1
Title: The influence of European match performance on stock prices of listed football clubs
Date: 11-7-2016
Name: Rick Zwarthoed
School: University of Amsterdam
Student number: 10374523
Supervisor: Drs. P.V. Trietsch
Study programme: Economics and Business
Specialization: Economics and Finance
2 Statement of Originality
This document is written by Student Rick Zwarthoed 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.
Abstract
This study examines the effect of European match performance on the stock prices of publically listed football clubs. This study contributes to existing literature by the fact that it uses observations from the period 2004-2016. Secondly, because it uses a continuous expectancy variable and a continuous match importance variable to measure the effect on abnormal returns, which isn’t done in previous studies. OLS-regressions are used to find an answer to the central question. 545 matches from the Champions League and Europa League (UEFA CUP) are examined. Results show a positive impact for wins on stock prices and a negative impact for draws and losses. It also shows a bigger impact for unexpected wins than expected wins and a bigger impact for unexpected losses than expected losses on stock prices. Level of expectancy is found to be positive and significant only for won matches. As a measure of match importance, stages of the tournament are found to be significant for four out of eight dummy variables. Lastly, an asymmetric relationship is found for unexpected wins and unexpected losses, where wins have a bigger impact on stock prices than losses. Two of the five results of the sub questions are in line with previous findings, two are partly in line with expectations and one is out of line with previous studies. Based on the findings of this study it is hard to draw reliable conclusions on the effect of match performance on the stock prices of publically listed football clubs.
3
Table of contents
1 Introduction ... 4 1.1 Motivation ... 4 1.2 Central Question ... 4 1.3 Literature ... 4 1.4 Sub questions ... 4 1.5 Data ... 5 1.6 Method ... 5 1.7 Structure ... 5 2 Literature review ... 62.1 Measuring abnormal returns... 6
2.2 Effect of match outcome on abnormal returns ... 7
2.2.1 Data used in previous studies ... 7
2.2.2 Effect of match outcome on abnormal returns ... 7
2.3 Explanatory variables ... 8
2.3.1 Type of match variables ... 8
2.3.2 Match venue variable ... 9
2.3.3 Goal difference variable ... 10
2.3.4 Expectancy variables ... 10
2.3.5 Variables for match importance ... 12
2.4 Asymmetric relationship between winning and losing ... 13
2.5 Summary of data, methods and outcomes from previous studies ... 13
3 Methodology ... 16
3.1 Data ... 16
3.2 Measuring abnormal returns... 16
3.3 Measuring the effect outcome of matches on abnormal returns ... 17
3.4 Measuring the effect of a dummy surprise variable on abnormal returns ... 18
3.5 Measuring the effect of level of expectancy on abnormal returns ... 19
3.6 Measuring the effect of the stage of the tournament on abnormal returns ... 19
3.7 Measuring an asymmetric relationship between wins and losses ... 20
4 Results ... 21
4.1 Estimated models for expected returns ... 21
4.2 Calculated abnormal returns ... 21
4.3 Effect of outcome on abnormal returns ... 21
4.4 Effect of dummy surprise variable on abnormal returns ... 22
4.5 Effect of level of expectancy on abnormal returns ... 24
4.6 Effect of stage of the tournament on abnormal returns ... 24
4.7 Asymmetric relationship between wins and losses ... 26
4.8 Comparison of expected relationships and obtained results ... 26
5 Conclusion ... 28
6 Reference list ... 30
4
1
Introduction
1.1
Motivation
In 1983 the first football club went public and since then other clubs have followed. An increase in earnings affects the stock price of a club, but does performance in matches also impact the stock price? This study examines matches from the period 2004 to 2016 and focuses on European matches, because according to studies from Scholtens (2009) and Renneboog & Vanbrabant (2000) the effect of winning or losing a match is bigger for European matches than for matches in the National competition. Besides a dummy expectancy variable, this study uses a continuous expectancy
variable, which isn’t done much in previous studies. Lastly the variable the stage of the tournament is used, which is only done in research from Edmans, Garcia & Norli (2007). They, however, use this variable on a dataset of international competitions for countries and not for European leagues. Therefore this study contributes to existing literature.
1.2
Central Question
Does the performance in European matches have influence on the stock price of publicly traded football clubs?
1.3
Literature
According to Scholtens (2009) losing or winning a match has an impact on the expected cash flows of the team and may affect the market value of the club. He finds a significant positive effect on
abnormal returns after a win and significant negative returns after a loss. Zuber et al. (2005) finds that the outcome of an European match has a bigger effect on abnormal returns than the same outcome of a match played in the National league. They argue that clubs benefit more from European matches through more gate receipts, prize money, advertising, broadcasting and merchandising. Scholtens (2009) finds that unexpected losses lead to bigger negative abnormal returns than expected losses and unexpected wins lead to bigger positive abnormal returns than expected wins. The results of Edmans, Garcia & Norli (2007) show that for losses, the further a game is played within a tournament, the bigger the negative abnormal returns are on the subsequent day. Bernile & Lyandres (2011) find that there is an asymmetric relationship between wins and losses, where losses have a bigger impact on stock prices than wins.
1.4
Sub questions
To provide an answer to the central question, it is divided into multiple sub questions. The first question is whether or not the outcome of a match has a significant effect on abnormal returns. The
5 second question is whether or not unexpected wins lead to higher significant positive abnormal returns than expected wins and whether or not an unexpected loss leads to significant lower negative abnormal returns. The third question is whether or not there is a asymmetric relationship between wins and losses. The fourth question is whether or not the level of expectancy has a significant effect on abnormal returns. The last question is whether or not the stage of the tournament matches are played in, has a significant effect on abnormal returns.
1.5
Data
To examine whether or not the outcome of a match has an effect on the stock price of a publicly traded football club, 545 observations are used from European matches. Observations are drawn from the period 2004-2016 and the clubs that are examined are AFC Ajax, Manchester United, Lazio Roma, AS Roma, Juventus, Benfica and Dortmund. Uefa.com is used to determine the outcome of the observed matches. Expectancy is based on betting quotation from oddsportal.com. Daily returns are calculated based on financial data from finance.yahoo.com and DataStream.
1.6
Method
In this study five regressions are done and 18 hypotheses are tested to provide answers to the sub questions. In the first regression the effect of match outcome (win/draw/loss) on abnormal returns is tested. In the second regression dummy expectancy variables are used and the effect of
(un)expected wins and (un)expected losses on abnormal returns are tested. In the third and fourth regression, a continuous expectancy variable is used and one regression is done on the effect of the level of expectancy of wins and on regression is done on the effect of the level of expectancy of losses on abnormal returns. In the last regression dummy variables for the stage of the tournament a match is won or lost in, are used and the effect of the stage of the tournament on the abnormal returns is tested. All five regressions are explained more extensively in the methodology.
1.7
Structure
This thesis proceeds as follows. In the next chapter a summary is given of the methods and findings of previous studies. Chapter 3 consists of a methodology where the methods of this study are further explained. In chapter 4 the empirical findings are discussed and in the last chapter conclusions are drawn.
6
2
Literature review
In the following chapter a summary of previous studies is given and an expectation for this study is derived. First the measurement of abnormal returns throughout previous studies is discussed. Thereafter the results from previous studies on the effect of match outcome on abnormal returns is discussed. Then different explanatory variables used in previous studies and their effect on abnormal returns are discussed and lastly, the asymmetric relationship between wins and losses is discussed.
