Arbitrage: An Undergraduate Introduction To Financial Mathematics

Arbitrage
An Undergraduate Introduction to Financial Mathematics

J. Robert Buchanan
2010
J. Robert Buchanan Arbitrage
Efcient Market Hypothesis
The Efcient Market Hypothesis has many forms but
essentially can be taken to mean
prices on all securities and products reect all known
information,
current prices are the best unbiased estimate of the value
of the security or product,
prices will adjust to any new information nearly
instantaneously,
an investor cannot outperform the market using known
information except through luck.
information,
instantaneously,
information,
instantaneously,
information,
instantaneously,
Simple Arbitrage Situation
Arbitrage arises from mis-priced nancial instruments.
Example
CostCo sells 100 stamps for $43.75.
USPS sell 100 stamps for $44.00.
Intuitive Idea
Imagine we will bet on the outcome of an experiment.
The Arbitrage Theorem states that either the probabilities of
the outcomes are such that
all bets are fair, or
there is a betting scheme which produces a positive gain
independent of the outcome of the experiment.
Odds
The odds against an outcome X are related to probabilities of
the outcome according to the formula:
n : m against = P(X) =
m
m + n
.
The odds for an outcome X are related to probabilities of the
outcome according to the formula:
n : m in favor = P(X) =
n
m + n
.
For a wager of m dollars on a event X with odds against of
n : m, if X occurs, we win n dollars, otherwise we lose our
investment.
Example (1 of 2)
Example
Suppose the odds against player A defeating player B in a
tennis match are 3 : 1 and the odds against player B defeating
player A are 1 : 1.
P(A wins) = 0.25 and P(B wins) = 0.5
Determine a betting strategy which guarantees a positive net
prot regardless of the outcome of the tennis match.
Example (2 of 2)
Betting strategy: wager $1 on player A and $2 on player B.
A wins: gain $3 on the rst bet and lose $2 on the second,
net gain of $1.
B wins: lose $1 on the rst bet and gain $2 on the second,
net gain of $1.
There is a positive payoff no matter which player wins.
Introduction to Linear Programming
Linear programming is a branch of mathematics concerned
with optimizing a linear function of several variables subject to
some set of constraints (linear equalities or inequalities) on the
variables.
Example (1 of 3)
Example
A bank may invest its deposits in loans which earn 6% interest
per year and in the purchase of stocks which increase in value
by 13% per year. Any un-invested amount is simply held by the
bank. Suppose that government regulations require that the
bank invest no more than 60% of its deposits in stocks. As a
good business practice the bank wishes to devote at least 25%
of its deposits to loans. Determine how the bank should
allocate its capital so as to maximize the total return on its
investments.
Example (2 of 3)
Assume the bank can invest a fraction x in loans and
fraction y in stocks.
The total return is therefore 0.06x + 0.13y.
The constraints are:
0.25 x
0 y 0.60
x + y 1
Example (2 of 3)
0.25 x
0 y 0.60
x + y 1
Example (2 of 3)
0.25 x
0 y 0.60
x + y 1
Example (3 of 3)
Feasible Region
0.06 x 0.13 y k
0.0 0.2 0.4 0.6 0.8 1.0
0.0
0.2
0.4
0.6
0.8
1.0
x
y
Optimal return occurs when x = 0.4 and y = 0.6.
Linear Programming Notation (1 of 3)
If c and x are vectors with n components each, the notation
c
T
x = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
represents a weighted sum of the components of x with the
weights being the components of c.
Constraints will be expressed in the form a
T
x z.
a
T
x z (a)
T
x z
a
T
x = z a
T
x z and (a)
T
x z
If c and x are vectors with n components each, the notation
c
T
x = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
represents a weighted sum of the components of x with the
weights being the components of c.
Constraints will be expressed in the form a
T
x z.
a
T
x z (a)
T
x z
a
T
x = z a
T
x z and (a)
T
x z
We write u < v if u
i
< v
i
for i = 1, 2, . . . , n. Similarly for
u > v,
u v, and
u v.
If 0 denotes the zero vector then x 0 is an example of a sign
constraint.
We write u < v if u
i
< v
i
for i = 1, 2, . . . , n. Similarly for
u > v,
u v, and
u v.
If 0 denotes the zero vector then x 0 is an example of a sign
constraint.
The processes of maximizing and minimizing c
T
x are
equivalent in the sense that c
T
x is a maximum if and only if
(c)
T
x is a minimum.
Suppose there are m inequality constraints:
a
T
1
x b
1
a
T
2
x b
2
.
.
.
a
T
m
x b
m
The processes of maximizing and minimizing c
T
x are
equivalent in the sense that c
T
x is a maximum if and only if
(c)
T
x is a minimum.
Suppose there are m inequality constraints:
a
T
1
x b
1
a
T
2
x b
2
.
.
.
a
T
m
x b
m
General Linear Program
Ax =
_
_
a
11
a
12
a
1n
a
21
a
22
a
2n
.
.
.
.
.
.
.
.
.
a
m1
a
m2
a
mn
_
_
_
_
x
1
x
2
.
.
.
x
n
_
_

