Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Change of measure and Girsanov theorem
80-646-08 Stochastic calculus I
Geneviève Gauthier
HEC Montréal
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
An example I
Let ( Ω , F , fF
t: 0 t T g , P ) be a …ltered probability space on which
a standard Brownian motion W
P= W
tP: 0 t T is constructed.
The stochastic process S = f S
t: 0 t T g represents the evolution of a risky security price and satis…es the stochastic di¤erential equation
dS
t= µS
tdt + σS
tdW
tP.
Let’s also assume that the interest rate r is constant. The discount factor is therefore
β ( t ) = exp ( rt )
which implies that d β ( t ) = r exp ( rt ) dt.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
An example II
Let’s set, for all 0 t T , Y
t= β
tS
ti.e. Y
trepresents the present value at time t of the risky security.
Using Itô’s lemma (more precisely the multiplication rule), we obtain
dY
t= ( µ r ) Y
tdt + σY
tdW
tP. Indeed,
dY
t= d β
tS
t= β
tdS
t+ S
td β
t+ d h β, S i
t= β
tµS
tdt + σS
tdW
tP+ S
t( r β
tdt )
= ( µ r ) β
tS
tdt + σβ
tS
tdW
tP.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
An example III
In its integral form, such a stochastic di¤erential equation becomes
Y
t= Y
0+ ( µ r )
Z
t 0Y
sds + σ Z
t0
Y
sdW
sP.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
Refresher
Itô process
Let W
Pbe a ( fF
tg , P ) Brownian motion.
An Itô process is a process X = f X
t: 0 t T g taking its values in R such that:
X
tX
0+
Z
t0
K
sds +
Z
t0
H
sdW
sPwith K = f K
t: 0 t T g and H = f H
t: 0 t T g , processes adapted to the …ltration fF
tg ,
P h R
T0
j K
sj ds < ∞ i = 1 P h R
T0
( H
s)
2ds < ∞ i = 1
Damien Lamberton and Bernard Lapeyre, Introduction au calcul
stochastique appliqué à la …nance, Ellipses, page 53.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
Example (suite) I
Recall that W
Pis a ( fF
tg , P ) Brownian motion.
In a risk-neutral world ( Ω , F , fF
t: t 0 g , Q ) , the stochastic process Y = f Y
t: 0 t T g should be a ( fF
tg , Q ) martingale.
Thus, under the risk-neutral measure, the trend of Y
should be nil, i.e. we want the drift coe¢ cient to be 0.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
Example (suite) II
Let’s set
W
tQ= W
tP+
Z
t0
γ
sds and note that
1
W
Qis not a P martingale (its expectation varies in time) and
2
dW
tQ= dW
tP+ γ
tdt . As a consequence
Y
t= Y
0+ ( µ r ) Z
t0
Y
sds + σ Z
t0
Y
sdW
sPY
t= Y
0+
Z
t0
( µ r σγ
s) Y
sds + σ Z
t0
Y
sdW
sQ.
In order to get rid of the drift term, it is su¢ cient to set µ r σγ
s= 0 , γ
s= µ r
σ .
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
Example (suite) III
Recall that
Y
t= Y
0+ σ Z
t0
Y
sdW
sQNote that, under the measure P , the process W
Qis not a standard Brownian motion since the law of W
tQunder the measure P is N
µσrt, t .
The process Y will not be a ( fF
tg , P ) martingale since the stochastic integral is constructed with respect to W
Qwhich is not a ( fF
tg , P ) martingale.
Indeed,
E
Ph W
tQi
= µ r
σ t
varies in time.
Change of measure
Example 1 Radon- Nikodym th.
Girsanov th.
Multidimensional References
Example (suite) IV
Recall that W
Pis a ( fF
tg , P ) Brownian motion, Y
t= Y
0+ σ
Z
t0
Y
sdW
sQwhere
W
Q( t ) = W
P( t ) + µ r σ t.
So we want to …nd the probability measure Q to be placed on the space ( Ω , F , fF
tg ) such that W
Qis a
Q standard Brownian motion.
