International Review of Financial Analysis

11 (2002) 139 – 158

Applying a three-factor defaultable term structure

model to the pricing of credit default options
Bernd Schmid*, Anna Kalemanova
RiskLab GmbH, Arabellastrasse 4, D-81925 Munich, Germany

Abstract

We show how to price (digital) credit default options and swaps based on a three-factor defaultable
term structure model. Basically, we need three pieces of information: the actual nondefaultable, the
defaultable, and the zero-recovery defaultable term structure. The first two pieces can be easily obtained
from observable market data using, e.g., the Nelson – Siegel methodology, the latter can be inferred from
the other two. We illustrate the whole pricing process, from model specification and parameter
estimation to the actual credit derivatives pricing. D 2002 Elsevier Science Inc. All rights reserved.

JEL classification: G13; E43

Keywords: Defaultable term structure model; Credit derivatives; Parameter estimation

1. Introduction

In this paper, we develop pricing formulas for (binary) credit default options/swaps based
on the Schmid and Zagst hybrid three-factor defaultable term structure model (see, e.g.,
Schmid & Zagst, 2000) and real-world data from German and Italian zero-coupon bonds.

1. First, we review the underlying defaultable term structure model. As a hybrid model, it
combines ideas of structural and reduced-form models (which can actually be shown to
coincide under certain conditions (see, e.g., Duffie & Lando, 2001)). One of the factors
that determine the credit spread is the so-called uncertainty index that can be understood

* Corresponding author. Tel.: +49-89-926-94-5111; fax: +49-89-926-94-5115.

E-mail address: risklab@gmx.de (B. Schmid).

1057-5219/02/$ – see front matter D 2002 Elsevier Science Inc. All rights reserved.
PII: S 1 0 5 7 - 5 2 1 9 ( 0 2 ) 0 0 0 7 2 - 8
140 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

as an aggregation of all information on the quality of the firm currently available: The
greater the value of the uncertainty process, the lower the quality of the obligor. A similar
idea to this uncertainty process was first introduced by Cathcart and El-Jahel (1998) in the
form of a so-called signaling process. In their model, default is explicitly driven by the
signaling process that is assumed to follow a diffusion process. The second underlying
process is the nondefaultable short rate that is assumed to follow a mean-reverting square
root process. The Schmid–Zagst model differs from Cathcart and El-Jahel in several
ways: First, they assume the underlying nondefaultable short rate to follow either a mean-
reverting Hull–White process or a mean-reverting square root process with time-
dependent mean reversion level. The uncertainty or signaling process is assumed to
follow a mean-reverting square root process. They start their considerations in the ‘‘real
world’’ instead of the ‘‘risk-adjusted world’’ and use Girsanov’s theorem for the change of
measure. Finally, in addition to the nondefaultable short rate and the uncertainty index,
they directly model the short rate spread process: they assume that the spread between a
defaultable and a nondefaultable bond is considerably driven by the uncertainty index but
that there may be additional factors that influence the level of the spreads: at least the
contractual provisions, liquidity, and the premium demanded in the market for similar
instruments have a great impact on credit spreads; they can easily relate credit spreads to
business cycles by replacing the uncertainty index by some index of macroeconomic
variables without any change of their theoretical framework. Their approach seems to be
reasonable in that credit spreads provide useful observable information on data upon
which pricing models can be based. In addition, the model can be fitted directly to match
the actual process followed by interest rate credit spreads. The analytical solution obtained
for defaultable bonds can be implemented easily in practice, as all the variables and
parameters can be implied from market data.
2. Second, we review how to estimate the parameters of the underlying processes.
Therefore, we use results from Schmid (2002) using Kalman filtering techniques (see,
e.g., Harvey, 1989; Oksendal, 1998, pp. 79–106). The method is applied to zero-coupon
bond data of German and Italian Government bonds between November 1, 1999 and
October 23, 2000. For that time frame, the exchange rates between the German and
Italian currencies have already been fixed due to the membership of the two countries to
the European Monetary Union. The German and Italian bonds in the sample are all
denominated in euro. Therefore, we do not have to consider any exchange rate risk in our
study. It is assumed that there is no default risk with the German bonds and the Italian
bonds to be risky. Germany and Italy are appropriate for the analysis because the two
countries are one of the most liquid markets for government bonds within the European
Monetary Union. The daily coupon bond prices (from November 1, 1999 until October
23, 2000) are provided by Reuters Information Services.
3. Based on results of Schmid (2002), we show formulas for the pricing of (binary) credit
default options and swaps. Therefore, we need the following data for all maturities T > 0:

the default-free term structure of bond prices P(t,T),

the defaultable term structure of bond prices Pd(t,T),

the defaultable term structure of bond prices under zero-recovery Pd,zero(t,T).
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 141

The first piece of information, the default-free term structure, is easily obtained.
Possible choices are government curves or swap curves in developed economies. The
second piece is the defaultable term structure of the reference credit. Ideally, it is obtained
directly from the prices of the reference credit’s bonds. Finally, the third piece of input
data is the defaultable bond prices under zero-recovery. These prices are usually
unobservable. However, we can derive the zero-recovery term structure of bond prices
from the default-free and defaultable term structures of bond prices. Finally, we analyze
all the pricing formulas.