_
_
b
1
b
2
.
.
.
b
m
_
_
= b.
The general form of a linear program will be
Maximize c
T
x subject to the constraints Ax b and
x 0.
Feasible Vectors and Cost Functions
Denition
Vector x is feasible if x 0 and Ax b.
Denition
If c is a vector of n components, then we dene
c
T
x = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
to be the cost function.
Denition
Vector x is an optimal solution if x is feasible and maximizes
the cost function.
Denition
Denition
c
T
x = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
Denition
the cost function.
Denition
Denition
c
T
x = c
1
x
1
+ c
2
x
2
+ + c
n
x
n
Denition
the cost function.
Example (1 of 2)
Example
Use the notion of the intersection of planes in R
3
to minimize
5x
1
+ 4x
2
+ 8x
3
subject to x
1
+ x
2
+ x
3
= 1 and x is feasible.
Example (2 of 2)
0.0
0.5
1.0
x
1
0.0
0.5
1.0
x
2
0.0
0.5
1.0
x
3
Constraint Set
0.0
0.5
1.0
x
1
0.0
0.5
1.0
x
2
0.0
0.5
1.0
x
3
Constraint Set
5x
1
4x
2
8x
3
k
Slack Variables
Many optimization problems may include inequality
constraints.
Inequality constraints can be converted to equality constraints
by introducing slack variables. Thus
x
1
+ x
2
+ x
3
1 becomes x
1
+ x
2
+ x
3
+ x
4
= 1,
where x
4
0 and takes up the slack to produce equality.
Example (1 of 2)
Example
Use the notion of the intersection of planes in R
3
to minimize
5x
1
+ 4x
2
+ 8x
3
subject to x
1
+ x
2
+ x
3
1 and x is feasible.
Example (2 of 2)
If the constraints are x
1
+ x
2
+ x
3
1 and x feasible, then the
set of points where the solution must be found would resemble
a tetrahedron with vertices at (0, 0, 0), (1, 0, 0), (0, 1, 0), and
(0, 0, 1).
Feasible Region
5x
1
4x
2
8x
3
k
0.5
0.0
0.5
1.0
x
1
0.5
0.0
0.5
1.0
x
2
0.5
0.0
0.5
1.0
x
3
Matrix Notation for Slack Variables
Suppose A is an mn matrix, x is a vector of n components,
and b is a vector of m components, then by augmenting x with
m slack variables and A with the mm identity matrix the
inequality constraint Ax b is equivalent to
_
_
a
11
a
12
a
1n
1 0 0
a
21
a
22
a
2n
0 1 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
a
m1
a
m2
a
mn
0 0 1
_
_
_
_
x
1
x
2
.
.
.
x
n
x
n+1
x
n+2
.
.
.
x
n+m
_
_
=
_
_
b
1
b
2
.
.
.
b
m
_
_
_
A I
m
_
x
x
_
= b.
Canonical Linear Form
The general linear problem can now be stated in equivalent
form
maximize c
T
x subject to Ax = b and x 0,
where
A is an m(n + m) matrix consisting of the original
constraint matrix augmented with the identity matrix, and
x is the previous solution vector augmented with the slack
variables.
This new form of the linear problem will be called the canonical
form.
Dual Problems
For every linear programming problem of the type discussed
above, there is an associated problem known as its dual.
Henceforth the original problem will be known as the primal.
These paired optimization problems are related in the following
ways.
Primal: Maximize c
T
x subject to Ax b and x 0.
Dual: Minimize b
T
y subject to A
T
y c and y 0.
