Forecasting Football

Master of Science Final Project
Forecasting
Football
Outcomes to
Invest in Betting
Markets
Business Administration
João Maria Telles Moniz Côrte-Real

2017/2018
Supervisor: Professor José Joaquim Dias Curto, PhD

Associate Professor at ISCTE Business School, Department of
Quantitative Methods for Management and Economics
Master of Science Final Project
Forecasting
Football
Outcomes to
Invest in Betting
Markets
Business Administration
João Maria Telles Moniz Côrte-Real

2017/2018
Supervisor: Professor José Joaquim Dias Curto, PhD

Associate Professor at ISCTE Business School, Department of
Quantitative Methods for Management and Economics
Forecasting Football Outcomes to Invest in Betting Markets
Acknowledgements
To Professor Dias Curto, for the patience and availability
To my father, and to my uncles and aunts from the mother’s side (especially to my
godparents João and Luísa), for supporting me through this journey
To my girlfriend Inês, the love of my life, for the understanding
To my aunts Isabel and Filipa, for loving me as a son
To my stepmother Domingas, for treating me as a son
To my mother, to my grandparents (Leonel, Graça, Inês and Filipe) and to my cousin

Cristina, for looking after me
To my sister and niece, Gracinhas, for the laughs
To my cousin António, for the unconditional brotherhood
To all my best friends, for the good times spent together

Table of Contents
1- Abstract ......................................................................................................................... I
2- Executive summary .................................................................................................... III
3- Introduction .................................................................................................................. 1
3.1- Forms of placing bets ............................................................................................ 2
3.1.1- Fixed-odds betting .......................................................................................... 2
3.1.2- Exchange betting ............................................................................................ 5
3.2- Legal context of the betting markets in Portugal .................................................. 6
3.3- Purpose of the project ............................................................................................ 9
4- Literature review ........................................................................................................ 10
4.1- Statistical models................................................................................................. 10
4.1.1- Poisson regression models ........................................................................... 10
4.1.2- Ordered probit/logit regression models ........................................................ 11
4.2- Fixed-odds betting markets efficiency ................................................................ 13
4.2.1- Arbitrage opportunities ................................................................................ 15
4.3- Home advantage .................................................................................................. 15
4.4- Money management methods.............................................................................. 19
4.4.1- Fixed-stake ................................................................................................... 19
4.4.2- Martingale .................................................................................................... 19
4.4.3- Kelly criterion .............................................................................................. 20
4.5- The importance of subjective factors .................................................................. 20
4.6- Conclusions about the literature review .............................................................. 22
5- Method ....................................................................................................................... 23
5.1- Poisson regression model .................................................................................... 23
5.1.1- Counting process and Poisson process definitions ....................................... 23
5.1.2- Poisson random variable .............................................................................. 24
5.1.3- Dataset .......................................................................................................... 25
5.1.4- Poisson regression ........................................................................................ 32
5.2- Betting strategies ................................................................................................. 39
5.2.1- Value strategy ............................................................................................... 40
5.2.2- Accuracy strategy ......................................................................................... 47
6- Discussion of the results ............................................................................................. 50
6.1- Portuguese online fixed-odds betting markets .................................................... 50
6.2- Betting record ...................................................................................................... 52
6.2.1- Portuguese online fixed-odds betting markets efficiency ............................ 53
6.2.2- The importance of our forecasting method in appraising the expected value
................................................................................................................................ 56
7- Conclusions ................................................................................................................ 59
7.1. Summarizing the value strategy .......................................................................... 59
7.1.1- Limitations and recommendations for further research ............................... 59
7.2- Summarizing the accuracy strategy..................................................................... 59
7.3- Portuguese online fixed-odds and markets efficiency......................................... 60
7.4- Final comment ..................................................................................................... 61
8- References .................................................................................................................. 62
8.1- Books ................................................................................................................... 62
8.2- Periodicals ........................................................................................................... 62
8.3- Unpublished works .............................................................................................. 64
8.4- Websites .............................................................................................................. 64
Annexes .......................................................................................................................... 67
Annex I- Watford vs Everton (match day 28) ............................................................ 67
Annex II- Arsenal vs Manchester City (match day 28) .............................................. 74
Annex III- Manchester City vs Chelsea (match day 29) ............................................ 80
Annex IV- Manchester United vs Liverpool (match day 30) ..................................... 87
Annex V- Liverpool vs Watford (match day 31) ....................................................... 94
Annex VI- Bournemouth vs West Bromwich Albion (match day 31) ..................... 102
Annex VII- Everton vs Manchester City (match day 32) ......................................... 110
Annex VIII- Arsenal vs Southampton (match day 33) ............................................. 118
Annex IX- Arsenal vs West Ham (match day 35) .................................................... 125
Annex X- Tottenham vs Watford (match day 36) .................................................... 133
Annex XI- Southampton vs Manchester City (match day 38) ................................. 141
1- Abstract
Our main objective was to develop a method which can be used to obtain long-term
returns on football bets. The forecasting of the matches was made using quantitative
(including a Poisson regression model) and qualitative (of subjective interpretation)
information. This approach was named value strategy, being able to generate a yield rate
of 19,88%, and to predict 8 out of 11 outcomes. However, the number of bets made is too
small to conclude that this strategy is capable of providing consistent earnings, in the
long-run.
The accuracy strategy consisted in betting on favourites who played at home, and
produced returns of 0,39%, not compensating for the risks assumed. If this strategy was
able to create revenues, it could indicate the presence of biases in the Portuguese betting
markets, and consequently, that these were semi-strong efficient. In practice, the accuracy
strategy broke even, therefore not suggesting that the Portuguese markets’ odds were
biased. However, if we did the same bets through Pinnacle (UK-based bookmaker), the
returns would be of 6,62%, hence indicating the existence of biases and semi-strong
efficiency.
Surprisingly, if one removes the overround, Portuguese bookmakers would have offered
better odds for the home wins than Pinnacle. This demonstrates the need of reviewing the
Portuguese online gambling legal framework, which is the main reason for the poor
competitiveness of Portuguese betting markets. As far as our research went, we could not
find any studies about these markets, hence we consider that our work adds an important
contribution to the existing literature.
Keywords: Betting Markets, Football, Forecasting, Sports Betting
JEL Classification System: Information and Market Efficiency, Sports Economics
O nosso principal objetivo foi desenvolver um método que possa ser usado para obter
retornos de longo prazo em apostas de futebol. A previsão dos jogos foi feita utilizando
informação quantitativa (incluindo um modelo de regressão Poisson) e qualitativa (de
interpretação subjetiva). A esta abordagem chamou-se estratégia de valor, tendo sido
capaz de originar retornos de 19,88%, e de prever 8 em 11 resultados. Contudo, o número
de apostas feitas é insuficiente para considerar que esta estratégia consegue gerar retornos
consistentemente, a longo prazo.
I
A estratégia de precisão consistiu em apostar nos favoritos que joguem em casa, e

conseguiu retornos de 0,39%, não compensando os riscos assumidos. Se esta estratégia
fosse capaz de criar retornos, podia indicar a tendenciosidade dos mercados de apostas
Portugueses, e consequentemente, que estes possuem eficiência semiforte. Na prática, a
estratégia de precisão ficou na mesma, não sugerindo, portanto, que as odds dos mercados
Portugueses são tendenciosas. No entanto, se fizéssemos as mesmas apostas através do
Pinnacle (casas de apostas Britânica), os retornos seriam de 6,62%, indicando a existência
de tendenciosidade e de eficiência semiforte.
Surpreendentemente, se removêssemos a margem implícita, as casas de apostas

Portuguesas teriam oferecido melhores odds do que o Pinnacle, para vitórias da equipa
caseira, o que demostra a necessidade de rever as leis do jogo online em Portugal, que
são a principal razão para a falta de competitividade dos mercados de apostas
Portugueses. Não conseguimos encontrar nenhum estudo acerca desses mercados, assim,
consideramos que o nosso trabalho acrescenta um importante contributo à bibliografia
existente.
Palavras-chave: Mercados de Apostas, Futebol, Projeções, Apostas Desportivas
JEL Classification System: Information and Market Efficiency, Sports Economics
II
2- Executive summary
A ideia de desenvolver este projeto emerge, essencialmente, do fascínio que o seu autor
sempre demonstrou por temas relacionados com o futebol, nomeadamente as apostas
desportivas (especialmente após a leitura do livro “Como ganhar com as apostas
desportivas”, de Paulo Rebelo). Durante o período em que frequentou as aulas do
mestrado em Gestão de Empresas, e apesar de não possuir grandes bases matemáticas,
interessou-se pela cadeira de Métodos Quantitativos, lecionada pelo Professor Dias Curto.
Daí surgiu a ambição de elaborar um método de previsão de resultados desportivos, que
lhe permitisse ganhar dinheiro ao mesmo tempo que desenvolvia as suas capacidades
analíticas.
A força motriz deste projeto foi a criação de um método de avaliação de jogos futuros
entre equipas de futebol, com base em ocorrências passadas, que servisse de suporte a
tomadas de decisão relativas a potenciais investimentos nos mercados de apostas a operar
legalmente em Portugal. Considerou-se extremamente importante que este sistema fosse
facilmente compreendido e aplicável (com maior ou menor sucesso) pela maioria dos
apostadores, o que diferencia este trabalho da maior parte das publicações existentes, que
se focam nas componentes teóricas e académicas, sendo por isso mais direcionadas a um
público de estudiosos e professores dos ramos estatísticos e económicos.
Após efetuar uma análise comparativa dos mercados de apostas nacionais e

internacionais, o autor apercebeu-se de limitações significativas no caso Português, tanto
ao nível dos tipos de mercados existentes, como da proposta de valor implícita nas odds
oferecidas, passando pelo número anormalmente pequeno de casas de apostas a operar
legalmente no nosso país. Posteriormente, apurou-se que estas limitações se devem, em
grande parte, à ineficiência dos decretos de lei que regulam o jogo online e as apostas
desportivas em Portugal, limitações essas que são prejudiciais aos interesses dos 3
principais “players” deste mercado: os apostadores, as casas de apostas e, em última
análise, o próprio Estado Português.
O passo seguinte passou por uma revisão literária extensiva, abrangendo todos os tópicos
que o autor julgou relevantes para ampliar o seu grau de conhecimento acerca deste tema,
facilitando assim a definição dos conteúdos a serem considerados na avaliação dos jogos.
Deste modo, procurou-se perceber se já existiam modelos econométricos que estimassem
probabilidades para os possíveis desfechos de um jogo de futebol, e a existirem, qual seria
III
o mais adequado. Analisaram-se também diversos artigos científicos, no sentido de aferir

o nível de eficiência dos mercados de odds fixas, quanto à informação publicamente
disponível.
Concluiu-se que o modelo de regressão Poisson era o mais indicado, por permitir estimar
probabilidades não só para as hipóteses de desfecho (vitória caseira, empate, ou vitória
forasteira), mas também para o número de golos a ser marcado por cada uma das equipas.
Concluiu-se também que, ao longo dos anos, foram detetadas de forma sistemática odds
tendenciosas nos mercados de apostas, indicando ineficiências potencialmente
exploráveis por parte dos apostadores. Para além disso, nesta fase do projeto, estudaram-
se igualmente outros aspetos que podem influenciar os acontecimentos das partidas, como
por exemplo, a importância da análise das variáveis qualitativas (informação dificilmente
quantificável), que para Rebelo, até é superior à da análise quantitativa.
Com base nas descobertas proporcionadas pela revisão literária, definiu-se um método de
avaliação composto por variáveis quantitativas (nas quais se incluem, para além do
modelo de Poisson, estatísticas descritivas da performance das equipas) e qualitativas (de
interpretação subjetiva, mas bem definidas), no sentido de identificar os jogos em que as
odds propostas pelos mercados de apostas Portugueses se enquadravam no conceito de
valor esperado positivo, que é definido e explicado por Rebelo na obra acima referida.
Uma vez que não foram encontrados quaisquer estudos relativos aos mercados de apostas
Portugueses, tomou-se como segundo objetivo deste projeto testar a eficiência dos
mesmos, e para isso, desenvolveu-se uma estratégia que consistia em apostar apenas nos
claros favoritos, quando estes jogavam em casa: se esta estratégia fosse capaz de gerar
retornos positivos, poderia indiciar a presença de alguma(s) das odds tendenciosas
recorrentemente detetadas nas publicações anteriormente consultadas.
Foram guardados registos de todas as apostas efetuadas, para posteriormente comparar as

odds oferecidas, nomeadamente em termos das odds médias para cada hipótese de
desfecho, bem como da margem média cobrada por cada uma das casas de apostas
(overround), com as odds oferecidas, para os mesmos jogos, por uma casa de apostas
sediada no Reino Unido (Pinnacle), a fim de avaliar o nível de competitividade dos
mercados de apostas Portugueses, face a uma referência internacional.
Após o final da época 2017/2018, analisaram-se os resultados, e tiraram-se as respetivas

ilações:
IV
• Através do da aplicação do método preditivo idealizado pelo autor, geraram-se

retornos na casa dos 20%, face ao investimento realizado. No entanto, devido ao
grau de detalhe com que os relatórios foram escritos, apenas foi possível realizar
11 apostas, o que é manifestamente pouco para se poder concluir que esse método
é capaz de gerar retornos positivos de forma consistente, no longo prazo.
• A estratégia relacionada com apostar em equipas caseiras amplamente favoritas

foi capaz de gerar retornos teoricamente positivos, embora praticamente nulos. Na
realidade, não se pôde concluir a existência de tendenciosidades exploráveis nas
odds propostas pelas casas de apostas autorizadas a operar no mercado Português.
• Em relação à competitividade dos mercados de apostas Portugueses, verificou-se

que, em média, as 4 casa de apostas consideradas neste estudo cobram um
overround bastante superior ao do Pinnacle (8,11% vs 2,75%). Se as mesmas
apostas em equipas caseiras favoritas fossem efetuadas através do Pinnacle, seria
possível obter retornos de 6,62%, que indicariam a presença de odds tendenciosas
exploráveis, e a consequente ineficiência dos seus mercados. Apesar de, em
média, o Pinnacle conseguir propor odds superiores para a vitória dos favoritos a
jogar em casa, se retirássemos o peso do overround, os mercados de apostas
Portugueses teriam fornecido, comparativamente, odds melhores. Uma vez que a
pesquisa previamente conduzida identificou a legislação em vigor como a
principal causa do overround excessivo nas odds portuguesas, este trabalho sugere
que a mesma deveria ser rapidamente revista, sob pena que cada vez mais
apostadores optem por apostar ilegalmente.
V
3- Introduction
“Determining the origin of sports betting is a difficult task. In its truest form, sports
betting has been around for thousands of years, as people have always wagered on the
outcome of an event between two competitors.” (Moody, 2017)
Although people have always been interested in betting on the outcome of sports events,
more recently sports betting started to be seen as an alternative form of investment, where
decision-making is based on numbers and probabilities instead of relying on gut-feeling.
In fact, “the rise of sports analytics, combined with better data and computational power,
has led to the creation of companies that specialize in betting on sport. (...) They pitch
sports betting as an alternative asset class, bringing potential for income and
diversification” (Kucharski, 2016).
“The UK betting industry (…) is leading the way internationally with the implementation
of a series of significant changes, [such as the] abolition of gambling tax for punters and
legitimization of online betting. In addition to the importance of this industry for the
economy per se, betting markets have received much attention in the academic literature
due to their similarities to financial markets. In a seminal paper, Thaler & Ziemba (1988)
were among the first to argue that betting markets may be better suited than financial
markets when testing for efficiency. The main advantage is that bets have a well-defined
period of life at the end of which their value becomes certain and this makes the testing
of market efficiency far less complicated” (Vlastakis et al., 2009: 427).
Besides that, notorious cases of successful punters and traders working on their own also
started to emerge, like Paulo Rebelo or Billy Walters, for example. In “Ganhar com as
Apostas Desportivas”, Rebelo (2012) states that data analysis and statistics are useful
tools to predict the outcomes of sports events, and also that it is possible to obtain positive
returns on betting just by running a quantitative analysis on both teams involved in a
certain match. However, he also claims that in his view, a qualitative analysis (subjective
interpretation of the news and factors that are very hard to quantify numerically, such as
injuries or suspensions of key players) adds far more value and, in case of conflict, should
outweigh the quantitative analysis.
According to the European Gaming and Betting Association (EGBA) report on sports
betting, in 2012 regulated betting accounted for 58 billion dollars, 14% of the total global
1
gambling yield. At the time, European countries represented the largest share of the global
betting market (41%), followed by Asia (39%). The EU fixed-odds sports betting gross
revenue was of 1379 million euros per year (on average), between 2010 and 2012. In
2013, European Sponsorship Association (ESA) said that commercial sponsorship of the
sport sector by gambling companies had become a significant source of sponsorship
funding for sports organizations, with 74 deals closed in 2012. From 2009 to 2013,
regulated gambling companies invested more than 60 million pounds in English Premier
League clubs, and appeared as the main sponsor in some well-known European clubs’
jerseys (like Bwin for Real Madrid and AC Milan, or Betclic for Juventus).
All around the world, different sports constitute an important part of every human society,
and these numbers demonstrate the growing importance of betting companies in the sport
industry, over the last decade. If the sports betting companies’ revenue is growing, one
can conclude that there is an increasing number of people betting. One can also conclude
that the clear majority of punters lose money to the bookmakers in the long term,
otherwise this would not be such an income generating activity. Very likely, this happens
because most punters do not base their betting decisions on a scientific study of the
variables that may affect the outcome of a match, but rather on their instinct and beliefs.
3.1- Forms of placing bets
3.1.1- Fixed-odds betting
Fixed-odds betting is one of the most popular forms of sports gambling. The bookmaker
determines the odds for a given future event and takes the opposite side of every
transaction. “The bettor is left with a take-it-or-leave-it decision: he can either hit the
market quotes or refrain from participating. This is why bookmaker markets are
sometimes called quote-driven markets, by analogy to the same setting in financial
markets” (Franck et al., 2010: 449-450).
In 1x2 market, the punters bet in one of the three possible outcomes of a match (home
win, draw or away win). The market prices of these outcomes are usually presented as
‘decimal odds’ (e.g. a certain bookmaker offers an odd of 1.45 for Arsenal beating West
Ham, at home), which represent the pay-out ratio of a winning bet (in this example, the
bookmaker pays 45 cents for each 1€ bet, if the home team wins). Fixed-odds can be
converted into probabilities (1/odd, as an odd is the inverse of probability), so in this
2
particular case, an odd of 1.45 for an Arsenal home win versus West Ham can be
interpreted as if the bookmaker attributes a probability of approximately 69% to Arsenal
winning the game (1/1.45 = 0.689). However, one must always have the bookmaker’s
overround in consideration (Franck et al., 2010).
“The sum of the probability-odds for the possible outcomes of an event is invariably
greater than one, which is the mechanism through which bookmakers collect commission.
The margin by which the sum of probability-odds exceeds one is known as the overround”
(Forrest et al., 2008: 159). The overround represents the specific expected profit for the
bookmaker, if bets are distributed in such way that identical amounts in prizes are paid to
the punters, whatever the match outcome. Once that it is very unlikely that bets are
distributed in such manner, the overround is simply an estimation of the bookmaker’s
expected return (Constantinou & Fenton, 2013). For example, let’s consider the odds
offered by Betclic for the Premier League match Southampton vs Tottenham, that will
take place in 21 st January of 2018:
• Home win- 4.90 (1/4.90 = 20,4%)
• Draw- 3.90 (1/3.90 = 25,6%)
• Away win- 1.60 (1/1.6 = 62,5%)
• 20,4% + 25,6% + 62,5% = 108,5%, in this specific case the bookmaker’s

overround is 8,5%
“In order to obtain the [bookmaker’s] prediction of the outcome, we assume that the
overround is equally distributed over the outcome probabilities. Therefore, we obtain the
market’s ‘implicit probabilities’ by a linear transformation”: (Franck et al., 2010: 449)
• Home win- (1/4.90) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 18,8%
• Draw- (1/3.90) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 23,6%
• Away win- (1/1.60) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 57,6%
• 18,8% + 23,6 % + 57,6% = 100%
The calculations for removing the overround are automatically done by an Excel
spreadsheet called “True Odds Calculator”, which is available for free download at
3
http://www.football-data.co.uk/
Rebelo (2012) explains the concept of positive expected value: when the probability of a
certain outcome converted into an odd is lower than the odd offered by the bookmaker.
In his opinion, the only way of consistently making a profit in the long-term is being able
to estimate the probability of an event’s outcome in an accurate way, and consequently
identify situations where the bookmakers’ mistakes of evaluation can be exploited. This
author also refers that finding value is less difficult if one specializes in certain leagues
or markets, because extended knowledge is essential when the objective is beating the
bookmaker, who has far more resources.
Among the Big 5 leagues (English Premier League, Spanish La Liga, Italian Serie A,
German Bundesliga and French Ligue 1), Premier League is the one where the money
from commercial and television rights is most equitably distributed for all its participants
(Macário, 2017). For example, the 2016/2017 winner Chelsea received only 1,6 times
more than the last placed Sunderland, which nonetheless still got 107,5 million euros
(Jornal Económico, 2017). This allows that every club is able to sign quality players and,
as a consequence of that, the general level of the competition increases, as well as the
unpredictability of its matches.
In the beginning of each season, there are typically 6 candidates to the top 4, which grants
qualification for the groups stage of Champions League (Arsenal, Chelsea, Liverpool,
Manchester City, Manchester United and Tottenham), and apart from these, nearly all
other teams might finish in the bottom 3, thus being relegated to the 2 nd division of English
football. Having this in consideration, the outcomes of Premier League’s matches would
be the most difficult and challenging to predict, hence we decided to specialize in 2
different betting markets of that league: 1x2 (bet on home win, draw or away win) and
under/over X goals in a match (e.g. if there will be more than 2.5 goals in a certain match).
The fixed-odds bets can be classified in 2 different types: single or multiple. In the first
case, the punter bets on one outcome of a single match, and if that outcome subsequently
occurs, he or she is rewarded with the return implicit in the odd (as explained above). In
a multiple bet, the punter bets separately on 2 or more outcomes from 1 or more matches,
and if all the bets made are accurate, he or she is rewarded as in the following example:
4
A punter bets 5€ on the following outcomes:
• Manchester United beats West Ham away, at 1.75
• Manchester United will score more than 1.5 goals, at 1.60
• A draw in Barcelona vs Real Madrid, at 3.20
If all these outcomes occur, the punter will be rewarded with:
• 5€ * 1.75 * 1.60 * 3.30 = 46,20€.
In other words, the punter is leveraging the return by multiplying the risks. Rebelo (2012)
also addresses this topic, saying that although many professional punters devalue the
usage of multiple bets, the possibility of leveraging is good, in itself. Even if this author
does not work with multiple bets himself, he understands its value in the betting markets.
Nowadays, it is possible to bet on sports events in real time (live bets), but the aim of this
project is to produce accurate forecasts based on past data and subjective information, so
this type of betting will not be considered. In order to minimize the risks, we decided not
to consider multiple bets as well.
Lastly, some bookmakers offer the option of making a single bet on 2 outcomes, in a
match. For instance, that Liverpool beats Leicester at home, and scores more than 3.5
goals. Because the risk is heightened, the odd is also higher (in most cases, higher than if
one makes a multiple bet on the 2 same outcomes). We believe that these markets may
offer some interesting opportunities to explore and therefore, our decision-making
process will take them into consideration.
3.1.2- Exchange betting
“In recent years, person-to-person exchange betting has evolved as a different betting
market structure. Here, individuals contract their opposing opinions with each other. On
an online platform, they can post the prices at which they are willing to place a bet either
on or against a given event. The latent demand for wagers is collected and presented in
the order book, which displays the most attractive odds, with the corresponding available
volume, in a canonical manner. Such a market design is often referred to as an order-
driven market. The bettor has the choice to either submit a limit order and wait for another
participant to match his bet or submit a market order and directly match an already offered
5
bet. As a result, there is a continuous double auction process taking place at the online
platform. If two bettors with opposing opinions agree on a price, their demands are
automatically translated into a transaction. After the bets have been matched, both of the
individuals hold a contract on a future cash flow. The size of the cash flow is determined
by the price of the contract, while the direction of the cash flow is tied to the outcome of
the underlying event. The provider of the platform charges a commission fee, which is
typically lower than the bookmaker’s overround, on the bettors’ net profits” (Franck et
al., 2010: 450).
“Online betting exchanges have experienced a fast boom. The odds analysed in this paper
are from Betfair, which is one of the most prominent bet exchange platforms. With a
weekly turnover of more than $50m and over two million registered users, Betfair
accounts for 90% of all exchange-based betting activity worldwide. It has been online
since 2000, and claims to process five million trades a day” (Franck et al., 2010: 450).
In the same document, the authors also explain that, at least in theory, bet exchanges
should be able to generate accurate forecasts, for three different reasons:
1. The betting exchange works in a way that rewards the value of information. In
this way, well informed traders should be able to generate higher average returns
than uninformed traders, because they are competing against each other.
2. The betting exchange encourages the truthful exposure of information. Based on

the information gathered, traders put money at stake, and by doing it, are revealing
their expectations of the outcome’s probability.
3. Through the price mechanism, the betting exchange delivers a proficient

algorithm, which is capable of gathering, storing and aggregating different
information in a dynamic way.
In practice, studies have demonstrated that bet exchanges provide very accurate forecasts.
In one of those studies, Croxon & Reade (2008) used high-frequency Betfair data in order
to test the efficiency for the arrival of goals, and concluded that prices incorporate the
relevant news swiftly and fully, indicating a high level of efficiency on Betfair odds.
3.2- Legal context of the betting markets in Portugal
The greatest difficulty anticipated is the lack of competition in the Portuguese online
6
betting business. The country’s online betting is regulated by the decree-law nº 66/2015,
which does not tax the punters winnings (unlike the financial markets), but taxes the
bookmakers in 8 to 16% over the volume of bets made (won or lost, which means that the
bookmakers must pay prizes and taxes for bets won by the punters). In contrast, the UK
government taxes the bookmakers in 15% of their profits (which are defined as stakes
received less winnings paid out), in the fixed-odds’ case, and as in Portugal, does not tax
the punters’ winnings at all (HM Revenue & Customs, 2017).
This taxation policy forces the bookmakers to offer lower odds, otherwise they would not
be able to run a profitable business. (Apostas Online, 2017). For example, on January 26
of 2018, Bet365 (UK based bookmaker) was offering an odd of 1.66 for Manchester
United win or draw away vs Tottenham (to be disputed on January 31 of 2018), while
Betclic.pt offered 1.48 (the difference is around 11%). Besides that, “only 39% of
Portuguese punters’ online betting turnover is spent with local operators, far below the
desired 80-90% channeling rate that most regulated online markets strive to achieve”
(Stradbrooke, 2017).
Physical stores specialized in sports betting are a more traditional alternative to online
fixed-odds betting (like the betting shops that exist in High Street, London). In Portugal,
physical betting is regulated by the decree-law nº 67/2015, which grants the State the
exclusive exploration of the physical betting. In its turn, the State concedes the monopoly
of physical betting to Santa Casa da Misericórdia (a charity organization owned by the
State), who introduced Placard in 2015, a physical way of placing bets in a bulletin shape,
which is sold by authorized distributors, such as stationery shops and other retail stores
that are not specialized in betting. The state justifies this exclusive concession of the
physical betting to Santa Casa with the argument that, in this way, it is easier to assure
the protection of the punters’ interests (e.g. to make sure that he or she receives the prize
in case of winning the bet), as well as controlling underage gambling and criminal
activities related with sports betting, such as money laundering (Batista, 2017).
Placard offers the possibility of betting in eight different sports (football, tennis,
basketball, handball, volleyball, rugby, ice hockey and American football) and, in spite
of its undeniable popularity and commercial success (over 1 200 000 people bet on
Placard, in two years), this game is very limitative to the punters, in the sense that it only
covers four different markets:1x2, 1x2 at half time, European handicap and under/over
7
2.5 goals in a match 1. Betting exchanges are not even considered in the current
Portuguese legal framework, and so our project is limited to fixed-odds betting markets.
Many important bookmakers (e.g. Bwin, Betfair, Bet365, Ladbrokes, etc.) consider this
legal environment unattractive and prefer not to ask for the permit that would allow them
to operate in the Portuguese market (Silva, 2014). Currently, Portugal has only four
licensed online bookmakers (Bet.pt, Casino Portugal, ESC and Betclic), while Denmark,
considered the best regulated market in Europe, has forty (Madremedia/Lusa, 2017). This
lack of competition between the bookmakers created an oligarchic market that is harmful
to the punters interests, because bookmakers do not feel much pressure to reward the risks
assumed by the punters with valuable odds (once that there are only 4 possible choices).
In March of 2018, a new online bookmaker obtained the permit to operate in Portugal
(Correio da Manhã, 2018). However, our project was already in progress when this
happened, and so we decided not to consider A Nossa Aposta in this work. For the same
reason, we also excluded Placard.pt, which is the online bookmaker owned by Santa Casa
da Misericórdia (as mentioned above), that entered the Portuguese online betting market
just before the beginning of the FIFA World Cup 2018 (Cortim, 2018).
On December 6 of 2017, the Remote Gambling Association (RGA) announced the

