School Holidays and Stock Market Seasonality. March, Forthcoming: Financial Management. Lily Fang INSEAD

Transcription

1 School Holidays and Stock Market Seasonality March, 2017 Forthcoming: Financial Management Lily Fang INSEAD Chunmei Lin Erasmus University Rotterdam and Tinbergen Institute Yuping Shao National University of Singapore 1

2 Abstract Using school holiday data from 47 countries, we document a strong link between school holidays and market return: market-level return in the month after major school holidays is 0.6% to 1% lower than other months. This explains but is not limited to the phenomenal September effect : In the U.S., September is the only month that exhibited negative average return in the past century; but the school-holiday effect remains even with monthly fixed effects. We explore the explanation that the effect is due to investor inattention during school holidays, which slows the incorporation of (negative) information in security prices. 2

3 Introduction In this paper, we document a novel asset pricing pattern: globally, market-wide returns are 0.6% to 1% lower in the months after major school holidays than other times, and we present evidence that this seasonality in stock returns is attributable to investor inattention during school holidays resulting in the slow incorporation of news, especially that of negative news. It has been widely known among Wall Street traders that September the month after the summer school holiday has historically been the worst performing month. This September effect is striking in magnitude: For example, an article in the Wall Street Journal titled How to play the September effect (Sep 6, 2013) points out that since 1896, the Dow Jones average return for the month of September has been -1.09%, while that of all other months has been +0.75%. A more recent article on the same topic illustrates the persistence of the September effect: It is the only month that has a negative average return for 20, 50, and 100 years. 1 What is also noted by these practitioner discussions is that the origin of the September effect is not well understood: The aforementioned 2013 WSJ article is accompanies by the following subtitle it s well known that September has been the worst month, on average, for stocks. But no one has come up with a plausible explanation of why that is. We contribute to the understanding of the September effect a remarkable yet hitherto undocumented phenomenon in the academic literature in two ways. First, we conjecture and present evidence that the September effect reflects a broader after school-holiday effect whereby market returns after major (long duration) school holidays are significantly lower than other times. Second, we propose an explanation to the after school holiday effect, and hence also the September effect that has long puzzled traders. We hypothesize that returns after major school holidays are low because during school holidays, the market is collectively less attentive 1 Some stock strategists brace for September swoon, the Wall Street Journal, Sep 1,

4 to news and as a result, information is incorporated into prices slowly. This effect is particularly strong for negative information because taking advantage of negative news is more difficult and requires more attention, a scarce cognitive resource (Kahneman (1973)), than positive news. Miller (1977) and Diamond and Verrecchia (1987) provide theoretical foundation that constraints on short selling would impede the impounding of negative information into stock prices. In this paper, we present evidence that inattention among professional investors curtail their arbitrage activities, compounding the effects of short-sale constraints. Prior literature shows a pre-holiday return effect. Lakonishok and Smidt (1988) and Ariel (1990) examine the daily returns around public holiday and find significant positive return on days immediately preceding public holidays. The after school-holiday low return effect we document here differs from the pre-holiday effect in two ways. First, the nature of the holiday we examine is different. Not only is our school holiday periods longer (they are at least a week in duration) than the public, statutory holidays studied before, but another important distinction is that during school holidays, the markets remain open, as opposed to during public and statutory holidays when the market is closed. Since markets remain open during the school holidays we examine, any reduced trading cannot be explained by traders inability to trade; it has to be due to traders being less attentive and less likely to trade. The second difference is that while we find low after school-holiday returns, the previous literature finds high pre-holiday returns. Our effect is robust to controlling for pre-holiday returns. Our empirical analysis consists of two main parts. We first establish the empirical fact that there is a broader after school-holiday effect which encompasses the September effect. To this end, we collected school holiday data using school calendars from 47 countries around the world. We find a striking pattern: returns are on average 1% lower in the months after major 4

5 school holidays. Importantly, this lower return is not driven by September alone, even though the September effect is pervasive in the northern hemisphere: Even when September is excluded, there is still a return gap of 0.6% between after-school-holiday months and other times, and the difference remains highly significant. To ascertain what we document is indeed an after school holiday effect rather than a spurious result, we conduct a number of validation tests that exploit exogenous variations in school calendars. For example, in the US, while the school year starts in September for many states, some states in the south have school years that begin in early August. For stocks headquartered in these states, we find that it is August, rather than September, that exhibit particularly low returns. In France, while some major school holidays have common, nationwide dates, others are staggered by region in order to reduce tourist congestion. We expect and find that the after-school-holiday effect is stronger after nationwide school holidays than regional ones. Moreover, we find that one region s after-holiday dates do not predict low returns for another region (a placebo test), unless the predictor is the last region that goes on holiday. These tests make the implicit assumption that investors exhibit local bias: local stocks are heavily held by local investors. Thus it is a joint test of local bias and the after-school-holiday effect. Local bias among professional investors has been documented in prior literature. For example, Coval and Moskowitz (1999) show that U.S. investment managers exhibit a strong preference for locally headquartered firms. Bok, Kang and Kim (2010) show that local institutional investors have information advantage and thus their trading play an important role in incorporating the information of local firms. After establishing a robust after school holiday effect, the second element in our empirical analysis is to examine potential explanations of the phenomenon. We hypothesize that 5

