th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Attainable contingent claims and replication strategies
80-646-08 Stochastic calculus I
Geneviève Gauthier
HEC Montréal
th.
Self-…nancing strategies
A riskless asset and a risky asset Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Self-…nancing trading strategies I
Assume that B = f B
t: 0 t T g and
S = f S
t: 0 t T g represent the evolutions of the riskless asset price and the risky asset price respectively.
Moreover, the riskless asset evolution is described by the di¤erential equation
dB
t= rB
tdt .
th.
Self-…nancing strategies
A riskless asset and a risky asset Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Self-…nancing trading strategies II
A trading strategy is a predictable portfolio process f ( φ
t, ψ
t) : 0 t T g
where
φ
t= the number of riskless asset shares held at time t and ψ
t= the number of risky asset shares held at time t . The value of such a trading strategy at time t is
V
t= φ
tB
t+ ψ
tS
t.
th.
Self-…nancing strategies
A riskless asset and a risky asset Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Self-…nancing trading strategies III
By applying the multiplication rule, we have that dV
t= φ
tdB
t+ B
td φ
t+ d h φ, B i
t+ ψ
tdS
t+ S
td ψ
t+ d h ψ, S i
t. De…nition
A trading strategy is said to be self-…nancing if dV
t= φ
tdB
t+ ψ
tdS
t,
i.e. the variations in the strategy value are only caused by the asset price variations and do not depend on the portfolio weight ‡uctuations. A strategy is therefore self-…nancing, if
B
td φ
t+ S
td ψ
t+ d h φ, B i
t+ d h ψ, S i
t= 0.
th.
Self-…nancing strategies
A riskless asset and a risky asset Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Self-…nancing trading strategies IV
De…nition
Let X be contingent claim. X is therefore a
non-negative-valued, F
Tmeasurable, random variable. The contingent claim X is attainable if there exists a self-…nancing trading strategy f ( φ
t, ψ
t) : 0 t T g such that
V
T= φ
TB
T+ ψ
TS
T= X .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Martingale representation theorem I
Let W = f W
t: 0 t T g be a Brownian motion constructed on a …ltered probability space
( Ω , F , fF
tg , Q ) 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 ) .
Let Z = f Z
t: 0 t T g be an Itô process such that Z
t Q=
p.s.Z
0+
Z
t0
H
sdW
swhere the fF
tg adapted process H = f H
t: 0 t T g
satis…es the conditions
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Martingale representation theorem II
E
Qh R
T0
H
s2ds i
< ∞ and
Q f ω 2 Ω : 9 t 2 [ 0, T ] such that H
t( ω ) = 0 g = 0.
Theorem
Martingale representation theorem.
Z is a ( Q , fF
tg ) martingale (Z is sometimes called a Brownian martingale, since the …ltration is generated by the Brownian motion);
moreover, for any other ( Q , fF
tg ) martingale M, there exists a predictable process φ = f φ
t: 0 t T g such that M
t= M
0+ R
t0
φ
sdZ
s.
That theorem will be useful when we must prove that a contingent claim is attainable.
Lamberton and Lapeyre, page 74, or Baxter and Rennie, page 78.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication
Example Multidimensional References Appendix
Steps in a contingent claim replication I
Find a martingale measure Q under which the process Z = B
1S for the present value of the underlying asset price is a martingale.
Construct the process M where
M
tE
QB
T1X jF
t, 0 t T .
Such a process is a martingale.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication
Example Multidimensional References Appendix
Steps in a contingent claim replication II
Using the martingale representation theorem, we know there exists a predictable process ψ such that
dM
t= ψ
tdZ
t= ψ
tS
tdB
t 1+ B
t 1dS
t= ψ
trB
t 1S
tdt + B
t 1dS
t.
