STAT3603 Stochastic Processes

Chapter 6
Brownian Motion and Its Applications

Wang Chen
stacw@hku.hk
Outline

▶ Review of Multivariate Normal Distribution

▶ Origin of Brownian Motion
▶ Statistical Properties of Brownian Motion
▶ Variations on Brownian Motion
▶ Pricing Stock Options

Reference
▶ IPM Chapter 10

2 / 37
Review of Multivariate Normal Distribution
An important distribution for studying Brownian Motion is the
Gaussian (Normal) Distribution.
▶ Univariate normal distribution: X ∼ N(µ, σ 2 )

1 (x−µ)2
f (x) = √ e− 2σ 2
2πσ 2  
1 1
= 1 1 exp − (x − µ)(σ 2 )−1 (x − µ)
2
(2π) 2 (σ ) 2 2

for −∞ < x < ∞.

E (X ) = µ and Var (X ) = σ 2 .
X −µ
Standardization: Y = ∼ N(0, 1),
σ2
pdf of Y : ϕ(y ) = √12π e −y /2
Ry
cdf of Y : P(Y < y ) = −∞ ϕ(x)dx = Φ(y ).

3 / 37
▶ Bivariate normal distribution:
     2 
X µx σx ρσx σy
∼N ,
Y µy ρσx σy σy2

f (x, y )
1 n 1
= exp − ×
2(1 − ρ2 )
q
2π σx2 σy2 (1 − ρ2 )
(x − µx )2 2ρ(x − µx )(y − µy ) (y − µy )2 o
 
− +
σx2 σx σy σy2
1
= 1 ×
2 σx2 ρσx σy 2
(2π) 2
ρσx σy σy2
(  2 −1  )
1
σx ρσx σy x − µx
exp − x − µx y − µy
2 ρσx σy σy2 y − µy

for −∞ < x, y < ∞.

4 / 37
The means, variances and covariance are expressed as
     
X E (X ) µx
E = =
Y E (Y ) µy
     2 
X Var (X ) Cov (X , Y ) σx ρσx σy
Var = =
Y Cov (X , Y ) Var (Y ) ρσx σy σy2

▶ Multivariate Normal Distribution X ∼ N(µ, Σ)

where X is an n × 1 random vector. The joint pdf is given by

 
− n2 − 12 1 ′ −1
f (x) = (2π) |Σ| exp − (x − µ) Σ (x − µ) , x ∈ Rn
2
Mean vector
E (X) = µ = (µi )n×1
Variance and Covariance matrix
Var (X) = Σ = (σij )n×n
5 / 37
Some Special Properties

1. Linear transformation
▶ Univariate X ∼ N(µ, σ 2 ) ⇒ aX + b ∼ N(aµ + b, a2 σ 2 ). In
particular, X σ−µ ∼ N(0, 1).
▶ Multivariate X ∼ N(µ, Σ) ⇒ AX + b ∼ N(Aµ + b , AΣA′ ).
In particular, Σ−1/2 (X − µ) ∼ N(0, I ) and
α′ X ∼ N(α′ µ, α′ Σα).
2. Marginal distribution, conditional distribution and independence
     2 
X µx σx ρσx σy
∼N ,
Y µy ρσx σy σy2

▶ X ∼ N(µx , σx2 ) and Y ∼ N(µy , σy2 ).

 
σx 2 2
▶ X |Y = y ∼ N µx + ρ (y − µy ), σx (1 − ρ ) .
σy
▶ X and Y are independent if and only if ρ = 0.
6 / 37
Origin of Brownian Motion

Historic Remarks
▶ 1827: A botanist, Robert Brown, observed the erratic
motion of grains of pollen suspended in a liquid. Later it was
found such irregular motion comes from the extremely large
number of collisions of the suspended pollen grains with
molecules of the liquid.

7 / 37
▶ 1905: Albert Einstein investigated on the physics and
mathematics theory of the random movement discovered by
Robert Brown. Such movement was later named as the
Brownian Motion.

8 / 37
▶ 1923: Norbert Wiener completed the mathematical theory of
Brownian motion. The standardized version of Brownian
motion is also named the Wiener process.