conclusions of a report made to determine how Portugal’s regulated online market (also
includes casino games and poker) was doing since the first locally licensed site began its
activity, in June of 2016. This report was based on a survey, which targeted a sample of
Portuguese online gamblers, and its results demonstrated that 68% of the respondents
were still using gambling sites not allowed to operate in Portugal. From these punters,
38% gambled solely through unlicensed websites, while 30% admitted using both
licensed and unlicensed online gambling platforms. When the respondents were asked
why they still use unlicensed websites, the main reason mentioned was the wish to access
better odds than the ones offered by the Portugal-licensed gambling platforms. The RGA
director of government relations, Pierre Tournier, claimed that the Portuguese regulated
market is “clearly failing to combat the unregulated market and change is much needed
to make the regulation work” (Stradbrooke, 2017).
1
For further information, please consult
https://www.jogossantacasa.pt/web/SCNoticias/newsDetail?contentId=22534.0&headerTitle=Comunicad
os%20de%20Imprensa
8
2017 was the first full calendar year since the regulation of the Portuguese online
gambling market. The SRIJ (Serviço de Regulação e Inspeção de Jogos do Turismo de
Portugal) official report on that year states that the market’s total revenue was of 122,6
million euros, from which the government took 54,3 million euros (44,2%), essentially
from the market’s punitive tax on sports betting turnover. The total sports betting revenue
hit 68,1 million euros in 2017 (55% of the overall online gambling market), and football
accounted for more than 75% of the wagers, followed by tennis with 12,4%, and
basketball with 8,5% (Stradbrooke, 2018).
3.3- Purpose of the project
The core purpose of this project is to create an uncomplicated method, that can be
understood and used by regular punters to obtain positive returns consistently, in the long-
term. This system intends to forecast the outcomes of football matches with accuracy, and
subsequently, to identify profitable opportunities in the Portuguese online fixed-odds
betting markets (1x2 and Under/Over X goals in a match).
The forecasting of the football matches will be done through a combination of analytic
methods and subjective interpretations of the aspects that can affect the outcomes of those
matches, with the aim of supporting our betting decisions. The decisions will be made by
comparing the probabilities calculated for a certain outcome with the odds offered by the
market, for that same outcome, using Rebelo’s (2000) concept of positive expected value
as the main point of reference.
Finally, we will also record the odds of the bets made in the Portuguese markets and
compare them with the ones provided by a British online bookmaker, that will serve as a
benchmark to evaluate the efficiency and competitiveness of the Portuguese online fixed-
odds betting markets.
9
4- Literature review
4.1- Statistical models
In the context of this project, we will not try to develop our own statistical models to
predict the outcomes of football matches, but simply to find good already developed
models and use them as a tool (among others) to forecast football events. Goddard (2005)
mentions two methods of modelling football outcomes: the first one is modelling the
probabilities for the goals scored and conceded by each team (and therefore the
probabilities for the outcomes, in an indirect way), and the second one is modelling the
outcomes’ probabilities (home win, draw, away win) directly. In the same document, the
author compared both methods and found them to yield similar performances in terms of
their ability to predict match results.
4.1.1- Poisson regression models
For the first method, although some articles have previously rejected the goal scoring
process as Poisson (e.g. Reep et al., 1971), Maher (1982) assumed independence between
the scores of both teams and, from the past goal-scoring data of the teams disputing the
top 4 divisions of English football, estimated attacking strength (home and away) and
defensive weakness (home and away) parameters for every team (4 parameters for each
team), through the method of maximum likelihood.
After comparing the expected goal scoring frequencies provided by the model with the
actual ones, goodness-of-fit tests concluded that, in spite of some minor but consistent
discrepancies, the Poisson regression model fitted relatively well into the goal scoring
process. In order to correct the effect of those small differences, which were explained by
the fact that the independence assumption was not completely valid, the author proposed
a bivariate Poisson regression model, with a correlation coefficient between both teams
scores of about 0.2. This modification improved the fit in a significant way.
Another interesting finding was that 2 parameters for each team were enough to describe
the teams’ attacking and defensive behaviour, because “although home ground advantage
is a highly significant factor, it applies with equal effect to all teams, and each team’s
inherent scoring power is diminished by a constant factor when playing away” (Maher,
1982: 113). It is important to highlight that this model was still not able to produce
forecasts for the outcomes of future matches.
10
Starting from Maher’s work, Dixon & Coles (1997) developed the Poisson regression
model into one that was actually capable of predicting future match results, with some
changes for enhancing the model’s predictive ability. The first one was about the
parameters of the model, that were calibrated to reflect the differences of quality between
divisions, which allowed to estimate the outcomes of English cups (FA Cup and League
Cup) matches. In the context of this project, we will not attempt to forecast cups events.
The second change is related with the fact that “in reality, a team’s performance tends to
be dynamic, varying from one time period to another (…) In particular, a team’s
performance is likely to be more closely related to their performance in recent matches
than in earlier matches” (Dixon & Coles, 1997: 272). To account for this, the authors
considered to formalize a stochastic development of the model’s parameters, however,
because this method requires that new parameters for each team would be estimated at
the time of making a bet, they preferred a more simplistic approach. It was assumed that
“parameters are, in a loose sense, locally constant through time and that historically
information is less valuable than recent information” (Dixon & Coles, 1997: 272). Having
this in consideration, the authors modified the likelihood function into a
“pseudolikelihood”, with a weighting function  which allowed for the historical data to
be down-weighted in the likelihood to a degree of choice.
4.1.2- Ordered probit/logit regression models
When it comes to modelling the probabilities for the outcomes directly, Goddard &
Asimakopoulos (2004) used a dataset of 10 seasons fitted into an ordered probit
regression model to forecast English football results. The authors justified the emphasis
on match outcomes rather than scores with simpler estimation procedures, and also with
the argument that fewer parameters were required.
The long-term performance of a team was modelled by building win-loss ratios from the
outcomes of the teams’ past matches, which were measured from the match results data
through distinct time intervals, with the objective of understanding how a team’s past
performance during different lengths of time explains its future performance, on average.
This was achieved by using those win-loss ratios (which were built with data relative to
England’s top 4 divisions, from 1989 to 1998) as the explanatory variables of an ordered
probit regression model. The most recent match data appeared to be the most relevant in
predicting future match results. In fact, it seemed that the impact of the historical
11
performance in the estimation accuracy diminished as a function of time in relation to the

predicted match.
Goddard & Asimakopoulos measured short-term performance levels through discrete

indicators, which were then inserted into the ordered probit model. Those indicators
reflected the team’s results in recent matches (a home win was encoded as 1, a draw as
0.5 and an away win as 0), and were used as explanatory variables together with the long-
term performance ratios. The short-term performance indicators importance to the
forecast accuracy decreased in the same way as the long-term performance level
indicators: the most recent matches have more influence in the forecasting of future
football matches. Empirical test results also demonstrated that generally, long-term
indicators contributed more to the prediction accuracy than short-term indicators.
This was the first model that quantified the predictive quality not only from data relative
to past outcomes, but also from other explanatory variables: importance of the match for
each team (e.g. avoiding relegation), participation in cup competitions and geographical
distance between the home grounds, all of which were tested positive and significant at
the 1% level.
Hvattum & Arntzen (2010) used the ELO system to rate each team’s current strength.
This system “was initially developed for assessing the strength of chess players, but has
been widely adopted in various other sports, including association football” (Hvattum &
Arntzen, 2010: 461). After computing the ratings for every team disputing the English
football top 4 leagues (calculated from historical match results, they are updated at the
end of each game), the rating difference between both teams involved in a match (acts as
a measure of relative performance) was used to derive covariates, which were then fitted
into an ordered logit regression model (works in the same way as the ordered probit, but
assumes logistic distribution instead of standardized normal distribution), in order to
estimate 1x2 probabilities.
“Hvattum & Arntzen compared the forecasting accuracy of the ELO model with the one
of Goddard & Asimakopoulos’ model. Based on the comparison with informational loss
and quadratic loss, the authors concluded that the ELO model produced more accurate
estimates for probability distributions. However, they were not able to determine which
model would be more profitable in betting. In fact, neither of the models were able to
produce abnormal returns when used in betting simulations. The authors also suggested
12
that the differences in forecasting accuracy between the two approaches are likely to
become smaller as sample size increases. Hence, they concluded that the ELO ratings are
a more efficient way of encoding past results when only short time periods are available
to calibrate the model” (Heino & Sillanpää, 2013: 38).
4.2- Fixed-odds betting markets efficiency
“In general terms, the theory of efficient markets is concerned with whether prices at any
point in time fully reflect available information” (Fama, 1970: 413). The author describes
3 types of information-related market efficiency: in a certain market, there is strong-form
efficiency when “all public and private information is incorporated in the security prices
and hence no one can consistently achieve abnormal returns in the market” (Heino &
Sillanpää, 2013: 11), semi-strong-form efficiency is when “all relevant public information
is quickly incorporated into security prices, and hence the opportunities for abnormal
returns disappear quickly” (Heino & Sillanpää, 2013: 11), and a market has weak-form
efficiency “as long as consistent abnormally large returns cannot be achieved by using
historical price [in this case, odds]” (Heino & Sillanpää, 2013: 11).
Kuypers (2000) defines abnormal returns as the ones that are better than the bookmakers’
take (the overround). According to the author, if one uses all the publicly available
information to estimate probabilities for the outcomes of a match and then is able to use
those estimations to get abnormal betting returns, it seems logical to conclude that the
markets at stake are semi-strong efficient. In the same publication, it is also argued that
from a theoretical point of view, even negative returns can be interpreted as an evidence
of semi-strong efficiency. For example, assuming that all the bookmaker’s overround
rates are equal and have a constant value of 9%, negative betting returns of -8% would
indicate that the punter has developed a better ability to estimate probabilities using
publicly relevant information than the market. Constantinou & Fenton (2013: 1) explain
that “a gambling market is usually described as being inefficient if there are one or more
betting strategies that generate profit, at a consistent rate, as a consequence of exploiting
market flaws”, and Buraimo et al. (2013: 182) “report striking evidence of semi-strong
efficiency in the UK fixed-odds football betting market using a reputable newspaper
tipster which offers probabilities of match outcomes rather than simple result indicators.”
An example of a well-documented inefficiency in the fixed-odds betting markets is the
13
favourite-longshot bias, which is a type of systemic error where favourites tend to win
more often than what is implied by the odds (Cain et al., 2000). The authors discovered
that this bias existed in the 1991/1992 odds of English football leagues, and Vlastakis et
al. (2009) found the same bias in a number of big online bookmakers. Based on the odds
offered by several bookmakers over a period of 7 football seasons (from 2005/06 to
2011/12), Constantinou & Fenton (2013) demonstrated the existence of a clear home
advantage bias , whereby the returns produced by betting on home wins were substantially
higher in comparison with the ones obtained by placing bets on away wins. In fact, the
results of this study showed that this bias is almost equally as strong as the favorite-
longshot bias. By using a dataset of 79 446 matches, disputed in 21 leagues of 11 different
European countries (from 2000 to 2001), Direr (2011) claims that would have been
possible to generate positive rates of return of 4.45% if best odds across all the
bookmakers were selected, and 2.78% if mean odds were selected, by following a strategy
of backing overwhelming favourites with odds of 1.19 (odds from 1.15 to 1.24 would
have generated positive returns).
Forrest et al. (2005) affirmed that, at the time, bookmakers set the odds approximately a
week before the matches and, although they reserved the right to adjust the odds until the
kick-off, they rarely did it, regardless of betting volumes and new information (e.g. an
injury of a key player). This allowed informed punters to exploit those market
inefficiencies, which could bring serious financial consequences for the bookmakers.
More recently, Buraimo et al. (2013: 175) claimed that “such a position became untenable
with the emergence of internet betting combined with increased global competition in
betting markets”, and classified “the adjustment of betting odds on football matches by
bookmakers up to kick-off” as a “relatively recent phenomenon”.
Levitt (2004) demonstrated that by intentionally setting odds that are not in consonance
with their outcome probabilities estimations, the bookmakers can exploit the punters’
sentiments and preferences, consequently increasing their profits. The author also
explains that the bookmakers must be careful in doing so, because well-informed punters,
who are able to estimate accurate probabilities for the possible outcomes of the matches,
can notice the distortions and develop strategies in order to use those distortions in their
benefit, generating positive returns.
14
4.2.1- Arbitrage opportunities
“An arbitrage opportunity is simply an opportunity whereby profit is guaranteed on the

basis of a negative profit margin which results by combining the odds published by the
various bookmaking firms. In particular, arbitrage opportunities depend on two factors:
a) the divergence in outcome probabilities between bookmaking firms and b) the profit
margin by each bookmaker. Negative profit margin is simply a scenario where a set of
probabilities is found (for a single match instance) in which the sum of the probabilities
within that set is < 1. Hence, profit for the bettor can be guaranteed if the bets are placed
such that the return is identical whatever the outcome” (Constantinou & Fenton, 2013:
3).
The existence of arbitrage opportunities is a classical sign of weak-form efficiency, as it

is possible to exploit those opportunities in order to obtain a sure profit just from
examining the historical prices, which in this case are the odds offered by the bookmakers.
These opportunities were detected in several publications, namely Buraimo et al.(2013)
Vlastakis et al. (2009) and Constantinou & Fenton (2013), however, as Buraimo et al.
explain, in order to exploit arbitrage one must have the possibility of choosing between a
vast number of bookmakers (as it happens in the English fixed-odds betting market). As
a consequence, in this work there will be no further discussion about this issue (there are
only 4 bookmakers operating legally in Portugal).
4.3- Home advantage
Home advantage is defined by Courneya & Carron (1992: 13) as “the consistent finding
that home teams win over 50% of the games played under a balanced home and away
schedule”, and it is a widely regarded matter in the literature related with sports events.
“Home advantage in football has long been established as an important factor in
determining the result of a game. Its existence is certain to affect the attitude of players,
coaches, referees, fans and the media alike. Surprisingly, and despite over 25 years of
research, the precise causes of home advantage and the way in which they operate are still
not well understood” (Pollard, 2008: 12). In the same work, the author reviews all the
hypothesis raised in order to explain this phenomenon: crowd effects, travel effects,
familiarity, referee bias, territoriality, specific tactics and psychological factors.
15
Figure 1- Percentage of home wins, draws and away wins worldwide, from 1888 to
2017 (Schoch, 2017)
Figure 2- Average of home and away goals scored per game worldwide, from 1888 to
2017 (Schoch, 2017)
16
Figure 3- Percentage of home wins, draws and away wins in the big 5 leagues, from
1888 to 2017 (Schoch, 2017)
When it comes to crowd effects, the connection between home advantage and crowd size
is somewhat questionable, because this advantage works even in matches with very little
attendance (Pollard & Pollard, 2005). Besides that, the proximity and intensity of support
can also influence what is happening on the pitch (Heuer & Rubner, 2009). It was never
understood if crowd effects have more positive influence over the home team or negative
influence over the away team (Pollard, 2008).
Likewise, there is no conclusive statistical evidence for a travel effect inconvenience to

the away team, due to the physical and psychological fatigue caused by a trip to another
team’s home ground. Distance between both teams’ stadiums has been researched as a
factor of disadvantage for the away team (both within the same country and
internationally), with paradoxical results (Pollard, 2006). Nevertheless, research
consistently discovered reductions of home advantage in the case of local derbies, where
no travel is required (Seckin & Pollard, 2008).
Through the analysis of the frequency of disciplinary cards and other referee decisions,
investigators also gathered significant evidence of a bias related with referee decisions
favouring the home team (Thomas et al., 2006). However, until now researchers were
unable to establish the cause for this bias.
17
The home team familiarity with the environment surrounding the match (e.g. stadium,
locker rooms, etc.) should also be considered as an important part of the home advantage.
Although familiarity is, by nature, difficult to investigate, some findings indicate it as a
probable aspect of home advantage. For example, playing on a synthetic field or on fields
with uncommonly large or small dimensions has been tested as an advantage for the team
who is familiar with those particular conditions (Pollard, 1986). Being familiar with
certain climate conditions and altitude also has a positive impact on the home team
chances of winning the match (McSharry, 2007).
Humans beings react aggressively to a feeling of territory invasion (in the same manner
as irrational animals do). This behaviour is called territoriality, and it seems logical to
assume that this may play a role in the home advantage phenomenon. Scientists were able
to demonstrate that home players increase their hormone activity before a game (Neave
& Wolfson, 2003), and also that teams grounded in countries or regions isolated and/or
with a history of conflict have higher levels of home advantage, probably due to an
enhanced sense of territoriality (Pollard, 2006).
Very often, managers change their tactical approach to the match, depending on whether
it is played at home or away. If the away team choose to approach the match in a more
cautious manner, possibly they are giving a territorial and psychological advantage to the
home team (Pollard, 1986). There are no solid scientific studies that relate tactics with
home advantage, despite a study attesting increased home advantage in the second leg of
European cup matches proposed tactical aspects as one of the explanations (Page & Page,
2007).
If managers and players are conscious of the home advantage phenomenon, it can affect
their mental attitude before and throughout the match. It is possible that psychological
factors are the main cause for home advantage, because even though there may be real
motives to explain this advantage, they are being heightened by the players’ beliefs,
creating a self-perpetuating phenomenon (Pollard, 1986). Neave and Wolfson (2005) did
a review of the home advantage in football, from a psychological point of view, and
concluded that at the end of the day, it is what is happening in the minds of referees,
managers and players that defines their actions, and consequently the outcome of the
match, as well as the importance of home advantage in that outcome. The only scientific
work about the psychological condition of footballers in relation to home advantage
18
seems to be an investigation conducted on five players (Waters & Lovell, 2002).
In his book, Rebelo (2002) claims that there is a slight correlation between the home
advantage and the position of a certain team in the league table. According to this author,
weaker teams are more dependent of the home advantage to win than stronger teams, once
that the latter are more consistent. This is the main conclusion of a study that he did for a
statistics course, in the year of 2005 (that same study is fully available at
https://www.academiadasapostas.com/blog/2011/01/o-factor-casa-no-futebol).
“One of the problems about researching home advantage in football is the fact that the
likely causes outlined above will be operating together, each interacting with the other in
ways that will be difficult to investigate, isolate and quantify. (…) Thus a researcher will
need to develop a strategy which either takes a multivariate approach, or which carefully
controls for possible confounding variables that are not the main focus of the study”
(Pollard, 2008: 13).
4.4- Money management methods
4.4.1- Fixed-stake
In this money management method, the punter defines a percentage of the initial money
to invest in each bet, independently of the risk taken or the reward offered by that bet.
Rebelo (2012) recommends this method due to its simplicity, and also because it assures
that the punter will never lose all the initial money. Simultaneously, it allows the punter
to exponentially increase that initial percentage, as his ability of betting accurately is
developed. One possibility is to redefine the initial percentage at the beginning of the
following month, based on the growth rate of the initial money. For example, if one begins
January with 100€ and has defined a 10% stake for each bet, at the beginning of February,
if the initial money grows to 110€, then the percentage at stake can be adjusted to 11%
until further review, and so on.
4.4.2- Martingale
Mansuy (2009) reviewed in detail the origins and the various senses of the word
martingale, with their respective etymologies, in mathematics, gambling, technology, and
vernacular language. From the gamblers perspective, the word goes back to the beginning
of the 18th century, when it was used to define a strategy of always betting as much as it
19
covers everything that was lost until that point, with the objective of recovering the all
the losses in just one move. Rebelo (2012) explains that, although in theory this is a
perfect method, as it virtually assures that anyone can make money in betting, he does not
recommend it because, in his view, what makes a punter win is the accuracy of his bets,
and not the money management method. If it is true that a good punter can lose all the
initial money for not having a good money management method, a bad punter will not
make money just for using a good money management method. According to this author,
if a punter cannot make money betting a fixed stake, he will not be able make money
consistently just by applying a martingale. On the contrary, a bad punter that falls into a
negative spiral of results would be multiplying the risk many times just to break even.
4.4.3- Kelly criterion
“The Kelly criterion (Kelly, 1956) is a vital tool in the armory of both portfolio investors
and gamblers. By maximizing logarithmic utility – simultaneously minimizing the risk of
ruin – Kelly provided the formula that gamblers with perfect probabilistic knowledge
must use to grow their bank at the largest expected rate” (O’Shaughnessy, 2011: 2).
“The basic Kelly criterion for a single option on a regular betting market gives the Kelly
Bet B as:
(1)
where M is the team’s market price [odd] and p is the gambler’s presumed probability of
the team winning. B is expressed as a percentage of the bettor’s bankroll, and a bet should
be placed if Mp>1. The formula is derived by maximizing log (expected bank) with
respect to the bet proportion B” (O’Shaughnessy, 2011: 2).
4.5- The importance of subjective factors
Forrest et al. (2005: 562) affirm that “it seems highly likely that odds-setters’ prices are
influenced by information of this kind. If the subjective information is used effectively,
it may contribute positively to the odds-setters’ forecasting performance”. Constantinou
et al. (2012: 1) presented a Bayesian framework “[graphical probabilistic belief network
20
that represents the conditional dependencies among uncertain variables, which can be
both objective and subjective] for forecasting Association Football matches [outcomes]
in which the subjective variables represent the factors that are important for prediction
but which historical data fails to capture”. “Pi-football” model produces match forecasts
“by considering generic factors for both the home and away team” (Constantinou et al.,
2012: 5). Those factors are: 1) relative team strength, 2) recent form, 3) psychological
factors and 4) fatigue.
“Components 2, 3 and 4 are predominantly dependent on subjective information. They

are used to revise the forecast from component 1. The outcome of each of the components
is mutually summarized in a single value (considering both teams) which we describe as
‘subjective proximity’. The subjective proximity is measured on a scale from 0 to 1. A
value equal to 0.5 indicates no advantage for either of the teams; a value less than 0.5
indicates an advantage for the home team, while a value greater than 0.5 indicates an
advantage for the away team” (Constantinou et al., 2012: 5).
For evaluating the performance of the model, the authors used both accuracy and
profitability measurements, and concluded that the subjective information improved the
forecasting ability of the model in a significant way. In terms of accuracy, they concluded
that the precision of quantitative forecasts was significantly inferior to the bookmaker’s
odds implicit probabilities, and also that subjective information improved the forecasts in
a way that matches the bookmaker’s estimations. This suggest that the bookmakers also
use information that is not captured by the standard statistical football data available to
the public, as suggested by Forrest et al. (2005).
For the 380 Premier League matches of the 2010/2011football season, it would be
possible to obtain profitability rates from 2.87% to 9.48% (at standard discrepancy levels
of 5% between the outcome’s probabilities provided by the model and the odds offered
by the multiple bookmakers considered), and from 8.86% to 35.63% (at higher standard
discrepancy levels, of 8% to 11%), respectively. Constantinou et al. (2012: 17) claim that,
at the time, “no other published work appears to be particularly successful at beating all
of the various bookmakers’ odds over a large period of time, which highlights the success
of pi-football”.
21
4.6- Conclusions about the literature review
After carefully reviewing the existent literature about sports betting, it is possible to take
some useful conclusions, which may help us in the development of a user-friendly method
to achieve positive returns on the Portuguese online fixed-odds betting markets.
Previously, we understood the importance of specialization to increase our chances of
beating the odds in a consistent way and therefore, we decided to concentrate in the
markets 1x2 and under/over X goals, of the English Premier League matches 2. We are
going to support our investing decisions with probabilities estimated by a regression
model, and also with descriptive statistics about the Premier League teams. Since we
decided to bet on both 1x2 and under/over X goals in a match, the first modelling method,
estimating probabilities for the expected goals to be scored and conceded by each team
(and therefore the outcomes probabilities, in an indirect way), seems to be the most
appropriate. That said, we will use a Poisson regression model applied to the estimation
of goal scoring probabilities.
When it comes to the efficiency of fixed-odds betting markets, we were unable to find
any literature about the Portuguese context, hence we think that any disclosures about this
topic would be a relevant contribution to the existing literature. In other countries, biases
such as the favorite-longshot bias or the home advantage bias were consistently
documented and therefore, we can attempt to develop betting strategies with the intention
of identify and exploit those biases in the Portuguese online fixed-odds betting markets.
If our strategies are able to generate positive returns, it can be seen as a strong evidence
that the Portuguese online fixed-odds betting markets are at least semi-strong efficient,
once that arbitrage opportunities (which indicate weak-form efficiency) are very unlikely,
due to the short number of bookmakers functioning legally in the country.
The initial amount of money available for this investment project is relatively small
(140,14€), and so we should use a fixed stake method (10% of the initial money on each
bet), as it virtually assures that we will not run out of funds until the end of the football
season. The importance of the qualitative analysis is highlighted by Rebelo (2012) and
confirmed by Constantinou et al. (2012). Eventhough this analysis is mostly subjective,
and consequently dependent of the interpretation and knowledge of each individual, the
2 Please consult the 4th page of our work
22
latters defined an evaluation method that can be applied to our forecasting method.
5- Method
5.1- Poisson regression model
As mentioned previously, the aim of this work is not to build our own statistical model.
From literature review, it was possible to conclude that the modelling method which
estimates probabilities for the expected goals to be scored and conceded by each team,
and indirectly, the probabilities for each possible outcome, was the one that made more
sense to apply here. Our research also identified the Poisson regression model as the most
widely used when the objective is to forecast the probabilities for the number of goals
that might be scored in a certain match. Moreover, although statistical modelling is an
integrating and important part of our work, it is not its core subject, but rather a supporting
element to our decision-making system. In fact, this paper is more targeted to regular
punters than to statisticians or academics. Consequently, this document will contain a
somewhat simplistic explanation about the building process of the model and how it is
able to produce accurate estimations for each possible score. If one has interest in the
statistical details of this model in the full extent, we suggest the further consultation of
Gardner's (2011) work, from where we retrieved the Poisson regression model used in
this project.
5.1.1- Counting process and Poisson process definitions
“For each real number t ≥ 0, let N(t) denote the number of times a specified event occurs
by time t. Note that N(t) is a nonnegative-integer valued random variable for each t ≥ 0.
The collection of random variables N(t): t ≥ 0 is called a counting process because it
counts the number of times the specified event occurs in time” (Weiss, 2006: 686).
“A counting process N(t) : t ≥ 0 is said to be a Poisson process with rate λ if the following
three conditions hold” (Weiss, 2006: 688):
The second condition in the definition states that the increments in time must be
23
independent, although when it comes to football, this may not be the case. If a match is
split into 2 increments, first half and second half, then it seems evident that, for several
reasons (e.g. an important player gets injured during the first half), what happened in the
first half might affect what will occur in the second half. It is even possible that football
matches themselves are not independent, as many dependencies can affect the outcome
of a match, such as differences in motivation between teams, resulting from success or
underachievement in the near past, for instance. However, for the purpose of modelling
football results, we will assume independence and take that a Poisson process N(t) has a
Poisson distribution (Gardner, 2011). A goodness-of-fit test on football scores’ data will
be performed later.
5.1.2- Poisson random variable
A random variable with Poisson distribution (X) can be defined as the number of
occurrences in a continuous interval: timespan (in our case), volume, etc. This distribution
can be used to represent different phenomena that happen in our daily lives (Curto, 2016),
such as the monthly number of traffic accidents, the daily number of phone calls received
by an individual or the total number of goals scored in a match.
In the context of our work, makes sense to consider a timespan of 90 minutes (plus the
stoppage time given in each match), in which the goals are scored randomly by both teams
(e.g. 1st goal scored by the minute 21, 2 nd goal scored by the minute 43, 3rd goal scored
by the minute 56, etc.). Furthermore, the values that the Poisson random variable can
assume must be non-negative integers, as it is impossible to score -2 goals or 5,5 goals,
for example. The probability mass function of a Poisson random variable is of the form:
(2)
For values of x = 0, 1, 2,…, n and where λ (expected value of X) is the only parameter
which characterizes this distribution (Curto, 2016).
The probability mass function of a Poisson random variable demonstrates that the
probability of the random variable X being a value x is equal to the right side of the
equation. Since it is used to calculate probabilities for the number of times that an event
24
might happen, and it is clear that the sum of all the probabilities for the number of times
that an event may occur is equal to 1, then (Gardner, 2011):
(3)
“for any value of λ, where λ is the expected value of the variable. A sum is used here as
a Poisson process and distribution uses discrete values, whereas in the continuous case,
this would be an integral as the area under the curve would be needed. However, in the
discrete Poisson case, it is only required that the number of occurrences at each time is
recorded as the data points, which can then be viewed in a bar chart. The graphical
representation of this will look different for differing values of λ. Lower values of lambda
will skew the density to the left as the expected value of the random variable is low
meaning that a higher percentage of low values will occur. (…) For a Poisson distribution,
the value λ represents the expected number of occurrences of an event for the variable in
question. A unique property of the Poisson distribution is that both the mean and the
variance are equal to this value of λ” (Gardner, 2011: 8).
5.1.3- Dataset
Rebelo (2012) claims that 3 seasons is the maximum timespan in which past data remains
relevant for predicting future events, because data from beyond that period does not
reflect the current strength of the teams involved in a match, due to the significant changes
that occurred meanwhile, such as player transfers or managerial changes, for example.
Dixon & Coles (1997) used that same data timespan in their model, and in Goddard &
Asimakopoulos' (2004) case, the model measured the long time performance from the
teams historical match data, which was only tested significant as a predictor of future
results until two seasons beyond the current one. We find very interesting that the
empirical allegations made by Rebelo correspond to the statistical testing results of
Goddard & Asimakopoulos’ work.
The 2015/2016 season of English Premier League was very atipic, with Leicester City
being crowned champions (when they historically struggled to avoid relegation), Chelsea
(champions in 2014/2015 and 2016/2017) ending the season ranked 9th and Liverpool
25
finishing in 8th place. On the other hand, from the end of the 2016/2017 season to the
beginning of the 2017/2018 season, only 3 clubs experienced managerial changes:
• Southampton- Claude Puel left the club, Mauricio Pellegrino joined in
• Crystal Palace- Sam Allardyce was replaced by Frank de Boer (who was fired
early in this season)
• Watford- Walter Mazzarri returned to Italy, and the ex-Hull City coach Marco
Silva was selected as his substitute
Given these theoretical reasons, it seems more appropriate to consider in the Poisson
regression model just the seasons 2016/2017 and 2017/2018, as this data timespan
appears to be the most significant for measuring the teams’ current attacking and
defensive strenghts. The historical data for both these seasons is fully available at
http://www.football-data.co.uk/, and when it comes to the current season, the data is
updated at each match day, as the season progresses. The descriptive statistics used in the
forecasting reports are only relative to the 2017/2018 season, for practical reasons (the
websites where those statistics are available present the information by season).
5.1.3.1- Goodness-of-fit
In order to use the Poisson regression model to estimate probabilities for the number of
goals that might be scored in a given match, and indirectly, for the possible outcomes of
that same match, it is imperative to assess if our dataset fits well into the Poisson
distribution. To this end, we decided to generate random numbers in Excel, set to be
Poisson distributed and with λ equal to the mean of our dataset (total number of goals /
total number of matches). Afterwards, we compared the histogram of this distribution
(left side) with another one representing the distribution of the actual number of goals
scored, in a certain timespan (right side). In the histograms, the x axis represents the bins
(in this case, the total number of goals scored in a certain match), while the y axis
represents the number of observations for each bin. As the season 2017/2018 progressed,
we needed to add new data at the end of each match day. In order to see if that new data
was fitting well into the Poisson distribution, we decided to divide the data in the
following pieces:
• 2016/2017 full season- λ = 2,8
26
• 2017/2018 full season- λ = 2,67894737
• 2016/2017 + 2017/2018 full seasons- λ = 2,73947368
• 2016/2017 + 1/4 of the 2017/2018 season- λ = 2,76210526
• 2016/2017 + 2/4 of the 2017/2018 season- λ = 2,75087719
• 2016/2017 + 3/4 of the 2017/2018 season- λ = 2,75639098
The results are presented below:
Figure 4- Random Poisson Distribution with λ = 2,8 vs Goals Distribution 2016/2017
Figure 5- Random Poisson Distribution with λ = 2,67894737 vs Goals Distribution