Observations
Primal: Maximize c
T
Dual: Minimize b
T
y subject to A
T
y c and y 0.
Note:
1
the process of maximization in the primal is replaced with
the process of minimization in the dual,
2
the unknown of the dual is a vector y with m components,
3
the vector b moves from the constraint of the primal to the
cost function of the dual,
4
the vector c moves from the cost of the primal to the
constraint of the dual,
5
the constraints of the dual are inequalities and there are n
of them.
Dual of the Dual
Theorem
The dual of the dual is the primal.
Proof
Starting with the dual problem,
Minimize b
T
y subject to A
T
y c and y 0.
We can re-write the dual in general form,
Maximize (b)
T
y subject to (A)
T
y c and y 0.
Now the dual of this problem (i.e., the dual of the dual) is
Minimize (c)
T
x subject to ((A)
T
)
T
x b and x 0.
This problem is logically equivalent to the problem
Maximize c
T
x subject to Ax b and x 0,
which is the primal problem.
Weak Duality Theorem
Theorem (Weak Duality Theorem)
If x and y are the feasible solutions of the primal and dual
problems respectively, then c
T
x b
T
y. If c
T
x = b
T
y then
these solutions are optimal for their respective problems.
Proof of Weak Duality Theorem
Feasible solutions to the primal and the dual problems must
satisfy the constraints Ax b with x 0 (for the primal
problem) and A
T
y c with y 0 (for the dual). Multiply the
constraint in the dual by x
T
x
T
A
T
y x
T
c c
T
x y
T
Ax.
Multiply the constraint in the primal by y
T
y
T
Ax y
T
b = b
T
y.
Directions of the inequalities are preserved because x 0 and
y 0. Combining these last two inequalities produces
c
T
x y
T
Ax b
T
y.
Therefore we have c
T
x b
T
y. If c
T
x = b
T
y then x and y must
be optimal since no x can make c
T
x larger than b
T
y and no y
can make b
T
y smaller than c
T
x.
Example (1 of 2)
Example
Primal: Maximize 4x
1
+ 3x
2
subject to x
1
+ x
2
2 and
x
1
, x
2
0.
Dual: Minimize 2y
1
subject to y
1
3, y
1
4, and y
1
0.
Example (2 of 2)
The minimum value of y
1
subject to the constraints must be
y
1
= 4. According to the Weak Duality Theorem then the
minimum of the cost function of the primal must be at least 8.
Applying the level set argument as before, the largest value of k
for which the level set 4x
1
+ 3x
2
= k intersects the set of
feasible points for the primal is k = 8.
Feasible Region
4x
1
3x
2
8
0.0 0.5 1.0 1.5 2.0
0.0
0.5
1.0
1.5
2.0
x
1
x
2
More on Duality
Theorem
Optimality in the primal and dual problems requires either
x
j
= 0 or (A
T
y)
j
= c
j
for each j = 1, . . . , n.
Proof
When x and y are optimal for their respective problems then
b
T
y = y
T
Ax = c
T
x
(y
T
A c
T
)x = 0
(A
T
y c)
T
x = 0.
Since x 0 and A
T
y c 0 then vector x must be zero in
every component for which vector A
T
y c is positive and vice
versa.
Example (1 of 4)
Example
Primal: Maximize c
T
x = 3x
1
+ 2x
2
x
3
+ 3x
4
subject to
x 0 and
_
1 1 1 0
2 0 1 1
_
_
_
x
1
x
2
x
3
x
4
_
_

