MScFE xxx [Course Name] - Module X: Collaborative Review Task

Revised: 09/09/2019

5
Module 5: The Black-Scholes Model................................................................. 4
1. Notes: An Introduction to the Black-Scholes Model ............................................... 5

2. Transcript: Girsanov’s Theorem ................................................................................... 10

3. Notes: Pricing Options ..................................................................................................... 13

4. Transcript: Pricing a Call Option.................................................................................. 17

5. Notes: Hedging ................................................................................................................... 20

6. Transcript: Pricing a Digital Option ........................................................................... 23

7. Notes: Generalized Black-Scholes.............................................................................. 26

8. Transcript: Generalized Black-Scholes Model ...................................................... 30

9. Notes: Problem set ............................................................................................................ 32

© 2019 - WorldQuant University – All rights reserved.

Module 5: The Black-Scholes Model

This Module introduces an important example of a complete market model: the Black-Scholes
Model. The module begins by describing the Black-Scholes model with one risky asset and one
riskless bank account that grows at a constant, continuously compounded rate of interest.
Then, the module shows how to price both vanilla and exotic derivatives in this simple model,
as well as how to derive the Black-Scholes partial differential equation. The module concludes
by discussing the multi-asset generalization of the Black-Scholes model.

5
The Black-Scholes Model is an important example of a complete market model. We will
begin with the simple model consisting of one stock 𝑆 and a riskless bank account 𝐵.
We will define all stochastic processes on a fixed time horizon [0, 𝑇], where 𝑇 > 0.

First, fix a filtered probability space (Ω, ℱ, 𝔽, ℙ) and a Brownian motion 𝑊. We assume
that 𝔽 = 𝔽𝑊 and ℱ = ℱ𝑇 . In this model, the stochastic processes 𝑆 and 𝐵 satisfy the
following SDEs:

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇 𝑑𝑡 + 𝜎 𝑑𝑊𝑡 ), 𝑑𝐵𝑡 = 𝑟𝐵𝑡 𝑑𝑡,

where 𝜇 ∈ ℝ, 𝜎 > 0 and 𝑟 > 0 are constants. Here 𝑟 is the constant continuously
compounded risk-free rate.

Solving both SDEs, we obtain:

1
(𝜇− 𝜎2 )𝑡+𝜎 𝑊𝑡
𝑆𝑡 = 𝑆0 𝑒 2 , 𝐵𝑡 = 𝐵0 𝑒 𝑟𝑡 .

We now turn to the discounted assets (1, 𝑋), where 𝑋 = 𝑆/𝐵. By Ito's Lemma, the SDE for
𝑋 is

𝑑𝑋𝑡 = 𝑋𝑡 ((𝜇 − 𝑟) 𝑑𝑡 + 𝜎𝑑𝑊𝑡 ).

To find an ELMM for this model, we will need a powerful result known as Girsanov's
theorem. To motivate this result, let 𝑍 be a standard normal random variable
(𝑍~𝑁(0, 1)) on (Ω, ℱ, ℙ) and 𝜇 be a real number. Define a new probability measure ℙ∗ on
(Ω, ℱ) by

𝑑ℙ∗ 1
𝜇𝑍− 𝜇2
∶= 𝑒 2 .
𝑑ℙ∗

Then ℙ is a probability measure that is equivalent to ℙ. Furthermore, the moment-

generating function of 𝑍 under ℙ∗ is given by

1 2 1 2 1 2
𝔼∗ (𝑒 𝛼𝑍 ) = 𝔼 (𝑒 𝛼𝑍 𝑒 𝜇𝑍−2𝜇 ) = 𝑒 −2𝜇 𝔼 (𝑒 (𝛼+𝜇)𝑍 ) = 𝑒 𝛼𝜇+2𝛼 ,

which is the moment-generating function of a 𝑁(𝜇, 1) random variable. Thus, by

choosing 𝜇 we can create a probability measure that “shifts" the mean of 𝑍 (from 0 to 𝜇)
but keeps the variance the same.

Girsanov's Theorem does something similar with Brownian motions. We will let ‖𝑥‖
denote the Euclidean norm of a vector 𝑥 in ℝ𝑑 .

Theorem 1.1 (Girsanov's Theorem) Let (𝛺, ℱ, 𝔽, ℙ) be a filtered space and 𝑊 =

𝑊 1 , … , 𝑊 𝑑 ) be a 𝑑-dimensional Brownian motion on this space. Let 𝛾 = (𝛾1 , … , 𝛾𝑑 ) be a
vector of progressive processes satisfying

𝑇
2
ℙ (∫ (𝛾𝑠𝑖 ) 𝑑𝑠 < ∞) = 1, 𝑖 = 1, … , 𝑑.
0

Define the continuous local martingale 𝑍 by

 𝑑 𝑡
1 𝑡
𝑍𝑡 ≔ exp (∑ ∫ 𝛾𝑠𝑖 𝑑𝑊𝑠𝑖 − ∫ ‖𝛾𝑠 ‖2 𝑑𝑠 ).
0 2 0
𝑖=1

Assume that 𝑍 is a martingale and define a new probability measure ℙ∗by

𝑑ℙ∗
= 𝑍𝑇 .
𝑑ℙ

Then the process 𝑊

̃ =𝑊 ̃ 𝑑 ) defined by
̃ 1, … , 𝑊

𝑡
̃𝑡𝑖 ≔ 𝑊𝑡𝑖 − ∫ 𝛾𝑠𝑖 𝑑𝑠
𝑊
0

is a 𝑑-dimensional Brownian motion on (𝛺, ℱ, 𝔽, ℙ∗ ).

