1
Faculty of Economics and Business
Study: Economics & Business
Field: Finance & Organization Economics
Efficiency of fixed-odds betting in
European Football
Bachelor thesis
by
Thomas Verbeek
0521310
Supervisor:
2
Abstract
This paper shows that by consistently betting on the best odds available amongst a group of bookmakers no profitable betting opportunities existed when betting on the favorite football (soccer) teams for a victory with odds at and below a threshold in the period 2011-2014. Results are based on fixed-odds bets of matches from 21 competitions from 11 countries in Europe. This paper extends earlier findings by Direr (2011) to recent years. Direr (2011) showed that profitable opportunities existed between 2000 and 2011 by betting on odds at or below a threshold.
The paper further shows that a favorite long-shot bias existed in both periods. This bias was smaller between 2011 and 2014 than between 2000 and 2011 and could not have been exploited in any profitable way. A distinction is further made between bets from different countries. Countries differ substantially in the amount of fans teams in certain leagues have and the income differences between clubs in different countries are large. This is assumed to lead to unequal betting volumes between leagues from different countries. The results show no profitable opportunities to be exploited in specific countries over both periods. The paper fits in the existing literature on betting markets and the favorite-longshot bias in particular.
3
Table of contents
Page number
1. Introduction 4
2. Explanation of fixed-odd betting and a review of the literature 6
2.1 Fixed-odds online betting 6
2.2 The betting market as a financial and information market 7
2.3 Favourite-longshot bias in sports betting markets 8
3. Data and methodology 12
4. Results 14
4.1 Odd thresholds at different time periods 14
4.2 Rates of return amongst countries 17
4.3 Regression 19 5. Discussion 20 6. Conclusion 22 Reference list 23
4
1. Introduction
This thesis explains the efficiency of fixed odds in the online betting industry in European football competitions. The online sports betting industry is a popular industry. In sports betting, one is forecasting the outcomes of sport events and placing a wager (a bet) on the outcome. There is a wide range of sport events on which bets can be placed. The total size of the global sports betting industry is difficult to estimate due to the lack of consistency in regulation and the size of the illegal sports betting market. Nevertheless, the estimated total size in 2013 was approximately 630 billion Euros1. William Hill, the biggest online bookmaker in the United Kingdom, had 1.5 billion Euros in profits while the total worth of the placed bets was over 22 billion Euros. Considering the amount of money that is circulating in this industry, surprisingly little academic research has been executed on this topic. To fill that void, this thesis aims to give an impression of the intriguing partly unexplored world of sports betting.
The online sports betting market is widely regarded as a specific type of financial market. My study builds on earlier literature in finance and especially on the favorite-longshot bias in online sports betting. It is linked to literature on the efficient market hypothesis, as described by Fama (1970) and Jensen (1978). If the betting market is weak-form efficient, no profit opportunities can be exploited by betting on odds within certain ranges.
The favorite-longshot bias is a phenomenon within the sports betting industry where on average, bettors overvalue “long shots“ and undervalue favorites. Yet this return may still be negative or unprofitable. It is a deviation from the market efficiency hypothesis. The favorite-longshot bias will be explained more thoroughly in the following section.
An important finding on the favorite-longshot bias and profitable betting opportunities has been proposed by Direr (2011), who wrote an influential article, stating that by consequently placing bets on favorite teams with odds at or below a threshold in national football leagues in Europe, a positive return could have been achieved between 2000 and 2011.
In this study, the results by Direr are extended to recent years (2011-2014). In addition, a distinction is made between countries in which the matches were played. The countries in our sample differ enormously in terms of revenues that the clubs within the
1
5
national leagues have. The big 5 European football leagues are included and some smaller leagues. The best teams from the big European countries have many fans, both nationally and internationally. It is likely that this also leads to much more betting on matches in those countries and on matches with the favorite teams involved accordingly. It is interesting to see whether the results indicate differences between the different countries’ profit opportunities. Data is used from 21 different national football leagues from 11 countries across Europe within a period of 14 years in total.
The research question of this thesis is:
Does the strategy that would have been profitable between 2000 and 2011 (according to Direr (2011)) of betting on favorites below a threshold of 1.21 hold for recent years (2011-2014) and can the strategy be made more profitable by differentiating between countries?
The results of this thesis show that the strategy that would have been profitable earlier was not profitable between 2011 and 2014. Even when differentiating between countries, the strategy remains significantly unprofitable in recent years. Between 2000 and 2011 structural betting on matches in some countries would have had a significant positive return, while this is no longer the case in recent years.