_
5
3
_
Example (2 of 4)
Dual: Minimize b
T
y = 5y
1
+ 3y
2
subject to
_
_
1 2
1 0
1 1
0 1
_
_
_
y
1
y
2
_

_
_
3
2
1
3
_
_
Example (3 of 4)
1 2 3 4 5 6
2.0
2.5
3.0
3.5
4.0
4.5
5.0
y
1
y
2
Optimal solution is at (y
1
, y
2
) = (3, 3) and has value 24.
Example (4 of 4)
Strict inequality is present in the second and third constraints
since
y
1
= 3 > 2
y
1
+ y
2
= 0 > 1.
Thus the second and third components of x in the primal
problem must be zero. Therefore the primal can be recast as
Primal: Maximize 3x
1
+ 3x
4
subject to x
1
0, x
4
0 and
_
1 1 1 0
2 0 1 1
_
_
_
x
1
0
0
x
4
_
_
=
_
x
1
2x
1
+ x
4
_

_
5
3
_
Thus x
1
= 5 and x
4
= 13, the maximum of the cost function for
the primal is 24 and it occurs at (x
1
, x
2
, x
3
, x
4
) = (5, 0, 0, 13).
Duality Theorem
Theorem (Duality Theorem)
If there is an optimal x in the primal, then there is an optimal y
in the dual and the minimum of c
T
x equals the maximum of
y
T
b.
Remark: before proving the Duality Theorem we must state a
lemma which will be used in the proof.
Lemma (Farkas Alternative)
Exactly one of the following two statements is true. Either
1
Ax b has a solution x 0, or
2
A
T
y 0 with b
T
y < 0 has a solution y 0.
Duality Theorem
Theorem (Duality Theorem)
If there is an optimal x in the primal, then there is an optimal y
in the dual and the minimum of c
T
x equals the maximum of
y
T
b.
Remark: before proving the Duality Theorem we must state a
lemma which will be used in the proof.
Lemma (Farkas Alternative)
Exactly one of the following two statements is true. Either
1
Ax b has a solution x 0, or
2
A
T
y 0 with b
T
y < 0 has a solution y 0.
Proof (1 of 6)
Primal: Maximize c
T
Dual: Minimize b
T
y subject to A
T
y c and y 0.
Assuming there are feasible solutions to each problem then we
can re-write the constraint of the dual as (A)
T
y c with
y 0. Thus according to the constraint on the primal, the
re-written constraint on the dual, and the conclusion of the
Weak Duality Theorem the following inequalities hold for
x, y 0.
Ax b
(A)
T
y c
c
T
x b
T
y 0
Proof (2 of 6)
These inequalities can be written in the block matrix form
_
_
A 0
0 A
T
c
T
b
T
_
_
_
x
y
_

_
_
b
c
0
_
_
.
According to the Farkas Alternative Lemma either this
inequality has a solution x, y 0 or the alternative
_
A
T
0 c
0 A b
_
_
_
u
v
_
_
0 and
_
b
T
c
T
0

_
_
u
v
_
_
< 0
has a solution u, v, 0.
Proof (3 of 6)
We may decompose this block matrix to derive the following
system of inequalities:
A
T
u +c 0, Av b 0, b
T
u c
T
v < 0
with u 0, v 0, and 0. If > 0 then this system of
inequalities is equivalent to the following system.
A
_
1
v
_
b
A
T
_
1
u
_
c
b
T
_
1
u
_
> c
T
_
1
v
_
Since u 0 and v 0 the vectors
1
u 0 and
1
v 0 as well.
Proof (4 of 6)
The rst two inequalities form a primal problem and its dual.
Primal: A
_
1
v
_
b
Dual: A
T
_
1
u
_
c
If we apply the Weak Duality Theorem, then it must be the case
that b
T
_
1
u
_
c
T
_
1
v
_
, contradicting the inequality:
b
T
_
1
u
_
> c
T
_
1
v
_
Therefore we know that = 0.
Proof (5 of 6)
Thus the Farkas Alternative simplies to the following system:
Av 0, A
T
u 0, and b
T
u < c
T
v
where u 0 and v 0. The last inequality implies that
b
T
u < 0 or c
T
v > 0. If b
T
u < 0 then the primal problem Ax b
has no feasible solution x 0. To see this note that together
the inequalities x 0, Ax b, and b
T
u < 0 imply that
(Ax)
T
b
T
x
T
A
T
b
T
x
T
_
A
T
u
_
b
T
u < 0.
However, x 0 and A
T
u 0 and thus x
T
_
A
T
u
_
0, a
contradiction.
Proof (6 of 6)
If c
T
v > 0 then the dual problem A
T
y c has no feasible
solution y 0. To see this note that together the inequalities
y 0, A
T
y c, and c
T
v > 0 imply that
_
A
T
y
_
T
c
T
y
T
A c
T
y
T
(Av) c
T
v > 0
y
T
(Av) < 0
However, y 0 and Av 0 and thus y
T
(Av) 0, a
contradiction.
Fundamental Theorem of Finance (1 of 2)
Assumptions and background:
Experiment has m possible outcomes numbered 1 through
m.
We can place n wagers (numbered 1 through n) on the
outcomes.
r
ij
is the return for a unit bet on wager i {1, 2, . . . , n}
when the outcome of the experiment is j {1, 2, . . . , m}.
Vector x = (x
1
, x
2
, . . . , x
n
) is called a betting strategy.
Component x
i
is the amount placed on wager i .
Return from a betting strategy is