Baxter and Rennie, pages 84-85.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication
Example Multidimensional References Appendix
Steps in a contingent claim replication III
So, let’s set
ψ
t= the number of risky asset shares held at time t and φ
t= the number of riskless asset shares held at time t where
φ
tM
tψ
tB
t 1S
t. The present value of such a strategy at time t is
B
t 1V
t= B
t 1( φ
tB
t+ ψ
tS
t)
= φ
t+ ψ
tB
t 1S
t= M
tψ
tB
t 1S
t+ ψ
tB
t 1S
t= M
t. Note that, at time T ,
V
T= B
TM
T= B
TE
QB
T1X jF
T= X .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication
Example Multidimensional References Appendix
Steps in a contingent claim replication IV
Recall :
dM
t= ψ
trB
t 1S
tdt + B
t 1dS
tφ
tM
tψ
tB
t 1S
tand V
t= B
tM
t. Is such a strategy self-…nancing?
dV
t= dB
tM
t= M
tdB
t+ B
tdM
t= φ
t+ ψ
tB
t 1S
tdB
t+ B
tψ
trB
t 1S
tdt + B
t 1dS
t= φ
tdB
t+ ψ
tB
t 1S
tdB
trψ
tS
tdt + ψ
tdS
t= φ
tdB
t+ rψ
tB
t 1S
tB
tdt r ψ
tS
tdt + ψ
tdS
t= φ
tdB
t+ ψ
tdS
twhich implies that the strategy indeed is self-…nancing.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example I
Consider the classic case. An asset price is modeled using a geometric Brownian motion
dS
t= µS
tdt + σS
td W f
twhere W f is a ( Ω , F , fF
t: t 0 g , P ) Brownian motion.
The riskless interest rate r is set constant.
The riskless asset is therefore
B
t= exp ( rt ) .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example II
In a risk-neutral world, the asset price is dS
t= rS
tdt + σS
tdW
twhere W is a ( Ω , F , fF
t: t 0 g , Q ) Brownian motion. Since the risk-neutral measure is unique, the market is complete and, as a consequence, all contingent claims C can be replicated.
Recall that a contingent claim is replicated if there exists a
self-…nancing trading strategy, the value of which at time
T is C .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example III
The present value of a contingent claim is a Q martingale. Indeed, if
M
t= E
Q[ β
TC jF
t] ,
then M
tis the present value at time t of the contingent
claim C and f M
t: t 0 g is a Q martingale.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example IV
Moreover, the present value of the risky asset is a Q martingale since
dB
t 1S
t= σB
t 1S
tdW
tand
E
QZ
T0
σB
t 1S
t 2dt
=
Z
T 0E
Qh
σB
t 1S
t 2i
dt (Fubini thm)
= σ
2Z
T0
E
QS
02exp σ
2t + 2σW
tdt
= S
02σ
2Z
T0
exp σ
2t dt = S
02e
Tσ21 < ∞ .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example V
Since B
t 1S
t: t 0 is a martingale, the di¤usion coe¢ cient of which, σB
t 1S
t, is always strictly positive, the representation theorem yields that there exists a predictable process ψ such that
dM
t= ψ
tdB
t 1S
t. Note that
dB
t 1S
t= σB
t 1S
tdW
timplies that
dM
t= ψ
tσB
t 1S
tdW
t. (1)
We now have the existence of a predictable process, but
we don’t know yet how to determine that process.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example VI
On the other hand, if we assume that the value of the contingent claim C can be expressed as a twice continuously di¤erentiable function of time and the risky asset price, i.e.