This has important implications for the academic work on the favorite-longshot bias and arbitrage opportunities in sports betting. It has also implications for bettors, who seek to exploit profitable betting opportunities.
The paper is structured as follows: section 2 explains the concept of fixed-odds betting and discusses relevant literature on betting markets and the favorite-longshot bias. In section 3 the data and methodology are described, while in section 4 the results are presented. A discussion follows in section 5. Lastly, in section 6 the conclusion is presented.
6
2. Explanation of fixed-odds betting and a review of the literature
In this section fixed-odds betting on football match outcomes is explained and an overview of the literature on betting markets and the favorite-longshot bias is presented. The section is structured as follows: the first subsection explains online betting and fixed-odds betting in particular and an example is given for clarification. In the second subsection similarities and differences between betting markets and financial markets are explained. The third subsection discusses relevant theory on the favorite-longshot bias.
2.1 Fixed-odds online betting
Bets on sport events can be placed at an online bookmaker. An online bookmaker functions as a market organizer for sports bets or wagers. In this role, a bookmaker accepts bets and makes sure it is able to make a profit in the long run. A wide variety of events occurring in a sports match can be bet on, being for instance end-results, half-time results, goal scorers and total goals scored.
In this thesis the focus is on fixed-odds betting on the end-result of a football match. It only matters whether a home team wins, a draw occurs or the away team wins. All three outcomes are possible in every match. An odd is fixed once a bet is placed. At that moment an amount of money is deducted from the client’s account and an agreement about a future cash flow from the bookmaker to the client is made when the bettor is right about the outcome. This agreement cannot be reversed or changed by the bookmaker. Bets can be placed at odds presented by a bookmaker. After a match is played, either an instant profit is made or a loss is incurred. The dataset that is used will be described further in a separate section.
The way odds are posted can be clarified by an example. The odds presented below were the odds posted by bookmaker Unibet for the three possible outcomes of a football match in the 2014-2015 football season in the highest Dutch football league ’de Eredivisie’: PSV wins: 1.06 Draw: 13.00 FC Dordrecht wins: 25.00 The odds above are posted in the decimal format, which is the most common format through Europe. Odds can be posted in various ways. Another frequently used format is the fractional
7
format, which is most common in the UK. A decimal odd of 1.50 is similar to a 1/2 fractional odd. A decimal odd of 1.50 means that 50 cents per 1 euro wagered is pure profit when a predicted outcome occurs, which is equal to 1 euro per 2 wagered (reported as 1/2 under the fractional format). And a 2.00 decimal odd is equal to a 1/1 fractional odd etc. The fractional format presents the instant profit that can be made per amount of money wagered. In this thesis, the decimal format will be used. Fractional odds have been translated into decimal odds for some bookmakers in the dataset that is used in this study. Other formats will not be discussed in this study.
In the example of the odds in the match above, betting €1 on PSV at Unibet will pay out €1.06 if PSV wins, which means an instant profit of €0.06 or 6%. Putting €1 on FC Dordrecht in the case that team wins, it will pay off €25, which means a profit of 24/1*100%=2400%. The odds deviate between different bookmakers. Another bookmaker, Bwin, pays out 1.05, 11.00 and 16.50 respectively, which is lower for every possible result. Adding up the three inverse values of the odds for the three possible outcomes gives a number larger than 1. This is to cover for the costs that bookmakers incur and the profit that is to be earned. The sum of the inverse values of the three outcomes is 106% for Unibet and 110% for Bwin. Therefore, the margin for Bwin in this example is larger than for Unibet.
In the above example, PSV is the favorite and FC Dordrecht is the outsider, also referred to as a longshot. The favorite-longshot bias is widely discussed in literature.
In the next subsection will be explained why sports betting is generally regarded as a type of financial and information market. In the last subsection, the favorite-longshot bias will be explained and some important papers will be discussed.
2.2 The betting market as a financial or information market.
Betting markets have attracted substantial attention by economists, as they are very suitable for testing market efficiency. Both financial and betting markets consist of a large number of participants, risking their wealth or capital on uncertain outcomes of future events. A lot of information about sports teams and events is widely available to the public, similar to companies and their respective stock prices.
Markets are called efficient if the prices of assets reflect information to the point where the marginal benefit of acting on information is not higher than the marginal costs of doing so (Jensen, 1978). This can be tested by taking historical prices of assets and seeing if
8
this does not lead to abnormal returns (Fama, 1970). Market efficiency was originally constructed for financial markets but also applies to wagering markets, as asset prices in financial markets can be replaced by match odds for betting markets.
