Hazard Function Approach: 2.1 The Toy Model

Chapter 2
Hazard Function Approach

We provide in this chapter a detailed analysis of the relatively simple case of the reduced form
methodology, when the ow of informations available to an agent reduces to the observations of
the random time which models the default event. The focus is on the evaluation of conditional
expectations with respect to the ltration generated by a default time with the use of the hazard
function. We study hedging strategies based on CDS and/or with DZC. We also present a model with
two default times. In the following chapters, we shall study the case when an additional information
ow - formally represented by some ltration F - is present, with the use of the hazard process.
2.1 The Toy Model
We begin with the simple case where a riskless asset, with deterministic interest rate (r(s); s 0)
is the only asset available in the default-free market.
The price of a risk-free zero-coupon bond with maturity T is B(0, T) = exp
_
_
T
0
r(s)ds
_
,
whereas its time t price B(t, T) is
B(t, T)
def
= exp
_
_
T
t
r(s)ds
_
.
Default occurs at time (where is assumed to be a positive random variable with density f,
constructed on a probability space (, G, P)). We denote by F the cumulative function of the r.v.
dened as F(t) = P( t) =
_
t
0
f(s)ds and we assume that F(t) < 1 for any t < T, where T is the
maturity date (Otherwise there exists t
0
< T such that F(t
0
) = 1, and default occurs a.s. before t
0
).
We emphasize that the risk is not hedgeable. Indeed, a random payo of the form 11
{T<}
cannot
be perfectly hedged with deterministic zero-coupon bonds which are the only tradeable assets in our
model. To hedge the risk, we shall assume later on that some defaultable asset is traded, e.g., a
defaultable zero-coupon bond or a CDS (Credit Default Swap).
Remark 2.1.1 It is not dicult to generalize the study presented in what follows to the case where
does not admit a density by dealing with the right-continuous version of the cumulative function.
The case where is bounded can also be studied along the same method. We leave the details to
the reader.
2.1.1 Defaultable Zero-coupon with Payment at Maturity
A defaultable zero-coupon bond (DZC in short)- or a corporate bond- with maturity T and
rebate R paid at maturity, consists of
19
20 CHAPTER 2. HAZARD FUNCTION APPROACH
The payment of one monetary unit at time T if default has not occurred before time T, i.e., if
> T,
A payment of R monetary units, made at maturity, if T, where 0 < R < 1.
Value of the defaultable zero-coupon bond
The value of the defaultable zero-coupon bond is dened as the expectation of discounted payos
D
(R,T)
(0, T) = E(B(0, T) [11
{T<}
+R11
{T}
])
= B(0, T) E(1 (1 R)11
{T}
)
= B(0, T)[1 (1 R)F(T)] , (2.1)
where the T in the superscript for D
(R,T)
means that the recovery R is paid at maturity T. In
fact, this quantity is a net present value and is equal to the value of the default free ZC, minus the
expected loss, computed under the historical probability. Obviously, this is not a hedging price.
The time-t value depends whether or not default has happened before this time. If default has
occurred before time t, the payment of R will be made at time T, and the price of the DZC is
RB(t, T).
If the default has not yet occurred, the holder does not know when it will occur. The value
D
(R,T)
(t, T) of the DZC is the conditional expectation of the discounted payo B(t, T) [11
{T<}
+
R11
{T}
] given the information:
D
(R,T)
(t, T) = 11
{t}
B(t, T)R + 11
t<

D
(R,T)
(t, T)
where the predefault value

D
(R,T)
is dened as
D
(R,T)
(t, T) = E
_
B(t, T) (11
{T<}
+R11
{T}
)
t <
_
= B(t, T)
_
1 (1 R)P( T
t < )
_
= B(t, T)
_
1 (1 R)
P(t < T)
P(t < )
_
= B(t, T)
_
1 (1 R)
F(T) F(t)
1 F(t)
_
(2.2)
Note that the value of the DZC is discontinuous at time , unless F(T) = 1 (or R = 1). In the case
F(T) = 1, the default appears with probability one before maturity and the DZC is equivalent to a
payment of R at maturity. If R = 1, the DZC is in fact a default-free zero coupon bond.
Formula (2.2) can be read as
D
(R,T)
(t, T) = B(t, T) EDLGD DP
where the Expected Discounted Loss Given Default (EDLGD) is dened as B(t, T)(1 R)
and the conditional Default Probability (DP) is
DP =
P(t < T)
P(t < )
= P( T|t < ) .
In case the payment is a function of the default time, say R(), the value of this defaultable zero-
coupon is
D
(R,T)
(0, T) = E
_
B(0, T) 11
{T<}
+B(0, T)R()11
{T}
_
= B(0, T)
_
P(T < ) +
_
T
0
R(s)f(s)ds
_
.
2.1. THE TOY MODEL 21
If the default has not occurred before t, the predefault time-t value

D
(R,T)
(t, T) satises
D
(R,T)
(t, T) = B(t, T)E( 11
{T<}
+R()11
{T}
t < )
= B(t, T)
_
P(T < )
P(t < )
+
1
P(t < )
_
T
t
R(s)f(s)ds
_
.
To summarize,
D
(R,T)
(t, T) = 11
11
{t<}
D
(R,T)
(t, T) + 11
11
{t}
R() B(t, T) .
Hazard function
We introduce the hazard function dened by
(t) = ln(1 F(t))
and its derivative (t) =
f(t)
1 F(t)
where f(t) = F
(t), i.e.,
1 F(t) = e
(t)
= exp
_
_
t
0
(s)ds
_
= P( > t) .
The quantity (t) is the hazard rate. The interpretation of the hazard rate is the probability that
the default occurs in a small interval dt given that the default did not occur before time t
(t) = lim
h0
1
h
P( t +h| > t) .
Note that is increasing.
Then, formula (2.2) reads
D
(R,T)
(t, T) = B(t, T)
_
1 F(T)
1 F(t)
+R
F(T) F(t)
1 F(t)
_
=

D(t, T) +R(B(t, T)

D(t, T)) ,
where
D(t, T) = exp
_
_
T
t
(r +)(s)ds
_
,
is the predefault value at time t of a DZC which pays one monetary unit at maturity, if the default
did not occur before maturity. Hence, the spot rate has to be adjusted by means of a spread (equal
to ) in order to evaluate DZCs.
The dynamics of

D
(R,T)
can be easily written in terms of the function as
d
t

D
(R,T)
(t, T) = (r(t) +(t))

D
(R,T)
(t, T) dt B(t, T)(t)R(t)dt
The dynamics of D
(R,T)
will be written in the next section.
If and R are constant, the credit spread is
1
T t
ln
B(t, T)
D
(R,T)
(t, T)
=
1
T t
ln
_
1 +R(e
(Tt)
1)
_
and goes to (1 R) when t goes to T.
The quantity (t, T) =
f(t,T)
1F(t,T)
where
F(t, T) = P( T| > t)
and f(t, T) dT = P( dT| > t) is called the conditional hazard rate. One has
F(t, T) = 1 exp
_
T
t
(s, T)ds .
In our setting,
1 F(t, T) =
P( > T)
P( > t)
= exp
_
T
t
(s)ds
and (s, T) = (s).
Lemma 2.1.1 The random variable () has an exponential law with parameter 1.
Proof: The Laplace transform of () is
E(e
()
) = E((1 F())
) = E(U
) =
1
1 +
where we have used that F(), hence 1 F() has a uniform law.
Remark 2.1.2 In case is the rst jump of an inhomogeneous Poisson process with deterministic
intensity ((t), t 0) (See Appendix if needed),
f(t) = P( dt)/dt = (t) exp
_
_
t
0
(s)ds
_
= (t)e
(t)
where (t) =
_
t
0
(s)ds and P( t) = F(t) = 1 e
(t)
, hence the hazard function is equal to the
compensator of the Poisson process, i.e. (t) = (t). Conversely, if is a random time with density
f, setting (t) = ln(1 F(t)) allows us to interpret as the rst jump time of an inhomogeneous
Poisson process with intensity the derivative of .
2.1.2 Defaultable Zero-coupon with Payment at Hit
Here, a defaultable zero-coupon bond with maturity T consists of
The payment of one monetary unit at time T if default has not yet occurred,
A payment of R() monetary units, where R is a deterministic function, made at time if
T.
Value of the defaultable zero-coupon
The value of this defaultable zero-coupon bond is
D
(R)
(0, T) = E(B(0, T) 11
{T<}
+B(0, )R()11
{T}
)
= P(T < )B(0, T) +
_
T
0
B(0, s)R(s)dF(s)
= G(T)B(0, T)
_
T
0
B(0, s)R(s)dG(s) , (2.3)
where G(t) = 1 F(t) = P(t < ) is the survival probability. Obviously, if the default has occurred
before time t, the value of the DZC is null (this was not the case for payment of the rebate at matu-
rity), and D
(R)
(t, T) = 11
t<

D
(R)
(t, T) where

D
(R)
(t, T) is a deterministic function (the predefault
2.1. THE TOY MODEL 23
price). The predefault time-t value

D
(R)
(t, T) satises
B(0, t)
D
(R)
(t, T) = E(B(0, T) 11
{T<}
+B(0, )R()11
{T}
|t < )
=
P(T < )
P(t < )
B(0, T) +
1
P(t < )
_
T
t
B(0, s)R(s)dF(s) .
Hence,
B(0, t)G(t)
D
(R)
(t, T) = G(T)B(0, T)
_
T
t
B(0, s)R(s)dG(s) .
In terms of the hazard function,
D
(R)
(0, T) = e
(T)
B(0, T) +
_
T
0
B(0, s)e
(s)
R(s)d(s) . (2.4)
The time-t value

D
(R)
(t, T) satises:
B(0, t)e
(t)

D
(R)
(t, T) = e
(T)
B(0, T) +
_
T
t
B(0, s)e
(s)
R(s)d(s) .
The process t D
(R)
(t, T) admits a discontinuity at time .
A particular case
If F is dierentiable, the function =
satises f(t) = (t)e

(t)
. Then,
D
(R)
(0, T) = e
(T)
B(0, T) +
_
T
0
B(0, s)(s)e
(s)
R(s)ds , (2.5)
=

D(0, t) +
_
T
0
D(0, s)(s)R(s)ds ,
and
D(0, t)
D
(R)
(t, T) =

D(0, T) +
_
T
t
D(0, s)(s)R(s)ds
with

D(0, t) = exp
_
_
t
0
[r(s) +(s)]ds
_
. The defaultable interest rate is r + and is, as expected,
greater than r (the value of a DZC with R = 0 is smaller than the value of a default-free zero-coupon).
The dynamics of

D
(R)
(t, T) are
d
D
(R)
(t, T) = {(r(t) +(t))
D
(R)
(t, T) R(t)(t))}dt .
The dynamics of D
(R)
includes a jump at time and will be computed in a next section.
Fractional recovery of treasury value
This case corresponds to R(t) = RB(t, T).
D
(R)
(t, T) = 11
t<
_
e
T
t
(r(s)+(s)) ds
+RB(t, T)
_
T
t
ds(s)e
s
t
(u)du
_
Fractional recovery of market value
Let us assume here that the recovery is R(t) = R
D
(R)
(t, T) where R is a constant (i.e. the recovery
is RD
(R)
(, T)). The dynamics of

D
(R)
is
d
D
(R)
(t, T) = {r(t) +(t)(1 R(t))}
D
(R)
(t, T)dt ,
hence
D
(R)
(t, T) = exp
_
_
T
t
r(s)ds
_
T
t
(u)(1 R(u))du
_
.
2.1.3 Implied probabilities
If defaultable zero-coupon bonds with zero recovery are traded in the market at price D
(R,)
(t, T),
the implied survival probability is Q
such that Q
( > T| > t) =
D
(R,)
(t,T)
B(t,T)
. Of course, this
probability may dier from the historical probability. The implied hazard rate is the function
(t, T) such that
(t, T) =

