PREDICTING EVENT RESULTS

SPORTS: HORSE RACING

Final year project

UNIVERSITY CARLOS III OF MADRID

HUPERIOTECHINCSSOLO
Department of Signal Theory and Communications

Author: Miriam Trujillo Sevillano

Tutor: Fernando Pérez Cruz
Title: Predicting results of sporting events: horse racing

Author: Miriam Trujillo Sevillano

Director: Fernando Pérez Cruz

THE COURT

President:

Vocal:

Secretary:

After the defense and reading of the Final Year Project was carried out on June 12, 2012
in Leganés, at the Higher Polytechnic School of the Carlos III University of Madrid, it is
agreed to grant it the QUALIFICATION of

VOCAL

SECRETARY PRESIDENT
Thanks

I would like to remember the people who have helped me get to the point where I
am today.

To begin with, I would like to thank my parents for all the support they have given
me, for not putting pressure on me at any time, giving me total freedom to do whatever I
wanted, and for helping me during the most difficult moments of my career.

To my brothers, for having been and continue to be an example to follow for me.
In addition to the help they have given me during many moments of my life. And, along
with them, my grandmother and my niece have given me moments of joy and relief in
the moments when I was most overwhelmed.

To my tutor, Fernando, for the help and dedication provided during the course of
the Project. And to Matilde, for assuming the role of the previous one, in his absence.

To my boyfriend, without whom this would not have been the same. Thank you
very much for being there at all times, and giving me the necessary encouragement to
trust myself.

To all my friends and classmates, for all the moments we have lived throughout
these years.
Thank you all.
Summary

Nowadays, sports betting is gaining great importance worldwide. It is increasingly

common for less specialized people to place bets, who see them as an escape route both
as fun and as a way to make a fortune.

In this Final Year Project we focus on horse racing betting. To do this, we will aim
to predict which horse will win in each race. And in this way, obtain benefits with our
bets.

We will build a database with South African races from February 2011 to June
2011, which have been extracted from the Formstar website [1].

Through the database, we will train a classifier using the Support Vector Machines
method, with the appropriate parameters, selected with cross validation. This way, when
we have new races, our system will be able to tell us the probabilities of each horse
winning the race, and we will be able to bet with this available information.
Index

Index

PREDICTING EVENT RESULTS SPORTS: HORSE RACING 1

Index 8
Table index XI
Index of Figures XIV
1. Goals 1
2. memory structure 2
CHAPTER 1 5
1.1. History of betting 5
1.2. Concepts used 6
1.3. Racing categories 9
Types of races depending on the terrain[8] 11
1.4. Types of bets 12
1.5. Bettor profile 16
Betting by age 17
Position to choose when we bet [17]. 17
1.6. Betting shops 17
1.6.1. ZEturf 18
1.6.2. Betfair 20
1.7. Related Posts 20
EPISODE 2 24
Model 24
2.1. Chosen model 24
2.1.1. Introduction 25
2.1.2. Support Vector Machine (SVM) for Classification 28
2.1.3.Forecasting race results 31
2.1.4. Empirical evaluation of the SVM/CL model 42
2.1.5. Conclusion 48
2.2. Applied model 48
2.2.1. Formstar website 49
 Index

2.2.2. Database 58
2.2.3. Inference 69
2.2.4. Odds and Q Factor 76
2.2.5. Kelly criterion 82
CHAPTER 3 91
Results 91
3.1. Choice of value of β 92
3.1.1. Probability of winning for a horse as a function of the value of β 92
For greater ease, let's show it graphically: 93
Horse 1 93
Horse 3 94
Horse 4 95
Horse 5 95
Horse 6 96
Horse 7 96
• Knight 11 98
• Knight 13 99
• Knight 15 100
horse 2 102
Horse 4 103
Horse 5 104
Horse 6 104
Horse 7 105
Horse 8 105
Horse 9 106
Horse 10 106
Horse 12 107
• Knight 14 108
3.1.2. Variation of gain as a function of β 110
3.1.3. Choosing the optimal value of β 114
CHAPTER 4 147
Conclusions and future lines of work 147
4.1. Conclusions 148
4.2. Future lines of work 149
CHAPTER 5 151
Budget 151
Index

5.1. Introduction 151

5.2. Personnel cost 152
5.3. Software cost 153
5.4. Hardware cost 153
5.5. Expendable material cost 154
5.6. Cost summary 154
References 157
Websites 157
Articles 159
5.1.
 Table index

Table index

CHAPTER 1. INTRODUCTION TO RACING

PREDICTING EVENT RESULTS SPORTS: HORSE RACING................................1
Index...................................................................................................................................8
Table index.......................................................................................................................XI
Index of Figures............................................................................................................XIV
1. Goals.......................................................................................................................1
2. memory structure....................................................................................................2
CHAPTER 1.......................................................................................................................5
1.1. History of betting.................................................................................................5
1.2. Concepts used......................................................................................................6
1.3. Racing categories.................................................................................................9
Types of races depending on the terrain[8]...............................................................11
1.4. Types of bets..........................................................................................................12
1.5. Bettor profile.....................................................................................................16
Betting by age...............................................................................................................17
Position to choose when we bet [17]........................................................................17
1.6. Betting shops.....................................................................................................17
1.6.1. ZEturf.........................................................................................................18
1.6.2. Betfair.........................................................................................................20
1.7. Related Posts.....................................................................................................20
Table index

EPISODE 2......................................................................................................................24
Model...............................................................................................................................24
2.1. Chosen model....................................................................................................24
2.1.1. Introduction................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...................................28
2.1.3.Forecasting race results....................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.............................................42
2.1.5. Conclusion..................................................................................................48
2.2. Applied model.......................................................................................................48
2.2.1. Formstar website........................................................................................49
2.2.2. Database..........................................................................................................58
2.2.3. Inference.....................................................................................................69
2.2.4. Odds and Q Factor.....................................................................................76
2.2.5. Kelly criterion............................................................................................82
CHAPTER 3.....................................................................................................................91
Results..............................................................................................................................91
3.1. Choice of value of β...........................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β.............92
For greater ease, let's show it graphically:............................................................93
Horse 1..................................................................................................................93
Horse 3..................................................................................................................94
Horse 4..................................................................................................................95
Horse 5..................................................................................................................95
Horse 6..................................................................................................................96
Horse 7..................................................................................................................96
• Knight 11............................................................................................................98
• Knight 13............................................................................................................99
• Knight 15..........................................................................................................100
horse 2.................................................................................................................102
Horse 4................................................................................................................103
Horse 5................................................................................................................104
Horse 6................................................................................................................104
Horse 7................................................................................................................105
Horse 8................................................................................................................105
Horse 9................................................................................................................106
Horse 10..............................................................................................................106
 Table index

Horse 12..............................................................................................................107
• Knight 14..........................................................................................................108
3.1.2. Variation of gain as a function of β..........................................................110
3.1.3. Choosing the optimal value of β..............................................................114
CHAPTER 4...................................................................................................................147
Conclusions and future lines of work.............................................................................147
4.1. Conclusions.....................................................................................................148
4.2. Future lines of work.........................................................................................149
CHAPTER 5...................................................................................................................151
Budget............................................................................................................................151
5.1. Introduction.....................................................................................................151
5.2. Personnel cost..................................................................................................152
5.3. Software cost...................................................................................................153
5.4. Hardware cost..................................................................................................153
5.5. Expendable material cost.................................................................................154
5.6. Cost summary..................................................................................................154
References......................................................................................................................157
Websites.....................................................................................................................157
Articles.......................................................................................................................159
Index of figures

Index of Figures

CHAPTER 1. INTRODUCTION TO RACING

OF HORSES
PREDICTING EVENT RESULTS SPORTS: HORSE RACING................................1
Index...................................................................................................................................8
Table index.......................................................................................................................XI
Index of Figures............................................................................................................XIV
1. Goals.......................................................................................................................1
2. memory structure....................................................................................................2
CHAPTER 1.......................................................................................................................5
1.1. History of betting.................................................................................................5
1.2. Concepts used......................................................................................................6
1.3. Racing categories.................................................................................................9
Types of races depending on the terrain[8]...............................................................11
1.4. Types of bets..........................................................................................................12
1.5. Bettor profile.....................................................................................................16
Betting by age...............................................................................................................17
Position to choose when we bet [17]........................................................................17
 Index of figures

1.6. Betting shops.....................................................................................................17

1.6.1. ZEturf.........................................................................................................18
1.6.2. Betfair.........................................................................................................20
1.7. Related Posts.....................................................................................................20
EPISODE 2......................................................................................................................24
Model...............................................................................................................................24
2.1. Chosen model....................................................................................................24
2.1.1. Introduction................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...................................28
2.1.3.Forecasting race results....................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.............................................42
2.1.5. Conclusion..................................................................................................48
2.2. Applied model.......................................................................................................48
2.2.1. Formstar website........................................................................................49
2.2.2. Database..........................................................................................................58
2.2.3. Inference.....................................................................................................69
2.2.4. Odds and Q Factor.....................................................................................76
2.2.5. Kelly criterion............................................................................................82
CHAPTER 3.....................................................................................................................91
Results..............................................................................................................................91
3.1. Choice of value of β...........................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β.............92
For greater ease, let's show it graphically:............................................................93
Horse 1..................................................................................................................93
Horse 3..................................................................................................................94
Horse 4..................................................................................................................95
Horse 5..................................................................................................................95
Horse 6..................................................................................................................96
Horse 7..................................................................................................................96
• Knight 11............................................................................................................98
• Knight 13............................................................................................................99
• Knight 15..........................................................................................................100
horse 2.................................................................................................................102
Horse 4................................................................................................................103
Index of figures

Horse 5................................................................................................................104
Horse 6................................................................................................................104
Horse 7................................................................................................................105
Horse 8................................................................................................................105
Horse 9................................................................................................................106
Horse 10..............................................................................................................106
Horse 12..............................................................................................................107
• Knight 14..........................................................................................................108
3.1.2. Variation of gain as a function of β..........................................................110
3.1.3. Choosing the optimal value of β..............................................................114
CHAPTER 4...................................................................................................................147
Conclusions and future lines of work.............................................................................147
4.1. Conclusions.....................................................................................................148
4.2. Future lines of work.........................................................................................149
CHAPTER 5...................................................................................................................151
Budget............................................................................................................................151
5.1. Introduction.....................................................................................................151
5.2. Personnel cost..................................................................................................152
5.3. Software cost...................................................................................................153
5.4. Hardware cost..................................................................................................153
5.5. Expendable material cost.................................................................................154
5.6. Cost summary..................................................................................................154
References......................................................................................................................157
Websites.....................................................................................................................157
Articles.......................................................................................................................159

Figura 3.1. Variation in the probability of winning for horse 1

PREDICTING EVENT RESULTS SPORTS: HORSE RACING................................1
Index...................................................................................................................................8
Table index.......................................................................................................................XI
Index of Figures............................................................................................................XIV
1. Goals.......................................................................................................................1
2. memory structure....................................................................................................2
 Index of figures

CHAPTER 1.......................................................................................................................5
1.1. History of betting.................................................................................................5
1.2. Concepts used......................................................................................................6
1.3. Racing categories.................................................................................................9
Types of races depending on the terrain[8]...............................................................11
1.4. Types of bets..........................................................................................................12
1.5. Bettor profile.....................................................................................................16
Betting by age...............................................................................................................17
Position to choose when we bet [17]........................................................................17
1.6. Betting shops.....................................................................................................17
1.6.1. ZEturf.........................................................................................................18
1.6.2. Betfair.........................................................................................................20
1.7. Related Posts.....................................................................................................20
EPISODE 2......................................................................................................................24
Model...............................................................................................................................24
2.1. Chosen model....................................................................................................24
2.1.1. Introduction................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...................................28
2.1.3.Forecasting race results....................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.............................................42
2.1.5. Conclusion..................................................................................................48
2.2. Applied model.......................................................................................................48
2.2.1. Formstar website........................................................................................49
2.2.2. Database..........................................................................................................58
2.2.3. Inference.....................................................................................................69
2.2.4. Odds and Q Factor.....................................................................................76
2.2.5. Kelly criterion............................................................................................82
CHAPTER 3.....................................................................................................................91
Results..............................................................................................................................91
3.1. Choice of value of β...........................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β.............92
For greater ease, let's show it graphically:............................................................93
Horse 1..................................................................................................................93
Horse 3..................................................................................................................94
Index of figures

Horse 4..................................................................................................................95
Horse 5..................................................................................................................95
Horse 6..................................................................................................................96
Horse 7..................................................................................................................96
• Knight 11............................................................................................................98
• Knight 13............................................................................................................99
• Knight 15..........................................................................................................100
horse 2.................................................................................................................102
Horse 4................................................................................................................103
Horse 5................................................................................................................104
Horse 6................................................................................................................104
Horse 7................................................................................................................105
Horse 8................................................................................................................105
Horse 9................................................................................................................106
Horse 10..............................................................................................................106
Horse 12..............................................................................................................107
• Knight 14..........................................................................................................108
3.1.2. Variation of gain as a function of β..........................................................110
3.1.3. Choosing the optimal value of β..............................................................114
CHAPTER 4...................................................................................................................147
Conclusions and future lines of work.............................................................................147
4.1. Conclusions.....................................................................................................148
4.2. Future lines of work.........................................................................................149
CHAPTER 5...................................................................................................................151
Budget............................................................................................................................151
5.1. Introduction.....................................................................................................151
5.2. Personnel cost..................................................................................................152
5.3. Software cost...................................................................................................153
5.4. Hardware cost..................................................................................................153
5.5. Expendable material cost.................................................................................154
5.6. Cost summary..................................................................................................154
References......................................................................................................................157
Websites.....................................................................................................................157
Articles.......................................................................................................................159
 Index of figures

Figure 3.12. Variation in the probability of winning horse 12

PREDICTING EVENT RESULTS SPORTS: HORSE RACING................................1
Index...................................................................................................................................8
Table index.......................................................................................................................XI
Index of Figures............................................................................................................XIV
1. Goals.......................................................................................................................1
2. memory structure....................................................................................................2
CHAPTER 1.......................................................................................................................5
1.1. History of betting.................................................................................................5
1.2. Concepts used......................................................................................................6
1.3. Racing categories.................................................................................................9
Types of races depending on the terrain[8]...............................................................11
1.4. Types of bets..........................................................................................................12
1.5. Bettor profile.....................................................................................................16
Betting by age...............................................................................................................17
Position to choose when we bet [17]........................................................................17
1.6. Betting shops.....................................................................................................17
1.6.1. ZEturf.........................................................................................................18
1.6.2. Betfair.........................................................................................................20
1.7. Related Posts.....................................................................................................20
EPISODE 2......................................................................................................................24
Model...............................................................................................................................24
2.1. Chosen model....................................................................................................24
2.1.1. Introduction................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...................................28
2.1.3.Forecasting race results....................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.............................................42
2.1.5. Conclusion..................................................................................................48
2.2. Applied model.......................................................................................................48
2.2.1. Formstar website........................................................................................49
2.2.2. Database..........................................................................................................58
2.2.3. Inference.....................................................................................................69
Index of figures

2.2.4. Odds and Q Factor.....................................................................................76

2.2.5. Kelly criterion............................................................................................82
CHAPTER 3.....................................................................................................................91
Results..............................................................................................................................91
3.1. Choice of value of β...........................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β.............92
For greater ease, let's show it graphically:............................................................93
Horse 1..................................................................................................................93
Horse 3..................................................................................................................94
Horse 4..................................................................................................................95
Horse 5..................................................................................................................95
Horse 6..................................................................................................................96
Horse 7..................................................................................................................96
• Knight 11............................................................................................................98
• Knight 13............................................................................................................99
• Knight 15..........................................................................................................100
horse 2.................................................................................................................102
Horse 4................................................................................................................103
Horse 5................................................................................................................104
Horse 6................................................................................................................104
Horse 7................................................................................................................105
Horse 8................................................................................................................105
Horse 9................................................................................................................106
Horse 10..............................................................................................................106
Horse 12..............................................................................................................107
• Knight 14..........................................................................................................108
3.1.2. Variation of gain as a function of β..........................................................110
3.1.3. Choosing the optimal value of β..............................................................114
CHAPTER 4...................................................................................................................147
Conclusions and future lines of work.............................................................................147
4.1. Conclusions.....................................................................................................148
4.2. Future lines of work.........................................................................................149
CHAPTER 5...................................................................................................................151
Budget............................................................................................................................151
 Index of figures

5.1. Introduction.....................................................................................................151
5.2. Personnel cost..................................................................................................152
5.3. Software cost...................................................................................................153
5.4. Hardware cost..................................................................................................153
5.5. Expendable material cost.................................................................................154
5.6. Cost summary..................................................................................................154
References......................................................................................................................157
Websites.....................................................................................................................157
Articles.......................................................................................................................159

Figure 3.23. Variation in the probability of winning for horse 8

PREDICTING EVENT RESULTS SPORTS: HORSE RACING................................1
Index...................................................................................................................................8
Table index.......................................................................................................................XI
Index of Figures............................................................................................................XIV
1. Goals.......................................................................................................................1
2. memory structure....................................................................................................2
CHAPTER 1.......................................................................................................................5
1.1. History of betting.................................................................................................5
1.2. Concepts used......................................................................................................6
1.3. Racing categories.................................................................................................9
Types of races depending on the terrain[8]...............................................................11
1.4. Types of bets..........................................................................................................12
1.5. Bettor profile.....................................................................................................16
Betting by age...............................................................................................................17
Position to choose when we bet [17]........................................................................17
1.6. Betting shops.....................................................................................................17
1.6.1. ZEturf.........................................................................................................18
1.6.2. Betfair.........................................................................................................20
1.7. Related Posts.....................................................................................................20
EPISODE 2......................................................................................................................24
Model...............................................................................................................................24
2.1. Chosen model....................................................................................................24
Index of figures

2.1.1. Introduction................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...................................28
2.1.3.Forecasting race results....................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.............................................42
2.1.5. Conclusion..................................................................................................48
2.2. Applied model.......................................................................................................48
2.2.1. Formstar website........................................................................................49
2.2.2. Database..........................................................................................................58
2.2.3. Inference.....................................................................................................69
2.2.4. Odds and Q Factor.....................................................................................76
2.2.5. Kelly criterion............................................................................................82
CHAPTER 3.....................................................................................................................91
Results..............................................................................................................................91
3.1. Choice of value of β...........................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β.............92
For greater ease, let's show it graphically:............................................................93
Horse 1..................................................................................................................93
Horse 3..................................................................................................................94
Horse 4..................................................................................................................95
Horse 5..................................................................................................................95
Horse 6..................................................................................................................96
Horse 7..................................................................................................................96
• Knight 11............................................................................................................98
• Knight 13............................................................................................................99
• Knight 15..........................................................................................................100
horse 2.................................................................................................................102
Horse 4................................................................................................................103
Horse 5................................................................................................................104
Horse 6................................................................................................................104
Horse 7................................................................................................................105
Horse 8................................................................................................................105
Horse 9................................................................................................................106
Horse 10..............................................................................................................106
Horse 12..............................................................................................................107
 Index of figures

• Knight 14..........................................................................................................108
3.1.2. Variation of gain as a function of β..........................................................110
3.1.3. Choosing the optimal value of β..............................................................114
CHAPTER 4...................................................................................................................147
Conclusions and future lines of work.............................................................................147
4.1. Conclusions.....................................................................................................148
4.2. Future lines of work.........................................................................................149
CHAPTER 5...................................................................................................................151
Budget............................................................................................................................151
5.1. Introduction.....................................................................................................151
5.2. Personnel cost..................................................................................................152
5.3. Software cost...................................................................................................153
5.4. Hardware cost..................................................................................................153
5.5. Expendable material cost.................................................................................154
5.6. Cost summary..................................................................................................154
References......................................................................................................................157
Websites.....................................................................................................................157
Articles.......................................................................................................................159
 Introduction

Introduction

Today the world of betting encompasses a large number of users who see in them
an opportunity to make money.

Our task will be to predict the winner of each horse race to obtain a profit. We are
going to use probability estimation methods, which will be useful when calculating the
possibility that each horse has of being the winner of a certain race.

If the above estimate deviates from the actual values used to establish the
winnings, there is the possibility of making a profit by betting on those that are lower
than the estimated probabilities.

1. Goals

The main objective of this Final Year Project is to build a system that is capable
of, based on a database with a certain number of races, predicting the results of
subsequent races.

1
Introduction

To do this, we first have to build a database that has the majority of possible races
to have reliable results. Another important issue is that, in addition to a large number of
horses, it is also interesting to have a lot of information about their characteristics.

Once we have the database, the next thing we have to do is train the machine. For
this there are several possibilities, such as using SVM ( Support Vector Machine ) or
doing it with Gaussian Processes. In our case we have chosen the first method, since the
article we have followed is based on it.

When we have the machine trained, we must enter the new (future) races and make
a forecast of their results.

What we have to do is maximize the profit, therefore we will look for the
parameters that best help us do this.

The main objective can be divided into several smaller ones:

• Creation of the database with the largest possible number of past races.

• Study of different models for the subsequent selection of the one that best
suits our needs.

• Implementation of the chosen model.

• Carrying out tests to check the validity of the model created.

2. memory structure
The memory is structured as follows:

• Chapter 1. Introduction to horse racing.

We will take a tour through the history of sports betting. We will also present
the concepts most used in horse racing, as well as the types of bets that can
be made.

2
 Introduction

On the other hand, we will present a series of articles that will help
understand the rest of the work.

• Episode 2. Model.

In this chapter we will study the model that we are going to follow, and we
will implement it to our particular data.

• Chapter 3. Results.

We will collect the results obtained by applying the model defined on our
database.

• Chapter 4. Conclusions and future lines of work.

In this chapter we will draw conclusions from the results obtained and
expose other lines of work that could be given.

• Chapter 5. Budget.

We will prepare the general budget of the costs associated with the Project.
To do this, the tasks carried out and the time we have invested in each of
them are shown. In addition to the necessary human resources, a detailed
report on the hardware equipment is included.

3
 CHAPTER 1. Introduction to horse racing

CHAPTER 1

Introduction to horse racing

This chapter is essential for understanding the work, since we present the topic
covered in it.

On the other hand, we frame horse racing within the world of sports betting.

In addition, we will make available to the reader the most important concepts,
which we will use throughout the Project, and a series of studies on horse racing that will
help to understand the rest of the work.

1.1. History of betting

Horse racing is one of the oldest known sports, and horse betting was soon linked
to it. Already in ancient China there was betting on animal races or fights, and in Arabia,
the passion for horses made races between them very successful. Chariot races were also
held in the Greek Olympic Games, a modality that was inherited by Roman culture [2].

5
CHAPTER 1. Introduction to horse racing

The rise of horse racing, and especially betting related to these races, takes place in
England. The 20th century is when we see a great growth in betting in this country.

In the United States, betting came due to the influence from England [3]. The
origin of a sports bet occurred when horse racing was at its peak and most popular,
especially among upper-class people. Later, with the American Civil War, racing
expanded to all economic sectors, causing popularity to increase considerably. However,
they lost their great fame with the arrival of team sports. The first to displace it was
baseball, followed by football.

In Spain it has been discovered in recent years and is becoming increasingly

famous. Without a doubt, the sport that attracts the most sports bets is the king of sports
in the country, soccer [4]. Many people dare to place bets on both the Spanish league and
foreign leagues or European competitions.

Nowadays the preferred means of placing bets in all countries is the Internet. The
website allows us to place bets with a single click, in addition to bringing us closer to
foreign competitions to make it easier for us to bet on them. Online sports betting is
undoubtedly a market with tremendous growth in our country and will surely continue to
grow a lot in the coming years [4].

1.2. Concepts used

To help understand this Project, we are going to include the definition of a series
of concepts [5][6] in alphabetical order.

All Weather (AW). Those racecourses that use synthetic sand instead of grass, allowing
horse racing despite inclement weather. Two types of compound are used in the United
Kingdom: Polytrack (Lingfield, Wolverhampton and Great Leighs) and Fibresand
(Southwell), the latter producing slower race paces.

Bankroll . Amount of money we have to be able to place our bets.

Blinkers . Equipment that is placed on the horse to restrict its lateral vision, helping it
maintain attention and avoiding possible distractions.

Bookie . Person who works in a betting house setting the odds on a sporting event. In

6
 CHAPTER 1. Introduction to horse racing

live bets, he is responsible for modifying them as the bet progresses. As well as closing
bets on a relevant event so that bettors do not bet knowing the result (for example, that a
goal is scored or that a horse abandons the race).

Bookmaker . Betting house, that is, a place where you can place sports bets.

Quota or odds . Amount set by the betting house for each of the bets that can be made.
Basing its value on the probability of it occurring or not.

Example: Real Madrid (1.35) - Draw (3.15) - Numancia (5.20). The odds would be each
of the numerical digits that accompany each bet. There are different types of fees:

• European or decimal quota: It is the one used in Spain. These odds are in the form of
an integer or decimal number (example: 1.50 or 3). To calculate the profit you
simply have to multiply the amount bet by the odds. For example: if we bet 10 euros
at 1.50, the profit is 5 euros.

• English quota: They follow the traditional British method. It is a fee in the form of a
fraction (example: 1/2 or 2/1). To calculate the profit, the amount bet must be
multiplied by the odds. For example: 10 euros bet at odds 5/1, the profit is 10 x 5/1 =
€50.

• American quota: They follow the traditional American method. It is an odds that can
be expressed in the form of a positive or negative number, starting from the point of
view of a bet of $100 (example: -200 or 200).

Draw . Only for smooth races. It refers to the starting box that the horse occupies before
the start of the race. Each horse's box is awarded after a draw and can have vital
importance in the future of the race, depending on the racecourse and the distance to be
covered.

R factor. Resistance that a horse has.

Form. Results obtained by a horse in the different races.

Going . Term that describes the state of racecourse grass. From slowest to fastest the
surfaces are named as follows: heavy (HY) – soft (SFT) – good to soft (GS) – good (G) –
good to firm (GF) – firm (F). This information can be very important when selecting our

7
CHAPTER 1. Introduction to horse racing

bet, given that not all horses are comfortable on all surfaces.

Handicap . Races in which the handicapper assigns each horse the weight it must carry
during the race, based on the results achieved by the horse in previous competitions. It is
intended that the horses leave the starting point under equal conditions. The theoretically
“better” specimens carry more weight while the “worse” ones can run more unloaded.
When a horse wins a race or finishes at least sixth place after the third unwinned race, it
is classified by assigning an Official Rating (OR). Based on the assigned OR, it is
decided in which level of races you can participate and what weight you must carry in it.

Handicapper . Official of the British Horseracing Authority (BHA), in charge of

deciding the Official Ratings (OR) and therefore the weights that each horse must carry
in a handicap race.

Jockey . It refers to the rider who leads the horse.

Official Rating (OR) : It is a way of measuring the worth of a horse based on its
performances. Depending on the OR of a horse, it may run in one race class or another.

Racecard . Program of the day's races. Nowadays, thanks to the Internet, they are
interactive and contain extensive information about all the races: previous races of the
horses, details of the trainer, the jockey , days since the last start.

Track probability . It is the opinion of bettors about the chance that a horse has of
winning a race.

Stake . It is the level of confidence that the bettor himself grants to a certain bet. The
lowest stake 1/10 (very little or almost no confidence) up to 10/10 (maximum level of
confidence). This level will set the guidelines that we should follow with our bankroll ,
distributing the stake in each of the bets in the best way. Example: If we give an 8/10
stake to an event we are granting it a high level of confidence and therefore, our amount
to bet on said event will also be high.

Starting Price (SP) . Price of the quotas of each competitor right at the moment the race
starts.

Value . This fact is considered when the house sets a disproportionate or exaggerated

8
 CHAPTER 1. Introduction to horse racing

quota with respect to the probabilities of a team or player of winning a match.

Weight allowances. Refers to certain discounts on the weight of the horse. Depending on
the number of victories of the rider, discounts may be requested.

1.3. Racing categories

At each racecourse we find that on the same day several types of races can be run
in which different horses participate, each of them has its particularities and must be
taken into account when betting. Only certain horses that meet certain conditions can
participate in each race, that is, any horse cannot perform any race.

In the case of our website [1], the careers are distinguished in [7]:

• Maiden. Horses or riders who have not yet won any of the races they have
run so far compete in this class. It is usually run by horses less than 2 years
old, and it is also very common to find some that have not yet participated in
any competition.

• Novice . It groups together horses or riders who are at a very early stage in
their career and who, therefore, have not yet reached a certain number of
achievements (normally they have not won more than a couple of times).

• Allowance . In this case there is a reduction in the weight assigned to a

horse to compensate for the inexperience of its rider.

• Conditions stakes . In these races, the weight that the horse must carry
depends on the age or sex of the horse.

• Handicap . Races in which the handicapper assigns each horse the weight
that must be carried during the race, based on the results achieved by the
horse in previous competitions. It is intended that the horses leave the starting
point under equal conditions. The theoretically better ones carry more weight
while the worse ones can run more unloaded. When a horse wins a race or
achieves at least sixth place from the third unwon race, it is classified by
assigning it an OR . Based on the assigned OR, it is decided in which level of
races you can participate and what weight you must carry in it.

9
CHAPTER 1. Introduction to horse racing

• Stakes . In these races the owner has to pay an entry fee for the horse to run.
That money is left as a pot for those who remain in the first positions.

Up to this point we have the races that are part of our database, however there are
other types:

• Horse racing in the United Kingdom is divided into different classes. The
lower the class number, the better quality it is. Thus, the best horses race in
Class 1 , while the horses with poorer results race in lower classes.

Class 1 is further divided into three groups and a fourth type of races called
Listed . Thus, Class 1 includes Group 1 races (the most important), Group 2,
Group 3 and Listed , listed in order of importance.

In the case of jumping, the Class 1 division, instead of a group, is called

Grade , so when it comes to show jumping, the Class 1 division is Grade 1
(the most important), Grade 2 , Grade 3 and Listed .

These data are important to take into account especially when we are going to
place the bet, because generally, in Class 1 races, there are no handicaps,
which means that all the horses run with practically the same weight (with
some exceptions). if mares compete against horses and by age, since the
youngest ones usually receive a little bonus).

This makes us see that in the most important races the horses run with equal
conditions, which a priori should make it easier to predict the potential of a
horse and translate it into a bet. On the other hand, in lower class races,
handicaps are usually run. In this type of races, in addition to correctly
evaluating the skill of a horse, we must be aware of the weight they will carry
and that all of this is reflected in the bet.

In the following table we can see the OR required to enter these races:

Class OR

Class 1 - Group 1 -
Class 1 - Group 2 -

1
0
 CHAPTER 1. Introduction to horse racing

Class 1 - Group 3 -
Class 1 - Listed 96-110
Class 2 86-110
Class 3 76-95
Class 4 66-85
Class 5 56-75
Class 6 46-65
Table 1.1 . Race class based on OR

The Nursery is a specific case of handicaps, with the same conditions as

normal handicap races but for 2-year-old horses.

Claimers races are generally low-level races, where all runners are subject to
a sale price established prior to the race by the horse's owners. This price
must be within a price range that is determined by the conditions of the race
and linked in turn to its quality and level. The weight carried by the horses is
also determined by the price that has been determined for the horse in
question, with the most expensive horses being those that carry the most
weight.

Sell races are usually similar to Claimers , with a similar operation except
that the price, instead of being set a priori, is determined by a public auction
that takes place after the end of the race. They are of a lower level than the
Claimers .

Apprentice races are races dedicated exclusively to novice riders, who have
not yet achieved 95 career victories. They tend to be infrequent races, and you
have to pay close attention to the weight allowances of the jockeys, since
depending on the number of victories they have achieved in their career they
may ask for a discount on the horse's weight. This discount is usually noted
with a number in parentheses next to the jockey 's name on racecards .

Types of races depending on the terrain[8]

• Flat : These are races without obstacles or smooth. These develop on grass ( flat
conventional) or on land ( standard ). Standard races are not all the same, as the

1
1
CHAPTER 1. Introduction to horse racing

material may be different and some may be faster than others. These races are also
called AW since they take place all year round. The distance ranges from 5 furlongs
to approximately 2 miles.

• Jumps : These are races with obstacles and depending on the type of fence they are
divided into two.

- Hurdles , which are fences that are not very high and flexible in case the horse
stumbles so that it does not have to abandon.

- Chase , are tall, wide fences (hedges). Depending on the racecourse it may be
more or less easy to overcome. Sometimes they have water or considerable
height and width.

