Anlong Li
Barclays Capital Inc.
New York, NY
Abstract: This paper studies the pricing of options whose payoffs are contin-
gent on constant maturity credit default swap (CMCDS) spreads. We extend
the convexity adjustment method for constant maturity swap (CMS) in interest
rates by modeling the swap rate and CDS spread either as a single factor (a
sum of the two) or as two factors separately. For the latter, we explicitly model
the correlation between interest rate and credit spread in deriving our results. It
is shown that when swap rate and CDS spread are independent of each other
and when the yield curve is relatively flat, the two-factor model gives the same
convexity adjustment as the one-factor model. Under lognormality assump-
tions, the model can produce analytic solutions for convexity adjustments as
well as for the valuation of CMCDS derivatives such as CDS swaptions, caps
and floors, and digital options.
1. Introduction
Similar to constant maturity swap (CMS) in interest rates, constant maturity credit default
swap (CMCDS) refers to a credit default swap where each premium payment is indexed to the
market spread of a CDS with constant tenor (maturity). The premium rate (or CDS spread) is
not known until the index (CDS spread) settles. For example, in a 10-year CMCDS indexed to
a constant maturity of 5 years, the premium rates (5 year CDS spreads) form a time series that
tracks the cost of buying 5 year default protection over the next 10 years.
In a CMCDS contract, each CDS spread is only applied to one premium payment which can
be viewed as the first premium in a forward CDS contract. In a forward-starting CDS contract,
however, the CDS spread is applied to all future premium payment periods. The valuation of
the first premium in isolation is much more difficult than that of all the premiums as a
whole. This is because that premiums at the CMCDS rate are earned over the entire term of
the CDS contract. There is a timing mismatch. A simple example of such mismatch is the
Libor-in-arrears swap studied by Li and Raghavan (1996).
Although CDS spreads are not martingales under the risk neutral measure associated with the
money market numeraire, we can make them so by changing the numeraire to the risky annu-
ity or PV01 of the underlying CDS. This approach has been used in the literature to justify the
use in practice of the Black (1976) model for pricing credit default swaptions.
Another way to view this is that, like CMS rate, the CMCDS rate is not a directly traded asset.
As a result, the forward CDS rate (the fair rate to enter into the forward-starting CDS today) is
not an unbiased estimate of the future CDS rate under the risky forward measure. The differ-
ence is known as the convexity adjustment.
For interest rate CMS, this adjustment is well-known. There are two general approaches to
this problem. The first is to use an arbitrage-free term structure model of interest rates where
the dynamics of the CMS rate is fully determined. The term structure model can be for the
short rate as in Li (1995), or the forward rate as in Li, Ritchken and Sankarasubramanian
(1995). The second approach, more common in practice, is to model swap rate directly as in
Pelsser (2001) and Hagan (2004).
In this paper, we will extend the second approach to the case of CMCDS. We first model the
sum of CMS and CMCDS rates in a one factor framework. Then we treat them as two corre-
lated processes. We showed that when the processes are independent of each other and the
yield curve is relatively flat, the two factor model gives the same convexity adjustment as the
one factor model.
Although each CDS rate is modeled as a diffusion on its own it does not mean that the model
admits arbitrage. It is somewhat difficult to reconcile the two approaches. For such an attempt,
we refer our readers to Brigo (2005). Another common practice that could introduce arbitrage
is to assume the CDS rate is lognormally distributed both before and after the convexity
adjustment. This can be avoided by apply the model directly to option valuation rather than
using the convexity adjustment, as we will do in this paper.
We proceed as follows. Section 2 reviews convexity adjustment methods for CMS. Section 3
extends the results to CMCDS, using both a one-factor and a two factor model. In Section 4
we discuss the valuation of CMCDS swaps and options, while Section 5 concludes the paper.
An alternative proof of the main result can be found in the appendix.
Consider an N-period swap starts at time T 0 with fixed payment dates Ti , i = 1, , N . Let
P ( t, T ) be the discount factor from t to T . The fair rate of this swap at time t T 0 is defined as
P ( t, T 0 ) P ( t, T N )
R s ( t ) = --------------------------------------------- (1)
As ( t )
where
N
As ( t ) = P ( t, T )
i i (2)
i=1
is the annuity (or DV01) of the swap, and i is the day-count-fraction for period [ T i 1, T i ] .