6 it is due to investor inattention and the consequent slowdown in the incorporation of news into prices. During school holidays, many traders are away from the market on vacation or in summer homes. They pay less attention to the market and are less likely to trade. This means that information will be incorporated in prices more slowly than other times. Coupled with the fact that negative information is more difficult to take advantage of than positive information, it follows that the slow incorporation of news into prices is particularly severe for negative news, leading to lower returns in the month after major school holidays. To test this investor inattention hypothesis, we first show that investors trading activities are significantly reduced during school holidays. Trading volume is 7.8% lower during school holidays than other times. Even around the release of earning news, trading volume is 2.4% lower during school holidays than other times. This evidence is consistent with Hong and Yu (2009) that trading is light during the summer season. Since trading volume is a direct manifestation of investor attention and desire to trade, this evidence is consistent with investors being collectively less attentive during school holidays. Second, we show that financial markets react to news, and especially negative news, more slowly when news is released during school holidays. To gauge the speed at which prices incorporate news, we calculate the Delayed Response Ratio (DRR) as in Dellavigna and Pollet (2009), which measures the percentage of a stock s total price reaction in the 30 days after an earnings announcement that occurs after the first two days (i.e., the drift component). We find that this delayed response ratio is about 60% higher for negative news announced during school holidays than other times. In contrast, the delayed price response to positive news is not 6

7 significantly different during school holidays compared to other times. 2 Thus, market reaction to negative news slows down significantly if news is announced during school holidays. Third, we compare the after-school holiday returns among stocks sorted by news type and trading volume. We find that the low return is concentrated among stocks that release news during school holidays and, among these, it is concentrated among stocks that release negative news and have lower trading volume during school holidays. Finally, we document that the after-school-holiday effect is stronger among larger stocks and stocks with higher institutional holdings. This cross-sectional pattern is in sharp contrast to many anomalies documented in the literature, which tend to be concentrated among small, illiquid stocks with low institutional holdings (e.g. Keim (1983), Reinganum (1983)). Typically, impediments to trade small, illiquid stocks prevent arbitrage activities to eliminate the anomaly. Thus at first glance, the fact that the after-holiday effect is stronger among large, high institutional holding stocks is puzzling. But this finding is precisely consistent with the investor inattention hypothesis. Even though large stocks are liquid and easy to trade, shorting activities, which require significant monitoring and cognitive resources, can be reduced by investor s inattention. In other words, inattention can serve as a soft short-sale constraint. Indeed, it is plausible that this type of cognition-based constraint can apply to institutional investors as well as to retail investors, whereas transactions costs may always be lower for institutional investors (i.e., traditional short-sale constraints are less binding for them). Inattention, therefore, acts as an 2 Prior research shows that prices generally respond more slowly to negative news than positive news. For instance, Hong, Lim, and Stein (1999) show that negative news is more slowly reflected in prices than positive news. Fang and Yasuda (2014) show that the market takes longer to incorporate negative analyst revisions than positive ones. In an intraday setting, Busse and Green (2003) show that it takes more time for the market to fully reflect negative information than positive information. Our evidence here shows that the change in the speed of price response is different between positive and negative news. 7

