Академический Документы
Профессиональный Документы
Культура Документы
t
N(0,1)
_
E[Z] = 0
var(Z) = t s.d. =
t
Z(T) Z(0) =
n
i=1
t =
n
i=1
Z
i
, where n =
T
t
_
E[Z(T) Z(0)] = 0
var(Z(T) Z(0)) = n t = T s.d. =
i=1
[g(t
i
) g(t
i1
)[, a = t
1
< t
2
< < t
n
= b,
where P is the set of all kinds of partitions (even n is allowed)
quadratic variation on [0, t] = t
[Z, Z](t) = [Z, Z]([0, t]) = lim
n
n
i=1
[Z(t
i
) Z(t
i1
)[
2
cov(Z(t), Z(s)) = E[Z(t)Z(s)] E[Z(t)]E[Z(s)] = E[Z(t)Z(s)]
(if s < t, Z(t) = Z(s) + Z(t) Z(s))
= E[Z
2
(s)] + E[Z(s)(Z(t) Z(s))] = E[Z
2
(s)] = var(Z(s)) = s = min(t, s)
(the covariance is the length of the overlapping time period (or the sharing path)
between Z(t) and Z(s))
Generalized Wiener process
dX = adt + bdZ (this form is also called stochastic dierential equation (SDE))
_
E[dX] = adt
var(dX) = b
2
dt s.d. = b
dt
dX N(adt, b
2
dt)
1-3
It o process (also called diusion process) (Ito, a Japanese mathematician, passed away in
2008 at the age of 93)
dX=a(X, t)dt+b(X, t)dZ
no more constants, so it is no more simple to derive E[dX] and var(dX)
For the stock price, it is commonly assumed to follow the below It o process
dS = Sdt + SdZ
dS
S
= dt + dZ (geometric Brownian motion, GBM)
dS
S
N(dt,
2
dt)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
d ln S
dS
=
1
S
d ln S =
dS
S
(Note that this dierential result is true only when S is
a real variable. This kind of dierentiation CANNOT be applied to stochastic
processes. The stochastic calculus is not exactly the same as the calculus for
real variables.)
In fact, the stock price follows the lognormal distribution based on the assumption
of the geometric Brownian motion, but it does not mean d ln S N(dt,
2
dt).
II. Itos Lemma
It os Lemma is in essence the Taylor series.
Taylor series: f(x, y) = f(x
0
, y
0
) +
f
x
(x x
0
) +
f
y
(y y
0
)
+
1
2!
(
2
f
x
2
(x x
0
)
2
+ 2
2
f
xy
(x x
0
)(y y
0
) +
2
f
y
2
(y y
0
)
2
) +
It os Lemma to derive the stochastic dierential equation:
Given dX = a(X, t)dt + b(X, t)dZ, and f(X, t) as a function of X and t, the following
stochastic dierential equation can be derived as follows.
df = (
f
t
+
f
X
a +
1
2
2
f
X
2
b
2
)dt + (
f
X
b)dZ,
where a and b are the abbreviations of a(X, t) and b(X, t).
1-4
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
The Itos Lemma holds under the following approximations:
(i)
dt
1
dt
dt
1.5
0
dt
2
0
.
.
.
(ii)
dZ dZ =?
According to the denition of dZ, dZ dZ =
2
dt.
N(0, 1)
var() = 1 E[
2
] (E[])
2
= 1 E[
2
] = 1 E[dZ
2
] = dt
In addition, var(dZ
2
) = var(
2
dt) = (dt)
2
var(
2
) 0 (because (dt)
2
0)
dZ dZ
a.s.
= dt
_
_
_
_
_
_
_
_
_
_
_
_
_
_
It os Lemma vs. the dierentiation of a real-variable function
For a real-variable f(t), if
df
dt
= g(t), we can interpret that with a innitesimal change
of dt, the change in f is g(t)dt, which is deterministic.
The interpretation of the It os Lemma: with a innitesimal change of dt, the change
in f is (
f
t
+
f
X
a +
1
2
2
f
X
2
b
2
)dt + (
f
X
b)dZ. Note that the rst term plays a similar role
as g(t)dt, but the second term tells us that the change in f is random.
