Chapter 1

Introduction

1.1 Fundamental Concepts in Finance

Consider a hypothetical situation that the financial market is formed by n basic com-
ponents called financial assets/instruments, denoted by S1 , . . . , Sn . One basic math-
ematical model can be defined as follows:
Definition 1.1. (Discrete time financial model).
i) At initial time 0, the values of the n assets are S0 = (S10 , . . . , Sn0 )0 .
ii) At maturity time T . There are m sets of possible values, or scenarios. For
scenario i (1 i m), the value of asset S j (1 j n) is denoted as Sij . The
information of the assets at T can be represented by the matrix
1 1
S1 S2 . . . Sn1

S2 S2 . . . Sn2
13 23
3
S = S1 S2 . . . Sn . (1.1)

.. .. ..
. . .
S1m S2m . . . Snm

We assume that Sij 0 for all i and j, since an asset has no liability. Bear
in mind that superscript (i ) indicates the possible scenarios and subscript ( j )
indicates the index of assets. 

In this book, we consider the pricing problem:

pricing problem
Suppose that there is a financial instrument F taking value F i , i = 1, . . . , m, at
each of the m scenarios in (1.1), what is the price/value of F at time 0, say F 0 ?
Before we answer this question, we first define some terms.

1
2 1 Introduction

Definition 1.2. (Terms in Finance)

Portfolio. A portfolio is a collection of assets. A portfolio can be written as

nj=1 w j S j , where w = (w1 , . . . , wn )0 is a vector called weight, which indicates a
holding of w j units of asset S j in the portfolio.
Payoff Vector. The vector that represents the value of a portfolio/financial in-
strument at different scenarios at maturity is called the payoff vector, or simply
payoff. For examples, F = (F 1 , . . . , F m )0 is the payoff vector of the financial in-
strument F, and Sw is the payoff vector of the portfolio nj=1 w j S j .
Complete. A market is said to be complete if every payoff vector is attainable
by some portfolio, i.e., for all F Rm , there exists w Rn such that F = Sw.
Hedging. A hedge is an investment position intended to offset potential losses/gains.
For example, if a person is holding one unit of F and he thinks it is too risky, he
can short a portfolio of 0.8w, where w satisfies F = Sw, so as to offset 80% of
the potential losses/gains. (such a w can be found if the market is complete)
Arbitrage. An arbitrage is a portfolio w that has non-positive initial value (w0 S0 ,
a scalar) and non-negative value at maturity (Sw, an m dimensional vector), ex-
cluding the case w0 S0 = 0 and Sw = 0. Mathematically, we have
0 0 0 0
w S 0 and |{z}
|{z} Sw > 0 or w
|{z} Sw 0 ,
S < 0 and |{z} (1.2)
Price Payoff Price Payoff

where
Sw 0 means all entries of Sw are non-negative.
Sw > 0 means all entries of Sw are non-negative with at least one element
being positive.
Sw  0 means all entries of Sw are positive. (used in Definition 1.10)
In words, arbitrage (for the first case) means that we can hold a portfolio without
any cost initially, while the portfolio is valuable in some scenarios at maturity.
Roughly speaking, arbitrage is a free lunch. In finance we assume that arbitrage
does not exist. Otherwise, demand and supply force will drive the prices back to
normal.

Remark 1.1. (No Arbitrage). We say that an arbitrage opportunity exists if there
exists w Rn such that (1.2) holds. Thus, to check whether an arbitrage opportunity
exists, we can verify the negation of (1.2), i.e., for all w Rn ,
0 0
Sw > 0 w
|{z} S >0
|{z} and |{z} w0 S 0 0
Sw 0 |{z} (1.3)
Payoff Price Payoff Price

This negation follows from the fact that for any two statements A and B, the two
statements

(A and B) , and B A ,
1.2 Principle of No Arbitrage Pricing 3

are logically equivalent. Here the notation A denotes the negation of statement A.
For simplicity, in this book we will verify the former case of no-arbitrage, the latter
case can be handled similarly.

1.2 Principle of No Arbitrage Pricing

In finance, the Principle of No Arbitrage Pricing is the key for pricing financial
instruments. The key idea is: To price any financial instrument F, we find a portfolio
nj=1 w j S j such that its payoff matches that of F at every scenario at T . Having
the same payoff structure at T , F and nj=1 w j S j should have the same initial price.
Otherwise, one can make a profit immediately by buying the cheaper one and selling
the more expensive one (see Exercise 1.12). Therefore, the price F 0 is the initial
values of the replicating portfolio, w0 S0 . Mathematically, we do the following steps:

Principle of no arbitrage pricing:

Step 1. (Replication). Given a target payoff vector F. Find a portfolio w =

(w1 , w2 , . . . , wn )0 such that
1 1 1
. . . Sn1

F S1 S2 w1
F 2 S2 S2 . . . Sn2 w2
3 13 23
F , F = S1 S2
. . . Sn3 w3
, Sw . (1.4)
.. .. .. .. ..
. . . . .
Fm S1m S2m . . . Snm wn

