(0, T ∗ ) < L[g](R + iu) × exp θs (zλsum λ(s)) ds .
π
0 0

Proof: Using the convolution representation (3.27) and performing Laplace

and inverse Laplace transformations as usual, we get
Z∞  
1∗ XTi
π0 = B(0, T ) < L[g](R + iu)MT ∗ (−R − iu) du.
π
0

It remains to derive an expression for the moment generating function. For

z ∈ C with <z = −R we have
" ZTi !#
XT T∗
MT ∗ i (z) = IEPT ∗ exp zλsum λ(s) dLs
0
ZTi
= exp θs (zλsum λ(s)) ds,
0
3.2 The Lévy Libor model 81

where the last line follows from proposition 1.9. 

Remark: We can also use the approximation employed in this section to price
caplets and floorlets. However, volatility structures that do not satisfy assump-
tion ([Link]) can then not be considered.

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 %

T1 65.656 41.876 21.027 7.667 2.015
T2 93.623 70.228 48.032 29.218 15.671
T3 91.603 68.989 48.043 30.599 17.832
T4 109.422 87.172 65.916 46.932 31.411
T5 106.807 85.187 64.677 46.481 31.617
T6 114.792 93.622 73.276 54.745 39.003
T7 111.898 91.275 71.474 53.461 38.164
T8 116.996 96.827 77.249 59.083 43.236
T9 113.930 94.266 75.135 57.332 41.774
strike 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 0.398 0.063 0.008 0.001 0.000
T2 7.461 3.202 1.261 0.463 0.161
T3 9.590 4.817 2.288 1.040 0.456
T4 19.834 11.896 6.830 3.783 2.035
T5 20.441 12.641 7.531 4.352 2.453
T6 26.569 17.389 10.997 6.758 4.057
T7 26.074 17.131 10.885 6.726 4.062
T8 30.318 20.450 13.333 8.447 5.224
T9 29.100 19.453 12.542 7.841 4.778

Table 3.1: Scenarios (d1) and (d2): Caplet prices in basis points.

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 % 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.01 20.01
T2 19.00 19.00 19.00 19.00 19.00 19.00 19.00 19.00 19.00 19.00
T3 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00 18.00
T4 17.00 17.00 17.00 17.00 17.00 17.00 17.00 17.00 17.00 17.00
T5 16.00 16.00 16.00 16.00 16.00 16.00 16.00 16.00 16.00 16.00
T6 15.00 15.00 15.00 15.00 15.00 15.00 15.00 15.00 15.00 15.00
T7 14.00 14.00 14.00 14.00 14.00 14.00 14.00 14.00 14.00 14.00
T8 13.00 13.00 13.00 13.00 13.00 13.00 13.00 13.00 13.00 13.00
T9 12.00 12.00 12.00 12.00 12.00 12.00 12.00 12.00 12.00 12.00

Table 3.2: Scenario (d2): BGM-implied volatilities of the caplet prices (in %)
using any of the two approximations.
82 Lévy models for effective rates

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 %

T1 65.717 42.047 20.617 7.054 2.296
T2 93.666 70.364 48.072 28.768 15.041
T3 91.663 69.103 47.990 30.191 17.348
T4 109.462 87.269 65.968 46.750 30.990
T5 106.848 85.271 64.704 46.307 31.261
T6 114.822 93.693 73.328 54.658 38.736
T7 111.924 91.337 71.519 53.384 37.934
T8 117.014 96.876 77.304 59.058 43.075
T9 113.943 94.307 75.186 57.318 41.640
strike 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 0.867 0.374 0.179 0.092 0.051
T2 7.305 3.542 1.781 0.938 0.518
T3 9.425 5.035 2.718 1.504 0.857
T4 19.420 11.729 6.967 4.134 2.477
T5 20.095 12.489 7.623 4.630 2.824
T6 26.231 17.142 10.940 6.895 4.329
T7 25.783 16.914 10.826 6.831 4.285
T8 30.063 20.212 13.207 8.466 5.367
T9 28.879 19.243 12.428 7.853 4.897

Table 3.3: Scenario (d3): Caplet prices in basis points using the
approximation of Eberlein and Özkan.

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 % 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 25.33 21.62 18.94 18.81 20.87 23.28 25.49 27.45 29.19 30.74
T2 22.48 20.56 19.12 18.32 18.25 18.79 19.64 20.58 21.52 22.41
T3 19.89 18.69 17.89 17.51 17.52 17.82 18.30 18.85 19.43 19.99
T4 18.67 17.77 17.14 16.76 16.62 16.67 16.86 17.14 17.47 17.82
T5 17.19 16.52 16.06 15.80 15.71 15.75 15.88 16.08 16.32 16.58
T6 16.08 15.52 15.13 14.89 14.78 14.77 14.84 14.96 15.12 15.30
T7 14.88 14.42 14.10 13.91 13.82 13.81 13.86 13.96 14.09 14.23
T8 13.85 13.44 13.15 12.97 12.87 12.84 12.86 12.92 13.01 13.12
T9 12.76 12.39 12.14 11.98 11.89 11.87 11.88 11.93 12.01 12.10

Table 3.4: Scenario (d3): BGM-implied volatilities of the caplet prices (in %)
using the approximation of Eberlein and Özkan.
3.2 The Lévy Libor model 83

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 %

T1 65.716 42.044 20.609 7.041 2.285
T2 93.665 70.362 48.066 28.756 15.024
T3 91.663 69.100 47.982 30.178 17.331
T4 109.461 87.268 65.964 46.740 30.976
T5 106.848 85.269 64.700 46.300 31.250
T6 114.822 93.692 73.326 54.653 38.729
T7 111.924 91.336 71.517 53.381 37.930
T8 117.014 96.876 77.304 59.057 43.073
T9 113.943 94.307 75.186 57.318 41.640
strike 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 0.859 0.369 0.175 0.090 0.049
T2 7.287 3.527 1.769 0.930 0.512
T3 9.407 5.020 2.706 1.494 0.850
T4 19.404 11.713 6.952 4.122 2.467
T5 20.082 12.476 7.612 4.620 2.816
T6 26.222 17.133 10.931 6.886 4.322
T7 25.777 16.908 10.821 6.826 4.281
T8 30.061 20.209 13.204 8.463 5.365
T9 28.879 19.243 12.428 7.853 4.897

Table 3.5: Scenario (d3): Caplet prices in basis points using the
approximation above.

strike 2.5 % 3.0 % 3.5 % 4.0 % 4.5 % 5.0 % 5.5 % 6.0 % 6.5 % 7.0 %
T1 25.29 21.59 18.92 18.79 20.84 23.23 25.43 27.38 29.10 30.65
T2 22.45 20.54 19.10 18.30 18.23 18.77 19.61 20.55 21.48 22.37
T3 19.87 18.67 17.87 17.49 17.50 17.81 18.28 18.83 19.40 19.96
T4 18.65 17.75 17.12 16.75 16.61 16.66 16.85 17.12 17.45 17.81
T5 17.18 16.51 16.05 15.79 15.70 15.74 15.87 16.07 16.31 16.57
T6 16.07 15.51 15.12 14.89 14.78 14.77 14.83 14.95 15.11 15.29
T7 14.88 14.42 14.10 13.91 13.82 13.81 13.86 13.96 14.08 14.23
T8 13.85 13.44 13.15 12.97 12.87 12.84 12.86 12.92 13.01 13.12
T9 12.76 12.39 12.14 11.98 11.90 11.87 11.88 11.93 12.01 12.10

Table 3.6: Scenario (d3): BGM-implied volatilities of the caplet prices (in %)
using the approximation above.
Chapter 4

The Lévy Libor model with

default risk

In this chapter we present a credit risk model that extends the Lévy Libor
model to defaultable market rates. Since the Libor market model is a special
case of the Lévy Libor model, our approach can also be seen as an extension of
the market model.
The first extension of the Libor market model to defaultable contracts was
done by Lotz and Schlögl (2000). They use a deterministic hazard rate to con-
struct a default time and then price defaultable forward rate agreements with
unilateral as well as with bilateral default risk. Their assumption of a deter-
ministic hazard rate is rather restrictive. In particular, it implies that the pre-
default value of a defaultable bond is a deterministic multiple of the respective
default-free bond price.
A second approach to extend the Libor market model was presented in
Schönbucher (1999a). Default-free forward Libor rates are modelled according
to the market model. In addition, the dynamics of defaultable forward Libor
rates, i.e. ratios of pre-default values of defaultable zero coupon bonds, are
specified. This specification is not done directly but via forward credit spreads
or, alternatively, by modelling forward default intensities. A problem in this
approach arises from the fact that their evolution is just specified; however,
dynamics of defaultable forward Libor rates (or forward credit spreads or for-
ward default intensities) cannot be modelled freely in this context. In fact, they
follow by arbitrage arguments from the specification of the default time and the
default-free forward Libor rates (for more details on this point we refer to sec-
tion 4.1). Consequently, dynamics of defaultable forward Libor rates can only
be “specified” by giving a pre-specification and then constructing a default
time that implies these dynamics. In the same way, Bielecki and Rutkowski
(1999, 2000) and Eberlein and Özkan (2003) have already extended the Heath–
Jarrow–Morton model to defaultable bonds, for driving Brownian motions and
Lévy processes respectively.
The model presented here is an extension of the Lévy Libor model. We
follow the idea of Schönbucher (1999a) and specify, in addition to default-free
forward Libor rates, the evolution of defaultable forward Libor rates, not di-
86 The Lévy Libor model with default risk

rectly but via forward default intensities. A pre-specification for their evolution
is given and a default time is then constructed that implies this evolution. The
resulting model is a generalization of Schönbucher’s Libor market model with
default risk to driving (non-homogeneous) Lévy processes. A formula for deriva-
tive pricing is presented that uses two analogons to forward measures, namely
defaultable forward measures (or survival measures) and restricted defaultable
forward measures. Finally, we price some of the most popular credit derivatives.
The chapter is organized as follows. Section 4.1 presents the details of the
model. The time of default is constructed in section 4.2. A condition which gua-
rantees that the evolution of the forward default intensities implied by the time
of default matches its pre-specification is established in section 4.3. Section
4.4 introduces defaultable forward measures as well as restricted defaultable
forward measures and presents a formula for derivative valuation. This formula
is used in the remaining sections to price some of the most heavily traded credit
derivatives, namely credit default swaps, total rate of return swaps, options on
defaultable bonds, credit spread options and credit default swaptions.

4.1 Presentation of the model

Let us introduce some notation first. We consider a fixed time horizon T ∗ and a
discrete tenor structure 0 = T0 < T1 < . . . < Tn = T ∗ with δk := Tk+1 − Tk for
k = 0, . . . , n − 1. We assume that default-free as well as defaultable zero coupon
bonds with maturities T1 , . . . , Tn are traded on the market. By B(t, Tk ) (resp.
B 0 (t, Tk )) we denote the time-t price of a default-free zero coupon bond (resp.
a defaultable zero coupon bond with zero recovery) with maturity Tk . Indicate
the time of default by τ and the pre-default values of the defaultable bonds by
B(·, ·), then we have
B 0 (t, Ti ) = 1l{τ >t} B(t, Ti ) and B(Ti , Ti ) = 1 for i ∈ {1, . . . , n}.
In what follows we are not going to model bond prices directly (it is only
assumed that the processes describing the evolution of the bond prices B(·, Ti )
and of the pre-default prices B(·, Ti ) are special semimartingales whose values
as well as all left hand limits are strictly positive). Instead, we are going to
specify the dynamics of forward Libor rates. The following notation (taken
from Schönbucher (1999a)) will be used:
• The default-free forward Libor rates are given by
 
1 B(t, Tk )
L(t, Tk ) := −1 (k ∈ {1, . . . , n − 1}).
δk B(t, Tk+1 )

• The defaultable forward Libor rates are given by

 
1 B(t, Tk )
L(t, Tk ) := −1 (k ∈ {1, . . . , n − 1}).
δk B(t, Tk+1 )

• The forward Libor spreads are given by

S(t, Tk ) := L(t, Tk ) − L(t, Tk ) (k ∈ {1, . . . , n − 1}).
4.1 Presentation of the model 87

• The default risk factors or forward survival processes are given by

B(t, Tk )
D(t, Tk ) := (k ∈ {1, . . . , n}).
B(t, Tk )

• The discrete-tenor forward default intensities are given by

 
1 D(t, Tk )
H(t, Tk ) := −1 (k ∈ {1, . . . , n − 1}).
δk D(t, Tk+1 )

We begin by building up the default-free part of the model. The dynamics

of default-free forward Libor rates are specified in the same way as in the Lévy
Libor model. Once again, we refer to Eberlein and Özkan (2005) for a detailed
construction. The most important properties of this model including those that
we will need in this chapter can be found in section 3.2.
∗
The model is driven by a d-dimensional non-homogeneous Lévy process LT
∗
with characteristics (0, c, F T ) on a complete stochastic basis (Ω, e Fe = FeT ∗ , F
e=
T ∗
(Fs )0≤s≤T ∗ , PT ∗ ). Moreover, L is assumed to satisfy assumption (SUP) from
e
section 1.3. The only difference to section 3.2 is that we put a slightly stronger
∗
restriction on LT (remember we only assumed (EM) before) in this chapter
and that we use the notation (Ω, e F,
e F,e PT ∗ ) instead of (Ω, F, F, PT ∗ ). All the
properties of the Lévy Libor model that have been established in section 3.2
remain valid (of course, we have to replace Ft by Fet in (3.17)). Our goal in what
follows is to include defaultable forward Libor rates in the Lévy Libor model.
At first sight, an evident way to build up the defaultable part of the model is
to specify the dynamics of defaultable forward Libor rates by an expression simi-
lar to (3.13). However, L(Tk , Tk ) < L(Tk , Tk ) implies B(Tk , Tk+1 ) > B(Tk , Tk+1 )
in which case there is an arbitrage opportunity in the market, provided that
B 0 (·, Tk+1 ) has not defaulted until Tk . It seems thus natural to specify the model
in such a way that defaultable forward Libor rates are always higher than their
default-free counterparts. This can be achieved by modelling forward Libor
spreads or forward default intensities as positive processes, instead of specify-
ing defaultable forward Libor rates directly. We can then get the defaultable
forward Libor rates through

L(t, Tk ) = S(t, Tk ) + L(t, Tk )

or
L(t, Tk ) = H(t, Tk )(1 + δk L(t, Tk )) + L(t, Tk ). (4.1)
Unfortunately, there is a problem in specifying the dynamics of H or S directly.
Suppose that we have already constructed a time τ describing the time of
default. Let us for a moment assume that τ is a stopping time with respect to
the filtration F
e (in the explicit construction which will follow τ will only be a
stopping time with respect to some larger filtration, but for the point we are
making here, this can be ignored). The terminal value of a defaultable bond is
given by
B 0 (Tk , Tk ) = 1l{τ >Tk } B(Tk , Tk ) = 1l{τ >Tk } .
 88 The Lévy Libor model with default risk

On the other hand, in the model for the default-free Libor rates the time-t price
of a contingent claim X paying 1l{τ >Tk } at Tk is given by

Xt := B(t, Tk )IEPTk [1l{τ >Tk } |Fet ] = 1l{τ >t} B(t, Tk )IEPTk [1l{τ >Tk } |Fet ].

