Você está na página 1de 143

# Master Course Exercises with Solutions

## Prof. Damiano Brigo

Mathematical Finance Section
Dept. of Mathematics

## Imperial College London

1 / 143

Content I
1

EXERCISES
Absence of Arbitrage in Black Scholes
Zero coupon bonds and rates
Spot, forward and swap rates
Put Call Parity Violation and Arbitrage
Transformations of Vasicek models
Dynamics for rt
Instantaneous Forward Rates and HJM drift condition in G2++
LIBOR fourth payoff
Swaptions
LIBOR MODEL CALIBRATION
Deterministic intensities / hazard rates
CIR model for default intensity
CVA for Bonds
CVA for Call option

## Imperial College London

2 / 143

Content II
CVA for Bonds portfolio
Risk Measures

3 / 143

EXERCISES

Scholes I

## Consider the Black and Scholes basic economy given by a bank

account and a stock, whose prices are given respectively by
dBt = rBt dt, B0 = 1,

dSt = St dt + St dWt , S0 = 1

## where r , , are positive constants and W is a brownian motion under

the physical measure P.
Consider the limiting case where = 0, ie the stock has no volatility.
Prove that if 6= r then one has arbitrage.

4 / 143

EXERCISES

## Exercise + Solutions: Absence of Arbitrage in Black

Scholes II
SOLUTION. To prove arbitrage we need to prove that there exists at
least one arbitrage opportunity, ie a self-financing trading strategy
(B , S ) such that
B0 B0 + S0 S0 = 0, BT BT + ST ST > 0 a.s.
for some maturity T > 0.
Let us start for the case where > r . We may easily establish an
arbitrage opportunity as follows.
At time 0 we buy S0 = 1 unit of stock and short-sell 1 unit of bank
account: B0 = 1.
This has cost zero as both positions have a value of 1 at time 0:
B0 B0 + S0 S0 = (1) 1 + 1 1 = 0.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

5 / 143

EXERCISES

## Exercise + Solutions: Absence of Arbitrage in Black

Scholes III
Integrating the bank account and stock equations
dBt = rBt dt,

dSt = St dt

## yields Bt = ert and St = et .

Since > r , our strategy is just to wait, as S grows with and B with r .
So now our strategy is to hold onto the initial amounts without buying
or selling anything.
Bt = 1, St = 1 for allt,

0 < t < T.

## It is obviously self financing as this does not involve any external

funding or any extraction of money from the accounts.
At maturity we have
(c) 2012-13 D. Brigo (www.damianobrigo.it)

6 / 143

EXERCISES

## Exercise + Solutions: Absence of Arbitrage in Black

Scholes IV
BT BT + ST ST = (1) erT + 1 eT > 0
since > r .
If < r we do the opposite: at time zero we buy one unit of Bank
account and go short one unit of stock, and then just wait. The
reasoning is analogous.
This exercise shows an important point: we cannot have two bank
accounts accruing at different rates, or we have immediately arbitrage.
Also, note that in this case the Radon Nykodym derivative dQ/dP
would not be well defined.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

7 / 143

EXERCISES

## Exercise + Solutions: Zero coupon bonds and rates I

Assume that the zero coupon bond term structure is given by
P(0, T ) = exp(kT 2 ), 0 < k ()
a) Write an expression for the LIBOR spot rate L(0, T ) as a function of
T.
b) Write an expression for the LIBOR forward rate F (0, T , T + 1) as a
function of T .
c) Study F (0, T , T + 1) as a function of k . Is it increasing, decreasing,
concave, convex, how does it behave at 0 and infinity?
d) Compute the swap rate S(0, 0, 2) = S0,2 (0) for a swap with annual
tenor in both legs, and compute the limits for k going to zero and
infinity of this swap rate.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

8 / 143

EXERCISES

## e) The continuously compounded spot rate at time t for maturity T is

defined as
1
R(t, T ) =
ln P(t, T ).
T t
Find an explicit expression for R(0, T ) when the bond follows (*) above.
f) By making use of point e), find the short rate r0 at time 0.

9 / 143

EXERCISES

## Exercise + Solutions: Zero coupon bonds and rates III

SOLUTIONS:
a)

L(0, T ) =

1
1
1
2
2
(1/P(0, T ) 1) = (1/ekT 1) = (ekT 1)
T
T
T

b)

F (0, T , T +1) =

1
2
2
(P(0, T )/P(0, T +1)1) = ekT /ek(T +1) 1
T +1T
= ek(2T +1) 1

c)
(c) 2012-13 D. Brigo (www.damianobrigo.it)

10 / 143

EXERCISES

## Exercise + Solutions: Zero coupon bonds and rates IV

The map
k 7 (ek(2T +1) 1)
is increasing, convex, at infinity goes to infinity, at 0 goes to 0.
d)
S(0, 0, 2) =

1 P(0, 2)
1 e4k
e4k 1
= k
=
1 P(0, 1) + 1 P(0, 2)
e + e4k
e3k + 1

## This tends to 0 for k going to 0, and to infinity for k going to infinity.

e)
R(0, T ) = kT
f)
r0 = lim R(0, T ) = lim kT = 0
T 0

T 0

11 / 143

EXERCISES

## Exercise with Solutions: Spot, fwd and swap rates I

Consider the following curve of zero coupon bonds for the maturities
T0 = 1y, T1 = 2y , . . . , T9 = 10y:
P(0, T0 ) = 0.961538462 P(0, T1 ) = 0.924556213
P(0, T2 ) = 0.888996359 P(0, T3 ) = 0.854804191
P(0, T4 ) = 0.821927107
a) Compute the forward swap rates S1,4 (0) and S2,4 (0).
b) Compute the forward libor rates F1 (0), F2 (0), ... and verify that the
swap rate S2,4 (0) is a weighted average of the forward LIBOR rates (in
particular, compute the weights).

12 / 143

EXERCISES

## Exercise with Solutions: Spot, fwd and swap rates II

SOLUTIONS:
a) We have an annual tenor structure where Ti Ti1 = 1 year. This is
important in defining all rates.
S1,4 (0) =

P(0, T1 ) P(0, T4 )
1 P(0, T2 ) + 1 P(0, T3 ) + 1 P(0, T4 )

Similarly we obtain
S2,4 (0) =

P(0, T2 ) P(0, T4 )
1 P(0, T3 ) + 1 P(0, T4 )

## Substitution of the given values for the Ps gives the result.

b) We know that
S2,4 (0) = w3 (0)F3 (0) + w4 (0)F4 (0),
(c) 2012-13 D. Brigo (www.damianobrigo.it)

13 / 143

EXERCISES

## Exercise with Solutions: Spot, fwd and swap rates III

where
w3 (0) = P(0, T3 )/(P(0, T3 ) + P(0, T4 )),
w4 (0) = P(0, T4 )/(P(0, T3 ) + P(0, T4 )).
Also,
F3 (0) =

1
1




P(0, T2 )
1 P(0, T3 )
1 ; F4 (0) =
1 .
P(0, T3 )
1 P(0, T4 )

14 / 143

EXERCISES

## Exercise + Solutions: Put Call Parity Violation and

Arbitrage I
Consider the Black and Scholes basic economy given by a bank
account and a stock, whose prices are given respectively by
dBt = rBt dt, B0 = 1,

dSt = St dt + St dWt , S0 = 1

## where r , , are positive constants and W is a brownian motion under

the physical measure P.
Consider two options that are at the money forward, namely having
strike
K = S0 erT ,
respectively a call option and a put option with maturity T and payout
(ST K )+ = max(ST K , 0) (Call), (K ST )+ = max(K ST , 0) (Put).
(c) 2012-13 D. Brigo (www.damianobrigo.it)

15 / 143

EXERCISES

## Exercise + Solutions: Put Call Parity Violation and

Arbitrage II

In this case, put call parity tells us that the initial value of the call
option, at time 0, must be equal to the initial value of the put option,
given that the forward contract value is zero (K is the at-the-money
forward strike that sets the forward price to zero).
Show that if this condition is violated, and for example
CallPrice0 = PutPrice0 + X
for a positive amount X > 0, one has arbitrage.

16 / 143

EXERCISES

## Exercise + Solutions: Put Call Parity Violation and

Arbitrage III
SOLUTION.
Since the price of the call is larger than the one of the put when they
should actually be the same, we can try by buying a put and
short-selling a call, and buying an at the money forward contract to
balance put minus call at maturity. We also buy some bank account
with the difference between the call and the put at time 0.
We enter into one position in a put option at time 0, and short sell one
call option at time 0, both options with strike K and maturity T . We
also enter into a forward contract at the same maturity with strike K .
This means that we accept to receive ST K at maturity (meaning that
if this quantity is positive we receive it, if it is negative we pay its
absolute value to our counterparty in the trade). We also buy an
amount X of bank account.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

17 / 143

EXERCISES

Arbitrage IV

## The cost of starting this strategy is:

We pay PutPrice0 to enter into the put option
we receive CallPrice0 = PutPrice0 + X by short-selling the Call
option
We pay X to buy a quantity X of bank account B0 at time 0.
We pay nothing to enter into the forward contract since its initial
cost is S0 KerT = 0.

18 / 143

EXERCISES

## Exercise + Solutions: Put Call Parity Violation and

Arbitrage V
These four operations have a total cost of
PutPrice0 CallPrice0 + X + 0 = 0
So it costs nothing setting up the strategy.
Following the initial setup, we just wait. This clearly preserves the self
financing condition since we do not inject external funds or extracts
funds from the strategy.
At maturity, we have the following cash flows:
We receive (K ST )+ from the Put option.
We pay (ST K )+ for the Call option we have been short-selling
We receive ST K from the forward contract
We have XerT in the bank account.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

19 / 143

EXERCISES

Arbitrage VI

## The total value of this strategy at T is hence

PutPayoutT CallPayoutT + FwdContractPayoutT + XBT =
(K ST )+ (ST K )+ + ST K + XerT =
= K ST + ST K + XerT = XerT > 0
So we have a self-financing trading strategy whose initial cost is zero
and that produces a positive final cash flow XerT in all scenarios.
Hence this is an arbitrage opportunity and the market is arbitrageable.