2.1
Measuring abnormal returns
In previous studies abnormal returns are calculated using different methods. To get daily returns all studies use the model 𝐷𝑅𝑖𝑡 = ln(
𝑃𝑡
𝑃𝑡−1) or an altered version of this model. Scholtens (2009) and Zuber et al. (2005) both use a market model to estimate the expected returns would the event not have occurred. The market model for team i’s expected return is 𝐸𝑅𝑖𝑡 = 𝛼𝑖+ 𝛽𝑖𝐷𝑅𝑚𝑘𝑡𝑖𝑡+ 𝜀𝑡.
Where 𝐸𝑅𝑖𝑡 is the expected return of team i on date t. 𝛼𝑖 is the alpha and 𝛽𝑖 is the beta of team i.
𝐷𝑅𝑚𝑘𝑡𝑡 is the daily return of the home market of team i on date t. 𝜀𝑡 is the error of the model
Subtracting the expected return from the realized returns gives the abnormal returns.
Renneboog & Vanbrabant (2000) use a version of CAPM: 𝐸(𝑅𝑖𝑡) = 𝐸(𝑅𝑓𝑡) + 𝛽𝑖[𝐸(𝑅𝑚𝑡) −
𝐸(𝑅𝑓𝑡) to estimate expected returns. Subtracting the expected returns from the realized returns
gives the abnormal returns. Benkraiem, Louhichi & Marques (2009) use bootstrapping techniques to calculate anticipated returns and subtract the anticipated returns from the realized returns to get abnormal returns.
For all three ways of calculating expected returns, an estimation period is needed. Scholtens (2009) estimates the parameters of the market model using a estimation period of 250 days. Zuber et al. (2005) use the trading days of the off-season (1 June – 31 July) to estimate the parameters. Renneboog & Vanbrabant (2000) use a 6 month estimation period to estimate the parameters of the CAPM. Benkraiem, Louhichi & Marques (2009) use an control period of 100 days, 20 days before the event (-120d, -20d).
Once abnormal returns are calculated, an event window is chosen, so that the abnormal returns in this event window can be examined. Scholtens (2009), Zuber et al. (2005) and Edmans, Garcia & Norli (2007) use an event window of the first trading day after a match is played. Benkraiem, Louhichi & Marques (2009) use an event window of 2 trading days before a match is played and the 2 first trading days after a match is played. Lastly Renneboog & Vanbrabant (2000) use an event window of the first 5 trading days after a match is played. In addition to that they examine a CAR (cumulative abnormal return) of the first 3 days and a CAR of the first 5 trading days. To test the abnormal returns found in the event window, a regression is done. To test these
7 findings, Scholtens (2009), Zuber et al. (2005) and Renneboog & Vanbrabant (2000) use an OLS-method. Edmans, Garcia & Norli (2007) and Floros (2014) use a Garch(1,1)-OLS-method.
2.2
Effect of match outcome on abnormal returns
After calculating the abnormal returns in the event windows, there has to be examined whether or not match outcomes have a significant effect on the returns of listed clubs stock prices. In the following two subsection data used in previous studies and the effect of match outcome on abnormal returns are discussed.
2.2.1 Data used in previous studies
To test the relationship between performance and abnormal returns, different teams, leagues and different periods are examined. Scholtens (2009) examines 1247 matches of 8 international football teams (Ajax, Dortmund, Lazio, Roma, Juventus, Sporting Lissabon and Porto) in the period 2000-2004. Zuber et al. (2005) examine 1027 matches of 10 football clubs in the Premier League listed on the London Stock Exchange in the period 1997-2000. Renneboog & Vanbrabant (2000) examine 840 matches of 17 football clubs from England and Scotland listed on the LSE or AIM (alternative investment market) in the period 1995-1998. Edmans, Garcia & Norli (2007) examine 1162 matches of 39 different countries in international competitions (World Cup, European Cup, Asian Cup, Copa America) from the period 1973-2004. Floros (2014) examines 179 European matches (Champions League and UEFA Cup) of Ajax, Benfica, Porto and Juventus in the period 2006-2011. Benkraiem, Louhichi & Marques (2009) examine 745 matches from 18 teams from 7 different European countries in the period 2006-2007.
2.2.2 Effect of match outcome on abnormal returns
Scholtens (2009) finds that a win leads to an abnormal return of 0.36%, a draw leads to an abnormal return of -1.10% and a loss leads to an abnormal return of -1.41% on the first trading day after the match is played. All abnormal returns being statistically significant on a 1%-level. He argues that losing or winning a match has an impact on the expected cash flows of the team and may affect the market value of the club. Therefore, wins lead to positive abnormal returns and losses to negative abnormal returns.
Zuber et al. (2005) only find one club where the matches lead to a significant abnormal return on the next day. They argue that their results provide evidence that there is little or no
relationship between game-related information and returns, and that this behaviour is not consistent with that observed in traditional markets around conventional corporate announcements expected to impact on cash flows and financial condition.
Renneboog & Vanbrabant (2000) argue that a loss and a draw reduce a clubs chances to play at a European level or to escape relegation and a win increases these chances and therefore
8 abnormal returns after a loss and draw are expected to be negative and positive after a win. They find that a win leads to an abnormal return of 0.921% that is significant on a 1%-level on the first trading day after the match is played. On the other trading days and the CAR 3 and CAR 5, they don’t find significant abnormal returns after a win. They find that a loss leads to an abnormal return of -0,429% and -0.435% that are significant on a 1%-level on the first and second trading day after a match is played. On the other trading days and the CAR 3 and CAR 5, they don‘t find significant abnormal returns after a loss. After a draw they only find a statistically significant abnormal return on a 1%-level of -0.607% on the first trading day after a match is played.
Benkraiem, Louhichi & Marques (2009) find no significant abnormal return after a win, but do find a significant abnormal return of -1.9% and -0.273% on the first trading day after a match that is respectively lost or drawn. They also find a significant abnormal return of 0.699% one day before a win. They explain this by the idea that supporter investors, who often hold a significant proportion of the listed teams’ capital, consider as a reference that their team will win. In other words, the win constitutes the norm. This is why the market punishes the defeat one day after the match. Conversely, it expects a victory one day before the match and does not reward it the day after.
Conclusions of previous studies differ, but most studies find a significant positive abnormal return after a match that is won and a significant negative abnormal return after a match that is drawn or lost. In this study methods used by Scholtens (2009) are mostly followed and it is expected that matches that are won lead to a significant positive abnormal return on the first trading after the match. And matches that are loss or drawn lead to a significant negative abnormal return on the first trading day after the match.
2.3
Explanatory variables
Different variables can be used to explain the level of abnormal returns after a team has played a match. In this section different variables used in previous studies are discussed. In the first
subsection, ‘type of match’-variables are discussed. In the second subsection, match venue variables are discussed. Thereafter goal difference variables are discussed. Fourthly, expectancy variables are discussed and lastly match importance variables are discussed.
2.3.1 Type of match variables
Most studies make a distinction between different leagues a match is played in. Every match of a club can be played in a different competition (National, Cup, European). Every one of these leagues has a different structure and different financial benefits. Match outcome within these leagues can therefore have different effects on abnormal returns.
Scholtens (2009) makes a distinction between matches played in National leagues and matches played in European leagues. He finds that in the National league, a win leads to an abnormal
9 return of 0.38% on the first trading day after the match is played. A loss leads to an abnormal return of -1.14% and a draw also leads to an abnormal return of -1.14%. All three are significant on a 5- or 1%-level. In the European leagues he finds that a win, loss, and draw lead to abnormal returns of respectively 0.22%, -2.14% and -0.94% on the first trading day after the match is played. Abnormal returns of losses and draws are significant on a 1%-level, and the abnormal return of wins are not significant.