To have a consistent model, we thus have to have

B 0 (t, Tk ) = 1l{τ >t} B(t, Tk )IEPTk [1l{τ >Tk } |Fet ]

and consequently (at least on {τ > t})

B(t, Tk ) = B(t, Tk )IEPTk [1l{τ >Tk } |Fet ]

or equivalently
D(t, Tk ) = IEPTk [1l{τ >Tk } |Fet ],
which immediately provides a formula for H (and also for S). In other words, as
soon as τ is specified we cannot freely choose the dynamics of H (or S). What
we can and will do in the sequel is the following: We give a pre-specification for
H and then construct τ in such a way that the dynamics of H implied by τ will
match this pre-specification. The following assumptions are made in addition
to (LR.1) and (LR.2):

(DLR.1): For any maturity Ti there is a deterministic function γ(·, Ti ) : [0, T ∗ ] →

Rd+ , which represents the volatility of the forward default intensity H(·, Ti ).
We suppose that γ(s, Tk ) = 0 for Tk < s ≤ T ∗ . Moreover, we require that
the functions λ(·, Ti ) from (LR.1) map to Rd+ and condition (3.12) is tight-
ened by assuming that
n−1
X
(|λj (s, Ti )| + |γ j (s, Ti )|) ≤ M for all s ∈ [0, T ∗ ] and j ∈ {1, . . . , d}.
i=1
(4.2)

(DLR.2): The initial term structure B(0, Ti ) (i ∈ {1, . . . , n}) of defaultable zero
coupon bond prices satisfies 0 < B(0, Ti ) ≤ B(0, Ti ) for all Ti as well as
L(0, Ti ) ≥ L(0, Ti ), i.e.

B(0, Ti ) B(0, Ti )
≥ .
B(0, Ti+1 ) B(0, Ti+1 )

To avoid confusion, let us denote by Hb the pre-specified forward default inten-

sities, which we postulate to be given by
 Zt Zt
√ T
H(t,
b Tk ) = H(0, Tk ) exp bH (s, Tk , Tk+1 ) ds + cs γ(s, Tk ) dWs k+1
0 0
Zt Z 
Tk+1
+ hγ(s, Tk ), xi(µ − ν )(ds, dx) (4.3)
0 Rd
4.2 Construction of the time of default 89

subject to the initial condition

 
1 B(0, Tk )B(0, Tk+1 )
H(0, Tk ) = −1 .
δk B(0, Tk )B(0, Tk+1 )
W Tk+1 and ν Tk+1 are defined in (3.18) and (3.20). The drift term bH (·, Tk , Tk+1 )
will be specified later. For the moment we only assume bH (s, Tk , Tk+1 ) = 0 for
Tk < s ≤ T ∗ , i.e. we require that H(t, b k , Tk ) for t ∈ [Tk , T ∗ ].
b Tk ) = H(T

4.2 Construction of the time of default

The construction of the default time will be done in the canonical way, that is
for a given F-hazard
e process Γ a stopping time τ on an enlarged probability
space will be constructed. We will do the construction for a general Γ first. The
key question then will be which particular hazard process to choose to make H
match H.b For more details on the canonical construction we refer to Bielecki
and Rutkowski (2002), from whom the notation is adopted.
Let Γ be an F-adapted,
e e F,
right-continuous, increasing process on (Ω, e PT ∗ )
satisfying Γ0 = 0 and limt→∞ Γt = ∞. Furthermore, let η be a random vari-
b F,
able on some probability space (Ω, b that is uniformly distributed on [0, 1].
b P)
Consider the product space (Ω, G, QT ∗ ) defined by
e × Ω,
Ω := Ω b G := Fe ⊗ F,
b QT ∗ := PT ∗ ⊗ P
b

and denote by F the trivial extension of F e to the enlarged probability space

(Ω, G, QT ∗ ), i.e. each A ∈ Ft is of the form A e×Ω b for some Ae ∈ Fet . We extend
all stochastic processes from the default-free part of the model to the extended
∗ ∗
probability space (by setting LT (e b ) := LT (e
ω, ω ω ) and similarly for all other
processes).
Define a random variable τ : Ω → R+ by
τ := inf{t ∈ R+ : e−Γt ≤ η}.
and denote Ht := σ 1l{τ ≤u} | 0 ≤ u ≤ t and Gt := Ft ∨ Ht for t ∈ [0, T ∗ ].

Then τ is a stopping time with respect to the filtration G := (Gs )0≤s≤T ∗ since
{τ ≤ t} ∈ Ht ⊂ Gt . Moreover, for 0 ≤ s ≤ t ≤ T ∗ we have (compare Bielecki
and Rutkowski (2002, (8.14)))
QT ∗ {τ > s|FT ∗ } = QT ∗ {τ > s|Ft } = QT ∗ {τ > s|Fs } = e−Γs , (4.4)
i.e. Γ is the F-hazard process of τ under QT ∗ .
∗
A question that arises naturally is whether or not LT is a non-homogeneous
Lévy process with respect to QT ∗ and the enlarged filtration G. To answer it,
we make use of the following lemma:
Lemma 4.1 Equation (4.4) implies each of the following equivalent conditions:
1. FT ∗ and Hs are conditionally independent given Fs under QT ∗ , i.e. for
any bounded FT ∗ -measurable random variable X and any bounded Hs -
measurable random variable Y we have
IEQT ∗ [XY |Fs ] = IEQT ∗ [X|Fs ] IEQT ∗ [Y |Fs ] (s ∈ [0, T ∗ ]).
90 The Lévy Libor model with default risk

2. For any bounded, FT ∗ -measurable random variable X

IEQT ∗ [X|Gs ] = IEQT ∗ [X|Fs ] (s ∈ [0, T ∗ ]) (4.5)

Proof: Let us show the equivalence first. Suppose condition 2 holds, then

IEQT ∗ [XY |Fs ] = IEQT ∗ [IEQT ∗ [X|Gs ] Y |Fs ] = IEQT ∗ [X|Fs ] IEQT ∗ [Y |Fs ]

for bounded random variables X and Y that are FT ∗ -measurable and Hs -

measurable respectively.
Now assume that condition 1 holds. Let X be a bounded FT ∗ -measurable
random variable and A := {A1 ∩ A2 : A1 ∈ Hs , A2 ∈ Fs }, then by assumption
for A ∈ A

IEQT ∗ [X1lA |Fs ] = 1lA2 IEQT ∗ [X1lA1 |Fs ] = IEQT ∗ [X|Fs ] IEQT ∗ [1lA |Fs ]

and consequently
Z
IEQT ∗ [X|Fs ] dQT ∗ = IEQT ∗ [IEQT ∗ [1lA |Fs ]IEQT ∗ [X|Fs ]]
A
Z
= IEQT ∗ [X1lA ] = X dQT ∗ .
A

Since A is a generator of Gs = Fs ∨ Hs that is closed under the formation of

finite intersections, a uniqueness result as e.g. Billingsley (1979, Theorem 34.1)
yields condition 2.
It remains to show that equation (4.4) implies any of the two conditions.
We show that it implies the first one. Since E := {{τ ≤ u} : 0 ≤ u ≤ s}
is a generator of Hs that is closed under the formation of finite intersections,
equation (4.4) together with the usual uniqueness result for σ-finite measures
yields for any B ∈ Hs

QT ∗ {B|FT ∗ } = QT ∗ {B|Fs }.

By algebraic induction, we get for any bounded Hs -measurable random variable

Y
IEQT ∗ [Y |FT ∗ ] = IEQT ∗ [Y |Fs ].
Hence,

IEQT ∗ [XY |Fs ] = IEQT ∗ [XIEQT ∗ [Y |FT ∗ ] |Fs ] = IEQT ∗ [X|Fs ] IEQT ∗ [Y |Fs ]

for each bounded FT ∗ -measurable random variable X. 

∗
Proposition 4.2 LT is a non-homogeneous Lévy process on the stochastic
∗
basis (Ω, GT ∗ , G, QT ∗ ) with characteristics (0, c, F T ).
4.2 Construction of the time of default 91

∗ ∗
Proof: LT is clearly an adapted, càdlàg process and satisfies LT0 = 0. Its
characteristic function is given by
Z
∗ ∗
IEQT ∗ [exp(iuLTt )] = exp(iuLTt (e b )) d(PT ∗ ⊗ P)(e
ω, ω b ω, ω
b ))
Ω×
e Ωb
Z
∗
= exp(iuLTt (e
ω )) dPT ∗ (e
ω)
Ω
e
∗
= IEPT ∗ [exp(iuLTt )].
∗
Hence, the characteristic function of LTt and thus also the characteristics of
∗ ∗ ∗
LT are preserved. It remains to show that LTt − LTs is independent of Gs
for s < t. Let B ∈ B d and A ∈ Gs , then using condition 2 of lemma 4.1 with
∗ ∗ ∗ ∗
X := 1lB (LTt − LTs ) and the fact that LTt − LTs is independent of Fs we get
Z
T ∗ T ∗ ∗ ∗
QT ∗ (A ∩ {(Lt − Ls ) ∈ B}) = 1lB (LTt − LTs ) dQT ∗
A
Z
∗ ∗
= IEQT ∗ [1lB (LTt − LTs )|Fs ] dQT ∗
A
Z
∗ ∗
= IEQT ∗ [1lB (LTt − LTs )] dQT ∗
A
∗ ∗
= QT ∗ (A)QT ∗ ({(LTt − LTs ) ∈ B}). 

In particular, each forward Libor rate L(t, Tk )0≤t≤Tk is a martingale with respect
to the filtration (Gs )0≤s≤Tk and the measure QTk+1 , which is constructed from
QT ∗ in the same way as PTk+1 is constructed from PT ∗ .
Γ is not only the F-hazard process of τ under QT ∗ , but also the F-hazard
process of τ under all other forward measures, as the following lemma shows:

Lemma 4.3 Γ is the F-hazard process of τ under QTk for all k ∈ {1, . . . , n}.

Proof: Fix a k and denote by ψ the (FTk -measurable) Radon–Nikodym deriva-

tive of QTk with respect to QT ∗ . From lemma 4.1 we know that FT ∗ and Hs are
conditionally independent given Fs under QT ∗ . Using the abstract Bayes rule
and this conditional independence (plus a dominated convergence argument)
we get

IEQT ∗ [ψ1l{τ >s} |Fs ]

QTk {τ > s|Fs } =
IEQT ∗ [ψ|Fs ]
IEQT ∗ [ψ|Fs ] IEQT ∗ [1l{τ >s} |Fs ]
=
IEQT ∗ [ψ|Fs ]
= e−Γs . 
92 The Lévy Libor model with default risk

Let us now turn to the question which hazard process Γ to choose to make
H match its pre-specification. As pointed out at the beginning of this chapter,
to have a consistent model we have to have

B 0 (t, Tk ) = B(t, Tk )QTk {τ > Tk | Gt } (4.6)

IEQTk [1l{τ >Tk } |Ft ]
= B(t, Tk )1l{τ >t} ,
IEQTk [1l{τ >t} |Ft ]

where the last equality follows from Bielecki and Rutkowski (2002, (5.2)). Let
IEQTk [1l{τ >Tk } |Ft ] IEQTk [1l{τ >Tk } |Ft ]
B(t, Tk ) := B(t, Tk ) = B(t, Tk ) , (4.7)
IEQTk [1l{τ >t} |Ft ] e−Γt

then h i
D(t, Tk ) = IEQTk eΓt −ΓTk Ft .
In particular,
 h i 
1

D(t, Tk )

1  IEQTk e−ΓTk Ft
H(t, Tk ) = −1 = h i − 1 . (4.8)
δk D(t, Tk+1 ) δk IE e−ΓTk+1 Ft
QTk+1

It is clear from the previous equation that, in order to make H match its pre-
specification H, b we only need to specify the hazard process Γ at the points Tk
for k ∈ {1, . . . , n} in a suitable way. The values of Γ in between these points do
not have an influence on the value of H. Moreover, we know that
h i h i
IEQTk e−ΓTk FTk−1 = IEQTk 1l{τ >Tk } FTk−1
h h i i
= IEQTk IEQTk 1l{τ >Tk } GTk−1 FTk−1
 
B(Tk−1 , Tk )
= IEQTk 1l{τ >Tk−1 } FTk−1
B(Tk−1 , Tk )
B(Tk−1 , Tk )
= e−ΓTk−1
B(Tk−1 , Tk )
= e−ΓTk−1 (1 + δk−1 H(Tk−1 , Tk−1 ))−1 ,

where the third equation follows from (4.7) and Bielecki and Rutkowski (2002,
(5.2)). We now define Γ recursively by setting Γ0 := 0,

ΓTk := ΓTk −1 + log(1 + δk−1 H(T

b k−1 , Tk−1 )) (k ∈ {1, . . . , n})
k−1
X
= log(1 + δl H(T
b l , Tl )), (4.9)
l=0

and for t ∈ (Tk−1 , Tk )

Γt := (1 − αk (t))ΓTk−1 + αk (t)ΓTk ,

where αk : [Tk−1 , Tk ] → [0, 1] is a continuous, strictly increasing function sat-

isfying αk (Tk−1 ) = 0 and αk (Tk ) = 1. Obviously Γ is a continuous, strictly
4.3 Specification of the drift 93

b ·) > 0 by construction), and F-adapted

increasing (since H(·, e process (since ΓTk
is FTk−1 -measurable) and can be used for the canonical construction.
e
It still has to be checked whether the implied dynamics of H match those
of H. Using (4.8) and (4.9) we get
b
!
1 1
H(t, T1 ) = −1
δ1 IEQT [e−ΓT2 +ΓT1 | Ft ]
2
 
1  1
= h i − 1
δ1 IE 1
|F
QT2 t
1+δ1 H(T
b 1 ,T1 )

or, written differently,

" #
1 1
IEQT2 Ft = .
1 + δ1 H(T
b 1 , T1 ) 1 + δ1 H(t, T1 )
 
1
Consequently, H(·, T1 ) meets its pre-specification if is a
1+δ1 H(t,T
b 1) 0≤t≤T1
QT2 -martingale. More generally we have the following result. Recall that H(t,
b Ti ) =
∗
b i , Ti ) for t ∈ [Ti , T ].
H(T
Q 
l 1
Lemma 4.4 H(·, Tk ) meets its pre-specification if i=1 1+δi H(t,T is
i ) 0≤t≤Tl
b
a QTl+1 -martingale for all l ∈ {1, . . . , k}.