• National Hunt Flat : It is a mix between normal flat races and obstacle races. They
appear in hurdle racetracks but are run without obstacles. They are usually a
preparation for horses to debut in the fences and their distance is 2 miles as a general
rule. The normal thing is to find a clear favorite with a low quota and many horses
without previous data.

1.4. Types of bets

As we have already seen, the favorite sports to bet on are football, basketball or
tennis. This is because they are the best-known sports in our country, the ones about
which we have the most information or the ones that are easiest for us to watch day after
day. Hence, for example, almost everyone knows that in a soccer match between FC
Barcelona and Sporting CF you have a much better chance of winning the first one.

However, horse racing betting is much more complex and unknown, which is why
bettors have to be more specialized in the subject to win money.

As a general rule, the order to follow when betting on horse races is: decide the
money you want to bet , the type of bet and the number of the horse(s) chosen. If the
money is not indicated, it is understood that you place the bet on the preset minimum.

In betting houses there are three large groups of bets [9] [10]:

1. Direct Horse Betting ; which includes betting on winner, second and third.

1
2
 CHAPTER 1. Introduction to horse racing

2. Combined Horse Betting ; which can be betting on exact, imperfect, trifecta and
quadrifecta.

3. Multiple Horse Betting ; where we find double, triple, quadruple, quintuple, triple
with exchange hitch and place and chain.

We go on to describe each of them:

• Bet on winner. It means that you bet on a horse that must finish first in the
race. They will only get paid if the chosen horse wins.

• Bet on second. In this case you bet on a horse, which can finish first or
second in the race. In this case you have two chances to win.

• Bet on third. Now the chances of winning are even greater, since the bettor
gets paid whether the chosen horse comes first, second or third.

In the three previous cases, the dividend is given for each euro bet, this value being
its base.

• Bet on exact. You have to choose the two horses that you think will finish
first and second in the race, in the exact order.

• Bet on imperfect. In this case, two horses are chosen that must finish first and
second in the race in any order.

In the two previous cases, the dividend is given for every two euros bet, this being
its base value.

• Bet on trifecta. The first, second and third horse must be chosen in order of
arrival at the racecourse. You win if the three horses are in the exact order
that was bet. The dividend is given for each euro bet, this value being its
base.

• Bet on quadrifecta. It consists of choosing the horses that finish first, second,
third and fourth, in exact order in the race.

• Double. Two horses are selected that must finish first in two consecutive
races. The dividend is given for every two euros bet at the racecourse, this
value being its base.

1
3
CHAPTER 1. Introduction to horse racing

• Triple. In this case there are three horses that are chosen. As in the previous
case, for the bettor to get the money, the three horses must finish first in three
consecutive races. The dividend is given for every two euros bet at the
racecourse

• Quaterna. Four horses are chosen to finish first in four races.

• Fivefold. It consists of guessing the winner of four consecutive races. The

dividend is given for every ten euros bet

• Triple with exchange, down payment and place. It is about guessing the
winners of three races. If the first one wins, it can be redeemed during the sale
of the second pass, and the same with the third. Optionally, you can enter the
bet of the first in the second or that of the third, with what is called
“connection voucher”. The dividend is given for every two euros bet

• Chain. Six horses must be identified that must finish first in six races on the
racecourse.

We also have a series of betting strategies that can bring us great benefits [11]:

• Placed. Betting placed on a horse means that it will be among the leading
positions. The number of positions to take into account depends on the
number of participants in the race. Usually in races of 4 runners or less this
option is not available. In races of 5 to 7 runners, the placed horses are the
first and second horses to reach the finish line, while in races of 9 to 15 they
are the first three and in races of 16 or more runners they are usually the first
4. This is not always true and it is something that we have to read in the rules
of each race, since in Maiden -type races, regardless of the number of horses,
it is only considered placed if it is among the first three positions.

• Back or Lay . This type of betting refers to betting for ( back ) or against (
lay ) a horse. Betting on backs is usually more complicated since you have to
guess the winner or place. If you opt for the lays , if our horse does not win
we will have the right bet. However, betting in favor brings greater benefits,
since the greater the risk, the greater the profits.

Lay placed . Strategy of betting against the favorite horse finishing in the

1
4
 CHAPTER 1. Introduction to horse racing

placing positions. In horse racing it is very common to find surprises, since

there are many factors that can influence the development of the race such as
the state of the ground, the possibility of tripping or falling and even the state
of the horse. horse at that time. Hence, it is very profitable to bet against those
horses that most have as favorites, since we can aspire to a great prize by
betting a small amount.

Keep in mind that in hurdle races, which tend to be more rugged, or in long
races, this strategy can lead to greater benefits.

Bet on the rider. Lately it is becoming common to think that there is more
advantage in betting on the runner than on the horse. Furthermore, it is much
easier to get hold of rider data than it is to interpret the pedigree, breeding or
terrain for a single horse. Hence, for bettors who do not have much
knowledge, it is a good betting strategy.

Bets on the totalizer or tote [12]. It is a pot where the odds are never known
until the moment the bets are closed. Thus, our fee will be determined by the
volume of money that has entered each of the participants in the race. The
more money that comes in for our horse, the less fee they will pay us, since it
is inversely proportional. In addition to the total pot, it is worth deducting the
commission that the tote receives, as an organization to defray expenses and
generate its profit, since at the end of the day it is another form of business.
Betting on the tote is usually interesting when we suspect that our horse is not
going to be very popular, since its fee will be high, and it is likely that the tote
will end up paying a higher fee. This is valid for both win and place bets.
However, with the appearance of betting exchange houses, these advantages
have been greatly reduced, reaching the point that it is not very worthwhile to
bet on the tote, at least in terms of winner and placed odds. The minimum bet
in this type of combination is two euros or two pounds, depending on the
country in which we are placing the bet. Obviously, as there are expected to
be many guessers, the payment we will receive as compensation is quite
small.

• Trading strategy. To understand it better, we are going to give the example of

1
5
CHAPTER 1. Introduction to horse racing

real estate buying and selling. The perfect operation to obtain profits is to buy
a house with the purpose of selling it later at a much higher price. In horse
betting, trading is the term used to obtain a profit derived from betting on a
horse to win a certain race ( back ), and then, throughout the race, betting
against that same horse ( lay ). . In this way we ensure that we will obtain
benefits no matter what happens in the race.

These strategies can be carried out before the start of the race or during it (
in-play ).

Another thing to keep in mind is that withdrawals are usually very frequent in
horse racing [13]. The main reason is usually that he does not enter the starting box (in
the case before the race) or that something happens during the course of the race that
causes him to abandon. Withdrawals minimize fees. In this way, if the horse withdrawn
is the one we had bet on, the money we bet is returned; if it is another horse, our quota is
reduced.

In some exchange houses they have a reduction factor for horses (in percentage), to
which the bettor has access. However, in traditional houses the reduction is made based
on the horse's quota at the time of abandonment.

1.5. Bettor profile

According to a survey carried out in the Basque Country [14] by the three betting
houses that operate in the BAC (Kiroljokoa, Reta and Garaipen-Victoria), the profile of
the bettor, broadly speaking, is that of a male between 25 and 45 years old, with a great
love for sports, a regular reader of specialized press and, therefore, knowledgeable about
current sports news. In addition, they tend to be daily Internet users, and often place bets
in the company of friends.

A study carried out by Betfair [15] tells us that of the bettors, women barely
exceed 15%.

In the following graph from Paysafecard [16] (a leading company in Europe in the
marketing of prepaid coupons for online purchases and a member of the AEDAPI) we
can see the distribution by age:

1
6
 CHAPTER 1. Introduction to horse racing

Betting by age

Figure 1.1. Amount of bets by age

As for the amount bet, they normally do not exceed €10 each time they play.
Furthermore, 60% play between 2 and 4 times a week, while 30% play only 4 times a
month.

In Spain, the star sport is football in all its variants, where you bet on who will
score the first goal to how many minutes the referee will count off, followed by tennis,
for which preference is given to placing bets following the live broadcast. .

We have to go to England or Australia to find that horse racing or rugby exceeds

50% of the bets.

Position to choose when we bet [17].

When betting we can choose between two basic profiles

• Conservative bettor. At this point we find bettors who do not make bets
higher than an odds of 1.5, hence the profits are more regular, but the benefits
are quite small. Furthermore, their bets will always be in accordance with all
the information they have.

• Risky bettor. In this case, the bettor avoids odds of less than 2.00. He will
lose more times than the previous one, but when he wins he will do so with a
much larger amount. This type of bettor is guided more by impulses or by the
heart.

1
7
CHAPTER 1. Introduction to horse racing

1.6. Betting shops

The concept of betting houses can encompass two large groups, traditional bookies
and so-called betting exchange houses [18] [19].

The majority of betting houses that currently exist on the market fall within the
group of traditional bookmakers . In them, the form of business is to interact directly
with the bettor, taking charge of the payment of the bet if it is a winner or keeping the
bettor's amount if it is a loser. Large investment groups are behind these, creating
important multinationals. The way to find benefits for this type of houses results from the
application of what we call quotas. That is, it is about playing with the odds, so that the
income from losing bets of bettors of one outcome is never less than the winning bet
payouts of bettors of the correct outcome. This is done by never distributing 100% of the
proceeds to the bet. Thus, no matter what happens in a game, the company will have
profits. This group includes many of the large well-known betting houses: Bwin, bet365,
Miapuesta, Interapuestas, William Hill, etc.

In the case of bet exchange, the bookie acts as an intermediary between 2 bettors.
They will be the ones who will agree on both the amount to bet and the odds for the
event in question. Here, the betting house will obtain the benefit through a commission
of a percentage that will be charged to the user who ends up being the winner of said bet.
This percentage ranges between 3% and 5% depending on the betting house. In this
group we can find Betfair as the main exponent and others less powerful such as RedBet
or Betsson (the latter also offers services like a traditional house).

The type of exchange is usually more profitable for the bettor, although it is also
usually a little more complex to use and requires a little more experience.

In the world of betting, the reputation and integrity of a bookmaker is its main
value and this will make bettors continue to trust that bookmaker or not, for future games
[20].

With the advent of the Internet, many bookmakers have an online page where you
can watch sporting events and bet live. Of these pages we are going to highlight the
following:

1
8
 CHAPTER 1. Introduction to horse racing

1.6.1. ZEturf

Founded in France in June 2005 [21]. It was the first online horse betting house. It
is considered the first complete offer that covers the entire European market for
international horse racing, achieving an average of more than three hundred thousand
daily visits.

Zeturf is only a mediator between the bettor and the bet, guaranteeing payment and
transparency in it.

To provide a greater degree of security, it has strict registration rules, inflexibility

in deposit and withdrawal policies, all supervised by

Barclays Bank which deals with financial transactions. Additionally, among its partners
it has Gamcare , which is an international organization that is responsible for supervising
matters related to abuses in money gambling, thus ensuring its transparency.

Additionally, on the website [22] we have all the information about the horses,
access to the forecasts, videos of the races and results.

Therefore, we can say that this betting house gives us a lot of security and has
adequate financial support while offering a multitude of data to help us place bets.

At ZEturf they have their own bets available [23]:

• The main race of the day: the ZE5 (quinté+). In ZE5, you must find the first 5
horses in the race. The race takes place every day at 1:50 p.m. French time
from Monday to Friday and at 3:08 p.m. on Saturdays and Sundays.

• The winner (bet said S). This bet covers the Gagnant and the Placé . If you
bet on a Gagnant horse, this horse has to win the race. If you bet Placé , you
have to finish the race among the first three places or first or second if there
are less than eight horses.

• The twin (bet said J). This can also be Gagnant or Placé . Here you choose
two horses. If you bet Gagnant , your two horses must finish first and second.
If you bet Placé , your two horses must finish the race in the first three
places.

1
9
CHAPTER 1. Introduction to horse racing

• The Exacta (said Ju bet). Here you choose two horses, and they must finish
the race in the first two places, in the chosen order.

• The trio (bet said Tr). You choose three horses, and these must be the first
three in the race.

• The fourth (bet said Zc). Here you must find the horse that will finish fourth
in the race.

• The trifecta (said Tri bet). Three horses are chosen, and they must finish the
race in the first three places, in the order you mentioned.

• The ZE 2/4 (bet said ZE 2/4). You choose two horses, and your two horses
must finish the race in the first four places.

• The quartet (bet said ZE4). You must find the first four horses in the race.

1.6.2. Betfair

It was founded in 1999, and is today one of the most important betting houses in
the world [24]. It is a great example of an intermediary between players, since in Betfair
the odds are set by the users, when betting against each other, and not by the house itself.

Betfair 's only task is to ensure that the operations are carried out, which is why it
keeps 5% of the bettors' profits.

For horse racing, one of the most commonly used methods is to carry out trading ,
which, as we have explained the types of bets, consists of betting in favor of a horse's
victory and later betting against it.

Furthermore, on its website [25] we can find information related to almost all
existing sports, as well as the possibility of placing bets on them.

1.7. Related Posts

In this section we are going to describe a series of articles that will help us better
understand the model on which we are going to base ourselves.

There is a paper [26] that investigates fundamental strategies for detecting and
2
0
 CHAPTER 1. Introduction to horse racing

exploiting public system errors in horse racing betting markets.

The realization of the Multinomial Logit model for horse racing processes is
estimated from a database of 200 races. To obtain a more efficient estimation of the
parameters, a recently developed procedure is used that exploits the information content
in the set of completion orders.

Variables in this probability model include horse and rider characteristics, as well
as several race-specific characteristics. In addition, sampling and retention procedures
are used to evaluate betting strategies.

Studies to evaluate the efficiency of horse racing betting markets are motivated by
similarities between horse racing and stock markets, such as uncertainty in future returns
on investments, presence of many participants, or the availability of variety. of
information related to research and participants [27][28][29].

This article looks for a profitable betting system by applying the pari-mutuel setup.
Betting systems have two components: a model of the horse racing process, and a betting
strategy. The racing process model is based on predicting the outcome of the race. Its
basis is to predict the probability of each horse winning the race. The betting strategy
uses those probabilities as input to a betting algorithm which determines the amounts to
bet on each horse.

In another article [30] the result of two years of application of the cross-validation
method of a handicap system is presented.

It is based on the performance of race data from an entire season being subjected to
discriminant analysis and classification criteria. The discriminant function and
classification criterion from the first year (Year 1) is applied to the data from the second
year (Year 2). Classification techniques are evaluated in terms of correct classification
percentage and returns on investment.

This study is designed to measure the relative effectiveness of the cross-validation

method across six different classification techniques. The term discrimination refers to
the statistical process that takes advantage of handicap racing systems that is capable of
differentiating between winners and non-winners. The term classification refers to the

2
1
CHAPTER 1. Introduction to horse racing

application of the initial rules to a second example of racing.

Data is obtained through observation of the performance of winning and losing

horses in two seasons (referred to as Year 1 and Year 2). He is not assuming that the two
meetings produce unique results that could not have occurred during another season. The
discrimination problem is based on the existence of a linear function that produces a
significant separation between winners and non-winners based on Year 1 data. The
classification problem takes the discriminant function and the classification criterion
from Year 1 and cross-validates it with the data from Year 2.

Data is obtained from two seasons of Golden Bear Raceway in Sacramento. Races
carried out between May and July. They are all one mile in distance and with eight or
nine horses per race

Studies of the efficiency of racing betting markets typically examine the returns on
winning bets [31]. The conclusion of this task is that there are trends in the probabilities
of winning, furthermore, those trends are insufficient to create a profitable strategy,
except in rare cases where there is a clear favorite.

Examining betting returns requires estimating the probability of each racing

outcome. Hausch, Ziemba, and Rubinstein (1981) applied the model described by
Harville (1973) to obtain such estimates. Harville's formulas have limitations described
by Hausch et al. This article describes a series of alternatives to Harville's model. The
resulting formulas derived from the alternative models give up the simplicity of the
Harville model in exchange for more precise estimates.

This article also describes the betting markets and the studies carried out on the
efficiency of the markets. In other sections they describe the Harville formulas, the
alternative formulas and the comparison with the two models.

Horse racing has great similarities to investing in the stock market [32]. In both
situations the future profits are not known with certainty, there are a large number of
participants, a lot of information is available, professional advice abounds, and each
participant has information about the activities of the other bettors (or investors). Horse
racing data allows us to take certain attitudes regarding risk and investment behavior.

2
2
 CHAPTER 1. Introduction to horse racing

This article studies a new example of horse racing results from Atlantic City, New
Jersey. The questions they study are:

1. Are market bets, determined by the public's betting behavior, the best data for
finishing order prediction?

2. Do starting bets on each horse reflect that horse's true probability of winning as
suggested by Baumol (1965) or is there a systematic tendency to overbet on
favorite horses as suggested by Rosett (1977), Ali (1977) and Snyder ( 1978)?

3. Since each bet is recorded almost immediately after it is placed, there may be
advantages for people who place bets minutes before the race begins. This strategy
minimizes the time in which the signal produced by the bet will be available to
other bettors. Is there any evidence that late bettors have better information than
early bettors?

2
3
EPISODE 2. Model

EPISODE 2

Model

Although horse racing is not one of the preferred sports for betting, there are
numerous models that can help us make a better estimate of a horse's probability of
winning.

Next, we are going to describe the model that has helped us carry out our Project.

Once the chosen model has been described, we will move on to implement it in our
example. So we will follow the guidelines set by him, but with the example that concerns
us.

2.1. Chosen model

Identifying winners of competitive events: An SVM-based
classification model for horse racing prediction [33]
The goal of horse racing modeling is to evaluate the efficiency of the information

2
4
 EPISODE 2. Model

available in betting markets.

The usual thing is to make a forecast of the runners' final positions through discrete
or continuous regression models. However, there is empirical evidence that indicates that
the information contained in the final positions is often unreliable.

To solve this problem, a classification based on paradigm modeling is proposed

that relies only on data that distinguishes winners from losers. To evaluate the
effectiveness of these models, an experiment was conducted using racing data from
England.

2.1.1. Introduction

Many studies explore horse racing betting markets because they share many
common characteristics with broader financial markets, including a large number of
participants and a wide range of factors that can influence horses [34][35][36] [37][38]
[39][40]. Furthermore, betting markets offer an important advantage over broader
financial markets, that is, they generate an unambiguous outcome (a winner) and an
associated rate of return within a finite time frame [36], and therefore provide an
objective reference point to measure the quality of an investment decision (e.g. a bet). As
a result, betting markets can provide clear insight into pricing issues that are more
complicated elsewhere [38] and the value of studying bettor decisions is reinforced by
the fact that these markets are in themselves important. For example, the turnover of

horse betting market in the United Kingdom in 2006 was €19.2 billion.

The prediction model is used when it is necessary to calculate the degree to which
bettors make efficient use of information when making their investment decisions.
Particularly, the models incorporate variables based on publicly available information,
used to estimate the horses' chances of winning. If these estimates allow profitable bets
to be made on a large number of future races, we can conclude that bettors do not
completely discard the information related to the attributes contained in the model [26]
[35][41][42].

Predicting the winner of a race has been shown to be important in explaining the

2
5
EPISODE 2. Model

relative strength of competitors within the race [26]. The Conditional Logit (CL) model
[43] has been proposed for this task, since, unlike common logistic regression that
considers each horse separately in each race, CL models a race as an entity and,
therefore, maintains a relationship between competing runners [26][44][45]. Recently,
Edelman (2007) [46] showed that the prediction accuracy of these models can be
improved if they are used in conjunction with more modern methods. This approach is
based on the two-stage philosophy of Benter (1994) [41] and uses Support Vector
Regression (SVR) to model the relationship between:

• Fundamental variables associated with the horses' recent performances and factors
related to current races (e.g. prize money, transported weight...).

• Final position of the horses.

The resulting forecasts are combined with the final market horse prices ( odds )
using CL in a second step. CL and SVR complement each other in the sense that CL
represents within-career competition, while SVR accommodates a large number of
potentially correlated variables, with low risk of overfitting and complex models with
non-linear relationships between attributes. , in a data-controlled manner.

This article develops a forecasting model that adopts the two stages of the
approximation model. However, while previous work on horse racing forecasting relies
primarily on regression methods [46][47], the model proposed here expresses Support
Vector Machines (SVM) to classify racing results. As will be explained later, theoretical
considerations as well as empirical results [48] cast doubt on the reliability of a key
component of the regression model, the order of importance of the final data. Taking a
ranking approach, the model focuses on distinguishing winning and non-winning horses
and avoiding excessive use of incorrect rankings, especially in the lower positions of the
ranking. Furthermore, a new model of pre Data processing is advised to introduce some
competitive ideas within the race, in the first stage modeling phase.

The main objective of this paper is to examine the effectiveness of the proposed
classification method for horse racing prediction, and to shed light on the marginal
contribution of the elements of this two-stage model. Finally, an empirical evaluation is

2
6
 EPISODE 2. Model

necessary to contrast the predictive performance proposed by the SVM-based model

with very competitive benchmarks [42][46]. The components of the predictive model
(the two-stage hierarchical approach, the new data from the pre-processing technique and
the non-linear model) are evaluated individually to confirm their adequacy.

The rest of the article is organized as follows: the SVM theory reviewed above,
describing the horse racing model and the details of the two stages of the model. Finally,
results of the empirical evaluation and conclusions.

2
7
EPISODE 2. Model

2.1.2. Support Vector Machine (SVM) for Classification

The SVM is a learning technique that facilitates linear and nonlinear binary
classification.

For example:

S= {( x i , y i )} i M
= 1

Where:

• x i ∈X⊆R N
with x being an N -dimensional vector

• and i ∈ { - 1, + 1 } the corresponding binary label

• M is the number of observations

SVM learns from the data of a functional model. f Λ ( x ) : X A { - 1, + 1 } . This

allows the estimation of the membership of observations not contained in S.

• The vector Λ includes the classifier parameters that are suitable in S at a stage
of building the models (training classifier).

SVM is inspired by statistical learning theory [49]. To arrive at a classification

model for S , the concept of maximum separation margin is implemented. That is, we try
to maximize the distance between cases that are close to the decision line that separates
the two classes [50]. It can be seen that by maximizing this margin the error in the future
data is minimized [49].

To build this linear classifier with maximum margin, the norm corresponding to
the hyperplane of the weight vector, w , must be minimized. Subject to the constraint that
the training data of each class resides on each side of the separation surface.

2
8
 EPISODE 2. Model

{x|WX,+b=1}

example of ciass -1
example of class +l
supporting hyperplane
border between class -1
and+1
support vector__

{x|Wx,+b=0}

Figure 2.1. Separation between two classes with the SVM classifier
[33]
With y i ∈ { - 1, + 1 } , this restriction can be put as [51]:
y i (( w ⋅x i ) + b ) ≥ 1, i = 1,..., M (2.1)

The examples that satisfy equation (2.1) are called support vectors and define the
orientation of the hyperplane.

To evaluate classification errors (i.e., examples that do not satisfy (2.1)),

auxiliary variables are introduced that relax the condition in (2.1) [50], ξ i . Therefore, to
build a maximum margin SVM classifier, we have to solve the following convex
quadratic programming problem:

1M

min 2 j| II + C
w
i ∑ξ (2.2)
w,b,ξ2 i=1

s . t .: y i (( w x i ) + b )⋅ ≥1−ξ i , i = 1,..., M

The variables w and b define the separation of the hyperplane, so the resulting
classifier has the form:

fΛ ={ w , b } ( x ) = sign ( (w *
⋅x)+b*) (2.3)

Where w* and b* represent the solution of equation (2.2).

2
9
EPISODE 2. Model

To construct a general nonlinear decision surface, the SVM map has the input data
in a high-dimensional feature space through a mapping feature function, Φ . The
construction of a separating hyperplane in this feature space leads to a nonlinear decision
boundary in the input space [49]. The ability of SVM to reveal non-linear relationships
between input variables by projecting the data into a higher dimensional space has been
demonstrated with known data sets [52]. For example, standard nonlinear classification,
such as the XOR problem or the classification of a checkerboard in black and white
regions, are solved with an SVM [50][53][54].

Mapping of the data is carried out to avoid many calculations in the transformed
feature space. Considering equation (2.2) with α i denoting the Lagrange multipliers [49]
[51]:

M 1M

max = ∑α ∑α α i − i j ⋅
y i y j ( x i x j ), α i=1 2i
,j=1 (2.4)
M

s.t.: ∑α ≤α ≤ i y i = 0; 0
i = 1
i C ∀ i = 1,..., M

Equation (2.4) contains the input data in the form of dot products. It can be called
the Kernel function, K , and can be used to calculate the dot product of the transformed
vectors in the input space:

K( , xi xj ) = Φ ( ) ⋅Φ ( ) xi xj (2.5)

The Kernel can be considered as a function to measure the distance between two
input vectors in nonlinear space. The resulting classifier is:

, _ (( _ ,

fΛ ={ α , b } ( x ) = sign 11 ∑α i
\ \ Í E SV
yiK(xi,x)I+b1,
(2.6)

With SV= {i|α i >0}

Where the set SV contains the support vectors.

In this article, the Radial Basis Gaussian Kernel Function (RBF) is used.

3
0
 EPISODE 2. Model

K ( x i , x j ) = exp (-γx i -xj II 2 ) , (2.7)

Which also presents fewer numerical difficulties, since the output values of the
Gaussian functions are between 0 and ∞ [49].

In order to solve a classification task with RBF-SVM, two parameters must be

determined:

• The regularization parameter, C , which controls the balance between margin

maximization and error-free classification of the training set.

• The smoothing parameter, γ , which determines the width of the Gaussian, and
thus the sensitivity of the distance measurement.

These parameters are calculated through the selection of different values for C and
γ, empirically evaluating the result with all possible combinations.

2.1.3.Forecasting race results

2.1.3.1. Background

Predictive models help examine the efficiency of horse racing betting markets.
Bettors' opinion of a horse's chance of winning is called track probability , q i
j
, and can
be obtained by

1
probability of winning for horse i in race j , u , by means of q i =
i
j j
.A
1+ui j

The market is efficient, from an information point of view, if bettors take into account all
the information available in it. Odds represent the market's best estimate of a horse's
chance of winning, and should reflect that horse's true probability in that race. Bets will
only be profitable if the odds given by the bookmakers are inaccurate. The goal of that
model is to accurately evaluate these probabilities based on publicly available
information. If it is shown that betting based on these probabilities makes a profit, it can
be said that the model was successful in making the information available to bettors.

3
1
EPISODE 2. Model

Decision making when betting on horse racing can be modeled as a discrete

process. The CL model [43] has appeared as an approach to study bettors' preferences
among different race participants. Unlike ordinary logical regressions, which treat each
data point (each horse) individually, CL maintains connections within the alternatives of
a choice set (that is, between the runners in a given race). This allows the identification
of information that affects the selection of the winning horse. CL returns a horse's
probability of winning, which is calculated relative to its competitors' probabilities. This
ability explains the reason for the popularity of CL in horse racing prediction [31][35]
[42][55].

The objective of the horse race prediction model CL is to predict a vector with the
probabilities of winning for race j , where the component p i
j
represents the model
estimating the probability of horse i of winning race j , and m j denotes the number of
runners in race j . To achieve this, the win rate or winningness index is defined, w i
j

wi =j
β⋅x i j
+ ε i j
(2.8)

Where:

• β is a vector of coefficients that measure the relative contribution of the input

variables contained in the vector.

• x i
j
describes runner i in race j .

• ε i
j
is the error term, it represents the information that goes unnoticed.

If w i
j
is defined so that the horse with the highest winningness index value wins
race j , then it can be shown that, if the errors are independent and distributed according
to the double exponential distribution, the probability of horse i of winning the stroke j is
given by the function CL [43]:

pi j = m e j xp ( β⋅x )
ij (2.9)

∑ exp (β⋅x ) i
j
i = 1

3
2
 EPISODE 2. Model

As Johnson et al. (2006) [35], the choice of this model allows the explonent β⋅
x i
j
to be interpreted directly as the ability of horse i . The model coefficients, β , are
estimated through a maximum likelihood procedure. In particular, given a data set of R

races, the joint probability L = L ( β ) is the probability of observing the results,

assuming that p is that of equation (2.9). Therefore

i
j

β
β←m β ax L ( β)=∏ R
pi =
j
∏ R
m ejxp
( ⋅ xij
)
(2.10)

j =

i=1
1 j =
1
∑
exp ( β⋅x ) i
j

Where x represents the winner of race j .

i
j

2.1.3.2. Two stages prior to the proposed model

Previous studies have shown that track probabilities are a good predictor of race
outcomes [56]. However, it can be detrimental to use them alongside variables that
describe a horse's abilities in a prediction model [41][46]. In particular, the dominant
impact of track probabilities can mask the other variables and negatively influence the
model [48].

3
3
EPISODE 2. Model

To alleviate this problem and guess the true influence of the fundamental
variables, a two-stage model is proposed. In one of these models, Benter (2003)
developed a first stage to predict the final position of the runners through a Multivariate
Linear Regression (MLR) using only fundamental variables. Track probabilities are not
being considered in this step. The estimation of the final positions is interpreted as an
evaluation of the riders' abilities, based on past performances, which can be observed
through the fundamental variables. These skills are then combined with track
probabilities using CL to estimate the probabilities of winning.

D represents a database of R past races with a number of runners M. D 1 and D 2

are disjoint subsamples of D containing runs R 1 and R 2 with runners M 1 and M 2 ,

respectively. x is the vector of the fundamental variables that describe horse i in race j .
i
j

And y is the final position. The two-stage procedure can be described:

i
j

TO
f MLR ( x ) = w ˆ ⋅ x+b
Step 1:
{w
M1

ˆ, b ˆ
←∑ ( y i -(w
+b
2
))

(2.11)

p j
i exp (β 1 f MLR (x )+β
i
j
2 q i
j
)
Step 2:
∑
m j

exp (β 1 f MLR (x )+β

i
j
2 q i
j
) i = 1

i = 1,..., M 2 ; j = 1,..., R 2

Where:

w and b represent the slope and intercept of the linear regression function.

TO
w ˆ and b are their respective square estimates, calculated through Step 1.

3
4
EPISODE 2. Model

Since the fundamental variables are processed with Step 1 and are summarized in f
MLR , the second step incorporates the CL model with only two inputs of the coefficients
β 1 and β 2 .

The index j is removed in the first step because the linear regression is not able to
exploit the information related to the race context. All brokers are considered
independent and their final positions are estimated only through the respective
fundamental variables. One way to overcome this restriction and take broker competition
into account is to replace the first step linear regression with a CL regression stage:

j exp ( α ˆ ⋅ x i j
)
fCL ( xi ) = mk
∑ exp ( α ˆ ⋅ x i j
) i = 1

Step 1:
j = 1,..., R 1
M 1
exp ( α ˆ ⋅ x j

) α ← max ∏ mk
Yo
,
i = 1
∑ exp ( α ˆ ⋅ x i j
) i = 1

(2.12)

Step 2: p j = exp ( β 1 f CL ( x i ) + β 2 q i j
) j
, j = 1,..., R

∑
Yo
mj 2

exp (β 1 f CL (x i
j
)+β 2 q i
j
) i = 1

This approach has been successfully applied in Sung et al. (2005) [42] and in Sung
and Johnson (2007) [48] is shown to outperform the single-step model.

Edelman (2007) [46] modifies this two-step model to overcome some limitations
of the CL and MLR algorithm, respectively. In particular, this technique deduces a
model to minimize the error of the forecasts in the training data. Consequently, they are
prone to model not only the structure but also the noise of the data (data overfitting),
especially if a large number of fundamental variables are processed [49]. Furthermore,
they are unable to explain nonlinear interactions between variables, unless predefined by
the modeler. Edelman fitted the model (2.11) using SVR [57] instead of MLR. He
further modified the original SVR procedure to allow multiple intersection terms. For 38
 EPISODE 2. Model

Therefore, a term b j is introduced in (2.11) instead of b to perform an effective analysis

of the race [46].

The three approaches described only differ in the first step, while the general idea
of the fundamental modeling variables, and the combination of the first step with the
track probabilities using the CL model, is identical.

MLR/CL refers to the original two-stage model (2.11), while CL/CL represents
(2.12) and SVR/CL the approximation of Edelman (2007) [46].

Note that the unmodified SVR algorithm is used, instead of Edelman's, as a

reference point in this study to obtain a clearer view on the competitive performance of
the classification-based model against the regression-based one and, thus, have reliability
in data classification.

