Introduction Contents

Week 4: Wiener processes, Ito’s lemma & Finance

Harjoat S. Bhamra

Imperial College

2015

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 1 / 44
 Introduction Contents

1 Introduction
Contents
Motivation

2 Wiener Processes (Brownian Motion)

Overview

3 Revision: Ito’s Lemma

4 Deriving the Black-Scholes Partial Differential Equation

Main Results
Overview
Exploiting the Binomial Model

5 American Option Pricing in a Binomial Tree

Some Limitations

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 2 / 44
 Introduction Motivation

Why do we care about Wiener Processes & Ito’s Lemma? I

Very useful for applying and extending the Black-Scholes-Merton Model.

Fischer Black, Robert C. Merton and Myron S. Scholes – pioneering formula for the
valuation of stock options.
Valuation methodology can be applied to practical problems in all finance, including
corporate finance.
Generated new types of financial instruments and facilitated more efficient risk management
in society.

Do you want to understand how to price and risk manage projects ranging from oil exploration to
Hollywood movies?
Do you want to understand how financial markets price risk and how to manage financial risks?
Or would you feel happier blindly applying models and hoping for the best?

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 3 / 44
 Wiener Processes (Brownian Motion) Overview

Brownian Motion in the Black-Scholes-Merton Model

Definition of Brownian Motion
Visualizing Brownian Motion
Black-Scholes Model
Different Ways of Computing the Option Price
Ito’s Lemma and Girsanov’s Theorem

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 4 / 44
 Wiener Processes (Brownian Motion) Overview

Black-Scholes Assumptions
You have seen a Brownian motion appear in the model of a non-dividend paying stock.
The stock price follows the process (under physical measure P)
expected rate of return
dSt z}|{
= µ · dt + σdZt (1)
St
The short selling of securities with full use of proceeds is permitted
There are no transactions costs or taxes.
All securities are perfectly divisible
There are no dividends during the life of the derivative.
There are no arbitrage opportunities.
Security trading is continuous
The risk-free rate of interest, r f , is constant and the same for all maturities
Z = (Z )t∈T is a standard Brownian motion under the physical measure P. What is the Brownian motion for? It is an attempt to
model small and frequent changes in the realized stock return over time, which (based on current information) cannot be
predicted
σdZt is the unexpected change in the stock’s return over the next instant (an infinitesimally small time interval of length dt)
=0
z }| {
Et [σdZt ] = σ Et [dZt ] = 0
=dt
z }| {
Vart [σdZt ] = σ 2 Vart [dZt ] = σ 2 dt
σ is the annualized stock return volatility
√
σ dt is the stock return volatility over the next instant (an infinitesimally small time interval of length dt)
√
σ T − t is the stock return volatility over the time interval [t, T ), where t and T are measured in units of years
Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 5 / 44
 Wiener Processes (Brownian Motion) Overview

Formal Definitions I

Definition 1
In continuous time, a stochastic process X is a collection of random variables (Xt )t∈T , where
T = [0, T ) or T ∈ [0, ∞). Each random variable Xt is a mapping from some underlying state
space, Ω, to the real line R, i.e. Xt : Ω → R.

Definition 2
Consider a continuous-time stochastic process Z = (Zt )t∈T , where T = [0, T ) or T = [0, ∞),
and the random variable Zt maps elements of some state space Ω to the real line. Z is a standard
Brownian motion (or standard Wiener process) under a probability measure P, if (under this
measure)
1 Z0 = 0
2 Zt ∼ N[0, t]
3 ∀t, s ∈ T s.t. s ≥ t, Zs − Zt is independent of Zt
4 all paths of Z are continuous

We can learn alot by playing with the definition of a standard Brownian motion.
√
Consider Zt+∆t − Zt . We know that Zt+∆t − Zt has the same distribution as  ∆t, where
 ∼ N[0, 1]. We can easily see that Et [Zt+∆t − Zt ] = 0

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 6 / 44
 Wiener Processes (Brownian Motion) Overview

