Understanding Betting Strategy

Kanika Saha1 and Ananya Lahiri†,2∗

1
Algo Lab CMI, India
2
Chennai Mathematical Institute, Chennai, India
arXiv:1706.01625v1 [stat.AP] 6 Jun 2017

05 June 2017

Abstract
In this paper, we present betting strategy of a football game using
probability theory. We know all betting houses offer slightly unfair
odds towards the player. Here we discuss a simple way to figure out
which betting house is offering relatively better odds compared to
others for English Premier League. However, this methodology can
be adopted for another league football match.

1 Introduction
It is believed that betting house always makes money in long run irrespec-
tive of their short term loss or gain. In this paper, we make an attempt to
understand this phenomenon with the concept of simple ‘expectation’ and
‘variance’ of probability theory. First, we will discuss what is fair game.
Let’s consider a simple game of English Premier League (EPL) where Manch-
ester United (ManU) is playing against Liverpool. Suppose a betting house
offers a game that if ManU wins with probability 0.606, then the player will
receive $0.65 from the betting house. On the other hand, if ManU loses with
probability 0.394 then the player has to pay $1 to the betting house.
Now player’s revenue scheme Rp is defined as follows:

0.65 with probability 0.606,
Rp =
−1 with probability 0.394,
and player’s expected revenue is
E(Rp ) = 0.65 × 0.606 − 0.394 = 0.
∗
†Corresponding author, E-mail: [email protected]

1
More about expectation and moments can be found in [1,2]. Now, we will
look at the revenue scheme and expected revenue for the betting house for
the same game.

−0.65 with probability 0.606,
Rb =
1 with probability 0.394.
Expected revenue for betting house is,
E(Rb ) = −0.65 × 0.606 + 0.394 = 0.
If we compare Rp and Rb , then we can see player’s loss is the gain of the
betting house and vice-versa. This type of game is known as ‘zero-sum’
game. Also, E(Rb ) = E(Rp ) = 0 means if the player and betting house
play this game several times, then ‘on-average’ neither betting house nor the
player will make or lose money. This is called ‘fair-game.’
Now we know that the betting house has an establishment cost and most
of the time these betting houses are the for-profit organization. Here the
question is how they are making money. An easy way of doing it, if the
betting house pays the player less than what they suppose to pay. That
means, in the fair game, if they think to pay the player $0.65, but to make
money, they will pay the player less than $0.65. For example, if they pay the
player $0.6 then the revenue scheme for the betting house is,

−0.6 with probability 0.606,
Rb =
1 with probability 0.394,
and the expected revenue for betting house is,
E(Rb ) = −0.6 × 0.606 + 0.394 = 0.0304.
It means if the player and betting house play this game several times, the
on-average betting house will make $0.03 or 3 cents from the player and the
player will lose the same amount because it is a zero-sum game. So to earn
money, in the long run, the betting house offers the player less than what is
fair. Rest of the paper is organized as follows. In section 2, we present the
strategy of betting houses using probability theory. In section 3, we discuss
the strategy for league football match. In section 4, we presented the data
analysis and showed how imputed cost of a betting house could be estimated
numerically.

2 Betting Strategy with Probability Theory

In this section, we present the strategy of the betting house. Suppose A is
an event with P (A) = p. If A happens then, the betting house will pay $r.

2
Otherwise, the betting house will receive $1. So the revenue scheme for the
betting house is

−r with probability p,
Rb =
1 with probability (1 − p),

and the expected revenue for betting house is,

E(Rb ) = −rp + (1 − p).

Now this game is fair game if E(Rb ) = 0, that is r = p1 − 1, and p1 is known as

‘decimal odds’. The variance of the revenue scheme is generally considered
as risk of a game. So for a fair game V ar(Rb ) = E(Rb2 ) − [E(Rb )]2 = E(Rb2 ).
Now

1 2
E(Rb2 ) = p − 1 + (1 − p)
p
h1 − p i
= (1 − p) +1
p
1−p 1
= = − 1 = r.
p p
Interestingly, for the fair game, r is the amount which betting house pays
for each dollar they receive. It turns out that r is also the measure of risk
for the same strategy as V ar(Rb ) = r, which is known as ‘fractional odds.’
To make a profit, in the long run, betting house pays $(r − ), where  > 0.
The revenue scheme is
−{( p1 − 1) − }

with probability p,
Rb =
1 with probability (1 − p),

and the expected revenue of betting house is,

1
E(Rb ) = −( − 1)p + p + (1 − p)
p
= −(1 − p) + p + (1 − p)
= p.

After simplification V ar(Rb ) = E(Rb2 ) − [E(Rb )]2 = r(1 − p)2 .