2. The underlying defaultable term structure model

To determine the prices of default (digital) options and swaps, it is essential to use a
defaultable term structure model. Therefore, we fix a terminal time horizon T*.
Uncertainty in the financial market is modeled by a complete probability space
(V,F,Q) and a filtration F=(Ft)0  t  T * of the three-dimensional standard Brownian
motion W ^r(t), W
^(t)0=(W ^s(t), W
^u(t)) satisfying the usual conditions (i.e., F0 is trivial and
contains all the Q-null sets of F and the filtration F is right continuous). In addition, we
assume FT * = F. The filtration F represents the arrival of information over time. We will
assume throughout that the price of a security at time t is zero if all dividend payments
after time t are zero. All prices are taken to be ex-dividend. The modeling takes already
place after measure transformation, i.e., we assume that Q is a martingale measure and all
discounted security price processes are local Q-martingales with respect R t to a suitable
rðlÞdl
numeraire. As numeraire, we choose the money market account BðtÞ ¼ e 0 , where r(t)
is the nondefaultable short rate. In the following, all processes are defined on the
probability space (V,F,Q).

Assumption 1: The dynamics of the nondefaultable short rate are given by the following
stochastic differential equation (SDE):
^
drðtÞ ¼ ½qr ðtÞ  a^ r rðtÞ dt þ sr rðtÞb dWr ðtÞ; 0  t  T *; ð1Þ

where a^r and sr>0 are positive constants, b = 0 or ½, and qr is a nonnegative valued
deterministic function.

This specification implies that the current rate r(t) is pulled toward (qr(t))/a^r with a speed of
adjustment a^r, and if b = ½, the instantaneous variance of the change in the rate is proportional
to its level.

Assumption 2: The development of the uncertainty index is given by the following SDE:
pffiffiffiffiffiffiffiffi
duðtÞ ¼ ½qu  au uðtÞ dt þ su uðtÞdWu ðtÞ; 0  t  T *; ð2Þ

where au and su>0 are positive constants and qu is a nonnegative constant.

142 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Assumption 3: The dynamics of the short rate spread (the short rate spread is supposed
to be the defaultable short rate minus the nondefaultable short rate) is given by the
following SDE:
pffiffiffiffiffiffiffi ^
dsðtÞ ¼ ½bs uðtÞ  a^ s sðtÞ dt þ ss sðtÞdWs ðtÞ; 0  t  T *; ð3Þ

where a^s, bs, and ss >0 are positive constants.

Given Assumption 1, the price of a nondefaultable zero-coupon bond is given by the

following proposition.

Proposition 4: Under Assumption 1, the time t value of a nondefaultable discount bond

with maturity T, P(t,T) = P(r,t,T), is given by (e.g., Cox, Ingersoll, & Ross, 1980; Hull &
White, 1990; Vasicek, 1977)

Pðr; t; T Þ ¼ Aðt; T ÞeBðt;TÞr ; ð4Þ

where A(t,T) and B(t,T) are defined by

8
> 1 ar ðT tÞ
^

< ar ½1  e
^ ; if b ¼ 0;
Bðt; T Þ ¼ and ð5Þ
> d ðTtÞ
: ðrÞ1eðrÞ r dr ðT tÞ ; if b ¼ 12 ;
k1 k2 e

8R 
>
>
T 1 2
s Bðt; T Þ2
 q ðtÞBðt; T Þ dt
> t 2 r
> r
>
>
>
>
>
< ¼ ln Pð0;T Þ @ lnPð0;tÞ
Pð0;tÞ  Bðt; T Þ @t if b ¼ 0;
lnAðt; TÞ ¼ ð6Þ
>
> s2
>  4ar3 ðear T  ear t Þ2 ðe2ar t  1Þ;
^ ^ ^

>
> ^
>
>
r
>
:  R T q ðtÞBðt; T Þdt;
>
if b ¼ 12 ;
t r

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ðxÞ ^
with dx ¼ a^ 2x þ 2s2x and k1=2 ¼ a2x  12 dx .

Eqs. (4)–(6) define the price of a discount bond at a future time t in terms of the short rate at a
time t and the prices of nondefaultable discount bond prices today. The latter can be calculated
from today’s term structure. The partial derivative (@ ln P(0,t))/@t in Eq. (6) can be
approximated by (lnP(0,t + e)  lnP(0,t  e))/2e where e is a small length of time. In addition,
if b = 0, the deterministic function qr(t) is given by

s2r ^

qr ðtÞ ¼ ft ð0; tÞ þ a^ r f ð0; tÞ þ ð1  e2ar t Þ; ð7Þ

2a^ r
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 143

where f(0,t) denotes the instantaneous forward rate as seen at time 0 for a contract maturing at
time t. If b = ½, qr(t) can be obtained iteratively from
Z t
 qr ðtÞBðt; tÞdt ¼ lnAð0; tÞ ð8Þ
0
by time discretization.
Under the assumption Cov(dW ^r(t), dW
^s(t)) = Cov(dW^r(t), dW
^u(t)) = Cov (dW^s(t), dW
^u(t)) = 0,
Schmid et al. (2000) generalized the result of Proposition 4 to the pricing of defaultable zero-
coupon bonds. Let T d denote the F-stopping time valued in [0,1), describing the stochastic
structure of the default time and assume fractional recovery of market value.1 Then the
following proposition holds:

Proposition 5: Assuming b = 0 and the dynamics specified by Eqs. (1)–(3), the value at
time t < t = min(T,T d), Pd(t,T) = Pd(r,s,u,t,T), of a defaultable discount bond is given by
d
ðt;T ÞsðtÞDd ðt;TÞuðtÞ
Pd ðt; T Þ ¼ Ad ðt; T ÞeBðt;T ÞrðtÞC ð9Þ

where
1  eds ðT tÞ
C d ðt; T Þ ¼ ðsÞ ðsÞ
; ð10Þ
k1  k2 eds ðTtÞ
2v0 ðt; T Þ
Dd ðt; T Þ ¼ ; ð11Þ
s2u vðt; T Þ
 
d 2qu  vðT ; T Þ 
lnA ðt; TÞ ¼ lnAðt; TÞ þ 2 ln : ð12Þ
su vðt; T Þ 
v(t,T) is defined by
j1 ds ðT tÞ  2dus fðk1 Þ
 a^ ðsÞ
a^u ðsÞ
vðt; T Þ ¼ ðs2u Þ 2ds þfðk1 Þ e F1 ðt; TÞ
j2
 a^u ðsÞ
2ds þfðk1 Þ

þ eds ðTtÞ F3 ðt; T Þ ; ð13Þ

with
ðsÞ ðsÞ ðsÞ ðsÞ
F1 ðt; T Þ ¼ Fðfðk1 Þ  fðk2 Þ; fðk1 Þ þ fðk2 Þ;
ðsÞ ðsÞ ðsÞ
1  2fðk1 Þ; k2 =k1 eds ðT tÞ Þ; ð14Þ

1
It is assumed that there is compensation in terms of equivalent defaultable bonds, which have not defaulted
yet, i.e., the recovery rate is expressed as a fraction of the market value of the defaulted bond just prior to default.
By equivalent, we mean bonds with the same maturity, quality, and face value. This model was mainly developed
by Duffie and Singleton (1997) and applied, e.g., by Schönbucher (2000).
144 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

ðsÞ ðsÞ ðsÞ ðsÞ

F3 ðt; T Þ ¼ Fðfðk1 Þ  fðk2 Þ; fðk1 Þ þ fðk2 Þ;
ðsÞ ðsÞ ðsÞ
1 þ 2fðk1 Þ; k2 =k1 eds ðTtÞ Þ; ð15Þ
and sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a^ 2u g þ 2bs s2u
fðgÞ ¼ : ð16Þ
4d2s g
In addition, j1 = j1(T,T) and j2 = j2(T,T), where

j1 ðt; T Þ ¼ z2 eds ðTtÞ F4 ðt; T Þ  x1 F3 ðt; T Þ; ð17Þ

j2 ðt; T Þ ¼ x2 F1 ðt; TÞ  z1 eds ðT tÞ F2 ðt; T Þ; ð18Þ

with ^  ðsÞ ðsÞ ðsÞ
au ðsÞ k f2 ðk2 Þ  f2 ðk1 Þ
x1=2 ¼  ds fðk1 Þ ; z1=2 ¼ ds 2ðsÞ ðsÞ
; ð19Þ
2 k1 1 2fðk1 Þ
and
ðsÞ ðsÞ ðsÞ ðsÞ
F2 ðt; T Þ ¼ Fð1  fðk1 Þ  fðk2 Þ; 1  fðk1 Þ þ fðk2 Þ;
ðsÞ ðsÞ ðsÞ
2  2fðk1 Þ; k2 =k1 eds ðTtÞ Þ; ð20Þ
ðsÞ ðsÞ ðsÞ ðsÞ
F4 ðt; T Þ ¼ Fð1 þ fðk1 Þ  fðk2 Þ; 1 þ fðk1 Þ þ fðk2 Þ;
ðsÞ ðsÞ ðsÞ
2 þ 2fðk1 Þ; k2 =k1 eds ðT tÞ Þ: ð21Þ
(x)
F(a,b,c,z) is the hypergeometric function.2 dx and k1/2 are defined in Proposition 4.
The solution for b = ½ can be found in Schmid (2002).