It’s interesting to see whether the best betting strategies that are chosen by investors or bettors, lead to positive returns. If rates of return differ between varying betting strategies, the information revelation property of wagering markets is invalid. Rational behavior by participants is another condition of an efficient market: if there are profitable opportunities that can be used for arbitrage but investors fail to do so, they do not act rationally.
In the next subsection the favorite-longshot bias and profitable opportunities are described.
2.3 Favorite-longshot bias in sports betting markets
In this subsection specific literature on the favorite-longshot bias in sports betting markets is discussed.
Various findings on the favorite-longshot bias and mispricing by bookmakers are described in the literature. Shin (1991, 1992, 1993) argues that bookmakers skew the odds to hedge themselves against the possible threat of bettors possessing superior information. Accordingly, bookmakers lower the odds of outcomes with low probability. Certain knowledge is only possessed by insiders and thus can be exploited in a way that is harmful to the bookmakers. He states that for these longshot outcomes the risk of losing large sums of money is highest.
Levitt (2004) finds little evidence that there are bettors who are able to beat the bookmakers systematically, even given the distorted prices that bookmakers set. Using a data set of NFL games, he states that bookmakers are just as good or better at predicting the outcomes of games compared to the most skilled gamblers. Bookmakers exploit their advantage by setting prices that make sure profits are higher than they would have been in the case they acted as market makers, by making sure the total dollar amount of the bets is equal on both sides (Levitt, 2004). There is a limit to the degree of distortion bookmakers can use to protect themselves to the few bettors who are as good as the bookmakers in predicting outcomes. According to Levitt (2004), this leads to a situation in which the most talented betting individuals are likely to be employed as odds makers by the bookmakers.
9
Evidence of the presence of a favorite-longshot bias in football in the United Kingdom is mixed. For example, Forrest, Goddard and Simmons (2005) do not find evidence of this bias in their study of nearly 10,000 matches between 1998 and 2003. In contrast to Deschamps and Gergaud (2007), they find a favorite-longshot bias to be present in English football between 2002 and 2006. Furthermore they come up with strategies that significantly improve return on investment with sports betting but still indicate that these strategies remain unprofitable, notwithstanding the existing longshot bias in their data. They also derive a more expanded strategy by using the variance in between odds of different bookmakers. Their idea is that a bookmaker underestimates the probability of an outcome when it differs substantially relative to other bookmakers and therefore could be offering hypothetical profitable opportunities that might be exploited. Their finding is that this strategy leads to higher, but still unprofitable returns.
Interestingly, in the same period Deschamps and Gergaud (2007) find some trends, one of which is a decrease in the margins of bookmakers. The main reason for this is the increased competition due to the entrance and the increased popularity of online betting companies. Another trend is the increase in the return on odds for draws. These trends are important to pay attention to, because the returns were already close to zero. If these trends pursue, profitable strategies may be expected in the years after.
In his study on European football matches, Direr (2011) finds that the expected return on longshot bets tends to be systematically lower than bets on favorites. His dataset comprises the odds of different bookmakers and the outcomes of 79,446 matches in football competitions in the period 2000-2011. He shows that a strategy of selecting odds at and under a threshold of 1.21 delivers profitable returns; if one selected the best odds amongst a group of bookmakers. His results have only exploited information contained in odds, did not rely on complex econometric models and are robust to out of sample tests (Direr, 2011). It’s interesting to find out whether his results hold for recent years and whether there are differences per country. With regard to the differences per country, one could argue twofold. At one hand, in popular competitions that generate a lot of betting traffic one could expect that the betting variance of end results is closer to normal distribution than in less popular competitions. At the other hand taking into account the high number of bets, a betting office is more prone to unexpected results and irrational behavior from their clientage. The main focus of this thesis is to investigate whether the only existing profitable betting strategy explained in academic work (Direr, 2011) to my best knowledge, can stand the test of time.
10
This leads to the following three hypotheses:
Hypothesis 1: The strategy proposed by Direr (2011) for the period 2000-2011 holds for 2011-2014.
Hypothesis 2a: Specific countries have significant positive rates of return. Hypothesis 2b: These significant positive returns hold over both periods.