20 / 143

EXERCISES

model I

## Consider the Vasicek process

dxt = k ( xt )dt + dWt , x0
where W is a brownian motion under the risk neutral measure.
a) Assume rt = xt + 0.01. Compute a formula for the risk neutral
probability that the short rate is negative at a time T .
b) Assume = 0 and set rt = (xt )3 . Find the stochastic differential
equation for the short rate.

21 / 143

EXERCISES

## Exercise with Solution: Transformations of the Vasicek

model II
SOLUTIONS: a) We know that in the Vasicek model the short rate at
time t conditional on time s is normally distributed with mean and
variance given by


E{x(t)|xs } = x(s)ek(ts) + 1 ek(ts)
i
2 h
1 e2k(ts) .
2k
It follows that the short rate rt = xt + 0.01 at time T conditional on time
0 is
VAR{x(t)|xs } =

rT Normal(0.01 + mT , VT2 )
(c) 2012-13 D. Brigo (www.damianobrigo.it)

22 / 143

EXERCISES

## Exercise with Solution: Transformations of the Vasicek

model III


i
2 h
mT = x(0)ekT + 1 ekT , VT2 =
1 e2kT
2k
The probability that a normal distribution is negative is
Q{rT < 0} = Q{Normal(0.01 + mT , VT2 ) < 0} =
= Q{0.01 + mT + VT Normal(0, 1) < 0} =
= Q{Normal(0, 1) < (0.01 + mT )/VT } = ((0.01 + mT )/VT )
b).
We use Itos formula.
1
d(xt )3 = 3xt2 dxt + 3 2xt dxt dxt =
2
(c) 2012-13 D. Brigo (www.damianobrigo.it)

23 / 143

EXERCISES

## Exercise with Solution: Transformations of the Vasicek

model IV
= 3xt2 k( xt )dt + 3xt2 dWt + 3xt 2 dt
d(xt )3 = 3[xt2 k( xt ) + xt 2 ]dt + 3xt2 dWt
1/3

## Set xt3 = rt , so that xt = rt

2/3

drt = 3[rt

. We have
1/3 2

k krt + rt

2/3

]dt + 3rt

dWt

By setting = 0 we obtain
1/3 2

drt = 3[krt + rt

2/3

]dt + 3rt

dWt

24 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt I

Consider the following model for the risk neutral dynamics of the short
rate:
drt = k dt + dWt , r0
where W is a Brownian motion under the risk neutral measure Q and
where k, and r0 are positive real numbers.
a) Compute the zero coupon bond price P(0, T ).
b) Compute the term structure of Interest Rates T 7 L(0, T )
associated with the model.
c) Is the dynamics realistic? List the advantages and the drawbacks of
this model.

25 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt II

SOLUTION.
a) Let us begin by solving the equation for the short rate:
drs = k ds + dWs .
Integrating both sides between 0 and t we obtain
rt r0 = k (t 0) + (Wt W0 )
so that
rt = r0 + kt + Wt ()
To compute the bond price we need to compute
"
!#
Z
T

P(0, T ) = E exp

rt dt

0
(c) 2012-13 D. Brigo (www.damianobrigo.it)

26 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt III

Let us begin with
Z

rt dt =?
0

## This is an integral with respect to dt, so we can use standard

integration by parts:
Z
0

T Z

rt dt = t rt

t drt

Z

Z
rt dt = T rT

or
Z

Z
rt dt = T rT

## (c) 2012-13 D. Brigo (www.damianobrigo.it)

t(kdt + dWt )
0

Z
t k dt

0
Master Exercises Prof. Brigo

t dWt
0
Imperial College London

27 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt IV

Now we also substitute rT from equation () written at t = T :
T

rt dt = T (r0 + kT + WT ) kT /2
0

t dWt
0

We have
T

rt dt = Tr0 + kT 2 + T WT kT 2 /2

or
Z

t dWt
0

dWt kT /2

rt dt = Tr0 + kT + T
0

## where we have substituted WT =

Z
X :=

t dWt
0

RT
0

dWt . We obtain

rt dt = Tr0 kT 2 /2

0
(c) 2012-13 D. Brigo (www.damianobrigo.it)

(T t) dWt
0

28 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt V

This last equation is the sum of a deterministic constant plus a Ito
integral. The Ito Integral has a deterministic integrand. Hence it is the
limit of a sum of independent gaussian random variables (increments
dW ) each multiplied by a constant, and so it is itself gaussian. By
using the Ito isometry, we may determine its mean and variance. We
know that
Z
T

(T t) dWt ] = 0

E[
0

and
Z

Z
(T t) dWt ] =

VAR[
0

2 (T t)2 dt = 2 T 3 / 3

## Hence we finally have

Z
P(0, T ) = E[exp(

rt dt)] = E[exp(X )]
0

29 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt VI

where X is a normal of mean = Tr0 kT 2 /2 and Variance
V 2 = 2 T 3 19/3. Hence, recalling the moment generating function of a
Normal random variable,
P(0, T ) = E[exp(X )] = exp( + V 2 /2)
we have
P(0, T ) = exp(Tr0 kT 2 /2 + 2 T 3 /6)
b). We need to compute



1
1
1
L(0, T ) =
1 =
exp(Tr0 + kT 2 /2 2 T 3 /6) 1
T P(0, T )
T
c). The dynamics is not very realistic, as we explain in the
(c) 2012-13 D. Brigo (www.damianobrigo.it)

30 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt VII

Advantages. The model is very tractable. One can make several
calculations, as we did before, and obtain bonds and interest rate
options in closed form. Since it is Gaussian, it is also very easy to
simulate.
rT is a gaussian random variable at all future times T . Hence it will
have a positive probability of taking negative values. Since
negative interest rates are not desirable, this is a drawback.
Depending on the value of the parameters, this probability can be
large.
We may compute it easily using formula ():

## Q(N(0, 1) < (r0 kt)/( t)) = ((r0 kt)/( t))

(c) 2012-13 D. Brigo (www.damianobrigo.it)

31 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt VIII

The Gaussian distribution has thin tails. Historical estimation
suggests that financial variables usually have fatter tails than in
the Gaussian case.
The future mean of the model, again from (), is
E[rt ] = r0 + kt
and keeps growing linearly in time with rate k. So our model has a
linearly increasing average interest rate that, after a while, will
reach unrealistically high expected values. The larger k, the faster
this unrealistic growth.

32 / 143

EXERCISES

Dynamics for rt

## Exercise with Solution: dynamics for rt IX

The future variance of the model, again from (), is
VAR[rt ] = 2 t.
The variance keeps growing and never reaches a steady state.
The model, in particular, is not mean reverting, as both mean and
variance tend to infinity as time grows.
It is an endogenous model, in that the initial zero curve
T 7 L(0, T ) is an output of the model rather than an input, and
the formula we derived for T 7 L(0, T ) suggests that the model
will only be able to reproduce very stylized initial term structures.

33 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition I

In the G2++ Model, the instantaneous-short-rate process is given by
r (t) = x(t) + y (t) + (t),

r (0) = r0 ,

## where the processes {x(t) : t 0} and {y(t) : t 0} satisfy

dx(t) = ax(t)dt + dW1 (t), x(0) = 0,
dy (t) = by (t)dt + dW2 (t), y (0) = 0,
where (W1 , W2 ) is a two-dimensional Brownian motion with
instantaneous correlation as from
dW1 (t)dW2 (t) = dt,

34 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition II

where r0 , a, b, , are positive constants, and 1 1. The
function is deterministic and well defined in the time interval [0, T ],
with T a given time horizon, typically 10, 30 or 50 (years).
We know that the bond price formula for G2++ model is given by
( Z
T
1 ea(T t)
P(t, T ) = exp
(u)du
x(t)
a
t
)
1 eb(T t)
1

y(t) + V (t, T ) .
b
2

35 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition III

where


2
2 a(T t)
1 2a(T t)
3
V (t, T ) = 2 T t + e

a
2a
2a
a


1 2b(T t)
3
2
2 b(T t)

+ 2 T t + e
b
2b
2b
b
"

ea(T t) 1 eb(T t) 1
+2
T t +
+
ab
a
b
#
e(a+b)(T t) 1

.
a+b

36 / 143

EXERCISES

## a) Write an expression for the instantaneous forward rate f (t, T ) in this

model.
b) Write the SDEs for the dynamics of f (t, T ) and, in particular, the
expression for the instantaneous volatility (t, T ) of f (t, T ).
c) Double check that the SDEs you found in point b) satisfy the HJM
drift condition for no arbitrage.

37 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition V

SOLUTION.
a) With reference to previous calculations done in the course, where
( Z
T
1 ea(T t)
(u)du
P(t, T ) = exp
x(t)
a
t
)
1
1 eb(T t)
y (t) + V (t, T ) ,

b
2
with

38 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition VI



2
2 a(T t)
1 2a(T t)
3
V (t, T ) = 2 T t + e

a
2a
2a
a


1 2b(T t)
3
2
2
e

+ 2 T t + eb(T t)
b
2b
2b
b
"

ea(T t) 1 eb(T t) 1
+2
T t +
+
ab
a
b
#
e(a+b)(T t) 1

,
a+b

39 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition VII

we now write
ln P(t, T )
(ZT
T

1 ea(T t)
1 eb(T t)
=
(u)du +
x(t) +
y (t)
T
a
b
t

1
V (t, T )
2
1 V
,
= (T ) + ea(T t) x(t) + eb(T t) y(t)
2 T

f (t, T ) =

40 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition VIII

where
 2 

V
2 
= 2 1 2ea(T t) + e2a(T t) + 2 1 2eb(T t) + e2b(T t)
T
a
b


a(T t)
b(T t)
+ 2
1e
e
+ e(a+b)(T t)
ab
2 2 
2
2 
= 2 1 ea(T t) + 2 1 eb(T t)
a
b 


a(T t)
1 eb(T t) .
1e
+ 2
ab
So we have
2
2 
a(T t)
1

e
2a2
2


2 

2 1 eb(T t)
1 ea(T t) 1 eb(T t) .
ab
2b

()