3. The data and parameter estimation

In the following analysis, we use observed data of credit spreads between AAA-rated
German and AA-rated Italian government bonds. We assume throughout that the German
bonds are default risk-free whereas the Italian bonds are defaultable. We consider a 12-month
time series of daily bond prices from November 1, 1999 until October 23, 2000 provided by
Reuters Information Services. All prices are denominated in euro, so we do not have to take

P ðaÞk ðbÞk zk
2
The hypergeometric function is usually denoted by F, has series expansion Fða; b; c; zÞ ¼ 1 k¼0 ðcÞk k!,
where (a)0 = 1, (a)n = a(a + 1)(a + 2). . .(a + n  1), n2N, and is the solution of the hypergeometric differential
equation z(1  z)y00+[c  + b + 1)z]y0  aby = 0. The hypergeometric function can be written as an integral
R 1(ab1
GðcÞ
Fða; b; c; zÞ ¼ GðbÞGðcbÞ 0 t ð1  tÞcb1 ð1  tzÞa dt; ðc > b > 0Þ, and is also known as the Gauss series or the
Kummer series.
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 145

care of possible currency risks involved in the credit spreads. Note further that our sample for
Germany and Italy only consists of pure discount bonds. We use a universe of 33 German and
36 Italian bonds. The range of maturities of the bonds spans from 0.36 to 20.8 years. We
apply the four-parameter approximation of Nelson and Siegel (1987) to infer the whole zero
rate curves from the bond data. To do that, we denote the zero rate at time t2[0,T] with
maturity T2[t,T*] for j2{Ger(many),I(taly)} by
ðT tÞ

1e b3

ðTtÞ
Rj ðt; T Þ ¼ bj0 ðtÞ þ ðbj1 ðtÞ þ bj2 ðtÞÞ ðT tÞ
 bj2 ðtÞe b3
; ð22Þ
b3
where
Ger I
PGer ðt; T Þ ¼ 100  eR ðt;T ÞðTtÞ
; Pd;I ðt; T Þ ¼ 100  eR ðt;T ÞðT tÞ : ð23Þ
The components in Eq. (22) determine the appropriate choices of weights that can be used to
generate zero curves of a variety of shapes. We estimate the parameters bij(t), i = 0, 1, 2,
j2{Ger,I}, and b3 for each day t from a nonlinear regression. It turns out that the best results
can be achieved by setting the constant parameter b3 equal to 3. After having estimated the
zero curves, we apply an outlier test to remove all the curves that could not be fitted well
enough by the Nelson and Siegel methodology. Fig. 1 shows the comovements of the rates
across time. According to the high correlations between Germany and Italy, the curves mainly
differ in their absolute level whereas the shapes are more or less all the same. The longer the
maturity of the German and Italian rates, the higher the level of the corresponding curves. To
further understand the structural difference between the German and Italian spot curves, we
investigate the spreads between them. The estimated spread processes for a 3-year maturity
are presented in Fig. 2.

Fig. 1. Three-year German, 3-year Italian, 6-year German, 6-year Italian, 9-year German, and 9-year Italian
zero rates.
146 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Fig. 2. Estimated credit spreads between Italian and German zero rates with a maturity of 3 years.

Then we use Kalman filtering methods to estimate the parameters of the three-factor
defaultable term structure model (for details see, e.g., Schmid, 2002). The application of
Kalman filtering methods to the parameter estimation of term structure models has the great
advantage that it allows the underlying state variables to be handled as completely
unobservable, whereas other methods often must use some short-term rates as proxies and
hereby introduce additional noise. The results of the Kalman filter estimation are summarized in
Tables 1 and 2.
For various tests on the quality of the estimations, see Schmid (2002).

4. The default option pricing formulas

4.1. Bond prices under zero-recovery

Suppose, we want to price an option on a defaultable zero-coupon (a so-called credit option).

If the option is knocked out at default of the zero-coupon bond, the buyer of the option receives
nothing. Hence, we can interpret the option as a defaultable investment with zero-recovery. We

Table 1
Estimated parameters for the German bond market (a^r = ar + lrsr2)
Parameter Estimation
qr 0.014413
ar 0.238205
sr 0.015581
lr  0.086076
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 147

Table 2
Estimated parameters for the Italian bond market (a^s = as + lsss2, a^u = au + lusu2)
Parameter Estimation Parameter Estimation
bsI 0.274800 qu 0.000031
asI 0.047687 au 0.068696
ssI 0.158324 su 0.030482
lsI  1.898668 lu  1.228143

show that we can determine the price of the option as the expected value of the promised cash
flow at maturity of the option discounted at risky discount rates. However, which risky discount
rates do we have to use? As the recovery rate of the option is different from the recovery rate of
the reference defaultable zero-coupon bond, the risky discount rates are not the same as in the
case of the pricing of defaultable zero-coupon bonds. Hence, we have to find a short rate credit
spread szero describing the credit spread process of an obligor that is equivalent to the issuer of the
zero-coupon bond (especially of the same quality) but with zero-recovery rate. Therefore, for
pricing credit derivatives such as credit options we need the following data for all maturities T>0:

the default-free term structure of bond prices P(t,T),

the defaultable term structure of bond prices Pd(t,T),

the defaultable term structure of bond prices under zero-recovery Pd,zero(t,T).

