Professional Documents
Culture Documents
Sample Exercises A
Sample Exercises A
it
Excercise 0: Review the exercises below and the related solutions, and signal possible
mistakes in the solutions proposed, motivating.
EXERCISE 1: CDS AND HAZARD RATES. Assume a CDS quoted spread is 400
basis points and the recovery is estimated to be 20%. Under the assumptions that i) the
premium leg of the CDS pays continuously and ii) the hazard rate is constant, find
d) If the recovery increases, but the spread remains the same, does the probability of
default increase of decrease? motivate
e) Is it true that if the CDS spread doubles and the recovery remains the same, then
the probability of default in ten years doubles?
EXERCISE 1: SOLUTIONS
RCDS
γ=
1 − REC
that holds for continuous payments in the premium leg and for constant hazard rates,
γ(t) = γ for all t. In our example
400bps
γ= = 0.04/0.8 = 0.05 = 5%
1 − 0.2
b) The probability of surviving time T in a hazard rate model with constant hazard
rate γ is
−γT −0.05·5 −0.25
Q(τ > T ) = e =e =e = ...
which is the probability of survival for five years.
−γT − R T
Q(τ ≤ T ) = 1 − e =1−e 1−REC
R
we see that if the recovery increases, 1 − REC decreases, so that 1−REC increases and
− R T
R
− 1−REC decreases, and e 1−REC still decreases since the exponential is increasing, and
− R T
finally −e 1−RECincreases as this is the opposite of a decreasing function. Adding 1
does not change things, so the function is increasing. Default probabilities increase if we
increase the recovery keeping the CDS rate fixed.
e) This is not true: in fact if now Rnew = 2Rold then γnew = 2γold follows easily
(why?) but this does not imply
Rather, it implies
which is quite different. Only for small γold, by expanding the exponential, we get
something close to twice the old default probability.
2. Describe the product you obtain by buying protection up to five years with a
standard CDS with fair running spread R0,5y (0) fixed at time 0 and selling protection on
the same name via a constant maturity CDS with a five year constant maturity in the
running spread. Is this position subject to default risk or spread movement risk? Why?
EXERCISE 2: SOLUTION.
The position is described by the two diagrams, as seen from point of view of A who
buys protection in the CDS (incoming protection arrow) and sells protection in the CMCDS
(exiting protection arrow):
CDS standard
CMCDS:
Since the two default or protection legs are the same, combining the two diagrams we
obtain the following diagram of cash flows:
CDS - CMCDS:
the net effect is thus that we (we are “A”) receive quarterly a spread that is
Ri−1,i+c(Ti−1) − R0,5y (0), i.e. the difference between the five year fair CDS spread
Ri−1,i+c(Ti−1) prevailing at the coupon date minus the five year CDS spread R0,5y (0)
fixed initially. When we enter such a position it is then clear that if the CDS spread
goes up in time, we will receive positive payments as Ri−1,i+c(Ti−1) − R0,5y (0) will
be positive, Ri−1,i+c(Ti−1) having increased over the initial R0,5y (0). On the contrary,
if CDS spreads go down, we will receive a negative difference, meaning that we actually
have to pay the opposite (positive) amount to B. We are clearly subject to spread risk,
but not to default risk as the default event itself does not impact us with the need to do
a protection payment. This is because the default legs in the CDS and CMCDS are the
same, so going long one and short the other one the cancel each other, which frees us
from default payments risk.
a) Write a formula as seen from time 0 for the probability that the firm defaults (by
T ).
b) Is the Default probability increasing or decreasing in the debt level K? Is the Default
probability increasing or decreasing in the initial firm value level V(0)? Motivate
c) What happens if the initial firm value is very large (tends to infinity)? What is the
behaviour of the default probability as a limit?
EXERCISE 3: SOLUTIONS.
2
log V (T ) = log V (0) + (r − k − σ /2)T + σW (T )
and √
2
log V (T ) = log V (0) + (r − k − σ /2)T + σ T N (0, 1)
where N (0, 1) is a standard normal random variable. This implies that the default event,
that in the Merton model can only happen at maturity, is written as {V (T ) ≤ L}, i.e.
the final firm value went below the debt L to be paid. We have that {V (T ) ≤ L} =
{log V (T ) ≤ log L} since log is an increasing function. Given the above equation for
V (T ) this translates into
( )
2
log(L/V (0)) − (r − k − σ /2)T
{log V (T ) ≤ log L} = N (0, 1) ≤ √
σ T
whose probability is
σ2 σ2
log V L(0) − (r − k − 2 )T log L
V (0) − (r − k − 2 )T
Q N (0, 1) ≤ √ = Φ √
σ T σ T
d) To compute the hazard rate in Merton model we need to compute the limit of the
above default probability formula, after substituting the simplification r − k − σ 2/2 = 0.