By changing the probability on the set Ω , we transform
the drift coe¢ cient so that the trend becomes zero and we
integrate with respect to a ( fF
tg , Q ) martingale. As a
result, the process Y will be ( fF
tg , Q ) martingale.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem I
A way to construct new probability measures on the measurable space ( Ω , F ) when we already have a
probability measure P existing on that space is as follows:
Let Y be a random variable constructed on the probability space ( Ω , F , P ) such that
8 ω 2 Ω , Y ( ω ) 0 and E
P[ Y ] = 1.
For all event A 2 F , I
Adenotes the indicator function of that event:
I
A( ω ) = 1 if ω 2 A 0 otherwise.
For all event A 2 F , let’s set
Q ( A ) = E
P[ Y I
A] .
Then Q is a probability measure on ( Ω , F ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem II
Proof. We must verify that (P1) Q ( Ω ) = 1,
(P2) 8 A 2 F , 0 Q ( A ) 1,
(P3) 8 A
1, A
2, ... 2 F such that A
i\ A
j= ∅ si i 6 = j,
Q S
i 1A
i= ∑
i 1Q ( A
i) .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem III
Veri…cation of (P1). But, since for all ω, I
Ω( ω ) = 1 and because we have assumed that E
P[ Y ] = 1,
Q ( Ω ) = E
P[ Y I
Ω] = E
P[ Y ] = 1,
which establishes condition ( P 1 ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem IV
Veri…cation of (P2). The second condition is just as easy to prove: since Y is a positive random variable, Y I
Ais a positive random variable too, and Q ( A ) = E
P[ Y I
A] 0.
Moreover,
Q ( A ) = E
P[ Y I
A] E
P[ Y I
Ω]
= E
P[ Y ]
= 1.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem V
Veri…cation of (P3). As we have established in an exercise in the …rst chapter, 8 A
1, A
2, ... 2 F such that A
i\ A
j= ∅ if i 6 = j ,
I
Si 1Ai= ∑
i 1
I
Ai.
As a consequence, Q [
i 1
A
i!
= E
Ph
Y I
Si 1Aii
= E
P"
Y ∑
i 1
I
Ai#
= ∑
i 1
E
P[ Y I
Ai]
= ∑
i 1
Q ( A
i) .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem VI
De…nition
Two probability measures P and Q constructed on the same measurable space ( Ω , F ) are said to be equivalent if they have the same set of impossible events, i.e.
P ( A ) = 0 , Q ( A ) = 0, A 2 F .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem VII
Question. Given two equivalent probability measures P and Q , does there exist a non-negative valued random variable Y such that
Q ( A ) = E
P[ Y I
A] ?
Note the di¤erence between such a problem and the result we have just proven.
In the latter, Y and P were given to us and we have constructed Q .
In this case, P and Q are given to us and we need to …nd Y , which is less easy.
The existence of such a variable is established in the next
theorem which is a version of the famous Radon-Nikodym
theorem.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem VIII
Theorem
Radon-Nikodym theorem . Given two equivalent probability measures P and Q constructed on the measurable space ( Ω , F ) , there exists a positive-valued random variable Y such that
Q ( A ) = E
P[ Y I
A] .
Such a random variable Y is often denoted by
ddQP.
Such a theorem still does not tell us how to …nd our
risk-neutral measure. Actually, it is the next result that
will provide us with the recipe to construct our measure
and it involves the Radon-Nikodym derivative.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem IX
A few thoughts about the discrete case
Assume that Ω only contains a …nite number of elements.
Let Y = β
TX be the present value of the attainable contingent claim X . Si F
0= f Ω , ∅ g , then its price at time t = 0 is
E
Q[ Y ] = ∑
ω2Ω
Y ( ω ) Q ( ω )
= ∑
ω2Ω
Y ( ω ) Q ( ω ) P ( ω ) P ( ω )
= E
PY Q
P
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional References
Radon-Nikodym theorem X
Consider the binomial market model: S
(1)represents the evolution of the riskless asset and S
(2)models a risky asset. The unique risk-neutral measure is denoted by Q , P being the ”real”measure.