Assumption 6: The zero-recovery short rate spread szero is given by:

sðtÞ
szero ðtÞ ¼ ; 0  t  T*; ð24Þ
1  wðtÞ

where s(t) is the short rate spread process defined in Eq. (3) and 0  w(t) < 1 is the
recovery rate process.
Rt Rt
Lemma 7: M d;zero ðtÞ ¼ 0 szero ðlÞdl  0 dHl is an Ft  (local) martingale. Ht is the
default indicator function of the reference credit asset.
Proof: See Schmid (2002).3

Now it can be shown that the following proposition is valid.

Proposition 8: Let EQ[|Y |]q] < 1 for some q>1. Under the zero-recovery assumption
(i.e., under the assumption that the contingent claim is knocked out at default of the
reference credit asset) and the stochastic processes specified for r, s, u, and szero, the
price process, F(t):
 RT  
 
YLT Ft ;
Q rðlÞdl
FðtÞ ¼ E e t 0  t < t; ð25Þ

3
The proof can be ordered from the author upon request (risklab@gmx.de).
148 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

where LT = 1  HT, is given by

FðtÞ ¼ V ðtÞ1ft<tg ; ð26Þ
where the adapted continuous process V is defined by
 RT  
zero
Q  t ðrðlÞþs ðlÞÞdl 

V ðtÞ ¼ E e Y  Ft ; 0tT ð27Þ

and V(t) = 0 for t  T, i.e., if there has been no default until time t, F(t) must equal the
expected value of riskless cash flows discounted at zero-recovery risky discount rates.
Eq. (27) has a unique solution in the space consisting of every semimartingale, J, such
that EQ[supt|Jt|q] < 1 for some q>1.
Proof: See Schmid (2002).4
Remark 9: Suppose we want to price a contingent claim that promises to pay off Y
at maturity time T of the contingent claim, if the reference credit asset has not
defaulted until then and zero in case of a default. Then the time t price of the
reference credit asset, given there has been no default so far, depends on the
stochastic processor r, s, and u. But in addition, the price
R T of the contingent claim
 ðrðlÞþszero ðlÞÞdl
depends on szero, because discounting is done with e t .

In the following, we assume that w(t) is a known constant, i.e., w(t) = w for all 0  t  T*.
Then the dynamics of the zero-recovery short rate spread are given by
pffiffiffiffiffiffiffiffiffiffiffiffiffi
dszero ðtÞ ¼ ½bzero
s uðtÞ  as s
zero
ðtÞ dt þ szero
s szero ðtÞdWs ðtÞ; 0  t  T *; ð28Þ
where
bs ss
bzero
s ¼ and szero
s ¼ pffiffiffiffiffiffiffiffiffiffiffiffi : ð29Þ
1w 1w
Now we can calculate the zero-recovery zero-coupon bond prices
d;zero
ðt;T Þszero ðtÞDd;zero ðt;T ÞuðtÞ
Pd;zero ðt; T Þ ¼ Ad;zero ðt; T ÞeBðt;T ÞrðtÞC ; ð30Þ

where Ad,zero(t,T), Cd,zero(t,T), and Dd,zero(t,T) are given by the corresponding formulas for
Ad(t,T), Cd(t,T), and Dd(t,T) with bs and ss substituted by bszero and sszero, respectively.

4.2. Default digital swaps and default digital put options

Definition 10: A default digital (put) option has a pre-agreed payoff at the time of default,
e.g., expressed as a fixed percentage of the notional amount or as absolute value. The

4
The proof can be ordered from the author upon request (risklab@gmx.de).
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 149

beneficiary pays the guarantor a lump-sum fee up-front. In case of a default digital swap,
the beneficiary pays the guarantor regular fees instead of a single up-front payment.

We consider a default digital put option that pays off 1 in case of a credit event. First, we
assume that the payoff takes place at maturity of the contract. If there has been no default until
time t, the time t, price of the contract is given by
 RT  
Q  t rðlÞdl

ddp
F ðtÞ ¼ E e 1fT d Tg  Ft ð31Þ

¼ Pðt; TÞ  Pd;zero ðt; TÞ;

where Pd,zero (t,T) is the bond price under a zero-recovery assumption. If the payoff takes
place at default of the reference credit asset and there has been no default until time t, the time
t price of the contract is given by
Z T R t~  
 
dHt~  Ft :
ddp Q rðlÞdl
F ðtÞ ¼ E e t ð32Þ
t

Hence, since LTFTddp = 0,

Z R t~ RT  
T
  
Lt Ftddp ¼E Q
e t
rðlÞdl
dHt~ þ e t
rðlÞdl
LT FTddp  Ft : ð33Þ

t

Applying a corollary5 from Duffie, Schroder, and Skiadas (1996, p. 1086) to Eq. (33) yields