2017/2018
27

2016/2017 + 1/4 of the 2017/2018 season

2016/2017 + 2/4 of the 2017/2018 season

2016/2017 + 3/4 of the 2017/2018 season
28

2016/2017 + 2017/2018
As one can observe, in all cases the shapes of the distributions look very similar to each
other. Another interesting observation is that the means of both seasons are also very
similar to each other, which indicates that the teams’ attacking and defensive abilities did
not change much from one season to the other. However, in order to assure that our
dataset fits into the Poisson distribution, a goodness-of-fit test must be performed.
In his work, Gardner (2011) uses the Chi-squared goodness-of fit test, as the Poisson
distribution and the football data are both discrete. The aim is to test if the observed data
follows a particular distribution (in this case, Poisson), and the hypothesis are formulated
in this way:
• H0: The total number of goals in a match is a random variable with Poisson
distribution
• H1: The total number of goals in a match has no Poisson distribution
Subsequently, the Chi-squared distribution test is calculated by this formula:
(4)
Where:
• Oi = Observed frequencies of the bins, calculated by the Excel function
29
=FREQUÊNCIA
• Ei = Expected observations under the distribution in question (in this case,

Poisson). The first step to obtain the values for Ei is to calculate the expected
probability for each one of the bins, under the Poisson distribution, using the Excel
function =DIST.POISSON. Then, it is necessary to multiply each result for the
respective Oi.
• n = Number of bins (in our case, 11)
Knowing the degrees of freedom (10), it is possible to calculate the p-value for the test,
using the Excel function =DIST.CHIQ.DIR, as the Chi-squared distribution is positive
(skewed to the right). If the p-value > 0,05, we do not reject H0, and consequently, we
can assume that the dataset has approximately a Poisson distribution. Once again, the
results are presented below:
Figure 10- Chi-squared goodness-of-fit test results for 2016/2017
Figure 11- Chi-squared goodness-of-fit test results for 2017/2018
30
Figure 12- Chi-squared goodness-of-fit test results for 2016/2017 + 1/4 of the
2017/2018 season
2017/2018 season
2017/2018 season
31
Figure 15- Chi-squared goodness-of-fit test results for 2016/2017 + 2017/2018
The results of the Chi-squared goodness-of-fit test are conclusive: once more, in all cases,
the p-value is far above the smallest level of significance at which the null
hypothesis would not be rejected (0,05). These values, in addition to the histograms, make
us feel quite comfortable to assume that our dataset fits approximately into the Poisson
distribution.
5.1.4- Poisson regression
After the assumption that our dataset fits into the Poisson distribution was validated by
the Chi-squared goodness-of-fit test, the Poisson regression model can be used to estimate
the probabilities for each possible score, in a given match.
The Poisson regression model is used as the “standard model for count data” (Cameron
& Trivedi, 1998: 9), which refers to the number of times that an event occurs, and follows
the same steps as a normal linear regression model (y = Xβ + ε), where the scalar variables
y and the explanatory variables X are used to estimate the parameters β. However, it is
slightly different as the values for count data need to be positive, and consequently,
exponentials need to be used in the Poisson regression model’s case (Gardner, 2011).
The Poisson regression model comes directly from the Poisson distribution by using the
same mass function and variables that describe the occurrences of the events. In this
model, yi are the scalar dependent variables which will correspond to the number of times
that the event in question occurs, and xi is a vector of linearly independent regressors,
that is, a variable which has an effect on the values that yi can assume (Cameron &
Trivedi, 1998). As the same authors explain (1998: 21), “the standard estimator for this
model is the maximum likelihood estimator”, and so the parameters β are estimated using
32
the method of maximum likelihood.
5.1.4.1- Applying the Poisson regression model to football data in R
As we explained in 5.1.3, the football data used in the Poisson regression model consists
on the outcomes of all Premier League’s matches played between 2016/2017 and
2017/2018. The raw historical data was retrieved from http://www.football-data.co.uk/,
and when it comes to 2017/2018, was updated at the end of each match day, as the season
progressed. In 5.1.3.1, we tested our data as fitting closely into the Poisson distribution
and therefore, usable in a Poisson regression model.
Gardner (2011) goes further by using the Poisson regression model to simulate a whole
season from the estimated parameters, and also to predict a future league table. The author
utilizes the statistical program R, having developed a code that fully operationalizes all
the steps necessary to the progression of his work. In the context of our research, we only
need to use the Appendixes B (“Parameter Estimation”) and C (“Probability Table”) of
his R code, and so we will not consider all the other segments that are not directly related
with obtaining the probabilities for each possible score, in a given match. The parts of the
R code used in this project were retrieved from
http://www1.maths.leeds.ac.uk/~voss/projects/2010-sports/Football.R.
First, we set a working directory in R (“R_tese”), where we created an Excel workbook

(“Data_MSc”), that contains a spreadsheet (“Folha4”) in which we organized the data in
4 different columns:
• Home team name
• Away team name
• Home team final score (FTHG)
• Away team final score (FTAG)
The following step was to load all the relevant R packages, including “openxlsx”, which
allows the R to read Excel files. After that, we imported the data from “Folha4” to our R
workspace.
Y corresponds to the data and, consequently, to the scores of the teams involved in our
dataset’s matches. As Yi is a vector, it can be created by including all of the teams’ scores
33
in it. Having this in consideration, if the data contains g number of matches, then the
length of Yi will be 2g, as there are 2 bits of data for each match, the home team’s score
and the away team’s score. It is clear that Yi ∈ N0 as the team’s scores will be an integer
of 0 or larger, with large numbers being very unlikely, and i = 1, ..., 2g. Given this
explanation, the vector will be of the form:
(5)
where ygi,j represents the number of goals scored by team i versus team j in game g
(Gardner, 2011).
The β is a vector containing the parameters for the model, which in this case, is of length
2n, as each of the n teams will be associated with an attacking and a defensive strength.
The first half of the vector will contain the attacking strengths (αk) of the n teams, and
the second half will contain the defensive strengths (γk) of the n teams, resulting in a
vector of the form (Gardner, 2011):
(6)
5.1.4.1.1- The home advantage parameter
Gardner (2011) also considers the possibility of adding a distinctive parameter between
the home team and the away team, as the teams playing at home benefit from a statistically
demonstrated advantage (as discussed in the subchapter 4.3 of our work). The author uses
the two-sample t-test to see if the mean of the home goals is significantly different from
the mean of the away goals. Testing the difference between the means of these two
samples will show if there is a clear difference between them (i.e. do teams score more
goals at home than away).
Although the t-test demands the variables to have normal distributions, Clarke & Cook
(2004: 416) state that “If X has a very skew distribution [e.g. Poisson], we need a sample
of considerable size (say, at least 200) before the distribution of X approaches reasonably
closely to a normal distribution”. The dataset used by Gardner (2011) includes all the
matches from the Premier League 2009-2010 (380), while our dataset includes all the
matches from the Premier League 2016-2017 and 2017-2018 (380 each), therefore the
34
variables can be assumed as normal distributed, and the t-test can be performed.
To test whether the difference between the means of the home goals and the away goals
is statistically significant or not, the following hypothesis was formulated:
• H0: μ1 = μ2
• H1: μ1  μ2
In Gardner’s (2011) case, the null hypothesis was strongly rejected at the 99% confidence
level, which implies that the difference between the means is statistically significant and
consequently, makes sense to add a home advantage parameter δ to the parameter vector
β, as it is an extra parameter that needs to be estimated from the data. As a consequence,
equation 6 becomes 7 and now has the length 2n + 1.
We decided to test if, in our case, the difference between the means of the home goals
and the away goals was also statistically significant. To do that, we used Microsoft Excel
instead of the statistical program R. According to Curto (2016), the first step is to assess
if the variances of both samples (home goals and away goals) differ in a statistically
significant way, using the “F-Test Two-Sample for Variances” from the “Data Analysis
Toolpack”. The hypothesis was:
• H0: The variance of the home goals is equal to the one of the away goals
• H1: The variance of the home goals is not equal to the one of the away goals
These were the results obtained:
Figure 16- F-test results for 2016/2017 Figure 17- F-test results for 2017/2018
35
At the 1% significance level, the null hypothesis was not rejected for the season
2016/2017, and rejected in the case of the 2017/2018 season, which meant that we
would need to perform a “t-Test: Two-Sample Assuming Equal Variances” to the
former, and a “t-Test: Two-Sample Assuming Unequal Variances” to the latter.
Figure 18- t-Test results for 2016/2017 Figure 19- t-Test results for 2017/2018
The results of our t-Test were exactly the same as the ones from the test executed by
Gardner (2011), that is, the null hypothesis was strongly rejected at the 1% significance
level, which confirmed the necessity of estimating a home advantage parameter.
5.1.4.1.2- Y, X and β
Gardner's (2011) next step was to formulate Y and X into the R language. To achieve this,
the author wrote the following code:
Figure 20- Gardner's (2011) code for Y and X
36
After that, the parameters for the attacking and defensive strengths of each team, as well
as the home advantage parameter, were estimated. “As Arsenal is the first team
alphabetically in 2009- 2010 Premiership season, Arsenal’s attack strength is being set to
0. (…) This actually makes sense because now the other 40 parameters (19 attack
strengths, 20 defence strengths and the home advantage) can be calculated by using this
as the benchmark. Since Arsenal is one of the best teams in the league, it can be expected
that only Manchester United and Chelsea (who were the only two teams to come above
Arsenal in the league in 2009-2010) will have strengths above 0 and the majority of other
strengths will be below 0. Using data inputted into R, it is possible to estimate the
parameters using a function called glm which stands for ‘generalized linear model’. The
Poisson regression model is a form of generalized linear model but with the Poisson
distribution taken into account (…). The glm function (…) estimates the 40 remaining
parameters, using the benchmark of the [Arsenal’s] parameter (…)” (Gardner, 2011: 30).
Figure 21- Poisson generalized linear model (glm) code, which was used by Gardner
(2011) to estimate the parameters
Figure 22- Estimated attack and defence parameters for each one of the Premier
League’s teams, plus the home advantage parameter (using data from the seasons
2016/2017 and 2017/2018)
37
5.1.4.2- Simulating games
To estimate the parameters, the scores from past games, Y, were used. Now these
estimates can be used to simulate future scores, by setting up the matrix X 3 for the games
to be simulated and putting the set of 41 parameters into order and named as the vector β,
they can be inserted into the formula Y = exp (Xβ), and the vector Y is predicted. This
vector will be formed as this (Gardner, 2011):
(7)
For example, considering the estimated parameters in figure 22, if we want to simulate a
match between Chelsea (home) and Arsenal (away), we should replace the symbols in
equation 8 for the respective numbers:
• λ Chelsea = exp (- 0.03930105 - (- 0.46511770) + 0.2862813)
• λ Arsenal = exp (0 - (- 0.17091337)
As it is explained in the subchapter 5.1.2 of our work, for a Poisson distribution, the value
λ represents the expected number of occurrences of an event, which in this case is the
expected number of goals scored by each one of the teams. Knowing the λ for each team,
it is possible to use the probability mass function of Poisson to calculate the expected
probability for each team to score 1, 2, 3, etc. goals, and from there, the expected
probability for each possible final score (0-0, 1-0, 1-1, 1-2, etc.). Gardner (2011)
developed this R code, which does this automatically, and presents the results in a table
format:
3 To understand this matrix in detail, please consult Gardner (2011: 22-28)
38
Figure 23- Gardner's (2011) code for each score’s probability table
Figure 24- R probability table for the match Chelsea vs Arsenal (using data from the
seasons 2016/2017 and 2017/2018)
In this example, if we sum all the probabilities for the home win (e.g. 1-0, 2-1, 2-0, etc.),
for the draw (e.g. 0-0, 1-1, 2-2, etc.) and for the away win (e.g. 0-1, 1-2, 2-3, etc.), we
obtain, indirectly, the probabilities for each possible outcome. Likewise, we can also sum
the probabilities for the match ending with under 1.5 goals (0-0 + 1-0 + 0-1), over 2.5
goals, etc.
5.2- Betting strategies
The initial 140,14€ will be split in half and applied to the following 2 betting strategies
in equal manner. At the end of the season, we will compare the performance of these
strategies in terms of accuracy and return.
39
5.2.1- Value strategy
As we have seen in the subchapter 3.3 of this work, due to the current situation of the
Portuguese fixed-odds betting markets, it may be challenging to find odds with positive
expected value. However, we believe that it is possible to do it, by applying the following
steps:
1. Look at the Premier League 1x2 and under/over X goals odds offered by the 4
bookmakers operating legally in Portugal, and identify the ones that offer a
relatively good reward for a relatively low risk (this should be done at least once
a week)
2. Run a quantitative analysis on the intended matches, using the Poisson regression
model and the descriptive statistics, with the aim of finding data patterns that can
support our decision-making procedure
3. Write a detailed forecasting report with the relevant results of the quantitative
analysis plus all the qualitative factors (tactics of both teams, psychological
moment, missing players, weather, etc.) that can influence the outcome of the
event and consequently, our decision of betting or not.
5.2.1.1- Forecasting report model
Introduction:
• Name of both teams
• Competition match day
• Date and time of the match
• Location: stadium and city
• Weather at the time of the match: precipitation, humidity and wind (source:
Google)
• Potential bets: markets that seem to have interesting odds
40
Probable line-ups: www.whoscored.com -> Previews -> Select the match from the list -
> Preview
Figure 25- Probable line-up Tottenham vs Arsenal (played at 10/02/2018)
Injuries and suspensions: www.whoscored.com -> Previews -> Select the match from
the list -> Preview
• List of all the players that are not available or are at risk for the match in question
Figure 26- Injuries and suspensions Stoke City vs Brighton (played at 10/02/2018)
Discipline table: This table is part of the tool Betting Data 2017-2018, which can be
downloaded for free at http://www.football-data.co.uk/. This tool analyses the raw data
available for download in the same website, and presents it in a way that can assist the
punter’s decision-making process.
• This table tells us the level of aggressiveness of both teams when playing at home,
away and overall, in comparison with the Premier League’s average (the right-
sided table refers to the home team disciplinary data, the table in middle indicates
the away team disciplinary data, and the green table shows the Premier League’s
average disciplinary data), measuring the average number of yellow cards, red
41
cards and booking points per match. In order to obtain the Premier League’s
average per match for a single team, one must divide the green table’s numbers in
half
• A yellow card worth 10 booking points and a red card corresponds to 25 booking
points (there is no distinction between direct red cards and double yellows)
Figure 27- Discipline table Everton vs Crystal Palace (played at 10/02/2018)
Tables: www.academiadasapostas.com -> Estatísticas -> England flag -> select match
from the list
• Head-to-head over the last 3 years table: allows to understand if one team has an
historical advantage over the other, maybe due to having a style of play that is
suited to exploit the other team’s weaknesses, or because of psychological factors
Figure 28- Head-to-head over the last 3 years Swansea vs Burnley (played at
10/02/2018)
42
• Premier league tables overall, home and away: allows to understand the
comparative strength of both teams, and the importance of home advantage for
each one of them
Figure 29- Premier League tables Swansea vs Burnley (played at 10/02/2018)
SoccerSTATS data: http://www.soccerstats.com/latest.asp?league=england -> select

match -> H2H
• Home/away performance: this table describes the percentage of points, goals

scored and goals conceded by each team at home and away
Figure 30- Home/away performance West Ham vs Watford (played at 10/02/2018)
• Points per game table: this table describes the relative form of each team, which
corresponds to the difference between the average points per game in the last 8
43
matches and the average points per game in all matches played so far in the
Premier League. If positive, it means that the team is performing better in the last
8 matches than in the overall season, if negative it means the opposite
Figure 31- Points per game West Ham (at 10/02/2018)
• Scoring table: This table provides all the information that is relevant to bet in the
goals related markets
Figure 32- Scoring table of West Ham (at 10/02/2018)
Style of play data: www.whoscored.com -> Previews -> Select the match from the list -
> Head to Head -> Team Statistics
• Team characteristics: provides a detailed description of every team’s strengths,

weaknesses and style of play. It allows to understand if any of each team’s
strengths match with the other team’s weaknesses, and vice-versa
44
Figure 33- Team characteristics Manchester City vs Leicester (played at 10/02/2018)
Subjective factors:
As mentioned earlier, these factors are very important, and having them in consideration
is essential when trying to produce accurate forecasts of what is likely to happen on the
pitch. However, they are hard to quantify and therefore left to subjective interpretation.
Constantinou et al. (2012) attribute key importance to three in particular: form,
psychological moment and fatigue. We believe that some of the descriptive statistics
present in this report describe relatively well the factor form, and so we will focus on the
factors psychological moment and fatigue. In the same work, the authors propose a system
for assessing these last two factors that we will adopt in our forecasting reports.
45
Figure 34- Evaluation method for the psychological moment of a certain team
(Constantinou et al., 2012: 19)
The point IV. of the Team Psychology evaluation will not be considered, as it was
previously discussed in the Head-to-head over the last 3 years table 4
Figure 35- Evaluation method for the fatigue of a certain team (Constantinou et al.,
2012: 19)
Poisson regression model:
The results of the Poisson regression model are presented in this Excel table, including
the probabilities for each possible score, the probabilities for 1x2, the probabilities for
under/over X goals and the respective odd conversion.
0 1 2 3 4 5 6 7+ Home
0 5,16% 7,02% 4,77% 2,16% 0,74% 0,20% 0,05% 0,01%
1 8,27% 11,26% 7,66% 3,47% 1,18% 0,32% 0,07% 0,02%
2 6,64% 9,03% 6,14% 2,79% 0,95% 0,26% 0,06% 0,01%
3 3,55% 4,83% 3,28% 1,49% 0,51% 0,14% 0,03% 0,01%
4 1,42% 1,94% 1,32% 0,60% 0,20% 0,06% 0,01% 0,00%
5 0,46% 0,62% 0,42% 0,19% 0,07% 0,02% 0,00% 0,00%
6 0,12% 0,17% 0,11% 0,05% 0,02% 0,00% 0,00% 0,00%
7+ 0,03% 0,05% 0,03% 0,01% 0,00% 0,00% 0,00% 0,00%
Away
5,16% 11,26% 6,14% 1,49% 0,20% 0,02% 0,00% 0,00%
Home win 32,50% Odd 3,08
Draw 24,27% Odd 4,12
Away win 43,23% Odd 2,31
Total 100,00%
Under 2.5 43,12% Odd 2,32
Over 2.5 56,88% Odd 1,76
Total 100,00%
4 Please consult the 41st page of our work
46
Figure 36- Poisson regression model results on Watford vs Everton (played at

24/02/2018)
Betting decision (based on the odds):
Here, the report’s content is compared with the potential bets, and afterwards a decision
is made about which markets worth a bet (if any).
5.2.2- Accuracy strategy
According to the document “The Football Betting Mastermind” (Betinfo24: 2010), it is

very unlikely that an individual punter is able to find value and beat the bookmakers in
the long term (in major leagues), because these have top-level statisticians and football
experts working for them, backed up by the best technology that money can buy
(consequently, the odds are pretty accurate). So, the only way of making consistent profits
is being able to identify the teams that are expected to win almost every game, and betting
high stakes on those teams (even if the odds are low).
The section 1 of that document explains how to identify the teams that have a very high
percentage of wins, dividing them in “hot teams” and “strong teams”. “Hot teams” are
defined as the ones that are very likely to win every match in the leagues where they
compete (except when playing against each other). By its turn, “strong teams” are defined
as the ones that are maintaining good performances as well, but are not as consistent as
the “hot teams”, especially when playing away (usually these teams are ranked between
3rd and 7th in the league table). As the “strong teams” do not perform as regularly as the
“hot teams”, bookmakers tend to offer better odds for their wins. However, the objective
of this particular strategy is to place bets as “certain” as possible, and therefore we will
rule the “strong teams” out.
The rules for discarding “hot teams” to bet on at each match day are also stated in “The
Football Betting Mastermind”:
• Discard “hot teams” that play against other “hot teams”
• Discard “hot teams” when playing away against a “strong team” (this rule is not
applicable to our case, as we decided to exclude the “strong teams”)
47
• Discard “hot teams” when odds offered for the win are < 1.15 (Betclic and Bet.pt
do not allow single bets with odds lower than 1.20)
Our research indicates that the home advantage is a factor that increases the chances of
winning considerably, so we decided to add another rule:
• Discard “hot teams” when playing away
In the major leagues of England, Spain, Germany, Italy, France and Portugal (the ones
that we know better), we identified these as “hot teams” (at 14/02/2018):
England:
• Manchester City- 1st place, 14 matches played at home, 13 wins, 1 draw, 0 losses
-> home win % of 92,86 (equivalent to an odd of 1.08), 50 goals scored and 10
conceded
Spain:
• Barcelona- 1st place, 11 matches played at home, 9 wins, 2 draws, 0 losses ->
home win % of 81,82 (equivalent to an odd of 1.22), 31 goals scored and 5
conceded
Germany:
• Bayern Munich- 1st place, 11 matches played at home, 10 wins, 1 draw, 0 losses -
> home win % of 90,91 (equivalent to an odd of 1.10), 34 goals scored and 9
conceded
Italy:
• Napoli- 1st place, 12 matches played at home, 9 wins, 2 draws, 1 loss -> home win
% of 75 (equivalent to 1.33 odd), 29 goals scored and 8 conceded
• Juventus- 2nd place, 12 matches played at home, 10 wins, 1 draws, 1 loss -> home
win % of 83,33 (equivalent to an odd of 1.20), 30 goals scored and 4 conceded
France:
• PSG- 1st place, 12 matches played at home, 12 wins, 0 draws, 0 losses -> home
win % of 100, 46 goals scored and 7 conceded
48
Portugal:
• Porto- 1st place, 11 matches played at home, 10 wins, 1 draw, 0 losses -> home
win % of 90,91 (equivalent to an odd of 1.10), 34 goals scored and 7 conceded
• Benfica- 2nd place, (10 matches played at home, 9 wins, 1 draw, 0 losses -> home
win % of 90 (equivalent to an odd of 1.11), 31 goals scored and 5 conceded
• Sporting- 3rd place, 12 matches played at home, 10 wins, 2 draws, 0 losses ->
home win % of 83,33 (equivalent to an odd of 1.20), 26 goals scored and 4
conceded
Average win %: 87,57 (equivalent to an odd of 1.14)
In conclusion, this strategy means that before each match day, we bet on every “hot team”
that plays at home against other teams that are not considered “hot”, if the reward offered
by the bookmakers in case of success is higher than 15% of the stake placed.
49
6- Discussion of the results
6.1- Portuguese online fixed-odds betting markets
Figure 37- Bookmakers’ odds from the Portuguese online betting markets (right side) vs
Pinnacle’s (British bookmaker) odds (left side)
If we compare the odds proposed by 4 bookmakers operating in the Portuguese online

betting markets with the ones offered by Pinnacle, the first thing to notice is the
substantial difference between the average overrounds: 8,11% practiced by the
Portuguese bookmakers, against the 2,75% taken by Pinnacle. This means that the
bookmakers’ commission 5 is 5,36 percentage points (pp) higher in the Portuguese online
betting markets’ case, and consequently, there should be considerably less opportunities
for the punters to exploit. Furthermore, Casino Portugal (CP) was the bookmaker that
5 Please consult the 3rd page of our work
50
most often offered the better odds (12 times), followed by Estoril Sol Casinos, or ESC,
(10 times), Bet.pt (7 times) and Betclic (5 times).
ESC can be considered a very sui generis bookmaker, because it sets odds that are, in
most cases, the highest in the Portuguese online betting markets. Its average overround is
equal to 5,01%, a lower value in comparison with the 8,26%, 9,42% and 10,05%, charged
by Bet.pt, Betclic and CP, respectively. However, if one is betting on a single outcome,
and the odd is higher than 1.20, it automatically drops to values worse than the ones
offered by the other 3 bookmakers. Because of this, we believe that makes sense to
compare the 3 Portuguese bookmakers’ and Pinnacle’s overrounds without considering
ESC. In that scenario, the difference is even bigger (6,65%), as the average overround of
Bet.pt, Betclic and CP is equal to 9,40%.
As it is explained in chapter 3.3, the lack of competitiveness in the Portuguese betting