We have
à !
log L
V (0)
Φ √
σ T
Q{τ ≤ T }
lim = lim =
T ↓0 T T ↓0 T
This would lead to a limit of the type zero over zero. This is because in the numerator,
the denominator tends to zero from the positive side, so that the fraction, with negative
numerator, tends to minus infinity. Taking into account that the normal cdf Φ tends to
zero when the argument of the numerator tends towards minus infinity, we have a limit of
the type zero over zero. We apply the limit theorem of De L’Hopital, and we get
à !à !
log L log L
V (0) V (0)
p √
σ T
− 2T 1√T σ
= lim =
T ↓0 1
1 − x2
p(x) = √ e 2 .
2π
√
If we set 1/ T = y , the limit becomes
L
à !
3
log V (0) log V L(0) y3
= lim −y p y = A lim =0
y↑+∞ 2σ σ y↑+∞ exp[(y 2 (log(L/V (0))/σ)2 )/2]
(for some constant A), since the exponential goes to infinity faster than any polynomial.
Q{τ ≤ T }
lim =γ
T ↓0 T
This is an important difference: basic structural models like Merton have no short
term credit spreads (the limit is zero). Intensity models instead have non-zero short term
credit spread. This is a modeling advantage of intensity models. It means that for very
short maturities the Merton model will have great difficulties in fitting nonzero spreads,
whereas the intensity model will have no problem.
Compute:
a) The probability of defaulting in one year, two years and three years.
c) assuming that the recovery is 0.5, interest rates are zero and the CDS sells protection
from today to three years, compute the price of the default or protection leg of the CDS.
EXERCISE 4: Solution
Compute Z Z
1 1
γ(t)dt = 0.02dt = 0.02 · 1 = 0.02
0 0
since γ(t) is constant and equal to 0.02 in the integration interval. Compute
Z 2 Z 1 Z 2 Z 1 Z 2
γ(t)dt = γ(t)dt+ γ(t)dt = 0.02dt+ 0.04dt = 0.02·1+0.04·1 = 0.06
0 0 1 0 1
Analogously, Z 3
γ(t)dt = 0.02 · 1 + 0.04 · 1 + 0.02 · 1 = 0.08
0
c) The default leg price is LGDE[D(0, τ )1{τ ≤3y}]. However, since interest rates are
zero, r = 0, all discounts are equal to one, D(s, u) = exp(−r(u − s)) = exp(0) = 1.
Hence we have
LGDE[D(0, τ )1{τ ≤3y}] = LGDE[1 · 1{τ ≤3y}] = LGDE[1{τ ≤3y}] = LGDQ(τ ≤ 3y) =
Excercise 0: Review the exercises below and the related solutions, and signal possible
mistakes in the solutions proposed, motivating.
where λ0, κ, µ, ν are positive constants. W is a brownian motion under the risk neutral
measure.
e) Is it true that, because of mean reversion, the variance of the intensity goes to zero
(no randomness left) when time grows to infinity?
f) Can you compute a rough approximation of the percentage volatility in the intensity?
h) Can you guess the spread of a CDS with 10y maturity with the above stochastic
intensity when the recovery is 0.35?
EXERCISE 1: Solutions.
b) We can refer to the formulas for the mean and variance of λT in a CIR model as
seen from time 0, at a given T . The formula for the variance is known to be (see for
Example Brigo and Mercurio (2006))
We can see that for k becoming large the variance becomes small, since the exponentials
decrease in k and the division by k gives a small value for large k. In the limit
lim VAR(λT ) = 0
κ→+∞
c) As the mean is
−κT −κT
E[λT ] = λ0e + µ(1 − e )
we clearly see that this is impacted by κ (indeed, ”speed of mean reversion”) and by µ
clearly (”long term mean”) but not by the instantaneous volatility parameter ν .
−κT −κT
lim λ0e + µ(1 − e )=µ
T →+∞
so that the mean tends to µ (this is why µ is called ”long term mean”).
e) In the limit where time goes to infinity we get, for the variance
These however do not take into account the important impact of κ in the overall volatility
of finite (as opposed to instantaneous) credit spreads and are therefore relatively useless.
hence 2κµ > ν 2 and trajectories are positive. Then we observe that the variance is very
small: Take T = 5y ,
which is much smaller of the level 0.04 at which the intensity refers both in terms of
initial value and long term mean. Therefore there is almost no randomness in the system
as the variance is very small compared to the initial point and the long term mean.
Hence there is almost no randomness, and since the initial condition λ0 is the same
as the long term mean µ0 = 0.04, the intensity will behave as if it had the value 0.04 all
the time. All future trajectories will be very close to the constant value 0.04.
RCDS
λ= ⇒ RCDS = λ(1 − REC) = 0.04(1 − 0.35) = 260bps
1 − REC
EXERCISE 2.Consider two standard normal random variables X1, X2 that are jointly
normal with correlation ρ. Write the copula functions for the following values of ρ:
a) ρ = 0
b) ρ = 1
c) ρ = 1/2
d) Write the copula for the random vector (X1, X23) when ρ = 1/2.