That is, we replicate the payoff of F in every scenario by holding

(S1 , S2 , . . . , Sn )0 with weight (w1 , . . . , wn )0 . The portfolio w is known as the
replicating portfolio.
Step 2: (Pricing). The price of the instrument, F 0 , is given by
n
F 0 = w0 S0 = w j S0j . (1.5)
j=1

The weight w of the replicating portfolio is found by solving the system of equa-
tions (1.4) with m equations and n unknowns. When does a solution exist? Is the
solution unique? This reduces to a linear algebra problem. Observe from (1.4) that
for any F, the weight w exists if F is in the column space of S. Thus the column
space of S is the key to the pricing problem.
For simplicity, we impose the following assumption. For the general case please
refer to the supplementary materials.
Assumption 1.1 (Linearly Independent Columns of S). In the market defined in
Definition 1.1, m = n and S has linearly independent columns.
4 1 Introduction

If the columns of S are linearly dependent, then at least one of the columns can
be written as a linear combination of other columns. Physically, it means that some
assets can be constructed from other assets (their final payoffs are the same in all
scenarios). If this happens, then the asset is redundant and may be ignored. In other
words, we can always reduce S to have linearly independent columns. The assump-
tion m = n is imposed for the market completeness:
Theorem 1.2. (Replicating Portfolio)
Under Assumption 1.1, the inverse of S exists (denoted by S1 ). For any given payoff
vector F, the weight of replicating portfolio is given by

w = S1 F.

Thus the market is complete.

Proof. Recall from linear algebra that every square matrix (m = n) with linearly
independent columns is invertible. Thus Assumption 1.1 implies the existence of
S1 . Since a replicating portfolio w satisfies F = Sw, multiplying S1 both sides
gives the conclusion. 

1.3 State price vector and risk neutral probabilities

In this section we introduce a new concept called state price vector, which gives a
probabilistic interpretation of the pricing procedure. To motivate the idea of state
price vector, we consider the following pricing formula:
Corollary 1.3 (Pricing) Under Assumption 1.1, for any payoff F and asset struc-
ture S and S0 , the price is given by

F 0 , (S0 )0 S1 F. (1.6)

Proof. Recall from (1.5) that given the replicating portfolio w, the price is given
by F 0 = w0 S0 Substituting the weight w = S1 F from Theorem 1.2 into (1.5), and
noting that w0 S0 = (S0 )0 w (w0 S0 is scalar), the result (1.6) follows. 

Motivated by (1.6), if we define the state price vector as

0 = (S0 )0 S1 , (1.7)

(an m dimensional vector), then (1.6) can be written as

m
F 0 = 0F = jF j . (1.8)
j=1

From (1.8), can be regarded as a probability distribution that gives weights to

the scenarios (F 1 , . . . , F m )0 . The price F 0 is regarded as the expected value of F
1.3 State price vector and risk neutral probabilities 5

under the discrete probability measure (1 , . . . , m )0 . Now the question is Is

really a probability distribution? Recall that any probability must be non-negative
and the total probability is 1. Thus has to satisfy (i) > 0 and (ii) mj=1 j = 1 in
order to be a probability distribution.
Proposition 1.4 Under Assumption 1.1, if there is no arbitrage opportunity in the
market, then the state price vector given in (1.7) satisfies > 0.

Proof. Define 1k,m = (0, . . . , 0, 1, 0, . . . , 0)0 to be an m dimensional vector with all

entries equal 0 except the k-th entry equals 1. The No arbitrage condition (1.3)
implies that if the payoff structure is F = 1k,m (gain $ 1 when scenario k happens),
then the price F 0 = 0 F = k must be positive, i.e., k > 0. Repeat the same argu-
ment for k = 1, . . . , m, we have > 0. In fact, we have proved the stronger result
 0. 

Remark 1.2. (Interpretation of State Price Vector.) As a by-product of the above

proof, it can be seen that k is the price of a product with final payoff of $ 1 at
scenario (state) k. This explains the name State Price.

We have just seen that (i) > 0 holds. However, (ii) mj=1 j = 1 does not hold
in general. Consider the following counter-example:
Example 1.1. Suppose that the first asset S1 is a risk free asset (e.g. bond) with S10 = 1
and S1i = 1 + r for i = 1, . . . , m. Since this risk free asset yields 1 + r in all scenarios,
the matrix S takes the form
(1 + r) S21 . . . Sn1

(1 + r) S2 . . . Sn2
2
3 3
S = (1 + r) S2 . . . Sn

.. .. ..
. . .
(1 + r) S2 . . . Snm
m

In particular, the first column of S is the vector (1 + r)1m , where 1m is an m dimen-

sional vector of ones. From the definition of inverse, S1 times the first column of
S gives (1, 0, . . . , 0)0 , thus S1 1m = (1/(1 + r), 0, . . . , 0)0 . Combining with (1.7), we
have
m
(1.7) S10 1
, i = 0 1m = (S0 )0 S1 1m = = ,
i=1 1+r 1+r

which is the discount factor of the risk free asset. Here, = m

i=1 i 6= 1 unless
r = 0. 

