1 Parameterization of Binomial Models and Derivation

of the Black-Scholes PDE.

Previously we treated binomial models as a pure theoretical toy model for our complete economy.
We turn to the issue of how to estimate parameters in the model and demonstrate the power of
the binomial model as a pricing tool – we will derive the famous Black-Scholes equation in this
binomial setting.

1.1 The Market:

Consider our stock:

Time at t, its value is S0

Time at t + δ, its pdf is p (S)

Here, we consider a very short time interval δ, i.e., δ ¿ 1.

therefore, at t = δ, the expected value of the stock price is

Z
E (S) = Sp (S) dS

and its variance is

£ ¤
V ar (S) = E (S − E (S))2
Z
= (S − E (S))2 p (S) dS

Next, we prescribe a model for our stock price movement:

1. Assumption 1: µ ¶
S
E = eµδ
S0
i.e., the rate of the expected return is µ.

1
 2. Assumption 2: µ ¶
S
V ar = σ2 δ
S0
where σ is the volatility.

N.B. Assumption 2 is a consequence of the following theorem, roughly stated, that if the total
volatility over time is bounded below away from zero and bounded above, and if there is always
some (finite) volatility over any time interval, then

V ar (S) ∼ δ

(see Neftci’s book, pp164-167).

Consider the binomial model

where p is a subjective probability.

1.2 Parameterization:
We want to match the mean and variance of this binomial model with those of our market, i.e.,

mean: pS0 u + (1 − p) S0 d = S0 eµδ

var: pS02 u2 + (1 − p) S02 d2 − [pS0 u + (1 − p) S0 d]2 = S02 σ 2 δ

N.B. V ar (X) = E (X 2 ) − [E (X)]2 . Therefore,

pu + (1 − p) d = eµδ (1)
pu2 + (1 − p) d2 − [pu + (1 − p) d]2 = σ 2 δ (2)

thus, we have 2 equations, 3 unknowns: p, u, d. There is a freedom of imposing constraints in our

parameterization. Eq. (1) gives
eµd − d
p=
u−d
Substitute this p into Eq. (2) yields

eµδ (u + d) − ud − e2µδ = σ 2 δ (3)

Next we are going to impose some specific constraints to obtain expressions for u, d, p.

2
1.2.1 The First Choice:
Impose
ud = 1.
To the accuracy of O (δ) , we have
√
u = e+σ√δ
d = e−σ δ

and the corresponding risk-neutral probability is

erδ − d
q =
u−d √
erδ − e−σ δ
= √ √
e+σ δ − e−σ δ
which is independent of µ.

The no-arbitrage principle dictates the following risk-neutral valuation of a contingent claim
with payoff f (ST ) :

The present value f0 = e−rδ [qf (S0 u, t + δ) + (1 − q) f (S0 d, t + δ)]

in which the rate of the expected market return µ does not appear.

A Deep Look at This Price: Since

f (S0 u, t + δ) = f (S0 + S0 u − S, t + δ)
= f (S0 + S0 (u − 1) , t + δ)
³ ³ √ ´ ´
= f S0 + S0 e+σ δ − 1 , t + δ ,

=⇒
n ³ ³ √ ´´ ³ ³ √ ´´o
f0 = e−rδ qf S0 + S0 e+σ δ − 1 + (1 − q) f S0 + S0 e−σ δ − 1

in which, for simplicity of notation, we have dropped t + δ. Taylor-expansion to O (δ) :

½ · ³ √ ´ 1 ∂2f ³ √ ´2 ¸
−rδ ∂f +σ δ 2 +σ δ
f0 = e q f (S0 ) + S0 e −1 + S e −1
∂S0 2 ∂S02 0
· ³ √ ´ 1 ∂2f ³ √ ´2 ¸¾
∂f −σ δ 2 −σ δ
+ (1 − q) f (S0 ) + S0 e −1 + S e −1
∂S0 2 ∂S02 0

in which all the derivatives are evaluated at time t + δ. Note that

3
 1. Since e−rδ [qS0 u + (1 − q) S0 d] = S0 , we have
√ √
qe+σ δ
+ (1 − q) e−σ δ
= erδ

