1 Student number: S2010224
Name: Dyon Bijl
Study Program: Msc Finance Supervisor: V. Angelini
The effect of outcomes of international soccer matches on national
market index returns
Abstract
This paper investigates whether the mood of investors has an effect on financial decision making, measured by national stock index returns. A change in mood is reflected by match outcomes of international soccer matches played by national teams. Results indicate that there is no systematic stock return reaction after wins and losses. However, small capitalization stocks seem more sensitive for soccer results compared to large capitalization stocks and TOTMK indexes, showing an abnormal return of -10 basis points on the first trading day following losses. Furthermore, controlling for predicted probabilities does not alter the results.
1. Introduction
An assumption often made in finance is that investors are rational (Ackert and Deaves, 2009). This means that investment decisions should not be influenced by factors which are not related to making the investment. However, there are several studies which find that a shift in the mood of investors leads to changes in their investment behavior. Factors which are found to have an effect on investors’ decisions are weather conditions such as daylight, sunshine, temperature and lunar cycles. Additionally, outcomes of sporting events are also found to have an effect on investors’ decisions1. These findings contrast the theory of rationality among investors, which could have implications for financial models which assume rationality, such as the Capital Asset Pricing Model (CAPM)2. Such anomalies in the market provide problems for the application of these models. Therefore research on behavioral biases may help understanding human behavior resulting in more suitable
1 These studies will be discussed in the literature review
2 models. Evidence of behavioral biases among investors provides support for the relatively new development of behavioral finance, which contrasts the rational decision-making approach of financial economics.
This paper will focus on the effect of sporting events, specifically, the outcomes of international soccer3 matches played by national teams, on the behavior of investors. These match results will be used as a variable to have an impact on mood, which in turn is theorized to influence investors’ decision making. The relationship hypothesized here is therefore an indirect one. The outcomes of soccer matches are theorized to have an effect on mood, after which mood impacts the decisions investors make. However, the relationship between sport outcomes and mood is already established in the psychology literature, which will be discussed in the literature review below. Therefore, the objective of this paper is to examine whether a relationship between mood and investors’ decision making exists. In doing this, this paper replicates the study of Edmans et al. (2007) for a more recent sample period. Whereas Edmans et al. (2007) investigates the time period between January 1973 to December 2004 this paper utilizes a time period from June 2004 to December 2013. These data are more recent and therefore provide an update of the image regarding investor sentiment. Interestingly, the recent financial crisis falls within this sample period and may provide insights into the reactions of investors during such a period. In contrast to Edmans et al. (2007) this paper also investigates the effect of friendly match outcomes. Although Edmans et al.( 2007) did not consider these matches to be relevant, this paper shows that friendly matches do trigger significant effects, however not in the hypothesized direction. Additionally, this paper applies a different method in calculating the effect of expected outcomes, determined by bookmakers, on stock price reactions.
This paper will contain the same hypotheses as Edmans et al. (2007) concerning the effect of soccer matches. As this paper will only analyse the effects of wins and losses, two main hypotheses arise. The hypothesis concerning wins is that they trigger positive abnormal returns the next trading day, which will be measured by log returns of national stock indexes. The hypothesis for losses is that they trigger negative abnormal returns the next
3
3 trading day. These hypotheses will be tested with both an event study and several regression models. Secondly, it is hypothesized that matches which are more important to the public than other matches will generate a larger effect. The importance of matches is reflected by their type, which are friendly, qualification, group or elimination matches, increasing with hypothesized impact on national index returns. Thirdly, this paper controls for the ex-ante predicted probabilities of the outcomes of matches. These probabilities are derived from odds provided by bookmakers. Hypothesized here is that an unexpected result has a larger impact on national stock returns compared to a rather expected outcome. Together with the main hypotheses this comes down to the hypotheses that unexpected wins trigger a larger positive abnormal return than expected wins and unexpected losses trigger a larger negative abnormal return than expected losses. This paper will continue with a literature review, after which the methodology will be discussed, followed by a section on data and finally, the results be discussed and a conclusion will be presented.
2. Literature review
4 papers is that there is a significant positive relationship between the amount of morning sunshine and stock returns. This finding supports the hypothesis that mood has an effect on stock prices where a ‘good’ mood leads to positive stock price reactions. Thirdly, the effect of temperature is examined by Cao and Wei (2005). They find larger returns when the temperature is low. They explain their finding by stating that investors become more aggressive at low temperatures and therefore are more risk-seeking causing larger returns. Fourthly, Yuan et al. (2006) investigates the effect of lunar cycles. They find international evidence that stock returns are significantly lower on days surrounding a full moon than on days surrounding a new moon. They state that lunar cycles affect mood which in turn leads to stock price reactions. Besides weather related variables also the effect of sporting outcomes on mood, and in turn stock prices, is investigated. These studies will be discussed next.
5 similar argument can be made for the effect of losses, where losses in an important match are likely to have a stronger effect on mood compared to unimportant matches.
Edmans et al. (2007) expands the research of sporting outcomes on stock returns by using different sports and different methodologies. Edmans et al. (2007) utilizes both an event study and a continuous variable approach to measure the effect of wins and losses on national indexes. They use a sample of over 1100 soccer matches with an additional 1500 matches in cricket, rugby, ice-hockey and basketball. Furthermore, the time period investigated ranges from January 1973 to December 2004. The results utilizing soccer matches will be discussed first. Their event study uses an estimation period consisting of all trading days within the time period which are not trading days associated with soccer matches. The average daily log return and daily standard deviation for the estimation period are 5.8 and 144.9 basis points, respectively. The average return on trading days following wins is 5.0 basis points which is not significantly different from their estimation period returns. In contrast, the mean return following losses is -18.4 basis points, which is significantly lower than 5.8 basis points, as was measured in their benchmark period. This result implies an asymmetrical effect because losses seem to have a larger impact on stock returns than wins. The negative effect of losses is also shown to be larger after world cup matches and elimination stage matches showing that the outcomes of more important matches have a stronger impact on stock prices. Similar results were found for the other sports. However, because sports matches are often played in weekends this means that the following trading day will be a Monday rather often. This means that there may be a day-of-the-week effect, influencing the results of the event study4. To control for this day-of-the-week effect, and several other market factors, they apply an econometric approach in which they estimate two models. Their first model regresses index returns on these market factors, after which the residuals are used as an input into their second model which tries to explain part of the residuals by the outcomes of sporting events. Their results are similar as in their event study, in which abnormal returns following wins were 1.6 basis points whereas the abnormal returns after losses was found to be -21.2 basis points. Confirming their previous results, they show that wins are not significantly different, in contrast to losses which are
6 significantly negative. Their results also show an increase of the effect when importance of matches increases, represented by abnormal returns of -13.1, -16.8 and -38.4 basis points after losses in close qualifying games, group matches and elimination games, respectively. These abnormal returns are all found to be significant at any level of alpha. Finally, they investigate whether the expected outcomes, calculated by converting Elo ratings, influence the size of stock price reaction. They hypothesize that a rather unexpected outcome will generate a larger abnormal return in the direction which was already stated by their main hypotheses. The intuition they provide for this hypothesis is that although investors are irrational, as was found earlier in their paper, they price in the expected match outcomes into their valuations of stocks, as predicted by market efficiency. Therefore, an unexpected result should trigger larger stock price reactions. However, they do not find such a relationship to be present.