Remark 1.3. Even if the risk free asset is not in the set of assets (S1 , . . . , Sn )0 , the
= 0 1m is by definition the price for a product with riskless payoff 1m . Thus
can be interpreted as a discount factor.
6 1 Introduction

Since (ii) does not hold, is not a probability distribution. However, it is inter-
esting to see from the above example that, = m i=1 i is the discount factor of the
risk free asset. If we rescale and rewrite (1.8) as
 0  
0 0 1 m
F = F= F= ,..., F, (1.9)

where = m e , / can be regarded as a probability distribution.

i=1 i , then

State Price Vector () and Risk Neutral Probabilities ():

e

Multiplying State Price Vector to the payoff F, i.e. 0 F gives the fair price.
, m i=1 i is the discount factor from time T to 0.
e , 1  0 is a probability distribution known as the

risk neutral probability.

Note that 0 = (S0 )0 S1 in (1.7) can only be defined when S1 exists, i.e., when
m = n. For general m, n, motivated by (1.7) and (1.8), we have the following defini-
tion.

Definition 1.3. (State Price Vector.) A state price vector is an m-dimensional vector
 0 satisfying

(S0 )0 = 0 S . (1.10)

The notation  means all the entries of are strictly positive. 

Remark 1.4. (Interpretation of Risk Neutral Probability) Rewrite (1.10), as (S0 )0 =

e 0 S where e = /. The j-th entry ( j = 1, . . . , n) of this equation is S0j =
i=1 i S j . If we regard m
m e i ei Sij /S0j as the expected gain of the j-th asset (under
i=1
risk neutral measure), then it can be seen that the gain equal to 1 for all assets.
This explains the name risk neutral.

From Definition 1.3, we have the following theorem under general situations,
without assuming m = n or completeness of the market.
Theorem 1.5. (Pricing)
Suppose that there exists a state price vector in the market. If a payoff F is replicable,
then the price is given by
F 0 = 0 F , Ee (F) ,
where Ee is the expectation under the risk neutral probability distribution .
e For
0
notational simplicity, sometimes we write F = E (F). Note that is not a proba-
bility distribution.

Proof. If F is replicable, then F = Sw for some w. By (1.3) and the principle of no

arbitrage pricing, the price is given by F 0 = (S0 )0 w = 0 Sw = 0 F. 
1.4 Physical Probability v.s. Risk Neutral Probability 7

Under Assumption 1.1 (m = n), the state price vector is constructed from the
principle of no-arbitrage pricing in (1.7). See Exercise 1.15. In fact, the opposite
also holds, i.e., if there is no arbitrage opportunity, then a state price vector exists.
However, the proof is much more difficult and is omitted.

Theorem 1.6. (No Arbitrage and State Price Vector). Consider the Discrete Time
Financial Model (1.1) where S has linearly independent columns. There is no arbi-
trage if and only if a state price vector in Definition 1.3 exists.

In practice, if one can show that every solution of in (1.10) has at least one
negative entry, then the existence of arbitrage opportunity is immediately implied
by Theorem 1.6. See Theorem 1.7d) and Exercise 1.18.

1.4 Physical Probability v.s. Risk Neutral Probability

Note that in the previous discussion the Physical probability about the price evolu-
tion has not been taken into account. This is because in the principle of No Arbitrage
Pricing, the asset structure (S0 and S) already controls the price of all replicable fi-
nancial instruments. To be precise, consider the following example.

Example 1.2. Suppose that the market has two assets S1 , S2 satisfying
   
1 1.2 10
S0 = , S= , P(Scenario 1 occurs) = 0.9 , P(Scenario 2 occurs) = 0.1.
1 1.2 1

Note that the first asset is a bond with 20% interest rate. Recall that a Call option is
a financial instrument that offers a chance to purchase the underlying asset for price
K (strike price) at time T . Consider the Call option F on the second asset S2 with
strike price K = 2. F is said to be a financial derivative of S2 . In the first scenario,
we can purchase S2 with $2 and sell it immediately for $10 to earn $8. In the second
scenario, we do nothing as the strike price is higher than the market price. Thus, the
payoff is given by  
8
F= .
0
Simple probability calculation suggests that the expected value of F at time T equals
to 8(0.9) + 0(0.1) = 7.2, discounting by the risk free rate gives the physical price
0 = $7.2/1.2 = $ 6. However, using (1.6), the risk neutral price can be calculated
Fpp
to be Frnp0 = (S0 )0 S1 F = $ 0.148 < 6.

