The prediction of stock returns with betting odds

The Prediction of Stock Returns with Betting Odds

This paper investigates the effect of sudden changes in the mood and behavior of investors on the stock market, based on psychological explanations and evidence. Football match results are one of the many factors that influence the mood and behavior of investors. The outcome of football matches is not random, but predictable. Bookmakers compute objective probabilities for all possible match outcomes. The quickly processed effect of match outcomes on investors’ mood is not only documented, but also predicted, with the help of these objective probabilities. Evidence is provided for the accurate predictions of bookmakers of a future negative market return when a country is expected to lose a match, whilst also providing correct predictions of negative short term returns on listed football clubs’ stock before domestic matches and positive short term returns before international football matches in the UEFA Champions League or Europa League.

University of Amsterdam Amsterdam Business School MSc Finance Asset Management Joris Hoogenbos 10288481 Ms E. Zhivotova 30-06-2017

2

Statement of Originality

This document is written by Student Joris Hoogenbos who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

3

Table of Contents

Abstract... 1

Statement of Originality ... 2

Table of Contents ... 3

List of Figures ... 5

1 Introduction ... 6

2 Related Literature ... 9

2.1 Efficiency of the Betting Market ... 9

2.2 The Sentiment Effect ... 10

2.3 Betting Biases ... 11

2.3.1 Favorite-Longshot Bias ... 12

2.3.2 Home-Field Advantage Bias ... 13

2.3.3 Draw Bias ... 13

2.3.3.1 Anchoring Bias ... 13

2.3.3.2 Regret Avoidance Bias ... 14

2.3.3.3 Optimism Bias ... 14

2.4 Explanation of the Value of Football Clubs ... 14

2.5 Sharp Bookmakers versus Soft Bookmakers ... 16

2.6 The Payout of Bookmakers to their Clientele ... 17

2.7 Reaction to Release Odds ... 18

3 Methodology ... 19

3.1 Estimating Abnormal Return ... 19

3.1.1 Clubs ... 19

3.1.2 Countries ... 20

3.2 Calculating Objective Probabilities from Betting Odds ... 21

3.3 The Prediction of Matches ... 23

3.4 Unexpected Outcomes ... 24

4

4 Data and Descriptive Statistics ... 26

4.1 Data ... 26

4.2 Summary Statistics ... 27

5. Results ... 30

5.1 The Effect of Matches on the Abnormal Return ... 30

5.2 The Effect of Matches on the Cumulative Abnormal Return ... 32

5.3 Unexpected Match Outcomes ... 35

5.4 The Prediction of Match Outcomes ... 37

5.5 Predicting the Stock Market ... 38

6 Robustness Checks ... 43

6.1 The Cumulative Abnormal Return after Matches ... 44

6.2 The Prediction of Stocks ... 44

6.3 Adding a Control Variable ... 46

6.4 Additional Results ... 49

7 Conclusion ... 50

Reference list ... 52

5

List of Figures

Table I ... 28

Table II ... 31

Table III... 33

Table IV ... 36

Table V ... 39

Table VI ... 43

Table VII ... 45

Table VII ... 47

Table IX ... 49

6

1. Introduction

Which factors influence the sentiment and behavior of people and investors? Investor sentiment is the attitude that investors have towards a particular financial market or security and it is an important determinant of the total effect on asset prices. Garcia (2013) used the fraction of positive and negative words from financial news in the New York Times as a proxy for investor sentiment, finding that stock returns can be predicted more easily using news’ content in recessions. Various psychological researches show evidence for the major factor that sentiment has on people’s lives. The major size that the football world has, together with the win-or-lose factor can cause major behavioral swings among large proportions of people within certain areas, or countries as a whole. These characteristics of the sport of football make this particular sport most compelling to use for this research.

French and Poterba (1991) reported a home bias, meaning that individuals are likely to be marginal investors in their domestic stock market. Edmans, Garcia and Norli (2007) not only found a statistically significant, negative effect as a market reaction to losses by national football teams, they also documented negative reactions on excess returns after incurred losses in other sports e.g. cricket, rugby and basketball games. The reactions to sports other than football had a smaller effect than football losses, but the effect was still statistically significant. Ashton, Gerrard and Hudson (2003) managed to find a significant relation between the general stock market index returns and the performance of the English national football team. Therefore, it seems that the portfolio worth of investors can really be affected by the results of football matches. A more direct effect of football match results is on the share value of listed football clubs. Tottenham Hotspur was the first football club to be listed on the stock exchange market in 1983, with various teams following their example. Tottenham Hotspur, among other teams, chose to be delisted from the stock market in 2011, implying that listing on the stock market is not always the wisest business decision for football clubs. Bernile and Lyandres (2011) analyzed returns of listed European football teams to find that investors are overly optimistic about their local teams’ chances of winning before an important match, only to be disappointed afterwards.

Baker and Wurgler (2007) showed that it is possible to measure investor sentiment. They found sentiment waves with clear, observable and important effects on firms and total market values. Football clubs have an actual meaning to the public and, mostly, their fans. This becomes clear from the given that fans allocate a considerable part of their personal budget and time into their club of choosing.

Investor sentiment and its effect on the stock market have been studied extensively in the economic literature. The result of football matches and other sport events influence investor

7 sentiment and, with that, indirectly affect the stock market. This paper will add to the existing literature by predicting stock returns, making use of betting odds. Betting odds are computed by bookmakers who then assign probabilities to each possible outcome of sport matches before the game is played. Betting odds could eventually be used as a proxy for the final result of sport matches. A profit could possibly be locked in by buying or selling certain stocks, based solely on the release of betting odds and their implied prediction for the match.

Palomino, Renneboog and Zhang (2009) noticed a conflicting strong market reaction to game results and a lack of reaction to the release of betting odds. They stated that this disparity in reaction could arise from the absence of informational content or surprise of betting information. Betting show the probabilities on each possible match outcome. Sauer (1998) found that betting odds are precisely calculated by bookmakers and might be useful as accurate predictions for the outcome of sporting events. The value of the sports-betting industry is estimated to be worth between $700 billion and $1 trillion per annum, with football accounting for an estimated 70 percent of these figures (BBC, 2013). Football is the sport that attracts the most viewers out of all sports and therefore it will be the sport of interest throughout this research.

It should be noticed that not every match has the same impact on the mood and therefore on the stock returns. The importance of a match for that team might be a noteworthy determinant for the magnitude of the effect on the stock market. Klibanoff, Lamont and Wizman (1998)

concluded that country-specific information that does not create a lot of media coverage, is only gradually incorporated into stock prices. Godinho and Cerqueira (2016) developed a measure that account for match importance to find a significant relation between the games’ outcomes and stock performances for most clubs considered in their research. They found that an unexpected match result leads to a higher improvement in stock returns. From an emotional standpoint, a person would probably be happier after a victory that was not expected beforehand than after an anticipated victory. Whether a victory is unexpected, can be tested using the betting odds.

Assuming that investor’s sentiment indeed does have an impact on stock returns, the question arises whether this impact will be significant and predictable, because the result of a sports contest is not random. It can be predicted, using expert predictions and inside information. These informational advantages are incorporated in betting odds, which are computed before each game and often turn out to be accurate predictors of such a match’ result.