As stated earlier, the relationship investigated is an indirect one because the variable mood intervenes the relationship between soccer outcomes and investors’ decisions. This paper examines the relationship between mood and investors’ decisions and therefore assumes an effect of soccer outcomes on mood. This assumption is supported by literature from the psychological field. For example, Hirt et al. (1992) find that the mood of fans and their self-esteem is influenced by outcomes of their favoured teams. Furthermore, fans estimated their own performance to be better in the future after a win of their team, and worse after a loss. Additionally, Wann et al. (1994) finds the similar result that fans experience an increased and decreased self-esteem after wins and losses respectively. Schwarz et al.(1987), Schweitzer et al. (1992) and Arkes et al. (1988) find similar results regarding optimism and increased self-esteem after wins, and pessimism and decreased self-esteem after losses. Because these studies provide evidence for a relationship between sporting outcomes and mood, results of this paper that are in line with the hypotheses will be assigned to mood.
7 Basking in Reflected Glory (BIRG) and Cutting of Reflected Failure (CORF) are examined by Cialdini et al. (1976) and Snyder et al. (1986), respectively. BIRG refers to identifying yourself with others who are successful, for example a sports team. On the other hand, CORF is dissociating yourself from those who are unsuccessful. These tactics of people may lead to a stronger mood effect after a win than after a loss, implying an asymmetrical outcome in which losses may not trigger a negative effect whereas wins do trigger a positive effect. Because the theory and empirical results contradict each other there will be no hypothesis concerning the direction of an asymmetrical effect.
8 3. Methodology
This paper applies two main methodologies to investigate the relationship between the outcomes of international soccer matches and national stock market returns. The paper will contain several event studies and regressions. The hypotheses to be tested are
1 Wins of national teams trigger a positive abnormal return on the next trading day following the match.
2 Losses of national teams trigger a negative abnormal return on the next trading following the match.
3 When the importance of matches increase, the hypothesized abnormal returns of wins and losses become larger (more positive for wins, more negative for losses). 4 unexpected wins trigger a larger positive abnormal return than expected wins and
unexpected losses trigger a larger negative abnormal return than expected losses.
The event study will follow the approach described by Brown and Warner (1980, 1985), who describe several methods for calculating the abnormal returns. As a robustness check, two of these methods will be utilized. The first method applied is the mean adjusted returns, which calculates the abnormal return by deducting the expected return from the actual return:
∈ R K, (1)
where ∈ is the abnormal return for stock i at day t, R is the actual return for stock i at day t and K is the expected return of stock i calculated as an average over a predetermined estimation period.
The second method is the market and risk adjusted returns, which calculates the abnormal return by:
∈ ∗ , (2)
9 The last model assumes that the CAPM model generates the expected returns (Brown and Warner (1980).
To calculate daily returns for stocks the following formula is applied:
, (3)
Where is the continuously compounded return of stock i on day t, LN is the natural logarithm and is the price of stock i at day t.
The events in this event study are the results of international soccer matches played by national teams on World Cups, European Cups, the Copa America, qualification matches for these cups and friendly matches. Specifically, only wins and losses are included because these outcomes are proven to have an effect on mood, whereas draws have not, as shown by Hirt et al. (1992). The selection procedure of teams and matches will be discussed in section 4. The stock returns which will be analysed are broad market returns of the relevant countries. So for example, when the Netherlands lose from Spain, the hypothesized effect of a loss will be tested on a local market index of the Netherlands. Additionally, the effect of the win of Spain will be tested on a local market index of Spain. Information on which indexes are included and the sampling period can be found in section 4.
10 take place before the result is established and therefore cannot be affected by mood swings caused by that particular outcome. Furthermore, it is not included into the estimation period as well because when the match is played before the close of the trading day it will contain the effects of the match into the estimation period. Furthermore, an event period longer than one day may suffer from event clustering. Event clustering is primarily present in cup matches because there could be only one or two days in-between matches. Therefore, a larger event period may capture the mood effects of two matches which biases the results (Brown and Warner (1980, 1985). Furthermore, studies in the psychological field which investigate the relationship between sporting outcomes and mood measure the relationship immediately following the sport outcomes. Because of this it is unknown how long the effect on mood will last. Therefore, only a small event period is utilized to prevent such errors. An event period of one day is also in line with Edmans et al. (2007) and Ashton et al. (2003).
Complementary, an econometric approach will be applied as a check on the results of the event study. Additionally, tests on the role of unexpectedness will be performed. These additional tests consist of regressions which control for ex-ante predicted probabilities by bookmakers which will be discussed below. The regression models applied in this paper follow Edmans et al. (2007), in which two models are utilized. The first model controls for several market factors after which model two regresses the residuals of model one and tries to explain part of the residuals by including dummies for wins and losses. Model 1 is specified as follows:
! " # $ % & ' , (4)
11 The only difference with Edmans et al. (2007) in model one is that this paper does not include the market factor which controls for non-weekend holidays. The factor is not included because of a lack of data availability. Furthermore, its effect on the results is likely to be minor because a calendar year only contains a few non-weekend holidays. Model 2 regresses the residuals from model 1 and is specified as follows:
( )*+, )-+. )/+0 )1+2 3*, 3-.