EXERCISE 2: Solutions.
As the variables are standard normal, if we call Φ the cdf of the standard normal we
know that
Φ(X1) = U1, Φ(X2) = U2
are uniform random variables. The copula is then defined as the multivariate distribution
of the uniforms,
C(u1, u2) = Q(U1 ≤ u1, U2 ≤ u2)
C(u1, u2) = Q(U1 ≤ u1, U2 ≤ u2) = by independence = Q(U1 ≤ u1)Q(U2 ≤ u1) = u1u2
c) For ρ = 1/2 we cannot invoke a special calculation; we obtain just the Gaussian
copula for correlation parameter 1/2, that cannot be written in closed form but only as an
integral of the related bivariate density.
d) We know the copula is invariant for transformations that preserve information, i.e.
invertible. Since
3
(X1, X2) 7→ (X1, X2 )
1/3
(Y1, (Y2) ) ← (Y1, Y2)
a) How many defaults are needed because the tranche starts experiencing some losses?
b) How many defaults are needed for the tranche to be completely wiped out?
EXERCISE 3: Solutions.
a) For the loss to be equal to at least 12%, thus impacting the tranche, we need to
have default of n names where
1
n× × (1 − 0.4) = n × 0.48% > 12%
125
This implies
Then the tranche will be impacted from the 26th default on.
b) For the tranche to be completely wiped out we need to have default of N names
where
1
N× × (1 − 0.4) = N × 0.48% > 22%
125
This implies
N > 22/0.48 = 45.83
Therefore the 12 − 22% tranche will be completely wiped out starting from the 46-th
default.
c) The maximum loss that the pool can experience is when all 125 names default. In
this case the total loss is
1
125 × × (1 − 0.4) = 0.6 = 60%
125
This means that with recovery 0.4 the loss will never reach 70%, not even if the whole
portfolio defaults.
Since there can be no default in the default leg of the tranche, the value of the default
leg of the tranche is 0.
Since the spread of the tranche is defined as the default leg divided by the tranche
DV01, and the latter is nonzero, we have that the resulting spread is zero. The only way
in the above setup to have the tranche being impacted at all, is to lower the recovery.
EXERCISE 4. We are given three CDS on names 1,2,3. Assume each single name
default is modeled with a constant hazard rate and that the CDS premium legs pay
continuously, so that the hazard rate formulas
RCDS,i
λi = , i = 1, 2, 3
1 − RECi
hold for the three default probabilities.
Assume the three names have all recovery 20% and they have spreads respectively
b) Suppose the CDS spreads are now very small. Find a further approximation for the
default leg price.
d) In which of the two cases a) and c) above is protection more expensive? Motivate
EXERCISE 4: Solution.
a) If i1 is the random variable indexing the first default time τ 1 = τi1 , the default leg
is
1
E[LGDi1 D(0, τ )1{τ 1≤1y}]
Now, since all LGD = 1 − REC = 1 − 0.2 = 0.8 and D(0, t) = 1 for all t being
interest rates zero, we have
1 1
0.8E[1{τ 1≤1y}] = 0.8Q(τ ≤ 1) = 0.8(1 − Q(τ > 1))
Now, Q(τ 1 > 1) is the probability that the smallest of the default times is larger than
1. But saying that the smallest is larger than one is the same as saying that all are larger
than one. So
1
Q(τ > 1) = Q(τ1 > 1 and τ2 > 1 and τ3 > 1) = ...
Since the defaults are independent, the probability becomes the product of probabilities,
b) If the CDS spreads are very small also the intensities will be very small. We have, using
x
e ≈1+x
for small x,
i.e. we get approximately the sum of the three CDS spreads as a cost of the default leg.
c) In general in the intensity model with constant hazard rates we can write
−1 ξ1 ξ2 ξ3
τ1 = Λ1 (ξ1) = , τ2 = , τ3 =
λ1 λ2 λ3
With a maximum dependence copula we have that the exponential triggers (on which the
copula is acting) are totally dependent, i.e. ξ1 = ξ2 = ξ3 = ξ . This means that
ξ ξ ξ
τ1 = , τ2 = , τ3 =
λ1 λ2 λ3
all with the same ξ . Hence we can see that the smallest τi is the one with the largest
λi, as we are dividing the same ξ for three possible λ’s. In our case the largest λ is λ3
coming from the largest spread RCDS,3, so that the smallest τ is τ3. This is true for every
possible scenario of ξ . Hence we have in this case
1
τ = τ3
This is obvious: If the three names are totally correlated, in a way it is enough to buy
protection from the riskiest one of them and one is also protected from the other two. Not
so if the correlation is smaller: in that case we have a larger premium to pay.