T
ln
D
(R,)
(t, T)
B(t, T)
=
(T) .
In the toy model, the implied hazard rate is not very interesting. In a general setting, the aim is to
obtain, for any s a process (t, s), t s such that
D
(R,)
(t, T) = B(t, T) exp
_
T
t
(t, s)ds .
This approach will be useful when he predefault price is stochastic.
2.1.4 Spreads
A term structure of credit spreads associated with the zero-coupon bonds S(t, T) is dened as
S(t, T) =
1
T t
ln
D
(R,)
(t, T)
B(t, T)
.
In our setting, on the set { > t}
S(t, T) =
1
T t
lnQ
( > T| > t) ,
whereas S(t, T) = on the set { t}.
2.2 Toy Model and Martingales
We now present the results of the previous section in a dierent form, following closely Dellacherie
([63], page 122). We keep the same notation for the cumulative function and the hazard function,
assumed to be continuous. We denote by (H
t
, t 0) the right-continuous increasing process H
t
=
11
{t}
and by (H
t
) its natural ltration. The ltration H is the smallest ltration which makes
a stopping time. The -algebra H
t
is generated by the sets { s} for s t (or by the r.v. t)
(note that the set { > t} is an atom). A key point is that any integrable H
t
-measurable r.v. H is
of the form H = h( t) = h()11
{t}
+h(t)11
{t<}
where h is a Borel function.
We now give some elementary tools to compute the conditional expectation w.r.t. H
t
, as presented
in Bremaud [32], Dellacherie [63], Elliott [80]. Note that if the cumulative distribution function F
is continuous, then, is a H-totally inaccessible stopping time. (See Dellacherie and Meyer [67] IV,
107.)
2.2. TOY MODEL AND MARTINGALES 25
2.2.1 Key Lemma
Lemma 2.2.1 If X is any integrable, G-measurable r.v.
E(X|H
s
)11
{s<}
= 11
{s<}
E(X11
{s<}
)
P(s < )
. (2.6)
Proof: The r.v. E(X|H
s
) is H
s
-measurable. Therefore, it can be written in the form E(X|H
s
) =
h( s) = h()11
{s}
+h(s)11
{s<}
for some function h. By multiplying both members by 11
{s<}
,
and taking the expectation, we obtain
E[11
{s<}
E(X|H
s
)] = E[E(11
{s<}
X|H
s
)] = E[11
{s<}
X]
= E(h(s)11
{s<}
) = h(s)P(s < ) .
Hence, h(s) =
E(X11
{s<}
)
P(s < )
gives the desired result.
Corollary 2.2.1 Assume that Y is H
-measurable, so that Y = h() for some Borel measurable

function h : IR
+
IR. If the hazard function of is continuous then
E(Y |H
t
) = 11
{t}
h() + 11
{t<}
_

t
h(u)e
(t)(u)
d(u). (2.7)
If admits the intensity function then
E(Y |H
t
) = 11
{t}
h() + 11
{t<}
_

t
h(u)(u)e
u
t
(v) dv
du.
In particular, for any t s we have
P( > s|H
t
) = 11
{t<}
e
s
t
(v) dv
and
P(t < < s|H
t
) = 11
{t<}
_
1 e
s
t
(v) dv
_
.
Proof: It suces to write
E(h()|H
t
) = 11
{t}
h() + 11
{t<}
E(h()11
{t<}
P(t < )
= 11
{t}
h() + 11
{t<}
e
(t)
_

t
h(s)dF(s)
and to note that
dF(s) = (1 F(s))d(s) = e
(s)
d(s)
2.2.2 Some Martingales
Proposition 2.2.1 The process (M
t
, t 0) dened as
M
t
= H
t

_
t
0
dF(s)
1 F(s)
= H
t

_
t
0
(1 H
s
)
dF(s)
1 F(s)
is a H-martingale.
Proof: Let s < t. Then:
E(H
t
H
s
|H
s
) = 11
{s<}
E(11
{s<t}
|H
s
) = 11
{s<}
F(t) F(s)
1 F(s)
, (2.8)
which follows from (2.6) with X = 11
{t}
.
On the other hand, the quantity
C
def
= E
__
t
s
(1 H
u
)
dF(u)
1 F(u)
H
s
_
,
is equal to
C =
_
t
s
dF(u)
1 F(u)
E
_
11
{>u}
H
s
= 11
{>s}
_
t
s
dF(u)
1 F(u)
_
1
F(u) F(s)
1 F(s)
_
= 11
{>s}
_
F(t) F(s)
1 F(s)
_
which, from (2.8) proves the desired result.
The function
_
t
0
dF(s)
1 F(s)
= ln(1 F(t)) = (t)
is the hazard function.
From Proposition 2.2.1, we obtain the Doob-Meyer decomposition of the submartingale H
t
as M
t
+
(t ). The predictable process A
t
=
t
is called the compensator of H.
In particular, if F is dierentiable, the process
M
t
= H
t

_
t
0
(s)ds = H
t

_
t
0
(s)(1 H
s
)ds
is a martingale, where (s) =
f(s)
1 F(s)
is a deterministic non-negative function, called the intensity
of .
Proposition 2.2.2 Assume that F (and thus also ) is a continuous function. Then the process
M
t
= H
t
(t ) follows a D-martingale.
x
Proposition 2.2.3 The process L
t
def
= 11
{>t}
exp
__
t
0
(s)ds
_
is a H-martingale and
L
t
= 1
_
]0,t]
L
u
dM
u
(2.9)
In particular, for t < T,
E(11
{>T}
|H
t
) = 11
{>t}
exp
_
_
T
t
(s)ds
_
.
Proof: We shall give 3 dierent arguments, each of which constitutes a proof.
a) Since the function is deterministic, for t > s
E(L
t
|H
s
) = exp
__
t
0
(u)du
_
E(11
{t<}
|H
s
) .
From the equality (2.6)
E(11
{t<}
|H
s
) = 11
{>s}
1 F(t)
1 F(s)
= 11
{>s}
exp((t) + (s)) .
Hence,
E(L
t
|H
s
) = 11
{>s}
exp
__
s
0
(u)du
_
= L
s
.
b) Another method is to apply integration by parts formula (see Appendix 8.4.2 if needed) to the
process L
t
= (1 H
t
) exp
__
t
0
(s)ds
_
dL
t
= dH
t
exp
__
t
0
(s)ds
_
+(t) exp
__
t
0
(s)ds
_
(1 H
t
)dt
= exp
__
t
0
(s)ds
_
dM
t
.
c) A third (sophisticated) method is to note that L is the exponential martingale of M (see Ap-
pendix), i.e., the solution of the SDE
dL
t
= L
t
dM
t
, L
0
= 1.
Proposition 2.2.4 Assume that is a continuous function. Then for any (bounded) Borel mea-
surable function h : IR
+
IR, the process
M
h
t
= 11
{t}
h()
_
t
0
h(u) d(u) (2.10)
is a H-martingale.
Proof: Notice that the proof given below provides an alternative proof of the rst part of Proposition
2.2.2. We wish to establish via direct calculations the martingale property of the process M
h
given
by formula (2.10). To this end, notice that formula (2.7) in Corollary 2.2.1 gives
E
_
h()11
{t<s}
| H
t
_
= 11
{t<}
e
(t)
_
s
t
h(u)e
(u)
d(u).
On the other hand, using the same formula, we get
J := E
_
_
s
t
h(u) d(u)
_
= E
_
h()11
{t<s}
+

h(s)11
{>s}
| H
t
_
where we set

h(s) =
_
s
t
h(u) d(u). Consequently,
J = 11
{t<}
e
(t)
_
_
s
t
h(u)e
(u)
d(u) +e
(s)
h(s)
_
.
To conclude the proof, it is enough to observe that Fubinis theorem yields
_
s
t
e
(u)
_
u
t
h(v) d(v) d(u) +e
(s)
h(s)
=
_
s
t
h(u)
_
s
u
e
(v)
d(v) d(u) +e
(s)
_
s
t
h(u) d(u)
=
_
s
t
h(u)e
(u)
d(u),
as expected.
Corollary 2.2.2 Let h : IR
+
IR be a (bounded) Borel measurable function. Then the process
M
h
t
= exp
_
11
{t}
h()
_
_
t
0
(e
h(u)
1) d(u) (2.11)
is a D-martingale.
Proof: In view of the preceding result applied to e
h
1, it is enough to observe that
exp
_
11
{t}
h()
_
= 11
{t}
e
h()
+ 11
{t}
= 11
{t}
(e
h()
1) + 1 .
Proposition 2.2.5 Assume that is a continuous function. Let h : IR

+
IR be a non-negative
Borel measurable function such that the random variable h() is integrable. Then the process
M
t
= (1 + 11
t
h()) exp
_
_
t
0
h(u) d(u)
_
(2.12)
is a H-martingale.
Proof: One notes that
M
t
= exp
_
_
t
0
(1 H
u
)h(u) d(u)
_
+ 11
t
h() exp
_
_

0
(1 H
u
)h(u) d(u)
_
= exp
_
_
t
0
(1 H
u
)h(u) d(u)
_
+
_
t
0
h(u) exp
_
_
u
0
(1 H
s
)h(s) d(s)
_
dH
u
From Itos calculus,
d
M
t
= exp
_
_
t
0
(1 H
u
)h(u) d(u)
_
((1 H
t
)h(t) d(t) +h(t)dH
t
)
= h(t) exp
_
_
t
0
(1 H
u
)h(u) d(u)
_
dM
t
.
It is useful to compare with the Doleans-Dade exponential of hM (see Appendix, Section 8.4.4).
Example 2.2.1 In the case where N is an inhomogeneous Poisson process with deterministic in-
tensity and is the rst time when N jumps, let H
t
= N
t
. It is well known that N
t
_
t
0
(s)ds
is a martingale (see Appendix). Therefore, the process stopped at time is also a martingale,
i.e., H
t

_
t
0
(s)ds is a martingale. Furthermore, we have seen in Remark 2.1.2 that we can
reduce our attention to this case, since any random time can be viewed as the rst time where an
inhomogeneous Poisson process jumps.
Exercise 2.2.1 In this exercise, F is only continuous on right, and F(t) is the left limit at point
t. Prove that the process (M
t
, t 0) dened as
M
t
= H
t

_
t
0
dF(s)
1 F(s)
= H
t

_
t
0
(1 H
s
)
dF(s)
1 F(s)
is a H-martingale.
2.2.3 Pricing and Trading Defaultable Claims
This section gives an overview of basic results concerning the valuation and trading of defaultable
claims. Here, we assume that the interest r is constant.
Recovery at maturity
Let S be the price of an asset which delivers only a recovery R() at time T where R is a deterministic
function. We know that the process dM
t
= dH
t
(1H
t
)
t
dt is a martingale where (t) = f(t)/G(t),
f is the density of and G(t) = Q( > t). Therefore,
e
rt
S
t
= E
Q
_
(R()e
rT
|G
t
_
= e
rT
11
<t
R() +e
rT
11
>t
E
Q
(R()11
t<<T
)
G(t)
= e
rT
_
t
0
R(u)dH
u
+e
rT
11
>t
S
t
where

S
t
is the predefault price given by the deterministic function
S
t
=
E
Q
(R()11
t<<T
)
G(t)
=
_
T
t
R(u)f(u)du
G(t)
Hence,
d
S
t
= f(t)
_
T
t
R(u)f(u)du
G
2
(t)
dt
R(t)f(t)
G(t)
dt =

S
t
f(t)
G(t)
dt
R(t)f(t)
G(t)
dt .
It follows that
d(e
rt
S
t
) = e
rT
_
R(t)dH
t
+ (1 H
t
)
f(t)
G(t)
_
S
t
R(t)
_
dt

S
t
dH
t
_
= (e
rT
R(t) e
rt
S
t
)(dH
t
(1 H
t
)(t)dt)
= e
rt
(e
r(Tt)
R(t) S
t
) dM
t
.
In that case, the discounted price is a martingale under the risk-neutral probability, and the price
S does not vanishes (as soon as R does not)
Recovery at default time
We can now write the dynamics of a defaultable zero-coupon bond with recovery R paid at hit,
assuming that M is a martingale under the risk-neutral probability.
Proposition 2.2.6 The risk-neutral dynamics of a DZC with recovery paid at hit is
dD
(R)
(t, T) =
_
r(t)D
(R)
(t, T) R(t)(t)(1 H
t
)
_
dt