3/0 312 4 , (5)
where ∈ are the residuals from model one, +, , +. , +0 and +2 are dummy variables which equals one when a match is won and zero when the match is lost for friendly, qualification, group and elimination matches, respectively. , , . , 0 and 2 are dummy variables which equals one when a match is lost and zero when a match is won for friendly, qualification, group and elimination matches, respectively. Finally, day t is defined as the next trading day after a soccer match is played.
In line with Edmans et al. (2007), panel corrected standard errors are used in model two, which consists of heteroscedasticity-consistent standard errors as developed by White (1980). Furthermore, model one is estimated with both an Ordinary Least Squares method (OLS) and a GARCH (1,1) model to allow for time-varying volatility. Additionally, when estimating the GARCH (1,1) model, normalized returns are used as an independent variable instead of the regular log returns. The normalized returns eliminate heterogeneity in volatility across countries. The returns are normalized per country and are calculated by the following formula:
6789:; < => ?⁄ , (6)
12 Finally, ex-ante predicted probabilities by bookmakers will be used to control for unexpectedness of the result. Recall that it is hypothesized that when unexpectedness rises, the effect of the match on the index will increase as well. To test this, additional variables are added to model two. The method applied here deviates from the method of Edmans et al. (2007) in which they use quadratic programming. To measure the effect of unexpectedness, interaction variables are added which comprise of win and loss dummies multiplied by the probabilities of a win. Whereas Edmans et al. (2007) calculate their own probabilities for 40% of the sample with the help of Elo ratings, this paper only uses probabilities provided by bookmakers, which is considered to be a more objective measure for the actual probability. However, these probabilities are not dramatically different because there is a correlation of 0.929 with probabilities derived from Elo ratings in their paper. The win probabilities are calculated by converting the odds (correcting for the compensation of bookmakers) by the following example:
Team A plays a match versus team B in which the odds are: A for a win for Team A, X for a draw and B for a win for team B. Now suppose we want to know the probability of a win for team A. The calculation is as follows:
A B AC / EA F G H, (7)
where A is the probability of a win for team A.
Because equation 5 distinguishes between different types of matches, the interaction variables which capture the unexpectedness must also be divided into these different types. Furthermore, they will be divided into wins and losses. Therefore, there are 8 interaction variables which capture unexpectedness (see equation 8).
( )*+, )-+. )/+0 )1+2 3*, 3-.
13 where is the normalized probability of a win for team i, calculated in the same method as in equation 6. Day t is defined as the next trading day after the match.
Because the hypothesis is that when unexpectedness increases the effect of a match outcome on returns increases, the hypothesis concerning the expected signs of the coefficients can be derived. Firstly, let us look at wins. The effect of a win, for example in a group match, in equation 8 is defined as: L∈
L)/ )/ " . The hypothesized effect of a win is positive, therefore )/ " >0 should hold. Additionally, if the predicted probability of a win is high, the effect of a win is expected to be smaller than when the probability was expected to be low. Therefore, when the probability increases, the effect of a win should decrease, so "<0 should hold. However, because the total effect of a win should still be positive, )/ must be larger than zero and have an absolute value higher than " . A similar argument holds for the effect of losses (again using group matches as an example) where the effect of group losses is defined as: L∈
L3/ 3/ J . Because the effect of a loss is hypothesized to be negative, the derivative as a whole should be negative. Furthermore, when the probability of a win increases, a loss becomes more unexpected. Therefore, J should be negative because it increases the effect of a loss, meaning a larger negative value for the derivate as a whole. Finally, 3/ should not be able to make the total derivative positive. This means that 3/ should be equal of smaller than zero because when the probability of a win is close to zero, J approaches zero, but the total effect should still be negative. Therefore 3/≤0. Equation 8 is estimated with both raw winning probabilities and with normalized winning probabilities. However, only results regarding normalized probabilities will be presented because of their similar outcomes.
4. Data
14 The selection of countries is based on the ranking on the official FIFA world ranking list5. To expect a mood effect it is required that soccer is a popular sport in the country. If soccer is not popular in the country it is less likely that there will be a mood effect after a particular outcome because people are not emotionally involved. The measure used here for popularity is the ranking on the FIFA world ranking list because when a country excels at a particular sport is likely that it also of national interest. In contrast, when a country is ranked low it is likely that the country is not interested in this particular sport. Furthermore, countries ranked higher on the list are more likely to participate at World and European Cups and therefore provide data for all types of matches. The selection criterion utilized is that a country has to be ranked within the top 30 on the FIFA world ranking list during the time period of 2004 to 2013. The country has to be present on every year’s January top 30. The consideration here using the top 30 is that soccer should be popular in the countries selected but there also have to be sufficient matches in the dataset to make statistical inferences. There are 13 countries which satisfy this criterion, which are: Argentina, Brazil, Croatia, England, France, Germany, Greece, Italy, Mexico, The Netherlands, Portugal and Spain and Uruguay. Because of a lack of equity data on Uruguay in the database of Thomson Reuters DataStream, it will be excluded from the dataset which leaves 12 countries into the dataset.
As mentioned above, the ranking lists of the time period January 2004 to January 2013 are used to determine included countries. This time frame aligns with this paper’s sample period. To extend on the paper of Edmans et al. (2007), a sample period is chosen which approximately starts at the end of their time frame. Secondly, the odds data which are used for measuring unexpectedness are available from June 2004 onwards which happens to coincide approximately with the end of the time frame of Edmans et al. (2007). The time frame utilized by Edmans et al. (2007) is from January 1973 to December 2004. This paper will therefore use more recent data by only including matches from June 2004 to December 2013. Then, all matches played by the selected countries which fall within this time frame are selected, excluding draws. All matches relevant for the study should be either a friendly, qualification, European cup, American cup or World Cup match. Because part of this study is
5
15 to determine whether different types of matches trigger varying degrees of the mood effect, it is necessary to distinguish matches by their matter of importance to the spectators. Whereas results of the friendly matches have no further implication and is of low interest to the public, cup matches are considered prestigious and important because of their media coverage and the status acquired from winning the title. Therefore, mood effects are more likely to be found after results of matches which are considered important than those which are not. Furthermore, within cup matches there are different types of matches to be distinguished. This paper follows the method applied by Edmans et al. (2007) and categorizes matches in qualification matches, group stage matches and elimination stage matches. In contrast to their paper, friendly matches will be included as well. A description of the match types is given in Appendix A. Data on results, game dates, game type and odds are acquired from http://www.oddsportal.com/. The odds are averages of the odds provided by several bookmakers. Depending on the match there may be different bookmakers providing odds for that particular match. Odds of the following bookmakers are included: 10Bet, bet-at-home, bet365, Betsson, BetVictor, bwin, MarathonBet, Pinnacle Sports, SBOBET, Titanbet, TonyBet, Unibet, William Hill and youwin.
