Gambling Cultures and Stock Price Volatility: A Cross-Country Analysis

Transcription

1 Gambling Cultures and Stock Price Volatility: A Cross-Country Analysis By Benjamin M. Blau a Abstract In this study, we test the hypothesis that favorable attitudes towards gambling in a particular country will increase the risk tolerance of firms and subsequently affect the level of volatility in stock prices. Using a clever research design that holds constant the structure of financial markets, we find that countries with more gaming institutions, higher gambling losses per adult, and legalized online gambling have less stable stock prices. These results are robust to different measures of volatility and controls for both firm-specific characteristics and macroeconomic conditions. To make stronger causal inferences, we use the adoption of legal online gambling in France as a natural experiment and test whether the volatility of stock prices is affected by changes in this law. Consistent with the idea that gambling attitudes lead to greater volatility, we find that the volatility of French stocks increases (relative to the volatility of non-french stocks) surrounding this event. a Blau is an Associate Professor in the Department of Economics and Finance in the Huntsman School of Business at Utah State University Old Main Hill, Logan Utah, ben.blau@usu.edu. Phone:

2 1. INTRODUCTION A growing body of both theoretical and empirical research suggests that gambling preferences by investors can influence asset prices in financial markets (Scheinkman and Xiong, 2003; Hong, Scheinkman, and Xiong, 2006; Barberis and Huang, 2008; Kumar, Page, and Spalt, 2011; Blau, Bowles, and Whitby, 2015). Shiller (2000) argues that increases in gambling activity affect culture and can change attitudes towards risk-taking (pp. 41), which will likely to influence investment in the stock market. Anecdotally, Shiller (2000) writes that from 1929 to 1933, volatility in the stock market, which was at some of the highest levels in U.S. history, corresponded with a period he termed as the gambling craze, suggesting that attitudes toward gambling are likely to be associated with higher levels of volatility in financial markets. Several explanations for this association exist. One relevant explanation is that in cultures with favorable attitudes toward gambling, managers of firms are likely to take greater risks, which could result in less stable stock prices. Another explanation could be linked to findings in Dorn and Sengmueller (2009) that show that investors with strong gambling preferences tend to trade stocks at a remarkable rate. Several studies document a positive relationship between trading activity and volatility (Schwert, 1989; Gallant, Rossi, and Tauchen, 1992; Karpoff, 1987; and Jones, Kaul, and Lipson, 1994). Perhaps excessive trading by investors in places with strong gambling cultures contributes to the link between gambling activity and volatility. Following these ideas, this study formally tests the assertion in Shiller (2000) that gambling attitudes and activity will lead to greater volatility in financial markets. In particular, we test whether or not countries with stronger gambling cultures have less stable stock prices. Identifying determinants of market volatility has important implications for several reasons. First, part of the foundation of financial economics is based on the idea of efficient 1

3 financial markets (Fama, 1970). Beginning with Shiller (1979, 1981), a number of studies show that stock prices are more volatile than expected according to standard asset pricing models (LeRoy and Porter, 1981; Pesando, 1979; Amsler, 1980; Singleton, 1980; Grossman and Shiller, 1981; Blanchard and Watson, 1982; and Flavin, 1983). Testing whether or not gambling activity/attitudes leads to higher volatility in stock prices presents an attempt to identify, in part, factors that lead to excess volatility. Further, our tests deviate away from more traditional explanations of volatility by examining how culture influences the stability of stock prices. The results from our tests can contribute to the general understanding of factors that influence the efficiency of markets. Second, the implications from our tests may contribute more broadly to our understanding of economic activity. Endogenous growth theory suggests that economic output is a function of aggregate capital stock in the economy. However, the level of capital depends on the level of financial development (Goldsmith (1969), McKinnon, (1973), Shaw (1973), and Pagano (1993)). In capital market equilibrium, gross savings in the economy will equal gross investment, which will lead to higher levels of aggregate capital. When financial markets are excessively volatile, the fraction of gross savings that flows into gross investment might decrease since risk-averse savers are less inclined to invest in capital markets. In this case, unusual volatility in financial markets might reduce gross investment, aggregate capital stock, and ultimately, economic output. Therefore, finding a link between gambling activity and the instability of asset prices has important implications that relate to economic growth. To properly test our hypothesis that countries with stronger gambling cultures will experience greater volatility in financial markets, we must account for the possibility that the structure of these markets in a particular country is endogenously determined, somehow, by the 2