Zuber et al. (2005) uses a dummy variable ‘TYPE’, being 0 if a match is played in the National League and 1 if a match is played in a Cup-tournament (FA Cup, Champions League, UEFA Cup etc.). They find the dummy variable to be significantly positive on a 5%-level. They explain this significance with the expectation that teams benefit from cup games through more gate receipts, prize money, advertising, broadcasting and merchandising.
Renneboog & Vanbrabant (2000) make a distinction between National league matches, European league matches, Cup Competition matches and promotion and relegation matches (games where the teams play for either promotion into a higher league, or relegation into a lower league) and find that abnormal returns on the first trading day after a game don’t differ that much between European league games, National league games and Cup competition matches, but increase
substantially with promotion and relegation games. They explain this by fact that following a promotion or relegation, expected cash flows will change massively.
Previous studies conclude that the type of match does have an effect on the level of abnormal returns. In this study however, only matches from European leagues are examined, so a ‘type of match’-variable isn’t used.
2.3.2 Match venue variable
Some studies examine whether the match is played at home or away. Playing a match at home is seen as an advantage (Zuber at al., 2005) and could therefore have an effect on abnormal returns. Zuber et al. (2005) argues that there is a home field advantage in football and uses a dummy variable ‘FIELD’ to capture this impact. The dummy variable is defined as 0 if the match is played away and 1 if the match is played at home. They don’t find any significance in their results.
Benkraiem, Louhichi & Marques (2009) also use a match venue (at home or away) variable. They divide the results in home wins/defeats/draws and away wins/defeats/draws. They find that losses at home lead to significant negative abnormal returns of -2.3% on the next trading day, while losses away generate significant negative abnormal returns of -1.7%. Also, the increase in stock prices prior matches is greater for wins at home than for wins away. These findings could be
explained by the fact that investors are more confident and take more risk around matches played at home. Thus, investors seem to buy the teams’ shares during the trading day preceding the matches.
10 This leads to an increase in prices before matches. They state that the failure to achieve the expected result (a win) leads to the negative abnormal returns mentioned above.
There aren’t a lot of studies that use a match venue variable, mostly because home advantage is already incorporated in an expectancy variable. The two studies that do distinct between home and away matches, differ in conclusions. In this study a match venue variable isn’t used, because home advantage is incorporated in the expectancy variable.
2.3.3 Goal difference variable
The amount of goals difference by which a game is won or lost could be seen as a performance measure (Zuber et al, 2005) or as an expectation for a club’s result in subsequent matches (Bell et al., 2012) and can therefore have an effect on abnormal returns.
Zuber et al. (2005) uses the variable ‘GOALDIFF’ as a measure of performance and defines it difference in goals scored by team i and opponent j. They expect a positive relationship between goal difference and abnormal returns, but they, however, don’t find the variable to be of any significant relevance.
Bell et al. (2012) argues that the size of the goal difference may alter expectations of the club’s results in subsequent games. To control for pre-game expectations of the goal difference, they create a goal difference surprise variable. This variable is computed relative to the benchmark of the club’s average goal difference in the five previous games. This variable is found to be significantly positive for 3 out of 19 teams observed, and isn’t significantly relevant when all the data is pooled. Few studies take into account a goal difference variable, implying that it is of little
importance. The ones that use the variable, don ‘t find it to be of significant relevance. Therefore, a goal difference variable isn’t used in this study.
2.3.4 Expectancy variables
Whether a result was expected or unexpected could have an effect on the level of the abnormal returns. Instinctively an unexpected win would lead to a higher positive abnormal return than an expected win and an unexpected loss would lead to a lower negative abnormal return than an expected loss. Some previous studies have used an expectancy variable.
Scholtens (2009) uses information from bookmakers and the methods of Palomino, Renneboog & Zhang (2005) to arrive at the expectations about the result. These expectations are used to create dummy variables for expected win, expected loss, unexpected win, unexpected loss, draw: win expected and draw: loss expected. He finds that in the National league an expected win and unexpected win both result in small positive abnormal returns. These returns aren’t statistically different from each other though. An expected loss surprisingly results in lower abnormal returns than unexpected losses. This difference is statistically significant. Lastly, he finds that a draw where a
11 win is expected results in significant negative abnormal returns and a draw where a loss is expected doesn’t result in a significant abnormal return. In the European leagues an expected win doesn’t result in a significant abnormal return, but an unexpected win does result in a significant positive abnormal return. Both expected and unexpected losses result in significant negative abnormal returns. Unexpected losses resulting into significant lower abnormal returns than expected losses and returns after losses in European games being significantly lower than returns after losses in National games. At last, a draw when a win was expected results in a significant negative abnormal return and a draw when a loss was expected doesn’t give a significant abnormal return. Scholtens (2009) argues that the fact that expected results lead to significant abnormal returns, suggests that the stock market does not account for all available information.
Zuber et al. (2005) also uses betting odds to represent expected game outcomes. These odds are converted into subjective winning probabilities (SWP). If for a team the SWP > 0.6, the team is expected to win, if the SWP < 0.4, the team is expected to lose and if the 0.4 < SWP < 0.6, the team is expected to draw. Thereafter they create two dummy variables named EWDL and ELDW. EWDL is defined as equalling one when the observed team is expected to win, but does not. ELDW is defined as equalling one when the observed team is expected to lose, but does not. In their study they don’t find any significant results for the dummy variables EWDL and ELDW.
Bell et al. (2012) states that the efficient market hypothesis requires that a club’s current share price should reflect all the information available to investors, including the expected results from prospective games (assuming this is price sensitive information). When the result of the game is known, the share price should adjust to reflect its unexpected component. Therefore, all the
information that affects the result of a game should be contained in market expectations, which can be proxied by the betting odds. They use the betting odds to calculate the ‘expected league points’ (EP), points that are expected to be earned in a match, and subtract these points from the realized points (3 for a win, 1 for a draw and 0 for a loss) and get the variable ‘surprise league points’ (S). This variable S is multiplied with two match importance variables ‘RV’ and ‘F’ and the variable S*RV*F is found to be positive for all the teams pooled, at a significance level of 1%. Stand-alone the variable S isn’t of significant relevance.
Several studies have used a dummy surprise variable, but only few have used a continuous expectancy variable. The way to determine whether a result is expected or unexpected differs among studies. In this study a dummy expectancy variable and a continuous expectancy variable are used. To determine whether wins and losses are expected or unexpected, the research of Zuber et al. (2005) is partly followed. The only difference is that matches are never expected to result in draws, because betting odds for winning and losing are never equal. So that every win or loss is either expected or unexpected. The only study using a continuous expectancy variable is from Bell et al.
12 (2012), but their research is limited to matches played in the national league, so their methods aren’t followed. In this study the continuous expectancy variable is also based on betting odds, with the realized quotation being the level of expectancy for a win or a loss. For the dummy variables it is expected that unexpected wins lead to higher abnormal returns than expected wins and unexpected losses lead to lower abnormal returns than expected losses. The effect of level of expectancy on abnormal returns is expected to be positive for wins and negative for losses.
2.3.5 Variables for match importance
The level of importance of a match could have an effect on the level of abnormal returns. Instinctively, the more important a match is, the more effect it is expected to have on abnormal returns. Several studies have used different measurements of match importance.
Zuber et al. (2005) argues that the place where teams finish in the English Premier League (EPL) may have important financial impacts. Only the top five teams in the EPL are allowed to
participate in European competitions in the subsequent season, which is a major source of additional broadcast, gate receipts, and merchandising revenues. The bottom three teams in the EPL are relegated to the English First Division. Relegated teams can expect to suffer falling revenues. Taking into account the importance of the place where teams finish, two dummy variables T5 and B3 are created. T5, equalling 1 if the team that played a match was at that time in the top 5, and equalling 0 if it wasn’t. B3, equalling 1 if the team that played a match was at that time in the bottom 3, and equalling 0 if it wasn’t. Surprisingly, both variables are found to be of no significant relevance.