16 for this type of index. The argument for using national indexes to measure the mood effect of domestic investors rests on the “home bias” found by French and Poterba (1991) as was discussed previously.
The descriptive statistics of the sample can be found in appendix A. This appendix shows the number of matches in each category. The total number of matches is 1112 of which 849 are wins and 263 are losses. The number of wins is relatively high caused by selecting only top ranked countries which are more likely to win than to lose. Additionally, because of the high frequency of friendly and qualification matches, they form the major share of the dataset representing 82% of the matches. However, there still remain 113 group matches and 88 elimination matches to be analysed. Additionally, appendix A shows that the matches are approximately evenly spread among the countries. Table 1 shows the descriptive statistics of the abnormal returns in the estimation windows for the mean adjusted event study. The mean abnormal return in the estimation windows are all zero rounded to four decimals. Furthermore, the Jarque-Bera statistic rejects a normal distribution of abnormal returns caused by skewness to the left and a leptokurtic character. Table 2 provides the descriptive statistics for the market and risk adjusted event study. The image is similar to that shown in table 1. There is still rejection of a normal distribution as shown by the high Jarque-Bera statistics. These results are typical for financial data.
5. Results
5.1. Results of the event study using the mean adjusted methodology
17 is therefore more suitable to make inferences on. Furthermore, the sign test is not sensitive for outliers because it is a non-parametric test. Finally, in contrast to the p-values, the sign test takes into account what the sign should be according to the hypotheses (positive for wins and negative for losses), therefore, a value close to 1 indicates a significant result in the opposite direction than expected.
From table 3 it becomes clear that only wins generate significant results in the expected direction. However, these results are only supported by the Student’s t-test, which is not a reliable measure in this case. The most striking result is at the bottom of table 3, which shows that elimination losses trigger positive abnormal returns of 10 basis points on the next trading day, supported by the values of the sign test which are 0.934, 0.986 and 0.986 for estimation periods of 500, 250 and 125 days, respectively. The values close to one indicate a significant result in the opposite direction than expected. This result is surprising because considering the hypotheses in this paper it is expected that elimination losses trigger negative abnormal returns, which even should exceed hypothesized negative abnormal returns of other losses. This result is the opposite as to what Edmans et al. (2007) find, who find that elimination losses trigger negative returns which is significantly different from abnormal returns after group games and qualification games which were also shown to be significantly negative.
19 Table 2
Descriptive statistics of the estimation windows of the market and risk adjusted event study. The mean, median, standard deviation, minimum and maximum are measured in abnormal returns. The estimation period is defined as the following interval where t=0 on match days [-250;-1].
indexes Large caps Totmk indexes Small caps Estimation period in days 250 250 250
Mean 0.000 0.000 0.000 Median 0.000 0.000 0.000 Standard deviation 0.013 0.010 0.010 Minimum -0.050 -0.038 -0.040 Maximum 0.050 0.037 0.036 Skewness -0.099 -0.162 -0.344 Kurtosis 4.785 2.532 2.472 Jarque-Bera 1779.576 89.772 173.927
From their paper it is not clear whether qualification wins trigger a significant positive abnormal return because their focus is on differences between the effect of wins and losses and not the effect of wins and losses separately. Finally, the sign tests show a counterintuitive result for friendly matches. Friendly wins show an average abnormal return of -10 basis points and losses show an average abnormal return of 10 basis points. The sign tests reveal that these friendly wins trigger negative abnormal returns which are significant for all estimation periods except for 750 days estimation. The sign test probabilities are 0.942, 0.953 and 0.962 for estimation periods of 500, 250 and 125 days, respectively. The positive abnormal return after friendly losses is significant at an alpha of 10% for the estimation period of 125 days. These result are in contrast to the hypotheses. Edmans et al. (2007) did not examine the effect of friendly matches because of their minor importance to the public.
20 Table 3
Results of the mean adjusted event study for large capitalization stock indexes. AR is defined as the abnormal return on the next trading day after the match, p-value is the probability attached to a parametric one-sided Student’s t-test and the sign-test is the probability attached to the non-parametric sign test. The sign test uses a binomial distribution which therefore does not assume a specific distribution concerning the abnormal returns.
*=significant at an alpha of 10%, **=significant at an alpha of 5%, ***=significant at an alpha of 1%.
The surprising result of friendlies found for the TOTMK indexes is confirmed in that wins trigger negative abnormal returns significant at an alpha of 5% for all estimation periods for
indexes Large caps
21 the sign tests. The effect of friendly losses is diminished compared to the TOTMK analysis because it is not significantly positive.
From this analysis can be concluded that the null hypothesis concerning wins and losses cannot be rejected using the mean adjusted approach. The results do not show a clear pattern of positive abnormal returns after wins and negative ones after losses. Instead, for some cases the results contradict the hypotheses. Finally, results do not significantly differ between different estimation periods. Because of this finding, the following analysis with the market and risk adjusted method will only use an estimation period of 250 trading days.
5.2 Results of the event study using the market and risk adjusted methodology
Results of the market and risk adjusted method can be found in table 4. This table has the same format as table 3 but contains results for an estimation period of 250 trading days only and for all index types. Let us first discuss the results of large caps. Because table 2 showed us that the abnormal returns of the market and risk adjusted method are also not normally distributed the inferences made from the Student’s t-test are not reliable. Therefore, most weight will be applied towards the sign test again. The results suggest that wins in general trigger negative abnormal returns which is significant at an alpha of 5% for the sign test. This is again the opposite as expected and does not indicate the hypothesized mood effect. The effect of friendlies is comparable to that found for the mean adjusted method but it is not significant. Additionally, an opposite effect is found for qualification wins which trigger negative abnormal returns of -20 basis points the next trading day, significant at an alpha of 5%.