8 alternative form of short sell constraint by reducing traders arbitrage activities and allowing for apparent price anomalies even among large, liquid stocks. Our paper makes a number of unique contributions to the literature. First, we document an asset pricing anomaly that has not been discussed in the academic literature but one that has long puzzled Wall Street professionals. Thus our empirical evidence is relevant to the academia and the practitioner world alike. Second, our work contributes to the literature on limited investor attention and its impact on asset returns. According to the seminal work of Kahneman (1973), attention is a scarce cognitive resource. A large literature has documented that investors are subject to limited attention (e.g. Barber and Odean, 2008; Gabaix et al., 2006; Hirshleifer and Teoh, 2003; Hirshleifer, et al, 2009; Hong and Stein, 1999; Hou and Moskowitz, 2005; Hou, Peng, and Xiong, 2008; Huberman and Regev, 2001; Peng and Xiong, 2006; Dellavigna and Pollet, 2009). But most existing evidence points to inattention among retail investors. Our examination of returns after school holidays provide evidence that investor inattention during school holidays is pervasive and affects professional traders and retail investors alike, which in turn affects asset returns. In this way, our paper joins the recent literature such as Goetzmann et al. (2015) that show that even institutional investors are subject to cognitive and behavioral biases. I. The After School Holiday Effect We collect US stock market data from CRSP for the period 1/ /2012. International stock return data for 47 countries are collected from MSCI Total Return Index and covers the period 1/ /2012. Table 1 reports the average stock returns by calendar month, for all 47 countries 8

9 in our sample. Out of the 47 countries, 37 are in the North Hemisphere. The first notable pattern is that the September column is dominated with negative average return figures. In fact, of the 37 North Hemisphere countries, 28 had average negative September returns. Looking at the global average, September is also the lowest return month with a negative return of -0.03%, while all other months have positive averages. Table 2 reports the average differences between each month s return and that of the other 11 months. Here an even clearer September effect emerges: This column is dominated by negative differences. Of the 37 North Hemisphere countries, only six exhibit a return difference between September and other months that is not negative: China, Columbia, India, Pakistan, Sri Lanka, and Turkey. Thus, the September effect is true for all Western and developed economies in the Northern Hemisphere. And the differences are almost always significant. Averaged across all countries, September returns are 1.62% lower than other months. We also note that August is also usually a low return month. As a comparison, January returns are on average 1.45% higher than other months. Thus, the September effect appears to be as strong as, if not stronger than the familiar January effect. 3 One obvious question is whether the September effect is just a statistical artifact. After all, among the 12 months in the year, there will be one month that has the lowest return. Maybe it is just by chance that it is September. If each month is equally likely to be the worst performing month, over the 86 year period for the US ( inclusive), we expect each month to be the worst performing one in about 7 years. In reality, September was the worst performing month for 12 years, the highest fraction and nearly twice as would be expected under the null. 3 Interestingly, January is not the best performing month. The data reveals a strong December effect (which may be called a Santa effect ): December is the month that shows the highest return, which is on average 1.99% higher than other months. In 24 of the 47 countries, the difference is significantly positive. 9

10 A related question is whether the September effect is driven by outliers. For example, some Septembers in recent memory were tumultuous times for the stock market, such as 2001, The worst five returns for September occurred in 1931 (-29%), 1937 (-13.5%), 1930 (- 12.4%), 1974 (-11%), and 1933 (-10.5%). Even after excluding all five, the September return is still on average 1% lower than other months, significant at the 5% level (t-stat 1.93). Table 3 examines the after-holiday returns in a regression setting using all the 47 countries. The baseline model in Panel A (column 1) shows that, controlling for year and country fixed effects, returns during the months after major school holidays is 1.1% lower than other month, with a t-stat of Even when we include month fixed effects (which severely biases against our hypothesis), the difference is still 0.6%, and highly significant (column 2). Column 3-6 divided the sample to two periods. It shows that the after holiday effect has decreases in recent years, yet remains statistically and economically significant. Next, we exclude September observations because we want to see that this is a robust result and not just driven by the pronounced September effect. Results in column 7 and 8 indicate that even after excluding September, the after-holiday month returns are still 0.6% lower than other times, with the result being significant at the 1% level. Thus, there is a strong after school-holiday effect and it is not just driven by September. Previous literature documents a positive return in the days before public holidays (Lakonishok and Smidt (1988) and Ariel (1990)). It is possible that what we document simply shows a mean reversion of the pre-holiday effect. To rule out this possibility, in Panel B, we repeat our baseline regression by using US daily returns and controlling for before and after public holidays using the Pre Public Holiday and Post Public Holiday dummies. This analysis is confined to the US data because we do not have public holiday information for all the 10

