You are on page 1of 19

Discrete-Time Option Pricing Theory

Eric Emer
Department of Mathematics
Massachusetts Institute of Technology
May 11, 2013
Abstract
This paper explains concepts related to Discrete-Time Option Pricing. It gives an
explanation of stock behavior, some of the key ideas regarding stock options, presents
the discrete-time Cox-Ross-Rubinstein Binomial Model, and then shows a derivation
of the key result: the Black-Scholes-Merton Model. This paper will focus only on the
pricing of European call options, which can only be exercised at their expiration date,
as opposed to American call options, which can be exercised at any time. We imply
that the theory for pricing a call option can be easily inverted and applied to the case
of a put option.
1 Introduction
Every day, many investors and traders issue buy and sell orders which ultimately de-
termine the price of a stock. Liquid markets allow stocks to settle at a particular price,
indicated by a discrete dollar value. In the nancial markets, there are many types
of derivatives which exist. One such derivative, discussed thoroughly in this paper, is
the European option. Rather than determining its price based purely on speculation
and trade history, we can intuitively infer that a derivative should have a fair price,
determinable by the parameters of the underlying asset. The motivation for studying
options is obvious: we have a complicated nancial derivative, and we can prot by
pricing it accurately. The intrinsic complexity of an option lies in the idea that, unlike
a stock, it has an expiration date. At the options expiration date, the stock will have
a xed value, so the option becomes easier to price as the option nears expiration, and
much harder to price farther away from expiration. In this paper, we will provide a
visualization of this idea in discrete-time, known as the Cox-Ross-Rubinstein (CRR)
Binomial Model.
Extensive research has been done on options valuation, but the foremost method
of determining the value of an option is the Black-Scholes Model, published in 1973.
In this paper, we will show a derivation of the Black-Scholes Model, a continuous time
1
option pricing model, from the discrete-time CRR Binomial Model. The motivation for
studying discrete-time option pricing is to achieve a more realistic behavior model of
options. In reality, trading is a discrete process rather than a continuous one. Trades
are executed at discrete points in time, at discrete price points. While it may be con-
venient to model the price of an option in continuous time, the notion of trading as a
continuous-time process is less realistic.
Determining the appropriate price of an option is a very common problem for
investors and traders. In this paper, we will give an overview of option pricing and its
parameters. Next, we will express the theoretical modeling of stock behavior which lays
the groundwork for option pricing theory. In addition, we will detail the CRR Binomial
Model for option valuation, a simple, yet elegant discrete-time option pricing model.
Finally, we will show that the CRR Binomial Model converges to the Black-Scholes
Model.
2 Dening Option Pricing
As stated earlier, we can intuitively infer that the option price and the stock price
depend on the same source of uncertainty, the price and movement of the underlying
asset. Therefore, there are six parameters which aect the price of a non-dividend-
paying option:
1. The current price, S
0
2. The strike price, K
3. The time until expiration, T
4. The volatility of the stock price,
5. The risk-free interest rate, r
We will examine what these parameters mean, and how they might aect the price
of an option. Consider a stock for the underlying asset S. The stock can be bought or
sold at times t, and the price per unit of stock varies with respect to t, and is given as
S
t
. For such a stock, we dene:
Denition 1 (Call option). A call option for S at expiration t
e
with strike price K
gives the owner the right but not the obligation to buy a unit of S from the seller at
price K at time t
e
.
Denition 2 (Put option). A put option for S at expiration t
e
with strike price K
gives the owner the right but not the obligation to sell a unit of S to the seller at price
K at time t
e
.
Lets consider the parameters as they apply to call options. At the time of ex-
piration, we can dene the payo, not including the price paid for the option, to be
= S
te
K. The value of a call option increases as the stock price increases. Also
visible from the payo equation, the value of the call option decreases as the strike price
2
increases. The value of the call option will increase as the time to expiration increases,
because this allows for more price movement. In laymans terms, the volatility , is the
degree of uncertainty about the future stock price. A higher volatility stock is more
likely to have large price swings. In the case of a call option, the owner of the option
benets from an increase in the stock price, and will have a downside risk limited to
the price paid for the call option. Therefore, the value of a call option increases as
the volatility increases. Finally, we examine the risk-free interest rate r. When r is
large, investors expected return on stocks will increase. As a result, the value of the
call option increases.
2.1 Lognormal Property of Stock Prices
We assume that percentage changes in a stocks price during a short period of time
are normally distributed. However, a stocks price cannot go below 0. Therefore, stock
prices have a lognormal distribution.
Figure 1: A lognormal distribution, bounded below by 0. Image credit: [4]
Consider a stock with parameters:
: Expected return on stock per year
: Volatility of stock price per year
Given our assumption, it would hold that, for a small time interval t:
mean return = t
standard deviation of return =
_
t
variance of return =
2
t
Suppose that during the interval t, the actual change in stock price is S from
S. Therefore,
S
S
denotes the percentage change in stock price. We can reasonably
approximate the percentage change in stock price by a normal distribution:
S
S
N(t,
2
t) (1)
and of course, the stock prices themselves are lognormally distributed.
3
ln S
t+t
N(ln S
t
+ (
1
2