4 gambling culture in the country. To overcome this possibility, we use a clever research design that holds constant the structure of the financial market while testing for the association between gambling activity and volatility. In particular, we examine the volatility of American Depositary Receipts (ADRs), which are securities that are traded on U.S. stock exchanges but represent shares of foreign countries. This research design allows us to condition the volatility of ADRs on the level of gambling activity and the favorability of gambling attitudes in the ADR home country while holding the market structure constant. We are not the first to use this research design. For example, Eleswarapu and Venkataraman (2006) find that the strength of political and legal institutions in an ADR home country leads to greater ADR liquidity in U.S. markets. We measure the strength of the gambling culture in an ADR home country in three ways. First, we gather data on the number of gaming institutions and calculate the ratio of gaming institutions to population (in billions). Second, we obtain the amount of gambling losses per adult in countries that were ranked in the top 10 in the world. Third, we determine whether or not a particular home country allows for legalized online gambling. Results from our univariate tests show a strong and positive relationship between these three measures of gambling culture in the home country and the volatility of ADRs. These results are robust to various measures of volatility. Of the three measures, gambling losses per adult in the home country seems to drive the direct relation between gambling culture and volatility. The findings from our univariate tests are robust to a number of different multivariate tests. Holding constant a variety of ADR-specific and home country-specific characteristics, results from these tests show gambling attitudes and activity in the home country is directly associated with the volatility of ADRs. The results from our multivariate tests are not only statistically significant, but the findings are also economically meaningful. For instance, a unit 3

5 increase in the number of gaming institutions in the home country is associated with an increase in volatility that ranges from 5.4% to 7.9% depending on how volatility is measured. When examining the relationship between gambling losses per adult in the home country and the volatility of ADRs, we find that a unit increase in gambling losses results in an increase in volatility of over 20%. Home countries that allow legal online gambling have ADR volatility that is about 8% higher than home countries that do not allow legal online gambling. These results are consistent with our hypothesis and suggest that countries with stronger gambling attitudes have less stable stock prices. Further, these results provide support for the assertion in Shiller (2000) that stronger gambling cultures can lead to greater risk-taking and, eventually, affect the level of volatility in financial markets. We recognize that observing a relationship between gambling activity/attitudes in ADR home countries and the volatility of ADRs is not tantamount to finding that stronger gambling cultures are causing greater volatility. Although unlikely, it is possible that greater volatility in ADRs causes greater gambling activity in the home country. In order to make stronger causal inferences, we examine the volatility of French ADRs surrounding the legalization of online gambling in France. Using this exogenous change in French law as a natural experiment seems to be an appropriate identification strategy. Results from univariate tests show that the legalization of online gambling in France led to a 25% increase in daily volatility in French ADRs. Using a difference-in-difference type approach, we examine relative volatility, or the difference between the volatility of French ADRs and the volatility of non-french ADRs, surrounding the legalization of online gambling. Perhaps more interesting than our first results, this second test shows that relative volatility increases more than 65% when moving from the pre-event period to the post-event period. Again, these 4

6 findings support the idea that causation flows from gambling activity/attitudes to volatility instead of the other way around. Recognizing that other factors may influence the level of volatility in French ADRs during both the pre- and post-event periods, we conduct a variety of multivariate tests. Again, we find that, after holding constant a number of ADR-specific characteristics, volatility for French ADRs increased substantially after online gambling was legalized in France. This is particularly true when comparing the volatility of French ADRs to the volatility of non-french ADRs and again indicates that strengthening the gambling culture in a particular country will cause an increase in volatility. Combined, the results from this study provide an important contribution to our understanding about how gambling cultures affect the volatility of stock prices. First, our findings provide support for Shiller s (2000) hypothesis. Second, our results contribute to the broad literature that attempts to identify factors that influence volatility (Black, 1976; Christie, 1982; Schwert, 1989; Chan and Lakonishok, 1995; Bekaert and Wu, 2000; and Wu, 2001). While this literature generally focuses on determinants of volatility that are more traditionally firm-specific, the results in our study suggest that more non-traditional factors, such as culture, influences the stability of stock prices. Third, as we have argued above, endogenous growth theory implies that volatility in capital markets can adversely influence economic growth. The rest of this paper follows. Section 2 provides a thorough discussion of the data used throughout the analysis. Section 3 presents our empirical tests and results. Finally, Section 4 provides some concluding remarks. 2. DATA DESCRIPTION 5

7 The data used in this analysis come from several common sources. First, we obtain the universe of ADRs from the Center for Research on Security Prices (CRSP) using share codes. We then cross check the ADRs with information from Bloomberg and find the home country for each ADR. From CRSP, we obtain share prices, stock returns, shares outstanding, trading volume, and bid-ask spreads from daily closing ask and bid prices. From this information, we are able to estimate market capitalization and Amihud s (2002) measure of illiquidity, which is the ratio of daily returns (in absolute value) scaled by volume (in 100,000s). We recognize the need to control for macroeconomic conditions in the ADR home countries so, from the World Bank, we obtain GDP per capita and unemployment rates. Finally, we obtain information regarding gambling attitudes from a number of different sources. From CasinoCity.com s worldwide gaming directory, we obtain the number of gaming institutions in each country. 1 From H2 Gambling Capital, we gather information about gambling losses per adult for the top 10 countries as of Finally, from a number of different sources that are cross-checked, we determine whether or not the ADR home country legally allowed online gambling as of The ADR and macroeconomic data is obtained for 2010 to 2012 and averaged across that time period so that our final sample consists of ADRs from 32 home countries. Table 1 provides a summary of the ADR home countries. We first list the number of ADRs for each home country and include the average volatility across ADRs. Throughout the analysis, we use three measures of volatility. First, we estimate the standard deviation of daily returns for ADRs from 2010 to The standard deviation represents the total volatility (Volt). Next, we estimate residual returns from the following four-factor model. Ri,t Rf,t = α + β1mrpt + β2smbt + β3hmlt + β4umdt + εi,t (1) 1 CasinoCity.com constantly updates the worldwide directory of gaming institutions so we are unable to determine the number of institutions as of 2010 but feel that this information, which was obtained in early 2014 is a good approximation of 2010 s number of gaming properties. 6