Edmans, Garcia & Norli (2007) use dummy variables to make a distinction between different stages in the tournament. They make a distinction between close qualification games, group games and elimination games. They find for losses, that the more important a game is (the further in the tournament), the more negative the effect of abnormal returns are on the first trading day after the match. For wins they don’t find significant results. They argue that the final stages of a football competition has the strongest mood effect, because such games receive the greatest media coverage and a loss in an elimination game, immediately sends a national team home.
Bell et al. (2012) uses two variable to determine match importance: ‘Rivalry (RV)’ and ‘Final Position (F) ’. RV being defined as the expected difference in two clubs final league positions. So the closer two clubs are expected to finish in the league, the bigger the rivalry is. The variable F is defined as a measurement of how close a match is played to the end of the season, and how it decreases the uncertainty of the final position of the club in the league. Bell et al (2012) argue that the market seeks to predict the future cash flows of the club, and the final league position is an important determinant of the club’s future cash flows. Therefore, the effect of a result on the final league position is important. In the regression, both variables are multiplied with each other and the
13 surprise variable and the variable S*F*RV is found to be significantly positive.
Several studies have used variables to measure the effect of the level of importance on abnormal returns, but the methods differ a lot. The methods used by Edmans, Garcia & Norli (2007) and Bell et al. (2012) lead to significant results. In this study only games in a European tournament are examined. Since the setup of the European tournaments are similar to those of World Cup and Continental cups, the method of Edmans, Garcia & Norli (2007) is followed, and dummy variables for different stages in the tournament are used. It is expected that the effect of wins on the level of abnormal returns is increasing with the stage of the tournament. And it is expected that the effect of losses on the level of abnormal returns is decreasing with the stage of the tournament.
2.4
Asymmetric relationship between wins and losses
In this section the asymmetric relation between wins and losses is discussed. When lost (won) matches have a relatively bigger impact on stock prices than won (lost) matches, there is an
asymmetric relationship. The studies of Scholtens (2009) and Benkraiem, Louhichi & Marques (2009 discuss such a relationship.
Scholtens (2009) finds that the response to defeat on stock prices was bigger than the response to winning. He explains this by the idea that the public is more sensitive to losing than to winning games. He also states that there is both theoretical background and empirical evidence that people respond emotionally stronger after defeats than after victories.
Edmans, Garcia & Norli (2007) find a large negative effect for losses and a smaller positive effect for wins and state that this is consistent with the inherent asymmetry between elimination wins and losses. While a loss leads to instant exit, a win merely advances the team to the next round. Thus, the attention of fans after a win may quickly turn to the next stage of matches. They argue that this may be exacerbated by an allegiance bias in fans’ expectations regarding game outcome. If fans overestimate the probability of a national team win, losses will have a particularly dramatic effect.
Bernile & Lyandres (2011) also find an asymmetric return pattern, where defeats have a bigger impact on stock prices than victories and state that this is consistent with findings of Edmans, Garcia & Norli (2007). They, however, don ‘t give an explanation for this findings.
A possible asymmetric relation between losses and wins is discussed in three studies and these studies all find a asymmetric relationship where losing matches has a bigger impact on stock price than winning matches. Therefore, in this study there is expected to be an asymmetric relationship, where a defeat has a bigger effect on stock prices than a win.
2.5
Summary of data, methods and outcomes from previous studies
In this section outcomes from previous studies, data and methods used by previous studies and an overview of expected relationships are given. This is all summarized in the tables below:
14 Table 1: Important significant abnormal returns and beta coefficients from previous studies
In this table, some abbreviations are used. NL means that a match was played in the National league, EL for European league, CC for Cup Competition. PM and RM means that the match was either a promotion match (PM) or a relegation match (RM). HG means that a game was played at home and AG that a game was played away. EG, GSG and CQG mean that a match was either an Elimination game (EG), Group stage game (GSG) or a Close qualification game (CQG). ***, **, * indicate statistical significance at the 1%, 5% and 10%-level respectively.
Authors Type of match Match venue Expectancy of match Match importance
Scholtens (2009) NL: 𝐴𝑅𝑤𝑖𝑛= 0.38% ** 𝐴𝑅𝑙𝑜𝑠𝑠= -1.14% *** 𝐴𝑅𝑑𝑟𝑎𝑤= -1.14% *** EL: 𝐴𝑅𝑙𝑜𝑠𝑠= -2.14%*** 𝐴𝑅𝑑𝑟𝑎𝑤= -0.94% *** NL: 𝐴𝑅𝑢𝑛𝑒𝑥𝑝𝑤𝑖𝑛= 0.23%** 𝐴𝑅𝑒𝑥𝑝𝑤𝑖𝑛= 0.39% ** 𝐴𝑅𝑢𝑛𝑒𝑥𝑝𝑙𝑜𝑠𝑠= -0.84% *** 𝐴𝑅𝑒𝑥𝑝𝑙𝑜𝑠𝑠= -1.87% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑤𝑖𝑛𝑒𝑥𝑝=-1.40% *** EL: 𝐴𝑅𝑢𝑛𝑒𝑥𝑝𝑤𝑖𝑛= 1.19% ** 𝐴𝑅𝑢𝑛𝑒𝑥𝑝𝑙𝑜𝑠𝑠= -3.07% *** 𝐴𝑅𝑒𝑥𝑝𝑙𝑜𝑠𝑠=-1.34% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑤𝑖𝑛𝑒𝑥𝑝=-1.32% *** Zuber et al. (2005) 𝐴𝑅𝐷𝑢𝑚𝑚𝑦𝐹𝐼𝐸𝐿𝐷=0.86% ** Bell et al. (2012) 𝛽𝑆∗𝐹∗𝑅𝑉= 0.016 *** 𝛽𝑆∗𝐹∗𝑅𝑉= 0.016 *** Renneboog & Vanbrabant (2000) NL: 𝐴𝑅𝑤𝑖𝑛𝑡=1= 0.95% *** 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -1.22% *** 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=2= -0.52% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑡=1= -0.53% *** CC: 𝐴𝑅𝑤𝑖𝑛𝑡=1= 0.84% *** 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -1.97% *** EL:𝐴𝑅𝑤𝑖𝑛𝑡=1= 0.99% ** 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -1.78% *** PM:𝐴𝑅𝑤𝑖𝑛𝑡=1= 3.19% ** 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -3.11% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑡=5= 0.85% *** RM: 𝐴𝑅𝑤𝑖𝑛𝑡=1= 5.77% * 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -6.47% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑡=1= -3.79% * Edmans, Garcia &
Norli (2007) EG: 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -18.2% *** GSG: 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -17.9 *** CQG: 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -11.6% ** Benkraiem, Louhichi & Marques (2009) HG:𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -2.32% *** 𝐴𝑅𝑤𝑖𝑛𝑡=−1= 0.86% *** AG: 𝐴𝑅𝑙𝑜𝑠𝑠𝑡=1= -1.68% *** 𝐴𝑅𝑑𝑟𝑎𝑤𝑡=1= -1.24% *** 𝐴𝑅𝑤𝑖𝑛𝑡=−1= 0.47% * Floros (2014) 𝐴𝑅𝑑𝑟𝑎𝑤𝐴𝑗𝑎𝑥= 0.50% ** 𝐴𝑅𝑑𝑟𝑎𝑤𝐵𝑒𝑛𝑓𝑖𝑐𝑎= 1.36% ** 𝐴𝑅𝑑𝑟𝑎𝑤𝐽𝑢𝑣𝑒𝑛𝑡𝑢𝑠= -0.9% ** 𝐴𝑅𝑙𝑜𝑠𝑠𝐽𝑢𝑣𝑒𝑛𝑡𝑢𝑠= -2.3% **
15 Table 2: a summary of data and methods used by previous studies
Authors Countries observed Period observed Event period Leagues Regression-method # Observations Scholtens (2009) Netherlands Germany Italy Portugal England 2000-2004 t=1 National European OLS N=1274
Zuber et al. (2005) England 1997-2000 t=1 National European
OLS N=1027
Bell et al. (2012) England 2000-2008 t=1 National OLS N=5187
Renneboog & Vanbrabant (2000) England, Scotland 1995-1998 t= [1,5] National European Cup OLS N=840
Edmans, Garcia & Norli (2007)
39 countries 1973-2004 t=1 International (World Cup, European Cup, Copa America, Asian Cup) GARCH(1,1) N=638 Benkraiem, Louhichi & Marques (2009) Denmark Netherlands England Germany Scotland Italy Portugal 2006-2007 t= [-2,2] National Wilcoxon-test N=745 Floros (2014) Netherlands Portugal Italy 2006-2011 t=1 European GARCH(1,1) N=179
Table 3: an overview of expected relationships
Subject Hypotheses
Effect of match outcome on abnormal returns
Won matches lead to positive return on the first trading day after the match. Lost and drawn matches to a negative return
Dummy expectancy variables
Unexpected wins lead to higher abnormal returns than expected wins and unexpected losses lead to lower abnormal returns than expected losses.