dðLt Ftddp Þ ¼ dHt þ rt Lt Ftddp dt þ dmt ð34Þ

for some martingale m. Together with Lemma 7, Eq. (34) becomes

dðLt Ftddp Þ ¼ szero ddp ~

t dt þ rt Lt Ft dt þ dm t ð35Þ

for some martingale m̃ with dm̃t = dmt + dMtd,zero. Finally, as

Z Z Z
ddp
Lt Ft dMt d;zero
¼ Lt Ft st dt  Lt Ftddp dHt
ddp zero
ð36Þ

hR i
5 T
Corollary 11 Let Y be a semimartingale satisfying E Q 0 jYt jdt < 1. Suppose that D is an adapted RCLL
process of integrable variation and x is a progressively measurable process. In addition, assume that the
properties of x guarantee that all processes
 are well defined. Then
 dY t =  dDt  Ytxtdt + dmt, t  T, for some
R T R u x dv RT 
martingale m, if and only if Yt ¼ EQ t e t v dDu þ e t v YT  Ft .
x dv
150 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

is a martingale by the arguments we have used before,

dðLt Ftddp Þ ¼ Lt Ftddp dHt  szero zero ddp

t dt þ ðrt þ st ÞLt Ft dt þ dmt
^
ð37Þ

ddp
¼ szero zero
t dt þ ðrt þ st ÞLt Ft dt þ dmt
^

for some martingale m̂ with dm̂ =  LtFtddpdMtd,zero. Eq. (37) holds because LtFtddpdHt = 0 By
applying, dm̃t Corollary 11 again, we get
Z R t~ RT  
T
 ðrðlÞþszero ðlÞÞdl zero ~  ðrðlÞþszero ðlÞÞdl 
Lt Ftddp ¼E Q
e t st~ dt þ e t LT FTddp  Ft : ð38Þ

t

Thus,
Z R t~  
ðrðlÞþszero ðlÞÞdl zero ~
T

Lt Ftddp ¼E Q
e t st~ dt  Ft ð39Þ
t
Z  R t~  
T
 ðrðlÞþszero ðlÞÞdl zero 
¼ E Q
e t st~  Ft dt;

t

as there is sufficient regularity to allow the interchange of expectation and integration.6 Using the
independence of r and szero, we finally get

Z  R t~  
T
~ zero 
Lt Ftddp ¼ Pðt; t ÞE e st  Ft dt~
Q  t s ðlÞdl zero 
ð40Þ
t

Z T
¼ Pd;zero ðt; t~Þf d;zero ðt; t~Þdt~;
t

which follows immediately from a well-known result (see, e.g., Sandmann & Sondermann,
1997) in nondefaultable interest rate theory, saying basically
 R t~    R t~   
Q rðlÞdl ~  
Q  t ðrðlÞdl  @ 

E e t rðt Þ  Ft ¼ E e  Ft lnPðt; T Þ  ~ ð41Þ
@T T¼t ;

h R  R t~  i hR  i R t~
6 T   ðrðlÞþszero ðlÞÞdl zero  ~ T  zero  ~  ðrðlÞþszero ðlÞÞdl zero
EQ t e t s ~
t dt  E Q
t s ~
t
dt < 1 and e t st~ is product measurable
on V  [t,T] (see, e.g., Duffie, 1992, p. 234).
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 151

and f d,zero(t,t̃ ) is the zero-recovery spread forward rate, i.e.,

@ Pd;zero ðt; t~ Þ
f d;zero ðt; t~ Þ ¼  ~ ln : ð42Þ
@t Pðt; t~ Þ

Hence, substituting in Eq. (42) yields

Z  
T
@ Pd;zero ðt; t~ Þ ~
Lt Ftddp ¼  Pd;zero ðt; t~ Þ ~ ln dt
t @t Pðt; t~ Þ

Z   ð43Þ
T
~ At~ ðt; t~ Þ
¼ Lt  P d;zero
ðt; T Þ þ P d;zero
ðt; t Þ  Bt ðt; t Þrt dt~:
~ ~
t Aðt; t~ Þ
We can conclude:

Theorem 12: If there has been no default until time t, the price of a default digital put
option paying off 1 in case of a credit event, is given by

Z  
T
~ At~ ðt; t~ Þ
Ftddp ¼1P d;zero
ðt; TÞ þ P d;zero
ðt; t Þ ~ ~
 Bt ðt; t Þrt dt~: ð44Þ
t Aðt; t~ Þ

Remark 13: 1. Eq. (43) is a closed form solution for the price of a default digital put:
(a) Pd,zero(t,t̃) is given by Eq. (30) for all t  t̃  T.
(b) At̃(t,t̃)=(@/@t̃)A(t,t̃) can be easily calculated from A(t,t̃) that is given by Eq. (6).
(c) Bt̃(t,t̃)=(@/@t̃)B(t,t̃) can be easily calculated from B(t,t̃) that is given by Eq. (5).
(d) Lt and rt are known at time t.

Fig. 3. Nondefaultable, defaultable, and zero-recovery term structures on October 23, 2000 (annual yields
compounded continuously).
152 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Fig. 4. Default digital put prices under the following assumptions (a) payoff at default of the reference credit asset
and (b) payoff at maturity of the reference credit asset.

2. Alternatively, without demanding any regularity conditions for the interchange of

expectation and integration, we can solve Eq. (39) by applying PDE techniques.