Other research has shown that bookmakers can increase their profits by distorting odds to benefit from bettors that bet irrational and therefore by sentiment. Bettors are called biased by sentiment if the demand spreads unevenly across the possible outcomes of an event although the odds represent true probabilities (Franck et al., 2013, p. 301). According to some articles, teams that are more popular in terms of fans can cause more sentimental betting (Forrest and Simmons, 2008; Franck et al, 2011). The idea behind this is that these sentimental bettor preferences go along with unexpected volumes of betting. The two main reasons behind sentimental betting can be either the longshot bias or the amount of bets on teams with a large fan base.
Bookmakers try to balance the liabilities by lowering odds on the bets with a relatively high betting volume and by increasing odds on the opposite side (Kuypers, 2000; Levitt, 2004). It is likely that the big-5 football nations in Europe (England, Spain, Germany, Italy and France) have more fans, followers and therefore bettors. Does this affect returns on matches from these big-5 countries in comparison to other, smaller countries?
Franck et al (2013) argue that bookmakers set the odds to reach optimal profits over a long period, rather than on every single bet. They monitor the betting behaviour of their customers and take their future betting behaviour into consideration. Positively biased odds with profit opportunities may be offered at first to increase the customer base. Early losses are to be expected for the bookmakers. In the long term however, bookmakers benefit from a larger customer base. Some of the new clients may contribute to the future profits of a bookmaker. Customers face costs for switching between bookmakers and thus start placing bets that are not optimal. Each customer’s betting history is monitored and discrimination is possible against bettors that are able to make a profit from betting. Their accounts may be closed or maximum bets can be imposed. The customer base that remains may be a very
11
profitable group for the bookmaker. Their findings show that bookmakers optimize over the long term, and are able to segment the market by discriminating against unprofitable and skilled bettors, such as arbitrageurs (Franck et al, 2013).
Lahvicka (2014) has three explanations for the favorite-longshot bias in sports betting markets. His first explanation is risk-loving on the side of bettors. The probability of a successful bet is smaller when betting on a longshot. But in case this happens, the excitement is presumably bigger. Bookmakers exploit this by decreasing odds on longshots. The second explanation is that bettors overestimate the probabilities of winning by longshots, which is again exploited by bookmakers. The third explanation is the existence of information asymmetry. If longshots are underestimated by bookmakers, they risk losing a large sum of money. This can be exploited by either better informed insiders or the general public reacting faster than bookmakers to new information (Lahvicka, 2014). Therefore, lower odds on longshots are reasons of protection, according to Lahvicka (2014).
12
3. Data and methodology
In this section the data and the methodology of this thesis are described. Odds of European football matches from 21 competitions in 11 countries have been studied, being the top four professional football leagues from England and Scotland, the top two leagues from Spain, Germany, Italy and France and the highest national leagues from the Netherlands, Belgium, Portugal, Greece and Turkey. These are the same countries and leagues that Direr (2011) has used in his earlier study.
In his study (Direr, 2011) the period taken is 2000-2011. This dataset includes 78,787 matches. In this thesis, the theory of Direr was used for analyzing data from three additional seasons, focusing on the recent period 2011-2014.
Odds from different bookmakers are available on football-data.co.uk for every season. In some years, for some matches no odds were available for any of the bookmakers included in the dataset. Thus 79,446 football matches lead to 78,787 football matches with at least 3 odds per match (an odd for a home team win, an away team win and a draw to occur) in the period 2000-2011.
In the period 2011-2014 21,642 matches are being studied. In this study, the best odd from the different odds posted by several bookmakers are taken. Very few matches have only one bookmaker to have posted its odds. Frequently, three or more bookmakers have posted their odds. It is possible having accounts with all the bookmakers included in the dataset. No fees are involved for opening an account. Bookmakers regularly offer sign-up bonuses to attract new customers. Some simple rules hold for collecting this bonus fee. These rules are usually in the form of minimal betting amounts that must be placed upon a pre-defined number of matches. When done, an amount of, for example, €50 is received. Unibet has attracted new customers in this way for years. This potential bonus fee will not be included in later profit calculations. The possible collection of this relatively small fee is presumed not to be the driver for a bettor who seeks arbitraging opportunities; the investor type.