Formal Definitions II

Now consider (Zt+∆t − Zt )2 , which has the same distribution as 2 ∆t. We can now compute
Et [(Zt+∆t − Zt )2 ]

=0
z }| {
2
Et [(Zt+∆t − Zt ) ] = Vart [Zt+∆t − Zt ] + (Et [Zt+∆t − Zt )])2 (2)
= ∆t (3)

It is tempting to try and compute the derivative of Brownian motion with respect to time via
the limit
Zt+∆t − Zt
lim (4)
∆t→0 ∆t
Zt+∆t −Zt
We know that ∆t
has the same distribution as
√
 ∆t 
= (5)
∆t (∆t)1/2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 7 / 44
 Wiener Processes (Brownian Motion) Overview

Formal Definitions III

We can see that the above expression is not well defined as ∆t → 0. Looking more closely,
we can see that

Zt+∆t − Zt ∞ with probability 1/2
lim = , (6)
∆t→0 ∆t −∞ with probability 1/2

which is not a function in the usual sense.

The√ problem behaviour exhibited above arises because the standard deviation of Zt+∆ − Zt
is ∆t. This means that over a small time interval the change in the Brownian motion
scales up by more than the size of the time interval. To make thispeasy to see, suppose
1 1
∆t = 100 . In this case, Zt+∆ − Zt has the same distribution as  1/100 = 10  and
√ 1 1
∆t = 10 > 100 = ∆t.
√
In general for ∆t ∈ (0, 1), we have ∆t > ∆t, so over small time intervals Brownian motion
can change by an amount which is large relative√ to the small time interval. As the time
interval gets smaller and smaller, the ratio of ∆t to ∆t gets larger and larger, implying
very large possible changes in the Brownian motion over very small time intervals. This fast
changing property of Brownian motion contributes to the bad behaviour of its time
derivative.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 8 / 44
 Wiener Processes (Brownian Motion) Overview

Visualizing Brownian Motion

1/31/2016 bm-sample-path.png (987×673)

https://quanttutorials.files.wordpress.com/2013/05/bm-sample-path.png 1/1

Try drawing a tangent to the above path at a given time! The self-similarity property of Brownian
motion makes this impossible.
To see what the path of Brownian motion looks like as you zoom in on it click here.
Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 9 / 44
 Wiener Processes (Brownian Motion) Overview

Option Pricing in the Black-Scholes Model

We know that for a European-style option with date-T payoff f (ST ), we can find it’s date-t
price by
1 finding the certainty equivalent of the date-T payoff using date-t information, i.e. the risk-neutral
expectation based on date-t information
Q
Et [f (ST )] (7)
2 discounting the certainty equivalent at the risk-free rate
−r (T −t) Q
e Et [f (ST )] (8)
Date-t option price (denoted by Vt )

Vt = e −r (T −t) EtQ [f (ST )] (9)

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 10 / 44
 Wiener Processes (Brownian Motion) Overview

Different Ways of Computing the Option Price

1 Use a binomial tree and work backwards in time from the date-T price, which is known to
be VT = f (ST ). Need to know how to compute risk-neutral probabilities in the tree (we
have done this!)
2 Deriving a partial differential equation for the option price (need Girsanov’s Theorem to
change from the physical measure to the risk-neutral measure Q and then need Ito’s Lemma)
Solve the partial differential equation numerically – the binomial model is one numerical method you
could use
Solve the partial differential equation by hand – not always easy and often impossible
3 Find the risk-neutral probability distribution of ST based on date-t information (this
involves Ito’s Lemma to solve the stochastic differential equation for S and Girsanov’s
Theorem to change from the physical measure to the risk-neutral measure Q. Then compute
EtQ [f (ST )] via
Monte-Carlo simulation: simulating different outcomes of ST and using them to compute the
risk-neutral mean of f (ST )
Direct calculation of EtQ [f (ST )] – not always easy and often impossible

We need Ito’s Lemma & Girsanov’s Theorem to implement option pricing models. These two
results are based on the mathematics of Brownian motion.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 11 / 44
 Wiener Processes (Brownian Motion) Overview

Ito’s Lemma

Where have you already used Ito’s Lemma?