Example 14: For the pricing of a default digital put option at October 23, 2000, we
use the nondefaultable German term structure and the defaultable Italian term
structure implied by the parameter estimates in Tables 1 and 2, the nondefaultable
short rate r = 0.042434, the short rate credit spread s = 0.001296, and an uncertainty
index value of u = 0.005112. The zero-recovery defaultable term structure for that
day is calculated as described in Section 4.1 based on a recovery rate assumption of
60%. Fig. 3 shows the three different term structures.

Based on these curves, in Fig. 4 we plot the default digital put prices for different
maturities. One of the curves is generated under the assumption that the payoff happens
immediately at default, the other curve assumes that the payoff happens at the maturity
of the reference credit asset. The price for the default protection that pays immediately at
default is always more expensive than the protection that pays at maturity of the
reference credit asset.

Fig. 5. Dependence of the default digital put prices on the short rate credit spreads (for maturities of 0.5, 1, and
1.5 years).
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 153

Fig. 5 shows (for several different maturities) how the default digital put prices based on
the parameters in Tables 1 and 2 depend on the short rate credit spread. For large short
rate credit spreads the default protection is more expensive than for small short rate
credit spreads.

4.3. Default swaps and default puts

Definition 15

1. A default option (put) is a credit derivative under which one party (the beneficiary)
pays the other party (the guarantor) a fixed amount (lump-sum fee up-front). This is
in exchange for the guarantor’s promise to make a fixed or variable payment in the
event of default in one or more reference assets to cover the full loss in default.
2. A default swap is a swap under which one party (the beneficiary) pays the other party
(the guarantor) regular fees, amounts that are based on a generic interest rate, called
the default swap spread or the default swap premium. This is in exchange for the
guarantor’s promise to make a fixed or variable payment in the event of default in one
or more reference assets to cover the full loss in default.

In practice, default swap and default put contracts differ in their specific default payments:

Replacement of the difference to an equivalent default-free bond (replacement type A).

That is, the payoff at the default time Td is

PðT d ; T Þ  Pd ðT d ; TÞ ¼ PðT d ; T Þ  wðT d ÞPd ðT d ; T Þ: ð45Þ

Replacement of the difference to par (replacement type B): if termination is triggered,
there is a payoff that is the difference between the face value and the market (at default) of
a reference credit asset (cash settlement). That is, the payoff at the time Td of default is

1  Pd ðT d ; TÞ ¼ 1  wðT d ÞPd ðT d ; T Þ: ð46Þ

A default swap on a defaultable coupon bond only pays off the difference between the
postdefault coupon bond price and par. There is only principal but not coupon protection. For
the case that a default happens, Figs. 6 and 7 show the cash flows of a default put and default
swap, respectively.

The pricing of a default swap consists of two problems. At origination, there is no

exchange of cash flows, and we have to determine the default swap spread S that makes the
market value of the default swap zero. After origination, the market value of the default swap
will change due to changes in the underlying variables. So given the default swap spread S,
we have to determine the current market value of the default swap. We assume throughout
that the credit swap counterparties (beneficiary and guarantor) are default-free, and that there
is a replacement to the difference of par (which is currently market standard).
154 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Fig. 6. Cash flows of a default put with a default event at time Td. The default put up-front fee is denoted by D.
Replacement of the difference to par: 1  Pd(Td,T). The reference credit asset is a defaultable zero-coupon bond.

4.3.1. The underlying reference credit asset is a defaultable zero-coupon bond

We assume that there has been no credit event until time t.

1. Default put options and replacement of the difference to par:

The time t price is given by

Z T R u  
 
ð1  ZðuÞÞdHu  Ft
dp Q rðlÞdl
F ðtÞ ¼ E e t
t

¼ F ddp ðtÞ  Pd ðt; TÞ þ Pd;zero ðt; TÞ: ð47Þ

2. Default put options and replacement of the difference to an equivalent default-free bond:

There is a simple no-arbitrage argument that can be used to determine the value of a credit
default option on a defaultable zero-coupon bond that—in case of a credit event—replaces
the difference to an equivalent default-free bond. Assume that an investor forms a portfolio
consisting of a defaultable zero-coupon bond with maturity T and a credit default option
that pays the difference to a default-free zero-coupon bond with maturity T in case of a
default. This construction eliminates the entire default risk from the defaultable zero-
coupon bond. Hence, the value of the portfolio must always be equal to the value of a
nondefaultable zero-coupon bond with maturity T, i.e.,
F dp ðtÞ ¼ Pðt; T Þ  Pd ðt; TÞ: ð48Þ

Eq. (48) is valid in the most general assumption on the processes r, s, and u.
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 155

Fig. 7. Cash flows of a default swap with a default event at time Td. The default swap spread is denoted by S.
Replacement of the difference to par: 1  Pd(Td,T). The underlying reference credit asset is defaultable zero-
coupon bond.