Accurate predictions of stock returns could help investors with creating a profitable investment strategy. This study is part of the literature that investigates implications of behavioral biases on asset pricing, which is documented thoroughly in psychological research. The main contribution of this paper is the link between investor’s sentiment and the stock market through

8 betting odds, using the predictability of sport matches’ outcomes to forecast the stock market for clubs and market indices. Betting odds then could be used as predictors for the stock price of listed football clubs and market indices of countries that were involved in those matches. The research question reads: Can betting odds be used as a predictor variable for stock returns?

The effect on abnormal returns is studied through an event study approach. The sample consists of 5,022 club and 1,105 country matches with the corresponding betting odds on all these games.

This paper is organized in the following way. Section I of this paper contains the introduction that describes a priori motivation and links investor sentiment with behavior and stock returns with predictions for football matches. In Section II, a literature background is given. This section shows findings of related research and takes on the ongoing debate in the literature about betting odds and the reaction of the stock market on matches that have been played. Section III shows the methods that are used. In Section IV, the sources of data are provided and a summary of statistics is shown. In the next part, Section V, the results are documented and the economic meaning of these results is discussed. In Section VI, the robustness of the results is examined and additional results are presented. In the last part, Section VII, conclusions are drawn from the literature and the results from this research. There will also be a discussion about the implications of the findings and directions for further research will be provided.

9

2. Related Literature

2.1 Efficiency of the Betting Market

In economic literature there has been an unsolved debate about overall market efficiency and whether anomalies in return are legitimate deviations from market efficiency. Shleifer (2000) and Schwert (2003) hold differing visions on this particular matter. An efficient market’s prices reflect all available information so that acting on information does not result in marginal benefits that exceed marginal costs (Jensen, 1978). Fama (1970) defined an efficient market as one that fully reflects all information that is available up until that point of time. Strangely enough, a profitable price-based betting rule has not been exploited thoroughly thus far, although it could be a

compelling issue for the literature about efficiency in finance. Unexploited profit opportunities are instructive about the degree of efficiency and the rationality of participants among information markets. Many researches have failed to bring clarity about the exact process that determines market efficiency, mainly because the tests that were conducted relied on a single asset-pricing model. Deviations from efficiency are easily written off to the theoretical model being unrealistic or lacking of important information. Although tests of efficiency in betting markets could face the same problem, the presence of this problem is less likely due to the simplicity of betting markets. Bets in the world of sports usually do only have two or three possible outcomes, while being constructed within a short period of time before the event takes place. This short period eliminates possible issues such as the time value of money. Pope and Peel (1989) did develop a linear probability model to prove that the betting market is efficient, concluding that the market is indeed efficient because of the absence of a strategy with a positive return after taxes. However, they were able to diminish the expected losses, which proved to them that odds cannot be seen as rational expectations. On the other hand, Deschamps and Gergaud (2008) concluded that the betting market must be an inefficient market, because of the abnormal return that is generated, using several different betting strategies. Bernile and Lyandres (2011) found an asymmetric reaction of the market to football games, with losses being associated with significant negative abnormal, postgame returns and wins being followed by returns of approximately zero.

All and all, the estimation of probabilities for match outcomes is a simpler process than determining the total stochastic behavior of financial assets and betting markets also have a bigger chance of being an efficient market, due to the immediate feedback that bettors receive (Thaler and Ziemba (1988)). This immediate feedback stimulates a fast learning process of the bettor. The market is said to be weak-form efficient when bettors are not able to earn a better risk-adjusted return or suffer a smaller loss by selecting a class of bets based on the odds. Semi-strong efficiency

10 tests use all publicly available information, like home field advantage, the current form of the teams of interest or the differing injury suffering between the teams and tests whether systematic

abnormal returns could be achieved. The efficiency of the football betting market will be scrutinized in this paper by testing whether betting odds can be used to forecast future short term returns from stocks, rather than from direct bets on the matches.

2.2 The Sentiment Effect

Sentiment has a local and global effect on the stock market. Baker, Wurgler and Yuan (2012) formed the idea that global sentiment and local sentiment might predict the time-series of cross-sectional returns between markets differently: High sentiment causes future returns to be low on relatively difficult to arbitrage and value stocks. The influence that the outcome of sport matches has on the sentiment of people has been documented in research thoroughly. This has led to various examples: College students from Indiana University evaluated their own performance significantly better after watching their college basketball team win a match than after watching a loss (Hirt, Zillmann, Erickson and Kennedy, 1992). Schweitzer, Zillmann, Weaver and Lutrell (1992) researched the probability of the occurrence of a war in Iraq in 1990 and found that students that supported the winning team on the televised American football game estimated the chance of the war happening significantly lower than students that supported the losing team. Arkes, Herren and Isen (1988) found that Ohio State University’s lottery tickets sales increased in days after victories of their football team. An example of the direct effect that sport matches can have on the lives of people is that in the three days after England’s penalty shootout loss against Argentina, heart attacks increased with 25% in England (Carroll, Ebrahim, Tilling, Macleod and Smith, 2002). White (1989) even detailed that elimination from the NFL playoffs caused a surge in homicides in the city where the teams come from. Another example of a negative reaction comes from Trovato’s (1998) article that found a rise in suicide numbers from Canadian inhabitants after suffering an early exit from the Stanley Cup.

These previous examples show that the outcome of sporting events could indeed have a direct impact on people, not only on sport related issues, but on their behavior in other affairs of life as well. Psychological literature makes a clear distinction between the mood after wins and losses. Where homicides, heart attacks and riots are effects after losses, the examples of major changes in behavior are less present following a win of the favorite team. The focus on gains and losses in utility from a certain reference point, rather than looking at the absolute values, comes from the famous prospect theory from Kahneman and Tversky (1979). This theory could explain the asymmetric reaction to wins and losses. Fans are overoptimistic about their chances, creating an upward biased

11 reference point on how the match will play out. A win would be a relatively small positive surprise, whereas a loss would be a relatively bigger negative surprise. Hence, the arising asymmetry in reactions to the outcome of a match persists. Fans could in this case be subject to the optimism or the anchoring bias, both biases will be explained later in this section, although Brown and Hartzell (2001) did not find support that the asymmetric reaction to match outcomes was caused by

investors being overly optimistic. Another possible explanation for the asymmetric response to wins and losses is the nature of the competition. In tournament format, a loss means that the team is eliminated, but a win does not guarantee winning a championship. As for the competition format, the team that ends up being champion always has far more wins than losses, so a loss hurts the chances of becoming champion more than a win harms the team’s chances. Large media coverage before and after a game would probably only enlarge the effects. Lee and Chiu (2016) suggest that not finding a positive reaction after wins can be explained, because Edmans et al. (2007) use only the close price, instead of the close and opening price to estimate excess stock returns, stating that the stock market is efficient, so it incorporates information about the stock market very fast. This asymmetric reaction leads to the first null hypothesis, which states that losses have an equal effect on stocks as wins. The alternative hypothesis states that losses have a bigger impact on stocks. The hypothesis will be split up in two categories: one for country football and one for club football. It could be argued that an unexpected win (loss) would have a greater positive (negative) impact on the stock return than an expected win (loss) does. Bell, Brooks, Matthews and Sutcliffe (2012) examined the impact of point surprises for English football clubs, with the pooled sample showing clear results. The individual clubs did not all show clear results after point surprises. Therefore, the second null hypothesis states that an unexpected result has the same impact on stock as an expected result, with the alternative hypothesis stating that an unexpected result has a greater impact on stock than an expected result.

