CREDIT RISK: MODELLING, VALUATION

AND HEDGING
Marek Rutkowski
Faculty of Mathematics and Information Science
Warsaw University of Technology
00-661 Warszawa, Poland
[email protected]
1. VALUE-OF-THE-FIRM APPROACH
2. INTENSITY-BASED APPROACH
3. MODELLING OF DEPENDENT DEFAULTS
4. CREDIT RATINGS AND MIGRATIONS
Winter School on Financial Mathematics
Oud Poelgeest, December 16-18, 2002
CREDIT RATINGS AND MIGRATIONS
1 Models Inputs
1.1 Term Structure of Credit Spreads
1.1.1 Credit Classes
1.1.2 Credit Spreads
1.1.3 Spot Martingale Measure P

1.1.4 Zero-Coupon Bonds

1.1.5 Conditional Dynamics of Bonds Prices
1.2 Recovery Schemes
2 Credit Migration Process
3 Defaultable Term Structures
3.1 Single Credit Ratings Case
3.1.1 Credit Migrations
3.1.2 Martingale Dynamics of a Defaultable ZCB
3.1.3 Risk-Neutral Representations
3.2 Multiple Credit Ratings Case
3.2.1 Credit Migrations
3.2.2 Martingale Dynamics of a Defaultable ZCB
3.2.3 Risk-Neutral Representations
3.3 Statistical Probability
3.3.1 Market Prices for Risks
3.3.2 Statistical Default Intensities
SELECTED REFERENCES
R. Jarrow, D. Lando and S. Turnbull: A Markov model for the
term structure of credit risk spreads. Review of Financial
Studies 10 (1997), 481523.
M. Kijima and K. Komoribayashi (1998) A Markov chain model
for valuing credit risk derivatives. Journal of Derivatives 6,
Fall, 97108.
D. Lando (2000) Some elements of rating-based credit risk
modeling. In: Advanced Fixed-Income Valuation Tools, J.
Wiley, Chichester, pp. 193215.
D. Lando (2000) On correlated defaults in a rating-based model:
common state variables versus simultaneous defaults. Preprint,
University of Copenhagen.
P. Schonbucher: Credit risk modelling and credit derivatives.
Doctoral dissertation, University of Bonn, 2000.
T.R. Bielecki and M. Rutkowski: Defaultable term structure:
Conditionally Markov approach. Preprint, 2000.
T.R. Bielecki and M. Rutkowski: Multiple ratings model of
defaultable term structure. Mathematical Finance 10 (2000).
R. Douady and M. Jeanblanc: A rating-based model for credit
derivatives, Preprint, 2002
1. MODELS INPUTS
Standard intensity-based approach (as, for instance, in Jarrow
and Turnbull (1995) or Jarrow, Lando and Turnbull (1997))
relies on the following assumptions:
existence of the martingale measure Q

is postulated,
the relationship between the statistical probability P and the
risk-neutral probability Q

derived via calibration,

credit migrations process is modelled as a Markov chain,
market and credit risk are separated (independent).
The HJM-type model of defaultable term structures with mul-
tiple ratings was proposed by Bielecki and Rutkowski (2000)
and Schonbucher (2000).
This approach:
formulates sucient consistency conditions that tie together
credit spreads and recovery rates in order to construct a risk-
neutral probability Q

and the corresponding risk-neutral in-

tensities of credit events,
shows how the statistical probability P and the risk-neutral
probability Q

are connected via the market price of interest

rate risk and the market price of credit risk,
combines market and credit risks.
1.1 Term Structure of Credit Spreads
We are given a ltered probability space (, F, P) endowed with
a d-dimensional standard Brownian motion W.
Remark. We may assume that the ltration F = F
W
.
For any xed maturity 0 < T T

the price of a zero-coupon

Treasury bond equals
B(t, T) = exp
_

_
T
t
f(t, u) du
_
,
where the default-free instantaneous forward rate f(t, T) pro-
cess is subject to the standard HJM postulate.
(HJM) The dynamics of the instantaneous forward rate f(t, T)
are given by, for t T,
f(t, T) = f(0, T) +
_
t
0
(u, T) du +
_
t
0
(u, T) dW
u
for some deterministic function f(0, ) : [0, T