Proof: The result for k = 1 has been proven above. Using (4.8), (4.9) and the
prerequisite we get for k > 1
hQ i
k−1
 
1
I
E
1  QTk i=0 1+δi H(T
b i ,Ti ) Ft
H(t, Tk ) = hQ i − 1
δk IE k 1
Ft
QTk+1 i=0 b 1+δi H(Ti ,Ti )
1 b Tk )) − 1) = H(t,
= ((1 + δk H(t, b Tk ). 
δk

Remember that we can still choose the drift coefficients bH (·, Tk , Tk+1 ) in (4.3)
in order to satisfy the prerequisite of the previous lemma. This choice is done
in the next section.

4.3 Specification of the drift

We specify the drift recursively starting  with b(·, T1 , T2 ). More
 precisely, we
look for a process b(·, T1 , T2 ) such that (1 + δ1 H(t, T1 ))
b −1 becomes a
0≤t≤T1
 T2 -martingale first. The next step
Q is to specify b(·, T2 , T3 ) in such a way that

−1 
(1 + δ1 H(t, T1 ))(1 + δ2 H(t, T2 ))
b b becomes a QT3 -martingale, and
0≤t≤T2
so on. Let us begin with two lemmata:
94 The Lévy Libor model with default risk

Lemma 4.5 Let X be a real-valued semimartingale with X0 = 0 and ∆X >

−1. Then
   
−1 c c 1 X
(E(X)) = E −X + hX , X i + −1+x ∗µ .
1+x

Proof: Using Kallsen and Shiryaev (2002, Lemma 2.6) we get

 
1 c c X
E(X) = exp X − hX , X i + (log(1 + x) − x) ∗ µ
2

Consequently, again using Kallsen and Shiryaev (2002, Lemma 2.6),

 
−1 1 c c X
(E(X)) = exp −X + hX , X i − (log(1 + x) − x) ∗ µ
2

1
= E − X + hX c , X c i − (log(1 + x) − x) ∗ µX
2
  
1 1
+ hX c , X c i + − 1 + log(1 + x) ∗ µX
2 1+x
   
c c 1 X
= E −X + hX , X i + −1+x ∗µ . 
1+x

Lemma 4.6 For k ∈ {2, . . . , n} and i ∈ {1, . . . , k − 1}

 Z• Z•
√
H(t,
b Ti ) = H(0, Ti )Et a(s, Ti , Tk ) ds + cs γ(s, Ti ) dWsTk
0 0
Z• Z   
hγ(s,Ti ),xi Tk
+ e − 1 (µ − ν )(ds, dx) ,
0 Rd

where
1
a(s, Ti , Tk ) := bH (s, Ti , Tk ) + hγ(s, Ti ), cs γ(s, Ti )i (4.10)
Z  2

hγ(s,Ti ),xi
+ e − 1 − hγ(s, Ti ), xi FsTk (dx).
Rd

and bH (s, Ti , Tk ) is given by (4.11).

Proof: From the default-free part of the model we know that

Zt k−1
!
T √ X
Wt i+1 = WtTk − cs α(s, Tl , Tl+1 ) ds
0 l=i+1

and
k−1
!
Y
ν Ti+1 (dt, dx) = β(s, x, Tl , Tl+1 ) ν Tk (dt, dx)
l=i+1
4.3 Specification of the drift 95

with α and β given by (3.19) and (3.21). Consequently, equation (4.3) implies

 Zt Zt
H √
H(t,
b Ti ) = H(0, Ti ) exp b (s, Ti , Tk ) ds + cs γ(s, Ti ) dWsTk
0 0
Zt Z 
Tk
+ hγ(s, Ti ), xi(µ − ν )(ds, dx) ,
0 Rd

where

bH (s, Ti , Tk ) := bH (s, Ti , Ti+1 ) (4.11)

k−1
* !+
X
− γ(s, Ti ), cs α(s, Tl , Tl+1 )
l=i+1
k−1
Z !
Y
− hγ(s, Ti ), xi β(s, x, Tl , Tl+1 ) − 1 FsTk (dx).
l=i+1
Rd

The claim now follows from Kallsen and Shiryaev (2002, Lemma 2.6). 

 
1
Proposition 4.7 is a QT2 -martingale if for s ∈ [0, T1 ]
1+δ1 H(t,T
b 1) 0≤t≤T1
 
H 1 1
b (s, T1 , T2 ) = Ys− − hγ(s, T1 ), cs γ(s, T1 )i (4.12)
2
!
ehγ(s,T1 ),xi − 1
Z
+ hγ(s, T1 ), xi − 1 ehγ(s,T1 ),xi − 1

FsT2 (dx),
1 + Ys−
Rd

δ1 H(s,T 1)
where Ys1 :=
b
.
1+δ1 H(s,T
b 1)

Proof: Lemma 4.6 gives us

 Z• Z•
√
H(t,
b T1 ) = H(0, T1 )Et a(s, T1 , T2 ) ds + cs γ(s, T1 ) dWsT2
0 0
Z• Z   
hγ(s,T1 ),xi T2
+ e − 1 (µ − ν )(ds, dx)
0 Rd

with
1
a(s, T1 , T2 ) = bH (s, T1 , T2 ) + hγ(s, T1 ), cs γ(s, T1 )i (4.13)
Z  2

+ ehγ(s,T1 ),xi − 1 − hγ(s, T1 ), xi FsT2 (dx).
Rd
96 The Lévy Libor model with default risk

Consequently, for Xt1 := 1 + δ1 H(t,

b T1 ) we have

dXt1 = δ1 dH(t,
b T1 )

1 √
1
= Xt− 1
Yt− a(t, T1 , T2 ) dt + Yt− ct γ(t, T1 ) dWtT2
Z   
+ Yt− 1
ehγ(t,T1 ),xi − 1 (µ − ν T2 )(dt, dx) .
Rd
Lemma 4.5 implies
 Z• Z•
1 √
(Xt1 )−1 = (X01 )−1 Et A(s, T2 ) ds − Ys− cs γ(s, T1 ) dWsT2
0 0
Z• Z   −1  
1 hγ(s,T1 ),xi T2
+ 1+ Ys− e −1 − 1 (µ − ν )(ds, dx) ,
0 Rd

where
1 1 2
A(s, T2 ) := − Ys− a(s, T1 , T2 ) + (Ys− ) hγ(s, T1 ), cs γ(s, T1 )i (4.14)
ehγ(s,T1 ),xi − 1
Z  
+ Ys− 1
ehγ(s,T1 ),xi − 1 − 1 hγ(s,T ),xi

FsT2 (dx).
1 + Ys− e 1 −1
Rd

b T1 ))−1 is a QT -local martingale if A(·, T2 ) ≡ 0. In this case

Thus, (1 + δ1 H(·, 2
it is also a martingale since it is bounded by 0 and 1 and therefore of class
[D] (compare Jacod and Shiryaev (2003, I.1.47c)). Combining A(·, T2 ) ≡ 0 with
(4.13) and (4.14) yields (4.12). 

More generally, we get the following proposition:

Q 
k−1 1
Proposition 4.8 i=1 is a martingale with respect to
b 1+δi H(t,Ti ) 0≤t≤Tk−1
QTk for k ∈ {2, . . . , n} if for all i ∈ {1, . . . , k − 1} and s ∈ [0, Ti ]
bH (s, Ti , Ti+1 ) = (4.15)
i
X j 1
Ys− hγ(s, Tj ), cs γ(s, Ti )i − hγ(s, Ti ), cs γ(s, Ti )i
2
j=1
i−1
!
j
X Ys−
+ i
hγ(s, Tj ), cs α(s, Ti , Ti+1 )i
j=1
Ys−
 
Z
ehγ(s,T i ),xi −1 Ti+1
+ hγ(s, Ti ), xi − Q 
  Fs (dx)
i j hγ(s,T ),xi
j=1 1 + Ys− e
j −1
Rd
Z
i −1
+(Ys− ) (β(s, x, Ti , Ti+1 ) − 1)
Rd
 i−1   −1 
j T
Y
hγ(s,Tj ),xi
× 1− 1 + Ys− e −1 Fs i+1 (dx),
j=1
4.3 Specification of the drift 97

δi H(s,Ti)
where Ysi :=
b
.
1+δi H(s,T
b i)

Proof: The proof is done in the appendix since it is calculationally intense. 

Note that we cannot just define bH (s, Ti , Ti+1 ) by (4.15) since the term
on the right hand side involves Ysi which depends on H(s, b Ti ) and thus on
H
b (·, Ti , Ti+1 ) itself. In other words, we have to deal with a stochastic differential
equation. Suppose that for every i ∈ {1, . . . , k − 1} there is a unique solution
to the SDE
Zt Zt
i √ T
h(t, Ti ) = h(0, Ti ) + f (s, h(s−, Ti )) ds + cs γ(s, Ti ) dWs i+1
0 0
Zt Z
+ hγ(s, Ti ), xi(µ − ν Ti+1 )(ds, dx) (4.16)
0 Rd

with
h(0, Ti ) := log H(0, Ti )
and
f i (s, x) := f1i (s) + f2i (s, x) + f3i (s, x) + f4i (s, x),
where
i−1
X δj eh(s−,Tj ) 1
f1i (s) := h(s−,T )
hγ(s, Tj ), cs γ(s, Ti )i − hγ(s, Ti ), cs γ(s, Ti )i
1 + δj e j 2
j=1
Z  
T
− ehγ(s,Ti ),yi − 1 − hγ(s, Ti ), yi Fs i+1 (dy),
Rd
δi ex
f2i (s, x) := hγ(s, Ti ), cs γ(s, Ti )i
1 + δi ex
i−1
!
1 + δi ex X δj eh(s−,Tj )
+ hγ(s, Tj ), cs α(s, Ti , Ti+1 )i
δi ex 1 + δ j eh(s−,Tj )
j=1
1 + δ ex Z
i
f3i (s, x) := (β(s, y, Ti , Ti+1 ) − 1) (4.17)
δi ex
Rd
i−1   −1 
δj eh(s−,Tj )  hγ(s,Tj ),yi

T
Y
1− 1+ h(s−,T )
e −1 Fs i+1 (dy),
j=1
1 + δj e j

and
Z   
δi ex  hγ(s,Ti ),yi −1
hγ(s,Ti ),yi
f4i (s, x) := e −1 1− 1+ e −1
1 + δi ex
Rd
i−1 
!
δj eh(s−,Tj )  hγ(s,Tj ),yi  −1
T
Y
× 1+ h(s−,T )
e −1 Fs i+1 (dy).
j=1
1 + δj e j
98 The Lévy Libor model with default risk

b Ti ) := exp h(s, Ti ) satisfies (4.3) with drift term bH (s, Ti , Ti+1 ) given
Then H(s, Q 
k−1 1
by (4.15). In this case proposition 4.8 yields that i=1 is
b 1+δi H(t,Ti ) 0≤t≤Tk−1
a QTk -martingale.
To prove that there is a unique solution to (4.16) we make use of the fol-
lowing theorem which is a direct consequence of Protter (1992, Theorem V.7)
(see also Protter (1992, Theorem V.6)):

Theorem 4.9 Assume a (one-dimensional) semimartingale Z with Z0 = 0 on

a complete stochastic basis (Ω, F, F, P) to be given and let f : R+ × Ω × R → R
be such that

1. for fixed x ∈ R, (t, ω) 7→ f (t, ω, x) is an adapted càglàd process, i.e. it has

left-continuous paths that admit right-hand limits.

2. there exists a finite random variable K such that for all t ∈ R+

|f (t, ω, x) − f (t, ω, y)| ≤ K(ω)|x − y|.

Then the stochastic differential equation

Zt
Xt = X0 + Zt + f (s, ·, Xs− ) ds,
0

where X0 is a constant, has a unique (strong) solution. This solution is a semi-

martingale.

Unfortunately, the functions f2i and f3i in (4.16) are not globally Lipschitz, i.e.
they do not satisfy condition 2 of the previous theorem. However, for the SDE
in consideration we can weaken this condition by assuming that f is locally
Lipschitz and satisfies a growth condition, as the following proposition shows:

PropositionR • √4.10 Assume

R • R we are given a d-dimensional special semimartin-
gale S := 0 cs dWs + 0 Rd x(µ − ν)(ds, dx) on a complete stochastic basis
(Ω, F, F, P), where W is a standard Brownian motion, c is deterministic, and
µ is the random measure associated with the jumps of S with (possibly non-
deterministic) compensator ν(ds, dx) = Fs (dx) ds. Suppose that σ : R+ → Rd
is a bounded function and let f : R+ × Ω × R → R be such that

1. for fixed x ∈ R, (t, ω) 7→ f (t, ω, x) is an adapted càglàd process.

2. for all r > 0 there is a real number Kr such that for all (t, ω) and all
x, y ∈ R with |x|, |y| ≤ r

|f (t, ω, x) − f (t, ω, y)| ≤ Kr |x − y| and |f (t, ω, x)| ≤ Kr .