2.3 Betting Biases

Bonner, Hugon and Walther (2007) evidently show that the reaction of investors to analysts’ forecasting depends on the media coverage around it. Bookmakers however, are not influenced by systematic optimism (Easterwood and Nutt, 1999). Bookmakers also do not have a conflict of interests like brokerage firm’s analysts possibly do (Michaely and Womack, 1999). The important sporting events are usually widely covered by the media, implying that the results are quickly incorporated in the share prices, while information without a lot of media coverage is only gradually incorporated in prices (Klibanoff et al., 1998).

12 whereas Forrest and Simmons (2002) did find such a bias from the bookmakers, which implies that bookmakers see value in offering less unfair bets for wagers, favoring teams that have more fans. Štrumbelj (2016) argues that this bias could arise from using the basic normalization method or the failure to account for the correlation between betting probabilities and difference in attendance. Graham and Stott (2008) found that there are systematic biases in bookmaker odds, which cannot be explained by an odds omitting variable or by extraneous information. Page (2009) found that a big majority of the bets that focus on the total goals scored, are on a total amount that lies above 2.5 goals, which teaches us that most bettors bet on an outcome that is more attractive for them, because they hope for an attractive match with many goals scored and bet on that desired outcome. This finding could mean that bettors have a bias for their favorite or local team as well, because that outcome would be most attractive for the local bettors as well.

The bounded rationality theory shows that individuals are constrained by information limits, causing the decision makers to make satisficing rather than optimizing choices in complex situations. Bounded rationality can explain why bettors start and continue betting and also cause inefficient betting, enabling researchers to predict betting behavior. The existing literature writes about several biases that bettors have, because these could influence the efficiency of the betting market, since placed bets impact the odds that are set by the bookmakers. Shin (1991) showed that bookmakers adjust the odds to collect enough revenue to pay off the winning bettors, regardless of the outcome. The most important examples of betting biases are described below:

2.3.1 Favorite-Longshot Bias

Hausch, Ziemba and Rubinstein (1981) were among the first to prove the existence of favorite-longshot bias in betting on horse races. People usually overweight probability on victory of longshots, teams that have a very low probability of winning, and underweight the probability of victory of the favorite team. This is the favorite-longshot bias, in which a tendency is observed for favorites to be under-bet and longshots to be over-bet, taking the objective probabilities of the occurrence of all events into account. On the other hand, Golec and Tamarkin (1991) and Gray and Gray (1997) found a bias towards the favorites instead of the longshots.

Favorite-longshot bias skews the betting odds, causing an increased return for bettors that bet on the favorites and not the longshots. Risk-seeking gamblers chose to bet on a longshot, even when they are aware of the correct probabilities of the outcome (Weitzman, 1965 & Rosett, 1965). Psychological research shows that bettors tend to prefer low bets with a high potential payout over less risky bets with a lower payoff (Hausch et al., 1981). People also tend to overestimate the likelihood of events with a small probability of occurring.

13

2.3.2 Home-Field Advantage Bias

Forrest and Simmons (2005) found that a coefficient for attendance significantly improves the probability of a victory from the home team. This effect is underestimated by many bettors. The underestimation of the effect of playing on home soil by bettors is known as home-field advantage bias. In the NFL, the largest American Football association of the world, the win percentage of home-field underdogs was 55,6% from 1973 until 1987. An underdog is a competitor thought to have little chance of winning a contest. Substantial profits could have been made on this in the NFL-point spread. Vergin and Sosik (1999) tested the NFL market as well, but could not find a profitable strategy. They did however draw the conclusion that matches with a wide, public focus were profitable when the bets were placed on the home team.

Dixon and Pope (2004) did find evidence on the existence of home-field bias in football as well, although Forrest and Simmons (2002) disagreed. In football, the advantage of playing home (64,5% of the games result in a win for the home team) is big relative to other sport, making it notable that there is so little support for home-field advantage bias. Feddersen, Humphreys and Soebbing (2013) use all-star votes as measure of popularity, concluding that betting odds on popular teams are less favorable, although this could be caused by over-betting. Popular teams have more fans and therefore a higher attendance, increasing their home-field advantage.

2.3.3 Draw Biases

Deschamps and Gergaud (2008) found a higher return for betting on draws than for betting on the home team or away team to win, with the odds for a draw increasing in every single year of their research. They concluded that people underweight the probability of a draw, creating

underpriced odds. Biased human behavior and heuristics predict the opposite effect. The draw bias can possibly be explained by the anchoring bias, the avoidance bias or the optimism bias.

2.3.3.1 Anchoring Bias

People tend to start their estimation of outcomes from a certain starting point, their anchor. Pompian (2006) describes the anchoring bias as people starting to collect information and

interpreting this information from their starting anchor of beliefs. People do not incorporate the information rationally, causing them to deviate too little from their anchor point, causing their estimations to cluster around that anchor point.

A bettor can find it hard to make a distinction in quality between two teams, making the most likely outcome a draw. Even when all outcomes are equally likely, bettor might find it optimal to bet on a draw, since draws usually give a higher payout than a win or loss from the home team.

14

2.3.3.2 Regret Avoidance Bias

Coricelli, Ebrahim, Tilling, Macleod and Smith (2005) found activation in humans brains when a situation of choice was won and relative deactivation when a situation of choice was lost. People can experience actual pain when the outcome of an event is worse than expected, so they want to avoid regretting their decision later. This is called regret aversion. Regret aversion can lead to indecisiveness, so bettors could end up betting on a draw irrationally often, just to avoid regretting their decision (Germeijs and de Boeck, 2002).

2.3.3.3 Optimism Bias

Optimism bias happens because people are overly optimistic regarding their own skills and presumably the skills of their favorite team. A signal that confirms their beliefs grows their

confidence, whereas a signal that differs from their beliefs does not decrease confidence. This is called self-attribution bias (Miller and Ross, 1975). Irrational behavior could be caused by optimism, even when a substantial amount is invested (Babad and Katz, 1991). Pompian (2006) shows that this phenomenon also arises in financial markets, where domestic stocks are overvalued compared to foreign stocks. In the betting world, this could apply to bettors betting on a win of their local team or, when it is a longshot, a draw.

International tournaments usually start with a group stage, after which the final stages will

be played in a knock out system, in which the losing team is eliminated from the tournament. Clubs are able to qualify for international tournaments, while countries can only play in international tournaments. When both teams are equal after regular time has been played, the teams will play an extra thirty minutes and when the teams are still tied after extra time, a penalty shootout will take place. In competition format, this is not the case, so in competitions a draw is a more important possible outcome for the matches. A draw harms the chances of achieving the highest possible for a team playing in a competition format, because the team will only be awarded one point, rather than the three points that a win is worth. Logically, the stocks of teams in competition format would be negatively affected after a draw, leading to the third hypothesis. The null hypothesis states that a draw has no impact on the stock of a team, with the alternative hypothesis stating that a draw has a negative impact on that team’s stock. This hypothesis will only be tested on clubs among domestic competitions.