W0 = 0
Wt has independent increment
Ws+t − Ws ∼ N(0, t)
Wt is continuous in t

9 / 37
Simple random walk model on one-dimension movements of
particle
▶ X (t) = position of the particle at time t
▶ {X (t), t ≥ 0} is a continuous time, continuous state space
stochastic process
▶ Assume a symmetric random walk model for the observable
process at discrete time points.
▶ Partition [0, t] into n equally distanced intervals, each with
length h = t/n.
▶ Zi = displacement at time ti = ih

1
P(Zi = −∆) = P(Zi = ∆) =
2
E (Zi ) = 0, Var (Zi ) = ∆2

10 / 37
 n
▶ X (t) =
P
Zi with t = nh
i=1

t 2
E (X (t)) = 0, Var (X (t)) = n∆2 = ∆
h
▶ By Central Limit Theorem (CLT), X (t) is asymptotically
distributed as normal:
∆2
X (t) ∼ N(0, σ 2 t) where σ 2 =
h
▶ It can be easily seen that the symmetric random walk model
has independent and stationary increments. So it is
reasonable to assume that {X (t), t ≥ 0} also has independent
and stationary increments.
▶ A key point of the asymptotic distribution is that the step size
√
∆ must be of order h, so that σ 2 is finite.

11 / 37
Definition
A continuous time stochastic process {X (t), t ≥ 0} taking values
in (−∞, ∞) is called a Brownian motion if the following
conditions are satisfied.
1. X (0) = 0
2. The process has stationary and independent increments.
3. For every t > 0, X (t) ∼ N(0, σ 2 t)
4. The sample path X (t) is continuous everywhere with
probability 1.

A Brownian motion with variance parameter σ 2 = 1 is called a

Standard Brownian Motion (or Wiener Process), usually
denoted as {B(t), t ≥ 0} (or {Wt , t ≥ 0}).

12 / 37
Since a general Brownian motion can be obtained by properly
scaling a standard Brownian motion, we can study the main
statistical properties simply based on standard Brownian motion
{B(t); t ≥ 0}.

Also, it is easy to see that

p
∆X (t) = X (t + ∆(t)) − X (t) = σ ∆(t)ϵ = σ∆B(t), (1)

where ϵ ∼ N(0, 1). Note that (1) is the discretized version of

Brwonian motion. Moreover, by letting ∆(t) → 0, it implies

dX (t) = σdB(t), (2)

which is the continuous version of Brownian motion.

13 / 37
Example
Find the probability of Brownian motion hitting A before −B,
where A > 0 and B > 0.

Solution: Recall that Brownian motion is a limit of the symmetric

random walk. The key point is to change the original question to
the ruin probability of a symmetric random walk with initial state
i. The ruin probability that, starting with i, the symmetric Markov
chain will eventually reach N before reaching 0 is i/N. Since

B A+B
i= , N= ,
∆ ∆
we have
B
P (up A before down B) = .
A+B

14 / 37
Statistical Properties of Brownian Motion
▶ Marginal Distribution
For any t > 0, B(t) ∼ N(0, t) with p.d.f.
1 x2
ft (x) = √ e − 2t , −∞ < x < ∞
2πt
▶ Joint Distribution
For any 0 = t0 < t1 < · · · < tn , the increments
Yi = B(ti ) − B(ti−1 ), i = 1, . . . , n are independently
distributed as
Yi ∼ N(0, ti − ti−1 )
so that the random vector Y = (Y1 , . . . , Yn )′ follows a
multivariate normal distribution:
 
t1 0 ··· 0
 0 t2 − t1 · · · 0 
Y ∼ N(0, ΣY ), ΣY =  .
 
. . ..
 .. .. .. 
. 
0 0 ··· tn − tn−1

15 / 37
▶ The random vector W = (B(t1 ), . . . , B(tn ))′ is a linear
transformation of Y:
    
Y1 1 0 ··· 0 Y1
 Y1 + Y2  1 1 · · · 0 Y2 
W=  =  .. .. . . ..   ..  = AY
    
..
 .   . . . .  . 
Y1 + · · · Yn 1 1 ··· 1 Yn

so that it also follows a multivariate normal distribution:

W ∼ N(0, AΣY A′ )
     
B(t1 ) 0 t1 t1 · · · t1
B(t2 ) 0 t1 t2 · · · t2  
W =  .  ∼ N  .  ,  . . .
     