3. there is a constant B1 such that for all (t, ω) and all x ∈ R

xf (t, ω, x) ≤ B1 (1 + x2 ).
4.3 Specification of the drift 99

Suppose further that there is a constant B2 such that for all (t, ω)
Z
hσ(t), ct σ(t)i + hσ(t), yi2 Fs (dy) ≤ B2 . (4.18)
Rd

Then the stochastic differential equation

Zt Zt
Xt = X0 + f (s, ·, Xs− ) ds + σ(s) dSs , (4.19)
0 0

where X0 is a constant, has a unique (non-exploding) solution which is a semi-

martingale.
Proof: The proof uses ideas of the proofs of theorems 2.2.3 and 2.3.3 in Reiss
(2003), where a similar statement is established for a deterministic function f
and a driving Brownian motion.
Let us show uniqueness of a solution first. Suppose X 1 and X 2 are two
solutions. To show that they are indistinguishable, it is enough to show that
they are modifications of each other since their paths are right continuous. Fix
a t ∈ R+ and define for n ∈ N
τn1 := inf{s ≥ 0 : |Xs1 | ≥ n}, τn2 := inf{s ≥ 0 : |Xs2 | ≥ n}.
Since the usual conditions hold, τn1 and τn2 are stopping times. Consequently,
τn := min(τn1 , τn2 ) is a stopping time that converges to infinity almost surely as
n → ∞. Hence,
 t∧τ 
Z n
 1 2
 1 2
IE |Xt∧τ n
− Xt∧τ n
| = IE  f (s, ·, Xs− ) − f (s, ·, Xs− ) ds 
0
 t∧τ 
Z n
≤ Kn IE  |Xs1 − Xs2 | ds
0
Zt
IE 1l{τn ≥s} |Xs1 − Xs2 | ds
 
= Kn
0
Zt
 1 2

≤ Kn IE |Xs∧τn
− Xs∧τn
| ds.
0
 1 2

We can apply Gronwall’s Lemma and conclude IE |Xt∧τ n
− Xt∧τn
| = 0. Thus,
1
Xt∧τ 2
= Xt∧τ almost surely for all n. Letting n → ∞ yields Xt1 = Xt2 almost
n n
surely.
To prove the existence statement, we use the previous theorem together
with a suitable cut-off scheme. For any R > 0 define

 f (s, ω,
x) for |x| ≤ R
x

2 − R
f (s, ω, R) for r < x < 2R

fR (s, ω, x) := x

 2 + R f (s, ω, −R) for − 2R < x < R
0 for |x| ≥ 2R.

100 The Lévy Libor model with default risk

 
Then fR satisfies the conditions of theorem 4.9 with K(ω) := max KR , KRR =:
K R . Denote by X R the (by theorem 4.9) unique solution of the SDE

Zt Zt
Xt = X0 + fR (s, ·, Xs− ) ds + σ(s) dSs
0 0
Zt Zt Zt Z
√
= X0 + fR (s, ·, Xs− ) ds + cs σ(s) dWs + hσ(s), xi(µ − ν)(ds, dx).
0 0 0 Rd

Introduce the stopping time τR := inf{t ≥ 0 : |XtR | ≥ R} and define

Xt∞ := XtR for t ≤ τR .

To check that X ∞ is well defined let 0 < R1 < R2 and τ := min(τR1 , τR2 ), then
(similarly as in the proof of uniqueness)

Zs
 
 
R1 R2
IE sup |XsR1 − XsR2 | = IE  sup fR1 (u, ·, Xu− ) − fR2 (u, ·, Xu− ) du 
0≤s≤t∧τ 0≤s≤t∧τ
0
 t∧τ 
Z
R1 R2
≤ IE  |fR1 (u, ·, Xu− ) − fR2 (u, ·, Xu− )| du
0
 t∧τ 
Z
≤ K R2 IE  |XuR1 − XuR2 | du
0
Zt  
≤ K R2 IE sup |XsR1 − XsR2 | du.
0≤s≤u∧τ
0

Again, we can apply Gronwall’s Lemma and conclude

 
IE sup |XsR1 − XsR2 | =0 for all t.
0≤s≤t∧τ

Hence, XtR1 and XtR2 coincide almost surely for t ≤ min(τR1 , τR2 ) and X ∞ is
well defined. It remains to show that limR→∞ τR = ∞ almost surely, since in
this case X ∞ is a solution to (4.19) and therefore a semimartingale.
4.3 Specification of the drift 101

Let h(x, t) := e−Bt (1 + x2 ) with B := 2B1 + B2 and YtR := h(XtR , t), then
by Itô’s formula

YtR − Y0R = e−Bt (1 + (XtR )2 ) − (1 + (X0 )2 )

Zt Zt
R −Bs
dXsR − B e−Bs 1 + (Xs−R 2


= 2Xs− e ) ds
0 0
Zt Zt Z
−Bs
+ e hσ(s), cs σ(s)i ds + e−Bs hσ(s), xi2 µ(ds, dx)
0 0 Rd
= local martingale
Zt 
+ e−Bs 2Xs− R R R 2


fR (s, ·, Xs− ) − B 1 + (Xs− )
0
Z 
+hσ(s), cs σ(s)i + hσ(s), xi2 Fs (dx) ds.
Rd

From condition 3 of the prerequisites and (4.18) we know that there is a localiz-
ing sequence (Tn )n≥1 such that the stopped process (Y R )Tn is a supermartingale
for all n. By the optional stopping theorem, the stopped process (Y R )Tn ∧τR is
a supermartingale for all n. Hence,
h
i
1 + (X0 )2 ≥ IE e−B(t∧Tn ∧τR ) 1 + (Xt∧T
R
n ∧τR
)2

≥ e−Bt (1 + R2 )P({τR ≤ t} ∩ {τR ≤ Tn }).

Taking the limes inferior (over n) on both sides and using Fatou’s Lemma we
obtain
1 + (X0 )2 ≥ e−Bt (1 + R2 )P({τR ≤ t}).

From limR→∞ 1 + R2 = ∞ we get limR→∞ P({τR ≤ t}) = 0. Since for R1 < R2

we have {τR2 ≤ t} ⊂ {τR1 ≤ t}, there exists for P-almost every ω and all t > 0 a
constant R0 (which may depend on ω and t) such that τR (ω) ≥ t for all R ≥ R0 .
This is equivalent to limR→∞ τR = ∞ almost surely. 

We can use the previous proposition to check that, at least in case the
driving process L is one-dimensional (d=1), the SDE (4.16) admits a unique
non-exploding solution:

Proposition 4.11 Assume d = 1. Suppose that γ(·, Ti ) is a càglàd function for

∗
each i ∈ {1, . . . , n − 1} and that the characteristics of LT are chosen in such
a way that f1i (·, ·, x), . . . , f4i (·, ·, x) have càglàd paths for each x ∈ R. Then the
stochastic differential equation (4.16) admits a unique (non-exploding) solution
for each i ∈ {1, . . . , n − 1}.
102 The Lévy Libor model with default risk

Proof: We use proposition 4.10 with

(Ω, F, F, P) := (Ω,
e F, e PT ),
e F,
i+1

Zt Zt Z
√ Ti+1
St := cs dWs + x(µ − ν Ti+1 )(ds, dx),
0 0 R

and σ(s) := γ(s, Ti ). Assumptions (SUP) and (DLR.1) imply (4.18). It remains
to verify the conditions 1-3 of proposition 4.10 for f i . Condition 1 is satisfied
by assumption. Conditions 2 and 3 can be checked separately for f1i , . . . , f4i .
Again, (SUP) and (DLR.1) yield that condition 2 holds for f1i , . . . , f4i and that
condition 3 holds for f1i and f4i . It remains to show that condition 3 is also
satisfied for f2i and f3i . For this purpose, it is sufficient to prove that there are
constants C2 , C3 such that for all (t, ω) and all x ∈ R

0 ≤ feji (t, ω, x) ≤ Cj (j ∈ {2, 3}),

δi ex
where feji (t, ω, x) := i
1+δi ex fj (t, ω, x).
The existence of the upper bound once
again follows from (SUP) and (DLR.1). Moreover, fe2i and fe3i are nonnegative
since α(·, Ti , Ti+1 ) and the integrand in (4.17) are nonnegative (at this point,
the assumption d = 1 is needed). 

Remark: To prove the existence of a solution to (4.16) for d > 1 we have

to put further restrictions on the characteristics of L to meet the growth
condition (condition 3) of proposition 4.10. For example, in the case of a
multivariate Gaussian model (i.e. Fs = 0 for all s ∈ [0, T ∗ ]) assuming that
hγ(s, Tj ), cs λ(s, Ti )i is nonnegative for all 1 ≤ j < i ≤ n − 1 will do the job.
This can be achieved by requiring that all entries in the matrices cs are non-
negative.

In the subsequent sections, we assume that the drift terms bH (·, Ti , Ti+1 ) are
chosen as described above and do not distinguish between H b and H anymore.

4.4 Defaultable forward measures

It is well known that pricing of derivatives in default-free interest rate mod-
els can often be facilitated considerably by changing numeraires, i.e. changing
measures, in a suitable way. In particular, forward measures prove to be useful
in many situations. Similarly, valuation of contingent claims in our model can
be simplified by using two counterparts to the default-free forward measures.
The first definition traces back to Schönbucher (1999a):

Definition 4.12 The defaultable forward (martingale) measure or survival mea-

sure QTi for the settlement day Ti is defined on (Ω, GTi ) by

dQTi B(0, Ti ) 0 B(0, Ti )

:= 0 B (Ti , Ti ) = 1l{τ >Ti } .
dQTi B (0, Ti ) B(0, Ti )
4.4 Defaultable forward measures 103

Equation (4.6) ensures that the preceding expression is indeed a density. QTi
corresponds to the choice of B 0 (·, Ti ) as a “numeraire”. We use quotation marks
since B 0 (·, Ti ) is not a strictly positive process with probability one. Conse-
quently, QTi is absolutely continuous with respect to QTi , but the two measures
are not mutually equivalent. In particular, the set A = {τ ≤ t} for t ∈ (0, Ti ]
has a strictly positive probability under QTi but zero probability under QTi .
The term “survival measure” is justified by the fact that
QTi (A ∩ {τ > Ti })
QTi (A) = = QTi (A|{τ > Ti }) (A ∈ GTi ),
QTi ({τ > Ti })

i.e. QTi can be regarded as the forward measure QTi conditioned on survival
until Ti . Once restricted to the σ-field Gt , the defaultable forward measure
becomes
dQTi B(0, Ti ) B(t, Ti ) B(0, Ti ) QT ({τ > Ti }|Ft )
= 1l{τ >t} = 1l{τ >t} i .
dQTi B(0, Ti ) B(t, Ti ) B(0, Ti ) QTi ({τ > t}|Ft )
Gt
0
The first equality follows from the fact that BB(·,T
(·,Ti )
i)
is a QTi -martingale, the
second equality from (4.7).
Another very useful tool in the context of derivative pricing is the re-
stricted defaultable forward measure, which has already been used in Bielecki
and Rutkowski (2002, Section 15.2). Note that the defaultable forward measure
restricted to the σ-field Ft is given by

dQTi B(0, Ti )
= QTi ({τ > Ti }|Ft )
dQTi B(0, Ti )
Ft

and denote by PTi the restriction of QTi to the σ-field FTi . This notation differs
slightly from the notation in the default-free part of the model where PTi was
defined on FeTi . However, this should not cause any confusion since FTi is the
trivial extension of FeTi .

Definition 4.13 The restricted defaultable forward (martingale) measure PTi

for the settlement day Ti is defined on (Ω, FTi ) by

dPTi B(0, Ti )
= QTi ({τ > Ti }|FTi ).
dPTi B(0, Ti )
We have an explicit expression for this density, namely
i−1
dPTi B(0, Ti ) −ΓT B(0, Ti ) Y 1
= e i = . (4.20)
dPTi B(0, Ti ) B(0, Ti ) k=0 1 + δ k H(T k , Tk )

Restricted to the the σ-field Ft this becomes (since i−1 1

Q
k=0 1+δk H(·,Tk ) is a PTi -
martingale)
i−1
dPTi B(0, Ti ) Y 1
= . (4.21)
dPTi Ft B(0, Ti ) 1 + δk H(t, Tk )
k=0
104 The Lévy Libor model with default risk

From equation (A.1) we get the representation

Z• i−1
dPTi l √
X
= ETi−1 − Ys− cs γ(s, Tl ) dWsTi
dPTi
0 l=1
•
i−1 
! !
Z Z Y  −1
l hγ(s,Tl ),xi Ti
+ 1+ Ys− e −1 − 1 (µ − ν )(ds, dx)
0 Rd l=1

with
δl H(s, Tl )
Ysl := .
1 + δl H(s, Tl )
Hence, the two predictable processes in Girsanov’s Theorem for semimartingales
(see Jacod and Shiryaev (2003, Theorem III.3.24)) associated with this change
of measure are
i−1 
X 
l
β(s) = − Ys− γ(s, Tl ) and
l=1
i−1
Y  −1
Y (s, x) = l
1 + Ys− ehγ(s,Tl ),xi − 1 .
l=1

We can conclude that

Zt X
i−1
T l √
Wti := WtTi + Ys− cs γ(s, Tl ) ds (4.22)
0 l=1

is a PTi -standard Brownian motion and the PTi -compensator of µ is given by

i−1   −1
Y T
Ti
ν (ds, dx) = l
1 + Ys− ehγ(s,Tl ),xi − 1 ν Ti (ds, dx) =: F s i (dx) ds.
l=1
(4.23)
Similar to the default-free part of the model, we have the following connection
between restricted defaultable forward measures for different settlement days:

Lemma 4.14 The defaultable Libor rate (L(t, Ti ))0≤t≤Ti is a PTi+1 -martingale
and
dPTi B(0, Ti+1 )
= (1 + δi L(t, Ti )) (0 ≤ t ≤ Ti ).
dPTi+1 B(0, Ti )
Ft

Proof: From equation (4.1) we get

(1 + δi L(t, Ti )) = (1 + δi H(t, Ti ))(1 + δi L(t, Ti ))

Yi i−1
Y
= (1 + δk H(t, Tk ))(1 + δi L(t, Ti )) (1 + δk H(t, Tk ))−1 .
k=0 k=0
4.4 Defaultable forward measures 105

Applying equations (3.17) and (4.21) yields

B(0, Ti+1 ) dPTi+1 B(0, Ti ) dPTi B(0, Ti ) dPTi

(1 + δi L(t, Ti )) =
B(0, Ti+1 ) dPTi+1 B(0, Ti+1 ) dPTi+1 Ft
B(0, Ti ) dPTi Ft
Ft

B(0, Ti ) dPTi
=
B(0, Ti+1 ) dPTi+1
Ft

and both statements are established. 

As mentioned above, (restricted) defaultable forward measures can be used

to determine prices of contingent claims. Consider a defaultable claim with a
promised payoff of X at the settlement day Ti and zero recovery upon default.
Then its time-t value is given by

πtX := 1l{τ >t} B(t, Ti )IEQTi [X1l{τ >Ti } |Gt ] (t ∈ [0, Ti ]).