] R, and
some F-adapted stochastic processes
: A R, : A R
d
,
where A = {(u, t) | 0 u t T

}.
1.1.1 Credit Classes
Suppose there are K 2 credit rating classes, where the K
th
class corresponds to the default-free bond.
For any xed maturity 0 < T T

, the defaultable instan-

taneous forward rate g
i
(t, T) corresponds to the rating class
i = 1, . . . , K 1. We assume that:
(HJM
i
) The dynamics of the instantaneous defaultable forward
rates g
i
(t, T) are given by, for t T,
g
i
(t, T) = g
i
(0, T) +
_
t
0

i
(u, T) du +
_
t
0

i
(u, T) dW
u
for some deterministic functions g
i
(0, ) : [0, T

] R, and
some F-adapted stochastic processes

i
: A R,
i
: A R
d
1.1.2 Credit Spreads
We assume that
g
K1
(t, T) > g
K2
(t, T) > . . . > g
1
(t, T) > f(t, T)
for every t T.
Denition 1 For every i = 1, 2, . . . , K1, the credit spread
equals s
i
(t, T) = g
i
(, T) f(, T).
1.1.3 Spot Martingale Measure P

The following condition excludes arbitrage across default-free

bonds for all maturities T T

and the savings account:

(M) There exists an F-adapted R
d
-valued process such that
E
P
_
exp
_
_
T

0

u
dW
u

1
2
_
T

0
|
u
|
2
du
__
= 1
and, for any maturity T T

, we have

(t, T) =
1
2
|

(t, T)|
2

(t, T)
t
where

(t, T) =
_
T
t
(t, u) du

(t, T) =
_
T
t
(t, u) du.
Let be some process satisfying Condition (M). Then the prob-
ability measure P

, given by the formula

dP

dP
= exp
_
_
T

0

u
dW
u

1
2
_
T

0
|
u
|
2
du
_
, P-a.s.,
is a spot martingale measure for the default-free term struc-
ture.
1.1.4 Zero-Coupon Bonds
The price of the T-maturity default-free zero-coupon bond
(ZCB) is given by the equality
B(t, T) := exp
_

_
T
t
f(t, u) du
_
.
Formally, the Treasury bond corresponds to credit class K.
Conditional value of T-maturity defaultable ZCB belonging
at time t to the credit class i = 1, 2, . . . , K 1, equals
D
i
(t, T) := exp
_

_
T
t
g
i
(t, u) du
_
.
We consider discounted price processes
Z(t, T) = B
1
t
B(t, T), Z
i
(t, T) = B
1
t
D
i
(t, T),
where B
t
is the usual discount factor (savings account)
B
t
= exp
_
_
t
0
f(u, u) du
_
.
Let us dene a Brownian motion W

under P

by setting
W

t
= W
t

_
t
0

u
du, t [0, T

].
1.1.5 Conditional Dynamics of Bonds Prices
Lemma 1 Under the spot martingale measure P

, for any
xed maturity T T

, the discounted price processes Z(t, T)

and Z
i
(t, T) satisfy
dZ(t, T) = Z(t, T)b(t, T) dW

t
,
where b(t, T) =

(t, T), and

dZ
i
(t, T) = Z
i
(t, T)(
i
(t) dt + b
i
(t, T) dW

t
)
where

i
(t) = a
i
(t, T) f(t, t) + b
i
(t, T)
t
and
a
i
(t, T) = g
i
(t, t)

i
(t, T) +
1
2
|

i
(t, T)|
2
b
i
(t, T) =

i
(t, T).
Remark 1 Observe that usually the process Z
i
(t, T) does not
follow a martingale under the spot martingale measure P