.. 
 ..   ..   .. .. .. . 
B(tn ) 0 t1 t2 · · · tn

16 / 37
 ▶ For any 0 < t1 ≤ t2 ,
     
B(t1 ) 0 t1 t1
∼N ,
B(t2 ) 0 t1 t2
▶ Covariance and Correlation
Cov (B(t1 ), B(t2 )) = t1 ,
r
t1
Corr (B(t1 ), B(t2 )) = .
t2
▶ Conditional Distribution For any 0 < t1 ≤ t2 ,
B(t2 )|B(t1 )=x1 ∼ N(x1 , t2 − t1 ),
 
t1 t1
B(t1 )|B(t2 )=x2 ∼ N x2 , (t2 − t1 ) .
t2 t2
Remark
Even though a Brownian motion is continuous everywhere, it is
nowhere differentiable. To see this, note that for t, h > 0,
 
B(t + h) − B(t) 1
∼ N 0, .
h h 17 / 37
Example
(Bicycle Race Between Two Competitors) Let Y (t) be the amount
of time (in s) by which the inside racer is ahead when 100t percent
of the race has been completed. Assume {Y (t), 0 ≤ t ≤ 1} is a
Brownian motion with σ 2 = 1.5. If the inside racer is leading by 1
second at the midpoint of the race, what is the probability that she
is the winner?

Solution:
1
P(Y (1) > 0|Y (0.5) = 1) = P(B(1) > 0|B(0.5) = √ )
1.5
 
1 Y (t)
B(1)|B(0.5)=1/√1.5 ∼ N √ , 0.5 , where B(t) =
1.5 σ
1 !
0 − 1.5
√
P(Y (1) > 0|Y (0.5) = 1) = 1 − Φ √ = 0.8759
0.5

18 / 37
Example
(continued) If the inside racer wins the race by a margin of 0.2
second, what is the probability that she was ahead at the midpoint?

Solution:

P(Y (0.5) > 0|Y (1) = 0.2) = P(B(0.5) > 0|B(1) = 0.1633)
 
0.5 0.5
B(0.5)|B(1)=0.1633 ∼ N (0.1633), (1 − 0.5)
1 1
≡ N(0.08165, 0.25)
 
0 − 0.08165
P(Y (0.5) > 0|Y (1) = 0.2) = 1 − Φ √ = 0.5649
0.25

19 / 37
Variations on Brownian Motion
For Brownian motion, the rate of change in mean is zero and the
rate of change in variance is σ 2 (a constant). In reality, the mean
and variance of a stochastic process can evolve over time.

Brownian Motion with Drift

Definition
{X (t) ; t ≥ 0} is a Brownian motion with drift if

X (t) = µt + σB (t) .

20 / 37
As for (1) and (2), the discretized version of Brwonian motion with
drift is
p
∆X (t) = X (t + ∆(t)) − X (t) = µ∆(t) + σ ∆(t)ϵ = µ∆(t) + σ∆B(t),
(3)

and the continuous version of Brwonian motion with drift is

dX (t) = µdt + σdB(t), (4)

that is, for Brownian motion with drift, the rate of change in mean
is µ and the rate of change in variance is σ 2 .

Conventionally, µ and σ are referred to as the drift and volatility

parameters, respectively.

21 / 37
Statistical properties of the Brownian motion with drift process can
be directly derived from those of the standard Brownian motion.

▶ Marginal Distribution
For any t > 0, X (t) ∼ N(µt, σ 2 t) with p.d.f.
1 (x−µt)2
ft (x) = √ e− 2σ 2 t , −∞ < x < ∞
2πσ 2 t
▶ Joint Distribution
     
X (t1 ) µt1 B(t1 )
X (t2 ) µt2  B(t2 )
 ..  =  ..  + σ  .. 
     
 .   .   . 
X (tn ) µtn B(tn )
   
µt1 t1 t1 · · · t1
µt2  t1 t2 · · · t2  
∼ N  .  , σ 2  . . .
   
.. 
 ..   .. .. .. . 
µtn t1 t2 · · · tn
22 / 37
      
X (t1 ) µt1 2 t1 t1
▶ For any 0 < t1 ≤ t2 , ∼N ,σ
X (t2 ) µt2 t1 t2
▶ Covariance and Correlation
Cov(X (t1 ), X (t2 )) = σ 2 t1 ,
r
t1
Corr(X (t1 ), X (t2 )) = .
t2
▶ Conditional Distribution For any 0 < t1 ≤ t2 ,
X (t2 )|X (t1 ) = x1 ∼ N(x1 + µ(t2 − t1 ), σ 2 (t2 − t1 ))
 
t1 2 t1
X (t1 )|X (t2 ) = x2 ∼ N x2 , σ (t2 − t1 )
t2 t2
Remark
The Brownian motion with drift process can be generalized to the
Ito (diffusion) process:

dX (t) = µ(X (t), t)dt + σ(X (t), t)dB(t), (5)

that is, both drift term and volatility term are allowed to depend
on X (t) and t. 23 / 37
Geometric Brownian Motion
Definition
{S (t) ; t ≥ 0} is a Geometric Brownian motion if