D
(R)
(t, T)dM
t
(2.13)
where M is the risk-neutral martingale M
t
= H
t

_
t
0
(1 H
s
)
s
ds.
Proof: From D
(R)
(t, T) = 11
t<

D
(R)
(t, T) = (1 H
t
)
D
(R)
(t, T) and the dynamics of

D
(R)
(t, T),
we obtain
dD
(R)
(t, T) = (1 H
t
)d
D
(R)
(t, T)

D
(R)
(t, T)dH
t
= (1 H
t
)
_
(r(t) +(t))
D
(R)
(t, T) R(t)(t)
_
dt

D
(R)
(t, T))dH
t
=
_
r(t)D
(R)
(t, T) R(t)(t)(1 H
t
)
_
dt

D
(R)
(t, T)dM
t
We emphazise that here, we are working under a risk-neutral probability. We shall see further on
how to compute the risk-neutral hazard rate from the historical one, using the Radon-Nikod ym
density. In that case, the discounted process is not a martingale under the risk-neutral probability.
The process S
t
e
rt
+
_
t
0
R(s)e
rs
(1H
s
)(s)ds is a martingale. The recovery has to be understood
as a dividend process, paid up time , at rate .
2.2.4 Representation Theorem
Proposition 2.2.7 Let h be a (bounded) Borel function. Then, the martingale M
h
t
= E(h()|H
t
)
admits the representation
E(h()|H
t
) = E(h()) +
_
t
0
(h(s)
h(s)) dM
s
,
where M
t
= H
t
(t ) and
h(t) =
1
G(t)
_

t
h(u)dG(u) =
1
G(t)
E(h()11
>t
) . (2.14)
Note that

h(t) = M
h
t
on {t < }. In particular, any square integrable H-martingale (X
t
, t 0) can
be written as X
t
= X
0
+
_
t
0
x
s
dM
s
where (x
t
, t 0) is a predictable process.
Proof: We give two dierent proofs.
First proof:
From Lemma 2.2.1
M
h
t
= h()11
{t}
+ 11
{t<}
E(h()11
{t<}
)
P(t < )
= h()11
{t}
+ 11
{t<}
e
(t)
E(h()11
{t<}
) .
An integration by parts leads to
e
t
E[h()11
{t<}
] = e
t
_

t
h(s)dF(s) =

h(t)
=
_

0
h(s)dF(s)
_
t
0
e
(s)
h(s)dF(s) +
_
t
0
E(h()11
{s<}
)e
(s)
d(s)
Therefore, since E(h()) =
_
0
h(s)dF(s) and M
h
s
= e
(s)
E(h()11
{s<}
) =

h(s) on {s < }, the
following equality holds on the set {t < }:
e
t
E[h()11
{t<}
] = E(h())
_
t
0
e
(s)
h(s)dF(s) +
_
t
0
h(s)d(s) .
Hence,
11
{t<}
E(h()|H
t
) = 11
{t<}
_
E(h()) +
_
t
0
(
h(s) h(s))
dF(s)
1 F(s)
_
= 11
{t<}
_
E(h())
_
t
0
(
h(s) h(s))(dH
s
d(s))
_
,
where the last equality is due to 11
t<
_
t
0
(
h(s) h(s))dH
s
= 0.
On the complementary set {t }, we have seen that E(h()|H
t
) = h(), whereas
_
t
0
(
h(s) h(s))(dH
s
d(s)) =
_
]0,]
(
h(s) h(s))(dH
s
d(s))
=
_
]0,[
(
h(s) h(s))(dH
s
d(s)) + (
h(
) h()) .
Therefore,
E(h())
_
t
0
(
h(s) h(s))(dH
s
d(s)) = M
H
(M
H
h()) = h() .
The predictable representation theorem follows immediately.
Second proof We start from the form of the conditional expectation
M
h
t
= E(h()|H
t
) = h()11
{<t}
+ 11
{>t}
e
(t)
_

t
h(u)dF(u)
=
_
t
0
h(s)dH
s
+ (1 H
t
)e
(t)
_

t
h(u)dF(u) =
_
t
0
h(s)dH
s
+ (1 H
t
)
h(t) .
We obtain, using that dF(t) = e
(t)
d(t) = e
(t)
(t)dt and using Itos formula and dM
t
= dH
t

(t)(1 H
t
)dt
dM
h
t
= h(t)dH
t
+ (1 H
t
)h(t)(t)dt
h(t) dH
t
(1 H
t
)
h(t)(t)dt
= (h(t)
h(t))dH
t
+ (1 H
t
)(h(t)
h(t))(t)dt = (h(t)
h(t))dM
t
Exercise 2.2.2 If is not continuous, prove that

E(h()|H
t
) = E(h())
_
t
0
e
(s)
(
h(s) h(s)) dM
s
.
2.2.5 Change of a Probability Measure
Let P
be an arbitrary probability measure on (, H
), which is absolutely continuous with respect

to P. We denote by the H
-measurable density of P
with respect to P
:=
dP
dP
= h() 0, P-a.s., (2.15)
where h : IR IR
+
is a Borel measurable function satisfying
E
P
(h()) =
_

0
h(u) dF(u) = 1.
We can use Girsanovs theorem. Nevertheless, we prefer here to establish this theorem in our
particular setting. Of course, the probability measure P
is equivalent to P if and only if the inequality

in (2.15) is strict P-a.s. Furthermore, we shall assume that P
( = 0) = 0 and P
( > t) > 0 for any

t IR
+
. Actually the rst condition is satised for any P
absolutely continuous with respect to P.

For the second condition to hold, it is sucient and necessary to assume that for every t
P
( > t) =
_
]t,[
h(u) dF(u) > 0,
Put another way, we assume that
g(t)
def
= e
(t)
E
_
11
>t
h()
_
= e
(t)
_
]t,[
h(u) dF(u) = e
(t)
P
( > t) > 0.
Then, the hazard function
of with respect to P
is well dened. Our goal is to examine

relationships between hazard functions
and . It is easily seen that in general we have
(t)
(t)
=
ln
_
_
]t,[
h(u) dF(u)
_
ln(1 F(t))
, (2.16)
since by denition
(t) = ln(1 F
(t)) where the c.d.f. F
of under P
is
F
(t) := P
( t) =
_
[0,t]
h(u) dF(u). (2.17)
Assume rst that F is an absolutely continuous function, so that the intensity function of under
P is well dened. Recall that is given by the formula
(t) =
f(t)
1 F(t)
.
The c.d.f. F
of under P
now equals
F
(t) := P
( t) = E
P
(11
t
h()) =
_
t
0
h(u)f(u) du.
so that F
follows an absolutely continuous function. Therefore, the intensity function
of the
random time under P
exists, and it is given by the formula
(t) =
h(t)f(t)
1 F
(t)
=
h(t)f(t)
1
_
t
0
h(u)f(u) du
.
To derive a more straightforward relationship between the intensities and
, let us introduce an
auxiliary function h
: IR
+
IR, given by the formula h
(t) = h(t)/g(t).
Notice that
(t) =
h(t)f(t)
1
_
t
0
h(u)f(u) du
=
h(t)f(t)
_
t
h(u)f(u) du
=
h(t)f(t)
e
(t)
g(t)
= h
(t)
f(t)
1 F(t)
= h
(t)(t).
This means also that d
(t) = h
(t) d(t). It appears that the last equality holds true if F is merely
a continuous function. Indeed, if F (and thus F
) is continuous, we get
d
(t) =
dF
(t)
1 F
(t)
=
d(1 e
(t)
g(t))
e
(t)
g(t)
=
g(t)d(t) dg(t)
g(t)
= h
(t) d(t).
To summarize, if the hazard function is continuous then
is also continuous and d
(t) =
h
(t) d(t).
To understand better the origin of the function h
, let us introduce the following non-negative

P-martingale (which is strictly positive when the probability measures P
and P are equivalent)
t
:=
dP
dP |H
t
= E
P
(|H
t
) = E
P
(h()|H
t
), (2.18)
so that
t
= M
h
t
. The general formula for
t
reads (cf. (2.2.1))
t
= 11
t
h() + 11
>t
e
(t)
_
]t,[
h(u) dF(u) = 11
t
h() + 11
>t
g(t).
Assume now that F is a continuous function. Then
t
= 11
t
h() + 11
>t
_

t
h(u)e
(t)(u)
d(u).
On the other hand, using the representation theorem, we get
M
h
t
= M
h
0
+
_
]0,t]
M
h
u
(h
(u) 1) dM
u
where h
(u) = h(u)/g(u). We conclude that
t
= 1 +
_
]0,t]
u
(h
(u) 1) dM
u
. (2.19)
It is thus easily seen that
t
=
_
1 + 11
t
v()) exp
_
_
t
0
v(u) d(u)
_
, (2.20)
where we write v(t) = h
(t)1. Therefore, the martingale property of the process , which is obvious

from (2.18), is also a consequence of Proposition 2.2.5.
Remark 2.2.1 In view of (2.19), we have
t
= E
t
_
_

0
(h
(u) 1) dM
u
_
,
where E stands for the Doleans exponential. Representation (2.20) for the random variable
t
can
thus be obtained from the general formula for the Doleans exponential. (See Appendix 8.4.4.)
We are in the position to formulate the following result (all statements were already established
above).
Proposition 2.2.8 Let P
be any probability measure on (, H
) absolutely continuous with respect

to P, so that (2.15) holds for some function h. Assume that P
( > t) > 0 for every t IR

+
. Then
dP
dP |H
t
= E
t
_
_

0
(h
(u) 1) dM
u
_
, (2.21)
where
h
(t) = h(t)/g(t), g(t) = e

(t)
_

t
h(u) dF(u),
and
(t) = g
(t)(t) with
g
(t) =
ln
_
_
]t,[
h(u) dF(u)
_
ln(1 F(t))
. (2.22)
If, in addition, the random time admits the intensity function under P, then the intensity function
of under P
satises
(t) = h
(t)(t) a.e. on IR
+
. More generally, if the hazard function
of under P is continuous, then the hazard function
of under P
is also continuous, and it

satises d
(t) = h
(t) d(t).
Corollary 2.2.3 If F is continuous then M
t
= H
t

(t ) is a H-martingale under P
.
Proof: In view Proposition 2.2.2, the corollary is an immediate consequence of the continuity
of
. Alternatively, we may check directly that the product U

t
=
t
M
t
=
t
(H
t
(t )) follows
a H-martingale under P. To this end, observe that the integration by parts formula for functions of
nite variation yields
U
t
=
_
]0,t]
t
dM
t
+
_
]0,t]
M
t
d
t
=
_
]0,t]
t
dM
t
+
_
]0,t]
M
t
d
t
+
ut
M
u
=
_
]0,t]
t
dM
t
+
_
]0,t]
M
t
d
t
+ 11
t
(
).
Using (2.19), we obtain
U
t
=
_
]0,t]
t
dM
t
+
_
]0,t]
M
t
d
t
+
11
t
(h
() 1)
=
_
]0,t]
t
d
_
(t )
(t ) + 11
t
(h
() 1)
_
+N
t
,
where the process N, which equals
N
t
=
_
]0,t]
t
dM
t
+
_
]0,t]
M
t
d
t
is manifestly a H-martingale with respect to P. It remains to show that the process
N
t
:= (t )
(t ) + 11
t
(h
() 1)
follows a H-martingale with respect to P. By virtue of Proposition 2.2.4, the process
11
t
(h
() 1) + (t )
_
t
0
h
(u) d(u)
is a H-martingale. Therefore, to conclude the proof it is enough to notice that
_
t
0
h
(u) d(u)
(t ) =
_
t
0
(h
(u) d(u) d
(u)) = 0,
where the last equality is a consequence of the relationship d
(t) = h
(t) d(t) established in

Proposition 2.2.8.
By virtue of Proposition 2.2.2 if
is a continuous function then the process M
= H
t
(t)
follows a H-martingale under P
. The next result suggests that this martingale property uniquely

characterizes the (continuous) hazard function of a random time.
Lemma 2.2.2 Suppose that an equivalent probability measure P
is given by formula (2.15) for some

function h. Let
: IR
+
IR
+
be an arbitrary continuous increasing function, with
(0) = 0. If
the process M
t
:= H
t

(t ) follows a H-martingale under P
, then
(t) = ln(1 F
(t))
with F
given by formula (2.21).