Consider the general case in which X is GTi -measurable and the common case of
an FTi -measurable promised payoff X. The following proposition is a corrected
version of Bielecki and Rutkowski (2002, Proposition 15.2.3):

Proposition 4.15 Assume that the promised payoff X is GTi -measurable and
integrable with respect to QTi . Then

πtX = 1l{τ >t} B(t, Ti )IEQT [X|Gt ] = B 0 (t, Ti )IEQT [X|Gt ].

i i

If X is FTi -measurable, then

πtX = 1l{τ >t} B(t, Ti )IEPT [X|Ft ] = B 0 (t, Ti )IEPT [X|Ft ].

i i

Proof: The first statement can be proved along the lines of Bielecki and
Rutkowski (2002, Proposition 15.2.3). For the second statement observe that

πtX = 1l{τ >t} B(t, Ti )IEQTi [X1l{τ >Ti } |Gt ]

IEQTi [X1l{τ >Ti } |Ft ]
= 1l{τ >t} B(t, Ti )
QTi {τ > t|Ft }
IEQTi [X1l{τ >Ti } |Ft ]
= 1l{τ >t} B(t, Ti )
QTi {τ > Ti |Ft }
IEPTi [XQTi {τ > Ti |FTi }|Ft ]
= 1l{τ >t} B(t, Ti )
QTi {τ > Ti |Ft }
= 1l{τ >t} B(t, Ti )IEPT [X|Ft ].
i

We used Bielecki and Rutkowski (2002, (5.2)) for the second equality, equation
(4.7) for the third and the abstract Bayes rule for the last equality. 
106 The Lévy Libor model with default risk

4.5 Recovery rules and bond prices

In the previous sections we specified the evolution of (ratios of pre-default val-
ues of) defaultable zero coupon bonds with zero recovery. In real markets how-
ever, dafaultable bonds usually have a positive recovery. In order to adapt our
model to this fact, we have to incorporate suitable recovery rules for bonds.
An overview on different kinds of recovery rules can be found in Bielecki and
Rutkowski (2002) and Schönbucher (1999b).
For defaultable zero coupon bonds we adopt the fractional recovery of trea-
sury value scheme, i.e. in case of a default before or at maturity T of the bond,
the owner receives a fixed fraction π ∈ [0, 1) of a default-free zero coupon bond
with the same notional (which we set equal to one for simplicity) and the same
maturity. To put it differently, the owner receives an amount of π at T . Conse-
quently, the time-T value of this bond is

B π (T, T ) := 1l{τ >T } + π1l{τ ≤T } = π + (1 − π)1l{τ >T } .

Its time-t value (t ≤ T ) is thus

B π (t, T ) = πB(t, T ) + (1 − π)1l{τ >t} B(t, T ).

In default-free interest rate models, a coupon bearing bond can be con-

sidered as a portfolio of zero coupon bonds. For defaultable coupon bonds the
situation is not quite as simple. A coupon bond can still be decomposed into
a series of zero coupon bonds, but it does not make much sense to assume
the same recovery rate π for all. Schönbucher (1999a, p. 14) remarks: “The
claim of a creditor on the defaulted debtor’s assets is only determined by the
outstanding principal and accrued interest payments of the defaulted loan or
bond, any future coupon payments do not enter the consideration.” We adopt
his recovery of par scheme for coupon bearing bonds:
Assumption (recovery of par). The recovery of a defaultable coupon bond
that defaults in the time interval (Tk , Tk+1 ] is given by the recovery rate π ∈ [0, 1)
times the sum of the notional and the accrued interest over (Tk , Tk+1 ]. It is paid
at Tk+1 .

Note that this assumption restricts recovery payments to the tenor dates. This
restriction is not strong for a number of reasons. We refer to Schönbucher
(1999a, Section 6.2) for a discussion.
Let us denote by eXk (t) the time-t value of receiving an amount of X at Tk+1
if and only if a default occurred in the time interval (Tk , Tk+1 ].

Lemma 4.16 Let X be FTk -measurable. Then, for t ≤ Tk

eX
k (t) = 1l{τ >t} B(t, Tk+1 )δk IEPT [XH(Tk , Tk )|Ft ].
k+1

Proof: We have

eX
k (Tk+1 ) = X1l{τ >Tk } − X1l{τ >Tk+1 } .
4.5 Recovery rules and bond prices 107

Receiving an amount of X1l{τ >Tk } at Tk+1 is equivalent to receiving an amount

of X1l{τ >Tk } B(Tk , Tk+1 ) at Tk . Combining this fact with proposition 4.15 yields
for t ≤ Tk
 
eX
k (t) = 1l{τ >t} B(t, Tk )I
E PT [XB(T , T
k k+1 )|F t ] − B(t, Tk+1 )I
E PT [X|F t ]
k k+1

= 1l{τ >t} B(t, Tk+1 )

 
× IEPT [(1 + δk L(Tk , Tk ))XB(Tk , Tk+1 )|Ft ] − IEPT [X|Ft ]
k+1 k+1

= 1l{τ >t} B(t, Tk+1 )δk IEPT [XH(Tk , Tk )|Ft ].

k+1

The second equality follows from the abstract Bayes rule, the third follows by
using equation (4.1). 

With the help of the preceding lemma we can deduce the time-0 price of a
defaultable coupon bond with m coupons of c that are promised to be paid at
the dates T1 , . . . , Tm as
m−1
X m−1
X
π
Bfixed (0; c, m):=B(0, Tm ) + cB(0, Tk+1 ) + π(1 + c)e1k (0)
k=0 k=0
m−1
X  
= B(0, Tm ) + B(0, Tk+1 ) c + π(1 + c)δk IEPT [H(Tk , Tk )] .
k+1
k=0

Similarly, the price of a defaultable floating coupon bond that pays an interest
rate composed of the default-free Libor rate plus a constant spread x can be
obtained. Suppose that the bond has m coupons, i.e. the bondholder is promised
to receive an amount of δk (L(Tk , Tk ) + x) at the dates Tk+1 for 0 ≤ k ≤ m − 1,
then its time-0 price is given by (using proposition 4.15)
m−1
X  
π
Bfloating (0; x, m) := B(0, Tm ) + δk B(0, Tk+1 ) x + IEPT [L(Tk , Tk )]
k+1
k=0
m−1  
L(T ,T )
X
+ π (1 + δk x)e1k (0) + δk ek k k (0)
k=0
m−1
X 
= B(0, Tm ) + δk B(0, Tk+1 ) x + IEPT [L(Tk , Tk )]
k+1
k=0
+ π(1 + δk x)IEPT [H(Tk , Tk )]
k+1

+ πδk IEPT [H(Tk , Tk )L(Tk , Tk )] .
k+1

Consequently, in order to price defaultable coupon bonds we need to evaluate

IEPT [H(Tk , Tk )], IEPT [L(Tk , Tk )] and IEPT [H(Tk , Tk )L(Tk , Tk )]:
k+1 k+1 k+1

Let us use the abbreviations

δi L(t, Ti ) δi H(t, Ti )
Vti := and Yti := .
1 + δi L(t, Ti ) 1 + δi H(t, Ti )
108 The Lévy Libor model with default risk

Combining the equations (4.3), (4.15), (3.13), (3.15), (4.22), and (4.23) yields

 Zt Zt
H √ T
H(t, Tk ) = H(0, Tk ) exp b (s, Tk , Tk+1 ) ds + cs γ(s, Tk ) dW s k+1
0 0
Zt Z 
Tk+1
+ hγ(s, Tk ), xi(µ − ν )(ds, dx) ,
0 Rd

where

X Ys− l Vk k−1
H 1 s−
b (s, Tk , Tk+1 ) = − hγ(s, Tk ), cs γ(s, Tk )i + k
hγ(s, Tl ), cs λ(s, Tk )i
2 Y s−
l=1
Z  
T
− ehγ(s,Tk ),xi − 1 − hγ(s, Tk ), xi F s k+1 (dx)
Rd
Z k 
Vs−   
hλ(s,Tk ),xi k hγ(s,Tk ),xi
+ k
e − 1 1 + Y s− e − 1
Ys−
Rd
k−1
!
Y  
T
× 1 + Ys− l
ehγ(s,Tl ),xi − 1 − 1 F s k+1 (dx)
l=1

as well as

 Zt Zt
L √ T
L(t, Tk ) = L(0, Tk ) exp b (s, Tk , Tk+1 ) ds + cs λ(s, Tk ) dW s k+1
0 0
Zt Z 
Tk+1
+ hλ(s, Tk ), xi(µ − ν )(ds, dx)
0 Rd

with

L
b (s, Tk , Tk+1 ) =
k
1 X
l
− hλ(s, Tk ), cs λ(s, Tk )i − Ys− hγ(s, Tl ), cs λ(s, Tk )i
2
l=1
Z  
T
− ehλ(s,Tk ),xi − 1 − hλ(s, Tk ), xi F s k+1 (dx)
Rd
k 
Z  !
 Y  
T
− ehλ(s,Tk ),xi − 1 l
1 + Ys− ehγ(s,Tl ),xi − 1 − 1 F s k+1 (dx).
l=1
Rd
4.5 Recovery rules and bond prices 109

Making use of Kallsen and Shiryaev (2002, Lemma 2.6) we get

H(t, Tk ) = (4.24)
Zt X
k−1 l Vk
Ys− s−
H(0, Tk ) exp k
hγ(s, Tl ), cs λ(s, Tk )i ds
l=1
Ys−
0
Zt k 
Vs−
Z   
hλ(s,Tk ),xi k hγ(s,Tk ),xi
+ k
e − 1 1 + Ys− e − 1
Ys−
0 Rd
k−1
! !
Y  
l hγ(s,Tl ),xi Tk+1
× 1 + Ys− e −1 −1 ν (ds, dx)
l=1
 Z• Z• Z  
√ T hγ(s,Tk ),xi

Tk+1
×Et cs γ(s, Tk ) dW s k+1 + e − 1 (µ − ν )(ds, dx) .
0 0 Rd

To obtain an expression for IEPT [H(Tk , Tk )] we approximate the stochastic

k+1
terms andi
Vs− by their deterministic initial values V0i and Y0i . Similar ap-
i
Ys−
proximations have been used by Brace, Gatarek, and Musiela (1997), Rebonato
(1998), Schönbucher (1999a), and Schlögl (2002). This yields

Zt X
k−1 l k
Y V 0 0
IEPT [H(Tk , Tk )] ≈ H(0, Tk ) exp hγ(s, Tl ), cs λ(s, Tk )i ds
k+1
l=1
Y0k
0
Zt
V0k  hλ(s,Tk ),xi
Z   
k hγ(s,Tk ),xi
+ e − 1 1 + Y0 e − 1
Y0k
0 Rd
k−1
! !
Y  
× 1 + Y0l ehγ(s,Tl ),xi − 1 − 1 νeTk+1 (ds, dx) ,
l=1

where νeTk+1 is an approximation for ν Tk+1 given by

k 
Y  −1
νeTk+1 (ds, dx) = 1 + Y0l ehγ(s,Tl ),xi − 1 (4.25)
l=1
n−1
Y    ∗
× 1 + V0l ehλ(s,Tl ),xi − 1 ν T (ds, dx).
l=k+1
110 The Lévy Libor model with default risk

Proceeding similarly yields

Zt X
k
IEPT [L(Tk , Tk )] ≈ L(0, Tk ) exp − Y0l hγ(s, Tl ), cs λ(s, Tk )i ds
k+1
0 l=1

Zt Z  
− ehλ(s,Tk ),xi − 1
0 Rd
k 
! !
Y  
hγ(s,Tl ),xi
× 1+ Y0l e − 1 − 1 νeTk+1
(ds, dx) .
l=1

To obtain IEPT [L(Tk , Tk )H(Tk , Tk )] observe that, since L(·, Tk ) is a PTk+1 -

k+1
martingale,
IEPT [1 + δk L(Tk , Tk )] = 1 + δk L(0, Tk ).
k+1

On the other hand, by equation (4.1),

IEPT [1 + δk L(Tk , Tk )] = IEPT [(1 + δk L(Tk , Tk ))(1 + δk H(Tk , Tk ))]

k+1 k+1

= 1 + δk IEPT [L(Tk , Tk )] + δk IEPT [H(Tk , Tk )]

k+1 k+1

+ (δk )2 IEPT [L(Tk , Tk )H(Tk , Tk )].

k+1

Consequently,

IEPT [L(Tk , Tk )H(Tk , Tk )] =

k+1
1  
L(0, Tk ) − IEPT [L(Tk , Tk )] − IEPT [H(Tk , Tk )] .
δk k+1 k+1

4.6 Credit-sensitive swap contracts

The market for credit derivatives has increased enormously in volume since the
first of these contracts have been introduced in the early 1990s. Their success
is due to the fact that they allow to transfer credit risk from one party to an-
other and therewith to manage the risk exposure. Long before 1990 products
with credit derivative-like features (as e.g. letters of credit or bond insurances)
have been traded. However, in contrast to the modern products, these contracts
did not allow to trade the credit risk protection separately from the risky un-
derlying. There are many publications describing various credit derivatives in
detail, among which are Schönbucher (2000), Bielecki and Rutkowski (2002)
and Schmid (2004). Information about the size of the credit derivatives’ market
as well as on the market share that different products have can also be found
in Schmid (2004).
The aim of this and the following section is to derive valuation formulae, in
our model framework, for some of the most popular and heavily traded credit
derivatives. Credit-sensitive swaps are considered now and credit options in
the next section. We use the notational convention that the credit derivative
4.6 Credit-sensitive swap contracts 111

contract is signed between two parties A (who will usually receive a payment
if a default occurs) and B (who pays in case of a default). The reference entity
(e.g. a corporate bond) is issued by a third party C.
If credit derivatives are traded over-the-counter, each party of the contract
is exposed to the risk that the other party cannot fulfill its obligations. In the
following, we assume that this counterparty risk can be neglected, i.e. only the
risk that the reference entity defaults is considered.

4.6.1 Credit default swaps

Credit default swaps can be used to insure defaultable assets against default.
The protection buyer A agrees to pay a fixed amount to the protection seller B
periodically until a pre-specified credit event (as e.g. the default of a bond issued
by a reference party C) occurs or the contract terminates. In turn, B promises to
make a specified payment to A that covers his loss if the credit event happens.
There are various types of default swaps differing in the specification of the
credit event as well as in the specification of the default payment.
Let us consider a standard default swap with maturity date Tm whose credit
event is the default of a coupon bond issued by C. The default payment is chosen
such that it covers the loss of A. More precisely, A receives an amount of

1 − π(1 + c) (fixed coupon bond)

or
1 − π(1 + δk (L(Tk , Tk ) + x)) (floating coupon bond)
at Tk+1 if a default happens in (Tk , Tk+1 ] for k ∈ {0, . . . , m − 1}. For this
protection A pays a fee s at the dates T0 , . . . , Tm−1 until default. Our goal is to
determine the default swap rate, i.e. the level of s that makes the initial value
of the contract equal to zero.
The time-0 value of the fee payments is
m
X
s B(0, Tk−1 ).
k=1

The initial value of the default payment equals

m
X
(1 − π(1 + c))e1k−1 (0)
k=1

for an underlying fixed coupon bond and

m  
L(T ,Tk−1 )
X
(1 − π(1 + δk−1 x))e1k−1 (0) − πδk−1 ek−1k−1 (0)
k=1

for a floating coupon bond. Consequently, the default swap rates are
m
1 − π(1 + c) X  
sfixed = Pm B(0, Tk )δk−1 IEPT [H(Tk−1 , Tk−1 )]
k=1 B(0, Tk−1 ) k=1
k
112 The Lévy Libor model with default risk

and
m 
1 X 
sfloating = Pm B(0, Tk )δk−1 (1 − π(1 + δk−1 x))
k=1 B(0, Tk−1 ) k=1

× IEPT [H(Tk−1 , Tk−1 )]−πδk−1 IEPT [H(Tk−1 , Tk−1 )L(Tk−1 , Tk−1 )] .
k k

The expectations in these equations can be obtained as in the previous section.