Continuous expectancy variables
The effect of level of expectancy on abnormal returns is positive for wins and negative for losses
Match importance The effect of wins on the level of abnormal returns is increasing with the stage of the tournament. The effect of losses on the level of abnormal returns is decreasing with the stage of the tournament
Asymmetric relationships
16
3
Methodology
In this chapter used data is discussed. Secondly, the measurement of abnormal returns is explained. After that, the method of measuring the effect of match outcome on abnormal return is explained. Then the method of measuring the effect of a dummy surprise variable on abnormal return is explained. Thereafter the method of measuring the effect of level of expectancy on abnormal return is explained. Afterwards the method of measuring the effect of the stage of the tournament on abnormal returns is explained. Lastly, the method of measuring an asymmetric relationship between wins and losses is explained.
3.1
Data
In this thesis seven clubs will be examined: AFC Ajax, Manchester United, Lazio Roma, AS Roma, Juventus, Benfica and Dortmund. These clubs are chosen, because they are all still publically listed in the year 2016 and have played in either the Europa league or the Champions League in their
observed periods. Financial data is collected from Datastream and Finance.yahoo.com. Match data from is taken from 2004-2016 from the website Uefa.com. For the data of AFC Ajax, AS Roma and Juventus the period 2004-2016 is used. Data from Manchester United is collected from the period 2012-2016, because stock prices of the club weren’t available before that time. The data of
Dortmund is collected from the period 2008-2016, Lazio from the period 2007-2016 and Benfica also from the period 2007-2016, all three because of unavailability of stock price data before that time. The dataset consists of 253 wins, 115 draws and 177 losses. Quotations on matches are collected from oddsportal.com. This site takes the average of quotation of several bookmakers. This makes the quotations reliable. The stage of the tournament is measured by creating dummy variables for different phases in the tournament.
3.2
Measuring abnormal returns
To measure abnormal returns, an estimation model has to be used. Firstly all the abnormal returns on are calculated with the method used in Zuber et al. (2005) and Scholtens (2009). The parameters for this model are estimated using an estimation period of 3 to 4 years using all the trading days in this period except the trading days where a European match was played on the previous day. This is partly done for simplicity reasons, because small estimation periods lead to unnecessary many estimation models. Estimation periods of 3 to 4 years are also used, so that there are a enough observations to get a significant and reliable estimate model. Only for Ajax an estimation period of 12 years is used, because using smaller estimation periods didn’t lead to significant estimation models. Returns of the market are drawn from equity indices of the home market. A summary is given in table 1 of the appendix.
17 For the use of the estimation model, daily returns of the clubs and markets are needed. Daily returns are calculated using the model 𝐷𝑅𝑖𝑡 = ln(
𝑃𝑡
𝑃𝑡−1). This goes for the daily returns of the clubs as well as the daily returns of the market. The equity index used for AFC Ajax is the AEX, for Manchester United the NYSE is used, for Lazio, AS Roma and Juventus the FTSEMIB is used, for Dortmund the XETRA is used and for Benfica the PSI-20 is used. This leads to the model for the abnormal returns: 𝐴𝑅𝑖𝑡 = 𝑅𝑅𝑖𝑡− (𝛼𝑖∆+ 𝛽𝑖∆𝐷𝑅𝑚𝑘𝑡𝑖𝑡). Where 𝑅𝑅𝑖𝑡 is the realized returns of team i on date t. 𝛼𝑖∆ the
estimated alpha of the model for team i, 𝛽𝑖∆ the estimated beta of the model for team i, and 𝐷𝑅𝑚𝑘𝑡𝑖𝑡
the daily return of the market of team i on date t.
To estimate the models, the methods of Zuber et al. (2005) are followed. The models are estimated with STATA. Once abnormal returns are calculated, an event window is chosen, so that the abnormal returns in this event window can be examined. In this study the research of Scholtens (2009), Zuber et al. (2005) and Edmans, Garcia & Norli (2007) is followed and an event window of the first trading day after a match is played is used.
To test the abnormal returns found in the event window, a regression is done. In this study the method of Scholtens (2009), Zuber et al. (2005), Renneboog & Vanbrabant (2000) and Bell et al. (2012) is followed and a OLS-regression is done.
3.3
Measuring the effect outcome of matches on abnormal returns
To find whether outcomes have significant effects on stock prices methods used by Scholtens (2009) are followed and an OLS-regression is done. The dependent variable is abnormal returns on date t and the dummy variables 𝐷𝑤𝑖𝑛𝑡−1 and 𝐷𝑙𝑜𝑠𝑠𝑡−1.: 𝐴𝑅𝑡 = 𝛼1+ 𝛽1𝐷𝑤𝑖𝑛𝑡−1+ 𝛽2𝐷𝑙𝑜𝑠𝑠𝑡−1+ 𝜀𝑡. α is the constant of the regression, 𝜀 the error-term and𝛽1 and 𝛽2are the beta’s. The regression gives the
value of the beta’s, the constant and the standard deviation. The dummy variable 𝐷𝑑𝑟𝑎𝑤𝑡−1 is omitted in the regression to prevent multicollinearity and is given by the constant. A t-test is done to see whether or not the results are significant.
In the first regression four hypotheses are tested. A t-test in STATA is done to see whether the null-hypotheses can be rejected. The hypotheses are summarized in the table below:
Table 4: Hypotheses tested in the first regression
Hypothesis number: 𝑯𝟎 𝑯𝟏
1 𝛽1= 0 𝛽1> 0
2 𝛽2= 0 𝛽2< 0
18
3.4
Measuring the effect of a dummy surprise variable on abnormal returns
In the second regression another regression is done, but different dummy variables are used. The previous dummy variables win and loss are divided in the dummy variables expected win, expected loss, unexpected win and unexpected loss. To determine whether a result is expected or unexpected, quotations from oddsportal.com are observed. Quotations on oddsportal.com are calculated using the average of several available bookmakers odds. The number of available bookmakers odds to calculate the average on, varies from 3 to 6 in 2004-2005, and 17 in 2015-16. Oddsportal.com gives 3 quotations on every match. One for a win of the home playing team (loss of the away playing team), one for a draw and one for a loss of the home playing team (win of the away playing team). The realized quotation is highlighted. The higher a quotation is, the more unexpected an outcome is. So if a bet is placed on a certain outcome of the match, the amount of money placed on the outcome multiplied by the quotation can be earned by the better.