11 countries in our sample. Following Lakonishok and Smidt (1988) and Ariel (1990), the set of public holidays includes New Year's Day, Martin Luther King Day, Presidents' Days (formerly Washington's Birthday), Memorial Day, July Fourth, Labor Day, Columbus Day, Veterans Day, Thanksgiving, and Christmas. Consistent with Lakonishok and Smidt (1988) and Ariel (1990), we find that returns during the three days before public holidays are indeed positive and often significant, but returns in the three days after public holidays (the Post Public Holiday dummy) are insignificant. And importantly, our After-school-holiday indicator remains negative and significant, consistent with our baseline findings. In the US, the Labor Day (the first Monday in September) is typically associated with the formal conclusion of the summer season, especially in the north east part of the country. Public schools and universities start the new academic year after this holiday. If the weak market level return we document is related to school holidays, we should see that the so called September effect in the US is concentrated in the days after the Labor Day. To test this, for each year between 1962 and 2012, we identified the exact date of the Labor Day. In Table 4, we split September dates into before and after Labor Day, and test that the weak September return is concentrated after the Labor Day. Univariate results confirm this conjecture. In fact, the average daily returns for the days before Labor Day is positive. The average daily returns for days after the Labor Day are mostly negative and the differences are significant. Thus not all September days are equal when it comes to returns: The weak returns are concentrated after the Labor Day. II. Regional Analysis 11

12 To rule out the possibility that what we document above is a spurious result rather than being truly related to the holiday cycle, we conduct a number of validation tests that exploits exogenous variations in the school calendar. A. US South Although the school year starts in September for many states in the US, in some states in the south, school calendars are historically different and start in August owing to agricultural demands. 4 We collected actual school start dates for each state, and identified 6 states with an early August start: Georgia (Aug 7), Indiana (Aug 6), Nevada (Aug 2), Oklahoma (Aug 5), Tennessee (Aug 5), and Hawaii (Aug 5). If the September effect is truly related to the end of the summer holidays, then we would expect that for stocks from these states, the September effect would be an August effect instead. To test this, we classified companies by their head-quarter locations. Companies that are headquartered in one of the 6 states above are coded with a South indicator. We hypothesize that in a return regression, the interaction term between South and August would be significantly negative, while the interaction term between South and September would be weaker. This means that August, rather than September, is a particularly low return months for these companies. Note that this test makes the implicit assumption that investors exhibit local bias: stocks headquartered in one state are disproportionally held by investors from that state, who follow the local holiday calendar. But there is substantial prior evidence supporting the notion of local bias even among professional investors (e.g., Covel and Moskowitz (1999)). 4 The typical school year has 180 days, usually beginning in September and ending in June. But in some southern states, the school year begins in August and end in May, a legacy of the agrarian calendar that ensured students were out of school in time for the planting season. ( School in August gets low grade, WJS, Aug 19, 2012.) 12

13 Table 5 reports the results. In Panel A, we tabulate the average monthly returns for stocks headquartered in the six southern states. One immediate observation is August is the lowest value-weighted return month with an average return of -0.14%. Panel B presents the regression results using all US stocks. We note that September and August both exhibit significantly lower returns. Consistent with our conjecture, the interaction term between August and South is significantly negative, but the interaction term between September and South is not. The coefficient suggests that returns are lower than other times by 39 basis points on average, and the effect is significant at the 1% level. Thus for Southern state companies, low returns occur in August rather than September, as we expect. B. French school zones As a second validation test, we exploit exogenous school calendar variations in France. France has five major school holidays during a school year: Christmas, winter, spring, summer, and All-Saints Day. In order to reduce tourist congestion, France is divided into three regions that follow staggered holiday schedules for the winter and spring holidays, which take place in late Feb/early March, and late April/early May respectively. While the school zones have changed prior to 1993, it has remains the same since 1993 and is as depicted in Figure 1. All three regions, however, enjoy the same dates for the Christmas, summer holidays, and All-Saints holidays. For the staggered winter and spring holidays, Region C (which contains Paris, the French capital) goes on holiday first, followed by Region B, then Region A. This setting allows us to test a number of predictions. First, we expect the after-school holiday effect to be stronger for the three nationwide holidays than the regional holidays. Second, if we perform a placebo test which regresses one region s returns on the other two regions holiday schedule, we do not expect Region B or C s after holiday dummies to predict Region 13

14 A s returns because these regions go on holiday before A and their after school holiday dummies would not line up correctly with the actual after holiday dates for Region A. On the other hand, Region A s after school holiday indicator may predict the other two Regions returns because Region A is the last to go on holiday. Table 6 presents the French regional test results. For the purpose of this test, we collected specific school holiday dates for each region from 1993 to In Panel A, we focus on the difference between common school holidays and regional school holidays. The results indicate that the after-school-holiday effect is generally stronger after common holidays than regional holidays: for both zone A and C, the after common holiday indicator is negative and significant, and of larger magnitude and significance than the after regional school holiday indicator. Although for Zone B, while the regional school holiday indicator is significantly negative, the common school holiday indicator is not. Panel B presents the placebo tests where we regress one region s stock returns on another s after-school-holiday indicators. The results are as we expect: Neither zones B and C s after-school-holiday indicators are significantly negative in zone A s regression because both regions go on holiday before zone A, so the after-school-holiday indicator is not correctly picking up Zone A s after-school-holiday dates. But in zone B and C s regressions, the zone A after-school-holiday indicator is significantly negative. Overall the tests in this subsection provide strong support to the notion that the anomaly we document is truly an after schoolholiday effect. C. Chinese New Year Our third validation test utilizes the Chinese New Year (CNY). While this holiday is observed in a number of countries, it is often quite short (1-2 days in duration) and hence not in 14