Odds are usually available for all the bookmakers 3-5 days before a match take place. The strategy that Direr (2011) describes is to compare odds of different bookmakers for a certain match and then take the best odd amongst them. This strategy is presumed to be the optimal strategy. Money can easily be transferred to an account with a bookmaker, using a money transfer site (MTS). Depositing money by bank transfer can take up to a week. Often there is less time between the moment that an odd is placed by a bookmaker and the moment
13
that the match takes place. Money will not always be in someone’s account with a bookmaker in time when a regular bank transfer is used. Therefore MTSs are being used. There are different money transfer sites, for example Neteller, Moneybookers and Skrill. Transferring money is not free and the costs differ per MTS. Exact costs of transferring and withdrawing money is outside the scope for this study. Direr (2011) shows that by selecting odds amongst a group of bookmakers a higher return can be achieved than sticking to one bookmaker, even when accounting for costs of using money transfer sites.
Having different accounts with different bookmakers and using different MTSs is a timely process and it takes time and effort to organize. This thesis does not account for these costs. The focus is on maximum instant rates of return that can be gained by selecting best odds. In terms of a strategy of maximizing the instant rate of return after transaction costs (transferring and withdrawing costs etc.) a strategy of selecting best odds is the optimal strategy.
Different tables provide the necessary insights. Instant rates of return are reported in absolute and in relative terms (see Table 1 in the Results section). Absolute return is the net amount of euros, based on the strategy of betting the same amount of money on every match. For example, placing bets of €1 consistently on odds at or below a threshold of 1.21 during 2000-2011 gives a profit in absolute terms of €46.55 over 1,119 bets of €1. This is a return in relative terms of 4.16%.
Furthermore an additional linear regression is used on the total sample to indicate the presence of a favorite-longshot bias in both periods (2000-2011 and 2011-2014). For this part, the odds are translated into implied odds. An implied odd is the inverse value of an odd divided by the three inverse values for the three odds. In the example of PSV vs. Dordrecht, the implied odd for a win for PSV is:
(1/1.06) / {(1/1.06)+(1/13)+(1/25)} = 0.8897.
All the odds of the home teams in the whole sample of each period have been translated into implied odds, creating the variable ‘Implied Probability Home Team’. A dummy variable ‘Home Team won’ was created, represented by a 1 or a 0 (1 stands for a home team win, 0 for the home team not winning). The dummy variable ‘Home Team won’ is regressed on ‘Implied Probability Home Team’.
14
4. Results
In this section the results are presented, divided into 2 subsections. In the first subsection an overview of the rates of return when betting on odds below a threshold are presented for several thresholds. Tables for the two periods are presented and tables with t-statistics are provided. The periods range from 2000 to 2011 and from 2011 to 2014. No distinction is made between countries in this subsection. All bets below a certain threshold from all countries are treated similarly and an equal weight is put on every bet.
In the second subsection a distinction is made between rates of return for different countries at a threshold of 1.21. Again, a view at two different periods is given and the results are discussed shortly.
4.1 Odd thresholds at different time periods.
In table 1 rates of return for bets at and below a threshold for the period 2000-2011 are presented. This is a reproduction of the table that Direr (2011) has presented in his seminal article. It was important to check for accuracy, as there are minor differences between his findings over that period and mine. For example, Direr (2011) finds 661 matches and 1,118 respectively for the thresholds of 1.19 and 1.21. I find 662 and 1,119 matches respectively. These differences are of no influence to his or my results. Therefore I state that both our results are accurate, based on the dataset we both used. His strategy is simple and to maximize rates of return by betting on odds below a threshold. The same strategy is used for this thesis. It can be seen from Table 1 that the instant rate of return is maximized at a threshold of 1.19. The instant return (assuming that the best bet was picked and with no transaction and other costs) is 4.45%. Direr (2011) presents reasons, for example reinvesting the proceeds, that make betting on a threshold of 1.21 more optimal. A higher annually compounded rate of return would have been achieved by doing this. In this thesis no attention is devoted to compounding rates of return or reinvestment. It can be seen in table 1 that at a threshold of 1.21 a substantial larger amount of matches is included, compared to a threshold level of 1.19 (1119 matches compared to 662 matches). On average over an 11-year period this means around 2 matches per week that satisfy the condition at a threshold of 1.21. At that threshold, an instant rate of return of 4.16% on average could have been achieved in the period 2000-2011.
15
Table 1. Odd threshold and return (2000-2011).