ω S(1)0 (ω) S(2)0 (ω)
! S(1)1 (ω) S(2)1 (ω)
! S(1)2 (ω) S(2)2 (ω)
!
P Q dQdP
ω1 (1;2)0 (1,1;2)0 (1,21;1)0 14 0,360 1,44
ω2 (1;2)0 (1,1;2)0 (1,21;3)0 14 0,540 2.16
ω3 (1;2)0 (1,1;4)0 (1,21;1)0 14 0,015 0,06
ω4 (1;2)0 (1,1;4)0 (1,21;5)0. 14 0.085 0,34
The Radon-Nykodym derivative is somewhat the memory
of the change of measure. For each path, it remembers
how we have changed weights.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem I
Let’s focus on a bounded time interval: t 2 [ 0, T ] . Let W = f W
t: t 2 [ 0, T ] g represent a Brownian motion constructed on a …ltered probability space
( Ω , F , fF
tg , P ) such that the …ltration fF
tg is the one generated by the Brownian motion, plus it includes all zero-probability events, i.e. for all t 0,
F
t= σ ( N and W
s: 0 s t ) .
The next theorem will enable to construct our risk- neutral
measures.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem II
Theorem
Cameron-Martin-Girsanov theorem. Let
γ = f γ
t: t 2 [ 0, T ] g be a fF
tg predictable process such that
E
Pexp 1 2
Z
T 0γ
2tdt < ∞ . There exists a measure Q on ( Ω , F ) such that (CMG1) Q is equivalent to P
(CMG2)
ddQP= exp h R
T0
γ
tdW
t 12R
T0
γ
2tdt i
(CMG3) The process f W = n W f
t: t 2 [ 0, T ] o de…ned as f
W
t= W
t+ R
t0
γ
sds is a ( fF
tg , Q ) Brownian motion.
(ref. Baxter and Rennie, page 74; Lamberton and Lapeyre, page 84)
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem III
The condition E
Ph
exp
12R
T0
γ
2tdt i
< ∞ is a su¢ cient
but non-necessary condition. It is know as the Novikov
condition.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem IV
Consider the stochastic di¤erential equation dX
t= b ( X
t, t ) dt + a ( X
t, t ) dW
twhere W represents a Brownian motion on the …ltered probability space ( Ω , F , fF
tg , P ) .
We assume that the drift and di¤usion coe¢ cients are such that there exists a unique solution to the equation, which we denote X .
We want to …nd a probability measure Q , such that, on
the space ( Ω , F , fF
tg , Q ) , the drift of X is e b ( X
t, t )
instead of b ( X
t, t ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem V
Let’s go!
dX
t= b ( X
t, t ) dt + a ( X
t, t ) dW
t= e b ( X
t, t ) dt + a ( X
t, t ) b ( X
t, t ) e b ( X
t, t ) a ( X
t, t )
! dt + a ( X
t, t ) dW
tprovided that a ( X
t, t ) is di¤erent from 0.
= e b ( X
t, t ) dt + a ( X
t, t ) d W
t+
Zt0
b ( X
s, s ) e b ( X
s, s ) a ( X
s, s ) ds
!
= e b ( X
t, t ) dt + a ( X
t, t ) d W f
twhere f
W
t= W
t+
Z
t0
γ
sds and γ
t= b ( X
t, t ) e b ( X
t, t )
a ( X
t, t ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Girsanov theorem VI
If E
Ph
exp
12R
T0
γ
2tdt i
< ∞ then by the
Radon-Nikodym and Cameron-Martin-Girsanov theorems, Q ( A ) = E
Pexp
Z
T0
γ
tdW
t1 2
Z
T0
γ
2tdt I
A, A 2 F and W f = n f W
t: t 2 [ 0, T ] o is a ( F , Q ) Brownian
motion.
