Local Football Sentiment and Returns from Stocks: Evidence from Europe

Размещено на http://www.allbest.ru/

Размещено на http://www.allbest.ru/

Local Football Sentiment and Returns from Stocks: Evidence from Europe

Introduction

stock evidence football

Behavioral approaches to the economic theories play a very important role in understanding economics as a whole. Even first classic traditional economists like Adam Smith, Irving Fisher or John Maynard Keynes pay their attention to the problem of individual's bounded rationality. Smith in his various papers wrote about human misevaluation of own labor skills which leads to the underpricing wages; Fisher in his book The Theory of Interest (1930) developed a `money illusion' phenomena and based the whole theory of nominal interest rates on it; Keynes in his book The General Theory of Employment, Interest and Money (1936) coined the term `animal spirits', that refers to the human behavior in financial decision making. As one can see, behavioral concepts were very broadening among different economists through the whole history of economic thought.

The rationale behind this idea is that humans are not `the ideal utility maximization machines' and it's absolutely normal for people to make irrational decisions, which may be controversial to the existing economic theories. That's why so many economists try to expand existing theories with behavioral notions or to create new ones with help of it, because it leads our knowledge of economics closer to reality. The main contributors to the behavioral theory in the economic field can be surely named as Daniel Kahneman and Amos Tversky. Their scientific collaboration made behavioral economics very popular among economists and directed a new way of development of economic thought. Their famous findings contained a developed knowledge of the presence of heuristics and biases and their following systematic and predictable errors; framing effect and its influence on rational choice; a critique of the traditional expected utility in a form of its inconsistency to represent a decision-making choices under risk as a model and hence developing an alternative `prospect theory' model and its further development into the utility model of both uncertainty and risk attitudes and a lot more. As one can see, all of these findings served as a `turning point' for future economic thoughts and till now a great amount of papers are still contributing to the new inventions in behavioral economics theory. The brightest example of the fact that behavioral economics is still very actual subject now is the Nobel Prize in Economic Sciences in 2017 which was awarded to Richard H. Thaler "for his contributions to behavioral economics". Richard Thaler's contribution to behavioral economics was not only in form of development the theoretical background in his scientific papers, but in implementing the existing knowledge into different parts of economics, especially into financial markets in order to describe the evidence of human's irrationality in the investors' decision-making processes. In his early works he argued about market efficiency and find the evidence of investors' overreaction phenomena in stock markets. The empirical evidence was provided for the fact that the returns of stocks which have experienced extreme capital losses in the past tended to outperform the portfolio with stocks that have experienced extreme capital gains in the past on 25% on average (DeBondt and Thaler (1985)). It means that the investor's vision of stocks is biased towards its capital gains and extreme losses that leads to the underestimation of the current stock even though it's contradictory to the efficient market hypothesis which assumes a complete rationality of investors. Also, some of his studies were contributed to the human's perception of risk, like the observation and confirmation of the `endowment effect' - the fact that people often demand much more to give up an object than they would be willing to pay to acquire it (Kahneman, Knetsch and Thaler (1991)). Moreover, Thaler didn't stop on the theorizing financial market's psychological distortions and brought its theory into the real world by founding its own asset management fund to “pioneered the application of behavioral finance in investment management”.

As one can see the behavioural approaches to economic study play a very important role. Financial markets are amongst the potential fields of economics when one could explore distortions from a traditional theory which can be a subject of behavioural studies. One of such areas is abnormal returns from stocks that happened because of wrong asset evaluation. Of course, there could be some cases when asset mispricing happens because of wrong data disclosure or misinterpretation of financial results but because of millions of investors taking part in everyday trading such cases are quite unlikely. Contrary, everyday excessive gains or losses can be alternatively explained using behavioural explanations that distort investors' rationality and hence leads to misleading market outcomes in stock markets.

This paper is trying to specify one of the many potential reasons that can influence investor mood by primarily concerning with studying some additional behavioural variables that influence the market outcome in national stock exchanges. By moving the explanatory subject into financial markets, current study is implementing econometric analysis of abnormal returns using football sentiment as a tool. The football was chosen as an area that can potentially cause deviations from rationality among investors because football matches are quite suitable as exogenous variables, since they do not carry any economic sense (thus, there is no room for reverse causality problem), while at the same time they are strong determinants of individual's mood.