4.6.2 Total rate of return swaps

Total rate of return swaps (sometimes also called total return swaps) belong
to the class of synthetic securitizations. The total return of some reference
entity (e.g. a defaultable coupon bond, a basket of assets, etc.) is exchanged for
periodic fixed or floating payments.
As an example, let us consider a total return swap with maturity date Tm
and a fixed coupon bond issued by some reference party C that matures at TM
(m ≤ M ≤ n) as reference entity. The payer party A agrees to pay the total
return of the bond to the receiver party B in return for periodic fixed payments
during the lifetime of the contract. The total return consists of the coupons as
well as of the change in the value of the bond, i.e. the difference between the
bond’s price at T0 and maturity date of the contract or time of default of the
bond (whatever comes first). The change in the bond’s value can (and in case of
a default before the swap terminates usually will) be negative. If the reference
entity defaults, the swap contract terminates. Note that B carries the price risk
(including credit risk) of the reference coupon bond. From his point of view,
the swap is similar to a synthetic purchase of the bond.
The payment streams of this swap are as follows:

• If no default occurs in (Tk , Tk+1 ] (0 ≤ k ≤ m − 1), B receives an amount

of (c − s) at Tk+1 , where s is the fixed periodic payment of B and c is the
coupon of the underlying bond.

• If a default occurs in (Tk , Tk+1 ] (0 ≤ k ≤ m − 1), B receives an amount

π
of π(1 + c) − Bfixed (0; c, M ) at Tk+1 and the swap then terminates.

π
• If no default occurs until Tm , B receives an amount of Bfixed (Tm ; c, M ) −
π
Bfixed (0; c, M ) at Tm .

From B’s point of view, the initial value of the contract is

m
X m
X
π
(c − s) B(0, Tk ) + (π(1 + c) − Bfixed (0; c, M )) e1k−1 (0)
k=1 k=1
+v0π (m, M, c) − B(0, Tm )Bfixed
π
(0; c, M ).

Here, v0π (m, M, c) denotes the value at time-0 of receiving, at Tm , an amount

4.6 Credit-sensitive swap contracts 113

π
of Bfixed (Tm ; c, M ) if no default occurs until Tm . Note that

π
1l{τ >Tm } Bfixed (Tm ; c, M ) = 1l{τ >Tm } B(Tm , TM )
M
X −1  
+ B(Tm , Tk+1 ) c + π(1 + c)δk IEPT [H(Tk , Tk )|FTm ] .
k+1
k=m

Hence, by proposition 4.15,


v0π (m, M, c) = B(0, Tm )IEPT B(Tm , TM )
m

M
X −1  
+ B(Tm , Tk+1 ) c + π(1 + c)δk IEPT [H(Tk , Tk )|FTm ] .
k+1
k=m

Using lemma 4.14 we get for k ∈ {m, . . . , M − 1}

" k #
B(0, Tk+1 ) Y 1 + δl L(0, Tl )
IEPT [B(Tm , Tk+1 )] = IE
m B(0, Tm ) PTm l=m 1 + δl L(Tm , Tl )
B(0, Tk+1 ) B(0, Tk+1 )
= IEPT [1] =
B(0, Tm ) k+1 B(0, Tm )
and similarly
  B(0, Tk+1 )
IEPT B(Tm , Tk+1 )IEPT [H(Tk , Tk )|FTm ] = IE [H(Tk , Tk )].
m k+1 B(0, Tm ) PTk+1
Thus,
M
X −1  
v0π (m, M, c) = B(0, TM ) + B(0, Tk+1 ) c + π(1 + c)δk IEPT [H(Tk , Tk )] .
k+1
k=m

We can determine the fixed periodic payment s that makes the initial value of
the contract equal to zero as
m
!−1  m
X X
π
s = c− B(0, Tk ) (Bfixed (0; c, M ) − π(1 + c)) e1k−1 (0)
k=1 k=1

π
+B(0, Tm )Bfixed (0; c, M ) − v0π (m, M, c) .

Similar formulae can be derived for the cases that the reference entity is a
floating coupon bond or the periodic payment is floating.

4.6.3 Asset swaps

An asset swap package or asset swap is a combination of a fixed coupon bond
issued by some reference party C and a fixed-for-floating interest rate swap.
Holding an asset swap is thus similar to a position in a floating coupon bond
issued by C. However, the interest rate swap usually remains in force even if
the bond defaults. For more details and a pricing formula that can be employed
in our model framework we refer to Schönbucher (1999a).
114 The Lévy Libor model with default risk

4.7 Credit options

The term credit option refers to calls and puts on floating rate or fixed rate
bonds issued by corporations as well as to options on asset swap packages,
credit default swaps and on many other credit derivatives. The purpose of this
section is to price some popular credit options within our model framework
under the following restriction on the volatility functions:
Assumption ([Link]). The volatility structures factorize in the following
way: for i ∈ {1, . . . , n − 1}

λ(s, Ti ) = λi σ(s) and γ(s, Ti ) = γi σ(s) (0 ≤ s ≤ Ti )

where λi and γi are positive constants and where σ : [0, T ∗ ] → Rd+ does not
depend on i.

This condition allows us to derive approximate pricing formulae that can numer-
ically be evaluated fast. As in the previous section we neglect the counterparty
risk.

4.7.1 Options on defaultable bonds

Options on defaultable bonds are indeed traded, as Schmid (2004, p. 147) re-
marks: “For quite a big number of liquid bonds, such as some of the Latin Amer-
ican Brady bonds and bonds of large US corporates, there is a well-developed
bond option market”. To keep formulae (relatively) simple, we examine Euro-
pean call options on defaultable zero coupon bonds. The employed techniques
can also be used to price options on coupon bearing bonds.
Let us consider a call with maturity Ti and strike K ∈ (0, 1) on a defaultable
zero coupon bond with maturity Tm (i < m ≤ n). The payoff at maturity is
either 1l{τ >Ti } (B π (Ti , Tm ) − K)+ or (B π (Ti , Tm ) − K)+ depending on whether
or not the option is knocked out at default. In the following, we only consider
the situation that the option is knocked out at default. The other case can be
treated similarly.
The time-Ti payoff of the option is given by

πTCO
i
(K, Ti , Tm ) := 1l{τ >Ti } (B π (Ti , Tm ) − K)+

+
= 1l{τ >Ti } πB(Ti , Tm ) + (1 − π)B(Ti , Tm ) − K
 m−1
Y
= 1l{τ >Ti } π (1 + δl L(Ti , Tl ))−1
l=i
m−1
Y +
−1
+ (1 − π) (1 + δl L(Ti , Tl )) −K .
l=i

To price the call we use Laplace transform methods and derive a convolution
4.7 Credit options 115

representation of the option value first. Proposition 4.15 yields

π0CO := π0CO (K, Ti , Tm )

 m−1Y
= B(0, Ti )IEPT π (1 + δl L(Ti , Tl ))−1
i
l=i
m−1
Y + 
−1
+ (1 − π) (1 + δl L(Ti , Tl )) −K
l=i
 m−1
Y  m−1
Y
= B(0, Tm )IEPT (1 + δl L(Ti , Tl )) π (1 + δl L(Ti , Tl ))−1
m
l=i l=i
m−1
Y + 
−1
+ (1 − π) (1 + δl L(Ti , Tl )) −K
l=i
 m−1
Y
= B(0, Tm )IEPT π (1 + δl H(Ti , Tl )) + (1 − π)
m
l=i
m−1
Y + 
−K (1 + δl L(Ti , Tl ))(1 + δl H(Ti , Tl )) .
l=i

Combining the equations (3.13), (3.15), (4.3), and (4.15) with (3.18) and (3.20)
δi L(t,Ti ) δi H(t,Ti )
and again using the abbreviations Vti := 1+δ i L(t,Ti )
and Yti := 1+δ i H(t,Ti )
yields
for k ∈ {1, . . . , n − 1}

 Zt Zt 
L ∗ ∗
L(t, Tk ) = L(0, Tk ) exp b (s, Tk , T ) ds + λ(s, Tk ) dLTs
0 0

with

bL (s, Tk , T ∗ ) =
n−1
1 X j
− hλ(s, Tk ), cs λ(s, Tk )i − Vs− hλ(s, Tj ), cs λ(s, Tk )i
2
j=k+1
 
Z   n−1
Y ∗
−  ehλ(s,Tk ),xi − 1 β(s, x, Tj , Tj+1 ) − hλ(s, Tk ), xi FsT (dx)
j=k+1
Rd

and

 Zt Zt 
H ∗ ∗
H(t, Tk ) = H(0, Tk ) exp b (s, Tk , T ) ds + γ(s, Tk ) dLTs
0 0
116 The Lévy Libor model with default risk

with

bH (s, Tk , T ∗ ) = (4.26)
k
X j 1
Ys− hγ(s, Tj ), cs γ(s, Tk )i − hγ(s, Tk ), cs γ(s, Tk )i
2
j=1
k−1 n−1
!
j k
X Ys− Vs− X j
+ k
hγ(s, T j ), c s λ(s, T k )i − Vs− hλ(s, Tj ), cs γ(s, Tk )i
j=1
Ys− j=k+1

Qn−1 
ehγ(s,Tk ),xi − 1 β(s, x, T , T )
Z
j=k+1 j j+1 ∗
−  Qk   − hγ(s, Tk ), xi FsT (dx)
j hγ(s,Tj ),xi − 1


Rd j=1 1 + Ys− e
Z n−1
Y
k −1
+ (Ys− ) (β(s, x, Tk , Tk+1 ) − 1) β(s, x, Tj , Tj+1 )
j=k+1
Rd
 k−1
Y  −1  ∗
j hγ(s,Tj ),xi
× 1− 1 + Ys− e −1 FsT (dx).
j=1

As in the previous section, we approximate the stochastic terms Vs− i and Y i

s−
in the drift terms bL (s, Tk , T ∗ ) and bH (s, Tk , T ∗ ) by their deterministic initial
values and call the resulting (deterministic) drifts bL ∗ H ∗
0 (s, Tk , T ) and b0 (s, Tk , T )
respectively (remember that Vs− i is also contained in β(s, x, T , T
i i+1 )). Then, due
to the assumption on the volatility structure, we get
m−1
"
Y  γ
l

CO H
π0 = B(0, Tm )IEPT π 1 + δl H(0, Tl ) exp XT + Bl
m σsum i
l=i
m−1
Y   λ 
l
+ (1 − π) − K 1 + δl L(0, Tl ) exp XTi + BlL
σsum
l=i
 +
! #
  γ
l H
1 + δl H(0, Tl ) exp XT + Bl
σsum i

with
m−1
X
σsum := (λl + γl ),
l=i
ZTi m−1 ZTi
∗ ∗
X
XTi := (λ(s, Tl ) + γ(s, Tl )) dLTs = σsum σ(s) dLTs ,
0 l=i 0

and
ZTi ZTi
∗ ∗
BlL := bL
0 (s, Tl , T ) ds, BlH := bH
0 (s, Tl , T ) ds.
0 0
Note that the option price depends on the distribution of one random variable
only, namely on the distribution of XTi with respect to PTm . Assume that this
4.7 Credit options 117

distribution possesses a Lebesgue-density ϕ (see section 2.2 for a discussion on

this assumption). The option price can then be written as a convolution, namely
Z
CO
π0 = B(0, Tm ) g(−x)ϕ(x) dx = B(0, Tm )(g ∗ ϕ)(0) (4.27)
R

with g(x) := (v(x))+ and

m−1
Y 
 γl H
v(x) := π 1 + δl H(0, Tl ) exp − x + Bl
σsum
l=i
m−1
Y   λl 
+ (1 − π) − K 1 + δl L(0, Tl ) exp − x + BlL
σsum
l=i
  γl 
H
1 + δl H(0, Tl ) exp − x + Bl .
σsum
The next step is to determine the bilateral Laplace transform of g. Observe that
we can write v as
m−1
Y   λl 
v(x) = 1 + δl L(0, Tl ) exp − x + BlL
σsum
l=i
  γl 
H
1 + δl H(0, Tl ) exp − x + Bl
σsum
m−1 −1
Y  λl
× π 1 + δl L(0, Tl ) exp − x + BlL
σsum
l=i
m−1
Y   λl 
+ (1 − π) 1 + δl L(0, Tl ) exp − x + BlL
σsum
l=i
!
  γl −1
1 + δl H(0, Tl ) exp − x + BlH −K .
σsum
v has a unique zero Z since the first m − i factors on the right-hand side are
positive and the last factor is continuous, strictly increasing, takes positive as
well as negative values, and hence has a unique zero. Consequently,
g(x) = v(x)1l[Z,∞) (x).
Note that v(x) can also be written as a finite sum of expressions of the type
“c1 exp(−c2 x)” with c1 ∈ R and c2 ∈ [0, 1]. For z ∈ C with < z > 0 we get
Z
c1 −Z(z+c2 )
e−zx (c1 e−c2 x )1l[Z,∞) (x) dx = e .
z + c2
R

Hence, the Laplace transform of g exists for all z ∈ C with < z > 0 and a closed
form expression (depending on Z) can be derived. However, since the number of
summands of the above form in v increases exponentially as (m − i) increases,
a numerical evaluation of the Laplace transform is (at least for large values
of (m − i)) more appropriate. Putting pieces together, we obtain the following
formula for the option price:
118 The Lévy Libor model with default risk