2.
³ √ ´2
+σ δ
e −1 ≈ σ2δ
³ √ ´2
e−σ δ − 1 ≈ σ2δ

Therefore, accurate to O (δ) , we obtain

· ¸
−rδ ∂f rδ ∂f 1 2 2 ∂2f
f0 = e f (S0 , t + δ) + S0 e − S0 + σ δ S0 2
∂S0 ∂S0 2 ∂S0

Using
¡ ¢
e−rδ ≈ 1 − rδ + O δ 2
¡ ¢
e+rδ ≈ 1 + rδ + O δ 2

leads to · ¸
∂f 1 2 2 ∂2f
f0 = f (S0 , t + δ) + δ −rf (S0 , t + δ) + rS0 + σ S0 2 (4)
∂S0 2 ∂S0
accurate upto O (δ) . This formula gives the relation between the values of the contingent claim
at time t and time t + δ.

1.2.2 Black-Scholes Equation:

Finally, we note that Eq. (4) can be rewritten as

f (S0 , t + δ) − f0 ∂f 1 ∂2f
− rf (S0 , t + δ) + rS0 + σ 2 S02 2 = 0
δ ∂S0 2 ∂S0

As δ → 0, we arrive at the famous Black-Scholes PDE:

∂f ∂f 1 ∂2f
− rf + rS + σ2S 2 2 = 0
∂t ∂S 2 ∂S
where the subscript 0 is dropped.

4
1.2.3 The Second Choice:
1
p=
2

Now Eqs. (1) and (2) become

u + d = 2eµδ
u2 + d2 = 2σ 2 δ + 2e2µδ

To the order of O (δ), the solutions are

√
u = e(µ− 2 σ )δ+σ√δ
1 2

d = e(µ− 2 σ )δ−σ δ
1 2

For this binomial model, the associated risk-neutral probability is

√
e(r−µ+ 2 σ )δ − e−σ
1 2
erδ − d δ
q= = √ √
u−d e+σ δ − e−σ δ
It appears that all parameters, u, d, q depend on µ. Does the risk-neutral pricing also depend on
µ? Let us examine this question. Consider a contingent claim with payoff f (ST ) , according to
the no-arbitrage valuation, its present value is

f0 = e−rδ [qf (S0 u, t + δ) + (1 − q) f (S0 d, t + δ)]

n ³ ³ √ ´´ ³ ³ √ ´´o
= e−rδ
qf S0 + S0 e (µ− 12 σ 2 )δ+σ δ
− 1 + (1 − q) f S0 + S0 e(µ− 12 σ2 )δ−σ δ
−1 .

Again, for simplicity of notation, t + δ is dropped from the second line above. Taylor expansion
yields, up to O (δ) ,
½ · ³ √ ´ 1 ∂2f ³ √ ´2 ¸
−rδ ∂f (µ− 12 σ2 )δ+σ δ 2 (µ− 12 σ2 )δ+σ δ
f0 = e q f (S0 , t + δ) + S0 e − 1 + S0 2 e −1
∂S0 2 ∂S0
· ³ √ ´ 1 ∂2f ³ √ ´2 ¸¾
∂f (µ− 12 σ 2 )δ−σ δ 2 ( µ− 12 σ2 )δ−σ δ
+ (1 − q) f (S0 , t + δ) + S0 e − 1 + S0 2 e −1 (5)
∂S0 2 ∂S0

in which all the derivatives are evaluated at time t + δ. Note that

1. Again because e−rδ [qS0 u + (1 − q) S0 d] = S0 , we have

√ √
qe(µ− 2 σ )δ+σ + (1 − q) e(µ− 2 σ )δ−σ
1 2 1 2
δ δ
= erδ

5
2.
³ √ ´2
e (µ− 12 σ 2 )δ±σ δ
−1
· ·µ ¶ ¸ ¸2
1 2 √ 1 2
≈ 1+ µ− σ δ±σ δ + σ δ−1
2 2
³ √ ´2
= µδ ± σ δ
³ ´
2 3/2
= σ δ+O δ