We can now apply Girsanov's theorem to find an ELMM for X above. Let γ be a one-
dimensional process such that

𝑇
ℙ (∫ 𝛾𝑠2 𝑑𝑠 < ∞) = 1.
0

Define 𝑍 by

𝑡
1 𝑡 2
𝑍𝑡 ≔ exp (∫ 𝛾𝑠 𝑑𝑊𝑠 − ∫ 𝛾𝑠 𝑑𝑠).
0 2 0

Then 𝑍 = {𝑍𝑡 : 0 ≤ 𝑡 ≤ 𝑇} is a UI martingale with 𝔼(𝑍𝑡 ) = 1 for every 0 ≤ 𝑡 ≤ 𝑇 and we can

therefore define a measure ℙ∗ on (Ω, ℱ) by

𝑑ℙ∗
= 𝑍𝑇 .
𝑑ℙ
Then the process 𝑊
̃ defined by

𝑡
̃𝑡 ≔ 𝑊𝑡 − ∫ 𝛾𝑠 𝑑𝑠
𝑊
0

is a Brownian motion with respect to ℙ∗. Now we can rewrite the SDE for 𝑋 in terms of
̃ as
𝑊

̃𝑡 + 𝜎𝛾𝑡 𝑑𝑡)
𝑑𝑋𝑡 = 𝑋𝑡 ((𝜇 − 𝑟) 𝑑𝑡 + 𝜎 𝑑𝑊𝑡 ) = 𝑋𝑡 ((𝜇 − 𝑟)𝑑𝑡 + 𝜎𝑑𝑊

̃𝑡 ).
= 𝑋𝑡 ((𝜇 − 𝑟 + 𝜎𝛾𝑡 )𝑑𝑡 + 𝜎𝑑𝑊

Hence, for 𝑋 to be a ℙ∗ local martingale, we need to choose

𝜇−𝑟
𝛾𝑡 = − .
𝜎

(𝜇−𝑟)
The quantity 𝜎
is often called the market price of risk.

𝜇−𝑟
So, let us pick 𝛾𝑡 = − . Then the SDE for 𝑋 is
𝜎

̃𝑡 ,
𝑑𝑋𝑡 = 𝑋𝑡 𝜎 𝑑𝑊

a local martingale under ℙ∗. (In fact, 𝑋 is a true martingale, so ℙ∗ is a martingale

measure.) We can therefore conclude that this model satisfies NFLVR by the first
fundamental theorem of asset pricing (FTAP I).

Now we move on to uniqueness of the measure. By the martingale representation of

Brownian motion, we observe that 𝑋 satisfies PRP with respect to ℙ∗, hence the market
is complete and the ELMM measure is unique.

Using these results, we can then price any contingent claim 𝐻 using the formula
 𝜋(𝐻) = 𝔼∗ (𝑒 −𝑟𝑇 𝐻).

We will go through some examples in the next sections.

Hi, in this video we introduce the Black-Scholes model and illustrate how to apply
Girsanov’s theorem to finding an ELMM in this model.

So, the one-dimensional Black-Scholes model says that the undiscounted stock price, 𝑆,
evolves according to the following stochastic differential equation:

𝑑𝑆𝑡 = 𝑆𝑡 (𝑀𝑑𝑡 + 𝜎𝑑𝑊𝑡 ).

𝑆
If we take the discounted stock price, which is 𝑒 𝑟𝑡𝑡 , where 𝑟 is the constant risk-free rate,

then the stochastic differential of 𝑋𝑡 will be equal to 𝑋𝑡 ((𝜇 − 𝑟) 𝑑𝑡 + 𝜎𝑑𝑊𝑡 ). So, we are
just removing 𝑟 from the drift itself. 𝑊 is the Brownian motion and we will assume that
the probability space contains this Brownian motion, 𝑊. Written in full:

𝑑𝑆𝑡 = 𝑆𝑡 (𝑀𝑑𝑡 + 𝜎𝑑𝑊𝑡 )

𝑆𝑡 𝑆𝑡
𝑋𝑡 = = 𝑟𝑡
𝐵𝑡 𝑒

𝑑𝑋𝑡 = 𝑋𝑡 ((𝜇 − 𝑟)𝑑𝑡 + 𝜎 𝑑𝑊𝑡 ).

Now, we want to find a ELMM for this model and we are going to apply Girsanov’s
theorem to do that.

Girsanov’s theorem says that if we have a new probability measure, let’s call it ℙ∗, that
is equivalent to ℙ and has the following Radon-Nikodym derivative, or density with
𝑇 1 𝑇
𝜃𝑠 𝑑𝑊𝑠 − ∫𝑂 𝜃𝑠2 𝑑𝑠
respect to ℙ, 𝑒 − ∫0 2 , then, this new stochastic process, 𝑊
̃𝑡 , which equals 𝑊𝑡
𝑡
plus ∫0 𝜃𝑠 𝑑𝑠, is a ℙ∗ Brownian motion. Written in full:
 𝑑ℙ∗ 𝑇 1 𝑇 2
= 𝑒 − ∫0 𝜃𝑆 𝑑𝑊𝑆 −2 ∫𝑂 𝜃𝑠 𝑑𝑠
𝑑ℙ
𝑡
̃𝑡 = 𝑊𝑡 + ∫ 𝜃𝑠 𝑑𝑠.
𝑊
0

So, 𝑊 itself, the original Brownian motion, need not be a Brownian motion under this
new probability measure, ℙ∗, but Girsanov’s theorem gives us a way of transforming the
old Brownian motion into a new Brownian motion, 𝑊
̃𝑡 .

Now, we want to try and choose this stochastic process, 𝜃, such that this new measure
that we get, ℙ∗, makes the discounted stock price, 𝑋, a local martingale. So, in other
words, it does not have any drift in its term.

So, let’s express this Brownian motion, 𝑊, in terms of 𝑊

̃𝑡 . We will get:

̃𝑡 − 𝜃𝑡 𝑑𝑡)),
𝑑𝑋𝑡 = 𝑋𝑡 ((𝜇 − 𝑟)𝑑𝑡 + 𝜎(𝑑𝑊

which simplifies to

̃𝑡 ).
𝑋𝑡 ((𝜇 − 𝑟 − 𝜎𝜃𝑡 )𝑑𝑡 + 𝜎𝑑𝑊

We want to choose 𝜃𝑡 so that the equation above is driftless and it is clear that we have
to make sure that the above is 0. So, we need 𝜇 − 𝑟 − 𝜎𝜃𝑡 = 0, which implies that 𝜃𝑡
𝜇−𝑟
should be equal to 𝜎
. In other words, it’s independent of 𝑡 itself.