Proof: The Bayes rule implies
E
P
(M
t
|H
s
) =
E
P
(M
t
|H
s
)
E
P
(|H
s
)
=
1
s
E
P
(M
t

t
|H
s
)
and thus
E
P
(M
t
|H
s
) =
E
P
_
(H
t

(t ))(H
t
h() + (1 H
t
)g(t))
H
s
_
H
s
h() + (1 H
s
)g(s)
,
or equivalently
E
P
(M
t
|H
s
) =
E
P
_
H
t
h() H
t
(t )h() (1 H
t
)
(t )g(t)
H
s
_
H
s
h() + (1 H
s
)g(s)
.
This means that
E
P
(M
t
|H
s
) =
J
H
s
h() + (1 H
s
)g(s)
,
where we write
J = E
P
_
H
t
h() H
t
(t )h() (1 H
t
)
(t )g(t)
H
s
_
.
We obtain
J = H
s
h() H
s
()h() (1 H
s
)(1 F(s))
1
E
P
_
11
{s<t}
(
() 1)h() +11
{>t}
(t)g(t)
_
and thus the martingale condition E
P
(M
t
|H
s
) = M
s
, is equivalent to the following equality
(1 H
s
)(1 F(s))
1
E
P
_
11
{s<t}
(
() 1)h() + 11
{>t}
(t)g(t)
_
=
(s)(1 H
s
)g(s).
Therefore, for every s t we have
E
P
_
11
{s<t}
(
() 1)h() + 11
{>t}
(t)g(t)
_
=
(s)(1 F(s))g(s)
so that
_
t
s
(
(u) 1)h(u) dF(u) +
(t)g(t)(1 F(t)) =
(s)
_

s
h(u) dF(u),
and nally,
_
t
s
(
(u) 1) dF
(u) +
(t)(1 F
(t)) =
(s)(1 F
(s)).
After simple manipulations involving the integration by parts, we get for s t
_
t
s
(1 F
(u)) d
(u) = F
(t) F
(s),
and since
(0) = F
(0) = 0, we nd that
= ln(1 F
(t)).
Representation Theorem
We now recall a suitable version of the predictable representation theorem and we shall present
a dierent proof. For any RCLL function

h : IR
+
IR such that the random variable

h() is
integrable, we set

M
t
= E
Q
(
h() | H
t
) for every t IR
+
. It is clear that

M is an H-martingale under
Q. The following version of the martingale representation theorem is well known (see, for instance,
Blanchet-Scalliet and Jeanblanc [27], Jeanblanc and Rutkowski [121] or Proposition 4.3.2 in Bielecki
and Rutkowski [23]).
Proposition 2.2.9 Assume that G is continuous and

h is an RCLL function such that the random
variable

h() is Q-integrable. Then the H-martingale

M admits the following integral representation
M
t
=

M
0
+
_
]0,t]
(
h(u) g(u)) dM
u
, (2.23)
where the continuous function g : IR
+
IR is given by the formula
g(t) =
1
G(t)
E
Q
_
11
{>t}
h()
_
=
1
G(t)
_

t
h(u) dG(u). (2.24)

Remark 2.2.2 It is easily seen that on the set {t } we have g(t) =

M
t
. Therefore, formula
(2.23) can also be rewritten as follows
M
t
=

M
0
+
_
]0,t]
_
h(u)

M
u
_
dM
u
=

M
0
+
_
]0,t]
_
h(u)

M(u)
_
dM
u
, (2.25)
where

M = g is the unique function such that

M
t
11
{>t}
=

M(t)11
{>t}
for every t IR
+
.
Lemma 2.2.3 Let M
1
and M
2
be arbitrary two H-martingales under Q. If for every t [0, T] we
have 11
{t<}
M
1
t
= 11
{t<}
M
2
t
then M
1
t
= M
2
t
for every t [0, T].
Proof: We have M
i
t
= E
Q
(h
i
() | H
t
) for some functions h
i
: IR
+
IR such that h
i
() is
Q-integrable. Using the well known formula for the conditional expectation
E
Q
(h
i
() | H
t
) = 11
{t}
h
i
() 11
{t<}
1
G(t)
_

t
h
i
(u) dG(u) = 11
{t}
h
i
() + 11
{t<}
g
i
(t),
and the assumption that 11
{t<}
M
1
t
= 11
{t<}
M
2
t
, we obtain the equality g
1
(t) = g
2
(t) for every
t [0, T] (recall that Q( > t) > 0 for every t [0, T]). Therefore, we have
_

t
h
1
(u) dG(u) =
_

t
h
2
(u) dG(u), t [0, T].
This immediately implies that h
1
(t) = h
2
(t) on [0, T], almost everywhere with respect to the dis-
tribution of , and thus we have h
1
( T) = h
2
( T), Q-a.s. Consequently, M
1
t
= M
2
t
for every
t [0, T].
2.2.6 Incompleteness of the Toy model
In order to study the completeness of the nancial market, we rst need to dene the tradeable
assets.
If the market consists only of the risk-free zero-coupon bond, there exists innitely many e.m.ms.
The discounted asset prices are constant, hence the set Q of equivalent martingale measures is the
set of probabilities equivalent to the historical one. For any Q Q, we denote by F
Q
the cumulative
function of under Q, i.e.,
F
Q
(t) = Q( t) .
The range of prices is dened as the set of prices which do not induce arbitrage opportunities. For
a DZC with a constant rebate R paid at maturity, the range of prices is equal to the set
{E
Q
(B(0, T)(11
{T<}
+R11
{<T}
)), Q Q} .
This set is exactly the interval ]RB(0, T), B(0, T)[. Indeed, it is obvious that the range of prices is
included in the interval ]RB(0, T), B(0, T)[. Now, in the set Q, one can select a sequence of proba-
bilities Q
n
which converge weakly to the Dirac measure at point 0 (resp. at point T) (the bounds
are obtained as limit cases: the default appears at time 0
+
, or never). Obviously, this range is too
large to be ecient. (See Hugonnier for a generalization of this result)
2.2.7 Risk Neutral Probability Measures
It is usual to interpret the absence of arbitrage opportunities as the existence of an e.m.m. . If DZCs
are traded, their prices are given by the market, and the equivalent martingale measure Q, chosen
by the market, is such that, on the set {t < },
D
(R)
(t, T) = B(t, T)E
Q
_
[11
T<
+R11
t<T
]
t <
_
.
Therefore, we can characterize the cumulative function of under Q from the market prices of the
DZC as follows.
Zero Recovery
If a DZC with zero recovery of maturity T is traded at a price D
(R)
(t, T) which belongs to the
interval ]0, B(t, T)[ , then, under any risk-neutral probability Q, the process B(0, t)D
(R)
(t, T) is a
martingale (for the moment, we do not know if the market is complete, so we can not claim that
the e.m.m. is unique), the following equality holds
D
(R)
(t, T)B(0, t) = E
Q
(B(0, T)11
{T<}
|H
t
) = B(0, T)11
{t<}
exp
_
_
T
t
Q
(s)ds
_
where
Q
(s) =
dF
Q
(s)/ds
1 F
Q
(s)
. It is obvious that if D
(R)
(t, T) belongs to the range of viable prices
]0, B(0, T)[, the process
Q
is strictly positive (and the converse holds true). The process
Q
is the
Q-intensity of . Therefore, the value of
_
T
t
Q
(s)ds is known for any t as soon as there are DZC
bonds for each maturity, and the unique risk-neutral intensity can be obtained from the prices of
DZCs as r(t) +
Q
(t) =
T
lnD
(R)
(t, T)|
T=t
.
Remark 2.2.3 It is important to note that there is no relation between the risk-neutral intensity
and the historical one. The risk-neutral intensity can be greater (resp. smaller) than the historical
one. The historical intensity can be deduced from observation of default time, the risk-neutral one
is obtained from the prices of traded defaultable claims.
Fixed Payment at maturity
If the prices of DZCs with dierent maturities are known, then from (2.1)
B(0, T) D
(R,T)
(0, T)
B(0, T)(1 R)
= F
Q
(T)
where F
Q
(t) = Q( t), so that the law of is known under the e.m.m.. However, as noticed
in Hull and White [107], extracting default probabilities from bond prices [is] in practice, usually
more complicated. First, the recovery rate is usually non-zero. Second, most corporate bonds are not
zero-coupon bonds.
Payment at hit
In this case the cumulative function can be obtained using the derivative of the defaultable zero-
coupon price with respect to the maturity. Indeed, denoting by
T
D
(R)
the derivative of the value
of the DZC at time 0 with respect to the maturity, and assuming that G = 1 F is dierentiable,
we obtain from (2.3)
T
D
(R)
(0, T) = g(T)B(0, T) G(T)B(0, T)r(T) R(T)g(T)B(0, T) ,
where g(t) = G
(t). Therefore, solving this equation leads to

Q( > t) = G(t) = (t)
_
1 +
_
t
0
T
D
(R)
(0, s)
1
B(0, s)(1 R(s))
((s))
1
ds
_
,
where (t) = exp
__
t
0
r(u)
1 R(u)
du
_
.
2.2.8 Partial information: Due and Landos model
Due and Lando [71] study the case where = inf{t : V
t
m} where V satises
dV
t
= (t, V
t
)dt +(t, V
t
)dW
t
.
Here the process W is a Brownian motion. If the information is the Brownian ltration, the time
is a stopping time w.r.t. a Brownian ltration, therefore is predictable and admits no intensity.
We will discuss this point latter on. If the agents do not know the behavior of V , but only the
minimal information H
t
, i.e. he knows when the default appears, the price of a zero-coupon is,
in the case where the default is not yet occurred, exp
_
_
T
t
(s)ds
_
where (s) =
f(s)
G(s)
and
G(s) = P( > s), f = G
, as soon as the cumulative function of is dierentiable. Due and Lando

have obtained that the intensity is (t) =
1
2
2
(t, 0)
f
x
(t, 0) where f(t, x) is the conditional density
of V
t
when T
0
> t, i.e. the dierential w.r.t. x of
P(V
t
x, T
0
> t)
P(T
0
> t)
, where T
0
= inf{t ; V
t
= 0}. In
the case where V is an homogenous diusion, i.e. dV
t
= (V
t
)dt + (V
t
)dW
t
, the equality between
Due-Lando and our result is not so obvious. See Elliott et al. [81] for comments.
2.3 Pricing and Trading a CDS
We are now in the position to apply the general theory to the case of a particular class contracts,
specically, credit default swaps. We work throughout under a spot martingale measure Q on
(, G
T
). We shall work under additional assumption that the interest rate r is null. Subsequently,
these restrictions will be relaxed.
2.3.1 Valuation of a Credit Default Swap
A stylized credit default swap is formally introduced through the following denition.
Denition 2.3.1 A credit default swap with a constant rate and recovery at default is a defaultable
claim (0, A, Z, ), where Z
t
R(t) and A
t
= t for every t [0, T]. An RCLL function R : [0, T]
IR represents the default protection, and a constant IR represents the CDS rate (also termed the
spread, premium or annuity of a CDS).
We shall rst analyze the valuation and trading credit default swaps. We denote by F the
cumulative distribution function of the default time under Q, and we assume that F is a continuous
function, with F(0) = 0 and F(T) < 1 for some xed date T > 0. Also, we write G = 1 F to
denote the survival probability function of , so that G(t) > 0 for every t [0, T]. For simplicity
of exposition, we assume in this section that the interest rate r = 0, so that the price of a savings
account B
t
= 1 for every t. Note also that we have only one tradeable asset in our model (a savings
account), and we wish to value a defaultable claim within this model. It is clear that any probability
measure Q on (, H
T
), equivalent to Q, can be chosen as a spot martingale measure for our model.
The choice of Q is reected in the cumulative distribution function F (in particular, in the default
intensity if F is absolutely continuous).
Ex-dividend Price of a CDS
Consider a CDS with the rate , which was initiated at time 0 (or indeed at any date prior to the
current date t). Its market value at time t does not depend on the past otherwise than through the
level of the rate . Unless explicitly stated otherwise, we assume that is an arbitrary constant.
Unless explicitly stated otherwise, we assume that the default protection payment is received at
the time of default, and it is equal R(t) if default occurs at time t, prior to or at maturity date T.
In view of (4.37), the ex-dividend price of a CDS maturing at T with rate is given by the
formula
S
t
() = E
Q
_
11
{t<T}
R()
H
t
_
E
Q
_
11
{t<}
_
( T) t
_
H
t
_
, (2.26)
2.3. PRICING AND TRADING A CDS 39
where the rst conditional expectation represents the current value of the default protection stream
(or the protection leg), and the second is the value of the survival annuity stream (or the fee leg).
Note that in Lemma 2.3.1, we do not need to specify the inception date s of a CDS. We only
assume that the maturity date T, the rate , and the protection payment R are given.
Lemma 2.3.1 The ex-dividend price at time t [s, T] of a credit default swap started at s, with
rate and protection payment R() at default, equals
S
t
() = 11
{t<}
1
G(t)
_
_
T
t
R(u) dG(u)
_
T
t
G(u) du
_
. (2.27)
Proof: We have, on the set {t < },
S
t
() =
_
T
t
R(u) dG(u)
G(t)