2.4 Explanation of the Value of Football Clubs

Renneboog and Vanbrabant (2000) found that only 7 out of 20 of the listed football clubs were profitable at the time of their research. A reason for this lack of profitability is the excessive

15 salaries that are paid to star players of the clubs since the passing of the Bosman Verdict by the European Court of Justice in December 1995. The Bosman Verdict stated that regulations of FIFA and UEFA regarding the transferring of football players are not in accordance with Article 48 of the Treaty of Rome, which prohibited players whose contracts terminated to move to a different club without payment to the club that employed the player most recently. The verdict ignited a wave of players that wait until the end of their contract to move to other teams. This impacts the book value of football clubs, since players rightly can be seen as assets ever since, being booked as Immaterial Fixed Assets with transfer money being capitalized if a player is sold for an amount higher than it was bought for. Clubs can opt to offer longer contracts while intending to sell players before the ending of their contracts in order to receive transfer funds. This strategy has not proven to be profitable because of the higher sums that have to be paid up front with this strategy. This paper researches the direct effect of matches on stocks. Because almost all matches are played outside of the transfer window, the time that players can move from one club to another, it is assumed that the buying and selling of players is already incorporated in the stock prices, when the matches are played.

Listed sport clubs differ from commercial firms in the sense that there is an immediate impact on share price performance based on weekly information, rather than quarterly results. A reason for a football club to conduct an IPO is to attract more funds, which could be used to buy star players, invest in an improved youth development system or to expand its stadium. IPOs have not been a direct success in most cases, as 8 out of 12 first clubs on the London Stock Exchange saw a decline in value within the first month (Renneboog & Vanbrabant, 2000). Consequently, in recent years did many of the English and Scottish football clubs, that were listed, decide to delist from the stock exchange, leave the public market and go back into the private market. Tottenham Hotspur, the first football club to be listed on the stock exchange market, was one of the teams that opted to delist. Their chairman Daniel Levy believed that future success for his club could be achieved through increasing the capacity of the club’s stadium by involving considerable additional capital

expenditure. The chairman explained that the AIM listing restricted their capability to generate additional financing that could be used for the future development of the club, which contradicts the original reasoning to opt for listing.

When the outcome of sporting events indeed does have a significant influence on the mood and investing behavior of investors, it can have indirect effect on the stock market of that particular country or region as well, because of funding going into the area of that club. A won match not only improves the position of a team, it also gives the team a better chance to end the season as

16 club, not only because of the receiving of extra proceeds from television rights, but also because of the prospect to a starting bonus from the UEFA Champions League or UEFA Europa League. In the Champions League season of 2016/2017, approximately € 1.319 billion was shared between the participants, per UEFA (2016). Each victory and qualification for a further stage in the competition earns additional funding from the UEFA. Continued good (bad) performances might also result in higher (lower) attendance for matches, higher (lower) catering and merchandise incomes and in the long run more (less) sponsoring income, which lead to improved (deteriorated) stock performance.

The structuring of the shareholders of football clubs usually results in stable performances. Football clubs usually have only a few controlling shareholders, a lot of individual investors and some institutional investors, with the individual investors being supporters of the club that hold the shares as their way of supporting their club or to enjoy other fringe benefits, like ticket discounts or

subscriptions. There is a consistent lack of short-term speculators. These findings result in lower share price fluctuations (Renneboog & Vanbrabant, 2000). This unusual structure, with highly loyal investors, changes the interpretation of the impact on current and future financial results.

2.5 Sharp Bookmakers versus Soft Bookmakers

Bookmakers demand a premium, which lies around 25%, according to Graham and Stott’s (2008) paper. Lower premia (up to 12%) would provide bettors with systematic profits when they would exploit the information from betting odds and chose a selected number of games (Spann & Skiera, 2009). Premia have to be accounted for, when the probabilities of matches’ outcomes are computed. However, Graham and Stott (2008) and Spann and Skiera (2009) used the odds of soft bookmakers in their calculations, whereas this paper uses a sharp bookmaker, which demands lower premia than the bookmakers from the aforementioned papers did. Bookmakers demanding a premium all but rules locking in systematic profits, according to Spann and Skiera (2009).

The biases that were mentioned earlier in this section could all influence the odd that is set eventually. Nonetheless, not all bookmakers have the same strategy to become profitable. There are two types of bookmakers; sharp bookmakers and soft bookmakers. Both types differ in a lot of ways. Sharp bookmakers automatically update odds, whereas odds are manually updated by soft

bookmakers. This can be explained by the strategy of sharp bookmakers, whose focus is fully based on generating high turnover, whilst keeping their margins low. On the other hand do soft

bookmakers profit from recreational players, gamblers and accumulators. An accumulator is a combination of multiple matches and is mostly used to boost the potential payout of the bet, of which an example is given earlier in this paper. Sharp bookmakers also do not put any limits on the betting account of bettors, whereas soft bookmakers limit bettors to certain amounts that can be

17 bet and prevent professional bettors from entering their website. The manual updating of the soft bookmakers gives bettors arbitrage opportunities, since not all bookmakers immediately change their odds and could not be harmed by bettors exploiting the opportunity, since it does not exist solely on their website. In this paper, odds computed as accurate as possible are desired, so the odds of sharp bookmakers are preferred. Sharp bookmakers have higher payout ratios, because of their low margins and immediate updating. It must be noted that sharp bookmakers are still profitable, implying that the betting biases might still be present, but their effect is minimized. Pinnacle, the bookmaker that is used throughout this research, is a sharp bookmaker that averages a payout ratio of 96,1% in the dataset that is used, which is considerably higher than soft bookmakers’ payout ratios. The lowest payout ratio within this sample denotes 89,6% and the highest payout ratio denotes 99,9%. The high payout ratio of Pinnacle proves that this company sets the most accurate odds, making them best applicable for this research. The accuracy of the betting odds will be tested before these odds can be used to predict stock returns. Accurate odds do assign the most realistic probabilities to the possible outcomes. The implied match outcome effect on stock returns could possibly be predicted with the forecasts of the bookmakers, so the fourth null hypothesis states that betting odds cannot predict short term stock returns, with the alternative hypothesis stating that betting odds can predict short term stock returns.

2.6 The Payout of Bookmakers to their Clientele

In Europe, apart from the UK, odds are usually displayed in numerical form. An example of how the payout of bookmakers to bettors will be given next: In April 2017, a heavily favored Ajax played a home game against an out of form sc Heerenveen in the Eredivisie, the highest Dutch domestic competition. The closing odd for a victory of Ajax was 1.19, with a draw and a loss having an odd of 8.18 and 16.00, respectively. When a bettor would have bet € 100 on an Ajax win, the bookmaker would have returned 100*1.19 = € 119, since Ajax indeed did win this match with a score of 5-1. The bettor would have made a profit of € 19. Betting € 100 on a draw or a sc Heerenveen win would have paid out a lot more, namely € 818 or € 1600, respectively. The high probability of Ajax defeating sc Heerenveen gives the bettor a lower return. Risky bettors might have opted for a bet on a sc Heerenveen win, but in this case, the bettor would have lost all his money, giving a net return of minus € 100,-. Risky bettors could be seen as the risky investors of the betting world.