Odd threshold Number of bets Number of wins Win frequency (%) Instant return (absolute) Instant return (%) 1.13 205 186 90.73 0.38 0.19 1.14 242 221 91.32 3.28 1.36 1.15 410 374 91.22 11.19 2.73 1.16 445 408 91.69 15.62 3.51 1.17 615 563 91.54 26.93 4.38 1.18 660 603 91.36 29.13 4.41 1.19 662 605 91.39 29.49 4.45 1.20 1112 994 89.39 46.29 4.16 1.21 1119 1000 89.37 46.55 4.16 1.22 1418 1244 87.73 45.23 3.19 1.23 1453 1278 87.96 51.90 3.57 1.24 1456 1281 87.98 52.62 3.61 1.25 2107 1792 85.05 40.37 1.92 1.26 2115 1799 85.06 41.19 1.95 1.27 2179 1846 84.72 36.88 1.69
Table 2 presents t-statistics for the return at a threshold of 1.19 and 1.21 for the period 2000-2011. It can be seen that the returns are significant at a 1% significance level. Note that the t-statistic for the return at a threshold of 1.21 is slightly higher than at 1.19.
Table 2. t-statistics for threshold 1.19 and 1.21 (2000-2011).
Threshold Average return (%) (%) SD Number of matches t-statistic
1.19 4.45 32.22 662 3.56
1.21 4.16 36.11 1119 3.85
Table 3 presents results over the period 2011-2014. A rate of return of 0.36% on average could have been achieved (assuming again that the best bet was chosen without extra costs involved) if a strategy of betting at odds of 1.19 or lower was continued. At 1.21 the strategy would have had an average return of 0.69%. If we would stick to the strategy to maximize instant rates of return, the optimal threshold level over this period would have been to bet on all bets with an odd of 1.13 and below. This would have led to an instant rate of return of 3.57% on average.
16
Table 3. Odd threshold and return (2011-2014).
Odd
threshold Number of bets Number of wins frequency (%) Win Instant return (absolute) return (%) Instant
1.13 102 96 94.12 3.64 3.57 1.14 134 123 91.79 2.42 1.81 1.15 170 154 90.59 2.07 1.22 1.16 177 159 89.83 0.87 0.49 1.17 224 197 87.95 -1.67 -0.74 1.18 261 230 88.12 0.27 0.10 1.19 271 239 88.19 0.98 0.36 1.20 375 326 86.93 1.38 0.37 1.21 381 332 87.14 2.64 0.69 1.22 473 402 84.99 -3.96 -0.84 1.23 496 420 84.68 -4.83 -0.97 1.24 502 425 84.66 -4.63 -0.92 1.25 639 536 83.88 -2.88 -0.45 1.26 660 554 83.94 -1.2 -0.18 1.27 684 573 83.77 -1.07 -0.16
Again a table with t-statistics is provided (table 4). It can be seen that at neither threshold a significant profitable return could have been achieved.
Table 4. t-statistics for threshold 1.13, 1.19 and 1.21 (2011-2014).
Threshold Average return (%) (%) SD Number of matches t-statistic
1.13 3.57 26.10 102 1.38
1.19 0.36 36.95 271 0.16
1.21 0.69 38.93 381 0.35
Thus, no significant positive instant rates of return could have been achieved by betting on odds below a threshold by selecting best odds out of a range of bookmakers (disregarding costs involved doing so) if the strategy that would have been profitable between 2000 and 2011 would have been continued. The strategy that was slightly profitable on average over the period 2000-2011 was not significantly profitable over the period 2011-2014. Therefore, hypothesis 1 does not hold.
17
4.2 Rates of return amongst countries
In this subsection the differences of odds at and below a threshold of 1.21 between countries are reported. It is important to note again that odds from 21 competitions from 11 different countries are included. For some countries this means that matches from more than one league are included. In order to not unnecessarily complicate the analysis, no distinction is made between matches of different leagues per country. Almost all matches satisfying the 1.21 threshold level in countries with more than one league in the sample are from the highest leagues. Table 5 presents an overview of the results for the different countries between 2000 and 2011.
Table 5. Average return per country at threshold 1.21 (2000-2011).
