Financial Mathematics Research Project Report

Lev Udaltsov
115368956
Supervisor - The Ghost of George Boole
Coordinator - Rachel Riley

Could I borrow a euro for the locker please?

A criticism of cash based locker room
systems in a contact-less world

School of Mathematical Sciences

University College Cork
Ireland
27th March 2019
 Abstract

Option pricing is a well covered area, with many approaches available. One of the best
known methods for pricing of options was proposed by Cox, Ross, and Rubenstein in 1979:
the binomial tree model. Their approach works well for both European options and American
options, but has to be tweaked for American options, to allow for the possibility of early exercise.
After reviewing the pricing of European options on binomial tree models, this project proceeds
with studying the pricing of American options, and provides a method for visualising how the
value of the option evolves within the model.

Contents

1 Introduction 2

2 Notation and Assumptions 3

3 Options 4
3.1 Review of Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2 European and American options . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4 The Binomial Tree Model 6

4.1 Intuition behind the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.2 1-Step Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.3 2-Step and n-Step model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.4 Pricing put options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.5 Time step refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.6 Comparison to continuous random walk . . . . . . . . . . . . . . . . . . . . . . . 18

5 Adjusting for dividends 22

6 Advantages and Limitations 23

7 Calculation of Prices 25

8 Visualisation of Binomial Tree 27

9 Conclusion 31

1
1 Introduction

Options are a financial instrument, similar to futures contracts, which have been officially traded
since 1973. They are sold by an option writer to an option buyer. The buyer then has the right, not
the obligation, to buy or sell the underlying asset at a pre-determined price at a specific date.
The approach to pricing of options underwent a massive change in 1973 when Fischer Black and
Myron Scholes presented the first completely satisfactory equilibrium option pricing model. These
"path-breaking" articles have laid the foundation for many academic studies that followed [2]. The
Black-Scholes model, however, is extremely mathematically involved as it is derived from a partial
differential equation known as the Black-Scholes equation. It is frequently used by option market
participants, but as a "black box" calculator - with known inputs and outputs, with little regard for
the mathematics behind the process.
In their paper, Cox, Ross and Rubenstein [2] put forward a much more intuitive and simple
approach to the pricing of options. They originally got the idea from William Sharpe, who introduced
the approach in his book Investments (1978) [13]. Their approach involves breaking down the
stock price movements to a discrete binomial process, and pricing the option using simple arbitrage
arguments.
This paper seeks to explore how the binomial method can be used to price American Options, and
compares this approach to alternative option pricing methods. The existing literature is combined
into a concise overview of how the method works, with different variations discussed and explained.
After doing this, it is shown how the process can be represented visually through binomial trees. This
visualisation is used to highlight the benefits of the model. The visualisation is implemented using
Python, with code that I have developed as part of the project, using calculations and formulations
proposed by D. Higham [4].

2
2 Notation and Assumptions

Imagine a market where two types of asset can be traded: one risk-free security, such as a bank
deposit, and one risky security, such as a stock. We will denote the value of a risk-free security A
at time t by A(t). Similarly, the value of a risky security S at time t will be given by S(t). At any
given time t, the value of the securities, at that time, will be known. At any time T > t, S(T) will be
unknown, while the value A(T) will be known, as guaranteed by the bank.
Consider now a portfolio Π that contains x risky securities, and y risk-free securities. The value
of Π at time t can be written as
Π(t) = xS(t) + y A(t). (1)

With this portfolio in mind, we can establish the main assumption that will be the primary
driving force behind our arguments for the model used; The No-Arbitrage Principle: There is no
portfolio Π with initial value Π(0) = 0 such that Π(T) ≥ 0 with probability 1 and Π(T) > 0 with
non-zero probability, for any time T > 0.
In other words, if the initial value of a portfolio is zero, Π(0) = 0, and Π(T) ≥ 0, then Π(T) = 0
with probability 1. It means an investor cannot obtain a risk-less profit on any portfolio that has no
initial investment [1]. There are other ways of formulating the No-Arbitrage Principle, but this is
the one we will focus on as it will be used to establish relationships between stock market price
fluctuations and the risk-free interest rate. It will also be used to make arguments for when it is or is
not profitable to exercise an option.
The aforementioned risk-free security A(t) will also have a key assumption relating to it, which
is the following:
The same constant interest rate r will apply for lending and borrowing money without risk, and
continuous compounding will be used.
In terms of investing in the risk-free asset, it means that a sum of money y invested for a length
of time T will return a sum of yerT . This is an essential point to keep in mind as it will be used in
conjunction with The No-Arbitrage principle to dictate how rational investors will act, and how they

3
will seek out profit from their trades and transactions.

3 Options

3.1 Review of Options

An option is the right, but not the obligation, to buy or sell an asset, called the underlying asset
on, or before, a specified future time T for price K, fixed in advance, known as the strike price or
exercise price. There are various types of options out there, but the two most popular options are
call and put options. There are various underlying securities based on which options can be written.
These include stock options, index options, Forex (currency) options, and futures options, but they
all operate on the same basic principle of the buyer/seller essentially making a bet on how they
believe the market will be different in the future. We will focus primarily on stock options.
A call option is a contract whereby the holder is given the right to buy the underlying asset at an
agreed upon price in the future. An investor would purchase such an option when they believe that
the price of the security is likely to rise in the future.
A put option is a contract whereby the holder is given the right to sell the underlying security at
an agreed upon price in the future. It can be viewed as the opposite of a call option, as an investor
would purchase it when they believe that the price of the security is likely to fall in the future.
Suppose that we let C(T) stand for the price of a call option at time T, and let P(T) stand for the
price of a put option at time T. Their payoff functions are defined as;

C(T) = max(S(T) − K, 0),

(2)
P(T) = max(K − S(T), 0).

where S(T) is the value of the security at time T, and K is the agreed upon price that the contract
was written for, which is called the strike price.
The payoffs in both cases are random variables dependent on S(T) which is the price of the
underlying asset at time T.

