The FIFA World Cup Effect on Stock Markets
As one of the world’s most watched sporting events, the Fédération Internationale de Football Association (FIFA) World Cup is a global soccer competition between nations from around the world. In fact, the FIFA World Cup attracts more than three billion viewers.[1] While not very popular in America, soccer, especially during the World Cup, attracts large portions of viewers from several countries, especially when their own nation’s team competes in this event. Many of these viewers are investors who react emotionally to match outcomes, affecting stock markets . This phenomenon is known as the “World Cup effect.” For example, an investor may react optimistically whenever his/her country wins a World Cup match, but also may react pessimistically whenever his/her country loses. The financial behaviors exhibited during the World Cup may be useful to current (and potential) investors who wish to understand the effects that the World Cup impose on markets.
My study expands on recent literature that investigate sports sentiment and market returns. My main contribution focuses on investor sentiment (also called market sentiment) from the 2014 and 2018 FIFA World Cups using stock market index returns of participating countries. Stock market index returns provide a mesure for analyzing investor sentiment of individual countries. While extensive evidence, which I review below, show that soccer results have a significant effect on investor mood, I plan to verify these tendencies using more recent data.
Since market sentiment affects asset prices in the short run, I hypothesize that, on average, a win during the World Cup generates significant positive returns for that country, while a loss generates significant negative returns for that country. Furthermore, since a draw in the World Cup often decreases the probability of a country advancing into the elimination stage more than if they had won the match, I hypothesize that a tie generates significant negative returns for that country, although not as severe than if that country had lost the match. I also predict more extreme returns for subsequent World Cup match rounds.
The remainder of the paper is structured as follows. Section I begins with a brief literature review regarding previous studies that investigate the link between soccer and market returns. In Section II, I explain the development of my hypotheses and models. Section III describes the data and their statistical assumptions. Sections IV and V present and evaluate the empirical results. Section VI summarizes my findings and concludes.
I. Literature Review
Several literary studies suggest sporting events influence stock markets through market sentiment. In fact, in 2007, Edmans, Garcia, & Norli investigated the effect of investor sentiment on stock returns in their study, “Sports Sentiment and Stock Returns.” Using international soccer results from 1973 through 2004, they documented a strong stock market decline after losses, but found no significant relationship between wins and stock market inclines. The authors believed the asymmetric tournament format explained the insignificance for returns after wins. That is, a win simply advances a country into the next stage, while a loss results in complete elimination from the tournament. Thus, I may expect index returns after losses to be more extreme than index returns after wins.
Motivated by the previous work mentioned above, Kaplanski & Levy (2010) expanded on the study, offering a way to exploit the U.S. market, implying market inefficiency. Although soccer is not very popular in the United States, the authors claimed that a large portion of the U.S. market includes foreign investors and believed the sentiment effect is not limited to local markets, thus making the U.S. market an accurate benchmark for analyzing this hypothesis. Adopting a dynamic time-series model similar to that found in Edmans et al. (2007), they found an average negative return over the World Cup’s effect period. Although I do not aggregate my returns, I may expect an average negative daily return for all days during the World Cup. To ensure the market is in fact exploitable, they tested for spurious correlation, and employed several methods to verify spurious correlation did not account for their results.
For example, they tested for the possibility that one of the World Cups occurred in an exceptionally bad year for markets. In my study, I capture this effect through my variable, year. I also employ a chow test to determine if the returns from each World Cup differ significantly. These authors also suspected a seasonal effect occurring in June and July to influence returns but added a dummy variable to control for this. Similar to my study, I include dummy variables controlling for day of the week to capture any calendar effect, but later find that days of the week have no effect on index returns.[2] Ultimately, they advised investors to leave the market and invest in 3-Month Treasury Bills during active World Cup days. While my study does not focus on an exploitation strategy, my results may support their recommendation.
Similar to Kaplanski & Levy (2010), Palomino, Reneboog, & Zhang (2009) discussed market efficiency through betting odds. As a proxy for match outcome expectations, the authors predicted betting odds would provide an excellent measure for predicting soccer match results. Using returns of soccer clubs listed on the London Stock Exchange, they concluded betting odds remarkably predicted match outcomes; however, investors ignore these odds likely due to their lack of information salience. Moreover, these authors claimed these betting odds contained new information to investors.