2
)t,
2
t)
which implies,
ln
S
t+t
S
t
N((
1
2

2
)t,
2
t) (2)
2.2 Volatility
It is important to dene the notion of volatility when computing option prices. This
is the parameter which makes the pricing of the derivative unique from the pricing of
the stock. Volatility can be computed over any selected time interval . We observe
the stock price at xed intervals: every minute, every hour, every day, every year, or
any other xed interval. To estimate the volatility of a stock, we dene:
n + 1: the discrete number of observations (0, 1, . . . , n);
S
i
: the observed stock price at the end of the i
th
interval, i = 0, 1, . . . , n;
: the length of the time interval in years.
We say,
u
i
= log
S
i
S
i1
, i = 1, . . . , n
Where u
i
is the continuously compounded return for day i. In order to estimate the
volatility, we must also compute the empirical variance.
s
2
=
1
n 1
n

i=1
(u
i
u)
2
, where u =
1
n
n

i=1
u
i
From (1), we recall that the empirical variance s
2
, is in fact an estimate for
2
. As a
result, we deduce the volatility, ,
s
2
=
2

=
s

Calculating the volatility of a stock can be a computationally intensive task for large
n, but it is critically important to option pricing.
2.3 Approximating Option Prices by Bounding
In this section, we will briey show the property of upper and lower bounds on a
call option. The bounds do not at all depend on the uncertain movement of the stock
price. Therefore, if the price of an option were to exceed the bounds, then a trader could
make a riskless prot. Because option pricing theory relies on the assumption of no
arbitrage, we can generate bounds based on fullling the requirement of no arbitrage.
For an underlying asset S, whose stock can be bought or sold at times t, and whose
price per unit of stock is given by S(t), we have:
4
Property 1 (Upper Bound). C(t
0
) S(t
0
)
where C(t
0
) is the price of the call option, evaluated at time t
0
, and S(t
0
) is the price of
the stock at that same time. If the value of the call option were to exceed this bound,
then a trader could make a riskless prot by simply buying the stock and selling the
call option. Similarly, we have,
Property 2 (Lower Bound). C(t
0
) max
_
S(t
0
) Ke
rT
, 0
_
where Ke
rT
represents the present value of K dollars. Namely, K dollars invested at
the risk-free rate r over a period T, the time until the options expiration. If the price
of the call option were to fall below S(t
0
)Ke
rT
, a trader could make a riskless prot
by buying the call and selling the stock. We also observe that an options value cannot
be negative. At worst, an options value will go to zero, and it will expire without
any value. Therefore, assuming no arbitrage, we conclude that we have bounds on the
price of the call option:
max
_
S(t
0
) Ke
rT
, 0
_
C(t
0
) S(t
0
)
3 Behavior Model of Stock Prices
In the eld of nancial derivatives, we expect that the price of the derivative will
depend on the uncertain movement of the stock price. In the case of options, we must
devise a behavior model for the price of the underlying asset, the stock, as a parameter
for our option pricing model. In this section, we will detail the assumptions required for
pricing options, and explain the Geometric Brownian Motion Model of stock behavior.
Given our market assumptions and a reliable model for modeling stock behavior, we
can begin to develop our model for pricing options in discrete time.
3.1 Market Assumptions in Option Pricing Theory
Most basic option pricing is premised on the notion of a slightly idealized market.
That is, a market where the following assumptions hold:
1. The stock price follows the model given in 3.2, with and constant.
2. The market has innite liquidity. Market participants may buy or sell units of S
at any time.
3. The market has innite depth. Each purchase or sale of S does not aect the
price of S.
4. There are no transaction costs.
5. There are no riskless arbitrage opportunities available.
6. The risk-free rate, r, is the constant and remains the same for all maturities.
5
3.2 Discrete-Time Geometric Brownian Motion
We consider the model of a random walk in discrete time. This model is in fact, a
fairly reasonable assumption for most small moves, and explains a lot of market behav-
ior. We assume that our variable t is not continuous, and that it will increase in discrete
steps of size t. Lets consider what this means. For a given t, we will have that S
t
evaluates to some xed monetary value. Notice, we also assumed that the markets
have innite depth. Therefore, regardless of how many buyers and sellers there are at
time t, we assume the price does not move. For example, in the case where t is equal
to one day, that would mean that the price only changes once each day. As we trim
down the size of this interval to the more accurate truth, milliseconds, it can become a
good model. In practice, orders are issued by traders at discrete times, where the stock
has a xed value for that small time period, even if that time period is only milliseconds.
Using the random walk model, we can visualize the basic idea of how the future
price S
t+t
is dependent on the current price S
t
,
S
t+t
= S
t
+
t