8 The dependent variable is the daily excess return (difference between the CRSP raw return and the daily risk-free rate) for each ADR i on day t. The independent variables include the Fama-French (1996) three factors (market risk premium MRP, small-minus-big risk factor SMB, and the high-minus-low risk factor HML) along with the up-minus-down factor. These factors are obtained from Wharton Research Data Services. After estimating this four-factor model for each ADR, we then calculate the standard deviation of residual returns (εi,t) in order to provide an approximation for idiosyncratic volatility (IdioVolt). For our third measure of volatility, we fit daily ADR returns to a Garch(1,1) model: σt 2 = ηvl + αµt βσt-1 2 (2) where VL is the long-run variance, σt-1 2 is the prior day s volatility of returns, and µt-1 2 is volatility of residual returns. To obtain the long-run forecasted variance σt 2, we estimate the following equation: σt 2 = ω + αµt βσt-1 2 (3) Here, we can estimate parameters for ω, α, and β and back out the long-run variance and the parameter η since η = 1 α β and ω= ηvl. The square root of this third measure of volatility is denoted as Garch(1,1) volatility (Garch(1,1)). Table 1 also reports the ratio of the number of gaming institutions per country relative to the country s population in billions (Gaming), whether or not the country is listed among the top 10 countries in the world in gambling losses per adult (Top10), the amount of gambling losses per adult (GambLoss), and whether or not online gambling in legally allowed (Online). As seen in Table 1, both China and the U.K. have the largest number of ADRs (27) while several countries only have one ADR. The table reports the average ADR volatility for each country according to each of our measures of volatility. Further, the table reports our approximations for gambling attitudes. Column [5] shows that Australia has 7

9 the most gaming institutions (relative to population in billions) and the highest gambling losses per adult. Further, Australia has legalized online gambling. It appears that China has the least favorable attitude towards gambling as the number of gaming institutions is the lowest in the sample of ADR home countries and China does not allow online gambling. We also note that in the last row of Table 1, we report the results for the entire sample. 3. EMPIRICAL TESTS AND RESULTS In this section of the paper, we present some results from univariate tests. We then report the findings from our multivariate analysis, where we control for ADR specific characteristics and macroeconomic variables. Here, we test whether or not favorable attitudes towards gambling in the ADR home country affects the volatility of ADRs. Finally, we report some robustness tests where we use the organization of a French gambling commission that legalized online gambling in France as a natural experiment. Using this change as our identification strategy, we will be able to make stronger causal inferences regarding ADR volatility and gambling activity. 3.1 Univariate Tests In this subsection, we conduct some univariate tests to determine the relationship between gambling attitudes and our three measures of ADR volatility. Table 2 reports the correlation matrix for the following variables: the number of gaming institutions relative to the home country population in billions (Gaming); the gambling losses per adult for those countries that are listed in the top 10 zero otherwise (GambLoss); the total ADR volatility (Volt); the idiosyncratic volatility (IdioVolt); and Garch(1,1) volatility (Garch(1,1)). We also report p- values in brackets underneath each of the correlation coefficients. As seen in Table 2, we first find that Gaming and GambLoss are highly correlated (correlation = , p-value = <.0001). Second, we show in the first row that Gaming is directly correlated with Volt, IdioVolt, and 8

10 Garch(1,1). Each of these correlation coefficients are significant at the 1% level according to the reported p-values. These findings suggest that the number of gaming institutions in the home country is associated with higher volatility in the ADR regardless of how volatility is constructed. Third, when focusing on the correlation between GambLoss and our measures of volatility, we again find positive and significant correlations. Further, these coefficients are larger in magnitude than the corresponding coefficients in the first row suggesting that, if anything, the direct relation between GambLoss and ADR volatility is stronger than the direct relation between Gaming and ADR volatility. We also note that our measures of volatility are highly correlated, which is expected given their construction. Table 3 provides some additional univariate tests. Here, we examine mean volatility across discrete variables that attempt to capture the favorability of gambling attitudes in ADR home countries. Panel A reports mean volatility for ADRs with home countries ranked in the top 10 in gambling losses per adult (Top10). Each of our measures of volatility produce qualitatively similar results, so for brevity, we only discuss our findings in column [1]. In home countries that are ranked in the top 10, we find that the average ADR has total volatility of In unranked home countries, the average ADR has total volatility of only The difference, which is reported in the third row, is statistically different from zero (t-statistic = 4.74). Further, this difference is economically meaningful as a top 10 ranking is associated with nearly 44% higher volatility. Panel B shows the results when we report mean ADR volatility for those home countries with legalized online gambling and those with laws restricting online gambling. Again, the results are similar across measures of volatility so we only discuss our findings in column [1]. The average ADR with home countries that allow online gambling has volatility of 2.77% 9