σ2
  
S (t) = S(0) exp µ − t + σB(t) .
2

Geometric Brownian motion {S (t) ; t ≥ 0} is widely used to model

the stock price at time t.
▶ Let Z (t) = log S(t). Then, given S(0) = S0 , Z (t) is a
Brownian motion with drift process.
▶ The discretized version of Z (t) is
∆Z (t) = Z (t + ∆(t)) − Z (t) = (µ − σ 2 /2)∆(t) + σ∆B(t),
and the continuous version of Z (t) is
dZ (t) = (µ − σ 2 /2)dt + σdB(t),
that is, for the logarithm of price Z (t), the rate of change in
mean is µ − σ 2 /2 and the rate of change in variance is σ 2 .
24 / 37
▶ Geometric Brownian motion has the following continuous
version:
dS(t) = µS(t)dt + σS(t)dB(t). (6)
To see this, we need Ito’s lemma.
▶ Ito’s lemma (optional). Suppose that G (X (t), t) is a
differentiable function of X (t) and t, and X (t) satisfies (5).
Then,
σ 2 (X (t), t) ∂ 2 G
 
∂G ∂G
dG = µ(X (t), t) + + dt
∂x ∂t 2 ∂x 2
 
∂G
+ σ(X (t), t) dB(t),
∂x
where
∂G ∂G (x, t) ∂G ∂G (x, t)
= |x=X (t) , = |x=X (t) , and
∂x ∂x ∂t ∂t
∂2G ∂ 2 G (x, t)
= |x=X (t) .
∂x 2 ∂x 2
25 / 37
▶ Let G (S(t), t) = log S(t), where S(t) satisfies (6). Then,

∂G 1 ∂G ∂2G 1
= , = 0 and 2
=− 2 .
∂x S(t) ∂t ∂x S (t)

By Ito’s lemma, we have

σ 2 S 2 (t)
  
1 1
d log S(t) = µS(t) + − 2 dt
S(t) 2 S (t)
 
1
+ σS(t) dB(t)
S(t)
= (µ − σ 2 /2)dt + σdB(t),

which is the continuous version of Z (t). Therefore, (6) is the

continuous version of the geometric Brownian motion.

26 / 37
 ▶ By (6), the discretized version of S(t) is

∆S(t) S(t + ∆(t)) − S(t)

= = µ∆(t) + σ∆B(t), (7)
S(t) S(t)
which implies that µ is the expected (relative) return of the
stock per unit time, and σ is the standard deviation (or
volatility) of the (relative) return of the stock per unit time.
Note that σ is not the volatility of the stock price itself.

Given S(0) = S0 , the following statistical properties the geometric

Brownian motion process {S(t); t ≥ 0} are straightforward.
▶ Marginal Distribution
For any t > 0, log S(t) ∼ N(log S0 + (µ − σ 2 /2)t, σ 2 t)
▶ Moment properties
2
E(S(t)) = S0 e µt , E(S 2 (t)) = S02 e 2µt+σ t and
 2 
Var(S(t)) = S02 e 2µt e σ t − 1 .

27 / 37
Example
A stock price follows the geometric Brownian motion with an
expected return of 10% per annum and volatility 20% per annum.
The current price of he stock is $12. What is the probability that
the stock will decrease in price after 3 months?

Solution: Note that µ = 10%, σ = 20%, and S0 = 12. Then,

log S(0.25) ∼ N(log 12 + (0.1 − 0.22 /2) × 0.25, 0.22 × 0.25)

≡ N(2.505, 0.01).

Therefore, the desired probability is

P(S(0.25) < 12) = P(log S(0.25) < log 12)

 
log 12 − 2.505
=Φ √ = Φ(−0.201) = 0.4203.
0.01

28 / 37
Example
(continued) Determine the mean and volatility of the stock price
after one year.

Solution: It is easy to see that the mean of the stock price after
one year is

E(S(1)) = S0 e µt = 12 × e 0.1 = 13.262;

moreover, since
2
Var(S(1)) = S02 e 2µt (e σ t − 1) = 122 × e 2×0.1 (e 0.04 − 1) = 7.1779,

the volatility of the stock price after one year is

p √
Var(S(1)) = 7.1779 = 2.679.

29 / 37
Pricing Stock Options
A derivative is a financial instrument whose value depends on
values of other financial instruments. The instrument whose value
the derivative depends is called the “underlying”.