The bettor could have combined this bet with a bet on the Europa League match between Olympique Lyonnais and Besiktas, which was won by Olympique Lyonnais with a score of 2-1. The odds for this match were 1.43 for the home team winning, 5.03 for a draw and 8.01 for Besiktas coming up with the win. When the bettor would have bet on the combination of both Ajax and

18 Lyonnais winning, his € 100 would have turned into an accumulation of the odds: 100 * 1.19 * 1.43 = € 170,17. However, if the bettor would have bet on a combination of the two matches, but at least one of the outcomes was not predicted correctly, the bettor would have lost all his money. So, when the bettor would have bet on both Ajax and Besiktas winning, a net loss of € 100 would have been posted , despite predicting one match correctly. The existence of accumulating bets gives

bookmakers increased chances of being profitable.

2.7 Reaction to Release Odds

Palomino, Zhang and Renneboog (2005) compare market reaction to betting odds and match outcomes, finding that these types of information differ in three crucial ways. Firstly, betting odds are a type of information that has a very short life. The match is already played a few days after the release of odds, after which the odds contain no valuable information anymore. Secondly, odds contain opinions about the outcome of a match, whereas the result of a match clearly shows all information. The last main difference lies in the degree of salience. Match results are mentioned on many television shows, websites, radio and daily newspapers, whereas betting odds are exclusively available in bookmakers’ offices and websites. The release of odds has no impact whatsoever on investor’s sentiment, but the match results do have an impact. A possible explanation is that the predictions of investors are similar to the professionally computed odds, or at least have roughly equal expectations about which team is favored to win. Barberis, Shleifer and Vishny (1998) showed an under-reaction to the release of betting odds, when good news can be used to predict positive returns in the near future. Daniel, Hirshleifer and Subrahmanyam (1998) explain this lack of reaction to odds by investors being overconfident in private information. Priors are overweighed and

therefore beliefs are not updated according to the Bayesian rule. It is assumed throughout this research that stocks do not react to the release of betting odds.

19

3. Methodology

3.1 Estimating Abnormal Return

Daily total returns and daily market indices are used. When the total return indices were not available, the price indices are used. There was no significant difference between the returns based on return indices and price indices for countries that provided both measures. An event study will be performed on the abnormal returns.

3.1.1 Clubs

Football clubs’ shares are not traded in large flows, whereas market indices are. There is a so-called non-synchronicity between a football club’s stock return and the market return. For this reason, the return of football clubs considered differs from the return of the countries that were involved in a match. For the football clubs, three leads and three lags of market returns are added to the market model, as was done in Dimson’s paper (1979), resulting in the following formula:

𝑅(𝑐𝑙𝑢𝑏)_𝑖𝑡 = 𝛼_𝑖∑ 𝛽_𝑖𝑇𝑅_{𝑚,𝑡+𝑇}+ µ_𝑖𝑡 +3

𝑇−3 (1)

where RmT is the return on the MSCI World Index on day τ, serving as the market index. 𝑅(𝑐𝑙𝑢𝑏)𝑖𝑡 is the return of a football club, βiT and αi are estimated using OLS to construct ai and biτ, in order to

determine the abnormal return of club i on day t (ARit):

𝐴𝑅𝑖𝑡 = 𝑅(𝑐𝑙𝑢𝑏)𝑖𝑡− 𝑎𝑖− 𝑅𝑚,𝑡 ∑ 𝑏𝑖𝑇 +3

𝜏=−3 (2)

𝑅(𝑐𝑜𝑢𝑛𝑡𝑟𝑦)𝑖𝑡 = 𝛾0𝑖+ 𝛾1𝑖𝑅(𝑐𝑜𝑢𝑛𝑡𝑟𝑦)𝑖𝑡−1+ 𝛾2𝑖𝑅𝑚𝑡−1+ 𝛾3𝑖𝑅𝑚𝑡+ 𝛾4𝑖𝑅𝑚𝑡+1+ 𝛾5𝑖𝐷𝑡+ 𝜀𝑖𝑡 (7)

Football games do not take place every weekend of the year, so event clustering has to be

accounted for. A way to account for event clustering, is to test abnormal return’s significance with the Wilcoxon signed-rank test. This test has no distribution and is robust to event clustering.

From the abnormal returns, cumulative abnormal returns can be constructed, by adding multiple abnormal returns. The computation of the cumulative abnormal return is executed the following way:

20 where CAR(t,T) denotes the cumulative abnormal return from day t until day T. Day t is the first day of the event and day T the last day of the event. To test the effect of various match outcomes the cumulative abnormal return is regressed with OLS, with the following regression:

𝐶𝐴𝑅(𝑡, 𝑇) = 𝛾0+ 𝛾1𝑊𝑖𝑡+ 𝛾2𝐿𝑖𝑡+ 𝛾3𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + µ𝑖𝑡 (4)

where Controls consists of dummy variables for every year in which matches are played and dummy variables for every team that has played, as well as dummies for different stock exchanges on which the clubs are listed. For clubs, Home is a dummy for home matches, which equals one if the team plays home and zero otherwise. Also for clubs, International is a dummy for international matches, which equals one when the match was played in the Champions League or Europa League and zero if it was played in domestic competitions.

3.1.2 Countries

To compute the return for countries, the world market index is included, but with only one lag and lead, since some local markets might be lagging or leading the world market index. The following model is estimated:

𝑅(𝑐𝑜𝑢𝑛𝑡𝑟𝑦)𝑖𝑡 = 𝛾0𝑖+ 𝛾1𝑖𝑅(𝑐𝑜𝑢𝑛𝑡𝑟𝑦)𝑖𝑡−1+ 𝛾2𝑖𝑅𝑚𝑡−1+ 𝛾3𝑖𝑅𝑚𝑡+ 𝛾4𝑖𝑅𝑚𝑡+1+ 𝛾5𝑖𝐷𝑡+ 𝜀𝑖𝑡 (5)

where R(country)it is the continuously compounded daily local currency return on the MSCI World

Index of country i on day t, with the inclusion of R(country)it-1 being justified to account for

first-order serial correlation. Rmt the continuously compounded daily world market index on day t that is

included to control for potential correlation between local indices. Dt are dummy variables for

Mondays through Thursdays, with Fridays being left out to avoid the dummy variable trap. The residual from equation 5 is εit, which estimates the result of the country’s football matches using the

following model:

ɛ_𝑖𝑡 = 𝛿₀+ 𝛿₁𝑊_𝑖𝑡+ 𝛿₂𝐿_𝑖𝑡+ 𝑣_𝑖𝑡 (6)

where Wit = (W1it,W2it,W3it) is a dummy variable for a victory within a certain subgroup; group stage

matches, qualification matches and knock out matches. Lit = (L1it,L2it,L3it) is a dummy variable for

losses within the same subgroups. This model is estimated mutually and also includes country dummies, which interact with the explanatory variables within the model. One of the dummies is always excluded to avoid multicollinearity and the dummy variable trap. From this formula, the cumulative abnormal return is constructed in the same way as in equation 3. Equation 4 is also used,

21 but there are other control variables in the case of countries, except for the team variables, which are still used. The control variables are KO, Tournament, Group and Qualification, which are dummies for matches in the knockout stage, in a tournament, in the group of a tournament and in the qualifications, respectively. The dummies equal one if played in the stage of their dummy and zero if played in a different stage. One of the dummies is always excluded to avoid multicollinearity and the dummy variable trap. For the control variables, it would be expected that the more

important the match, the stronger the effect on the stocks. Also for the control variables, one would expect that more difficult matches would have a greater positive effect and a less negative effect, since the expectations would be adjusted to the level of the opponent.