11 compared to the average ADR with home countries with online gambling restrictions of 2.36%. In the third row of Panel B, the difference is 0.41% and is reliably different from zero (t-statistic = 2.37). Further, this difference is economically significant as ADRs with home countries with legal online gambling have volatility that is more than 17% higher than ADRs with home countries that do not. Taken together, results from our univariate tests in Tables 2 and 3 suggest that a stronger gambling culture in the home country is an important determinant, or at least an important correlate, of ADR volatility. 3.2 Multivariate Tests The results from the previous section suggest that the level of home country gambling activity/activity is directly associated with the volatility of ADRs. We recognize, however, that other factors may be influencing our results. In this section, we attempt to control for both ADRspecific and country-specific characteristics in a number of different multivariate tests. We begin by estimating the following equation using cross-sectional data made up of ADRs. VOLATILITYi,j= β0 + β1gamingj+ β2ln(gdp/cap)j+ β3ln(unemp)j+ β4nasdi + β5ln(mktcap)i,j + β6ln(price)i,j+ β7turni,j+ β8spreadi,j+ β9ln(illiq)i,j + εi,j (4) The dependent variable VOLATILITY is measured in different three ways. Ln(Volti,j) is the natural log of total volatility for ADR i in country j. Ln(IdioVolti,j) is the natural log of idiosyncratic volatility for ADR i in country j. Ln(Garch(1,1)i,j) is the natural log of Garch(1,1) volatility for ADR i in country j. We include as control variables the following: ln(gdp) is the natural log of GDP per capita in country j; ln(unemp) is the natural log of the unemployment rate in percent in each country j; NASD is an indicator variable capturing whether ADR i is listed on NASDAQ zero otherwise; ln(mktcap) is the natural log of average market capitalization (from 2010 to 2012) for each ADR on the last trading day of the year; ln(price) is the natural log of the average closing price (from 2010 to 2012) for each stock at the end of the 10

12 each year; Turn is the ratio of total trading volume scaled by the shares outstanding; Spread is the relative bid-ask spread in percent for each ADR; ln(illiq) is the natural log of Amihud s (2002) measure of illiquidity, which is the ratio of the absolute value of daily returns to trading volume (in 100,000s). The independent variable of interest is Gaming, which is the ratio of the number of gaming institutions to country population (in billions). To the extent that levels of home country gambling affects the volatility of ADRs, we expect a positive and significant coefficient on Gaming. We report t-statistics that are obtained from standard errors that are clustered across ADRs. Results from estimating equation (4) are reported in Table 4. The format of this table, and the three that follow, report the results for each of our measures of volatility. We also show the results from two specifications. First, we only include the home country characteristics as independent variables. Second, we present the results for the entire specification. We note, however, that in unreported results, we estimate a variety of different specifications including different combinations of control variables and find the results to be qualitatively similar those reported in this study. Columns [1] and [2] present the results when the dependent variable is Ln(Volt). We find in both columns that the estimate for Ln(GDP) is negative and reliably different from zero suggesting that home countries with lower GDP per capita have higher ADR volatility. Focusing on the other control variables in column [2], we find that lower priced ADRs with higher turnover, and larger bid-ask spreads tend to be more volatile. After controlling for these variables, we find that Gaming produces a positive and significant coefficient (estimate , t-statistic = 2.38). Qualitatively similar results are found in column [1]. Further, these findings are economically meaningful as a unit increase in Gaming is associated with 7.3% increase in 11

13 ADR volatility. The economic magnitude of these results is even greater in column [1]. These findings support our results from our earlier univariate tests and suggest that the level of gambling in the ADR home country is associated with higher levels of volatility. Columns [3] through [6] present the results when we look at idiosyncratic volatility and Garch(1,1) volatility as dependent variables. In general, the control variable produce estimates that are similar in sign and magnitude to the corresponding coefficients in columns [1] and [2]. There are a few differences however. First, we find that the indicator variable NASD produces a positive estimate that is reliably different than zero in columns [4] and [6] suggesting that NASDAQ listed stocks have higher volatility than stocks listed on other exchanges such as the NYSE. Second, we do not find that share turnover produces reliable coefficients in either columns [4] or [6]. Despite these differences, the coefficient on Gaming is positive and significant in each of the columns. In economic terms, results in columns [4] and [6] suggest that a unit increase in Gaming results in an increase in ADR volatility of 5.4% and 7.9%, respectively. These findings again indicate that the strength of the gambling culture in the home country is directly related to the volatility of ADRs. Next, we replicate our analysis but use a different proxy for favorable gambling attitudes in the ADR home country. In particular, we estimate the following equation. VOLATILITYi,j= β0 + β1gamblossj+ β2ln(gdp/cap)j+ β3ln(unemp)j+ β4nasdi + β5ln(mktcap)i,j + β6ln(price)i,j+ β7turni,j+ β8spreadi,j+ β9ln(illiq)i,j + εi,j (5) The dependent variable and independent variables are similar to those in equation (4). The only difference is that the independent variable of interest is GambLoss, which is the amount of gambling losses per adult in the ADR home country. Recall that GambLoss is equal to the amount of gambling losses only for those home countries that are ranked in the top 10 zero 12