4
3.2 European and American options

The key difference between American and European options lies in when the options can be exercised.
A European style option can only be exercised at the expiry date, T, while an American option can
be exercised at any time before, or on, the expiry date, t ≤ T. For the sake of clarity, from now on
European call and put options will be referred to as CE and PE respectively, while American call
and put options will be referred to as C A and PA.
Intuitively, we can establish the following inequalities;

CE ≤ C A, PE ≤ PA . (3)

This is because for European and American options with the same strike price K and the same expiry
date T, the American option gives the holder at least the same exercise rights as the European option.
For example, consider a European call option written on a certain stock, which fluctuates greatly
in price. It may happen that S(t) will be much greater than K at a certain time before the expiry
date, t ≤ T. With an American call option on the same stock, the holder would have the ability to
exercise the option at that time t, for price K, gaining a profit of S(t) − K. The same argument can
be made for put options, when the stock price S(t) goes much lower than the strike price K.
Although intuition says that CE ≤ C A, it is actually the case that CE = C A for an American and a
European call option, that doesn’t pay dividends, with the same strike price K and expiry date T.
Suppose that CE > C A: could an investor use this to make an arbitrage profit? The answer is yes.
One could sell an American call and buy a European call, investing the difference - C A − CE at the
interest rate r. If the American call is exercised at time t ≤ T, borrow a share and sell it for K to
settle your obligation as the writer of the option, investing the sum received K at the interest rate r.
Then, at time T, you can the European call you purchased to buy a share for K and close your short
position in that stock. Your net position at the end of the period will be

(C A − CE )erT + Ker(T−t) − K > 0.

5
If American option was not exercised at all, you will end up with the European option that will also
not be exercised for a profit of
(C A − CE )erT > 0.

Therefore, the value of an American call and European call must be equal.

4 The Binomial Tree Model

4.1 Intuition behind the model

Here is a simple example to illustrate the main argument of the model. Consider a non-dividend
paying stock with current price S(0) = 100, and suppose that at the end of a period of time T = 1 its
price must either be S(1) = 50 or S(1) = 200, with equal probability. A call option on this stock is
available with an exercise price of K = 100, expiring at the end of the period. The risk free interest
rate is 5%. In this scenario, we can use the No-Arbitrage Principle to deduce the price of this call
option; C(0) = C. Consider the following hedging strategy:

• Write (sell) 3 calls worth C each.

• Buy 2 shares worth 100 each.

• Borrow 95.12 at 5% compound interest to be paid back at the end of the period.

From the following table we can see the different outcomes for this transaction

Table 1: Outcomes of hedged transaction when selling calls

Price goes down Price goes up

Present date S(1) = 50 S(1) = 200
Write 3 calls 3C 0 −300
Buy 2 shares −200 100 400
Borrow 95.12 −100 −100
Total 0 0 0

6
 If there is no arbitrage to be made, the price of the call must satisfy the equation 3C −200+95.12 =
0. Therefore, the current value of the call, C(0), is C = 34.96. If the price of the call was greater
than this figure, the writer of the call would be making an instantaneous profit at the present date,
while his future net expenditure equaling 0 with a probability of 1. This violates the No-Arbitrage
Principle.
Consider now the opposite hedging strategy:

• Buy 3 calls worth C each.

• Sell 2 shares short worth 100 each.

• Invest 95.12 at 5% compound interest to be paid back at the end of the period.

Table 2: Outcomes of hedged transaction when buying calls

Price goes down Price goes up

Present date S(1) = 50 S(1) = 200
Buy 3 calls -3C 0 300
Sell 2 shares 200 −100 −400
Invest −95.12 100 100
Total 0 0 0

From this table we can see that if the price of the call option was less than 34.96, the person
using the hedging strategy will, again, making an instantaneous profit at the present time, without
the risk of losing any money on the transaction. Hence, the price of the call option must be equal to
34.96, like in the previous scenario.
This simple example illustrates how an appropriately levered position in stock will replicate the
future returns of a call. In other words, if we purchase/sell shares and borrow/invest money in a
certain way, we can effectively duplicate a pure position in calls [2]. Note that the only information
we needed to compute the exact price of the call was the risk-free interest rate r, the strike price
K, the underlying stock price S(0), and the range of upwards and downwards movement in the
stock price. We did not need to know the probability of upward or downward movement, which is

7
important to note. This means that regardless of your view of how the stock price will move, the
value of the call option will remain the same.

4.2 1-Step Model

To derive the framework of the fully fledged binomial model, we will first begin by looking at a
1-Step tree. In this setup, we first assume that the price follows a binomial process over discrete time
periods. We will say that the stock will either grow over this single time-step by a factor u, with
probability q, or decrease by a factor d, with probability 1 − q. We also require that d < er < u,
where r is the assumed continuously compounding risk free rate. If this inequality didn’t hold, there
would be an opportunity for arbitrage [1].
If we were to draw the previous example as a 1-Step binomial tree, it would look like this:

uS(0)
q

S(0)

(1 −
q)
dS(0)

We can use a similar diagram to look at the value of the call option going from one time-step to
the next. Let C be the current value of the call option, let Cu be the value of the call if the stock
price goes up to uS(0) and let Cd be the value of the call if the stock price drops to dS(0). Using the
payoff functions defined in (2) we can draw a diagram representing the value of the call option.

