You are on page 1of 31

The effects of past performance on

expected future performance.

.........................................................................

School of Mathematical and Physical Sciences

Department of Mathematics

Dissertation

Course: ................Financial Mathematics...............

Candidate Number: .......................204749............................

Title: ................The effects of past performance on expected

future performance....................

Supervisor: .................Dr Qi Tang...................

Date: ...........29 August 2019...............................

Number of Words: ............. ................................

1
In making this submission I declare that the information

contained on this cover sheet is correct and that the content

of this dissertation is my own work.

.........................................................................

Contents.

Abstract.

1. Introduction

2. Uses of Bayes’ theorem.

3. An alternative theorem.

4. Uses of this Theorem.

2
Abstract:

Bayes’ theorem is a well known theorem used to describe

the probability of an event using new evidence. It has many

applications, including medicine, artificial intelligence,

finance and gambling. The dissertation aims to show how

past performance can be used to determine future

performance and to discuss the use of Bayes’ theorem in

finance and gambling. Bayes’ theorem is stated. An

alternative theorem is proposed using the example of a

biased coin and looking at the probability of heads and tails.

Proof is given. The author concludes that his alternative

theorem has applications that Bayes’ theorem does not, for

example the probability of a team that is behind catching up

and the probability of a share price returning to its original

3
value. These are applications that Bayes’ theorem cannot do

on its own.

Introduction:

Bayes’ theorem is named after Reverand Thomas Bayes who

first used conditional probability in order to calculate limits

on an unknown parameter. [1] At the heart of the theorem is

the concept that existing predictions are effected by new

evidence. It is used in many spheres including artificial

intelligence. At it’s simplest the Baysian approach is a logical

way to use probability and reasoning to make decisions in

the face of uncertainty, for example in gambling.

In finance Bayes’ theorem can be used to calculate the risk of

4
lending money to potential borrowers. Bayes’ theorem

however can be used in many areas. Jeff Grover’s book

“Strategic Economic Decision Making: Using Baysian Belief

Networks to solve Computer Problems” contains many

examples of its application including gambling. The Baysian

belief network can enable casinos to maintain a balance

between winning and losing and therefore maintain

profitability.

When making a bet, whether on horses, the outcome of a

football match, or the fluctuation of a share price, Bayes’

theorem allows the bettor to anticipate a different outcome

due to changing circumstances. It allows the bettor to test the

probability of an outcome at any point in time according to

current evidence. [1]

Bayesian analysis is one of the popular methods of

calculating the probabilities of various outcomes in sports

5
betting. [2] The Bayes model remains an useful model for

Betting stratedgy. [3]

The aims of this dissertation are to show how past

performance can be used to generate predictions of future

performance, and to discuss the use of Bayes’ theorem in

finance and gambling.

The use of Bayes’ theorem will be discussed in chapter 2,

while an alternative theorem that the Author has devised is

proposed and explained in chapter 3. In chapter 4, uses for

this alternative theorem will be suggested.

6
Chapter 2: The uses of Bayes’

theorem.

When it comes to gambling, there are situations when future

performance is independent of past performance, such as a

coin that is known to be fair. And there are times when past

performance can tell you about future performance, such as

the performance of the stock market of a sports team. [1]

𝑝 𝑏 𝑎)𝑝(𝑎)
𝑝 𝑎 𝑏 =
𝑝(𝑏)

Using Bayes theorem you can adjust the expected probability

Of something happening from information about prior

occurrences.

You can also you it to work out the probability of something

being true via observation.

7
For example say you have a disease which has a frequency of

one in 100. And you have a 90% accurate test for the disease.

If the test is positive, how likely are you to have the disease?

Here p(a)=0.01 p(b)=0.01*.9+.99*.1=.108 p(b|a)=0.9

So p(a|b)= 0.9*0.01/.108=1/12.

Bayes theorem is often used for financial forecasting [4]

The Bayes’ theorem for continuous variables.

𝑓 𝑦 𝑥)𝑓(𝑥)
𝑓 𝑥 𝑦 =
𝑓 𝑦 𝑠 ) 𝑓 𝑠 𝑑𝑠

8
Chapter 3: An alternative theorem.

One question that can be asked is, when there are two

outcomes (a and b), and after a while one has happened

more then the other (a). Then what is the probability that

after a specified amount of time that it is b that has happened

more often?

This can apply to many areas. For example the probability of

a team that is behind a set of games overcoming their deficit.

In such a case the number of wins and losses could be used to

estimate the probability of each team winning or losing and

therefore for the team that is behind to catch up.

I propose an Alternative theorem.

9
.............................................................. ...........