It is tempting to think that F is worthy to buy if the price is less than $ 6. If F is

now selling at $ 5, say, should we buy it?
The answer is negative. Formally, we can argue that if F 0 = 5, there exists an
arbitrage opportunity (Exercise 1.14). To get more intuition, note that although it
seems to be a good deal to buy F at $ 5, we should not take that because S2 itself
is a much better deal! To see this, just notice that if you buy F, you pay $ 5 at the
8 1 Introduction

beginning and receive $ 8 or $ 0. but if you buy S2 , you pay $ 1 at the beginning and
receive $ 10 or $ 1 (pay less and get more).
In fact, in a complete market, every payoff structure can be replicated by trading
of assets {Si }i=1,...,n . In other words, the structure {S0 , S} controls the risk neutral
price of financial instruments with any payoff structure. If probability measure other
than the risk neutral measure (e.g. physical measure) is used for pricing, arbitrage
opportunities may occur.
To conclude, this example points out that pricing is governed by the pre-determined
structure of the assets, physically what will be happening is irrelevant. 

Example 1.3. In the previous example, S2 is a much better deal than F if F 0 =

5. In fact, one can check that the physical price of S2 should be $ (10(0.9) +
1(0.1))/1.2 = $ 7.58. How come it is selling at $ 1?
One may think in this way. In the computation of the physical price, it just
weights the payoff by the probability for each scenario but ignores the importance
of different scenarios. For example, if Scenario 2 is a state of bad health condition
and one needs $ 1.2 for the medical fee, then S1 , which guarantees a payoff of $ 1.2,
will worth more than S2 .
After all, the setting of asset price structure S0 and S boils down to how we
measure risk, which is a matter of taste/utility, and hence is not questionable unless
arbitrage opportunities exist. 

1.5 Binomial trees

One well known example of the discrete time financial model (1.1) is the binomial
model, or binomial tree:
Definition 1.4. (Binomial Model)
Binomial model is a discrete time financial model with n = m = 2. It consists of
two assets, bond (B) and stock (S) in the market
two scenarios, upward (u) and downward (d) movements of the stock (u > d).
Specifically,

erT su
 
S0 = (1, s)0 and S = . (1.11)
erT sd

In (1.11), s is the initial price of the stock and r is the continuously compounded
risk free interest rate for the bond, thus the accumulation factor is erT .
The properties of Binomial model are summarized in the following Theorem.
Theorem 1.7. In the binomial tree model,
a) (Replicating Portfolio). For any payoff structure F = (F 1 , F 2 )0 , (Fu , Fd )0 , the
replicating portfolio is given by
1.6 Multi-period Binomial models 9
 0
Fd u Fu d Fu Fd
w = S1 F = , ,
erT (u d) s(u d)

i.e., a holding of (Fd u Fu d)/erT (u d) unit of bonds and (Fu Fd )/s(u d)

units of stocks.
b) (Risk Neutral Probabilities). The state price vector is unique and is given by

0 1
0
= (S0 )0 S1 = erT d, u erT . (1.12)
erT (u d)

In particular, the discount factor is = 2j=1 j = erT and the risk neutral
probability is given by
0
erT d u erT

e=
, .
ud ud

c) (Pricing). The price of any instrument with payoff F = (Fu , Fd )0 is given by

Fu (erT d) + Fd (u erT )
F 0 = 0F = .
erT (u d)

d) (Arbitrage) Arbitrage opportunities exist if u > d > erT or d < u < erT .


Proof. Note that now m = n = 2 and u > d implies that Assumption 1.1 holds.
Thus, the explicit formulas in the previous sections can be used. First, a) follows
from Theorem 1.2. Next, it is easy to see that b) and c) follow from (1.7) and (1.8)
respectively. For d), suppose first that u > d > erT . Note that the the first entry of the
state price vector is negative. This implies that the payoff vector F = (1, 0)0 creates
an arbitrage opportunity (check). Intuitively, the stock now outperforms the bond in
every scenario, so we can short the bond and long the stock to obtain arbitrage. In
fact, from Theorem 1.2, the precise arbitrage portfolio is found to be w = S1 (1, 0)0 .
The case d < u < erT can be handled similarly. Alternatively, d) can be shown by
applying Theorem 1.6. 

1.6 Multi-period Binomial models

Obviously, a discrete time financial model is inadequate to describe the evolution of

an asset price, no matter how many scenarios are chosen, since the dynamic of the
price is not modeled. Therefore, to construct a more sophisticated market model, one
natural way is to bring together copies of binomial models to form a multi-period
Binomial model, or Binomial tree, see below.
10 1 Introduction

S0 u3

S0 u2

S0 u S0 u
up

S0 S0
dow
n
S0 d S0 d

S0 d 2

S0 d 3

Definition 1.5. (Multi-period Binomial Model).

In an N-period binomial model, the market is observable at time 0 = t0 < t1 < t2 <
< tN = T , where ti = i . Over each time interval [ti ,ti+1 ], the assets follow the
binomial model with a bond and a stock. In particular, the values of the bond and
stock at time ti are given by (Bi , Si ), where (B0 , S0 ) = (1, s) at time 0 and

Si1 u if market goes up
Bi = erti , Si = .
Si1 d if market goes down

Since at each time point (each branch of the tree) the market can go up or down,
the N-period multinomial model can handle 2N possible scenarios without specify-
ing 2N assets. For simplicity, sometimes we assume ud = 1. Thus some of the 2N
scenarios collapse to the same ones and there are only i + 1 distinct scenarios for
(Bi , Si ) at time ti (See the figure).

