Introduction to Ito’s Lemma

Wenyu Zhang

Cornell University
Department of Statistical Sciences

May 6, 2015

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 1 / 21

Overview

1 Background

2 Ito Processes

3 Ito’s Lemma

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 2 / 21

Background

Proved by Kiyoshi Ito (not Ito’s theorem on group theory by Noboru

Ito)
Used in Ito’s calculus, which extends the methods of calculus to
stochastic processes
Applications in mathematical finance e.g. derivation of the
Black-Scholes equation for option values

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 3 / 21

Ito Processes

Question
Want to model the dynamics of process X (t) driven by Brownian motion
W (t).

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 4 / 21

Ito Processes: Discrete-time Construction

T
Partition time interval [0,T] into N periods, each of length ∆t = N;
tn = n∆t
Xtn+1 = Xtn + µtn ∆t + σtn ∆Wtn
I drift µ
I volatility σ
I fluctuations ∆Wtn = Wtn+1 − Wtn ∼ N(0, ∆t)

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 5 / 21

Ito Processes: Discrete-time Construction

Summing the increments,

N−1
X N−1
X
XT = X0 + µtn ∆t + σtn ∆Wtn
n=0 n=0

Continuous-time analogue as N → ∞,
Z T Z T
?
XT →
− X0 + µt dt + σt dWt
0 0

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 6 / 21

Ito Processes: Discrete-time Construction

Regularity conditions for µt and σt

adapted to FtW
continuous in t R 
T
integrability conditions (σt ∈ M2 i.e. E 0 σt2 dt < ∞)
Riemann-Stieltjes integral
N−1
X Z T
µtn ∆t → µt dt
n=0 0

Ito integral
N−1 Z T
L2
X
σtn ∆Wtn −→ σt dWt
n=0 0

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 7 / 21

Ito Integrals

Question
RT
What is 0 σt dWt ?

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 8 / 21

Ito Integrals

Theorem (Existence and Uniqueness of Ito Integral)

Suppose that vt ∈ M2 satisfies the following: For all t ≥ 0,
A1) vt is a.s. continuous
A2) vt is adapted to FtW
RT
Then, for any T > 0, the Ito integral IT (v ) = 0 vt dWt exists and is
unique a.e.

Steps for proof

1 Construct a sequence
r  of adapted stochastic processes vn such that
RT 
kv − vn kM2 = E 0 |vn (t) − v (t)|2 dt → 0
2 Show that kIT (vn ) − IT (v )kL2 → 0
3 Show the a.s. uniqueness of the limit IT (v )

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 9 / 21

Ito Integrals: Example

Example (Ito Integral)

Z T N−1
X
Wt dWt with approximating sums Wtn ∆Wtn
0 n=0

N−1 N−1
X 
X 1 1
Wtn (Wtn+1 − Wtn ) = (Wt2n+1 − Wt2n ) − (Wtn+1 − Wtn )2
2 2
n=0 n=0
N−1
1 2 1X
= WT − (Wtn+1 − Wtn )2
2 2
n=0
L21 1
−→ WT2 − T as N → ∞
2 2

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 10 / 21

Ito Integrals: Example

Example (Riemann-Stieltjes Integral)

Z T
Gt dGt with G ∈ C 1 , G (0) = 0
0

Z T Z T
Gt dGt = Gt Gt0 dt
0 0
1
= GT2
2

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 11 / 21

Ito Integral: Properties
Linear in the integrand
Time-additive
Martingale
Proof
For t < T , increase the partition by an extra point tk = t.
"N−1 #
X
E [IT (vn ) − It (vn )|Ft ] = E σtn ∆Wtn |Ft
n=k
"N−1 #
X
=E E (σtn ∆Wtn |Ftn )|Ft
n=k
"N−1 #
X
=E σtn E (∆Wtn |Ftn )|Ft
n=k
=0
IT (vn ) is a martingale. Martingales are preserved under L2 limits.
Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 12 / 21
Ito Processes
Z t Z t
Xt − X0 = µs ds + σs dWs
0 0
SDE notation:

dXt = µt dt + σt dWt

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 13 / 21

Ito’s Lemma

Theorem (Ito’s Lemma)

Suppose that f ∈ C 2 . Then with probability one, for all t ≥ 0,

∂f 1 ∂2f
df (Xt ) = (Xt )dXt + (Xt )(dXt )2
∂x 2 ∂x 2
Z t Z t
1
f (Xt ) − f (X0 ) = f 0 (Xs )dXs + f 00 (Xs )ds
0 2 0
Explicit statement:

1 2 ∂2f
 
∂f ∂f
df (Xt ) = µt (Xt ) + σt 2 (Xt ) dt + σt (Xt )dWt
∂x 2 ∂x ∂x

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 14 / 21

Ito’s Lemma: Idea

Can be obtained heuristically by second order Taylor expansion of f

about Xt
(dXt )2 = (µt dt + σt dWt )2 term cannot be dropped
I (dWt )2 = dt
drop terms  dt
Z T N−1
X
p
(dt) = lim (∆t)p
0 N→∞
n=0
T p
 
= lim N
N→∞ N
= 0 as N → ∞ if p > 1

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 15 / 21

Ito’s Lemma: Idea

Z T N−1 Z T
2
2 L
X
2 2
(dWt ) = dt since (dWt ) = lim (∆Wtn ) = T = dt
0 N→∞ 0
n=0

"N−1 # N−1 N−1

X X X
2 2
E (∆Wtn ) = E (∆Wtn ) = ∆t = T
n=0 n=0 n=0
"N−1 #2 "N−1 # N−1
X X X
2 2
E (∆Wtn ) − T = Var (∆Wtn ) = Var (∆Wtn )2
n=0 n=0 n=0
N−1
X N−1
X
= (E (∆W )4 − [E (∆W )2 ]2 ) = (3(∆t)2 − (∆t)2 )
n=0 n=0
2
2T
= →0 as N → ∞
N

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 16 / 21

Ito’s Lemma: More details

More rigorously, for bounded continuous g ,

N−1 Z T
2
2 L
X
g (Wtn )(∆Wtn ) −→ g (Wt )dt as N → ∞
n=0 0

Proof RT
Since t 7→ g (Wt ) is a.s. continuous, N−1
P
n=0 g (Wtn )∆t → 0 g (Wt )dt.
WTS:
N−1
X  L2
g (Wtn ) (∆Wtn )2 − ∆t −→ 0

IN =
n=0

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 17 / 21

Ito’s Lemma: More details
nP o
N−1
1. E (IN2 ) = E 2 2
n=0 (g (Wtn )) [(∆Wtn ) − ∆t]
2

E g (Wtn )g (Wtm )[(∆Wtn )2 − ∆t][(∆Wtm )2 − ∆t]



= E E g (Wtn )g (Wtm )[(∆Wtn )2 − ∆t][(∆Wtm )2 − ∆t]|Ftm



= E g (Wtn )g (Wtm )[(∆Wtn )2 − ∆t]E (∆Wtm )2 − ∆t|Ftm

  

=0 since {Wt2 − t} is martingale

CT 2
2. E (IN2 ) ≤ N

N−1
X
E (IN2 ) ≤ kg k2∞ E [(∆Wtn )2 − ∆t]2 since g is bounded
n=0
N−1
X √
= kg k2∞ (∆t)2 E (W12 − 1)2 since ∆Wtn ∼ ∆tW1
n=0
N−1 2
X T CT 2
=C =
n=0
N N
Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 18 / 21
Ito’s Lemma: Example

Example (Ito’s Lemma)

Use Ito’s Lemma, write Zt = Wt2 as a sum of drift and diffusion terms.

Zt = f (Xt ) with µt = 0, σt = 1, X0 = 0, f (x) = x 2

dZt = df (Xt )
1
= f 0 (Xt )dXt + f 00 (Xt )(dXt )2
2
1
= 2Wt dWt + 2(dWt )2
2
= 2Wt dWt + dt

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 19 / 21

Ito’s Lemma: Higher dimensions

Ito’s Lemma
If Xt and Yt are Ito processes and f : R2 7→ R is sufficiently smooth, then

∂f ∂f
df (Xt , Yt ) = dXt + dYt
∂x ∂y
1 ∂2f 2 1 ∂2f 2 ∂2f
+ (dXt ) + (dYt ) + (dXt )(dYt )
2 ∂x 2 2 ∂y 2 ∂x∂y

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 20 / 21

References

Rene L. Schilling/Lothar Partzsch

Brownian Motion - An Introduction to Stochastic Processes (2012)

CUHK course notes (2013)

Chapter 6: Ito’s Stochastic Calculus

Karl Sigman
Columbia course notes (2007)
Introduction to Stochastic Integration

Wenyu Zhang (Cornell) Ito’s Lemma May 6, 2015 21 / 21