Solving the stochastic differential equation for the stock price in the Black-Scholes Model

dSt 1 2
= µdt + σdZt ⇒ St = S0 e µt e − 2 σ t+σZt (10)
St

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 12 / 44
 Wiener Processes (Brownian Motion) Overview

Girsanov’s Theorem

Where have you already used Girsanov’s Theorem?

Moving from the stock returns under the physical measure P to stock returns under the
risk-neutral measure P

dSt
= µdt + σdZt , Z = (Z )t∈T is a Brownian motion under P (11)
St
dSt
= rdt + σd Ẑt , Ẑ = (Ẑ )t∈T is a Brownian motion under Q (12)
St
µ−r
d Ẑt = dt + dZt (13)
σ

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 13 / 44
 Revision: Ito’s Lemma

Ito’s Lemma I

Lemma 3 (Ito’s Lemma)

Suppose f ∈ C 2,1 (R × R+ ) and X = (Xt )t∈T (T = [0, T ) or T = [0, ∞)) is a continuous-time
stochastic process with the integral representation
Z t Z t
Xt = X0 + µs ds + σs dZs , (14)
0 0

where Z = (Zt )t∈T is a standard Brownian motion. Then, we have

Z t ∂f (Xs , s)
Z t ∂f (Xs , s) 1
Z t ∂ 2 f (Xs , s) 2
f (Xt , t) = f (X0 , t) + ds + dXs + σs ds.
0 ∂t 0 ∂x 2 0 ∂x 2

It’s easier to restate Ito’s Lemma informally:

∂f (Xt , t) ∂f (Xt , t) 1 ∂ 2 f (Xt , t)

df (Xt , t) = dt + dXt + (dXt )2
∂t ∂x 2 ∂x 2

Ito’s Lemma is just a glorified Taylor series expansion. It stems from an attempt to understand
df (Xt ,t)
the total time derivative dt
. A Taylor series expansion gives us

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 14 / 44
 Revision: Ito’s Lemma

Ito’s Lemma II

∂f (Xt , t) ∂f (Xt , t)
df (Xt , t) = dt + dXt (15)
∂t ∂x
1 ∂ 2 f (Xt , t) ∂ 2 f (Xt , t) 1 ∂ 2 f (Xt , t)
+ (dt)2 + dXt dt + (dXt )2 (16)
2 ∂t 2 ∂x∂t 2 ∂x 2

df (Xt ,t)
Now naively compute dt
and see what happens as dt → 0

df (Xt , t) ∂f (Xt , t) ∂f (Xt , t) dXt

= + (17)
dt ∂t ∂x dt
1 ∂ 2 f (Xt , t) ∂ 2 f (Xt , t) 1 ∂ 2 f (Xt , t) (dXt )2
+ 2
dt + dXt + (18)
2 ∂t ∂x∂t 2 ∂x 2 dt

With dt → 0, dXt → 0, so

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 15 / 44
 Revision: Ito’s Lemma

Ito’s Lemma III

 not well defined, oops!

z}|{
df (Xt , t) ∂f (Xt , t) ∂f (Xt , t)  dZt  1 ∂ 2 f (X , t) (dX )2
t t
= + µt + σt + (19)
 
dt ∂t ∂x  dt  2 ∂x 2 dt

Now writing (14) in differential form

dXt = µt dt + σt dZt ,

we can see that

(dXt )2 (µt dt + σt dZt )2

=
dt dt
µ2t (dt)2 + 2µt σt dtdZt + σt2 (dZt )2
=
dt
2
2 (dZt )
→ σt , as dt → 0
dt

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 16 / 44
 Revision: Ito’s Lemma

Ito’s Lemma IV

Now (dZt )2 = (Zt+dt − Zt )2 has the same distribution as 2 dt, where  ∼ N[0, 1], which provides
some intuition for the heuristic rule
(dZt )2 = dt (20)

(dXt )2
which gives dt
= σt2 and hence

 not well defined, oops!