In order to expand the existing knowledge related to sport sentiment, this study focuses on firm-level analysis and tries to explore the relationship between a football club and locally headquartered company which is trading on the national stock exchange. The main contribution to the knowledge of football sentiment and its financial market's consequences was made by analysing mainly national football teams matches during major sport events (like FIFA World Cup or UEFA European Championship) and their influence on national market indexes or using very limited football clubs' sample to test some financial indicators (like exchange rate). However, not much attention was paid to micro-level analysis that specifies football club performance on unique company's return. This paper is trying to fill in such poorly discovered area and increase the knowledge of using football sentiment in prediction of abnormal returns' outcome. The study includes six different European countries of both developed and developing economic status and also controls for the firm-specific differences in the level of market capitalization. For such purposes, the sample was created which consists of 348034 daily market observations that are distributed among 225 companies on the period from 2012 to 2018. Using modified Edmans (2007) approach stocks' abnormal returns would be tested on the presence of significance among various football matches' dummy variables. There are 9614 football matches in the sample that attributes to the particular 43 football clubs. The football club level sample is also distributed among three major football club competitions - first national division, UEFA Champions League and UEFA Europa League - in order to control the difference between sentiment.

There are four main hypotheses in current study to be tested:

1. The main one concerns about the question of the general significance of local football sentiment; whether company's stock returns are affected by the outcome of the local football team game.

2. If the first hypothesis finds a statistical evidence of the presence of football sentiment in companies' stock return evaluation, then one should test the difference between behavioral effect among big cap and small cap stocks.

3. Based on the findings one should also check whether the initial reasonable assumption that `win' of football match is associated with the following positive return and vice versa is true and if it's not how one can formulate the general rule of market outcome based on win/loss outcome.

4. The fourth hypothesis relates to economic development; whether the effect of football sentiment is different in developed and developing countries.

1. Literature review

The theoretical background of investor psychology regarding asset pricing was provided by Hirshleifer (2001). This paper investigates the effect of investor mood on asset pricing. Author provided possible explanations to psychological effects that are relevant for studying investor behaviour in securities markets. Hirshleifer listed: heuristic simplification - cognitive resource constraints: read limited attention, processing power, memory constraints etc.; narrow framing (Kahneman and Lovallo (1993)) - overconfidence of investor in its own selecting portfolio skills, subjecting to `fashion', marketing stocks etc.; disposition effect (Shefrin and Statman (1985)) - an excessive propensity to hold on to securities that have declined in value and to sell `winner' stocks (ones that was observed with excessive capital gains); house money effect (Thaler and Johnson (1990)) - more disposition to take part in risky activities connected with money that was recently won; and a lot more. Such behavioural distortions could serve as fundamental variables that explain irrational investors' decisions in stock markets.

By getting closer to the understanding investor mood one should first suggest that this variable is not only theoretically significant but found some empirical support for it, too. Mehra and Sah (2002) viewed investor mood as two variables: investors' discount factors and investors' levels of risk-aversion. They consider these factors as subjective variables that could potentially influence the market trade outcomes. Authors found evidence that both of these sources promote volatility in the prices of stocks. Moreover, even small changes in investors' discount factors is followed by large fluctuations in equity prices. Another study in this area belongs to Etzioni (1988), who investigated the effect of so-called `N/A' (normative/affective) factors. The study argues the theoretical thesis that all of the individual's actions are fully based on rational foundations, while real-life cases suggested that it's not. Alternatively, the author stated, most economic choices are made based on emotional involvements and value commitments, while often excluding information processing. That's why investor mood is one of the most important variable in the investor's decision-making process. Forgas (1998) implemented three experiments and found that negative individual's mood decreases correspondence bias (or fundamental attribution error- FAE) - the tendency to explain the actions and behaviour of other people by internal factors, while own behaviour by external factors - and positive moods increase the FAE, because of the information-processing consequences of these affective states. In other words, a good mood is associated with fast and efficient decision-making. Thus, it's seen that based on psychological researches mood significantly affects the investor's decision-making process even though individual may be unconscious of its appearance and hence mood can be further considered to be an important subject of study to additional findings in that area.

The important mission to aggregate all existing studies about investor mood and financial markets was made by Shu (2010). The author tried to link the existing theories about psychological biases affecting the investor's behavior with the empirical evidence of such a phenomenon. With its own constructed asset pricing model, author came up to several important conclusions related to the subject. First of all, results in this study suggested that changes in investor mood are followed up by some market implications, meaning that the fluctuations in investor mood have a real impact on the final market outcomes. In previous studies, it was found that investor mood has a serious influence on two very important variables: risk attitude and time preferences, the study of Shu confirmed it and additionally stated that the returns of the assets are significantly affected by investor mood through these two variables. The study also investigated the volatility issues: changes in investor mood provoke fluctuations in the equity market in a form of increased volatility. A lot more suggestions were made in this study and all of these bring the author to one important conclusion: the inclusion of different investor mood variables into asset pricing model succeed to explain the various price and return anomalies in the financial markets. This implication is crucial for current diploma paper since it allows to be sure that the investigation of different components of investor mood is legitimate and can be further rewarded with significance in the explanation of asset returns.