2.1.3.3. An SVM model based on two stages.

The two-stage prediction model, which is developed in this article, is based on that
of Edelman (2007) [46]. It differs from this in that instead of carrying out a regression of
the final position of the horses in the first phase, the model proposed here uses an SVM
to obtain a classification model to identify the winner of a race. This is motivated by the
view that, in the context of horse racing, last positions do not necessarily carry important
information. The rules of racing require that jockeys want to ensure that the horse is in
the best possible position, but there is an incentive for them to do so when there is no
hope of victory. The reliability of the final positions, which form the core of the
regression-based model, is questionable. This view is supported by empirical results
from Sung and Johnson (2007) [48]. Our model focuses on only distinguishing winners
from non-winners.

The SVM/CL model can be defined as follows, where y denotes a binary win/no-
win variable indicator, rather than a final position.

3
9
EPISODE 2. Model

f svM ( x ) = 2 &iy i exp ( v x, - x II 2 )

|+bˆ
\ie SV )

{ α ˆ, bα } ← max =
M 1
ˆ
∑α i

Step 1: i = 1
x i - x K II 2
αα
∑
M
- 1 1
i K y i y K exp
)

2
i,k=1

M1

s.t. α y = 0; 0 ≤ α ≤ C ∀ i = 1,..., M
ii Yo 1
i=1

(2.13)

p j
i β 1 exp ( f ( x ))
SVM i
j

Step 2: mj

∑i=1
exp ( f ( x )) + β
SVM i
j
2 q i
j

i = 1,..., M 2 ; j = 1,..., R 2

The goal of the first step of the model is to estimate the likelihood of a winner.
Therefore, the sign function (2.6) is removed, to obtain a continuous output from SVM.
The resulting value, f SVM (x), is proportional to the distance of a data point (a horse) to
the separating hyperplane (between winners and losers), representing the reliability of a
point (a horse) belonging to a particular class. [49] (whether winner or loser). If a horse
has a high SVM it is more likely to win.

As in most previous models, the relative information of the horses competing

against each other, and prior knowledge of each race, is lost in step 1. The SVM is
unable to explain the competition within the race. However, it should be noted that the
overall objective of the prediction model is not to maximize the number of correct
winner predictions, but rather to maximize the profit. Profitable bets require an accurate
estimate of a horse's probability of winning. The output of the first step serves only as a
summary of a horse's ability (based on previous races), while the second step serves to
obtain information about the competition within the race.

What we want to try is for the SVM to be able to capture some elements of the
competition within the race in the first phase, to improve the final estimates. To this end,
a standardization of the career is proposed to complement the

4
0
 EPISODE 2. Model

data. Continuous variables are commonly standardized to mean zero and standard
deviation one, before applying the prediction model, to avoid numerical difficulties with
different ranges of values [58]. This is achieved by subtracting its mean value from the
variable and dividing by its deviation. This preprocessing is enhanced to take into
account, to some extent, competition within the race. In particular, data standardization
is carried out as follows:
j-j
(2.14)
x =
j it t
∀ i = 1,..., m
Item
σ j
t

Where:

• x it
j
( x ) denotes the new value of attribute t of horse i in race j .
it
j

The mean x t
j
and the standard deviation σ t
j
are calculated over the runners
in race j .

To show the intuition of race normalization, we consider two four-runner races,

with horse class as the single input.

Horse Race Database Class normalization normalization

Race 1 Horse 1 20 -0,234 -1,162
horse 2 40 0,435 -0,387
Horse 3 60 1,104 0,387
Horse 4 80 1,773 1,162

Race 2 Horse 1 1 -0,87 -1,162

horse 2 3 -0,803 -0,387
Horse 3 5 -0,736 0,387
Horse 4 7 -0,669 1,162
Table 2.1 . Comparison between career standardization and database standardization

4
1
EPISODE 2. Model

Horse class is an abstract measure of the horse's ability. The best horses have the highest
value.

2.1.4. Empirical evaluation of the SVM/CL model

2.1.4.1. Data and variables

Data for the empirical analyzes of this study were provided by Raceform Ltd. and refer to the
races held at the Goodwood racecourse in the United Kingdom between May 1995 and August
2000. This period was deliberately chosen from the largest online betting exchange, Betfair , which
was first announced in October 2000. The arrival of Betfair increased competition between
bookmakers (individual Betfair clients can act as bookmakers) and increased the number of
professional bets on the market. The advantage of using data from before September 2000 is that we
can refer to empirical results from before the arrival of Betfair [26]: 200 races before 1986 [46]: 300
races before 1995.

The data consists of 556 races with 5947 runners. The 400 races (4296 horses) conducted
before May 1999 are used to develop the prediction models, while the remaining 156 are reserved
for other test samples.

The key variables used are those included in Bolton and Chapman (1986) [26] and are
provided in Table 2.2. These variables have been in the public domain for nine years before 1995.

4
2
 EPISODE 2. Model

Independent variables Definition of variables

Variables related to the market

Ln(q j i ) The logarithm of the normalized track probabilities
Fundamental variables
pre_s_ra
The highest speed in the previous race that the horse has made
The average of the horse's fastest speed in the four previous races. With
avgsr4
value 0 if the horse has no past races
The average of the highest speed in the past races, with the same
disavesr distance, made by the horse. It has the value 0 when it is the horse's first
race
go_avesr Average speed level of the horse's past races in that
same ground
draw Starting position in current race

eps Total prizes won (finishing first, second or third) to date. And number of
races in which it happened
A value of 1 indicates a horse that has run three or four of the last
newdis four races with a distance of less than 80% of the current race distance. A
0 for anything else
weight The weight carried by the horse in the current race
win_run The percentage of races won by the horse
jnowin Number of jockey wins to date
jwinper The percentage of the jockey's victories to date

jst1miss A 1 indicates that other variables related to the rider have been lost. Or
otherwise
Table 2.2 . Variables used to carry out the analysis and their definition

2.1.4.2. Experimental setup

Continuous variables in the data set are normalized in a database.

The standardization procedures have been indicated in point 3.3.

A subsample of 200 runs (of the 400 from before May 1999) are used to build an SVM
classifier with RBF- Kernel (13). The parameters C and γ are determined by the means of five-fold
cross-validation [59]. The 200 runs are divided into five partitions of equal size and the recursive
SVM model is built with four partitions, evaluating the result on the remaining one. Performance is
measured in terms of the number of winners that were accurately predicted. 441 different

parameters are considered between the values log( C ) = { - 3, - 2,...,17 } and log( γ ) = { - 20, -
19,...,0 } [60]. The parameter values that lead to a greater number of correct winners identified
during cross-validation are retained, and a final SVM classifier is made with this fit based on 200

4
3
EPISODE 2. Model

runs.

The resulting SVM classification model is used to score the remaining 200 training races,
providing a skill index of the relative strength of each horse and based solely on the fundamental
variables. Subsequently, the skill index is grouped with the track probabilities using CL in step 2
(2.13).

In order to evaluate the profitability of the prediction model, the Kelly Betting Strategy [61] is
implemented. The Kelly Strategy identifies how much to bet on each horse.

If the proposed methodology produces a higher positive return than the other models, it can be
said that the SVM/CL approach adds value.

2.1.4.3. Comparative evaluation of the two-stage model.

Empirical evaluations examine the effectiveness of the SVM/CL model with the RBF Kernel
function (including the database and normalized variables). For example, the Kelly Strategy, based
on the prediction of the probability of winning from the proposed model, obtains a return of
30.58%.

Rate of return t value

R2 AUC
No reinvestment (%) With reinvestment β1 β2
(%)
SVM/CL 30,58 642,65 0,1323 4,85 10,32 0,762
SVR/CL 17,5 211,55 0,1238 2,86 9,57 0,757
CL/CL 1,74 -16,53 0,1231 2,64 10,53 0,759
Table 2.3. Empirical comparison of the three different two-stage models across 156 exclusion runs.
The two-step CL/CL model [42] and the SVR/CL procedure [46] are considered as
benchmarks. Applying these techniques to the same 156 races, a performance rate of 1.74% and
17.50% respectively is obtained (see table 2.3). Similar comparisons can be seen when reinvestment
of profits is allowed. The two vector-based support methods, SVM/CL and SVR/CL, produce a
significant increase in wealth (642.65% and 211.55%, respectively), while wealth decreases by
16.53%. when we use the CL/CL model.

The proposed SVM/CL method outperforms its two competitors, providing significantly
higher profits using the Kelly Criterion with and without reinvestment. The results of using the
Kelly method without reinvestment are a more reliable indicator of the success of the relative
model, in that the profits reaped with reinvestment can arise from the luck of the order in which

4
4
 EPISODE 2. Model

winners and losers occur. As a consequence, in subsequent analyzes we focus on results without
reinvestment.

It could also be argued that relying solely on the profitability of a particular model overlooks
other key performance indicators of a model. Therefore, Table 2.3 summarizes additional
performance indicators to facilitate subsequent comparison of the three methods.

The R² factor represents the percentage of the variation in the profitability of a model. In this
way, that of the SVM/CL model (0.132) exceeds that of its two competitors, indicating that the
generated probabilities contain more useful information on the fundamental variables. This is
confirmed when we examine the values of the t-statistic of B 1 , the CL coefficient associated with
the output of stage 1 (equations (2.11)-(2.13)). The capacity index obtained by processing the
fundamental variables in the first stage with an SVM has the highest t- value and can therefore be
considered the most informative.

Rate of return t value

No reinvestment With reinvestment R2 AUC
β1 β2
(%) (%)
Reference model 30,58 642,65 0,1323 4,85 10,32 0,762
proposed SVM/CL One-stage
CL model
-0,46 -48,37 - - - 0,737
SVM -22,57 -100,00 - - - 0,761

Two-stage model using

different input variables
Normalized database only 5,86 -116,33 0,12 1,14 11,870,756
Only Normalization by races
3,46 -0,04
0,123 3,26 10,720,756
Two-Stage SVM Model with
Linear Kernel
Linear SVM/CL 7,35 120,73 0,1193 0,188 13,20,756
Table 2.4. Empirical comparison of other models with the 156 exclusion races.

The discriminative power of a model in terms of its area under a receiving performance
characteristic curve (AUC) is also considered an indicator of performance (performance) [62]. The
AUC is a popular measure for classifier evaluation. For this application, it represents the probability
that a model assigns a higher chance of winning to a winning horse than to a losing one. Practical
AUC values range from 0.5 to 1 with higher values representing higher discriminative power. The
performance differences in terms of AUC between the three models are minor (see table 2.4). These

4
5
EPISODE 2. Model

results suggest that although SVM/CL is only slightly better in terms of identifying winners, it
especially excels at producing the most accurate winning probabilities and thus making more
profits.

The biggest difference between the approach proposed here and the work of Edelman (2007)
[46] is the use of range ordered data in model construction. The results suggest that classification
may be more reliable than regression for horse racing data. This view is supported by conducting a
formal test to examine whether races that exclude the final winner follow the same distribution as
races in which the winner is taken into account [63][64]. The corresponding statistical test ( x 1
2
3 =
21.20 ) indicates that this hypothesis should be rejected by 7%, providing further evidence of the
unreliability of the order of the final positions.

2.1.4.4. Examining the origin of profit

Despite the attractive performance of the SVM/CL model, SVM models are methods that do
not provide us with an explanation of the relationships between the data. Thus, additional
simulations are required to shed light on the origin of the profit with the Kelly Model-based betting
method. Furthermore, the fact that we use SVM classification in the first stage, we will have three
factors that affect the performance of the model; specifically, perform two-stage modeling that does
not employ the use of track probabilities , employ a pre-test approach processing a special data set
to capture some information about the competition within the race, and using a non-linear model
(e.g. Kernel RBF function), to distinguish between winners and losers. The results of these
experiments are summarized in Table 2.4.

2.1.4.4.1. Comparison between one- and two-stage models.

Regarding the two-stage model approach, Sung and Johnson (2007) [48] show that it is
superior to one-stage models when we use CL. Their results are confirmed by the data used in this
study: a one-stage CL model produces a loss of 0.46% during runs when the Kelly Strategy is
applied without reinvestment. However, the one-stage SVM model performs much worse, having a
loss of 22.57% during the same runs (Table 2.4). The winning probabilities produced by the single-
stage CL or SVM models are significantly lower than the respective two-stage models.

2.1.4.4.2. Comparison of linear and non-linear models.

The proposed SVM/CL model incorporates an RBF Kernel function to account for non-linear

4
6
 EPISODE 2. Model

relationships between fundamental variables. The superior performance of the SVM/CL over the
CL/CL model, which accommodates only linear relationships, indicates that the relationship
between independent variables and race outcomes is not linear. However, to get a clearer view of
this topic we are going to look at the performance of an SVM/CL model with a Linear Kernel (last
row of Table 4). A betting simulation of this model on the 156 reveals that the linear SVM/CL
model produces a lower return (7.35% without reinvestment) than that produced by the non-linear
SVM/CL model (30.58%). Similarly, the other performance indicators demonstrate the superiority
of the nonlinear model. Since the Kernel function is the only difference between these two models,
it can be concluded that the non-linear relationships that exist between independent variables should
be taken into account when modeling the outcome of the race.

2.1.4.5. Debate

It is important to remember that the only variables used in this study are those included in the
article by Bolton and Chapman (1986) [26]. It was expected that since they had been in the public
domain for many years, the public would have complete information on these variables. Despite
this, the CL/CL model manages to generate a small profit over the sample of 156 races (1.74%) if
the profits are not reinvested.

Both methods achieve a significant improvement in terms of benefit (SVM/CL: 30.58% and
SVR/CL: 17.50%). This result confirms the previous results of Edelman (2007) [46] that
incorporate a two-step model that allows including information on the fundamental variables, which
has not yet been taken into account by bettors. Furthermore, the proposed classification-based
approach, SVM/CL has significant additional improvements over an SVR/CL approach. This
confirms the suspicion that we cannot reliably make a forecast based on the regression of final
positions.

However, using classification has some theoretical drawbacks. In particular, the binary
win/lose objective variable is not independent for each horse. A multinominal SVM formulation
[65] could be considered as an alternative. This could involve defining a runner's finishing position
as a discrete target variable and building an SVM model that distinguishes horses that finish first,
second etc. However, such an approach has major drawbacks. In particular, races can include a
large number of riders and a class is needed for each possible finishing position.

A key issue in predicting the outcome of racing events lies within the competition of each
race. From a methodological point of view, the only procedure currently available capable of

4
7
EPISODE 2. Model

pleasing the relationships between runners is CL. On the other hand, the results of Benter (1994)
[41], Edelman (2007) [46] and those presented here indicate that ordinary prediction techniques that
consider each example as independent adapt well, if combined with CL within a second stage. In
fact, the combination with CL is essential as demonstrated in section 4.4.1. Consequently, further
improvements can be expected when it becomes possible to also model relationships between
horses in a race in stage one.

Other models, such as ordinal SVM [66] or Kernel logistic regression [67], could be
considered, since they follow the same ideas as the SVM model, but embody different types of loss
functions.

However, all of these procedures measure the loss over individual examples and aggregate
these values to form an overall measure of empirical error. This is the step where the dependencies
between examples are lost. Recent advances in the field of structural SVMs could offer an
alternative by allowing more complex loss functions that do not restrict individual examples. For
example, Joachims (2005) [68] develops an SVM that optimizes the AUC directly. This technique
appears to be a good candidate for the first stage of horse racing modeling in future work.

2.1.5. Conclusion

A two-stage methodology is proposed to predict outcomes of competitive events. This model

differs from the other two-stage models in that it considers ranking rather than regression in the first
stage to avoid problems with unreliable rank in the order of horses. Instead, it performs a data
standardization procedure within the race to give the SVM-based model information within the
racing competition. Empirical results have demonstrated the effectiveness of the model in offering
considerable accuracy improvements over competing alternatives.

The results indicate that, although horse racing betting models use similar fundamental
variables that have been in the public domain for many years [26][41][44], the betting public has
not yet fully deciphered the information contained in these variables. . This reflects the complexity
of the relationship between the fundamental variables and the outcome of the race, which in view of
the observed results, is prone to include non-linear iterations and could remain hidden from those
who populate the markets. In future work, the techniques of extracting training rules from SVM
classifiers [69] could be applied to explore the nature of the relationships between variables and
improve understanding of the information that individuals in these markets fail to decipher.

4
8
 EPISODE 2. Model

Furthermore, it would be interesting to carry out additional experiments using data after the arrival
of Betfair . Comparing such results to those presented here could help quantify the degree to which
the Internet has changed the focus of horse racing betting markets.

2.2. Applied model

In this section we will apply the model described above to our particular case, so that, first we
will explain how we have found the database of horse racing results, and later we will calculate the
chances that each horse has of winning in a given race. as well as the money we will bet on each
horse, and the total profit we will obtain with those bets.

2.2.1. Formstar website

We have already talked about some of the betting houses that exist for horse racing.

In our case we have based ourselves on data obtained from the Formstar website [1]. It is a
South African page that collects races of various types run in

4
9
EPISODE 2. Model

different racecourses. The choice was made after studying several web pages that collected
information about different horses. On this page we found variables similar to those presented to us
in the previous article, and it also contained numerous information about the characteristics of the
horses, riders or trainers.
We are going to describe the structure of the page in question, since at first glance it may
seem a little complicated to understand.

1. Main page, Home , where they give us some information related to Formstar, news, etc.

2. Meetings tab where the information that will interest us the most is found. As soon as we enter
it we see all the days of races that have been run and some that will be run in the coming days.

In all the windows of this tab some (or all) of the following options will appear:

^ It helps us go to the Races window, wherever we are.

^ It is used to see all the races run previously.

^ It is used to see all the races run with

posteriority.

^ Used to search for the career you want without having to scroll down
manually.

^ We get the betting statistics that have been the best throughout
over the years (since 2003). The studies are from the first, second, third
and a quarter.

5
0
 EPISODE 2. Model

^ It is one of the most important options, since it explains one by one

all the columns that contain each page, the meanings of the words used,
the numbers, the initials, etc.

It is used to go to the field window. This option is found in

Forecast and Result.
-

^ To move to the Result window.

^ Go to the Forecast
window.such as:
They give us information

• Race date.

• Where the race takes place.

• Type of terrain.

• If it has already been run (Result) or if it is still pending (Forecast).

• The race number.

• The number that the horse has on its bib number.

• Name of the horse on which the highest bet has been placed.

• The odds of that horse.

• If the race has already been run, the position in which the horse that was thought to win was
finally placed.

• And finally the starting price, that is, the fee that will be paid for that horse when the race
begins.

Next we can go into a race day and we find all the categories that have been run. That is, in one race
day several races of different or the same category could be held (at different times and with
different horses). We are in the Races window.

5
1
EPISODE 2. Model

In this case the different columns give us information about:

• Race number.

• Type of career.

• Length of the racecourse.

• Bets placed about that race.

• Strength of the horse (it is a factor calculated by the page itself).

• Number of horses participating.

• Name of the winning runner (in case it has already been run).

• Time taken by the winner to complete the entire race.

• Bib number of the horse that, according to the page itself, is most likely to win.

• Previous horse name.

• Odds calculated by FormStar.

• Final position of that horse.

• The current starting price.

The next window that we find (by clicking on the type of race you want) is the Forecast . On this
page we will have more information about the horses that have completed the race

• Predicted position of each horse.

Position in which it actually ends (if known).

Runner's bib number.

Last five results of the broker. Where 1 = first, 2 = second,…, 0 = tenth or worse.

Name of the horse. If the horse has not been bred locally, the origin is in parentheses.

If it has blinkers.

5
2
 EPISODE 2. Model

Starting position.

Handicap ranking of each horse. The bigger the better.

Rider's name.

Rider rating.

Total mass carried by the horse.

Rank made by the handicap official based on the merit of the horse.

Horse trainer.

Coach rating.

Rating of the trainer and rider combination.

FormStar Rating.

How many meters away is each horse predicted to be from the winner.

The odds calculated by FormStar of the runner winning in that race.

Starting price of the horse. F denotes the favorite (only if the result is known).

Races run by the horse.

Races won by the horse.

Races in which the horse finished placed.

Bets won (approximately) by the horse.

For Premium members there is the option to add a comment about the race.

We are also given the option to see more data about the horses by clicking on the one we want. The
data they offer us in the Form tab are:

Date.

5
3
EPISODE 2. Model

Racecourse. We are going to find some acronyms whose meaning is:

either Arlin: Arlington.

either Bftn: Bloemfontein.

either Clwd: Clairwood Park.

either Dbvl: Durbanville.

either Fair: Fairview.

or Flmg: Flamingo Park.

either Gosf: Gosforth Park.

either Grey: Greyville.

either Kenw: Kenilworth.

either Kimb: Kimberley.

either Maur: Mauritius.

either Nmkt: Newmarket.

either Rand: Randjiesfontein.

either Scott: Scottsville.

o TrfN: Turffontein (New course - inside track).

either Turf: Turffontein.

either VSnd: The Vaal (Sand).

either Vaal: The Vaal .

either Zimb: Zimbabwe (Borrowdale).

Ground conditions, whose meaning is:

5
4
 EPISODE 2. Model

either F: firm.

either G: good.

either H: dense

either A: firm.

either S: flexible/soft.

either And: soft.

Race type:

either AdvP: Advanced Plate.

either AlwP: Allowance Plate.

either Cndp: Conditions Plate.

either Fhcp: Durbanville.

either FSt2: 2yo Feature Stakes.

either FSt3: 3yo Feature Stakes.

either FStk: Feature Stakes.

either GrdP: Graduation Plate.

either HcpA: A Division Handicap (merit rated 99 and

over).

either HcpB: B Division Handicap (merit rated 91 to 98).

either HcpC: C Division Handicap (merit rated 83 to 90).

either HcpD: D Division Handicap (merit rated 75 to 82).

o HcpE: E Division Handicap (merit rated up to 74).

either JuvP: Juvenile Plate.

5
5
EPISODE 2. Model

either MJvP: Maiden Juvenile Plate.

either MdnP: Maiden Plate.

o NovP: Novice Plate.

either PrgP: Progress Plate.

either SupP: Superior Plate.

Power of each runner in the race.

Distance.

If the horse wears blinkers.

Starting position.

Rider's name.

Weight carried by the horse in kg.

Rank made by the handicap official based on the merit of the horse.

• Coach.

• Ranking of the runners with the least advantage according to Formstar (before the race).

• Number of runners.

• Final position

• How many runners are left from the first, if it is the first, how much is left from the
second.

• Runner with a disadvantage in the race according to FormStar.

• Time to complete the race (in seconds).

• FormStar speed ratings for runners.

5
6
 EPISODE 2. Model

• Corridor starting price. F is the favorite.

• The number of races completed by the horse.

• Number of races won.

• Number of times it has been placed.

• Bets won (approximately) by the horse.

From here we are going to briefly describe the rest of the tabs that we find on the Formstar
website, but that in this Project we have not taken into account to create the database.

3. The next tab we have is Horses . What we find in it is a summary table with a classification of
the twenty best horses. We are given information such as the situation within the ranking,
position in the last five races, name of the horse, description of the horse, races run, races won,
percentage of races won, number of times it has finished placed, percentage of the number of
times it has been placed, the Formstar rating, money won, whether the horse is still active or not
and the last race it had completed.

4. Next we have the Connections tab with information about Trainers and Jockeys .

The data they provide us from the riders and the trainers are the same. The position he occupies
within the classification, his name, races run, races he has finished first, second, third and
fourth (each one in a different column), percentage of races won, percentage of races in which
he has finished placed, bets won and Formstar ranking which is calculated in the case of
jockeys by taking an exponentially weighted average of the number of points of the horses
ridden by the jockey and in the case of trainers of the horses that have been trained by him.

5. The following tabs: Subscriptions , Support , Info and Login do not provide any relative data
that can be taken into account for creating the database.

2.2.2. Database

One of the main objectives of this Final Year Project was to create a database with the results
of past races. This is because, unlike in other sports, in which we are concerned, we do not have the
results in an easily accessible way. Therefore we had to obtain them from web pages that made the
information available to the user as the races were carried out, and, if possible, that also had a

5
7
EPISODE 2. Model

history of previous races to cover a greater amount of data.

To create the database it was necessary to create a program that would read from the chosen
web page (explained in the previous section).

The web page is treated like a file, so we must open it to start reading it, in addition to closing
it when we finish.

To make it easier to read, we have removed the bold (there were words that were in bold and
others were not, without following any logic), and we have also made a left justification of all the
columns. Therefore, all the data that interests us will be between the html code: <td class==left>
and </td>. In this way we will look for that code, and once we are placed in the exact place, we will
collect the data that will interest us.

We have not found any page that had exactly all the variables that have been presented to us
in the article, however, we can use others with the same results. The tab, of those described before,
that we are going to use is going to be Forecast .

To save the variables, or references to them, that will be part of our database, we are going to
create a numerical matrix.

The first variable in the file that will interest us will be the one in the second column, which is
the final position in which each horse finished. Since we are using past races to create the
database, this position will already be known.

Here the only thing we are going to differentiate is whether a horse has come in first place or
not, therefore we are going to create a column with a 1 in the winner's position and a 0 in the
rest. To do this, what we do is read the value, if it is equal to 1 we assign a 1 to the column of
a numerical matrix, which we will call result, if it is different from 1 we give it a 0. In the
following way:

Winner

Horse 1 0

horse 2 0

Horse 3 0

Horse 4 1

Horse 5 0

Horse 6 0

5
8
 EPISODE 2. Model

Horse 7 0

Horse 8 0

Horse 9 0
Table 2.5 . Matrix example of the winning horse
Another piece of information that we are going to take into account is going to be the last five
positions in which the horses that run the race have finished. This information is provided to us
by the fourth column of the file ( Form ). We are going to save each position in a different
column of the numerical matrix. That is, the first column will contain information about the
position in which a given horse was five races ago, the second four races ago, and so on. When
a horse has not completed the five races we will put the value 11 in that column. Furthermore,
when it has finished in tenth position or worse, we will fill it with a value of 10.

To assign it, what we have to do is check the length of what we are reading, with this we will
know how many previous runs it has done, and thus we will know if it has not done 5 runs
previously, so we will fill those positions with a value of 11. We will also have to make sure if
what we are analyzing is a 0 or another value, so that if it is a 0 we will assign a 10 to that
position of the numerical matrix and if it is another value we will fill it with ourselves.

Let's give an example: We assume that what we are reading is 503. As we see, the length is 3.
So we will fill the first two columns of the numerical matrix with an 11. The next one is a 5, it
is neither 0 nor empty, so we assign it directly. Since the posterior is a 0, in that case we will
put a 10 in that position. We can put 3 directly as well. This is how it would look:

Column 2 Column 3 Column 4 Column 5 Column 6

11 11 5 10 3
Table 2.6. Example of filling columns with Form variable

We are going to fill the next six columns of the numerical matrix with the handicap ranking of
each horse, the rider's rating, the total weight carried by the horse in kilograms, the horse's
merit classification according to Formstar , the trainer's rating and the combination of rider and
trainer qualifications. We assign these data directly to our numerical matrix, since in all cases
the data is given to us in the form of numbers.

The next piece of information that interests us is column sixteen of the file, this is the
classification of the horse according to Formstar . Since it is a number I can assign it directly
to the numerical matrix. If in any case we find that the variable poor is in a position, in the
corresponding position of the numerical matrix we will put the value -2.

5
9
EPISODE 2. Model

In column seventeen is the forecast of how many meters the horse has left behind the first,
which is also a number so we do not have to do anything to assign it.

Then we read column eighteen and nineteen from the file, here we find the horse's odds and its
starting price, respectively. They give us both data in “ab” format. What we want is the
relationship between them, so we are going to read from the beginning to the script and save it
in an auxiliary variable, and then from the script to the end, saving it in another variable. In
some cases instead of “-” we find a “/”, that is, “a/b”. In this case we look for the slash and
replace it with the hyphen, thus we proceed in the same way as if from the beginning the
format was “ab”. Once the two values are obtained, we divide the first by the second, that is, in
the example a divided by b.

For example:

Odds aux1 aux2 Column 15

7-2 7 2 7/2 = 3.5
Table 2.7 . Example of filling columns with quotas

Another thing to keep in mind is that there are horses in which, along with the odds and the SP,
we have an “F” denoting the favorite. For it, we are going to make a column whose values will
be 0 (if there is no F) and 1 (if there is F).

There are times when we find that due to some error, these two columns do not have any value,
that is, those cells are empty. In that case, what we have done is assign the word “NOTHING”
to the corresponding position of the html page in question, that is:

<td class==left></td> ^ <td class==left>NOTHING</td>

Therefore, we also have to check if what we are reading is the word “NOTHING” and if so,
we put a 0 in the four positions of our matrix (both the odds value and the SP, as well as the
corresponding F)

• Next, we again have three numerical values, which are R (races run by the horse), W (races
won) and P (number of times it has been placed).

• And finally, in column twenty-three of the file are the bets won by the horse. In this case the
data is given to us in decimal format with a comma and in order to assign it to the numerical
matrix we need it in decimal format with a point. To do this, we replace one with the other

6
0
 EPISODE 2. Model

and assign it with double format, for high numbers.

Now we are going to proceed to read other data that interests us from the file, to have
important information on the characteristics of the horses that perform the race, but that are found in
the title of the page. These are terrain, distance and R factor.

To do this we first have to place ourselves in the appropriate position in the file. We do this by
looking for the first word of the phrase and placing ourselves right where it ends. Also here
everything we read will be between the html codes <td class==left> and </td>.

• The first thing we are going to read is going to be the type of terrain. To do this, we are going
to look for the hyphens of the phrase in which it is immersed, which is in the header of the
web page, behind Meeting .

For example: Meeting: Sat 12-Feb-2011 - Kenilworth - Going: Good

Since what interests us is from the fourth script onwards, we place ourselves there. From the
fourth dash to the end there will always be the word Going : followed by the terrain, so we are
not going to read from the script itself, but rather a few positions later (to be specific, seven
positions) until the end. To assign it to the numerical matrix we have made a correspondence
as follows:

Land type Numerical

correspondence
good 1
Firm 2
Heavy 3
hard 4
Soft 5
Yield 6
Table 2.8 . Numerical correspondence with the terrain

The next two pieces of information are found in the same sentence, after the word Race , so
we have to search for that word and the hyphens only once.

In these last two cases we will have to distinguish between two situations:

o That the race is in the Division . Example: Race: 6 - D Div Handicap (MR75-82) - 1000m -

6
1
EPISODE 2. Model

R 57,040

o The rest of the races. Example: Race: 4 - Maiden Plate - 2500m - R 53,360

This is so, because our way of searching for the data that interests us will be by searching for
the script from which it is found, and in these two cases the number of scripts is different. Having
four scripts in the case of Division races, since there is one script immersed in the type of race, and
three scripts in the rest.

• The distance of the race is the one that comes first. Therefore, what we are looking for will be
between the third and fourth script in the Division races, and in the rest the information will
be between the second and third script.

For the R factor, we will have to distinguish it as well. In the first case it will be between the
fourth script and the end. And in the second it will be between the third and the end.

Summary mode:

o Division Race:

- Distance: between third and fourth dashes.

- R factor: between fourth script and finale.

o Rest of races:

- Distance: between second and third dashes.

- R factor: between third script and finale.

We attach a table as a summary of the above.

Column
Data
number
Horse position 5 races ago

1 10: Position ≥10

11: Unrun race

6
2
 EPISODE 2. Model

Horse position 4 races ago

2 10: Position ≥10

11: Unrun race

Horse position 3 races ago

3 10: Position ≥10

11: Unrun race

Horse position 2 races ago

4 10: Position ≥10

11: Unrun race

Previous race horse position

5 10: Position ≥10

11: Unrun race

6 drtg
7 Rider rating

6
3
EPISODE 2. Model

8 Weight (Kg)
9 M.R.
10 Coach rating
11 Rider-trainer combination
Fcast
12 -2: Poor

13 Meters from the first

14 Odds
Favorite Horse

15 1: Favorite
0: Not Favorite

16 SP
Favorite Horse

17 1: Favorite
0: Not Favorite

18
Running races
19 Races won
20 Number of times placed
21 Bets won
Land
1: good
2: Firm
3: Heavy
22
4: hard
5: Soft
6: Yield

23 Distance
24 R-factor
Table 2.9. Numeric array data summary
Below we are going to give an example of how the numerical matrix would look in a
particular race.