_
t (3)
where is the volatility constant, and
t
is a random variable that is equal to +1 or
1 with equal probability. Therefore, the model suggests that the price moves up or
down by

t at each time step t. This is the nature of the random walk.


However, this model is a bit too simple, and we address its shortcomings. The rst
shortcoming is that it does not account for drift. Drift is the idea that most assets,
although they can be risky, will tend to grow in value as time goes on. Therefore, we
introduce the drift term from 2.1,
S
t+t
= S
t
+ t +
t

_
t (4)
where is the expected rate of return on S per unit time.
Next, we address the second shortcoming. Our model in (4) permits prices to
become negative. As explained in 2.1, we can apply the lognormal property of stock
prices to counteract this. Thus, our model becomes,
log S
t+t
= log S
t
+ t +
t

_
t (5)
where is the expected relative increase in S per unit time, and
2
is the relative
increase in variance per unit time. This is our nal model. We observe that S
t+t
can
only take on two possible values, both of which depend on S
t
. We also notice that for
a very small interval t, we have that t

t, because lim
t0
+

t
t
= , and
so the drift term has a negligible eect. In (5), we have our Discrete-Time Geometric
Brownian Motion Model for the movement of the underlying asset S.
6
4 Binomial Models
4.1 Binomial Model for Stocks
The understanding of the the Binomial Model is intimately related to the Black-
Scholes Formula. The Brownian Motion dynamics outlined in the previous section
imply that if we slice a time period T time thinly into intervals of t, that it will
behave like a Binomial Tree, where we can reasonably assume only two directions of
movement for the stock price. The Binomial Model uses the assumptions detailed in
3.1. In this section, we will discuss the theoretical movement of a stock price using the
Binomial Model. Consider a stock whose initial price is S
0
. In the case of an option, we
select some discrete time interval t, and we work forward from the purchase date to
the expiration date. If the time to expiration is T, then we progress through N =
T
t
time steps.
We make a basic assumption: that at any time period, only two movements are
possible. Either the stock price will increase by a factor of u 1, or it will decrease by
a factor of 0 d 1. Therefore, if the current price is S
0
, then the future price will
be either S
up
= uS
0
, or S
down
= dS
0
. The probability of an upward move is q, and the
probability of a downward move is 1 q. An example is given in Figure 2. We use the
notion of volatility to compute these values,
u = e

t
d = e

t
=
1
u
and we include one more assumption. Namely, that risk-neutrality holds. This means
that the expected return on investing in stocks must be at least equal to the possible
return from the risk-free rate. So, given u, d, R, that:
q
R d
u d
Therefore, we observe that we can easily compute the stock price at a given time
step. At time step n = n
u
+ n
d
, we have n
u
upticks and n
d
downticks. Therefore, we
can quickly calculate:
S
n
= S
0
u
nun
d
(6)
4.2 Cox-Ross-Rubinstein Binomial Option Pricing Model
We are able to come up with a model for the price of the call option by working
backwards. Easiest, is to begin with the nal time step, the time of expiration. At the
time of expiration, the option either has an exercisable value, or it expires worthless.
We conclude that at expiration, we have:
C
T
= max {S
T
K, 0}
7
S
0
dS
0
uS
0
d
2
S
0
duS
0
udS
0
u
2
S
0
t 2t t t t
(
1