14 otherwise. 2 As before, we estimate this cross-sectional regression using OLS and account for clustered standard errors. Table 5 presents the regression results. The control variables produce estimates that are very similar to the corresponding estimates in the previous table. When focusing on the variable of interest, we find that GambLoss provides positive and significant coefficients across columns. Focusing on the full specification in column [2], a unit increase in gambling losses is associated with a 23.1% increase in total ADR volatility. The economic magnitude of the estimate for GambLoss in columns [4] and [6] is similar. These results corroborate the findings in Tables 2 and 3 and again suggest that the level of gambling in the home country directly affects the stability of ADR prices. We recognize that the discontinuity in the variable GambLoss might present issues when trying to make inferences. Therefore, instead of including GambLoss as the independent variable of interest, we include an indicator variable Top10, which captures those home countries that are ranked in the top 10 in gambling losses per adult. In particular, we estimate the following equation using cross-sectional data and report the results in Table 6. VOLATILITYi,j= β0 + β1top10j+ β2ln(gdp/cap)j+ β3ln(unemp)j+ β4nasdi + β5ln(mktcap)i,j + β6ln(price)i,j+ β7turni,j+ β8spreadi,j+ β9ln(illiq)i,j + εi,j (6) The control variables again produce coefficients that are similar in sign and magnitude to those in the previous table. Consistent with our findings in Table 5, we find that home countries that are ranked in the top 10 in gambling losses per adult have more volatile ADRs than home countries that do not. In economic terms, a home country that is ranked in the top 10 has total ADR volatility that is nearly 17% higher than the ADR volatility of home countries that are not 2 H2 Gambling Capital only makes the Top 10 rankings publically available. The entire ranking is proprietary. 13

15 ranked in the top 10. Columns [4] and [6] suggest that ranked home countries have ADR volatility that is 19% and 15.9%, respectively, higher than home countries that are not ranked. In our final test in this subsection, we examine the effect of legalized online gambling on ADR volatility by estimating the following regression. VOLATILITYi,j= β0 + β1onlinej+ β2ln(gdp/cap)j+ β3ln(unemp)j+ β4nasdi + β5ln(mktcap)i,j + β6ln(price)i,j+ β7turni,j+ β8spreadi,j+ β9ln(illiq)i,j + εi,j (7) The dependent and independent variables are similar to those in previous tables. The only exception is the variable Online, which is equal to one if the ADR home country has legalized online gambling zero otherwise. The results from this analysis are presented in Table 7. Again, the control variables produce coefficients that are similar to corresponding coefficients in previous tables. Consistent with our expectation, the estimates for Online are positive and generally significant. We note, however, that the coefficients on Online are only significant at the 10% level in columns [2] and [4]. Given that we only have observations, significance at the 10% level is relatively reliable. Not only are these results statistically significant, but the coefficients are also economically meaningful. For instance, in column [2], the coefficients suggest that home countries that legally allow online gambling have ADRs with volatility that is approximately 8% higher than ADR volatility in home countries with restrictions on online gambling. Similar results are found in column [4] and [6]. Taken together, Tables 3 through 7 support our contention that the favorability of gambling attitudes affects the stability of stock prices. 3.3 The Effect of Legalizing Online Gambling on ADR Volatility In previous tests, we have attempted to control for the possibility that the financial market structure might be endogenously determined in some way by the favorability of gambling attitudes, which might influence the inferences that we are able to make about gambling and 14

16 volatility. Therefore, we have tried to hold constant the financial market structure and robustly examine the volatility of ADRs while conditioning on the level of gambling in the home country. It is still possible that our inferences regarding causation are misguided. Said differently, greater volatility in ADRs might somehow cause greater gambling activity in the home country. To account for this possibility, we set out to find an appropriate identification strategy. On May 12 th, 2010, France adopted the French Regulatory Authority for Online Games, which essentially legalized online gambling (ARJEL). 3 This change in the regulation of online gambling, which we argue is exogenous, might provide a nice natural experiment for our tests. In this final subsection, we examine the volatility of French ADRs surrounding this particular event in attempt to make stronger causal inferences about the relationship between volatility and gambling activity. To begin, we provide some univariate tests. In particular, we examine daily Garch(1,1) volatility for French ADRs for the 121-day period surrounding the adoption of ARJEL. 4 Not only do we examine volatility of French ADRs (Garch(1,1)FR), but we also use a type of difference-in-difference approach and examine the volatility of French ADRs relative to the volatility of non-french ADRs (Garch(1,1)FR - Garch(1,1)Non-FR). Table 8 reports these two variables in both the pre-arjel period and the post-arjel period. Focusing on column [1], we find that the volatility of French ADRs in the 60-day pre-event period was 2.77% compared to 3.45% during the 61-day post event period. This difference (0.68%) is statistically different from zero (t-statistic = 6.96) and suggests that after the legalization of online gambling in France, 3 The synonym ARJEL stands for the French translation of Regulatory Authority for Online Games. In French, this authority is called Autorite de Regulation des Jeux En Ligne. 4 The results reported below are not dependent upon this particular choice of time window. In unreported tests, we examined the 61-day period surrounding ARJEL and the 181-day period surrounding ARJEL and find qualitatively similar results. We did not want to extend the sample time period beyond 181 days for fear that confounding effects might influence the conclusions that we draw. 15