Furthermore, Castellani, Pattitoni, & Patuelli (2015) elaborated on the notion of efficient markets. Similar to Palomino et al. (2009), they found that unexpected events generated more extreme market reactions than expected events. This suggests that any significant results I generate in my study are likely due to upsets, or unexpected results. Unlike the previous literature mentioned, they investigated further into match outcomes, and also found that wins corresponded to positive market reactions while losses corresponded to negative market reactions. They also determined a significant relationship between competition type (Champions League, UEFA Cup, etc.) and market returns. While my study does not focus on competition type, I later find significant differences in stock index returns associated with World Cup round type.[3]
II. Development of Testable Hypotheses
My dependent variable includes participating countries’ major stock index returns (either raw returns as a percentage or normalized returns as a percentage) for days during the 2014 and 2018 World Cups. Using stock index returns provide an accurate measure to determine the country’s overall market return on a specific day.
I categorize my independent variables into match outcome dummy variables, which list every possible World Cup match outcome, disregarding quarterfinal and consolation matches; date dummy variables, which include each weekday excluding Friday as well as, year, which indicates the World Cup; and lastly, global returns, which is the average stock index return for all countries in the dataset on day, t.[4]
Table 3 includes the expected sign for each coefficient. For each match outcome variable, the coefficient sign reflects the expectation corresponding to my hypotheses. Each expectation indicates expected returns for that match outcome against returns during a day where the country does not play a match (omitted). Also, as the calendar effect suggests, Fridays (omitted) exhibit particularly large returns compared to other days of the week, so I predict negative coefficients for each day dummy variable. Lastly, I expect the variable, year, to be negative, since average returns were less in 2018 than in 2014 (omitted).[5]
Table 4 presents each of my four models. For each of these models, I normalize my estimated regressions, either by setting my dependent variable as normalized returns or including global returns as an independent variable. Models 1 and 3 include global returns as an independent variable; however, Model 3 also includes day indicators. Conversely, Models 2 and 4 include normalized returns as a dependent variable, but Model 4 also includes day indicators.
Although I have 65 distinct matches with wins and 54 with losses, I only have 16 distinct days in which one country won and only 31 distinct days where one country lost, compared to 46 days I analyze stock index returns. With that said, I have a relatively small sample size. Furthermore, my model likely suffers from omitted variable bias. For example, markets may be driven by the same motivation that causes countries to perform well in the World Cup. My model also does not capture any foreign investment. Many countries included in the dataset have global stock markets, thus stock index returns may not fully represent overall investor emotion for a particular country. For example, if the United States loses a World Cup match, I assume United States’ investors will react pessimistically. However, due to the large foreign holdings in the New York Stock Exchange (NYSE), these foreign investors likely react indifferently after a United States’ loss; therefore, NYSE Composite returns may not fully reflect the attitude of United States’ investors.[6]
III. Data and Statistical Assumptions
I limit my sample to only include stock market returns for days during the 2014 and 2018 World Cups. My data correspond to the economic variables proposed by the theory, since each variable includes all possible World Cup match outcomes, as well as returns for stock indices of various countries. I compiled cross-sectional data from Investing.com, a financial portal that provides information and data about global financial markets, which I consider a reliable source. I also retrieved World Cup results from FIFA.com.
Table 1 shows each country and its index I use in my dataset. For the majority of the observations, index returns reflect the next day’s returns after the match has been played, unless a country’s market is still open after the match has ended, which in this case, I use returns for the same day the match has been played. Unfortunately, data were not available for all participating World Cup countries. If data were unavailable or significantly sparse for days during the World Cup, I did not include those countries in my dataset, which may lead to sampling error. Table 2 lists these countries.
Table 5 summarizes the sample statistics. For all days during the 2014 and 2018 World Cups, there is approximately a -0.052 percent actual daily market index return. Returns can also be volatile during this time, shown by the wide the range of returns. Half of the data for actual returns lie between -0.48 percent and 0.42 percent, and the data are slightly negatively skewed.[7]
Tables 6 and 7 present the summarized statistics when controlling for match outcome as well as the year, respectively. On average, daily stock index returns are less for wins than losses, which refutes my hypothesis; however, stock index returns are far more volatile after wins. Furthermore, daily stock index returns do not differ much between the 2014 and 2018 FIFA World Cups (-0.053 percent and -0.050 percent, respectively).
Since my dataset contains more observations of matches with wins than losses, this may suggest that losing countries are likely developing countries. To explain, developing countries likely do not have established stock exchanges; therefore, stock index returns would not exist. In my case, since there are more wins than losses, I can assume that the some losing countries are likely developing countries.
IV. General Empirical Results
Table 8 presents estimated regression results for Models 1, 2, 3, and 4. When comparing Models 2 and 4, both the adjusted R² and the Aikake Information Criterion (AIC) suggest Model 2 has greater explanatory power. Similarly, when comparing Models 1 and 3, Model 1 has greater explanatory power, shown by the greater adjusted R² and lower AIC. My model explains more variability of the response data than Edmans et al. (2007), who generated an adjusted R² of 0.15.