Remark 1.5. (State Price Vector at each branch). Suppose that at time ti a branch
of the tree takes value (erti , Si ). The binomial model at this branch can be expressed
in matrix form (1.11) by
 rt 
e i+1 Si u
S0 = (erti , Si )0 and S = . (1.13)
erti+1 Si d

Note that ti+1 ti = . From (1.7), the state price vector is given by
1
0 = (S0 )0 S1 =


er d, u er
er (u d)
= (u , d ) , (1.14)

say. By construction, the state price vector is common for all branches.
1.6 Multi-period Binomial models 11

From Remark 1.5, the pricing procedure can be performed by repeating the same
idea in the single period model at each branch of the tree. The procedure is known
as Backward Induction on tree.

Backward Induction on Binomial tree

Suppose that we want to price a product F where only the values of FN (the
price of F at the N-th period, i.e. time T ) at each of the 2N scenarios are known.
The pricing procedure proceeds as follows:
i) Set j = N. At each of the 2 j1 branches of the tree in the ( j 1)-th period,
use a single period model to compute Fj1 .
ii) Repeat step i) for j = N 1, N 2, . . . , 1.
iii) The price is given by F0 .

Example 1.4. Suppose that the asset price evolves according to the tree shown be-
low, where the values without brackets are the asset prices.
133.1
(0)
121
(0)
110 108.9
(2.725) (0)
100 99
(7.475) (5.45)
90 89.1
(12.225) (10.9)
81
(19)
72.9
(27.1)
To illustrate the method, suppose that the risk-free interest rate is zero. What is
the price of a three month American put option with strike price K = 100?
Solution: First note that this is a particular example of (1.13) where r = 0, S0 =
100, u = 1.1 and d = 0.9. From (1.14), it can be seen that the state price vector for
all branch is = (0.5, 0.5)0 . Using the backward induction of trees, we have:
1. The values of the option at time 3, reading from top to bottom, are 0, 0, 10.9
and 27.1, respectively.
2. At time 2, we must consider two possibilities: the value if we exercise the option
immediately, and the value if we hold the option until the next period. For the
top node it is easy since the values are zero in both cases. For the second node,
if we exercise the option, then the value is 1. On the other hand, if we hold the
option, then from the analysis of the single step binary model, the value of the
12 1 Introduction

option is the expected value under the risk-neutral probabilities of the claim at
time 3, i.e. 0 0.5 + 10.9 0.5 = 5.45. As 5.45 > 1, we should hold the option
and the value of the option should be 5.45. For the bottom node, using the same
reasoning, the value is max(19, 0.5(10.9 + 27.1)) = 19.
3. Now consider the two nodes at time 1. The value at the top node is max(0, 0.5(0+
5.45) = 2.725. For the bottom node, the option worths max(10, 0.5(5.45 +
19)) = 12.225.
4. Finally, at time 0, the price is given by max(0, 0.5(2.725 + 12.225)) = 7.475.
Thus, the option price at time 0 is 7.475. The price at each node is shown in the
brackets of the figure above.

To describe the details of the multi-period binomial model, we define the notion
of path.
Definition 1.6. (Path) For an N-period binomial model and t = 1, . . . , N, the path
t = (z1 , . . . , zt )0 is a t-dimensional vector specifying the evolution of the market up
to time t, where for j = 1, . . . ,t,

1 if market goes up,
zj =
1 if market goes down.

Notice that once N is given, we know t for t N.

Example 1.5. Using the dynamic of Example 1.4, if the stock price goes up in the
first two periods, then 2 = (1, 1)0 .In this case, we denote S1 ( 2 ) = 110, S2 ( 2 ) =
121; F1 ( 2 ) = 2.725, F2 ( 2 ) = 0. However, S3 ( 2 ) and F3 ( 2 ) are unknown.

The concept of risk neutral probability or state price vector can be generalized
from single period model to multiple period model by the following theorem.
Theorem 1.8. (State Price Vector and Pricing in Multi-period Binomial model)
Let F j , Nj , j = 1, . . . , 2N be the 2N possible payoffs of F at time N and their cor-
N
responding paths, respectively. Define the State Price Vector = ( 1 , . . . , 2 )0 ,
where
N
j j (Nj ) = u I{zt =1} + d I{zt =1}


t=1

is the state price associated with the path Nj = (z1 , . . . , zn )0 , IB is the indicator function
of the set B and (u , d ) are defined in (1.14). Given the payoff vector F =
N
(F 1 , . . . , F 2 )0 , the price of F at t = 0 is

2N
F0 = j F j , E (F) , (1.15)
j=1

where E is defined as the expectation under the discrete distribution .

1.6 Multi-period Binomial models 13

Proof. From the operation of the Backward reduction algorithm, the payoff F j is
multiplied by the discounted risk neutral probabilities u or d at each period, cor-
responding to the market going up or down along the path. The product of all the
N branches along the path Nj is exactly j . Summing over all the contributions of
the 2N payoffs gives (1.15). 