Therefore, the Radon-Nikodym derivative corresponding to any ELMM is given by:

𝑑ℙ∗ 𝜇−𝑟 1 𝜇−𝑟 2

𝑊𝑇 − ( ) 𝑇
= 𝑒 𝜎 2 𝜎 ,
𝑑ℙ
after evaluating the corresponding integral. So, that is the Radon-Nikodym derivative
and, with that probability measure, the discounted stock price is a local martingale.
Hence, this is an ELMM. As we can see, it is unique — there is no other value of 𝜃𝑡 that
will make this a local martingale.

Now that we have looked at Girsanov’s theorem, in the next video, we are going to price
a call option in the Black-Scholes model.
In this section we illustrate how to price a call option and a put option using the Black-
Scholes model.
First, we consider a call option on 𝑆 with a strike price 𝐾. This is a derivative whose
payoff is

𝐻 = (𝑆𝑇 − 𝐾)+ .

Under the ELMM ℙ∗ , the dynamics of 𝑋 are

̃𝑡 ,
𝑑𝑋𝑡 = 𝑋𝑡 𝜎 𝑑𝑊

where 𝑊
̃ is a ℙ∗-Brownian motion. Hence the dynamics of 𝑆 are

̃𝑡 ).
𝑑𝑆𝑡 = 𝑆𝑡 (𝑟 𝑑𝑡 + 𝜎 𝑑𝑊

Solving this yields

0
1 2
𝑆𝑡 = 𝑆0 𝑒 (𝑟 − ̃𝑡 .
𝜎 ) 𝑡 + 𝜎𝑊
2

We now calculate the price of 𝐻.

1 +
(𝑟− 𝜎2 )𝑇+𝜎𝑊
̃𝑇
𝜋(𝐻) = 𝔼∗ (𝑒 −𝑟𝑇 (𝑆𝑇 − 𝐾)+ ) = 𝑒 −𝑟𝑇 𝔼∗ ((𝑆𝑇 − 𝐾)+ ) = 𝑒 −𝑟𝑇 𝔼∗ ((𝑆0 𝑒 2 − 𝐾) )

∞ +
1
(𝑟− 𝜎 2 )𝑇+𝜎√𝑇𝑧 1 1 2
= 𝑒 −𝑟𝑇 ∫ (𝑆0 𝑒 2 − 𝐾) 𝑒 − 2𝑧 𝑑𝑧,
−∞ √2𝜋
where we have used the fact that ℙ∗ 𝑊
̃ 𝑇 = ℙ√𝑇𝑧 , where 𝑍 ~ 𝑁(0, 1). To evaluate this
∗

integral, we note that the integrand is only non-zero when

1
(𝑟− 𝜎 2 )𝑇+𝜎√𝑇𝑧
𝑆0 𝑒 2 > 𝐾,

which is equivalent to

𝐾 1
ln (𝑆 ) − (𝑟 − 2 𝜎 2 ) 𝑇
0
𝑧= ∶= −𝑑2 .
𝜎√𝑇

Hence,

∞ 1
(𝑟− 𝜎2 )𝑇+𝜎√𝑇𝑧 1 1 2
𝜋(𝐻) = 𝑒 −𝑟𝑇
∫ (𝑆0 𝑒 2 − 𝐾) 𝑒 −2𝑧 𝑑𝑧
−𝑑2 √2𝜋
= 𝑆0 Φ(𝑑1 ) − 𝐾𝑒 −𝑟𝑇 Φ(𝑑2 )

after completing the square, where

𝑑1 ≔ 𝑑2 + 𝜎√𝑇

and Φ is (as usual) the standard normal CDF.

In general, if 0 ≤ 𝑡 ≤ 𝑇, then the price of 𝐻 at time 𝑡 is

𝔼∗ (𝑒 −𝑟(𝑇−𝑡) (𝑆𝑇 − 𝐾)+ |ℱ𝑡 ) = 𝑆𝑡 Φ(𝑑1 (𝑡)) − 𝐾𝑒 −𝑟(𝑇−𝑡) Φ(𝑑2 (𝑡)),

where

𝑆 1
ln( 𝑡 )+(𝑟+ 𝜎 2 )(𝑇−𝑡)
𝑑1 (𝑡) = 𝐾 2
𝜎 √𝑇−𝑡
and 𝑑2 (𝑡) = 𝑑1 (𝑡) − 𝜎√𝑇 − 𝑡.
Note that 𝑑𝑖 (0) = 𝑑𝑖 for 𝑖 = 1,2.

We now price a put option. For that, note that

𝑆𝑇 − 𝐾 𝑆𝑇 > 𝐾
(𝑆𝑇 − 𝐾)+ − (𝐾 − 𝑆𝑇 )+ = { = 𝑆𝑇 − 𝐾.
𝑆𝑇 − 𝐾 𝑆𝑇 ≤ 𝐾

Therefore, the put option price (𝜋𝑝) is related to the call option price (𝜋𝐶) by the
equation

𝜋𝐶 − 𝜋𝑃 = 𝔼∗ (𝑒 −𝑟𝑇 (𝑆𝑇 − 𝐾)) = 𝑆0 − 𝐾𝑒 −𝑟𝑇 .

This relationship is called the put-call parity.

Now let us consider some non-vanilla options. First we price a digital (binary) option
that pays 1 if the terminal stock price value exceeds a pre-specified threshold 𝐾 > 0
and 0 otherwise. This option is sometimes called a cash-or-nothing call and has a
payoff 𝐻𝐶 that can be represented as

𝐻𝐶 = 𝐼{𝑆𝑇 >𝐾} .

The price of this derivative is given by

𝔼∗ (𝑒 −𝑟𝑇 𝐼{𝑆𝑇 >𝐾} ) = 𝑒 −𝑟𝑇 ℙ∗ (𝑆𝑇 > 𝐾) = 𝑒 −𝑟𝑇 (1 − Φ(−𝑑2 )) = 𝑒 −𝑟𝑇 Φ(d2 ).

Now we consider an asset-or-nothing call, whose payoff is

𝐻𝐴 = 𝑆𝑇 𝐼{𝑆𝑇 >𝐾} .