In order to be specific in the conclusion of the previous paragraph, I would like to review some of the major works that were done on this subject in a matter of observing some different variables related to investor mood. There is a great variety of studies that pick some of the mood variables and then test them on financial markets for significance. One of the `traditional' attempt to explain mood variable goes through the weather. In the study of Hirshleifer and Shumway (2003) the sunshine metrics were examined. The authors collected the data of total sky cover (SKC index) in 26 countries for the period 1982-1997 and with strictly controlling the seasonal effect the results suggested the evidence of the presence of strong significance in the correlation between sunshine and stock returns. The study of Cao and Wei (2005) is focused on the temperature outside. Authors took the data of daily temperature from all over the world from the period 1962-2001 and tested it with available market indexes. The results suggested the overall negative correlation between daily temperature and stock market returns, which was consistent with previously formed hypothesis that lower daily temperature leads to higher stock returns (more individual's aggression - more aggressive risk-taking), while higher daily temperature effect is ambiguous (since high daily temperature causes both apathy and aggression). Another part of studying investor mood implications on market returns is trying to find evidence in air pollution (Levy and Yagil (2011)). Driven by the common knowledge that air pollution influences negatively on mood, researchers found a significantly negative relation between air pollution and stock return. Some findings in investor mood variables are really exotic. For example, Wang, Lin and Chen (2010) chose lunar cycle variable as the one that contributes to the investor mood and foundempirical evidence that lunar cycle has a negative influence on stock returns and positive influence on stock volatility. While the other study found that the increased attendance to comedy movies on weekends is related to a decrease in stock returns on the Mondays (Lepori (2015)). All of these studies present great evidence for variation of studied subjects in investor mood variable. Due to the complexity of human nature, there is a great variety in factors that influence our mood and therefore our behavior.

All of this narration with empirical examples creates evidence of the presence of investor mood and thus makes the subject of this work relevant for the investigation. In the previous examples, we've seen the different approaches to investor mood in terms of different factors like weather, temperature, etc. This work is trying to bring more evidence for another field of investor mood investigation - the football. Football is the most popular sport in Europe in terms of its fan base and news coverage. It covers a lot of space around every individual and that's why football is still a subject of investigation in a scientific area in a sense of its influential power. One of the most significant explanation that football can be a variable that explains investor mood: football doesn't coincide with economic consequences. If some of the teams lose or win the match, it doesn't result in the market revaluation of the asset, because it doesn't have a direct influence on economics. In order to get more clear evidence of football's influential role in people's minds, one need to look more closely to the psychological works that were done in this area. The early paper of Schwarz, Strack, Kommer and Wagner (1987) used football as a tool in order to measure an individual's life satisfaction at the exact moment. Authors chose two games from the World Cup 1982 and tried to find out respondents' happiness and life satisfaction shortly before or after the broadcasting of the games. The results suggested that individuals reported more positive evaluations after the German team had won a game than before, but less positive evaluations when the game was tied. This study brings clear evidence of the significance of football games on post-game individual's consciousness. Another study brings more evidence to the post-game individual's behavior. Bizman and Yinon (2002) found out that self-esteem and positive emotions were higher while negative emotions lower, after the team wins the match than after team loses the match. Also, football fans were found to associate more with the team after the team wins its game than after the team fails to do so. The study of Wann, Dolan, McGeorge and Allison (1994) tested several hypotheses about athletic games sentiment and fans' self-identification with the playing teams. For fans with high self-identification with the team fans there was found an increase in pre- to postgame positive emotions following a win and an increase in negative emotions following a loss. Given that among investors there is definitely some number of football supporters, the review psychological studies about the relation between football and mood gives us a clear link to the following empirical investigation of football sentiment in the financial markets.