f ( t, S
t) , then Itô’s lemma implies that df ( t, S
t)
= ∂f
∂t ( t, S
t) dt + ∂f
∂s ( t, S
t) dS
t+ 1 2
∂
2f
∂s
2( t, S
t) d h S i
t= ∂f
∂t ( t, S
t) dt + ∂f
∂s ( t, S
t) ( rS
tdt + σS
tdW
t) + 1
2
∂
2f
∂s
2( t , S
t) σ
2S
t2dt
= ∂f
∂t ( t, S
t) + rS
t∂f
∂s ( t, S
t) + σ
2
2 S
t2∂
2f
∂s
2( t, S
t) dt + σS
t∂f
∂s ( t, S
t) dW
t.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example VII
Using Itô’s lemma again (multiplication rule), we …nd the stochastic di¤erential equation for the present value of the contingent claim:
dM
t= dB
t 1f ( t, S
t)
= f ( t, S
t) dB
t1+ B
t 1df ( t, S
t)
= rf ( t, S
t) B
t 1dt + B
t 1df ( t, S
t)
=
∂f∂t
( t, S
t) + rS
t∂f
∂s
( t, S
t) +
σ2
2
S
t2∂2f
∂s2
( t, S
t) rf ( t, S
t) B
t1dt +
σBt 1S
t∂f∂s
( t, S
t) dW
t.th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example VIII
Recall that in equation (1), we had
dM
t= ψ
tσB
t 1S
tdW
t. We now have
dM
t=
∂f∂t
( t, S
t) + rS
t∂f
∂s
( t, S
t) +
σ2
2
S
t2∂2f
∂s2
( t, S
t) rf ( t, S
t) B
t 1dt +
σBt 1S
t∂f∂s
( t, S
t) dW
t.So the drift coe¢ cients and the di¤usion coe¢ cients in these two equations must be the same, i.e.
∂f
∂t ( t, S
t) + rS
t∂f
∂s ( t, S
t) + σ
2
2 S
t2∂
2f
∂s
2( t, S
t) rf ( t, S
t) = 0 and
σB
t 1S
t∂f
∂s ( t, S
t) = ψ
tσB
t 1S
t.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example IX
By isolating ψ
tin the equation σB
t 1S
t∂f
∂s ( t, S
t) = ψ
tσB
t 1S
t, we …nd
ψ
t= ∂f
∂s ( t, S
t) .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example X
We must therefore hold ψ
t=
∂f∂s( t, S
t) risky asset shares and
φ
t= M
tψ
tB
t 1S
triskless asset shares.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example XI
Now assume that the contingent claim is a call option C = max ( S
TK , 0 ) . We know that the value of such a contingent claim at time t is
f ( t, S
t) = S
tN ( d ( S
t)) Ke
r(T t)N d ( S
t) σ p T t where N ( ) is the cumulative distribution function of the standard normal law and
d ( s ) =
ln s ln K + r +
σ22( T t )
σ ( T t )
1/2.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example XII
Its present value is M
t= B
t 1f ( t, S
t)
= e
rtS
tN ( d ( S
t)) Ke
rTN d ( S
t) σ p
T t
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example XIII
But, the numbers of risky asset and riskless asset shares are respectively
ψ
t= ∂f
∂s ( t, S
t)
= N ( d ( S
t)) and φ
t= M
tψ
tB
t 1S
t= Ke
rTN d ( S
t) σ p
T t
The derivatives are calculated in the Appendix 1.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example XIV
The value of the strategy at time t is V
t= ψ
tS
t+ φ
tB
t= N ( d ( S
t)) S
tKe
r(T t)N d ( S
t) σ p
T t .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example
Multidimensional References Appendix
An example XV
It is possible to verify that the strategy does replicate the call option since, at time T ,
d ( S
T) = lim
t!T
ln S
Tln K + r
σ22( T t ) σ ( T t )
1/2= 8 <
:
∞ if S
TK
∞ if S
T< K So, the value of our strategy is
V
T=
limt!T
N ( d ( S
T)) S
TKe
r(T t)N d ( S
T) +
σp T t
=
8<:
S
TK
ifS
TK
0 ifS
T< K
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
Martingale representation theorem I
Obviously, all of this exists in a multidimensional version!
Let W = W
(1), ..., W
(n)be a Brownian motion in n dimensions constructed on a …ltered probability space ( Ω , F , fF
tg , Q ) 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(1), ..., W
s(n): 0 s t .