q
)
q
q
2
(
1

q
)
q
(
1

q
)
q
(
1

q
)
2
Figure 2: A two-step binomial tree model of stock price.
Working backwards, we soon uncover that exactly one time period prior to expiration,
we have a one-step binomial model, given in Figure 3.
C
tt
C
d
= max {0, dS K}
C
u
= max {0, uS K}
(1

q
)
q
Figure 3: Calculating the price of a call option exactly one period before expiration.
We are left to investigate the question of computing C
tt
. To do this, we invoke
a portfolio argument. Consider the risk-free rate r. We dene R := e
rt
. This means
that one has the ability to invest their money in a bond and collect interest rate r over
time period t, and R is the value of 1 dollar invested at the risk-free rate over time t.
We presume that B
0
= 1, and B
t
= B
0
e
rt
. We suppose that the portfolio has m
shares of stock S, and B dollars worth of bonds. In Figure 4, we create a binomial tree
model for this portfolio to illustrate.
8
t t t
(mS +B)
(mdS +RB)
(muS +RB)
(1

q)
q
Figure 4: Calculating the price of a portfolio of stocks and bonds using the one-step binomial
model.
Because we have modeled a world under exactly the same pretenses as the initial world
for obtaining the price of the call option one period before expiry, we can choose m and
B to replicate the payo of the call option. There absolutely must be some combination
of stock or cash (bonds) equal to the value of the call option. This is implied by our
no-arbitrage assumption. We construct the following linear system in m, B:
muS + RB = C
u
(7)
mdS + RB = C
d
(8)
solved using the substitution method, we get the unique solution,
m

=
C
u
C
d
(u d)S
, B

=
uC
d
dC
u
(u d)R
(9)
for which we conclude that a portfolio with m

shares and B

dollars in bonds will


result in the same cashow as the call option. Regardless of the movement of the
underlying, the portfolio and the call option will have the same payo. We have
assumed no arbitrage, so this should hold indenitely. In the Binomial Model, each
node has exactly two immediate children. Therefore, we say, for any parent node C:
C = m

S + B

=
C
u
C
d
(u d)S
S +
uC
d
dC
u
(u d)R
(10)
In the case of determining C
Tt
, the price of the call exactly one time step prior to
expiration, we plug in our max functions for C
u
, C
d
as shown in Figure 3.
C = m

S + B

=
C
u
C
d
(u d)S
S +
uC
d
dC
u
(u d)R
=
RC
u
RC
d
+ uC
d
dC
u
(u d)R
=
(R d)C
u
+ (u R)C
d
(u d)R
=
(Rd) max{0,uSK}+(uR) max{0,dSK}
(ud)
R
9
C =
Rd
ud
max {0, uS K} +
uR
ud
max {0, dS K}
R
(11)
Notably, this nal equation for C does not depend on q. Therefore, the probabil-
ity of the underlying assets movement in one direction or another is irrelevant to the
pricing of the option. This is why we can later relate this model to the Geometric
Brownian Motion discussed earlier. Referring to (5), we see that the random variable