6
4
EPISODE 2. Model

Column 1 Column 2 Column 3 Column 4 Column 5 Column 6 Column 7 Column 8 Column 9 Column 10 Column 11 Column 12 Column 13
Horse 1
2 11 11 11 11 25 -27 60 0 25 -30 1213 0
horse 2
7 6 9 2 11 -3 25 60 68 39 4 1049 1,2
Horse 3
10 3 11 11 11 -165 -27 60 0 62 -26 1032 1,33
Horse 4
10 8 3 4 11 -81 -20 60 70 -62 -25 998 1,58
Horse 5
4 6 4 11 11 -78 19 60 75 -62 -14 933 2,06

Horse 6 10 10 7 9 10 -52 64 60 55 -18 -108 700 3,77

Horse 7
10 7 9 8 10 -112 18 60 66 -79 -37 628 4,3
Horse 8
2 10 4 10 10 -102 -77 60 61 -79 -168 567 4,75
Horse 9
10 11 11 11 11 -55 38 60 0 -192 -18 513 5,14

Table 2.10. Example of numeric matrix with a particular race (columns 1 to 13)
Column 14 Column 15 Column 16 Column 17 Column 18 Column 19 Column 20 Column 21 Column 22 Column 23 Column 24

Horse 1 2,5 0 1,4 1 1 0 1 10,672 1 1200 53,36

horse 2
4,5 0 3,5 0 4 0 1 10,672 1 1200 53,36
Horse 3
5 0 6,5 0 2 0 1 5,336 1 1200 53,36

Horse 4 5,5 0 6,5 0 4 0 2 8,004 1 1200 53,36

Horse 5 7 0 5,5 0 3 0 2 5,918 1 1200 53,36

Horse 6
20 0 22 0 15 0 5 43,5 1 1200 53,36
Horse 7
25 0 25 0 9 0 1 11,6 1 1200 53,36
Horse 8
40 0 50 0 29 0 16 108,518 1 1200 53,36
Horse 9
66 0 14 0 1 0 0 0 1 1200 53,36

Table 2.10. Example of numeric matrix with a particular race (columns 14 to 24)
68
 EPISODE 2. Model

2.2.3. Inference

The main objective of this section is to obtain, from our database calculated in the
previous point, the probabilities that each horse has of winning in each race.

To test the model, what we have done is have all the E Division category races
whose racecourse is 1400 meters long. In this way we have 40 races, each one with a
certain number of horses.

The first thing we have to do is divide the data set of the 40 races into the
following subsets:

• Training data. They are those that are used to train the machine in order for it
to better adjust to our variables.

• Test or test data. These data will allow us to validate how good the analysis is
in another subset not included in the previous one.

To divide the data set we have used two techniques that are based on cross-
validation, also known as cross-validation . It is a technique used to evaluate the results
of a statistical analysis, and ensure that they are independent of the partition between
training and test data. It is used to estimate how accurate a model is.

Cross validation [72] is a way to evaluate a model with a hypothetical test data set,
when we do not have the set itself.

There are many types of cross validation, however the ones we are going to use
are:

1. Cross validation of K iterations (CV). In this type of validation, the sample data is
divided into K subsets, one of which is used as test data, while the rest (K-1) are
used as training data. The cross-validation process is repeated during K iterations,
with each of the possible test subsets. The choice of the number of iterations
depends on the size of the data set.

This method is very precise, since we evaluate from K combinations of training and
test data. It has a disadvantage, which is that it is quite slow from a computational

6
9
EPISODE 2. Model

point of view.

In our case, because we have 40 data samples, what we are going to do is divide the
set into five subsets, each with eight different runs. Thus, the training set will have
four of the subsets and the test set will have one.

So each iteration will be:

• Iteration 1.

o Training set. Values between 9 and 40.

o Test set. Values between 1 and 8.

• Iteration 2.

o Training set. Values between 1 and 8, and between 17 and 40.

o Test set. Values between 9 and 16.

• Iteration 3.

o Training set. Values between 1 and 16, and between 25 and 40.

o Test set. Values between 17 and 24.

• Iteration 4.

o Training set. Values between 1 and 24, and between 33 and 40.

o Test set. Values between 25 and 32.

• Iteration 5.

o Training set. Values between 1 and 32.

o Test set. Values between 33 and 40.

2. Leave-one-out (LOO) cross-validation . In this case, what we are going to do is

separate the data so that for each iteration we have only one test sample and we use

7
0
 EPISODE 2. Model

the rest as training data.

In this case the error is very low, however it is very computationally expensive,
since it has to carry out a large number of iterations, as many as there are samples
we have.

In our case we will have 40 iterations, these being:

• Iteration 1.

o Training set. Values between 1 and 39.

o Test set. Value 40.

• Iteration 2.

o Training set. Values between 1 and 38 and 40.

o Test set. Value 39.

• Iteration K.

o Training set. Values between 1 and 40 except the K value.

o Test set. K value.

Once we have divided the data between the test set and the training set, we are
going to perform an SVM on the training sets.

SVM [71] is a set of supervised learning algorithms (technique for deducing a

function from training data). Given a set of training samples we can label the classes and
train the SVM to build a model that predicts the class of a new sample. An SVM is a
model that represents sample points in space, and is responsible for separating the classes
by as wide a space as possible. The new (test) samples will be classified into one class or
another depending on their proximity to them. The more separation there is between
classes, the better the classification will be.

The way the SVM works is by constructing a hyperplane that optimally separates
the two different sets of classes. This optimal separation is because the SVM looks for

7
1
EPISODE 2. Model

the hyperplane that has the maximum distance with the points that are closest to it. In
this way, the points of the vector that are labeled as belonging to one category are on one
side of the hyperplane and those of the other category are on the other side. The vector
formed by the points closest to the hyperplane is called the support vector.

Figure 2.2 . Example of hyperplane with SVM function

7
2
 EPISODE 2. Model

The construction of the hyperplane must be carried out avoiding two phenomena:

Underfitting: it is based on obtaining less statistical information from the

samples than we could obtain. That is, we made a simpler border than it
should.

Overfitting: we overfit the data, building an excessively complex frontier. In

this way we lose the ability to generalize, that is, for the data we already have
(training data) it works very well, but for the new data (test data) it will not
be valid.

In our development we have used the two-stage model described in equation

(2.13) of the article presented above. That is to say,

f svM ( x ) = 2 &iy i exp ( v x, - x II 2 )

|+bˆ
\ie SV )

{ α ˆ, bα } ← max =
M 1
ˆ
∑α i

Step 1: i = 1
x i - x K II 2
αα
∑
M
- 1 1
i K y i y K exp )

2
i,k=1

M1

s.t. α y = 0; 0 ≤ α ≤ C ∀ i = 1,..., M
ii Yo 1
i=1

j= β exp ( f ( x ))
1 SVM i
j

p i
mj
,

Step 2:
∑ exp ( f SVM ( x )) + β
i
j

i=1
2 q i
j

i = 1,..., M 2 ; j = 1,..., R 2

To perform the SVM we have used the svmtrain function [70], which has as
parameters:

• Boxconstraint or c parameter. It is the smoothing parameter. In our case the

parameter c will take the values 1, 2.5, 4, 10, 25, 50 and 100. We have chosen
several values, very different from each other, to see how the result varies

• KernelCacheLimit. Value that specifies the cache memory size that we are going to
use to execute our program. The larger it is, the faster the execution will be,

7
3
EPISODE 2. Model

however, the more memory it will use to the point that execution can be interrupted.
Therefore, a compromise must be reached between speed and resource consumption.
We have used the value 1000 as the average cache memory size.

• Kernel_Function . At this point is where we assign the Kernel function that we are
going to use to perform the training. There are several types:

or Linear. It uses a linear Kernel function, that is, a scalar product will be created.

or Quadratic. Quadratic kernel function.

or Polynomial. Polynomial kernel function. By default, grade 3, if you want another

grade, you must specify it.

or RBF. Radial Base Gaussian Kernel Function, like the one we have described in
the article in point 3.1. As we have said that we are going to follow the model,
our Kernel function is going to belong to this class.

• Rbf_Sigma. In Kernel RBF functions we are going to need a scale factor. This is
where we will put said factor.

To train the machine, we are going to use various sigma values. These are 0.4145;
0.829; 1,658; 3,316; 6,632; 13,26 and 15. As in the case of the parameter c , here we
have also chosen very different values.

• Method . It refers to the method that we are going to use to find the separation
hyperplane. There are several options:

or QP. It refers to quadratic programming.

or SMO. Sequential Minimum Optimization, which is an improved SVM algorithm

that works with large quadratic problems, which divides them into smaller
problems. In this way the computing time is less and the generalization error is
minimized. We are going to use this method, since this way it will be resolved
more quickly.

or LS. It is the least squares method.

When we use the SMO method, we have to introduce a series of parameters, these

7
4
 EPISODE 2. Model

will be:

• Display . Specifies the level of information about optimization iterations that show
how the program runs. It has options:

o Off . It does not report any type of information.

o Iter . Make a report every five hundred iterations.

o Final . It is only informed when the program finishes executing.

• MaxIter . It is an integer that specifies the number of iterations to be performed.

have our program. In this case there is also a compromise to be made, because if the
number is too small and this limit is exceeded before the algorithm converges, then
the algorithm stops and returns an error. However, if the number of iterations is very
large, the program will take a long time to execute. In our case this number will be
200,000, since it is a very large program, and therefore, with this number we ensure
that we will not have any execution errors.

Once we have performed the SVM function, we have to perform a Kernel function
[71]. We are going to have to use this due to the computational limitations of linear
learning machines. The representation through Kernel functions is based on projecting
the information to a higher dimensional feature space, which increases the computational
capacity of linear learning machines. That is, we will map the input space to a new space
of higher dimensionality.

7
5
EPISODE 2. Model

Having this data we can now move on to the second stage of the model we have
been following. That is, calculate, through what was previously found, the probability
that each horse has of winning within each race.

p j
i β 1 exp ( f ( x ))
SVM i
j

Step 2: mj

∑i=1
exp ( f ( x )) + β
SVM i
j
2 q i
j

i = 1,..., M 2 ; j = 1,..., R 2

To calculate the value of β 1 and β 2 we will take into consideration that β 2 = (1 -

β 1 ) . In this way, we are going to encounter different situations:

• P 1 = 1^yp 2 = 0 . In this case, we will only take into account the results calculated
through the SVM.

• p 2 = 1 and P 1 = 0 . Here we will only use the predictions made by

The experts.

• 0 < ^1 < 1 . In this case we are going to calculate the probabilities with both
predictions. Giving more weight to one or the other depending on how large β 1 is.

2.2.4. Odds and Q Factor

The calculation of the odds based on the probabilities of each horse is done with:

O= QW
-1 (2.14)
ipi

In which:

• Or i are the betting house's odds .

• P i in the probability of horse i of winning the race according to the market.

• Q is the percentage as much as one of what is distributed. That is, there is a

so many percent that the betting house keeps, and the Q is the remainder,
which will be distributed among all the winners of the bets.

7
6
 EPISODE 2. Model

• W=1

If we compare it with the same formula in the case of football,

O= QW
Yo Pi

We find that in this case we do not have a factor of 1 subtracting, this is so, because in the
case of horse racing, the betting houses, when you make a winning combination, in addition to
giving you the corresponding amount for having won, they return what you bet. . For example, if
you bet 0.30 cents on the horse that is first, if the profit is €1, the total you will win will be €1.30.
In the case of football you would only win 0.30 cents (you don't get your bet back).

On the other hand we also know that:

∑
i=1
Pi=1 (2.15)

That is, the sum of the probabilities of winning for each horse in a race must be 1.

Therefore, we have a system of N+1 equations (2.14 and 2.15) with N+1 unknowns ( Q and
P N ).

If we solve (2.14) we have:

( O 1 + 1) ⋅ p 1 − Q = 0
( O 2 + 1) ⋅ p 2 - Q = 0

( O N + 1) ⋅ p N − Q = 0

Therefore, we are going to solve these equations through a matrix system in the following
way:

7
7
 EPISODE 2. Model

(O1 + 1 0 0 -1 > 'p1 (0

Or 2 + 1 0 -1 ^ )
0:
0 • : — :
0 0 0 OR N + 1 PN 0
s 1
1 1 0
> l Q) l
1
)

-1 (0 ( p 1)
' Or 1 +1 0 l 0 -1 )
)
0 Or 2 + 10 : -1
: : 0 : • : — :
0 0 0 ORN + 1 0 PN
^1 1 l 1 0) l1 lQ)
>
That is, to calculate the vector of P and Q we have to do the inverse of the
odds matrix.

In this way we would already have all the necessary data to calculate our probability of a
horse winning in a certain race.

Let's give an example of what was shown above, so that it is understood.

better.

We have a race with ten horses. Each horse has the following odds
according to Formstar :

Horse Horse Horse 3 Horse Horse Horse 6 Horse Horse 8 Horse 9 Horse
1 2 4 5 7 10

Odds 3,5 3,8 9 6,5 5,5 13 21 16 34 67

Table 2.11. Example with the odds of the horses in a race

If we substitute in the last matrix of the development:

7
8
 EPISODE 2. Model

^ 3,5 0 -11
+1 0 3,8 +1
L

0
0
-1
(0
1 1
(p 1
• 0 • • — •
0 0 0 67 +1 0 p 10
1 1
1 1 0J <
1Q
We will have to: 1J J

-1
^ 4,5 0 L 0 ( 01
0 4,8 0
0 •
• • 0
■ 68
0 0 0 .1
1 >

Horse 1. p i = 0.2679

Horse 2. p2 = 0.1705

Horse 3. p3 = 0.1443

Horse 4. p4 = 0.0938

Horse 5. p5 = 0.0938

Horse 6. p6 = 0.0853

Horse 7. p7 = 0.0721

Horse 8. p8 = 0.0447

Horse 9. p9 = 0.0184

Horse 10. p10 = 0.0093

Q value: Q = 0.9378 . That is, the Formstar betting house keeps 6.22% of the amount
that users bet on it.

7
9
 EPISODE 2. Model

Horse Horse Horse Horse Horse Horse Horse Horse Horse Horse
1 2 3 4 5 6 7 8 9 10

p 0,2679 0,1705 0,1443 0,0938 0,0938 0,0853 0,0721 0,0447 0,0184 0,0093

Table 2.12. Example with website odds for each horse in a race

Once we have the value q we can substitute it into equation (2.13) and, from
i
j

In this way, we calculate our chances of a horse winning a race.

For the example above, the final probabilities for each horse will be:

Horse 1. p1 = 0.0721

Horse 2. p2 = 0.0519

Horse 3. p3 = 0.0648

Horse 4. p4 = 0.0600

Horse 5. p5 = 0.0440

Horse 6. p6 = 0.0884

Horse 7. p7 = 0.0805

Horse 8. p8 = 0.0773

Horse 9. p9 = 0.1833

Horse 10. p10 = 0.2777

Horse Horse Horse Horse Horse Horse Horse Horse Horse Horse
1 2 3 4 5 6 7 8 9 10

p 0,0721 0,0519 0,0648 0,0600 0,0440 0,0884 0,0805 0,0773 0,1833 0,2777

Table 2.13. Example with our probabilities, calculated through the odds , for each horse in a race

8
0
 EPISODE 2. Model

To have more profits, one of the objectives was that the probability of winning for each
horse, calculated through the SVM, was different from that calculated in the betting house on
which we are based. In this way, if we find a horse that has, according to us, more possibilities than
those considered by the rest, our benefits will be greater.

In this example we find:

Horse Horse Horse Horse Horse Horse Horse Horse Horse Horse
1 2 3 4 5 6 7 8 9 10
Model
probability
with SVM 0,0721 0,0519 0,0648 0,06 0,044 0,0884 0,0805 0,0773 0,1833 0,2777
Probability
Formstar 0,2679 0,1705 0,1443 0,0938 0,0938 0,0853 0,0721 0,0447 0,0184 0,0093
Table 2.14. Comparison of our odds with Formstar 's, for each horse, in a race

■ Model probability with

SVM
□ Formstar Probability

Figure 2.3. Comparison of our odds with Formstar's, for each horse, in a race

As we see, it would be more profitable for us to bet on horses 6, 7, 8, 9 and 10. Above all, if
we bet on horses 9 and 10, and one of those two turns out to be the winner, our profit would be
very great, since we are betting on a horse that in reality, for the rest of the bettors, had almost no
chance of winning. win, and therefore, our quota will be multiplied by a greater number.

8
1
 EPISODE 2. Model

2.2.5. Kelly criterion

The Kelly Criterion [61] is a very useful tool in the world of betting, since it tells us how
much we should bet to obtain profits.

The Kelly Criterion formula was developed in 1956 by John Kelly [72], an American
scientist whose objective was to investigate the best option for managing a monetary portfolio. It is
used in the stock market to create stock portfolios based on the expected profits of each security
and the estimated probabilities for a specific period of time. It is also used by expert Poker and
BlackJack players, as well as by betting house players and in general, all kinds of games.

It is designed to maximize the growth of our available money or bankroll in the long term,
estimating the optimal amounts to play in each bet or stake . For this method to be profitable, we
must be able to estimate the odds better than those of the bookmakers. The more different our odds
are from those of the bookmakers, the greater our profit will be.

There are bettors who consider this criterion too risky, since its effectiveness depends on our
predictions being better than those of the bookmakers, and that is very complicated. For this
reason, variants called Kelly Fractional are used in which you bet half or a quarter of what is
recommended, thus minimizing the risk of losing a little, but the benefits obtained in the long term
are smaller.

Something very important about using the Kelly Criterion is that using this strategy we will
never go bankrupt, since the amount to bet is determined by a percentage of the bankroll .

There are other bankroll management systems, but none revolutionized the world of betting
as much as the Kelly Criterion.

To use the Kelly Criterion, the first thing we have to do is calculate our probabilities (in our

case applying an SVM). We also need the returns from the bookmaker ( α ) where we intend to
stake .

This criterion can be summarized in the following stages [73]:

1. Taking the returns of the betting houses and the probability calculated by us previously as
known, we have to multiply between both, for each of the cases, and order them from highest

8
2
 EPISODE 2. Model

to lowest. so that

p( s)⋅ ≥p(s+1)⋅
αs αs + 1

Being s a certain event.

2. When we have the indices ordered, we must calculate b . This being the smallest value greater
than 0 resulting from:

1p
F= t
t = Event 1, Event 2... Eventot
t
1- σ t

Where:

tt

pt = ∑p (s), σ t = ∑ and F0=1

1 1 αs

3. Finally we calculate the vector of the stakes , or what is the same, the vector that returns the
amount of money that we are going to bet on each event.

a(s)=p(s)-b
αs
To further explain it, we are going to illustrate it with the example of a race that has ten
horses.

We start with the following values:

α ^ Odds for each horse in a race, given by Formstar .

p ^ Probabilities obtained using the SVM

Horse Horse Horse Horse Horse Horse Horse Horse Horse Horse
1 2 3 4 5 6 7 8 9 10

p 0,0721 0,0519 0,0648 0,0600 0,0440 0,0884 0,0805 0,0773 0,1833 0,2777

8
3
 EPISODE 2. Model

α 3,5 3,8 9 6,5 5,5 13 21 16 34 67

Table 2.15. Values of the probabilities obtained with SVM and the odds of each horse in a race for the Kelly Criterion

Once we have these two data, we must multiply them both and order them from greatest to
least.

Horse 1: 0.00721 ⋅ 3.5 = 0.2525

Horse 2: 0.0519 ⋅ 3.8 = 0.1971

Horse 3: 0.0648 ⋅ 9 = 0.5824

Horse 4: 0.0600 ⋅ 6.5 = 0.3901

Horse 5: 0.0440 ⋅ 5.5 = 0.2422

Knight 6: 0.0884 ⋅ 13 = 1.1488

Knight 7: 0.0805 ⋅ 21 = 1.6910

Knight 8: 0.0773 ⋅ 16 = 1.2366

Knight 9: 0.1833 ⋅ 34 = 6.2331

Knight 10: 0.2777 ⋅ 67 = 18.6068

In an orderly manner they would be: Horse 10 > Horse 9 > Horse 7 > Horse 8 >
Horse 6 > Horse 3 > Horse 4 > Horse 1 > Horse 5 > Horse 2

Then we have to apply the following formula

8
4
 EPISODE 2. Model

1- p
Ft = t = Horse 1, Horse 2... Horse
t

1 -
We do not know the order of t in advance, in our case it depends on the
number of horses running the race to be studied.

• Knight 10
1 - O 2777
F ( Horse 10) = 0, 777 = 0,7332
1- 1
67

Horse 9

1-0,2777-0,1833
F ( Knight 9) = = 0,5640
67+34

Horse 7

1-0,2777-0,1833-0,0805
F ( Knight 7) = 1
= 0,5049
1
67 + 34 + 21

Horse 8

1-0,2777-0,1833-0,0805-0,0773
F ( Knight 8) = = 0,4508
1
1-
67+34+21+16

• Knight 6
1-0,2777-0,1833-0,0805-0,0773-0,0804
F ( Knight 6) = 1 = 0,3809
1
67+34+21+16+13

8
5
EPISODE 2. Model

Horse 3

1 - 0,2777 - 0,1833 - 0,0805 - 0,0773 - 0,0804 - 0,0648

F ( Horse 3) = , , , , , ,

1
1-
67+34+21+16+13+8

F ( Horse 3) = 0.3469

Horse 4

1-0,2777-0,1833-0,0805-0,0773-0,0804-0,0648-0,0600
F ( Knight 4) =

67+34+21+16+13+8+6,5

F ( Horse 4) = 0.3337

Horse 1

1-0,2777-0,1833-0,0805-0,0773-0,0804-0,0648-0,0600-0,0721
F ( Horse 1) =

67+34+21+16+13+8+6,5+5,5

F ( Horse 1) = 0.4400

Horse 5

1-0,2777-0,1833-0,0805-0,0773-0,0804-0,0648-0,0600-0,0721-0,0440
F ( Horse 2) =

67+34+21+16+13+8+6,5+5,5+3,8

F ( Horse 2) = 1.4357

We calculate the value of b , this is 0.3337, and it is also located in position F 7 .

8
6
 EPISODE 2. Model

Finally we have to obtain the stakes . We are going to bet up to the horse that in the
ordered vector was in position 7, therefore, we will bet on Horse 10, Horse 9, Horse 7, Horse
8, Horse 6, Horse 3 and Horse 4. So we are left with:

Horse 1: a = 0
• 1

• Horse 2: a = 0
2

• Horse 3:

b 0 3337
0,3337
- = 0,0276
9
a =p
3 3 α 3 = 0,0647 -
Horse 4:
•
b 0 3337
, 7
- = 0,0087
6,5
a =p
4 4 α 4 = 0,0440 -

• Horse 5: a 5 = 0

• Horse 6:

b
-
0 3337
, 7
= 0,0627
a =p
6 6 α 6 =0,0884- 13

• Horse 7:

b 03337
0,3337
- = 0,0646
21
a =p
7 7 α 7 = 0,0805 -
Horse 8:
•
b 0 3337
, 7
- = 0,0564
a =p = 0,0773 - 16
Horse 9:
8 8

α 8

8
7
EPISODE 2. Model

a = p - b = 0.1833 - 0.3337 = 0.1735

PREDICTING EVENT RESULTS SPORTS: HORSE RACING.......................................1
Index..........................................................................................................................................8
Table index..............................................................................................................................XI
Index of Figures....................................................................................................................XIV
1. Goals...............................................................................................................................1
2. memory structure............................................................................................................2
CHAPTER 1..............................................................................................................................5
1.1. History of betting........................................................................................................5
1.2. Concepts used.............................................................................................................6
1.3. Racing categories........................................................................................................9
Types of races depending on the terrain[8]......................................................................11
1.4. Types of bets.................................................................................................................12
1.5. Bettor profile.............................................................................................................16
Betting by age......................................................................................................................17
Position to choose when we bet [17]................................................................................17
1.6. Betting shops.............................................................................................................17
1.6.1. ZEturf................................................................................................................18
1.6.2. Betfair................................................................................................................20
1.7. Related Posts.............................................................................................................20
EPISODE 2..............................................................................................................................24
Model.......................................................................................................................................24
2.1. Chosen model...........................................................................................................24
2.1.1. Introduction.......................................................................................................25
2.1.2. Support Vector Machine (SVM) for Classification...........................................28
2.1.3.Forecasting race results...........................................................................................31
2.1.4. Empirical evaluation of the SVM/CL model.....................................................42
2.1.5. Conclusion.........................................................................................................48
2.2. Applied model...............................................................................................................48
2.2.1. Formstar website...............................................................................................49
2.2.2. Database.................................................................................................................58
2.2.3. Inference............................................................................................................69
2.2.4. Odds and Q Factor.............................................................................................76

8
8
 EPISODE 2. Model

2.2.5. Kelly criterion....................................................................................................82

CHAPTER 3............................................................................................................................91
Results.....................................................................................................................................91
3.1. Choice of value of β..................................................................................................92
3.1.1. Probability of winning for a horse as a function of the value of β....................92
For greater ease, let's show it graphically:....................................................................93
Horse 1..........................................................................................................................93
Horse 3..........................................................................................................................94
Horse 4..........................................................................................................................95
Horse 5..........................................................................................................................95
Horse 6..........................................................................................................................96
Horse 7..........................................................................................................................96
• Knight 11....................................................................................................................98
• Knight 13....................................................................................................................99
• Knight 15..................................................................................................................100
horse 2........................................................................................................................102
Horse 4........................................................................................................................103
Horse 5........................................................................................................................104
Horse 6........................................................................................................................104
Horse 7........................................................................................................................105
Horse 8........................................................................................................................105
Horse 9........................................................................................................................106
Horse 10......................................................................................................................106
Horse 12......................................................................................................................107
• Knight 14..................................................................................................................108
3.1.2. Variation of gain as a function of β.................................................................110
3.1.3. Choosing the optimal value of β......................................................................114
CHAPTER 4..........................................................................................................................147
Conclusions and future lines of work....................................................................................147
4.1. Conclusions.............................................................................................................148
4.2. Future lines of work................................................................................................149
CHAPTER 5..........................................................................................................................151
Budget....................................................................................................................................151

8
9
EPISODE 2. Model

5.1. Introduction.............................................................................................................151
5.2. Personnel cost.........................................................................................................152
5.3. Software cost...........................................................................................................153
5.4. Hardware cost.........................................................................................................153
5.5. Expendable material cost........................................................................................154
5.6. Cost summary.........................................................................................................154
References.............................................................................................................................157
Websites.............................................................................................................................157
Articles...............................................................................................................................159

In this way our vector a(s) would be:

a ( s ) = [ 0 0 0.0276 0.0087 0 0.0627 0.0646 0.0564 0.1735 0.2727 ]

The winner of this race was horse number one. We did not place any bet on that horse,
therefore, in this race we would have lost what we bet, which is €0.6663.

9
0
 CHAPTER 3. Results

CHAPTER 3

Results

In this chapter we are going to show the results that we have obtained by developing
the model described in the previous chapter.

The data on which we have based have been races carried out between February and
June 2011. Additionally, as we mentioned before, all the data has been taken from the
Formstar page that has career information in South Africa.

The study was based on E Division 1400 meter distance races. We have analyzed the
method making a distinction by distance so that the results obtained were more accurate.

To calculate profits we have a bankroll of €1 per race, regardless of the number of

races we have.

In addition, we are going to do the analysis with two types of cross-validation:

• Cross-validation of K iterations.

• Leave-One-Out Cross Validation.

91
CHAPTER 3. Results

3.1. Choice of value of β

As we have seen before, the SVM function will vary depending on the value that β
takes.

Therefore, in this section we will try to choose the optimal value of β .

3.1.1. Probability of winning for a horse as a function of the value of β

We are going to show how the probability that each horse has of winning, in a race,
varies depending on the choice made of β . To do this we are going to show a race that
contains 15 horses. We are going to do the study from the point of view of CV and then with
LOO.

3.1.1.1. CV Study

First of all, we will study how β influences if we are using cross validation of K
iterations.

Let's see how the probability of winning for each horse varies as we vary the value of β
. Once we have all the probabilities, and knowing who the winner of the race was, we can
decide what is the value of β that would have led us to place a greater bet on that horse.

92
 CHAPTER 3. Results

Beta=0 Beta=0.05 Beta=0.1 Beta=0.2 Beta=0.4 Beta=0.6 Beta=0.7 Beta=0.8 Beta=0.9 Beta=1
Horse 1 0,0667 0,0655 0,0641 0,0609 0,0529 0,0436 0,0388 0,0341 0,0296 0,0254
horse 2 0,0667 0,0633 0,0599 0,0532 0,0403 0,0290 0,0241 0,0198 0,0161 0,0129
Horse 3 0,0667 0,0634 0,0601 0,0536 0,0410 0,0297 0,0248 0,0205 0,0167 0,0134
Horse 4 0,0667 0,0689 0,0711 0,0748 0,0798 0,0809 0,0798 0,0777 0,0747 0,0710
Horse 5 0,0667 0,0639 0,0611 0,0553 0,0436 0,0326 0,0277 0,0232 0,0192 0,0156
Horse 6 0,0667 0,0616 0,0567 0,0477 0,0324 0,0209 0,0165 0,0128 0,0098 0,0075
Horse 7 0,0667 0,0614 0,0564 0,0471 0,0317 0,0202 0,0158 0,0122 0,0093 0,0070
Horse 8 0,0667 0,0666 0,0663 0,0651 0,0604 0,0532 0,0490 0,0445 0,0399 0,0354
Horse 9 0,0667 0,0699 0,0731 0,0792 0,0896 0,0960 0,0975 0,0977 0,0967 0,0945
Horse 10 0,0667 0,0653 0,0638 0,0603 0,0519 0,0423 0,0375 0,0328 0,0283 0,0241
Horse 11 0,0667 0,0751 0,0844 0,1055 0,1589 0,2270 0,2660 0,3076 0,3513 0,3963
Horse 12 0,0667 0,0688 0,0709 0,0744 0,0790 0,0796 0,0784 0,0761 0,0730 0,0692
Horse 13 0,0667 0,0711 0,0755 0,0845 0,1018 0,1165 0,1221 0,1264 0,1291 0,1303
Horse 14 0,0667 0,0668 0,0668 0,0661 0,0624 0,0558 0,0518 0,0474 0,0429 0,0383
Horse 15 0,0667 0,0683 0,0698 0,0721 0,0742 0,0725 0,0702 0,0672 0,0634 0,0591

Table 3.1. Example of the winning probabilities of each horse for each β with cross validation

For greater ease, let's show it graphically:

Horse 1

Figure 3.1. Variation of the probability of winning for horse 1 as a function of β with
cross validation

93
CHAPTER 3. Results

horse 2

Figure 3.2. Variation of the probability of winning for horse 2 as a function of β

with cross validation

Horse 3
Probability

Figure 3.3. Variation of the probability of winning for horse 3 as a function of β

with cross validation

94
 CHAPTER 3. Results

Horse 4

Figure 3.4. Variation in the probability of winning for horse 4 as a function of

cross-validated β

Horse 5

Figure 3.5. Variation of the probability of winning for horse 5 depending on

of β with cross validation

95
CHAPTER 3. Results

Horse 6

Figure 3.6. Variation of the probability of winning for horse 6 depending on

of β with cross validation

Horse 7
Probabi

Figure 3.7. Variation of the probability of winning for horse 7 depending on

of β with cross validation

96
 CHAPTER 3. Results

Horse 8

Figure 3.8. Variation in the probability of winning for horse 8 depending

on
of β with cross validation
Probability

Figure 3.9. Variation of the probability of winning for horse 9 as a function of β

with cross validation

97
CHAPTER 3. Results

Horse 10

Figure 3.10. Variation of the probability of winning for horse 10 depending on

of β with cross validation

• Knight 11

Figure 3.11. Variation in the probability of winning for horse 11 depending on

of β with cross validation

98
 CHAPTER 3. Results

Horse 12

Figure 3.12. Variation of the probability of winning for horse 12 depending on

of β with cross validation

• Knight 13

Figure 3.13. Variation in the probability of winning for horse 13 depending on

of β with cross validation

99
CHAPTER 3. Results

• Knight 14

Figure 3.14. Variation of the probability of winning for horse 14 depending on

of β with cross validation

• Knight 15

P
r
o
b
a

Figure 3.15. Variation of the probability of winning for horse 15 depending on

of β with cross validation

10
0
 CHAPTER 3. Results

As we see, the previous graphs do not follow a pattern. We have the example of
horse 1, horse 2, horse 3, horse 5, horse 6, horse 7, horse 8, horse 10 and horse 14 that
have their maximum with β = 0 and from there they go down until they reach the
minimum at β = 1 . Then horses 4, 12 and 15 start at a high value with β = 0 , but,
instead of going down, they go up, having their maximum at β = 0.6 for the first two and
β = 0.4 for the last, for, then go down to its minimum at β = 1 . And finally there are
horses 9, 11 and 13 which at β = 0 have their smallest value and ascend until they have
the maximum around β = 1 . The way of ascending and descending varies for each
horse, just as horse 1 descends almost in a straight line, horse 2 does so exponentially.
Likewise, horse 13 ascends logarithmically and horse 14 almost follows a straight line.