The most fundamental study that links football results and financial markets through investor mood was the paper provided by Edmans, Garcнa and Norli (2007). The author integrated a new investor mood variable that was based on international soccer results. Authors used a sample of international soccer results from January 1973 through December 2004 which included football matches from the World Cup, European Championship, Copa America, and Asian Cup with a corresponding 39 countries. Also, the corresponding market indices for given countries were collected. As a result, all football matches were used as mood events which used to be an influential factor of investors' mood at a particular time. In order to measure the effect of football matches, authors looked at the changes in daily local currency return on a broadly based stock market index for each country on the next after the match. In order to do this, they've created a two-stage model, which modified version will be used in the current study. In the end, the result suggested a presence of significantly strong negative stock market reaction to losses by national soccer teams. In monthly terms, the excess returns associated with a soccer loss exceed 7%. This paper provides a fundamental analysis of football sentiment: what is it in relation to investor sentiment, how to measure and interpret the results etc. Moreover, it constitutes the legitimacy of studying football sentiment as an important variable in investigating investor mood and its influence on markets. The other papers on similar issues confirm the results of the previous study. For example, Ashton, Gerrard and Hudson (2003) looked at the relation between the performance of the England football team and subsequent daily changes in the FTSE 100 index. The results suggested that win and loss in the game by the England national team is followed by positive and negative market returns, respectively.

The papers that are described above used a national team games and the following stock market index as a subject of investigation of football sentiment. Nevertheless, there exists a variety of different instruments that can be used as a football sentiment. Some studies are focused not only on national-level games but on the club-level, too. For example, the study of Eker, Berument and Dogan (2007) took three most popular Turkish clubs: Fenerbahce, Galatasaray and Besiktas and put the game outcomes of these teams in UEFA Champions League into the model to measure its effect on the exchange rate of the Turkish lira against the U.S. dollar. The results showed that the win of each of examined Turkish football team against foreign rivals in corresponding European cup significantly increases the exchange rate depreciation of the Turkish lira against the U.S. dollar. Another study of Berument, Ceylan, and Onar (2013) used the same data: performance of three most popular Turkish clubs (Fenerbahce, Galatasaray and Besiktas) in UEFA Champions League and how it influences the individual's risk perception in the form of BIST 100 returns. Authors used specific EGARCH model for testing their hypothesis. The results of this study suggested that wins led to higher asset returns and lower risk aversion on the following business day of the Borsa Istanbul, while loss or tie is followed by lower returns and higher risk aversion. The paper by Fung, Demir, Lau and Chan (2015) used Edmans, Garcнa and Norli (2007) approach to test the different variations of football sentiment. With a sample of international matches of Turkey National team and UEFA Champions League games of the same three most popular Turkish clubs, the authors tried to investigate how football sentiment varies with changing data/variables. They came up to a conclusion that when changing win/loss dummies with unexpected win/loss variables, removing Monday matches, dropping sports-related firms, and sorting portfolio returns by market capitalization and past returns, the sports-sentiment effect disappeared. However, a significant negative loss effect was observed as well as the evidence that sporting events have a larger impact on stock return volatility for firms with smaller market capitalization and lower past returns.

As one can see, the studies concerning football sentiment all came up to conclusion of the significant presence of football sentiment in the asset return model. However, a very few attention was paid to the firm-level analysis: most of all studies related to the football sentiment examined the national team's effect on the market indices with only further inclusion of the limited number of clubs. Nevertheless, the area of football sentiment relation to asset pricing is worth mentioning in the academic environment. The economic logic of relation between football sentiment and the firm-level asset returns is the following: given the well-known fact that the football team supporters come from the city, where the club is based and the fact, that there is a bias in the form of investor's local stock preferences while forming portfolio (Coval and Moskowitz (1999)) we have a strong connection between local investors, local companies and local football teams in form of behavioral effects. While the behavioral effect from non-local investors can be explained by the identification of the club performance with the company performance.

The remarkable paper in the related subject was made by Chang, Chen, Robin, Chou and Lin (2012). The authors tried to implement a firm-level analysis of sport sentiment by examining the returns of the locally headquartered Nasdaq firm with behavioral sentiment in the form of NFL team from the same geographic area performance. They showed that a loss of the local NFL team is followed by significantly lower next-day stock returns for locally headquartered firms. Their results suggested that the outcomes of local sports teams influence investor sentiment, which significantly affects the returns of localized trading stocks. Another finding indicated that the negative effects of local team losses on local firms' returns are significantly stronger for small firms.

By taken the previous paper as an inspiration and a fundamental core for current work, I want to expand the knowledge of local football sentiment in the following ways. First of all, the previous study was taken in America and with a different kind of sport. Due to regional specificity, it's hard to aggregate its results on the different areas; in U.S. football (soccer) is not even a leading sport event. That's why the author took NFL for their sport sentiment. By moving the actions to the European region and implementing football as a tool, I will expand the knowledge of local football sentiment in that region. Secondly, the study of Chang, Chen, Robin, Chou and Lin (2012) uses a simple Fama-French 3 Factor model, which is not really appropriate to use in the different European regions due to the lack of data and inconsistency. That's why in this study I will use another modified approach in order to be consistent with the region. Thirdly, no other study tried to bring light on the different effects of behavioral biases in firm-specific asset pricing in the different countries. This study will examine 6 countries, in which half of them are developed and the other one is developing and tries to measure the difference in the effects.