Let ( , , t, Q ) be the filtered probability space on a finite time interval [ 0, T * ] . The two
equivalent martingale measures Q T and Q A are associated with numraires P ( t, T ) and A s ( t ) ,
respectively. Since P ( t, T0 ) P ( t, T N ) is the price difference between two traded securities, fol-
lowing Harrison and Kreps (1979) or Harrison and Pliska (1981), R s ( t ) in Equation (1) is a
martingale under Q A . Thus
EQA [ Rs ( t ) 0 ] = Rs ( 0 ) (3)
From now on we will drop the conditioning -algebra 0 whenever no confusion arises. In
practice, it is convenient to assume R s ( t ) follows a geometric Brownian motion
s2
R s ( t ) = R s ( 0 ) exp s W s ( t ) ------ t (4)
2
where W s ( t ) is a standard Brownian motion under Q A , and s is the volatility of the swap rate.
This leads naturally to the popular Black (1976) model for pricing European swaptions.
However, to price CMS options, we are interested in the distribution of R s ( t ) under the termi-
nal measure Q T . For example, the present value of a payment in the amount of R s ( T0 ) at time
T T 0 is then
P ( 0, T )E Q T [ R s ( t ) ] (5)
As ( 0 ) P ( t, T )
E Q T [ R s ( t ) ] = -----------------
- E R ( t ) ---------------- (6)
P ( 0, T ) Q A s As ( t )
for t T 0 T . Unlike in Equation (3), the expected swap rate in Equation (6) under the terminal
measure Q T is no long the forward rate R s ( 0 ) . The difference,
C cms = E QT [ R s ( T 0 ) ] R s ( 0 ) (7)
As an example, we look at the street-standard model where we assume a flat yield curve
with rate R s . This leads to
1. Many authors, including Hagan (2004), have derived the convexity adjustment based on similar assumptions.
N 1
Rs
G ( R s ) = ---------------------------
- 1 1 -
------------------ (9)
( 1 + 1 Rs ) 1 + i Rs
i=1
Even with a model as simple as in Equation (9), the calculation of Equation (6) is rather com-
plicated. A first order Taylors expansion of Equation (8) is often used:
G ( R s ( t ) ) G ( R s ( 0 ) ) + G' ( R s ( 0 ) ) ( R s ( t ) R s ( 0 ) ) (10)
As ( 0 )
- E { R ( T ) [ G ( R s ( 0 ) ) + G' ( R s ( 0 ) ) ( R s ( T 0 ) R s ( 0 ) ) ] } R s ( 0 )
C cms = -----------------
P ( 0, T ) Q A s 0
As ( 0 )
= G' ( R s ( 0 ) ) -----------------
- E [ R 2 ( T ) R s2 ( 0 ) ]
P ( 0, T ) Q A s 0 (11)
We choose the Taylors expansion at R s ( t ) = R s ( 0 ) because it is the most likely future value.
To avoid specifying G ( R s ( t ) ) , we can assume it is simply a linear function of the swap rate.
That is
G ( Rs ( t ) ) + s Rs ( t ) (12)
As ( 0 )
C cms = s -----------------
- E [ R 2 ( T ) R s2 ( 0 ) ] (13)
P ( 0, T ) QA s 0
To determine the coefficients and s , we need only to fix two points on the line. Let call this
the fixed-point method. As shown in Figure 1, the first point we fix is R s ( t ) = 0 . In this case
all discount bond prices go to unity provided that interest rates are non-negative during the
swap period. Therefore
N 1
= G(0) =
i . (14)
i=1
P ( 0, T -)
----------------- = s + s Rs ( 0 ) (15)
As ( 0 )
Then
N 1
1 P ( 0, T )
s = ------------- ------------------
Rs ( 0 ) As ( 0 ) i
(16)
i=1
G ( Rs ( t ) )
+ s Rs ( t )
P ( 0, T )
------------------
As ( 0 )
Rs ( t )
Rs ( 0 )
With the lognormal assumption in Equation (4), we have the following convexity adjustment.
N 1
As ( 0 )
C cms = 1 ------------------
P ( 0, T ) i R s ( 0 ) [ exp ( s2 T 0 ) 1 ] (17)
i=1
The same result can be obtained using the linear swap rate model of Pelsser (2001) where
P ( t, T )
---------------- = a + b s R s ( t ) . (18)
As ( t )
for all time T > t , not just the particular time T 0 we are interested interest here. Linear swap
rate is a much stronger assumption than necessary for computing convexity adjustments.