17 French ADRs had an increase in volatility. In economic terms, the increase in French ADR volatility represents nearly a 25% increase. The difference between French ADR volatility and non-french ADR volatility (Garch(1,1)FR - Garch(1,1)Non-FR) provides us with an opportunity to examine a treatment group of ADRs (French ADRs) and a control group of ADRs (non-french ADRs). In column [2], we find that this measure of relative volatility increases from the pre- to the post-event period, which is again consistent with our expectations. For instance, the relative volatility for French ADRs was.41% during the pre-arjel period and.68% during the post-arjel period. The difference (.27%), which represents a 65.9% increase in relative volatility, is statistically different than zero and economically meaningful. This simple univariate test seems to suggest that causation flows from gambling activity to volatility and not the other way around. We recognize the possibility that other factors might be influencing the level of French ADR volatility during this time period. Therefore, we attempt to control for these factors in a multivariate setting. Table 9 reports the results from estimating the following equation using pooled data that consists of ADRs across the 121-day period. Volatilityi,j,t or Relative Volatilityi,j,t = β0 + β1arjelt + β2nasdi + β3ln(mktcap)i,j,t + β4ln(price)i,j,t + β5turni,j,t + β6spreadi,j,t + β7ln(illiq)i,j,t + εi,j,t (8) The dependent variable in columns [1] and [2] is the natural log of Garch(1,1) volatility for ADR i in France on day t (Ln(Garch(1,1)i,j,t) FR ). The dependent variable in columns [3] and [4] is the difference in the natural log of Garch(1,1) volatility for ADR i in France on day t and the natural log of the average Garch(1,1) volatility for Non-French ADRs (Ln(Garch(1,1)i,j,t) FR - Ln(Garch(1,1)i,j,t) Non-FR ). We include as independent variables the following: ARJEL is an indicator variable capturing the post-arjel period (the 61 days on or after May 12 th, 2010) zero otherwise; NASD is an indicator variable capturing whether ADR i is listed on NASDAQ 16

18 zero otherwise; ln(mktcap) is the natural log of market capitalization for each ADR on each trading day of the sample time period; ln(price) is the natural log of the closing price for each stock at the end of the each day; Turn is the ratio of total trading volume scaled by shares outstanding for each day; Spread is the relative bid-ask spread in percent for each ADR on each day; ln(illiq) is the natural log of Amihud s (2002) illiquidity measure for each ADR on each day. We note that we do not include macroeconomic characteristics given that there is no variation in these variables during our sample time period. Following the format of previous tables, we report in the first column the results from including only the country-specific variables, which in this case is only the indicator variable ARJEL. The second column reports the results from the full specification. In column [1], we find that this univariate regression supports the results from our univariate tests in the previous table as the variable ARJEL produces a positive and significant estimate (estimate = , t-statistic = 9.81). In economic terms, the post-event period saw a 28.8% increase in daily volatility in French ADRs. Column [2] shows qualitatively similar results as the coefficient on ARJEL is (t-statistic = 15.28). Again, the economic magnitude of this coefficient is also statistically significant. Holding these other variables constant, the post-arjel period saw a 21.2% increase in French ADR volatility. We also note that during this period, NASDAQ exchange listing, share turnover, and bid-ask spreads were positively related to French ADR volatility while market cap, share prices, and Amihud s (2002) measure of illiquidity were negatively related to French ADR volatility. Next, we turn our attention towards the relative volatility of French ADRs (Ln(Garch(1,1)i,j,t) FR - Ln(Garch(1,1)i,j,t) Non-FR ). The univariate regression in column [3] shows that the post-event period saw an increase in relative volatility of nearly 13%. Column [4], which 17

19 reports the results from the full specification, shows that the magnitude of the coefficient on ARJEL decreases markedly. However, we still find that the estimate is positive and reliably different from zero (estimate = , t-statistic = 3.93). In economic terms, the results indicate that, holding the other independent variables equal, the post-arjel period saw an increase in French ADR volatility relative to non-french ADR volatility of 5.3%. These findings again support the idea that removing restrictions on online gambling, which we argue resulted in greater gambling activity in the ADR home country, increased the level of volatility in French ADRs relative to the volatility of non-french ADRs. 4. CONCLUSION A growing consensus in prior research suggests that volatility in financial markets is greater than expected according to traditional asset pricing models (Shiller, 1979; Singleton, 1980; Grossman and Shiller, 1981; LeRoy and Porter, 1981; Shiller, 1981; Blanchard and Watson, 1982; and Flavin, 1983). Given these results, a considerable portion of the literature has been devoted to identifying determinants, or at least correlates, of volatility (Black, 1976; Christie, 1982; Schwert, 1989; Chan and Lakonishok, 1995; Bekaert and Wu, 2000; and Wu, 2001). These studies, however, focus on firm-specific or macroeconomic characteristics. We deviate away from these more traditional determinants of volatility and test the hypothesis that cultures with favorable attitudes toward gambling will exhibit less stable stock prices. Shiller (2000) argues that gambling activity can influence our culture and change individual attitudes toward taking risk in other areas, such as financial decision making. To the extent that this is true, in places with stronger gambling cultures, firms may be less risk averse which might be reflected in greater volatility in stock prices. The main objective of this study is test this hypothesis. 18