Using the results in points 1-2 above, accurate to O (δ) , we obtain

· ¸
∂f 1 2 2 ∂2f
f0 = (1 − rδ) f (S0 , t + δ) + S0 rδ + σ δ S0 2
∂S0 2 ∂S0
· ¸
∂f 1 2 2 ∂2f
= f (S0 , t + δ) + δ −rf (S0 , t + δ) + rS0 + σ S0 2 (6)
∂S0 2 ∂S0
Note that
1. Due to the order shown in point 2 above, we only need to evaluate q upto O (1) to ensure
the entire expression (5) is valid up to the order O (δ) . Therefore,
√
e(r−µ+ 2 σ )δ − e−σ δ
1 2

q = √ √
e+σ δ − e−σ δ
· ¸
1 (µ − r) √
= 1− δ + O (δ)
2 σ
1 ³ ´
= + O δ 1/2
2
2. Eq. (6) is exactly the same price as we obtained in the case of the first choice, i.e., ud = 1.
In particular, we note that, although the parameters, u, d, q, now depend on µ, the rate
of the expected market returned. The risk-neutral pricing is still independent of µ – it
disappears into higher order terms, which will have no bearings on our no-arbitrage price.
3. We can again obtain the same Black-Scholes PDE, which describe how the value of a
contingent claim will evolve in time.
4. A real miracle: different parameterizations lead to the same Black-Scholes PDE under the
no-arbitrage principle for a given market.
5. For different markets (described by different values of µ), as long as the market volatility
σ is the same, we will get the same price for the option. The rate of the expected market
return is irrelevant. This is a far stronger result than the case where, for a given set of
u, d, no-arbitrage pricing is independent of p as in our theoretical binomial toy model of
economy.

6
1.3 The Simplest Binomial Model
Note that
√
e(r−µ+ 2 σ )δ − e−σ δ
1 2

q = √ √
e+σ δ − e−σ δ
· ¸
1 (µ − r) √
= 1− δ + O (δ) .
2 σ

If we choose
µ=r
then
1
q= + O (δ)
2
The O (δ)-term in q = 12 + O (δ) will not affect our risk-neutral pricing or the derivation of
the Black-Scholes PDE because its combined contribution with f (Su, t + δ) and f (Sd, t + δ) to
the price is higher order than O (δ) . Since p = 1/2, we can simply choose p = q = 1/2. For
risk-neutral valuation, the simplest binomial model for a short duration δ with a price accuracy
within O (δ) is

7
2 Multiperiod Binomial Trees

2.1 Setup:
1. 2 securities:

(a) a risky asset (e.g., stock without dividend);

(b) a risk-free asset (e.g., bond).

2. A series of times:

0, δt, 2δt, · · · , Nδt = T

at which trades take place.
3. A binomial tree of possible states for stock prices

and constant interest rate r (can be easily generalized to rk for time interval kδt.
1. Since the market permits no-arbitrage, we have
s2j < erδt sj < s2j+1 ∀j
This is an example of dynamically complete market.

2.2 Generalization:

8
 Recombinant Tree:

Note that: At time step n,

1. There are 2n states for the non-recombinant tree;

2. There are (n + 1) states for the recombinant tree.

This fact gives rise to the numerical advantage of recombinant trees.

Note that the parameterization with u, d gives a naturally recombined tree:

Risk-neutral valuation for each node:

fnow = e−rδt (qfup + (1 − q) fdown )

erδt Snow − Sdown
q =
Sup − Sdown
with the replicating portfolio:
fup − fdown
Stock : ∆ shares, ∆ =
Sup − Sdown
Bond : fnow − ∆Snow

Note that φ ≡ ∆.

9
2.2.1 Evaluation is just "working backward through the tree":

Example:

1
For simplicity, r = 0, and the price movement is assumed to be such that q = 2
at every node
for this tree.

What is the value of a European call option with K = 100 at T = 3δt?

Note that the value at maturity is
(ST − 100)+ = 60, 20, 0, 0
At the node with S = 140:
1 1
f = qfup + (1 − q) fdown = × 60 + × 20 = 40
2 2
Working backward, we conclude the value of the option is 15.