3
3 Strategy for League Football Match
There are three mutually exclusive outcomes in a league football match.
The outcomes are (i) Hometeam win, (ii) Away team win, (iii) Draw. Let
us consider the EPL match between ManU Vs. Liverpool, where ManU is
Hometeam. Suppose for this match betting house calculates the probabilities
0.6, 0.15 and 0.25 for three outcomes, namely Hometeam wins, Away team
wins and draw respectively. The corresponding decimal odds are 1/0.6 =
1.66, 1/0.15 = 6.66 and 1/0.25 = 4.00.
In the previous section, we noticed that the betting house would never
reveal these fair odds. They will always announce odds which are less than
the fair value so that they can stay in profit. For instance, if the betting house
offers odds of 1.57, 6.57 and 3.87 against the respective events of Hometeam
wins, Away team wins and draw, then the revised revenue scheme for ManU
to win the match with announced odds of 1.57 is,

H −(1.57 − 1) with probability 0.6,
Rb =
1 with probability 0.4.
The expected revenue for the betting house is,
E(RbH ) = −0.57 × 0.6 + 0.4 = 0.058.
Similarly, the revised revenue scheme for the Liverpool to win the match with
announce odds of 6.57 is,

A −(6.57 − 1) with probability 0.15,
Rb =
1 with probability 0.85,
and the expected revenue for betting house is,
E(RbA ) = −5.57 × 0.15 + 0.85 = 0.0145.
Likewise, the revised revenue scheme that match will be draw with announce
odds of 3.87 is,

D −(3.87 − 1) with probability 0.25,
Rb =
1 with probability 0.75,
and the expected revenue for betting house is,
E(RbD ) = −2.87 × 0.25 + 0.75 = .0325.
Expected revenue of betting house for all events are positive. Mathemat-
ically, we can show when the announce odds are converted to probabilities;
they usually add up to more than 1 to keep the house on benefit.

4
      
1 1 1 1 1 1
Let us assume, PH
− = PH∗ , PA
− = PA∗ and PD
− = PD∗

Therefore,
1 − PH 1
= ∗
PH PH
PH
= PH∗
1 − PH
PH < PH∗

In the same way we can show PA < PA∗ and PD < PD∗ . Now adding both
sides we can conclude,
PH + PA + PD < PH∗ + PA∗ + PD∗ .
Of course probabilities of a fair game sums up to 1 i.e. PH + PA + PD = 1,
hence evidently PH∗ + PA∗ + PD∗ > 1.

4 Data Analysis
Let us discuss the former analysis above with EPL data. Data is accessible
on http://www.football-data.co.uk/englandm.php. In the table (1)
we present the decimal odds of 10 different EPL matches from the popular
betting house Bet365. First three columns, ‘B365H’, ‘B365D’ and ‘B365A’
represent the decimal odds (i.e. 1/p) of three events that are (i) home team
wins, (ii) draw and (iii) away team wins; from the betting house Bet365. It is
visible that total probability of the three mutually exclusive events of a given
match is greater than one which violates the thumb rule that probabilities
add up to one. It is inevitable that sum of the probabilities will be greater
than one from our results described in the previous section. However, if there
are two betting houses, we can say that one house is offering better odds to
the players if its sum of probabilities is closer to one.
Let us consider a match held on 8th August 2015, between Bournemouth
and Aston Villa where the former team is Hometeam and later one is Away
team. We look at the betting houses for this match, the following table 5 (for
the six betting houses, namely Bet365 (B365), Bet&Win (BW), Interwetten
(IW), Ladbrokes (LB), William Hill (WH), VC Bet (VC) ). Here we can
compare the announce odds whichever is near to 1 that betting the house is
offering fair. As per the table 5, B365 is offering fair odds compared with
others. The betting house Interwetten is offering worst odds for this match.

5
4.1 Imputed Cost of a Betting House
We can show different ways of calculating Imputed cost of the betting house.
Here we are using additive model for the same. In this model, revenue of
betting house is described as,
−{( p1 − 1) − }

with probability p,
Rb =
1 with probability (1 − p).
As discussed earlier, to make a profit, betting houses will always offer an
amount which is less than what they suppose to pay. Here  is that prof-
itable amount. Suppose the betting houses announce odds are PH∗ , PA∗ and
PD∗ respectively for Hometeam, Away team and Draw of a match. These
announce odds can be defined as below:
 1  1  1  1  1  1
− = ∗, − = ∗, − = ∗
PH PH PA PA PD PD
Now we all know that probabilities of three mutually exclusive events add
up to 1 for fair game, i.e. PH + PA + PD = 1. But PH∗ + PA∗ + PD∗ > 1. Let
us say δ is the difference between sum of these two probabilities. Then