If the realized quotation is smaller than the quotation of the opposite result, the outcome was expected. And if the realized quotation is bigger than the quotation of the opposite result, the outcome was unexpected. So if for instance the quotation for Ajax winning a game was 2.10 and the quotation for losing that game was 3.50 and Ajax did in fact win, the outcome (win) was expected. Only wins and losses are observed, because draws are never expected by bookmakers. This method gives 41 unexpected wins, 74 unexpected losses, 212 expected wins and 103 expected losses. Table 5: number of expected/unexpected observations
For the second regression the model has to be altered. The dependent variable in the regression is still abnormal returns on date t, but independent variables are now 𝐷𝑤𝑖𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1,
𝐷𝑤𝑖𝑛𝑢𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1,𝐷𝑙𝑜𝑠𝑠𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1 and 𝐷𝑙𝑜𝑠𝑠𝑢𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1. The beta’s are 𝛽3, 𝛽4, 𝛽5 and 𝛽6, and 𝜀𝑡 is the
error term. 𝛼2 is again the omitted dummy variable 𝐷𝑑𝑟𝑎𝑤𝑡−1. The model looks as follows:
𝐴𝑅𝑡 = 𝛼2+ 𝛽3𝐷𝑤𝑖𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1+ 𝛽4𝐷𝑤𝑖𝑛𝑢𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1+ 𝛽5𝐷𝑙𝑜𝑠𝑠𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1+ 𝛽6𝐷𝑙𝑜𝑠𝑠𝑢𝑛𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑𝑡−1+ 𝜀𝑡.
There are two hypotheses tested in this regression. A t-test in STATA is done to see whether the null-hypotheses can be rejected. The hypotheses are summarized in the table below:
# observations Unexpected Expected
Wins 41 212
19 Table 6: Hypotheses tested in the second regression
Hypothesis number: 𝑯𝟎 𝑯𝟏
4 𝛽4= 𝛽3 𝛽4 > 𝛽3
5 𝛽6= 𝛽5 𝛽6< 𝛽5
3.5
Measuring the effect of level of expectancy on abnormal returns
In the third and fourth regression, the level of expectancy on abnormal returns is measured. Every win or loss is given a value based on betting quotation to determine the level of expectancy. So, if a realized quotation on a win was 5.20, then this was more unexpected than if the realized quotation was 2.10. The same idea holds for losses. This gives two separate regressions for wins and losses. The model for wins looks as follows: 𝐴𝑅𝑤𝑖𝑛𝑡= 𝛼3+ 𝛽7𝐿𝑂𝐸𝑡−1+ 𝜀𝑡. Where 𝐿𝑂𝐸𝑡−1 is the level of
expectancy of a won match and 𝛽7 is the beta for this variable. The model for losses looks as follows:
𝐴𝑅𝑙𝑜𝑠𝑠𝑡 = 𝛼4+ 𝛽8𝐿𝑂𝐸𝑡−1+ 𝜀𝑡. Where 𝐿𝑂𝐸𝑡−1 is the level of expectancy of a lost match and 𝛽8 is
the beta for this variable
There are two hypotheses tested in these regressions. A t-test in STATA is done to see whether the null-hypotheses can be rejected. The hypotheses are summarized in the table below Table 7: Hypotheses tested in the third and fourth regression
Hypothesis number: 𝑯𝟎 𝑯𝟏
6 𝛽7= 0 𝛽7 > 0
7 𝛽8= 0 𝛽8 < 0
3.6
Measuring the effect of the stage of the tournament on abnormal returns
In the last regression, the effect of the stage of the tournament on abnormal returns is measured. Methods used by Edmans, Garcia & Norli (2007) are followed, with the adaption that the variable ‘Elimination Games’ is divided in ‘Round of 32 or 16 matches (R032/16M)’ and ‘Quarter-, Semi- or Final matches (QSFM)’, because there are generally more matches played in the knockout-stage in European leagues than in International Cups. Dummy variables are created for all the stages, and for matches won and lost. This gives the following model:𝐴𝑅𝑡 = 𝛼5+ 𝛽9𝐷𝑄𝑀𝑊𝑡−1+ 𝛽10𝐷𝑄𝑀𝐿𝑡−1+ 𝛽11𝐷𝐺𝑆𝑀𝑊𝑡−1+ 𝛽12𝐷𝐺𝑆𝑀𝐿𝑡−1+ 𝛽13𝐷𝑅𝑂32/16𝑀𝑊𝑡−1 + 𝛽14𝐷𝑅𝑂32/16𝑀𝐿𝑡−1+ 𝛽15𝐷𝑄𝑆𝐹𝑀𝑊𝑡−1+ 𝛽16𝐷𝑄𝑆𝐹𝑀𝐿𝑡−1+ 𝜀𝑡
20 Table 8: A summary of matches won and loss in different stages of an European league
#observations Wins Losses
Qualification matches (QM) 47 10
Group stage matches (GSM) 129 91
Round of 32 or 16 matches (RO32/16M) 56 49
Quarter-, Semi- or Final matches (QSFM) 21 27
In this regression, all beta coefficients are tested on significant difference from zero. There are eight hypotheses tested in this regression. A t-test in STATA is done to test if the beta’s of the model are significantly different from zero.
3.7
Measuring an asymmetric relationship between wins and losses
To find out if there is an asymmetric relationship between wins and losses, a t-test is done on the absolute difference between wins and losses in the first regression and between (un)expected wins and (un)expected losses in the second regression. There are three hypotheses tested, which are stated in the table below:
Table 9: Hypotheses tested on asymmetric relationship between wins and losses
Hypothesis number: 𝑯𝟎 𝑯𝟏
16 |𝛽1| = |𝛽2| |𝛽1| < |𝛽2|
17 |𝛽3| = |𝛽5| |𝛽3| < |𝛽5|
21
4
Results
In this chapter results from STATA regressions are discussed and hypotheses are or are not rejected. First the estimated models for expected returns are discussed, secondly a summary of the obtained abnormal returns is given. Thereafter results from the five regressions are discussed and the null-hypotheses are or are not rejected. Lastly, a comparison of expected relationships and obtained results is given.
4.1
Estimated models for expected returns
To estimate the models, a regression is with STATA is done. The daily returns of the market are used as the independent variable and the daily returns of the club are used as the dependent variable. This is done 14 times, on 7 different teams and 14 estimation periods. A summary of the results are given in table 2 of the appendix.
Only beta’s are given, because none of the alpha’s in the regression were significant. All beta’s are significant and are multiplied by the return of the market on date t to get the expected returns of the clubs on date t.