22 in line with our main hypotheses but there still is no obvious relationship between match outcomes and stock returns.
Table 4
Results of the market and risk adjusted event study for an estimation period of 250 days. AR is defined as the abnormal return, p-value is the probability of a one-sided t-test and sign-test is the probability attached to the sign test. If the probability of the sign test is close to one it indicates a relationship which is the opposite than expected. This is caused by the characteristics of the sign test. Therefore, when the sign test probability is for example 0.96, this indicates a significant relationship at an alpha of 5% in the opposite direction as to what was hypothesized.
Indexes Large caps TOTMK indexes Small caps
Estimation period 250 250 250 Total Wins AR -0.001 0.000 0.000 p-value 0.003*** 0.303 0.114 Sign-test 0.986 0.698 0.655 Losses AR 0.000 0.000 -0.001 p-value 0.282 0.417 0.097* Sign-test 0.597 0.500 0.034** Friendlies Wins AR -0.001 0.000 -0.001 p-value 0.132 0.266 0.083* Sign-test 0.820 0.414 0.898 Losses AR 0.001 0.001 0.000 p-value 0.384 0.221 0.396 Sign-test 0.829 0.952 0.269 Qualification Wins AR -0.002 0.000 -0.001 p-value 0.012** 0.266 0.149 Sign-test 0.961 0.853 0.718 Losses AR -0.003 -0.002 -0.002 p-value 0.205 0.231 0.116 Sign-test 0.179 0.070* 0.000*** Group Wins AR 0.000 0.002 0.002 p-value 0.254 0.019** 0.086* Sign-test 0.545 0.070* 0.003*** Losses AR -0.001 0.001 0.001 p-value 0.181 0.311 0.335 Sign-test 0.500 0.500 0.279 Elimination Wins AR -0.002 -0.001 0.000 p-value 0.132 0.276 0.378 Sign-test 0.894 0.894 0.500 Losses AR 0.000 -0.002 -0.005 p-value 0.472 0.141 0.073* Sign-test 0.434 0.033** 0.012**
23 Finally, the results of the small capitalization stocks. In contrast to the other indexes there is a significant abnormal return of -10 basis points after losses in general, significant at an alpha of 5%. Furthermore, as was the case with TOTMK indexes, significant results are found for qualification losses, elimination losses and group wins, all in the expected direction. Qualification losses show an average abnormal return of -20 basis points, significant at an alpha of 1%. Elimination losses show an average abnormal return of -50 basis points, significant at an alpha of 5%. Finally, group wins produce an average abnormal return of 20 basis points, significant at an alpha of 1%. It is only for this part of the event study that there are no results indicating an opposite effect than expected. The fact that small stocks generate more reasonable results in line with the expectation may be because of their larger home bias. Since they are owned by a relatively larger share of domestic investors it could be that the mood effect is primarily present in small stocks and not so much in other indexes. This result is in line with Edmans et al. (2007) who also finds a stronger effect for small cap indexes.
24 5.3 Results of the econometric approach
Table 5, 6 and 7 show the regression results of model two for large capitalization stocks, TOTMK indexes and small capitalization stocks, respectively. Panel A shows the results using the raw returns with an Ordinary Least Squares estimation. Panel B shows the results for normalized returns utilizing a GARCH estimation. The returns are normalized per country with equation 6. The coefficients in Panel A are abnormal returns caused by that particular variable. However, because in panel B the coefficients are for normalized returns, we need to multiply the coefficients with the average daily standard deviation of the market indexes to get the abnormal return. When multiplied, the results for both methods can be compared (Edmans et al., 2007).
Table 5 shows a surprising result, namely that wins overall have an average abnormal return of -0.119, significant at an alpha of 5% in panel A. This result is similar in panel B. The daily standard deviation of large capitalization stocks in this paper’s sample is 1.605. Therefore, the abnormal return for overall wins is -0.0846*1.605= -0.136. This not only contradicts the hypothesized relation, but in fact shows an opposite relationship. In panel B, an opposite relation is also found for qualification wins, which is in line with the results of the event study. The abnormal return for this particular event is -0.1427*1.605= -0.229, significant at an alpha of 5%.
Table 6, which uses TOTMK index returns as an independent variable does not show a result considering wins and losses in general. However, in line with the event study results it does show a positive effect of group wins, which remains when doing a GARCH estimation with normalized returns. For the raw returns, an abnormal return of 0.233 is found, which is significant at an alpha of 5%. In contrast to the event study, it does not find an effect for qualification and elimination losses.
25 From these results we can conclude that the null hypotheses concerning wins and losses cannot be rejected. Although for specific cases there are results in line with our hypothesis, it is not frequent enough to reject the null hypotheses from the introduction. From this automatically follows that the null hypotheses concerning an increasing effect when importance of the match increases also cannot be rejected. These results are in contrast to Edmans et al. (2007) who do find a significant loss effect and an increasing effect when importance increases. These results can be found in the literature review.
Table 5
Results of model two for the econometric approach for large capitalization stocks. ) and 3 are defined as the coefficients representing the abnormal return following wins and losses, respectively. In panel B, the abnormal returns can be calculated by multiplying the betas by the daily standard deviation of the stock indexes, which is 1.605 for large capitalization stocks. N is the number of matches in each category and the p-value is the significance value of the associated coefficient.
Wins Losses
Matches N ) p-value N 3 p-value Panel A: raw data using OLS
All matches 846 -0.119 0.049* 263 -0.096 0.406 Friendlies 344 -0.075 0.388 134 0.037 0.791 Qualification 373 -0.160 0.118 58 -0.515 0.167 Group 78 -0.101 0.406 35 -0.101 0.560 Elimination 51 -0.141 0.516 36 0.090 0.646
Panel B: normalized returns in a GARCH estimation All matches 846 -0.085 0.024** 263 -0.077 0.289
Friendlies 344 -0.037 0.489 134 -0.006 0.946 Qualification 373 -0.143 0.027** 58 -0.340 0.135 Group 78 -0.028 0.685 35 -0.044 0.697 Elimination 51 -0.068 0.605 36 0.051 0.683
*=significant at an alpha of 10%, **=significant at an alpha of 5%, ***=significant at an alpha of 1%.