3.2 Calculating Objective Probabilities from Betting Odds

Probabilities from odds set by bookmakers are usually measures of short-run outcome uncertainty. Štrumbelj (2016) analyzed the methods that are most used to derive probability forecasts from betting odds. This research shows that basic normalization produces biased probabilities, with biases that are approximately two thirds larger than those derived by Shin probabilities, concluding that Shin probabilities should be the preferred method for the deriving of probabilities from betting odds. However, Štrumbelj (2016) uses soft bookmakers exclusively in this paper, whereas Pinnacle as a sharp bookmaker has a higher payout ratio that calculates its odds in a different way, minimizing the described problems.

Odds have the power to predict match outcomes and can be used to construct probabilities for the different outcomes of the match that is to be played, namely a probability for a win, a draw or a loss of the team that is scrutinized at that time. The same method as used by Bell et al. (2012) is adopted in order to convert odds into probabilities regarding match outcome. Let oio be the odd for

a match outcome for team i. In this way oio denotes a proxy for the normalized belief of the

bookmaker about this match. With this method, the objective normalized probabilities PWi, PDi, PLi

are estimated. The formulas of the win, draw and loss probabilities are given below:

𝑃𝑊_𝑖=

1 𝑜

1 𝑜

+ 1 𝑜

+ 1 𝑜

1 𝑜

1 𝑜

+ 1 𝑜

+ 1 𝑜

1 𝑜

1 𝑜

+ 1 𝑜

+ 1 𝑜

22 where oiW, oiD, oiL represent the outcome of a win, draw and loss respectively.

The profit of the bookmaker is determined by the margin that it sets, which is determined by the denominators in the first three formulas. The bookmakers intend to minimize their risk and always generate a profit, whereas the bettors always have a negative expected return, unless an arbitrage opportunity arises. Arbitrage opportunities never occur within the same bookmaker, but only between multiple bookmakers. A single bookmaker is never harmed when an arbitrage

opportunity arises, even when the bookmaker is one of the bookmakers where such an opportunity is existent, because their own model is still based on minimizing risk and generating profit. The arbitrage opportunity is virtually non-existent for that single bookmaker, because the odds of other bookmakers are irrelevant for the achieving of their own goals. From equations 7, 8 and 9, the tenth equation is estimated:

𝑃𝐷𝐼𝐹𝐹_𝑖= 𝑃𝑊_𝑖− 𝑃𝐿𝑖 (10)

where PDIFFi is the difference between the win probability and the loss probability and therefore

measures the uncertainty of the match, because every match has other precisely calculated outcome probabilities. The higher the absolute value of this measure, the more likely a certain outcome is. A positive value implies that the chance on a victory is higher than the chance on a loss. The more PDIFFi drifts up from zero, the more likely a victory will be for the team of interest. When PDIFFi approaches zero, the match outcome will become more and more uncertain, with a value of zero meaning that the bookmakers cannot make a distinction between the two teams. In this case, odds for a victory and loss should roughly be equal and ideally lie at 3.00, but in reality would lie slightly below 3.00, so the bookmaker is still able to lock in a profit. The probabilities are still objective, because the margin is equal for all possible outcomes, thus its effect can be omitted, making the normalized probabilities still equal to one third for each outcome.

Palomino et al. (2009) went on to create dummy variables based on a similar variable. The dummy variables describe the possibilities of a strong expectancy to win, a weak expectancy to win, a strong expectancy to lose and a strong expectancy to win, with the values for all dummy variables chosen arbitrarily. Adopting this way of constructing these variables, allows for measuring the difference in effect of expected and unexpected wins and losses. An unexpected win or loss should have a greater impact than an expected win or loss do, as suggested by rational asset pricing, because expected results should already be incorporated in stock prices, when the betting market is efficient.

In this paper, similar variables are constructed as in the paper of Palomino et al. (2009), although the numbers differ slightly. The four dummy variables are composed, using bookmakers’ expectations by applying equations 1 through 4. Each dummy variable has two specifications:

23 specification (a) is based on PDIFFi and specification (b) on PWi and PLi. Note that PWi , PLi and PDIFFi

not only contain information about the probability to win or lose, but the probability to draw as well, so the dummy variables contain all available information. The following criteria for the dummies are chosen:

- SF (strongly favored to win): SF(a) equals one, when PDIFFi > 0.35 and zero when otherwise.

SF(b) equals one, when PWI > 0.60 ánd PLI < 0.20 and zero otherwise.

- WF (weakly favored to win): WF(a) equals one, when PDIFFi lies in a domain of {0;0.35}.

WF(a) is zero otherwise. WF(b) equals one, if PWi > PLi ánd PWi lies in a domain of

{0.35;0.60}. WF(b) is zero otherwise. Conclusively, in this case a win is more to be expected than a loss.

- WL (weakly expected to lose): WL(a) equals one, when PDIFFi lies in a domain of {-.35;0} and

zero otherwise. WL(b) equals one, when PLi lies in a domain of {0.35;0.60} ánd PWi < PLi. WL(b) is zero otherwise. Conclusively, in this case a loss is more to be expected than a win.

- SL (strongly expected to lose): SL(a) equals one, when PDIFFi < -0.35 and zero when

otherwise. SL(b) equals one, when PWi < 0.20 ánd PLi > 0.60 and zero otherwise.

3.3 The Prediction of Matches

Various regressions are run to analyze whether the betting odds from this sample are accurate predictors of the outcome of matches, most importantly to make a distinction between the probabilities of wins and losses. The following regressions are run, using the Probit model:

𝑊_𝑖𝑡 = 𝛾₀+ 𝛾₁𝑃𝐷𝐼𝐹𝐹_𝑖+ 𝛾₂𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝜐_𝑖𝑡 (11)

𝑊_𝑖𝑡 = 𝛾₀+ 𝛾₁𝑆𝐹(𝑎) + 𝛾₂𝑆𝐿(𝑎) + 𝛾₃𝑊𝐹(𝑎) + 𝜐_𝑖𝑡 (12)

𝑊_𝑖𝑡 = 𝛾₀+ 𝛾₁𝑆𝐹(𝑏) + 𝛾₂𝑆𝐿(𝑏) + 𝛾₃𝑊𝐹(𝑏) + 𝜐_𝑖𝑡 (13)

where Wit = (W1it,W2it,W3it) is a dummy variable for a victory within a certain subgroup; group stage

matches, qualification matches and knock out matches for countries and home matches, away matches, domestic matches, European matches and combinations of these matches for clubs. Lit = (L1it,L2it,L3it) is a dummy variable for losses within the same subgroups. The Controls are the same as

mentioned in the CAR equation above. SF(a), SL(a), WF(a), SF(b), SL(b) and WF(b) are dummies based on the criteria described in the “Calculating objective probabilities from betting odds” paragraph. These dummies equal one when the match odds met their exact criteria and were zero otherwise. The same regressions are run with Lit as dependent variable, instead of Wit, only with Lit as

24 dependent variable, WL(a) and WL(B) replace WF(a) and WF(b).Draws are not predicted, since draws have less influence on the stock prices and are more difficult to predict. The Probit model is used, because the dependent variables can only take up two values; one or zero.