_
_
T
t
udG(u) +TG(T)
G(t)
t
_
=
1
G(t)
_
_
T
t
R(u) dG(u)
_
TG(T) tG(t)
_
T
t
udG(u)
_
_
.
Since
_
T
t
G(u) du = TG(T) tG(t)
_
T
t
udG(u), (2.28)
we conclude that (2.27) holds.
The ex-dividend price of a CDS can also be represented as follows (see (4.38))
S
t
() = 11
{t<}
S
t
(), t [0, T], (2.29)
where

S
t
() stands for the ex-dividend pre-default price of a CDS. It is useful to note that formula
(2.27) yields an explicit expression for

S
t
(), and that

S() follows a continuous function, provided
that G is continuous.
2.3.2 Market CDS Rate
Assume now that a CDS was initiated at some date s t and its initial price was equal to zero. Since
a CDS with this property plays an important role, we introduce a formal denition. In Denition
2.3.2, it is implicitly assumed that a recovery function R is given.
Denition 2.3.2 A market CDS started at s is a CDS initiated at time s whose initial value is
equal to zero. A T-maturity market CDS rate (also known as the fair CDS spread) at time s is the
level of the rate = (s, T) that makes a T-maturity CDS started at s valueless at its inception. A
market CDS rate at time s is thus determined by the equation S
s
((s, T)) = 0, where S is dened
by (2.26). By assumption, (s, T) is an F
s
-measurable random variable (hence, a constant if the
reference ltration is trivial).
Under the present assumptions, by virtue of Lemma 2.3.1, the T-maturity market CDS rate
(s, T) solves the following equation
_
T
s
R(u) dG(u) +(s, T)
_
T
s
G(u) du = 0,
and thus we have, for every s [0, T],
(s, T) =
_
T
s
R(u) dG(u)
_
T
s
G(u) du
. (2.30)
Remarks 2.3.1 Let us comment briey on a model calibration. Suppose that at time 0 the market
gives the premium of a CDS for any maturity T. In this way, the market chooses the risk-neutral
probability measure Q. Specically, if (0, T) is the T-maturity market CDS rate for a given recovery
function R then we have
(0, T) =
_
T
0
R(u) dG(u)
_
T
0
G(u) du
.
Hence, if credit default swaps with the same recovery function R and various maturities are traded
at time 0, it is possible to nd the implied risk-neutral c.d.f. F (and thus the default intensity
under Q) from the term structure of CDS rates (0, T) by solving an ordinary dierential equation.
Standing assumptions. We x a maturity date T, and we write briey (s) instead of (s, T). In
addition, we assume that all credit default swaps have a common recovery function R.
Note that the ex-dividend pre-default value at time t [0, T] of a CDS with any xed rate
can be easily related to the market rate (t). We have the following result, in which the quantity
(t, s) = (t) (s) represents the calendar CDS market spread (for a given maturity T).
Proposition 2.3.1 The ex-dividend price of a market CDS started at s with recovery R at default
and maturity T equals, for every t [s, T],
S
t
((s)) = 11
{t<}
((t) (s))
_
T
t
G(u) du
G(t)
= 11
{t<}
(t, s)
_
T
t
G(u) du
G(t)
, (2.31)
or more explicitly,
S
t
((s)) = 11
{t<}
_
T
t
G(u) du
G(t)
_
_
T
s
R(u) dG(u)
_
T
s
G(u) du
_
T
t
R(u) dG(u)
_
T
t
G(u) du
_
. (2.32)
Proof: To establish equality (2.32), it suces to observe that S
t
((s)) = S
t
((s)) S
t
((t)), and
to use (2.27) and (2.30).
Remark 2.3.1 Note that the price of a CDS can take negative values.
Forward Start CDS
A representation of the value of a swap in terms of the market swap rate, similar to (2.31), is well
known to hold for default-free interest rate swaps. It is particularly useful if the calendar spread is
modeled as a stochastic process. In particular, it leads to the Black swaption formula within the
framework of Jamshidians [112] model of co-terminal forward swap rates.
In the present context, it is convenient to consider a forward start CDS initiated at time s [0, U]
and giving default protection over the future time interval [U, T]. If the reference entity defaults
prior to the start date U the contract is terminated and no payments are made. The price of this
contract at any date t [s, U] equals
S
t
() = E
Q
_
11
{U<T}
R()
H
t
_
E
Q
_
11
{U<}
_
( T) U
_
H
t
_
. (2.33)
Since a forward start CDS does not pays any dividends prior to the start date U, the price S
t
(), t
[s, U], can be considered here as either the cum-dividend price or the ex-dividend price. Note that
since G is continuous, the probability of default occurring at time U equals zero, and thus for t = U
the last formula coincides with (2.26). This is by no means surprising, since at time T a forward
start CDS becomes a standard (i.e., spot) CDS.
If G is continuous, representation (2.33) can be made more explicit, namely,
S
t
() = 11
{t<}
1
G(t)
_
_
T
U
R(u) dG(u)
_
T
U
G(u) du
_
.
A forward start market CDS at time t [0, U] is a forward CDS in which is chosen at time t in
such a way that the contract is valueless at time t. The corresponding (pre-default) forward CDS
rate (t, U, T) is thus determined by the the following equation
S
t
((t, U, T)) = E
Q
_
11
{U<T}
R()
H
t
_
E
Q
_
11
{U<}
(t, U, T)
_
( T) U
_
H
t
_
= 0,
which yields, for every t [0, U],
(t, U, T) =
_
T
U
R(u) dG(u)
_
T
U
G(u) du
.
The price of an arbitrary forward CDS can be easily expressed in terms of and (t, U, T). We
have, for every t [0, U],
S
t
() = S
t
() S
t
((t, U, T)) = ((t, U, T) ) E
Q
_
11
{U<}
_
( T) U
_
H
t
_
,
or more explicitly,
S
t
() = 11
{t<}
((t, U, T) )
_
T
U
G(u) du
G(t)
.
Under the assumption of a deterministic default intensity, the formulae above are of rather limited
interest. Let us stress, however, that similar representations are also valid in the case of a stochastic
default intensity, where they prove useful in pricing of options on a forward start CDS (equivalently,
options on a forward CDS rate).
Case of a Constant Default Intensity
Assume that R(t) = R is independent of t, and F(t) = 1 e
t
for a constant default intensity
> 0 under Q. In this case, the valuation formulae for a CDS can be further simplied. In view of
Lemma 2.3.1, the ex-dividend price of a (spot) CDS with rate equals, for every t [0, T],
S
t
() = 11
{t<}
(R )
1
_
1 e
(Tt)
_
.
The last formula (or the general formula (2.30)) yields (s) = R for every s < T, so that the market
rate (s) is independent of s. As a consequence, the ex-dividend price of a market CDS started at
s equals zero not only at the inception date s, but indeed at any time t [s, T], both prior to and
after default). Hence, this process follows a trivial martingale under Q. As we shall see in what
follows, this martingale property the ex-dividend price of a market CDS is an exception, rather than
a rule, so that it no longer holds if default intensity is not constant.
2.3.3 Price Dynamics of a CDS
Unless explicitly stated otherwise, we consider a spot CDS and we assume that
G(t) = Q( > t) = exp
_
_
t
0
(u) du
_
,
where the default intensity (t) under Q is a strictly positive deterministic function. We rst focus
on the dynamics of the ex-dividend price of a CDS with rate started at some date s < T.
Lemma 2.3.2 The dynamics of the ex-dividend price S
t
() on [s, T] are
dS
t
() = S
t
() dM
t
+ (1 H
t
)( R(t)(t)) dt, (2.34)
where the H-martingale M under Q is given by the formula
M
t
= H
t

_
t
0
(1 H
u
)(u) du, t IR
+
. (2.35)
Hence, the process

S
t
(), t [s, T], given by the expression
S
t
() = S
t
() +
_
t
s
R(u) dH
u

_
t
s
(1 H
u
) du (2.36)
is a Q-martingale for t [s, T]. Specically,
d
S
t
() =
_
R(t) S
t
()
_
dM
t
=
_
R(t) (1 H
t
)
S
t
()
_
dM
t
. (2.37)
Proof: It suces to recall that
S
t
() = 11
{t<}
S
t
() = (1 H
t
)
S
t
()
so that
dS
t
() = (1 H
t
) d
S
t
()

S
t
() dH
t
.
Using formula (2.27), we nd easily that we have
d
S
t
() = (t)
S
t
() dt + ( R(t)(t)) dt. (2.38)
In view of (2.35) and the fact that S
() =

S
() , the proof of (2.34) is complete. To prove the

second statement, it suces to observe that the process N given by
N
t
= S
t
()
_
t
s
(1 H
u
)( R(u)(u)) du =
_
t
s
S
u
() dM
u
is an H-martingale under Q. But for every t [s, T]
S
t
() = N
t
+
_
t
s
R(u) dM
u
,
so that

S() also follows an H-martingale under Q. Note that the process

S() given by (2.36)
represents the cum-dividend price of a CDS, so that the martingale property

S() is expected.
Equality (2.34) emphasizes the fact that a single cash ow of R() occurring at time can be
formally treated as a dividend stream at the rate R(t)(t) paid continuously prior to default. It is
clear that we also have
dS
t
() =
S
t
() dM
t
+ (1 H
t
)( R(t)(t)) dt. (2.39)
In some instances, it can be useful to reformulate the dynamics of a market CDS in terms of
market observables, such as CDS spreads.
Corollary 2.3.1 The dynamics of the ex-dividend price S
t
((s)) on [s, T] are also given as
dS
t
((s)) = S
t
((s)) dM
t
+ (1 H
t
)
_
_
T
t
G(u) du
G(t)
d
t
(t, s) (t, s) dt
_
. (2.40)
Proof: Under the present assumptions, for any xed s, the calendar spread (t, s), t [s, T] is a
continuous function of bounded variation. In view of (2.34), it suces to check that
_
T
t
G(u) du
G(t)
d
t
(t, s) (t, s) dt = ((s) R(t)(t)) dt, (2.41)
where d
t
(t, s) = d
t
((t) (s)) = d(t). Equality (2.41) follows by elementary computations.
Trading a Credit Default Swap
We shall show that, in the present set-up, in order to replicate an arbitrary contingent claim Y
settling at time T and satisfying the usual integrability condition, it suces to deal with two traded
assets: a CDS with maturity U T and a constant savings account B = 1. Since one can always
work with discounted values, the last assumption is not restrictive.
According to Section 4.3.4, a strategy
t
= (
0
t
,
1
t
), t [0, T], is self-nancing if the wealth
process V (), dened as
V
t
() =
0
t
+
1
t
S
t
(), (2.42)
satises
dV
t
() =
1
t
_
dS
t
() +dD
t
_
=
1
t
d
S
t
(), (2.43)
where S() is the ex-dividend price of a CDS with the dividend stream D , and so,