This is a derivative whose payoff at maturity is equal to the value of the stock price
(𝑆𝑇 ) if the stock price is greater than 𝐾 and zero otherwise. Its price is given by
 𝔼∗(𝑒−𝑟𝑇 𝐻𝐴) = 𝑒−𝑟𝑇 𝔼∗ (𝑆𝑇 𝐼{𝑆𝑇>𝐾}) = 𝑆0Φ(𝑑1),

through a similar calculation to the call option price.

Note that the payoff of a call option 𝐻 is related to the digital options above by

𝐻 = 𝐻𝐴 − 𝐾𝐻𝐶 ,

hence the relationship between the prices too.

Finally, we mention that there are put option equivalents of these. Define them yourself
and find their prices.
Hi, in this video we derive the price of a call option in the Black-Scholes model.

Consider a European call option whose payoff, 𝐻, is given by (𝑆𝑇 − 𝐾)+ , so the positive
part of 𝑆𝑇 − 𝐾.

We are assuming the one-dimensional version of the Black-Scholes model, in the sense
that the stock price, under the real-world probability measure, evolves according to the
following stochastic differential equation when 𝑊 is a Brownian motion:

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇𝑑𝑡 + 𝜎𝑑𝑊𝑡 ).

In the last video, we showed how to find an ELMM, and, in that case, the stock price will
have the following stochastic differential equation:

̃𝑡 )
𝑑𝑆𝑡 = 𝑆𝑡 (𝑟𝑑𝑡 + 𝜎𝑑𝑊

where, 𝑊
̃ is a Brownian motion under the ELMM ℙ∗.

To calculate the price of 𝐻, we have to find the expected value under ℙ∗of the
discounted value of 𝐻, which is shown by:

1 +
(𝑟− 𝜎2 )𝑇+𝜎𝑊
̃𝑇
𝐸 ∗ = (𝑒 −𝑟𝑇 (𝑆0 𝑒 2 − 𝐾) ).

Now, the above is an expectation and the only random variable is 𝑊

̃ 𝑇 , which has a

normal distribution with mean 0 and variance 𝑇. So, we can express this in terms of
a standard normal random variable, which we write as:
 1 +
(𝑟− 𝜎2 )𝑇+𝜎√𝑇𝑧
𝑒 −𝑟𝑇 𝐸 ∗ ((𝑆0 𝑒 2 − 𝐾) )

where 𝑍 has a standard normal distribution.

This can be rewritten as:

∞ 1
(𝑟− 𝜎 2 )𝑇+𝜎√𝑇𝑧
𝑒 −𝑟𝑇 ∫ (𝑆0 𝑒 2 − 𝐾)+ ,
−∞

which we multiply by the density of 𝓏, (𝓏 is standard normal), so, written in full, that
will give us:

∞ 𝑧2
−𝑟𝑇
1
(𝑟− 𝜎 2 )𝑇+𝜎√𝑇𝑧 +
1 −
𝑒 ∫ (𝑆0 𝑒 2 − 𝐾) 𝑒 2 𝑑𝓏.
−∞ √2𝜋

To evaluate this expectation, we must just find the region where this is positive,
because this is 0 when 𝐾 is greater than this quantity here, and evaluate the integral
over that region, which, as shown in the notes, gives us:

𝑆0 ∅(𝑑1 ) − 𝑒 𝑟𝑇 𝐾∅(𝑑2 ).

In the above equation:

• ∅ is the CDF of a standard normal,
2
𝑆 1𝜎
𝐼𝑛 ( 0 )+(𝑟+ )𝑇
𝐾 2
• 𝑑1 = 𝜎√𝑇
, and

• 𝑑2 = 𝑑1 − 𝜎√𝑇.

So, that’s the expression for 𝑑1 and 𝑑2 , which gives us the price of the call option at time
0.
Now that we have priced a call option, in the next video, we are going to look at how to
price a digital option.
In this section we derive the hedging formula in the Black-Scholes model and also
derive the Black-Scholes PDE.

Consider an option whose payoff 𝐻 is of the form 𝐻 = ℎ(𝑆𝑇 ) for some Borel measurable
function ℎ: ℝ → ℝ. We want to find a trading strategy (𝑣0 , 𝜑) such that 𝑉𝑇 ((𝑣0 ; 𝜑)) = 𝐻.

We will first find a trading strategy in the discounted assets (1, 𝑋) that replicates the
discounted derivative 𝐻
̃ ∶= 𝑒 −𝑟𝑇 𝐻 = 𝑒 −𝑟𝑇 ℎ(𝑒 𝑟𝑇 𝑋𝑇 ) = 𝑔(𝑋𝑇 ) and then show that the same

strategy applied to the original assets (𝐵, 𝑆) replicates 𝐻.

Consider the martingale 𝑀 = {𝑀𝑡 ∶ 0 ≤ 𝑡 ≤ 𝑇} defined by 𝑀𝑡 ∶= 𝔼∗ (𝑔(𝑋𝑇 )|ℱ𝑡 ). Since 𝑋 is

a Markov process, we have

𝑀𝑡 = 𝔼∗ (𝑔(𝑋𝑇 )|ℱ𝑡 ) = 𝔼∗ (𝑔(𝑋𝑇 )|𝑋𝑡 ),

hence by the Doob-Dynkin theorem, there exists a function 𝐹 ∶ [0, ∞) × ℝ → ℝ such that

𝑀𝑡 = 𝐹(𝑡, 𝑋𝑡 ).

By Ito’s Lemma, we have

𝜕𝐹 𝜕𝐹 1 𝜕2𝐹 𝜕𝐹 1 2 2 𝜕 2 𝐹 𝜕𝐹
𝑑𝑀𝑡 = 𝑑𝑡 + 𝑑𝑋𝑡 + 𝑑〈𝑋〉𝑡 = ( + 𝜎 𝑋𝑡 ) 𝑑𝑡 + 𝜎𝑋𝑡 𝑑𝑊𝑡 ,
𝜕𝑡 𝜕𝑥 2 𝜕𝑥 2 𝜕𝑡 2 𝜕𝑥 2 𝜕𝑥

hence
 𝜕𝐹 1 2 2 𝜕 2 𝐹
+ 𝜎 𝑋𝑡 =0
𝜕𝑡 2 𝜕𝑥 2

since 𝑀 is a martingale.

