Sei sulla pagina 1di 34

Chapter 6

Ito’s Stochastic Calculus

6.1 Introduction

When Bachelier ﬁrst applied Wiener process on modeling the ﬂuctuation of asset prices, the price of an asset at time t, X t , has an inﬁnitesimal increment dX t propor- tional to the increment dW t of the Wiener process, i.e.,

dX t = σ dW t ,

where σ is a positive constant. As a result, an asset with initial price X(0) = x worths

X t = x+σW t

at time t. This model suffers from one serious ﬂaw: for any t > 0 the price X t can be

negative with non-zero probability (but actual stock prices are never negative). To tackle this problem, successors assumed that the relative price dX t /X t of an asset is

proportional to dW t , i.e.,

dX t = σX t dW t .

(6.1)

Although this equation looks like a differential equation, traditional methods are no longer applicable because the paths of W t are not differentiable (Theorem 5.2).

A way around the obstacle was found in the 1940s by Ito, who gave a rigorous

meaning to (6.1) by writing it as

t

X t = x+σ 0 X s dW s ,

(6.2)

where the integral with respect to W t on the right-hand side is called the Ito stochas- tic integral. In this section we discuss the deﬁnition and properties of stochastic integral, and explore the applications to pricing ﬁnancial products.

107

108

6.2 Ito Stochastic Integral

6 Ito’s Stochastic Calculus

6.2.1 Motivation

t

Motivated by (6.2), we construct the Ito stochastic integral in the form f (s) dW s for some stochastic process/random function f (s) ( f (s, ω ), to be precise). We follow an approach similar to constructing Riemann integral, i.e., deﬁne the integral by the limit of the discretized version

0

n1

i=0 f(s i )(W t i+1 W t i ) ,

(6.3)

where s i [t i ,t i+1 ]. The major differences between Riemann and Ito integrals are

1. Riemann integration results in a real number, but Ito integration results in a random variable (since W t is random). Thus, while deﬁning Riemann integral involves convergence of real numbers, deﬁning Ito integral in (6.3) requires convergence of random variables, which is considerably more difﬁcult.

2. If a Riemann integral exists, then s i can be an arbitrary point in [t i ,t i+1 ] since the upper and lower Riemann sum converge. However, in Ito integral, the limit will be different depending on the choice of s i . This is due to the non-zero quadratic variation of the Brownian motion W t . See Exercise 6.1.

To avoid the ambiguity in 2), the deﬁnition of stochastic integrals will ﬁx the choice s i = t i for each i in the approximating sum (6.3). The choice s i = t i is natural if we regard f (t) as the trading strategy and W t as the stock price: For the i + 1-th period [t i ,t i+1 ], the trading strategy should only depend on the information up to time t i . Hence, f (t i ) units are invested and a proﬁt of f (t i )(W t i+1 W t i ) is made.

Therefore, n1 f(t i )(W t i+1 W t i ) represents the total proﬁt.

Remark 6.1. (Previsible) To be precise, in deﬁning stochastic integral we require the integrand f (t) to be previsible or predictable, i.e., f (t) is F t for all t where F t ≡ ∪ s<t F s and F t = σ({W s , s t}). However, it can be shown that if f (t) is continuous and adapted to F t , then f (t) is automatically previsible. Since we mainly deal with continuous integrand, we do not distinguish between previsible and adapted process.

i=0

6.2.2 Ito Integral

Ito integral is a random variable since W t and the integrand f (t ) are random. To ensure regularity of the Ito integral (such as the existence of the ﬁrst and the second moments), we restrict f (t) to the following class of stochastic processes.

to be the class of

Deﬁnition 6.1. (M

2

T

and M 2 Stochastic Processes) Denote M

2

T

stochastic processes f (t), t 0, such that

6.2

Ito Stochastic Integral

109

E

0

T |f(t)| 2 dt < .

Let M 2 be the class of stochastic processes f (t) such that f (t) M

T

2 for any T > 0.