markets created an oligarchy “that is harmful to the punters interests, because bookmakers
do not feel much pressure to reward the risks taken by the punters with valuable odds”.
On the contrary, the British markets have so much offer that in order to stay competitive,
the bookmakers have to provide valuable odds, and therefore, to reduce drastically their
commissions, that is, the overrounds. Besides that, if we compare the average odds for
the 3 possible outcomes, in all cases, Pinnacle pays more than the 4 Portuguese
bookmakers (home win: 1.34 vs 1.31; draw: 6.33 vs 5.64 and away win: 12.49 vs 10.71).
Another explanation for the higher overrounds in the Portuguese bookmakers’ case is the
punitive tax of 8 to 16% over the volume of bets made, which means that they must pay
prizes and taxes for bets won by the punters, while the UK government taxes the
bookmakers in 15% of their profits (HM Revenue & Customs, 2017). This legal
framework “forces” the Portuguese bookmakers to increase the overround, and decrease
the odds, reducing the value offered to the market.
It is known that competitive markets deliver higher value to its customers than
uncompetitive markets, however, our investigation could not find any records related with
the Portuguese markets’ odds. Thus, we believe that this document adds a valid
contribution to the existing literature, as it provides numbers which support the idea that
Portuguese betting markets are, in fact, less competitive (at least in comparison with the
ones from Great Britain), and as a consequence, it seems reasonable to conclude that it is
more difficult for its customers (punters) to consistently profit, in the long-run.
51
In a survey that served as the basis for the Remote Gambling Association report
(Stradbroke, 2017), which was made to determine how Portugal’s regulated online market
was doing since its regulation, “68% of the respondents were still using gambling sites
not allowed to operate in Portugal. From these punters, 38% gambled solely through
unlicensed websites, while 30% admitted using both licensed and unlicensed online
gambling platforms. When the respondents were asked why they still use unlicensed
websites, the main reason mentioned was the wish to access better odds than the ones
offered by the Portugal-licensed gambling platforms” 6. These results suggest that punters
feel that Portuguese odds do not reward the risks taken in a fair way, and many prefer to
use unlicensed bookmakers, breaking the law and diminishing the government’s returns
on the sports betting business.
The findings of our study appear to coincide with the conclusions drawn in the Remote
Gambling Association’s reports (2017 and 2018), which claim that although the tax on
sports betting turnover is a significant source of income for the Portuguese government,
the current Portuguese betting regulation is harmful to both the punters’ and the
bookmakers’ interests, and in the words of RGA director Pierre Tournier, is “clearly
failing to combat the unregulated market, (…) change is much needed to make the
regulation work” (Stradbrooke, 2017). Hence, our research indicates that, from a long-
term strategic perspective, reviewing the decrees of Law nº 66 and 67/2015 would benefit
all 3 parts involved: the punters, the bookmakers and the Portuguese government itself.
6.2- Betting record
Figure 38- Betting record of the value strategy
6 Please consult the subchapter 3.3 of our work
52
Figure 39- Betting record of the accuracy strategy (4 Portuguese bookmakers vs Pinnacle)
As explained in 5.2, we started with an initial amount of 70,07€ for each one of the
strategies adopted. With the value strategy, we were able to obtain 13,93€, which
corresponds to a Return on Investment (ROI) of 19,88%. The betting accuracy rate is
equal to 72,73%, as we accurately predicted 8 out of 11 outcomes. In the Portuguese
online fixed-odds betting markets, the accuracy strategy was able to produce 0,28€, which
is equivalent to a ROI of 0,39%. This was accomplished by correctly forecasting 27 out
of 34 results (accuracy rate of 79,41%). At last, if we have used Pinnacle’s website to
execute the accuracy strategy, our money would have grown by 6,62%, that is, more 4,71€
than we began with. For convenience, from this point on, we will refer to this as if it was
another different strategy, simply called Pinnacle.
6.2.1- Portuguese online fixed-odds betting markets efficiency
“In general terms, the theory of efficient markets is concerned with whether prices at any
point in time fully reflect available information” (Fama, 1970: 413). The author describes
3 types of information-related market efficiency: strong-form efficiency, semi-strong-
form efficiency and weak-form efficiency (these concepts are defined and explained in
53
the beginning of subchapter 4.2). Kuypers (2000) defines abnormal returns as the ones
that are better than the bookmakers’ overround): if one uses all the publicly available
information to estimate probabilities for the outcomes of a match, and then is able to get
abnormal returns from it, makes sense to conclude that the markets at stake are semi-
strong efficient.
Theoretically, even negative returns can suggest semi-strong efficiency. For example,
assuming that all the bookmaker’s overround rates are equal and have a constant value of
9%, negative returns of -8% would indicate that the punter has developed a better aptitude
to estimate outcome probabilities than the market. In another publication, Constantinou
& Fenton (2013: 1) explain that “a gambling market is usually described as being
inefficient if there are one or more betting strategies that generate profit, at a consistent
rate, as a consequence of exploiting market flaws” 7. According to these theories, either
the accuracy strategy, the value strategy and Pinnacle were able to produce abnormal
returns, once that all the yield rates produced by following these strategies were positive.
In reality, the bookmakers’ overrounds are not equal and constant, thus one can use the
average overround as a benchmark to evaluate the returns obtained. The accuracy strategy
and Pinnacle would have to generate returns of at least - 8,10% and - 2,74%, respectively,
to fit in the Kuypers (2000) definition of “abnormal returns”. The former produced 0,39%
and the latter 6,62%, which are values clearly above the threshold set by this author, and
that, as he claims, indicate the existence of some form of semi-strong efficiency among
these 2 markets. We did not record the odds offered for each outcome of the matches used
in the value strategy, and as a consequence of that, we were unable to calculate the average
overround taken by the bookmakers, in this case. However, as this strategy generated a
ROI of 19,88%, it would certainly fit in the Kuypers (2000) notion of “abnormal returns”.
As it is stated in the subchapter 4.2 of our work, “an example of a well-documented

inefficiency in the fixed-odds betting markets is the favourite-longshot bias, which is a
type of systemic error where favourites tend to win more often than what is implied by
the odds (Cain et al., 2000). Vlastakis et al. (2009) found this bias in a number of big
online bookmakers. Based on the odds offered by several bookmakers over a period of 7
football seasons (from 2005/06 to 2011/12), Constantinou & Fenton (2013) demonstrated
54
the existence of a clear home advantage bias, whereby the returns produced by betting on
home wins were substantially higher in comparison with the ones obtained by placing
bets on away wins. In fact, the results of this study showed that this bias is almost equally
as strong as the favorite-longshot bias. By using a dataset of 79 446 matches, disputed in
21 leagues of 11 different European countries (from 2000 to 2001), Direr (2011) claims
that would have been possible to generate positive rates of return of 4.45% if best odds
across all the bookmakers were selected, and 2.78% if mean odds were selected, by
following a strategy of backing overwhelming favourites with odds of 1.19 (odds from
1.15 to 1.24 would have generated positive returns)”.
The accuracy strategy is about betting on home team favorites when the bookmakers offer
odds equal or higher than 1.15 8. If this strategy was able to produce positive returns, it
could suggest the presence of at least one of these biases (or both at the same time) in the
Portuguese online fixed-odds betting markets. As mentioned above, the accuracy strategy
was able to get 0,28€, which in theory can be seen as a positive return. However, being
the return so close to 0, one cannot say that such a low-income generation compensates
the degree of risk taken, and therefore, one cannot conclude that the Portuguese betting
markets have some form of inefficiency resulting from these biases, which can be
exploited by the punters to profit consistently, in the long-term.
When it comes to the British online bookmaker, the situation is different. Betting in the
same matches using Pinnacle’s platform would have provided 4,71€, which is equivalent
to a ROI of 6,62%, hence better than the one obtained by Direr (2011), if best odds were
chosen (although our number of observations is incomparably smaller). These numbers
are clearly above the 0 threshold and consequently, it is possible to interpret them as a
sign that at least one of these biases may exist in the odds offered by this bookmaker.
Figure 37 compares the odds of 4 Portuguese bookmakers vs Pinnacle, with and without
the overround. If one looks at odds without the overround (right side of the tables), it is
possible to perceive that Pinnacle offered better average odds for draws (6.51 vs 6.07)
and away wins (12.83 vs 11.49), while the Portuguese bookmakers offered better average
odds for home wins (1.41 vs 1.37). In contrast, if one considers the overround, Pinnacle
offered better average odds for the home wins than the 4 Portuguese bookmakers (1.34
8 Please consult the subchapter 5.2.2 of our work
55
vs 1.31).
This suggests that, if the favorite-longshot bias and/or the home advantage bias exist in
Pinncale’s odds, they also exist in the odds offered by the 4 Portuguese bookmakers. The
problem is that the average overround is much higher in the latter’s case (8,11% vs
2,75%), nullifying the possibility to take advantage of those biases, in a way that benefits
the punters’ interests. This is a clear example of the issues associated with the lack of
competitiveness in the Portuguese fixed-odds betting markets, which is largely related
with the country’s current online gambling laws (as discussed in 6.1).
6.2.2- The importance of our forecasting method in appraising the expected value
The importance of doing a proper qualitative analysis to all the determinants of a match
that are not easily quantifiable has been discussed over the course of our paper. In
Rebelo’s (2012) opinion, this type of evaluation is even more important than the
quantitative analysis, and in case of conflict, should prevail over the latter. Forrest et al.
(2005) claim that it is very probable that bookmakers consider subjective factors
(determinants that are important for prediction but which historical data fails to capture)
when setting the odds, and Constantinou et al. (2012) developed a Bayesian framework,
which took into account both objective and subjective variables. The model was evaluated
though accuracy and profitability measurements, and it was concluded that the subjective
information improved significantly its performance. At the date, these authors even stated
that, in terms of profitability, and over such a long period of time (one full season), no
other published work appeared to be as successful as their own 9.
In the value strategy, we did not follow some set of pre-determined rules to decide which
bets to make. Instead, our focus was to maximize the expected value of each bet, through
a combination of analytic methods and subjective interpretations of the information that
is not expressed in numbers 10. The bets made were individually selected after going
through a detailed analysis, with well-defined criteria. That analysis is presented in the
form of a report, whose contents are described in the subchapter 5.2.1.1. By following
this strategy, we were able to obtain a ROI of 19,88% (as mentioned above), therefore
19,49 pp and 13,26 pp higher than the ones from the accuracy strategy and Pinnacle,

10 Please consult the subchapter 5.2.1 of our work
56
respectively, although the betting accuracy rate was 6,68% lower in the value strategy’s
case (72,73% vs 79,41%). Furthermore, the latter’s average odd is considerably high
(1.64) in comparison with the ones from other 2 strategies (1.31 and 1.34).
Starting from Rebelo’s (2000) concept of positive expected value 11, the average odd of
the value strategy implies substantially more risk exposure than the accuracy strategy or
Pinnacle’s mean odds, at least according to the bookmakers’ assessment. On the other
hand, it also entails higher returns when the bets made are accurate. The average profit
obtained by the accuracy strategy corresponds to 0,01€ (virtually none), which means that
its expected returns, based on past trades, is approximately null. The standard deviation
is 4,34€, that is, the observations tend to deviate from the mean about that value, up or
down. Investors can use the standard deviation as a measure of risk: the more
unpredictable the price action and the wider the range, the greater the risk. The coefficient
of variation characterizes the relation between the standard deviation and the mean,
determining how much risk is assumed in comparison with the amount of return expected:
lower values indicate a better risk-return trade-off. Once that the average is so close to 0,
in this case we should not consider the coefficient of variation, as it is too high for the
interpretation to be meaningful.
Pinnacle would have generated an average profit of 0,14€, with a standard deviation equal
to 4,61€. The value strategy, by its turn, produced, on average, 1,27€ for each bet placed,
and its standard deviation is 6,05€, thus suggesting a higher risk than the one from
Pinnacle (and consequently, also than the one from the accuracy strategy).
These results seem to indicate that, through a combination of analytic methods and
subjective interpretations of all the aspects that can affect the outcomes of the matches
selected, it is possible to beat the bookmakers which operate in the Portuguese online
fixed-odds betting markets. Despite having a lower accuracy rate in comparison to the
accuracy strategy (and Pinnacle), the value strategy produced much higher returns.
Therefore, our forecasting method appears to increase the average returns while helping
to better appraise and control the degree of risk assumed, hence maximizing the expected
value of the bets made. However, the subjectivity of this strategy requires knowledge and
specialization to predict the events accurately, and as a consequence of that, obtain
11 Please consult the 4th page of our work
57
positive returns.
Nonetheless, it is important to highlight that we only have 11 observations for the value
strategy, which is less than 3 times the number of observations for the accuracy strategy
(and Pinnacle). Having this in consideration, we cannot take definitive conclusions about
the contribution added by our forecasting method, and thus about the value strategy’s
ability to generate profits in the long-term. The reason for such a small number of
observations is that each forecasting report, being so extensive, took several days to write,
which constitutes a problem to the user-friendliness of our method. In order to clearly
demonstrate to the readers how to perform this analysis, we had to do it in this way,
however, the punters can use our reports as a model to do their own assessment, without
needing to write such a detailed evaluation of all the situations that may condition the
outcome of a football game.
58
7- Conclusions
7.1. Summarizing the value strategy
The main objective of this project was to develop an uncomplicated method, which can
be used by ordinary punters to consistently obtain long-term positive returns. To achieve
that, our method would have to be able to accurately predict the outcomes of matches
with potentially valuable odds, within the Portuguese online fixed-odds betting markets.
The forecasting of the football games was made combining quantitative information
(including a Poisson regression model) with a qualitative analysis (thus mostly
subjective) of the chosen matches. This approach was named value strategy, being able
to generate a Return on Investment of nearly 20%, and to correctly predict 8 out of 11
results (which corresponds to an accuracy rate of 72,73%).
7.1.1- Limitations and recommendations for further research
Although the value strategy was able to produce significant yields, the number of bets
made is too small to conclude that it is, in fact, capable of providing consistent earnings
in the long-run. Besides that, due to the subjective nature of decision-making, it would be
interesting to perceive if our forecasting method works for other punters and competitions
(once that we focused on the Premier League’s matches). Therefore, we will divulgate
this paper in betting forums, and as a suggestion for further research, we challenge all the
interested punters to use our forecasting report as a benchmark to make betting decisions,
and subsequently, to give feedback about the returns obtained over a period of, at least, 1
entire season.
7.2- Summarizing the accuracy strategy
The accuracy strategy, which is based on “The Football Betting Mastermind” (Betinfo24:
2010), consisted in betting on the teams which were expected to win almost every match,
once that according to this document, it is very unlikely that an individual punter is able
to find enough value, in the major leagues’ odds, to consistently beat the bookmakers.
Summing up, this strategy is driven by accuracy, and not by finding positive expected
value. The value strategy provided returns of 0,39%, thus not compensating for the level
of risk assumed.
In spite of having an accuracy rate of 79,41%, that is higher than the one from the value
59
strategy, the returns obtained by the former were incomparably smaller, which suggests
that making an individual appreciation of the matches confers protection against risk
exposure, while increasing the expected returns. Comparatively, the value strategy’s
average odd implies much higher risks, while its returns imply a much higher expected
value, in relation to the accuracy strategy. Perhaps if we have used the pre-determined
rules of the latter to identify the matches with betting potential and then, from those,
carefully selected the less risky alternatives, we would have been closer to the accuracy
rate of 90%, which according to “The Football Betting Mastermind” (Betinfo24: 2010),
is the benchmark to get significantly positive returns.
7.3- Portuguese online fixed-odds and markets efficiency
As we decided to bet only on overwhelming favourites who played at home, if the

accuracy strategy was able to create positive returns, it would indicate the existence of
the well-documented favourite-longshot and/or home advantage biases in the Portuguese
online fixed-odds betting markets, and consequently, that these markets were semi-strong
efficient. In practice, the accuracy strategy broke even, therefore not suggesting
exploitable biases in the odds proposed by the 4 Portuguese bookmakers. However, if we
have made the same bets through an UK-based online bookmaker, the rate of return would
be equal to 6,62%, hence indicating the presence of this biases in the odds set by Pinnacle.
Amazingly, if we remove the weight of the overround from the odds, the Portuguese
bookmakers would have offered better odds for the home wins than Pinnacle. This
example clearly expresses the necessity of reviewing the Portuguese gambling legal
framework (as has been claimed by punters and betting agencies), which was identified
as the main reason for the poor competitiveness of Portuguese fixed-odds betting markets,
in comparison with other countries. As far as our research went, we could not find any
literature, studies nor even a record of past odds, about the Portuguese fixed-odds betting
markets. Therefore, we consider that our work adds an important contribution, in the
sense that it provides information about the odds offered by the Portuguese online betting
markets in general, and also about each one of the 4 Portuguese bookmakers considered
in this paper. Moreover, it offers a comparative analysis with the odds proposed by an
UK-based online bookmaker, which highlights the lack of competitiveness of the
Portuguese betting context.
60
7.4- Final comment
In spite of the difficulties encountered, if one is willing to invest a considerable amount

of time and effort, we believe that it is indeed possible to find profitable opportunities
within the Portuguese online fixed-odds. Rebelo (2012) states that the ability to accurately
forecast football events continuously improves with practice, although not everyone is
able to do it, being necessary to possess an innate gift, as it happens with professional
sports players, for example. However, the number of licenced online bookmakers has
been growing in Portugal, and as a consequence of that, the value implied by the odds is
expected to increase, as well, creating new opportunities to exploit.
61
8- References
8.1- Books
• Cameron, A. C., & Trivedi, P. K. 1998. Regression analysis of count data.

Cambridge (UK): Cambridge Univeristy Press.
• Clarke, G. M., & Cook, D. 2004. A basic couse in statistics. London: Arnold.
• Curto, J. 2016. Estatística, muitas aplicações em Excel e poucas fórmulas. Lisbon:
GUIDE- Artes Gráficas, Lda.
• Neave, N. & Wolfson, S. 2005. The Home Advantage: Psychological and
Physiological Factors in Soccer. In D. Lavalee, J. Thatcher & M. Jones (Eds.), Coping
and Emotion in Sport: 131-148. Hauppauge (NY): Nova Science Publishers.
• Rebelo, P. 2012. Ganhar com as apostas desportivas. Lisbon: Marcador.
• Weiss, N. A. 2006. A course in probability. Boston: Pearson Education Inc.
8.2- Periodicals
• Buraimo, B., Peel, D., & Simmons, R. 2013. Systematic Positive Expected Returns
in the UK Fixed Odds Betting Market: An Analysis of the Fink Tank Predictions.
International Journal of Financial Studies, 1(4): 168–182.
• Cain, M., David, L., & Peel, D. 2000. The favourite-longshot bias and market
efficiency in UK football betting. Scottish Journal of Political Economy, 47(1): 25–
36.
• Constantinou, A. C., & Fenton, N. E. 2013. Profiting from arbitrage and odds biases
of the European football gambling market. The Journal of Gambling Business and
Economics, 7(2): 41–70.
• Constantinou, A. C., Fenton, N. E., & Neil, M. 2012. Pi-football: A Bayesian network
model for forecasting Association Football match outcomes. Knowledge-Based
Systems, 36: 322–339.
• Courneya, K. S. & Carron, A. V. 1992. The Home Advantage in Sport Competitions:
A Literature Review. Journal of Sport & Exercise Psychology, 14: 13-27.
• Direr, A. 2011. Are Betting Markets Efficient ? Evidence from European Football
Championships. Applied Economics, 45(3): 343–356.
• Dixon, M. J., & Coles, S. G. 1997. Modelling Association Football Scores and
Inefficiencies in the Football Betting Market. Journal of the Royal Statistical
Society: Series C (Applied Statistics), 46(2): 265–280.
• Fama, E. F. (1970). Efficient Capital Markets- A Review of Theory and Empirical
Work. The Journal of Finance, 25(2): 383–417.
• Forrest, D., Goddard, J., & Simmons, R. 2005. Odds-setters as forecasters: The case
62
of English football. International Journal of Forecasting, 21(3): 551–564.

• Forrest, D., McHale, I., & McAuley, K. 2008. “Say It Ain’t So”: Betting-Related
Malpractice in Sport. International Journal of Sport Finance, 3(3): 156–166.
• Franck, E., Verbeek, E., & Nüesch, S. 2010. Prediction accuracy of different market
structures- bookmakers versus a betting exchange. International Journal of
Forecasting, 26(3): 448–459.
• Goddard, J. 2005. Regression models for forecasting goals and match results in
association football. International Journal of Forecasting, 21(2): 331–340.
• Goddard, J., & Asimakopoulos, I. 2004. Forecasting Football Results and the
Efficiency of Fixed-odds Betting. Journal of Forecasting, 23(1): 51–66.
• Heuer, A., & Rubner, O. 2009. Fitness, chance, and myths: An objective view on
soccer results. European Physical Journal B, 67(3): 445–458.
• Hvattum, L. M., & Arntzen, H. 2010. Using ELO ratings for match result prediction
in association football. International Journal of Forecasting, 26(3): 460–470.
• Kelly, J. L. 1956. A New Interpretation of Information Rate. Bell System Technical
Journal, 35(4): 917–926.
• Kuypers, T. 2000. Information and efficiency: An empirical study of a fixed odds
betting market. Applied Economics, 32(11): 1353–1363.
• Levitt, S. D. 2004. Why are gambling markets organized so differently from financial
markets? Economic Journal, 114(3): 223–246.
• Maher, M. J. 1982. Modelling association football scores. Statistica Neerlandica,
36(3): 109–118.
• Mansuy, R. 2009. The origins of the word “martingale.” Electronic Journal for
History of Probability and Statistics, 5(169): 105–113.
• McSharry, P. E. 2007. Altitude and athletic performance: Statistical analysis using
football results. British Medical Journal, 335(7633): 1278–1281.
• Neave, N., & Wolfson, S. 2003. Testosterone, territoriality, and the “home
advantage.” Physiology and Behavior, 78(2): 269–275.
• Page, L., & Page, K. 2007. The second leg home advantage: Evidence from European
football cup competitions. Journal of Sports Sciences, 25(14): 1547–1556.
• Pollard, R. 1986. Home advantage in soccer: A retrospective analysis. Journal of
Sports Sciences, 4(3): 237–248.
• Pollard, R. 2006. Worldwide regional variations in home advantage in association
football. Journal of Sports Sciences, 24(3): 231–240.
• Pollard, R. 2008. Home Advantage in Football: A Current Review of an Unsolved
Puzzle. The Open Sports Sciences Journal, 1(1):
• Pollard, R., & Pollard, G. 2005. Long-term trends in home advantage in professional
63
team sports in North America and England. Journal of Sports Sciences, 23(4): 337–
350.
• Reep, C., Pollard, R., & Benjamin, B. 1971. Skill and Chance in Ball Games. Journal
of the Royal Statistical Society, 134(4): 623–629.
• Seckin, N., & Pollard, R. 2008. Home Advantage in Turkish Professional Soccer.
Perceptual and Motor Skills, 107: 51–54.
• Thaler, R. H., & Ziemba, W. T. 1988. Anomalies: Parimutuel Betting Markets:
Racetracks and Lotteries. Journal of Economic Perspectives, 2(2): 161–174.
• Thomas, S., Reeves, C., & Smith, A. 2006. English Soccer Teams Aggressive
Behaviour When Playing Away From Home. Perceptual and Motor Skills, 1023: 17–
320.
• Vlastakis, N., Dotsis, G., & Markellos, R. N. 2009. How efficient is the European
football betting market? Evidence from arbitrage and trading strategies. Journal of
Forecasting, 28(5): 426–444.
• Waters, A., & Lovell, G. 2002. An Examination of the Homefield Advantage in a
Professional English Soccer Team from a Psychological Standpoint. Football
Studies, 5(1): 46–59.
8.3- Unpublished works
• Gardner, J. A. 2011. Modeling and Simulating Football Results. Unpublished

assignement in mathematics, University of Leeds School of Mathematics, UK.
Retrieved from https://www1.maths.leeds.ac.uk/~voss/projects/2010-
sports/JamesGardner.pdf
• Heino, O., & Sillanpää, V. 2013. Forecasting football match results - A study on
modeling principles and efficiency of fixed-odds betting markets in football.
Unpublished Master of Science dissertation, Aalto University School of Business,
Helsinki. Retrieved from
http://epub.lib.aalto.fi/fi/ethesis/pdf/13445/hse_ethesis_13445.pdf
8.4- Websites
• Apostas Online. (2017). Porque razão há tão poucas casas de apostas em Portugal?
Retrieved November 18, 2017, from https://apostasonline.pt/porque-razao-ha-tao-
poucas-casas-de-apostas-em-portugal/
• Batista, Rodrigo. (2017). Apostas desportivas: O estado actual. Retrieved January 21,
2018, from https://blogvisaodemercado.pt/2017/12/apostas-desportivas-o-estado-
actual/
• Betinfo24. (2010). The Football Betting Mastermind. Retrieved November 16, 2017,
from http://betinfo24.co.uk/FootballMastermindB24.pdf
• Correio da Manhã. (2018). Site “A Nossa Aposta” vai ter apostas desportivas.
Retrieved May 27, 2018, from http://www.cmjornal.pt/tv-media/detalhe/site-a-nossa-
64
aposta-vai-ter-apostas-desportivas
• Cotrim, António. (2018). Site de apostas “placard.pt” já abriu e vai canalizar lucros
para causas sociais. Retrieved June 14, 2018, from
https://observador.pt/2018/06/11/site-de-apostas-placard-pt-ja-abriu-e-vai-canalizar-
lucros-para-causas-sociais/
• European Gaming and Betting Association. Sports Betting: Commercial and Integrity
Issues. Retrieved November 22, 2017, from http://www.egba.eu/media/Sports-
Betting-Report-FINAL.pdf
• HM Revenue & Customs. (2017). General Betting Duty, Pool Betting Duty and
Remote Gambling Duty. Retrieved September 28, 2018, from
https://www.gov.uk/guidance/general-betting-duty-pool-betting-duty-and-remote-
gaming-duty
• Jornal Económico. (2017). Direitos de TV na Premier League: último classificado
encaixou mais do que o Benfica. Retrived October 24, 2018, from
https://jornaleconomico.sapo.pt/noticias/direitos-de-tv-na-premier-league-ultimo-
classificado-encaixou-mais-do-que-benfica-166640
• Kucharski, Adam. (2016). Betting and investment both require skill and luck.
Retrieved November 15, 2017, from https://www.ft.com/content/a472722e-0e25-
11e6-b41f-0beb7e589515
• Macário, José. (2017). Direitos televisivos: como são distribuídos nas principais
ligas? Retrived October 24, 2018, from
https://jornaleconomico.sapo.pt/noticias/direitos-televisivos-como-sao-distribuidos-
nas-principais-ligas-193225
• Madremedia/Lusa. (2017). Jogo 'online': Associação internacional pede "revisão
urgente" do regulamento português. Retrieved November 18, 2017, from
http://24.sapo.pt/atualidade/artigos/jogo-online-associacao-internacional-pede-
revisao-urgente-do-regulamento-portugues
• Moody, Allen. (2017). History of Sports Betting. Retrieved November 15, 2017, from
https://www.thoughtco.com/history-of-sports-betting-3116857
• O’Shaughnessy, D. 2011. Optimal Exchange Betting Strategy for Win-Draw-Loss
Markets. Retrieved December 12, 2017, from
http://www.rankingsoftware.com/research/OptimalBettingStrategyWinDrawLoss.pd
f
• Schoch, David. (2017). Home-field Advantage. Retrieved January 24, 2018, from
http://soccerverse.com/articles/home_field_advantage
• Silva, Ana Rute. (2014). Operadores dizem que impostos sobre jogo online tornam
“inviável” acesso ao mercado. Retrieved November 18, 2017, from
https://www.publico.pt/2014/07/03/economia/noticia/operadores-dizem-que-
impostos-sobre-jogo-online-tornam-inviavel-acesso-ao-mercado-1661529
• Stradbrooke, Steven. (2017). RGA: Portugal’s regulated online gambling market a
failure. Retrived July 5, 2018, from
65
https://calvinayre.com/2017/12/06/business/portugal-onling-gambling-market-
failure/
• Stradbrooke, Steven. (2018). Portugal’s online gambling market closed out 2017 on
a high. Retrieved July 5, 2018, from
https://calvinayre.com/2018/02/16/business/portugal-online-gambling-record-
revenue/
66
Annexes
Annex I- Watford vs Everton (match day 28)
• 24/02/2018 at 5:30 pm, Vicarage Road Stadium (Watford)
• Weather: 3ºC; precipitation probability of 0%; humidity percentage of 58% and

wind speed of 21 km/h
• Analysis: Weather as perfect as it gets to play football in the UK, at this time of
the year. It allows for a fast circulation of the ball, and consequently for an open
match.
Potential bets:
• Over 2.5 goals in the match- 1.88
• Both teams score- 1.71 (yes) and 1.80 (no)
• Watford or draw and over 1.5 goals in the match- 1.70
• Watford or draw and both teams score- 2.20
Probable line-ups:
• Watford- 3-4-2-1: Gomes; Mariappa, Prodl and Holebas; Femenía, Capoue,

Doucouré and Zeegelaar; Deulofeu and Richarlison; Deeney
• Everton- 4-2-3-1: Pickford; Coleman, Keane, Williams and Martina; Gueye and
Schneiderlin; Walcott, Davies and Sigurdsson; Tosun
Injuries and suspensions:
• Watford- Out: Hoban, Catchcart and Chalobah; doubtful: Kabasele and Cleverly
(all injured)
• Everton- Out: Baines, McCarthy, Stekelenburg, Mangala; doubtful: Coleman,

Gueye and Jagielka (all injured)
• Analysis- No key players missing on both teams, as many of these are virtually
chronically injured.
67
Discipline table:
• Watford (average per match at home)- Yellow cards: 1.9, red cards: 0.2, booking
points: 23.1
• Everton (average per match away)- Yellow cards: 1.4, red cards: 0.1, booking
points: 15.8
• Premier League (team average per match overall)- Yellow cards: 1.6, red cards:
0.1, booking points: 16.7
• Analysis- At home, Watford is more aggressive than the average of Premier

League teams, while Everton is below the average in terms of aggressiveness,
when playing away. This means that Watford may win more duels and recover
the ball more swiftly, but on the other hand, The Hornets are probably going to be
more exposed to disciplinary conditioning through yellow or red cards.
Head-to-head over the last 3 years:
• 5 matches- 2 wins for Everton (40%), 2 draws (40%) and 1 win for Watford
(20%); 4 times over 1.5 (80%) and 1 time under (20%); 3 times over 2.5 (60%)
and 2 times under (40%); neither of the teams won away
• Analysis- There is not a significant bias favouring any of the teams.
Premier league tables overall, home and away:
• Overall- Watford: 11th place, 30 points, 37 goals scored and 47 conceded;

Everton: 9th place, 34 points, 32 goals scored and 46 conceded
• Home- Watford: 18th place, 16 points, 20 goals scored and 26 conceded
• Away- Everton: 17th place, 8 points, 9 goals scored and 28 conceded
• Analysis- Both teams performed in a very similar way this season, with a big
difference: while Watford does not rely much on the home advantage (won 16
points at home and 14 when playing away), Everton is hugely dependent on this
factor (only 1 win away from Goodison Park). This can influence the match in
Watford’s favour.
68
Home/away performance tables:
• Watford- won 53% of its points at home and 47% when playing away, scored
54% of its goals at home and 46% when playing away, conceding 55% of its goals
at home and 45% away
• Everton- won 76% of its points at home and 24% when playing away, scored
• Premier League- On average, 57% of the goals were scored at home and 43%
were scored away
• Analysis- This table provides numerical evidence that Everton is very reliant on
the home advantage, even in comparison with the Premier League’s average.
Points per game tables:
• Watford has a relative form of -0.11, while Everton scores -0.26. The obvious
conclusion is that none of the teams have performed good recently, but Everton
has been even less consistent than Watford
Scoring tables:
• Watford scores, on average, 1.45 goals per match at home, while Everton
concedes 1.70 goals per match when playing away. If we average these numbers,
we obtain Watford’s “scoring strength”, which corresponds to 1.62 goals. Besides
that, Watford’s matches at home ended with over 2.5 goals 69% of the times
• Everton scores, on average, 1.19 goals per match when playing away, while
Watford concedes 2.00 goals per match at home. If we average these numbers, we
obtain Everton’s “scoring strength”, which corresponds to 1.60 goals. Besides
that, Everton’s away matches ended with over 2.5 goals 46% of the times
• Analysis- If we sum Watford and Everton scoring strengths, we obtain a total of

