MFDC 10 D
MFDC 10 D
MFDC 10 D
u
= q
u
S
u
e
rT
/S
and
q
d
= q
d
S
d
e
rT
/S,
then q
u
> 0, q
d
> 0, and
C
S
= q
u
C
u
S
u
+q
d
C
d
S
d
1 = q
u
+q
d
B
S
= q
u
B
u
S
u
+q
d
B
d
S
d
.
(4)
Thus, we observe that qs are new probabilities and
the ratios are martingales with respect to these new
probabilities.
This is an example of changing the numraire (denom-
inator).
4
The result of Example 10.1 can be general-
ized.
If there is no arbitrage, then given any (non-
dividend-paying) asset B there exist a prob-
ability measure Q
B
such that S(t)/B(t) is a
Q
B
-martingale.
That is,
S(t)
B(t)
= E
B
t
_
S(T)
B(T)
_
, (5)
where T t and E
B
t
denotes conditional ex-
pectation w.r.t Q
B
, given information up to
time point t.
Thus we have the pricing equation
S(t) = B(t)E
B
t
_
S(T)
B(T)
_
. (6)
5
Remark 10.2 If we select B(t) = e
rt
risk free bond, we
get the traditional pricing equation
S(t) = e
rt
E
t
_
S(T)
e
rT
_
= e
r(Tt)
E
t
[S(T)], (7)
where the expectation is with respect to the risk-
neutral martingale measure.
This change of numraire technique has es-
pecially applications in pricing interest rate
derivatives.
It is also useful in pricing stock options as
well.
6
Example 10.2: (Exchange option) Consider an Euro-
pean option that allows to exchange an asset U for
another asset V . The volatilities of U and V are
U
and
V
, the correlation between the assets is , and
the expiration day is T.
If we choose U as the numraire (denominator) in (6),we
obtain
V (t) = U(t)E
U
t
_
V (T)
U(T)
_
. (8)
The payo function of the option at the expiration is
C(T) = max [V (T) U(T), 0]. (9)
Thus,
C(0) = U(0)E
U
_
max[V (T) U(T), 0]
U(T)
_
(10)
or
C(0) = U(0)E
U
_
max
_
V (T)
U(T)
1, 0
__
. (11)
7
U and V are both GBM processes (here non-dividend-
paying), which implies (in the risk-neutral world)
V (t)
U(t)
=
V (0)
U(0)
exp
_
1
2
(
2
V
2
U
)t +
U
W
V
T
V
W
U
T
_
. (12)
From this we nd that the volatility of the U/V -process
is
=
_
2
U
+
2
V
2
U
V
. (13)
From (12) we see that the ratio is again log-normal.
Thus, using the properties of the log-normal distribu-
tion, we get
C(0) = U(0)
_
E
U
_
V (T)
U(T)
_
(d
1
) (d
2
)
_
, (14)
where
d
1
=
log(V
0
/U
0
) +
1
2
2
T
T
(15)
and
d
2
= d
1
T. (16)
8
Because V (T)/U(T) is Q
U
-martingale,
E
U
[V (T)/U(T)] = V (0)/U(0)
and (14) reduces to
C
0
= V
0
(d
1
) U
0
(d
2
), (17)
where C
0
= C(0), U
0
= U(0), and V
0
= V (0).
9