Taking into account that the winning horse was number 3, we see that the highest
probability for this horse is obtained with β = 0 .

Therefore, in this case the optimal β would be β = 0 .

3.1.1.2. LOO Studio

In this case, we are going to use the Leave-One-Out technique. We will see how
the probability varies depending on the value we choose for β .

Beta=0 Beta=0.05 Beta=0.1 Beta=0.2 Beta=0.4 Beta=0.6 Beta=0.7 Beta=0.8 Beta=0.9 Beta=1
Horse 1 0,066 0,065 0,063 0,059 0,051 0,041 0,036 0,032 0,028 0,024
horse 2 7 0,066 2 0,062 6 0,058 9 0,050 0 0,035 5 0,024 9 0,019 5 0,015 4 0,012 7 0,010
Horse 3 7 0,066 4 0,062 2 0,057 1 0,049 7 0,034 3 0,023 8 0,018 9 0,015 7 0,011 1 0,009
Horse 4 7 0,066 2 0,067 8 0,068 4 0,069 7 0,068 3 0,064 8 0,061 0 0,058 9 0,055 4 0,051
Horse 5 7 0,066 7 0,063 5 0,059 3 0,053 5 0,040 6 0,028 8 0,024 6 0,020 1 0,016 5 0,013
Horse 6 7 0,066 3 0,060 9 0,054 0 0,043 1 0,027 9 0,016 2 0,012 1 0,009 5 0,007 5 0,005
Horse 7 7 0,066 4 0,061 5 0,056 9 0,046 5 0,030 4 0,019 5 0,015 4 0,012 1 0,009 3 0,007
Horse 8 7 0,066 3 0,066 1 0,066 6 0,065 9 0,061 6 0,054 4 0,050 0 0,046 2 0,042 1 0,038
Horse 9 7 0,066 7 0,069 5 0,072 5 0,077 0 0,086 3 0,090 5 0,092 5 0,092 6 0,092 6 0,091
Horse 10 7 0,066 6 0,066 5 0,066 7 0,065 0 0,060 9 0,054 1 0,050 5 0,046 1 0,042 1 0,038
Horse 11 7 0,066 7 0,072 5 0,079 4 0,092 8 0,122 1 0,154 2 0,170 2 0,186 2 0,203 3 0,219
Horse 12 7 0,066 8 0,070 1 0,073 6 0,080 2 0,091 0 0,099 3 0,101 7 0,103 0 0,104 2 0,105
Horse 13 7 0,066 1 0,072 5 0,078 0 0,091 1 0,119 1 0,148 8 0,162 7 0,177 8 0,191 2 0,205
Horse 14 7 0,066 5 0,069 6 0,071 4 0,076 0 0,083 0 0,086 6 0,086 1 0,086 3 0,085 2 0,083
Horse 15 7 0,066 3 0,069 9 0,072 4 0,078 1 0,088 4 0,094 7 0,096 3 0,097 3 0,097 6 0,097
7 9 9 7 3 5 4 4 7 2
Table 3.2. Example of the probabilities of winning for each horse for each β with Leave-One-Out

The representation of the probability of winning for each horse as a function of β

would look like this:

10
1
CHAPTER 3. Results

Horse 1

Figure 3.16. Variation of the probability of winning for horse 1 depending on

of β with Leave-One-Out

horse 2

Figure 3.17. Variation of the probability of winning for horse 2 depending on

of β with Leave-One-Out

10
2
 CHAPTER 3. Results

Horse 3

Figure 3.18. Variation of the probability of winning for horse 3 depending on

of β with Leave-One-Out

Horse 4

Figure 3.19. Variation of the probability of winning for horse 4 depending on

of β with Leave-One-Out

10
3
CHAPTER 3. Results

Horse 5

Figure 3.20. Variation of the probability of winning for horse 5 depending on

of β with Leave-One-Out

Horse 6

Figure 3.21. Variation in the probability of winning for horse 6 depending on

of β with Leave-One-Out

10
4
 CHAPTER 3. Results

Horse 7

Figure 3.22. Variation of the probability of winning for horse 7 depending on

of β with Leave-One-Out

Horse 8

Figure 3.23. Variation in the probability of winning for horse 8 depending on

of β with Leave-One-Out

10
5
CHAPTER 3. Results

Horse 9

Figure 3.24. Variation of the probability of winning for horse 9 depending on

of β with Leave-One-Out

Horse 10

Figure 3.25. Variation of the probability of winning for horse 10 depending on

of β with Leave-One-Out

10
6
 CHAPTER 3. Results

• Knight 11

Figure 3.26. Variation in the probability of winning for horse 11 depending on

of β with Leave-One-Out

Horse 12

Figure 3.27. Variation of the probability of winning for horse 12 depending on

of β with Leave-One-Out

10
7
CHAPTER 3. Results

• Knight 13

Figure 3.28. Variation in the probability of winning for horse 13 depending on

of β with Leave-One-Out

• Knight 14

Figure 3.29. Variation of the probability of winning for horse 14 depending on

of β with Leave-One-Out

10
8
 CHAPTER 3. Results

• Knight 15

Figure 3.30. Variation of the probability of winning of horse 15 as a function of β with Leave-One-Out

In this case we can also affirm that each horse does not vary its profit in the same
way, as a function of β . In this case, horse 1, horse 2, horse 3, horse 5, horse 6, horse 7,
horse 8 and horse 10 have their maximum with β = 0 and from there they go down until
they reach the minimum at β = 1 . Horses 4 and 14 start at a high value with β = 0 , but
instead of going down, they go up, having their maximum at β = 0.2 and β = 0.7
respectively, and then go to their minimum at β = 1 . And finally, horses 9, 11, 12, 13
and 15, which at β = 0 have their smallest value and ascend until they have the
maximum around β = 1 . The way of ascending and descending varies for each horse,
just as horse 1 descends almost in a straight line, horse 2 does so exponentially, and
horse 8 does so logarithmically. Likewise, horse 9 ascends logarithmically and horse 11
almost follows a straight line.

Taking into account that the winning horse was number 3, we see that the highest
probability for this horse is obtained with β = 0 .

Therefore, in this example the optimal β would be β = 0 .

Furthermore, if we compare it with the result using CV, we see that it is the same,

10
9
CHAPTER 3. Results

so, in this example, we cannot decide between which of the two methods leads us to a
better result.

3.1.2. Variation of gain as a function of β

In this section we will see how the profit varies, for each race, as a function of β ,
in order to decide which value of β is the best, in order to obtain greater benefits.

To do this, we are going to carry out the study with each of the methods that we
have used to carry out the training of our program.

3.1.2.1. CV Study

The profit for each race, as a function of β, will be:

11
0
 CHAPTER 3. Results

Career Beta=0 Beta=0.05 Beta=0.1 Beta=0.2 Beta=0.4 Beta=0.6 Beta=0.7 Beta=0.8 Beta=0.9 Beta=1
1 -0,9414 -0,8974 -0,8533 -0,7651 -0,5888 -0,8701 -0,6973 -0,5246 -0,9636 -0,9101
2 -0,9368 -0,9126 -0,8884 -0,8401 -0,7433 -0,6465 -0,5981 -0,8961 -0,9118 -0,8017
3 -0,9418 -0,9173 -0,8929 -0,844 -0,7463 -0,6486 -0,9534 -0,8096 -0,8525 -0,7503
4 -0,9427 -0,8995 -0,8563 -0,7699 -0,5971 -0,8532 -0,6723 -0,8238 2,7785 0,4709
5 -0,942 -0,8971 -0,8523 -0,7626 -0,5833 -0,8338 -0,6525 -0,721 -0,8853 -0,3623
6 -0,9486 -0,9043 -0,8599 -0,7713 -0,5939 -0,8534 -0,6698 -0,7824 -0,8701 -0,191
7 -0,9442 -0,9035 -0,8628 -0,7814 -0,6187 -0,8581 -0,6777 5,1565 3,4452 1,734
8 -0,9204 -0,8844 -0,8484 -0,7765 -0,6325 -0,916 -0,7571 -0,8318 -0,9354 -0,4858
9 46,3937 44,290 42,1878 37,98 29,5701 21,1584 16,9524 12,7466 8,5407 4,3348
10 -0,9369 8 -0,9064 -0,8759 2 -0,8149 -0,6929 -0,5708 -0,9123 -0,7601 -0,9851 -0,6412
11 -0,9527 -0,9251 -0,8974 -0,8422 -0,7317 -0,6212 -0,566 -0,9109 -0,7701 0,5575
12 -0,9322 -0,8938 -0,8554 -0,7785 -0,6248 -0,9388 -0,7799 -0,9163 -0,9951 -0,8808
13 -0,9373 -0,8944 -0,8514 -0,7655 -0,5938 -0,422 -0,9184 -0,7778 -0,9537 -0,9371
14 -0,925 -0,9057 -0,8863 -0,8476 -0,7702 -0,6928 -0,6541 -0,9026 -0,9415 -0,7361
15 -0,9178 -0,8832 -0,8486 -0,7793 -0,6408 -0,9698 -0,8159 -0,9112 -0,9368 -0,3907
16 -0,9353 -0,9037 -0,872 -0,8086 -0,6818 -0,9824 -0,8297 -0,7696 -0,9551 -0,2129
17 -0,9427 -0,9151 -0,8876 -0,8326 -0,7224 -0,6123 -0,913 -0,766 -0,8744 -0,7335
18 -0,9573 -0,9128 -0,8682 -0,7792 -0,6012 -0,9996 -0,8307 -0,6618 -0,9384 -0,0838
19 -0,9276 -0,8861 -0,8446 -0,7616 -0,5955 -0,9338 -0,7651 -0,8592 -0,976 -0,156
20 -0,9484 -0,9041 -0,8598 -0,7712 -0,594 -0,8581 -0,8138 -0,9356 -0,903 -0,8743
21 -0,937 -0,906 -0,8749 -0,8128 -0,6886 -0,5644 -0,8508 -0,9373 -0,9791 -0,8289
22 -0,9293 -0,887 -0,8446 -0,76 -0,5906 -0,8092 -0,9539 -0,6712 -0,9962 -0,5194
23 -0,9367 -0,9012 -0,8657 -0,7947 -0,6528 -0,5109 -0,8635 -0,9721 -0,9039 -0,9456
24 -0,9286 -0,884 -0,8393 -0,7501 -0,5717 -0,8204 -0,7626 -0,8613 -0,9601 -0,7933
25 -0,9335 -0,893 -0,8524 -0,7712 -0,6089 -0,8984 -0,7306 -0,8243 -0,9003 -0,9084
26 -0,9393 -0,9211 -0,903 -0,8666 -0,7939 -0,7213 -0,6849 -0,8866 -0,936 0,3087
27 -0,9462 -0,91 -0,8738 -0,8014 -0,6565 -0,9349 -0,7689 -0,8827 -0,8588 -0,7438
28 -0,9607 -0,9159 -0,871 -0,7813 -0,6019 -0,9487 -0,7808 8,0238 4,2251 0,4264
29 -0,9342 -0,8916 -0,8491 -0,7639 -0,5936 -0,8283 -0,6441 -0,7479 -0,9785 -0,2925
30 -0,9356 -0,9047 -0,8738 -0,8119 -0,6883 -0,5646 -0,8881 -0,7423 -0,8058 -0,9117
31 -0,9409 -0,9001 -0,8592 -0,7774 -0,6138 -0,9869 -0,8314 -0,9232 -0,8842 -0,9834
32 -0,9539 -0,9096 -0,8652 -0,7765 -0,5992 -0,9585 -0,7931 -0,6278 -0,931 -0,7913
33 -0,9146 -0,8815 -0,8484 -0,7821 -0,6496 -0,9575 -0,8059 -0,9222 -0,8846 -0,6743
34 -0,9409 -0,9084 -0,8759 -0,811 -0,6811 -0,9847 -0,8298 -0,9199 -0,9631 -0,642
35 -0,9446 -0,9088 -0,873 -0,8014 -0,6582 -0,9497 -0,779 -0,8616 -0,7729 -0,5815
36 -0,9557 -0,9127 -0,8697 -0,7837 -0,6117 -0,8926 -0,7139 -0,8895 -0,9814 0,1755
37 -0,9099 -0,8774 -0,8448 -0,7798 -0,6497 -0,9755 -0,8302 -0,6848 13,332 9,7608
38 -0,9515 -0,9081 -0,8647 -0,7779 -0,6044 -0,6019 -0,903 -0,7553 -0,8722 -0,9372
39 -0,9354 -0,9058 -0,8762 -0,817 -0,6987 -0,5803 -0,8489 -0,9548 -0,9258 -0,2587
40 -0,9562 -0,9189 -0,8816 -0,8071 -0,6579 -0,5087 -0,8923 -0,741 -0,9857 -0,8638

Table 3.3. Example of the profit of each race as a function of â with cross validation

As we see, except for a few values (in bold), the rest are all losses.

11
1
CHAPTER 3. Results

We can make a summary table of the above, with the total profit for all the races
we have bet on, and its variation with β :

Beta 0 0,05 0,1 0,2 0,4 0,6 0,7 0,8 0,9

1
Revenu 9,7779 9,0985 8,42 7,0621 4,346 -9,9203 -13,6804 -4,4393 0,154 -3,0548
Table 3.4. Total profit for each â with cross validation

We made a graphic representation to see it more clearly.

Figure 3.31. Variation of total profit as a function of â with cross validation

As we can see, the highest benefit would be achieved with β = 0 . With this value
we would obtain a profit of €9.7779. If we take into account that we bet €1 per race, the
total profit will be €49.7779.

3.1.2.2. LOO Studio

We are going to use the same procedure for the case of Leave-One-Out .

First of all we will show the table with the profit we obtain for each race.

11
2
 CHAPTER 3. Results

Caree Beta=0 Beta=0. Beta=0 Beta=0 Beta=0 Beta=0 Beta=0 Beta=0Beta=0 Beta=
1 -0,9414 -0,8992 - - - - - - - -
2 -0,9368 -0,9116 0,8569
- 0,7724
- 0,6034
- 0,9383
- 0,7769
-0,583 0,8862
- 0,9362
- 0,928
-
3 -0,9418 -0,9099 0,8863
- 0,8357
- 0,7346
- 0,6335
- - 0,8985
- 0,9181
- 0,813
-
4 -0,9427 -0,9007 0,878
- 0,8143
- 0,6868
- 0,5593
- 0,8824
- 0,7283
5,7339 0,7776
3,5 0,631
1,28
5 - -0,9013 0,8587
- 0,7747
- 0,6066
- 0,8625
- 0,6831
- - 073
- 05
-
6 0,942
-0,9486 -0,9053 0,8606
- 0,7792
- 0,6163
- 0,9036
- 0,7339
- 0,8282
- 0,9811
- 0,564
-
7 -0,9442 -0,9082 0,8621
- 0,7756
- 0,6026
- 0,8644
- 0,6827
- 0,8259
4,9947 0,9442
3,2 0,063
1,53
8 -0,9204 0,8722
-0,8827 - 0,8001
- 0,656
- 0,9158
- 0,7451
- - 633
- 19
-
9 46,3937 44,398 0,8451
42,403 0,7697
38,414 0,6189
30,434 0,9457
22,455 0,7918
18,465 0,8721
14,475 0,9682
10,485 0,099
6,49
10 -0,9369 8 -0,9001 - 9 - 2 -7 - 2 - 4 - 7 - 9-
62
11 -0,9527 -0,9316 0,8632
- 0,7895
- 0,642
- 0,9631
- 0,7939
-0,985 0,9054
- 0,8022
- 0,760
0,09
12 -0,9322 -0,8955 0,9106
- 0,8686
- 0,7845
- 0,7003
- - 0,8381
- 0,915
- 63-
13 -0,9373 -0,8956 0,8588
- 0,7854
- 0,6385
- 0,9397
- 0,7808
- 0,915
- 0,9394
- 0,858
-
14 - -0,8986 0,8538
- 0,7704
- 0,6035
- 0,4366
- 0,8733
- 0,9631
- 0,9099
- 0,854
-0,65
15 0,925
-0,9178 -0,8794 0,8723
- 0,8196
- 0,7141
- 0,6087
- 0,9384
-0,739 - 0,7971 0,8911
- -
16 -0,9353 -0,9014 0,841
- 0,7642
- 0,6105
- 0,9039
- - 0,8566
- 0,8944
- 0,498
0,03
17 -0,9427 -0,9148 0,8676
- 0,7998
- 0,6642
- 0,9563
- 0,7993
- 0,9781
- 0,937
- 44
-
18 -0,9573 -0,9129 0,8869
- 0,8312
- 0,7198
- 0,6083
-0,984 0,9359
- 0,7921
- 0,9136
- 0,749
-
19 -0,9276 -0,8897 0,8685
- 0,7797
- 0,6021
- - 0,8125
- 0,9654
- 0,9119
- 0,189
-
20 -0,9484 -0,9028 0,8518
- 0,7759
- 0,6242
- 0,9896
- 0,8302
- 0,9697
- 0,957
- 0,431
-
21 - - 0,8573
- 0,7662
- 0,854
- 0,8687
- 0,6848
- 0,8774
- 0,8358
- 0,954
-
22 0,937
-0,9283 0,905 0,8731
-0,8871 - 0,8091
- 0,6812
- 0,9869
- 0,8335
- 0,9174
- 0,847
- 0,720
-
23 -0,9367 -0,9067 0,8449
- 0,7604
- 0,5916
- 0,8146
- 0,9717
- 0,6915
- 0,9137
- 0,473
-
24 -0,9286 -0,8832 0,8768
- 0,817
- 0,6973
- 0,5776
- 0,9227
-0,789 0,7801- 0,9331
- 0,960
-
25 -0,9335 -0,8926 0,8378
- 0,7471
- 0,5655
- 0,9519
- - 0,878
- 0,7694
- 0,751
-
26 -0,9393 -0,9259 0,8517
- 0,7698
- 0,606
- 0,8912
- 0,7222
- 0,809
- 0,9114
- 0,877
0,13
27 -0,9462 -0,9059 0,9125
- 0,8857
- 0,8321
- 0,7785
- 0,7517
- 0,9644
- 0,8441
- 13-
28 -0,9607 - 0,8657
- 0,7851
- 0,6241
- 0,8898
- 0,7163
- 0,8232 0,8212
8,188 4,4 0,630,631
29 -0,8919 0,8734
-0,9342 0,917 - 0,786
- 0,6113
- 0,4367
- 0,8579
- - 098
- 17
-
30 -0,9356 - 0,8495
- 0,7649
- 0,5956
- 0,8396
- 0,6573
-0,884 0,7494
- 0,7613
- 0,403
-
31 -0,8998 0,8784
-0,9409 0,907 - 0,8212
- 0,7068
- 0,5924
-0,951 - 0,7376
- 0,8083
- 0,876
-
32 -0,9539 -0,9089 0,8587
- 0,7764
- 0,6119
- - 0,7895
- 0,9206
- 0,8645
- 0,983
-
33 -0,9146 -0,8845 0,8639
- 0,7738
- 0,5937
- 0,9548
- 0,7889
- 0,9539
- 0,9627
- 0,778
-
34 -0,9409 -0,9051 0,8544
- 0,7942
- 0,6738
- 0,9468
- 0,8775
- 0,9546
- 0,9401
- 0,313
-
35 -0,9446 - 0,8692
- 0,7976
- 0,6544
- 0,9596
- 0,8005
- 0,8876
- 0,9495
- 0,611
-
36 -0,9137 0,8873
-0,9557 0,916 - 0,8301
- 0,7156
- 0,6011
- 0,9222
- 0,771
- 0,8319
- 0,659
0,10
37 -0,9099 -0,8738 0,8716
- 0,7875
- 0,6194
- 0,8731
- 0,6911
- 0,8245
14,569 0,9596
10, 496,84
38 -0,9515 -0,9095 0,8377
- 0,7656
- 0,6212
- 0,9448
- 0,7943
- - 9 707
- 39-
39 -0,9354 - 0,8674
- 0,7833
- 0,6151
- 0,9508
- 0,7865
- 0,9597
- 0,9402
- 0,911
-
40 -0,9562 0,901- 0,8665
- 0,7977
- 0,6599
- 0,9459
- 0,7853
- 0,9159
- 0,8737
- 0,425
-
0,924
Table 0,8918 0,8273 0,6983 0,5694 0,9576 0,8156 0,9969
3.5. Example of the profit of each race based on â with Leave-One-Out 0,908

In this case, as in the previous one, for almost all races we have a loss of money.
The only ones we win are the ones shown in bold.

11
3
CHAPTER 3. Results

As a summary we will say that:

Beta 0 0,05 0,1 0,2 0,4 0,6 0,7 0,8 0,9 1
Revenue 9,7789 9,1989 8,6199 7,4622 4,8773 -9,5836 -12,8863 17,711 1,012 -3,1824
Table 3.6. Total profit for each â with Leave-One-Out

Graphically we can see it as:

Figure 3.32. Variation of total gain as a function of â with Leave-One-Out

In this case, the optimal value of β to maximize profit would be given with β = 0.8,
where we have a profit of €17,711. Therefore, if we take into account that we have bet
€1 per race, we will end our bets with a total of €57,711.

Therefore, we will have the best result with β = 0.8 and the LOO method.

3.1.3. Choosing the optimal value of β

As we have seen in the previous sections, depending on the gain we could say that
the best value of β is β = 0.8 , and, furthermore, with the LOO method we obtain
superior results than with CV.

However, to draw a final conclusion, we are going to focus more on the bets we
place through the Kelly Criterion.

To do this, we are going to choose the parameters c = 25 and sigma = 15 , which

11
4
 CHAPTER 3. Results

are the ones that we have seen give us better results.

First of all, we are going to show the tables where we collect:

• Total amount bet on each race, that is, the sum of everything we bet on each
horse in a race.

• The profit we get in each race.

Next, we will present the vectors obtained with the Kelly Criterion, named “a”, for
the parameter β . This way we will observe how the number of horses we bet on in each
race varies, depending on said parameter.

The way we are going to present it is, for each type of validation, we will first
show two tables, one of those mentioned above, each one with a different value of β .
And, next, we will put the vectors of the Kelly bets belonging to each table.

3.1.3.1. CV Study

11
5
CHAPTER 3. Results

Total
Guy bets Revenue
Guy Total bets Revenue

1 0,9414 0 1 0,8974 0

2 0,9368 0 2 0,9126 0

3 0,9418 0 3 0,9173 0
4 0,9427 0 4 0,8995 0

5 0,942 0 5 0,8971 0

6 0,9486 0 6 0,9043 0

7 0,9442 0 7 0,9035 0

8 0,9204 0 8 0,8844 0

9 0,9279 47,3216 9 0,8858 45,1766

I.C
0,9369 0 10 0,9064 0
.
11 0,9527 0 11 0,9251 0

12 0,9322 0 12 0,8938 0

13 0,9373 0 13 0,8944 0

14 0,925 0 14 0,9057 0

15 0,9178 0 15 0,8832 0

16 0,9353 0 16 0,9037 0

17 0,9427 0 17 0,9151 0

18 0,9573 0 18 0,9128 0

19 0,9276 0 19 0,8861 0

20 0,9484 0 20 0,9041 0

21 0,937 0 21 0,906 0

22 0,9293 0 22 0,887 0

23 0,9367 0 23 0,9012 0

24 0,9286 0 24 0,884 0

25 0,9335 0 25 0,893 0

26 0,9393 0 26 0,9211 0

27 0,9462 0 27 0,91 0

28 0,9607 0 28 0,9159 0

29 0,9342 0 29 0,8916 0

30 0,9356 0 30 0,9047 0

31 0,9409 0 31 0,9001 0

32 0,9539 0 32 0,9096 0

33 0,9146 0 33 0,8815 0

34 0,9409 0 34 0,9084 0

35 0,9446 0 35 0,9088 0

36 0,9557 0 36 0,9127 0

37 0,9099 0 37 0,8774 0

38 0,9515 0 38 0,9081 0

39 0,9354 0 39 0,9058 0

40 0,9562 0 40 0,9189 0

Total 37,54 47,3216 Total 36,0781 45,1766

37 β = 0,05

11
6
 CHAPTER 3. Results

Table 3.7. Cross-validation: total bets and winnings for each type, with â = 0 and â = 0.05

11
7
CHAPTER 3. Results

Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,9414 0 0 0
to
1 0 0 0 0 0 0 0 0 0 0,9368
to
2 3 0 0 0 0 0 0 0 0 0,941 0
to 4 0 0 0 0 0 0 0 0 8 0 0,9427 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,942
to 6 0 0 0 0 0 0 0 0 0 0 0 0,9486 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,9442
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,9204 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9279
to
10 0 0 0,936 0 0 0 0 0 0 0
at
11 0 09 0 0 0 0 0 0 0 0,9527
at 0 0 0 0 0 0 0 0 0 0 0 0 0,9322 0
at
1213 0 0 0 0 0 0 0 0 0 0 0 0 0,9373 0
at
14 0 0 0 0 0 0 0 0 0 0 0,925
at
15 0 0 0 0 0 0 0 0,917 0 0 0 0 0 0
at 0 0 0 0 0 0 0 8 0 0 0 0 0 0,9353
16
at 0 0 0 0 0 0 0 0 0 0 0,9427 0 0
17
at 0 0 0 0 0 0 0 0 0 0,9573
18
at 0 0 0 0 0 0 0 0 0 0,9276
19
to 0,9484 0 0 0 0 0 0 0 0 0
20
at 0 0 0 0 0 0 0 0 0 0 0 0 0,937 0
at
2122 0 0 0 0 0 0 0 0 0,929 0 0 0 0 0 0 0
at
23 0 0 0 0 0 0 0 0 3 0,936 0 0
at
24 0 0 0 0 0 0 0,928 0 7 0 0 0 0 0 0
at
25 0 0,933 0 0 0 0 6 0 0 0 0 0 0 0 0
at
26 05 0 0 0 0 0 0 0 0 0,9393
at
27 0 0 0 0 0 0 0 0,946 0 0 0 0 0 0
at 0 0 0 0 0 0 0 2 0 0 0,9607 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,9342 0
29
at 0 0 0 0 0 0 0 0,935 0 0
30
at 0 0 0 0 0 0 0 6 0 0 0,9409 0 0 0 0
31
at 0 0 0 0 0 0 0 0 0,953 0 0 0
32
at 0,914 9
33 0 0 0 0 0 0 0 0 0 0 0 0 0 0
at 0 0 0 0 0 0 6 0 0 0 0 0 0 0 0 0 0,9409
at
3435 0 0 0 0 0 0 0 0 0,944 0
at
36 0 0 0 0 0 0 0 0 6 0 0,9557
at
37 0 0 0 0 0 0 0 0 0,909 0 0 0 0 0
at
38 0 0 0 0 0 0 0 0 9 0 0,9515 0 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,9354 0 0 0
at 0 0 0 0 0 0 0 0 0,956
40
2

Table 3.8. Cross-validation: bets we make for each horse, depending on the type, with β = 0

11
8
 CHAPTER 3. Results

• β = 0,05
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,8974 0 0 0
to 0 0 0 0 0 0 0 0 0 0,9126
to
2 0 0 0 0 0 0 0 0 0,917 0
to
3 0 0 0 0 0 0 0 0 3 0 0,8995 0 0 0
to
4 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8971
to
5 0 0 0 0 0 0 0 0 0 0 0 0,9043 0 0
6 7
at 0 0 0 0 0 0 0 0 0 0 0 0 0,9035
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,8844 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8858
to 0 0 0 0 0 0 0 0 0 0,9064
10
at 0 0 0 0 0 0 0 0 0 0,9251
11
at 0 0 0 0 0 0 0 0 0 0 0 0 0,8938 0
12
at 0 0 0 0 0 0 0 0 0 0 0 0 0,8944 0
13
at 0 0 0 0 0 0 0 0 0 0 0,9057
14
at
15 0 0 0 0 0 0 0 0,883 0 0 0 0 0 0
at
16 0 0 0 0 0 0 0 2 0 0 0 0 0 0,9037
at
17 0 0 0 0 0 0 0 0 0 0 0,9151 0 0
at
18 0 0 0 0 0 0 0 0 0 0,9128
at
19 0 0 0 0 0 0 0 0 0 0,8861
to 0,9041 0 0 0 0 0 0 0 0 0
at
2021 0 0 0 0 0 0 0 0 0 0 0 0 0,906 0
at 0 0 0 0 0 0 0 0 0,887 0 0 0 0 0 0 0
22
at 0 0 0 0 0 0 0 0 0,901 0 0
23
at 0 0 0 0 0 0 0,884 0 2 0 0 0 0 0 0
24
at 0 0,893 0 0 0 0 0 0 0 0 0 0 0 0
25
at 0 0 0 0 0 0 0 0 0 0,9211
26
at
27 0 0 0 0 0 0 0 0,91 0 0 0 0 0 0
at
28 0 0 0 0 0 0 0 0 0 0,9159 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,8916 0
at
2930 0 0 0 0 0 0 0 0,904 0 0
at
31 0 0 0 0 0 0 0 7 0 0 0,9001 0 0 0 0
at
32 0 0 0 0 0 0 0 0 0,909 0 0 0
at 0 0 0 0 0 0 0 0 6 0 0 0 0 0 0 0,8815
at
33 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9084
34
at 0 0 0 0 0 0 0 0 0,908 0
35
at 0 0 0 0 0 0 0 0 8 0 0,9127
36
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8774
37
at 0 0 0 0 0 0 0 0 0 0,9081 0 0 0 0
38
at
39 0 0 0 0 0 0 0 0 0 0 0,9058 0 0 0
at
40 0 0 0 0 0 0 0 0 0,918
9

Table 3.9. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.05

11
9
 CHAPTER 3. Results

β = 0,1 β = 0,2
Guy Total bets Revenue
Guy Total bets Revenue
1 0,8533 0
1 0,7651 0
2 0,8884 0
2 0,8401 0
3 0,8929 0
3 0,844 0
4 0,8563 0
4 0,7699 0
5 0,8523 0
5 0,7626 0
6 0,8599 0
6 0,7713 0
7 0,8628 0
7 0,7814 0
8 0,8484 0
8 0,7765 0
9 0,8438 43,0316 0,75 9
9 6 38,7416
10 0,8759 0
10 0,8149 0
11 0,8974 0
11 0,8422 0
12 0,8554 0
12 0,7785 0
13 0,8514 0
13 0,7655 0
14 0,8863 0
14 0,8476 0
15 0,8486 0
15 0,7793 0
15 0,872 0
16 0,8086 0
17 0,8876 0
17 0,8326 0
18 0,8682 0
18 0,7792 0
19 0,8446 0
19 0,7616 0
0,85 9
20 8 0
20 0,7712 0
21 0,8749 0
21 0,8128 0
22 0,8446 0 22 0,76 0
23 0,8657 0 23 0,7947 0
24 0,8393 0 24 0,7501 0
25 0,8524 0 25 0,7712 0
25 0,903 0 26 0,8666 0
27 0,8738 0 27 0,8014 0
28 0,871 0 28 0,7813 0
29 0,8491 0 29 0,7639 0
30 0,8738 0 30 0,8119 0
0,85 9
31 2 0 31 0,7774 0
32 0,8652 0 32 0,7765 0
33 0,8484 0 33 0,7821 0
34 0,8759 0 34 0,811 0
35 0,873 0 35 0,8014 0
36 0,8697 0 36 0,7837 0
37 0,8448 0 37 0,7798 0
38 0,8647 0 38 0,7779 0
39 0,8762 0 39 0,817 0
40 0,8816 0 40 0,8071 0
Total 34,6116 43,0316 Total 31,6795 38,7416

Table 3.10. Cross-validation: total bets and profits for each type, with β = 0.1 and β = 0.2