3.22 goals expected in the forthcoming match. Besides that, if we average both
teams’ percentage of matches that ended with over 2.5 goals, we get a 57.5%
probability that this match in particular ends with over 2.5 goals, which can be
converted into an odd of 1.74.
69
Team characteristics:
• Both teams are strong at stealing the ball from the opposition and coming back
from losing positions (Everton is even stronger in the latter)
• Watford is very weak at protecting the lead and defending counter-attacks. This
team uses an offside trap defensive strategy, which means that the defensive line
must be positioned high in the field. As Everton has very fast players (e.g.
Walcott), and attacks mostly through long balls, there might be some space to
exploit behind Watford’s defensive line
• Everton is very weak at defending against skilful players. Watford’s front line is
rich in players with such characteristic (e.g. Deulofeu and Richarlison)
• Both teams are weak at avoiding fouling in dangerous areas, so this match might
be prone to harmful individual mistakes from both sides
• Watford likes to attack down the left, while Everton likes to attack down the right,
so we can expect most of the action to be developed on that wing. However,
Everton is very weak when it comes to defend attacks down the flanks
Team psychology:
• Team spirit and motivation- As this is a direct confrontation between two teams
with the same ambitions in terms of classification, and with just 4 points
separating Watford from Everton, both teams should be highly motivated to win
the match
• Confidence- In spite of being highly motivated for this match in particular, both
teams are very inconsistent, as they easily go from an incredible victory to a
surprising defeat in the following match day. Besides that, both teams have
problems in their defensive processes, hence we believe that their confidence
levels should be classified as low
• Managerial impact- Both clubs have experienced managerial sackings over the
course of the season. Everton sacked Ronald Koeman after the defeat at home
against Arsenal (2-5), on the 22nd October, that left the Liverpool based side in
relegation zone (18th place). On the other hand, Marco Silva was fired on the 20th
70
January, after suffering a categorical defeat against Leicester (2-0 away). In spite
of being criticized by many for his outdated style of play, Sam Allardyce was able
to put Everton in 9th place, with 5 wins, 4 draws and 4 losses. Because of this clear
improvement in Everton’s results, we consider that he had a high impact on the
team’s performance. New Watford’s boss Javi Gracia started by drawing with no
goals away against Stoke, achieved an incredible win at home against Chelsea (4-
1), but then lost away to direct competitor West Ham (2-0). So far, we consider
that his managerial impact on Watford has been normal
Team fatigue:
• Toughness of previous match- Watford loss 2-0 away against West Ham, while
Everton won 3-1 at home against Crystal Palace. At match day 26, Everton had
31 points, Watford had 30, West Ham and Crystal Palace both had 27, and none
was safe in terms of avoiding relegation. The toughness of these matches is
considered high because the teams involved were direct competitors for the safety
of the mid-table places
• Resting- Both teams are very rested, as the last Premier League’s match day
finished at 12/02/2018
• National team participation- The last round of international matches was in

November of 2017, and so this will not be considered
71
0 1 2 3 4 5 6 7+ Home
0 5,16% 7,02% 4,77% 2,16% 0,74% 0,20% 0,05% 0,01%
1 8,27% 11,26% 7,66% 3,47% 1,18% 0,32% 0,07% 0,02%
2 6,64% 9,03% 6,14% 2,79% 0,95% 0,26% 0,06% 0,01%
3 3,55% 4,83% 3,28% 1,49% 0,51% 0,14% 0,03% 0,01%
4 1,42% 1,94% 1,32% 0,60% 0,20% 0,06% 0,01% 0,00%
5 0,46% 0,62% 0,42% 0,19% 0,07% 0,02% 0,00% 0,00%
6 0,12% 0,17% 0,11% 0,05% 0,02% 0,00% 0,00% 0,00%
7+ 0,03% 0,05% 0,03% 0,01% 0,00% 0,00% 0,00% 0,00%
Away
5,16% 11,26% 6,14% 1,49% 0,20% 0,02% 0,00% 0,00%
Draw 24,27% Odd 4,12
Total 100,00%
Under 2.5 43,12% Odd 2,32
Over 2.5 56,88% Odd 1,76
Total 100,00%
• Analysis- Our Poisson regression model grants more chances to an away win, in
contrast with the bookmaker’s odds, and the descriptive statistics as well. This is
probably because in 2016/2017, Everton performed very good, finishing in 7th
place, while Watford ended 17th, just 6 points above the relegation zone. It is
possible to detect a data pattern that is relevant for betting on over 2.5 goals, for
which the model provides a probability that is very similar to the one provided by
the descriptive statistics (56,88% and 57,50%, respectively).
Betting decision:
• Our analysis indicates that we can expect an open match, as both teams should be
very motivated to win, in spite of the low confidence levels that are a consequence
of a very inconsistent defensive process. Usually, Everton has troubles when
playing away, thus we believe in a home win or draw. Watford and Everton have
attacking players with high quality, that can be decisive against most of the
Premier League’s teams. Our descriptive scoring analysis determines a probability
of 57.5% to the match ending with over 2.5 goals, and the Poisson regression
model determines a 56.88% probability to the same outcome. According to our
72
judgement, this means that an offered odd above 1.76 on over 2.5 goals would
have positive expected value. Therefore, we think that betting on over 2.5 goals
at 1.88 is the best possible choice.
73
Annex II- Arsenal vs Manchester City (match day 28)
• 01/03/2018 at 7:45 pm, Emirates Stadium (London)
• Weather: -2ºC; precipitation probability of 48%; humidity percentage of 64% and

• Analysis: Acceptable weather conditions to play football in the UK, at this time
of the year. The match was at risk due to the eastern cold wave that painted in
white the Emirates Stadium surrounding area, however, it is equipped with
undersoil heating, so there is no threat of the pitch being frozen and, as a
consequence of that, unplayable.
Potential bets:
• Manchester City- 1.75
Probable line-ups:
• Arsenal- 4-3-3: Cech; Bellerín, Koscielny, Mustafi and Kolasinac; Ramsey,

Elneny and Wilshere; Ozil, Aubameyang and Mkhitaryan
• Manchester City- 4-3-3: Ederson; Walker, Stones, Laporte and Danilo; De

Bruyne, Gundogan and D. Silva; B. Silva, Aguero and Sané
• Arsenal- Out: Cazorla, Lacazette and Monreal (all injured)
• Manchester City- Out: Mendy and Fernandinho (injured), Delph (suspended);

doubtful: Sterling (injured)
• Analysis- 2 important players injured on each side: Lacazette and Monreal for
Arsenal, Fernandinho and Sterling for Manchester City.
Discipline table:
• Arsenal (average per match at home)- Yellow cards: 1.2, red cards: 0.0, booking
points: 12.3
• Manchester City (average per match away)- Yellow cards: 1.9, red cards: 0.1,
74
booking points: 21.2
• Analysis- At home, Arsenal is less aggressive than most Premier League’s teams,
while Manchester City is way above the average in terms of aggressiveness, when
playing away. This means that The Citizens may win more duels and recover the
ball more swiftly, but on the other hand, they are probably going to be more
exposed to disciplinary conditioning through yellow or red cards.
• 11 matches- 4 wins for Arsenal (36,36%), 4 draws (36,36%) and 3 wins for
Manchester City (27,28%)
• Analysis- There is not a significant bias favouring any of the teams. On the 25 th
of February, Manchester City won the Carabao Cup by defeating Arsenal 3-0 at
Wembley, in a match were The Citizens clearly overpowered The Gunners.
• Overall- Arsenal: 6th place, 45 points, 51 goals scored and 36 conceded;

Manchester City: 1 st place, 72 points, 79 goals scored and 20 conceded
• Home- Arsenal: 3rd place, 32 points, 36 goals scored and 14 conceded
• Away- Manchester City: 1st place, 32 points, 29 goals scored and 10 conceded
• Analysis- Manchester City has been, by far, the strongest team in the Premier
League this season, placing 1 st with the best goal difference at home and away.
On the other hand, Arsenal has been very consistent at home, being able to match
Manchester City’s away performances. However, Arsenal relies too much on the
home advantage, as around 71% of its Premier League points were earned when
playing at the Emirates Stadium.
• Arsenal- won 71% of its points at home and 29% when playing away, scoring
also 71% of its goals at home and 29% when playing away, conceding 39% of its
75
goals at home and 61% away
• Manchester City- won 56% of its points at home and 44% when playing away,
scored 63% of its goals at home and 37% when playing away, conceding 50% of
its goals at home and 50% away
were scored away
• Analysis- This table provides numerical evidence that Arsenal is very reliant on
Manchester City achieved very good results away, and even better results at home,
hence we can assume that the home advantage has its influence on the team’s
performance, but The Citizens do not depend on it to obtain good results.
• Arsenal has a relative form of -0.29, while Manchester City scores -0.54. At first
glance, it seems weird that Arsenal has a less negative relative form in comparison
with City. This happens because Manchester City has been very consistent over
the course of the season, only losing against Liverpool, and drawing against
Everton, Crystal Palace and Burnley. All these matches are included in the last 8
matches, with the exception of Manchester City vs Everton
Scoring tables:
• Arsenal scores, on average, 2.77 goals per match at home, while Manchester City
we obtain Arsenal’s “scoring strength”, which corresponds to 1.77 goals
• Manchester City scores, on average, 2.33 goals per match when playing away,
while Arsenal concedes 1.08 goals per match at home. If we average these
numbers, we obtain Manchester City’s “scoring strength”, which corresponds to
1.71 goals
• Both teams like to play a possession-based football, using short passes to control
the game in the opposition’s half, but usually Wenger adapts Arsenal style of play
76
when facing Guardiola’s teams. Therefore, we should expect Manchester City to

dominate this match, against a more speculative Arsenal. City defends high on the
pitch, using the offside trap strategy. Although Manchester City has this process
very well automatized, the team is prone to individual mistakes (like Arsenal), so
there might be some space to exploit behind The Citizens’ defensive line, as
Arsenal has very fast offensive players (e.g. Aubameyang and Mkhitaryan)
• Arsenal creates a lot of scoring chances, and Manchester City is very weak when
it comes to stopping opponents from creating chances (the opposite is also truth).
Arsenal is strong at finishing scoring chances, while Manchester City is very
strong
• Manchester City and Arsenal are very strong at creating chances using through
balls. However, Arsenal is weak at defending against through balls
• City is strong in counter-attacks, while Arsenal is weak at defending them
• Arsenal is weak at protecting the lead, but is strong at coming back from losing
positions
Team psychology:
• Team spirit and motivation- Pep Guardiola promotes a squad rotation policy, so
that every player feels useful and integrated. There are no cases (at least publicly
known) of indiscipline or internal issues between the players and the manager.
The way the team is performing (and winning) provides strong evidence of a
positive team culture. Apart from that, Manchester City should be highly
motivated to win the Premier League as soon as possible, in order to then focus
its full attention on the latter stages of the Champions League. On the contrary,
Arsenal is 8 points above the 7th placed Burnley and 10 points below the
Champions League spots, so the team’s faith is more or less defined. Arséne
Wenger is a spent manager, that can no longer produce good results and inculcate
a positive spirit among the players. Nonetheless, Arsenal is hurt in its pride after
the humiliating defeat in the League Cup final, and surely will try to clean up the
image left in that match. For those reasons, we classify Arsenal’s team spirit and
motivation as normal
77
• Confidence- In 2017/2018, Manchester City has been perhaps the best team in
Europe, only losing to Shakhtar Donetsk (Champions League), Wigan (FA Cup)
and Liverpool (Premier League), and tying against Everton, Crystal Palace and
Burnley (all for Premier League). On the last 25th of February, City won 3-0 to
Arsenal, conquering the League Cup. On these circumstances, we believe it is fair
to say that The Citizens’ confidence levels must be very high. Arsenal, by its turn,
is performing below the expectations, ranking 6th in the Premier League, and
coming from traumatic losses in the Northern London derby (vs Tottenham) and
in the League Cup final (vs Manchester City). Therefore, Arsenal’s confidence
levels should be low at this point
• Managerial impact- Although Arsenal do not win the Premier League since
2003/2004, Arséne Wenger has been in charge for over 22 years. For better or
worse, his impact on the club has to be considered very high. Guardiola arrived at
Manchester and after a mediocre first season, has been able to turn Manchester
City into one of the most complete teams in Europe. The financial investment in
the squad was huge, but nowadays City plays in the image of its coach, and is
definitely one of the best sides that he built in the entire career. That being said,
we have to consider Guardiola’s impact on the club as very high, as well
Team fatigue:
• Toughness of previous match- Arsenal loss the final of the League Cup 3-0
against Manchester City. Considering this, we classify the toughness of Arsenal’s
last match as the highest, and the toughness of Manchester City’s last match as
low
• Resting- Both teams last match was against each other, for the League Cup final
(4 days ago), as mentioned above. From today’s probable line-up, the only players
that rested were the goalkeeper Ederson, and defenders John Stones and Aymeric
Laporte. Bernardo Silva started on the bench, but replaced the injured
Fernandinho by the minute 52. Arsenal rested Petr Cech, plus the defender
Kolasinac (who entered by the minute 26, replacing the injured Monreal) and the
midfielder Elneny. Aaron Ramsey was replaced by Iwobi, on the 73 rd minute
78
0 1 2 3 4 5 6 7+ Home
0 3,32% 5,19% 4,06% 2,12% 0,83% 0,26% 0,07% 0,02%
1 6,11% 9,56% 7,48% 3,90% 1,53% 0,48% 0,12% 0,03%
2 5,62% 8,80% 6,88% 3,59% 1,40% 0,44% 0,11% 0,03%
3 3,45% 5,40% 4,22% 2,20% 0,86% 0,27% 0,07% 0,02%
4 1,59% 2,48% 1,94% 1,01% 0,40% 0,12% 0,03% 0,01%
5 0,58% 0,91% 0,72% 0,37% 0,15% 0,05% 0,01% 0,00%
6 0,18% 0,28% 0,22% 0,11% 0,04% 0,01% 0,00% 0,00%
7+ 0,06% 0,09% 0,07% 0,04% 0,02% 0,00% 0,00% 0,00%
Away
3,32% 9,56% 6,88% 2,20% 0,40% 0,05% 0,00% 0,00%
Draw 22,41% Odd 4,46
Total 100%
• Analysis- Our Poisson regression model grants more chances to an away win,
which is expectable, as Manchester City performed considerably better than
Arsenal over the last 2 seasons. Nevertheless, the model says that betting on a
City’s win would only have positive expected value if the bookmakers offer an
odd over 2.25, which is considerably higher than the best odd offered by the
Portuguese bookmakers (1.75).
Betting decision:
• Overall, our analysis suggests that Manchester City is, these days, a much more
reliable, confident and motivated team than Arsenal (as the League Cup match
confirms), and in spite of the latter’s good performances at home, City should win
this match, with more or less difficulties. If The Citizens score first, Arsenal will
be forced to risk more, exposing themselves to counter attacks. The level of
fatigue is the same for both teams, but Manchester City defensive balance can be
affected by the absence of Fernandinho. Although our Poisson regression model
indicates that an odd of 1.75 is very distant from the positive expected value
threshold (2.25), we still believe that betting on a Manchester City’s win offers a
relatively good reward for the level of risk assumed.
79
Annex III- Manchester City vs Chelsea (match day 29)
• 04/03/2018 at 4:00 pm, Ethiad Stadium (Manchester)

match.
Potential bets:
Probable line-ups:
• Manchester City- 4-3-3: Ederson; Walker, Kompany, Otamendi and Danilo; De

Bruyne, Gundogan and D. Silva; Sterling, Aguero and Sané
• Chelsea- 3-4-2-1: Courtois; Azpilicueta, Christensen and Rudiger; Moses, Kanté,

Fàbregas and Alonso; Willian and Hazard; Morata
• Manchester City- Out: Mendy (injured) and Delph (suspended); doubtful:

Sterling, Walker and Fernandinho (all injured)
• Chelsea- Doubtful: Bakayoko, Barkley and D.Luiz (all injured)
• Analysis- Manchester City has 3 important players in doubt, while Chelsea has 3
medium-importance players in doubt
Discipline table:
• Manchester City (average per match at home)- Yellow cards: 1.3, red cards: 0.1,
• Chelsea (average per match away)- Yellow cards: 1.1, red cards: 0.1, booking
points: 13.2
• Premier League (average per match)- Yellow cards: 1.6, red cards: 0.1, booking
80
points: 16.6
• Analysis- Both teams are considerably below the Premier League average in
terms of aggressiveness. It seems unlikely that cards or injuries might condition
the outcome of this match.
• 8 matches- 3 wins for Manchester City (37,5%), 3 wins for Chelsea (37,5%) and
2 draws (25%)
• Analysis- There is not a significant bias favouring any of the teams.
• Overall- Manchester City: 1 st place, 75 points, 82 goals scored and 20 conceded,

Chelsea: 5th place, 53 points, 50 goals scored and 25 conceded
• Home- Manchester City: 1st place, 40 points, 50 goals scored and 10 conceded
• Away- Chelsea: 3rd place, 24 points, 26 goals scored and 15 conceded
• Analysis- Manchester City has been, by far, the strongest team in the Premier
League this season, placing 1 st with the best goal difference, at home and away.
On the other hand, Chelsea has been performing more or less in equal manner at
home and away, so it is fair to assume that none of the teams is relying hugely on
the home advantage factor.
• Chelsea- won 55% of its points at home and 45% when playing away, scoring
were scored away
81
• Analysis- This table provides numerical evidence indicating that neither of the
teams is very reliant on home advantage. Manchester City achieved very good
results away, and even better results at home, hence we can conclude that the home
advantage has its influence on the team’s performance (specially on the number
of goals scored), but City do not depend on it to obtain good results. Regarding
Chelsea’s performance, it is possible to say exactly the same, with the difference
that the overall results obtained are not nearly as good as the ones achieved by
Manchester City. In this case, the influence of the home advantage on Chelsea’s
performance is mostly related with suffering less goals at home.
• Manchester City has a relative form of -0.55, while Chelsea scores -0.51. At first
glance, it seems weird that Chelsea has a less negative relative form in comparison
with City. This happens because Manchester City has been very consistent over
the course of the season, only losing against Liverpool, and drawing against
Everton, Crystal Palace and Burnley. All these matches are included in the last 8
matches, with the exception of Manchester City vs Everton
Scoring tables:
• Manchester City scores, on average, 3.57 goals per match at home, while Chelsea
we obtain Manchester City’s “scoring strength”, which corresponds to 2.32 goals
• Chelsea scores, on average, 1.86 goals per match when playing away, while
Manchester City concedes 0.71 goals per match at home. If we average these
numbers, we obtain Chelsea’s “scoring strength”, which corresponds to 1.29 goals
• Both teams like to play a possession-based football, using short passes to control
the game in the opposition’s half, but usually Conte adapts Chelsea style of play
when facing Guardiola’s teams. Therefore, we should expect Manchester City to
dominate possession in this match, against a more speculative Chelsea. City
defends high on the pitch, using the offside trap strategy. Although Manchester
City has this process very well automatized, the team is prone to individual
82
mistakes, so there might be some space to exploit behind The Citizens defensive
line. However, Chelsea is very weak in avoiding offside.
• Manchester City is strong at attacking set pieces, but Chelsea is equally strong at
defending them
• Chelsea likes to attack down the left, while Manchester City prefers to attack down
the right, so we can expect most of the action to be developed on that wing
• Both teams are weak at disputing aerial duels
• Chelsea is strong at creating scoring chances, while The Citizens are very weak at
stopping opponents from creating scoring chances
Team psychology:
positive team culture. Besides that, Manchester City should be highly motivated
to win the Premier League as soon as possible, in order to then focus its full
attention on the latter stages of the Champions League. Chelsea is the current
Premier League champion but the results have been disappointing this season. The
possible causes for this decline are, most likely, related with the selling of top
players, such as Matic or Diego Costa, which were not replaced by new players
with, at least, the same level in terms of quality. In addition, Antonio Conte is
having problems with the board, and simultaneously, with key players like
Courtois, Morata, David Luiz or Hazard. Due to these internal issues, plus the
results below the expectations, we have to classify Chelsea’s team spirit and
motivation as low
Burnley (all for Premier League). On these circumstances, we believe it is fair to
say that The Citizens’ confidence levels must be very high. Chelsea, on the other
83
hand, is going through a complicated period, dropping down to 5th in the Premier
League, out of the 2018/2019 Champions League qualifying places. On top of
that, probably Chelsea will be eliminated from the 2017/2018 Champions League,
after an upsetting 1-1 at home against Barcelona. The team will play the FA Cup
quarter-finals, still having a chance to win that competition, which is clearly not
enough for a club with Chelsea’s ambitions. Given the overall context of the
team’s behaviour in this season, we think that its current confidence level should
be classified as low
• Managerial impact- Guardiola arrived at Manchester and, after a mediocre first

season, has been able to turn Manchester City into one of the most complete teams
in Europe. The financial investment in the squad was huge, but nowadays City
plays in the image of its coach, and is definitely one of the best sides that he built
in the entire career. That being said, we have to consider Guardiola’s impact on
the club as very high. Antonio Conte achieved an astonishing Premier League
winning in his first season as Chelsea’s manager, 7 points ahead of the 2 nd ranked
Tottenham Hotspurs. Conte introduced the 3-4-2-1 tactic in the English top
division, which is commonly used in Italy and was subsequently adopted by
Arsenal, Watford and occasionally by Manchester rivals, Tottenham and
Liverpool, for example. Although this season is not being nearly as good as the
past one, we still have to consider the impact of Antonio Conte on Chelsea as high,
at least
Team fatigue:
• Toughness of previous match- Manchester City won comfortably 3-0 vs Arsenal

at Emirates, while Chelsea lost 2-1 against direct competitors Manchester United,
at Old Trafford, in a match where The Blues scored first. Having this scores in
consideration, we decided to rate the toughness of previous matches as medium
for Manchester City, and as the highest for Chelsea
• Resting- In City’s last match, from the probable line-up for today, the only player
that rested was Sterling, because he was injured. In this match, Sterling is expected
to replace Bernardo Silva in the starting 11. During the match, Guardiola
substituted Walker for Zinchenko (72 nd minute), Aguero for Touré (82 nd minute)
and D.Silva for Jesus (87th minute). Against Manchester United, Antonio Conte
84
only rested Fàbergas, although the latter was called to replace Drinkwater on the
81st minute. Apart from that, Conte replaced Hazard for Pedro (73 rd minute) and
Moses for Giroud (78th minute)

• Analysis- Our Poisson regression model grants more than 50% chances to a home
win. Nevertheless, the model says that betting on a Manchester City win would
only have positive expected value if the bookmakers would offer an odd over 1.92,
which is considerably higher than the best odd offered by the Portuguese
bookmakers (1.57). It is important to understand that Chelsea’s chances might be
overestimated, as they performed considerably better in 2016/2017 than in this
season, while City’s chances, on the contrary, are probably underestimated (as its
performance improved a lot in the current season).
Betting decision:
• The 0-3 away win against Arsenal just added more substance to our view that,
these days, Manchester City is the best team in the Premier League (and quite
possibly, in the world) by far. The difference to the other teams is enormous, as
The Citizens have the best (and more expensive) squad (in terms of quality and
quantity), the best manager, the most balanced and refined style of play (best
85
attack and defence, both at home and away). Manchester City is also an extremely
confident team, that never renounces on its playing principles, regardless of the
opponent’s name or status. With the exception of the away match against
Liverpool, every single team that played against Manchester City just assumed its
superiority and gave away the control of the match, trying not to suffer a heavy
loss. From our standpoint, every City match on which the bookmakers are offering
odds of around 1.5 for its win, delivers a great deal in terms of risk-reward trade-
off.
86
Annex IV- Manchester United vs Liverpool (match day 30)
• 10/03/2018 at 12:30 pm, Old Trafford (Manchester)

match.
Potential bets:
• Manchester United or Draw- 1.40
• Manchester United or Draw and over 1.5 goals- 1.80
Probable line-ups:
• Manchester United- 4-2-3-1: De Gea; Valencia, Bailly, Smalling and Young;

Matic and McTominay; Mata, Lingard and Sánchez; Lukaku
• Liverpool- 4-3-3: Karius; Arnold, Matip, Van Dijk and Robertson; Chamberlain,
Henderson and Can; Salah, Firmino and Mané
• Manchester United- Out: Herrera, Blind, Jones, Rojo, Martial and Pogba (all
injured)
• Liverpool- Doubtful: Robertson and Henderson (all injured)
• Analysis- Manchester United has 6 important players injured (especially Martial

and Pogba), while Liverpool has 2 potential starters in doubt.
Discipline table:
• Manchester United (average per match at home)- Yellow cards: 1.6, red cards:
• Liverpool (average per match away)- Yellow cards: 1.4, red cards: 0.1, booking
points: 16.1
87
• Analysis- Both teams are slightly below the Premier League average in terms of
aggressiveness (at home, Manchester United is equal to the average in terms of
yellow cards, but has not seen any red cards so far). It seems unlikely that cards
or injuries might condition the outcome of this match.
• 10 matches- 5 wins for Manchester United (50%), 4 draws (40%) and 1 win for
Liverpool (10%); 7 times over 1.5 (70%) and 3 time under (30%); Liverpool never
won away, while United did it 3 times. The last 4 matches between these 2 teams
ended up drawn
• Analysis- Given these numbers, it is fair to conclude that there is a certain level
of bias favouring Manchester United, in a match that represents arguably the
biggest historical rivalry in the English football.
• Overall- Manchester United: 2 nd place, 62 points, 56 goals scored and 22

conceded, Liverpool: 3rd place, 60 points, 67 goals scored and 32 conceded
• Home- Manchester United: 2 nd place, 35 points, 31 goals scored and 6 conceded
• Away- Liverpool: 2nd place, 27 points, 34 goals scored and 22 conceded
• Analysis- Both teams have been performing very consistently at home and away.
United defends better than Liverpool, and Liverpool attacks better than United,
but the overall goal difference is virtually the same (+34 goals for Manchester and
+35 goals for Liverpool).
• Manchester United- won 57% of its points at home and 43% when playing away,
• Liverpool- won 55% of its points at home and 45% when playing away, scoring
88
were scored away
• Analysis- This data tells us that both teams won around 55% of their points at
home, and 45% when playing away (10% more when playing at home).
Liverpool’s ability to score goals is nearly the same at home and away, while
Manchester United is near the Premier League average, but both teams suffer
considerably more goals when playing away (around 30%-70%).
• Manchester United has a relative form of +0.11, while Liverpool scores +0.31.
Both teams are experiencing good levels of recent form, but Liverpool even more
than Manchester United
Scoring tables:
• Manchester United scores, on average, 2.20 goals per match at home, while
Liverpool concedes 1.60 goals per match when playing away. If we average these
numbers, we obtain Manchester United’s “scoring strength”, which corresponds
to 1.90 goals. Besides that, United’s matches at home ended with over 1.5 goals
73% of the times
• Liverpool scores, on average, 2.33 goals per match when playing away, while
Manchester United concedes 0.47 goals per match at home. If we average these
numbers, we obtain Liverpool’s “scoring strength”, which corresponds to 1.40
goals. Besides that, Liverpool’s away matches ended with over 1.5 goals 93% of
the times
• Analysis- If we sum United and Liverpool scoring strengths, we obtain a total of

3.30 goals expected in the forthcoming match. Besides that, if we average both
teams’ percentage of matches that ended with over 1.5 goals, we get an 83%
converted into an odd of 1.21.
89
• Both teams like to use short passes to control the game in the opposition’s half.
• Liverpool uses an offside trap defensive strategy, which means that the defensive
line must be positioned high in the field. As Manchester United has very fast and
skilful players (e.g. Sánchez, Lingard, Lukaku or Rashford), and is good in
counter-attacking, there might be some space to exploit behind Liverpool’s
defensive line
• Both teams often attempt though balls
• Liverpool is weak at protecting the lead, while United is strong at coming back
from losing positions
• Manchester United is weak at stopping opponents from creating chances, while

Liverpool is very strong at creating scoring chances
• Both teams have many similarities, as United is good and Liverpool even better
at shooting from direct free kicks, finishing scoring chances, creating long shot
opportunities, creating chances through individual skill and counter attacking
Team psychology:
• Team spirit and motivation- Both teams’ spirit and motivation have to be very
high, as they are only separated by 2 points, and this is historically the most
electrifying rivalry in the Premier League
• Confidence- Although Manchester United is not being able to compete with