t
is negligible. Consider in the binomial, the expected rate of return of the stock,
= uq +d(1 q) 1. This is dependent on q. Therefore, we see that we actually need
not even know the stocks expected rate of return in order to calculate the value of C.
The relevant values in this equation that are not givens are u, d. As shown earlier,
in order to compute u, d, we need the volatility, . We employ a probability trick.
We wish to apply risk-neural conditions and as a result, dene p :=
Rd
ud
, but only if
R d < u d, so as to restrict p to the domain (0, 1). We prove this inequality using
the no-arbitrage assumption.
Lemma 1.
d e
rt
u
Proof. This lemma is implied by the no-arbitrage assumption and the ability to short-
sell. We consider the rst case, that e
rt
> u. Then, the value of investing cash at the
risk-free rate for one time period exceeds the value maximum return on the stock in one
time period. Therefore, the purchase of a bond by short selling results in immediate
prot. This contradicts the no-arbitrage assumption. Also consider the second case,
that e
rt
< d. Then, investors can make a risk-free prot by purchasing the stock
while being nanced by credit. This also contradicts the no-arbitrage assumption.
Having proven Lemma 1, we may write:
C =
1
R
(pC
u
+ (1 p)C
d
) (12)
We note that 0 < p < 1 takes the form of a probability. Under risk-neutral circum-
stances, the expected return on the stock is equal to the risk-free rate of return. In
which case,
(uq + d(1 q))S = RS
Which can be rearranged to,
p = q =
R d
u d
(13)
As promised, we move backwards, calculating the value of the option now two periods
prior to expiration. Figure 2 shows the two-step binomial tree, which is indeed what
we use to track the stock price. Figure 5 shows the two-step binomial tree for the value
of the call option.
We recall (11), where we calculated C at one period prior to expiration. Applying
(12), we get:
10
C
C
d
C
u
C
dd
= max {0, d
2
S K}
C
du
= max {0, duS K}
C
ud
= max {0, udS K}
C
uu
= max {0, u
2
S K}
t 2t t t t
(
1

q
)
q
q
2
(
1

q
)
q
(
1

q
)
q
(
1

q
)
2
Figure 5: A two-step binomial tree model of the option price. The root, C, is representative
of C
t2t
C
u
=
pC
uu
+ (1 p)C
du
R
(14)
C
d
=
pC
du
+ (1 p)C
dd
R
(15)
We apply (10). Namely, that under the conditions of no arbitrage, we have that:
C = m

S + B

=
1
R
(pC
u
+ (1 p)C
d
) (16)
Plugging in (14) and (15), as well as using the values from Figure 5, we are able to get
the price of a parent node in terms of only its grandchildren:
C =
1
R
_
p
pC
uu
+ (1 p)C
ud
R
+ (1 p)
pC
du
+ (1 p)C
dd
R
_
=
1
R
2
(p
2
C
uu
+ 2p(1 p)C
ud
+ (1 p)
2
C
dd
)
C =
p
2
max
_
0, u
2
S K
_
+ 2p(1 p) max {0, duS K} + (1 p)
2
max
_
0, d
2
S K
_
R
2
(17)
We see in (17) the equation for C two periods prior to expiration. Just like (11), this
is dependent on only: S, K, u, d, R, and T. The process of repeated substitution for
11
periods 3, . . . , n periods prior to expiry will allow us to obtain the relevant theorem for
the value of a call option. We get
C
n
=
1
R
n
_
n

j=0
_
n
j
_
p
j
(1 p)
nj
max
_
0, u
j
d
mj
S K
_
_
(18)
Where
_
n
j
_
represents the number of possible paths the stock can take to reach a certain
node in the tree, p is the probability of an upward move under risk neutral conditions,
u is one plus the return per period on the stock if it goes up, d is one plus the return
per period on the stock if it goes down, and R is one plus the risk-free rate per period.
Therefore, the summation gives us the expected payout of the option at the time of
expiration, under risk-neutral conditions. The fraction
1
R
n
discounts the term to the
present value.
We can simplify a bit further. Dene g(j) = max
_
0, u
j
d
nj
S K
_
. We recognize
that for some j, g(j) = 0. Therefore, we dene a to be the smallest integer such that
u
a
d
na
S > K, and a n (19)
As a result
g(j) =
_
0 : j < a
u
j
d
nj
S K : j a
and applying this simplication, we get the theorem
Theorem 1 (Cox-Ross-Rubinstein Model, n periods prior to expiry).
C
n
=
1
R
n
_
n

j=a
_
n
j
_
p
j
(1 p)
nj
(u
j
d
nj
S K)
_
where u
a
d
na
S > K, and a n
5 Derivation of Black-Scholes as a Limit of the
Binomial Model
5.1 Restructuring the CRR Binomial Model in the Form
of Black-Scholes
In this section we will derive the Black-Scholes Model from the CRR Binomial
Model. The typical proof from Cox, Ross, and Rubinstein shows that in the binomial
model, as the number of time periods N , and the length of each time period
t 0, then the Cox-Rubinstein-Ross Binomial Model will converge to the Black-
Scholes Model. The proof given in [CRR] relies on a specic case of the Central Limit
Theorem. A much better and more general proof of the convergence of the binomial
model to the Black-Scholes model was written by Hsia (1983). In Hsias proof, we need
not make assumptions about the probability, p. We will give a detailed discussion of
this proof. We start by stating our objective, given by the Black-Scholes Model,
12
C = SN(d
1
) Ke
rT
N(d
2
) (20)
d
1
=
log
S
K
+ (r +
1
2