Recall that a random variable X is in L 2 , or X L 2 , if E|X| 2 < . Both M 2 and L 2 are related to the existence of second moment, but M 2 is for a stochastic process and L 2 is for a random variable. Since the Ito integral is a random variable and the integrand is a random function (stochastic process), we need to deﬁne a measure of length and a mode of conver- gence in deﬁning the integral by a limit of a discretization in (6.3).

2 Norms) For a random variable X and a stochastic

Deﬁnition 6.2. (L 2 and M

T

process f f (t), the L 2 and M

T

2 norm are given respectively by

X L 2 = E(X 2 )

and

f M

2

T

= E

0

T

 |f(t)| 2 dt . (6.4)

Deﬁnition 6.3. (L 2 , M

2 and M 2 Convergences) A sequence of random vari-

T

ables {X n } converges in L 2 to X if

X n X L 2 = E(|X n X| 2 ) 0.

as n . A sequence of random functions / stochastic processes { f n (t)} con-

verges in M

T

to f if for any T ,

2

f n f M

2

T

= E

0

T |f n (t)f(t)| 2 dt 0.

Similarly, { f n (t)} converges in M 2 to f if { f n (t)} converges to f in M

Using Deﬁnitions 6.1 and 6.3, we deﬁne the Ito Integral on the class of stochastic process M 2 as follows:

2 for all T .

T

Deﬁnition 6.4. (Ito Integral) For any T > 0 and any stochastic process f M 2 , the stochastic integral of f on [0, T ] is deﬁned by

n1

I T ( f ) = lim j=0 f(t j ) W(t j+1 )W(t j ) ,

n

where (0 = t 0 < t 1 < ··· < t n1 < t n = T ) is any partition of [0, T ] with max j |t j t j1 | → 0 as n . We write I T ( f ) as

I T (f) =

0

T

f

(t) dW t ,

or

0

T

f

dW .

110

6 Ito’s Stochastic Calculus

The following theorem justiﬁes the above deﬁnition of Ito integral.

Theorem 6.1. (Existence and Uniqueness of Ito Integral) Suppose that a function f M 2 satisﬁes the following assumptions: For all t 0,

A1)f (t) is almost surely continuous, i.e., P(lim ε0 |f(t +ε)f(t)| = 0) = 1, A2)f (t) is adapted to the ﬁltration {F t }, where F t = σ({W s ,s t}).

Then, for any T > 0, the Ito integral

I T (f) = 0

T

f (t) dW t

exists and is unique almost everywhere.

Proof. The proof proceeds in three steps:

1) Construct a sequence of adapted stochastic processes f n such that f f n M 2

0.

Show that I T (f n )I T (f) L 2 0.

2)

3) Show the a.s. uniqueness of the limit I T (f).

To begin,

1) First we ﬁnd a sequence of M

T

2 functions f 1 , f 2 ,

E(

T |f n (t)f(t)| 2 dt) 0. Deﬁne

0

such that f f n M

2

T

=

f n (t) = n

0

k

n

k1

n

f (s) ds if t [

k , k+1

n

n

otherwise.

) for k = 1, 2,

,

[T n] 1,

(6.5)

Then by construction f n (t) is adapted to F t and is a step function for each ω. (Function of this kind is called random step function. Since it is random, to be precise, it should be denoted by f n (t) = f n (t, ω)). Note that

k

n

k+1

n

|f n (t)| 2 dt = n

k

n

k1

n

f

(t)dt

2

k

n

k1

n

|f(t)| 2 dt , a.s.

(6.6)

by Cauchy Schwartz (CS) inequality. The above inequality holds almost surely as CS is applied for each ω in f (t, ω).

Next we show f f n M

T |f n (t)f(t)| 2 dt) 0. By the a.s. continuity

of f (t), it can be shown that

2 = E(

T

0

n

0

lim

T

|f n (t)f(t)| 2 dt = 0

a.s.