To apply the Itos Lemma is like to take the dierentiation for stochastic processes.
Based on the result of dZ dZ = dZ
2
= dt, it is straightforward to infer that the quadratic
variation of the Wiener process over [0, t], i.e., [Z, Z](t) = [Z, Z]([0, t]) = lim
n
n
i=1
[Z(t
i
)
Z(t
i1
)[
2
, equals t.
Example 1 of applying the It os Lemma: f = ln S, dS = Sdt + SdZ
d ln S = (0 +
1
S
S
1
2
1
S
2
2
S
2
)dt +
1
S
SdZ
= (
2
2
)dt + dZ
Dene ln S
T
= ln S
T
ln S
t
and Z
Tt
= Z(T) Z(t)
ln S
T
= (
2
2
)(T t) + Z
Tt
ln S
T
ln S
t
N((
2
2
)(T t),
2
(T t))
ln S
T
N(ln S
t
+ (
2
2
)(T t),
2
(T t))
The stock price is lognormal distributed.
1-5
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
Another derivation: apply the stochastic integral on the both side of the equation
_
T
t
d ln S
=
_
T
t
(
2
2
)d +
_
T
t
dZ()
[
T
t
= (
2
2
)(T t) + (Z()[
T
t
)
Z(T) Z(t) Z
Tt
N(0, T t)
ln S
T
ln S
t
N((
2
2
)(T t),
2
(T t))
Example 2: f = S Ke
r(Tt)
(f is the value of a forward agreement)
df = (S rKe
r(Tt)
)dt + SdZ
Example 3: F = Se
r(Tt)
(F is the forward price of a stock)
dF = ( r)Fdt + FdZ
It os Lemma for multiple variates
dS
S
=
S
dt +
S
dZ
S
(foreign stock price)
dX
X
=
X
dt +
X
dZ
X
(exchange rate: 1 foreign dollar = X domestic dollars)
Dene f = S X (the value of a foreign stock share in units of domestic dollars)
df=[
f
t
+
f
S
S
S +
f
X
X
X +
1
2
2
f
S
2
2
S
S
2
+
1
2
2
f
X
2
2
X
X
2
+
2
f
SX
XS
S
X
S X]dt +
f
S
S
SdZ
S
+
f
X
X
XdZ
X
df = [
S
XS +
X
XS +
XS
X
SX]dt +
S
XSdZ
S
+
X
XSdZ
X
df
f
= (
S
+
X
+
XS
X
)dt +
S
dZ
S
+
X
dZ
X
(because f = SX)
_
_
_
_
_
_
_
_
dZ
S
dZ
X
=
S
dt
X
dt =
S
X
dt
E[dZ
S
dZ
X
] = E[
S
X
]dt =
XS
dt
var(dZ
S
dZ
X
) = (dt)
2
var(
S
X
) 0
dZ
S
dZ
X
a.s.
=
XS
dt
1-6
III. Stochastic Integral
Stochastic integral (or called Ito intergral or It o calculus): allows one to integrate one
stochastic process (the integrand) over another stochastic process (the integrator). Usu-
ally, the integrator is a Wiener process.
Integral over a stochastic process:
_
b
a
X()dZ(), where X() can be a deterministic
function or a stochastic process, and dZ() is a Wiener process. (vs. integral over a
variable:
_
b
a
f(y)dy, where f(y) is a deterministic function of the variable y.)
Stochastic integral for simple deterministic processes
If X() is a deterministic process, given any value of t, the value of X() can be known
exactly. Therefore, in an innitesimal time interval, (t
i1
, t
i
], the value of X() can be
approximated by a constant C
i
. The term simple means to approximate the process
by a step function. (In contrast, if X() is a stochastic process, given any value of , we
only konw the distribution of possible values for X().)
Figure 1-2
( ) X
0
t
1
t
2
t
3
t
4
t
5
t
1
C
2
C
3
C
4
C
5
C
T 0
For simple deterministic processes, we can dene the stochastic integral as follows. (This
denition is similar to the rectangle method to dene the integral over a variable.)