In practice, we don’t need to determine the measure Q . It
is su¢ cient for us to know it exists, and to know the
stochastic di¤erential equation of the process of interest
on the space ( Ω , F , fF
tg , Q ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 I
Girsanav theorem
Let’s go back to the Black-Scholes market model. The stochastic process Y = f Y
t: 0 t T g constructed on the space ( Ω , F , fF
tg , P ) used to construct the
Brownian motion represents the evolution of the present value of a risky security where
dY
t= ( µ r ) Y
tdt + σY
tdW
tP.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 II
Girsanav theorem
But, in a risk-neutral world ( Ω , F , fF
tg , Q ) , the trend of Y should be zero, i.e. we want the drift coe¢ cient to be zero. Thus
dY
t= ( µ r ) Y
tdt + σY
tdW
tP= σY
tµ r
σ dt + σY
tdW
tP= σY
td W
tP+ µ r
σ t = σY
tdW
tQwhere
W
tQW
tP+ µ r
σ t = W
tP+
Z
t 0µ r
σ ds.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 III
Girsanav theorem
In the present case,
8 s, γ
s= µ r
σ .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 IV
Girsanav theorem
Recall the Cameron-Martin-Girsanov theorem. Let γ = f γ
t: t 2 [ 0, T ] g be a fF
tg predictable process such that
E
Pexp 1 2
Z
T0
γ
2tdt < ∞ . There exists a measure Q on ( Ω , F ) such that (CMG1) Q is equivalent to P
(CMG2)
ddQP= exp h R
T0
γ
tdW
t 1 2R
T0
γ
2tdt i
(CMG3) The process f W = n W f
t: t 2 [ 0, T ] o de…ned as W f
t= W
t+ R
t0
γ
sds is a ( fF
tg , Q ) Brownian motion.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 V
Girsanav theorem
Let’s verify that the condition on the process γ is indeed satis…ed:
E
Pexp 1 2
Z
T0
γ
2tdt
= E
P"
exp 1 2
Z
T 0µ r σ
2
dt
!#
= exp 1 2
µ r σ
2
T
!
< ∞ .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 VI
Girsanav theorem
Let’s apply Girsanov theorem : d Q
d P = exp
Z
T0
γ
tdW
tP1 2
Z
T 0γ
2tdt
= exp
" Z
T 0
µ r
σ dW
tP1 2
Z
T0
µ r σ
2
dt
#
= exp
"
µ r
σ W
TP1 2
µ r σ
2
T
# .
This implies that Q [ A ] = E
P"
exp µ r
σ W
TP1 2
µ r σ
2
T
! I
A#
, A 2 F .
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 VII
Girsanav theorem
Moreover, under the measure Q , the evolution of the present value of the risky security satis…es the equation
dY
t= σY
tdW
tQwhere W
Qis a Q Brownian motion.
We can also deduce the stochastic di¤erential equation satis…ed by the evolution of the risky security price S :
dS
t= µS
tdt + σS
tdW
tP= µS
tdt + σS
td W
tQµ r σ t puisque W
tQW
tP+ µ r
σ t
= µS
tdt + σS
tdW
tQσS
tµ r σ dt
= rS
tdt + σS
tdW
tQ.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 VIII
Girsanav theorem
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 IX
Girsanav theorem
Note that we don’t really need to calculate Q , we simply need to know that it exists then toestablish what is the equation satis…ed by the processus of interest, i.e. the evolution of the risky security price. Indeed, on ( Ω , F , fF
tg , Q ) ,
dS
t= rS
tdt + σS
tdW
tQwhere W f is a Q Brownian motion.
But the unique solution to that equation is S
t= S
0exp r σ
22 t + σW
tQ.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 X
Girsanav theorem
Since the price of a call option, the strike price of which is K and the maturity of which is T , is given by
E
Q[ exp ( rT ) max ( S
TK ; 0 )]
= E
Qexp ( rT ) max S
0exp r σ
22 T + σW
TQK ; 0
= E
Qmax S
0exp σ
22 T + σW
TQK exp ( rT ) ; 0
=
Z∞∞
max S
0exp σ
22 T + σz Ke
rT; 0 f
Z( z ) dz.