12
0
CHAPTER 3. Results

• β = 0,1
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,8533 0 0 0
to
1 0 0 0 0 0 0 0 0 0 0,8884
to
2 3 0 0 0 0 0 0 0 0 0,892 0
to 4 0 0 0 0 0 0 0 0 9 0 0,8563 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8523
to 6 0 0 0 0 0 0 0 0 0 0 0 0,8599 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,8628
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,8484 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8438
to
10 0 0 0 0 0 0 0 0 0 0,8759
at
11 0 0 0 0 0 0 0 0 0 0,8974
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,8554 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,8514 0
at
14 0 0 0 0 0 0 0 0 0 0 0,8863
at
15 0 0 0 0 0 0 0 0,848 0 0 0 0 0 0
at 0 0 0 0 0 0 0 6 0 0 0 0 0 0,872
16
at 0 0 0 0 0 0 0 0 0 0 0,8876 0 0
17
at 0 0 0 0 0 0 0 0 0 0,8682
18
at 0 0 0 0 0 0 0 0 0 0,8446
19
to 0,8598 0 0 0 0 0 0 0 0 0
20
at 0 0 0 0 0 0 0 0 0 0 0 0 0,8749 0
at
2122 0 0 0 0 0 0 0 0 0,844 0 0 0 0 0 0 0
at
23 0 0 0 0 0 0 0 0 6 0,865 0 0
at
24 0 0 0 0 0 0 0,839 0 7 0 0 0 0 0 0
at
25 0 0,852 0 0 0 0 3 0 0 0 0 0 0 0 0
at
26 04 0 0 0 0 0 0 0 0 0,903
at
27 0 0 0 0 0 0 0 0,873 0 0 0 0 0 0
at 0 0 0 0 0 0 0 8 0 0 0,871 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,8491 0
29
at 0 0 0 0 0 0 0 0,873 0 0
30
at 0 0 0 0 0 0 0 8 0 0 0,8592 0 0 0 0
31
at 0 0 0 0 0 0 0 0 0,865 0 0 0
32
at 2
33 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8484
at
34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8759
at
35 0 0 0 0 0 0 0 0 0,873 0
at
36 0 0 0 0 0 0 0 0 0 0,8697
at
37 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8448
at
38 0 0 0 0 0 0 0 0 0 0,8647 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,8762 0 0 0
at
39 0 0 0 0 0 0 0 0 0,881
40
6

Table 3.11. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.1

12
1
CHAPTER 3. Results

• β = 0,2
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,7651 0 0 0
to
1 0 0 0 0 0 0 0 0 0 0,8401
to
2 3 0 0 0 0 0 0 0 0 0,844 0
to 4 0 0 0 0 0 0 0 0 0 0,7699 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7626
to 6 0 0 0 0 0 0 0 0 0 0 0 0,7713 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,7814
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,7765 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7596
to
10 0 0 0 0 0 0 0 0 0 0,8149
at 0 0 0 0 0 0 0 0 0 0,8422
at
1112 0 0 0 0 0 0 0 0 0 0 0 0 0,7785 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,7655 0
at
14 0 0 0 0 0 0 0 0 0 0 0,8476
at
15 0 0 0 0 0 0 0 0,779 0 0 0 0 0 0
at 0 0 0 0 0 0 0 3 0 0 0 0 0 0,8086
16
at 0 0 0 0 0 0 0 0 0 0 0,8326 0 0
17
at 0 0 0 0 0 0 0 0 0 0,7792
18
at 0 0 0 0 0 0 0 0 0 0,7616
19
to 0,7712 0 0 0 0 0 0 0 0 0
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,8128 0
at
22 0 0 0 0 0 0 0 0 0,76 0 0 0 0 0 0 0
at
23 0 0 0 0 0 0 0 0 0,794 0 0
at 0 0 0 0 0 0 0,750 0 7 0 0 0 0 0 0
at
2425 0 0,771 0 0 0 0 1 0 0 0 0 0 0 0 0
at
26 02 0 0 0 0 0 0 0 0 0,8666
at
27 0 0 0 0 0 0 0 0,801 0 0 0 0 0 0
at 0 0 0 0 0 0 0 4 0 0 0,7813 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,7639 0
29
at 0 0 0 0 0 0 0 0,811 0 0
30
at 0 0 0 0 0 0 0 9 0 0 0,7774 0 0 0 0
31
at 0 0 0 0 0 0 0 0 0,776 0 0 0
32
at 5
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7821
at
3334 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,811
at
35 0 0 0 0 0 0 0 0 0,801 0
at
36 0 0 0 0 0 0 0 0 4 0 0,7837
at
37 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7798
at
38 0 0 0 0 0 0 0 0 0 0,7779 0 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,817 0 0 0
at 0 0 0 0 0 0 0 0 0,807
40
1

Table 3.12. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.2

Table 3.13. Cross-validation: total bets and profits for each type, with β = 0.4 and β = 0.6
12
2
 CHAPTER 3. Results

β = 0,4 β = 0,6
Guy Total bets Total
Revenue Guy bets
Revenue

1 0,5888 0 1 0,8701 0

2 0,7433 0 2 0,6465 0
3 0,7463 0 3 0,6486 0
4 0,5971 0 4 0,8532 0
5 0,5833 0 5 0,8338 0
6 0,5939 0 6 0,8534 0
7 0,6187 0 7 0,8581 0
8 0,6325 0 8 0,916 0
9 0,5914 30,1615 9 0,8658 22,0242
I.C
. 0,6929 0 10 0,5708 0
11 0,7317 0 11 0,6212 0

12 0,6248 0 12 0,9388 0
13 0,5938 0 13 0,422 0
14 0,7702 0 14 0,6928 0
15 0,6408 0 15 0,9698 0
16 0,6818 0 16 0,9824 0
17 0,7224 0 17 0,6123 0
18 0,6012 0 18 0,9996 0
19 0,5955 0 19 0,9338 0
20 0,594 0 20 0,8581 0
21 0,6886 0 21 0,5644 0
22 0,5906 0 22 0,8092 0
23 0,6528 0 23 0,5109 0
24 0,5717 0 24 0,8204 0
25 0,6089 0 25 0,8984 0
26 0,7939 0 26 0,7213 0
27 0,6565 0 27 0,9349 0
28 0,6019 0 28 0,9487 0
29 0,5936 0 29 0,8283 0
30 0,6883 0 30 0,5646 0
31 0,6138 0 31 0,9869 0
32 0,5992 0 32 0,9585 0
33 0,6496 0 33 0,9575 0
34 0,6811 0 34 0,9847 0
35 0,6582 0 35 0,9497 0
36 0,6117 0 36 0,8926 0
37 0,6497 0 37 0,9755 0
38 0,6044 0 38 0,6019 0
39 0,6987 0 39 0,5803 0
40 0,6579 0 40 0,5087 0
Total 25,81 30,1615 Total 31,94 22,0242
55 45

12
3
CHAPTER 3. Results

• β = 0,4
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,5888 0 0 0
to
1 0 0 0 0 0 0 0 0 0 0,7433
to
2 3 0 0 0 0 0 0 0 0 0,746 0
to 4 0 0 0 0 0 0 0 0 3 0 0,5971 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,5833
to 6 0 0 0 0 0 0 0 0 0 0 0 0,5939 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,6187
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,6325 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,5914
to
10 0 0 0 0 0 0 0 0 0 0,6929
at
11 0 0 0 0 0 0 0 0 0 0,7317
at 0 0 0 0 0 0 0 0 0 0 0 0 0,6248 0
at
1213 0 0 0 0 0 0 0 0 0 0 0 0 0,5938 0
at
14 0 0 0 0 0 0 0 0 0 0 0,7702
at
15 0 0 0 0 0 0 0 0,640 0 0 0 0 0 0
at 0 0 0 0 0 0 0 8 0 0 0 0 0 0,6818
16
at 0 0 0 0 0 0 0 0 0 0 0,7224 0 0
17
at 0 0 0 0 0 0 0 0 0 0,6012
18
at 0 0 0 0 0 0 0 0 0 0,5955
19
to 0,594 0 0 0 0 0 0 0 0 0
20
at 0 0 0 0 0 0 0 0 0 0 0 0 0,6886 0
at
2122 0 0 0 0 0 0 0 0 0,590 0 0 0 0 0 0 0
at
23 0 0 0 0 0 0 0 0 6 0,652 0 0
at
24 0 0 0 0 0 0 0,571 0 8 0 0 0 0 0 0
at
25 0 0,608 0 0 0 0 7 0 0 0 0 0 0 0 0
at
26 09 0 0 0 0 0 0 0 0 0,7939
at
27 0 0 0 0 0 0 0 0,656 0 0 0 0 0 0
at 0 0 0 0 0 0 0 5 0 0 0,6019 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,5936 0
29
at 0 0 0 0 0 0 0 0,688 0 0
30
at 0 0 0 0 0 0 0 3 0 0 0,6138 0 0 0 0
31
at 0 0 0 0 0 0 0 0 0,599 0 0 0
32
at 2
33 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6496
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6811
at
3435 0 0 0 0 0 0 0 0 0,658 0
at
36 0 0 0 0 0 0 0 0 2 0 0,6117
at
37 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6497
at
38 0 0 0 0 0 0 0 0 0 0,6044 0 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,6987 0 0 0
at 0 0 0 0 0 0 0 0 0,657
40
9

Table 3.14. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.4

Table 3.13. Cross-validation: total bets and profits for each type, with β = 0.4 and β = 0.6
12
4
 CHAPTER 3. Results

• β = 0,6
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,417 0,4531 0 0
to 0 0 0 0 0 0 0 0 0 0,6465
to
2 0 0 0 0 0 0 0 0 0,648 0
to
3 0 0 0 0 0 0 0 0 6 0 0,4347 0 0,4185 0
to
4 0 0 0 0 0 0 0 0 0 0 0 0,4235 0 0,4103
to
5 0 0 0 0 0 0 0 0 0 0 0 0,4252 0 0,4282
6 7 0 0 0 0 0 0 0 0,396 0 0 0 0 0,4619
at
at 8 0 0 0 0 0 0 0 2 0 0 0 0 0 0,4949 0,4211
at 9 0 0 0 0 0 0 0 0 0 0 0 0,434 0 0,4318
to 0 0 0 0 0 0 0 0 0 0,5708
10
at 0 0 0 0 0 0 0 0 0 0,6212
11
at 0 0 0 0 0 0 0 0 0 0 0 0 0,4781 0,4607
12
at 0 0 0 0 0 0 0 0 0 0 0 0 0,422 0
13
at 0 0 0 0 0 0 0 0 0 0 0,6928
14
at
15 0 0 0 0 0 0 0 0,516 0 0 0 0 0 0,4537
at
16 0 0 0 0 0 0 0 1 0 0 0 0,4231 0 0,5593
at
17 0 0 0 0 0 0 0 0 0 0 0,6123 0 0
at
18 0 0 0 0 0 0,551 0 0 0 0,4482
at
19 0 0 0,484 0 0 4 0 0 0 0 0,4496
to 0,4279 02 0 0 0 0 0 0 0 0,4302
at
2021 0 0 0 0 0 0 0 0 0 0 0 0 0,5644 0
at 0 0 0 0 0 0 0 0 0,429 0 0 0 0 0,3795 0 0
22
at 0 0 0 0 0 0 0 0 7 0,510 0 0
23
at 0 0 0 0 0 0,414 0,405 0 9 0 0 0 0 0 0
24
at 0 0,455 0 0 0 6 0 8 0 0 0 0 0 0 0 0,4429
25
at 5
0 0 0 0 0 0 0 0 0 0,7213
26
at
27 0 0 0 0 0 0 0 0,518 0 0 0 0,4169 0 0
at
28 0 0 0 0 0 0 0 0 0 0,4304 0 0 0 0,5183
at 0 0 0 0,397 0 0 0 0 0 0 0,4312 0
at
2930 0 0 0 1 0 0 0 0 0,564 0 0
at
31 0 0 0 0 0 0 0 6 0 0 0,4583 0 0 0 0,5286
at
32 0 0 0 0 0 0 0 0 0,429 0 0 0,5292
at 0 0 0 0 0 0 0,431 0 3 0 0 0 0 0 0 0,5257
at
33 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0,4291 0,5556
34
at 0 0 0 0 0 0 0 0,420 0,529 0
35
at 0 0 0 0 0 0 0,443 2 0 5 0 0,4488
36
at 0 0 0 0 0 0 8 0 0 0 0 0,4492 0 0 0,5263
37
at 0 0 0 0 0 0 0 0 0 0 0 0,6019 0 0
38
at
39 0 0 0 0 0 0 0 0 0 0 0,5803 0 0 0
at
40 0 0 0 0 0 0 0 0 0,508
7

Table 3.15. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.6

Table 3.10. Cross-validation: total bets and profits for each type, with β = 0.1 and β = 0.2

12
5
CHAPTER 3. Results

β = 0,7 β = 0,8
Guy betting Revenue
totals Guy Total bets Revenue
1 0,6973 0
1 0,5246 0
2 0,5981 0
2 0,8961 0
3 0,9534 0
3 0,8096 0
4 0,6723 0
4 0,8238 0
5 0,6525 0
5 0,721 0
6 0,6698 0
6 0,7824 0
7 0,6777 0
7 0,7968 5,9533
8 0,7571 0
8 0,8318 0
9 0,6916 17,644
9 0,7602 13,5068
10 0,9123 0
10 0,7601 0
11 0,566 0
11 0,9109 0
12 0,7799 0
12 0,9163 0
13 0,9184 0
13 0,7778 0
14 0,6541 0
14 0,9026 0
15 0,8159 0
15 0,9112 0
16 0,8297 0
16 0,7696 0
17 0,913 0
17 0,766 0
18 0,8307 0
18 0,6618 0
19 0,7651 0
19 0,8592 0
20 0,8138 0
20 0,9356 0
21 0,8508 0
21 0,9373 0
22 0,9539 0
22 0,6712 0
23 0,8635 0
23 0,9721 0
24 0,7626 0
24 0,8613 0
25 0,7306 0
25 0,8243 0
26 0,6849 0
26 0,8866 0
27 0,7689 0 27 0,8827 0
28 0,7808 0 28 0,8369 8,8607
29 0,6441 0 29 0,7479 0
30 0,8881 0 30 0,7423 0
31 0,8314 0 31 0,9232 0
32 0,7931 0 32 0,6278 0
33 0,8059 0 33 0,9222 0
34 0,8298 0 34 0,9199 0
35 0,779 0 35 0,8616 0
36 0,7139 0 36 0,8895 0
37 0,8302 0 37 0,6848 0
38 0,903 0 38 0,7553 0
39 0,8489 0 39 0,9548 0
40 0,8923 0 40 0,741 0
Total 31,32 17,644 Tota 32,7601 28,3208
44 l

12
6
 CHAPTER 3. Results

• β = 0,7
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,328 0,3693 0 0
to
1 0 0 0 0 0 0 0 0 0 0,5981
to
2 3 0 0 0 0 0 0 0 0 0,610 0,3433
to 4 0 0 0 0 0 0 0 0 1 0 0,346 0 0,3263 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0,3332 0 0,3193
to 6 0 0 0 0 0 0 0 0 0 0 0 0,3346 0 0,3352
at 7 0 0 0 0 0 0 0 0,298 0 0 0 0 0,379
7
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,4216 0,3355
at 9 0 0 0 0 0 0 0 0 0 0 0 0,3456 0 0,346
to
10 0 0 0 0 0 0 0 0 0,382 0,5299
at
11 0 0 0 0 0 0 0 0 4 0 0,566
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,4001 0,3798
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,3419 0,5765
at
14 0 0 0 0 0 0 0 0 0 0 0,6541
at
15 0 0 0 0 0 0 0 0,444 0 0 0 0 0 0,3716
at 0 0 0 0 0 0 0 3 0 0 0 0,3347 0 0,495
16
at 0 0 0 0 0 0 0 0 0 0 0,5642 0,3488 0
17
at 0 0 0 0 0 0,475 0 0 0 0,3557
18
at 0 0 0,401 0 0 0 0 0 0 0,3632
19
to 9 0,473
0,3404 0 0 0 0 0 0 0 0
20
at 0 0 0 0 0 0 4 0 0 0 0 0 0 0,5075 0,3433
at
2122 0 0 0 0 0 0 0 0 0,35 0 0 0,3125 0 0,2914 0 0
at
23 0 0 0 0 0 0 0 0 0,448 0,4153 0
at
24 0 0 0 0 0 0 0,317 0 2 0 0 0 0 0,4451 0
at
25 0 0,372 0 0 0 0 5 0 0 0 0 0 0 0 0,358
at
26 06 0 0 0 0 0 0 0 0 0,6849
at
27 0 0 0 0 0 0 0 0,444 0 0 0 0,3247 0 0
at 0 0 0 0 0 0 0 2 0 0 0,3395 0 0 0 0,4413
28
at 0 0 0 0,299 0 0 0 0 0 0 0,3442 0
29
at 0 0 0 9 0 0 0 0 0,509 0 0,3789
30
at 0 0 0 0 0 0 0 2 0 0 0,3754 0 0 0 0,456
31
at 0 0 0 0 0 0 0 0 0,339 0 0 0,4536
32
at 0,348 5
33 0 0 0 0 0 0 0 0 0 0 0 0 0 0,4577
at
34 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0,34 0,4898
at
35 0 0 0 0 0 0 0 0,324 0,454 0
at
36 0 0 0 0 0 0 0,352 4 0 6 0 0,361
at
37 0 0 0 0 0 0 9 0 0 0 0 0,3701 0 0 0,4601
at
38 0 0 0 0 0 0 0 0 0 0,3523 0 0,5507 0 0
at 0 0 0 0 0 0 0 0 0 0 0,5275 0 0,3214 0
at
39 0 0 0 0 0 0 0 0,451 0,440
40
6 7

Table 3.17. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.7

Table 3.10. Cross-validation: total bets and profits for each type, with β = 0.1 and β = 0.2

12
7
 CHAPTER 3. Results

• β = 0,8
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,239 0,2856 0 0
to
1 0 0 0 0 0 0 0 0 0,341 0,5548
3 0,558
to
2 3 0 0 0 0 0 0 0 0 0,2511
to 4 0 0 0,314 0 0 0 0 0 5 0 0,2654 0 0,2438 0
to 5 0 06 0 0 0 0 0 0 0 0 0 0,249 0,24 0,232
to 6 0 0 0 0 0 0 0 0 0 0,2789 0 0,2499 0 0,2536
at 7 0 0 0 0 0 0,283 0 0,212 0 0 0 0 0,3006
5 7
at 8 0 0 0 0 0 0 0 0 0 0 0 0,2266 0,3518 0,2534
at 9 0 0 0 0 0 0 0 0 0 0 0 0,2665 0,2289 0,2648
to
10 0 0 0 0 0 0 0 0 0,295 0,4644
at 0 0 0 0 0 0 0 0 7 0,384 0,5268
at
1112 0 0 0 0 0 0 0 0 1 0,286 0 0 0 0,3264 0,3032
at
13 0 0 0 0 0 0 0 0 7 0 0 0 0 0,2555 0,5223
at
14 0 0 0 0 0 0 0 0 0,279 0 0,6236
at
15 0 0 0 0 0 0 0 0,379 0,234 0 0 0 0 0,2968
at 0 0 0 0 0 0 0 9 0 5 0 0 0 0,338 0,4316
16
at 0 0 0 0 0 0 0 0 0 0 0,5074 0,2586 0
17
at 0 0 0 0 0 0,398 0 0 0 0,2632
18
at 0,332 6 0,239
0 0 0 0 0 0 0 0,2872
19
to 8 2 0,415
0,2571 0 0 0 0 0 0 0 0,2632
20
at
21 0 0 0 0 0 0 3 0 0 0,229 0 0 0 0,4476 0,2604
at
22 0 0 0 0 0 0 0 0 3 0,261 0 0 0,2161 0 0,1941 0 0
at
23 0 0 0 0 0 0 0 0 0,380 0,3444 0,2469
at 0 0 0 0 0 0,246 0,233 0 8 0 0 0 0 0,3815 0
at
2425 0 0,294 0 0 0 0 8 0 0 0 0 0 0,2512 0 0,2782
at
26 09 0 0 0 0 0 0,234 0 0 0,6525
at
27 0 0 0 0 0 0 1 0 0,374 0 0 0 0,2394 0,2687 0
at 0 0 0 0 0 0 0 6 0 0,216 0,252 0 0 0 0,3688
28
at 0 0 0 0,216 0,268 0 0 0 1 0 0 0,2627 0
29
at 0 0 0 5 0 7 0 0 0 0,445 0 0,2964
30
at 0 0 0 0 0 0 0 9 0 0,237 0,2962 0 0 0 0,3895
31
at 5 0,249
0 0 0 0 0 0 0 0 0 0 0,378
32
at 0,269 8 0,257
0 0 0 0 0 0 0 0 0 0 0 0 0,395
at
3334 0 0 0 0 0 0 9 0 0,236 3 0 0 0 0 0 0 0,2569 0,4264
at 0,232 6 0,241 0,388
35 0 0 0 0 0 0 0
at
36 0 0 0 0,335 0 0 2 0,273 3 0 1 0 0,2802
at
37 0 0 0 4 0 0 0 9 0 0 0 0 0,291 0 0 0,3938
at
38 0 0 0 0 0 0 0 0 0 0,2646 0 0,4907 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,4716 0,2473 0,2359 0
at 0 0 0 0 0 0 0 0,376 0,365
40

Table 3.18. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.8

12
8
CHAPTER 3. Results

β = 0,9

Table 3.19. Cross-validation: total bets and profits for each type, with β = 0.9 and β = 1
12
9
CHAPTER 3. Results

Total Guy Total bets Revenue

Guy bets Revenue
1 0,9636 0 1 0,9101 0

2 0,9118 0 2 0,8017 0

3 0,8525 0 3 0,9535 0,2032

4 0,885 3,6635 4 0,9053 1,3762

5 0,8853 0 5 0,7429 0,3806

6 0,8701 0 6 0,5871 0,3961

7 0,9526 4,3978 7 0,6501 2,3841

8 0,9354 0 8 0,631 0,1452

9 0,8551 9,3958 9 0,5683 4,9031

10 0,9851 0 10 0,7563 0,1151

11 0,7701 0 11 0,875 1,4325

12 0,9951 0 12 0,8808 0

13 0,9537 0 13 0,9371 0

14 0,9415 0 14 0,7631 0,027

15 0,9368 0 15 0,5314 0,1407

16 0,9551 0 16 0,9205 0,7076

17 0,8744 0 17 0,7335 0

18 0,9384 0 18 0,9325 0,8487

19 0,976 0 19 0,9416 0,7856

20 0,903 0 20 0,8743 0

21 0,9791 0 21 0,9212 0,0923

22 0,9962 0 22 0,5194 0

23 0,9039 0 23 0,9456 0

24 0,9601 0 24 0,7933 0

25 0,9003 0 25 0,9084 0

26 0,936 0 26 0,8727 1,1814

27 0,8588 0 27 0,861 0,1172

28 0,9219 5,147 28 0,8543 1,2807
29 0,9785 0 29 0,8242 0,5317

30 0,8058 0 30 0,952 0,0403

31 0,8842 0 31 0,9834 0
32 0,931 0 32 0,7913 0

33 0,8846 0 33 0,8663 0,192

34 0,9631 0 34 0,642 0
35 0,7729 0 35 0,8309 0,2494
36 0,9814 0 36 0,7087 0,8842
37 0,8256 14,1576 37 0,9716 10,7324
38 0,8722 0 38 0,9372 0
39 0,9258 0 39 0,9863 0,7276

40 0,9857 0 40 0,8638 0
Total 36,6077 36,7617 Total 32,929 29,8749
7

Table 3.19. Cross-validation: total bets and profits for each type, with β = 0.9 and β = 1
13
0
 CHAPTER 3. Results

• β = 0,9
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0,4498 0,1561 0,211 0 0,1467
to
1 0 0 0 0 0 0 0,136 0 0,267 0,5074
to
2 3 0 0 0 0,172 0 0 8 0 0 6 0,512 0,1677
to 4 0 0 0,239 5 0 0 0 0 0,162 3 0 0,1826 0,1409 0,1587 0
to 5 0 09 0 0 0 0 0,194 9 0 0 0,2197 0 0,1672 0,158 0,1461
to 6 0 0,167 0 0 0,165 0 3 0 0 0 0,2018 0 0,1647 0 0,171
4 2 0,209 0,219 0,128
at 7 0 0 0 0 0 0 0 0,1727 0 0,2228
4 6 1
at 8 0 0 0,138 0 0 0 0 0 0 0 0,1998 0,1434 0,2824 0,1718
0 0 0 0 0,145 0 0 0 0 0 0,1908 0,1885 0,1461 0,1842
at 9
to
10 0 0 0,176 0 5 0 0 0 0,163 0,228 0,4176
at
11 0 03 0 0 0 0 0 2 0 0,302 0,4681
at 0 0 0 0 0 0 0,143 0 0,208 0 0,1656 0 0,2519 0,2258
at
1213 0 0 0 0 0 0 8 0,131 0 0 0 0,1754 0 0,1722 0,4744
at
14 0 0 0 0 0,153 0 7 0 0 0,200 0 0,5872
at
15 0 0 0,129 0 7 0 0 0 0,313 6 0,150 0,1234 0 0 0 0,2199
at 0 08 0 0 0 0 0 5 0 2 0,149 0 0,1663 0,2685 0,3707
16
at 0 0 0 0 0 0 0 0,128 6 0 0 0,4555 0,1783 0,112
17
at 0 0 0 0 0 0,344 0,18 6 0,223 0 0,19
18
at 0,264 5 0,142 0,160 9 0,196
0 0 0 0 0 0,2122
19
to 7 6 3 0,346 2 0,210
0,1699 0 0 0 0 0 0 0,1762
20
at 0 0 0 0 0 0 8 0 1 0 0,141 0 0 0,2737 0,387 0,1767
at
2122 0 0 0 0 0 0,115 0 0 7 0,185 0,257 0 0,1375 0 0,1099 0 0,1911
at
23 0 0 0 0 0 5 0 0 0,169 2 0,309 0,2683 0,1567
at
24 0 0 0 0 0 0,164 0,150 0 9 0,181 0 0 0,1465 0,3179 0
at
25 0 0,216 0 0 0 4 0 1 0 0,175 2 0 0,1419 0 0,1685 0 0,198
at
26 08 0 0 0 0 00 0,148 1 0,170 0,6176
at
27 0 0 0 0 0 0 0,302 4
0 0 0,2216 0 0,1507 0,1836 0
at 0 0 0 0 0 0 0,128 9 0 0,125 0,1649 0,206 0 0 0,2968
28
at 0 0 0 0,139 0,200 0 7 0 0,145 5 0 0,1564 0,1838 0,1531
29
at 0 0 0 1 0 8 0 0,201 0 3 0,386 0 0,2184
30
at 2 2 0,148
0 0 0 0 0 0 0 0 0,215 0,2007 0 0 0,3197
31
at 8 0,166
0 0 0 0 0 0 0 0 0,1592 0,2889 0,3163
32
at 0,188 6 0,174
33 0 0 0 0 0 0 0 0 0 0 0 0,1926 0,3293
at 0 0 0 0 0 0 5 0 0,151 2 0 0 0,1338 0,1432 0 0 0,1726 0,3625
at
3435 0 0 0 0 0 0 0,138 0,149 0,316 0,1685
at
36 0 0 0 0,262 0 0,147 9 0,192 5 0,181 0 0,1978
at
37 0 0 0 6 0 0 2 0 1 0 7 0 0 0,2776 0,2162 0 0 0,3318
at
38 0 0 0 0 0 0 0 0 0 0,1809 0,2554 0,4359 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,413 0,1606 0,1479 0,2043
at 0 0 0,247 0 0 0,132 0 0,310 0,295
40
7 7 3

Table 3.20. Cross-validation: bets we make for each horse, depending on the type, with
β = 0.9

13
1
 CHAPTER 3. Results

Horse1 Horse2 Horse3

Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0,0074 00 0,058 0,008 0,004 0 0,033 0,092 0,0215 0,3937 0,0683 0,129 0,0365 0,0574
to 0 0 0,022 1 0
7
0 0,043 0,043
2 0,044 3 0,190
0,4581
to
2 0 0,036 3 0,023 0,093 0,020 6 0,058 0,149 5 0,016 2 0,467 0,0871
to
3 09 0 4 0,163 8 0,108 6 0,065 0,031 2 0,062 6 0,077 9 0,058 0,0988 0,0529 0,0724 0,1146
to
4 0 0 2 0,058 8 0,034 0 3 0,023 3 0,105 6 0,060 4 0,125 0,1344 0 0,0778 0,067 0,0556
to
5 0,0609 0,062 6 0 6 0 0,060 7 0,027 0,058 3 0,016 9 0 0,1029 0,0443 0,0722 0,0076 0,0741
6
at 7 06 0 0 0,085 2 0 3 0,113 3 0,121 7 0,025 0,060 0 0,0671 0,0383 0,1381
0 0,036 3 0,058
0 0,013 5 0,01 5 0,041 9 0,004 4 0 0,0222 0,1055 0,0482 0,2069 0,084
at 8
at 9 0 08 0 6 0,018 2 0,045 0 3 0,000 3 0,029 0,137 0 0,0958 0,0958 0,0487 0,0961
to 0 0,014 0,080 1 0,001 5 0 0,021 9 0,087 6 0,065 8 0,137 0,3482
10
at 0 4 0,006 1 0,014 6 0,033 0 4 0 6 0,030 5 0,130 5 0,240 0,4193
11
at 05 06 0 1 0,005 0,009 0,025 9 0,052 2 0,072 4 0,124 0,0527 0,0776 0,1406 0,1745 0,1455
12
at 0 0 0,024 1 0,009 3 0,040 3 0,005 7 0,041 8 0,025 7 0,037 0,0029 0,0903 0,1491 0,0871 0,4226
13
at 1 7 7 0,055 3 8 8 0,049 7 0,111
0 0 0 0 0 0 0,0027 0,5445
14
at
15 0 0 0,023 0 5 0 0 0,029 2 0,234 2 0,049 0,0196 0 0,0235 0,0216 0,1304
at
16 0 0 0,011 0,032 0,011 0,043 7 0,040 3 0,091 3 0,061 0,0472 0,0802 0,1936 0,3076
03 0 3 0,025 5 0,021 4
at 7 0,120 4 0,031 3 0,024
17 0,009 0 0 0,3988 0,0883 0,0136
at
18 0 0,084 0,109 6 0,031 5 0,068 0,276 7 0,093 9 0,142 1 0,021 0,1046
0,2293 9 0,098 6 0,189 6 0 1 0,036 3 0,053
at 5 0,072 3 0,112 6 0,019
19 0,1311
0,0828 2 0 2 0,061 0,078 1 0,019
to 0,103 9 0,278 3 0,126 5 0,033 0,0894
0 5 0,025 5 0,077 6 0,001 4 0,026
at 7 0,008 5 0,052 9 0,052
2021 0 0,0172 0,047 0,1977 0,325 0,091
at 0 0,016 1 0 3 0,000 9 0,068 2 0,006 4 0 3 0 1 0,092 0,1541 0,0424 0,0353 0 0,0084 0,0154 0,0791
22
at 0,0166 6 0,070 0,059 7 0 8 0,031 6 0,091 0,080 0,087 0,241 0,1955 0,073
23
at 01 0,003 0,068 0,043 4 0,028 4 0,073 5 0,058 0,027 1 0,088 0,0213 0,0278 0,0523 0,2428 0,0567
24
at 0,0534 6 0,137 2 0,048 3 0,060 9 0 7 0 6 0,053 7 0,091 4 0,122 0,0545 0,0332 0,0839 0,0529 0,1165
25
at 4 5 0,012 8 0,090 0,011 0,006 4 0,058 8 0,030 1 0,082
0 0 0,5808
26
at 5 0,022 9 0,077 3 0,015 5 0,019 2 0,231 5
27 0 0 0,019 0 0,1417 0,0744 0,0627 0,0993 0,097
at
28 0 0,023 7 0,026 4 0,081 6 0,008 6 0,021 0,034 6 0,036 0,031 0,0756 0,1198 0,0414 0,1316 0,2218
at 0,1436 0 9 0,190 6 0,043 6 0,110 6 0,007 5 0,070 7 0,045 2 0 0,0597 0,0972 0,056
at
2930 0 0,030 1 0,020 4 0,005 2 0,108 2 0,125 9 0,088 9 0,327 0,103 0,1424
0,0327 9 0 5 0,052 8 0,010 0,128 3 0,033 8 0,007
at 0,021 3 0,063
31 0,1359 0,1216 0,0223 0,1004 0,2533
05 0 9 0,017 1 0,008 7 0,036 6 0,076
at 4 0,077
32 0,0571 0 0,0657 0,211 0,2415
at 0 0 0,012 3 0,019 0,007 0,081 6 0,107 9 0,027 2 0,091 0,0123 0,0771 0,0308 0,0224 0,1119 0,264
at
33 0 0,010 8 0 5 0,025 7 0 6 0 5 0 1 0,048 6 0,014 0,0728 0,0289 0,0394 0,0105 0,0173 0,0795 0,295
34
0 4 0,003 0,330 9 0,053 1 0,065
at 0,041 2 0,248
0 0 0,0872
35
0 6 0,005 8 0 6 0,043 8 0,098 2 0,083
at 0,173 7 0,196
0 0,108
36
04
at 0,017 1 0,009 0,001 7 0,002 1 0,017 9 0,068 5 0,035
0 0,2104 0,1414 0,1733 0,0245 0,2699
37
at 3 6 0,032 7 0,006 4 0,071 4 0,007 2 0,011 5 0,026
0 0 0 0,0963 0,1795 0,38 0,0478 0,0772
38
at 0,103 5 0,042 7 0,022 9 0,040 3 0,001 8 0,100
2 0,007
39 0,0133 0 0,0353 0,3561 0,076 0,0616 0,125
at 9 0,162 2 0,085 1 0,003 7 0,035 5 0,014 7 0,235 9 0,219
40 0,107 0
9 6 6 1 3 5 8

Table 3.21. Cross-validation: bets we make for each horse, depending on the type, with β = 1

As we see, the largest difference between what we bet and what we win occurs
with β = 0 . With that value we have a profit of €9.7779. Therefore, in this case we could
say that the optimal value of β would be 0.