_
T
0
X()dZ() =
n
i=1
C
i
(Z(t
i
) Z(t
i1
)) N(0,
n
i=1
C
2
i
(t
i
t
i1
))
(In fact, there should be a term lim
n
in front of each
n
i=1
. It is omitted for simplicility.)
(Note that the result of a stochastic intergral is a distribution, and we are interested in
the mean and variance of this distribution.)
1-7
In the above equation, the reason for the nal normal distribution:
1. The sum of normally distributed random variables is still a normally distributed random
variable.
2. The mean for the resulting random variable is the sum of the mean of all normally
distributed random variables.
3. The variance for the resulting random variable is the sum of the variances of all nor-
mally distributed random variables because all normally distributed random variables are
independent.
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
(i) According to the above denition, if X(t) = 1, the result of the stochastic integral is
consistent with the denition of the Wiener process.
_
T
0
X()dZ() =
_
T
0
dZ() = Z()[
T
0
= Z(T) Z(0) N(0, T)
=
n
i=1
(Z(t
i
) Z(t
i1
)) N(0,
n
i=1
(t
i
t
i1
)) = N(0, T)
(ii) Alternative way to calculate the variance of the result of the stochastic integral.
var(
_
XdZ) = E[(
_
XdZ)
2
] (E[
_
XdZ])
2
= E[(
_
XdZ)
2
] = E[(
n
i=1
C
i
(Z(t
i
) Z(t
i1
)))
2
]
=
n
i=1
n
j=1
C
i
C
j
E[(Z(t
i
) Z(t
i1
))(Z(t
j
) Z(t
j1
))]
calculate the squared term in the expectation, and then apply the distributive property of
the expectation over the addition and scaler multiplication
=
n
i=1
C
2
i
(t
i
t
i1
)
because cov(Z(t
i
) Z(t
i1
), Z(t
j
) Z(t
j1
)) = 0, and var(Z(t
i
) Z(t
i1
)) = t
i
t
i1
Consider a simple predictable process: in the time interval (t
i1
, t
i
], the constant C
i
is
replaced by a random variable
i
, which depends on the values of Z(t) for t t
i1
, but
not on values of Z(t) for t > t
i1
. Therefore, X(t) is dened as follows.
X(t) = I
{t|t=0}
+
n
i=1
i
I
{t|t
i1
<tt
i
}
,
where I is a indicator function and is a constant. The corresponding stochastic intergral
is dened as follows.
_
T
0
X()dZ()
n
i=1
i
(Z(t
i
) Z(t
i1
)).
1-8
The reason for the name predictable:
1. The value of X(t) for (t
i1
, t
i
],
i
, is determined based on the information set formed
by Z(t) until t
i1
, denoted by T
t
i1
. It is also called that
i
is T
t
i1
-measurable. (See
Figure 1-3)
2. In contrast, the value of Z(t
i
) Z(t
i1
) will not realize until the time point t
i
, i.e.,
this value will be known based on the information set T
t
i
. In other words, Z(t
i
) is
T
t
i
-measurable. (See Figure 1-3)
3. Therefore, we say that X(t) is predictable since we know its realized value just
before the time point at which Z(t) is realized.
4. In the continuous-time model, Z(t) is T
t
-measurable (the realized value is known at
t). For any process that we can know its realized value just before t, we call this process
to be T
t
-measurable and thus predictable.
Figure 1-3
( ) X t
1 i
t
i
t
i
1 i
t
i
t
( ) Z t
1
( ) ( )
i i
Z t Z t
i=1
Z(t
i1
)I
{t|t
i1
<tt
i
}
( lim
n
X
n
(t) = Z(t) almost surely)
_
T
0
X
n
()dZ() =
n
i=1
Z(t
i1
)(Z(t
i
) Z(t
i1
))
=
1
2
n
i=1
[(Z(t
i
))
2
(Z(t
i1
))
2
(Z(t
i
) Z(t
i1
))
2
]
=
1
2
(Z(T))
2
1
2
(Z(0))
2
1
2
n
i=1
(Z(t
i
) Z(t
i1
))
2
_
_
_
_
_
_
_
_
_
_
Similar to the derivation of the Itos Lemma that E[dZ
2
] = dt and
var(dZ
2
) 0 when n (dt 0), (Z(t
i
) Z(t
i1
))
2
converges to
t
i
t
i1
in probability if [t
i
t
i1
[ is very small. So, we can
conclude that when n (dt 0), lim
n
n
i=1
(Z(t
i
) Z(t
i1
))
2
= T.