26 remains present in all subsamples. Most of these results are significant at an alpha of 5%. Furthermore, appendix E shows a similar outcome in which the qualification wins remain to have a negative effect. However, panel B in appendix E has a lack of observations to make a solid conclusion for non-Europe.
Table 6
Results of model two for the econometric approach for the TOTMK indexes. ) and 3 are defined as the abnormal return following wins and losses, respectively. In panel B, the abnormal returns can be calculated by multiplying the betas by the daily standard deviation of the stock indexes, which is 1.420 for TOTMK indexes. N is the number of matches in each category and the p-value is the significance value of the associated coefficient.
Wins Losses
Matches N ) p-value N 3 p-value Panel A: raw data using OLS
All matches 842 -0.001 0.985 261 -0.028 0.730 Friendlies 344 -0.020 0.772 133 0.069 0.488 Qualification 369 -0.035 0.612 58 -0.228 0.302 Group 78 0.234 0.010** 34 0.062 0.724 Elimination 51 0.018 0.909 36 -0.148 0.495
Panel B: normalized returns in a GARCH estimation All matches 842 -0.016 0.608 261 -0.034 0.551
Friendlies 344 -0.017 0.717 133 0.0307 0.677 Qualification 369 -0.046 0.351 58 -0.167 0.276 Group 78 0.135 0.033** 34 0.017 0.875 Elimination 51 -0.016 0.881 36 -0.105 0.455
*=significant at an alpha of 10%, **=significant at an alpha of 5%, ***=significant at an alpha of 1%.
27 positive sign, indicating that unexpectedness does not affect the size of the abnormal return. These results are in line with Edmans et al. (2007) who also do not find a relationship between unexpectedness and stock market returns.
Table 7
Results of model two for the econometric approach for small capitalization stocks. ) and 3 are defined as the abnormal return following wins and losses, respectively. In panel B, the abnormal returns can be calculated by multiplying the betas by the daily standard deviation of the stock indexes, which is 1.404 for small caps. N is the number of matches in each category and the p-value is the significance value of the associated coefficient.
Wins Losses
Matches N ) p-value N 3 p-value Panel A: raw data using OLS
All matches 636 -0.0504 0.4066 190 -0.0879 0.4329 Friendlies 225 -0.0502 0.5877 95 -0.0121 0.9284 Qualification 304 -0.0800 0.4204 40 -0.4582 0.2089 Group 63 0.0349 0.7505 26 0.0131 0.9474 Elimination 44 0.0306 0.8795 29 0.0844 0.6997
Panel B: normalized returns in a GARCH estimation All matches 636 -0.0250 0.5071 190 -0.0868 0.1664
Friendlies 225 -0.0762 0.1884 95 -0.0197 0.8194 Qualification 304 -0.0204 0.7298 40 -0.1890 0.1297 Group 63 0.1301 0.1550 26 0.0178 0.9045 Elimination 44 -0.0165 0.9057 29 -0.2594 0.1872
28 Table 8
Results of model 2 with the Garch (1,1) estimation utilizing normalized returns and normalized probabilities. is defined as the coefficient on the interaction variables which multiply winning probabilities with dummies for wins and losses, the p-value is the probability of a joint significance Wald test testing whether the combination of the coefficients of the win or loss dummy together with the probability of a win is significant. ) and 3 are defined as win and loss dummies, respectively and is the coefficient of the interaction variables. The total effect of wins and losses can be calculated by summing the coefficients of the win or loss dummy and .
Wins Losses
Matches N ) p-value N 3 p-value
Panel A: Large Caps
All matches 846 -0.0942 0.0394 0.0453* 263 -0.1150 -0.0487 0.5654 Friendlies 344 -0.0399 0.0342 0.6187 134 0.0578 0.0791 0.8269 Qualification 373 -0.2012 0.0987 0.0508* 58 -0.4566 -0.2466 0.3062 Group 78 -0.0178 0.1349 0.2425 35 0.1234 0.1846 0.3563 Elimination 51 -0.1233 -0.0703 0.7329 36 -0.1288 -0.1711 0.6168
Panel B: TOTMK indexes
All matches 842 -0.0260 0.0427 0.3916 261 -0.0019 0.0410 0.7218 Friendlies 344 -0.0188 0.0182 0.8954 133 0.0147 -0.0199 0.9108 Qualification 369 -0.1170 0.1194 0.0809* 58 -0.1425 0.0516 0.4864 Group 78 0.1432 0.1137 0.0208** 34 0.2444 0.2552 0.0545* Elimination 51 -0.1460 -0.1647 0.4599 36 0.1818 0.2734 0.3858
Panel C: Small Caps
All matches 636 -0.0306 0.0239 0.6716 190 -0.0328 0.0709 0.3381 Friendlies 225 -0.0757 -0.0101 0.4211 95 -0.0525 -0.0413 0.9181 Qualification 304 -0.0697 0.0827 0.4357 40 -0.2183 -0.0672 0.2951 Group 63 0.1341 0.0442 0.2378 26 0.2697 0.3386 0.0702* Elimination 44 -0.1762 -0.1877 0.5612 29 0.4233 0.6133 0.1610
*=significant at an alpha of 10%, **=significant at an alpha of 5%, ***=significant at an alpha of 1%.