where f
Z( ) represents the probability density function of a normal random variable with zero expectation and variance T . The rest of the calculation is a pure application of the
properties of the normal law.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 XI
Girsanav theorem
Since
S
0exp σ
22 T + σz > Ke
rT, σ
2
2 T + σz > ln Ke
rTS
0= ln K rT ln S
0, z >
ln S
0ln K + r
σ22T
σ d
2p
T
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 XII
Girsanav theorem
Z
∞∞
max S
0exp σ
22 T + σz Ke
rT; 0 f
Z( z ) dz
=
Z
∞d2p T
S
0exp σ
22 T + σz Ke
rTf
Z( z ) dz
=
Z
∞d2
pT
S
0exp σ
22 T + σz f
Z( z ) dz Z
∞d2
pT
Ke
rTf
Z( z ) dz
= S
0Z
∞d2p T
p 1 2π
p 1
T exp z
22T σz + σ
2T
22T dz
Ke
rTZ
∞d2
pT
p 1 2π
p 1
T exp z
22T dz
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 XIII
Girsanav theorem
= S
0Z
∞d2
pT
p 1 2π
p 1
T exp ( z σT )
22T
! dz
Ke
rTZ
∞d2p T
p 1 2π
p 1
T exp z
22T dz Let’s set u = z p σT
T and v = p z T
= S
0Z
∞d2 σp T
p 1
2π exp u
22 du Ke
rTZ
∞ d2p 1
2π exp v
22 dv
= S
01 N d
2σ p
T Ke
rT( 1 N ( d
2))
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 XIV
Girsanav theorem
where N ( ) is the cumulative distribution function of a
standard normal random variable.
Change of measure Radon- Nikodym th.
Girsanov th.
Example 1
Multidimensional References
Example 1 XV
Girsanav theorem
But the symmetry of N implies that 1 N ( x ) = N ( x ) . Then
S
01 N d
2σ p
T Ke
rT( 1 N ( d
2))
= S
0N d
2+ σ p
T Ke
rT( N ( d
2))
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 I
Let’s assume that W
Pand W f
Prepresent two standard Brownian motions constructed on the …ltered probability space ( Ω , F , fF
tg , P ) .
Note that n e
B
tP: t 0 o where B e
tρW
tP+
q
1 ρ
2W f
tPis a standard Brownian motion such that
Corr
PW
tP, B e
tP= ρ.
Exercise: prove it.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 II
The instantaneous exchange rate
dC
t= µ
CC
tdt + σ
CC
tdW
tPenables us to model the number of Canadian dollars per unit of foreign currency at any time.
Suppose also that the stochastic di¤erential equation dS
t= µ
SS
tdt + σ
SS
td B e
tP= µ
SS
tdt + σ
SρS
tdW
tP+ σ
Sq
1 ρ
2S
td W f
tPmodels the evolution of a foreign risky asset price.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 III
Lastly, the Canadian instantaneous interest rate r and the foreign instantaneous interest rate v are assumed to be constant. As a consequence, the discount factor is
β
t= exp ( rt ) .
and the value in foreign currency of an initial investment
equal to one foreign currency unit is B
t= exp ( vt ) .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 IV
Let’s put ourselves in the shoes of a Canadian investor.
C
tS
tgives us the Canadian dollar value of a risky asse at time t ,
C
tB
tgives us the Canadian dollar value, at time t , of one foreign currency unit invested in a foreign bank account U
t= β
tC
tS
tgives us the Canadian dollar present value of the risky asset at time t.
V
t= β
tC
tB
tgives us the Canadian dollar present value, at time t, of one foreign currency unit invested in a foreign bank account.