2. Data

The empirical part of this paper consists of two main types of data: financial and football. Financial data consists of firm-specific market data and market-specific market data, dependent on the implemented model. Firm-specific market data contains daily stock data: the open/close price and the volume of trading. Market-specific data includes the everyday value of the market index of a particular country. All of this information was collected from the open sources: finance.yahoo.com, stooq.com (Polish analogue of finance.yahoo), tradingview.com, investing.com). Football data is the detailed provision of football match outcome. It was collected from the open sources, too: football-data.co.uk/data.php (for national championships), sport.niv.ru/archives/archives.pl (for European Cups). From these sites, one can download raw data, which consists of non-numerical components (team names, the name of division, etc.) as well as numerical ones (score of the match, date of the match). This raw data was transformed into the dummy variable format, which is applicable for further empirical investigation.

Six countries were chosen for the empirical research in this paper: England, France, Germany, Russia, Turkey, Poland. The rationale for choosing these countries is the following. Firstly, for this research one needs to have both developed and developing countries. England, France, Germany are the main contributors of The MSCI Europe Index, which constitutes their economic dominance in the region, while Russia, Turkey, Poland are the main contributors of The MSCI Emerging Markets Europe Index and that is the evidence of their leading positions among developing countries in the region. Choice of the leaders of developed and developing countries in this research has several important features. It allows to speak about the aggregation of results: if some effects on stock prices were found to be significant in areas that make up from 60% to 89% of a broad index, then one can constitute this result to be significant for the whole area, too. Moreover, the chosen countries are also presented in the highest competition table based on UEFA Rankings: England, Germany and France are placed in the 2^nd, 4^th and 5^th place, respectively, while Russia, Turkey and Poland stand on 6^th, 10^th and 25^th place, respectively. The mix of developed financial institutions and high popularity of football (which also reflects in the success on the European Cup arena) in presented countries allows to interpret results of the research without considerable constraints.

As it was mentioned earlier one of the key features of this study is to analyze the effect that local football clubs cause on local companies' stocks. So, one should create a link between a firm and a football club that played in the local area. For those reasons the following step-by-step methodology was implemented:

1. For each country select 100 companies of both types - with small and big capitalization. The selection proceeds through the record of companies listed on each country's national stock exchange and sorted by the value of their market capitalization.

2. After creating the relevant lists of companies - sort them randomly and then proceed with a one-by-one evaluation of company's headquarters city location using open sources like Bloomberg, Thompson Reuters, official sites and others.

3. One should choose the company that, firstly, headquartered in the city of observed country and, secondly, this city has a football club that was playing in the highest national football division during the observed period. The time period is 2012-2018, as it provides the most actual data and the amount of observations within that period is enough to make a study.

4. Proceed with this step-by-step companies' selection until the number of companies will reach 21 or all of the companies from the list of 100 would end.

After the described manipulations, the following results were obtained (see Table 1).

As one can see, the sample mean of daily observations per big cap company is different in comparison to the one of small cap company. A simple two-sample t-test for difference of means shows that the sample means are different at 1% significance level. The possible explanation of such difference is that big cap stocks are usually referred to long living companies, that are stable during the long periods of time. Contrary, small cap stocks referred to volatile companies, that are not stable on long time periods. Some companies from the sample list were eliminated or excluded from the national stock exchange trading before the end of observing period, while some of small cap companies were only originated in the observed period (after 2012).

It should also be noticed that the sample mean of daily observations per big cap company is different in comparison to the one of small cap company. A simple two-sample t-test for difference of means shows that the sample means are different at 1% significance level. The possible explanation of such difference is that big cap stocks are usually referred to long living companies, that are stable during the long periods of time. Contrary, small cap stocks referred to volatile companies, that are not stable on long time periods. Some companies from the sample list were eliminated or excluded from the national stock exchange trading before the end of observing period, while some of small cap companies were only originated in the observed period (after 2012).

After the sample of companies was created, one now has the links between the company and the city that should be transformed into the company - football club relationship. This relationship is described in Table 2. It can be noticed that the distribution of companies between cities is not equal: most companies locate their headquarters in the capitals of their countries. This is a reasonable outcome since capitals usually acquire most of the economic activity in the region, while also providing gains in terms of easier access to necessary operations. In this sample, 76% of all companies are concentrated in 21% of cities (capitals).