z}|{
df (Xt , t) ∂f (Xt , t) ∂f (Xt , t)  dZt 
= + µt + σt (21)
 

dt ∂t ∂x  dt 

1 ∂ 2 f (Xt , t) 2
+ σt (22)
2 ∂x 2
| {z }
does not vanish because Brownian motion is sufficiently fast

dZt
dt
is not well-defined, so multiply through by dt

∂f (Xt , t) ∂f (Xt , t) 1 ∂ 2 f (Xt , t) 2

df (Xt , t) = + (µt dt + σt dZt ) + σt dt (23)
∂t ∂x 2 ∂x 2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 17 / 44
 Revision: Ito’s Lemma

Ito’s Lemma V

I prefer to write the above expression as

∂f (Xt , t) ∂f (Xt , t) 1 ∂ 2 f (Xt , t)

df (Xt , t) = dt + dXt + (dXt )2 (24)
∂t ∂x 2 ∂x 2

and use the heuristic rule (20) to figure out that in the continuous-time limit (dt → 0)
(dXt )2 = σt2 dt.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 18 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma I

Assume the price of a non-dividend paying stock at date-t is given by St , and that under the
physical probability measure P

dSt
= µdt + σdZt ,
St

where µ and σ are constants, and Z is a standard Brownian motion under P.

1 Find an expression for d ln St and use it to show that
µt − 1 σ 2 t+σZt
St = S0 e e 2
(this is just solving the stochastic differential equation for the stock price)
1
2 Hence, using the fact that for a normal random variable Y , E [e Y ] = e E [Y ]+ 2 Var [Y ] , show that
−µt
S0 = e E0 [St ].
3 Give a financial interpretation of the above expression.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 19 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma II

We start by applying Ito’s Lemma to find d ln St :

∂(ln St ) 1 ∂ 2 (ln St )
d ln St = dSt + (dSt )2
∂St 2 ∂St2
1
= St−1 dSt + (−St−2 )(dSt )2
2
dSt 1
= − St−2 (dSt )2
St 2
1 dSt 2
 
dSt
= − .
St 2 St
 2
dSt
Now using our heuristic rules St
= σ 2 dt. Therefore

1
d ln St = µdt + σdZt − σ 2 dt
2
 
1
= µ − σ 2 dt + σdZt .
2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 20 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma III

t is a dummy variable so replace it with u:

 
1
d ln Su = µ − σ 2 du + σdZu .
2

Now integrate wrt u from u = 0 to u = t:

Z t  Z t Z t
1 2
d ln Su = µ − σ du + σ dZu
0 2 0 0
 
1 2
ln St − ln S0 = µ − σ t + σ(Zt − Z0 )
2
 
St 1 2
ln = µ− σ t + σZt .
S0 2

Applying the exponential function to each side, we obtain

St 1 2
= e (µ− 2 σ )t+σZt . (25)
S0
Hence, we obtain
1 2
St = S0 e µt e − 2 σ t+σZt
.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 21 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma IV

Question: why did we solve the stochastic differential equation for the stock price by
considering ln St ? The answer lies in how we would have solved the non-stochastic
differential equation

dS(t)
= µS(t) (26)
dt
You probably remember that the solution is ln S(t) = ln S(0) + µt, so a sensible strategy for
solving the stochastic differential equation is to work with ln St
We now want to calculate E0 [St ], i.e. the physical expectation of the date-t stock price,
conditional on date-0 information.

1 2
E0 [St ] = S0 e µt E0 [e − 2 σ t+σZt
]
µt− 21 σ 2 t
= S0 e E0 [e σZt ].

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 22 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma V

Now, note that σZt is a normal random variable, and so
=0
z }| {
1
E0 [e σZt ] = e E0 [σZt ] + 2 Var0 [σZt ]
1
= e 2 Var0 [σZt ]
1 2
= e 2σ Var0 [Zt ]

1 σ2 t
=e 2 .