4.2
Calculated abnormal returns
To calculate abnormal returns, expected returns are subtracted from realized returns. A summary and descriptive statistics of the abnormal returns is given in the table below:
Table 10: Summary of abnormal returns
Variable Abnormal returns
#observations 545
Mean -0.0061
Std. Dev. 0.0345
Min -0.2395
Max 0.2350
4.3
Effect of outcome on abnormal returns
Using the data and methods discussed in the methodology, the first regression is done. The summary of the output is given in the table below:
22 Table 11: Output STATA regression 1: effect of wins/losses/draws on abnormal returns
Variable Coefficient Std. Error t-value P > |t|
Win 𝜷𝟏 0.0096 0.0031 3.06 0.002
Loss 𝜷𝟐 -0.0074 0.0033 -2.25 0.025 Draw 𝜶𝟏 -0.0081 0.0020 -3.95 0.000
The results of the first regression summarized in table 11, show that a win leads to an abnormal return of 0.96% and a loss leads to an abnormal return of -0.74% on the first trading day after the match is played. This can be explained by the idea of Scholtens (2009), who argues that losing or winning a match has an impact on the expected cash flows of the team and may affect the market value of the club. The results show that a draw also leads to an abnormal return of -0.81% on the first trading day after the match is played. This can be explained by the idea of Renneboog & Vanbrabant (2000), who argue that draws increase the probability of elimination in an European league. All three variables are significant at a 5%-level. A
The findings of this regression are in line with the expectations drawn in chapter 2. All three null-hypotheses stated in chapter 3.3 are therefore rejected.
4.4
Effect of dummy surprise variable on abnormal returns
Using the data and methods discussed in the methodology, the second regression is done. The summary of the output is given in the table below:
Table 12: Output STATA regression 2: effect of dummy surprise variable on abnormal returns
Variable Coefficient Std. Error t-value P > |t|
Exp. win 𝜷𝟑 0.0060 0.0032 1.88 0.131
Unexp. win 𝜷𝟒 0.0278 0.0066 4.17 0.000
Exp. loss 𝜷𝟓 -0.0021 0.0034 -0.60 0.472 Unexp. loss 𝜷𝟔 -0.0149 0.0052 -2.86 0.002
Draw 𝜶𝟐 -0.0081 0.0020 -3.94 0.009
The results of the second regression summarized in table 12, show that an expected win results in an abnormal return of 0.60% and an expected loss results in an abnormal return of -0.21% on the first trading day after a match. Neither of these variables are found to be significantly
relevant. This can be explained by the theory of Bell et al. (2012) who state that the efficient market hypothesis requires that a club’s current share price should reflect all the information available to
23 investors, including the expected results from prospective games (assuming this is price sensitive information). So if a result of a match was expected, stock prices of the club shouldn’t change and there are no significant abnormal returns on the first trading day after the match to be found.
The results of the regression also show that an unexpected win results in an abnormal return of 2.78% and a unexpected loss results in an abnormal return of -1.46% on the first trading day after a match. Both are significant on a 1%-level. These results fall in line with the theory of Bell. et al (2012), since they state that when the result of the game is known, the share price should adjust to reflect its unexpected component.
The results of the t-tests on the difference between unexpected win and unexpected win and between unexpected loss and expected loss are shown in the tables below:
Table 13: Output STATA: a t-test on the differences between the effect of unexpected win and expected win
# observations Mean Std. Error t-value P (𝜷𝟒> 𝜷𝟑) Unexpected win Expected win 𝜷𝟒 𝜷𝟑 41 212 0.0278 0.0060 0.0066 0.0032 Combined 253 0.0095 0.0029 Difference 0.0217 0.0078 2.77 0.003
Table 14: Output STATA: a t-test on the differences between the effect of unexpected loss and expected loss
# observations Mean Std. Error t-value P (𝜷𝟔< 𝜷𝟓) Unexpected Loss Expected Loss 𝜷𝟔 𝜷𝟓 74 103 -0.0149 -0.0020 0.0052 0.0034 Combined 177 -0.0074 0.0030 Difference -0.0129 0.0060 -2.15 0.016
The results of table 13 and 14 show that 𝛽4 is significantly higher than 𝛽3 on a 1%-level and
that 𝛽6 is significantly lower than 𝛽5 on a 5%-level. This falls in line with the expectation and is
explained by the previous mentioned theory of Bell et. al (2012) about the efficient market hypothesis.
The findings of this regression fall in line with expectation drawn in chapter 2. Therefore, the two null-hypotheses stated in chapter 3.4 are rejected.
24
4.5
Effect of level of expectancy on abnormal returns
Using the data and methods discussed in the methodology, the third and fourth regression are done. The summary of the output is given in the tabled below:
Table 15: Output STATA regression 3: Level of expectancy of wins on abnormal returns
Variables Coefficient Std. Error t-value P > |t|
Level of expectancy 𝜷𝟕 0.0064 0.0020 3.26 0.001
Table 16: Output STATA regression 4: Level of expectancy of losses on abnormal returns
Variables Coefficient Std. Error t-value P > |t|
Level of expectancy 𝜷𝟖 -0.0010 0.0012 -0.81 0.419
The result of the fourth regression, summarized in table 15, shows that every 1-point increase in betting odds (increase in unexpectedness) of a realized win, increases the abnormal returns on the first trading day after the match with 0.64 percent points. This result is significant on a 1%-level. This falls in line with expectations and can be explained by the theory of Palomino,
Renneboog & Zhang (2009), who state that market reactions reflect rational expectations on future firm value, investors would price the expected outcome of games before a game is played.
Therefore, the market reactions to wins or defeats should be stronger, the lower the probability of those outcomes.
The result of the fifth regression, summarized in table 16, no significant relation is found between the unexpectedness of losses and the level of abnormal returns. This doesn’t fall in line with expectations drawn in chapter 2. This may be explained with the theory that markets are very fast in processing good news about game outcomes and slower in incorporating bad news (Palomino, Renneboog & Zhang, 2009). So perhaps the relationship between level of expectancy and abnormal returns isn’t significant on the first day after a match is played, but is significant on some subsequent day.
The findings of the third regression do fall in line with expectations and the findings of the fourth regression don’t fall in line with expectations. Therefore null-hypothesis number 6, stated in chapter 3.5, can be rejected and null-hypothesis number 7 cannot be rejected.
4.6
Effect of stage of the tournament on abnormal returns
Using the data and methods discussed in the methodology, the third and fourth regression are done. The summary of the output is given in the tabled below:
25 Table 15: Output STATA regression 5: The effect of the stage of the tournament on abnormal returns
Variables Coefficient Std. Error t-value P > |t|
Qualification matches won 𝜷𝟗 0.0099 0.0038 2.61 0.009 Qualification matches lost 𝜷𝟏𝟎 -0.0230 0.0111 -2.07 0.039
Group stage matches won 𝜷𝟏𝟏 0.0119 0.0039 3.01 0.003
Group stage matches lost 𝜷𝟏𝟐 -0.0047 0.0039 -1.21 0.225 Round of 32 and 16 matches won 𝜷𝟏𝟑 0.0082 0.0062 1.32 0.187 Round of 32 and 16 matches lost 𝜷𝟏𝟒 -0.0016 0.0042 -0.37 0.714 Quarter-, Semi-, Final matches won 𝜷𝟏𝟓 -0.0017 0.0094 -0.18 0.855 Quarter-, Semi-, Final matches lost 𝜷𝟏𝟔 -0.0215 0.0099 -2.17 0.030
The results of regression 5, summarized in table 17, show that only four of the eight dummy variables result in a significant abnormal return. Firstly, it shows that winning a qualification match leads to an abnormal return of 0.99% and losing a qualification match leads to an abnormal return of -2.3% on the first day after the match is played on a 1%- and 5%-level. The table shows that for group stage matches only a win leads to a significant abnormal return of 1.19% on the first day after the match is played on a 1%-level. For round of 32 and 16 matches no significant returns are found. For quarter-, semi- and final matches only a defeat results in a significant abnormal return of -2.15% on the first day after the match on a 5%-level.