Once the link between company and city was created, the next step would be the construction of the company-football club relationship. The sample data of such relationships is described in Table 3. Following the numbers from the table, the same trend can be found among football clubs' distribution between cities. 47% of football clubs in the sample are located in 21% of cities, which is also can be explained with economic reasoning: capitals have more population that is ready to support the club and more facilities to allow basement of several clubs within one city.

It is also important to mention why there is a separation between the competition in which each club is participating. The point is that, firstly, the format of the competition is different: in national championships, each team play with each other during the whole football season (usually from August till May), so the amount of games is much more comparing to the European Cups. UEFA Europa League and UEFA Champions league has the different format: they both consists of qualification rounds (up to 8 games), 6 games in the group stage and maximum 7 games in the knockout stage (in Europa League - up to 9 games). Most clubs don't play such amount of games, since some of them skip the qualification round (while some of them fail to go through qualifications), many of them failed to pass the group stage (the club should earn either first or second place in the group in order to go to knockout stage) and other are being knocked out in the final stage of tournament so only two teams are reaching the final. Secondly, the European Cups are attracting much more audience all around the world, so the emotional effect of games in that competition is assumed to have much more effect on people since it's a very prestigious tournament to play. For example, the English Premier League, the most viewed national football competition in the world, has the peak audience of 3.24 million viewers in the season 2017/2018, while UEFA Champions League knockout stage in season 2015/2016 generates global audiences of between 112 and 200 million viewers during live match broadcasts. It is reasonable to control the difference between signals from both types of competition due to its potential difference in `emotional power'. It can be also mentioned that UEFA Europa League and UEFA Champions league are counted apart. That's because of different level of prestige of such tournaments: in UEFA Champions League there are teams that earn the highest places in their national divisions, while UEFA Europa League carries teams that were just behind national division's triumphant.

3. Empirical framework

The current study takes the main idea of implementing a firm-level analysis of sport sentiment from Chang, Chen, Chou and Lin (2012). Similarly, the current study introduces a link between the company and the sport club in order to examine the behavioural effect of sport matches on the financial market area. However, the current study uses another econometric model for testing hypotheses. Chang, Chen, Chou and Lin (2012) first regress stock returns on the Fama-French three factors, a momentum factor, lag stock and market returns, and Monday and January dummies. After collecting the residuals, they then regress them on the win and loss dummies and test the significances of the dummy variables. However, the current study will use another model for several reasons:

1) Fama-French 3 or 5 factor model proved its consistency and relevancy in U.S. stock market; however, it doesn't perform on the same level in the European area (see Bauer, Cosemans and Schotman (2010)).

2) For this study it was crucial to use the model which can work reasonably well in 6 different European countries and two different types of firms within them. It was also one of the reasons to choose less complex model.

3) The construction of data for Fama-French factors is quite complex and requires a lot of skillful work that alone will be enough for standalone study.

Given all of these three reasons, it was decided to use modified Edmans (2007) approach, that satisfies all three reasons above: relevancy, generalization among different economic areas and relatively easy to implement. It is also worth mentioning that even Chang, Chen, Chou and Lin (2012) model was partially based on Edmans (2007) approach.

The modified Edmans (2007) model that was implemented for the current study has the following structure. An econometric approach tests the null hypothesis that the company's stock returns are unaffected by the outcome of the local football team game. That hypothesis, if true, would constitute the rationality of investors and the inability of behavioural factors like football sentiment to influence the investment decision, that will put another evidence of market efficiency. The alternative hypothesis is that the outcome of matches with local football team leads to some market reaction - the win of local team leads to positive mood of investor, while loss of local team leads to the opposite results, thus the state of mood is translated into market investment decisions - which brings a controversial evidence of investors' rationality and efficiency of the markets, meaning that investment decision can be affected by behavioral factors.

Initially, it's reasonable to assume that `win' in the match of local football team is associated with the following good investor's mood, while `lose' result with a bad one. However, this cannot be easily translated into financial markets, as the psychology literature suggests that the mood sentiment effect can be absolutely different: Hirshleifer and Shumway (2003) findings state that positive investor mood has a positive impact on stock returns, while Guven (2009) lab evidence tells the opposite story: positive mood induces people to behave more cautiously and avoid risk in order to preserve their good emotional states and thus some studies suggested that good mood leads to the decreasing stock returns on markets. So, one of the additional goals of the study is to check which of psychological approaches are consistent with firm-specific football sentiment outcome.