Thus, since 𝑀𝑇 = 𝐹(𝑇, 𝑋𝑇 ) = 𝑔(𝑋𝑇 ) = 𝐻

̃ , we have

̃ ) + ∫𝑇 𝜕𝐹 𝑑𝑋𝑡 ,
̃ = 𝔼∗ (𝐻
𝐻 0 𝜕𝑥

which means that (𝔼∗ (𝐻

̃ ), 𝜑) where

𝜕𝐹
𝜑𝑡 ≔ (𝑡, 𝑋𝑡 )
𝜕𝑥

is a replicating strategy for 𝐻

̃ . The corresponding holding in the riskless asset 𝐵 is

𝜂𝑡 = 𝐹(𝑡, 𝑋𝑡 ) − 𝜑𝑡 𝑋𝑡 .

Now we show that the same strategy − applied to (𝐵, 𝑆) instead of (1, 𝑋) − replicates
𝐻 as well. Let 𝑉 be the value of this strategy. Then

̃ 𝐵𝑇 = 𝐻.
𝑉𝑇 = 𝜂 𝑇 𝐵𝑇 + 𝜑𝑇 𝑆𝑇 = (𝐹(𝑇, 𝑋𝑇 ) − 𝜑𝑇 𝑋𝑇 ) 𝐵𝑇 + 𝜑𝑇 𝑆𝑇 = 𝐵𝑇 𝐹(𝑇, 𝑋𝑇 ) = 𝐻

It also follows that 𝑒 −𝑟𝑡 𝑉𝑡 = 𝐹(𝑡, 𝑋𝑡 ), hence 𝑉𝑡 = 𝑒 𝑟𝑡 𝐹(𝑡, 𝑋𝑡 ) = 𝑒 𝑟𝑡 𝐹(𝑡, 𝑆𝑡 𝑒 −𝑟𝑡 ) =: 𝑉(𝑡, 𝑆𝑡 ) for
some function 𝑉: [0, ∞) × ℝ → ℝ. Using the chain rule we see that

𝜕𝐹 𝜕𝑉
= .
𝜕𝑋𝑡 𝜕𝑆𝑡

Hence, since 𝐹 satisfies the PDE

 𝜕𝐹 1 2 2 𝜕 2 𝐹
+ 𝜎 𝑋𝑡 = 0,
𝜕𝑡 2 𝜕𝑋𝑡2

then 𝑉(𝑡, 𝑆𝑡 ) satisfies the PDE

𝜕𝑉 1 2 2 𝜕 2 𝑉 𝜕𝑉
+ 𝜎 𝑆𝑡 2 + 𝑟𝑆𝑡 − 𝑟𝑉(𝑡, 𝑆𝑡 ) = 0,
𝜕𝑡 2 𝜕𝑆𝑡 𝜕𝑆𝑡

together with the boundary condition 𝑉(𝑇, 𝑆𝑇 ) = ℎ(𝑆𝑇 ). This equation is the celebrated
Black-Scholes PDE.

Let 𝐻 be a call option with strike 𝐾. We know that for this derivative,

𝑉(𝑡, 𝑆𝑇 ) = 𝑆𝑡 Φ(𝑑1 (𝑡)) − 𝐾𝑒 −𝑟(𝑇−𝑡) Φ(𝑑2 (𝑡)),

where

𝑆 1
ln (𝐾𝑡 ) + (𝑟 + 2 𝜎 2 ) (𝑇 − 𝑡)
𝑑1 (𝑡) = and 𝑑2 (𝑡) = 𝑑1(𝑡) − 𝜎√𝑇 − 𝑡 .
𝜎√𝑇 − 𝑡

The hedging strategy 𝜑 is given by

𝜕𝑉
𝜑𝑡 = = Φ(𝑑1 (𝑡)).
𝜕𝑆𝑡

Check this as an exercise.

Hi, in this video we introduce a digital option and show how to price it in the Black-
Scholes model.

A digital option is a derivative whose payoff is of the following form: 𝐻 = 𝐼{𝑆𝑇 >𝐾} . So it

pays the value 1 if 𝑆𝑇 is greater than 𝐾 and 0 otherwise. This is the indicator.

We can draw it in a diagram, where the x-axis is 𝑆𝑇 and the y-axis is 𝐻. When 𝑆𝑇 is less
than 𝐾, the option pays the value 0 and, as soon as 𝑆𝑇 is greater than 𝐾, it pays the value
1. In comparison to a call option on the other hand, it pays the value 0 when 𝑆𝑇 is less
than 𝐾 and it pays 𝑆𝑇 minus 𝐾 when 𝑆𝑇 is greater than 𝐾.

To find the price of 𝐻, we have to find the expected value under a risk-neutral measure
of the discounted derivative 𝐻; or the discounted payoff of 𝐻, which is equal to:

𝐸 ∗ (𝑒 −𝑟𝑇 𝐻) = 𝑒 −𝑟𝑇 𝐸 ∗ (𝐻).

We can simplify this to: 𝑒 −𝑟𝑇 𝐸 ∗ (𝐼{𝑆𝑇 >𝐾} ) under the risk-neutral measure.

Note that the expected value of an indicator is just the probability of the set itself. So,
the risk-neutral probability when 𝑆𝑇 is greater than 𝐾 is equal to:

𝑒 −𝑟𝑇 ℙ∗ (𝑆𝑇 > 𝐾).

Now, 𝑆𝑇 is equal to 𝑆0 under the risk-neutral measure, as follows:

 1
(𝑟− 𝜎2 )𝑇+𝜎𝑊
̃𝑇
𝑆𝑇 = 𝑆0 𝑒 2

Therefore, if you substitute that, we get:

1
(𝑟− 𝜎2 )𝑇+𝜎𝑊
̃𝑇
𝑆𝑇 > 𝐾 ⟺ 𝑆0 𝑒 2 > 𝐾,

which is equal to, taking the second part of the equation, dividing by 𝑆0 and taking the
logarithms,

1 𝐾
̃ 𝑇 > 𝐼𝑛 ( ).
(𝑟 − 𝜎 2 ) 𝑇 + 𝜎𝑊
2 𝑆0

This can be further simplified to:

𝐾 1
̃ 𝑇 > 𝐼𝑛 ( ) − (𝑟 − 𝜎 2 ) 𝑇.
𝜎𝑊
𝑆0 2

We then divide by 𝜎 root 𝑇, because this is a normal random variable. So, if we replace
this with this, we get:

𝐾 1
𝐼𝑛 (𝑆 ) − (𝑟 − 2 𝜎 2 ) 𝑇
0
ℙ∗ (𝑆𝑇 > 𝐾) = ℙ∗ (𝑍 > ),
𝜎√𝑇

because of the standard deviation of the beginning point.