3.4 Unexpected outcomes

To test whether unexpected outcomes have a greater effect on stocks than other matches, a t-test on matches with an unexpected outcome will be compared by the t-values of the computed CAR(1,2). Adding a condition to equation 4, will determine the effect of an unexpected outcome. This gives the following equations for wins and losses. Draws are excluded, since an unexpected win or loss is always more unexpected than a draw. To elaborate, when a team is heavily favored to win a match, the odd for a loss will be higher than the odd for a draw. The same goes for a heavy underdog. The equations include a condition, which ensures that the outcome was not expected ex ante:

𝐶𝐴𝑅(𝑡, 𝑇|𝑃𝐷𝐼𝐹𝐹_𝑖 < 0) = 𝛾₀+ 𝛾₁𝑊_𝑖𝑡+ 𝛾₂𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝑢_𝑖𝑡 (14)

𝐶𝐴𝑅(𝑡, 𝑇|𝑃𝐷𝐼𝐹𝐹_𝑖> 0) = 𝛾₀+ 𝛾₁𝐿_𝑖𝑡+ 𝛾₂𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝑢_𝑖𝑡 (15)

where Wit = (W1it,W2it,W3it) is a dummy variable for a victory within a certain subgroup; Controls

include team dummies, year dummies, stock exchange dummies and dummies for European

matches, domestic matches, home and away matches and combinations of the aforementioned. The regression for unexpected losses is computed the same way. Lit = (L1it,L2it,L3it) is a dummy variable for

losses within the same subgroups as for the wins. Controls are the same as in the formula for unexpected wins. Subsequently, the coefficients are compared to each other with the following z-test: 𝑧 =(x̅𝑢𝑛−x̅𝑛𝑜) − (µ𝑢𝑛− µ𝑛𝑜) √𝜎2𝑢𝑛⁄𝑛𝑢𝑛+ 𝜎 2 𝑛𝑜 _𝑛 𝑛𝑜 ⁄ (16)

where x̅𝑢𝑛 and x̅𝑛𝑜 are the means of the unexpected outcome and the “normal” outcomes, which depicts the whole sample. Variances are denoted with σ2 and µ𝑢𝑛 and µ𝑛𝑜 depict the unknown means. The test is designed that when µ𝑢𝑛− µ𝑛𝑜 has a value of zero, there is no significant difference between the tested values.

25

3.5 Predicting Stock Returns with Betting Odds

The assumption that the stock market does not react to the release of betting odds is made, based on researches of Barberis et al. (1998), Daniel et al. (1998) and Palomino et al. (2009). To test whether the betting odds have predictive power on stock returns, the following regressions are run:

𝐶𝐴𝑅(𝑡, 𝑇) = 𝛾₀+ 𝛾₁𝑆𝐹(𝑎) + 𝛾₂𝑆𝐿(𝑎) + 𝛾₃𝑊𝐹(𝑎) + 𝛾₄𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝑢_𝑖𝑡 (17)

𝐶𝐴𝑅(𝑡, 𝑇) = 𝛾0+ 𝛾1𝑆𝐹(𝑏) + 𝛾2𝑆𝐿(𝑏) + 𝛾3𝑊𝐹(𝑏) + 𝛾4𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝑢𝑖𝑡 (18)

where the independent variables and the control variables are again dummy variables for each team, stock exchange and year and dummies for Tournament, Group matches, European matches and Home matches. Significant coefficients for the independent variables would mean that stock returns can be predicted with betting odds for certain matches. The betting odds would then have to meet the criteria of the significant predictors. To test for multicollinearity, a variance inflation factor (VIF) test is done.

26

4.

Data and Descriptive Statistics

4.1 Data

Returns from stock indices are obtained from Datastream. Returns are preferably computed making use of total return indices. If return indices are not available, the price indices are used instead. All returns are measured in the local currency. The dates of the return vary from three days before until three days after the matches are played. Three World cups and three Euro cups are investigated for the “country” part of the paper, including qualification matches. The qualifications for World Cup 2018 have not been completed at the moment of writing, so the final tournament cannot be used throughout this paper. After exclusion of matches without closing odds, this totals a number of 1,105 matches from August 2006 up until March 2017. Countries from whom the market index could not be retrieved throughout the whole sample of interest, are excluded. Exceptions are non-European countries, because tournaments without European teams are not part of this research.

The same rules are applicable for the stock returns of the listed football clubs of interest, but these are not all listed on the same stock exchanges. Some clubs are listed on multiple stock

exchanges. Ajax, Borussia Dortmund, Juventus and Porto are all listed on multiple stock exchanges, but they have in common that they are all listed on the Stuttgart Stock Exchange, so this exchange is used. All Turkish clubs (Galatasaray, Fenerbahce and Besiktas) are only listed on Borsa Istanbul Exchange. Arsenal F.C. is a special case. Although Arsenal F.C. is publicly held, it is not listed on a stock exchange, with a single share being priced very highly. Arsenal’s stock is not influenced by weekly results and is infrequently traded, because it is owned by a parent company, Arsenal Holdings plc. For these reasons, Arsenal F.C. is excluded from the dataset. Olympique Lyonnais is also excluded from the dataset, because their betting odds could not be delivered by the provider of the dataset. Renneboog and Vanbrabant (2000) found that victories are more rewarded in increases of share price for listed clubs on the London Stock Exchange (LSE), compared to clubs listed on the Alternative Investment Market (AIM), while defeats of AIM listed clubs leads to larger price reductions. Previous researches, like the paper op Cain, Law and Peel (2000) and Renneboog and Vanbrabant (2000) were solely based on the UK football market. The wave of delisting UK based football clubs causes that an assumption needs to be made. Given the different stock exchanges that football clubs are listed on, the assumption that the effect does not differ between stock exchanges has to be made. The number of stock exchanges is still minimized.

For the listed clubs, all results in their domestic competition and in the European

tournaments are considered. The data start from the 2007-2008 season and end in April of the 2016-2017 season. Exceptions on this rule are the Scottish Premiership, the Italian Serie A and the

27 European tournament, UEFA’s Champions League and Europa League. The data from the former start in the 2008-2009 season, the Serie A starting from the 2010-2011 season and the European cups both starting from the 2011-2012 season. The data total a number of 5,022 matches, 859 European matches and 4,163 domestic matches, after exclusion of matches without closing odds.

All betting odds are retrieved from www.indatabet.com. This website uses the odds from Pinnacle, a sharp bookmaker. Their odds are computed very precisely, with average payout ratios of 96,1% for opening odds and closing odds. Odds are available for all matches of all competitions, totaling full-time results of 6,127 matches between August 2006 and April 2017. Both opening and closing odds are mentioned in the data. Closing odds should contain more information, regarding injuries and other determining factors for the outcome of the match, and regarding the predictions that bettors make for these matches. This learning process however does not show in an increased payout ratio, which is the same for both opening and closing odds, but this could be due to good modelling by the bookmaker as well. The match results are from Pinnacle as well. The results and odds are double checked with results odds from Oddsportal, a website that specializes in current and historical betting odds. The results are excluded when contradicting and the odds are excluded when the normalized probabilities differ by more than ten percent.