41 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition IX

formula to eat x(t),
Since dx(t) = ax(t)dt + dW1 (t), if we apply Itos
we have
d(eat x(t)) = aeat x(t)dt + eat dx(t)
= aeat x(t)dt aeat x(t)dt + eat dW1 (t) = eat dW1 (t).
Integrate from time 0 to t for both side, we get
Z t
eat x(t) =
eas dW1 (s),
0
t

Z
x(t) =

Z
y(t) =

## eb(ts) dW2 (s).

0
(c) 2012-13 D. Brigo (www.damianobrigo.it)

42 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition X

Substitute x(t) and y (t) into the formula () we have just derived for
f (t, T ), and we have
2
2
2 
2 
a(T t)
b(T t)
1

e
1

2a2
2b2 


1 ea(T t) 1 eb(T t)

ab
Z t
Z t
a(ts)
+
e
dW1 (s) +
eb(ts) dW2 (s).

f (t, T ) = (T )

43 / 143

EXERCISES

formula:
b) Apply Itos
df (t, T ) =

f
f
f
1 2f
1 2f
dt +
dXt +
dYt +
dX
dX
+
dYt dYt
t
t
t
X
Y
2 Xt2
2 Yt2
+

2f
dXt dYt .
X Y

## Now we can easily calculate different terms as follows

 2


f
2 a(T t) 
1 ea(T t) + eb(T t) 1 eb(T t)
=
e
t
a
b



a(T t) 
b(T t)
1e
+ eb(T t) 1 ea(T t) )
+ e
b
b
a(T t)
b(T t)
+ ae
x(t) + be
y (t),
f
= ea(T t) ,
X
(c) 2012-13 D. Brigo (www.damianobrigo.it)

f
= eb(T t) ,
Y

44 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XII

and

2f
2f
2f
=
=
= 0.
X Y
Xt2
Yt2

Thus,
f
dt aea(T t) xt dt + ea(T t) dW1,t
t
beb(T t) yt dt + eb(T t) dW2,t
 2
 2


a(T t) 
=
e
1 ea(T t) + eb(T t) 1 eb(T t)
a
b

a(T t) 
1 eb(T t)
+ e
b

i

+ eb(T t) 1 ea(T t) dt
b
+ ea(T t) dW1,t + eb(T t) dW2,t .

df (t, T ) =

45 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XIII

We denote as f (t, T ) the drift of this model, and
f (t, T ) as the
volatility of the model, and the dynamics of f (t, T ) can be written as


dW1,t
df = f (t, T )dt +
f (t, T )
,
dW2,t
where the volatility can be written explicitly as a row vector



## f (t, T ) = ea(T t) , eb(T t) .

This volatility vector however does not represent the whole covariance
of the system, since correlation is also present in the brownian
motions, as we have
dhW1 , W2 it = dW1,t dW2,t = dt.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

46 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XIV

To have a vector volatility representing the entire instantaneous
covariance structure of the system of interest rates, we now
incorporate the correlation into the volatility vector. This is achieved
through a Cholesky decomposition technique.
In dimension 2, this is as follows. We construct
another two

dimensional Brownian Motion B1,t , B2,t (not to be confused with the
bank account B) such that B1,t and B2,t are independent. Assume
W1,t = B1,t
W2,t = B1,t +

p
1 2 B2,t .

## It is immediate to check that the single processes W1 and W2 defined

this way are standard Brownian motions in dimension one, and also
that their quadratic covariation is . Hence we have constructed a
(c) 2012-13 D. Brigo (www.damianobrigo.it)

47 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XV

representation of our two correlated Brownian motion in terms of two
independent ones.
Then, we have



 1
0
dB
1,t
a(T t)
b(T t)
p
df = f (t, T )dt + e
, e
.
dB2,t
1 2

## and by carring out the matrix product in front of dB, we

 obtain the
vector volatility of the model with respect to B1,t , B2,t :

 1
0
p
f (t, T ) = e
, e
1 2



p
= ea(T t) + eb(T t) , 1 2 eb(T t) .


a(T t)

b(T t)

## This vector volatility embeds the whole instantaneous covariation of

the system.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

48 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XVI

c) To check that this model satisfies the HJM drift condition for no
arbitrage, we need to check that
!
Z
T

f (t, T ) = f (t, T )

f (t, s)0 ds

where we have a matrix product in the right hand side and where the
notation 0 denotes transposition. Also, the time integral of vectors is
defined componentwise, namely by integrating each component
function.

49 / 143

EXERCISES

## Exercise: G2++ model and HJM drift condition XVII

We have easily
Z
f (t, T )

!0

f (t, s)ds
t

= f (t, T )

b(T t)
b e

eb(T t) 1


a ea(Tt) 1

12
b

 !
1

 2


2 a(T t) 
e
1 ea(T t) + eb(T t) 1 eb(T t)
a
b



a(T t) 
b(T t)
+ eb(T t) 1 ea(T t)
+ e
1e
b
b
= f (t, T ).

50 / 143

EXERCISES

## Exercise with Solution: LIBOR fourth payoff I

Consider a contract paying in T4 =four years the fourth power of the
LIBOR rate that resets in T3 =three years plus a positive strike K > 0,
if the total of this sum is positive, and zero otherwise.
a) Compute the pricing formula of this product, namely




B(0)
B(0)
4
+
4
+
(L(T3 , T4 ) + K ) = E
(F4 (T3 ) + K )
V =E
B(T4 )
B(T4 )
in a LIBOR market model setting
dF4 (t) = 4 F4 (t)dZ4 , Q 4
as a function of the constant volatility 4 , of the initial forward F4 (0), of
the bond P(0, T4 ) and of the strike K .
b) Is this product sensitive to volatility of interest rates? Motivate
(c) 2012-13 D. Brigo (www.damianobrigo.it)

51 / 143

EXERCISES

## c) If you answered Yes to the previous question, compute the vega

sensitivity of the product price to volatility, namely
V=

V
4

52 / 143

EXERCISES

## Exercise with Solution: LIBOR fourth payoff III

SOLUTION:
a)
We change the numeraire from B to the T4 -forward measure
numeraire P(, T4 ):




B(0)
P(0, T4 )
4
+
4
+
B
4
(F4 (T3 ) + K ) = E
(F4 (T3 ) + K )
E
B(T4 )
P(T4 , T4 )
= P(0, T4 )E 4 [(F4 (T3 )4 + K )+ ].
Now, since the rate F is positive and K is positive too, we have that the
sum F4 (T3 )4 + K is positive in all scenarios,
F4 (T3 )4 + K 0.
It follows that the positive part in the option term is not necessary,
namely
(c) 2012-13 D. Brigo (www.damianobrigo.it)

53 / 143

EXERCISES

## Exercise with Solution: LIBOR fourth payoff IV

(F4 (T3 )4 + K )+ = max(F4 (T3 )4 + K , 0) = F4 (T3 )4 + K
in all scenarios. Hence our pricing formula reduces to
V = P(0, T4 )E 4 [(F4 (T3 )4 + K )+ ] = P(0, T4 )E 4 [F4 (T3 )4 + K ] =
= P(0, T4 )(E 4 [F4 (T3 )4 ] + K ).
All we have to compute then is
E 4 [(F4 (T3 )4 ].
Since we know that the dynamics of F4 under numeraire P(., T4 ) is
dF4 = 4 F4 dZ ,
by Itos formula
(c) 2012-13 D. Brigo (www.damianobrigo.it)

54 / 143

EXERCISES

## Exercise with Solution: LIBOR fourth payoff V

1
d(F4 )4 = 4(F4 )3 dF4 + 4 3 (F4 )2 dF4 dF4
2
d(F4 )4 = 4(F4 )3 4 F4 dZ + 6F42 42 F42 dZdZ
d(F4 )4 = 4(F4 )4 4 dZ + 6F44 42 dt
Set Yt = (F4 (t))4 . From the last equation we have
dY = (642 )Ydt + (44 )YdZ .
This is a Black Scholes geometric brownian motion of type
dY = aYdt + bYdZ
(a = 642 , b = (44 )) whose expected value is
E[Y (T )] = Y0 eaT
(c) 2012-13 D. Brigo (www.damianobrigo.it)

55 / 143

EXERCISES

## Therefore the price is

P(0, T4 )(E 4 [F4 (T3 )4 ] + K ) = P(0, T4 )(E[Y (T3 )] + K ) =
= P(0, T4 )(Y0 eaT3 + K )
2

56 / 143

EXERCISES

## Exercise with Solution: LIBOR fourth payoff VII

b) We see clearly that the price
V = P(0, T4 )(F4 (0)4 e

42

T3

+ K)

## depends on the volatility 4 , so the answer is yes.

c)
h
i
4 e642 T3 + K )

P(0,
T
)(F
(0)
4
4
V
=
V=
4
4
h
i
2
F4 (0)4 e64 T3
2
= P(0, T4 )
= P(0, T4 )F4 (0)4 e64 T3 12 4 T3 .
4

57 / 143

EXERCISES

Swaptions

## Exercise with Solutions: Swaptions I

Consider the following curve of zero coupon bonds for the maturities
T0 = 1y, T1 = 2y , . . . , T9 = 10y:
P(0, T0 ) = 0.961538462 P(0, T1 ) = 0.924556213
P(0, T2 ) = 0.888996359 P(0, T3 ) = 0.854804191
P(0, T4 ) = 0.821927107
a) Compute the forward swap rates S1,4 (0) and S2,4 (0).
b) Compute the forward libor rates F1 (0), F2 (0), ... and verify that the
swap rate S2,4 (0) is a weighted average of the forward LIBOR rates (in
particular, compute the weights).
c) Consider an option to enter a payer swap first resetting at T2 = 3y
and lasting up to T4 = 5y with a fixed rate equal to K = 5%. Let the
volatility of the underlying swap rate be 20%. Compute the price of the
related option.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

58 / 143

EXERCISES

Swaptions

## Exercise with Solutions: Swaptions II

d) Same as c) but the underlying swap is a receiver swap and the fixed
rate is K = 3%.