Now consider the time t price of a contingent claim V ( t, R s ( T 0 ), T ) on CMS rate R s ( T 0 ) with
payment date T . Then
V ( t, R s ( T 0 ), T ) = P ( 0, T )E QT [ V ( T 0, R s ( T 0 ), T ) ]
P ( T 0, T )
= A s ( 0 )E Q A -------------------- V ( T 0, R s ( T 0 ), T )
As ( T0 )
= A s ( 0 )E QA [ ( + s R s ( T 0 ) )V ( T 0, R s ( T 0 ), T ) ]
(19)
Now consider a forward starting credit default swap (CDS) that has a premium payment
schedule identical to that of the interest swap in the previous section. Let P ( t, T ) be the risky
discount factor (i.e., the product of risk free discount bond price and survival probability). The
fair CDS spread R cds ( t ) is defined as
I { > t } E Q L def I { T0 TN } exp r ( u ) du
t
R cds ( t ) = --------------------------------------------------------------------------------------------------------- (20)
A cds ( t )
where is the time of default, L def = 1 R is the loss given default, R is the recovery rate, and
A cds ( t ) = I { > t }
[ t P ( t, T ) + a ( t, T
i i i i 1, Ti ) ] (21)
i=1
is the risky annuity for the premium leg with a i ( t, T i 1, T i ) as the present value of accrued pre-
mium. The expectation in Equation (20) is taken under the risk neutral measure Q .
Since the numerator in Equation (20) is the market price of the default payment for the CDS,
then, following Li (2007), the rate R cds ( t ) is a martingale under the martingale measure Q A
associate with numraire A cds ( t ) . That is
dQ M M ( t ) A cds ( 0 )
----------- = ----------------- -----------------
- (23)
dQ A A cds ( t ) M ( 0 )
We also define the risky swap rate as the fixed rate for the corresponding interest swap that
terminates at default. That is,
P ( t, T 0 ) E Q I { T 0 TN } exp r ( u ) du P ( t, T N )
t
R s ( t ) = I { > t } ------------------------------------------------------------------------------------------------------------------------------------ (24)
A cds ( t )
Consider a risky discount bond that matures at time T T 0 but vanishes if default occurs at
before time T 0 . Let P T0 ( t, T ) be the value of such risky bond at time t . Then the time- t value of
a payment at time T T 0 in the amount of R cds ( T 0 ) is
P T0 ( t, T )E Q [ R cds ( T 0 ) ] (25)
T
dQ M M ( t )P T0 ( 0, T )
----------- = ---------------------------------
- (26)
dQ T t
M ( 0 )P T0 ( t, T )
Then
dQ A cds ( 0 )P T0 ( t, T )
---------T- = --------------------------------------- (27)
dQ A A cds ( t )P T0 ( 0, T )
In general, both the swap rate R s ( t ) and CDS spread R cds ( t ) are stochastic. To evaluate the
convexity adjustment above, we extend the approach in CMS by assuming the Radon-
Nikodym derivative in Equation (23) is function of R s ( t ) and R cds ( t ) . We will treat R s ( t ) and
R cds ( t ) either as two separate factors or as a combined single factor in the rest of this section.
P T 0 ( t, T )
--------------------- = G ( R cs ( t ) ) (29)
A cds ( t )
R cs ( t ) = R s ( t ) + R cds ( t ) . (30)
G ( R cs ( t ) ) G ( R cs ( 0 ) ) + G' ( R cs ( 0 ) ) [ R cs ( t ) R cs ( 0 ) ] (31)
2. Alternatively, we can use the CDS spread R cds ( t ) instead of the total risky yield R cs ( t ) . The convexity, under
certain assumptions, turns out to be the same, as shown in Appendix A.