15 our main school holiday data set (this is the case for Hong Kong and Singapore). However as the CNY is culturally the most important holiday in the Chinese tradition, even though the official holiday is short, businesses and individuals often unofficially take extra days off. Thus, if there is an after school holiday effect, we expect similar mechanisms to be at work and result in low returns after the Chinese New Year in regions where this festival is observed. 5 Table 7 presents the results. For this test, we collected market return data for China, Taiwan, Singapore, and Hong Kong. Results in this table strongly indicate that returns are particularly low after the CNY holiday in these countries/regions: it is lower than other times by an average of %. This effect is in fact larger and more significant than the other school holiday effect in these countries. We acknowledge that the CNY effect we documented here may be related to the cultural New Year effect documented in Bergsma and Jiang (2015), which shows that stock markets tend to outperform in days surrounding a cultural New Year due to positive holiday moods. The low post-cny returns could be a reversal of strong pre New Year returns. Nevertheless, the after other-school-holiday dummy remains significant in the regression, indicating that the broader after school-holiday effect is robust. Overall, the evidence in this section provide strong evidence that returns are low after major (school) holidays. This phenomenon is global, and is not just driven by September. In the next section we examine hypotheses that can potentially explain this phenomenon. III. Explaining the After Holiday Effect A. The Investor Inattention Hypothesis 5 In mainland China, the Chinese New Year holiday is accompanied by a market closure as well. 15

16 The investor inattention hypothesis maintains that the after-holiday low returns is due to investor inattention during holidays, which results in slow incorporation of information, particularly that of negative information. In order to test this hypothesis, we break it down into basic components. First, a necessary condition for the hypothesis to hold is that investors exhibit inattention during school holidays. Second, we need to show that it is the inattention to news, and particularly to negative news, that drives the low return after school holidays. To test that investors are indeed less attentive during school holidays, we follow prior literature and examine volume data. The investor inattention hypothesis predicts that overall market volume is expected to fall during holidays to reflect less intense trading. Second, investors would react less to news than normal days. Our basic analytical framework is the following panel regression model: (1) where the dependent variable is the natural log of the daily turnover. Holiday is the key indicator variable of interest and equals 1 for school holidays and zero otherwise. Firm-level control variables include the explanatory variables that affect trading activities documented in Chordia, Huh and Subrahmanyam (2007): RET+ as the monthly return of an individual stock if positive, and zero otherwise; RET as the monthly return if negative, and zero otherwise; book-to-market ratio (BM), log of a firm s book-to-market ratio; Logprice; SIZE, natural log of a firm s market capitalization of equity; AGE as natural log of the number of months since a firm was listing on an exchange; LogANA as the natural log of the number of analysts who follow a company; monthly forecast dispersion; earnings surprise (ESURP); portfolio betas (PBETA); leverage as the debt-to-asset ratio. Earnings surprise volatility (ESV) is the company s earnings surprise 16

17 standard deviation during the previous four years. Chordia, Huh and Subrahmanyam (2007) s framework is cross sectional in nature, while we have a panel dataset. Thus we further include year and firm fixed effect in the model, and standard errors are clustered by date. It is important to note that trading volume may be generally lower during school holidays not only due to investor inattention, but also due to lighter news flow. Therefore, in implementing the regression in Equation (1), we not only compare volume between school holidays and non-school holidays, but also do the same analysis conditioned on news flow. That is, we compare trading volume around earnings announcements during school-holidays and nonschool holidays. Table 8 shows the regression results. The first column reports the results for all trading days; the remaining three columns report the results for earnings announcement days only, separately for all news, positive news, and negative news. The school holiday indicator has a negative and significant coefficient in all columns. On average, turnovers are reduced by 7.8% during school holidays compared to non-holidays. This is consistent with Hong and Yu (2009) which reports that trading volume during the summer drops by about 9%. Conditioned on earnings news, we find that there is still a significant drop in trading volume by 2.4 % overall, 1.7% for positive earnings news and 2.9% for negative earnings news. Overall, our volume analysis provides support to the notion that the market is collectively less attentive during school holidays. To demonstrate the second element of the investor inattention hypothesis namely that it is investor inattention that contributes to the low after-holiday returns, we further decompose this into two testable predictions. First, we expect prices to react to news more slowly during school holidays than other times. This effect should be particularly strong for negative news because 17