Therefore

E0 [St ] = S0 e µt
S0 = e −µt E0 [St ]

The date-0 value of the stock is the physical expected future value, discounted back using
the continuously compounded discount rate µ, which is the expected return (per unit time)
on the stock.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 23 / 44
 Revision: Ito’s Lemma

Valuing a non-dividend paying stock: application of Ito’s Lemma VI

Assume the price of a non-dividend paying stock at date-t is given by St , and that under the
risk-neutral probability measure Q

dSt
= rdt + σdZtQ ,
St

where r and σ are constants, and Z Q , is a standard Brownian motion under Q.

1 Find an expression for d ln St and use it to show that
rt − 1 σ 2 t+σZt
Q
St = S0 e e 2

1
2 Hence, using the fact that for a normal random variable Y , E [e Y ] = e E [Y ]+ 2 Var [Y ] , show that
−rt Q
S0 = e E0 [St ].
3 Give a financial interpretation of the above expression.
You can do this yourself!

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 24 / 44
 Deriving the Black-Scholes Partial Differential Equation Main Results

Overview I

The date-t price of a derivative (European or American) with terminal payoff f (ST ) satisfies
the following recursion (which goes backwards in time), which we call the fundamental
valuation equation

early exercise payoff

z }| {
Vt = max( f (St ) , EtQ [e −rdt Vt+dt ]) (27)
| {z }
continuation value

Date-t price, V (St , t), of an American-style option with date-T payoff f (ST ) is given by the
following parabolic partial differential equation when early exercise is not optimal

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0, V (St , t) > f (St ) (28)
∂t ∂St 2 ∂St2

In the early exercise region, V (St , t) = f (St )

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 25 / 44
 Deriving the Black-Scholes Partial Differential Equation Main Results

Overview II

For a European-style option which expires at date-T , early exercise is not possible, so we
have

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0, V (ST , T ) = f (ST ) (29)
∂t ∂St 2 ∂St2

All derivatives written on the stock satisfy the fundamental partial differential equation

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0 (30)
∂t ∂St 2 ∂St2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 26 / 44
 Deriving the Black-Scholes Partial Differential Equation Overview

Steps

We shall use the Binomial Model to see that under the risk-neutral measure Q

dSt
 
Q
Et = rdt (31)
St
We shall use the Binomial Model to see that the date-t price of a derivative (European or American) with terminal payoff
f (ST ) satisfies the following recursion (which goes backwards in time), which we call the fundamental valuation equation

early exercise payoff

z }| {
Q −rdt
Vt = max( f (St ) , Et [e Vt+dt ]) (32)
| {z }
continuation value

We shall apply Ito’s Lemma to the fundamental valuation equation to show that the date-t price of a derivative on the
stock satisfies the fundamental pde

1 2 2 ∂2 ∂ ∂
σ S V (S, t) + rS V + V (S, t) − rV (S, t) = 0 (33)
2 ∂S 2 ∂S ∂t

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 27 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral expected returns I

Exploit the Binomial Model to get intuition

Stock price tree

1 u u2 u3 u4 ... u N−1 uN
 
 0 d ud u2 d u3 d ... u N−2 d u N−1 d 
0 0 d2 ud 2 u2 d 2 ... u N−3 d 2 u N−2 d 2
 
 
d3 ud 3 u N−4 d 3 u N−3 d 3
 
 0 0 0 ... 
S = S0  d4 u N−5 d 4 u N−4 d 4
 
0 0 0 0 ... 
 
 .. .. .. .. .. 

 . . . ... ... ... . . 

 0 0 0 0 0 ... d N−1 ud N−1 
0 0 0 0 0 ... 0 dN

S = [S(i, j)]i,j∈{1,...,N+1}

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 28 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral expected returns II

Stock price vector at date (j − 1)∆t is S j

1 u
   
uN
 
0 d
 u N−1 d
   
0 0
    
S 1 = S0   , S 2 = S0   , . . . S N+1 = S0  (34)
     
..
.. ..