This is equal to:

𝐾 1
𝐼𝑛 (𝑆 ) − (𝑟 − 2 𝜎 2 ) 𝑇
0
1 − ℙ∗ (𝑍 < ).
𝜎√𝑇
And, using the properties of the standard normal distribution, we see that this is equal
to:

1 − ∅(−𝑑2 ) = ∅(𝑑2 ),

where 𝑑2 was calculated in the previous video.

Now that we have looked at the price of a digital option, in the next video we’re going to
look at the general Black-Scholes model.
We now extend the Black-Scholes model to multiple assets.
Consider 𝑑 > 1 assets 𝑆 = (𝑆 1 , … , 𝑆 𝑑 ) whose prices evolve according to the following

SDEs:

𝑚
𝑗
𝑑𝑆𝑡𝑖 = 𝑆𝑡𝑖 (𝜇𝑖 𝑑𝑡 + ∑ 𝜎𝑖𝑗 𝑑𝑊𝑡 ) 𝑖 = 1, … , 𝑑,
𝑗=1

where 𝜇𝑖 , 𝜎𝑖𝑗 are constants and 𝑊 = (𝑊 1 , … , 𝑊 𝑚 ) is an 𝑚-dimensional Brownian motion

process (〈𝑊 𝑖 , 𝑊𝑗 〉𝑡 = 𝛿𝑖𝑗 𝑡).

These SDEs can also be written succinctly as

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇 𝑑𝑡 + 𝜎 𝑑𝑊𝑡 )

for appropriately defined matrices 𝜇 and 𝜎.

We will assume that 𝔽 = 𝔽𝑊 and ℱ = ℱ𝑇 .

We now attempt to find an ELMM in this model by applying Girsanov's theorem. Let 𝛾 =
(𝛾1 , … , 𝛾 𝑚 ) be a 𝑊-integrable vector process so that the positive local martingale

𝜀((𝛾⦁𝑊))is a true martingale. Define the measure ℙ∗ by

𝑚
𝑑ℙ∗ 𝑇
1 𝑇
∶= 𝜀((𝛾⦁𝑊))𝑇 = exp (∑ ∫ 𝛾𝑠𝑖 𝑑𝑊𝑠𝑖 − ∫ ‖𝛾𝑠 ‖2 𝑑𝑠).
𝑑ℙ 0 2 0
𝑖=1
Then 𝑊
̃ = (𝑊 ̃ 𝑚 ) is an 𝑚-dimensional Brownian motion with respect to ℙ∗, where
̃ 1, … , 𝑊

𝑡
̃𝑡𝑖 ≔ 𝑊𝑡𝑖 − ∫ 𝛾𝑠𝑖 𝑑𝑠,
𝑊 𝑖 = 1, … , 𝑚.
0

Substituting these equations to the SDEs for the discounted asset 𝑋𝑡 = 𝑒 −𝑟𝑡 𝑆𝑡 , we get

𝑚 𝑚 𝑚
𝑗 𝑗 𝑗
̃𝑡 ).
𝑑𝑋𝑡𝑖 = 𝑋𝑡𝑖 ((𝜇𝑖 − 𝑟) + ∑ 𝜎𝑖𝑗 𝑑𝑊𝑡 ) = 𝑆𝑡𝑖 ((𝜇𝑖 − 𝑟 + ∑ 𝛾𝑡 𝜎𝑖𝑗 ) 𝑑𝑡 + ∑ 𝜎𝑖𝑗 𝑑𝑊
𝑗=1 𝑗=1 𝑗=1

So, for ℙ∗ to be an ELMM, we need to choose 𝛾 so that

𝑚
𝑗
𝜇𝑖 − 𝑟 + ∑ 𝛾𝑡 𝜎𝑖𝑗 = 0 𝑖 = 1, … , 𝑑.
𝑗=1

This system of equations can have no solutions, a unique solution, or infinitely many
solutions, depending on the relationship between 𝑚, 𝑑, 𝑟, 𝜎𝑖𝑗 , and 𝜇𝑖 .

Let us look at a concrete example. Consider two stocks 𝑆 and 𝑈 whose prices have the
following dynamics:

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇𝑆 𝑑𝑡 + 𝜎𝑆 (𝜌𝑑𝑊𝑡1 + √1 − 𝜌2 𝑑𝑊𝑡2 ))

and

𝑑𝑈𝑡 = 𝑈𝑡 (𝜇𝑈 𝑑𝑡 + 𝜎𝑈 𝑑𝑊𝑡1 ),

where 𝑊 1 and 𝑊 2 are independent Brownian motion processes and 𝜇𝑆 , 𝜇𝑈 , 𝜎𝑆 , 𝜎𝑈 , 𝜌 are

all constants with |𝜌| < 1. Define a new process 𝑊 3 by
 𝑡 𝑡
𝑊𝑡3 ≔ ∫ 𝜌𝑑𝑊𝑠1 + ∫ √1 − 𝜌2 𝑑𝑊𝑠2 = 𝜌𝑊𝑡1 + √1 − 𝜌2 𝑊𝑡2 .
0 0

Then 𝑊 3 is also a Brownian motion process, with 〈𝑊 3 , 𝑊 1 〉𝑡 = 𝜌𝑡, and the pair of SDEs
can be rewritten as

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇𝑆 𝑑𝑡 + 𝜎𝑆 𝑑𝑊𝑡3 ) and 𝑑𝑈𝑡 = 𝑈𝑡 (𝜇𝑈 𝑑𝑡 + 𝜎𝑈 𝑑𝑊𝑡1 ).