_
T
0
Z()dZ() = lim
n
_
T
0
X
n
()dZ() =
1
2
(Z(T))
2
1
2
T
1-10
Properties of It o Integral:
(i)
_
T
0
(X()+Y ())dZ() =
_
T
0
X()dZ()+
_
T
0
Y ()dZ() (distributive property)
(ii)
_
T
0
I
[a,b]
()dZ() = Z(b) Z(a), 0 < a < b < T
(iii) E[
_
T
0
X()dZ()] = 0
(iv) var(
_
T
0
X()dZ()) = E[(
_
T
0
X()dZ())
2
] =
_
T
0
E[X()
2
]d (It o Isometry)
Find E[
_
T
0
Z()dZ()] and var(
_
T
0
Z()dZ()).
(i) E[(Z(T))
2
] = var(Z(T)) + E[Z(T)]
2
= T
E[
_
T
0
Z()dZ()] = E[
1
2
(Z(T))
2
1
2
T] = 0
(Property (iii) can be applied to obtaining the identical result directly.)
(ii) var(
_
T
0
Z()dZ()) =
1
4
var((Z(T))
2
)
=
1
4
E[(Z(T))
4
] E[(Z(T))
2
]
2
=
1
4
3T
2
T
2
=
T
2
2
_
_
_
_
If x N(,
2
), then E[x
4
] =
4
+ 6
2
2
+ 3
4
.
Since Z(T) N(0, T), we can derive E[(Z(T))
4
] = 3T
2
.
Apply Property (iv) to nding var(
_
T
0
Z()dZ()) as follows:
var(
_
T
0
Z()dZ()) =
_
T
0
E[(Z())
2
]d =
_
T
0
d =
1
2
2
[
T
0
=
T
2
2
1-11
IV. Solve Stochastic Dierential Equations with Stochastic Integral
How to solve X(t) systematically through the stochastic integral is the major application
of the stochastic integral.
Given dX(t) = X(t)dt + dZ(t)
. .
, solve X(t).
Ornstein-Uhlenbeck process
_
X(t) (X, t)
(X, t)
According to the stochastic integral, X(t) should satisfy
X(t) = X(0)+
_
t
0
(X, )d +
_
t
0
(X, )dZ()
_
_
_
_
_
_
However, (X, t) is a function of X(t), so (X, t) is a stochastic process as well.
Moreover, since the value of (X, t) is unknown due to the unsolved X(t). Thus,
we cannot derive X(t) by applying the stochastic integral directly.
Dene Y (t) = X(t)e
t
dY (t) = e
t
dX(t) + e
t
X(t)dt (through the It os Lemma)
= e
t
[X(t)dt + dZ(t)] + e
t
X(t)dt
= e
t
dZ(t)
Y (t) = Y (0) +
_
t
0
e
dZ()
dU(t)
U(t)
= (t)dt + (t)dZ(t)
(The U(t) is similar to S(t), so we can apply the results on p.1-6 to solve U(t).)
U(t) = U(0) exp(
_
t
0
(()
1
2
2
())d +
_
t
0
()dZ())
. .
(1)
(ii) Consider X(t) = U(t) V (t), and U(0) = 1 and V (0) = X(0),
where dU(t) = (t)U(t)dt + (t)U(t)dZ(t)
dV (t) = a(t)dt + b(t)dZ(t)
The integration by parts for stochastic processes:
U(t)V (t) U(0)V (0) =
_
t
0
V ()dU() +
_
t
0
U()dV () + [U, V ](t),
where [U, V ](t)= lim
n
n
i=1
(U(t
i
)U(t
i1
))(V (t
i
)V (t
i1
)) (quadratic covariation).
d[U, V ](t) = dU(t) dV (t) =
U
V
dt (there is no product of drift terms because
they are all with (dt)
2
or (dt)
1.5
, which is too small relative to dt)
stochastic product rule:
dX(t) = dU(t) V (t) + U(t) dV (t) + d[U, V ](t),
where d[U, V ](t) = dU(t) dV (t) = (t)U(t)b(t)dt
Substitute dU(t) and dV (t) into the above equation, and compare with dX(t).
b(t) U(t) = (t), a(t) U(t) = (t) (t) (t)
b(t) =
(t)
U(t)
, a(t) =
(t)(t)(t)
U(t)
V (t) = V (0) +
_
t
0
() ()()
U()
d +
_
t
0
()
U()
dZ()
. .