The relatively small positive abnormal returns after a qualification win, falls in line with expectations drawn in chapter 2, the relatively big negative return after a qualification doesn’t. This can be explained by the fact that with a qualification for a European competition, the club can generate additional funds from selling the broadcasting rights of this competition. While the broadcasting rights of the Champions League are centrally marketed by the UEFA, every participant in the UEFA Cup competition markets the broadcasting rights of its home games individually (Stadtmann, 2009). A loss of a qualification would most likely lead to a team not qualifying for an European competition, which means a miss of those additional funds.
A slightly higher positive abnormal return of group match wins than qualification match wins, falls in line with expectations. The insignificant returns after a group match loss however don’t. The insignificant returns after round of 32 and 16 matches won or loss also weren’t expected. Lastly, the insignificant returns of winning a quarter-, semi- or final match weren’t expected too. And the relatively big negative abnormal return after a lost quarter-, semi- or final match do fall in line with expectations. Therefore four of the null-hypotheses stated in chapter 3.6 are rejected and four of the null-hypotheses are not rejected.
26
4.7
Asymmetric relationship between wins and losses
The results of table 11 show that a win has a bigger effect on stock prices than a loss, however this difference isn’t significant. The results of table 12 show that the effect on stock prices is bigger for unexpected wins than for unexpected losses. And that expected wins and expected losses both don’t result in any significant effect. This doesn’t fall in line with expectations. A t-test on the absolute differences between the effect of unexpected win and unexpected loss is shown in the table below: Table 18: Output STATA: a t-test on the absolute differences between the effect of unexpected win and unexpected loss
# observations Mean Std. Error t-value P (𝜷𝟒> 𝜷𝟔) Unexpected win Unexpected loss |𝜷𝟒| |𝜷𝟔| 41 74 0.0278 0.0149 0.0066 0.0052 Combined 115 0.0195 0.0042 Difference 0.0129 0.0084 1.54 0.065
The result of table 18 shows that |𝛽4| is significantly higher than |𝛽6| on a 10%-level. This
can be explained by the findings of Palomino, Renneboog & Zhang (2009), who state that markets are very fast in processing good news about game outcomes (most of the impact of a victory is incorporated in the share prices during the first trading day) and somewhat slower in incorporating bad news (defeats).
None of the findings are in line with expectations from chapter 2, for wins and losses from regression 1 and expected wins and expected losses from regression 2, no significant differences are found. For unexpected wins and unexpected losses from regression 2 an significant opposite results was found to what was expected. Therefore all three hypotheses stated in chapter 3.7 are rejected.
4.8
Comparison of expected relationships and obtained results
In this section a table is shown between the expectations & hypotheses stated in chapter 2 and whether or not the obtained results of this study fall in line with the hypotheses. A summary is given in the table below:
27
Table 19: a comparison of expected relationships and obtained results
Subject Hypotheses Result from study
Effect of match outcome on abnormal returns
Won matches lead to positive return on the first trading day after the match. Lost and drawn matches to a negative return
In line with expectations and hypotheses
Dummy expectancy variables
Unexpected wins lead to higher abnormal returns than expected wins and unexpected losses lead to lower abnormal returns than expected losses.
In line with expectations and hypotheses
Continuous expectancy variables
The effect of level of expectancy on abnormal returns is positive for wins and negative for losses
In line with expectations and hypotheses for wins and differs from the hypothesis for losses
Match importance The effect of wins on the level of abnormal returns is increasing with the stage of the tournament. The effect of losses on the level of abnormal returns is decreasing with the stage of the tournament
Differs from expectations and hypotheses
Asymmetric relationships
A defeat has a bigger impact on stock prices than a win
Differs from expectations and hypotheses
28
5
Conclusion
This study examines the effect of European match performance on the stock prices of publically listed football clubs in the period 2004 - 2016. This is done by observing abnormal returns on the first trading day after a match is played. To provide an answer to the central question, five sub research questions are answered.
The obtained results for the first sub question show that a win lead to a significant positive abnormal return on the first trading day after a match and a draw and a loss to a significant negative abnormal return. Results from this study are in line with results from with findings from previous studies. The theory that supports this findings is from Renneboog & Vanbrabant (2000), who argue that a loss and a draw reduce a clubs chances to play at a European level or to escape relegation and a win increases these chances and therefore abnormal returns after a loss and draw are expected to be negative and positive after a win.
The obtained results for the second sub question show that an unexpected win indeed does lead to a significantly higher abnormal return than an expected win and that an unexpected loss leads to a significantly lower abnormal return than an expected loss. These results are in line with findings from Scholtens (2009) and differ from the findings of Zuber et al. (2005). The theory that supports the findings of this study is the efficient market hypothesis. Prices should contain all available information, so when the result of the game is known, the stock price should adjust to reflect its unexpected component (Bell et al., 2012).
The obtained results for the third sub question show that there only is an asymmetric relationship between unexpected win and unexpected loss, where unexpected wins have a bigger impact on stock price than unexpected losses, which is opposite to findings from previous studies. These findings of this study can be explained by the theory of Palomino, Renneboog & Zhang (2009), who state that markets are very fast in processing good news about game outcomes and slower in incorporating bad news (defeats).
The obtained results for the fourth sub question show that only for wins an significant relation between unexpectedness and abnormal returns is found. The significant effect of level of expectancy on abnormal returns for wins, falls in line with the rational expectation theory. Market reactions to wins or defeats should be stronger, the lower the probability of those outcomes
(Palomino, Renneboog & Zhang, 2009). The effect for losses isn’t significant, which doesn’t fall in line with the rational expectation theory. This may again be explained by the idea that markets are slower in incorporating bad news (Palomino, Renneboog & Zhang, 2009). And that perhaps the relationship between level of expectancy and abnormal returns for losses isn’t significant on the first
29 day after a match is played, but is significant on some subsequent day.
The obtained result for the last sub question show that for wins two out of four stages have a significant effect and for losses also two out of four stages have a significant effect on abnormal returns. This doesn’t fall in line with results from Edmans, Garcia & Norli (2007).
Conclusively, performance in European matches does have an influence on stock prices of publically listed football clubs. Winning a match unexpectedly impacts the stock price positively on the first trading day after a match, where losing unexpectedly or drawing a match impacts the stock price negatively. The impact of winning is bigger than that of losing. The more unexpected a win was, the bigger the impact on stock prices is. This relation however doesn’t hold up for losses. Losing a match in a qualification round or a quarter-, semi- or final has a negative impact on stock prices, and winning a match in the qualification round or group stage has a positive impact on stock prices. Based on the findings of this study it is hard to draw reliable conclusions on the effect of match performance on the stock prices of publically listed football clubs.
The focus of this study was on matches played in European leagues. But there was no distinction made between matches played in the Champions League or the Europa League (UEFA CUP). Secondly, only the results of all the teams pooled has been examined, where it could have been interesting to make a distinction between the observed teams. Other shortcomings of this study, are in the definitions of the surprise variables. The dummy expectancy variables and
continuous expectancy variables are simple and could be defined better and linked more to previous research. As a measure of match importance, stages of the tournament are used as dummy
variables. This leads to multiple dummy variables for both losses and wins. To improve the results of the study, a continuous variable could be used that is better defined and more linked to previous research. The multiple dummy variables also lead to a relatively short number of observations for some stages of the tournament and that could’ve been the reason for insignificant result in that regression.
For further research it is advised to add the findings for each club observed and make a distinction between matches played in the Champions league and Europa league, because the first can be seen as more important than the latter. Secondly it is advised to define the expectancy variable in another way that links more to previous research. Furthermore, it is important find a way to define the match importance variable as a continuous variable. It is also advised to get more observations of matches played in the final stages of the tournament, because these are theoretically the most important.