The Ramsey Regression Equation Specification Error Test (RESET) test results are shown on Table 9. Unfortunately, I reject the null hypothesis for all models and acknowledge my models suffer from misspecification but cannot attribute a specific reason to misspecification. I cannot strongly support the results I generate from my models.
Although all my models are incorrectly specified, for theoretical and statistical reasons, I consider Model 1 as my best model. While Models 1 and 3 provide the best fit and mostly support theoretical expectations, I conclude that the day variables are not jointly significant, a major reason I consider Model 1 as my best model.[8]
While prior literature does not evaluate stock index returns at each specific World Cup round, generally, my results correspond to those found by Edmans et al. (2007) and Castellani et al. (2015), who each report positive coefficients for wins and negative coefficients for losses. However, Edmans et al. (2007) also record negative normalized returns for group stage wins.[9]
Table 10 presents tests relating to the classical assumptions: The White and Breusch-Pagan-Godfrey (BPG) Tests for heteroskedasticity, and tests for multicollinearity. Although I fail to reject the null for the White test, I reject the null for all BPG tests, and conclude possible heteroskedasticity for Model 1.[10] Using White corrected standard errors to correct for heteroskedasticity, both group stage wins and semifinal wins lose their 5 percent significance; however, the variable, final losses, becomes significant at the 1 percent level for Models 1 and 2, and significant at the 5 percent level for Models 3 and 4. Lastly, since all Pearson correlation coefficients are less than 0.8 and all variance inflation factors (VIFs) are less than 5, I do not consider multicollinearity an issue.
Table 11 presents the results after eliminating outliers. When eliminating outliers from my dataset, I generate significant coefficients for group stage wins, round of 16 wins, round of 16 losses, and semifinal losses. Coefficients for semifinal wins and final wins become zero, implying all observations pertaining to these dummy variables are outliers. Moreover, the semifinal loss coefficient becomes positive, and the year coefficients becomes negative. Although outliers significantly change my results, I lack strong theoretical evidence to omit outliers; therefore, further interpretation of Model 1 will include outliers.
V. Interpretation of the Empirical Results
When considering Model 1, as expected, all coefficients estimate negative stock index returns after losses. A country should expect fewer returns after losing World Cup matches than if they had not played a World Cup match that day. Furthermore, coefficients pertaining to winning elimination matches estimate greater returns than if a country had not played that day. However, group stage wins estimate a -0.33 percent fewer daily stock index return than days when a country does not play. This result is still negative when disregarding outliers. Although disconcerting, this finding is consistent with Edmans et al. (2007), who also documented -3.4 basis points for normalized stock returns after group stage wins. This coefficient may be explained for a couple of reasons, First, a win in the group stage does not always warrant promotion into the elimination stage. For instance, a team that has already been eliminated from the World Cup may still win a group stage game, but investors likely will not react optimistically after this win. Another reason for this unexpected result may be due to trading volume. As people disengage from the stock market to focus on watching soccer, decreases in demand often lead to stock index declines. As mentioned earlier, many countries in my dataset have global markets, thus later in the tournament when fewer teams remain, although one country may still be active, investors from other eliminated countries likely focus on trading again (and may even trade in an active country) instead of watching soccer. This likely increases the trading volume, thus the demand for stocks. This effect follows a logarithmic distribution, where eliminated countries diminish marginally. I also generate one-tailed significance for round of 16 wins and (10 percent significance) and final losses (1 percent significance) and conclude there is a relationship between these match outcomes and stock index returns.[11]
I estimate negative coefficients for my day variables, as expected. These results imply returns are less during these days compared to Friday. Interestingly, I obtain a positive coefficient for my year variable, when I expected a negative coefficient.
Table 12 presents various statistical tests regarding my dataset. As I explained earlier, I found no joint significance between the day variables in my model, given a 0.01 F-statistic. This suggests that the calendar effect does not pose any significant implications towards stock index returns during these World Cups. Since many World Cup matches are played on weekends, Mondays’ returns likely capture investor sentiment from weekend matches. Additionally, in line with Palomino et al. (2009), who conclude statistically significant positive market reactions following wins, I conclude wins are jointly significant, given the F-statistic (3.39); however, Palomino et al. (2009) also conclude significant negative reactions following losses, while I cannot. Furthermore, ties provide no significance in determining stock market returns.[12]
While prior literature does not focus on specific rounds in the World Cup, I conclude that round of 16 wins generate greater stock index returns than group stage wins.[13] I attribute this result likely to negative returns exhibited after group stage wins versus positive returns for round of 16 wins. Other than this result, I conclude that subsequent rounds do not result in more extreme returns as I hypothesized.