6. Conclusion
29 and group wins triggered returns in line with the hypotheses. However, when turning to the other indexes these results disappear. The reason for this particular finding may be because of a larger ‘’home bias’’ in small capitalization stocks. A larger home bias should result in a stronger effect of sports sentiment because a larger share of domestic investors is affected by the result of their national team. Therefore, it seems to be the case that small
capitalization stocks are more sensitive for sports sentiment compared to large capitalization stocks and TOTMK indexes. When examining the results of the econometric approach it shows that they are similar in that no systematic effects are found following wins and losses. On the one hand, significant signs in the theorized direction are present in some specific cases, on the other hand, there are also results implying the opposite relationship in a statistically significant matter. Therefore, controlling for market factors and using a
continuous sample does not change the results by a significant degree. The final part of the study was to examine whether the predicted probabilities of bookmakers have an impact on the size of the effect of wins and losses. This seemed not to be the case, as Edmans et al. (2007) find as well. The overall conclusion from this paper is that, assuming an effect of sport outcomes on the mood of investors, mood does not have an impact on the decision making of investors. However, these results are in sharp contrast to the results of Ashton et al. (2003) and Edmans et al. (2007) who perform similar studies. Furthermore, it is also in contrast to the studies that were discussed in the literature review which use other
measures for mood such as weather conditions and lunar cycles. One reason for the differing results could come from the fact that a large part of this paper’s sample period is in the crisis which started in 2008. In contrast to Edmans et al. (2007), this paper also included friendly matches which may not trigger an effect because of their minor importance to the public. However when turning the attention to the other match types the results can still be compared with Edmans et al. (2007). Furthermore, Edmans et al. (2007) select their
qualification matches with a “closeness in ability” criteria, meaning that the teams should be approximately of equal ability. This paper does not select matches with this criteria. The argument for this is that the winning probabilities should correct for these unequal matches. When unequal matches are left out of the sample the effect of the variable ‘winning
probabilities’ is diminished because the range of winning probabilities is narrowed.
31
7. Appendix
Appendix A
Descriptive statistics of the dataset. This table presents the number of matches within each category. Friendly matches are matches not part of a cup or tournament, Qualification matches are played to determine which teams qualify for Cups. Group matches are matches which are played in the first stage of Cups, often referred to as a round-robin tournament or single round-robin tournament, in which teams meet each other twice or once, respectively. When the group stage is finalized the proceeding teams play elimination rounds in which a loss leads to a direct elimination of the tournament. A tie means overtime and, if there is still no winner, eventually penalties to decide a winner.
Country matches wins losses friendlies qualification group elimination
32 Appendix B
Results of the mean adjusted event study for the TOTMK indexes and the small capitalization stocks. AR is defined as the abnormal return on the next trading day after the match, p-value is the
probability attached to a parametric one-sided Student’s t-test and the sign-test is the probability attached to the non-parametric sign test. The sign test uses a binomial distribution which therefore does not assume a specific distribution concerning the abnormal returns.
indexes TOTMK indexes Small caps
33 Appendix C
Results of the event study for the market and risk adjusted method for TOTMK indexes and an estimation period of 250 days. This table only contains results on the subsamples Europe, non-Europe, without Greece, without Argentina, without brazil, without Mexico, and Europe without Greece.
Indexes Europe
34 Appendix D
Results of the robustness check leaving out one country at the time with GARCH (1,1) estimation using normalized returns of large capitalization stocks. the abnormal returns can be calculated by multiplying the betas by the daily standard deviation of the stock indexes, which is 1.605 for large capitalization stocks. N is defined as the number of matches in each category.
Wins Losses
Matches N ) p-value N 3 p-value
Panel A: without Argentina
Friendlies 309 -0.0637 0.2459 121 -0.0629 0.5223 Qualification 354 -0.1270 0.0411** 49 -0.3089 0.2161 Group 70 -0.0217 0.7671 35 -0.0378 0.7360 Elimination 49 -0.0703 0.6059 32 0.0753 0.5582
Panel B: Without Brazil
Friendlies 290 -0.0288 0.6210 125 -0.0306 0.7421 Qualification 363 -0.1296 0.0367** 55 -0.3312 0.1627 Group 70 0.0241 0.7203 34 -0.0383 0.7404 Elimination 46 -0.0384 0.7683 33 0.0562 0.6741
Panel C: Without Croatia
Friendlies 323 -0.0304 0.5735 126 0.0015 0.9876 Qualification 339 -0.1048 0.1100 50 -0.4271 0.0875*
Group 74 -0.0400 0.5820 32 -0.0031 0.9785 Elimination 51 -0.0693 0.5971 35 0.0660 0.5994
Panel D: Without England
Friendlies 318 -0.0251 0.6541 122 -0.0320 0.7421 Qualification 341 -0.1499 0.0269** 53 -0.3537 0.1498 Group 71 -0.0258 0.7311 34 -0.0338 0.7636 Elimination 50 -0.0691 0.6077 32 0.0258 0.8510
Panel E: Without France
Friendlies 315 -0.0318 0.5720 122 0.0522 0.5868 Qualification 343 -0.1330 0.0538* 52 -0.3849 0.1226 Group 75 -0.0304 0.6701 30 -0.0084 0.9460 Elimination 48 -0.0883 0.5273 33 0.0719 0.5776
Panel F: Without Germany
Friendlies 313 -0.0271 0.6340 121 -0.0101 0.9171 Qualification 338 -0.1413 0.0361** 57 -0.3212 0.1624 Group 70 -0.0578 0.4352 32 -0.1090 0.3329 Elimination 42 -0.0535 0.7299 32 0.0176 0.8713
Panel G: Without Greece
Friendlies 331 -0.0549 0.3087 123 -0.0018 0.9855 Qualification 334 -0.1852 0.0060*** 52 -0.2994 0.2030 Group 76 -0.0233 0.7420 28 -0.0901 0.4223 Elimination 48 -0.0946 0.4923 35 0.0209 0.8672