59 / 143

EXERCISES

Swaptions

## Exercise with Solutions: Swaptions III

Solutions
The first two points have been solved in an earlier exercise above.
c) The annuity numeraire is C2,4 (0) = 1 P(0, T3 ) + 1 P(0, T4 ). We also
have
d1,2 =

ln(0.04/0.05) 21 (0.2)2 3
ln(S2,4 (0)/K ) 21 2 T2

=
.
T2
0.2 3

## Substituting in Blacks formula we have

C2,4 (0)[S2,4 (0)(d1 ) K (d2 )]
and inserting the given numbers we obtain the result.
d) Similar to c). One obtains (for example by put call parity)
C2,4 (0)[K (d1 ) S2,4 (0)(d2 )]
(c) 2012-13 D. Brigo (www.damianobrigo.it)

60 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration I

Consider the following tenor structure:
T0 = 1y, T1 = 2y , T2 = 3y, . . . , T5 = 6y .
Consider the associated forward LIBOR rates Fi (t) = F (t; Ti1 , Ti ),
i = 1, . . . , 6, whose instantaneous volatility we denote by i (t).
Consider the Caplet volatilities vTCaplet
=: vi1 for the caplet resetting at
i1
Ti1 with maturity Ti .
a) Given the following caplet volatilities
v0 = 0.1; v1 = 0.12; v2 = 0.15;
v3 = 0.14; v4 = 0.13; v5 = 0.12
compute the LIBOR model vols i (t) consistent with these data in case
we assume i (t) = ik for t [Tk1 , Tk ] :
(c) 2012-13 D. Brigo (www.damianobrigo.it)

61 / 143

EXERCISES

Inst. Vols
Fwd : F1 (t)
F2 (t)
..
.
FM (t)

t (0, T0 ]
1
2

(T0 , T1 ]
1

(T1 , T2 ]

...
...
...

(TM2 , TM1 ]

...
M

...
M1

...
M2

...
...

...
1

## b) Repeat the same calculation if v5 = 0.08 and notice an important

point on the output you obtain.
c) Given the same input as in part a), compute the LIBOR model
volatilities under the assumption i (t) = i .
Inst. Vols
Fwd : F1 (t)
F2 (t)
..
.
FM (t)

t (0, T0 ]
1
2

(T0 , T1 ]
2

(T1 , T2 ]

...
...
...

(TM2 , TM1 ]

...
M

...
M

...
M

...
...

...
M

62 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration III

d) Again, repeat the same calculation as in c) but with v5 = 0.08. Do
you find the same problems as in point b) ?
e) Given the following separable volatilities, i (t) = i ik for
t [Tk1 , Tk ),
Inst. Vols
Fwd : F1 (t)
F2 (t)
..
.
FM (t)

t (0, T0 ]
1 1
2 2

(T0 , T1 ]
2 1

(T1 , T2 ]

...
...
...

(TM2 , TM1 ]

...
M M

...
M M1

...
M M2

...
...

...
M 1

## compute the caplet volatilities v1 , . . . , v5 :

1 = 1, 2 = 1.1, 3 = 1.2, 4 = 0.9, 5 = 0.8, 6 = 1.1
1 = 0.1, 2 = 0.12, 3 = 0.12, 4 = 0.09, 5 = 0.08, 6 = 0.11
(c) 2012-13 D. Brigo (www.damianobrigo.it)

63 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration IV

f) With the same volatilities as in e), plot also the evolution of the term
structure of caplet volatilities in time up to five years in the future.
g) Find the exponential Full rank instantaneous correlation structure
i,j = + (1 )e|ji| ,
with = 0.2 implying 1,2 = 0.9.
h) With instantaneous correlations as in g) and volatilities as in e),
compute the terminal correlation matrix in three years (at time T2 ) and
say whether it looks acceptable.

64 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration V

SOLUTION:
a) We know that the T0 T1 caplet volatility in the LIBOR model is v0 ,
where
Z T0
Z
1 1y 2
1
2
2
v0 =
1 (t) dt =
1 dt = 12 ,
T0 0
1 0
so that 1 = v0 = 0.1.
Similarly, the T1 T2 caplet volatility is v1 , where
v12
1
= (
2

1y

Z 2y
Z T1
1
1
2
(
(
2 (t)2 dt) =
=
2 (t) dt) =
T1 0
2y 0
2

2y

2 (t) dt +
0

1y

Z 2y
Z
1 1y 2
2 (t) dt) = (
12 dt)
2 dt +
2 0
1y

=
(c) 2012-13 D. Brigo (www.damianobrigo.it)

1 2
( + 22 ),
2 1

65 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration VI

from which
2 =

2v12 12 = 0.1371

Then, analogously,
v22 =
and
3 =

1 2
( + 22 + 12 )
3 3

q
3v2 22 12 = 0.1967,

and similarly
4 = 0.1044, 5 = 0.0781,

6 = 0.0436.

b). With v5 = 0.08 we follow the same procedure, but when we reach
q

6 = 6v52 52 42 32 22 12 = 0.0461 =
(c) 2012-13 D. Brigo (www.damianobrigo.it)

66 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration VII

= 0.2147i
the square root of a negative number, i.e. an imaginary number,
unacceptable as a volatility. Then in this case the chosen
parameterization of volatilities cannot be calibrated to caplet volatilities.
c) As before, we know that the T0 T1 caplet volatility in the LIBOR
model is v0 , where
v02

1
=
T0

Z
0

T0

1
1 (t) dt =
1
2

1y

21 dt = 21 ,

so that 1 = v0 = 0.1.
Similarly, the T1 T2 caplet volatility is v1 , where
v12

Z T1
Z 2y
1
1
2
=
(
2 (t) dt) =
(
2 (t)2 dt) =
T1 0
2y 0

67 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration VIII

=

1
(
2

1y

2 (t)2 dt +

2y

1y

2 (t)2 dt) =

1
(
2

1y

Z
0

22 dt +

2y

1y

22 dt)

1
(222 ) = 22 ,
2

from which
2 = v1 = 0.12
Then, analogously,
v22 =

1 2
( + 23 + 23 )
3 3

and
3 = v2 = 0.15,
and similarly
4 = v3 = 0.14, 5 = v4 = 0.13,
(c) 2012-13 D. Brigo (www.damianobrigo.it)

## Master Exercises Prof. Brigo

6 = v5 = 0.12
Imperial College London

68 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration IX

d) This time v5 = 0.08 gives no problem, since 6 = v5 = 0.08 is fine.
The parameterization never breaks down if the input caplet
volatilities make sense, whereas the parameterization can give
problems for steep caplet volatility graphs, as seen in case b) above.
e) By carrying out multiplications, i ([Tk1 , Tk )) = i ik , as in
Inst. Vols
Fwd : F1 (t)
F2 (t)
..
.
FM (t)

t (0, T0 ]
1 1
2 2

(T0 , T1 ]
2 1

(T1 , T2 ]

...
...
...

(TM2 , TM1 ]

...
M M

...
M M1

...
M M2

...
...

...
M 1

0.1
0.132
0.144
0.081
0.064
0.121

0.11
0.144
0.108
0.072
0.088

0.12
0.108
0.096
0.099

0.09
0.096
0.132

0.08
0.132

0.11

69 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration X

Then
v12

Z T1
Z 2y
1
1
2
(
(
=
2 (t) dt) =
2 (t)2 dt) =
T1 0
2y 0
Z
Z 2y
1 1y
2
= (
2 (t) dt +
2 (t)2 dt) =
2 0
1y
Z
Z 2y
1 1y
2
= (
(2 2 ) dt +
(2 1 )2 dt)
2 0
1y

or, basically, the sum of the squares of the elements in the second row
of the above matrix, each squared being multiplied by the relevant year
fraction (1y in this case):
v12 =

1
(1 0.1322 + 1 0.112 ) = 0.01476, v1 = 0.1215
2

70 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XI

Similarly, v22 is obtained by adding up the squares of the third row and
dividing by 3:
v22 =

1
(1 0.1442 + 1 0.1442 + 1 0.122 ) = 0.01862, v2 = 0.13646
3

and so on.
f) Term structure of caplet volatilities in one year, i.e. at T0 = 1y . From
the formula in an earlier Lecture:
Z Th1
1
2
V (T0 , Th1 ) =
h2 (t)dt, h > 1.
Th1 T0 T0
In particular,
1
V (T0 , T1 ) =
T1 T0
2

T1

T0

22 (t)dt

1
=
1

2y

1y

22 (t)dt =

71 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XII

Z

2y

(2 1 )2 dt = (2 1 )2 = (0.11)2

1y

## so that V (T0 , T1 ) = 0.11.

1
V (T0 , T2 ) =
T2 T0
2

T2

T0

32 (t)dt

Z
1 3y 2
= (
(t)dt) =
2 1y 3

Z
Z 3y
1 2y 2
= (
(t)dt +
32 (t)dt) =
2 1y 3
2y
Z
Z 3y
1 2y
2
= (
(3 2 ) dt +
(3 1 )2 dt) =
2 1y
2y
=

1
((3 2 )2 + (3 1 )2 ) = 0.01757
2

72 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XIII

so that V (T0 , T2 ) = 0.132544. Similarly,
V (T0 , T3 ) = 0.1023, V (T0 , T4 ) = 0.0866, V (T0 , T5 ) = 0.1136
This completes the term structure of caplets volatilities in one year, i.e.
at T0 .
0.11, 0.1325, 0.1023, 0.0866, 0.1136
Notice: it is enough to add the squares (times the relevant year
fraction) in each row of the ziggurat matrix up to the end, starting
from the second column, then dividing by the number of years and
taking square roots. Fast and simple in excel.
For the term structure in two years, i.e. at T1 = 2y , the exercise is
similar:
Z T2
Z
1
1 3y 2
V 2 (T1 , T2 ) =
32 (t)dt =
(t)dt =
T2 T1 T1
1 2y 3
(c) 2012-13 D. Brigo (www.damianobrigo.it)

73 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XIV

Z

3y

(3 1 )2 dt = (3 1 )2 = (0.12)2

2y

## so that V (T1 , T2 ) = 0.12.

1
V (T1 , T3 ) =
T3 T1
2

T3

T1

42 (t)dt

Z
1 4y 2
= (
(t)dt) =
2 2y 4

Z
Z 4y
1 3y 2
= (
(t)dt +
42 (t)dt) =
2 2y 4
3y
Z
Z 4y
1 3y
2
= (
(4 2 ) dt +
(4 1 )2 dt) =
2 2y
3y
=