Cu = max(uS(0) − K, 0)
q

(1 −
q)
Cd = max(dS(0) − K, 0)

If we were to construct a portfolio containing x shares of stock and y dollars invested in a

8
risk-free asset, this would cost us xS(0) + y at time t = 0. At the end of the same 1-Step time period,
the possible values of this portfolio would be

xuS(0) + er y
q

xS(0) + y A(0)

(1 −
q)
xdS(0) + er y

Since the choice of x and y in this portfolio is completely up to us, we can pick these amounts so
that the value of the portfolio at the end of the time-step is equal to the value of the call option at
that same time. This gives us the following equalities:

xuS(0) + er y = Cu

xdS(0) + er y = Cd

Assuming S(0), u, d, Cu, Cd and r to be constant, and solving these equations for x and y we find

Cu − Cd
x=
(u − d)S(0)
(4)
uCd − dCu
y=
(u − d)er
If x and y are chosen this way, we call the resulting portfolio a "hedging portfolio". By using the
No-Arbitrage Principle again, it can be shown that the current value of the call, C, cannot be less
than or greater than the hedging portfolio, i.e C = xS(0) + y [2]. Subbing the previously derived
values for x and y into this equation gives us

Cu − Cd uCd − dCu
C= +
(u − d) (u − d)er
 r (5)
u − er
   
e −d
= Cu + Cd e−r .
u−d u−d

This quality holds true if this value is greater than S − K. However, if this value is less than
S − K, then the value of the call option is C = S − K. This is important to note, as this difference

9
will be taken into account when pricing the value of the American options later on. If the option
were a European option, the equation in (5) would always hold as there is no early exercise allowed.
Equation (5) can also be simplified by setting

er − d u − er
p := and 1 − p := ,
u−d u−d

giving us

C = [pCu + (1 − p)Cd ]e−r , (6)

which is the exact formula for a call option one period prior to expiration, given in terms of S, K, u, d
and r. This formula has several notable features. Firstly, it doesn’t rely on the probability q of
the stock price rising/falling, as it doesn’t appear in the formula. This means that if two different
investors disagreed on how they believed the stock would move, they could agree on the current price
C when calculated in this way. Secondly, it doesn’t depend on an investors attitude towards risk.
They could be risk-averse or risk-seeking, but the only assumption we made is that they would prefer
more wealth over less wealth. Thirdly, the only random variable that the formula depends on is the
stock price itself. It is not, theoretically, affected by how the rest of the market moves. If another
pricing formula involving other variables was submitted as giving a different market price for the
option, one could show that this price is incorrect based on arbitrage arguments, as shown above.
er −d
Finally, one of the most important features of the model is that the value p := u−d can be seen
as a probability, because it is always greater than zero and less than one, based on the inequality
d < er < u. If an investor were risk-neutral, it is the value that q would have. To see this, we can set
the expected return on a unit of stock worth S(0) equal to the return we would get from investing the
same amount in a risk free asset.

quS(0) + (1 − q)dS(0) = er S(0)

10
dividing across by S(0) and solving for q yields

er − d
q= = p.
u−d

This is a notable feature, as it means the price of a call option can be interpreted as the expectation
of its discounted future value in a risk-neutral world. Applying the formula to the example in 4.1,
we can price the option directly without the need for awkwardly constructing a replicating portfolio
and using arbitrage arguments, and it will give us the same answer

S(0) = 100

u = 2, d = 0.5
er − d e0.05 − 0.5
p= = = 0.3675
u−d 2 − 0.5
Cu = 100, Cd = 0

C = [0.3675 · 100 + 0.6325 · 0]e−0.05 = 34.96.

4.3 2-Step and n-Step model

The next logical step is to expand this 1-Step procedure to more steps. For a 2-Step binomial process,
the stock can take three different values, and the tree will look like this-

u2 S(0)
2
q

uS(0) (1 −
q q )q

S(0) duS(0)
q
(1 − q)
(1 −
q)
dS(0)

(1 −
q) 2
d 2 S(0)

11
 Doing the same for the value of the call we get the following-

Cuu = max(u2 S(0) − K, 0)

2
q

Cu (1 −
q q )q

C Cud = max(udS(0) − K, 0)

(1 − (1 − q )q
q)
Cd

(1 −
q) 2
Cdd = max(d 2 S(0) − K, 0)

Cuu stands for the value of the call after two periods if the stock price moves upwards twice, and
Cdd stands for the value of the call after two periods if the stock price moves downwards twice. Cdu
is the value of the call after two periods if the stock price moves upward once, and downward once.
Note that Cdu = Cud as it is irrelevant whether the stock price goes up and then down, or down and
then up.
We can look at the last period after the two "nodes" Cu and Cd as 2 individual 1-Step processes,
which gives us the exact same situation as before. Therefore we can say that

Cu = [pCuu + (1 − p)Cdu ]e−r ,

and
Cd = [pCdu + (1 − p)Cdd ]e−r .

In the exact same way as in the previous section, we can construct a portfolio of x stocks and y
invested in a risk-free asset, and use Equation (4) again, with new values of Cu and Cd .
Substituting these values of Cu, Cd into Equation (6), gives rise to the following equation for the
price of a call option at time 0

12
 C = [p2Cuu + 2p(1 − p)Cud + (1 − p)2Cdd ]e−2r

= [p2 max(u2 S(0) − K, 0) + 2p(1 − p)max(duS(0) − K, 0) (7)

+ (1 − p)2 max(d 2 S(0) − K, 0)]e−2r

This formula, like the one for a 1-Step process, depends on S, K, u, d, and r, but is also affected
by the number of time-steps taken; n. In the example above, n = 2, but we can extend this approach
to any number of periods, as it is a recursive procedure with the same logic and arbitrage argument
applied at each step. The generalised valuation formula is then
" n   #
Õ n!
C= pi (1 − p)n−i max(ui d n−i S(0) − K, 0) e−rn . (8)
i=0
i!(n − i)!