. This
feature is related to the fact that it does not represent the
(discounted) price of a tradable security.
1.2 Recovery Schemes
Let Y denote the cash ow at maturity T and let Z be the
recovery process (an F-adapted process). We take K = 2.
FRTV: Fractional Recovery of Treasury Value
Fixed recovery at maturity scheme. We set Z
t
= B(t, T) and
thus
Y = 11
{>T}
+ 11
{T}
.
FRPV: Fractional Recovery of Par Value
Fixed recovery at time of default. We set Z
t
= , where is a
constant. Thus
Y = 11
{>T}
+ B
1
(, T) 11
{T}
.
FRMV: Fractional Recovery of Market Value
The owner of a defaultable ZCB receives at time of default a
fraction of the bonds market value just prior to default. We
set Z
t
= D(t, T), where D(t, T) is the pre-default value of
the bond. Thus
Y = 11
{>T}
+ D(, T)B
1
(, T) 11
{T}
.
2 CREDIT MIGRATION PROCESS
We assume that the set of rating classes is K = {1, . . . , K},
where the class K corresponds to default. The migration
process C will be constructed as a (nonhomogeneous) condi-
tionally Markov process on K. Moreover, the state K will be
the unique absorbing state for this process.
Let us denote by F
C
t
the -eld generated by C up to time
t. A process C is conditionally Markov with respect to the
reference ltration F if for arbitrary s > t and i, j K we have
Q

_
C
t+s
= i | F
t
F
C
t
_
= Q

(C
t+s
= i | F
t
{C
t
= j} ) .
The probability measure Q

is the extended spot martingale

measure.
The formula above will provide the risk-neutral conditional pro-
bability that the defaultable bond is in class i at time t + s,
given that it was in the credit class C
t
at time t.
We introduce the default time by setting
= inf {t R
+
: C
t
= K}.
For any date t, we denote by

C
t
the previous bonds rating.
3 DEFAULTABLE TERM STRUCTURE
3.1 Single Rating Class (K = 2)
We assume the FRTV scheme (other recovery schemes can also
be covered, though).
Our rst goal is to derive the equation that is satised by the
risk-neutral intensity of default time.
Intensity of Default Time
We introduce the risk-neutral default intensity
1,2
as a so-
lution to the no-arbitrage equation
(Z
1
(t, T) Z(t, T))
1,2
(t) = Z
1
(t, T)
1
(t).
It is interesting to notice that for = 0 (zero recovery) we
have simply

1,2
(t) =
1
(t), t [0, T].
On the other hand, if we take > 0 then the process
1,2
is
strictly positive provided that
D(t, T) > B(t, T), t [0, T].
Recall that we have assumed that D(t, T) < B(t, T).
3.1.1 Credit Migrations
Since K = 2, the migration process C lives on two states.
The state 1 is the pre-default state, and the state 2 is the
absorbing default state. We may and do assume that C
0
= 1.
We postulate that the conditional intensity matrix for the pro-
cess C is given by the formula

t
=
_
_
_

1,2
(t)
1,2
(t)
0 0
_
_
_
.
For = 0, the matrix takes the following simple form

t
=
_
_
_

1
(t)
1
(t)
0 0
_
_
_
.
The default time now equals
= inf {t R
+
: C
t
= 2 }.
It is dened on an enlarged probability space
(

, F
T

, Q

) := (

, F
T

F, P

Q)
where the probability space (

,

F, Q) is large enough to support
a unit exponential random variable, say. Then
= inf {t R
+
:
_
t
0

1,2
(u) du }.
Hypotheses (H)
All processes and ltrations may always be extended past the
horizon date T

by constancy.
We set H
t
= 11
{t}
and we denote by H the ltration gener-
ated by the process H:
H
t
= (H
u
: u t).
In other words, H is the ltration associated with the observa-
tions of the default time.
It is clear that in the present setup
G = F H.
It is not dicult to check that the hypotheses (H.1)-(H.3) hold
in the present context.
In the general case of a model with multiple ratings, the ltra-
tion H will be generated by the migrations process C, that is,
we shall set
H
t
= (C
u
: u t).
Due to the judicious construction of the migration process C,
the hypotheses (H.1)-(H.2) remain valid in the case of multiple
ratings.
3.1.2 Martingale Dynamics of a Defaultable ZCB
Thanks to the consistency equation, the process
M
1,2
(t) := H
t

_
t
0

1,2
(u)(1 H
u
) du
is a martingale under Q

relative to the enlarged ltration G.