The components of such a Brownian motion are
independent from each other.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
Martingale representation theorem II
Let Z
(i)= n Z
t(i): 0 t T o
be an Itô process such that
Z
t(i)Q=
p.s.Z
0(i)+
∑
n j=1Z
t 0H
s(i,j)dW
s(j)where, for all i 2 f 1, ..., n g and j 2 f 1, ..., n g , the fF
tg adapted process H
ij= f H
ij( t ) : 0 t T g satis…es the conditions
E
QZ
T0
H
s(i,j) 2ds < ∞ and
Q (
ω 2 Ω : 9 t 2 [ 0, T ] such that H
t( ω ) H
t(i,j)( ω ) is singular
)
= 0.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
Martingale representation theorem III
Theorem
The martingale representation theorem
For all i 2 f 1, ..., n g , Z
(i)is a ( Q , fF
tg ) martingale.
Moreover, for any other ( Q , fF
tg ) martingale M, there exist n predictable processes
φ
(i)= n φ
(ti): 0 t T o such that
M
t= M
0+
∑
n i=1Z
t0
φ
(si)dZ
s(i).
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
Martingale representation theorem IV
Such a theorem will be useful when we must prove that a contingent claim is attainable in a market model
comprising multiple risky assets.
Baxter and Rennie, page 162
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general I
X
(1), ...., X
(M)represent the M risk factors in our model.
Some of these sources of risk may be tradable assets (stocks, bonds, etc.), some others are not (volatility, foreign exchange rate, interest rate, etc.). Assume their dynamics are described as follows:
dX
t(m)= µ
t(m)X
t(m)dt +
∑
K k=1σ
(tm,k)X
t(m)dW
t(k)where the K risk-neutral Brownian motions are independent,
µ
(m)and σ
(m,k)are predictable processes meeting certain
regularity conditions that ensure the existence of a solution.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general II
There are N tradable assets, S
t(n)= f
n( X
t) , which are functions of risk factors. If the functions f
nare su¢ ciently smooth, Itô’s lemma implies that
dS
t(n)=
∑
M m=1∂fn
∂x(m)
(
Xt) dX
t(m)+
1 2∑
M m=1∑
M j=1∂2
f
n∂x(m)∂x(j)
(
Xt) d
DX
(m),X
(j)Et
=
∑
M m=1∂fn
∂x(m)
(
Xt)
µ(m)tX
t(m)dt +
∑
M m=1∂fn
∂x(m)
(
Xt)
∑
K k=1σ(m,k)t
X
t(m)dW
t(k)+
1 2∑
M m=1∑
M j=1∂2
f
n∂x(m)∂x(j)
(
Xt)
∑
K k=1σ(m,kt )σ(j,kt )
X
t(m)X
t(j)dt.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general III
Since S
t(n)is a tradable asset, its risk-neutral dynamics require the drift to satisfy
r
tS
t(n)=
∑
M m=1∂fn
∂x(m)
(
Xt)
µ(m)tX
t(m)+
12
∑
M m=1∑
M j=1∂2
f
n∂x(m)∂x(j)
(
Xt)
∑
K k=1σ(m,kt )σ(j,kt )
X
t(m)X
t(j).As a consequence, dB
t 1S
t(n)= B
t 1∑
K k=1∑
M m=1∂f
n∂x
(m)( X
t) σ
(tm,k)X
t(m)!
dW
t(k). where B
t= exp R
t0
r
sds . If
EQ 2 4ZT
0
∑
M m=1∂fn
∂x(m)
(
Xs)
σs(m,k)X
s(m)!2
ds
3 5<
∞,n
B
t 1S
t(n)o
0 t T
is a Q martingale.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general IV
The goal is to hedge a derivative product, the value of which at time t is g ( X
t) . If g is su¢ ciently smooth, Itô’s lemma implies that
dB
t1g (
Xt) = B
t 1∑
K k=1∑
M m=1∂g
∂x(m)
(
Xt)
σ(m,k)tX
t(m)!
dW
t(k).th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general V
If the number of tradable assets is equal to the number of Brownian motions, N = K , and if
Q 8<
:9t2[0,T] such that Ht
∑
M m=1∂fn
∂x(m)(Xt)σ(tm,k)Xt(m)
!