2
)T

T
(21)
d
2
=
log
S
K
+ (r
1
2

2
)T

T
(22)
where, S is the current stock price, K is the strike price, r is the risk-free rate, T is the
time until expiration, and
2
is the variance of the returns of the stock compounded
continuously. Now, we recall the result of our binomial model, given in Theorem 1.
We can divide the formula from Theorem 1 into two parts, giving
C
n
= S
_

n
j=a
_
n
j
_
p
j
(1 p)
nj
u
j
d
nj
R
n
_
KR
n
_
n

j=a
_
n
j
_
p
j
(1 p)
nj
_
(23)
We let the two terms in parentheses be dened by variables
B
1
=

n
j=a
_
n
j
_
p
j
(1 p)
nj
u
j
d
nj
R
n
(24)
B
2
=
n

j=a
_
n
j
_
p
j
(1 p)
nj
u
j
(25)
Noting that B
1
is now a binomial distribution. Which allows us to simplify to
C = SB
1
Ke
rt
B
2
(26)
The form of the CRR Binomial Model given by the equations of (24), (25), and (26)
is what will we use in our proof of convergence to the Black-Scholes Model.
5.2 Proof of Convergence of the CRR Binomial Model to
the Black-Scholes Model
Our goal is to show convergence to the Black-Scholes formula. In order to achieve
this goal, we need only show that B
1
and B
2
converge to N(d
1
) and N(d
2
). Before
beginning the proof, we will give a rough outline for it. Our initial goal is to show that
B
1
converges to d
1
. We will get a discrete value for the minimum number of upticks,
a, for a call to expire with positive value. We will invoke the DeMoivre-LaPlace Limit
Theorem and show that B
1
is a normal distribution. We will evaluate the expectation
and variance for the change in stock price over time T, and we will apply these in our
calculation of the size of the downtick, d. We will take the limit of d as our number
of intervals n . We will obtain the particular p, p

, the probability of an uptick


in each distribution, which give us the expectation of change in stock price which
we evaluated previously. Then we will apply properties of lognormally distributed
13
random variables to show that that B
1
converges to N(d
1
). We can then use p and
the expectation of change in stock price to show that B
2
converges to N(d
2
). To do
all of this, we begin by expanding (19) by dening it in our lognormal world.
u
a
d
na
S > K
a log u + (n a) log d + log S > log K
a log u + nlog d + log S > log K
a(log u log d) > log K log S nlog d
a >
log
K
S
nlog d
log
u
d
and we get a lower bound on a. However, we also require a to be an integer in the
iteration of the summation, therefore, we add to the computation in order to get an
integer. Therefore, we dene:
a =
log
K
S
nlog d
log
u
d
+ (27)
We wish to nd the number of upward ticks in order for the option to expire in the
money. Namely, the number of upward ticks such that the option has a positive payo.
We propose that we can introduce , because in the limit it will converge to 0, since
there will be an innite number of integer steps.
We conclude from the DeMoivre-LaPlace Limit Theorem that a binomial distri-
bution will converge to the normal distribution if np as n . Consider
B
1
as written in (24). Applying the DeMoivre-LaPlace Limit Theorem, we see that
in order to show that B
1
converges to a normal distribution, we need to show that
B
1

_

a
f(j)dj. Recall that j represents the number of upticks in the stock price,
and that a is the minimum number of upticks required for the call option to have value
greater than 0.
Lemma 2. B
1

_

a
f(j)dj
Proof. We seek to prove Lemma 2. We observe that f(j) is the probability density
function for a normal distribution. Because f(j) is not standard normal, we are required
to convert using z-scores. Namely, by dening, z =
jE[j]

j
. This would yield:
_

a
f(j)dj =
_

d
f(z)dz, where d =
j E[j]

j
However, Hsia cleverly denes
d =
(j E[j])

j
(28)
permitting us to write the equivalent expression
_
d

f(z)dz
14
So we write
B
1

_

a
f(j)dj =
_
d

f(z)dz, which is indeed equivalent to N(d)