Recall that for any t [0, T] we have
D(t, T) = exp
_

_
T
t
g(t, u) du
_
and that D(t, T) is interpreted as the pre-default value of a T-
maturity defaultable ZCB that is subject to the FRTV scheme.
In other words, D(t, T) is understood as the value of a T-
maturity defaultable ZCB conditioned on the event: the bond
has not defaulted by the time t.
Recall that
Z
1
(t, T) = B
1
t
D(t, T)
and
Z(t, T) = B
1
t
B(t, T).
Auxiliary Process

Z(t, T)
We introduce an auxiliary process

Z(t, T), t [0, T],

Z(t, T) = 11
{>t}
Z
1
(t, T) + 11
{t}
Z(t, T).
It can be shown that

Z(t, T) satises the SDE (A)
d

Z(t, T) = Z
1
(t, T)b
1
(t, T) 11
{>t}
dW

t
+ Z(t, T)b(t, T) 11
{t}
dW

t
+ (Z(t, T) Z
1
(t, T)) dM
1,2
(t).
Notice that

Z(t, T) follows a G-martingale under Q

.
This leads to construction of an arbitragefree model of the
defaultable term structure and to risk-neutral representation
for the price of the defaultable bond.
We introduce the price process through the following denition.
Denition 2 The price process D
C
(t, T) of a T-maturity
ZCB is given by
D
C
(t, T) = B
t

Z(t, T).
3.1.3 Risk-Neutral Representations
Proposition 1 The price D
C
(t, T) of a defaultable ZCB
satises
D
C
(t, T) = 11
{>t}
D(t, T) + 11
{t}
B(t, T).
D
C
(t, T) = 11
{C
t
=1}
exp (
_
T
t
g(t, u) du)
+ 11
{C
t
=2}
exp (
_
T
t
f(t, u) du).
Moreover, the risk-neutral valuation formula holds
D
C
(t, T) = B
t
E
Q

(B
1
T
11
{T}
+ B
1
T
11
{>T}
| G
t
).
Furthermore
D
C
(t, T) = B(t, T) E
Q
T
( 11
{T}
+ 11
{>T}
| G
t
)
where Q
T
is the T-forward measure associated with Q

.
Special cases:
For = 0, we obtain D
C
(t, T) = 11
{>t}
D(t, T).
For = 1, we have, as expected, D
C
(t, T) = B(t, T).
Default-Risk-Adjusted Discount Factor
The default-risk-adjusted discount factor equals

B
t
= exp (
_
t
0
(r
u
+
1,2
(u)) du)
and we set

B(t, T) =

B
t
E
P

(

B
1
T
| F
t
).
We consider a bond with FRTV.
Proposition 2 We have
D
C
(t, T) = B(t, T) + (1 ) 11
{>t}

B(t, T)
and thus
D
C
(t, T) = B(t, T) (1 )
_
B(t, T) 11
{>t}

B(t, T)
_
.
Interpretation:
A decomposition of D
C
(t, T) of the price of a defaultable
ZCB into its predicted post-default value B(t, T) and the
pre-default premium D
C
(t, T) B(t, T).
A decomposition D
C
(t, T) as the dierence between its
default-free value B(t, T) and the expected loss in value
due to the credit risk. From the buyers perspective: the
price D
C
(t, T) equals the price of the default-free bond mi-
nus a compensation for the credit risk.
3.2 Multiple Credit Ratings Case
We work under the FRTV scheme. To each credit rating i =
1, . . . , K1, we associate the recovery rate
i
[0, 1), where

i
is the fraction of par paid at bonds maturity, if a bond
belonging to the i
th
class defaults.
As we shall see shortly, the notation