4.2 Summary Statistics

Table I shows information about the total number of games and the results that are included in this research. The mean daily log stock returns and their standard deviations are given the first trading day after matches. It totals 6,127 matches, divided over 1,107 country matches and 5,022 club matches. For country football, only losses are followed by a negative stock response, 2.8 basis points, whereas draws and wins are followed by a positive stock response of 17 basis points and 7.5 basis points, respectively. After matches on tournaments, the return is negative with 11 basis points, mainly due to a negative mean daily log stock return of approximately 34 basis points, following a victory in the knockout stages. The standard deviation of wins on tournaments is higher than after draws and losses, with 1.48%. The strongest stock response after country matches come following a knockout stage loss and a group stage draw, with -34.6 and 34.3 basis points, respectively. Knockout losses also have the highest standard deviation, with 2.11%.

For club football, the absolute average mean daily stock returns after match days are higher than for country football. The same applies for the standard deviations. Losses are followed by a stronger negative return, than the positive return after wins. Losses average a negative stock return on the following day in every category. The average mean daily log stock return after a loss is -1.31%, with a standard deviation of 3.86%. The strongest stock response is after a home loss in a European competition, which is followed by a negative stock return of 2.39%, varying from a

28

Table I

Number of Wins, Draws and Losses in Country and Club Football and Percent Mean Daily Return on the First Trading Day after Matches

The table reports the number of wins, draws and losses in country football and club football and the percent mean daily return on the first working day after the match has been played. The standard deviations are also in percentages. For country football, it shows the number of wins, draws and losses per stage from August 2006 until March 2017 in the World Cups and Euro Cups, covering the tournaments from Euro 2008 up until the qualifications for World Cup 2018. For club football, it shows the number of wins, draws and losses from August 2007 until April 2017 in the domestic competitions of Netherlands, Germany, England, Portugal and Turkey. The data from the Scottish competition are from August 2008, the data from the Italian competition from August 2010, with the Champions League and Europa League starting in August 2011 and all the data from the aforementioned competitions end in April 2017. The mean returns reported in the tables are log daily returns the next working day after the match. The returns are retrieved from Datastream. Qualification matches are matches in a competition to qualify for the final tournament of the Euro or World Cup. Tournament matches are played in the Euro or World Cup, divided between Group- and Knock out matches, with Knock out matches beginning after the group stage, where a loss means elimination. European competition matches are played in the Champions League or Europa League.

return of -23.33% to 5.47%. Draws average a negative stock return on the following day in every category as well, but the effect is weaker than after losses. This negative response in club football is expected, because a draw earns the team only one point, where a win earns the team three points. Therefore, the result in points of a draw is closer to that of a loss than that of a win. The strongest negative stock return comes after a home draw in European competitions, with an average of -1.05%, with a standard deviation of 9,25%. The average mean daily log stock return after a draw is

All matches Country 1105 -0.028 1.409 Club 5,022 -0.122 5.710

Wins Draws Losses

N Mean SD N Mean SD N Mean SD

Panel A: Country matches (40 countries) Stage Qualification 412 -0.142 1.316 193 -0.128 1.393 193 -0.018 1.243 Tournament 146 -0.110 1.858 47 -0.343 1.314 110 -0.045 1.542 Group 97 -0.006 1.756 47 -0.343 1.314 74 -0.117 1.107 Knock out 49 -0.343 2.045 0 -0 0 40 -0.346 2.110 Total 558 -0.075 1.480 240 -0.170 1.378 307 -0.028 1.360

Panel B: Club matches (15 clubs)

Home 1699 -0.260 3.629 492 -0.456 14.124 321 -1.533 3.929 Away 1218 -0.715 4.002 610 -0.626 3.154 682 -1.211 3.827 European competition 380 -0.088 3.393 218 -0.690 6.779 261 -1.649 5.008 Home 245 -0.262 2.746 99 -1.051 9.247 87 -2.389 4.259 Away 135 -0.722 4.265 119 -0.390 3.646 174 -1.279 5.316 Domestic competition 2537 -0.504 3.850 884 -0.515 10.320 742 -1.196 3.363 Home 1454 -0.348 3.751 393 -0.306 15.113 234 -1.214 3.759 Away 1083 -0.714 3.970 491 -0.683 3.024 508 -1.187 3.167 Total 2917 -0.450 3.795 1102 -0.550 9.720 1003 -1.314 3.861

29 -55 basis points, with a standard deviation of 9,72%. Wins average a positive stock return on the following day in every category, with a home victory in European competitions as exception. The average mean daily log stock return after a win is 45 basis points, with a standard deviation of 3.80%. The strongest positive stock return comes after an away victory in European competitions, with a mean daily log stock return of roughly 72 basis points and a standard deviation of 4,27%.

30

5. Results

The effect that football match outcomes have on stocks and the predictability of these matches and stocks, is researched in this paper.

5.1 The Effect of Matches on the Abnormal Return

The abnormal returns are computed and tested first. The returns are taken from the first trading day after the game, and when matches are played during a trading day, the next trading day is used. Although Lee and Chiu argued that investors would not make irrational trading decisions after a depressing loss, Edmans et al. (2007) concluded that there is an asymmetric reaction on market indices after wins and losses in country football, with especially a lacking of a positive response after wins. In the abnormal returns as shown in Table II, a similar effect is found. There are no significant stock responses in any category after wins, whereas losses have an overall highly significant, negative impact on the abnormal return of market indices. Losses have a negative effect on stocks, with a significance level of 1%. A loss is followed by a decrease in return of approximately 29 basis points on the market index the following day. This can possibly be explained by supporters of teams being overconfident by their chances before a match, making a loss a negative surprise, whereas a victory would have been more expected and therefore psychologically a smaller surprise. One might argue that this overconfidence is possibly already incorporated in the stock prices. The biggest effect for countries was found in qualification matches, where a loss is followed by a

decrease of approximately 42 basis points. Possibly, this is caused by a higher amount of low quality teams, from whom fans build up their hopes, only to be disappointed after another lost match. Another possible explanation comes from countries that usually have a team with a lot of quality, but lose. Fans from this team will be more disappointed with a loss in the qualification stages, which is normally a stage of the process that is easily overcome. A third possible explanation is that qualification matches have more statistical power in this case, because other papers report a significant negative effect in the other subcategories of losses as well. A tournament loss also has a highly significant negative impact on the abnormal return of market indices, a decrease of

approximately 2 basis points.

For clubs, the effects of match results are stronger, as documented in Table II. This makes sense, because match results are of a greater importance for football clubs’ worth than for countries as a whole. Most shareholders are supporters of a club, whereas market indices have a more diverse set of investors. For clubs, match outcomes in all subcategories for wins and losses have a significant impact on stocks. Wins have a positive effect, whereas losses have a negative effect. All results are highly significant at the 1% level. Away wins have a stronger positive effect than home