     . 
 .   . 
dN
0 0

Date-t stock prices are just risk-neutral expecations of date-(t + ∆t) prices discounted at
the risk-free rate

q 1−q 0 0 0 ... 0 0
 
 0 q 1−q 0 0 0 0 0 
1−q
 
 0 0 q 0 ... 0 0 
0 0 0 q 1−q ... 0 0
 
 
−r ∆t 
S j−1 =e 0 0 0 0 q ... 0 0  Sj (35)


.. .. .. .. ..
 
 

 . . . ... ... ... . . 

 0 0 0 0 0 ... q 1−q 
0 0 0 0 0 ... 0 q
| {z }
=Φ
Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 29 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral expected returns III

Φ is a matrix of state prices

S j−1 = ΦS j (36)
Stepping away from the Binomial Model, we can write the above expression as

St = EtQ [e −rdt St+dt ] (37)

We now simplify the RHS

St+dt = St + dSt (38)

1 2 2
e −rdt = 1 − rdt + r dt + . . . (39)
2!

1 2 2
e −rdt St+dt = (1 − rdt + r dt + . . .)(St + dSt ) (40)
2!
1 1
= St − rSt dt + r 2 St dt 2 + dSt − rdtdSt + r 2 dt 2 dSt + . . . (41)
2! 2!
EtQ [e −rdt St+dt ] = St − rSt dt + EtQ [dSt ] + ... (42)
|{z}
order strictly greater than 1 in dt
Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 30 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral expected returns IV

Substituting the above expression into St = EtQ [e −rdt St+dt ] gives

St = St − rSt dt + EtQ [dSt ] + ... (43)

|{z}
order strictly greater than 1 in dt

EtQ [dSt ] = rSt dt + ... (44)

|{z}
order strictly greater than 1 in dt
 
dSt
EtQ = rSt + ... (45)
dt |{z}
order strictly greater than 0 in dt

dt → 0 (continuous-time limit)
 
dSt
EtQ = rSt (46)
dt
Using risk-neutral probabilities, the expected rate of return on the stock is the risk-free rate.
Using risk-neutral probabilities, the expected risk premium on the stock is zero.
What would you do if, using risk-neutral probabilities, the expected risk premium on the
stock were not zero?

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 31 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative I

Recall that we can price the stock within the binomial model

S j−1 = ΦS j (47)
This applies more generally and makes it easy to code up option pricing problems using the
binomial tree approach
Suppose we have a European-style option with payoff at date-T given by some function of
the date-T stock price, f (ST )
We approximate f (ST ) via an N + 1-dimensional column vector

f (S(1, N + 1))
 
 f (S(2, N + 1)) 
f (S(3, N + 1))
 
V N+1 = (48)
 
..

 
 . 
f (S(N + 1, N + 1))

The function f operates on each element of the vector S N+1 –this is easy to implement in
MATLAB
Finding the option price at date-(T − ∆t) is easy

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 32 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative II

V N = ΦV N+1 (49)
In general we can use the matrix of states prices to compute the date- (j − 1)∆t option price
from the date-j∆t option price

V j = ΦV j+1 (50)
Extending this to deal with American-style options is easy – all we have to have to is
compare the early exercise payoff with the continuation value. At date-(j − 1)∆t, the early
exercise payoff is f (Sj−1∆t ), which we approximate by the vector

f (S(1, j))
 
 f (S(2, j)) 
f (S(3, j))
 
gj =  (51)
 
..

 
 . 
f (S(N + 1, j))

The date-(j − 1)∆t American option price is given by the maximum of the early exercise
payoff and the option’s continuation value

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 33 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative III

early exercise payoff

z}|{
V j = max( gj , ΦV j+1 ) (52)
| {z }
continuation value

the max operation is carried out element-by-element, which is easy to do in MATLAB

Once again, we can step away from the binomial model. We write

continuation value
z }| {
Vt = max( f (St ) , EtQ [e −rdt Vt+dt ]) (53)
| {z }
early exercise payoff

We call the above expression the fundamental valuation equation

We want to simplify the continuation value

EtQ [e −rdt Vt+dt ] = e −rdt EtQ [Vt+dt ] (54)

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 34 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative IV