Example 1.6. (Risk Neutral Probabilities)

Summing over all the j s, we recover the N-period discount factor

2N N
, j = CkN uk dNk = (u + d )N = erN = erT ,
j=1 k=0

thus the Risk Neutral Probability vector in the N-period Binomial model is simply -

1
= .

t
u

Lastly, recall that hedging is an action to reduce risk by taking an investment

position intended to offset potential losses/gains. In a multi-period Binomial model,
the hedging of F can be performed by adjusting the replicating portfolio at the be-
ginning of each period.
Example 1.7. (Replicating Portfolio in Multi-period Binomial model)
Given that at period k the asset values are (Bk , Sk ) = (erk , s) and the value of an
instrument is Fk . Suppose that the two possible values of F at period k + 1 are found
to be Fk+1 = (Fk+1u , F d )0 . Then, using Theorem 1.7, the weight of the replicating
k+1
portfolio is given by
!0
d u Fu d Fu Fd
Fk+1 k+1 k+1 k+1
wk+1 = , . (1.16)
er(k+1) (u d) s(u d)

d
Note that s, Fk+1 u
and Fk+1 are known given the path k of the market. In other
words, to hedge for time k + 1, the weight wk+1 has to be specified at time k using
the information k . One can verify the correctness of the hedging by showing that
the hedging portfolio takes exactly the same value as Fk+1 at time k + 1 (Exercise).

We end this section with the general concept of Self-financing portfolio, which
means that there is no external infusion or withdrawal of money over time, from
which we can deduce the Binomial Representation, which describes the evolution
of the replicating portfolio with respect to the assets. In the following discussion,
Vk R, Sk Rn and Fk R are the values (at the end of period k) of the portfolio,
the assets and the financial product that we want to price, respectively. Note that
they represent the variables only, they are not the matrices that specify the value at
each state.
14 1 Introduction

Definition 1.7. (Self-Financing Portfolio)

Suppose that a portfolio V holds a weight wk on assets valued Sk at the k-th period
(during time t ((k 1) , k ]). By construction, the portfolio at time k is of value

Vk = w0k Sk . (1.17)

Immediately after time k (denoted by k + ), the investor changes the weight from
wk to wk+1 . The portfolio is said to be self-financing if

w0k+1 Sk = w0k Sk , (1.18)

i.e., the value of the portfolio remains unchanged when the weight is changed from
wk to wk+1 at time k + . 

Example 1.8. In this example we verify that the replicating portfolio in Example
1.7 is self-financing. Note that Definition 1.7 is for general multi-period models.
For multi-period Binomial model, using the notation of Example 1.7 we have Sk =
(Bk , Sk )0 . From the principle of no-arbitrage pricing, Fk is the value of the replicating
portfolio at the beginning of the (k + 1)-th period, implying that

Fk = w0k+1 Sk . (1.19)

On the other hand, since the portfolio V is constructed to replicate F, we have Vk =

Fk for all k. Combining with (1.17) and (1.19), we have w0k+1 Sk = Fk = Vk = w0k Sk ,
i.e., the self-financing property.

Example 1.9. (Binomial Representation). From (1.17) and (1.18), one can check
that the portfolio V satisfies

Vk+1 = Vk + w0k+1 (Sk+1 Sk ) , (1.20)

which describes how the portfolio depends on the underlying assets that replicate it.
Iterating the calculation gives

Vk+1 = Vk1 + w0k+1 (Sk+1 Sk ) + w0k (Sk Sk1 )

=
k
= V0 + w0j+1 (S j+1 S j ) , (1.21)
j=0

which can be interpreted as a discrete stochastic integral of the weight w with re-
spect to the asset values S.
1.7 Extensions to Continuous Model 15

1.7 Extensions to Continuous Model

It is clear that a continuous model is more appropriate to describe the real situations.
One natural way to achieve an approximately continuous model is to use a discrete
model with very small time periods. We will use the following particular multi-
period binomial model:
1. The time period of each interval is of length .
2. The cash bond takes the form Bk = er k .
3. The initial stock price is S0 .
4. The stock price process is modeled by a binomial tree. If the stock price at the
end of the k-th period is Sk , then over the next time period it moves to new
values
n o 
1 if up,
S(k+1) = Sk exp + Zk+1 , where Zk+1 = (1.22)
1 if down.

Remark 1.6. Note that the above model is a particular case in Definition 1.5, with

u = exp( + ) , and d = exp( ) . (1.23)

Roughly speaking, the stock is expected

to grow with the rate exp( ) with volatil-
ity (fluctuation) of size exp( ).