1
((4 2 )2 + (4 1 )2 ) = 0.009882
2

74 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XV

so that V (T1 , T3 ) = 0.0994.
1
V (T1 , T4 ) =
T4 T1
2

T4

T1

52 (t)dt

Z
1 5y 2
(t)dt) =
= (
3 2y 5

Z
Z 4y
Z 5y
1 3y 2
2
= (
(t)dt +
5 (t)dt +
52 (t)dt) =
3 2y 5
3y
4y
Z 3y
Z 4y
Z 5y
1
2
2
(5 3 ) dt +
(5 2 ) dt +
(5 1 )2 dt) =
= (
3 2y
3y
4y
1
((5 3 )2 + (5 2 )2 + (5 1 )2 ) = 0.008277333
3
so that V (T1 , T4 ) = 0.09098. And so on, with V (T1 , T5 ) = 0.11911.
The term structure of caplet volatilities in 2y (at time T1 ) is
=

## 0.12, 0.0994, 0.09098, 0.11911.

(c) 2012-13 D. Brigo (www.damianobrigo.it)

75 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XVI

Notice that it is enough to add the squares (times the relevant year
fraction) in each row of the ziggurat matrix up to the end, starting
from the third column, then dividing by the number of years and taking
square roots. Very fast in excel.
The term structure in three years is computed similarly:
0.09, 0.099363, 0.1251
and in Four years:
0.08, 0.1215
while, finally, in five years we just have V (T4 , T5 ) = 0.11.
g) If = 0.2, then
1,2 = + (1 )e|21| = + (1 )e0.2 .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

76 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XVII

Since we know that 1,2 = 0.9, we solve
0.9 = + (1 )e0.2
in . We obtain = 0.4483.
h). To compute the terminal correlations in three years (time T2 ) we
use the formula
R T2

i (t)j (t) dt
.
qR
T2 2
T2 2
0 i (t)dt
0 j (t)dt

## Corr(Fi (T2 ), Fj (T2 )) = i,j qR

Let us compute for example Corr(F3 (T2 ), F5 (T2 )), given by the above
formula with i = 3 and j = 5. We need
3,5 = + (1 )e|53| = 0.4483 + (1 0.4483)0.6703
(c) 2012-13 D. Brigo (www.damianobrigo.it)

77 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XVIII

= 0.8181
Compute the numerator:
3y

1y

Z
3 (t)5 (t) dt =

## 3 (t)5 (t) dt+

0

2y

3y

Z
3 (t)5 (t) dt +

+
1y

3 (t)5 (t) dt =
2y

1y

2y

3 3 5 5 dt +

=
0

3 2 5 4 dt+
1y

3y

3 1 5 3 dt =

+
2y

= 3 3 5 5 + 3 2 5 4 + 3 1 5 3 =
(c) 2012-13 D. Brigo (www.damianobrigo.it)

78 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XIX

By substituting from the above ziggurat matrix we get = 0.0311. Again,
it is faster to multiply the corresponding terms in the third and fifth rows
of the ziggurat up to the third column included to get a faster
evaluation:
0.144 0.064 + 0.144 0.072 + 0.12 0.096.
RT
Let us compute the denominators. Notice that 0 2 32 (t)dt is three
times the squared T2 T3 caplet volatility, which is v22 as computed in
RT
point e). We get 0 2 32 (t)dt = 3v22 = 3 0.01862 = 0.05586. Take
square root to get 0.2363 as the first term in the denominator. For the
second term, we need to compute
Z
0
(c) 2012-13 D. Brigo (www.damianobrigo.it)

3y

52 (t)dt

79 / 143

EXERCISES

## Exercise with Solutions: LIBOR model calibration XX

It is enough to sum the squares (times the relevant year fractions) of
the fifth row (i.e. 5 ) in the ziggurat matrix of point e) up to the third
column (i.e. up to three years) included. We get
Z
0

3y

## Square root of this gives us 0.136. Our final calculation is thus

Corr(F3 (T2 ), F5 (T2 )) = 0.8181

0.0311
= 0.7917
0.2363 0.136

Once you have computed the other terminal correlations in three years
and formed the matrix, you check whether the columns are decreasing
when moving away from the diagonals. You also check there are no
negative entries.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

80 / 143

EXERCISES

## Deterministic intensities / hazard rates I

Assume that, following calibration to CDS, we have a piecewise
constant hazard rare (t) that takes the following values.
(t) = 0.02 for 0 <= t < 1y
(t) = 0.04 for 1y <= t < 2y
(t) = 0.02 for 2y <= t < 3y
Compute:
a) The probability of defaulting in one year, two years and three years.
b) The probability of surviving in two 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.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

81 / 143

EXERCISES

## Deterministic intensities / hazard rates II

Solutions.
a) Probability of default in T year is
T

Z
Q( T ) = 1 exp(

(t)dt) =?
0

Compute
1

(t)dt =
0

Compute
Z

Z
(t)dt =

Z
(t)dt +

Z
0.02dt +

(t)dt =
0

0.04dt =
1

## = 0.02 1 + 0.04 1 = 0.06

(c) 2012-13 D. Brigo (www.damianobrigo.it)

82 / 143

EXERCISES

Analogously,
Z

0

## We can now compute the probabilities of defaulting in 1, 2 and 3 years

respectively as
1 exp(0.02), 1 exp(0.06), 1 exp(0.08)
b) The probability of surviving in two years is
Z

Q( > T ) = exp(

(t)dt)
0

## for T = 2y and hence exp(0.06).

(c) 2012-13 D. Brigo (www.damianobrigo.it)

83 / 143

EXERCISES

## c) The default leg price is LGDE[D(0, )1{ 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, )1{ 3y } ] = LGDE[11{ 3y} ] = LGDE[1{ 3y} ] = LGDQ( 3y ) =

= LGD(1exp(0.08)) = (1REC)(1exp(0.08)) = 0.5(1exp(0.08))

84 / 143

EXERCISES

## The CIR model for Credit Risk I

Assume we are given a stochastic intensity process of CIR type,

## dyt = ( yt )dt + yt dW (t)

where y0 , , , are positive constants. W is a brownian motion under
the risk neutral measure.
a) Increasing increases or decreases randomness in the intensity?
And ?
b) The mean of the intensity at future times is affected by k? And by ?
c) What happens to mean of the intensity when time grows to infinity?
d) Is it true that, because of mean reversion, the variance of the
intensity goes to zero (no randomness left) when time grows to infinity?
(c) 2012-13 D. Brigo (www.damianobrigo.it)

85 / 143

EXERCISES

## e) Can you compute a rough approximation of the percentage volatility

in the intensity?
f) Suppose that y0 = 400bps = 0.04, = 0.3, = 0.001 and
= 400bps. Can you guess the behaviour of the future random
trajectories of the stochastic intensity after time 0?
g) Can you guess the spread of a CDS with 10y maturity with the
above stochastic intensity when the recovery is 0.35?

86 / 143

EXERCISES

## The CIR model for Credit Risk III

SOLUTIONS.
a) We can refer to the formulas for the mean and variance of yT in a
CIR model as seen from time 0, at a given T . The formula for the
variance is known to be (see course slides or, for example, Brigo and
Mercurio (2006))
VAR(yT ) = y0

2 T
2
(e
e2T ) + (1 eT )2

## whereas the mean is

E[yT ] = y0 eT + (1 eT )
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
(c) 2012-13 D. Brigo (www.damianobrigo.it)

87 / 143

EXERCISES

lim VAR(yT ) = 0

## so that for very large there is no randomness left.

We can instead see that VAR(yT ) is proportional to 2 , so that if

## increases randomness increases, as is obvious from yt being the

instantaneous volatility in the process y.
b) As the mean is
E[yT ] = y0 eT + (1 eT )
we clearly see that this is impacted by (indeed, speed of mean
reversion) and by clearly (long term mean) but not by the
instantaneous volatility parameter .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

88 / 143

EXERCISES

## c) As T goes to infinity, we get for the mean

lim y0 eT + (1 eT ) =

T +

so that the mean tends to (this is why is called long term mean).

89 / 143

EXERCISES

## The CIR model for Credit Risk VI

d) In the limit where time goes to infinity we get, for the variance
lim [y0

T +

2 T
2
2
(e
e2T ) + (1 eT )2 ] =

2
2

So this does not go to zero. Indeed, mean reversion here implies that
as time goes to infinite the mean tends to and the variance to the
2
constant value 2
, but not to zero.

90 / 143

EXERCISES

## The CIR model for Credit Risk VII

e) Rough approximations of the percentage volatilities in the intensity
would be as follows. The instantaneous variance in dyt , conditional on
the information up to t, is (remember that VAR(dW (t)) = dt)
VAR(dyt ) = 2 yt dt
The percentage variance is


dyt
2
2 yt
VAR
dt
= 2 dt =
yt
yt
yt
and is state dependent, as it depends on yt . We may replace yt with
either its initial value y0 or with the long term mean , both known. The
two rough percentage volatilities estimates will then be, for dt = 1,
s
s
2

= ,
=
y0
y0

91 / 143

EXERCISES

## These however do not take into account the important impact of in

the overall volatility of finite (as opposed to instantaneous) credit
spreads and are therefore relatively useless.