18 negative news is more difficult to take advantage in general. Second, it should be the stocks that release negative news during school holidays, and especially those that investors are inattentive about those that are lighted traded during school holidays that contribute to the low after school-holiday returns. To test the first prediction that prices react to news, and especially negative news more slowly during school holidays, we calculate a Delayed Response Ratio (DRR) for earnings announcements and compare this between news released during holidays and those outside. Following Dellavigna and Pollet (2009), we calculate the DRR as follows: (2) where CAR stands for Cumulative Abnormal Returns. In other words, if firm i releases a piece of earnings news on date t, then the DRR is the ratio between the cumulative price response between day t+2 and t+30, and the total cumulative price response between day t and t+30; it is the fraction of the total response that is delayed (beyond the initial 2 days). We then regress these DRR measures on the school holiday indicator and control variables. The control variables include a Friday indicator since Dellavignal and Pollet (2009) shows a significant delay in price response to news released on Fridays, and also the magnitude of the earnings surprise. The results are reported in Table 9. The table shows that the school holiday indicator is significantly positive indicating more delayed response to news in the regression for negative earnings news, but not for positive news. The magnitude of the delay is large: The coefficient of for DRR[+2, +30] means delayed response is 56.7% higher for negative news released during holidays than other times, controlling for firm and time effects, and for Friday releases. DRR[+7, +28] yields similar results. Thus, the market s speed of responding to negative news is slowed down significantly 18

19 during school holidays, but such a slowdown is not observed for positive news, consistent with our hypothesis. Table 10 directly examines which type of stock contributes to the low after schoolholiday returns. In Panel A, we sort stocks by news type and tabulate their after school-holiday returns. This panel shows that the low after- school-holiday return is mainly attributable to news stocks: While news stocks have an average daily return of -0.02% in the 30 days after school holidays, no news stocks have an average daily return of +0.11%. Among news stocks, the negative return is stronger among negative news stocks than positive news stocks: The DGTWbenchmarked returns are -0.09% and -0.07% for the two groups, respectively, the difference being significant at the 5% level. In Panel B, we double sort stocks by news type and trading volume during school holidays and report their average returns in the 30 days after school holidays. Consistent with our expectation, we find that the low returns are concentrated not only among negative news stocks, but especially those that also experience low trading volume during school holidays. For instance, focusing on DGTW-benchmarked returns (the bottom panel), the daily returns for the negative news stocks with low trading during school holidays was % per day while it is 0 for negative news stocks but experience high trading volume during school holidays. Overall, results in this section shows that 1) trading volume is significantly reduced during school holidays; 2) market s reaction to negative news (but not positive news) is significantly slower during school holidays than other times; 3) stocks that release negative news during school holidays which also experience light trading exhibit the lowest returns in the after school-holiday period. This set of evidence provides strong support for the investor inattention hypothesis. 19

20 IV. Cross-sectional patterns of the after-holiday effect The previous subsection presented evidence consistent with the investor inattention hypothesis. But it still leaves open the following question: if this pricing pattern reflects market inefficiency, why is it not eliminated by profit motivated arbitrageurs? At least part of the answer, we believe, is precisely that professional traders are paying less attention to information release during holiday and therefore less likely to act on arbitrage opportunities. If this is the case, one implication is that the after school holiday effect may be particularly strong among larger stocks with higher institutional ownership. Since such stocks are liquid and easy to trade, such a prediction appears counter intuitive at first glance. But the key mechanism that drives the after school-holiday effect is investor inattention, rather than physical impediments to trade. Therefore as long as investors are inattentive, price anomalies can exist even for easy-to-trade stocks. In this way, inattention acts as a soft, rather than hard, trading constraint that prevents arbitrage to drive price efficiency. We test this prediction by examining the strength of the effect in different subsample of stocks sorted by size and institutional ownership. Table 11 presents the results. Panels A-D reports the after-holiday regression coefficient for portfolios of stocks sorted, by size, Amihud Illiquidity, institutional holdings, and analyst coverage, respectively. Panel A shows that among large stocks, the after-holiday returns are 80 basis points lower than other times. In contrast, the effect is not present for medium sized firms; for small firms, the result reverses: these stocks do not exhibit the after-holiday low return effect; instead, their returns are significantly higher after holidays than other times. The overall conclusion is that the after-holiday anomaly is stronger among large stocks than small stocks. Results in Panel B, sorting stocks on Amihud illiquidity 20