This formula makes intuitive sense if you consider the coefficient i= 0 i !(n−i)! at the start of the
n!
Ín

formula. This is the same as ni and can be viewed as the binomial coefficient. It means that

movements that deviate less from the stock price S(0) are given more weight, as they are "more
likely" to occur given the set-up of the model.
To clean up this formula, we can focus on all of the outcomes of the binomial process that
actually result in S − K > 0. In every binomial model, the prices at the end of the final period
will be distributed in order from highest to lowest, starting at un S(0) and going down to d n S(0).
Therefore, there will be a certain number a where for every i < a, we have ui d n−i S(0) − K ≤ 0,
and so max(ui d n−i S(0) − K, 0) = 0. At the same time, for every i ≥ a, max(ui d n−i S(0) − K, 0) =
ui d n−i S(0) − K. Disregarding all the outcomes for i < a, because they will have a value of 0, Equation
(8) can be written as
" n   #
Õ n!
C= pi (1 − p)n−i (ui d n−i S(0) − K) e−rn (9)
i=a
i!(n − i)!

This equation can be re-arranged further, with some additional replacements, and eventually one

13
would arrive at the following simplification

" n    i n−i  #
n! ud
p?i (1 − p?n−i
Õ
C = S(0)
i=a
i!(n − i)! ern
" n   #
Õ n!
− Ke−rn pi (1 − p)n−i ,
i=a
i!(n − i)!

where p? = ( ur )p. To understand why this substitution makes sense, it is enough to see that

  i   n−i
ui d n−i

u d
i
p (1 − p) n−i
= p (1 − p) = p?i (1 − p?)n−i .
ern r r

Replacing
n    i n−i 
Õ n! ?i ? n−i u d
p (1 − p )
i=a
i!(n − i)! ern

and
n  
Õ n!
pi (1 − p)n−i
i=a
i!(n − i)!

with
φ[a; n, p?]

and
φ[a; n, p]

respectively, where φ[a; n, p] is the binomial distribution function, we end up at the Cox-Rubenstein
Formula for Call Option Pricing [2].

C = Sφ[a; n, p?] − Ke−rn φ[a; n, p] (10)

In 3.2 it was shown that the price of an American and European call was equal, provided they do
not pay dividends and have the same strike price K and expiry date T. Therefore, Equation (10) can
be used both for American and European options.

14
4.4 Pricing put options

Put options can be priced using the exact same process and arbitrage arguments as a call option.
Although the payoff function at the end is different from that of a call option, the binomial formulation
can still be applied. Using the definition for the payoff on a put option given in Equation (2), the one
step model for the value of a put option can be viewed as

Pu = max(K − uS(0), 0)
q

(1 −
q)
Pd = max(K − dS(0), 0)

Just like in section 4.2 we can set up a portfolio with x stocks and y invested in a risk-free asset
which will have the same end of period values as the put. Using the same series of steps as done in
section 4.2, it can be shown that

P = [pPu + (1 − p)Pd ]e−r (11)

er −d
is the value of the option at the start of the period (with p = u−d as before) [2]. However, this is
only if P > K − S, and P = K − S otherwise. This is important when pricing American put options,
because of the possibility of early exercise. Therefore we cannot express the price of the American
put option in a simple form, the same way we can for the European option.
The following example will illustrate in detail how the possibility of early exercise will affect the
price of an American put option. Consider a put option on a stock with a current price of S(0) = 30,
with expiry date T = 3 and where u = 1.1, d = 0.9, and r = 0.01. The risk neutral probability p of a
e0 .01−0.9
price going up would be 1.1−0.9 = 0.5503. The price evolution of the asset in the binomial tree
model would look like this, with the payoff, K − S, given at the end.

15
 K = 33

39.93 Payoff = 0

36.3

33 34.49 Payoff = 0

30 31.35

28.5 29.79 Payoff = 3.21

27.08

25.73 Payoff = 7.27

t= 0 1 2 3

Using the risk neutral probability p and discounting the expectation, we can calculate that at
time t = 2, the value of the put call would be 0 if the price at that time is 36.3, 1.43 if the price at
that time is 31.35, and 4.99 if the price at that time is 27.08. These are the expected values of the
put call if the investor waits and does not exercise the option until time T = 3. If he exercises the
option at t = 2, the value of the put is then either 0, 1.65 or 5.92, if you directly calculate K − S at
that time. So in the American option setting, the rational investor would not wait until the expiry
date to exercise the option.
Notably, if the price of the stock dropped to 28.5 after the first time-step, the investor would
exercise it then for a payoff of 4.5, as the risk neutral expectation of the price would only be
(0.5503 ∗ 1.65 + 0.4497 ∗ 5.92)e−0.01 = 3.57. Had the stock gone up in price to 33, its value at that
time would be (0.4497 ∗ 1.65)e−.01 = 0.7346. In this scenario, the investor would not have exercised
the option and waited to see how it moves. Going back to time 0, the price of the option at the time
of writing would be (0.5503 ∗ 0.7346 + 0.4497 ∗ 4.5)e−0.01 = 2.40 If a European option was written
with the exact same parameters, its value would be 1.68. This explains why the possibility for earlier
exercise makes American put options inherently more valuable than European put options, and why

16
there isn’t a fixed simple formula that one can employ to calculate its price.