We wish to …nd a measure Q , such that the stochastic
processes U and V are ( fF
tg , Q ) martingales.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 V
Recall:
dC
t= µ
CC
tdt + σ
CC
tdW
tP, dB
t= vB
tdt
d β
t= r β
tdt
First, let’s determine the stochastic di¤erential equation satis…ed by the Canadian dollar present value of one foreign currency unit invested in a foreign bank account V = βCB under the measure P . Itô’s lemma allows us to write
dC
tB
t= C
tdB
t+ B
tdC
t+ d h B, C i
t= C
t( vB
tdt ) + B
tµ
CC
tdt + σ
CC
tdW
tP= ( v + µ
C) C
tB
tdt + σ
CC
tB
tdW
tP.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 VI
Thus,
dV
t= d β
tC
tB
t= β
tdC
tB
t+ C
tB
td β
t+ d h β, CB i
t= β
t( v + µ
C) C
tB
tdt + σ
CC
tB
tdW
tP+ C
tB
t( r β
tdt )
= ( µ
C+ v r ) β
tC
tB
tdt + σ
Cβ
tC
tB
tdW
tP= ( µ
C+ v r ) V
tdt + σ
CV
tdW
tP.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 VII
Recall:
dS
t= µ
SS
tdt + σ
SρS
tdW
tP+ σ
Sq
1 ρ
2S
td W f
tP, dC
t= µ
CC
tdt + σ
CC
tdW
tP,
dB
t= vB
tdt,
d β
t= r β
tdt .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 VIII
Second, let’s determine the stochastic di¤erential equation satis…ed by U = βCS under the measure P . Itô’s lemma allows us to write
dC
tS
t= C
tdS
t+ S
tdC
t+ d h S , C i
t= C
tµ
SS
tdt + σ
SρS
tdW
tP+ σ
Sq
1 ρ
2S
td W f
tP+ S
tµ
CC
tdt + σ
CC
tdW
tP+ σ
SρS
tσ
CC
tdt
= ( µ
S+ µ
C+ σ
Sσ
Cρ ) C
tS
tdt + ( σ
Sρ + σ
C) C
tS
tdW
tP+ σ
q
1 ρ
2C
tS
td W f
tP.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 IX
Applying Itô’s lemma again, we obtain
dU
t= d β
tC
tS
t= β
tdC
tS
t+ C
tS
td β
t+ d h β, CS i
t= β
t( µ
S+ µ
C+ σ
Sσ
Cρ ) C
tS
tdt + ( σ
Sρ + σ
C) C
tS
tdW
tP+ σ
Sp
1 ρ
2C
tS
td W f
tP+ C
tS
t( r β
tdt )
= ( µ
S+ µ
C+ σ
Sσ
Cρ r ) β
tC
tS
tdt + ( σ
Sρ + σ
C) β
tC
tS
tdW
tP+ σ
Sq
1 ρ
2β
tC
tS
td W f
tP= ( µ
S+ µ
C+ σ
Sσ
Cρ r ) U
tdt + ( σ
Sρ + σ
C) U
tdW
tP+ σ
Sq
1 ρ
2U
td f W
tP.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 X
We have
dV
t= ( µ
C+ v r ) V
tdt + σ
CV
tdW
tP, dU
t= ( µ
S+ µ
C+ σ
Sσ
Cρ r ) U
tdt
+ ( σ
Sρ + σ
C) U
tdW
tP+ σ
Sq
1 ρ
2U
td W f
tPwhich allows us to write
dV
t= ( µ
C+ v r σ
Cγ
t) V
tdt + σ
CV
td W
tP+
Z t 0
γ
sds dU
t= µ
S+ µ
C+ σ
Sσ
Cρ r
( σ
Sρ + σ
C) γ
tσ
Sp
1 ρ
2e γ
tU
tdt + ( σ
Sρ + σ
C) U
td W
tP+
Zt 0
γ
sds + σ
Sq
1 ρ
2U
td W f
tP+
Zt0
e γ
sds .
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 XI
So, we want to solve the linear system
µ
C+ v r σ
Cγ
t= 0 µ
S+ µ
C+ σ
Sσ
Cρ r ( σ
Sρ + σ
C) γ
tσ
Sq
1 ρ
2γ e
t= 0
the unknowns of which are γ
tand γ e
t. In matrix form, we write
σ
C0
σ
Sρ + σ
Cσ
Sp 1 ρ
2γ
te
γ
t= µ
C+ v r
µ
S+ µ
C+ σ
Sσ
Cρ r .