One important feature, that Edmans (2007) mentioned in his model is that “under the null hypothesis, soccer outcomes are uncorrelated with asset prices. This, in turn, implies that the effects of soccer should be consistently estimated with any model of stock returns--even one that is completely misspecified”. This notion proves the legitimacy of the chosen model.

The two-step econometric approach starts with the estimation of the following regression (1):

where is an index of firms, is the stock return on day t, is the lag variable for stock return, is a set of daily dummy variables for the days of the week: Monday, Tuesday, Wednesday, and Thursday (so that ) in order to control for daily effect, is the national stock market index return on day t, is the lag variable of the return of national stock market index. The return of the national index was included to control for the correlation between individual stock value and the market index portfolio value attributed to systematic risk that is well documented in the literature. The lagged stock return, , is included to account for first-order serial correlation. From equation (1) for each company we extract estimated residuals, which represent abnormal returns that should be the results from football-sentiment effects. If we denote to be residuals from (1), the estimation of the effect of the outcome of local football clubs' matches can be done using the following regression model (2):

where is a set of multiple dummy variables of football matches' outcome. The outcomes of football matches were transformed to be actual for day t, however, in fact, they were usually played in day t-1 or t-2. The point is that football matches that are playing during the working days started in the evening, when all stock markets are closed, so it's assumed that the mood effect from match outcome is translating on the next day, too. For football matches that are played in the weekends the mood effect is translated on Monday trading day.

The reason of such multiple differentiation in is the following. As it was mentioned before, the division between national competitions and European cups are necessary because of different status of these matches and the power of sentiment that can be different, too (due to more audience attention to European cups). The division inside European cup rounds was implemented to control the effect on different rounds, since the goals of teams vary according to their class. For some biggest clubs like, for example, Bayern Munich the only goal in UEFA Champions League is to win this cup, so probably local fans won't pay much attention on group stage of competition (assuming, that it should pass this stage easily), while demanding very high results in the final stage of competition. Controversially, for some smaller clubs, for example, Legia Warsaw even the fact of participation in group stage of UEFA Champions League will earn maximum attention of local fans because the probability of pass it is quite low. One other important feature is the inclusion of SUCCESS/FAIL variables. Due to specific rules on final stages of European cups the teams are playing two matches in each round and passing the round depends on two matches' aggregation result. It could lead to the situation that team can lose second match in the round (which is a `bad' sentiment) or just simply play draw (which is `neutral' sentiment), while at the same time pass the round due to higher aggregation result (which is a `good' sentiment). The inclusion of such types of variables will control described outcomes.

All in all, the following two-stage model will be implemented on 12 datasets: 6 countries and each consists of big cap stocks and small cap stocks. According to the findings of Banz (1981) one could expect more broaden and powerful football sentiment effect among small cap stocks comparing to big cap stocks. It is also expected to find evidence of bigger behavioral effect on stocks of developing countries, since the amount of trading volumes on their stock exchange is less than in developed countries and their stocks are concentrated more in local investors' portfolios.

4. Empirical results

The two-step panel data regression model predicts the significant football dummy variables in 4 datasets out of 12. The descriptive interpretation of results is provided:

In Table 4 one could see the results for Big Cap Germany companies. As the theoretical probability to find significance variables is higher among European Cups games (because they have fewer games in the sample and attracts a lot of attention), for this dataset it was chosen to drop games in national championships (as they provide very poor values of p-value) and concentrated only on Champions League and Europa League. As a result, the only significant variable here is WOL_CL_GRL - Wolfsburg loses its match on the group stage of UEFA Champions League. The coefficient before the dummy variable is negative, so the outcome is consistent with Hirshleifer and Shumway (2003) findings. It is the only significant variable that was found in Big Cap companies' datasets.

From Table 5 one could see a 5% significance for LEGIA_L - Legia Warsaw loses its match in Ekstrakslasa (Poland national division). The significance of variables of national divisions outcome has more confidence in this statistical outcome since there are a lot more matches in the sample. Given the fact that most of the Polish firms are concentrated in Warsaw and hence are assumed to be dependent from Legia football sentiment, the result suggests that for most firms in Poland in the current study sample the `lose' of Legia in Ekstraklasa is followed by negative returns in the nearest trading day, which is also consistent with Hirshleifer and Shumway (2003).