4.5 Time step refinement

In all the previous sections so far, it would have been understandable to automatically associate each
time-step taken with a particular length of time, such as a day, or a week. If the time-steps were
really representing such lengths of time, you would be right to say that the model is a complete
oversimplification. Firstly, because a day from now a stock may have more than just two possible
values. Secondly, the market is open for trading more than once a day, and trading takes place almost
continuously.
However, while the Binomial Tree Method is built upon a very simple 1-Step model, it possesses
the flexibility to meet these concerns. Each time-step can be taken to represent any unit of time, be
it an hour or even a minute. Then, trading can take much more frequently and the stock can take on
a multitude of values by the end of the n-Step period.
With this in mind, we have to be careful in choosing the parameters when modelling the stock
price movements. Suppose that T is the fixed length of calendar time to expiration, and we choose
to construct a model that has n periods until expiration, each of length δt. Then, clearly, δt = Tn .
The more frequently that trading occurs, the greater n is, and so the smaller δt becomes. We must
then adjust the variables r, u, and d so that we obtain empirically realistic results.
Previously, we took it for granted that each time-step would have an associated risk-free rate of
interest r, which was compounded continuously. Now, we need to differentiate between the interest
rate over the calendar time, and the rate that will be effective over each time-step. We will continue
to let r denote the interest rate over the time period T, as defined in Section 2, and let r̂ denote
the rate over a single time-step. Due to our assumption about the interest rate being compounded
continuously, it is clear that erT = (er̂ )n , and so r̂ = rT/n or r̂ = rδt. By construction, the interest
rate r̂ is dependent on n so that the total return over elapsed time T is independent of n.

17
4.6 Comparison to continuous random walk

In all of the examples so far, the choices of parameters u, d, and by construction p, have been
completely arbitrary. There are a number of different ways one can derive these parameters, and
here we will focus on the way proposed by Wilmott et al., (1995). In his book, The Mathematics
of Financial Derivatives, he shows it useful to derive the parameters in a way that arises from a
discrete random walk model of the underlying security.
It has been shown that binomial models can be seen as particular cases of explicit finite-difference
methods. This, paired with the assumption of risk neutrality, means we can rely on Black-Scholes
analysis to formulate the models [3]. Therefore, we can choose values for u, d and p such that the
discrete random walk represented by the binomial tree method and the continuous random walk
have the same expectation and variance.
Suppose that the value of the asset is S(m) at the time-step mδt. We equate the expected value
and variance of S(m + 1) for the binomial tree model, and the continuous risk-neutral random walk.
In a risk-neutral world, the price of a stock can be given by the stochastic differential equation

dS
= σdX + r dt. (12)
S

In this equation, dS
S is the measure of the return on the asset. It is broken down into two main
components. Firstly, r dt is the predictable, deterministic return on the money invested in a risk-free
asset, at rate r. The second term, σdX models the random change in the asset price in response
to unpredictable movements in the market. It is represented by a random sample drawn from a
normal distribution, dX, multiplied by the volatility, σ, which measures the standard deviation of
the returns.
The term dX is important as it contains the randomness in the movement of the stock price,
which is definitely something that needs to be taken into account. Here it is known as the Wiener
process, and it has the following properties [3]

• dX is a random variable, drawn from the normal distribution;

18
 • the mean of dX is zero;

• the variance of dX is dt.

These features of dX are important as they allow to compute the probability density function of S.
Through some manipulations and stochastic some calculus, using Ito’s Lemma it can be shown [3]
that the probability density function of S is

1
e−(log(S/S0 )−(µ− 2 σ
1 2 )t)/2σ 2 t
√ (13)
σS 2πt

Using this we can calculate the expected value of S(m + 1) given that the price of the stock at
time mδt is S(m);

∫ ∞  
Ec [S(m + 1)|S(m)] = S p S(m), mδt; S , (m + 1)δt dS0 = erδt S(m),
0 0
0

where p(S, t; S0, t 0) is the probability density function for the risk-neutral random walk defined in
Equation (13).
The expected value of S(m + 1) given S(m) under the binomial tree model is given by

Eb [S(m + 1)|S(m)] = (pu + (1 − p)d)S(m).

Putting the last two results together gives

pu + (1 − p)d = erδt, (14)

er δt −d
which makes sense, as rearranging this equation to find p we see that p = u−d . As before, this can
be seen as the risk-neutral probability of the stock going up in value to uS(m), except over a shorter
time step, so we go from r to r̂ = rδt.
The variance of the stock price at S(m + 1), given S(m), V[S(m + 1)|S(m)] is given by

E[S(m + 1)2 |S(m)] − E[S(m + 1)|S(m)]2 .

19
It can be shown that under the continuous random walk,

∫ ∞  
2 )δt
Ec [S(m + 1) |S(m)] =
2
(S ) p S(m), mδt; S , (m + 1)δt dS0 = e(2r+σ
0 2 0
(S(m))2,
0

and so the corresponding variance is

Vc [S(m + 1)|S(m)] = e2rδt (eσ

2 δt
− 1)(S(m))2 .