Proposition 4.17 Suppose that the distribution of XTi possesses a Lebesgue-

XT
density. Denote by M Tmi the PTm -moment generating function of XTi . Choose
XT
an R > 0 such that M Tmi (−R) < ∞. Then the price of the call is approximately
given by
Z∞  
1 XTi
π0CO (K, Ti , Tm ) = B(0, Tm ) < L[g](R + iu)M Tm (−R − iu) du, (4.28)
π
0

where L[g] denotes the bilateral Laplace transform of g. Furthermore, we have

ZTi  
XT m m
M Tmi (−R − iu) ≈ exp θs (f (s) − (R + iu)σsum σ(s)) − θs (f (s)) ds
0

with
m−1 n−1
X δl H(0, Tl ) X δl L(0, Tl )
f m (s) := − γ(s, Tl ) + λ(s, Tl ).
1 + δl H(0, Tl ) 1 + δl L(0, Tl )
l=1 l=m

Proof: Using the convolution representation (4.27) and performing Laplace

and inverse Laplace transformations, we get
Z∞  
1 XTi
π0CO = B(0, Tm ) < L[g](R + iu)M Tm (−R − iu) du.
π
0

It remains to derive an expression for the moment generating function. Observe

that
  ZTi 
XT ∗
M Tmi (z) = IEPT exp zσsum σ(s) dLTs
m
0
  ZTi 
m T∗
= IEPT ∗ ZTi exp zσsum σ(s) dLs ,
0

where (Ztm )0≤t≤Tm denotes the density process of PTm with respect to PT ∗
(which is of course a PT ∗ -martingale), given by
m−1 n−1
Y 1 + δl H(0, Tl ) Y 1 + δl L(t, Tl )
Ztm :=
1 + δl H(t, Tl ) 1 + δl L(0, Tl )
l=1 l=m
m−1
Y 1 + δl H(0, Tl )
= R 
t T ∗ + drift
l=1 1 + δ l H(0, T l ) exp 0 γ(s, T l ) dLs
R ∗

t
Y 1 + δl L(0, Tl ) exp 0 λ(s, Tl ) dLTs + drift
n−1
× .
1 + δl L(0, Tl )
l=m
4.7 Credit options 119

We use an approximation that has already been employed in the derivation of

caplet prices
 in the
 default-free Lévy Libor model. Note that 1 + ε exp(x) ≈
ε
(1 + ε) exp 1+ε x for small absolute values of x and approximate (compare
(3.24))
 Zt 
m m T∗ m
Zt ≈ exp f (s) dLs + Dt ,
0

where
m−1 n−1
m
X δl H(0, Tl ) X δl L(0, Tl )
f (s) := − γ(s, Tl ) + λ(s, Tl )
1 + δl H(0, Tl ) 1 + δl L(0, Tl )
l=1 l=m

and the drift term Dm is chosen in such a way that the PT ∗ -martingale property
of Z m is preserved, i.e.

  Zt −1 
∗
Dtm := log IEPT ∗ exp f m
(s) dLTs .
0

Hence, for z ∈ C with < z = −R we obtain

 ZTi −1
XT m ∗
M Tmi (z) ≈ IEPT ∗ exp f (s) dLTs
0
 ZTi 
m T∗
× IEPT ∗ exp (f (s) + zσsum σ(s)) dLs
0
ZTi  
m m
= exp θs (f (s) + zσsum σ(s)) − θs (f (s)) ds,
0

where the last line follows from proposition 1.9. 

4.7.2 Credit spread options

The following definition is taken from Schmid (2004, p. 148):

Definition 4.18 A credit spread call (put) option with maturity T and strike
spread K on a defaultable bond B π (·, U ) with maturity U ≥ T gives the holder
the right to buy (sell) the defaultable bond at time T at a price that corresponds
to a yield spread of K above the yield of an otherwise identical non-defaultable
bond B(·, U ).

Let us consider a call that is knocked out at default with maturity Ti and strike
spread K on the defaultable bond B π (·, Tm ) (i < m ≤ n). Its time-Ti value is
120 The Lévy Libor model with default risk

given by
 +
−(Tm −Ti )K
πTCSO
i
(K, T i , T m ) := 1l{τ >Ti } B π
(T i , T m ) − e B(T i , T m )
 +
= 1l{τ >Ti } (1 − π)B(Ti , Tm ) − (e−(Tm −Ti )K − π)B(Ti , Tm )

In the following, only the case π < e−(Tm −Ti )K is considered. In any other case,
the call will always be exercised (and is therefore no real option). Proceeding
similarly as in the previous section we get

π0CSO (K, Ti , Tm ) = B(0, Tm )(g ∗ ϕ)(0)

with g(x) := (v(x))+ and

v(x) :=
m−1
Y 
−(Tm −Ti )K
 γl
(1 − π) − (e − π) 1 + δl H(0, Tl ) exp − x+ BlH .
σsum
l=i

Here, ϕ, σsum , and BlH are defined as in section 4.7.1. Since v is continuous,
strictly increasing and takes negative as well as positive values it has a unique
zero. We can conclude, as in the previous section, that the bilateral Laplace
transform of g exists for all z ∈ C with < z > 0. By applying exactly the same
arguments as before we arrive at formula (4.28) for the price of the call. Hence,
the only difference in the pricing formulae for a call on a defaultable bond and
a credit spread call lies in the different Laplace transforms L[g].

4.7.3 Credit default swaptions

A credit default swaption gives its holder the right to enter a credit default swap
at some pre-specified time and swap rate. These options are often embedded in
other credit derivatives (e.g. as an extension option in a credit default swap).
For more details we refer to Schönbucher (1999a).
Let us consider a credit default swaption that is knocked out at default
with strike rate S and maturity Ti on a default swap that terminates at Tm
(i < m ≤ n) with an underlying fixed coupon bond. Its time-Ti value is

m−1
!
X
πTCDS
i
(S, Ti , Tm ) := 1l{τ >Ti } +
(s(Ti ; Ti , Tm ) − S) B(Ti , Tk ) ,
k=i

where s(t; Ti , Tm ) denotes the forward default swap rate at time t. Note that

m−1
X
1l{τ >Ti } s(Ti , Ti , Tm ) B(Ti , Tk ) =
k=i
m−1
X 
1l{τ >Ti } (1 − π(1 + c)) B(Ti , Tk+1 )δk IEPT [H(Tk , Tk )|FTi ] .
k+1
k=i
4.7 Credit options 121

As before, we approximate the stochastic terms Vs− i and Y i in (4.24) by their

s−
deterministic initial values V0i and Y0i and obtain
IEPT [H(Tk , Tk )|FTi ] ≈ C i,k H(Ti , Tk )
k+1

with
ZTk X
k−1
i,k Y0l V0k
C := exp hγ(s, Tl ), cs λ(s, Tk )i ds
Y0k
Ti l=1
ZTk Z
V0k  hλ(s,Tk ),xi 
k

hγ(s,Tk ),xi

+ e − 1 1 + Y0 e − 1
Y0k
Ti Rd
k−1
! !
Y  
× 1 + Y0l ehγ(s,Tl ),xi − 1 − 1 νeTk+1 (ds, dx)
l=1

and νeTk+1 given by (4.25). By proposition 4.15

π0CDS := π0CDS (S, Ti , Tm )
m−1
"
X 
= B(0, Ti )IEPT (1 − π(1 + c)) B(Ti , Tk+1 )δk C i,k H(Ti , Tk )
i
k=i
m−1
X + #
−S B(Ti , Tk )
k=i
"
(1 − π(1 + c))δm−1 C i,m−1 H(Ti , Tm−1 )
= B(0, Ti )IEPT Qm−1
l=i (1 + δl L(Ti , Tl ))(1 + δl H(Ti , Tl ))
i

m−2
!+ #
X (1 − π(1 + c))δk C i,k H(Ti , Tk ) − S
+ Qk −S .
k=i l=i (1 + δl L(Ti , Tl ))(1 + δl H(Ti , Tl ))

We proceed similarly as in the derivation of the price for a call on a defaultable

bond and define σsum , XTi , BlH , and BlL as before. Assuming that the distribu-
tion of XTi with respect to PTi possesses a Lebesgue-density ϕ we obtain
π0CDS = B(0, Ti )(g ∗ ϕ)(0)
with g(x) := (v(x))+ and
(1 − π(1 + c))δm−1 C i,m−1 H(0, Tm−1 ) exp − γσm−1 H


sum
x + Bm−1
v(x) := Qm−1 λl
x + BlL



l=i 1 + δl L(0, Tl ) exp − σsum
1
× Qm−1 γl
x + BlH



l=i 1 + δl H(0, Tl ) exp − σsum
X (1 − π(1 + c))δk C i,k H(0, Tk ) exp − σγk x + BkH − S
m−2


sum
+ Qk λl L



k=i l=i 1 + δ l L(0, T l ) exp − σsum x + Bl
!
1
× Qk γl

− S.
H
l=i 1 + δl H(0, Tl ) exp − σsum x + Bl
122 The Lévy Libor model with default risk

Note that v is continuous, tends to −S as x → −∞ and to −(m−i)S as x → ∞.

Consequently, g has compact support and the bilateral Laplace transform of g
exists for all z ∈ C. In a numerical evaluation of the Laplace transform, for large
values of m − i, we can save computational time by applying the multiplication
scheme
m−2
X k
Y
ck dl = di (ci + di+1 (ci+1 + di+1 (. . . (cm−3 + dm−2 cm−2 )))).
k=i l=i

We obtain the following formula for the price of the swaption:

Proposition 4.19 Suppose that the distribution of XTi possesses a Lebesgue-
XT
density. Denote by M Ti i the PTi -moment generating function of XTi . Choose
XT
an R ∈ R such that M Ti i (−R) < ∞ (e.g. R = 0). Then the price of the credit
default swaption is approximately given by
Z∞  
1 XT
π0CDS (K, Ti , Tm ) = B(0, Ti ) < L[g](R + iu)M Ti i (−R − iu) du, (4.29)
π
0

where L[g] denotes the bilateral Laplace transform of g. Furthermore, we have

ZTi  
XT
M Ti i (−R − iu) ≈ exp θs (f i (s) − (R + iu)σsum σ(s)) − θs (f i (s)) ds
0

with
i−1 n−1
X δl H(0, Tl ) X δl L(0, Tl )
f i (s) := − γ(s, Tl ) + λ(s, Tl ).
1 + δl H(0, Tl ) 1 + δl L(0, Tl )
l=1 l=i

Proof: As before, we use the convolution representation of the swaption price

combined with Laplace and inverse Laplace transformation methods to get
(4.29). The expression for the moment generating function follows as in the
proof of proposition 4.17. 

4.8 Conclusion
A generalization of the Libor market model with default risk by Schönbucher
(1999a) and extension of the Lévy Libor model due to Eberlein and Özkan
(2005) has been introduced, the Lévy Libor model with default risk. A pricing
formula for derivatives has been established which uses two counterparts to
forward measures, namely defaultable forward mesures and restricted defaultable
forward measures. Using this formula, we deduced approximate pricing solutions
for some popular credit derivatives. A topic for future research is the extension
of the model to rating classes.
Appendix A

Proof of Proposition 4.8

Qk−1 1
First, we derive a condition ensuring the martingale property of i=1 1+δi H(·,Ti)
b
that involves the terms bH (·, T1 , Tk ), . . ., bH (·, Tk−1 , Tk ):
Q 
k−1 1
Lemma A.1 i=1 is a QTk -martingale if
b 1+δi H(t,Ti ) 0≤t≤Tk−1

k−1
X
i H
Ys− b (s, Ti , Tk )
i=1
k−1 k−1
1X i X
i j
= − Ys− hγ(s, Ti ), cs γ(s, Ti )i + Ys− Ys− hγ(s, Ti ), cs γ(s, Tj )i
2
i=1 i,j=1
j≥i

k−1
Z X k−1
Y  −1 
i i hγ(s,Ti ),xi
FsTk (dx).


+ Ys− hγ(s, Ti ), xi −1+ 1+ Ys− e −1
i=1 i=1
Rd

δi H(s,Ti)
where Ysi :=
b
.
1+δi H(s,T
b i)

Proof: Let us denote Xti := 1 + δi H(t,

b Ti ), then using lemma 4.6, the fact that
dXti = δi dH(t,
b Ti ), and Kallsen and Shiryaev (2002, Lemma 2.6) we get

Z• Z•
i √
Xti = X0i Et i
Ys− a(s, Ti , Tk ) ds + Ys− cs γ(s, Ti ) dWsTk
0 0
Z• Z  
!
i hγ(s,Ti ),xi Tk
+ Ys− e − 1 (µ − ν )(ds, dx)
0 Rd
Zt Zt
i √
= X0i exp D(s, Ti , Tk ) ds + Ys− cs γ(s, Ti ) dWsTk
0 0
Zt Z   
!
+ i
log 1 + Ys− ehγ(s,Ti ),xi − 1 (µ − ν Tk )(ds, dx) ,
0 Rd
124 Proof of Proposition 4.8

where

D(s, Ti , Tk )
i 1 i 2
:= Ys− a(s, Ti , Tk ) − (Ys− ) hγ(s, Ti ), cs γ(s, Ti )i
Z  2
    
+ log 1 + Ys− i
ehγ(s,Ti ),xi − 1 − Ys− i
ehγ(s,Ti ),xi − 1 FsTk (dx)
Rd
(4.10) i H 1 i i 2


= Ys− b (s, Ti , Tk ) + Ys− − (Ys− ) hγ(s, Ti ), cs γ(s, Ti )i
Z  2 
  
i hγ(s,Ti ),xi
+ log 1 + Ys− e − 1 − Ys− hγ(s, Ti ), xi FsTk (dx).
i

Rd

Consequently, using Kallsen and Shiryaev (2002, Lemma 2.6) once again,

k−1
!−1
Y Xti
i=1
X0i
Zt X
k−1 Zt X
k−1
i √
= exp − D(s, Ti , Tk ) ds − (Ys− cs γ(s, Ti )) dWsTk
0 i=1 0 i=1
t  k−1 !
Z Z Y   
− log i
1 + Ys− ehγ(s,Ti ),xi − 1 (µ − ν Tk )(ds, dx)
0 Rd i=1
• Z• X
k−1
i √
Z
= Et A(s, Tk ) ds − (Ys− cs γ(s, Ti )) dWsTk (A.1)
0 0 i=1
Z• Z  k−1
Y  −1  !
i hγ(s,Ti ),xi Tk
+ 1+ Ys− e −1 − 1 (µ − ν )(ds, dx)
0 Rd i=1

with

k−1 k−1
X 1 X i j
A(s, Tk ) := − D(s, Ti , Tk ) + Ys− Ys− hγ(s, Ti ), cs γ(s, Tj )i
2
i=1 i,j=1
Z k−1
Y  −1
+ i
1 + Ys− ehγ(s,Ti ),xi − 1 −1
i=1
Rd
 k−1
Y   
!
i hγ(s,Ti ),xi
+ log 1+ Ys− e −1 FsTk (dx).
i=1

Qk−1 −1
i=1 (1 + δi H(t, Ti )) is a QTk -local martingale if A(s, Tk ) = 0 for all s. In this
b
case it is also a martingale since it is bounded by 0 and 1 and therefore of class
[D] (compare Jacod and Shiryaev (2003, I.1.47c)). Plugging in the expression
 125

for D(·, Ti , Tk ) yields

A(s, Tk ) =
k−1 k−1
X
i H 1X i
− Ys− b (s, Ti , Tk ) − Ys− hγ(s, Ti ), cs γ(s, Ti )i
2
i=1 i=1
k−1
j
X
i
+ Ys− Ys− hγ(s, Ti ), cs γ(s, Tj )i
i,j=1
j≥i

k−1
Z X k−1
Y  −1 
i i
ehγ(s,Ti ),xi − 1 FsTk (dx).