Panel H: Without Italy
35
Appendix C continued
Group 73 -0.0482 0.4884 33 -0.0281 0.8074 Elimination 45 -0.1062 0.3424 34 0.0636 0.6265
Panel I: Without Mexico
Friendlies 309 -0.0495 0.3517 115 0.0352 0.6897 Qualification 346 -0.1477 0.0279** 48 -0.2178 0.2634 Group 75 -0.0013 0.9851 29 -0.0180 0.8824 Elimination 49 -0.0168 0.8992 34 0.0059 0.9618
Panel J: Without Netherlands
Friendlies 320 -0.0424 0.4496 125 -0.0166 0.8609 Qualification 332 -0.1452 0.0372** 55 -0.3676 0.1195 Group 69 -0.0107 0.8851 33 -0.0403 0.7312 Elimination 46 -0.0643 0.6449 31 0.0522 0.7144
Panel K: Without Portugal
Friendlies 324 -0.0361 0.5215 125 0.0252 0.7820 Qualification 338 -0.1521 0.0296** 54 -0.3577 0.1398 Group 68 -0.0491 0.5306 32 -0.0561 0.6454 Elimination 46 -0.0754 0.6022 30 0.0898 0.5121
Panel L: Without Spain
Friendlies 303 -0.0264 0.6394 128 -0.0075 0.9361 Qualification 336 -0.1566 0.0247** 56 -0.3411 0.1461 Group 67 -0.0547 0.4676 33 -0.0693 0.5527 Elimination 41 -0.0723 0.6333 35 0.0612 0.6295
36 Appendix E
Results of robustness check for Europe and non-Europe, with GARCH (1,1) estimation using normalized returns of large capitalization stocks. abnormal returns can be calculated by multiplying the betas by the daily standard deviation of the stock indexes, which is 1.605 for large capitalization stocks. N is defined as the number of matches in each category
Wins Losses
Matches N ) p-value N 3 p-value Panel A: Europe Friendlies 220 -0.0807 0.1625 93 -0.0617 0.5125 Qualification 317 -0.1147 0.0592* 36 -0.0983 0.6267 Group 59 0.0753 0.2678 28 -0.0050 0.9685 Elimination 42 0.0277 0.8387 27 0.0270 0.8476 Panel B: non-Europe Friendlies 124 0.0516 0.5474 41 0.0986 0.5770 Qualification 56 -0.3418 0.0419** 22 -0.4115 0.2829 Group 19 -0.2899 0.0573* 7 -0.1536 0.5129 Elimination 9 -0.3265 0.2589 9 0.1171 0.6607
37 References
Ackert, L. F., & Deaves, R. (2009). Behavioral Finance: Psychology, Decision-Making, and Markets: Psychology, Decision-Making, and Markets. Cengage Learning.
Arkes, H. R., Herren, L. T., & Isen, A. M. (1988). The role of potential loss in the influence of affect on risk-taking behavior. Organizational behavior and human decision processes, 42(2), 181-193.
Ashton, J. K., Gerrard, B., & Hudson, R. (2003). Economic impact of national sporting success: evidence from the London stock exchange. Applied Economics Letters, 10(12), 783-785. Brown, S. J., & Warner, J. B. (1980). Measuring security price performance. Journal of financial Economics, 8(3), 205-258.
Brown, S. J., & Warner, J. B. (1985). Using daily stock returns: The case of event studies. Journal of financial economics, 14(1), 3-31.
Cao, M., & Wei, J. (2005). Stock market returns: A note on temperature anomaly. Journal of Banking & Finance, 29(6), 1559-1573.
Chang, S. C., Chen, S. S., Chou, R. K., & Lin, Y. H. (2008). Weather and intraday patterns in stock returns and trading activity. Journal of Banking & Finance, 32(9), 1754-1766.
Cialdini, R. B., Borden, R. J., Thorne, A., Walker, M. R., Freeman, S., & Sloan, L. R. (1976). Basking in reflected glory: Three (football) field studies. Journal of personality and social psychology, 34(3), 366. Cross, F. (1973). The behavior of stock prices on Fridays and Mondays. Financial analysts journal, 67-69.
Dahlquist, M., & Robertsson, G. (2001). Direct foreign ownership, institutional investors, and firm characteristics. Journal of Financial Economics, 59(3), 413-440.
Edmans, A., Garcia, D., & Norli, Ø. (2007). Sports sentiment and stock returns. The Journal of Finance, 62(4), 1967-1998.
French, K. R., & Poterba, J. M. (1991). Investor diversification and international equity markets (No. w3609). National Bureau of Economic Research.
French, K. R. (1980). Stock returns and the weekend effect. Journal of financial economics, 8(1), 55-69.
Hirshleifer, D., & Shumway, T. (2003). Good day sunshine: Stock returns and the weather. The Journal of Finance, 58(3), 1009-1032.
38
Kamstra, M. J., Kramer, L. A., & Levi, M. D. (2000). Losing sleep at the market: The daylight saving anomaly. American Economic Review, 1005-1011.
Kang, J. K., & Stulz, R. (1997). Why is there a home bias? An analysis of foreign portfolio equity ownership in Japan. Journal of Financial Economics, 46(1), 3-28.
Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. The review of economics and statistics, 13-37.
Markowitz, H. (1952). Portfolio selection*. The journal of finance, 7(1), 77-91.
Mossin, J. (1966). Equilibrium in a capital asset market. Econometrica: Journal of the Econometric Society, 768-783.
Merton, R. C. (1987). A simple model of capital market equilibrium with incomplete information. The journal of finance, 42(3), 483-510.
Saunders Jr, E. M. (1993). Stock prices and Wall Street weather. American Economic Review, 83(5), 1337-45.
Schwarz, N., Strack, F., Kommer, D., & Wagner, D. (1987). Soccer, rooms, and the quality of your life: Mood effects on judgments of satisfaction with life in general and with specific domains. European Journal of Social Psychology, 17(1), 69-79.
Schweitzer, K., Zillmann, D., Weaver, J. B., & Luttrell, E. S. (1992). Profile: Perception of threatening events in the emotional aftermath of a televised college football game. Journal of Broadcasting & Electronic Media, 36(1), 75-82.
Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk*. The journal of finance, 19(3), 425-442.
Snyder, C. R., Lassegard, M., & Ford, C. E. (1986). Distancing after group success and failure: Basking in reflected glory and cutting off reflected failure. Journal of Personality and Social Psychology, 51(2), 382.
Treynor, J. L. (1961). Market value, time, and risk. Unpublished manuscript, 95-209.
Yuan, K., Zheng, L., & Zhu, Q. (2006). Are investors moonstruck? Lunar phases and stock returns. Journal of Empirical Finance, 13(1), 1-23.
Wann, D. L., Branscombe, N. R., & Dolan, T. J. _, Kimberly K. McGeorge, Julie A. Allison
(1994),“Relationships Between Spectator Identification and Spectators’ Perceptions of Influence, Spectators’ Emotions, and Competition Outcome,”. Journal of Sport and Exercise Psychology, 16, 347-364.