Say you have a biased coin where the probability of heads

and tails are p and 1 - p, respectively and the value of p is

unknown. If after n flips, the numbers of heads exceeds the

number of tails by m, what is the probability that after t flips

that the number of head’s exceed the number of tails? How

does this vary when the variables are changed?

Let’s first assume that the initial possible probability

distribution for p is a uniform one.


!!! !!!
After n flips there have been heads and tails. We
! !

will review the probability of the number of tails being equal

to the number of heads after t flips.

10
If we fix t and m, what value of n will maximize the

probability of having the same number of tails and heads

after t flips?

....... .............................................................

.....

We look at the probability of there being m more heads then

tails after n flips, then at the probability that given this, that

the number of heads and tails will be equal after t flips.

............................................................... ..........

For an unbiased coin, the probability of m more heads then

!! !!! !!!
tails after n flips is where a = and b = .
(!! ∗ !! ∗ ! ! ) ! !

For a biased coin with a probability of p for heads the

probability of m more heads then tails after n flips


!
!! ∗ ! ! ∗ ! – !
is .
!! ∗ !!

11
To have the same number of heads and tails after t flips, there

!–!!!
has to be =
!

!!!!!
0.5t - a heads and = 0.5t - b tails.
!

The probability of the remaining t - n flips resulting in the

number of heads and tails being equal is

! – ! ! ∗ ! !.!! ! ! ∗ ( ! – ! ) !.!! ! !
!.!! ! ! ! ∗ !.!! – ! !

!–!!! !–!!!
!–! !∗! ! ∗(!–!) !
= !–!–! !–!!!
!∗ !
! !

Here p is unknown. We will assume that all possible values of

p from, 0 to 1 are of equal likelihood. How can we determine

what the probabilities involved are?

The observed frequency of an outcome can be used to

estimate the outcomes underlying probability.

12
This can be used to estimate to true probability of heads

becoming more common then tails at the end.

Bayes’ theorem states that the probability of a given b = the

probability of b given a multiplied by the probability of a

divided by the probability of b.

Whenever there is a head you can therefore adjust the

probability of p towards 1 and whenever there is a tails you

can adjust the probability of p towards 0.

Let L(a, b) mean that after a flips there have been b heads.

And Let L(a, b: d, f) mean that after a flips there have been b,

and after d flips there have been f heads, where a > d.

P(L(d,f)|L(a,b:d,f))= 1 =P(L(d,f)|L(a,b:d,f))

13
!!! !!!
If there have been heads and tails, then probability
! !

of their ending up the same number of heads and tails after t


!!!
flips is 𝑝(𝐿(𝑡, 𝑡/2)| 𝐿(𝑛, )) =
!

𝑚 + 𝑛 𝑡 𝑡
𝑃 𝐿 𝑛, 𝐿 𝑡, 𝑃 𝐿 𝑡,
2 2 2
𝑚 + 𝑛
𝑃 𝐿 𝑛, 2

𝑛 + 𝑚 𝑡
𝑃 𝐿 𝑛, 𝐿 𝑡,
2 2

!
Can be determined by the following. If you have heads and
!

! !!
tails there are ! ! possible ways to arrange them. The
! !∗ !
! !

number of ways to arrange the flips so that the first n flips so


!!!
contain heads is (the number of permutations for
!

!!!
choosing flips out of n.) (the number of permutations
!

!–!! !
for choosing flips out of t - n. )
!

𝑛! 𝑡 − 𝑛 !

𝑛 + 𝑚 𝑛 − 𝑚 𝑡−𝑛−𝑚 𝑡– 𝑛 + 𝑚
2 ! 2 ! 2 ! 2 !

14
𝑛! 𝑡 − 𝑛 !
=𝑛 + 𝑚 𝑛 − 𝑚 𝑡−𝑛−𝑚 𝑡– 𝑛 + 𝑚
2 ! 2 ! 2 ! 2 !

!!! !
This gives 𝑃 𝐿 𝑛, 𝐿 𝑡, =
! !

!! ! ! ! !
! ! ! ! ! ! !!!!! ! – ! ! !
! ! ! !
! ! ! !
=
!!
! !
!∗ !
! !

𝑡 !
! 𝑛! 𝑡 − 𝑛 !
2
𝑛 + 𝑚 𝑛 − 𝑚 𝑡−𝑛−𝑚 𝑡– 𝑛 + 𝑚
2 ! 2 ! 2 ! 2 ! ∗ 𝑡!

! ! ! !
! ! !! ∗ ! ! ∗ ! ! ! ! !! ! !
!
𝑝 𝐿 𝑡, = ! ! ! 𝑑𝑝 = ! ! =
! !∗ ! ! ∗ !!! ! !!!
! ! !