21 measure, is qualitatively the same: the effect is concentrated among liquid stocks. Results in Panels C and D are not significant, but they still reveal a monotonic decline in the magnitude of the after-holiday return effect from high analyst coverage and high institutional holdings, to low coverage and low holdings. The consistent picture that emerges from this cross-sectional analysis is that the afterschool-holiday effect is stronger among large, liquid stocks than among small, illiquid stocks. This pattern stands in sharp contrast to many anomalies documented in the empirical literature. Typically, return anomalies are strongest among small, opaque, illiquid stocks. The fact that the after-holiday low return anomaly is particularly strong among large liquid stocks suggests that the effect cannot be explained by conventional, physical impediments to trade. Instead, inattention, a cognitive constraint, can explain the lack of arbitrage activities even for large, liquid stocks. Conclusions In this paper, we document a strong asset pricing anomaly returns in the month after major school holidays are 0.6% to 1% lower than other time. This phenomenon is global. It is related to, but broader than, the September effect (that September has notoriously low returns historically) which is widely known among Wall Street traders. Exploiting exogenous variations in school calendars, we show that there is a genuine effect that returns are indeed lower after major (school) holidays. We hypothesize and provide evidence that the after-holiday low returns is due to investor inattention during holidays, resulting in news, and especially negative news, being incorporated slowly into prices. Consistent with the investor inattention hypothesis, we find that trading is 21

22 significantly reduced during holidays. Furthermore, we find that the low after school-holiday returns are driven by stocks that release negative news, and especially those that also experience light trading, during school holidays. Finally, we document that the after school holiday low return anomaly is stronger among larger stocks with higher institutional holdings, which is in sharp contrast to many anomalies that are more prevalent among small, illiquid stocks with low institutional ownership. We believe that this cross-sectional pattern provides further credence to the investor inattention hypothesis: The apparent arbitraging opportunity is not quickly eliminated even for large, liquid stocks may be precisely due to the fact that investors are less attentive and less likely to act on any arbitrage opportunity. 22

23 References Ariel, Robert A. High Stock Returns Before Holidays: Existence And Evidence On Possible Causes. The Journal of Finance 45.5 (1990): Barber, Brad, and Terrance Odean, 2008, All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors, Review of Financial Studies 21, Bergsma, Kelley, and Danling Jiang, 2016, Cultural New Year Holidays And Stock Returns Around The World, Financial Management 45, 3-35 Bok Baika, Jun-Koo Kang,Jin-Mo Kim, 2010, Local Institutional Investors, Information Asymmetries, And Equity Returns, Journal of Financial Economics 97-1, Chordia, Tarun, Sahn-Wook Huh, and Avanidhar Subrahmanyam, The Cross-Section Of Expected Trading Activity. Review of Financial Studies 20.3 (2007): Coval, Joshua D., and Tobias J. Moskowitz. 1999, Home Bias at Home: Local Equity Preference in Domestic Portfolios, Journal of Finance 54, DellaVigna, Stefano, and Joshua Pollet, 2009, Investor Inattention and Friday Earnings Announcements, Journal of Finance 64, Diamond, D. W. and R. E. Verrecchia, 1987, Constraints on Short-Selling and Asset Price Adjustment to Private Information, Journal of Financial Economics 18, Gabaix, Xavier, David Laibson, Guillermo Moloche and Stephen Weinberg. The Allocation of Attention: Theory and Evidence, American Economic Review, Vol. 96, , Hirshleifer, David and Siew Hong Teoh. Limited Attention, Information Disclosure, and Financial Reporting, Journal of Accounting and Economics, Vol. 36, , Hirshleifer, D., Lim, S.S., Teoh, S.H.: Driven to Distraction: Extraneous Events and Underreaction to Earnings News, Journal of Finance 64, Hong, Harrison and Jialin Yu, 2009, Gone Fishin: Seasonality in Trading Activity and Asset Prices, Journal of Financial Markets 12, Hong, Harrison and Jeremy Stein. A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets, Journal of Finance, Vol. 54, , Hou, Kewei, and Tobias Moskowitz, 2005, Market Frictions, Price Delay, And The Cross- Section Of Expected Returns, Review of Financial Studies 18,