PH∗ + PA∗ + PD∗ − (PH + PA + PD ) = δ

PH∗ PA∗ PD∗
⇒ PH∗ + PA∗ + PD∗ − ( + + )=δ
1 + PH∗ 1 + PA∗ 1 + PD∗
PH∗ PA∗ PD∗
⇒ PH∗ − + P ∗
A − + P ∗
D − =δ
1 + PH∗ 1 + PA∗ 1 + PD∗
1 1 1
⇒ PH∗ (1 − ) + P ∗
A (1 − ) + P ∗
D (1 − )=δ
1 + PH∗ 1 + PA∗ 1 + PD∗
(PH∗ )2 (PA∗ )2 (PD∗ )2
⇒ + + = δ.
1 + PH∗ 1 + PA∗ 1 + PD∗
Clearly we can estimate  from the above equation as PH∗ , PA∗ , PD∗ and δ
are already known to us. Let us solve the equation for  using the given EPL
data. Considering the first match for B365 betting house where {PH∗ } = 0.50,
{PA∗ }=0.25, {PD∗ }=0.28 and δ=1.03-1=0.03. Now solving the above equation
with the help of this numerical values,
2
0.50 0.252 0.282
( 1+(0.50) + 1+(0.25)
+ 1+(0.28)
) = 0.03.
Here we can see the equation is cubic in . Instead solving analytically which
is complicated, we are introducing R programming to optimise . Therefore
calculate the value of  is 0.065 or following the revenue model we can say
$0.065 per dollar is the profitable amount for the betting house.

6
4.2 Comparative study of Imputed cost among differ-
ent betting houses:
Let us consider the result presented in Table (5) and figure (5).

1. All betting houses are limited to $0.10 seems like there is some regula-
tory policy

2. For 95% of the matches B365 charges between $.04 to $.06.

3. For majority games and approximately for all matches BW and IW

respectively charge $0.10.

4. Same behavior we can see for LB also.

5. But VC never charges greater than $.09. Moreover, for majority matches,
they charge between $.04 to $.06.

6. B365 and VC has similar kind of behavior.

5 Conclusion
In this paper, we presented how the strategy of the betting houses can be
explained with the probability odds. From the real data, it is visible that
for all betting houses considered here, the sum of the probability for the
win, loss, and draw of a match are greater than one. We also presented
how the cost of a game can be imputed numerically. Therefore, with this
understanding, one realizes how the betting houses are consistently making
money and accordingly decides whether to invest one’s money on the betting
house.

References
[1] Feller, W. An Introduction to Probability Theory and Its Applications.
Wiley Series in Probability and Statistics, Volume 1, 1968.

[2] Rao, C. R, Linear Statistical Inference and its Applications. Wiley Series
in Probability and Statistics, 2nd Edition ,ISBN: 978-0-471-21875-3, 2001.

Appendix: Figures and Tables

7
Figure 1: Imputed Cost of 380 matches by Different betting Houses

8
 Table 1: Table here
B365H B365A B365D P(B365H) P(B365A) P(B365D) Sum of prob
1 2.00 4.00 3.60 0.50 0.25 0.28 1.03
2 1.36 11.00 5.00 0.74 0.09 0.20 1.03
3 1.70 5.50 3.90 0.59 0.18 0.26 1.03
4 1.95 4.33 3.50 0.51 0.23 0.29 1.03
5 1.65 6.00 4.00 0.61 0.17 0.25 1.02
6 2.55 3.00 3.30 0.39 0.33 0.30 1.03
7 1.29 12.00 6.00 0.78 0.08 0.17 1.03
8 2.88 2.70 3.30 0.35 0.37 0.30 1.02
9 3.40 2.30 3.40 0.29 0.43 0.29 1.02
10 5.75 1.67 4.00 0.17 0.60 0.25 1.02

Table 2: Compare Betting Odds Across Betting Houses for the Same Match
Hometeam Awayteam Draw P(Hometeam) P(Awayteam) P(Draw) Sum of pro
B365 2.00 4.00 3.60 0.50 0.25 0.28 1.0
BW 2.00 3.70 3.30 0.50 0.27 0.30 1.0
IW 2.10 3.30 3.30 0.48 0.30 0.30 1.0
LB 2.05 4.00 3.30 0.49 0.25 0.30 1.0

Table 3: Summary of Imputed Cost of Six Betting Houses over the 380
League Matches
B365 BW IW LB WH VC
Mean 0.06 0.09 0.10 0.09 0.10 0.05
Median 0.06 0.10 0.10 0.09 0.10 0.05
Maximum 0.10 0.10 0.10 0.10 0.10 0.09
Minimum 0.03 0.04 0.09 0.04 0.06 0.03
Sd 0.01 0.01 0.00 0.02 0.01 0.01