20 Following Eleswarapu and Venkataraman (2006), we use a clever research design that accounts for the possibility that the structure of the market is endogenously determined by the strength of the gambling culture in a particular country. More specifically, we examine the volatility of ADRs while conditioning on the level of gambling activity/attitudes in the home country. Doing so allows us to hold constant the structure of the market while testing whether stronger gambling cultures determine the stability of stock prices. Results show a statistically significant and economically meaningful relationship between gambling activity/attitudes in the home country and ADR volatility. This relation is robust to a number of controls for ADRspecific and macroeconomic-specific characteristics as well of various measures of volatility. In economic terms, a unit increase in the number of gaming institutions in the home country is associated with an increase in volatility of around 7%. Similarly, a unit increase in the amount of gambling losses per adult in the home country results in an increase in ADR volatility of more than 20%. We also find that home countries that allow for legalized online gambling have ADR volatility that is approximately 8% higher than home countries that legally restrict online gambling. Recognizing the possibility that our causal inferences are misguided, we attempt to find an appropriate identification strategy. On May 12 th, 2010, France adopted a law allowing for legal online gambling. Using this change in French law, which we argue is exogenous in our tests, as a natural experiment, we examine the volatility of French ADRs. First, we find that the daily volatility of these ADRs increase approximately 25% during the period surrounding the event. Second, and perhaps more importantly, we use a difference-in-difference type approach and find that the volatility of French ADRs relative to the volatility of non-french ADRs 19

21 increases more than 60% during the post-event period. These findings support the idea that gambling activity/attitudes cause greater volatility. Combined, our findings contribute to the literature by (i) providing support for the assertion in Shiller (2000) and (ii) identifying a non-traditional determinant of volatility (gambling culture). Further, the results in this study have broad and important implications relating to economic growth. Endogenous growth theory suggesting that economic output is, in part, driven by the quality of financial markets. When capital markets are more volatile, riskaverse savers are less likely to invest, which can lower the level of aggregate capital stock and subsequently economic growth. In the framework of our study, countries with stronger gambling cultures may have a more difficult time converting gross savings to gross investment. 20

22 REFERENCES Amihud, Y., Illiquidity and Stock Returns: Cross-Section and Time-Series Effects. Journal of Financial Markets 5, Barberis, N., and M. Huang, Stocks as Lotteries: The Implications of Probability Weighting for Security Prices. American Economic Review 98, Bekaert, G., and G. Wu, Asymmetric Volatility and Risk in Equity Markets. Review of Financial Studies 13, Blanchard, O.J., and M.W. Watson, Bubbles, Rational Expectations, and Financial Markets. In Crises in the Economic and Financial Structure (ed. P. Wachtel). Lexington: Lexington Books. Blau, B.M., T.B. Bowles, R.J. Whitby, Gambling Preferences, Options Markets, and Volatility. Forthcoming, Journal of Financial and Quantitative Analysis. Black, F., Studies of Stock Price Volatility Changes, in: Proceedings of the 1976 Meetings of the Business and Economics Statistics Section, American Statistical Association, Chan, L.K.C., and J. Lakonishok, The Behavior of Stock Prices around Institutional Trades. Journal of Finance 50, Christie, A.A., The Stochastic Behavior of Common Stock Variances: Value, Leverage, and Interest Rate Effects. Journal of Financial Economics 10, Dorn, D., and P. Sengmueller, Trading as Entertainment. Management Science 55, Eleswarapu, V., and K. Venkataraman, The Impact of Legal and Political Instituions on Equity Trading Costs: A Cross-Country Analysis. Review of Financial Studies 19, Fama, E.F., Efficient Capital Markets: A Review of Theory and Empirical Work. Journal of Finance 25, Fama, E.F., and K.R. French, Multifactor Explanations of Asset Pricing Anomalies. Journal of Finance 51, Flavin, M.A., Excess Volatility in the Financial Markets: A Reassessment of the Empirical Evidence. Journal of Political Economy 91, Gallant, A.R., P.E. Possi, and G.E. Tauchen, Stock Prices and Volume. Review of Financial Studies 5, Goldsmith, R.W., Financial Structure and Development (Yale University Press. New Haven, CT). 21