+ Ys− hγ(s, Ti ), xi − 1 + 1 + Ys−
i=1 i=1
Rd


Now assume that the drift terms bH (·, Ti , Ti+1 ) satisfy (4.15). Then, using
equations (4.11) and (3.20) we obtain

bH (s, Ti , Tk ) =
i
X j 1
Ys− hγ(s, Tj ), cs γ(s, Ti )i − hγ(s, Ti ), cs γ(s, Ti )i
2
j=1
i−1 k−1
! * !+
j
X Ys− X
+ i
hγ(s, Tj ), cs α(s, Ti , Ti+1 )i − γ(s, Ti ), cs α(s, Tl , Tl+1 )
j=1
Ys−
l=i+1

Qk−1 
ehγ(s,Ti ),xi − 1 β(s, x, T , T )
Z
l=i+1 l l+1 Tk
+ hγ(s, Ti ), xi − Qi 
  Fs (dx)
j hγ(s,T ),xi
d j=1 1 + Ys− e
j −1
R
Z k−1
Y
i −1
+ (Ys− ) (β(s, x, Ti , Ti+1 ) − 1) β(s, x, Tl , Tl+1 )
l=i+1
Rd
 i−1   −1 
j
Y
1− 1 + Ys− ehγ(s,Tj ),xi − 1 FsTk (dx)
j=1

= bH
1 (s, Ti , Tk ) + bH
2 (s, Ti , Tk ) + bH
3 (s, Ti , Tk ),

where

i
Xj 1
bH
1 (s, Ti , Tk ) := Ys− hγ(s, Tj ), cs γ(s, Ti )i − hγ(s, Ti ), cs γ(s, Ti )i,
2
j=1
i−1
!
j
X Y s−
bH
2 (s, Ti , Tk ) := i
hγ(s, Tj ), cs α(s, Ti , Ti+1 )i
j=1
Ys−
k−1
* !+
X
− γ(s, Ti ), cs α(s, Tl , Tl+1 ) ,
l=i+1
126 Proof of Proposition 4.8

and
Z 
i −1
bH
3 (s, Ti , Tk ) := (Ys− ) i
Ys− hγ(s, Ti ), xi
Rd
i 
Y  −1  k−1
j
Y
hγ(s,Tj ),xi
+ 1 + Ys− e −1 −1 β(s, x, Tl , Tl+1 )
j=1 l=i+1
i−1 
Y  −1  k−1 
j
Y
hγ(s,Tj ),xi
− 1 + Ys− e −1 −1 β(s, x, Tl , Tl+1 ) FsTk (dx).
j=1 l=i

Note that
k−1 k−1 X
k−1
j
X X
i H
Ys− b2 (s, Ti , Tk ) = 1l{j≤i−1} Ys− hγ(s, Tj ), cs α(s, Ti , Ti+1 )i
i=1 i=1 j=1
k−1 X
X k−1
i
− 1l{i≤l−1} Ys− hγ(s, Ti ), cs α(s, Tl , Tl+1 )i
i=1 l=1
=0

and
k−1
X k−1
Z X
i H i


Ys− b3 (s, Ti , Tk ) = Ys− hγ(s, Ti ), xi − 1
i=1 i=1
Rd
k−1
Y  −1 
j
+ 1 + Ys− ehγ(s,Tj ),xi − 1 FsTk (dx).
j=1

Hence, k−1 i H
P
i=1 Ys− b (s, Ti , Tk ) satisfies the prerequisite of lemma A.1 and the
claim is proven.
References

Applebaum, D. (2004). Lévy Processes and Stochastic Calculus. Cambridge

University Press.
Artzner, P. and F. Delbaen (1992). Credit risk and prepayment option.
ASTIN Bulletin 22, 81–96.
Barndorff-Nielsen, O. E. (1977). Exponentially decreasing distributions for
the logarithm of particle size. Proceedings of the Royal Society London A
353 , 401–419.
Barndorff-Nielsen, O. E. (1998). Processes of normal inverse Gaussian type.
Finance and Stochastics 2, 41–68.
Bertoin, J. (1996). Lévy processes. Cambridge University Press.
Bielecki, T. and M. Rutkowski (1999). Modelling of the defaultable term
structure: conditionally Markov approach. Working paper.
Bielecki, T. and M. Rutkowski (2000). Multiple ratings model of defaultable
term structure. Mathematical Finance 10, 125–139.
Bielecki, T. R. and M. Rutkowski (2002). Credit Risk: Modeling, Valuation
and Hedging. Springer.
Billingsley, P. (1979). Probability and Measure. Wiley.
Björk, T. (2004). Arbitrage Theory in Continuous Time (2nd ed.). Oxford
University Press.
Björk, T., G. Di Masi, Y. Kabanov, and W. Runggaldier (1997). Towards a
general theory of bond markets. Finance and Stochastics 1, 141–174.
Black, F. and J. C. Cox (1976). Valuing corporate securities: some effects of
bond indenture provisions. Journal of Finance 31, 351–367.
Black, F. and M. Scholes (1973). The pricing of options and corporate liabil-
ities. Journal of Political Economy 81, 637–654.
Brace, A., D. Gatarek, and M. Musiela (1997). The market model of interest
rate dynamics. Mathematical Finance 7, 127–155.
Brigo, D. and F. Mercurio (2001). Interest Rate Models: Theory and Practice.
Springer.
Carr, P., H. Geman, D. Madan, and M. Yor (2002). The fine structure of asset
returns: An empirical investigation. Journal of Business 75, 305–332.
128 REFERENCES

Carr, P. and D. Madan (1999). Option valuation using the fast Fourier trans-
form. The Journal of Computational Finance 2, 61–73.
Cont, R. and P. Tankov (2003). Financial Modeling with Jump Processes.
Chapman & Hall/ CRC Press.
Dellacherie, C. and P. A. Meyer (1978). Probabilities and Potential. Amster-
dam: North Holland.
Duffie, D. and K. J. Singleton (1999). Modeling term structures of defaultable
bonds. Review of Financial Studies 12, 687–720.
Eberlein, E. (2001). Application of generalized hyperbolic Lévy motions to
finance. In O. E. Barndorff-Nielsen, T. Mikosch, and S. Resnick (Eds.),
Lévy Processes: Theory and Applications, pp. 319–337. Birkhäuser Verlag.
Eberlein, E., J. Jacod, and S. Raible (2005). Lévy term structure models:
no-arbitrage and completeness. Finance and Stochastics 9, 67–88.
Eberlein, E. and W. Kluge (2004). Exact pricing formulae for caps and swap-
tions in a Lévy term structure model. Preprint No. 86, Freiburg Center
for Data Analysis and Modelling, University of Freiburg.
Eberlein, E. and W. Kluge (2005). Valuation of floating range notes in Lévy
term structure models. Mathematical Finance. Forthcoming.
Eberlein, E., W. Kluge, and A. Papapantoleon (2005). Symmetries in Lévy
term structure models. Working paper, University of Freiburg.
Eberlein, E. and F. Özkan (2003). The defaultable Lévy term structure: rat-
ings and restructuring. Mathematical Finance 13, 277–300.
Eberlein, E. and F. Özkan (2005). The Lévy Libor model. Finance and
Stochastics 9, 327–348.
Eberlein, E. and S. Raible (1999). Term structure models driven by general
Lévy processes. Mathematical Finance 9, 31–53.
Eberlein, E. and E. A. v. Hammerstein (2004). Generalized hyperbolic and
inverse Gaussian distributions: limiting cases and approximation of pro-
cesses. In R. C. Dalang, M. Dozzi, and F. Russo (Eds.), Seminar on
Stochastic Analysis, Random Fields and Applications IV, Progress in
Probability 58, pp. 221–264. Birkhäuser Verlag.
Geman, H., N. El Karoui, and J.-C. Rochet (1995). Changes of numeraire,
changes of probability measures and pricing of options. Journal of Applied
Probability 32, 443–458.
Geske, R. (1977). The valuation of corporate liabilities as compound options.
Journal of Financial and Quantitative Analysis 12, 541–552.
Heath, D., R. A. Jarrow, and A. Morton (1992). Bond pricing and the term
structure of interest rates: A new methodology for contingent claims val-
uation. Econometrica 60, 77–105.
Heston, S. (1993). A closed-form solution for options with stochastic volatility
with applications to bond and currency options. The Review of Financial
Studies 6, 327–343.
REFERENCES 129

Hubalek, F. and L. Krawczyk (1998). Simple explicit formulae for variance-

optimal hedging for processes with stationary independent increments.
Unpublished manuscript.
Jacod, J. and A. N. Shiryaev (2003). Limit Theorems for Stochastic Processes
(2nd ed.). Springer.
James, J. and N. Webber (2000). Interest Rate Modelling. Wiley.
Jamshidian, F. (1989). An exact bond option pricing formula. The Journal
of Finance 44, 205–209.
Jamshidian, F. (1997). Libor and swap market models and measures. Finance
and Stochastics 1, 293–330.
Jamshidian, F. (1999). Libor market model with semimartingales. Working
paper, NetAnalytic Ltd.
Jarrow, R. A. and S. M. Turnbull (1995). Pricing derivatives on financial
securities subject to credit risk. The Journal of Finance 50, 53–85.
Kallsen, J. and A. N. Shiryaev (2002). The cumulant process and Esscher’s
change of measure. Finance and Stochastics 6, 397–428.
Küchler, U. and E. Naumann (2003). Markovian short rates in a forward
rate model with a general class of Lévy processes. Discussion paper 6,
Sonderforschungsbereich 373, Humboldt Universität zu Berlin.
Lando, D. (1997). Modelling bonds and derivatives with default risk. In
M. Dempster and S. Pliska (Eds.), Mathematics of Derivative Securities,
pp. 369–393.
Longstaff, F. A. and E. S. Schwartz (1995). A simple approach to valuing
risky fixed and floating rate debt. The Journal of Finance 50, 789–819.
Lotz, S. and L. Schlögl (2000). Default risk in a market model. Journal of
Banking and Finance 24, 301–327.
Madan, D. and E. Seneta (1987). Chebyshev polynomial approximations and
characteristic function estimation. Journal of the Royal Statistical Society
Series B 49, 163–169.
Madan, D. and E. Seneta (1990). The variance gamma model for share market
returns. Journal of Business 63, 511–524.
Madan, D. and H. Unal (1998). Pricing the risks of default. Review of Deriva-
tives Research 2, 121–160.
Madan, D. and H. Unal (2000). A two-factor hazard-rate model for pricing
risky debt and the term structure of credit spreads. Journal of Financial
and Quantitative Analysis 35, 43–65.
Merton, R. (1974). On the pricing of corporate debt: the risk structure of
interest rates. The Journal of Finance 29, 449–470.
Miltersen, K. R., K. Sandmann, and D. Sondermann (1997). Closed form
solutions for term structure derivatives with log-normal interest rates.
The Journal of Finance 52, 409–430.
130 REFERENCES

Musiela, M. and M. Rutkowski (1997). Continuous-time term structure mod-

els: forward measure approach. Finance and Sochastics 1, 261–291.
Musiela, M. and M. Rutkowski (1998). Martingale Methods in Financial Mod-
elling (2nd ed.). Springer.
Navatte, P. and F. Quittard-Pinon (1999). The valuation of interest rate digi-
tal options and range notes revisited. European Financial Management 5,
425–440.
Nunes, J. P. V. (2004). Multifactor valuation of floating range notes. Math-
ematical Finance 14, 79–97.
Özkan, F. (2002). Lévy Processes in Credit Risk and Market Models. Ph. D.
thesis, University of Freiburg.
Protter, P. (1992). Stochastic Integration and Differential Equations (2. ed.).
Springer.
Raible, S. (2000). Lévy processes in finance: theory, numerics, and empirical
facts. Ph. D. thesis, University of Freiburg.
Rebonato, R. (1998). Interest-Rate Option Models (2. ed.). Wiley.
Reiss, M. (2003). Stochastic differential equations. Lecture Notes, Humboldt
University Berlin.
Samuelson, P. A. (1965). Rational theory of warrant pricing. Industrial Man-
agement Review 6, 13–32.
Sandmann, K., D. Sondermann, and K. R. Miltersen (1995). Closed form
term structure derivatives in a Heath–Jarrow–Morton model with log-
normal annually compounded interest rates. Proceedings of the Seventh
Annual European Futures Research Symposium Bonn, 1994 , 145–165.
Sato, K.-I. (1999). Lévy Processes and Infinitely Divisible Distributions. Cam-
bridge University Press.
Schlögl, E. (2002). A multicurrency extension of the lognormal interest rate
market models. Finance and Stochastics 6, 173–196.
Schmid, B. (2004). Credit Risk Pricing Models: Theory and Practice (2. ed.).
Springer.
Schönbucher, P. J. (1999a). A Libor market model with default risk. Working
paper, University of Bonn.
Schönbucher, P. J. (1999b). A tree implementation of a credit spread model
for credit derivatives. Working paper, University of Bonn.
Schönbucher, P. J. (2000). Credit Risk Modelling and Credit Derivatives. Ph.
D. thesis, University of Bonn.
Schoutens, W. (2003). Lévy Processes in Finance: Pricing Financial
Derivates. Wiley.
Shiryaev, A. N. (1999). Essentials of Stochastic Finance: Facts, Models, The-
ory. World Scientific Publishing.
REFERENCES 131

Turnbull, S. (1995). Interest rate digital options and range notes. Journal of
Derivatives, Fall, 92–101.
Zhou, C. (1997). A jump-diffusion approach to modeling credit risk and valu-
ing defaultable securities. Working paper, Federal Reserve Board, Wash-
ington DC.