S() = S() +D
is the corresponding cum-dividend price process. As usual, we say that a strategy replicates a
contingent claim Y if V
T
() = Y . On the set { t T} the ex-dividend price S() equals zero,
and thus the total wealth is necessarily invested in B, so that it is constant. This means that
replicates Y if and only if V
T
() = Y .
Lemma 2.3.3 For any self-nancing strategy we have, on the set { T},
V () := V
() V
() =
1
(R()

S
()). (2.44)
Proof: In general, the process
1
is G-predictable. In our model,
1
is assumed to be an RCLL
function. The jump of the wealth process V () at time equals, on the set { T},
V () =
1
S +
1
D =
1

S,
where
S() = S
() S
() =
() (recall that the ex-dividend price S() drops to zero at

default time) and manifestly
D = R().
2.3.4 Hedging of Defaultable Claims
An H
T
-measurable random variable Y is known to admit the following representation
Y = 11
{T}
Z() + 11
{T<}
X, (2.45)
where Z : [0, T] IR is a Borel measurable function, and X is a constant. For deniteness, we shall
deal with claims Y such that h is an RCLL function, but this formal restriction is not essential.
Using results of Section 2.2.4
E(V
T
|H
t
) = V
t
= Z
11
{t}
+ 11
{t<}
1
G
t
_
XG
T
+
_
T
t
Z
s
dG
s
_
=
_
t
0
Z
t
dH
s
+ (1 H
s
)
1
G
t
_
XG
T
+
_
T
t
Z
s
dG
s
_
hence dV
t
= (Z
t
g)dM
t
with g(t) =
1
G
t
(
_
T
t
Z
s
dG
s
+XG
T
).
Our aim is to provide an hedging strategy for Y , using a CDS. In view of Lemma 2.3.2, the
dynamics of the price S() are
dS
t
() = S
t
() dM
t
+ (1 H
t
)( R(t)(t)) dt.
From Corollary 4.3.1, we know that the wealth V () of any admissible self-nancing strategy
0
,
1
built on savings account and CDS, is an H-martingale under Q. Furthermore, using the dynamics
obtained in Section 2.2.4, we have
dV
t
() =
1
t
(S
t
R
t
)dM
t
(2.46)
Then, by identication
Proposition 2.3.2 Assume that the inequality

S
t
() = R(t) holds for every t [0, T]. Let
1
be an
RCLL function given by the formula
1
t
=
h(t) g(t)
R(t)

S
t
()
, (2.47)
and let
0
t
= V
t
()
1
t
S
t
(), where the process V () is given by (2.43) with the initial condition
V
0
() = E
Q
(Y ), where Y is given by (2.45). Then the self-nancing trading strategy = (
0
,
1
) is
admissible and it is a replicating strategy for a defaultable claim (X, 0, Z, ), where X = c(T) and
Z
t
= h(t).
Replication of a Defaultable Claim
We now give a dierent proof, based on the representation theorem.
To deal with a claim Y given (2.45), we shall apply Proposition 2.2.9 to the function

h, where
h(t) = h(t) for t T and

h(t) = c(T) for t > T (recall that Q( = T) = 0). In this case, we obtain
g(t) =
1
G(t)
_
_
T
t
h(u) dG(u) +c(T)G(T)
_
, (2.48)
and thus for the process

M
t
= E
Q
(Y | H
t
), t [0, T], we have
M
t
= E
Q
(Y ) +
_
]0,t]
(h(u) g(u)) dM
u
(2.49)
with g given by (2.48). Recall that

S() is the pre-default ex-dividend price process of a CDS with
rate and maturity T. We know that

S() is a continuous function of t if G is continuous.
Proposition 2.3.3 Assume that the inequality

S
t
() = R(t) holds for every t [0, T]. Let
1
be an
RCLL function given by the formula
1
t
=
h(t) g(t)
R(t)

S
t
()
, (2.50)
and let
0
t
= V
t
()
1
t
S
t
(), where the process V () is given by (2.43) with the initial condition
V
0
() = E
Q
(Y ), where Y is given by (2.45). Then the self-nancing trading strategy = (
0
,
1
) is
admissible and it is a replicating strategy for a defaultable claim (X, 0, Z, ), where X = c(T) and
Z
t
= h(t).
Proof: The idea of the proof is based on the observation that it is enough to concentrate on the
formula for trading strategy prior to default. In view of Lemma 2.3.2, the dynamics of the price
S() are
dS
t
() = S
t
() dM
t
+ (1 H
t
)( R(t)(t)) dt.
and thus we have, on the set { > t},
dS
t
() = d
S
t
() =
_
(t)
S
t
() + R(t)(t)
_
dt. (2.51)
From Corollary 4.3.1, we know that the wealth V () of any admissible self-nancing strategy is
an H-martingale under Q. Since under the present assumptions dB
t
= 0, for the wealth process
V () we obtain, on the set { > t},
dV
t
() =
1
t
(d
S
t
() dt) =
1
t
(t)
_
R(t)

S
t
()
_
dt. (2.52)
For the martingale

M
t
= E
Q
(Y | H
t
) associated with Y , in view of (2.49) we obtain, on the set
{ > t},
d
M
t
= (t)(h(t) g(t)) dt. (2.53)
We wish to nd
1
such that V
t
() =

M
t
for every t [0, T]. To this end, we rst focus on the
equality 11
{t<}
V
t
() = 11
{t<}
M
t
for pre-default values. Since (t) is assumed to be strictly positive,
a comparison of (2.52) with (2.53) yields
1
t
=
h(t) g(t)
R(t)

S
t
()
, t [0, T]. (2.54)
We thus see that if V
0
() =

M
0
then also 11
{t<}
V
t
() = 11
{t<}
M
t
for every t [0, T]. As usual,
the second component of a self-nancing strategy is given by (2.42), that is,
0
t
= V
t
()
1
t
S
t
(),
where V () is given by (2.43) with the initial condition V
0
() = E
Q
(Y ). In particular, we have that
0
0
= E
Q
(Y )
1
0
S
0
().
To complete the proof, that is, to show that V
t
() =

M
t
for every t [0, T], it suces to compare
the jumps of both processes at time (both martingales are stopped at ). It is clear from (2.49)
that the jump of

M equals

M = h() g(). Using (2.44), we get for the jump of the wealth
process
V () =
1
(R()

S
()) = h() g(),

and thus we conclude that V
t
() =

M
t
for every t [0, T]. In particular, is admissible and
V
T
() = V
T
() = h( T) = Y , so that replicates a claim Y . Note that if = (0) then
S
0
((0)) = 0, so that
0
0
= V
0
() = E
Q
(Y ).
Let us now analyze the condition

S
t
() = R(t) for every t [0, T]. It ensures, in particular,
that the wealth process V () has a non-zero jump at default time for any the self-nancing trading
strategy such that
1
t
= 0 for every t [0, T]. It appears that this condition is not restrictive, since
it is satised under mild assumptions.
Indeed, if > 0 and R is a non-increasing function then the inequality

S
t
() < R(t) is valid
for every t [0, T] (this follows easily from (2.26)). For instance, if (t) > 0 and the protection
payment R > 0 is constant then it is clear from (2.30) that the market rate (0) is strictly positive.
Consequently, formula (2.26) implies that

S
t
((0)) < R for every t [0, T], as was required. To
summarize, when a tradeable asset is a market CDS with a constant R > 0 and the default intensity
is strictly positive then the inequality holds. Let us nally observe that if the default intensity
vanishes on some set then we do not need to impose the inequality

S
t
() = R(t) on this set in order
to equate (2.52) with (2.53), since the desired equality holds anyway.
It is useful to note that the proof of Proposition 2.3.3 was implicitly based on the following
observation. In our case, Lemma 2.2.3 can be applied to the following H-martingales under Q:
M
1
= V (), that is, the wealth process of an admissible self-nancing strategy and M
2
=

M, that
is, the conjectured price of a claim Y , as given by the risk-neutral valuation formula.
The method presented above can be extended to replicate a defaultable claim (X, A, Z, ), where
X = c(T), A
t
=
_
t
0
a(u) du and Z
t
= h(t) for some RCLL functions a and h. In this case, it is natural
to expect that the cum-dividend price process
t
associated with a defaultable claim (X, A, Z, ), is
given by the formula, for every t [0, T],
t
= 11
{t<}
M
t
+ 11
{t<}
1
G(t)
_
T
t
a(u)G(u) du +
_
t
0
h(u) dH
u
+
_
t
0
a(u)(1 H
u
) du, (2.55)
where

M
t
= E
Q
(Y | H
t
) with Y is given by (2.45). Let us denote by
t
the corresponding ex-dividend
price, that is:
t
= 11
{t<}
M
t
+ 11
{t<}
1
G(t)
_
T
t
a(u)G(u) du. It is rather straightforward to verify
that
t
satises
t
= E
Q
(Y ) +
_
T
0
a(t)G(t) dt +
_
(0,t]
(h(u)
u
) dM
u
, t [0, T],
so that it is a martingale. Consequently, the dynamics of
t
are
d
t
= (h(t)
t
) dM
t
, t [0, T].
From this, or directly from (2.55), we see that the pre-default dynamics of process
t
are
d
t
= d
M
t
+(t)a(t) dt = (t)
_
h(t) g(t) a(t)
_
dt = (t)
_
h(t)

(t)
_
dt, t [0, ),
where we set a(t) = (G(t))
1
_
T
t
a(u)G(u) du and

(t) is the pre-default value of
t
. Note that
a(t) represents the pre-default value of the future promised dividends associated with A. Therefore,
arguing as in the proof of Proposition 2.3.3, we nd the following expression for the component
1
of a replicating strategy for a defaultable claim (X, A, Z, )
1
t
=
h(t) g(t) a(t)
R(t)

S
t
()
, t [0, T]. (2.56)
It is easy to see that the jump condition at time , mentioned in the second part of the proof
of Proposition 2.3.3, is also satised in this case. In fact, it is enough to observe that
=
h()
= h() g() a().

Remark 2.3.2 Of course, if we take as (X, A, Z, ) a CDS with rate and recovery function R ,
then we have h(t) = R(t) and g(t) +a(t) =

S
t
(), so that clearly
1
t
= 1 for every t [0, T].
The following immediate corollary to Proposition 2.3.3 is worth stating.
Corollary 2.3.2 Assume that