Country Number of bets Number of bets won Average bets per season Instant return (%) SD (%) t-statistic England 77 68 7.0 4.07 38.19 0.94 Germany 25 21 2.3 -0.32 44.45 -0.04 Spain 79 74 7.2 8.99 28.72 2.78*** Italy 110 100 10.0 5.10 33.65 1.59 France 5 5 0.5 18.80 2.68 15.67*** Netherlands 152 135 13.8 3.81 37.10 1.27 Belgium 79 70 7.2 4.74 37.87 1.11 Portugal 76 67 6.9 3.99 38.46 0.90 Scotland 212 191 19.3 4.89 35.01 2.04** Greece 218 194 19.8 2.18 36.23 0.89 Turkey 86 75 7.8 2.51 39.60 0.59 Total 1119 1000 102 4.16
Note: ** and *** indicate significance at a 5% and 1% confidence level respectively
It can be seen that average returns from only three countries were significantly positive when someone would have regularly placed bets on matches with a threshold of 1.21. These countries are Spain, France and Scotland. The reason for a high t-value for French matches is that out of the five bets satisfying the threshold all five were successful. France had only a few matches within the eleven years of competition where there was a highly favorite team with an odd of 1.21 or lower. This gives a very low variance and standard deviation accordingly. Remember that a return on a single bet is either negative (-100% when losing) or positive (up to a maximum of 21%). Variances across returns are thus
18
larger when bets with negative returns are involved for a country. Thus, in the case of France, the t-value is not a good indicator.
Therefore, when looking at the odds of matches from a country individually, only betting consequently on Spanish and Scottish matches in the period 2000-2011 would have led to significant positive returns. The Scottish league was eminently known for the big difference between the two top teams (Celtic and Glasgow Rangers) and the rest of the league. Almost all odds satisfying the threshold of 1.21 include matches with those teams involved in Scotland.
In table 6, the results for the period 2011-2014 are presented. Many returns for matches in countries individually are now negative. Only Portugal has a significant positive return, but again, just as France in 2000-2011, there were very few matches involved, only seven. No significant positive returns could have been achieved by betting on matches from individual countries specifically. Possible arbitrage strategies that could have been exploited between 2000 and 2011 by betting on matches from specific countries such as Spain or Scotland were not exploitable between 2011 and 2014.
Table 6. Average return per country at threshold 1.21 (2011-2014).
Country Number of bets Number of bets won Average bets per season return (%) Instant SD (%) t-statistic
England 29 25 9.7 1.76 41.47 0.23 Germany 32 29 10.7 5.26 34.57 0.86 Spain 88 80 29.3 2.26 32.77 0.65 Italy 14 12 4.7 1.93 43.21 0.17 France 10 8 3.3 -4.8 50.20 -0.3 Netherlands 53 44 17.7 -3.6 44.14 -0.59 Belgium 5 4 1.7 -0.06 52.58 -0.26 Portugal 46 42 15.3 6.83 33.51 1.38 Scotland 55 43 18.3 -9.73 48.25 -1.5 Greece 42 38 14.0 5.43 34.79 1.01 Turkey 7 7 2.3 11.57 4.79 6.39*** Total 381 332 127 0.69
Note: *** indicates significance at a 1% confidence level
This leads us to answering hypothesis 2a and 2b. Based on above evidence there’s no country where a significant positive return could have been achieved over both periods. Only when looking individually at countries, bets on Spanish and Scottish matches showed
19
significant positive returns between 2000 and 2011. Thus, hypothesis 2a is true for Spain and Scotland between 2000 and 2011. Hypothesis 2b does not hold
4.3 Regression
In table 7 and 8, one can see that the favorite-longshot bias was present during both periods. Between 2000 and 2011 this bias was larger. This can be seen by the coefficient ‘Implied Probability Home team’. Both coefficients in both regressions are significant.
Table 7. Favorite-longshot bias (2000-2011)
Home team won Coef. Std.Err. t P>t ConfInterval] [95%
Implied Probability Home
team 1.0952 0.0128 85.86 0.0000 [1.0702 1.1202] Constant -0.0342 0.0060 -5.72 0.0000 [-0.0460 -0.0225]
Table 8. Favorite-longshot bias (2011-2014)
Home team won Coef. Std. Err. t P>t [95%
ConfInterval]
Implied Probability Home
team 1.0477 0.0217 48.30 0.0000 [1.0052 1.0902] Constant -0.0201 0.0102 -1.98 0.0480 [-0.0401 0.0002]
20
5. Discussion
It can be seen from the results that betting on odds at 1.21 or lower was not significantly profitable between 2011 and 2014, contrary to 2000-2011. There are possible reasons for this. Bookmakers might have decreased odds that they would previously have quoted higher. The average number of matches per year satisfying the threshold criterion has increased from 102 in the period 2000-2011 to 127 between 2011 and 2014. This could mean that matches that were previously quoted at, for example, 1.16, were now quoted at 1.13. By this, average return over 100 successful bets for a bettor decreases by 3 percentage points. Odds that bookmakers would previously have quoted above 1.21 could now be quoted below 1.21. By lowering odds consequently by a few percentage points, profit for a bettor decreases by a few percentage points, assuming the winning percentage of those bets remains constant.