For the binomial method, we have

Eb [S(m + 1)2 |S(m)] = (pu2 + (1 − p)d 2 )(S(m))2 (15)

and so, using the result in Equation (14) the corresponding variance is

Vb [S(m + 1)|S(m)] = pu2 + (1 − p)d 2 − e2rδt (S(m))2 (16)

Equations (15) and (16) are two equations with three unknowns, so to solve them we require an
additional constraint. The two popular choices are either setting u = 1
d or p = 12 .
If we choose u = 1
d as our constraint, Wilmott et al. show that solving Equations (15) and (16)
give us the following choices for u, d and p

√ √ erδt − d
d = A− A2 − 1, u = A + A2 − 1, p = , (17)
u−d

where
 
1 −rδt (r+σ 2 )δt
A= e +e .
2

This choice gives rise to a tree that is symmetric about the initial price, with the initial
price S(0) being reoccurring every second time step. This is because after an upward then
an downward movement, or an downward then an upward movement, the price will become
udS(0) = duS(0) = d d1 S(0) = S(0).

20
 If we choose the constraint p = 12 , solving (14) and (15) for u and d gives us the following

p
d = erδt 1 − eσ δt − 1 ,
2

p
u = erδt 1 + eσ δt − 1 ,
2
(18)

1
p= .
2

It is relevant to point out that for either choice of constraint, it is preferred that the time-steps taken
are not too large. In the former case, where u = d1 , if too large a time step is taken, p or 1 − p may
become negative, in which case the binomial method will fail. On the other hand, if we choose
p= 1
2 and too large a time step is taken, then d will become negative, and again the method will not
work. So in terms of parameter choice, the smaller the time step - the better. We will later see that
smaller time steps also lead to more accurate results when calculating the true price of the option.
It is notable that in these equations we also have a new σ parameter, which we didn’t have before.
As mentioned above, this is the standard deviation of the returns, also known as the volatility. The
use of volatility is somewhat limiting, as the volatility of an asset will rarely remain constant over
the life of the option. There are different ways to estimate the volatility of an asset, the popular
choices being using implied volatility, historical volatility, and looking at volatility as a function of
time. Wilmott et al. discusses these in length, but we will be assuming σ to be a constant for the
purposes required here.
In 4.2 we discussed how the binomial tree method works regardless of both an investors attitude
towards risk, and their view of which way the stock price is likely to move. However, the standard
deviation of the returns is something we should include in the model. If the underlying security for
the option is the price of gold, we would expect the standard deviation to be low, as opposed to the
standard deviation of the stock price of a new start up. This difference in variation of prices should
be included in the modelling of the asset price. Cox, Ross, and Rubenstein (1979) [2] included
this by setting the variance of the continuously compounded rate of return of the assumed price
movement equal to with that of the actual stock price, which led to a slightly different choices for u,
d and p. D. Higham (1995) [4] had a different approach again, which led to different parameters

21
1, but all of the above are based on the same principle of equating the expectation and standard
deviation of returns of the binomial model with the expectation and standard deviation of the actual
returns.
The use of volatility is somewhat limiting, as the volatility of an asset will rarely remain constant
over the life of the option.

5 Adjusting for dividends

So far we have looked at options on underlying assets that do not pay any dividends. We should,
however, bring them into consideration as options are often written where the underlying asset pays
a dividend. Most stable companies offer dividends to shareholders, and a dividend can be seen as a
shareholder’s share of the company profits. In the US, for example, exchange-traded stock options
generally have less than 8 months to maturity and the dividends payable during the life of the option
can usually be predicted with reasonable accuracy [6].
One can easily accommodate a constant dividend yield Do paid on the underlying asset. The
effective risk-free growth rate becomes r − D0 rather than r, and so the SDE describing the stock
price becomes
dS
= (r − Do )dt + σdX.
S

Therefore we can replace r by r − D0 in the tree construction phase, as proposed by Wilmott [3].
For the case u = 1/d we get

√ √ e(r−D0 )δt − d
d = A− A2 − 1, u = A + A2 − 1, p = , (19)
u−d

where
 
1 −(r−D0 )δt (r−D0 +σ 2 )δt
A= e +e .
2
√ √
1In his book, Higham chooses p to be equal to 12 , and his values for u and d are eσ δt+(r− 2 σ δt+(r− 12 σ 2 )δt ,
1 2
)δt and e−σ
respectively. It can be shown the values in (17) are very close to this when δt is very small.

22
 Similarly, if p = 1/2, parameters described in (17) become

p
d = e(r−D0 )δt 1 − eσ δt − 1 ,
2

p
u = e(r−D0 )δt 1 + eσ δt − 1 ,
2
(20)

1
p= .
2
For a discrete dividend payment, we will consider how the ex-dividend date affects payment. If
you purchase a stock on its ex-dividend date or after, you will not receive the next dividend payment,
as you effectively did not invest in it on time. This is important, as when dividends are expected, we
can no longer assert that an American call option will not be exercised early. The payment of a
dividend effectively causes the stock price to jump down, making the option less attractive. So since
the payment of a dividend changes the price of a stock, it may be optimal to exercise an American
call immediately before an ex-dividend date.
This is important for us as it means that it means we should consider the dividends carefully
when pricing American options. In 3.2 it was shown that an American call option should never be
exercised early in the absence of dividends. To extend this argument in the case of dividend paying
stock, Hull (2000) shows, using arbitrage arguments, that it can only be optimal to exercise at a
time immediately before the stock goes ex-dividend. There exists a formula suggested by Roll [8],
Geske [9], and Whaley [10], for valuing American calls when there is only one ex-dividend date.
This involves the cumulative bivariate normal distribution function and is more involved than the
standard Black-Scholes. However, it still lacks the transparency and ease of use that the binomial
tree model possesses.