n,k
is singular 9=
;=0,
then the martingale representation theorem implies the existence of predictable processes ϕ
(n)such that
dB
t 1g (
Xt)
=
∑
K n=1ϕ(nt )
dB
t 1S
t(n)= B
t 1∑
K n=1ϕ(n)t
∑
K k=1∑
M m=1∂fn
∂x(m)
(
Xt)
σ(m,k)tX
t(m)!
dW
t(k)= B
t 1∑
K k=1∑
M m=1∑
K n=1ϕ(n)t ∂fn
∂x(m)
(
Xt)
!
σ(m,kt )
X
t(m)!
dW
t(k).th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional
References Appendix
In general VI
We obtain two equations for dB
t 1g ( X
t) . By setting equal the processes in front of each Brownian, we …nd
∑
M m=1∂g
∂x(m)
(
Xt)
σ(m,kt )X
t(m)=
∑
K n=1ϕ(n)t
∑
M m=1∂fn
∂x(m)
(
Xt)
σt(m,k)X
t(m)! ,
which is equivalent to
∑
M m=1"
∂g
∂x
(m)( X
t)
∑
K n=1ϕ
(tn)∂f
n∂x
(m)( X
t)
#
σ
(tm,k)X
t(m)= 0. (2) When x
(m)= f
m, the problem reduces to
∑
M m=1∂g
∂x
(m)( X
t) ϕ
t(m)( X
t) σ
(tm,k)X
t(m)= 0 and we conclude that ϕ
(tn)=
∂f∂gn
( X
t) .
More generally, the system of linear equations (2) needs to be
solved.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
References
Martin Baxter and Andrew Rennie (1996). Financial Calculus, an introduction to derivative pricing, Cambridge university press.
Damien Lamberton and Bernard Lapeyre (1991).
Introduction au calcul stochastique appliqué à la …nance,
Ellipses.
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Annexe I
In this appendix, we show the following:
ψ
t= ∂f
∂s ( t, S
t)
= N 0
@ ln S
tln K + r +
σ22( T t ) σ ( T t )
1/21 A
and
φ
t= M
tψ
tB
t 1S
t= Ke
rTN d ( S
t) σ p
T t
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Annexe II
Recall that
f ( t, S
t) = S
tN ( d ( S
t)) Ke
r(T t)N d ( S
t) σ p T t where
d ( s ) = ln ( s ) ln ( F
i) + r +
12σ
2( T t ) σ p
T t .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Annexe III
As a consequence
∂f
∂s
( t, s )
= N ( d ( s )) + s
∂∂s
N ( d ( s )) Ke
r(T t) ∂∂s
N d ( s )
σp T t
= N ( d ( s )) + sn ( d ( s ))
∂∂s
d ( s ) Ke
r(T t)n d ( s )
σp
T t
∂∂s
d ( s )
= N ( d ( s )) +
hsn ( d ( s )) Ke
r(T t)n d ( s )
σp
T t
i ∂∂s
d ( s )
.where n ( ) is the probability density function of a standard
normal random variable. We will show that the term inside [ ]
is nil
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Annexe IV
Indeed,
n d ( s )
σp T t
= p
12πexp 1
2
d ( s )
σp T t
2= p
12πexp 1
2
( d ( s ))
2+
σp
T td ( s )
12σ2
( T t )
= p
1 2πexp1
2
( d ( s ))
2+
ln( s )
ln( K ) + r +
12σ2( T t )
1
2σ2
( T t )
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix
Annexe V
which implies that
Ke
r(T t)n d ( s ) σ p T t
= p 1
2π exp 1
2 ( d ( s ))
2+ ln ( s )
= p s
2π exp 1
2 ( d ( s ))
2= sn ( d ( s )) .
th.
Self-…nancing strategies Martingale representation th.
Steps in a contingent claim replication Example Multidimensional References Appendix