(6.7)

Let Y n = also that

T |f n (t) f(t)| 2 dt. Note that (6.7) means that lim n Y n a.s. = 0. Note

0

6.2

Ito Stochastic Integral

111

Y n 2

0

T

T

|f(t)| 2 +|f n (t)| 2 dt 4 0 |f(t)| 2 dt Y ,

¯

a.s.

where the second inequality follows from summing over all k in (6.6). Since Y n

0, |Y n | ≤ Y a.s. and E Y < by the deﬁnition of M Convergence Theorem (DCT) to obtain E(Y n ) 0, i.e.,

T

¯

¯

2

, we can apply Dominant

f f n M

2

T

= E

0

T

|f n (t)f(t)| 2 dt 0 .

(6.8)

2) To show the existence of Ito Integral, we need to show that I T (f n ) =

0

T f n (t) dW t

converges to an element in L 2 . Since f n (t) is a step function taking constant

values in each interval [

k , k+1

n

n

), we can write

I T (f n ) =

k1 f n

k

n

W k+1 W k .

n

n

Note that

=

=

I T (f n )

2

L 2

E

k1 f n

k

n

W k+1 W k

n

n

2

E

k1 j1 f n

∑ ∑

k

n

f n

j

n

W k+1 W k W j+1 W

n

n

n

n

j

k1 E f n

E 0

T

k

n

k

n

2 E W k+1 W k 2

n

n

2

1

n

(Independent increment and f n

2

) = t)

=

= k1 E f n

=

(Stationary increment and E(W

t

f n (t) 2 dt

(Riemann integral)

=

f n

2

M

2

T

.

k

n

F k )

n

(6.9)

Similarly, it can be shown that

2

I T (f n )I T (f m ) L 2 = f n f m

2

M

2

T

From (6.8), for any ε > 0, there is an N such that f f n Thus for n, m > N,

.

2

M

2

T

(6.10)

< ε 2 for all n > N.

112

6 Ito’s Stochastic Calculus

2

I T (f n )I T (f m ) L 2 = f n f m

2

M

2

T

f n f

2

M

2

T

< ε 2 + ε 2 = ε .

+ f m f

2

M

2

T

(Triangular Inequality)

The sequence {I T (f n )} n1 with the property I T (f n )I T (f m ) L 2 0 for n, m is called a (Cauchy sequence). It is well known in mathematical analysis that any Cauchy sequence in L 2 has a limit. Call the limit I T (f). 3) Finally, we show the uniqueness of the limit. Suppose that there are two se-

|f(t)

(j) ), j = 1, 2, converge to

2

quences of stochastic processes { f

i

(j)

f

n

| 2 dt) n

(1)

(2)

} i1 and { f

i

T

} i1 satisfying E(

0

−→ 0 for j = 1, 2. We need to show that I T (f

n

the same limit. Introduce a new sequence

(1)

{g i } i1 = {f

1

(2)

, f

1

(1)

, f

2

(2)

, f

2

(1)

, f

3

,

}.

−→ 0. Therefore, the ar-

gument in 2) shows that I T (g n ) converges to some limit. Note that if a sequence

converges, then every subsequence converges to the same limit (Exercise 6.2).

Thus I T (f

of uniqueness. Note that the above uniqueness is proved in L 2 sense, i.e., I T (f (1) )I T (f (2) ) 0, since

By construction, {g i } i0 satisﬁes E(

T |f(t)g n | 2 dt)

0

n

(1)

n

(2)

) and I T (f

n

) do converge to the same limit, completing the proof

2 L 2 =

I T (f (1) )I T (f (2) ) L 2

2

I T (f (1) )I T (g n ) L 2 + I T (g n )I T (f (2) )

2

2

L 2

for all n and the quantity on the right is arbitrarily small. However, I T (f (1) )I T (f (2) ) L 2 = 0 implies that I T (f (1) ) = I T (f (2) ) almost surely (See Exercise 6.13). Thus the deﬁnition of Ito’s integral is unique a.s

2

Example 6.1. To show the existence of

0

T W t dW t , from Theorem 6.1 we need to

show that the Wiener process W t belongs to M 2 . Since for all T

E

0

T

|W t | 2 dt =

0

T

E(|W t | 2 )dt =

0

T

t dt < .