23 Grossman, S.J., and R.J. Shiller, The Determinants of the Variability of Stock Market Prices. American Economic Review Papers and Proceedings 71, Hong, H., J.A. Scheinkman, and W. Xiong, Asset Float and Speculative Bubbles. Journal of Finance 61, Jones, C.M., G. Kaul, and M.L. Lipson, Transactions, Volume, and Volatility. Review of Financial Studies 7, Karpoff, J.M., The Relation between Price Changes and Trading Volume: A Survey. Journal of Financial and Quantitative Analysis 22, Kumar, A., J.K. Page, and O.G. Spalt, Religious Beliefs, Gambling Attitudes, and Financial Market Outcomes. Journal of Financial Economics 102, LeRoy, S.F., and R.D. Porter, The Present-Value Relation: Tests Based on Implied Variance Bounds. Econometrica 49, McKinnon, R.I., Money and Capital in Economic Development (Brookings Institution, Washington, DC). Pagano, M., Financial Markets and Growth. European Economic Review 37, Scheinkman, J.A., and W. Xiong, Overconfidence and Speculative Bubbles. Journal of Political Economy 111, Schwert, G.W., Why Does Stock Market Volatility Change over Time? Journal of Finance 44, Shiller, R.J., The Volatility of Long-Term Interest Rates and Expectations Models of the Term Structure. Journal of Political Economy 87, Shiller, R.J., Do Stock Prices Move Too Much to Be Justified by Subsequent Changes in Dividends? American Economic Review 71, Shiller, R.J., Irrational Exuberance. Princeton University Press, Princeton, NJ. Shaw, E.S., Financial Deepening in Economic Development (Oxford University Press, New York, NY). Singleton, K.J., Expectations Modes of the Term Structure and Implied Variance Bounds. Journal of Political Economy 88, Wu, G., The determinants of asymmetric volatility. Review of Financial Studies 14,

24 Table 1 Financial Market Volatility and Gambling Attitudes by ADR Home Country The table reports some summary statistics regarding the volatility of the ADRs in each home country and different proxies for gambling attitudes. Column [1] shows the number of ADRs included in the sample. Volt is the total volatility of the ADR. IdioVolt is the idiosyncratic volatility of each ADR. Garch(1,1) is the conditional expected volatility obtained from fitting daily ADR returns to a Garch(1,1) model. These measures of volatility are estimated using daily returns and average annual volatility from 2010 to Gaming is the ratio of the number of gaming institutions to country population (in billions). Top10 determines whether or not the ADR home country is listed in the top 10 countries with the most gambling losses per adult. GambLoss is the number of gambling losses per adult for the countries that ranked in the top 10. Online shows whether or not online gambling is allowed in the ADR home country. No. ADRs Volt IdioVolt Garch(1,1) Gaming Top10 GambLoss Online [1] [2] [3] [4] [5] [6] [7] [8] Australia ,288 Belgium No 0 Brazil No 0 No Chile No 0 No China No 0 No Denmark No 0 No Finland No France No 0 Germany No 0 Greece No Hungary No 0 India No 0 No Indonesia No 0 No Ireland Israel No 0 No Italy Japan No 0 No Luxemberg No 0 No Mexico No 0 No Netherlands No 0 No NewZealand No 0 Norway No Peru No 0 No Phillipines No 0 Portugal No 0 No Russia No 0 South Africa No 0 No South Korea No 0 No Spain Sweden No 0 No Switzerland No 0 U.K No 0 No All Obs N/A N/A 23

25 Table 2 Univariate Correlation The table reports Pearson correlation coefficients for two continuous proxies for gambling attitudes and our three measures of ADR volatility. Gaming is the ratio of the number of gaming institutions to country population (in billions). GambLoss is the number of gambling losses per adult for the ADR home countries that ranked in the top 10. Volt is the total volatility of the ADR. IdioVolt is the idiosyncratic volatility of each ADR. Garch(1,1) is the conditional expected volatility obtained from fitting daily ADR returns to a Garch(1,1) model. In brackets, we report p-values that provide the results from tests that coefficients are different than zero. Gaming GambLoss Volt IdioVolt Garch(1,1) [1] [2] [3] [4] [5] Gaming GambLoss Volt IdioVolt Garch(1,1) *** [<.0001] *** [0.0002] *** [<.0001] *** [0.0012] *** [<.0001] *** [<.0001] *** [0.0016] *** [<.0001] *** [<.0001] *** [<.0001]

26 Table 3 ADR Volatility and Gambling Attitudes This table reports volatility for countries with more and with less favorable attitudes towards gambling. Top10 is an indicator variable capturing whether or not the ADR home country is listed in the top 10 countries with the most gambling losses per adult. Online is an indicator variable that captures whether or not online gambling is allowed in the ADR home country. Volt is the total volatility of the ADR. IdioVolt is the idiosyncratic volatility of each ADR. Garch(1,1) is the conditional expected volatility obtained from fitting daily ADR returns to a Garch(1,1) model. At the bottom of each panel, we report the difference between volatility in ADRs in home countries with the most favorable attitudes towards gambling and volatility in ADRs in home countries with the least. We also report t-statistics testing whether the differences are statistically different from zero. Panel A. Gambling Losses Per Adult Volt IdioVolt Garch(1,1) [1] [2] [3] Top Non-Top Difference *** (4.74) Panel B. Online Gambling Online *** (4.46) *** (3.94) Non-Online Difference ** (2.37) ** (1.96) ** (2.27) 25