n
i =1
x
i
r
ij
.
Fundamental Theorem of Finance (2 of 2)
Theorem (Arbitrage Theorem)
Exactly one of the following is true: either
1
there is a vector of probabilities y = y
1
, y
2
, . . . , y
m
for
which
m
j =1
y
j
r
ij
= 0, for each i = 1, 2, . . . , n, or
2
there is a betting strategy x = x
1
, x
2
, . . . , x
n
for which
n
i =1
x
i
r
ij
> 0, for each j = 1, 2, . . . , m.
Proof (1 of 4)
Let the n-tuple (x
1
, x
2
, . . . , x
n
) be the betting strategy.
Vector x = x
1
, x
2
, . . . , x
n
, x
n+1
where the (n + 1)st
element is the payoff for the betting strategy.
We would like to maximize x
n+1
under the constraint that
n
i =1
x
i
r
ji
x
n+1
for j = 1, 2, . . . , m. This is equivalent to
n
i =1
x
i
(r
ji
) + x
n+1
0 for j = 1, 2, . . . , m.
c = 0, . . . , 0
. .
n components
, 1 and b = 0, . . . , 0
. .
m components
The maximization problem above can be stated as Maximize

c
T
x = x
n+1
subject to Ax b and x 0.
Proof (2 of 4)
Ax =
_
_
r
11
r
12
r
1n
1
r
21
r
22
r
2n
1
.
.
.
.
.
.
.
.
.
.
.
.
r
m1
r
m2
r
mn
1
_
_
_
_
x
1
x
2
.
.
.
x
n
x
n+1
_
_
0
0
.
.
.
0
_
_
.
The dual of this primal problem can be stated as Minimize
b
T
y = 0 subject to A
T
y c and y 0. The unknown y is a
vector of m components. Since b = 0 the cost function is
minimized at 0.
Proof (3 of 4)
The dual problem is that of minimizing 0 subject to y 0 and
_
_
r
11
r
21
r
m1
r
12
r
22
r
m2
.
.
.
.
.
.
.
.
.
r
1n
r
2n
r
mn
1 1 1
_
_
_
_
y
1
y
2
.
.
.
y
m1
y
m
_
_
=
_
_
0
0
.
.
.
0
1
_
_
.
This is equivalent to the system of equations,

m
j =1
y
j
r
ji
= 0 for
i = 1, 2, . . . , n;

m
j =1
y
j
= 1, and y
j
0 for j = 1, 2, . . . , m. The
dual problem is feasible with a minimum value of zero if and
only if y is a probability vector for which all bets have an
expected return of zero.
Proof (4 of 4)
Suppose the dual problem is feasible, the primal problem is
also feasible since x
i
= 0 for i = 1, 2, . . . , n + 1 satises the
inequality constraints of the dual. According to the Duality
Theorem then the maximum of the primal problem is zero
which means no guaranteed prot is possible. In other words if
(1) is true then (2) is false.
Now suppose the dual problem is not feasible. According to the
Duality Theorem the primal problem has no optimal solution.
Therefore zero is not the maximum payoff, thus there is a
betting strategy which produces a positive payoff. Thus we see
that if (1) is false then (2) is true.

Arbitrage: An Undergraduate Introduction To Financial Mathematics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Arbitrage: An Undergraduate Introduction To Financial Mathematics

Uploaded by

Copyright:

Available Formats

Arbitrage

An Undergraduate Introduction to Financial Mathematics

The maximization problem above can be stated as Maximize

You might also like