S
t
() = R(t) for every t [0, T]. Then the market is complete, in
the sense, that any defaultable claim (X, A, Z, ), where X = c(T), A
t
=
_
t
0
a(u) du and Z
t
= h(t)
for some constant c(T) and RCLL functions a and h, is attainable through continuous trading in a
CDS and a bond. The cum-dividend arbitrage price
t
of such defaultable claim satises, for every
t [0, T],
t
= V
t
() =
0
+
_
]0,t]
(h(u)
u
) dM
u
,
where
0
= E
Q
(Y ) +
_
T
0
a(t)G(t) dt,
with Y given by (2.45). Its pre-default price is (t) = g(t) +a(t) + A
t
, so that we have, for every
t [0, T]
t
= 11
{t<}
( g(t) +a(t) +A
t
) + 11
{t}
(h() +A
) = 11
{t<}
(t) + 11
{t}
.
Case of a Constant Default Intensity
As a partial check of the calculations above, we shall consider once again the case of constant
default intensity and constant protection payment. In this case, (0) = R and S
t
((0)) = 0 for
every t [0, T], so that
dV
t
() =
1
t
R dt =
1
t
(0) dt. (2.57)
Furthermore, for any RCLL function h, formula (2.54) yields
1
t
= R
1
_
h(t) +e
t
_
T
t
h(u) d
_
e
u
_
c(T)e
T
_
. (2.58)
Assume, for instance, that h(t) = R for t [0, T[ and c(T) = 0. Then (2.58) gives
1
t
= e
(Tt)
.
Since S
0
((0)) = 0, we have
0
0
=
0
(Y ) = V
0
() = R(1 e
T
). In view of (2.57), the gains/losses
from positions in market CDSs over the time interval [0, t] equal, on the set { > t},
V
t
() V
0
() = R
_
t
0
1
u
du = R
_
t
0
e
(Tu)
du = Re
T
_
e
t
1
_
< 0.
Suppose that default occurs at some date t [0, T]. Then the protection payments is collected, and
the wealth at time t becomes
V
t
() = V
t
() +
1
t
R = R(1 e
T
) Re
T
_
e
t
1
_
+Re
(Tt)
= R.
The last equality shows that the strategy is indeed replicating on the set { T}. On the set
{ > T}, the wealth at time T equals
V
T
() = R(1 e
T
) Re
T
_
e
T
1
_
= 0.
Since S
t
((0)) = 0 for every t [0, T], we have that
0
t
= V
t
() for every t [0, T].
Short Sale of a CDS
As usual, we assume that the maturity T of a CDS is xed and we consider the situation where the
default has not yet occurred.
1. Long position. We say that an agent has a long position at time t in a CDS if he owns at time
t a CDS contract that had been created (initiated) at time s
0
by some two parties and was sold to
the agent (by means of assignment for example) at time s. If s
0
= s then the agent is an original
counter-party to the contract, that is the agent owns the contract from initiation. If an agent owns
a CDS contract, the agent is entitled to receive the protection payment for which the agent pays
the premium. The long position in a contract may be liquidated at any time s < t < T by means of
assignment or osetting.
2. Short position. We stress that the short position, namely, selling a CDS contract to a dealer,
can only be created for a newly initiated contract. It is not possible to sell to a dealer at time t a
CDS contract initiated at time s
0
< t.
3. Osetting a long position. If an agent has purchased at time s
0
s < T a CDS contract
initiated at s
0
, he can oset his long position by creating a short position at time t. A new contract
is initiated at time t, with the initial price S
t
((s
0
)), possibly with a new dealer. This short position
osets the long position outstanding, so that the agent eectively has a zero position in the contract
at time t and thereafter.
4. Market constraints. The above taxonomy of positions may have some bearing on portfolios
involving short positions in CDS contracts. It should be stressed that not all trades involving a CDS
are feasible in practice. Let us consider the CDS contract initiated at time t
0
and maturing at time
T. Recall that the ex-dividend price of this contract for any t [t
0
, T[ is S
t
((t
0
)). This is the
theoretical price at which the contract should trade so to avoid arbitrage. This price also provides
substance for the P&L analysis as it really marks-to-market positions in the CDS contract.
Let us denote the time-t position in the CDS contract of an agent as
1
t
, where t [t
0
, T].
The strategy is subject to the following constraints:
1
t
0 if
1
t
0
0 and
1
t

1
t
0
if
1
t
0
0. It
is clear that both restrictions are related to short sale of a CDS. The next result shows that under
some assumptions a replicating strategy for a claim Y does not require a short sale of a CDS.
Corollary 2.3.3 Assume that

S
t
() < R(t) for every t [0, T]. Let h be a non-increasing function
and let c(T) h(T). Then
1
t
0 for every t [0, T].
Proof: It is enough to observe that if h be a non-increasing function and c(T) h(T) then it
follows easily from the rst equality in (2.24) that for the function g given by (2.48) we have that
h(t) g(t) for every t [0, T]. In view of (2.50), this shows that
1
t
0 for every t [0, T].
2.4 Successive default times
The previous results can easily be generalized to the case of successive default times. We assume in
this section that r = 0.
2.4.1 Two times
Let us rst study the case with two random times
1
,
2
. We denote by
(1)
= inf(
1
,
2
) and
(2)
= sup(
1
,
2
), and we assume, for simplicity, that P(
1
=
2
) = 0. We denote by (H
i
t
, t 0)
the default process associated with
i
, (i = 1, 2), and by H
t
= H
1
t
+H
2
t
the process associated with
two defaults. As before, H
i
is the ltration generated by the process H
i
and H is the ltration
generated by the process H. The -algebra G
t
= H
1
t
H
2
t
is equal to (
1
t) (
2
t). It is useful
to note that G
t
is strictly greater than H
t
.
Exemple: assume that
1
and
2
are independent and identically distributed. Then, obviously,
for u < t
P(
1
<
2
|
(1)
= u,
(2)
= t) = 1/2 ,
hence (
1
,
2
) = (
(1)
,
(2)
).
Computation of joint laws
A H
1
t
H
2
t
-measurable random variable is equal to
- a constant on the set t <
(1)
,
- a (
(1)
)-measurable random variable on the set
(1)
t <
(2)
, i.e., a (
1
)-measurable
random variable on the set
1
t <
2
, and a (
2
2
t <
1
- a (
1
,
2
2
t.
We note G the survival probability of the pair (
1
,
2
), i.e.,
G(t, s) = P(
1
> t,
2
> s) .
We shall also use the notation
g(s) =
d
ds
G(s, s) =
1
G(s, s) +
2
G(s, s)
where
1
G is the partial derivative of G with respect to the rst variable.
We present in a rst step some computations of conditional laws.
P(
(1)
> s) = P(
1
> s,
2
> s) = G(s, s)
P(
(2)
> t|
(1)
= s) =
1
g(s)
(
1
G(s, t) +
2
G(t, s)) , for t > s
We also compute conditional expectation in the ltration G = H
1
H
2
: For t < T
P(T <
(1)
|H
1
t
H
2
t
) = 11
t<
(1)
P(T <
(1)
)
P(t <
(1)
)
= 11
t<
(1)
G(T, T)
G(t, t)
P(T <
1
|H
1
t
H
2
t
) = 11
t<
1
P(T <
1
|H
2
t
)
P(t <
1
|H
2
t
)
+ 11
1
<t
= 11
t<
1
_
11
t<
2
P(T <
1
, t <
2
)
P(t <
1
, t <
2
)
+ 11
2
t
P(T <
1
|
2
)
P(t <
1
|
2
)
_
+ 11
1
<t
= 11
t<
1
_
11
t<
2
G(T, t)
G(t, t)
+ 11
2
<t
P(T <
1
|
2
)
P(t <
1
|
2
)
_
+ 11
1
<t
2.4. SUCCESSIVE DEFAULT TIMES 49
P(
(2)
T|H
1
t
H
2
t
) = 11
t<
(1)
P(t
(1)
<
(2)
< T)
P(t <
(1)
)
+ 11
1
t<
2
P(t <
2
< T|
1
)
P(t <
2
|
1
)
+11
2
t<
1
P(t <
1
< T|
2
)
P(t <
1
|
2
)
+ 11
(2)
<t
.
The computation of P(T <
1
|
2
) can be done as follows: the function h such that P(T <
1
|
2
) =
h(
2
) satises
E(h(
2
)(
2
)11
2
<t
) = E((
2
)11
2
<t
11
T<
1
)
for any function . This implies that (assuming that the pair (
1
,
2
) has a density f)
_
t
0
dvh(v)(v)
_

0
duf(u, v) =
_
t
0
dv(v)
_

T
duf(u, v)
or
_
t
0
dvh(v)(v)
2
G(0, v) =
_
t
0
dv(v)
2
G(T, v)
hence, h(v) =

2
G(T,v)
2
G(0,v)
.
We can also write
P(T <
1
|
2
= v) =
P(T <
1
,
2
dv)
P(
2
dv)
=
1
P(
2
dv)
d
dv
P(
1
> T,
2
> v) =

2
G(T, v)
2
G(0, v)
hence, on the set
2
< T,
P(T <
1
|
2
) = h(
2
) =

2
G(T,
2
)
2
G(0,
2
)
In the same way, for T > t
P(
1
T <
2
|H
1
t
H
2
t
)11
{
1
t<
2
}
= 11
{
1
t<
2
}
(
1
)
where satises
E((
1
)11
1
t<T<
2
) = E((
1
)(
1
)11
{
1
t<
2
}
)
for any function . In other terms
_
t
0
du(u)
_

T
dvf(u, v) =
_
t
0
du(u)(u)
_

t
dvf(u, v)
or
_
t
0
du(u)
1
G(u, T) =
_
t
0
du(u)(u)
1
G(u, t) .
This implies that
(u) =

1
G(u, T)
1
G(u, t)
P(
1
T <
2
|H
1
t
H
2
t
)11
{
1
t<
2
}
= 11
{
1
t<
2
}
1
G(
1
, T)
1
G(
1
, t)
.
Value of credit derivatives
We introduce dierent credit derivatives
A defaultable zero-coupon related to the default times D
i
delivers 1 monetary unit if
i
is greater
that T: D
i
(t, T) = E(11
{T<
i
}
|H
1
t
H
2
t
)
A contract which pays R
1
is one default occurs before T and R
2
if the two default occur before T:
CD
t
= E(R
1
11
{0<
(1)
T}
+R
2
11
{0<
(2)
T}
|H
1
t
H
2
t
)
We obtain
D
1
(t, T) = 11
{
1
>t}
_
11
{
2
t}
2
G(T,
2
)
2
G(t,
2
)
+ 11
{
2
>t}
G(T, t)
G(t, t)
_
(2.59)
D
2
(t, T) = 11
{
2
>t}
_
11
{
1
t}
1
G(
1
, T)
2
G(
1
, t)
+ 11
{
1
>t}
G(t, T)
G(t, t)
_
(2.60)
CD
t
= R
1
11
{
(1)
>t}
_
G(t, t) G(T, T)
G(t, t)
_
+R
2
11
{
(2)
t}
+R
1
11
{
(1)
t}
(2.61)
+R
2
11
{
(2)
>t}
_
I
t
(0, 1)
_
1

2
G(T,
2
)
2
G(t,
2
)
_
+I
t
(1, 0)
_
1

1
G(
1
, T)
1
G(
1
, t)
_
(2.62)
+I
t
(0, 0)
_
1
G(t, T) +G(T, t) G(T, T)
G(t, t)
__
(2.63)
where by
I
t
(1, 1) = 11
{
1
t,
2
t}
, I
t
(0, 0) = 11
{
1
>t,
2
>t}
I
t
(1, 0) = 11
{
1
t,
2
>t}
, I
t
(0, 1) = 11
{
1
>t,
2
t}
More generally, some easy computation leads to
E(h(
1
,
2
)|H
t
) = I
t
(1, 1)h(
1
,
2
) +I
t
(1, 0)
1,0
(
1
) +I
t
(0, 1)
0,1
(
2
) +I
t
(0, 0)
0,0
where
1,0
(u) =
1
1
G(u, t)
_

t
h(u, v)
1
G(u, dv)
0,1
(v) =
1
2
G(t, v)
_

t
h(u, v)
2
G(du, v)
0,0
=
1
G(t, t)
_

t
_

t
h(u, v)G(du, dv)
The next result deals with the valuation of a rst-to-default claim in a bivariate set-up. Let us
stress that the concept of the (tentative) price will be later supported by strict replication arguments.
In this section, by a pre-default price associated with a G-adapted price process , we mean here the
function such that
t
11
{
(1)
>t}
= (t)11
{
(1)
>t}
for every t [0, T]. In other words, the pre-default
price and the price coincide prior to the rst default only.
Denition 2.4.1 Let Z
i
be two functions, and X a constant. A FtD claim pays Z
1
(
1
) at time
1
if
1
< T,
1
<
2
, pays Z
2
(
2
) at time
2
if
2
< T,
2
<
1
, and X at maturity if
1

2
> T
Proposition 2.4.1 The pre-default price of a FtD claim (X, 0, Z,
(1)
), where Z = (Z
1
, Z
2
) and
X = c(T), equals
1
G(t, t)
_
_
T
t
Z
1
(u) G(du, u)
_
T
t
Z
2
(v)G(v, dv) +XG(T, T)
_
.
Proof: The price can be expressed as
E
Q
(Z
1
(
1
)11
{
1
T,
2
>
1
}
|H
t
) +E
Q
(Z
2
(
2
)11
{
2
T,
1
>
2
}
|H
t
) +E
Q
(c(T)11
{
(1)
>T}
|H
t
).
The pricing formula now follows by evaluating the conditional expectation, using the joint distribu-
tion of default times under the martingale measure Q.
Comments 2.4.1 Same computations appear in Kurtz and Riboulet [139]
2.5. FINANCIAL MARKET 51
2.4.2 Poisson Jumps
Suppose that the default times are modeled via a Poisson process with intensity h. (See the appendix
for denitions and main properties of Poisson processes) The terminal payo is