To find an ELMM for (𝑆, 𝑈), we have to solve the following system of equations

𝜇𝑆 − 𝑟 + 𝛾𝑡1 𝜎𝑆 𝜌 + 𝛾𝑡2 𝜎𝑆 √1 − 𝜌2 = 0
𝜇𝑈 − 𝑟 + 𝛾𝑡1 𝜎𝑈 = 0

for 𝛾𝑡1 and 𝛾𝑡2 , which can easily be seen to have a unique solution.

Under the unique ELMM ℙ∗, the dynamics of 𝑆 and 𝑈 are given by

̃𝑡1 + √1 − 𝜌2 𝑑𝑊
𝑑𝑆𝑡 = 𝑆𝑡 (𝑟 𝑑𝑡 + 𝜎𝑆 (𝜌𝑑𝑊 ̃𝑡2 ))

and

̃𝑡1 ),
𝑑𝑈𝑡 = 𝑈𝑡 (𝑟 𝑑𝑡 + 𝜎𝑈 𝑑𝑊

Where 𝑊
̃ 𝑖 𝑖 = 1,2 are ℙ∗ Brownian motions.

We again define a third ℙ∗ Brownian motion 𝑊

̃ 3 as

𝑡 𝑡
̃𝑡3 ≔ ∫ 𝜌 𝑑𝑊
𝑊 ̃𝑠1 + ∫ √1 − 𝜌2 𝑑𝑊
̃𝑠2 = 𝜌𝑊
̃𝑡1 + √1 − 𝜌2 𝑊
̃𝑡2 ,
0 0

and rewrite the SDE for 𝑆 as

 ̃𝑡3 ).
𝑑𝑆𝑡 = 𝑆𝑡 (𝑟 𝑑𝑡 + 𝜎𝑆 𝑑𝑊

Now, let us price options on this model. First, options that depend only on one asset can
be priced using exactly the same formulae from the one-dimensional model. We
consider the following exchange option whose payoff 𝐻 is

𝐻 = max {𝑆𝑇 − 𝑈𝑇 , 0}.

This is an option to exchange one asset for another at maturity. To price this option, we
first solve the SDEs for 𝑆 and 𝑈 to get

1 1 2
(𝑟− 𝜎𝑆2 )𝑇+𝜎𝑆 𝑊
̃𝑇3 (𝑟− 𝜎𝑈 ̃𝑇1
)𝑇+𝜎𝑈 𝑊
𝑆𝑇 = 𝑆0 𝑒 2 𝑎𝑛𝑑 𝑈𝑇 = 𝑈0 𝑒 2 .

𝑆
Then by writing 𝐻 = max {𝑈𝑇 − 1,0} and considering the process 𝑉 = 𝑆/𝑈, one can use a
𝑇

modified version of the one-dimensional Black-Scholes call option pricing formula

derived in the previous sections to get

𝜋(𝐻) = 𝑆0 Φ(𝑑1 ) − 𝑈0 Φ(𝑑2 ),

where

𝑆 1
𝑙𝑛 (𝑈0 ) + 2 𝜎 2 𝑇
0
𝑑1 = , 𝑑2 = 𝑑1 − 𝜎√𝑇 𝑎𝑛𝑑 𝜎 ≔ √𝜎𝑆2 + 𝜎𝑈2 − 2𝜌𝜎𝑆 𝜎𝑈 .
𝜎√𝑇
Hi, in this video we introduce a multidimensional version of Black-Scholes model.

Recall that the one-dimensional Black-Scholes model says that the stock price evolves
according to the following geometric Brownian motion:

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇𝑑𝑡 + 𝜎𝑑𝑊𝑡 )

where 𝜇 and 𝜎 are constants and 𝑊 is a Brownian motion.

We now want to generalize this to the case where, instead of one stock or risky asset, 𝑠,
we have 𝑑 of them, where 𝑑 is finite but greater than or equal to 2, in this case. Written
in full:

𝑠(𝑠1 , … , 𝑠 𝑑 ) 𝑑 ≥ 2.

The Black-Scholes analog of that model still uses something similar to geometric
Brownian motion, then. What we need is a vector of Brownian motion. In other words,
we need 𝑚-dimensional Brownian motion processes from 𝑊 1 up to 𝑊 𝑚 , which are
independent. The covariation between them is as follows:

〈𝑊 𝑖 , 𝑊𝑗 〉𝑡 = 𝛿𝑖𝑗𝑡 .

This means that it is 𝑡 if 𝑖 is equal to 𝑗, and 0 otherwise.

The multidimensional Black-Scholes model says that the stock prices of 𝑆 1 up

to 𝑆 𝑑 all evolve according to the following SDEs: 𝑑𝑆𝑡𝑖 is equal to 𝑆𝑡𝑖 times 𝜇𝑖 𝑑𝑡 −
where each stock has its own drift − plus, and then in volatility terms, just a
combination of the Brownian motion processes, which are the sum from 𝑗 = 1 up to
𝑚 of 𝜎𝑖𝑗 𝑑𝑊𝑗 𝑡 . That is the stochastic differential equation when 𝑖 = 1 up to 𝑑. Written in

full:

𝑚
𝑖 𝑖
𝑑𝑆𝑡 = 𝑆𝑡 (𝜇𝑖 𝑑𝑡 + ∑ 𝜎𝑖𝑗 𝑑𝑊𝑗 𝑡 ).
𝑗=1

We normally write this in vector notation as:

𝑑𝑆𝑡 = 𝑆𝑡 𝜇𝑑𝑡 + 𝜎𝑑𝑊𝑡 ,

where these are now matrices instead of vectors themselves.

This model is guaranteed to have an ELMM if 𝑚 is greater than or equal to 𝑑. There are
cases when 𝑚 is less than 𝑑 and there is no ELMM.

The model is complete − in other words, there is only one ELMM that is guaranteed if
𝑚 = 𝑑. Of course, there are cases where this doesn’t hold and the model is still
complete, but we won’t explore those cases.
Consider the BS model with 𝜇 = 0.1, 𝑟 = 0.06, and 𝜎 = 0.25. Then what is 𝐵12 equal to?