We apply Ito’s Lemma to simplify the continuation value

Vt+dt = V (St+dt , t + dt) = V (S, t) + dV (S, t) (55)

∂ ∂ 1 ∂2
= V (S, t) + V (S, t)dt + V (S, t)dS + V (S, t)(dS)2 (56)
∂t ∂S 2 ∂S 2

Compute EtQ [Vt+dt ]

∂2
 
∂ ∂ 1
EtQ [Vt+dt ] = V (S, t) + V (S, t) + rS V (S, t) + σ 2 S 2 2
V (S, t) dt (57)
∂t ∂S 2 ∂S

Compute EtQ [e −rdt Vt+dt ]

EtQ [e −rdt Vt+dt ] = (1 − rdt + . . .) [V (S, t) (58)

∂2
  
∂ ∂ 1
+ V (S, t) + rS V (S, t) + σ 2 S 2 2
V (S, t) dt (59)
∂t ∂S 2 ∂S

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 35 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative V

Expanding, we obtain

EtQ [e −rdt Vt+dt ] = (60)

∂2
 
∂ ∂ 1
V (S, t) + V (S, t) + rS V (S, t) + σ 2 S 2 2
V (S, t) − rV (S, t) dt (61)
∂t ∂S 2 ∂S
+terms of order strictly greater than 1 in dt (62)

The recursive valuation equation Vt = max(f (St ), EtQ [e −rdt Vt+dt ]) becomes
! !
∂V (S, t) ∂V (S, t) 1 2 2 ∂ 2 V (S, t)
V (S, t) = max f (S), V (S, t) + + rS + σ S − rV (S, t) dt + . . .
∂t ∂S 2 ∂S 2
(63)

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 36 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative VI

Suppose that earlyexercise it not optimal, i.e.

∂ 2 V (S,t)

∂V (S,t) ∂V (S,t)
f (S) < V (S, t) + ∂t
+ rS ∂S + 12 σ 2 S 2 ∂S 2 − rV (S, t) dt + . . .. Letting
dt → 1, we see that in the continuous-time limit, f (S) < V (S, t).
In this case the valuation equation becomes

∂ 2 V (S, t)
 
∂V (S, t) ∂V (S, t) 1
V (S, t) = V (S, t) + + rS + σ2 S 2 − rV (S, t) dt + . . .
∂t ∂S 2 ∂S 2
(64)
1 2 2 ∂ 2 V (S, t)
 
∂V (S, t) ∂V (S, t)
0= + rS + σ S − rV (S, t) dt + . . . (65)
∂t ∂S 2 ∂S 2
∂V (S, t) ∂V (S, t) 1 ∂ 2 V (S, t)
0= + rS + σ2 S 2 − rV (S, t) (66)
∂t ∂S 2 ∂S 2
+ terms of order strictly greater than 0 in dt (67)

Letting dt → 0, we obtain the following pde outside of the early exercise region

∂V (S, t) ∂V (S, t) 1 ∂ 2 V (S, t)

+ rS + σ2 S 2 − rV (S, t) = 0, f (S) < V (S, t) (68)
∂t ∂S 2 ∂S 2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 37 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative VII

Now suppose that early exercise is optimal,

∂ 2 V (S,t)

∂V (S,t) ∂V (S,t)
f (S) ≥ V (S, t) + ∂t
+ rS ∂S
+ 12 σ 2 S 2 ∂S 2
− rV (S, t) dt + . . ..
We then obtain

V (St , t) = f (St ) (69)

To summarize, we have the following pde outside the early exercise region

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0, V (St , t) > f (St ) (70)
∂t ∂St 2 ∂St2

In the early exercise region, V (St , t) = f (St )

The boundary between the non-early exercise and early exercise regions is called a free boundary
(because we have to find it as part of the pricing problem). Suppose the time at which the stock
price first reaches the early exercise region, i.e. touches the free boundary starting from a position in
the non-early exercise region is given by τ . Mathematically, we have

τ = inf{t > 0 : V (St , t) ≤ f (St )} (71)

We can write the option price as
Q −r (τ −t)
V (St , t) = Et [e f (Sτ )] (72)