The dataset of Small Cap Turkey companies provides the biggest number of significant variables (Table 6) - 6 football dummy variables were found to be statistically significant at 5-10% significance level. Their description is the following:

- BESIKTAS_W - Besiktas wins its match in Super Lig (Turkey national division)

- FENER_W - Fenerbahce wins its match in Super Lig

- FENER_L - Fenerbahce loses its match in Super Lig

- GALAT_L - Galatasaray loses its match in Super Lig

- KASIM_W - Kasimpasa wins its match in Super Lig

- BUY_CL_L - Istanbul BB loses its match in UEFA Champions League

As one can see there are 5 clubs in significance zone at which four of them are proved to be remarkable on national division's level. These results are consistent with previous findings in Borsa Istanbul that suggested that football sentiment based on `Big 3' Turkish football clubs (Galatasaray, Fenerbahce, Besiktas) can be useful in determining various financial variables. However, the analysis of coefficient before football club dummies brings controversial results: `win'/`lose' football match outcome can be associated with both positive and negative market reaction. The interpretation of such evidence is difficult, one possible explanation is the differentiation among `fans' groups (i.e one specific type of fans supports Besiktas, other - Fenerbahce) but it's the only way to test such hypothesis is experimental study.

The most significant variable in this study was found in Small Cap Germany companies' dataset (Table 7). BAYERN_L - Bayern Munich loses its match in Bundesliga (Germany national division) - is significant at 1% significance level. The effect of current sentiment is so strong that even if one omits all other variables the coefficient and significance won't change (Table 8).

Overall results are rejected the previously stated null hypothesis that company's stock returns are unaffected by the outcome of the local football team game. There is clear statistical evidence that outcome of matches with local football team leads to some market reaction - the win or loss of local team translates into market investment decisions through the changes in investors' mood. However, it's impossible to formulate the general effect of win or loss of football team on stocks' return: the result of the study suggests that football sentiment effect can be absolutely opposite. Nevertheless, even the psychological literature couldn't give a unique answer, so both positive or negative stock reactions can be explained by findings of either Hirshleifer and Shumway (2003) or Guven (2009).

The study also proved the expected differentiation of sentiment effect between big cap and small cap companies: only one variable was found to be significant among all big cap companies' datasets and its effect is obviously less than the one observed in small cap stocks.

The results support the initial hypothesis that bigger football sentiment effect should be found in stocks of developing countries (6 out of 8 football clubs which are proved to be significant are from developing countries). Again, it can be explained by the fact that the amount of trading volumes on their stock exchange is less than in developed countries and their stocks are concentrated more in local investors' portfolios.

It is also worth mentioning one important feature from the results. 5 out of 8 football clubs which were found to be significant are so-called `national leaders' in terms of support. Bayern is the most popular club in Germany, Legia is the most supported club in Poland, while `Big 3' Turkish clubs are the most popular clubs in Turkey with the greatest fan base in the country. It means that there is more possibility to find football sentiment among clubs that concentrates the most supporters inside the country.

5. Critical review and discussion

One of the most popular critique of empirical studies are concerning the sample. This paper is not an exception, so there are several ways that can help to improve this study. First of all, in observing the aggregative results about differences in developed and developing countries one can fairly noticed that this study is limited in its countries' sample. More accurate results require more broaden analysis in increasing set of European countries. Also one could expect to mention that more companies can be added to the sample to improve study. Moreover, an addition of new companies can somehow control the geographical bias that most of firms in this study is concentrated in the country's capital which is followed that some football clubs are linked to much greater amount of firms, while other football clubs sometimes have only one company in their observations, which in some cases can't be enough to make a confident statistical conclusion about their football sentiment effect (see Table 2). Another important implication about potential expansion of the sample is the case of European Cups. The point is that some big clubs are consistently taking part in them every year, so the time period of 2012-2018 is enough to construct good set of European Cups' dummies. However, some smaller clubs are played in such tournaments only once or twice during the observed period, so the number of observations for them can be too small to conclude the presence/absence of local football sentiment. With increasing time period and the following inclusion of European Cups' observations for some countries one could expect the results to be more significant than there are now, while for the others - to see the smoothening of sentiment effect and its further disappearing.

Even though this study tries to expand the existing knowledge of football sentiment, there is still a room for further expertise. One of the possible way for further research is to implement analysis of `unexpected' football results which theoretically can have stronger investor mood implication. The most suitable tool for such analysis is the inclusion of betting odds in the model. However, the study of Fung, Demir, Lau and Chan (2015) suggests that the unexpected outcomes have no impact on firm excess returns. Nevertheless, the study uses different techniques of linking football club performance with stock returns, so maybe for such model it could work. Another area of potential expansion is to somehow control for the region specificity. From the Table 1 and 2 one could see that, for example, Germany has the most football clubs in the sample, because companies are headquartered all around the country, while at the same time all British firms in the sample are concentrated only in London. The regional effect could also play important role in explaining investor mood effect.