Thus W t belongs to M 2 . Also, we have seen from Chapter 5 that W t satisﬁes As- sumptions A1 and A2 of Theorem 6.1. Hence the existence of the Ito integral

T

0

W t dW t is justiﬁed.

Example 6.2. We derive the formula

1 2 T directly from deﬁni-

tion, i.e., by approximating the integrand by random step functions. Fix T > 0 and

t

0

W t dW t = 1 2 W

2

T

T

n = iT

i

n

, set

6.3

Properties of the Stochastic Integral

113

f n (t) =

n1

i=0

W t

n

i

1

[t

n

i

,t

i+1 ) (t).

n

Then the sequence f 1 , f 2 ,··· ∈ M

T

2 approximates f , since

E

0

|f(t)f n (t)| 2 dt =

=

=

=

n1

i=0

n1

i=0

1

n

n

t

E |W t W t

n

i+1

(t t

n

i

) dt

n

i+1 t i

n

2

n

i

as n .

t

i

t

n1

n

i

t

n

i+1

| 2 dt

i=0 t

2

1 T 2

2 0

From Theorem 6.1, lim n I T (f n ) exists in L 2 sense. To ﬁnd an explicit formula of the limit, note from the equation a(b a) = 1 2 (b 2 a 2 )1 2 (ba) 2 that

I T (f n ) =

n1

i=0

W t

n

i

W t

n

i+1 W t i

n

= 1 2 n1

i=0

W

2

t

n

i+1 W i

2

t

n

n1

i=0

(W t

=

1

2 W

2

T

1

2

n1

i=0

(W t

n

i+1 W t i

n

) 2

n

i+1 W t i

n

1

2 W T

2

)

2

1

2 T

in L 2 as n . The last convergence follows from the fact that the quadratic variation of Wiener process is T . Therefore, we conclude that

0

T

W t dW t = I T (f) =

1

2 W

2

T

1

2 T .

6.3 Properties of the Stochastic Integral

The basic properties of the Ito integral are summarized in the following theorem:

Theorem 6.2. The following properties hold for any f , g M 2 , any α, β R and any 0 s < T :

a) Linearity:

114

6 Ito’s Stochastic Calculus

0

T

(α f (t) + β g(t)) dW (t) = α

0

T

T

f (t) dW (t) + β 0 g(t) dW (t); (6.11)

b) Isometry:

E

0

T

f (t) dW (t) 2 = E

0

T

|f(t)| 2 dt ;

c) Martingale Property:

E

0

T

f (t) dW (t) F s = 0 f (t) dW (t).

s

In particular, E

T

0

f (t) dW (t) = 0.

(6.12)

Proof. a) If f and g belong to M 2 , then from the proof of Theorem 6.1, they can be approximated by some sequences f 1 , f 2 , ··· and g 1 ,g 2 , ··· . From Deﬁnition 6.4, it is clear that, for each n,

I(α f n +βg n ) =

n1

j=0

(α f(t

n

j

)+βg(t

n

j

)) W t n j+1 W t

n

j

= αI(f n )+βI(g n ),

where {t

n

j

} j=0,1,

is the “common” grid where both f n and g n are constant in

). Taking limit on both sides of this equality as n , we

,n

 n j each interval (t n j−1 ,t obtain which is (6.11).

I(α f +βg) = αI(f)+βI(g),

b) The word isometry means an equality under different metrics. Note that (6.12) can be written as I T (f) L 2 = f M . Hence, the Ito integral connects the two

2

T

metrics L 2 and M

T

2 . The proof of (6.12) follows by constructing the f n in (6.5)

2 0. Combining with (6.9) in

such that I T (f n ) I T (f) L 2 0, f n f M

the proof of Theorem 6.1, the result follows.