A cds ( 0 )
- E Q { R cds ( T 0 ) [ G ( R cs ( T 0 ) ) G ( R cs ( 0 ) ) ] }
C cmcds = ---------------------
P T 0 ( 0, T ) A
A cds ( 0 )
- G' ( R cs ( 0 ) )E { R cds ( T 0 ) [ R s ( T 0 ) R c ( 0 ) + R cds ( T 0 ) R cds ( 0 ) ] }
= ---------------------
P T0 ( 0, T ) QA
(32)
A cds ( 0 )
- G' ( R cs ( 0 ) )E Q [ R cds
C cmcds = --------------------- 2 (T ) R 2 (0)]
0 cds (33)
P T0 ( 0, T ) A
In general, we G' ( R cs ( 0 ) ) is not available because we do not know the function form of
G ( R cs ( t ) ) . Similar to the case of CMS, we use a linear approximation for G ( R cs ( t ) ) :
G ( R cs ( t ) ) cs + cs [ R s ( t ) + R cds ( t ) ] (34)
We further assume that those approximations are exact in expectation under Q A . That is
P T0 ( 0, T )
---------------------- = cs + cs [ R s ( 0 ) + R cds ( 0 ) ] (35)
A cds ( 0 )
A cds ( 0 )
- cs E { R cds ( T 0 ) [ R cs ( T 0 ) R cs ( 0 ) ] }
C cmcds = --------------------- (36)
P T0 ( 0, T ) QA
Now lets use the fixed point method to determine the coefficients in the approximation. Note
that Equation (35) implies the approximating straight line in Equation (34) intersects with the
curve line in Equation (29) at the forward point R cs ( t ) = R cs ( 0 ) . We need only to select another
intersection to fix the straight line. Figure 2 shows the following two choices.
cs A cds ( 0 ) R cs ( T 0 )
- E Q R cds ( T 0 ) -----------------
C cmcds = 1 ------------------------- -1 (37)
P T 0 ( 0, T ) A
R cs ( 0 )
If the CDS spread R cds ( T 0 ) and risky swap rate R s ( T 0 ) are un-correlated, then
cs A cds ( 0 ) 2 (T ) R2 (0)
R cds 0 cds
C cmcds = 1 -------------------------- E Q --------------------------------------------- (38)
P T0 ( 0, T ) A R cs ( 0 )
G ( R cs ( t ) )
Approximation 2
P T0 ( 0, T ) Approximation 1
---------------------
-
A cds ( 0 )
Rs ( t )
0 Rs ( 0 ) R cs ( 0 )
An this point, the credit spread reduces to zero. As a result, all risky discount bond reduce to
risk free discount bond. Therefore from Equation (34) we have
P ( 0, T )
------------------ = cs + cs R s ( 0 ) . (39)
As ( 0 )
P ( 0, T )
cs = ------------------ cs R s ( 0 ) , (40)
As ( 0 )
and
1 P T0 ( 0, T ) P ( 0, T )
cs = ------------------ ---------------------
- ------------------ . (41)
R cds ( 0 ) A cds ( 0 ) As ( 0 )
A cds ( 0 ) P ( 0, T ) 2 (T ) R2 (0)
R cds 0 cds
C cmcds = 1 ---------------------- ------------------ E Q --------------------------------------------- (42)
P T0 ( 0, T ) A s ( 0 ) A
R cds ( 0 )
Since R cs ( t ) is distributed around its forward value R cs ( 0 ) , the second approximation is gener-
ally better for R cs ( t ) above the risky swap rate R s ( 0 ) as shown in Figure 2. Theoretically we
can use all three points {0, R s ( 0 ) , R cs ( 0 ) } to obtain a quadratic approximation for G ( R cs ( t ) )
instead of the linear approximation we have discussed here. But this may not be necessary in
practice as other modeling errors can overshadow any marginal gains from a quadratic
approximation.
P T0 ( t, T )
- = G ( R cds ( t ), R s ( t ) )
-------------------- (43)
A cds ( t )
It can be approximated by its tangent line at the forward risky swap rate R s ( 0 ) and CDS rate
R cds ( 0 ) :
Again because the first order partial derivatives G 1 and G 2 are generally not available, we use
a linear approximation as follows
Both sides of Equation (45) are Q A martingales. Taking expectation under Q A yields
P T0 ( 0, T )
---------------------- = + cds R cds ( 0 ) + s R s ( 0 ) (46)
A cds ( 0 )
(47)
2
R s ( t ) = R s ( 0 ) exp s W s ( t ) -----s- t (48)
2
and
cds
2
R cds ( t ) = R cds ( 0 ) exp cds W cds ( t ) ---------
- t (49)
2
respectively, where W s ( t ) and W cds ( t ) are standard Brownian motions with correlation under
Q A , s and cds are the corresponding volatilities.