6 Advantages and Limitations

The binomial tree method for the valuation of options is regarded as one of the most important
concepts in modern financial theory. It is often compared to the Black-Scholes model in terms of
its advantages and disadvantages, as the Black-Scholes is the other extremely popular approach to

23
pricing options. Some of the features of the binomial tree model that make it appealing are the
multiple period view, the transparency of the model, and the simplicity behind its mathematical
construction.
In regards to the pricing of American options, the multiple period view of the binomial tree
method stands out as its key advantage. Using this model, the user can visualise the change in
underlying asset price from period to period and as such decide on the best decision to make
in different points in time. As illustrated in the last section, this allows us to take into account
the early exercise possibility to maximise the profit, and hence increases the value of the option.
In comparison, the Black-Scholes model cannot account for this. It most commonly used as a
calculation with given inputs and outputs, and lacks the flexibility granted by the multi period
characteristic of the binomial tree method.
Transparency is another feature of the binomial tree method that is extremely useful when pricing
an option. For example, the Black-Scholes model has five inputs:

• Risk-free rate,

• Exercise price,

• Current price of the asset,

• Expiry time/time to maturity,

• Implied volatility.

When the Black-Scholes model is used, it is simply calculating the price based on the 5 inputs above.
The impacts of these factors hard to understand with no intuitive visualisation of the process. With
the binomial tree method, one can see how the price changes from time step to time step. Although
it may be tedious, it is also possible to alter the inputs and parameters between time steps based on
how you think the market may change (i.e. change in interest rate). The Black-Scholes model does
not allow for this type of alteration.

24
 The original Black-Scholes model also does not allow to account for discrete dividends being
paid out. There are, however, different variations of the model that do alter the formula to adjust for
this, such as the Roll-Geske-Whaley method to price American options mentioned in the previous
section.
The most obvious limitation of the binomial tree model is the computational intensity of the
process. Black-Scholes based models require a few computations at most, where for the binomial
model, there are many calculations done for each time step, both going forward in time, and backward
in time when pricing the model. Higham (2004) also showed that the smaller the step size, the
greater the accuracy, so one will always be trading off accuracy for computational efficiency, and vice
versa. At present time, however, this isn’t too much of a concern. With the processing capabilities of
modern processors, such calculations do not take long to compute, even for very large values of M.
Finite difference methods are another widely used model for pricing options. They have the
same level of complexity as those solved by tree approaches [6], and are only applied when the other
approaches are inappropriate. Examples of this are when interest rates are changing or when there
is a time linked dividend being paid. When standard assumptions are applied, the explicit technique
encompasses the binomial and trinomial tree methods [11]. It has also been shown that under a
certain choice of parameters, the binomial tree model becomes a special case of the explicit finite
difference method [12].

7 Calculation of Prices

When calculating the prices of options, we will follow the methodology outlined by Higham in An
Introduction to Financial Option Valuation [4]. This approach follows the same basic principle as
the Cox, Ross and Rubenstein, but is presented and explained through Matlab. For the purposes of
this project, the code has been adapted for Python.
The basic idea, again, is that we calculate the price of the option at time t = 0 by calculating the
payoff at expiry time t = T, and recursively working backwards. The value of the stock at each time

25
t = i can take on i different values, so we will denote the stock price S at time i and position n by
Sni = d i−n un S00 , where S00 = S(0) from our original formulation. Then, the value of an American
option t time i and position n, both put and call, is given by

VnM = φ(SnM ),
(21)
Vni = i+1
max(φ(Sni ), e−rδt [pVn+ 1 + (1 − p)Vni+1 ]),

where φ is the payoff function dependent on the type of option, 0 ≤ n ≤ i, and 0 ≤ i ≤ M − 1. Here,
at each time step, we consider the maximum between exercising the option at time i for φ(Sni ), and
the discounted expected payoff of the option at time i + 1 under the risk neutral probability. This is
the same formulation as given in equations (6) and (11), except generalised. Here, we also consider
the immediate payoff function, φ, as we are concerned with American options that grant the holder
the possibility of early exercise.
We will calculate the value of the option at each time step under the two different constraints
explored in 4.6, where either u = 1/d or p = 1/2.The two different parameter choices we will be
using have are described in (16) and (17). As such, the inputs our calculations require are the
following

• S(0), the initial stock price at time t = 0,

• K, the strike price,

• T, the expiry date,

• r, the risk free interest rate,

• σ, the volatility of the stock,

• M, the number of time steps desired,

• D, the continuous dividend rate.

Much literature has been devoted to establishing that the error in various methods tends to zero
as M → ∞ [4], and very little attention was initially paid to the rate of convergence. Leisen and

26
Reimer (1996) [5], developed a general convergence rate theory, and Walsh (2003) [7] provided a
more detailed analysis.
There are other ways of calculating option prices that one could compare this method with.
Higham (2004), for example, provides graphs of the error vs. step size when comparing the
pricing of European options using the Black-Scholes method and the binomial method, to show
the convergence of the binomial method to the Black-Scholes method, for large values of M [4].
This, however, beyond the scope of this project. Wilmott et al. (1995) also provide tables of figures
comparing the two methods, for different values of T and M.
One of the main benefits of this code, is that could easily be altered if one would wish for u and
d to be deterministic functions of time. If an investor believes that these upward and downward
movements will vary in magnitude over the lifetime of the option, this can be accounted for. Similarly,
the risk free interest rate could be adjusted to be a function of time rather than a constant.