Why this is the correct price?

Because it can be replicated at every trading time.

The replication (in one possible path) – Consider that a bank sells the option:

1. At time t = 0,
25 − 5
∆ = = 0.5 shares of stock
120 − 80
stock : ∆ × S = 0.5 × 100 = 50
Bond : fnow − ∆ × S = 15 − 50 = −35

10
 i.e.,the back sells the option for $15
borrows $35
uses these $15 + $35 = $50
1
to buy shares of the stock.
2

2. Next step: say, the stock goes up to 120. the the new ∆ is
40 − 10
∆= = 0.75
140 − 100
Need additional 0.25 shares of stock to be purchased at the present value of stock, $120.

=⇒ Need $120 × 0.25 = $30 borrowed

Debt = $35 + $30 = $65.

3. If the stock, say, goes up to $140, then

60 − 20
∆ = =1
160 − 120
Need another 0.25 shares at $140/share :
0.25 × 140 = $35
Debt : $35 + $65 = $100

4. If the stock, say, goes down to $120 (at maturity), then,

The debt = $100, which matches the strike K

Portfolio: $120
|{z} − $100 = $20, which replicates the option claim.
1 Share of Stock

Thus, replicating the claim. That is, the bank sells the option at t = 0; At time t = T, it
could deliver one share of stock, collect K = $100, pay off the loan. No gain, no loss!

Conclusion: The portfolio is self-financing at each trading, i.e., the total value of the
portfolio before and after each trade are the same.

This can be again seen as follows:

For simplicity, assume interest rate r = 0. Notation: At tick-time i,

A portfolio Πi : ∆i+1 stock

ψi+1 = fi − ∆i+1 Si

1. Start:
Π0 : ∆1 S0 + ψ1 = f0

11
2. One tick:

Π1 : worth ψ1 + ∆1 S1
= ψ1 + ∆1 S0 − ∆1 S0 + ∆1 S1
= f0 + ∆1 (S1 − S0 )
f1 − f0
= f0 + (S1 − S0 )
S1 − S0
= f1

rebalance ∆ such that

f1 = ∆2 S1 + ψ2

3. Tick-time 2:

Π2 : worth ∆2 S2 + ψ2
= f1 + ∆2 (S2 − S1 )
f2 − f1
= f1 + (S2 − S1 )
S2 − S1
= f2

4. At t = T :

ΠT : ∆T −1 ST + ψT −1
= fT −1 + ∆T −1 (ST − ST −1 )
= fT
which produces the claim.

Thus, the replicating process is self-financing.

12
2.3 Valuation Formula:

2.3.1 Case of General Trees:

erδt S1 − S2
q1 =
S3 − S2
rδt
e S2 − S4
q2 =
S5 − S4
rδt
e S3 − S6
q3 =
S7 − S6
Therefore,

f (3) = e−rδt [q3 f (7) + (1 − q3 ) f (6)]

f (2) = e−rδt [q2 f (5) + (1 − q2 ) f (4)]

=⇒
f (1) = e−rδt [q1 f (3) + (1 − q1 ) f (2)]
= e−2rδt [q1 q3 f (7) + q1 (1 − q3 ) f (6)
(1 − q1 ) q2 f (5) + (1 − q1 ) (1 − q2 ) f (4)]

i.e.,

Initial value of a claim

X
= e−rNδt (the probability of the path associated with a final state i) × (payoff of state i)
final states

13
2.3.2 Case of Recombinant Trees

erδt − d
q =
u − d" #
f (1) = e−2rδt q2 f (7) + 2q (1 − q) f (5) + (1 − q)2 f (4)
⇑ recombined paths

For the N-step:

The present value of an option with payoff f (ST ):
N ·µ
X ¶ ¸
−rNδt N k N−k ¡ k N−k
¢
e q (1 − q) f S0 u d
k
k=0
µ ¶
N
where is the number of ways of having k steps up and N − k steps down in a total N
k
time-steps.
e.g., A European call has the present value:
N ·µ
X ¶ ¸
−rNδt N k N−k ¡ k N−k
¢
e q (1 − q) S0 u d −K +
k
k=0

14