Lastly, since I include data from both the 2014 and 2018 World Cups, I predicted returns during the 2014 World Cup would differ from returns during the 2018 World Cup. Performing a Chow Test, I generate an F-statistic, 1.64.[14] Thus, I fail to reject the null hypothesis, and conclude a structural break does not exist in my data. In other words, returns do not differ between the two World Cups. Figure 1 graphically depicts this result. While global returns were lower during 2018 than 2014, I conclude no significant difference during these time periods. I support this conclusion, given the year variable is close to zero. This may suggest that, in 2018, cumulative negative returns primarily came from periods not during the World Cup, since cumulative returns were positive during the 2014 World Cup (0.89 percent).[15]
VI. Summary and Conclusion
Expanding on prior research, I analyze the effect of investor sentiment on stock index returns from the two most recent FIFA World Cups. Generally, my hypotheses match the results. As my model suggests, positive stock index returns are expected following all matches ending in a win, except for group stage wins. Likewise, for all matches ending in a loss, negative index returns are expected. However, Edmans et al. (2007) also documented negative returns for World Cup group stage wins. These negative returns for group stage wins may be attributed to the asymmetric format of the World Cup or even trading volume decreases. Furthermore, stock index returns are not consistent with World Cup match round. Unexpectedly, semifinal wins are expected to generate the highest stock index returns, while group stage wins generate the lowest stock index returns. However, there are major limitations, which may produce these perplexing results, such as sample size and omitted variables. Unfortunately, my model is not very robust, given the change in coefficient characteristics after dropping outliers.
Interestingly, the average daily stock index return for days during the World Cup is negative. Although the purpose of my research is not to provide a trading strategy, these results may be noticeable to investors. Consistent with negative average returns found by Kaplanski & Levy (2010), who recommend being out of the market during the World Cup period, investors may be able to exploit the market during the World Cup if overall average returns continue to be negative during this event.
[1] According to an audit, FIFA reported a record 3.572 billion people who watched the 2018 World Cup in Russia.
[2] Also referred to as the day-of-the-week effect, the weekend effect, or the Monday effect, the calendar effect theorizes that certain months and days exhibit larger returns than others.
[3] Refer to footnote 13.
[4] Due to the structure of the FIFA World Cup, quarterfinal and consolation matches are played on either Friday evenings or Saturdays, times when markets are closed; therefore, data is unavailable.
[5] Returns for the FTSE All-World Index were 2.2 percent in 2014 versus -11.3 percent in 2018.
[6] Kaplanski & Levy (2010) report value of foreign holdings of U.S. corporate equity securities.
[7] Disregard high skewness values for match dummies. The frequency of these matches decline, thus explaining the increased skewness.
[8] Generating an F-Statistic of 0.01, I fail to reject the null hypothesis, and conclude day variables are not jointly significant. See Table 12.
[9] Edmans et al. (2007) only distinguish group stage games from elimination games.
[10] For White test, X 2,30 (24.08) with p-value, 0.7683. BPG Tests 1 and 2 are evaluated against X 2,1 and BPG Test 3 is evaluated against X 2,2.
[11] P-value for one-tailed t-tests for round of 16 wins and final losses are 0.07 and 0.001, respectively.
[12] T-statistic (0.57) implies group stage matches do not significantly differ from zero.
[13] T-statistic is 1.80 (5 percent significant) for difference in returns between round of 16 wins and group stage wins.
[14] F-Statistic (1.64) < F-critical (1.75) using df = 12 from Table B-2.
[15] Refer to footnote 5. See https://www.investing.com/indices/ftse-all-world-historical-data for historical FTSE All-World Index data.
Castellani, M., Pattitoni, P., & Patuelli, R. (2015). Abnormal Returns of Soccer Teams: Reassessing the Informational Value of Betting Odds. Journal of Sports Economics, 16(7), 735–759.
Edmans, A., Garcia, D., & Norli, O. (2007). Sports Sentiment and Stock Returns. Journal of Finance, 62(4), 1967–1998.
FIFA.com. (n.d.). FIFA. Retrieved from https://www.fifa.com/
Kaplanski, G., & Levy, H. (2010). Exploitable Predictable Irrationality: The FIFA World Cup Effect on the U.S. Stock Market. Journal of Financial and Quantitative Analysis, 45(2), 535–553.
Palomino, F., Renneboog, L., & Zhang, C. (2009). Information Salience, Investor Sentiment, and Stock Returns: The Case of British Soccer Betting. Journal of Corporate Finance, 15(3), 368–387.
World Indices. (n.d.). Retrieved from https://www.investing.com/indices/world-indices