,
(2) where V (0) = X(0)
X(t) = U(t) V (t) = (1) (2)
1-13
Brownian bridge (pinned Brownian motion):
dX(t) =
bX(t)
Tt
dt + dZ(t), 0 t T, X(0) = a
(t) =
b
Tt
, (t) =
1
Tt
, (t) = 1, (t) = 0
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
U(t) = U(0)exp(
_
t
0
(()
1
2
2
())d +
_
t
0
()dZ())
= exp(
_
t
0
1
T
d) = exp(ln(T )[
t
0
)
= exp(ln
Tt
T
) =
Tt
T
b(t) =
T
Tt
a(t) =
b
Tt
01
Tt
T
=
bT
(Tt)
2
V (t) = V (0)
..
+
_
t
0
bT
(T )
2
d
. .
+
_
t
0
T
T
dZ()
X(0) = a
bT
Tt
b
X(t) = U(t) V (t) =
Tt
T
[a +
bT
Tt
b + T
_
t
0
1
T
dZ()]
X(t) = a(1
t
T
) + b
t
T
+ (T t)
_
t
0
1
T
dZ(), 0 t < T, lim
tT
X(t) = b
Figure 1-5
( ) X t
t
T
b
a
lim ( )
t T
X t b
0
_
_
_
_
_
_
E[X(t)] = a(1
t
T
) + b
t
T
var(X(t)) = t
t
2
T
=
Ttt
2
T
=
t(Tt)
T
cov(X(t), X(s)) = min(s, t) st/T
The Brownian bridge is suited to formulate the process of the zero-coupon bond price
because the bond price today is known and the bond value is equal to its face value on
the maturity date. The disadvantage of fomulating the bond price to follow the Brownian
bridge is that the zero-coupon bond price could be negative due to the normal distribution
of dZ(t) in dX(t).
1-14
Given X(t) = a(1
t
T
) + b
t
T
+ (T t)
_
t
0
1
T
dZ(),
prove (i) var(X(t)) =
t(Tt)
T
.
(ii) cov(X(t), X(s)) = s
st
T
(if t > s).
(i) According to the fourth property of It o integral, that is,
var(
_
T
0
X()dZ()) =
_
T
0
E[X()
2
]d, we can derive
var(X(t)) = (T t)
2
_
t
0
(
1
T
)
2
d = (T t)
2
((T )
1
[
t
0
)
= (T t)
2
(
1
Tt
1
T
) =
t(Tt)
T
(ii) cov(X(t), X(s))
= cov(X(s) + X(t) X(s), X(s)) (assume s < t)
= var(X(s)) + cov(X(t) X(s), X(s))
=
s(Ts)
T
+ cov((T t)
_
t
0
1
T
dZ() (T s)
_
s
0
1
T
dZ(), (T s)
_
s
0
1
T
dZ())
_
_
_
_
_
_
_
(T t)
_
t
0
1
T
dZ() (T s)
_
s
0
1
T
dZ()
Ts=Tt+ts
= (T t)
_
t
s
1
T
dZ() (t s)
_
s
0
1
T
dZ()
=
s(Ts)
T
(t s)(T s) var(
_
s
0
1
T
dZ())
=
s(Ts)
T
(t s)(T s)(
_
s
0
(
1
T
)
2
d)
=
s(Ts)
T
(t s)(T s)(
1
(Ts)
1
T
)
=
sTs
2
T
(t s)(T s)(
s
T(Ts)
)
=
sTs
2
st+s
2
T
= s
st
T
Introduction to Stochastic Calculus with Applications, Klebaner, 2005
1-15