Manchester City for the Premier League title, the team has been performing
consistently this season, being in 2 nd place (the only loss at home was against
Manchester City), and still in the FA Cup quarter-finals, in which United will play
against Brighton, at Old Trafford. In the Champions League, United passed the
group stage for the first time since Alex Ferguson’s departure (2012/2013), facing
Sevilla at home for a place in the quarter-finals (0-0 away). Liverpool has been
performing nearly as good as Manchester United, placing 3rd in the league, 2
points behind The Red Devils, and already with a guaranteed spot among the best
8 teams of the 2017/2018 Champions League, after defeating Porto 0-5 at Estádio
90
do Dragão. The Reds were eliminated in the 4 th round of the FA Cup by West
Brom (2-3 at home), but on the other hand, were the only team in the Premier
League that was able to defeat Manchester City (4-3 at home). Given this context,
we believe that it is fair to classify both teams’ confidence levels as high
• Managerial impact- Klopp arrived at Liverpool in October of 2015, in

replacement of Brendan Rodgers, and in his first season (incomplete), although
The Reds finished 8th in the Premier League, he was able to reach the Europa
League final, which was lost to Sevilla (1-3). About 2 and a half years later,
Klopp’s “rock’ n roll” style of football is fully implemented at Merseyside:
Liverpool defends high, pressing a lot, and attacks at the speed of light, as
Borussia did when Klopp was in charge. To achieve that, some very high
investments in new players had to be made (e.g. Salah, Mané, Wijnaldum or Van
Dijk), but these days, there is no doubt in considering Klopp’s managerial impact
on Liverpool as high, not being very high just because he did not win any
silverware, so far. José Mourinho came to Manchester United with the club
immersed in a deep crisis, after 2 years of heavy investments without returns (poor
results), with Van Gaal at the helm. Mourinho brought a more pragmatic approach
to the club, focusing on a solid and compact defensive process, then trying to
exploit the opponent’s mistakes over the course of the match. In 2 years, the
Portuguese spent around 340 million euros in new players, such as Pogba (105
million euros, at the time the world transfer fee record), Alexis Sánchez, Lukaku,
Ibrahimovic or Matic, and won the FA Community Shield, the League Cup and
the Europa League in the first season. This season has been a disappointing one,
as the team was expected to fight for the Premier League title, but is just in 2nd
place, 16 points below Manchester City. The Red Devils have always been known
(at least in Ferguson’s era) for playing fast, attacking football (similar to Klopp’s
style), and José Mourinho’s cautious and cynical style of play does not convince
neither the fans or the club’s historical players. Because Mourinho has been able
to implement his ideas (either people like them or not) and to win some titles, we
consider that he is having a high managerial impact in the club, as well
Team fatigue:
• Toughness of previous match- Liverpool won comfortably 2-0 vs Newcastle at
91
Anfield Road, while Manchester United won 2-3 against Crystal Palace away, in
a match where Palace was winning 2-0 at half-time, and United only scored the
winning goal after the 90th minute. Having this in consideration, we consider the
toughness of previous matches as low for Liverpool, and as the highest for
Manchester United
• Resting- In United’s last match, the only player that fully rested was Bailly.
McTominay was replaced at half-time by Rashford, and on the 67th minute,
Mourinho decided to change Young for Mata, and Valencia for Luke Shaw.
Against Newcastle, none of the players that compose the Liverpool’s probable
line-up for Old Trafford rested entirely, but Matip only entered by the minute 88,
in replacement of Firmino. Besides that, Mané was replaced by Lallana (74 th
minute), and Chamberlain was replaced by Milner, also on the 88 th minute.
Overall, both teams rested a central defender, and United was able to rest a
midfielder for 45 minutes, and both full-backs for about 20 minutes, while
Liverpool rested a winger for 15 minutes

92
• Analysis- Our Poisson regression model grants 74,34% chances to a home win or
draw, and 74,40% to a home or away win, values that can be converted into an
odd of around 1.35. The model also grants a chance of 72,64% for the match to
end with over 1.5 goals, which can be converted into an odd of 1.38. We believe
that, in this case, the results given by the model should be pretty accurate, as both
teams are performing considerably better this season than in the last one, and have
not changed their managers or styles of play.
Betting decision:
• Our analysis suggests that, over the last 3 years, there has been a head-to-head
bias favouring Manchester United in the direct confrontations against Liverpool,
although in the last match between these teams (0-0 at Anfield), Liverpool
completely overpowered United, missing several opportunities to score. In 10
matches, Liverpool never won away, the last 4 ended up drawn, and 7 had over
1.5 goals. In 2017/2018, The Red Devils have been very strong at home, only
losing to Manchester City, and defeating top 6 rivals Chelsea and Tottenham.
However, injuries have been affecting United severely this season, with key
French players Martial and Pogba unavailable for this match. The scoring table
says that there is a probability of 83% that the match ends with over 1.5 goals (odd
of 1.21), and the Poisson regression determines a probability of 72,64% (odd of
1.38) for the same outcome. The model also attributed a likelihood of 74,34%
(odd of 1.35) to United’s win or draw. About the match, our expectation is that
neither of the teams will risk too much (at least in the beginning), because they
are only separated by 2 points, and so a draw can be an interesting outcome for
both (especially for United). Liverpool uses a risky style of play, with high
pressure, but with the exception of Van Dijk, has not enough individual quality
on the defence to do so, and can be exposed to United’s counter attacks. Mourinho
knows this, and probably will adopt the usual cautious tactics, waiting for the right
moment to strike. The Red Devils must be careful though, because they are weak
in stopping opponents from creating scoring chances, and Liverpool’s front line
is absolutely devastating. Having all this in consideration, we believe that an
offered odd of 1.80 for a Manchester United win or draw, plus over 1.5 goals in
the match delivers positive expected value, therefore worthing a bet.
93
Annex V- Liverpool vs Watford (match day 31)
• 17/03/2018 at 5:30 pm, Anfield Road (Liverpool)

• Analysis: Good weather to play football in the UK, at this time of the year. The
precipitation probability of 0% allows for a fast circulation of the ball, and
consequently for an open match.
Potential bets:
• Liverpool and over 2.5 goals- 1.51
• Liverpool and under 4.5 goals- 1.58
Probable line-ups:
• Liverpool- 4-3-3: Karius; Arnold, Matip, Van Dijk and Robertson; Wijnaldum,
Henderson and Can; Salah, Firmino and Mané
• Watford- 4-2-3-1: Karnezis; Janmaat, Prodl, Mariappa and Britos; Doucouré and
Capoue; Femenía, Pereyra and Richarlison; Deeney
• Liverpool- None
• Watford- Out: Hoban, Catchcart, Chalobah, Kabasele, Cleverley and Deulofeu

(all injured)
• Analysis- Watford has 5 important players injured (especially Kabasele and

Cleverley), and also a key attacking player injured, Gerard Deulofeu.
Discipline table:
• Liverpool (average per match at home)- Yellow cards: 1.0, red cards: 0.0,
• Watford (average per match away)- Yellow cards: 1.3, red cards: 0.1, booking
points: 15.5
94
• Analysis- Both teams are below the Premier League average in terms of
aggressiveness, therefore, it seems unlikely that cards or injuries might condition
• 5 matches- 3 wins for Liverpool (60%), 1 draw (20%) and 1 win for Watford
(20%); 3 times over 2.5 (60%) and 2 times under (40%); 2 times over 4.5 (40%)
and 3 times under (60%)
• Analysis- Given these numbers, it is fair to conclude that there is a certain level
of bias favouring Liverpool, however, the last match for Premier League ended
up in a draw, with Watford scoring the equalizing goal at the stoppage time.
• Overall- Liverpool: 4th place, 60 points, 68 goals scored and 34 conceded,

Watford: 10th place, 36 points, 39 goals scored and 50 conceded
• Home- Liverpool: 5th place, 33 points, 33 goals scored and 10 conceded
• Away- Watford: 8th place, 14 points, 17 goals scored and 24 conceded
• Analysis- This table provides support to the idea that Liverpool is a much stronger
team than Watford, investing a lot more to be placed among the Champions
League places (top 4), while Watford is just a team that fights to achieve its
stability in the Premier League. Overall, Liverpool has more 24 points, 29 goals
scored and less 16 conceded than The Hornets, which is a very considerable
difference.
• Liverpool- won 55% of its points at home and 45% when playing away, scoring
• Watford- won 61% of its points at home and 39% when playing away, scoring
95
were scored away
• Analysis- Liverpool won 10% more points at home than away, and Watford 21%,
which means that Watford is 11% more dependent on the home advantage factor
than Liverpool. In terms of goals scored, it is fair to say that Watford is consonant
with the Premier League average, while Liverpool scores virtually the same
number of goals at home and away (49%-51%), maybe due to the immense
strength of its offensive line. However, Liverpool concedes a lot more goals when
playing away from Anfield (29%-71%), probably because its defensive line is not
very reliable, and opponents tend to attack more when playing at their own
grounds. Watford, by its turn, concedes approximately the same number of goals
at home and away (52%-48%).
• The Reds have a relative form of +0.00, while Watford scores +0.18, which means
that Liverpool’s level of recent form is neutral, and Watford’s is positive
Scoring tables:
• Liverpool scores, on average, 2.20 goals per match at home, while Watford
we obtain Liverpool’s “scoring strength”, which corresponds to 1.90 goals.
Besides that, Liverpool’s matches at home ended with over 2.5 goals 53% of the
times, while 80% of those matches ended with less than 4.5 goals
• Watford scores, on average, 1.13 goals per match when playing away, while
Liverpool concedes 0.67 goals per match at home. If we average these numbers,
that, Watford’s away matches ended with over 2.5 goals 53% of the times, while
87% of those matches ended with less than 4.5 goals
• Analysis- If we sum Liverpool and Watford scoring strengths, we obtain a total

of 2.80 goals expected in the forthcoming match. Besides that, if we average both
96
teams’ percentage of matches that ended with over 2.5 goals, we get an 53%
converted into an odd of 1.87. When it comes to the probability that this match in
particular ends with less than 4.5 goals, the average obtained is 83,5%, which
corresponds to an odd of 1.20.
• Liverpool likes to play a possession-based football, using short passes to control

the game in the opposition’s half
• Both teams are strong at stealing the ball from the opposition, but have different
styles of doing it. Watford can be considered an aggressive team, while Liverpool
is non-aggressive. Besides that, Watford is strong in aerial duels, while Liverpool
is weak
• Liverpool is weak at protecting the lead, while Watford is very weak. However,
the latter is strong at coming back from losing positions
• Liverpool is strong in attacking set pieces, while Watford is weak at defending

them
• Watford uses an offside trap defensive strategy, which means that the defensive
line must be positioned high in the field. As Liverpool has a very fast and skilful
offensive triangle (Mané, Firmino and Salah), and is very good in counter-
attacking, there might be some space to exploit behind Watford’s defensive line,
especially because The Hornets are very weak at defending counter attacks
Team psychology:
• Team spirit and motivation- Watford suffered upsetting early eliminations from
the domestic cups, so the only focus now is to avoid relegation from the Premier
League as soon as possible, preferably placing within the first half of the table.
The team seems united enough around Gracia, although some important players
expressed, at the time, their dissatisfaction about Silva’s dismissal (e.g.
Richarlison). On the other hand, Liverpool needs to guarantee a Champions
League qualification place (among the top 4), ideally as close as possible to
Manchester City. Klopp is a charismatic coach, highly passionate about football
97
and a great motivator, as well. He has the support of the team, and will not allow
any kind of relaxation until the end of the season. Both teams are in a process of
establishing themselves at a higher level (Liverpool as serious title contenders,
Watford as a mid-table team), consequently, they need to consolidate their growth
by winning as often as possible. That being said, we consider that the team spirit
and motivation of Liverpool and Watford must be high
• Confidence- Liverpool has been performing consistently this season, placing 4 th

in the Premier League, and already with a guaranteed spot among the best 8 teams
of the 2017/2018 Champions League, after defeating Porto 0-5 at Estádio do
Dragão. The Reds were eliminated in the 4th round of the FA Cup by West Brom
(2-3 at home), but on the other hand, were the only team in the Premier League
that was able to defeat Manchester City (4-3 at home). Given this context, we
believe that it is fair to classify Liverpool’s level of confidence as high. The
Hornets have been performing much better in 2017/2018 than in the last season,
where they ranked 17th, the last place above relegation zone. Currently, Watford
ranks 10th, 8 points above relegation, when there are still 7 matches left to play.
In the national cups, Watford was surprisingly eliminated from the League Cup
by Bristol (2-3 at home), in the 3rd round, and defeated by Southampton (1-0 away)
in the 4th round of FA Cup. Considering this, we think that Watford’s level of
confidence should be classified as medium
• Managerial impact- Klopp arrived at Liverpool in October of 2015, in

replacement of Brendan Rodgers, and in his first season (incomplete), although
The Reds finished 8th in the Premier League, he was able to reach the Europa
League final, which was lost to Sevilla (1-3). About 2 and a half years later,
Klopp’s “rock’ n roll” style of football is fully implemented at Merseyside:
Liverpool defends high, pressing a lot, and attacks at the speed of light, as
Borussia did when Klopp was in charge. To achieve that, some very high
investments in new players had to be made (e.g. Salah, Mané, Wijnaldum or Van
Dijk), but these days, there is no doubt in considering Klopp’s managerial impact
on Liverpool as high, not being very high just because he did not win any
silverware, so far. Marco Silva was fired on the 20 th January, after suffering a
categorical defeat against Leicester (2-0 away), being replaced by Javi Gracia,
98
who started by changing the usual formation from 3-4-2-1 to 4-2-3-1. Besides that,
some players that were important under Silva’s management, like Hughes,
Carrillo or Gray, lost relevance to others, such as Roberto Pereyra or Troy Deeney.
The Spanish coach began his spell by drawing away, against Stoke (0-0). After
that, he achieved an incredible home win against Chelsea (4-1), but then lost away
to direct competitor West Ham (2-0). In the last 3 matches, Watford was able to
beat Everton and West Brom (both 1-0) at home, but lost 3-0 away to Arsenal in
the last match day. Since Javi Gracia took the helm, Watford has been stable in
the 10th place, 4 points below 9th ranked Everton, therefore we think that, so far,
his managerial impact at Vicarage Road has been normal
Team fatigue:
• Toughness of previous match- Liverpool lost the North-West Derby vs

Manchester United (2-1), at Old Trafford. It was an even and intense match, that
must have risen the fatigue levels of both teams (United lost to Sevilla the chance
of qualifying for the Champions League quarter-finals, 3 days later, at home).
Arsenal gave no possibilities to Watford, winning 3-0 at Emirates, although the
outcome does not reflect what really happened in the game, as The Hornets
disputed the match, had several scoring chances, but were not effective when it
came to finish those occasions. Having this in consideration, we classify the
toughness of previous matches as high for The Hornets, and as the highest for
Liverpool
• Resting- In Liverpool’s last match, from the probable line-up for today, the only
player that fully rested was Matip. Arnold was replaced on the 80 th minute by
Wijnaldum, and 4 minutes later, Klopp risked even more for the equalizing goal,
changing Robertson for Solanke. Against Arsenal, none of the players that
compose Watford’s probable line-up for Anfield Road rested entirely, but Britos
only entered by the minute 80, in replacement of Janmaat. Besides that, Femenía
was replaced by Hughes (62 nd minute), and Pereyra was replaced by Okaka, 5
minutes later

99
• Analysis- Our Poisson regression model grants 81,55% chances to a home win,
which can be converted into an odd of 1.23. The model also grants a chance of
75,04% to the match ending with over 2.5 goals, which corresponds to an odd of
1.33. The probability estimated for the match having less than 4.5 goals is 64,37%,
which is equivalent to an odd of 1.55. We believe that, in this case, the results
given by the model should be relatively accurate, as both teams are performing
considerably better this season than in the last one.
Betting decision:
• Liverpool will approach this match in its full strength, while Watford has 6 players
currently injured. Even if Watford had no injuries, Liverpool would still have a
squad with much more individual quality. Furthermore, The Reds play at home,
and so they are expected to dominate and win this match, as The Hornets have
defensive flaws that can be exploited. In fact, both teams are strong in recovering
the ball, but Watford has problems in controlling the space behind its defensive
100
line, and Liverpool is very strong in the offensive transitions. Besides that, our
analysis indicates that there is a bias in the head-to-head confrontations between
these teams, in Liverpool’s favour. The Poisson regression model attributes
81,55% chances to a home win, which is equivalent to an odd of 1.23. The model
also grants probabilities of 75,04% to this match having more than 2.5 goals (odd
of 1.33), and of 64,37% to this match ending with less than 4.5 goals (odd of 1.55).
Additionally, the sum of both teams’ “scoring strengths” is equal to 2.80 goals.
Although, at this stage, it seems more logical to bet on Liverpool’s win and over
2.5 goals, 80% of Liverpool’s matches at home, and 87% of Watford’s matches
away ended with less than 4.5 goals. Considering this, we believe that betting on
Liverpool and under 4.5 goals in the match, has a positive expected value when
the offered odd is of 1.58.
101
Annex VI- Bournemouth vs West Bromwich Albion (match day 31)
• 17/03/2018 at 3:00 pm, Dean Court Stadium (Bournemouth)

match.
Potential bets:
• Bournemouth- 1.82
Probable line-ups:
• Bournemouth- 4-4-1-1: Begovic; Francis, S.Cook, Aké and Daniels; Smith,

L.Cook, Gosling and Stanislas; King; Wilson
• West Bromwich- 4-4-2: Foster; Dawson, Hegazi, Evans and Gibbs; Phillips,
Krychowiak, Livermore and Brunt; Rodriguez and Rondón
• Bournemouth- Out: Mings; doubtful: Pugh (all injured)
• West Bromwich- Out: Chadli, Morrison, Sturridge, and Barry; doubtful: Evans
(all injured)
• Analysis- Bournemouth has a mid-importance player in doubt (Pugh), while West

Bromwich will have to deal with the absences of Nacer Chadli, Gareth Barry and
probably also Johnny Evans (very important players). Besides that, the skilful but
“chronically” injured Daniel Sturridge will not be fit to play this match, as well.
Discipline table:
• Bournemouth (average per match at home)- Yellow cards: 1.6, red cards: 0.1,
• West Bromwich (average per match away)- Yellow cards: 1.9, red cards: 0.1,
102
• Premier League (average per match)- Yellow cards: 1.6, red cards: 0.1, booking
points: 16.5
• Analysis- Both teams are above the Premier League’s average in terms of
aggressiveness, especially West Bromwich. Therefore, it seems likely that at least
a high number of yellow cards can be shown during the 90 minutes. Furthermore,
red cards or injuries might occur, conditioning the outcome of the match. As WBA
has a very physical and combative style of play (the squad was made at the image
of Tony Pullis, the coach that began the season), we believe that its excess of
impetuosity can work as an advantage for Bournemouth.
• 6 matches- 3 wins for Bournemouth (50%), 2 wins for West Bromwich (33,3%)
and 1 draw (16,7%)
• Analysis- Over the last 3 years, Bournemouth won 50% of the matches between
these teams, however, the last 2 were won by WBA, who was never able to win
away. Considering this, we think that the equation is balanced and consequently,
there is not a significant bias favouring any of the teams.
• Overall- Bournemouth: 12nd place, 33 points, 35 goals scored and 48 conceded,

West Bromwich: 20th place, 20 points, 23 goals scored and 47 conceded
• Home- Bournemouth: 15th place, 19 points, 21 goals scored and 25 conceded
• Away- West Bromwich: 20th place, 7 points, 7 goals scored and 23 conceded
• Analysis- This table allows to understand that, in one hand, Bournemouth is

performing as a mid-table team over the course of the 2017/2018 season, currently
placing 6 points above relegation zone, and 4 points below the first-half of the
table. On the other hand, West Bromwich’s season has been disappointing to say
the least, as The Baggies have a reasonably good squad, and were expected to
finish around mid-table. The reality, however, is that West Brom has been the
worst team in the Premier League so far, losing the last 6 matches in a row. The
103
West Bromwich Albion’s current situation is very concerning, as the club ranks
in the bottom place of the division, already 8 points below the non-relegation zone,
when there are only 8 matches left to play.
• Bournemouth- won 58% of its points at home and 42% when playing away,
• West Bromwich- won 65% of its points at home and 35% when playing away,
scoring also 70% of its goals at home and 30% when playing away, conceding
51% of its goals at home and 49% away
were scored away
• Analysis- By analysing the numerical data provided in this table, two main
conclusions can be made: the first one is that both teams depend on the home
advantage to win points and score goals, which is normal in low-to-mid table
teams, but WBA’s dependence on this factor is huge. Even so, West Bromwich
results at home have been very poor, which anticipates serious difficulties in the
forthcoming away matches. The second conclusion is that both Bournemouth and
West Brom are conceding virtually the same number of goals at home and away,
which can be interpreted as a sign of weakness in their defensive processes, as
each one of these teams are suffering goals in a consistent way, regardless of the
situation.
• Bournemouth has a relative form of +0.40, while WBA scores -0.17. In fact,
Bournemouth did not begin well the season, but is experiencing good recent form,
only losing twice in the last 11 matches. West Bromwich, by its turn, is in a
dreadful moment, with 6 losses and a draw in the last 7 matches. Having this in
consideration, the fact that the score for its relative form is only of -0.17, is a
striking evidence of how terrible The Baggies season has been, until this moment
Scoring tables:
104
• Bournemouth scores, on average, 1.40 goals per match at home, while West
Brom concedes 1.53 goals per match when playing away. If we average these
numbers, we obtain Bournemouth’s “scoring strength”, which corresponds to 1.47
goals
• West Bromwich scores, on average, 0.47 goals per match when playing away,
while Bournemouth concedes 1.67 goals per match at home. If we average these
numbers, we obtain WBA’s “scoring strength”, which corresponds to 1.07 goals
• Bournemouth is strong at coming back from losing positions, while West

Bromwich is very weak at protecting the lead
• Bournemouth take long shots often, while West Bromwich is weak at defending
them
• Bournemouth uses a non-aggressive approach, as it is weak in aerial duels, while

West Bromwich Albion uses an aggressive approach, and it is strong in aerial
duels, hence we expect West Brom to win most of the disputed balls
• Bournemouth likes to attack down the left, while WBA likes to attack down the
right, so we can expect most of the action to be developed on that wing. However,
West Brom is very weak when it comes to defend attacks down the flanks
Team psychology:
• Team spirit and motivation- Bournemouth’s motivation to win this match

should be considered very high, as this is a great opportunity to win 3 points at
home, against a weakened opponent. Those points would be a valuable
contribution to the achievement of the club’s objectives for this season: the main
one is to avoid relegation, and after that is guaranteed, it would be nice to assure
a place in the first half of the table. The team is going through a good moment,
and at least to our knowledge, there are no public cases of indiscipline or
instability at Dean Court. On the other hand, The Baggies desperately need points
in order to escape from relegation. West Brom is under overwhelming pressure,
and that is surely affecting the players’ minds, as some of the recent lost matches
were not as negative as the final scores suggest. This season, everything seems to
105
go wrong with West Bromwich and as we know, the team spirit cannot be good
when results do not show up for such a long period of time. Regarding this, the
WBA’s team spirit and motivation has to be classified as very low
• Confidence- The 2017/2018 season began terribly for Bournemouth, with 4

defeats in 4 matches, however the team’s performance improved a lot after the
defeat vs Manchester City (4-0 away), including incredible wins against Arsenal
(2-1 at home) and Chelsea (0-3 away). In the last 10 matches for the Premier
League, Bournemouth only lost twice. When it comes to its performance in the
domestic cups, Bournemouth was surprisingly defeated by Championship teams:
in the 3rd round of the League Cup, losing 2-3 at home vs Preston, and also in the
3rd round of the FA Cup (3-0 away vs Millwall). WBA’s performance in the
Premier League has been a complete disaster, as they were only able to win 3
matches, so far: 1-0 to Bournemouth in the 1 st match day (at home), 0-1 to Burnley
in the 2nd match day (away), and the only win under Pardew’s management was
against Brighton, at home (2-0). In the FA Cup, The Baggies eliminated Liverpool
in the 4th round (2-3 at Anfield Road), but loss in the next round against
Southampton (1-2 at home). West Bromwich adventure in the League Cup ended
after losing naturally vs Manchester City (1-2 at home). Given these results, we
consider Bournemouth’s level of confidence for this match as high, and West
Brom’s as very low
• Managerial impact- Eddie Howe is a manager that is fully identified with

Bournemouth’s culture. Howe began and finished his career as a player at Dean
Court, and after that, the Englishman also began his career as a manager in the
club. After a brief spell at Burnley, he returned to Bournemouth, achieving
promotions from Sky Bet League 1 to Championship (2012/2013), and from there
to the Premier League, in 2014/2015. In the English main division, Howe lead the
team to a 16th place in 2015/2016, when few expected Bournemouth to avoid
relegation, and then to a surprising 9th place in 2016/2017. This season,
Bournemouth is in 10th place, playing an entertaining fast pace football that mixes
British and Dutch football philosophies. Today, Eddie Howe is widely considered
the most promising English manager (has been linked with a move to Arsenal, in
replacement of long-time manager Arsène Wenger), and of course, his managerial
106
impact on Bournemouth must be considered very high. West Bromwich began the
season with Tony Pullis at the helm, which is a manager that is known for
promoting a very physical and direct style of play, relying more on muscle than
skill. Pullis built the team in his image, hiring several players that match the
characteristics mentioned above (for example, Krychowiak or Jay Rodriguez).
Although Tony Pullis is criticized for his outdated style of play, the truth is that
he has been able to achieve relatively good results, specially during his tenure in
Stoke City. However, this season the results were not so good, and Pullis was
sacked after being defeated by Chelsea (0-4 at home), leaving the club just one
point above the relegation zone. In his replacement, the board nominated Alan
Pardew, a man who had successful experiences as a manager in the Premier
League, being able to save Newcastle and Crystal Palace from relegation.
However, Pardew is not being able to implement his ideas for WBA’s football,
probably because these are quite different from Pullis vision, and The Baggies
went down to the last place in the League, now already 8 points below the
“salvation zone”. Considering this, we believe that Pardew’s managerial impact
should be evaluated as low, not being very low just because things were already
not so bright when he arrived at The Hawthorns Stadium
Team fatigue:
• Toughness of previous match- In the last match day, both teams lost 1-4 at home:
Bournemouth against top 4 candidates Tottenham, and West Bromwich against
Leicester City. Given these outcomes, we have to consider the toughness of
previous matches as the highest
• Resting- In Bournemouth vs Tottenham, from the probable line-up for today, the
only player that began as a substitute was King, although he replaced Mousset at
the 67th minute. On the same minute, Eddie Howe also replaced Daniels for Ibe.
Finally, Stanislas was replaced for Defore (minute 76). Against Leicester, Nyom
replaced the injured Evans, and Burke was preferred over Rodriguez to begin the
match. Alan Pardew unsuccessfully tried to change the course of the events by
replacing Krychowiak for Field (minute 59), Phillips for Rodriguez (minute 70)
and Burke for Robson-Kanu (minute 83). After observing the starting elevens and
substitutions made by each one of the teams in the last match day, it seems
107
reasonable to conclude that they have more or less equal resting levels

win. Nevertheless, the model says that betting on a Bournemouth’s win would
only have positive expected value if the bookmakers would be willing to offer an
odd over 1.97, which is higher than the best odd offered by the Portuguese
bookmakers (1.82). However, one must consider the possibility that the model is
overestimating WBA’s chances, as they performed considerably better in
2016/2017 than in the current season.
Betting decision:
• The information contained in this report suggests that West Bromwich Albion is
going through the worst moment of the season, and that avoiding relegation would
be considered almost a miracle, as the 6 losses in the last 6 matches attest. As
things were not bad enough already, The Baggies have some important absences
confirmed, such as the cases of Sturridge, Gareth Barry or Chadli, (plus the
probable absence of Johnny Evans) while Bournemouth has not significant
absences for this match. Besides that, West Brom is a very aggressive team, and
it is likely that a high number of cards will be shown, conditioning the
performance of the team. Over the last 3 years, WBA was never able to beat
108
Bournemouth away from home, although The Baggies have won the last 2
matches between these teams. Bournemouth plays at Dean Court and is in a good
moment of form, so we believe that betting on a home win at 1.82 has positive
expected value.
109
Annex VII- Everton vs Manchester City (match day 32)
• 31/03/2018 at 5:30 pm, Goodison Park (Liverpool)

match.
Potential bets:
• Manchester City and over 1.5 goals- 1.45
• Manchester City and below 4.5 goals- 1.60
Probable line-ups:
• Everton- 4-2-3-1: Pickford; Coleman, Keane, Jagielka and Baines; Gueye and
Rooney; Walcott, Davies and Bolasie; Tosun
• Manchester City- 4-3-3: Ederson; Walker, Kompany, Otamendi and Delph; De

Bruyne, Fernandinho and D.Silva; Sterling, Jesus and Sané
• Everton- Out: McCarthy, Stekelenburg, Sigurdsson, Mangala (all injured) and