92 / 143

EXERCISES

## The CIR model for Credit Risk IX

f) First we check if the positivity condition is met.
2 = 2 0.3 0.04 = 0.024;

2 = 0.0012 = 0.000001

## hence 2 > 2 and trajectories are positive. Then we observe that

the variance is very small: Take T = 5y ,
VAR(yT ) = y0

2 T
2
(e
e2T ) + (1 eT )2 0.0000006.

Take the standard deviation, given by the square root of the variance:

## STDEV(yT ) 0.0000006 = 0.00077.

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.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

93 / 143

EXERCISES

## Hence there is almost no randomness, and since the initial condition

y0 is the same as the long term mean = 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.
g) In a constant intensity model the CDS spread can be approximated
by
y=

RCDS
RCDS = y (1 REC) = 0.04(1 0.35) = 260bps
1 REC

94 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds I

Consider a default-free zero coupon bond with final maturity T in a
market with constant-in-time and random interest rates r that are
uniformly distributed in [0, R]. Assume interest rates are independent
of the default time of the counterparty.
Denote by P(t, T ) the price of the bond at time t for maturity T .
Assume that the bond portfolio is held at time t by a party A and that
the payment of the notional 1 at maturity T is expected from a
counterparty C.
Assume further that the probability of default of C is associated to an
intensity model with random intensity that is constant in time and
uniformly distributed in the interval [0, L]. Clearly, is independent of r .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

95 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds II

a) Compute the Credit Valuation Adjustment (CVA) for the price of the
bond as seen from A.
b) Compute the CVA sensitivity
CVA
R
to the interest rates range R.
c) Compute the CVA limit when the intensity range contracts to 0,
namely L 0, and comment your finding.
d) Compute the CVA limit when the interest rates range contracts to 0,
namely R 0, and comment your finding.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

96 / 143

EXERCISES

## e) Discuss the CVA calculation in a setting as above but where we no

longer assume r to be uniformly distributed. Instead, we assume that
the instantaneous interest rate r now follows a Vasicek process
drt = k( rt )dt + dWr (t),

r0

## with k, , and r0 positive constants and where Wr is a Brownian

motion under the risk neutral measure and is independent of .
f) Briefly discuss as we could modify the setup in point e) above to
include wrong way risk analysis.

97 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds IV

Solution.
a). We compute
CVA = (1 Rec)E0 [1{ <T } D(0, )(NPV( ))+ ] =
where NPV ( ) is the residual NPV of the bond at the default time of
the counterparty. The residual NPV in our case is the residual value of
the bond at time for maturity T , which is P(, T ).
CVA = (1 Rec)E0 [1{ <T } D(0, )(P(, T ))+ ] =
Since a zero coupon bond price is always positive, the positive part
(...)+ can be removed without affecting our equation:
= (1 Rec)E0 [1{ <T } D(0, )P(, T )] =
Remembering the general definition of zero coupon bond price, namely
(c) 2012-13 D. Brigo (www.damianobrigo.it)

98 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds V

P(t, T ) = Et [D(t, T )], we have
= (1 Rec)E0 [1{ <T } D(0, )E [D(, T )]] =
Since D(0, ) is measurable for E , we can bring it inside the expected
value:
= (1 Rec)E0 [1{ <T } E [D(0, )D(, T )]] =
= (1 Rec)E0 [1{ <T } E [D(0, T )]] =
where we have used D(0, s)D(s, t) = D(0, t). Then we notice that also
1{ <T } is measurable for E , and hence can be brought inside the
expected value
= (1 Rec)E0 [E [1{ <T } D(0, T )]] =

99 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds VI

Now we use the tower property of conditional expectation:
E0 [E [X ]] = E0 [X ], so that
= (1 Rec)E0 [1{ <T } D(0, T )] = (1 Rec)Q( < T )E0 [D(0, T )] =
where we have used independence of and D(0, T ), which follows
from the independence of r and . We obtain
= (1 Rec)Q( < T )P(0, T )
Hence to compute CVA we just need to compute Q( < T ) and
P(0, T ).
We start from
Z
Q{ T } = E[exp(

t dt)] = E[exp(T )] =
0

100 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds VII

since in our case is constant in time. Then
Z L
Z L
1
1
=
exp(xT )p (x)dx =
exp(xT ) dx =
(1 exp(LT ))
L
LT
0
0
where we have used the fact that is uniformly distributed (with
density 1/L) in [0, L]. Then Q{ < T } = 1 Q{ T } above.
The bond price is similarly computed:
Z
P(0, T ) = E0 [exp(

rt dt)] = E0 [exp(rT )] =
0

where we have used the fact that interest rates are constant. We get
Z
=
0

1
1
exp(xT ) dx =
(1 exp(RT ))
R
RT

101 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds VIII

where the calculation is completely analogous to the one for .
By substituting the above expressions we finally obtain the CVA
formula


1
1
CVA = (1 REC) 1
(1 exp(LT ))
(1 exp(RT ))
LT
RT
b)
Set



1
A(L) := (1 REC) 1
(1 exp(LT )) .
LT
B(R) :=

1
(1 exp(RT )).
RT

102 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds IX

A(L) is the part of CVA that does not depend on R. CVA is given by
CVA = A(L) B(R) = A(L)

1
(1 exp(RT )).
RT

## Then by the differentiation rule for a quotient of two functions

CVA
1
= A(L) 2 (RT exp(RT ) + exp(RT ) 1)
R
R T
c)
When L 0 we can compute the limit
lim A(L)B(R) = B(R) lim A(L) =
L0

L0


1
= B(R)(1 REC) lim 1
(1 exp(LT ))
LT
L0
(c) 2012-13 D. Brigo (www.damianobrigo.it)

103 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds X



1
= B(R)(1 REC) 1 lim
(1 exp(LT )) =
L0 LT
Now we can use well known limits of the exponential function, or
expand the exponential function in Taylor series around 0, or again use
De LHopital. In any case we obtain
= B(R)(1 REC)(1 1) = 0.
This result is natural: If the uniform distribution of collapses towards
0, given that dt is the local probability of default around [t, t + dt) for
all t, it follows that when L 0 one has zero probability of defaulting
around [t, t + dt) for all ts. Hence default cannot happen with zero
intensity and CVA is zero because we are pricing counterparty default
risk in a situation where default cannot happen.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

104 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds XI

d)
When R 0 we can compute the limit
lim A(L)B(R) = A(L) lim B(R) =

R0

R0

1
(1 exp(RT )) =
R0 RT

= A(L) lim

## Again, we can use well known limits of the exponential function, or

expand the exponential function in Taylor series around 0, or again use
De LHopital. In any case we obtain
= A(L)1 = A(L)
This is also natural. If interest rates collapse to zero, with R going to
zero, then the zero coupon bond will be worth 1 whatever the maturity.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

105 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds XII

Hence CVA reduces to the probability of default times (1-REC), given
that the residual value of the bond will be 1 regardless of when default
happens.
e) In the Vasicek case for r , given that r is still independent of , all
steps done in point a) until
... to compute CVA we just need to compute Q( < T ) and P(0, T )
are still valid. We still have
CVA = (1 Rec)Q( < T )P(0, T )
except that now P(0, T ) will be the bond price given by the Vasicek
model, namely
P(0, T ) = A(0, T ) exp(B(0, T )r0 )
(c) 2012-13 D. Brigo (www.damianobrigo.it)

106 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds XIII

(write the expressions for A and B from the lectures and write the final
CVA total formula).
f)
To introduce wrong way risk, we need to introduce dependence
between the default event and the Bond underlying interest rates. To
do that, we need to correlate and r . Since we are in an intensity
framework, we need to correlate a possibly random intensity to interest
rates.
A possible way to do that would be to use a CIR++ model for the
intensity: define the default intensity t of the counterparty as
t = yt + (t)

107 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds XIV

where y follows a CIR modela and is a positive deterministic shift
used to fit exactly market implied default probabilities coming either
from CDS or Defaultable Bonds.
If we assume y follows a CIR dynamics under the risk neutral
probability measure Q, we can write

## dyt = ( yt )dt + yt dWy (t).

Now to correlate intensity and rates we introduce a non-null quadratic
covariation between Wr and Wy , namely
dhWr , Wy it = dt.
This creates a link between r and . If is negative, then when the
intensity increases, making default more likely1 , r will tend to
decrease, due to the negative correlation. When r decreases, the
(c) 2012-13 D. Brigo (www.damianobrigo.it)

108 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds XV

Bond price P increases. Let us take the extreme case and set
correlation to = 1. We have
P CVA Amplified.
Due to the totally negative correlation, when increases and default
becomes more likely, we have that the bond increases too, so that the
option embedded in the CVA payoff goes more in the money. Hence
CVA will be bigger due to the effect of correlation. This means that in
this case wrong way risk is given by correlation = 1.

## remember that the intensity is a local probability of default conditional on

not having defaulted earlier
(c) 2012-13 D. Brigo (www.damianobrigo.it)

109 / 143

EXERCISES

## Exercise with Solution: CVA for Call option I

Consider a call option C with final maturity T on a default-free equity
stock S in a market with constant and deterministic interest rate r .
Assume the stock S follows a geometric brownian motion with volatility
under the risk neutral measure and with initial value S0 at time 0.
Assume that the call portfolio is held at time t by a party A and, if the
option is exercised, the underlying S will be delivered at maturity T by
a counterparty C at the agreed strike price K .
Assume further that the probability of default of C is associated to an
intensity model with random intensity that is constant in time and
uniformly distributed in the interval [0, L]. Assume is independent of
the stock price S. The recovery rate associated to default of C is
assumed to be a deterministic constant REC.
Compute the Credit Valuation Adjustment (CVA) for the price of the call
as seen from A.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

110 / 143

EXERCISES

## Exercise with Solution: CVA for Call option II

Solution.
We compute
CVA = (1 Rec)E0 [1{ <T } D(0, )(NPV( ))+ ] =
where NPV ( ) is the residual NPV of the call at the default time of
the counterparty. The residual NPV in our case is the expected value
at time of the discounted payoff of the call, which is
NPV( ) = E [D(, T )(ST K )+ ].
CVA = (1 Rec)E0 [1{ <T } D(0, )(E [D(, T )(ST K )+ ])+ ] =

111 / 143

EXERCISES

## Exercise with Solution: CVA for Call option III

Since the discount factor and the payoff of a call are always positive,
the outer positive part (...)+ can be removed without affecting our
equation:
= (1 Rec)E0 [1{ <T } D(0, )E [D(, T )(ST K )+ ]]
= (1 Rec)E0 [1{ <T } E [D(0, )D(, T )(ST K )+ ]]
= (1 Rec)E0 [1{ <T } E [D(0, T )(ST K )+ ]] =
where we also used the fact that D(0, ) is measurable for E and the
property D(0, s)D(s, t) = D(0, t). Now, noticing that 1{ <T } is also
measurable for E we can then apply the tower property of conditional
expectation E0 [E [X ]] = E0 [X ]
= (1 Rec)E0 [E [1{ <T } D(0, T )(ST K )+ ]] =
= (1 Rec)E0 [1{ <T } D(0, T )(ST K )+ ] =
(c) 2012-13 D. Brigo (www.damianobrigo.it)