However, looking at what we bet on each horse in each race, we can say that up to
β = 0.6 we usually bet only on one horse per race, and, furthermore, with β = 0 up to β =
0.7 we are only making profits for a career, even if the benefit is very great. This carries a
great risk, since if the horse on which we have bet is not the winner, a situation that can
occur with great probability, we will have a great loss.

Let's give an example, assuming equal probability that each horse wins, and taking

into account that it is a race with ten horses. The probability that each horse wins will be

13
2
 CHAPTER 3. Results

10 = 0.1 and, therefore, the probability that we will obtain profits by betting only on one

horse will be 10%, compared to 90% that we will lose money. Therefore, we can

conclude that it is very risky to bet on only one horse in each race.

If we look at the betting values for each horse with β = 0.9 and β = 1 we see that
the number of horses bet on in each race increases considerably.

Furthermore, with β = 0.9 we obtain a profit of €0.154. If we compare it with the

€9.7 of β = 0 we see that it is a very low value. On the other hand, in this case we have a
better chance of guessing the winner of a race, since we are betting on a greater number
of horses.

If we continue with the previous example, assuming that we bet on four horses. The
probability that we win will be 40% and that we will not be right will be 60%. If we
compare it with 10%, we see that, in this case, our bet carries a lower level of risk.

Therefore, from our point of view, it would be better to use a value of β of 0.9,
since we have profits, and we also ensure that we will not have a large number of losses
if we are not correct in our prediction. Thus, our optimal value of β will be 0.9.

On the other hand, risky bettors (section 1.5.) could decide that the best value to bet
is 0.

In conclusion we can say that, depending on the type of bettor we are, we will
choose one value of β or another.

3.1.3.2. LOO Studio

13
3
CHAPTER 3. Results

β=0 β = 0,05
Total Total
Guy Revenue Guy Revenue
bets bets
1 0,941 0 1 0,899 0
4
2 0,936 0 2 2 0,911 0
3 8 0,941 0 3 6 0,909 0
4 8 0,942 0 4 9 0,900 0
5 7 0,94 0 5 7 0,901 0
2
6 0,948 0 6 3 0,905 0
7 6 0,944 0 7 3 0,908 0
8 2 0,920 0 8 2 0,882 0
9 4 0,927 47,3216 9 7 0,88 45,2868
I. 9 0,936 0 1 8 0,900 0
C. 1 9 0,952 0 0 1 1 0,931 0
1 1 7 0,932 0 1 1 6 0,895 0
2 1 2 0,937 0 2 1 5 0,895 0
3 1 3 0,92 0 3 1 6 0,898 0
4 1 5 0,917 0 4 1 6 0,879 0
5 1 8 0,935 0 5 1 4 0,901 0
6 1 3 0,942 0 6 1 4 0,914 0
7 1 7 0,957 0 7 1 8 0,912 0
8 1 3 0,927 0 8 1 9 0,889 0
9 2 6 0,948 0 9 2 7 0,902 0
0 2 4 0,93 0 0 2 8 0,90 0
1 2 7 0,928 0 1 2 5 0,887 0
2 2 3 0,936 0 2 2 1 0,906 0
3 2 7 0,928 0 3 2 7 0,883 0
4 2 6 0,933 0 4 2 2 0,892 0
5 2 5 0,939 0 5 2 6 0,925 0
6 2 3 0,946 0 6 2 9 0,905 0
7 2 2 0,960 0 7 2 9 0,91 0
8 2 7 0,934 0 8 2 7 0,891 0
9 3 2 0,935 0 9 3 9 0,90 0
0 3 6 0,940 0 0 3 7 0,899 0
1 3 9 0,953 0 1 3 8 0,908 0
2 3 9 0,914 0 2 3 9 0,884 0
3 3 6 0,940 0 3 3 5 0,905 0
4 3 9 0,944 0 4 3 1 0,91 0
5 3 6 0,955 0 5 3 6 0,913 0
6 3 7 0,909 0 6 3 7 0,873 0
7 3 9 0,951 0 7 3 8 0,909 0
8 3 5 0,935 0 8 3 5 0,90 0
9 4 4 0,956 0 9 4 1 0,92 0
0Tota 2 37,54 47,3216 0Total 4 36,0879 45,2868
l 27
Table 3.22. Leave-One-Out: Total bets and winnings for each type, with β=0 and β=0.05

13
4
 CHAPTER 3. Results

Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,9414 0 0 0
to 2 0 0 0 0 0 0 0 0 0 0,9368
to 3 0 0 0 0 0 0 0 0 0,9418 0
to 4 0 0 0 0 0 0 0 0 0 0,9427 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,942
to 6 0 0 0 0 0 0 0 0 0 0 0 0,9486 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,9442
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,9204 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9279
to
10 0 0 0,9369 0 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,9527
11
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,9322 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,9373 0
at 0 0 0 0 0 0 0 0 0 0 0,925
14
at
15 0 0 0 0 0 0 0 0,9178 0 0 0 0 0 0
at
16 0 0 0 0 0 0 0 0 0 0 0 0 0,9353
at 0 0 0 0 0 0 0 0 0 0 0,9427 0 0
17
at
18 0 0 0 0 0 0 0 0 0 0,9573
at
19 0 0 0 0 0 0 0 0 0 0,9276
to 0,9484 0 0 0 0 0 0 0 0 0
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,937 0
at 0 0 0 0 0 0 0 0 0,9293 0 0 0 0 0 0 0
22
at 0 0 0 0 0 0 0 0 0,9367 0 0
23
at
24 0 0 0 0 0 0 0,9286 0 0 0 0 0 0 0
at 0 0,9335 0 0 0 0 0 0 0 0 0 0 0 0
25
at 0 0 0 0 0 0 0 0 0 0,9393
26
at
27 0 0 0 0 0 0 0 0,9462 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,9607 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,9342 0
29
at
30 0 0 0 0 0 0 0 0,9356 0 0
at 0 0 0 0 0 0 0 0 0 0,9409 0 0 0 0
31
at
32 0 0 0 0 0 0 0 0 0,9539 0 0 0
at 0 0 0 0 0 0 0,9146 0 0 0 0 0 0 0 0
33
at
34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9409
at
35 0 0 0 0 0 0 0 0 0,9446 0
at 0 0 0 0 0 0 0 0 0 0,9557
36
at
37 0 0 0 0 0 0 0 0 0,9099 0 0 0 0 0
at
38 0 0 0 0 0 0 0 0 0 0,9515 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,9354 0 0 0
39
at
40 0 0 0 0 0 0 0 0 0,9562

Table 3.23. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0

13
5
CHAPTER 3. Results

• β = 0,05
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,8992 0 0 0
to 2 0 0 0 0 0 0 0 0 0 0,9116
to 3 0 0 0 0 0 0 0 0 0,9099 0
to 4 0 0 0 0 0 0 0 0 0 0,9007 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9013
to 6 0 0 0 0 0 0 0 0 0 0 0 0,9053 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,9082
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,8827 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,888
to 0 0 0 0 0 0 0 0 0 0,9001
at
1011 0 0 0 0 0 0 0 0 0 0,9316
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,8955 0
at 0 0 0 0 0 0 0 0 0 0 0 0 0,8956 0
13
at
14 0 0 0 0 0 0 0 0 0 0 0,8986
at
15 0 0 0 0 0 0 0 0,8794 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0 0 0,9014
16
at
17 0 0 0 0 0 0 0 0 0 0 0,9148 0 0
at
18 0 0 0 0 0 0 0 0 0 0,9129
at 0 0 0 0 0 0 0 0 0 0,8897
19
to
20 0,9028 0 0 0 0 0 0 0 0 0
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,905 0
at 0 0 0 0 0 0 0 0 0,8871 0 0 0 0 0 0 0
22
at
23 0 0 0 0 0 0 0 0 0,9067 0 0
at
24 0 0 0 0 0 0 0,8832 0 0 0 0 0 0 0
at 0 0,8926 0 0 0 0 0 0 0 0 0 0 0 0
25
at
26 0 0 0 0 0 0 0 0 0 0,9259
at
27 0 0 0 0 0 0 0 0,9059 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,917 0 0 0 0
28
at
29 0 0 0 0 0 0 0 0 0 0 0,8919 0
at
30 0 0 0 0 0 0 0 0,907 0 0
at 0 0 0 0 0 0 0 0 0 0,8998 0 0 0 0
31
at
32 0 0 0 0 0 0 0 0 0,9089 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8845
33
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,9051
34
at
35 0 0 0 0 0 0 0 0 0,916 0
at 0 0 0 0 0 0 0 0 0 0,9137
36
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8738
37
at
38 0 0 0 0 0 0 0 0 0 0,9095 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,901 0 0 0
39
at
40 0 0 0 0 0 0 0 0 0,924

Table 3.24. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.05

13
6
 CHAPTER 3. Results

β = 0,1
Total
Guy Revenue
bets
1 0,856 0
29 0,886 0
33 0,07 0
40 0,858 0
57 0,050 0
66 0,062 0
71 0,072 0
82 0,045 0
91 0,040 43,252
1 1 0,063 0
0 1 2 0,910 0
1 1 6 0,050 0
2 1 0 0,053 c
3 1 0 0,072 0
4 1 3 0,04 0
5 1 1 0,067 0
5 1 6 0,006 0
7 1 9 0,050 0
0 1 5 0,051 0
9 2 0 0,057 0
0 2 3 0,073 0
1 2 1 0,044 0
2 2 9 0,076 0
3 2 0 0,037 0
4 2 0 0,051 c
5 2 7 0,912 0
5 2 5 0,065 0
7 2 7 0,073 0
0 2 4 0,049 0
9 3 5 0,070 0
0 3 4 0,050 0
1 3 7 0,063 c
2 3 9 0,054 0
3 3 4 0,069 0
4 3 2 0,007 0
5 3 3 0,071 0
5 3 6 0,037 0
7 3 7 0,067 0
0 3 4 0,066 0
9 4 5 0,091 0
0Total 0 34,6321
β = 0,2
Table 3.25. Leave-One-Out: total bets and winnings for each type, with β = 0.1 and β = 0.2

13
7
 CHAPTER 3. Results

Total
Guy Revenue
bets
1 0,772 0
24 0,035 0
37 0,014 0
43 0,774 0
57 0,779 0
62 0,775 0
76 0,000 0
01 0,769 0
97 0,760 39,1325
1 3 0,709 0
0 1 5 0,060 0
1 1 6 0,705 0
2 1 4 0,770 0
3 1 4 0,019 0
4 1 6 0,764 0
5 1 2 0,799 0
6 0

Table 3.25. Leave-One-Out: total bets and winnings for each type, with β = 0.1 and β = 0.2

13
8
 CHAPTER 3. Results

1 0,031 0
7 1 2 0,779 0
0 1 7 0,775 0
9 2 9 0,766 0
0 2 2 0,009 0
1 2 1 0,760 0
2 2 4 0,01 0
3 2 7 0,747 0
4 2 1 0,769 0
5 2 0 0,005 0
6 2 7 0,705 0
7 2 1 0,70 0
8 2 6 0,764 0
9 3 9 0,021 0
0 3 2 0,776 0
1 3 4 0,773 0
2 3 0 0,794 0
3 3 2 0,797 0
4 3 6 0,030 0
5 3 1 0,707 0
6 3 5 0,765 0
7 3 6 0,703 0
0 3 3 0,797 0
9 4 7 0,027 0
0Total 3 31,720 39,1825
3

Table 3.25. Leave-One-Out: total bets and winnings for each type, with β = 0.1 and β = 0.2

13
9
 CHAPTER 3. Results

• β = 0,1
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,8569 0 0 0
to 2 0 0 0 0 0 0 0 0 0 0,8863
to 3 0 0 0 0 0 0 0 0 0,878 0
to 4 0 0 0 0 0 0 0 0 0 0,8587 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8606
to 6 0 0 0 0 0 0 0 0 0 0 0 0,8621 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,8722
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,8451 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8481
to
10 0 0 0 0 0 0 0 0 0 0,8632
at 0 0 0 0 0 0 0 0 0 0,9106
11
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,8588 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,8538 0
at 0 0 0 0 0 0 0 0 0 0 0,8723
14
at
15 0 0 0 0 0 0 0 0,841 0 0 0 0 0 0
at
16 0 0 0 0 0 0 0 0 0 0 0 0 0,8676
at 0 0 0 0 0 0 0 0 0 0 0,8869 0 0
17
at
18 0 0 0 0 0 0 0 0 0 0,8685
at
19 0 0 0 0 0 0 0 0 0 0,8518
to 0,8573 0 0 0 0 0 0 0 0 0
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,8731 0
at
22 0 0 0 0 0 0 0 0 0,8449 0 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0,8768 0 0
23
at
24 0 0 0 0 0 0 0,8378 0 0 0 0 0 0 0
at
25 0 0,8517 0 0 0 0 0 0 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,9125
26
at
27 0 0 0 0 0 0 0 0,8657 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,8734 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,8495 0
29
at
30 0 0 0 0 0 0 0 0,8784 0 0
at 0 0 0 0 0 0 0 0 0 0,8587 0 0 0 0
31
at
32 0 0 0 0 0 0 0 0 0,8639 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8544
33
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8692
34
at
35 0 0 0 0 0 0 0 0 0,8873 0
at 0 0 0 0 0 0 0 0 0 0,8716
36
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0,8377
37
at
38 0 0 0 0 0 0 0 0 0 0,8674 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,8665 0 0 0
39
at 0 0 0 0 0 0 0 0 0,8918
40

Table 3.26. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.1

14
0
 CHAPTER 3. Results

• β = 0,2
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 0 0 0 0 0 0 0 0 0 0 0 0,7724 0 0 0
to
1 0 0 0 0 0 0 0 0 0 0,8357
to
2 3 0 0 0 0 0 0 0 0 0,814 0
to 4 0 0 0 0 0 0 0 0 3 0 0,7747 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7792
to 6 0 0 0 0 0 0 0 0 0 0 0 0,7756 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,8001
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,7697 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7683
to
10 0 0 0 0 0 0 0 0 0 0,7895
at 0 0 0 0 0 0 0 0 0 0,8686
at
1112 0 0 0 0 0 0 0 0 0 0 0 0 0,7854 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,7704 0
at
14 0 0 0 0 0 0 0 0 0 0 0,8196
at
15 0 0 0 0 0 0 0 0,764 0 0 0 0 0 0
at 0 0 0 0 0 0 0 2 0 0 0 0 0 0,7998
16
at 0 0 0 0 0 0 0 0 0 0 0,8312 0 0
17
at 0 0 0 0 0 0 0 0 0 0,7797
18
at 0 0 0 0 0 0 0 0 0 0,7759
19
to 0,7662 0 0 0 0 0 0 0 0 0
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,8091 0
at
22 0 0 0 0 0 0 0 0 0,760 0 0 0 0 0 0 0
at
23 0 0 0 0 0 0 0 0 4 0,817 0 0
at 0 0 0 0 0 0 0,747 0 0 0 0 0 0 0
at
2425 0 0,769 0 0 0 0 1 0 0 0 0 0 0 0 0
at
26 08 0 0 0 0 0 0 0 0 0,8857
at
27 0 0 0 0 0 0 0 0,785 0 0 0 0 0 0
at 0 0 0 0 0 0 0 1 0 0 0,786 0 0 0 0
28
at 0 0 0 0 0 0 0 0 0 0 0,7649 0
29
at 0 0 0 0 0 0 0 0,821 0 0
30
at 0 0 0 0 0 0 0 2 0 0 0,7764 0 0 0 0
31
at 0 0 0 0 0 0 0 0 0,773 0 0 0
32
at 8
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7942
at
3334 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7976
at
35 0 0 0 0 0 0 0 0 0,830 0
at
36 0 0 0 0 0 0 0 0 1 0 0,7875
at
37 0 0 0 0 0 0 0 0 0 0 0 0 0 0,7656
at
38 0 0 0 0 0 0 0 0 0 0,7833 0 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,7977 0 0 0
at 0 0 0 0 0 0 0 0 0,827
40
3

Table 3.27. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.2

14
1
CHAPTER 3. Results

β = 0,4 β = 0,6
Guy Total Revenue Guy Total Revenue
bets bets
1 0,603 0 1 0,938 0
24 0,734 0 23 0,633 0
36 0,686 0 35 0,559 0
48 0,606 0 43 0,862 0
56 0,616 0 55 0,903 0
&3 0,602 0 66 0,864 0
76 0,65 0 74 0,915 0
86 0,618 0 88 0,945 0
99 0,608 31,043 97 0,916 23,3719
10 7 0,64 4 0 10 7 0,963 0
11 2 0,784 0 11 1 0,700 0
12 5 0,638 0 12 3 0,939 0
13 5 0,603 0 13 7 0,436 0
14 5 0,714 0 14 6 0,608 0
15 1 0,610 0 15 7 0,903 0
1 5 0,664 0 16 9 0,956 0
& 17 2 0,719 0 17 3 0,608 0
18 8 0,602 0 18 3 0,98 0
19 1 0,624 0 19 4 0,989 0
20 2 0,85 0 20 6 0,868 0
21 4 0,681 0 21 7 0,986 0
22 2 0,591 0 22 9 0,814 0
23 6 0,697 0 23 6 0,577 0
24 3 0,565 0 24 6 0,951 0
25 5 0,60 0 25 9 0,891 0
26 6 0,832 0 26 2 0,778 0
27 1 0,624 0 27 5 0,889 0
28 1 0,611 0 28 8 0,436 0
29 3 0,595 0 29 7 0,839 0
30 6 0,706 0 30 6 0,592 0
31 8 0,611 0 31 4 0,95 0
32 9 0,593 0 32 1 0,954 0
33 7 0,673 0 33 8 0,946 0
34 8 0,654 0 34 8 0,959 0
35 4 0,715 0 35 6 0,601 0
36 6 0,619 0 36 1 0,873 0
37 4 0,621 0 37 1 0,944 0
38 2 0,615 0 38 8 0,950 0
39 1 0,659 0 39 8 0,945 0
40 9 0,698 0 40 9 0,569 0
Total 3 26,1661 Tota 4 32,9555 23,3719
l

Table 3.28. Leave-One-Out: total bets and winnings for each type, with β = 0.4 and β = 0.6

14
2
 CHAPTER 3. Results

• β = 0,4
Horse1 Horse2 Horse3 Horse4 Horse5 Horse lo6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse lo12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,6034 0 0 0
to 2 0 0 0 0 0 0 0 0 0 0,7346
to 3 0 0 0 0 0 0 0 0 0,6868 0
to 4 0 0 0 0 0 0 0 0 0 0,6066 0 0 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6163
to 6 0 0 0 0 0 0 0 0 0 0 0 0,6026 0 0
at 7 0 0 0 0 0 0 0 0 0 0 0 0 0,656
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,6189 0
at 9 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6087
to
10 0 0 0 0 0 0 0 0 0 0,642
at 0 0 0 0 0 0 0 0 0 0,7845
11
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,6385 0
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,6035 0
at 0 0 0 0 0 0 0 0 0 0 0,7141
14
at
15 0 0 0 0 0 0 0 0,6105 0 0 0 0 0 0
at
16 0 0 0 0 0 0 0 0 0 0 0 0 0,6642
at 0 0 0 0 0 0 0 0 0 0 0,7198 0 0
17
at
18 0 0 0 0 0 0 0 0 0 0,6021
at
19 0 0 0 0 0 0 0 0 0 0,6242
to 0,584 0 0 0 0 0 0 0 0 0
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,6812 0
at
22 0 0 0 0 0 0 0 0 0,5916 0 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0,6973 0 0
23
at
24 0 0 0 0 0 0 0,5655 0 0 0 0 0 0 0
at
25 0 0,606 0 0 0 0 0 0 0 0 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0,8321
26
at
27 0 0 0 0 0 0 0 0,6241 0 0 0 0 0 0
at
28 0 0 0 0 0 0 0 0 0 0,6113 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,5956 0
29
at 0 0 0 0 0 0 0 0,7068 0 0
at
3031 0 0 0 0 0 0 0 0 0 0,6119 0 0 0 0
at
32 0 0 0 0 0 0 0 0 0,5937 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6738
33
at
34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6544
at
35 0 0 0 0 0 0 0 0 0,7156 0
at 0 0 0 0 0 0 0 0 0 0,6194
36
at
37 0 0 0 0 0 0 0 0 0 0 0 0 0 0,6212
at
38 0 0 0 0 0 0 0 0 0 0,6151 0 0 0 0
at 0 0 0 0 0 0 0 0 0 0 0,6599 0 0 0
39
at
40 0 0 0 0 0 0 0 0 0,6983

Table 3.29. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.4

14
3
CHAPTER 3. Results

• β = 0,6
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,4393 0,499 0 0
to 2 0 0 0 0 0 0 0 0 0 0,6335
to 3 0 0 0 0 0 0 0 0 0,5593 0
to 4 0 0 0 0 0 0 0 0 0 0,4489 0 0,4135 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0,4434 0 0,4602
to 6 0 0 0 0 0 0 0 0 0 0 0 0,4381 0 0,4263
at 7 0 0 0 0 0 0 0 0,398 0 0 0 0 0,5179
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,4753 0,4704
at 9 0 0 0 0 0 0 0 0 0 0 0 0,4584 0 0,4583
to
10 0 0 0,4452 0 0 0 0 0 0 0,5179
at
11 0 0 0 0 0 0 0 0 0 0,7003
at 12 0 0 0 0 0 0 0 0 0 0 0 0 0,4983 0,4414
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,4366 0
at
14 0 0 0 0 0 0 0 0 0 0 0,6087
at
15 0 0 0 0 0 0 0 0,47 0 0 0 0 0 0,4339
at 16 0 0 0 0 0 0 0 0 0 0 0,4235 0 0,5328
at
17 0 0 0 0 0 0 0 0 0 0 0,6083 0 0
at
18 0 0 0 0 0 0,5352 0 0 0 0,4488
at
19 0 0 0,4965 0 0 0 0 0 0 0,4932
to 0,4134 0 0 0 0 0 0 0 0 0,4553
20
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,5598 0,4271
at
22 0 0 0 0 0 0 0 0 0,4312 0 0 0 0 0,3833 0 0
at
23 0 0 0 0 0 0 0 0 0,5776 0 0
at 24 0 0 0 0 0 0 0,4007 0 0 0 0 0 0,5512 0
at
25 0 0,4511 0 0 0 0 0 0 0 0 0 0 0 0,4401
at 26 0 0 0 0 0 0 0 0 0 0,7785
at
27 0 0 0 0 0 0 0 0,4694 0 0 0 0,4204 0 0
at 28 0 0 0 0 0 0 0 0 0 0,4367 0 0 0 0
at
29 0 0 0 0,4053 0 0 0 0 0 0 0,4343 0
at
30 0 0 0 0 0 0 0 0,5924 0 0
at
31 0 0 0 0 0 0 0 0 0 0,4549 0 0 0 0,4961
at 32 0 0 0 0 0 0 0 0 0,4212 0 0,5336 0
at
33 0 0 0 0 0 0 0,3857 0 0 0 0 0 0 0 0,5611
at
34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,4441 0,5156
at
35 0 0 0 0 0 0 0 0 0,6011 0
at 36 0 0 0 0 0 0 0,4134 0 0 0,4597
at
37 0 0 0 0 0 0 0 0 0 0 0,4609 0 0 0,4838
at
38 0 0 0 0 0 0 0 0 0 0,4544 0,4965 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,5304 0 0,4154
at
40 0 0 0 0 0 0 0 0 0,5694

Table 3.30. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.6

14
4
 CHAPTER 3. Results

β = 0,7
Total
Guy Revenue
bets
1 0,776 0
29 0,58 0
33 0,882 0
44 0,683 0
51 0,733 0
&9 0,682 0
77 0,745 0
81 0,791 0
98 0,750 1.9,2163
1 9 0,793 0
0 1 9 0,98 0
1 1 5 0,780 0
2 1 8 0,873 0
3 1 3 0,938 0
4 1 4 0,73 0
5 1 9 0,799 0
& 1 3 0,935 0
7 1 9 0,812 0
8 1 5 0,830 0
9 2 2 0,684 0
0 2 8 0,833 0
1 2 5 0,971 0
2 2 7 0,922 0
3 2 7 0,78 0
4 2 9 0,722 0
5 2 2 0,751 0
6 2 7 0,716 0
7 2 3 0,857 0
8 2 9 0,657 0
9 3 3 0,88 0
0 3 4 0,789 0
1 3 5 0,788 0
2 3 9 0,877 0
3 3 5 0,800 0
4 3 5 0,922 0
5 3 2 0,691 0
6 3 1 0,794 0
7 3 3 0,786 0
8 3 5 0,785 0
9 4 3 0,957 0
0Tota 6 32,102 19,2163
l 6
β = 0,8

Table 3.31 . Leave-One-Out: total bets and winnings for each type, with β = 0.7 and β = 0.8

14
5
CHAPTER 3. Results

Guy Revenue
Total bets
1 0,8862 0
2 0,8985 0
3 0,7283 0
4 0,76 6,5019
5 8 0,8282 0
6 0,8259 0
7 0,8693 5,864
8 0,8721 0
9 0,8247 15,3004
10 0,9054 0
11 0,8381 0
12 0,91 0
13 5 0,9631 0
14 0,7971 0
15 0,8566 0
16 0,9781 0

14
6
 CHAPTER 3. Results

17 0,7921 0
18 0,9654 0
19 0,9697 0
20 0,8774 0
21 0,9174 0
22 0,6915 0
23 0,7801 0
24 0,87 0
25 8 0,80 0
26 9 0,9644 0
27 0,8232 0
28 0,9314 9,1194
29 0,7494 0
30 0,7376 0
31 0,9206 0
32 0,9539 0
33 0,9546 0
34 0,8876 0
35 0,77 0
36 1 0,8245 0
37 0,9576 15,5275
38 0,9597 0
39 0,9159 0
40 0,8156 0
Total 34,602 52,3132
2

Table 3.31 . Leave-One-Out: total bets and winnings for each type, with β = 0.7 and β = 0.8

14
7
CHAPTER 3. Results

• β = 0,7
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,354 0,4229 0 0
to 2 0 0 0 0 0 0 0 0 0 0,583
to 3 0 0 0 0 0 0 0 0 0,5069 0,3755
to 4 0 0 0 0 0 0 0 0 0 0,3626 0 0,3205 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0,3564 0 0,3775
to 6 0 0 0 0 0 0 0 0 0 0 0 0,3497 0 0,333
at 7 0 0 0 0 0 0 0 0,3008 0 0 0 0 0,4443
at 8 0 0 0 0 0 0 0 0 0 0 0 0 0,3987 0,393
at 9 0 0 0 0 0 0 0 0 0 0 0 0,3741 0 0,3768
to 0 0 0,3545 0 0 0 0 0 0 0,4394
10
at
11 0 0 0 0 0 0 0,3136 0 0 0,6714
at
12 0 0 0 0 0 0 0 0 0 0 0 0 0,4236 0,3572
at
13 0 0 0 0 0 0 0 0 0 0 0 0 0,3583 0,515
at 14 0 0 0 0 0 0 0 0 0,3714 0 0,5669
at 1
0 0 0 0 0 0 0 0,3906 0 0 0 0 0 0,3484
5
at
16 0 0 0 0 0 0 0 0 0 0 0,3351 0 0,4642
at
17 0 0 0 0 0 0 0 0 0 0 0,5601 0,3758 0
at 18 0 0 0 0 0 0,4561 0 0 0 0,3564
at
19 0 0 0,4163 0 0 0 0 0 0 0,414
to
20 0,32 0 0 0 0 0 0 0 0 0,3648
at
21 0 0 0 0 0 0 0 0 0 0 0 0 0,4945 0,339
at 22 0 0 0 0 0 0 0 0 0,352 0 0 0,3235 0 0,2962 0 0
at
23 0 0 0 0 0 0 0 0 0,5257 0,397 0
at
24 0 0 0 0 0 0 0,3878 0 0 0 0 0 0,4812 0
at 25 0 0,3675 0 0 0 0 0 0 0 0 0 0 0 0,3547
at 26 0 0 0 0 0 0 0 0 0 0,7517
at
27 0 0 0 0 0 0 0 0,3874 0 0 0 0,3289 0 0
at
28 0 0 0 0 0 0 0 0 0 0,3569 0 0 0 0,501
at
29 0 0 0 0,3095 0 0 0 0 0 0 0,3478 0
at 30 0 0 0 0 0 0 0 0,541 0 0,343
at
31 0 0 0 0 0 0 0 0 0 0,3715 0 0 0 0,418
at 32 0 0 0 0 0 0 0 0 0,3301 0 0,4588 0
at
33 0 0 0 0 0 0 0 0 0,3768 0 0 0 0 0 0,5007
at 34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0,3574 0,4431
at 3
0 0 0 0 0 0 0,3658 0 0,5564 0
5
at
36 0 0 0 0 0 0 0,3175 0 0 0,3736
at
37 0 0 0 0 0 0 0 0 0 0 0,3838 0 0 0,4105
at 38 0 0 0 0 0 0 0 0 0 0,3691 0,4174 0 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,4597 0 0,3256 0
at
40 0 0 0 0 0 0 0 0,4462 0,5114

Table 3.32. Leave-One-Out : bets we make for each horse, depending on the type, with β = 0.7

14
8
 CHAPTER 3. Results

• β = 0,8
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0 0,2714 0,3509 0 0,264
to 2 0 0 0 0 0 0 0 0 0,3607 0,5379
to 3 0 0 0 0 0 0 0 0 0,4405 0,2878
to 4 0 0 0 0 0 0 0 0 0 0,2827 0,2501 0,2353 0
to 5 0 0 0 0 0 0 0 0 0 0 0 0,2758 0,2536 0,2987
to 6 0 0 0 0 0 0 0 0 0 0,306 0 0,2677 0 0,2522
at 7 0 0 0 0 0 0,2792 0 0,2149 0 0 0 0 0,3752
at 8 0 0 0 0 0 0 0 0 0 0 0 0,2272 0,3257 0,3192
at 9 0 0 0 0 0 0 0 0 0 0 0 0,299 0,2257 0,3
to
10 0 0 0,2779 0 0 0 0 0 0,2526 0,3749
at
11 0 0 0 0 0 0 0,213 0 0 0,6251
at 12 0 0 0 0 0 0 0 0 0,2843 0 0 0 0,3533 0,2774
at
13 0 0 0 0 0 0 0,2299 0 0 0 0 0 0,2766 0,4566
at
14 0 0 0 0 0 0 0 0 0,2854 0 0,5117
at
15 0 0 0 0 0 0 0 0,3195 0,2658 0 0 0 0 0,2713
at 16 0 0 0 0 0 0 0 0 0 0 0,2534 0,326 0,3988
at
17 0 0 0 0 0 0 0 0 0 0 0,5026 0,2894 0
at
18 0 0 0 0 0 0,3932 0,2941 0 0 0,2781
at 19 0 0 0,351 0 0 0 0,272 0 0 0,3467
to 0,236 0 0 0 0 0 0,3483 0 0 0,2931
20
at
21 0 0 0 0 0 0 0 0 0,2291 0 0 0 0,4328 0,2555
at
22 0 0 0 0 0 0 0 0 0,2634 0 0 0,2286 0 0,1995 0 0
at
23 0 0 0 0 0 0 0 0 0,4642 0,3159 0
at 24 0 0 0 0 0 0,2334 0,2224 0 0 0 0 0 0,4222 0
at
25 0 0,289 0 0 0 0 0 0 0 0 0 0,2458 0 0,2743
at 26 0 0 0 0 0 0 0,2355 0 0 0,7289
at
27 0 0 0 0 0 0 0 0,3097 0 0 0 0,2442 0,2693 0
at 28 0 0 0 0 0 0 0 0 0,2224 0,272 0 0 0 0,4371
at 29 0 0 0 0,2268 0,256 0 0 0 0 0 0,2666 0
at
30 0 0 0 0 0 0 0 0,4823 0 0,2553
at
31 0 0 0 0 0 0 0 0 0,2811 0,2924 0 0 0 0,3471
at 32 0 0 0 0 0 0 0 0 0,2436 0 0,3937 0,3166
at
33 0 0 0 0 0 0 0,2093 0 0,302 0 0 0 0 0 0,4432
at
34 0 0 0 0 0 0 0 0,2377 0 0 0 0 0 0 0,2768 0,373
at
35 0 0 0 0 0 0 0,275 0 0,496 0
at 36 0 0 0 0 0 0 0,2321 0,2984 0 0,2939
at
37 0 0 0 0 0 0 0 0 0 0,3045 0,3113 0 0 0,3418
at
38 0 0 0 0 0 0 0 0 0 0,2888 0,345 0,3259 0 0
at
39 0 0 0 0 0 0 0 0 0 0 0,3948 0,2797 0,2415 0
at
40 0 0 0 0 0 0 0 0,3698 0,4458