T

c) Following the proof of Theorem 6.1, we can ﬁnd a random step function f n approximating f such that I T (f n )I T (f) L 2 0. W.L.O.G., we can assume s = t k for some k. (adding an extra point still gives a step function). Thus

6.4

Ito’s Lemma

115

E(I T (f n )|F s )

= E j=1 f(t j1 )(W t j W t j1 )

n

F s

= j=1 f(t j1 )(W t j W t j1 )+E j=k+1 f(t j1 )(W t j W t j1 )|F s

k

n

k

= j=1 f(t j1 )(W t j W t j1 )

=

I s (f n ) .

(by independent increment and t k = s)

(6.13)

The same argument in the proof of Theorem 6.1 yields I s (f n )I s (f) L 2 0. On the other hand, we have E(I T (f n )|F s )E(I T (f)|F s ) L 2 0, since

E (E(I T (f n )|F s )E(I T (f)|F s )) 2

= E (E(I T (f n )I T (f)|F s )) 2

E E((I T (f n )I T ( f )) 2 |F s )

= E((I T (f n )I T ( f )) 2 )

0.

(Jensen’s Inequality)

(Tower Property)

By taking limit on both sides of Equation (6.13), we have

E(I T (f)|F s ) = I s ( f ) ,

completing the proof.

6.4 Ito’s Lemma

(6.14)

In this section we prove the Ito’s Lemma which is the foundation of mathematical ﬁnance and stochastic calculus.

6.4.1 The case F(t,W t )

Theorem 6.3. (Ito’s Lemma) Suppose that F(t, x) is a real valued function with continuous partial derivatives F t (t,x), F x (t, x) and F xx (t, x) for all t 0 and x R. Assume also that the process F x (t,W t ) belongs to M 2 . Then F(t,W t ) satisﬁes

=

F(T,W T )F(0,W 0 )

0

T

F t (t,W t )+ 1 2 F xx (t,W t ) dt + 0

T

F x (t,W t ) dW t ,

a.s.

(6.15)

116

6 Ito’s Stochastic Calculus

In differential notation, (6.15) can be written as

dF(t,W t ) = F t (t,W t )+ 2 F xx (t,W t ) dt + F x (t,W t ) dW t .

1

Remark 6.4

a) Compare (6.16) with the usual chain rule

(6.16)

dF(t,x t ) = F t (t,x t )dt + F x (t,x t )dx t

for a differentiable function x t . The additional term called the Ito correction.

2 1 F xx (t,W t ) dt in (6.16) is

b) Equation (6.16) is often written in the abbreviated form

dF = F t + 2 F xx dt + F x dW t .

1

(6.17)

Strictly speaking, the differential notation (6.16) does not make sense due to the non-differentiability of Brownian motion (dW t is undeﬁned).

Proof. We ﬁrst prove the case where F t , F x , F xx are all bounded by some C > 0.

Consider a partition 0 = t

by W

n

0

n

< t n < ··· < t

by

n

n

n

i

= T of [0, T ], where t

W ; and t

n

i+1 t i

n

n

by

n

i

n = iT

i

1

n

.

Denote W t

n

i

i

; the increments W

n

i

,t i+1 ] such that

n

n

i+1 W i

n

t. Using Taylor’s

in each

expansion, there is a point W

interval [t

n

i

in each interval [W

i

,W i+1 ] and a point t

n

n

i

F(T,W T )F(0,W 0 )

n1

=

i=0 F(t

n

i+1 ,W

i+1 )F(t

n

n

i

n

,W

i

)

=

=

=

n1

i=0 F(t

n

i+1 ,W

i+1 )F(t

n

n

i

,W i+1 ) +

n

n1

i=0 F(t

n

i

,W i+1 )F(t

n

n

i

n

,W

i

)

n1

i=0

n1

i=0

F t ( ,W

t

n

i

n

i+1 ) i

n

F t ( ,W

t

n

i

n

i+1 ) i

n

t

t

+

+

n1

i=0

F x (t

n

i

n

,W

i

)

n

i

W +

1

2

n1

i=0

F xx (t

n

i

,

W

n

i

)(

n

i

1

2

n1

i=0

F xx (t

n

i

n

,W

i

)

n

i

t +

n1

i=0

F x (t

n

i

n

,W