The solution is
γ
t= µ
C+ v r σ
Ce
γ
t= µ
Sv + σ
Sσ
Cρ σ
Sp 1 ρ
2ρ ( µ
C+ v r ) σ
Cp 1 ρ
2Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 XII
So, let’s set W
tQ= W
tP+ R
t0
γ
sds and W f
tQ= f W
tP+ R
t0
γ e
sds where
γ
s= µ
C+ v r σ
Ce
γ
s= µ
Sv + σ
Sσ
Cρ σ
Sp 1 ρ
2ρ ( µ
C+ v r ) σ
Cp 1 ρ
2.
We can then write dV
t= σ
CV
tdW
tQdU
t= ( σ
Sρ + σ
C) U
tdW
tQ+ σ
Sq
1 ρ
2U
td W f
tQ.
Is it possible to …nd a measure Q such that W
Qand W f
Qare ( fF
tg , Q ) Brownian motions simultaneously?
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Girsanav theorem I
Let W = W
(1), ..., W
(n)be a Brownian motion with dimension n, i.e. its components are independent standard Brownian motions on the …ltered probability space
( Ω , F , fF
tg , P ) Theorem
Cameron-Martin-Girsanov theorem. For all i 2 f 1, ..., n g , γ
(i)= γ
t(i): 0 t T is a fF
tg predictable process such that
E
Pexp 1 2
Z
T0
γ
(ti)2
dt < ∞ .
There exists a measure Q on ( Ω , F ) such that
(CMG1) Q is equivalent to P
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Girsanav theorem II
(CMG2)
ddPQ= exp ∑
ni=1R
T0
γ
(ti)dW
t(i) 12R
T0
∑
ni=1γ
(ti)2
dt (CMG3) For all i 2 f 1, ..., n g , the process
f
W
(i)= W f
t(i): 0 t T de…ned as f
W
t(i)= W
t(i)+ R
t0
γ
(si)ds is a ( fF
tg , Q ) Brownian motion.
(ref. Baxter and Rennie, page 186)
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 (continued) I
Since the functions γ and γ e are constant, the Novikov condition is satis…ed and Girsanov theorem
(multidimensional version) allows to conclude there exists
a martingale measure Q such that W
Qand W f
Qare
( fF
tg , Q ) Brownian motions.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 (continued) II
An interesting fact, on the space ( Ω , F , fF
tg , Q ) , we have the stochastic di¤erential equation satis…ed by the instantaneous exchange rate
dC
t= µ
CC
tdt + σ
CC
tdW
tP= µ
CC
tdt + σ
CC
td W
tQµ
C+ v r σ
Ct
= ( r v ) C
tdt + σ
CC
tdW
tQ.
The di¤erence between the domestic and foreign
instantaneous interest rates can be recognized in the drift
coe¢ cient.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 (continued) III
Again on the space ( Ω , F , fF
tg , Q ) , the stochastic di¤erential equation for the evolution of the risky asset Canadian dollar price is
dCtSt
= (µS+µC+σSσCρ)CtSt dt + (σSρ+σC)CtSt dWtP+σS
q
1 ρ2CtSt dWftP
= (µS+µC+σSσCρ)CtSt dt
+ (σSρ+σC)CtSt d WtQ µC+v r σC
t
+σS
q
1 ρ2CtSt d WftQ µS v+σSσCρ σS
p1 ρ2
ρ(µC+v r) σC
p1 ρ2
! t
!
= rCtSt dt+ (σSρ+σC)CtSt dWtQ+σS
q
1 ρ2CtSt dWftQ
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References
Example 2 (continued) IV
where the last equality is obtained by simplifying the drift coe¢ cient
( µ
S+ µ
C+ σ
Sσ
Cρ ) ( σ
Sρ + σ
C) µ
C+ v r σ
Cσ
Sq
1 ρ
2µ
Sv + σ
Sσ
Cρ σ
Sp 1 ρ
2ρ ( µ
C+ v r ) σ
Cp 1 ρ
2!
.
Change of measure Radon- Nikodym th.
Girsanov th.
Multidimensional Example 2 Girsanov Example 2 (continued) Example 3 References