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 38 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative VIII

The above expression is useful conceptually, but not for valuation. We sometimes call it the
probabilistic solution, because it is written in terms of a risk-neutral expectation.
For a European-style option which expires at date-T , early exercise is not possible, so we
have

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0, V (ST , T ) = f (ST ) (73)
∂t ∂St 2 ∂St2

The probabilistic solution for the European-style option price is given by

V (St , t) = Et [e −r (T −t) f (ST )] (74)

For European-style options the probabilistic solution is useful for computation. We can use
Monte-Carlo simulation to estimate the risk-neutral expectation. This is useful when we go beyond
the existing model and have a derivative which depends on the prices of several stocks, making the
partial differential equation approach and extensions of the binomial tree hard to implement.
For any derivative on the stock, we have the fundamental partial differential equation

∂V (St , t) ∂V (St , t) 1 ∂ 2 V (St , t)

+ rSt + σ 2 St2 − rV (St , t) = 0 (75)
∂t ∂St 2 ∂St2

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 39 / 44
 Deriving the Black-Scholes Partial Differential Equation Exploiting the Binomial Model

Risk-neutral valuation of a derivative IX

The pde is parabolic (useful to know if you want to do some further work on solution techniques)
The type of derivative contract tells us the boundary conditions for this pde, which includes early
exercise features
Given the boundary conditions, we can solve for the derivatives price backwards in time on a
binomial tree

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 40 / 44
 American Option Pricing in a Binomial Tree Some Limitations

An American-style vanilla put I

We shall use the binomial tree to solve the parabolic partial differential equation with a free
boundary (sounds cool!)
Stock price tree

1 u u2 u3 u4 ... u N−1 uN
 
 0 d ud u2 d u3 d ... u N−2 d u N−1 d 
0 0 d2 ud 2 u2 d 2 ... u N−3 d 2 u N−2 d 2
 
 
d3 ud 3 u N−4 d 3 u N−3 d 3
 
 0 0 0 ... 
S = S0  d4 u N−5 d 4 u N−4 d 4
 
0 0 0 0 ... 
 
. . . . .
.. .. .. .. ..
 

 ... ... ... 

 0 0 0 0 0 ... d N−1 ud N−1 
0 0 0 0 0 ... 0 dN
The date-(j − 1)∆t American option price is given by the maximum of the early exercise
payoff and the option’s continuation value

early exercise payoff

z}|{
V j = max( gj , ΦV j+1 ) (76)
| {z }
continuation value

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 41 / 44
 American Option Pricing in a Binomial Tree Some Limitations

An American-style vanilla put II

We can build a matrix of prices for the American put

V11 V12 V13 V14 V15 ... V1,N V1,N+1

 
 0 V22 V23 V24 V25 ... V2,N V2,N+1 
 
 0 0 V33 V34 V35 ... V3,N V3,N+1 
0 0 0 V44 V45 ... V4,N V4,N+1
 
 
V= 0 0 0 0 V55 ... V5,N V5,N+1
 

.. .. .. .. ..
 
 

 . . . ... ... ... . . 

 0 0 0 0 0 ... VN,N VN,N+1 
0 0 0 0 0 ... 0 VN+1,N+1
At each node in the tree, we can check whether or not it is optimal to exercise early
We can approximate the free boundary

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 42 / 44
 American Option Pricing in a Binomial Tree Some Limitations

An American-style vanilla put III

140

130

120

110

100

90

80

70
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 43 / 44
 American Option Pricing in a Binomial Tree Some Limitations

An American-style vanilla put IV

Close to ‘today’, the tree does not have many nodes for stock prices – our approximation to
the free boundary is poor in this part of the tree
This means that when an option approaches expiry, the free boundary is approximated poorly.
Take a look at Pricing Put Options With The Binomial Method on the Wolfram website.
Can reduce this problem using other techniques such as finite difference methods –
equivalent to using trinomial tree.

Harjoat S. Bhamra Week 4: Wiener processes, Ito’s lemma & Finance 2015 44 / 44