24 Hou, Kewei, Lin Peng, and Wei Xiong, 2008, A Tale Of Two Anomalies: The Implication Of Investor Attention For Price And Earnings Momentum, Working paper, Ohio State University. Huberman, Gur, and Tomer Regev. Contagious Speculation and a Cure for Cancer: A Nonevent that Made Stock Prices Soar, Journal of Finance, Vol. 56, , Kahneman, D.,1973, Attention and Effort, Englewood Cliffs, NJ: Prentice-Hall. Keim, Donald B., 1983, Size Related Anomalies and Stock Return Seasonality, Journal of Financial Economics 12, Lakonishok, Josef, and Seymour Smidt. "Are Seasonal Anomalies Real? A Ninety-Year Perspective." Review of Financial Studies 1.4 (1988): Miller, E.M., 1977, Risk, Uncertainty, and Divergence of Opinion, Journal of Finance 32, Neissner, Marina, 2015, Strategic Disclosure Timing and Insider Trading, working paper, Yale School of Management. Peng, Lin and Wei Xiong. Investor Attention, Overconfidence and Category Learning, Journal of Financial Economics, Vol. 80, pp , Reinganum, Marc R., 1983, "The Anomalous Stock Market Behavior of Small Firms in January: Empirical Tests for Tax-loss Selling Effects," Journal of Financial Economics 12,

25 Figure 1. French school regions This figure presents the three French school regions A, B, and C, after Prior to 1993, the zoning was different but it has been the same since Zone A consists of: Academies of Caen, Clermont-Ferrand, Grenoble, Lyon, Montpellier, Nancy-Metz, Nantes, Rennes, Toulouse. Zone B consists of Academies of Aix-Marseille, Amiens, Besançon, Dijon, Lille, Limoges, Nice, Orleans-Tours, Poitiers, Reims, Rouen, Strasbourg, Zone C consists of Academies of Bordeaux, Créteil, Paris, Versailles. For the winter (Feb/March) and spring (April/May) holidays, Region C goes on holiday first, followed by B, and then by A. All three regions have the same dates for the Christmas, summer, and All-Saints holidays. 25

26 Table 1: Average Return by Calendar Month This table reports the average return (in %) by calendar month for 47 countries. US data is from CRSP and covers the period 1/ /2012. International data is from MSCI Total Return Index and covers 1/ /2012. Country Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec ARGENTINA AUSTRALIA AUSTRIA BELGIUM BOTSWANA BRAZIL CANADA CHILE CHINA COLUMBIA CZECH DENMARK FINLAND FRANCE GERMANY GREECE HONGKONG HUNGARY INDIA INDONESIA IRELAND ISRAEL ITALY JAPAN KOREA MALAYSIA MEXICO NETHERLANDS NEWZEALAND NORWAY PAKISTAN PERU PHILIPPINES POLAND PORTUGAL RUSSIA SINGAPORE SOUTHAFRICA SPAIN SRILANKA SWEDEN SWITZERLAND TAIWAN THAILAND TURKEY UK US Average

27 Table 2. Market Seasonality: Return Differences by Calendar Month This table reports the difference (in %) between the average return for each calendar month and the average return for other months for 47 countries. US data is from CRSP and covers the period 1/ /2012. International data is from MSCI Total Return Index and covers 1/ /2012. *, **, *** indicates that the difference is statistically distinguishable from 0 using a two-tailed test, at the 10%, 5%, and 1% level, respectively. Country Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec ARGENTINA ** -7.12* 5.42*** AUSTRALIA * * *** AUSTRIA * ** *** BELGIUM 1.47* *** -1.54*** *** *** BOTSWANA -1.22* *** *** BRAZIL 7.98* * ** * CANADA *** CHILE * CHINA -4.2* COLUMBIA * *** CZECH ** ** ** DENMARK 2.25** *** * 1.47** FINLAND * -2.72** * 3.65* FRANCE *** ** GERMANY ** * -2.41*** ** GREECE 3.49* * HONGKONG *** ** -2.29* ** HUNGARY * ** ** INDIA *** INDONESIA * * IRELAND * 2.23** * *** *** ISRAEL * ITALY 3.24** ** -1.69* ** JAPAN ** *** KOREA * MALAYSIA ** *** MEXICO NETHERLANDS * 1.59** *** *** NEWZEALAND ** *** *** * NORWAY 2.18** *** *** PAKISTAN * *** PERU PHILIPPINES ** POLAND ** PORTUGAL ** RUSSIA ** SINGAPORE 2.82* *** -2.06** ** SOUTHAFRICA ** SPAIN 1.52* *** SRILANKA * SWEDEN 1.86* 1.8* ** -3.15*** -3.04*** SWITZERLAND *** *** TAIWAN * THAILAND ** TURKEY * UK ** ** ** US ** ** 27