!!! !!!
!!! ! !! ∗ ! ! ∗ ! ! ! !
𝑃 𝐿 𝑛, = ! ! ! ! ! ! ! 𝑑𝑝 =
! !∗ !
! !

!!! !!!
!! ! ! !
! !
!!! !!! =
!!! ! ! ! (! ! !)
! !

𝑡 𝑚 + 𝑛
𝑃(𝐿(𝑡, )| 𝐿(𝑛, )) =
2 2

15
!!! ! !
! ! !, ! !, ! ! !,
! ! !
!!!
=
! ! !,
!

! !
! !! ! ! ! ! !
!

!!! !!! !–!!! !–!!! ! ! !
! ! ! ! ! ! ! ! ∗ !!
! =
(! ! !)

! !
! !!! ! !!! !
!
!!! !!! !–!!! !–!!!
! ! ! ! ∗ (! ! !)!
! ! ! !

Now, how does changing n affect this probability?

!!!!! ! ! !!! ! !
! ! !!!, ! !, ! ! !, ! ! !, ! !, ! ! !,
! ! ! ! ! !
!!!!!
− !!!
=
! ! ! ! !, ! ! !,
! !

! !
! (! ! !)! ! – ! ! ! !
!
!!!!! !–!–! !–!–!–! !–!–!!!
-
! ! ! !(! ! !)!
! ! ! !

! !
! (! ! !)! ! ! ! !
!
!!! !–! !–!–! !–!!!
! ! ! !(! ! !)!
! ! ! !

! !
– ! !!! ! ! ! ! ! ! !! – ! ! ! ! ! ! ! !–! !–!–! !
!
!–!!! !!!!! !–!–! !–!!!
=
! ! ! ! ! (! ! !)!
! ! ! !

16
!!! ! !
! !!! ! ! !, ! !, !! ! ! !, !!
! ! !
𝑃(𝐿(𝑡, − 𝑎)| 𝐿(𝑛, )) = !!!
.
! ! ! ! !,
!

!
𝑃 𝐿 𝑡, − 𝑎 =
!

! ! ! !
!! !! !! !! ! !! !
! !! ∗ ! ! ∗ !!! ! ! ! !
! ! ! 𝑑𝑝 = ! ! = ,
!! !∗ !! ! !! !∗ !! ! !!! ! !!!
! ! ! !

!!! !
𝑃 𝐿 𝑛, =
! (! ! !)

𝑛 + 𝑚 𝑡
𝑃 𝐿 𝑛, 𝐿 𝑡, − 𝑎
2 2

! !
If you have − 𝑎 heads and + 𝑎 tails there are
! !

!!
! ! possible ways to arrange them. The number of
!! !∗ !! !
! !

ways to arrange the flips so that the first n flips so contain


!!! !!!
heads is (the number of permutations for choosing
! !

flips out of n.) (the number of permutations for choosing

!–!! !
− 𝑎 flips out of t - n. )
!

𝑛! 𝑡 − 𝑛 !
𝑛 + 𝑚 𝑛 − 𝑚 ∗ 𝑡−𝑛−𝑚 𝑡– 𝑛 + 𝑚
2 ! 2 ! ( 2 − 𝑎)! ( 2 + 𝑎)!

17
𝑛! 𝑡 − 𝑛 !
=𝑛 + 𝑚 𝑛 − 𝑚 𝑡−𝑛−𝑚 𝑡– 𝑛 + 𝑚
2 ! 2 ! ( 2 − 𝑎)! ( 2 + 𝑎)!

!!! !
This gives 𝑃 𝐿 𝑛, 𝐿 𝑡, =
! !

!! ! ! ! !
!!! !!! !–!!! !–!!!
! !( ! !)! ( ! !)!
! ! ! !
=
!!
! !
!! !∗ !! !
! !

! !
! ! ! ! ! ! !! ! ! ! !
! !
!!! !!! !–!!! !–!!!
! !( ! !)! ( ! !)! ∗!!
! ! ! !

𝑡 𝑚 + 𝑛
𝑃(𝐿(𝑡, )| 𝐿(𝑛, )) =
2 2
!!! ! !
! ! !, ! !, ! !! ! !, ! !
! ! !
!!!
=
! ! !,
!

18
! !
!! ! ! ! ! !! ! ! ! ! !
! ! ∗
!!! !!! !–!!! !–!!! !!!
! ! ! !( ! ! !)! ( ! ! !)! ∗!!
! =
(! ! !)

! !
! ! ! ! ! !(!!!)! ! ! ! !
! !
!!! !!! !–!!! !–!!!
! !( ! !)! ( ! !)! ∗ (! ! !)!
! ! ! !