8 Visualisation of Binomial Tree

Using the process described above, it was possible to implement an algorithm to recursively price
options, with a binomial tree being plotted and saved as a .png file in the same file directory where
the code is saved. A full few trees have been printed on full pages at the end of the paper for
reference.
The code is broken up into four sections. Firstly each cell prices either an American call or
an American put. Secondly, for each of these we either have u = 1/d p = 1/2 (AmericanCall1,
AmericanPut1) or u = 1/d (AmericanCall2, AmericanPut2) . The general concept is to show three
separate values at each node in the binomial tree;

• The value of the stock at that time and position,

• The instantaneous profit of exercising the option at that time and position,

• The expected value of the option if you do not exercise and wait until the next time step.

27
 Using this, potential investors can view how the value of the option will change based on how
the underlying price changes. They can also see how changes in the inputs and parameters affect the
price of the option. Figure 1 on page 29 shows the evolution of the price of an American put with
S(0) = 100, K = 98, T = 1, r = 0.06, D = 0.03, σ = 0.1, and p = 1/2. Figure 2 on page 30 shows
the same setup, except for an American call.
This code is similar to what can be found in the DerivaGem software which is mentioned and
discussed in Hull (2000). The implementation presented here is a lot more flexible, however, as it
is implemented in Python as opposed to using Macros in an Excel spreadsheet. It is completely
transparent and adjustable, and can be altered to compare to different models, or to price other
instruments. The reader is encouraged to take a look at the code and try it out for themselves. There
are countless different possibilities for what stocks you can price using this method, and countless
ways you can adjust and tweak the parameters to obtain different results.

28
 Figure 1: American Put 1, p = 1/2.

Here we can see the evolution of an American put. Interestingly, it can be noted that once the
price of the asset dips below approximately 90, that it will always be optimal to exercise early,
whereas in the earlier stages of the process it was always more advantageous to hold onto the asset.
If the strike price K was higher, it would be more profitable to exercise early more often.

29
 Figure 2: American Call 1, p = 1/2.

Here we can see for the American call, even though there is a dividend payment, it is never
optimal to exercise the option early. By changing the initial parameters a bit we can see that in
different situations it would sometimes be profitable to exercise early. If the dividend payment D
was greater, for example, there would be times before the expiry date when immediate exercise is
more profitable than holding on to the option.

30
9 Conclusion

It has been shown here that the pricing of options can be broken solely into a set of arbitrage
considerations, provided that the stock price movements conform to a discrete binomial process.
Importantly, the probability of an upward or downward movement do not enter the valuation formula.
Therefore, we obtain the same result regardless of what q depends on, be it on current or past
stock prices, or on other random variables. One could also account for certain imperfections in the
binomial tree model, such as differential borrowing and lending rates, and margin requirements [2].
these can be shown to produce upper and lower bounds on options prices, outside of which riskless
profitable arbitrage would be possible.
With this model in mind, we showed how it can be used to price American options. At first
concerned with American call options, we showed how the value of an American call is equal to that
of a European call, when they have the same underlying asset, expiry date and strike price, given
that the risk free rate is the same. We then explained how when valuing American put options this is
not the case, and that since early exercise may be optimal, the price of an American put is greater or
equal to that of a European put.
We have also shown that the comparison of the binomial tree model to the continuous random
walk model, discussed by Willmott et al. [3], is extremely useful when deriving the parameters
required to build the binomial tree model. Using different constraints, full sets of values can be
derived that, again, do not rely on the actual probability q of the stock price going up or down over
one time step.
The advantages and limitations of the model were also discussed, particularly the multi-period
view and the transparency of the model. These advantages are what allowed us to then provide a
visualisation of the model, so that it could be seen directly how the price of an option is affected by
the evolution of the underlying asset price. This is done by providing code that calculates the stock
prices at each time step, as well as the accompanying option values at that time step. A few toy
examples of such trees are included here, with code provided explaining the process.

31
Implementation

All implementations were written in Python and are publicly available on GitHub.

Acknowledgement

I would like to thank my supervisor, Dr. Claus Köstler, for giving me guidance and direction over
the course of the project.

References

[1] M. Capinski and T. Zastawniak Mathematics for Finance, An Introduction to Financial

Engineering, Springer, 2011.

[2] J.C. Cox, S.A. Ross and M. Rubenstein Option Pricing: A Simplified Approach, Journal of
Financial Economics 7, 1979.

[3] P. Wilmott, S. Howson, J. Dewynne, The Mathematics of Financial Derivatives, A Student

Introduction, Cambridge University Press, 1995.

[4] D. J. Higham, An Introduction to Financial Option Valuation, Cambridge University Press,

2004.

[5] D. Leisen and M. Reimer, Binomial models for option valuation, examining and improving
convergence, Applied Mathematical Finance, 3:319-346, 1996.

[6] J. Hull, Options, futures, and other derivatives, 4th edn. Englewood Cliffs, Prentice Hall India,
2000.

[7] J. B. Walsh, The rate of convergence of the binomial tree scheme, Finance and Stochastics,
7:337, 2003.

32
 [8] R. Roll, An analytic valuation formula for unprotected American call options on stocks with
known dividends, Journal of Financial Economics, 5:251-258, 1977.

[9] R. Geske A note on an analytical valuation formula for unprotected American options on
stocks with known dividends, Journal of Financial Economics, 7:375-380, 1979.

[10] R. E. Whaley, On the valuation of American call options on stocks with known dividends,
Journal of Financial Economics, 9:207-211, 1981.

[11] E. Shwartz, M. Brennan, Finite difference methods and jump processes arising in the pricing
of contingent claims: A synthesis, Journal of Financial and Quantitative Analysis 13 461-474,
1978.

[12] M. Rubenstein, On the relation between binomial and trinomial option pricing models, Journal
of Derivatives, 8 47-50, 2007.

[13] W. F. Sharpe, Investments, Prentice Hall, 1978.

33