C

indicates the rating of

the bond just prior to default. Thus, the cash ow at maturity
is
X = 11
{>T}
+

11
{T}
.
To simplify presentation we let K = 3 (two dierent credit
classes) and we let
i
[0, 1) for i = 1, 2. The results carry
over to the general case of K 2.
3.2.1 Credit Migrations
Risk-neutral intensities of credit migrations
1,2
(t),
1,3
(t),
2,1
(t)
and
2,3
(t) are specied by the no-arbitrage condition:

1,2
(t)(Z
2
(t, T) Z
1
(t, T)) +
1,3
(t)(
1
Z(t, T) Z
1
(t, T))
+
1
(t)Z
1
(t, T) = 0,

2,1
(t)(Z
1
(t, T)

Z
2
(t, T)) +
2,3
(t)(
2
Z(t, T) Z
2
(t, T))
+
2
(t)Z
2
(t, T) = 0.
If the processes
1,2
(t),
1,3
(t),
2,1
(t) and
2,3
(t) are non-
negative, we construct a migration process C, on some enlarged
probability space (

, G, Q

), with the conditional intensity ma-

trix
(t) =
_
_
_
_
_
_
_

1,1
(t)
1,2
(t)
1,3
(t)

2,1
(t)
2,2
(t)
2,3
(t)
0 0 0
_
_
_
_
_
_
_
where
i,i
(t) =

j=i

i,j
(t) for i = 1, 2. Notice that the
transition intensities
i,j
follow F-adapted stochastic processes.
The default time is given by the formula
= inf{ t R
+
: C
t
= 3 }.
3.2.2 Martingale Dynamics of a Defaultable ZCB
We set H
i
(t) = 11
{C
t
=i}
for i = 1, 2, and we let H
i,j
(t) repre-
sent the number of transitions from i to j by C over the time
interval (0, t].
It can be shown that the process
M
i,j
(t) := H
i,j
(t)
_
t
0

i,j
(s)H
i
(s) ds, t [0, T],
for i = 1, 2 and j = i, is a martingale on the enlarged proba-
bility space (

, G, Q

).
Auxiliary Process

Z(t, T)
We introduce the process

Z(t, T) as a solution to the following
SDE (A)
d

Z(t, T) = (Z
2
(t, T) Z
1
(t, T)) dM
1,2
(t)
+ (Z
1
(t, T) Z
2
(t, T)) dM
2,1
(t)
+ (
1
Z(t, T) Z
1
(t, T)) dM
1,3
(t)
+ (
2
Z(t, T) Z
2
(t, T)) dM
2,3
(t)
+ H
1
(t)Z
1
(t, T)b
1
(t, T) dW

t
+ H
2
(t)Z
2
(t, T)b
2
(t, T) dW

t
+ (
1
H
1,3
(t) +
2
H
2,3
(t))Z(t, T)b(t, T) dW

t
,
with the initial condition

Z(0, T) = H
1
(0)Z
1
(0, T) + H
2
(0)Z
2
(0, T).
The process

Z(t, T) follows a martingale on (

, G, Q

), and
thus Q

is called the extended spot martingale measure.

The proof of the next result employs the no-arbitrage condition.
Lemma 2 For any maturity T T

, we have

Z(t, T) = 11
{C
t
=3}
Z
C
t
(t, T) + 11
{C
t
=3}

C
t
Z(t, T)
for every t [0, T].
Price of a Defaultable ZCB
We introduce the price process of a T-maturity defaultable ZCB
by setting D
C
(t, T) = B
t

Z(t, T) for any t [0, T].
In view of Lemma 2, the price of a defaultable ZCB equals
D
C
(t, T) = 11
{C
t
=3}
D
C
t
(t, T) + 11
{C
t
=3}

C
t
B(t, T)
with some initial condition C
0
{1, 2}. An analogous formula
can be established for an arbitrary number K of rating classes,
namely,
D
C
(t, T) = 11
{C
t
=K}
D
C
t
(t, T) + 11
{C
t
=K}