Solution:
First, fix a filtered probability space (Ω, ℱ, 𝔽, ℙ) and a Brownian motion 𝑊. We assume
that 𝔽 = 𝔽𝑊 and ℱ = ℱ𝑇 . In this model, the stochastic processes 𝑆 and 𝐵 satisfy the
following SDEs:

𝑑𝑆𝑡 = 𝑆𝑡 (𝜇 𝑑𝑡 + 𝜎𝑑𝑊𝑡 ), 𝑑𝐵𝑡 = 𝑟𝐵𝑡 𝑑𝑡,

where 𝜇 ∈ ℝ, 𝜎 > 0, and 𝑟 > 0 are constants. Here 𝑟 is the constant continuously
compounded risk-free rate.

Solving both SDEs, we obtain:

1
(𝜇− 𝜎2 )𝑡+𝜎𝑊𝑡
𝑆𝑡 = 𝑆0 𝑒 2 , 𝐵𝑡 = 𝐵0 𝑒 𝑟𝑡 .

Thus, we just have to substitute in 𝐵𝑡 as follows:

𝐵12 = 𝐵0 𝑒 𝑟∗12 = 𝐵0 𝑒 0.06∗12 = 𝐵0 𝑒 0.72 .

Consider the BS model with 𝑆0 = 100, 𝜇 = 0.1, 𝑟 = 0.06, 𝑇 = 1, and 𝜎 = 0.25. What is
the price of a call option with strike price 𝐾 = 110?
Solution:
Following the lecture notes, the call price is defined as:

𝔼∗ (𝑒 −𝑟(𝑇−𝑡) (𝑆𝑇 − 𝐾)+ |ℱ𝑡 ) = 𝑆𝑡 Φ(𝑑1 (𝑡)) − 𝐾𝑒 −𝑟(𝑇−𝑡) Φ(𝑑2 (𝑡)),

where

𝑆 1
ln ( 𝐾𝑡 ) + (𝑟 + 2 𝜎 2 ) (𝑇 − 𝑡)
𝑑1 (𝑡) = and 𝑑2 (𝑡) = 𝑑1 (𝑡) − 𝜎√𝑇 − 𝑡.
𝜎√𝑇 − 𝑡

Note that 𝑑𝑖 (0) = 𝑑𝑖 for 𝑖 = 1, 2. Applying the above equations to our case,

100 1
𝑙𝑛 (110) + (0.06 + 2 0.252 ) (1)
𝑑1 (𝑡) = = −0.0162407 and 𝑑2 (𝑡) = 𝑑1 (𝑡) − 0.25√1
0.25√1
= −0.2662407.

Finally, we apply (for 𝑡 = 0)

𝔼∗ (𝑒 −𝑟(𝑇−𝑡) (𝑆𝑇 − 𝐾)+ |ℱ𝑡 ) = 𝑆𝑡 Φ(𝑑1 (𝑡)) − 𝐾𝑒 −𝑟(𝑇−𝑡) Φ(𝑑2 (𝑡)) = 8.42966.

Consider the BS model with 𝑆0.5 = 112, 𝜇 = 0.2, 𝑟 = 0.04, 𝑇 = 1, and 𝜎 = 0.30. Consider a
call option with a strike price of 𝐾 = 100 and its corresponding hedging strategy
(𝑣0 , 𝜑). Then what is 𝜑0.5 equal to?

Solution:
Let 𝐻 be a call option with strike 𝐾. We know that for this derivative,
 𝑉(𝑡, 𝑆𝑡 ) = 𝑆𝑡 Φ(𝑑1 (𝑡)) − 𝐾𝑒 −𝑟(𝑇−𝑡) 𝜎(𝑑2 (𝑡)),

where
𝑆 1
ln ( 𝑡 ) + (𝑟 + 𝜎 2 ) (𝑇 − 𝑡)
𝑑1 (𝑡) = 𝐾 2 and 𝑑2 (𝑡) − 𝜎√𝑇 − 𝑡.
𝜎√𝑇 − 𝑡

The hedging strategy 𝜑 is given by

𝜕𝑉
𝜑𝑡 = = Φ(𝑑1 (𝑡)).
𝜕𝑆𝑡

In our case, 𝑑1 is equal to:

𝑆 1 112 1
In ( 𝐾𝑡 ) + (𝑟 + 2 𝜎 2 ) (𝑇 − 𝑡) In (100) + (0.04 + 2 0.32 ) (0.5)
𝑑1 (0.5) = = = 0.734583,
𝜎√𝑇 − 𝑡 0.3√0.5

and

𝜕𝑉
𝜑0.5 = = Φ(𝑑1 (0.5)) = 0.76870.
𝜕𝑆𝑡

Consider the BS model with 𝑆0.5 = 108, 𝜇 = 0.1, 𝑟 = 0.06, 𝑇 = 1, and 𝜎 = 0.25. What is the

price of an asset-or-nothing call option with strike price 𝐾 = 110 at time 0.5?

Solution:
This is a derivative whose payoff at maturity is equal to the value of the stock price (𝑆𝑇 )

if the stock price is greater than 𝐾 and zero otherwise. Its price is given by

𝐶 = 𝑆0 ∗ Φ(𝑑1),
where

𝑆 1 118 1
ln ( 𝑡 )+(𝑟+ 𝜎 2 )(𝑇−𝑡) ln ( )+(0.06+ 0.252 )(0.5)
𝐾 2 110 2
𝑑1 (0.5) = 𝜎 √𝑇−𝑡
= 0.25√0.5
= 0.154296,

and

𝐶0.5 = 𝑆0.5 ∗Φ(𝑑1) = 108 ∗ Φ(0.154296) = 60.6217.

Consider the BS model with 𝑆0 = 120, 𝜇 = 0.2, 𝑟 = 0,04, 𝑇 = 1, and 𝜎 = 0.30. Compute the
price of a derivative with payoff 𝐻 = max {𝑆𝑇 , 80}.

Solution:
First of all, we have to notice that the payoff 𝐻 = max {𝑆𝑇 . 80} is equivalent to:

𝐻 = max { 𝑆𝑇 , 80} = 80 + max {𝑆𝑇 − 80,0},

where the payoff max {𝑆𝑇 − 80,0} represent a call option with strike 80. The call option

price today is equal to (applying Black-Scholes): 44. Therefore, the price of the payoff
should be equal to:

𝑃𝑟𝑖𝑐𝑒 = 80 ∗ 𝑒 −0.04∗1 + 44 = 120.86,

which is the solution that we were looking for.