Table 3.33. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.8

14
9
CHAPTER 3. Results

β = 0,9
Total
Guy Revenue
bets
1 0,936 0
2 2 0,918 0
1
3 0,777 0
4 6 0,90 4,4103
5 3 0,981 0
& 1 0,944 0
7 2 0,816 4,0797
8 4 0,968 0
9 2 0,95 11,4369
10 1 0,802 0
11 2 0,91 0
12 5 0,939 0
4
13 0,909 0
14 9 0,891 0
15 1 0,894 0
1 4 0,93 0
& 17 7 0,913 0
6
18 0,911 0
19 9 0,95 0
20 7 0,835 0
21 8 0,84 0
22 7 0,913 0
7
23 0,933 0
24 1 0,769 0
25 4 0,911 0
26 4 0,844 0
27 1 0,821 0
2
28 0,919 5,3296
29 8 0,761 0
30 3 0,808 0
31 3 0,864 0
32 5 0,962 0
7
33 0,940 0
34 1 0,949 0
35 5 0,831 0
36 9 0,959 0
37 6 0,724 11,431
38 2 0,940 2 0
39 2 0,873 0
40 7 0,996 0
9
Total 35,6757 36,6877
β=1
Table 3.34. Leave-One-Out: total bets and winnings for each type, with β = 0.9 and β = 1

15
0
CHAPTER 3. Results

Total
Guy Revenue
bets
1 0,928 0
21 0,813 0
38 0,941 0,310
48 0,929 7 2,210
57 0,795 2 0,230
66 0,678 7 0,615
78 0,698 2,230
88 0,775 7 0,676
98 0,672 1 7,168
1 7 0,772 9 0,011
0 1 1 0,844 2 0,940
1 1 5 0,85 8 0
2 1 8 0,90 0,050
3 1 5 0,65 6 0
4 1 0,611 0,113
5 1 5 0,94 0,976
6 2 4

Table 3.34. Leave-One-Out: total bets and winnings for each type, with β = 0.9 and β = 1

15
1
CHAPTER 3. Results

1 0,749 0
7 1 5 0,929 0,740
8 1 8 0,94 5 0,511
9 2 3 0,954 4 0
0 2 1 0,957 0,236
1 2 6 0,473 7 0
2 2 8 0,960 0
3 2 5 0,751 0
4 2 2 0,877 0
5 2 2 0,917 1,049
6 2 8 0,835 1 0,204
7 2 8 0,869 8 1,501
8 2 7 0,826 4 0,423
9 3 9 0,930 0,054
0 3 9 0,983 5 0
1 3 1 0,778 0
2 3 9 0,824 0,511
3 3 1 0,756 1 0,144
4 3 4 0,748 7 0,088
5 3 1 0,692 5 0,797
6 3 1 0,974 7,818
7 3 4 0,91 3 0
8 3 1 0,982 0,556
9 4 4 0,90 8 0
0Total 8 32,5981 30,1721

Table 3.34. Leave-One-Out: total bets and winnings for each type, with β = 0.9 and β = 1

15
2
 CHAPTER 3. Results

• β = 0,9
Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0 0 0 0 0 0 0 0 0 0 0,2896 0,1882 0,2779 0 0,1805
to 2 0 0 0 0 0 0 0,1402 0 0,2895 0,4884
to 3 0 0 0 0,1877 0 0 0 0 0,3801 0,2098
to 4 0 0 0,196 0 0 0 0 0,1819 0 0,204 0,1696 0,1515 0
to 5 0 0 0 0 0 0 0,1838 0 0 0,2072 0 0,1967 0,1726 0,2208
to 6 0 0,1888 0 0 0,166 0 0 0 0 0,2338 0 0,1852 0 0,1704
at 7 0 0 0 0 0 0,1943 0 0,1222 0 0 0,1965 0 0,3034
at 8 0 0 0,1493 0 0 0 0 0 0 0 0,1768 0,1437 0,2529 0,2455
at 9 0 0 0 0 0,1832 0 0 0 0 0 0,1742 0,2259 0,1434 0,2243
to
10 0 0 0,1906 0 0 0 0,3119 0 0 0,2997
at
11 0 0 0 0 0 0 0,1264 0 0,2008 0,5878
at 12 0 0 0 0 0 0 0,1438 0 0,204 0 0,1143 0 0,2813 0,196
at
13 0 0 0 0 0 0 0,1402 0 0 0 0,1808 0 0,1934 0,3955
at
14 0 0 0 0 0,2167 0 0 0 0,2112 0 0,4632
at 15 0 0 0,147 0 0 0 0 0,246 0,1861 0,1235 0 0 0 0,1918
at 16 0 0 0 0 0 0 0 0 0,1896 0 0,1673 0,2488 0,3313
at
17 0 0 0 0 0 0 0 0,131 0 0 0,4504 0,2133 0,1189
at
18 0 0 0 0 0 0,32 0,2085 0,1925 0 0,1909
at
19 0 0 0,2781 0 0 0 0,1892 0,2163 0 0,2734
to 0,1461 0 0 0 0 0 0,2713 0,2086 0 0,2098
20
at
21 0 0 0 0 0 0 0 0 0,1393 0 0,1699 0 0,3687 0,1691
at
22 0 0 0 0 0 0 0 0 0,1856 0,2449 0 0,1485 0 0,1138 0 0,2209
at
23 0 0 0,1404 0 0 0 0 0 0,4085 0,2434 0,1408
at 24 0 0 0 0 0 0,1431 0,1311 0 0 0 0 0,1407 0,3545 0
at 2
0 0,2107 0 0 0 0 0 0,216 0 0,1273 0 0,1632 0 0,1942
5
at
26 0 0 0 0 0 0 0,1436 0 0 0,7005
at
27 0 0 0 0 0 0 0 0,2303 0 0,2489 0 0,1568 0,1852 0
at 28 0 0 0 0 0 0 0 0 0,13 0,1857 0,2326 0 0 0,3715
at
29 0 0 0 0,1403 0,1736 0 0 0 0 0,1302 0,1838 0,1334
at
30 0 0 0 0 0 0,2086 0 0,4272 0 0,1725
at
31 0 0 0 0 0 0 0 0 0,1973 0,2106 0,1849 0 0 0,2717
at 32 0 0 0 0 0 0 0 0,2464 0,155 0 0,324 0,2373
at
33 0 0 0 0 0 0 0,1208 0 0,225 0 0 0 0 0,2104 0,3839
at
34 0 0 0 0 0 0 0 0,1534 0,1483 0 0,1495 0 0 0 0,1956 0,3027
at 35 0 0 0 0 0 0 0,1961 0 0,4427 0,1931
at 36 0 0 0 0,1752 0 0,2004 0,1468 0,223 0 0,2142
at
37 0 0 0 0 0 0 0 0 0 0,2241 0,2329 0 0 0,2672
at
38 0 0 0 0 0 0 0 0 0 0,2056 0,2688 0,2473 0 0,2185
at
39 0 0 0 0 0 0 0 0 0 0 0,3264 0,1968 0,1539 0,1966
at
40 0 0 0,1763 0 0 0,1342 0 0,3016 0,3848

Table 3.35. Leave-One-Out: bets we make for each horse, depending on the type, with β = 0.9

15
3
CHAPTER 3. Results

Horse1 Horse2 Horse3 Horse4 Horse5 Horse6 Horse7 Horse8 Horse9 Horse10 Horse11 Horse12 Horse13 Horse14 Horse15 Horse16
to 1 0,0103 0 0 0,0412 0,008 0,0025 0 0,0369 0,0894 0,0362 0,2171 0,1045 0,2041 0,0822 0,0958
to 2 0 0 0,0181 0,0056 0 0,0509 0,047 0,0405 0,2146 0,437
to 3 0 0,0565 0,0147 0,1148 0,0465 0,0643 0,1607 0,0241 0,323 0,1372
to 4 0 0 0,1147 0,1232 0,0384 0,021 0,0678 0,099 0,0292 0,1227 0,085 0,0645 0,1641
to 5 0 0 0,0355 0,0234 0 0,0179 0,0908 0,0469 0,134 0,1184 0 0,1093 0,0816 0,1378
to 6 0,0946 0,087 0 0 0,0616 0,0276 0,0504 0,0159 0 0,1391 0,022 0,0952 0,0116 0,0738
at 7 0 0 0 0,0502 0 0,1062 0,0765 0,027 0,029 0,0043 0,1068 0,0681 0,2307
at 8 0 0 0,0545 0,0425 0,0615 0,0158 0,084 0,0144 0 0,0229 0,085 0,0517 0,1757 0,1676
at 9 0,0061 0,0282 0 0,0285 0,0873 0,0288 0,0119 0,0034 0,0777 0 0,0773 0,1373 0,0456 0,1406
to
10 0 0,0014 0,1069 0,0462 0 0,0255 0,2414 0,0474 0,0752 0,2282
at
11 0 0,0132 0,0228 0,036 0 0 0,0305 0,0855 0,112 0,5446
at 12 0,0044 0 0 0 0,0139 0,0085 0,0539 0,087 0,1212 0,0677 0,0215 0,1592 0,2078 0,1129
at
13 0 0,0113 0,031 0,0115 0,0564 0 0,0517 0,0364 0,0314 0,0092 0,0969 0,1232 0,1107 0,3354
at
14 0 0 0 0 0,1214 0 0 0,0033 0,1202 0 0,4051
at
15 0,0164 0 0,0482 0 0,019 0 0,1036 0,1632 0,0941 0,0246 0 0,0019 0,0378 0,1029
at 16 0 0 0,0084 0,0236 0,0066 0,0498 0,0769 0,0856 0,107 0,0651 0,0821 0,1727 0,2642
at
17 0 0 0 0,0218 0,0198 0,002 0,1006 0,0333 0,0334 0 0,3925 0,1262 0,0199
at
18 0 0,0741 0,113 0,0484 0,0822 0,2504 0,1264 0,1087 0,0198 0,1067
at
19 0,0823 0,0639 0,2051 0 0,0439 0,0791 0,1063 0,1363 0,026 0,2
to 20 0,0603 0 0,0674 0,1053 0,0293 0,1691 0,2051 0,1368 0,0464 0,1344
at
21 0 0 0,0415 0,0814 0,0123 0,0315 0,0215 0,0653 0,0531 0,0286 0,0876 0,1412 0,3074 0,0861
at
22 0 0 0 0,0035 0,0417 0,0073 0 0 0,0934 0,1434 0,0057 0,049 0 0,0136 0,0011 0,1151
at
23 0,018 0,0617 0,0538 0 0,0251 0,0965 0,0629 0,0706 0,3504 0,1673 0,0542
at 24 0 0,0019 0,0479 0,0343 0,0295 0,0558 0,0425 0,0147 0,1085 0,0114 0,0075 0,0532 0,2908 0,0531
at
25 0,0837 0,1298 0,033 0,0378 0 0 0,0512 0,1347 0,1253 0,0366 0,0108 0,0767 0,0462 0,1114
at 26 0 0 0,0099 0,0807 0,0062 0,0011 0,0581 0,0289 0,0575 0,6754
at
27 0 0 0,0299 0,0648 0,0273 0,0299 0,0173 0,1512 0 0,1727 0,0805 0,0699 0,1015 0,0908
at 28 0 0,0094 0,0124 0,0444 0 0,0094 0,0394 0,0174 0,0366 0,099 0,15 0,0286 0,118 0,3051
at
29 0,107 0 0,2248 0,0563 0,0943 0,046 0,0564 0,0427 0 0,0469 0,1021 0,0504
at
30 0 0,0404 0,0444 0,0078 0,08 0,1325 0,0426 0,373 0,1195 0,0907
at
31 0,053 0 0,0403 0,0334 0,0212 0,1103 0,0371 0,0108 0,1179 0,1313 0,1048 0,0307 0,0917 0,2005
at 32 0,0117 0 0 0,0147 0,0013 0,0046 0,033 0,1591 0,0636 0,0899 0,2487 0,1523
at
33 0 0 0,0341 0,0502 0,028 0,0105 0,0303 0,009 0,1455 0,0063 0,0184 0,0264 0,016 0,1268 0,3226
at
34 0 0,0148 0,0096 0,0233 0,0214 0 0 0,0569 0,0512 0,0893 0,0526 0,0342 0,0414 0,0235 0,1082 0,2299
at
35 0,0134 0 0,0783 0 0 0,0148 0,1091 0,0427 0,3846 0,1053
at 36 0 0,0117 0 0,0764 0,0191 0,1033 0,0481 0,13 0,1771 0,1264
at
37 0 0 0,0198 0,0077 0,0028 0,0073 0,0324 0,1154 0,046 0,1533 0,1617 0,1996 0,0285 0,1999
at
38 0 0 0 0,028 0,0167 0,0538 0,0166 0,0177 0,0614 0,1223 0,1925 0,1685 0,097 0,1365
at
39 0,0196 0 0,0795 0,0571 0,0304 0,0721 0,0023 0,0837 0,0106 0,0611 0,261 0,1175 0,0694 0,118
at
40 0,131 0 0,0851 0,0937 0 0,0384 0,0132 0,2265 0,32

Table 3.36. Leave-One-Out: bets we make for each horse, depending on the type, with β = 1

As we see, the largest difference between what we bet and what we win occurs
with β = 0.8 . With that value we have a profit of €17.71. Therefore, in this case we
could say that the optimal β value would be 0.8.

Furthermore, unlike in the case of cross validation, where with the value that
brought us the most benefit we only bet on one horse in each race, with β = 0.8, we are
betting on two or three horses, depending on the race.

With β = 0.9 we bet on three or four horses, depending on the race, and we obtain

15
4
 CHAPTER 3. Results

a profit of €1. If we compare it with the €17 of β = 0.8 we see that it is a very low value.
Even so, we would have more chances of getting it right than if β took the value of 0.8.

However, in this case it is not so risky to use the value of 0.8, and we are also
having a very large profit. Therefore, in this case we could conclude that the optimal
value of β would be 0.8 regardless of the type of bettor we are.

15
5
 CHAPTER 4. Conclusions and future lines of work

CHAPTER 4

Conclusions and future lines of

work

In this chapter we are going to have two tasks:

• The first is to draw conclusions from the results shown in the previous
chapter.

• And the second is to give some guidelines on how the Project we are dealing
with could be improved or expanded.

14
7
CHAPTER 4. Conclusions and future lines of work

4.1. Conclusions
As we have seen in the previous chapter, depending on the parameters that we use
to perform the SVM function we will obtain higher or lower profits and even losses in
some races.

As we are using the Kelly Criterion, we see that when we have losses, they are
very small, so that it does not have a very significant impact on profits.

We have been able to see that, if we used cross validation, there would be two

values of β that would be optimal depending on the type of bettor we are. With β = 0 ,
which means that we will only take into account the probabilities calculated by Formstar
, that is, here the SVM does not play any role , we would obtain more benefits. In total
we would have a profit of €49.8, however, we only bet on one horse in a race, which is

too risky. Therefore, from our point of view it would be better to use the value of β=
0.9 , since it also brings us benefits, although they are significantly lower, and also, by
betting on a greater number of horses, we have more chances of being right. the winner.

With β = 0.8 and using Leave-One-Out validation we obtain a profit of €57.71,

which is the highest profit we have obtained from all the tests carried out. Furthermore,
in this case we are betting on a higher number of horses per race, so we could conclude
that it would not be so risky to use this value.

Furthermore, we point out that between the two methods used, the optimal thing is

to use Leave-One-Out, since with this method and β = 0.8 we maximize the benefit,
which is our goal. This is because with this method we are using more training samples
and, as we have said, the more runs we use to train the machine, the better results we will
achieve.

Undoubtedly, the benefit depends on the bankroll we have. By betting €1 per race
we obtain a total profit of €57.71. If instead of €1, we used €10 per race, the profit would
be €577.1.

14
8
 CHAPTER 4. Conclusions and future lines of work

4.2. Future lines of work

As improvements for a future Project based on this, we can propose:

• Perform the experiments with a larger database. If we can train our program
with a greater number of races, it will be more adjusted to the results so that
with the new races we can obtain greater benefits. Furthermore, it is
necessary since the results of LOO and CV should not be very different,
which they are in our results, therefore we need more data to be able to say
that the model works.

• We may also take other information into account, such as the rider's or
trainer's position in the rankings.

• In our case we have studied the races that have in common that they are run
on a 1400 meter racecourse. We could also have made the distinction based
on the type of terrain on which the race is held. And even if we have a very
large database, we could combine the length of the racecourse with the type
of terrain.

• In this case we have carried out the analysis with SVM, another option would
be to use Gaussian Processes [70].

• We have handled the RBF Kernel type, however there are more possibilities
such as using a linear, quadratic or polynomial Kernel .

Furthermore, to avoid repetition, we are going to indicate a line of work that we

tried to follow, and that was not correct. To have a greater number of training runs, we
decided to use all races that were between 1200 meters and 1600 meters. The results they
gave us were not satisfactory, since we were not making profits. With this we have
realized that it is not possible to treat horses that run on racecourses with different
lengths in the same way.

14
9
 CHAPTER 5. Budget

CHAPTER 5

Budget

In this chapter we are going to address the budget of this Final Year Project. To do
this, we are going to show the different phases that have led to it becoming a reality, in
addition to the cost of material and personnel associated with the implementation of this
system.

Finally we will make a summary of it, using the template provided by the Carlos
III University.

5.1. Introduction
We can divide the work phases into:

• Phase 1. Analysis of information and study of the different options available.

• Phase 2. Creation of the database.
• Phase 3. Model development.
• Phase 4. Obtaining results.

15
1
CHAPTER 5. Budget

• Phase 4. Documentation.

The total time has been 13 months, with a variable time distribution between the
different phases followed for the realization. A continuous effort has been involved,
although the load line has been irregular throughout the entire process. From the summer
period until the end of the Project, the time spent has been greater than in school periods.
We have followed the following distribution of hours between phases:

• Phase 1. 90 days * 2 hours/day = 180 hours

• Phase 2. 140 days * 2 hours/day = 280 hours
• Phase 3. 50 days * 3 hours/day = 150 hours
• Phase 4. 45 days * 3 hours/day = 135 hours
• Phase 5. 65 days * 3 hours/day = 195 hours

Below we are going to make a breakdown of the different costs associated with the
Project, to facilitate the subsequent reading of the template provided by the Carlos III
University of Madrid.

5.2. Personnel cost

The staff will be different depending on the phases we have carried out during the
Project. Except in Phase 2 and Phase 3, which in both cases required a programmer.
Therefore, to calculate the cost associated with the programmer, we are going to add the
hours of the two phases mentioned above.

15
2
 CHAPTER 5. Budget

Position No. hours Hourly cost Total (€)

Analyst 180 30 € 5.400 €
Programmer 430 25 € 10.750 €
Testing technician 135 22 € 2.970 €
Documentation manager 195 15 € 2.925 €
22.045 €
Table 5.1. Personnel cost

5.3. Software cost

In this section we will include the cost of the applications and tools used during the
completion of the Project, both for programming and for its documentation.

Description Cost (€)

Microsoft Office 2007 290 €
Professional
Matlab 2008 License 300 €
590 €
Table 5.2. Software cost

5.4. Hardware cost

In this case we will include all the hardware that has been necessary to carry out
the development of the Project. All equipment used was the property of the author.

Description Cost (€) % dedicated Time (months) Depreciation Total cost (€)
usage
HP PC 800 € 100% 13 60 173,33 €
HP Mouse and Keyboard 40 € 100% 13 60 8,67 €
Samsung TFT Monitor 150 € 100% 13 60 32,50 €
Verbatim 4 GB Pendrive 12 € 100% 13 60 2,60 €
HP Deskjet 3050 Printer 89 € 100% 1 60 1,48 €
218,58 €
Table 5.3. Hardware cost

15
3
CHAPTER 5. Budget

5.5. Expendable material cost

These expenses include the consumption of printer ink made to print
documentation in Phase 1 and Phase 5. In addition, we also include office supplies such
as pens or notebooks.

Description Cost (€)

Ink cartridges 100 €
Office supplies 15 €
115 €
Table 5.4. Expendable material cost
5.6. Cost summary
We summarize the expenses calculated above, including the 20% indirect costs as
indicated in the budget template.

Description Total costs (€)

Staff 22.045 €
Software 590 €
Hardware Amortization 219 €
Expendable material 115 €
Indirect costs (20%) 4.594 €
27.562 €
Table 5.5. Cost summary

Apart from this final result, we have included a second table adding a cost of 18%
in value added taxes (VAT).

Description Total costs (€)

Total without VAT 27.562 €
Total with VAT (18%) 4.961 €
32.523 €
Table 5.6. Total budget without and with VAT

The total budget of this Project amounts to €32,523.

The design engineer,

Signed Miriam Trujillo Sevillano

15
4
 CHAPTER 5. Budget

CARLOS III UNIVERSITY OF MADRID Higher Polytechnic School

PROJECT BUDGET

1 .- Author:
Miriam Trujillo Sevillano
2 .- Department:
Signal theory
3 .- Project description:
- Qualification Predicting results of sporting events: Horse racing
- Duration (months) 13
Indirect cost rate: 20%

4 .- Total budget of the Project (values in Euros):

32,523.00 Euros

5.- Budget breakdown (direct costs)

STAFF

NIF (do not fill out -

Dedication (men Cost man
Last name and name for information Category Cost (Euro) Sign in accordance
table ) month
only)
Miriam Trujillo Sevillano Analyst 1,37 3.941,60 5.400,00
Miriam Trujillo Sevillano Programmer 3,28 3.277,44 10.750,00
Miriam Trujillo Sevillano Testing technician 1,03 2.883,49 2.970,00
Miriam Trujillo Sevillano Documentation 1,49 1.963,09 2.925,00
0,00
Men month 7.17 Total 22.045,00
1075.5 hours
1 Man month = 131.25 hours. Maximum annual dedication of 12 men per month (1575 hours)
a)

Annual maximum for PDI of the Carlos III University of Madrid of 8.8 man months (1,155 hours)

EQUIPMENT

Depreciation
% Dedicated project Attributable
Description Cost (Euro) Dedication (months) period
usage cost d)

HP PC 800 € 100 13 6 173,33

Mouse and keyboard HP TFT Monitor 40 € 100 13 0 6
Samsung Pendrive Verbatim 4 Gb HP Deskjet 150 € 100 13 0 8,67
3050 Printer 12 € 100 13 6 32,50
89 € 100 10 2,60
6 1,48
0 0,00
1091 60
Total 218,58
d)
Amortization calculation formula:
TO A = number of months from the billing date in which the equipment is used
B = depreciation period (60 months)
xCxD C = cost of equipment (without VAT)
b D = % of usage dedicated to the project (usually 100%)

OUTSOURCING OF TASKS

Description Company Attributable cost

Total 0,00

OTHER DIRECT COSTS OF THE PROJECT e)

Description Company Attributable costs

Software Cost Expendable Materials 590,00
115,00

Total 705.00 pts above, for example: fungible,

e)
This chapter of expenses includes all expenses not contemplated in the conce

trips and diets, others,…

6.- Cost summary

Budget
Budget Total Costs Total Costs

Staff 22.045
Amortization 219
Subcontracting of tasks Operating costs 0
705
Indirect costs
4.594
Total 27.562

Table 5.7. Budget template

15
5
 References

References

Websites

[1] http://formstar.net/racing/index.php

[2] http://www.apostar.ws/historia.php

[3] http://www.retrofootball.com/blog/204/

[4] http://concursodeapuestas.com/historia-de-las-apuestas-deportivas/

[5] http://www.apuestascaballos.net/glosario-carreras-caballos

[6] http://www.rinconapuestas.com/diccionario/

[7] http://www.apuestascaballos.net/apuestas-caballos/tipos-carrera-categoria-maiden-
handicap-stakes-listed-claimers-sellers.html

[8] http://www.apuestacaballos.com/tipos-de-carreras

[9] http://tienespoker.com/apuestas-hipicas-online.php

[10] http://www.elgalope.com/hipodromo.html

[11] http://www.webapuestas.com/guia-tutorial-apuestas/apuestas- sports/caballos/guia-

15
7
References

apuestas-caballos.html

[12] http://www.apuestascaballos.net/tipos-de-apuestas/apuestas-al-tote-1-totewin-y-
tote-place.html

[13] http://www.webapuestas.com/guia-tutorial-apuestas/apuestas-
sports/caballos/consejos-apuestas-caballos.html

[14]
http://elrecreativo.com/index.php?option=com_content&view=article&catid=110%3An
oticias-del-recreativo&id=10543%3Apais-vasco-el-perfil-del-apostante-varon-de-entre-
25-y -45-year-old-sports-fan&Itemid=849

[15] http://www.baquia.com/posts/el-futbol-y-el-tenis-los-preferidos-de-los-apostantes-
espanoles

[16] http://www.lookandbet.com/2009/03/perfil-del-usuario-de-apuestas-online.html

[17] http://armasdejuego.com/apuestas-deportivas/estrategias-trucos-y-consejos/el-
profile-de-un-apostante.html

[18] http://www.soloapuestasdeportivas.es/%C2%BFcomo-Funcan-las-casas-de-
puestas/7

[19] http://capital2max.com/archives/508

[20] http://es.wikipedia.org/wiki/Corredor_de_apuestas

[21] http://apuestasdecaballos.com.es/apuesta-con-zeturf-primera-casa-de-apuestas-
hipicas-on-line/

[22] http://www.zeturf.com/es/

[23] http://apostarcarrerascaballosfrancia.blog.com.es/2011/12/04/como-apostar-en-las-
cambios-francesas-12254295/

[24] http://www.webapuestas.com/casas-apuestas/betfair/index.html

[25] http://www.betfair.com/es/

15
8
 References

[70] http://es.wikipedia.org/wiki/Validaci%C3%B3n_cruzada

[71] http://es.wikipedia.org/wiki/M%C3%A1quinas_de_vectores_de_soporte

[72] http://www.mathworks.es/help/toolbox/bioinfo/ref/svmtrain.html

[73] http://www.webapuestas.com/guia-tutorial-apuestas/gestion-bankroll/criterio-
kelly.html

Articles

[26] Searching for positive returns at the track: a multinomial logit model for
handicapping horse races. Bolton RN, Chapman, RG, 1986.

[27] Financial Theory and Corporate Policy. Copeland, T.E., Weston, J.F., 1979,

[28] Efficient Capital Markets: A Review of Theory and Empirical Work. Fame, EF,
1970

[29] Securities Market Efficiency in an Arrow-Debreu Market. Rubinstein, M., 1975.

[30] An empirical cross-validation of alternative classification strategies applied to

harness racing data for win bets. Ludlow, L.H.

[31] Estimating the probabilities of the outcomes of a horse race (alternatives to the
Harville formulas). Stern, H.S. [46] Adapting support vector machine methods for horse
oddsrace prediction. Edelman, D., 2007

[32] Racetrack betting and informed behaviour. Peter Asch, Burton G. Malkiel and
Richard E. Quandt, 1982.

[33] Identifying winners of competitive events: A SVM-based classification model for

horserace prediction. Stefan Lessman, Ming-Chien Sung, Johnnie EV Johnson, 2009.

[34] Transactions costs, market inefficiencies and entries in a racetrack betting model.
Hausch, D.B., Ziemba, W.T., 1985

[35] Exploring decision markers' use of price information in a speculative market.

Johnson, J.E.V., Jones, O., Tang, L., 2006

15
9
References

[36] Insider trading, herding behavior and market plungers in the British horse-race
betting market. Law, D., Peel, D.A., 2002

[37] Why are gambling markets organized so differently from financial markets? Levitt,
S.D., 2004

[38] Te economics of wagering markets. Sauer, R.D., 1998

[39] Inside information in a betting market. Schnytzer, A., Shilony, Y., 1995

[40] Information efficiency in betting markets. Vaughan Williams, L., 1999

[41] Computer based horse race handicapping and wagering systems: A report. Benter,
W., 1994

[42] Searching for semi-strong form inefficiency in the UK racetrack betting market.
Sung, MC, Johnson, J.E.V., Bruce, A.C., 2005

[43] Conditional logit analysis of qualitative choice behavior. McFadden, D., 1973

[44] Still searching for positive returns at the track: Empirical results from 2000 Hong
Kong races. Chapman, R.G., 1994

[45] Multinomial Probit Models for Competitive Horse Racing. Gu, M.G., Huang, C.,
Benter, W., 2003

[46] Adapting support vector machine methods for horserace odds prediction. Edelman,
D., 2007

[47] Advances in the mathematical modeling of horse race outcomes. Benter, W., 2003

[48] Comparing the effectiveness of one and two step confitional logit models for
predicting outcomes in a speculative market. Sung, M., Johnson, J.E., V., 2007

[49] The Nature of Statistical Learning Theory. Vapnik, V.N., 1995

[50] An introduction to Support Vector Machines and other Kernel-based Learning

Methods. Cristianini, N., Shawe-Taylor, J., 2000

[51] A tutorial on support vector machines for pattern recognition. Burges, C.J.C., 1998

16
0
 References

[52] Benchmarking least squares support vector machine classifiers. Van Gestel, T.,
Suykens, JAK, Baesens, B., Viane, S., Vanthienen, J., Dedene, G., De Moor, B.,
Vandewalle, J., 2004

[53] Predictions in marketing using the support vector machine. Cui, D., Curry, D.,
2005

[54] Least square supports vector machine classifiers. Suykens, JAK, Vanderwalle, J.,
1999

[55] Subjective information and market efficiency in a betting market. Figlewski, S.,
1979

[56] Investigating the roots of the favorite-longshot bias: An analysis of supply and
demand side agents in parallel betting markets. Bruce, A.C., Johnson, J.E.V., 2000

[57] A tutorial on support vector regression. Smola, A.J., Schölkopf, B., 2004

[58] Neural Networks for Pattern Recognition. Bishop, C. M., 1995

[59] Cross-validation choice and assessment of statistical predictions. Stone, M., 1974

[60] A Practical Guide to Support Vector Classification. Hsu, C.W., Chang, C.C., Lin,
C.J., 2003

[61] A New Interpretation of Information Rate. Kelly, J.L., 1956.

[62] An introduction to ROC analysis. Fawcett, T., 2006

[63] Exploiting rank ordered choice set data within the stochastic utility model.
Chapman, R.G., Staelin, R., 1982

[64] Transferability of disaggregated mode choice models. Watson, P.L., Westin, R.B.,
1975

[65] A comparison of methods for multi-class support vector machines. Hsu, C.W., Lin,
C.J., 2002

[66] Support vector learning for ordinal regression. Herbrich, R., Graepel, T.,
Obermayer, K., 1999

16
1
References

[67] A fast dual algorithm for kernel logistic regression. Keerthi, SS, Duan, KB,
Shevade, SK, Poo, AN, 2005

[68] A support vector method for multivariate performance measures. Joachims, T.,
2005

[69] Comprehensible credit scoring models using rule extraction from support vector
machines. Martens, D., Baesens, B., Van Gestel, T., Vanthienen, J., 2007

[74] c. K. YO. Williams and D. Barber. Bayesian classification with Gaussian

processes, 1998.

16
2