R s ( t ) = R cds
( t ) ( t ) (50)
R s ( 0 ) = E Q [ R s ( t ) ] E Q R cds
( t )E ( t )
Q
(51)
T T T
and
E Q R cds 1 + (t)
E Q [ R cds ( t )R s ( t ) ] E Q R cds
1 + ( t )E ( t ) = R ( 0 ) ----------------------------
QT s
T
(t)
- (52)
A T E Q R cds
T
To find the coefficients s and cds in this approximation, we use the fixed-point method. For
simplicity, we only consider the second approximation used in the one factor case. That is, we
pick the fixed point of R cs ( t ) = R s ( 0 ) . This is equivalent to set R cds ( t ) = 0 . Then from Equation
(45) we have
P ( t, T )
---------------- + s R s ( t ) (54)
As ( t )
This is identical to Equation (12) in the CMS case where and s are determined by Equa-
tion (14) and Equation (16) respectively.
1 P T 0 ( 0, T )
cds = ------------------ ---------------------
- ( + s Rs ( 0 ) )
R cds ( 0 ) A cds ( 0 )
1 P T0 ( 0, T ) P ( 0, T )
- ------------------ s [ R s ( 0 ) R s ( 0 ) ]
= ------------------ ---------------------
R cds ( 0 ) A cds ( 0 ) As ( 0 )
(55)
Note that for relatively flat yield curves3, the credit contingent swap rate R s ( 0 ) is very close to
the regular swap rate R s ( 0 ) . If we set R s ( 0 ) = R s ( 0 ) , cds reduces to cs in Equation (41) as in
the one-factor case.
When CDS rate R cds ( T 0 ) and risky swap rate R s ( T 0 ) are independent of each other, the second
term in Equation (47) drops out (Note that E QA R s ( T 0 ) = R s ( 0 ) because risky swap rate is a mar-
tingale under the risky annuity measure). Then
3. If yield curve is deterministic, then R s ( 0 ) = R s ( 0 ) even if the yield curve is not flat.
A cds ( 0 )
- cds [ E Q R cds
C cmcds = --------------------- 2 (T ) R 2 (0)]
0 cds (56)
P T0 ( 0, T ) A
In this section, we apply the model in the previous section to the valuation of options on CDS
spread. Although convexity adjustment is not necessary for credit default swaptions, we
present the result for it here anyway out of completeness. Analytic solutions are available
when R s ( T 0 ) and R cds ( T 0 ) are jointly lognormal.
A credit default swaption is the right to enter into a CDS with a specific premium rate X (the
strike) at some future date T 0 (the expiry). The option to sell protection (i.e., to receive premi-
ums in a CDS) can be valued as
[ X R cds ( T 0 ) ] + A cds ( T 0 ) dQ M
V s ( 0 ) = M ( 0 )E Q ----------------------------------------------------------
- -----------
M ( T0 )
T0
A
dQ A
= A cds ( 0 )E { [ X R cds ( T 0 ) ] + }
QA
(57)
where W QA ( t ) is a standard Wiener process under the measure Q A . As a result, R cds ( t ) is log
normally distributed with mean R cds ( 0 ) and its logarithm has a variance of
( u, T , T
t
t2 ( T 0, T N )t = 2
0 N ) du . (59)
0
We call t ( T 0, T N ) the term volatility of R cds ( t ) . Then Equation (57) reduces to the Black for-
mula. The option to buy protection can be priced similarly.
Without loss of generality, we consider only CMCDS forwards (or one-period CMCDS)
because multi-period CMCDS is just a portfolio of CMCDS forwards. Assume the CDS
spread R cds ( t ) resets at time t = T 0 and is paid at future time T T 0 . The present value of this
one-period CMCDS payment is
V cmcds ( 0 ) = P T0 ( 0, T )E QT [ R cds ( T 0 ) ]
= P T0 ( 0, T ) [ R cds ( 0 ) + C cmcds ]
(60)
Again we consider only CMCDS caplets, which are one-period CMCDS caps. Assume the
CMCDS rate R cds ( t ) resets at time t = T 0 and the positive difference between R cds ( t ) and
strike K is paid at some future time T T 0 . The present value of this option is
V cap ( 0 ) = P T 0 ( 0, T )E QT { [ R cds ( T 0 ) K ] + }
P T 0 ( t, T )
= A cds ( 0 )E Q --------------------
- [ R cds ( T 0 ) K ] +
A cds ( t )
A
When R s ( T 0 ) and R cds ( T 0 ) are jointly lognormal, the expectation in the last equation can be
evaluated analytically. It involves two-dimensional cumulative Normal distributions.