We let S

be the price of the stock at the time of expiration. After n time periods
and j upticks, we know from (6) that
S

S
= u
j
d
nj
. Taking the log, we see some
familiar equations,
log
S

S
= j log u + (n j) log d
= j log
u
d
+ nlog d
We observe the expectation that we get from our binomial model,
E
_
log
S

S
_
= E[j] log
u
d
+ nlog d (29)
E[j] =
E
_
log
S

S
_
nlog d
log
u
d
(30)
Investigating the conditional variance will also help us classify the distribution. We
get
Var
_
log
S

S
_
=
_
Var[j]
__
log
u
d
_
2
(31)
yielding
Var[j] =
Var
_
log
S

S
_
(log
u
d
)
2
(32)
Applying (27), (28), (30) and (32), we explicitly write d. Earlier, in (28) we allowed
ourselves to express the downtick as,
d =
a +E[j]

j
Inserting (28), (30) and (32), we can write:
E[j] a =
E
_
log
S

S
_
+ log
S
K
log
u
d
log
u
d
,
j
=
_
Var(log
S

S
)
log
u
d
(33)
Which, we plug in to nd d,
d =
log
S
K
+E
_
log
S

S
_
log
u
d
_
Var
_
log
S

S
_
(34)
15
Now, we observe the properties of a binomial distribution. We conrm that for a
binomial distribution, with probability per outcome q, bin(n, j, q), that Var[j] = nq(1
q). We see that
_
Var[j] =
_
nq(1 q)
d =
log
S
K
+E
_
log
S

S
_
_
Var
_
log
S

S
_

log
u
d
_
q(1 q)

n
(35)
Where it is apparent that lim
n
log
u
d

q(1q)

n
= 0. Therefore, we eliminate the second
term. With just the rst term of (35), we have a lognormal process. We know that for
a lognormal process, we have that log
S

S
=
2
T. Giving
lim
n
d =
log
S
K
+E
_
log
S

S
_
_
Var
_
log
S

S
_
=
log
S
K
+E
_
log
S

S
_

T
We still need to obtain d
1
and d
2
as shown in the Black-Scholes Formula. We need
d = d
1
when the probability of an uptick is p

, and we need d = d
2
when the probability
of an uptick is p. Note that d here is referring to the distribution, not the size of the
downtick (also denoted by d throughout this paper). Therefore, we need to obtain p, p

such that:
E
_
log
S

S
_
=
_
(r +
1
2

2
)T : for probability p

(r
1
2

2
)T : for probability p
We begin by considering the rst case, p

. We recall from (13) that p =


Rd
ud
. And
so p

=
u
R
p. Solving for R, we get
R =
1
p

1
u
+ (1 p

)
1
d
(36)
where we recall that R = e
rt
. Therefore, if T = nt, then R
n
= e
rT
. Therefore,
R
n
= e
rT
=
_
1
p

1
u
+ (1 p

)
1
d
_
n
(37)
We now observe that
S
S

is the initial price of the stock, n time periods before expiration,


divided by the price of the stock at expiration. With S = S
0
, and S

= S
n
, this can
be written as the product:
S
S

=
S
0
S
1
S
1
S
2
. . .
S
n1
S
n
=
n

j=1
S
j1
S
j
(38)
16
and the expectation of this expression is
E
_
S
S

_
=
n

j=1
E
_
S
j1
S
j
_
(39)
where, we notice that we are able to make this claim because the variables are inde-
pendent. If we apply our denition of p

for B
1
, then we can write the expectation in
a way that only depends on p

, u, d, and n.
E
_
S
j1
S
j
_
= p

1
u
+ (1 p

)
1
d
(40)
E
_
S
S

_
=
n

j=1
_
p

1
u
+ (1 p

)
1
d
_
=
_
p

u
+
1 p

d
_
n
(41)
E
_
S
S

_
1
=
_
p

u
+
1 p

d
_
n
(42)
Now, we recall (37), which gives us e
rT
in terms of p

, u, and d. We can use (37) to


show
e
rT
= E
_
S
S

_
1
T log r = log E
_
S
S

_
(43)
Now, we are using the log of the expectation. We know from 2.1 that
S

S
will
be lognormally distributed. In addition, we know from the properties of logs that the
inverse,
S
S