Without loss of generality, we assume T = 1. Similar to the previous section, we

partition the time interval [0, T ] = [0, 1] into n intervals of length , but later we let
0, n such that n = 1. From (1.22), it can be easily checked that at time
t = m (the end the of the m-th period),
( )
m
St = S0 exp m + Zk . (1.24)
k=1

Note that all the randomness of St is from m k=1 Zk . Note also that the physical prob-
ability for the stock to go up or down is not important. Substitute (1.23) into (1.14),
the risk neutral probabilities (without discounting) of going up (Zk = 1) and down
(Zk = 1) are respectively

r er e
u e , P(Zk = 1) , p = ,
e + e

d er , P(Zk = 1) = 1 p .

Recall from the basic calculation of Bernoulli random variable that for each k,

E(Zk ) = p (1 p) = 2p 1 ,
Var(Zk ) = p + (1 p) (2p 1)2 = 4p(1 p) .
16 1 Introduction

Normalizing m
k=1 Zk , we can rewrite (1.24) as
( !)
m
Zk (2p 1) p
St = S0 exp m + p 4p(1 p) + m(2p 1)
k=1 4p(1 p)
( )
m
2p 1 p
= S0 exp m + m + 4 p(1 p) Yk , (1.25)
k=1

where Yk , Zk (2p1) is a random variable with mean 0 and variance 1.

4p(1p)
Next we investigate the limit of (1.25) as 0. Using a second order Taylors
expansion around 0 on each of the four exponential functions in p (i.e. ex = 1 + x +
x2 /2 + . . .), we have the approximation

(r 12 2 ) +
p . (1.26)
2
See Exercise 1.20. Therefore, if we approximate the continuous time situation by
taking 0, we have
 
1 2p 1 1 1
p , + 2 r . (1.27)
2 2

For the limit of the random part m

k=1 Yk , we borrow two powerful theorems from
probability theory:
Functional Central Limit Theorem (FCLT)
Suppose that {yk } is a sequence independent random variables with mean 0
[nt]
and variance 1. Define the partial sum process SY (t) = k=1 Yk . Here, SY (t) is
a random function from t (0, ) to R. As n , the rescaled partial sum
satisfies
[nt]
SY (t) k=1 yk D
, Wt ,
n n

where Wt is the Brownian Motion (Wiener process) with index t.

Continuous Mapping Theorem

D
Suppose that Xn X as n and f () is continuous function on the space of
X. Then
D
f (Xn ) f (X) .

as n .
The precise definition of Brownian Motion will be given in the latter chapter.
One useful property is that Wt N(0,t).
1.7 Extensions to Continuous Model 17

Note that by construction, n = 1 and m = t. Thus we can write = n1 and

m = nt. Combining with (1.27), FCLT and the Continuous Mapping Theorem, the
limit of (1.25) is seen to be
  
D 1 2
St S0 exp r t + Wt , (1.28)
2

for which we call the Continuous Time Pricing Model, or the Black-Scholes
Model, for asset St .

Remark 1.7. i) The name pricing model comes from the fact that probabilistic
dynamic of St used for pricing is the risk neutral probability. Again, the phys-
ical probability is never taken into account.
ii) The continuous compounding Bt = ert is already a continuous time pricing
model for the bond.


Using the Continuous Time Pricing Model (1.28), we can readily price one par-
ticular kind of financial derivative: the derivative which payoff is a function of ST ,
the underlying price at maturity.
Theorem 1.9. (Pricing under Continuous Time Model) Consider a financial in-
strument F under the Continuous Time Pricing Model. If the payoff of F at T is
FT = f (ST ), then the price F 0 is given by
Z    
0 rT 1 2
F = E ( f (ST )) = e f S0 exp r t +x t (x) dx ,
2

where t (x) is the density of the Normal distribution N(0,t).

Proof. It follows from the definition of expectation E () and the definition of the
random variable ST given in (1.28). 

Example 1.10. (Black-Scholes Call Option Price).

Recall that the European Call option with pre-specified strike price K has a payoff
structure f (ST ) = (ST K)+ at maturity time T . Thus Theorem 1.9 is applicable,
and the price of the Call option is given by (after some straightforward but tedious
algebra)
Z     +
1
F 0 = erT S0 exp r 2 T +x K T (x) dx
2
= N(d1 )S0 N(d2 )KerT ,

hwhere  is the distribution

 N() i function of a standard Normal random variable, d1 =
S0 2

ln K + (r + /2)T / T and d2 = d1 T . This pricing formula is first
developed by Black and Scholes in 1973.
18 1 Introduction

Note that, this section determined the risk neutral probability distribution for St
for pricing purpose, which is analogous to evaluating the state price vector in the
discrete time models. The analog for other results such as Replicating Portfolios
and Representations requires more machinery in probability theory, for which we
will acquire in latter chapters.