In a CMCDS digital option, assume the CMCDS rate R cds ( t ) resets at time t = T 0 and $1 is
paid at some future time T T0 if R cds ( t ) is higher than K . The present value of this option is
P T 0 ( T 0, T )
= A cds ( 0 )E Q ------------------------
- I{R (T ) > K }
A A
cds 0 ( T ) cds 0
= A cds ( 0 )E Q { [ + cds R cds ( T 0 ) + s R s ( T 0 ) ]I { Rcds ( T0 ) > K } }
A
(62)
again when R s ( T 0 ) and R cds ( T 0 ) are lognormally distributed, the expectation here can be evalu-
ated using two-dimensional cumulative Normal distributions.
5. Conclusion
In this paper, we extended the convexity adjustment method used in constant maturity swap
(CMS) to constant maturity credit default swaps (CMCDS). The basic idea is that relative
prices of traded securities are martingale under some equivalent measure and the price of risk-
free or risky bond which is traded can be approximated as quadratic functions of CMS or
CMCDS. Such convex relationship between price and yield together with the change of mea-
sure leads to simple solutions to convexity adjustments. Analytical solution are available
under lognormality assumptions for CMS and CMCDS rates.
In the case of CMCDS, risky bond yield consists of both risky swap rate and CDS spread. We
modeled the sum of the two as a whole (one factor) first and then separately (as two factors)
afterwords. In the latter, we explicitly considered the correlation between interest rate and
credit spread in deriving our results. We showed that when the two are independent and the
yield curve is relatively flat, the two factor model can be reduced to one factor one.
Finally we applied our convexity adjustment method to the valuation of credit default swap-
tions and options on CDS spreads. Again, under lognormal assumptions closed form solutions
have been obtained.
References
Black, F., 1976 The Pricing of Commodity Contracts Journal of Financial Economics, vol.3,
pp.167-179
Brigo, D., 2005 Constant Maturity Credit Default Swap Pricing with Market Models Work-
ing Paper, (http://www.damianobrigo.it)
Hagan, P., 2004 Convexity Conundrums: Pricing CMS Swaps, Caps and Floors in The Best
of Wilmott.
Harrison, J. M. and S. Pliska, 1981, Martingale and Stochastic Integrals in the Theory of
Continuous Trading, Stochastic Processes and Their Applications, 11, 215-260.
Li, A., 1995, A One-Factor Lognormal Markovian Interest Rate Model: Theory and
Implementation. Advances in Futures and Options Research, Vol. 8, 1995 Available at
SSRN: http://ssrn.com/abstract=938801
Li, A, 2007, A Jump Diffusion Model for Contingent CDS Valuation (August 31, 2007).
Available at SSRN: http://ssrn.com/abstract=1012676
Li, A., P. Ritchken, and L. Sankarasubramanian, 1995, Lattice Models for Pricing American
Interest Rate Claims, Journal of Finance, Vol 50 No 2.
Pelsser, A., 2001, Mathematical Foundation of Convexity Correction, Available at SSRN:
http://ssrn.com/abstract=267995
P T0 ( t, T )
- = G ( R cds ( t ) ) + cds R cds ( t )
-------------------- (63)
A cds ( t )
P T0 ( 0, T )
---------------------- = + cds R cds ( 0 ) (64)
A cds ( 0 )
Then
A cds ( 0 )
E T [ R cds ( t ) ] = ---------------------
- E Q [ R cds ( t ) ( + cds R cds ( t ) ) ]
P T0 ( 0, T ) A
A cds ( 0 )
= R cds ( 0 ) + E Q ( cds ) --------------------- - { E Q [ R cds
2 (t)] R 2 (0)}
cds
A
P T0 ( 0, T ) A
(65)
A cds ( 0 )
C cmcds = E T [ R cds ( t ) ] R cds ( 0 ) = E Q ( cds ) ---------------------- { E Q [ R cds
2 (t)] R 2 (0)}
cds (66)
A
P T0 ( 0, T ) A
Using the fixed point at R cds ( t ) = 0 we have = G ( 0 ) . We know that when R cds ( t ) = 0 , all
risky discount bonds reduce to risk free discount bonds. Therefore from Equation (63) we
have
P ( 0, T )
= G ( 0 ) = ------------------ . (67)
As ( 0 )
1 P T0 ( 0, T ) P ( 0, T )
E Q ( cds ) = ------------------ ---------------------
- ------------------ (68)
A R cds ( 0 ) A cds ( 0 ) As ( 0 )