, will also be lognormally distributed. We also must apply the lemma:


Lemma 3. For a lognormally distributed random variable X,
log E[X] = E[log X] +
1
2
Var[log X]
Proof. This follows from
S

S
= e
x
log
S

S
= x
from 2.1 we know
E[X] = e
+
1
2

2
log E[X] = +
1
2

2
which is equivalent to the statement of the lemma, since = E[log X] and
2
=
Var[log X].
log E[X] = E[log X] +
1
2
Var[log X]
17
Our next step will be to apply Lemma 3 to (43), this gives
T log r = log E
_
S
S

_
T log r = E
_
log
S
S

_
+
1
2
Var
_
log
S
S

_
T log r = E
_
log
S

S
_
+
1
2
Var
_
log
S

S
_
T log r = E
_
log
S

S
_
+
1
2
Var
_
log
S

S
_
(44)
which we conclude from linearity of expectation, and the fact that Var[log X] =
Var[log X]. We rearrange the terms to get
E
_
log
S

S
_
= T log r +
1
2
Var
_
log
S

S
_
(45)
and we recall, also from 2.1 that Var
_
log
S

S
_
=
2
T. We apply this equality and get:
E
_
log
S

S
_
= (log r +
1
2

2
)T (46)
If we recall that log r is the discounting rate. This shows that B
1
converges to N(d
1
).
The next task is to show that B
2
converges to N(d
2
). Recall,
B
2
=
n

j=a
_
n
j
_
p
j
(1 p)
nj
u
j
and that p =
Rd
ud
. We rearrange the terms in order to obtain
R = (pu + (1 p)d)
We know that S
j
= S
j1
u with probability p, and S
j
= S
j1
d with probability 1 p,
just as in Figure 2. Therefore,
E
_
S

S
_
=
n

j=1
E
_
S
j
S
j1
_
=
n

j=1
pu + (1 p)d
= (pu + (1 p)d)
n
= R
n
The result is that log E
_
S

S
_
= T log r. We apply (45) and get that
E
_
log
S

S
_
(47)
18
and we observe that Var
_
log
S

S
_
=
2
T, yielding the key result,
E
_
log
S

S
_
= (log r
1
2

2
)T (48)
where once again log r is the discounting rate, and so we have B
2
converging to N(d
2
).
Therefore, as the length of the time periods taken in the Binomial Model approach
zero and the number of time periods goes to innity, we have that the model converges
to Black-Scholes.
6 Conclusion
In this paper we have shown methods for option pricing in discrete time. Discrete-
time option valuation provides a better model than continuous-time option valuation,
because option trading is in fact a discrete-time process. However, some concerns
remain. One obvious drawback of the Geometric Brownian Motion Model of Black-
Scholes and the uptick/downtick model of the CRR Binomial Model is that they only
allow for the stock price to move in two ways. The inaccuracy of these models is
exacerbated when the stock price deviates greatly from the behavior models given.
Another issue with Black-Scholes and the Binomial Model is that they both require
the user to compute the parameter of volatility . Computing can, in fact, be a
computationally intensive procedure. Lastly, the models detailed in this paper make
several idealistic assumptions regarding market activities. For those interested, there
exists extensive literature and research whose aim is to relax the idealized assumptions
of the models explained in this paper. Those adjustments build immediately upon the
content of this paper.
References
[1] Chance, D.M., "Convergence of the Binomial to the Black-Scholes Model",
LSU.edu, 1-9.
[2] Cox, J. C., S. A. Ross and M. Rubinstein. "Option Pricing: A Simplied Ap-
proach." Journal of Financial Economics 7 (1979), 229-263.
[3] Hsia, C. C. "On Binomial Option Pricing." The Journal of Financial Research 6
(1983), 41-46.
[4] Hull, J. C., "Options, Futures, and Other Derivatives" Prentice Hall (2009), 280-
320.
[5] Jarrow, R. A. and A. Rudd, "Option Pricing." Homewood, Illinois: Irwin, 1983.
[6] Nielsen, L. T., "Understanding N(d
1
) and N(d
2
): Risk-Adjusted Probabilities in
the Black-Scholes Model", INSEAD (1992), 1-16.
[7] Rendlemen, R. and Bartter, B., "Two State Option Pricing." The Journal of
Finance 34 (1979), 1093-1110.
19

You might also like