Williams (suspended); doubtful: Holgate (injured)
• Manchester City- Doubtful: Delph and Aguero (all injured)
• Analysis- Besides the chronically injured McCarthy, Everton has 2 important

players out for this match (Sigurdsson and Williams), and one in doubt (Holgate).
Manchester City, by its turn, has 2 players in doubt, being one of them very
important to the team’s attacking dynamics (Sergio Aguero).
Discipline table:
• Everton (average per match at home)- Yellow cards: 1.6, red cards: 0.1, booking
110
points: 17.7
• Analysis- Both teams are above the Premier League average in terms of
aggressiveness (especially Manchester City), therefore, it seems possible that
cards or injuries might condition the outcome of this match.
• 5 matches- 2 wins for Everton (22%), 4 draws (45%) and 3 wins for Manchester
City (33%); 7 times over 1.5 (78%) and 2 times under (22%); 3 times over 2.5
(33%) and 6 times under (67%); 9 times under 4.5 (100%)
• Analysis- The head-to-head record seems pretty even, nevertheless one must not
forget that in the summer of 2017, both teams did the biggest investments of their
history. As a consequence of that, Manchester City is performing like never
before, and Everton is having an average season. That being said, we think that
the head-to-head data over the last 3 years does not truthfully reflects the
difference of strength between these teams (Manchester City is incomparably
stronger).
• Overall- Everton: 9th place, 40 points, 37 goals scored and 50 conceded,

• Home- Everton: 7th place, 29 points, 25 goals scored and 18 conceded
• Analysis- These numbers support the fact that Manchester City is the best team
in the Premier League, and of course, a much stronger team than Everton. Overall,
The Citizens have more 41 points, 50 goals scored and less 30 conceded than
Everton, which is a huge difference.
111
• Everton- won 72% of its points at home and 28% when playing away, scoring
were scored away
• Analysis- Manchester City gained 6% more points at home than away, in contrast
with Everton, that won 44% more points at home than away from Goodison Park.
This data allows to understand that Everton depends hugely on the home
advantage factor to win, while Manchester City does not. Furthermore, Everton
depends on this factor not only to win, but also to score and prevent goals. On the
contrary, Manchester City conceded exactly the same number of goals at home
and away, which in this case, can be interpreted as a sign of a well-implemented
defensive process, as The Citizens have the best defence of all the Premier
League’s teams, and play exactly the same way independently of being at home
or away. They score 20% more when playing at Ethiad Stadium, which can be
explained by the “fear” that Guardiola’s pupils instil to the vast majority of their
opponents (including rivals such as Manchester United), apart from the home
advantage effect on its own.
• The Toffees have a relative form of +0.23, while City scores -0.32. At first glance,
it may seem weird that Everton has a better level of relative form than Manchester
City. This is because City has been very consistent over the course of the season,
only losing against Liverpool, and drawing against Everton, Crystal Palace and
Burnley. The defeat and one of these draws happened over the last 8 matches.
Everton began the season performing very poorly, under Ronald Koeman’s
guidance. After the Dutchman was dismissed, Sam Allardyce took charge and the
results improved significantly, as Everton went from relegation zone to mid-table.
112
From the last 8 matches, Everton won 4, lost 3 and tied 1
Scoring tables:
• Everton scores, on average, 1.67 goals per match at home, while Manchester City
we obtain Everton’s “scoring strength”, which corresponds to 1.17 goals. Besides
that, Everton’s matches at home ended with more than 1.5 goals 80% of the times,
with over 2.5 goals 53% of the times, and with less than 4.5 goals 87% of the
times
while Everton concedes 1.20 goals per match at home. If we average these
numbers, we obtain City’s “scoring strength”, which corresponds to 1.74 goals.
Besides that, The Citizens’ away matches ended with more than 1.5 goals 80% of
the times, with over 2.5 goals 53% of the times, and with less than 4.5 goals 80%
of the times
• Analysis- If we sum Everton and Manchester City scoring strengths, we obtain a

total of 2.91 goals expected in the forthcoming match. Besides that, Everton and
City matches finished with more than 1.5 goals 80% of the times, which is
equivalent to an odd of 1.25. If we average both teams’ percentage of matches that
ended with over 2.5 goals, we get an 53% probability that this match in particular
ends with over 2.5 goals, which can be converted into an odd of 1.87. When it
comes to the probability of this match in particular ends with less than 4.5 goals,
the average obtained is 83,5%, which corresponds to an odd of 1.20.
• Manchester City is very strong at attacking down the wings, while Everton is very
weak at defending against these type of attacks
• Everton is very weak at defending against skilful players, while The Citizens are
strong at creating chances through individual skill, as they have very technical
players, such as De Bruyne, David Silva or Leroy Sané
• Manchester City is strong at attacking set pieces, while Everton is weak at

defending them
113
• Both teams are weak in aerial duels, and also at avoiding individual errors
• Manchester City style is all about keeping possession of the ball, using short
passes to control the game in the opposition’s half. The Toffees like to play in
their own half, are weak at keeping possession of the ball, but strong at stealing
the ball from the opposition
• City likes to attack down the left, while Everton likes to attack down the right, so
we can expect most of the action to be developed on that wing
• City defends high on the pitch, using the offside trap strategy. As Everton likes to
attack using long balls, and has some fast players in its offensive line (e.g. Walcott,
Bolasie or Niasse), there might be some space to exploit behind The Citizens
defensive line. However, Manchester City has this process very well automatized,
being the team that suffered less goals in the League
Team psychology:
• Team spirit and motivation- This season, Everton invested massive sums in new
players, targeting to a Champions League qualification spot (top 4), and also to a
good performance in Europa League. Those goals are now out of reach, but The
Toffees are still required to achieve the best possible place in the Premier League
(realistically, a 7th place would be nice). In one hand, the club has to deal with the
disappointment of missing the season’s primary ambitions, and the demotivation
of not competing for a real objective, but on the other hand, Everton’s supporters
are very demanding, and the team has to secure a place above mid-table, to
continue its process of solidification as a contender for European qualification.
Given this situation, we consider Everton’s team spirit and motivation as medium.
Pep Guardiola promotes a squad rotation policy, so that every player feels useful
and integrated. There are no cases (at least publicly known) of indiscipline or
internal issues between the players and the manager. The way the team is
performing (and winning) provides strong evidence of a positive team culture.
Besides that, Manchester City should be highly motivated to win the Premier
League as soon as possible, in order to then focus its full attention on the latter
stages of the Champions League
114
Burnley (all for Premier League). On these circumstances, we believe it is fair to
say that The Citizens’ confidence levels must be very high. Everton has its
position in the Premier League perfectly stabilized: 9th place, relatively far from
both relegation and European qualification zones. Also, the team was eliminated
from Europa League (group stage), FA Cup (2-1 defeat vs Chelsea away) and
League Cup (also a 2-1 defeat, in The Merseyside derby vs Liverpool, at Anfield
Road). Although Everton has been performing below the initial expectations for
this season, The Tofees have improved a lot since Koeman’s departure (at least in
terms of results), and so their confidence levels should be classified as medium
• Managerial impact- Everton sacked Ronald Koeman after the defeat at home
against Arsenal (2-5), on the 22nd October, that left the Liverpool based side in
relegation zone (18th place). Sam Allardyce succeeded Koeman as Everton’s
manager, and was able to put Everton in 9 th place, with 7 wins, 4 draws and 6
losses. Because of this clear improvement in Everton’s results, we consider that
he had a high impact on the team’s performance. Guardiola arrived at Manchester
and after a mediocre first season, has been able to turn Manchester City into one
of the most complete teams in Europe. The financial investment in the squad was
huge, but nowadays City plays in the image of its coach, and is definitely one of
the best sides that he built in the entire career. That being said, we have to consider
Guardiola’s impact on the club as very high
Team fatigue:
• Toughness of previous match- The last matches of both teams were against
Stoke (away), because Manchester City did not play the match day 31, in which
The Citizens were supposed to meet Brighton, at the Ethiad Stadium. Because of
that, we will not classify the toughness of Manchester City’s last match. Everton,
by its turn, had a tough match at The Britannia Stadium, although Stoke had a
player sent off by the minute 30 (Charlie Adam). Everton scored first by Tosun
(69th minute), but even with 10 players, Stoke was able to equalize the score. The
winning goal was scored, once again, by Tosun (84th minute). Considering this,
we decided to classify the toughness of Everton’s last match as very high
115
• Resting- The last match day of Premier League ended at 17/03/2018 (14 days
ago). Besides that, Manchester City’s last match was delayed, and so this part of
the report will not be considered to forecast what is likely to happen in this match
• National team participation- From the Manchester City’s probable line-up for
this match, the only player that was not called to the most recent double round of
international matches was Fabian Delph, who is in doubt due to an injury. If Delph
does not recover, he will be replaced by Zinchenko, who was called to play for
Ukraine’s national team. Therefore, the level of national team participation of the
Manchester City’s starting 11 can be classified either as many or all of its players.
In Everton’s case, the level of national team participation of its starting 11 should
be classified either as half-team or many, as 6 of its players have been called to
play for their respective national teams: Pickford for England, Coleman for
Republic of Ireland, Gueye for Senegal, Bolasie for Congo, Tom Davies for
England U21 and Tosun for Turkey
116
• Analysis- The Poisson regression model grants 59,12% chances to a home win,
which can be converted into an odd of 1.69. The probability given by the Poisson
regression to the match ending with more than 1.5 goals is 81,37%, which is
equivalent to an odd of 1.23. The model also grants a chance of 59,63% to this
match ends with over 2.5 goals, which corresponds to an odd of 1.68. The
probability estimated for the match having less than 4.5 goals is 80,01%, which is
equivalent to an odd of 1.25. It is important to understand that Everton’s chances
might be overestimated, as they performed considerably better in 2016/2017 than
in 2017/2018, while City’s chances, on the contrary, are probably underestimated
(as its performance improved a lot in the current season).
Betting decision:
• Manchester City is the best team in the Premier League (and quite possibly, in the
world) by far. The difference to the other teams is enormous, as The Citizens have
the best (and more expensive) squad (in terms of quality and quantity), the best
manager, the most balanced and refined style of play (best attack and defence,
both at home and away). Manchester City is also an extremely confident team,
that never renounces on its playing principles, regardless of the opponent’s name
or status. With the exception of the away match against Liverpool, every single
team that played against Manchester City just assumed its superiority and gave
away the control of the match, trying not to suffer a heavy loss. From our
standpoint, every City’s match on which the bookmakers are offering odds of
around 1.5 for its win, delivers a great deal in terms of risk-reward trade-off.
117
Annex VIII- Arsenal vs Southampton (match day 33)

of the year. It is likely that rains during the match, but UK-based players are used
to that situation. Probably, it will not have a huge impact on the outcome of the
event.
Potential bets:
• Arsenal- 1.50
Probable line-ups:
• Arsenal- 4-2-3-1: Cech; Bellerín, Mustafi, Chambers and Kolasinac; Elneny and
Ramsey; Ozil, Wilshere and Welbeck; Aubameyang
• Southampton- 4-2-3-1: McCarthy; Soares, Stephens, Hoedt and Bertrand;

Lemina and Romeu; Tadic, Davis and Redmond; Austin
• Arsenal- Out: Cazorla, Mkhitaryan and Ospina (all injured)
• Southampton- None
• Analysis- Besides the “chronically” injured Santi Cazorla, Arsenal has 1

important player missing (Mkhitaryan), while Southampton has all of its players
available for this match.
Discipline table:
points: 12.5
• Southampton (average per match away)- Yellow cards: 1.9, red cards: 0.0,
118
while Southampton is above the average in terms of aggression, when playing
away (although red cards were not shown to any of its players, so far). This means
that The Saints may win more duels and recover the ball more swiftly, but on the
other hand, they are probably going to be more exposed to disciplinary
conditioning through yellow cards.
• 10 matches- 4 wins for Arsenal (40%), 2 draws (20%) and 4 wins for
Southampton (40%)
• Analysis- There is not a significant bias favouring any of the teams, as each one
was able to win the same number of matches against the other, at home and away.
However, it is important to take into account that this season, Southampton is
performing considerably worse than usually, as The Saints are fighting to avoid
relegation, when over the last 4 seasons they finished always within the first half
of the table.
• Overall- Arsenal: 6th place, 51 points, 58 goals scored and 41 conceded;

Southampton: 18th place, 28 points, 29 goals scored and 47 conceded
• Away- Southampton: 13th place, 12 points, 13 goals scored and 26 conceded
• Analysis- Arsenal is one of the teams in the Premier League that depends more
on the home advantage factor to score goals and win matches. Southampton, by
its turn, has been the second worst team of the Premier League at home. Even so,
The Saints conquered more points at home, also having a better goal difference
than when playing away from the St Mary’s Stadium. Overall, Arsenal conceded
only 6 goals more than Southampton, but has a much stronger attack, composed
by several world-class players, such as Aubameyang, Lacazette, or Ozil. As a
119
consequence of that, Arsenal has nearly the double of goals scored and points
obtained.
• Southampton- won 57% of its points at home and 43% when playing away,
were scored away
Southampton scored 10% more and conceded 10% less goals at home. Also, the
team achieved 14% more points at St Mary’s, in spite of being the second worst
team at home, just above West Bromwich. Given so, we find these numbers
consistent with the performance below the expectations that Southampton is
demonstrating this season.
• Arsenal has a relative form of -0.15, while Southampton scores -0.02. This means
that in the last 8 matches, Southampton conquered less points on average, than the
average of points conquered over the course of the entire season, so far.
Nonetheless, Arsenal has a relative form 0.13 points lower than The Saints. That
being said, it is easy to understand that Southampton is facing a better moment
than The Gunners, who lost 4 of their last 8 matches
Scoring tables:
• Arsenal scores, on average, 2.63 goals per match at home, while Southampton
we obtain Arsenal’s “scoring strength”, which corresponds to 2.18 goals
120
• Southampton scores, on average, 0.87 goals per match when playing away, while
Arsenal concedes 1.06 goals per match at home. If we average these numbers, we
obtain Southampton’s “scoring strength”, which corresponds to 0.97 goals
• Arsenal likes to play a possession-based football, using short passes to control the
game in the opposition’s half. However, Southampton is strong at stealing the ball
from the opposition
• Both teams are weak in aerial duels, and also at defending counter-attacks
• Southampton is weak and Arsenal is even weaker at avoiding individual errors
• Arsenal is weak at defending attacks down the wings, and Southampton likes to
play with width, specially down the left wing. In contrast, Arsenal prefers to
conduct its attacks through the middle
• Southampton is weak at protecting the lead, while Arsenal is strong at coming

back from losing positions
• Arsenal is strong at creating long shot opportunities, while Southampton is very

weak at defending against long shots
• Arsenal is strong at finishing scoring chances, while Southampton is weak
Team psychology:
• Team spirit and motivation- In the Premier League, Arsenal is 5 points above
the 7th placed Burnley and 5 points below the 5th ranked Chelsea. Arséne Wenger
is a spent manager, that can no longer produce good results and inculcate a
positive spirit among the players. The veteran Frenchman has a contract for
another season, but will only remain as Arsenal’s manager if he is able to win the
Europa League, and to perform well in the last matches of the current season. The
big question is if whether the players want Arsène to remain in charge or not, and
because of this interrogation, we classify The Gunners’ team spirit and motivation
as medium. Arsenal is the 3rd best team in the League at home, and will face an
opponent that is struggling to leave relegation zone. In fact, The Saints are
desperate to get points, in order to secure a place in the next edition of the Premier
121
League, so they should be highly motivated to obtain at least 1 point from this
match
• Confidence- Arsenal is performing below the expectations in the Premier League,

ranking 6th, 13 points below the 4th place, that this season grants access to the
2018/2019 Champions league groups stage. Although The Gunners failed to
defend the FA Cup title, losing away to Nottingham Forrest in the 3rd round, they
were able to reach the League Cup final (lost 3-0 to Manchester City), and still
have good chances to qualify for the semi-finals of the Europa League, after
defeating CSKA in the first leg (4-1), at The Emirates Stadium. The Saints are
having their worst season since they got promoted back to the Premier League, in
2012/2013. The club has a relatively good squad, and was expected to finish
comfortably above mid-table, as it has been the case over the last years. The fact
is that Southampton is in relegation zone, and had to sack Mauricio Pellegrino,
whose spell at St Mary’s should be described as terrible, to say the least. When it
comes to the national cups, The Saints were eliminated from the League Cup by
Wolverhampton, in the 2 nd round (2-0 at home), but managed to qualify for the
semi-finals of the FA Cup, which is their last hope of saving the season. After
evaluating the situation of both clubs, we believe that Arsenal’s confidence should
be classified as medium, and Southampton’s as very low
worse, his impact on the club has to be considered very high. Mauricio Pellegrino
was fired in the match day 31, after losing 3-0 away vs Newcastle, leaving
Southampton in 17th place, just 1 point above the relegation zone. After that, Mark
Hughes took charge and in his first match, he was able to qualify Southampton
for the semi-finals of the FA Cup, defeating Wigan away (2-0). However, in the
next match, Southampton categorically lost 3-0 away vs Newcastle, and was sent
to relegation zone. Because Mark Hughes did only 2 matches at the helm of
Southampton, we decided that it would be unfair to classify his managerial impact
on the club
Team fatigue:
• Toughness of previous match- In the last match day, Arsenal beat Stoke City 3-
122
0, at home. However, Stoke’s defence was a tough nut to crack, as The Gunners
only scored the first goal by the minute 75, through a penalty kick converted by
the newly-signed Pierre-Emerick Aubameyang. Southampton, by its turn, was
destroyed by West Ham United, losing 3-0 away, with all the goals being scored
within the first 45 minutes of the match. Having this in consideration, we consider
Arsenal’s toughness of previous match as medium, and Southampton’s as the
highest
• Resting- In the match versus Stoke, Petr Cech was unavailable, due to a groin
injury, and Ospina was selected to replace him. Wenger also preferred to use
Monreal as left full-back (played the entire match), instead of Kolasinac. Against
West Ham, Southampton used a more offensive 4-4-2 formation, with Hojbjerg
(box-to-box midfielder) instead of Oriol Romeu (defensive midfielder), and the
Italian striker Gabbiadini instead of the attacking midfielder Davis. As Arsenal is,
at least in theory, a stronger opponent than West Ham, it makes sense that Mark
Hughes tries to reinforce the midfield area, risking less than against The
Hammers. During the match, left-winger Redmond was replaced by Boufal
(minute 67). In conclusion, from the probable line-ups for today, the number of
players that rested was 2 for each side
• National team participation- The last round of international matches finished at

28/03/2018, and so this will not be considered
123
• Analysis- Our Poisson regression model grants more than 70% chances to an
Arsenal’s win, which can be converted into an odd of 1.42. However,
Southampton chances might be overestimated, as the team is performing much
worse this season than in the last one. If we made our decision just by looking at
these numbers, betting on Arsenal at 1.50 would surely have positive expected
value.
Betting decision:
• In 2017/2018, Southampton is performing highly below the initial expectations,

currently facing the serious possibility of being relegated to the Championship.
Because of that, the team must feel motivated (and pressured) to get points, but
on the other hand, the players confidence levels should be very low, at this point.
In contrast, Arsenal is in a much more neutral situation, as the club has its position
in the Premier League perfectly stabilized (which is not a good thing, in this case).
The Gunners have been a very consistent team at home, and have world-class
players in the attack, so we can expect them to dominate the possession of the ball,
with Southampton trying to hold their offensive line, and then exploiting
Arsenal’s defensive weaknesses through the wings, with crosses targeting the
head of powerful striker Charlie Austin. Our Poisson regression model establishes
the threshold of positive expected value at 1.42, which emphasizes even more our
belief that backing an Arsenal’s win at 1.50, given the context of both teams, has
positive expected value.
124
Annex IX- Arsenal vs West Ham (match day 35)

of the year. It is likely that rains during the match, but UK-based players are used
to that situation. Probably, it will not have a huge impact in the outcome of the
event.
Potential bets:
• Arsenal- 1.41
• Arsenal and over 1.5 goals- 1.58
Probable line-ups:
• Arsenal- 4-2-3-1: Cech; Bellerín, Mustafi, Chambers and Kolasinac; Elneny and
Xhaka; Nelson, Iwobi and Welbeck; Aubameyang
• West Ham- 3-4-2-1: Hart; Rice, Ogbonna and Cresswell; Zabaleta, Kouyaté,
Noble and Masuaku; Mário and Lanzini; Arnautovic
• Arsenal- Out: Cazorla and Mkhitaryan (all injured)
• West Ham- Out: Obiang, Reid, Byram and Antonio; doubtful: Collins (all
injured)
• Analysis- Besides the “chronically” injured Santi Cazorla, Arsenal has 1

important player missing (Mkhitaryan). West Ham, by its turn, will miss the
contributions of mid-importance players like the defender Winston Reid, the
winger Michail Antonio or the midfielder Pedro Obiang.
Discipline table:
125
points: 15.3
• West Ham (average per match away)- Yellow cards: 2.0, red cards: 0.1, booking
points: 22.8
while West Ham is highly above the average in terms of aggression, when playing
away. This means that The Hammers may win more duels and recover the ball
more swiftly, but on the other hand, they will probably be more exposed to
disciplinary conditioning through yellow and red cards.
• 8 matches- 5 wins for Arsenal (62,5%), 2 draws (25%) and 1 win for West Ham
(12,5%); 6 times over 1.5 (75%) and 2 (25%) times under
• Analysis- Considering the outcomes of the matches between these 2 teams over
the last 3 years, it is possible to identify a bias favouring Arsenal. The last match
ended with an Arsenal’s home win (1-0), which eliminated West Ham from the
Lague Cup. However, the last Premier League match between Arsenal and West
Ham ended in a tie (0-0), at the London Stadium. When it comes to the number of
goals, although the vast majority of the matches ended with over 1.5 goals, the
last 2 matches (both in 2017/2018) ended 0-0 and 1-0, in Arsenal’s favour.
• Overall- Arsenal: 6th place, 54 points, 62 goals scored and 45 conceded; West
Ham: 14th place, 35 points, 41 goals scored and 59 conceded
• Away- West Ham: 15th place, 12 points, 21 goals scored and 38 conceded
• Analysis- Arsenal is one of the teams in the Premier League that depends more
on the home advantage factor to score goals and win matches, as 41 out of 54
points, and 45 out of 62 goals were obtained at home. Besides that, Arsenal
126
conceded 16% more goals when playing away from the Emirates. On the other
hand, The Hammers scored virtually the same number of goals at home and away,
although they conceded 28% more goals when playing away. Also, almost 2/3 of
West Ham’s league points were achieved at the London Stadium. This attests
West Ham’s dependence on the home advantage factor, especially when it comes
to its aptitude of defending solidly. In the Premier League table, West Ham is
ranked 8 places below Arsenal, with less 19 points, less 21 goals scored and more
14 goals conceded. In fact, The Hammers have the worst defence away from
home, and the third worst defence overall (just above Watford and Stoke). As
Arsenal has a very strong offensive line, composed by several world-class players,
such as Aubameyang, Lacazette, or Ozil, we anticipate that West Ham may
experience severe difficulties at the Emirates Stadium.
• West Ham- won 66% of its points at home and 34% when playing away, scored
were scored away
the home advantage, even in comparison with Premier League’s average. As
mentioned above, West Ham scored 49% of its goals at home and 51% when
playing away. Also, the team conceded 28% more goals and achieved 32% less
points away from the London Stadium. Summing up, both teams are considerably
dependent on the home advantage factor, but Arsenal is even more dependent than
The Hammers, which is interesting once that, at least in theory, better teams
should be less dependent of the home advantage to win.
127
• Arsenal has a relative form of -0.14, while West Ham scores -0.06. This means
that in the last 8 matches, West Ham conquered less points on average, than the
average of points conquered over the course of the entire season, so far.
Nonetheless, Arsenal has a relative form 0.09 points lower than The Hammers.
That being said, it is easy to understand that The Gunners are facing an even worse
moment than West Ham United
Scoring tables:
• Arsenal scores, on average, 2.65 goals per match at home, while West Ham
we obtain Arsenal’s “scoring strength”, which corresponds to 2.45 goals. Besides
that, Arsenal’s matches at home ended with over 1.5 goals 94% of the times
• West Ham scores, on average, 1.24 goals per match when playing away, while
Arsenal concedes 1.12 goals per match at home. If we average these numbers, we
obtain West Ham’s “scoring strength”, which corresponds to 1.18 goals. Besides
that, West Ham’s away matches ended with over 1.5 goals 94% of the times, as
well
• Analysis- If we sum Arsenal and West Ham scoring strengths, we obtain a total
of 3.63 goals expected in the forthcoming match. Besides that, if we average both
teams’ percentage of matches that ended with over 1.5 goals, we get a 94%
converted into an odd of 1.06
• Arsenal likes to play a possession-based football, using short passes to control the
game in the opposition’s half. West Ham, by its turn, uses the typical British style
of play, often attempting long balls and crosses to the head of an offensive
reference, such as Andy Carroll. As Arsenal defends high on the pitch, using the
offside trap strategy, there might be some space to exploit behind The Gunners
defensive line
• Both teams are weak at avoiding offside, and also at defending counter-attacks
• Arsenal is weak at defending attacks down the wings, and West Ham likes to play
128
with width. In contrast, Arsenal prefers to conduct its attacks through the middle
• Arsenal is weak at protecting the lead, while both teams are strong at coming back
from losing positions
• Arsenal is strong at attacking and defending set pieces, while West Ham is weak
at defending set pieces
Team psychology:
• Team spirit and motivation- In the Premier League, Arsenal is 2 points above
the 7th placed Burnley and 6 points below the 5th ranked Chelsea. Arsène Wenger
is a spent manager, that can no longer produce good results and inculcate a
positive spirit among the players. For that reason, the veteran Frenchman will not
finish his contract (expires in the summer of 2019), leaving Arsenal’s helm by the
end of this season. Arsenal is the 3rd best team in the League at home, and will
face an opponent that is struggling to avoid relegation. In fact, The Hammers need
to get points in order to secure a place in the next edition of the Premier League,
hence they should be highly motivated to obtain at least 1 point from this match.
Furthermore, this is a London derby, which by itself is a factor of motivation, due
to the historical rivalry between 2 teams that are based in the same city.
Consequently, we believe that both teams’ spirit and motivation should be
considered high
• Confidence- Arsenal is performing below the expectations in the Premier League,

ranking 6th, 13 points below the 4th place, that this season grants access to the
2018/2019 Champions league groups stage. Although The Gunners failed to
defend the FA Cup title, losing away to Nottingham Forrest in the 3rd round, they
were able to reach the League Cup final (lost 3-0 to Manchester City), and still
have the chance to qualify for the final of Europa League, after drawing with
Atlético de Madrid in the first leg of the semi-finals (1-1), at The Emirates
Stadium. West Ham, by its turn, is fighting to avoid relegation, placing 14 th in the
league with 35 points (7 points above the relegation zone). The Hammers’
performances in this season have been very irregular, as they have been capable
of the best and the worst. In the domestic cups, West Ham was eliminated by
Wigan in the 4th round of the FA Cup, and by Arsenal in the quarter-finals of the
129
League Cup. After evaluating the situation of both clubs, we believe that Arsenal’s
confidence should be classified as medium, and West Ham’s as low
worse, his impact on the club has to be considered very high. This summer, West
Ham invested a considerable sum in well-known players, such as Arnautovic or
Chicharito, targeting for a place in the top half of the table, maybe even a
European qualification spot. The expectations were high, but West Ham began the
season terribly, and Slaven Bilic was fired after being defeated by Liverpool (1-4
at home, in the match day 11), leaving The Hammers in relegation zone. In his
replacement, the board appointed David Moyes, who was able to take the club out
of the relegation zone, although a place among the teams that will dispute the
2018/2019 edition of the Premier League is not secured yet. In spite of Moyes’
ability to improve the club’s situation in the league table, the reality is that West
Ham has a fairly good squad, and should been performing much regularly. For
that reason, we decided to classify Moyes’ managerial impact on the club as
medium
Team fatigue:
• Toughness of previous match- In the last match day Arsenal scored first, but
eventually loss 2-1 to Newcastle, at St. James Park. West Ham, by its turn,
disappointingly tied at home with Stoke City, in a match where the giant veteran
Peter Crouch scored first for The Potters (by the minute 79), but the home team
was able to equalize 10 minutes later, by Andy Carroll, who entered the pitch just
4 minutes earlier. Having this in attention, we consider that the toughness of
previous matches was high, in both cases
• Resting- In the match versus Newcastle, Arsène Wenger used a 4-3-3 formation,
with an offensive line composed by Lacazette (as striker), Aubameyang and Iwobi
(as wingers, instead of Nelson and Welbeck). Besides that, the Frenchman
reinforced the midfield with Willock playing alongside Elneny and Xhaka.
Wenger also preferred to use Monreal as the left full-back (played the entire
match), instead of Kolasinac, with Chambers playing as right full-back and Rob
Holding as central defender, instead of Bellerín. During the match, Willock was
130
replaced by Welbeck (68th minute), Chambers was replaced by Maitland-Niles

(78th minute) and Iwobi was replaced by Nketiah (86 th minute). Against Stoke
City, West Ham began with the same 11 as the probable line-up for this match,
but with Fernandes instead of Lanzini. During the match, Fernandes was replaced
by Chicharito (minute 76), Masuaku was replaced by Lanzini (minute 81) and
João Mário was replaced by Andy Carroll (minute 86)