Among many instruments, options convey the right, but not the
obligation, to engage in a future transaction on some underlying
security.
▶ Call option gives its holder the right to buy a certain amount
of an underlying asset by a certain date for a certain price.
▶ Put option gives its holder the right to sell a certain amount
of an underlying asset by a certain date for a certain price.
▶ The buying or selling price is known as the exercise price or
strike price.
▶ The date that an option expires is called expiration date or
maturity date.

30 / 37
Two majority styles of options
▶ American option: may be exercised at any time before
maturity
▶ European option: can be exercised only at the maturity date

Below, we mainly focus on European options based on the

underlying stock S(t), which is a geometric Brownian motion.

Let K be the strike price and T the maturity date. Then, the
payoff of the options is
▶ Call option: (S(T ) − K )+ = max(S(T ) − K , 0)
▶ Put option: (K − S(T ))+ = max(K − S(T ), 0)

One of the most important questions in finance is how to price the

option.

31 / 37
▶ The price is calculated in a risk-neutral world such that
 
S(T )
E |FT −∆ = S(0),
e rT
where FT −∆ contains all past information, and r is the
risk-free interest rate. In a risk-neutral world, there has no
arbitrage opportunity, that is, no sure-win strategy exists.
▶ By noting that E(S(t)) = S0 e µt conditional on S(0) = S0 , it
is not hard to see that µ = r in a risk-neutral world. This
means that
σ2
  
S(t) = S(0) exp r − t + σB(t) (8)
2
in a risk-neutral world.
▶ Conventionally, we use EQ to denote the expectation in a
risk-neutral world. A fair price of the option is
 
Payoff of the option
C (0) = EQ , (9)
e rT
where S(t) satisfies (8).
32 / 37
Theorem
(Black-Scholes formula) Suppose that S(t) is a geometric
Brownian motion. Given S(0) = S0 (current price of the stock),
the price of a European call option with expiration date T and
strike price K is given by
 √ 
C (0) = S0 Φ σ T + b − Ke −rT Φ (b) ,

where
log (S0 /K ) + (r − σ 2 /2)T
b= √ .
σ T

Proof: By (9), conditional on S(0) = S0 , the price is

 
max(S(T ) − K , 0)
C (0) = EQ
e rT
1 h i
= rT EQ max(S0 e Y (T ) − K , 0) ,
e
where Y (T ) = log[S(T )/S0 ].
33 / 37
By (8), we know that Y (T ) is normal with mean µT and variance
σ 2 T , where µ = r − σ 2 /2. Hence,
Z ∞
1 1 2 2
C (0) = rT max(S0 e y − K , 0) √ e −(y −µT ) /2σ T dy
e 2πσ T2
Z−∞∞
1 1 2 2
= rT (S0 e y − K ) √ e −(y −µT ) /2σ T dy .
e log(K /S0 ) 2πσ 2 T
 √ 
Changing of a variable w = (y − µT ) / σ T yields
Z ∞ √ Z ∞
1 σw T −w 2 /2 1 2 /2
C (0)e rT
= S0 e µT
√ e e dw −K √ e −w dw
2π a 2π a

where
log (K /S0 ) − µT
a= √ .
σ T

34 / 37
Now,
Z ∞ √ Z ∞ √ 2
1 σw T −w 2 /2 σ 2 T /2 1
√ e e dw = e √ e −(w −σ T ) /2 dw
2π a 2π
 a √  
σ 2 T /2
= e P N σ T,1 ≥ a
2
 √ 
= e σ T /2 P N (0, 1) ≥ a − σ T
2
  √ 
= e σ T /2 P N (0, 1) ≤ − a − σ T
2
 √ 
= e σ T /2 Φ σ T − a .

So we have
2
 √ 
C (0)e rT = S0 e µT +σ T /2 Φ σ T − a − K Φ (−a)
 √ 
= S0 e rT Φ σ T − a − K Φ (−a) .

Finally, the result holds by letting b = −a.

35 / 37
Remark
By the observable value of C (0) from the market, we can calculate
σ according to Black-Scholes formula. This obtained value of σ is
called implied volatility. Ideally, the implied volatility should be
unchanged across different choices of S0 /K and T . However, this
is not the case in real world, where the implied volatility has the
“smile” pattern (usually called volatility smile).

36 / 37
This implies that BS formula may not be the case in many
circumstance. The unsatisfactory performance of BS formula could
be due to
(i) S(t) may have a time-varying volatility (say σ(t)) instead of a
constant volatility σ.

(ii) S(t) may not be a geometric Brownian motion (e.g, S(t) can
be the jump diffusion process).

37 / 37