1.8 Geometric Brownian Motion

A well-known model for the dynamic of financial assets is to describe the asset
return as a Geometric Brownian Motion:
dSt
= dt + dWt , (1.29)
St
where
St is the value (price) of a financial asset at time t.
dSt /St is the return (the % change in price).
is the drift.
is the volatility.
Wt is a Brownian Motion.
In fact, it can be shown that (1.29) is equivalent to the continuous time model (1.28).
Therefore, (1.29) can be regarded as another formulation for deriving the pricing
procedure in continuous time. Which such formulation, it is easier to price more
complicated products and constructing replicating portfolios.
However, there is considerable amount of probability theory behind (1.29), for
which we will try to fully understand before putting the equation into practical use.
For examples,
What is a Brownian Motion? How to understand Brownian Motion as a random
function? What are its properties?
What is the relationship between Wt and Wt+ ?
The Brownian motion satisfies Wt N(0,t). How to justify its existence?
What is it meant by differentiating a Brownian Motion in (1.29)?
Does the continuous analog of (1.21) exist? If yes, is it related to an integral
with respect to the Brownian motion? What is the meaning of integrating with
respective to a Brownian Motion?
This course aims to provide a solid mathematical treatment to mathematical finance
based on modern probability theories. Starting from the fundamental issue of pre-
cisely defining a random variable, we build on concepts including sigma field, prob-
ability measure, Lesbeque integration, conditional probabilities and expectations,
martingales, stopping times, Ito Calculus and stochastic integrals. After acquiring
all the relevant machineries, we can establish a unified theory for mathematical fi-
nance.
1.9 Exercises 19

1.9 Exercises

Exercise 1.10 Let (S11 , S21 , S12 , S22 ) = (1, 2, 3, 6)0 .

i)Write down the matrix S.
ii)For the portfolio with weight w = (3, 4)0 , find the payoff vector.
iii)Find a weight for the portfolio such that the payoff is F = (3, 9)0 .
iv) Can you find a weight for the portfolio with payoff F = (4, 9)0 ? Is the market
complete?
v) If the initial price is S0 = (1, 4)0 . Is there any arbitrage opportunity?
vi) If the initial price is S0 = (1, 3)0 . Is there any arbitrage opportunity?
Exercise 1.11 consider

83 8 3 3.001
S0a = (1, 1)0 , Sa = 3 10 , S0b = (1, 1, 1)0 , Sb = 3 10 9.999 .
79 79 9

Is there any arbitrage opportunity for the system S0a , Sa ? Is there any arbitrage op-
portunity for the system S0b , Sb ?
Exercise 1.12 Given that

2 3
S = 1 4 , F = (5, 5, 5)0 , S0 = (2, 1)0 .
5 0

i) How many assets are there in this market?

ii) How many scenarios are there in this market?
iii) What is the value of the 2nd asset under the 1st scenario?
iv) What is the value of the third asset under the 1st scenario?
v) Is the market complete? Why?
vi) What is the price of F at time 0?
vii) If F 0 = 3.5, does arbitrage opportunity exist? If yes, explain how could one
take the arbitrage opportunity.
viii) What is the replicating portfolio?
Exercise 1.13 Suppose that S1 , S2 have payoffs (10, 10, 3)0 and (9, 9, 2)0 in 3
possible scenarios at T .
i) If S0 = (1, 1)0 . Is there any arbitrage opportunity?
ii) If S0 = (1, 0.8)0 . Is there any arbitrage opportunity?

For any asset A with initial value A0 > 0 and payoff (A1 , . . . , Am )0 , one way to
quantify risk is by the criterion

Aj
Risk = min .
j=1...,m A0
20 1 Introduction

iii) Find the Risk for portfolio Va = S1 .

iv) Find the Risk for the portfolio Vb = S1 S2 .
v) Can you construct a portfolio using S1 and S2 with minimum risk?

Exercise 1.14 Find the replicating portfolio for F in Example 1.2. If F 0 = 5, state
how you take the arbitrage opportunity.
Exercise 1.15 Prove Theorem 1.6.
Exercise 1.16 Consider the eight combinations of the following three pairs of dis-
joint events:
i) {m > n}, {m < n}.
ii) { the market is complete},{ the market is not complete}
iii) {no arbitrage opportunity exists.}, { arbitrage opportunity does exist.}
For each of the eight cases, give an example of the market in terms of S, F and
S0 , with columns of S being linearly independent. (Hint: not all combinations are
possible.)
Exercise 1.17 Under the setting of Example 1.4 and Definition 1.7, write down the
value of Vk , Fk and Sk for different k along the path (1, 1, 1)0 . Do the Vk , Fk , Sk
satisfy equations (1.17) to (1.21)?

Exercise 1.18 Consider a future contract with strike price K = 105 and maturity
T = 1. Suppose the initial price of the underlying stock is S0 = 100 and the one-
period risk-free rate is r = 30%. The factor u and d are 1.2 and 0.9 respectively.
If we model the dynamic by the one-period binomial tree model, then prove that
arbitrage exists by
i) constructing a portfolio explicitly;
ii) using Theorem 1.6;
iii) using Theorem 1.7.

Exercise 1.19 Note that the choice of u = exp( z), where z = 1, is not arbi-
trary. Repeat the same derivation for (1.28) using u = exp( ( )q z) for q < 1/2 and
q > 1/2. What is the resulting limit?
Exercise 1.20 Show equation (1.26).

Exercise 1.21 Give the detailed calculations for Example 1.10.