112 / 143

EXERCISES

## Exercise with Solution: CVA for Call option IV

We assume independence of and S,
= (1 Rec)E0 [1{ <T } ]E0 [D(0, T )(ST K )+ ] =
= (1 Rec)Q( < T ) CallPrice0 .
Hence, to compute CVA we just need to compute Q( < T ), and the
call price at t = 0. We already know from the previous exercise that
Q( < T ) = 1

1
(1 exp(LT ))
LT

## Assuming the Black-Scholes model, we know that the price of a call is

CallPrice0 = S0 N(d1 ) KerT N(d2 )

113 / 143

EXERCISES

## Exercise with Solution: CVA for Call option V

where
ln(S0 /K ) + (r + 2 /2)T
p
(T )
p
d2 = d1 (T )

d1 =

## By substituting the above expressions we finally obtain the CVA

formula


1
CVA = (1 REC) 1
(1 exp(LT )) (S0 N(d1 ) KerT N(d2 ))
LT

114 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio I

Consider two default-free zero coupon bond withs final maturities T1
and T2 = T , respectively, T1 < T2 , in a market with interest rates that
are independent of credit and default risk.
Denote by P(t, T ) the price of a default-free bond at time t for maturity
T.
Assume that the bonds are held at time t by a bank B and that the
payment of the notionals 1 at maturities T1 and T2 is expected from a
counterparty C.
Try to answer all questions below in terms of default probabilities,
recovery rates, and zero coupon bond prices P(0, Ti ). If at some point
you need information that is not in this dataset, please specify at that
point what further information you would need.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

115 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio II

a) Compute the total Credit Valuation Adjustment (CVA) seen from the
Bank for the two bonds position, assuming that they are not in a netting
agreement with C.
b) Compute the total Credit Valuation Adjustment (CVA) seen from the
Bank for the two bonds position, assuming that they are in a netting
agreement with C.
c) Comment on the differences between a) and b).
d) Repeat points a), b) and c) but in case now we have the following
two bond positions: Short one bond with maturity T1 , long one bond
with maturity T2 .
e) Repeat points a), b) and c) but in case now we have the following
two bond positions: Long one bond with maturity T1 , short one bond
with maturity T2 .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

116 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio III

Solution.
a). If there is no netting agreement, the losses on the two positions,
due to default risk, are to be priced separately, and then they are
We need then to compute CVA for the first bond and CVA for the
second bond, and add them up.
We have seen in a previous exercise that the price of CVA at time 0 for
a zero coupon default-free bond with maturity T is
CVA = (1 Rec)Q( < T )P(0, T ).
This formula is derived exactly in the same way as before. It is enough
to assume that interest rates and default are independent for this
formula to hold (rederive the formula and check this). Hence it holds
also in this case. The total CVA for the two bonds is then
(c) 2012-13 D. Brigo (www.damianobrigo.it)

117 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio IV

CVATOT = CVA1 + CVA2 ,

## CVA1 = (1Rec)Q( < T1 )P(0, T1 ), CVA2 = (1Rec)Q( < T2 )P(0, T2 ).

By substituting the default probabilities formulas derived in the earlier
exercise for the uniform intensity case, we conclude.
b) If there is a netting agreement, the residual values on the two
positions are to be netted at default, before checking positivity.
It is best to write the discounted cash flows of the netted portfolio at a
time t < T2 : we obtain, recalling that T = T2 ,
(t, T ) = 1tT1 D(t, T1 ) 1 + D(t, T2 ) 1.
In other terms, we need to take into account that after T1 , the first bond
expires, and the notional is paid back. Hence it makes a difference
(c) 2012-13 D. Brigo (www.damianobrigo.it)

118 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio V

whether we are looking at a case before T1 or after T1 . Before T1 we
have two bond positions, after T1 we only have one, namely the T2
maturity bond.
We compute the CVA of the two bond positions under netting as
follows:
CVA = (1 Rec)E0 [1{ <T } D(0, )(E [(, T )])+ ] = ...
Now we pause a second to compute
E [(, T )] = E [1 T1 D(, T1 ) 1 + D(, T2 ) 1] =
= 1 T1 E [D(, T1 ) 1] + E [D(, T2 ) 1] = 1 T1 P(, T1 ) + P(, T2 ) 0,
where we have used the fact that 1 T1 is E -measurable, and the
definition of zero coupon bond price. We see that the residual NPV is
(c) 2012-13 D. Brigo (www.damianobrigo.it)

119 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio VI

always non-negative, given that it is the sum of two positive zero
coupon bonds. It follows that
(E [(, T )])+ = (1 T1 P(, T1 )+P(, T2 ))+ = 1 T1 P(, T1 )+P(, T2 )
so that, going back to our CVA calculation, writing LGD = 1-Rec,
... = (1 Rec)E0 [1{ <T } D(0, )(1 T1 P(, T1 ) + P(, T2 ))] =

= LGD E0 [1{ <T } D(0, )1 T1 P(, T1 )]+LGD E0 [1{ <T } D(0, )P(, T2 ))] =
= LGD E0 [1{ <T1 } D(0, )P(, T1 )]+LGD E0 [1{ <T } D(0, )P(, T2 ))] = ...
where we have used the fact that
1{ <T1 } 1{ <T2 } = 1{ <T1 } .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

120 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio VII

The second expected value is the CVA on a single T2 bond position,
and is the same as the CVA2 we found in point a) above.
We can pause for a second to make the first expected value explicit, as
E0 [1{ <T1 } D(0, )(E (D(, T1 )))] = E0 [E (1{ <T1 } D(0, )D(, T1 ))] =
= E0 [E (1{ <T1 } D(0, T1 ))] = E0 [1{ <T1 } D(0, T1 )] = Q( < T1 )P(0, T1 )
where we have used the tower property of expectations and the
independence between default and interest rates discounts D(0, T1 ).
Putting all the pieces together:
... = LGD Q( < T1 )P(0, T1 ) + LGD E0 [1{ <T } D(0, )P(, T2 ))]
So we conclude
CVANETTING = CVA1 + CVA2
(c) 2012-13 D. Brigo (www.damianobrigo.it)

121 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio VIII

exactly as in the previous case a).
c) There is no difference between a) and b). The reason is that, being
the NPV of the two assets positive or zero in all scenarios, and never
negative, the two asset values never offset each other. There is no
scenario where one asset is positive and the other one is negative,
giving some benefit to netting. Hence netting is irrelevant here.
d)
We restart from a), and call this d.a)
d.a) If there is no netting agreement, the losses on the two positions,
due to default risk, are to be priced separately, and then they are
We need then to compute CVA for the first (short) bond and CVA for
the second (long) bond, and add them up.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

122 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio IX

For the second (long) bond, we still have
CVA = (1 Rec)Q( < T2 )P(0, T2 ) = CVA2 .
This formula is derived exactly in the same way as before. For the first
(short) bond, the price of the position E [(, T )] at default time is
P(, T1 )
0

if

T1 ,

if

## > T1 (bond expired before default).

Then the residual NPV is always negative or zero for the first bond, so
that when we take its positive part,
(E [(, T )])+ = (P(, T1 )1{ T1 } )+ = 0,

123 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio X

so that the CVA for the first bond position is zero, since the positive
part of the residual NPV is zero in all scenarios. So in this case
CVATOT = CVA2 .
d.b)
If there is a netting agreement, the residual values on the two positions
are to be netted at default, before checking positivity.
It is best to write the discounted cash flows of the netted portfolio at a
time t < T2 : we obtain, recalling that T = T2 ,
(t, T ) = 1tT1 D(t, T1 ) 1 + D(t, T2 ) 1.
The first bond, with maturity T1 , has a minus because we are in a short
position in that bond. The second bond has a plus because we are in a
long position.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

124 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XI

In other terms, we need to take into account that after T1 , the first bond
expires. Hence it makes a difference whether we are looking at a case
before T1 or after T1 . Before T1 we have two bond positions, after T1
we only have one, namely the T2 maturity bond.
We compute the CVA of the two bond positions under netting as
follows:
CVA = (1 Rec)E0 [1{ <T } D(0, )(E [(, T )])+ ] = ...
Now we pause a second to compute
E [(, T )] = E [1 T1 D(, T1 ) 1 + D(, T2 ) 1] =
= 1 T1 E [D(, T1 ) 1] + E [D(, T2 ) 1] = 1 T1 P(, T1 ) + P(, T2 ),

125 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XII

where we have used the fact that 1 T1 is E -measurable, and the
definition of zero coupon bond price. It follows that
(E [(, T )])+ = (1 T1 P(, T1 ) + P(, T2 ))+ =
= (1 T1 P(, T1 ) + (1 T1 + 1 >T1 )P(, T2 ))+ =
= (1 T1 (P(, T1 ) P(, T2 )) + 1 >T1 P(, T2 ))+
This last expression can take two different values depending on
whether T1 or > T1 . We see that
(E [(, T )])+ = (P(, T1 ) + P(, T2 ))+ = 0 if T1
(E [(, T )])+ = (P(, T2 ))+ = P(, T2 ) if > T1 .
The first row 0 is due to the fact that a zero coupon bond with a
shorter maturity is always worth more than a zero coupon bond with a
(c) 2012-13 D. Brigo (www.damianobrigo.it)

126 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XIII

longer maturity, as we get the money back earlier. Hence
P(, T1 ) > P(, T2 ), which also implies P(, T2 ) P(, T1 ) < 0 and
hence (P(, T2 ) P(, T1 ))+ = 0.
We can write this in a single equation as
(E [(, T )])+ = 1{ >T1 } P(, T2 ).
Going back to our CVA calculation, writing LGD = 1-Rec,
... = LGD E0 [1{ <T } D(0, )1{ >T1 } P(, T2 )] =
= LGD E0 [1{T1 <T2 } D(0, )P(, T2 ) = ...
where we have used 1{ <T } 1{ >T1 } = 1{ <T2 } 1{ >T1 } = 1{T1 < <T2 } .
We continue as usual
... = LGD E0 [1{T1 < <T2 } D(0, )E [D(, T2 )]] =
(c) 2012-13 D. Brigo (www.damianobrigo.it)

127 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XIV

= LGD E0 [E [1{T1 < <T2 } D(0, )D(, T2 )]] =
= LGD E0 [E [1{T1 < <T2 } D(0, T2 )]] = LGD E0 [1{T1 < <T2 } D(0, T2 )] =
= LGD E0 [1{T1 < <T2 } ]E0 [D(0, T2 )] = LGD Q{T1 < < T2 }P(0, T2 ) =
= LGD(Q{ < T2 } Q{ < T1 })P(0, T2 ) =
= CVA2 LGD Q{ < T1 }P(0, T2 ).
We conclude
CVANETTING = CVA2 LGD Q{ < T1 }P(0, T2 ).
d.c) There is now a relevant difference between d.a) and d.b). The
reason is that now the two assets offset each other in several
scenarios. When default happens before T1 , the short T1 bond has a
negative value that dominates, in absolute value, the positive bond T2 .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

128 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XV

Hence the total NPV in this case is negative and there is no
contribution to the CVA payoff. When instead default happens after T1
and before T2 , the first bond is gone and we only have the second
(positive) bond T2 , which is not offset by any other cash flow.
Therefore netting benefits us only in scenarios where default is before
T1 . This is confirmed if we compute the difference between CVATOT
and CVANETTING , giving us the price benefit of netting:
CVATOT CVANETTING = CVA2 (CVA2 LGD Q{ < T1 }P(0, T2 )) =
= LGD Q{ < T1 }P(0, T2 ).
This is the netting benefit.