T
i
T
(1 R(T
i
)),
where R is a deterministic function valued in [0, 1]. The value of this payo is E(
T
i
T
(1R(T
i
))).
In the case of constant R(s) = R, we get
E
_
_

T
i
T
(1 R(T
i
))
_
_
= E
_
(1 R)
N
T
_
= exp
_
R
_
T
0
h(s)ds
_
.
In the general case,
E
_
_

T
i
T
(1 R(T
i
))
_
_
= E
_
_
exp
_
_
sT
ln(1 R(s))N
s
)
_
_
_
_
= E
_
exp
_
_
T
0
ln(1 R
s
)dN
s
__
.
Hence (See the appendix)
E
_
_

T
i
T
(1 R(T
i
))
_
_
= exp
_
_
T
0
R(s)h(s)ds
_
.
2.5 Financial Market
We now assume that some nancial market presents tradable assets. The total surprise occurs when
there are no information on the possible occurrence of default on the market, i.e. when the default
time is independent of the prices (we do not assume for the moment that a defaultable asset is
traded). If F = (F
t
, t 0) is the ltration generated by the prices, the quantity
G
P
(t, T) = P( > T|F
t
) = P( > T)
does not depend on t.Of course, this is not a reasonable hypothesis, however, this is a nice example
to understand the tools. It can be mentioned that in practise, a lot of models are in fact reduced to
this case (in particular, intensity models where the intensity is assumed to be deterministic)
The function G
P
(t, T) represents the survival probability of default, using the knowledge of the
market (which here corresponds, from the independence assumption to no information). We denote
by F
P
the cumulative function of , i.e.
F
P
(t) = P( t) = 1 G
P
(t) .
We assume in that follows that F
P
is dierentiable, and we denote its derivative (the density of )
by f
P
. More interesting , is the knowledge of the probability of the occurrence of the default, given
that the default has not occurred before time t, that is, for t < T
P( > T| > t) =
P( > T)
P( > t)
=
1 F
P
(T)
1 F
P
(t)
=
G
P
(T)
G
P
(t)
We shall denote
t
= exp
_
t
0
r
s
ds and
t
T
= exp
_
T
t
r
s
ds the discount factor, which, in case of
stochastic interest rate, diers from the price of a zero-coupon bond.
2.5.1 Hazard Function
We introduce
P
(t) = ln(1 F
P
(t)) the hazard function, an increasing function of the form
P
(t) =
_
t
0

P
(s)ds. The process H
t

_
t
0

P
(s)ds where
P
(s)ds =
f
P
(s)
1 F
P
(s)
can be easily shown to be a P martingale.
Of course, the hazard function depends on the choice of the probability, and the process
M
t
def
= H
t

_
t
0
(s)ds
where (s)ds =
f(s)
1F(s)
with F(t) = Q( t) is a Q martingale.
In that setting, one can not hedge defaultable claims, using the primary market, i.e., assets
traded in the underlying market, where the information does not include the knowledge of default
time. If this primary market is arbitrage free, there exists a risk neutral probability Q
F
. If the risk
neutral probability Q
F
is unique, this primary market is complete and arbitrage free. However, for
the moment, we can not extend Q
F
to a risk-neutral probability for the defaultable market: we recall
that a risk-neutral probability is such that the discounted prices are martingales, and we do not have
yet prices for defaultable claims. Hence, the market including defaultable assets is incomplete.
The risk-neutral survival probability of default Q is evaluated from the price of defaultable securities.
2.5.2 Defaultable zero-coupon bonds
Now, if a DZC of maturity T is traded in the market at price D(t, T), then the risk neutral probability
can be deduced from that price. Indeed, from the fundamental theorem of asset pricing, the value
of a DZC is the conditional expectation of discounted payo, i.e., noting that D(T, T) = 11
T<
, and
that
E
Q
(D(T, T)|F
T
) = E
Q
(11
T<
|F
T
) = Q(T < ) = G(T)
D(t, T) = E
Q
(D(T, T)
t
T
|G
t
) = 11
t<
E
Q
(D(T, T)
t
T
|F
t
)
E
Q
(11
t<
|F
t
)
= 11
t<
E
Q
(E
Q
(D(T, T)|F
T
)
t
T
|F
t
)
E
Q
(11
t<
|F
t
)
= 11
t<
G(T)E
Q
(
t
T
|F
t
)
G(t)
= 11
t<
G(T)B(t, T)
G(t)
Here Q is the pricing probability, chosen by the market. Of course, to avoid arbitrage opportuni-
ties, the restriction of Q to the primary market must be equal to Q
F
(or to one of the riskneutral
probability if there is no uniqueness of these probabilities)
2.5.3 Deterministic interest rate and deterministic recovery
In the particular case of deterministic interest rate, the dynamics of D are
dD(t, T) =
G(T)
G(t)
B(t, T)dH
t
(1 H
t
)B(t, T)
G(T)
G(t)
2
dG(t)
+(1 H
t
)
G(T)
G(t)
r(t)B(t, T)dt
= r(t)D(t, T)dt D(t
, T)(dH
t
(t)dt) = r(t)D(t, T)dt D(t
, T)dM
t
More generally, the price of a corporate bond with deterministic recovery R is
D
(R)
(t, T) = 11
t<
_
G(T)
G(t)
B(t, T) +
1
G(t)
_
T
t
B(t, s)R(s)dF(s)
_
2.5. FINANCIAL MARKET 53
The risk-neutral dynamics of a corporate bond is
dD
(R)
(t, T) =
_
r(t)D
(R)
(t, T) R(t)(t)(1 H
t
)
_
dt D
(R)
(t
, T)dM
t
The value at time t of a defaultable payo h(S
T
) is
E
Q
(11
>T
h(S
T
)
t
T
|G
t
) = 11
t<T
B(t, T)
E
Q
(11
>T
h(S
T
)|F
t
)
Q( > t)
In our model, the independence assumption and the hypothesis that the interest rate is deterministic
implies that
E
Q
(11
>T
h(S
T
)
t
T
|F
t
) = B(t, T)E
Q
(h(S
T
)E
Q
(11
>T
|F
T
)|F
t
)
= B(t, T)G
T
E
Q
(h(S
T
)|F
t
)
Particular case r = 0
In that case, we deduce that the price of a defaultable claim is the price of the default free payo
times the price of a defaultable ZC
E
Q
(11
>T
h(S
T
)|G
t
) = D(t, T)E
Q
(h(S
T
)|F
t
)
Assuming that a DZC is traded in the market, as well as the underlying asset S and that the
underlying default free market is complete, an hedging strategy for the defaultable payo is to hold
L
t
t
units of the underlying and E(h(S
T
)|F
t
)G
1
T
units of DZC. Indeed, the price of the defaultable
claim is
Y
t
= D(t, T)V
t
= 11
t<
G(t)
1
G(T)E
Q
(h(S
T
)|F
t
)
hence, denoting by the hedging strategy for the default free contingent claim h(S
T
), i.e. the
process such that
h(S
T
) = h +
_
T
0
s
dS
s
dY
t
= V
t
dD(t, T) +D(t, T)dV
t
= V
t
dD(t, T) +D(t, T)
t
dS
t
2.5.4 Stochastic interest rate
In the case of stochastic interest rate, we assume that the dynamics of the default free ZC is
dB(t, T) = B(t, T) (r(t)dt +(t, T)dW
t
)
Then, the dynamics of D is
dD(t, T) =
G(T)
G(t)
B(t, T)dH
t
(1 H
t
)B(t, T)
G(T)
G(t)
2
dG(t)
+(1 H
t
)
G(T)
G(t)
d
t
B(t, T)
= D(t, T)(r(t)dt +(t, T)dW
t
) D(t
, T)(dH
t
(t)dt)
= r(t)D(t, T)dt +D(t
, T)((t, T)dW
t
dM
t
)
Pricing defaultable contingent claims:
We now consider a defaultable contingent claim of payo 11
>T
h(S
T
) and whose the price is
E
Q
(11
>T
t
T
h(S
T
)|G
t
) = 11
t<
G(t)
1
G(T)E
Q
(h(S
T
)
t
T
|F
t
)
= 11
t<
Q( > T|t > )E
Q
(h(S
T
)
t
T
|F
t
)
= 11
t<
exp(
_
T
t
(s)ds) E
Q
(h(S
T
)
t
T
|F
t
) .
The price of the defaultable contingent claim if the product of the price of the corresponding default
free payo and of the conditional survival probability. A more interesting way of writing this equality
is
E
Q
(11
>T
t
T
h(S
T
)|G
t
) = 11
t<
E
Q
(h(S
T
) exp(
_
T
t
(r
s
+(s))ds)|F
t
)
The price of a defaultable contingent claim is the price of the same promised payo h(S
T
) in a world
where the interest rate is r +.
Hedging defaultable contingent claims:
We assume that the primary market is complete. Let

h
t
= E
Q
(11
>T
t
T
h(S
T
)|G
t
) be the discounted
price of the defaultable claim, and

h
F
t
= E
Q
(h(S
T
)
t
T
|F
t
) be the discounted price of the payo
h(S
T
). Then,

h
t
= 11
t<
G(T)
G(t)
h
F
t
= L
t
h
F
t
with L
t
= 11
t<
G(T)
G(t)
. In particular, if

D(t, T) and

B(t, T)
are the discounted prices of DZC and ZC

D(t, T) = L
t

B(t, T). Using that dL
t
= L
t
dM
t
, one gets
d
h
t
=
h
F
t
L
t
dM
t
+L
t
d
h
F
=
1
B(t, T)
_
h
F
t
d
D(t, T) L
t
h
F
t
d
B(t, T)
_
+L
t
d
h
F
t
.
The primary market being complete d
h
F
t
=
t
d
S
t
. The following portfolio, constructed on the DZC,
the ZC and the underlying asset
1
B(t, T)
h
F
t
,
1
B(t, T)
h
F
t
L
t
, L
t
t
is self-nancing since
1
B(t, T)
h
F
t

D(t, T)
1
B(t, T)
h
F
t

B(t, T) +L
t
S
t
=

h
t

Hazard Function Approach: 2.1 The Toy Model

Uploaded by

Copyright:

Available Formats

You might also like

Hazard Function Approach: 2.1 The Toy Model

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hazard Function Approach: 2.1 The Toy Model

Uploaded by

Copyright:

Available Formats

Chapter 2

Hazard Function Approach

satises f(t) = (t)e

-measurable, so that Y = h() for some Borel measurable

Proposition 2.2.5 Assume that is a continuous function. Let h : IR

Exercise 2.2.2 If is not continuous, prove that

be an arbitrary probability measure on (, H

), which is absolutely continuous with respect

is equivalent to P if and only if the inequality

( > t) > 0 for any

absolutely continuous with respect to P.

is well dened. Our goal is to examine

and . It is easily seen that in general we have

(t)) where the c.d.f. F

follows an absolutely continuous function. Therefore, the intensity function

exists, and it is given by the formula

is also continuous and d

, let us introduce the following non-negative

and P are equivalent)

(u) = h(u)/g(u). We conclude that

(t)1. Therefore, the martingale property of the process , which is obvious

be any probability measure on (, H

) absolutely continuous with respect

( > t) > 0 for every t IR

(t) = h(t)/g(t), g(t) = e

is also continuous, and it

. Alternatively, we may check directly that the product U

(t) d(t) established in

is a continuous function then the process M

. The next result suggests that this martingale property uniquely

is given by formula (2.15) for some

(t ) follows a H-martingale under P

given by formula (2.21).

(u) 1)h(u) dF(u) +

h(u) dG(u). (2.24)

(t). Therefore, solving this equation leads to

, as soon as the cumulative function of is dierentiable. Due and Lando

() , the proof of (2.34) is complete. To prove the

() (recall that the ex-dividend price S() drops to zero at

h(t) = h(t) for t T and

()) = h() g(),

= h() g() a().

You might also like