It is interesting that the rates of return have declined after the article of Direr (2011) was published. This could be an indication that the online bookmakers adjusted their odds in order to mitigate the potential risk of the exploitable favorite-longshot bias by bettors. However, this should be subject to further research.
Another possible explanation of the lower returns between 2011 and 2014 might be due to increased competitive balance in competitions. A reason for this increased competitive balance can be the exorbitant total flow of money from television rights distributed to the teams in, for example, the Premier league. Subsequently, this could potentially lead to similar quoted odds but less successful bets for favorite teams at or below the threshold level. Bookmakers might not have adjusted their odds accordingly. But again, further research to prove this is necessary.
From the regression it could be seen that the favorite-longshot bias was smaller between 2011 and 2014 than between 2000 and 2011. A presence of a favorite-longshot bias is no guarantee for profit to be made. The bias needs to be sufficiently large to compensate for the margin that bookmakers hold to make a profit. In the example of PSV vs. Dordrecht, the odd for a PSV win was 1.06 and the implied odd of a win of PSV was 0.8897 (as explained before). Remember that the implied odds for a match are the inverse values for each outcome divided by the sum of the three inverse values of the outcomes. In order to start making profit on bets of 1.06, teams with these odds must win at least 1/1.06=94.34% of the time. This means that if a team with an implied odd of 0.8897 wins more than 88.97% of the
21
time, a favorite bias is indicated in the regression statistics; however it only starts to become profitable as it starts winning more than 94.34% of the matches.
22
6. Conclusion
This thesis fits in the literature on the favorite-longshot bias and the functioning of efficient markets. This study fits in the literature of betting markets as a special type of financial market and the favorite-longshot bias in particular.
The main finding of this thesis is that the finding by Direr (2011), who showed that profit could be made by consistently betting according to a threshold strategy, did not stand the test of time. Changing solely the years to 2011-2014, this research shows that no profits could be generated in recent years by betting at odds of and below a threshold of 1.21. Even when looking at countries in isolation; no profit could be made. A favorite-longshot bias is still present, in the sense that favorites win more often than predicted by their implied odds. However, this bias could not have been exploited in a profitable manner in recent years.
Different possible explanations have been presented for the disappearance of the profitable favorite-longshot bias that was present in earlier years. However, these need to be investigated further. With the increased popularity of betting exchanges, where on average better odds are presented for the match outcomes, it is highly questionable whether profit opportunities are to be exploited by placing bets with a bookmaker in the future. More research into betting markets and betting exchanges is needed to measure inefficiency or profit opportunities in these markets.
23
Reference list
Deschamps, B. and Gergaud, O. (2007). Efficiency in Betting Markets: Evidence from English football. The Journal of Prediction Markets, 1, pp. 61–73
Direr, A. (2013). Are betting markets efficient? Evidence from European football championships. Applied Economics, 45, 343–56.
Fama, E. F. (1991). Efficient Capital Markets: A review of Theory and Empirical Work,
Journal of finance, 25, 383-417.
Forrest, D., Goddard, J., and Simmons, R.(2005). ‘Odds-setters as forecasters: the case of English football’. International Journal of Forecasting, 21, 551-564.
Franck, C., Nüesch, S. and Verbeek, E. (2011). Sentimental preferences and the organizational regime of betting markets. Southern Economic Journal, 78, 502-18
Franck, C., Nüesch, S. and Verbeek, E. (2013). Inter-market arbitrage in betting, Economica,
80, 300-325.
Jensen, M. C. (1978). Some anomalous evidence regarding market efficiency, Journal of
Financial Economics, 6, 95-101.
Lahvička, J. (2014). What causes the favourite-longshot bias? Further evidence from tennis.
Applied Economics Letters, Vol. 21, No. 2, 90–92
Levitt, S. (2004). How Do Markets Function? An Empirical Analysis of Gambling on the National Football League, Economic Journal, 114, 2043-2066.
Rossi, M. (2011). International Journal of Sport Finance, vol. 6, issue 4, pp. 317-334
Shin, H. A. (1991). Optimal betting odds against insider traders. Economic Journal, 101, 1179–85.
24
Shin, H. A. (1992). Prices of state contingent claims with insider traders, and the favourite– longshot bias. Economic Journal, 102, 426–35.
Shin, H. A. (1993). Measuring the incidence of insider trading in a market for state-contingent claims. Economic Journal, 103, 1141–53.