3. Default swaps:

In case of a credit default swap there are regular payments S (the credit swap spread)
instead of an up-front fee F0dp. The value of paying F0dp at the origination of the credit
default put option must be the same as paying S at some predefined times ti, i = 1, . . ., m,
are all possible payment dates.
X
m
F dp ð0Þ ¼ S Pd;zero ð0; ti Þ: ð49Þ
i¼1

This is equivalent to a credit swap spread of

F dp ð0Þ
S¼X
m : ð50Þ
d;zero
P ð0; ti Þ
i¼1

Example 16: We use the same data as in Example 14 and plot the prices for default put
prices for replacement types A and B (see Fig. 8). In case of replacement type B, the
default protection is always more expensive than with replacement type A. The longer
the maturity of the default put, the higher the default probability of the reference credit
asset the more expensive the contract.

4.3.2. The underlying reference credit asset is a defaultable fixed rate bond

In the following, we want to ignore any accrued premiums that may be outstanding at the
time of default. In addition, we assume that there has been no credit event until time t.
156 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Fig. 8. Default put prices dependent on the time to maturity of the default protection contract and the replace-
ment type.

1. Default put options and replacement to the difference to par:

The reference credit asset is a coupon bond with discrete coupon payments ci occurring at
dates t < ti  T, i = 1, . . ., n, and maturity T. Then the pricing argument for the default put is
exactly the same as in the case of the zero-coupon bond, and we get

Fcdp ðtÞ ¼ F ddp ðtÞ  Pcd ðt; T Þ þ Pcd;zero ðt; T Þ; ð51Þ

where
Xn
Pcd;zero ðt; T Þ ¼ ci Pd;zero ðt; ti Þ þ Pd;zero ðt; TÞ: ð52Þ
i¼1

2. Default put options and replacement of the difference to an equivalent default-free bond:

We can use the same no-arbitrage argument as in the case of zero-coupon bonds and get

Fcdp ðtÞ ¼ Pc ðt; TÞ  Pcd ðt; TÞ: ð53Þ

Fig. 9. Default put prices dependent on the time to maturity of the default protection contract and the replace-
ment type.
 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158 157

3. Default swaps:

The credit swap spread can be calculated by the same argument as in the case of zero-
coupon bonds. Hence, if there are regular payments S at some predefined times ti, i = 1, . . .,
m, until a default happens, S is given by

Fcdp ð0Þ
S¼X
m : ð54Þ
d
P0 ð0; ti Þ
i¼1

Example 17: We use the same data as in Example 14 and assume that the reference
credit asset pays annual coupons of 5%. In case of coupon bonds, the two curves in Fig.
9 are closer together than in the case that the reference credit asset is a zero-coupon
bond (Fig. 8).

5. Summary and conclusion

We have shown how to apply a defaultable three-factor term structure model to the pricing
of (digital) credit default options and swaps. We have considered different recovery rate
treatments and have analyzed the prices deduced from the theoretical model and real-world
data. The methodology can be easily applied to the pricing of other credit derivative contracts
(see, e.g., Schmid, 2002).

References

Cathcart, L., & El-Jahel, L. (1998). Valuation of defaultable bonds. Journal of Fixed Income, 8(2), 65 – 78.
Cox, J., Ingersoll, J., & Ross, S. (1980). An analysis of variable rate loan contracts. Journal of Finance, 15,
389 – 403.
Duffie, D. (1992). Dynamic asset pricing theory. Princeton, NJ: Princeton University Press.
Duffie, D., & Lando, D. (2001). Term structures of credit spreads with incomplete accounting information.
Econometrica, 69(3), 633 – 664.
Duffie, D., Schroder, M., & Skiadas, C. (1996). Recursive valuation of defaultable securities and the timing of
resolution of uncertainty. Annals of Applied Probability, 6(4), 1075 – 1090.
Duffie, D., & Singleton, K. J. (1997). Modeling term structures of defaultable bonds. Working paper, Graduate
School of Business, Stanford University.
Harvey, A. C. (1989). Forecasting, structural time series models and the Kalman filter. Cambridge: Cambridge
University Press.
Hull, J., & White, A. (1990). Pricing interest-rate-derivative securities. Review of Financial Studies, 3(4), 573 – 592.
Nelson, C., & Siegel, A. (1987). Parsimonious modeling of yield curves. Journal of Business, 60, 473 – 489.
Oksendal, B. (1998). Stochastic differential equations—an introduction with applications. Berlin: Springer.
Sandmann, K., & Sondermann, D. (1997). A note on the stability of lognormal interest rate models and pricing of
eurodollars futures. Mathematical Finance, 7(2), 119 – 126.
Schmid, B. (2002). Pricing credit linked financial instruments. Lecture notes in economics and mathematical
systems, vol. 516. Berlin: Springer.
158 B. Schmid, A. Kalemanova / International Review of Financial Analysis 11 (2002) 139–158

Schmid, B., & Zagst, R. (2000). A three-factor defaultable term structure model. Journal of Fixed Income, 10(2),
63 – 79.
Schönbucher, P. J. (2000). Credit risk modeling and credit derivatives. PhD thesis, Rheinische Friedrich-Wil-
helms-Universität Bonn.
Vasicek, O. (1977). An equilibrium characterization of the term structures. Journal of Financial Economics, 7,
117 – 161.