𝑡 𝑚 + 𝑛
𝑃(𝐿(𝑡, < )| 𝐿(𝑛, )) =
2 2

!.!"(! – ! ! !) ! !!!
!!!
𝑃(𝐿(𝑡, − 2𝑖)| 𝐿(𝑛, ))
! !

!!! ! !
!.!" ! – ! ! ! ! ! !, ! !, ! !! ! ! !, ! !!
! ! !
!!! !!!
! ! !,
!

𝑡 𝑡
!.!" ! – ! ! ! 2 – 2𝑎 ! 2 + 2𝑎 ! 𝑛+1 ! 𝑡 − 𝑛 !
!!! 𝑛+𝑚 𝑛−𝑚 𝑡–𝑛−𝑚
2 ! 2 ! 2 − 𝑎 ! 𝑡 – 𝑛2+ 𝑚 + 𝑎 ! ∗ 𝑡 + 1 !

19
20
Chapter 4: Uses for this alternative theorem.

This theorem has applications that Bayes’ theorem doesn’t.

This theorem can be used to work out the probability of a

team that is behind a set of games overcoming their deficit.

It can be used to determine the probability of a process that

has two possible outcomes and has had one of them happen

more then the other returning to having had the same

number of each out come happen.

It can be used to determine the probability of a share price

returning to it’s original value. These are applications that

Bayes’ theorem cannot do on it’s own.

21
Graphs.

t=40;

m=4;

clear f;

for I=m/2:(t-2)/2

n=2*I;

clear d;

d=0;

for a=0:floor (0.25 * (t-m-n))

b=factorial (t/2-2*a) * factorial (t/2+2*a);

b=b * factorial (n+1) * factorial (t-n);

c=factorial ((n+m)/2) * factorial ((n-m)/2);

c=c * factorial ((t-n-m)/2-2*a) * factorial ((t-

n+m)/2+2*a) * factorial (t+1);

d(a+1)=b/c;

22
end

f(n/2-1)=sum(d);

end

P=linspace(4,38,18);

plot(P,f);

title('graph for m=4');

xlabel('n');

ylabel('probability');

23
t=40;

m=6;

clear g;

for I=m/2:(t-2)/2

n=2*I;

clear d;

d=0;

for a=0:floor(0.25 * (t-m-n))

24
b=factorial(t/2-2*a) * factorial(t/2+2*a);

b=b * factorial(n+1) * factorial(t-n);

c=factorial((n+m)/2) * factorial((n-m)/2);

c=c * factorial((t-n-m)/2-2*a) * factorial((t-

n+m)/2+2*a) * factorial(t+1);

d(a+1)=b/c;

end

g(n/2-2)=sum(d);

end

P=linspace(6,38,17);

plot(P,g);

title('graph for m=6');

xlabel('n');

ylabel('probability');

25
t=40;

m=8;

clear h;

for I=m/2:(t-2)/2

n=2*I;

26
clear d;

d=0;

for a=0:floor(0.25 * (t-m-n))

b=factorial(t/2-2*a) * factorial(t/2+2*a);

b=b * factorial(n+1) * factorial(t-n);

c=factorial((n+m)/2) * factorial((n-m)/2);

c=c * factorial((t-n-m)/2-2*a) * factorial((t-

n+m)/2+2*a) * factorial(t+1);

d(a+1)=b/c;

end

h(n/2-3)=sum(d);

end

P=linspace(8,38,16);

plot(P,h);

title('graph for m=8');

xlabel('n');

ylabel('probability');

27
As you can see, increasing the value of m decreases the

chance that the number head will end up equal to or more

then the number of tails at the end.

The probability is maximized when n is just over half t.

28
Conclusion.

All in all, Bayes’s theorem has many uses in finance, gambling

and medicine.

The author’s alternative theorem also has additional uses.

29
References.

[ 1 ] Jeff Grover. Strategic Economic Decision Making- Using

Baysian Belief Networks to Solve Complex Problems.

Springer, NY, 2013.

[ 2 ] Pinnacle.com. Baysian Analysis and Sports Betting,

Pinnacle 2014 betting

articles. https://www.pinnacle.com accessed 29 August,

2019.

[ 3 ] M.G Londoño and A.R. Hassan Sports Betting Odds: A

source for Empirical Bayes.

http://www.eafit.edu.co/programas-

academicos/pregrados/ingenieria-matematica/practicas-

investigativas/Documents/sports-betting-odds.pdf

30
accessed on 29 August, 2019.

[4] https://www.investopedia.com/articles/financial-

theory/09/bayesian-methods-financial-modeling.asp

31

You might also like