129 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XVI

e)
We restart from a), and call this e.a)
e.a) If there is no netting agreement, the losses on the two positions,
due to default risk, are to be priced separately, and then they are
We need then to compute CVA for the first (long) bond and CVA for the
second (short) bond, and add them up.
For the first (short) bond, we still have
CVA = (1 Rec)Q( < T1 )P(0, T1 ) = CVA1 .
This formula is derived exactly in the same way as before. For the
second (short) bond, similarly to what we have seen in d.a), we see

130 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XVII

that the residual value at default is always negative, so that this yields
a zero CVA. So in this case
CVATOT = CVA1 .
e.b)
If there is a netting agreement, the residual values on the two positions
are to be netted at default, before checking positivity.
It is best to write the discounted cash flows of the netted portfolio at a
time t < T2 : we obtain, recalling that T = T2 ,
(t, T ) = 1tT1 D(t, T1 ) 1 D(t, T2 ) 1.
The second bond, with maturity T2 , has a minus because we are in a
short position in that bond. The first bond has a plus because we are
in a long position.
(c) 2012-13 D. Brigo (www.damianobrigo.it)

131 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XVIII

In other terms, we need to take into account that after T1 , the first bond
expires. Hence it makes a difference whether we are looking at a case
before T1 or after T1 . Before T1 we have two bond positions, after T1
we only have one, namely the T2 maturity bond.
We compute the CVA of the two bond positions under netting as
follows:
CVA = (1 Rec)E0 [1{ <T } D(0, )(E [(, T )])+ ] = ...
Now we pause a second to compute
E [(, T )] = E [1 T1 D(, T1 ) 1 D(, T2 ) 1] =
= 1 T1 E [D(, T1 ) 1] E [D(, T2 ) 1] = 1 T1 P(, T1 ) P(, T2 ),

132 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XIX

where we have used the fact that 1 T1 is E -measurable, and the
definition of zero coupon bond price. It follows that
(E [(, T )])+ = (1 T1 P(, T1 ) P(, T2 ))+ =
= (1 T1 P(, T1 ) (1 T1 + 1 >T1 )P(, T2 ))+ =
= (1 T1 (P(, T1 ) P(, T2 )) 1 >T1 P(, T2 ))+
This last expression can take two different values depending on
whether T1 or > T1 . We see that
(E [(, T )])+ = (P(, T1 ) P(, T2 ))+ if T1
(E [(, T )])+ = (P(, T2 ))+ = 0 if > T1 .

133 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XX

This is the same as
(E [(, T )])+ = P(, T1 ) P(, T2 ) if T1
(E [(, T )])+ = (P(, T2 ))+ = 0 if > T1
because the positive part in the first row is not acting. Indeed, the
difference P(, T1 ) P(, T2 ) is already positive in all scenarios, as it is
the difference of a more valuable bond with a less valuable one, as we
explained earlier in d.b).
We can write this in a single equation as
(E [(, T )])+ = 1{ <T1 } (P(, T1 ) P(, T2 )).
Going back to our CVA calculation, writing LGD = 1-Rec,
... = LGD E0 [1{ <T } D(0, )1{ <T1 } (P(, T1 ) P(, T2 ))] =
(c) 2012-13 D. Brigo (www.damianobrigo.it)

134 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XXI

= LGD E0 [1{ <T1 } D(0, )(P(, T1 ) P(, T2 ))] =
= LGD E0 [1{ <T1 } D(0, )E [D(, T1 ) D(, T2 )]] =
= LGD E0 [1{ <T1 } E [D(0, )D(, T1 ) D(0, )D(, T2 )]] =
= LGD E0 [1{ <T1 } E [D(0, T1 ) D(0, T2 )]] =
= LGD E0 [E [1{ <T1 } (D(0, T1 ) D(0, T2 ))]] =
= LGD E0 [1{ <T1 } (D(0, T1 ) D(0, T2 ))] =
= LGD E0 [1{ <T1 } ] E0 [D(0, T1 ) D(0, T2 )] =
= LGD Q{ < T1 }[P(0, T1 ) P(0, T2 )].
We conclude CVANETTING = LGD Q{ < T1 }[P(0, T1 ) P(0, T2 )] =
= CVA1 LGD Q{ < T1 }P(0, T2 ).
(c) 2012-13 D. Brigo (www.damianobrigo.it)

135 / 143

EXERCISES

## Exercise with Solution: CVA for Bonds portfolio XXII

e.c) There is now a relevant difference between e.a) and e.b). The
reason is that now the two assets offset each other in several
scenarios. When default happens before T1 , the short T1 bond has a
positive value that dominates, in absolute value, the negative bond T2 .
Hence the total NPV in this case is positive but less than if we had only
the first bond. When instead default happens after T1 and before T2 ,
the first bond is gone and we only have the second (negative) bond T2 ,
which being negative gives no contribution to the CVA payout.
Therefore netting benefits us only in scenarios where default is before
T1 . This is confirmed if we compute the difference between CVATOT
and CVANETTING , giving us the price benefit of netting:
CVATOT CVANETTING = CVA1 (CVA1 LGD Q{ < T1 }P(0, T2 )) =
= LGD Q{ < T1 }P(0, T2 ).
(c) 2012-13 D. Brigo (www.damianobrigo.it)

## Imperial College London

136 / 143

EXERCISES

Risk Measures

Risk Measures I
Consider the dynamics of an equity asset price S in the Black and
Scholes model, under both probability measures P (the Physical or
Historical measure) and Q (the risk neutral measure).
a) Define Value at Risk (VaR) for a time horizon T with confidence
level for a general portfolio.
b) Compute VaR for horizon T and confidence level for a portfolio
with N units of equity, where the equity price follows the Black Scholes
process above.
c) Explain at least one drawback of VaR as a risk measure
d) Is the equity dynamics you used for VaR the same you would have
used to price an equity call option in Black Scholes?

## Imperial College London

137 / 143

EXERCISES

Risk Measures

Risk Measures II
Solutions.
a)
VaR is related to the potential loss on our portfolio over the time
horizon T . Define this loss LT as the difference between the value of
the portfolio today (time 0) and in the future T .
LT = Portfolio0 PortfolioT .
VaR with horizon T and confidence level is defined as that number
q = qT , such that
P[LT < q] =
so that our loss at time T is smaller than q with P-probability .
In other terms, it is that level of loss over a time T that we will not
exceed with probability . It is the P-percentile of the loss distribution
over T .
(c) 2012-13 D. Brigo (www.damianobrigo.it)

138 / 143

EXERCISES

Risk Measures

## Risk Measures III

b)
In Black Scholes the equity process follows the dynamics
dSt = St dt + St dWt ,
where , are positive constants and W is a brownian motion under
the physical measure P.
We know that ST can be written as



1 2
(1)
ST = S0 exp
T + WT ,
2
and recalling the distribution of WT ,




1 2
ST = S0 exp
T + T N (0, 1)
2
(c) 2012-13 D. Brigo (www.damianobrigo.it)

## Imperial College London

(2)

139 / 143

EXERCISES

Risk Measures

Risk Measures IV
so that in our case LT = N(S0 ST ), namely





1 2
LT = NS0 1 exp
T + T N (0, 1)
2
Hence







1 2
q
= P[LT < q] = P
1 exp
T + T N (0, 1)
<
2
NS0





1 2
q
=P
T + T N (0, 1) > ln 1
2
NS0




q
1 2
ln 1 NS

T
2
0
=

= P N (0, 1) >
T
(c) 2012-13 D. Brigo (www.damianobrigo.it)

## Imperial College London

140 / 143

EXERCISES

Risk Measures

Risk Measures V


ln 1

q
NS0


ln 1

q
NS0


12 2 T

= 1
T




q
ln 1 NS
12 2 T
0

=
T
So we have obtained

=
or
1 () =
(c) 2012-13 D. Brigo (www.damianobrigo.it)


ln 1


12 2 T

q
NS0


12 2 T

## Imperial College London

141 / 143

EXERCISES

Risk Measures

Risk Measures VI
and therefore


  

1 2
q
1
exp T () + T = 1
2
NS0
 



1 2
1
q = NS0 1 exp T () + T
2
c) VaR is not subadditive, hence it does not recognize the benefit of
diversification. Also, VaR ignores the structure of the loss distribution
after the percentile. So if 99% VaR is 10 billions, we can have the
remaining 1% loss concentrated
(i) either on 10.1 billions,
(ii) or on 10 trillions,
(c) 2012-13 D. Brigo (www.damianobrigo.it)

142 / 143

EXERCISES

Risk Measures

## as two stylized cases, without VaR being able to tell us anything on

whether we are in case (i) or (ii).
d) No the dynamics is not the same, to price an option we need to use
the risk neutral dynamics, where the drift parameter of S is replaced
by the risk free rate r of the bank account.

143 / 143