• Analysis- Our Poisson regression model grants almost 75% chances to an

Arsenal’s win, which can be converted into an odd of 1.33. However, West Ham
chances might be overestimated, as the team is performing worse this season than
in the last one. The model also attributes a probability of 89,33% to this match
ends with more than 1.5 goals, which corresponds to an odd of 1.12. If we sum
1.33 + 0.12 (the threshold of positive expected value for an Arsenal’s win, plus
the threshold for the match ending with more than 1.5 goals), we obtain an odd of
1.46, which means that according to this model, betting on Arsenal at 1.41, and
on Arsenal and over 1.5 goals at 1.58 would have positive expected value.
131
Betting decision:
• In 2017/2018, West Ham is performing below the initial expectations, currently

facing the possibility of being relegated to the Championship. Because of that, the
team must feel motivated (and pressured) to get points, especially due to the
rivalry inherent to a derby, and the importance that the fans attribute to win against
a club from the same city (this last part is even more true for Arsenal, once that
this is the last derby in Wenger’s era). In contrast, Arsenal is in a much more
neutral situation, as the club has its position in the Premier League perfectly
stabilized (which is not a good thing, by the way). The historical head-to-head
over the last 3 years is clearly favourable to Arsenal, and both teams’ dependence
on the home advantage factor also favours Arsenal, in this specific case
(furthermore because the League’s second-best attack at home will face-off the
worst defence when playing away). Besides that, The Gunners have been a very
consistent team at home, and have world-class attacking players, thus we can
expect them to dominate the possession of the ball, with West Ham trying to hold
their offensive line, and then exploiting Arsenal’s defensive weaknesses through
crosses delivered from the wings, and also through poisonous counter attacks. The
scoring table analysis tells us that there is a 94% probability that this match ends
with over 1.5 goals (equivalent to an odd of 1.06), and our Poisson regression
model establishes the threshold of positive expected value at 1.33 for an Arsenal’s
win, and at 1.12 to the match ends with more than 1.5 goals. Considering all these
factors, we believe that the possibility of returning 58% on this investment seems
highly attractive.
132
Annex X- Tottenham vs Watford (match day 36)
• 30/04/2018 at 8:00 pm, Wembley Stadium (London)

• Analysis: Great weather to play football in the UK, at this time of the year. It is
unlikely that rains, which allows for a fast circulation of the ball, and consequently
for an open match.
Potential bets:
• Under 3.5 goals in the match- 1.58
Probable line-ups:
• Tottenham- 4-2-3-1: Lloris; Trippier, Sánchez, Verthongen and Davies; Dier and
Dembelé; Eriksen, Alli and Son; Kane
• Watford- 4-2-3-1: Karnezis; Janmaat, Mariappa, Kabasele and Britos; Doucouré

and Capoue; Femenía, Hughes and Richarlison; Deeney
• Tottenham- Out: Winks (injured)
• Watford- Out: Hoban, Chalobah, Cleverley, Deulofeu and Okaka; doubtful:

Prodl, Carrillo and Pereyra (all injured)
• Analysis- Watford has 4 long-time injured players (Hoban, Chalobah, Cleverley

and Deulofeu), plus 3 rotation players in doubt (Prodl, Carrillo and Pereyra), while
Tottenham will only miss its young talent Harry Winks (who is injured since the
20th of March).
Discipline table:
• Tottenham (average per match at home)- Yellow cards: 1.0, red cards: 0.0,
• Watford (average per match away)- Yellow cards: 1.3, red cards: 0.1, booking
133
points: 15.9
• Analysis- Both teams are below the Premier League average in terms of
aggressiveness, therefore, it seems unlikely that cards or injuries might condition
• 5 matches- 4 wins for Tottenham (80%) and 1 draw (20%); 3 times under 3.5
(60%) and 2 times over (40%)
• Analysis- Given these numbers, it is fair to conclude that there is a clear bias
favouring Tottenham, however, the last Premier League’s match between these
teams ended up in a draw, at Vicarage Road.
• Overall- Tottenham: 4th place, 68 points, 66 goals scored and 31 conceded,

Watford: 12th place, 34 points, 42 goals scored and 60 conceded
• Home- Tottenham: 5th place, 34 points, 32 goals scored and 12 conceded
• Away- Watford: 11th place, 14 points, 17 goals scored and 30 conceded
• Analysis- These numbers demonstrate that Tottenham is a much stronger team,

investing a lot more to be placed among the Champions League places (top 4),
while Watford is just a team that fights for achieving its stability in the Premier
League. Overall, Tottenham has the double of Watford’s points, more than that in
terms of the goals scored, and conceded virtually half as much goals as The
Hornets did, which is a very considerable difference.
• Tottenham- won 50% of its points at home and 50% when playing away, scoring
134
• Watford- won 63% of its points at home and 37% when playing away, scoring
were scored away
• Analysis- By looking at these numbers, it is possible to conclude that Tottenham

does not rely on the home advantage factor to get points, and also to score goals,
which is a sign of consistency. However, Tottenham depends on the home
advantage to avoid conceding goals. In the summer of 2017, Tottenham’s board
made an effort to keep all the most important players from the previous season,
but sold the right full-back Kyle Walker to Manchester City, for a fee that was, at
the time, the highest ever paid for a defender. Besides that, the left full-back
Danny Rose was injured most of the season, and the newly-signed young prodigy
Davinson Sánchez needed some time to adjust to his new team, and to the English
style of football, as well. This conjugation of factors provides an explanation for
the lack of defensive reliability shown in some of The Spurs’ matches. Watford,
by its turn, is significantly dependent on the home advantage to obtain points, and
is slightly above the Premier League’s average when it comes to the number of
goals scored at home. The Hornets conceded exactly the same number of goals at
home and away, which in this case suggests defensive permeability, as Watford
has the 3rd worst defence of all the Premier League’ teams.
• Tottenham has a relative form of +0.38, while Watford scores -0.09. This means
that in the last 8 matches, the latter conquered less points on average, than the
average of points conquered over the course of the entire season, so far. On the
other hand, Tottenham has a positive relative form, that is +0.47 points higher
than the one relative to The Hornets. That being said, it is easy to understand that
The Spurs are facing a much better moment than Watford
Scoring tables:
• Tottenham scores, on average, 2.00 goals per match at home, while Watford
135
we obtain Tottenham’s “scoring strength”, which corresponds to 1.88 goals.

Besides that, Tottenham’s matches at home ended with under 3.5 goals 69% of
the times
• Watford scores, on average, 1.00 goals per match when playing away, while
Tottenham concedes 0.75 goals per match at home. If we average these numbers,
that, Watford’s away matches ended with under 3.5 goals 71% of the times
• Analysis- If we sum Tottenham and Watford scoring strengths, we obtain a total

of 2.76 goals in the forthcoming match. Besides that, if we average both teams’
percentage of matches that ended with under 3.5 goals, we get an 70% probability
that this match in particular ends with under 3.5 goals, which can be converted
into an odd of 1.43.
• Tottenham likes to play a possession-based football, using short passes to control

the game in the opposition’s half. However, Tottenham is also very strong in
counter attacks, often using through balls to create scoring chances. In contrast,
Watford is very weak at defending counter attacks. As both teams use an offside
trap defensive strategy, there might be some space to exploit behind the defensive
lines’ backs, although The Spurs are weak at avoiding offside
• Tottenham is strong at attacking and defending set pieces, while The Hornets are
just weak at defending them
• Tottenham is strong at protecting the lead, while Watford is very weak
• Watford preferentially conducts its attacks down the left flank, where its most
talented player, Richarlison, is deployed. On the contrary, Tottenham usually
attacks through the middle
Team psychology:
• Team spirit and motivation- Watford suffered upsetting early eliminations from
domestic cups, but at this stage has already mathematically avoided relegation
(the main goal for this season). Now, the only focus is to achieve the best possible
136
classification in the Premier League, preferentially placing within the first half of
the table. The team seems united enough around Gracia, although some important
players expressed, at the time, their dissatisfaction about Silva’s dismissal (e.g.
Richarlison), and so we consider that Watford’s team spirit and motivation should
be classified as medium, at this point. This has been an unusual season for
Tottenham, who is playing all of its home games at the national Wembley
Stadium, while the new White Hart Lane is under construction. Although The
Spurs still have the support of its fans, being temporarily separated from their
home ground probably reduces the home advantage benefits to the team’s spirit
and motivation. However, Tottenham is clearly favourite and still has chances of
reaching the 2nd place. Therefore, its players should be highly motivated to win
this match
• Confidence- The Hornets have been performing much better in 2017/2018 than
in the last season, where they ranked 17th, the last place above relegation zone. In
the national cups, Watford was surprisingly eliminated from the League Cup by
Bristol (2-3 at home), in the 3rd round, and defeated by Southampton (1-0 away)
in the 4th round of FA Cup. Additionally, in the last 6 matches for the Premier
League, Watford lost 4 times and tied twice, hence we think that the team’s level
of confidence should be classified as low. When it comes to the internal cups,
Tottenham suffered a surprising early elimination from the League Cup, at the
hands of city rivals West Ham (2-3 at home), but reached the semi-finals of the
FA Cup (eliminated by Manchester United). In the Champions League,
Tottenham passed the group stage and then had an electrifying clash against
Italian titans Juventus, in which The Spurs performed incredibly well, but
eventually ended up being eliminated (2-3 on aggregate), perhaps due to the
team’s lack of experience. At the moment, Tottenham is 4 th in the Premier League,
5 points above rivals Chelsea, and 3 points below Liverpool (with a game in hand).
As one can observe, the difference of scenarios between these teams is huge, and
so, Tottenham should approach this match very confident in obtaining a good
score
• Managerial impact- Mauricio Pochettino began his coaching career at

Barcelona, being appointed as the manager of RCD Espanyol (the last team where
137
he played). After two and a half seasons, his work caught Southampton’s
attention, and he moved on to the Premier League. After just one season at St
Mary’s (2013/2014), the attractive style of play that Pochettino was able to
implement granted him a place as the head coach of Tottenham Hotspurs, a club
with higher ambitions, but struggling to reach Champions League qualification
spots (6th place in 2013/2014). In his first season at White Hart Lane, Pochettino
repeated the 6th place, but The Spurs’ board decided to give him another shot. In
the next 2 seasons (2015/2016 and 2016/2017), the Argentinian put The Spurs
back on track, placing 3rd and 2nd respectively, and fighting for the Premier League
title until the last match days. Furthermore, this season Tottenham qualified for
the knockout stages of the Champions League for the first time since 2010/2011
(lost dramatically to last year’s finalist Juventus). With Pochettino, Tottenham
achieved consistent results, playing stylish football while investing relatively less
than its direct rivals, consequently, his managerial impact must be considered very
high. Marco Silva was fired on the 20th January, after suffering a categorical away
defeat against Leicester (2-0), being replaced by Javi Gracia, who started by
changing the usual formation from 3-4-2-1 to 4-2-3-1. Besides that, some players
that were important under Silva’s management, like Hughes, Carrillo or Gray, lost
relevance to others, such as Roberto Pereyra or Troy Deeney. The Spanish coach
began his spell by drawing away, against Stoke (0-0). After that, he achieved an
incredible win at home against Chelsea (4-1), but then lost away to direct
competitor West Ham (2-0). In the following 3 matches, Watford was able to beat
Everton and West Brom (both 1-0) at home, but then lost 3-0 away vs Arsenal,
and 5-0 vs Liverpool (both times when playing away). Since Javi Gracia took the
helm, Watford has been stable in the mid-table, therefore we think that his
managerial impact at Vicarage Road has been normal
Team fatigue:
• Toughness of previous match- In the previous match, Watford and Crystal

Palace tied 0-0 at Vicarage Road. The Hornets were superior to Palace, but were
unable to materialize that superiority in goals, although Crystal Palace had some
chances of its own. On the same day (21 st of April), Tottenham and Manchester
United disputed a place in the FA Cup final, at Wembley. The Spurs scored first,
138
by Dele Alli (11th minute), dominated possession (64%) and had more attempts
on goal, but were less effective, and allowed United to turn things in its favour,
with goals scored by Alexis Sánchez (24 th minute) and Ander Herrera (62nd
minute). Having all this in consideration, we decided to classify Tottenham’s
toughness of last match as the highest, and Watford’s as medium
• Resting- Against Crystal Palace, the only players from Watford’s probable line-
up that rested entirely were the full-backs, Janmaat and Britos. Besides that,
Pereyra was replaced by Richarlison (55th minute), Okaka was relaced by Femenía
(70th minute) and Hughes was replaced by Gray, 9 minutes later. Tottenham began
the semi-final of the FA Cup exactly with the same probable line-up for the match
against Watford, but replaced Ben Davies for Lucas Moura as a response to the
goal scored by Herrera (68th minute), Wanyama for Dembelé (10 minutes later)
and Lamela for Son (by the minute 86, in a final attempt to obtain the equalizing
goal)

139
win, less than 10% to a draw, and only 4,23% to an away win. Although the model
strongly indicates that Tottenham has very high probabilities of winning, when it
comes to the number of goals in the match, the probabilities for under and over
3.5 goals are more or less even. In fact, the model estimates a probability of
51,85% for the former (equivalent to an odd of 1.93), and of 48,15% for the latter
(corresponds to an odd of 2.08). If we based our decision only on the data provided
by the Poisson regression, betting on under 3.5 goals in this match at 1.58 could
not be seen as a reasonable choice.
Betting decision:
• After carefully analysing all the factors that may influence the outcome of this
match, one thing that becomes clear is The Spurs’ favouritism to win. As a matter
of fact, unlike Watford, Tottenham has no relevant players missing or at risk for
this match, and is going through a positive moment of recent form. In addition,
the historical head-to-head over the last 3 years suggests that there is a bias in
favour of the home team. The Poisson regression grants Tottenham 85,78%
chances of winning this match, which is equivalent to an odd of 1.17, and a
probability of 51,85% that this match ends with under 3.5 goals, which
corresponds to an odd of 1.93. However, the scoring tables tell us that there is a
70% probability that the match ends with less than 3.5 goals, which can be
converted into an odd of 1.43. Moreover, the sum of both teams’ scoring strengths
is equal to 2.76 goals. We believe that Tottenham will win this match, but will not
try very hard to score a lot of goals, in an attempt to manage the players’ effort,
as many of them will be called for the FIFA World Cup 2018. Furthermore, we
expect Watford to approach this match with caution, aware of Tottenham’s goal-
scoring ability. In this case, the subjective analysis will outweigh the Poisson
quantitative analysis in our decision-making process. In combination with the data
retrieved from the scoring tables, it makes us feel confident in the success of
betting on under 3.5 goals, at 1.58.
140
Annex XI- Southampton vs Manchester City (match day 38)
• 13/05/2018 at 3:00 pm, St Mary’s Stadium (Southampton)

• Analysis: Weather as perfect as it gets to play football in the UK, allowing for a
fast circulation of the ball, and consequently for an open match.
Potential bets:
Probable line-ups:
• Southampton- 3-4-2-1: McCarthy; Stephens, Yoshida and Hoedt; Soares,

Romeu, Hojbjerg and Bertrand; Redmond and Tadic; Austin
• Manchester City- 4-3-3: Ederson; Walker, Stones, Otamendi and Mendy; De

Bruyne, Fernandinho and Gundogan; Sterling, Jesus and Sané
• Southampton- Out: Davis (injured) and Boufal (internal disciplinary issues);

doubtful: Lemina and Bednarek (both injured)
• Manchester City- Out: Aguero (injured)
• Analysis- Southampton will miss 2 important players: the central midfielder

Steven Davis and the attacking midfielder Soufiane Boufal, who was banished
from The Saints’ first team by Mark Hughes, after refusing to enter the pitch
against Chelsea (2-3 loss, at home). Additionally, Mario Lemina and Bednarek
are both having physical problems, and consequently in doubt for this match. On
the other hand, Manchester City will only miss its striker Sergio Aguero. Although
this is an important casualty, The Citizens have Gabriel Jesus, who is more than
capable of performing at a similar level.
Discipline table:
141
• Southampton (average per match at home)- Yellow cards: 1.4, red cards: 0.0,
• Analysis- At home, Southampton is less aggressive than most Premier League’s

teams, while Manchester City is way above the average in terms of aggression,
when playing away. This means that The Citizens may win more duels and
recover the ball more swiftly, but on the other hand, they are probably going to be
more exposed to disciplinary conditioning through yellow or red cards.
• 7 matches- 1win for Southampton (14%), 1 draw (14%) and 5 wins for
Manchester City (72%); 7 times over 1.5 (100%)
• Analysis- Given these numbers, it is fair to conclude that there is a clear bias
favouring Manchester City. On top of that, every single match between these
teams over the last 3 years ended with over 1.5 goals.
• Overall- Southampton: 17th place, 36 points, 28 goals scored and 56 conceded,

• Home- Southampton: 19th place, 19 points, 20 goals scored and 25 conceded
• Analysis- These numbers demonstrate that Manchester City is the best team in
the Premier League, and of course, a much stronger team than Southampton.
Overall, The Citizens have more 61 points, 77 goals scored and less 29 conceded
than Southampton, which is a huge difference.
142
• Southampton- won 53% of its points at home and 47% when playing away,
were scored away
• Analysis- This data allows one to understand that neither of the teams depend
much on the home advantage factor to gain points, and also to score and prevent
goals.
• The Saints have a relative form of -0.03, while City scores -0.17. At first glance,
it may seem weird that Southampton has a better level of relative form than
Manchester City. This is because The Citizens have been very consistent over the
course of the season, only losing against Liverpool and Manchester United, and
drawing against Everton, Crystal Palace, Burnley and Huddersfield. The defeat vs
United and the draw vs Huddersfield happened over the last 8 matches.
Southampton performed very poorly under Pellegrino’s guidance. When the
Argentinian was dismissed, Mark Hughes took charge and the results improved
significantly, as Southampton was able to leave relegation zone, after winning 0-
1 against Swansea (away). In the last 8 matches, The Saints won 2, lost 4 and tied
2
Scoring tables:
• Southampton scores, on average, 1.11 goals per match at home, while

Manchester City concedes 0.72 goals per match when playing away. If we average
these numbers, we obtain Southampton’s “scoring strength”, which corresponds
to 0.92 goals. Besides that, Southampton’s matches at home ended with more than
1.5 goals 72% of the times
143
while Southampton concedes 1.39 goals per match at home. If we average these
numbers, we obtain City’s “scoring strength”, which corresponds to 1.92 goals.
Besides that, The Citizens’ away matches ended with more than 1.5 goals 83% of
the times
• Analysis- If we sum Southampton and Manchester City scoring strengths, we

obtain a total of 2.84 goals in the forthcoming match. Moreover, if we average
both teams’ percentage of matches that ended with over 1.5 goals, we get a 78%
converted into an odd of 1.28
• Both teams are weak in aerial duels
• Manchester City is very strong at creating scoring chances, especially by using

through balls, and also by using the individual skill of its players. Furthermore,
City is very strong at finishing scoring chances, while Southampton is weak
• Man City is very strong at creating long shot opportunities, while The Saints are
very weak at defending against long shots
• Manchester City is very strong at attacking and defending set pieces, and strong
at shooting from direct free kicks
• Manchester City style is related with keeping possession of the ball, using short
passes to control the game in the opposition’s half. Moreover, The Citizens defend
high on the pitch, using a very well automatized offside trap strategy, which
allows them to be the team that suffered the less number of goals in the League,
this season
Team psychology:
positive team culture. The Citizens already assured the main objective of winning
144
the Premier League, but even so, they should be highly motivated to win it in
style, targeting for an unprecedented record of 100 points. Although
Southampton’s season was a huge disappointment to its supporters, the club was
able to duck the menace of relegation, which due to the team’s underperformance,
became the main objective for 2017/2018. In spite of the Soufiane Boufal’s
incident, the group showed union, tenacity and mental strength when pursuing the
permanence in the Premier League, especially when was able to recover from the
traumatic losses in the matches against Arsenal and Chelsea, obtaining a positive
final sequence that culminated in the amazing win against Swansea. However, in
this final match, there is nothing else to fight for, and consequently, there might
be some propensity to relaxation. Given so, we consider that, for this match in
particular, Southampton’s levels of motivation should be low
• Confidence- In 2017/2018, The Citizens have been perhaps the best team in
Europe, winning the Premier League comfortably in the match day 34, after Man
United surprising defeat vs West Bromwich, at Old Trafford. As a matter of fact,
Manchester City only lost to Shakhtar Donetsk (Champions League), Wigan (FA
Cup), Liverpool (twice for the quarter-finals of the Champions League and once
for the Premier League) and Manchester United (Premier League), and tied
against Everton, Crystal Palace, Burnley and Huddersfield (all for Premier
League). On these circumstances, we believe it is fair to say that Man City’s
confidence levels must be very high. On the contrary, The Saints are having their
worst season since they got promoted back to the Premier League, in 2012/2013.
The club has a relatively good squad, and was expected to finish comfortably
above mid-table, as it has been the case over the last years. The fact is that
Southampton barely avoided relegation, and had to sack Mauricio Pellegrino,
whose spell at St Mary’s should be described as terrible, to say the least. When it
comes to the national cups, The Saints were eliminated from the League Cup by
Wolverhampton, in the 2 nd round (2-0 at home), but managed to qualify for the
semi-finals of the FA Cup, in which they were eliminated by Chelsea (2-0 at
Wembley). For those reasons, we believe that the team’s confidence levels should
be classified as low
• Managerial impact- Mauricio Pellegrino was fired in the match day 31, after
145
losing 3-0 away vs Newcastle, leaving Southampton in 17 th place, just 1 point

above the relegation zone. After that, Mark Hughes took charge and in his first
match, was able to qualify Southampton for the semi-finals of the FA Cup,
defeating Wigan away (2-0). However, in the next match, Southampton
categorically lost 3-0 away vs Newcastle, and was sent to relegation zone. After
that, a sequence of 3 losses in a row anticipated the worst scenario for The Saints,
although they were a tough nut to crack against strong opponents like Arsenal and
Chelsea. In the last 4 matches for the Premier League, Southampton was able to
bounce back, achieving 2 wins and 2 draws, including a dramatic victory in the
last match against Swansea, that guaranteed Southampton the relegation
avoidance. Because of that, and also because the team improved significantly the
quality of its football, we decided to classify Hughes’ managerial impact in
Southampton as high. Guardiola arrived at Manchester and after a mediocre first
season, has been able to turn Manchester City into one of the most complete teams
in Europe. The financial investment in the squad was huge, but nowadays City
plays in the image of its coach, and is definitely one of the best sides that he built
in the entire career. That being said, we have to consider Guardiola’s impact in
the club as very high
Team fatigue:
• Toughness of previous match- The Citizens’ last match took place on the 9th of
May, at the Ethiad Stadium, where they achieved a comfortable win against
Brighton (3-1). As usual, Man City dominated its opponent, and the final victory
was never at risk, therefore, we decided to rate the toughness of City’s last match
as low. One day earlier, at Wales, Southampton disputed the first place above the
relegation zone with Swansea. In a well-balanced match, Southampton was able
to strike the only goal by the minute 72. Gabbiadini, who entered the match just 4
minutes earlier, was the goal scorer. Due to the importance of what was there at
stake, and also because of the evenness of the match itself, we considered the
toughness of Southampton’s last match as the highest
• Resting- Against Brighton, Manchester City started with a completely different

defensive line, in comparison with the probable line-up for this match: Danilo,
Kompany, Laporte and Zinchenko began the match, instead of Walker, Stones,
146
Otamendi and Mendy, although the latter replaced Zinchenko by the minute 76.
Besides that, Guardiola decided to spare the key midfielder Kevin De Bruyne, that
was replaced by the veteran Yaya Touré (who was substituted by Nmecha, on the
86th minute). 3 minutes earlier, the Catalan manager already have changed Jesus
for Diaz. When it comes to Southampton, the only difference between the starting
11 of the game against Swansea and the probable line-up for this match is that
Yoshida is expected to substitute Bednarek (who is in doubt, as mentioned above).
During the match at the Liberty Stadium, Redmond was replaced for Shane Long
by the minute 64, and 4 minutes later, the defender Bednarek was replaced by the
striker Manolo Gabbiadini. These changes were made with the aim of providing
the firepower needed to score, and were proven successful, as Gabbiadini netted
the winning goal that assured to The Saints a place in the next edition of the
Premier League. The final substitution was made by the minute 83, when Mark
Hughes decided to replace Tadic for McQueen

147
• Analysis- The Poisson regression model grants 67,28% chances to an away win,
which can be converted into an odd of 1.49. The probability given by the Poisson
regression to the match ends with more than 1.5 goals is 78,16%, which is
equivalent to an odd of 1.28. It is important to understand that Southampton’s
chances might be overestimated, as they performed much better in 2016/2017 than
in this season, while City’s chances, on the contrary, are probably underestimated
(as its performance improved a lot in the current season).
Betting decision:
• Manchester City is the best team in the Premier League (and quite possibly, in the
world) by far. The difference to the other teams is enormous, as The Citizens have
the best (and more expensive) squad (in terms of quality and quantity), the best
manager, the most balanced and refined style of play (best attack and defence,
both at home and away). Manchester City is an extremely confident team, that
never renounces on its playing principles, regardless of the opponent’s name or
status. With the exception of the away matches against Liverpool, every single
team that played against Manchester City just assumed its superiority and gave
away the control of the match, trying not to suffer a heavy loss. Both teams already
accomplished their main goals: Man City was crowned Premier League champion
and Southampton avoided relegation, after an uncommon struggled season.
However, City still wants to be the first team to surpass the barrier of the 100
points in the league. Also, the Poisson regression analysis allocates a probability
of 67,28% to an away win (odd of 1.49), and of 78,16% (odd of 1.28) to the match
ends with over 1.5 goals. The scoring table analysis grants the same probability
when it comes to the total number of goals in this match, plus a scoring potential
of 2.84 goals, resulting from the sum of both teams’ scoring strengths. Given this
context, our decision will be to bet on Manchester City and over 1.5 goals at 1.55,
in an attempt to maximize the return on investment, as we are very confident that
more than 1.5 goals will occur.
148

Forecasting Football

Uploaded by

Copyright:

Available Formats

Forecasting Football

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Forecasting Football

Uploaded by

Copyright:

Available Formats

What is the purpose of the project?

What is the purpose of the project?

What statistical models are discussed in the literature review?

What statistical models are discussed in the literature review?

Master of Science Final Project

João Maria Telles Moniz Côrte-Real

Supervisor: Professor José Joaquim Dias Curto, PhD

João Maria Telles Moniz Côrte-Real

Supervisor: Professor José Joaquim Dias Curto, PhD

To Professor Dias Curto, for the patience and availability

To my girlfriend Inês, the love of my life, for the understanding

To my aunts Isabel and Filipa, for loving me as a son

To my stepmother Domingas, for treating me as a son

To my mother, to my grandparents (Leonel, Graça, Inês and Filipe) and to my cousin

To my sister and niece, Gracinhas, for the laughs

To my cousin António, for the unconditional brotherhood

To all my best friends, for the good times spent together

Keywords: Betting Markets, Football, Forecasting, Sports Betting

JEL Classification System: Information and Market Efficiency, Sports Economics

A estratégia de precisão consistiu em apostar nos favoritos que joguem em casa, e

Surpreendentemente, se removêssemos a margem implícita, as casas de apostas

Palavras-chave: Mercados de Apostas, Futebol, Projeções, Apostas Desportivas

JEL Classification System: Information and Market Efficiency, Sports Economics

Após efetuar uma análise comparativa dos mercados de apostas nacionais e

o mais adequado. Analisaram-se também diversos artigos científicos, no sentido de aferir

Foram guardados registos de todas as apostas efetuadas, para posteriormente comparar as

Após o final da época 2017/2018, analisaram-se os resultados, e tiraram-se as respetivas

• Através do da aplicação do método preditivo idealizado pelo autor, geraram-se

• A estratégia relacionada com apostar em equipas caseiras amplamente favoritas

• Em relação à competitividade dos mercados de apostas Portugueses, verificou-se

3.1- Forms of placing bets

3.1.1- Fixed-odds betting

• Home win- 4.90 (1/4.90 = 20,4%)

• Draw- 3.90 (1/3.90 = 25,6%)

• Away win- 1.60 (1/1.6 = 62,5%)

• 20,4% + 25,6% + 62,5% = 108,5%, in this specific case the bookmaker’s

• Home win- (1/4.90) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 18,8%

• Draw- (1/3.90) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 23,6%

• Away win- (1/1.60) * (1/ ((1/4.90) + (1/3.90) + (1/1.60))) = 57,6%

• 18,8% + 23,6 % + 57,6% = 100%

A punter bets 5€ on the following outcomes:

• Manchester United beats West Ham away, at 1.75

• Manchester United will score more than 1.5 goals, at 1.60

• A draw in Barcelona vs Real Madrid, at 3.20

If all these outcomes occur, the punter will be rewarded with:

• 5€ * 1.75 * 1.60 * 3.30 = 46,20€.

3.1.2- Exchange betting

2. The betting exchange encourages the truthful exposure of information. Based on

3. Through the price mechanism, the betting exchange delivers a proficient

3.2- Legal context of the betting markets in Portugal

On December 6 of 2017, the Remote Gambling Association (RGA) announced the

3.3- Purpose of the project

4.1- Statistical models

4.1.1- Poisson regression models

4.1.2- Ordered probit/logit regression models

performance in the estimation accuracy diminished as a function of time in relation to the

Goddard & Asimakopoulos measured short-term performance levels through discrete

4.2- Fixed-odds betting markets efficiency

An example of a well-documented inefficiency in the fixed-odds betting markets is the