C
t
B(t, T).
Properties of D
C
(t, T):
D
C
(t, T) follows a (Q

, G)-martingale, when discounted by

the savings account.
In contrast to the conditional price processes D
i
(t, T),
the process D
C
(t, T) admits discontinuities, associated with
changes in credit quality.
It represents the price process of a tradable security: the
defaultable ZCB of maturity T.
3.2.3 Risk-Neutral Representations
Recall that
i
[0, 1) is the recovery rate for a bond which is
in the i
th
rating class prior to default.
The price process D
C
(t, T) of a T-maturity defaultable ZCB
equals
D
C
(t, T) = 11
{C
t
=3}
B(t, T) exp (
_
T
t
s
C
t
(t, u) du)
+ 11
{C
t
=3}

C
t
B(t, T)
where s
i
(t, u) = g
i
(t, u) f(t, u) is the i
th
credit spread.
Proposition 3 The price process D
C
(t, T) satises the risk-
neutral valuation formula
D
C
(t, T) = B
t
E
Q

C
T
B
1
T
11
{T}
+ B
1
T
11
{>T}
| G
t
).
It is also clear that
D
C
(t, T) = B(t, T) E
Q
T
(

C
T
11
{T}
+ 11
{>T}
| G
t
)
where Q
T
stands for the T-forward measure associated with
the extended spot martingale measure Q

.
3.3 Statistical Probability
We shall now change, using a suitable generalization of Gir-
sanovs theorem, the measure Q

to the equivalent probability

measure Q.
In the nancial interpretation, the probability measure Q will
play the role of the statistical probability.
It is thus natural to postulate that the restriction of Q to the
original probability space necessarily coincide with the statis-
tical probability P for the default-free market.
Condition (L): We set
dQ
dQ

= L
T

, Q

-a.s.,
where the Q

-local positive martingale L is given by the formula

dL
t
= L
t

t
dW

t
+ L
t
dM
t
, L
0
= 1,
and the Q

-local martingale M equals

dM
t
=

i=j

i,j
(t) dM
i,j
(t)
=

i=j

i,j
(t) (dH
i,j
(t)
i,j
(t)H
i
(t) dt)
for some processes
i,j
> 1.
3.3.1 Prices for Market and Credit Risks
For any i = j we denote by
i,j
> 1 an arbitrary nonnegative
F-predictable process such that
_
T

0
(
i,j
(t) + 1)
i,j
(t) dt < , Q

-a.s.
We assume that E
Q

(L
T

) = 1, so that the probability measure

Q is well dened on (

, G
T

).
Financial interpretations:
The process corresponds to the market price of interest
rate risk.
Processes
i,j
represent the market prices of credit risk.
Let us dene processes
Q
i,j
by setting for i = j

Q
i,j
(t) = (
i,j
(t) + 1)
i,j
(t)
and

Q
i,i
(t) =

j=i

Q
i,j
(t).
3.3.2 Statistical Default Intensities
Proposition 4 Under an equivalent probability Q, given by
Condition (L), the process C is a conditionally Markov process.
The matrix of conditional intensities of C under Q equals

Q
t
=
_
_
_
_
_
_
_
_
_
_
_
_

Q
1,1
(t) . . .
Q
1,K
(t)
. . . . .

Q
K1,1
(t) . . .
Q
K1,K
(t)
0 . . . 0
_
_
_
_
_
_
_
_
_
_
_
_
.
If the market price for the credit risk depends only on the cur-
rent rating i (and not on the rating j after jump), so that

i,j
=
i,i
=:
i
for every j = i
then
Q
t
=
t

t
, where
t
= diag [
i
(t)] with
i
(t) =
i
(t)+1
is the diagonal matrix (see, e.g., Jarrow, Lando and Turnbull
(1997).
Important issues:
Valuation of defaultable coupon-bonds.
Modelling of correlated defaults (dependent migrations).
Valuation and hedging of credit derivatives.
Calibration to liquid instruments.