Investor attention and commonalities across asset pricing anomalies

Transcription

1 Investor attention and commonalities across asset pricing anomalies This version: December, 2016 Abstract We comprehensively exam the effects of investor s attention of individual stocks on daily financial market anomalies. The weekday effect of anomalies coincides with the seasonality of investor attention. Using stocks reaching price limits as exogenous shocks, we find that most of anomalies are stronger for stocks attracting relatively higher investor attention. Investor attention may help to explain financial market anomalies beyond investor sentiment, because noise traders invest in stocks that catch excessive attention and make arbitrage hard and mispricing persistent.

2 1. Introduction Classical asset pricing theories indicate that only the loadings of systematic risk factors such as market risk, SMB (Small Minus Big), HML (High Minus Low) and liquidity factor can explain the cross-sectional variation of expected return (see Sharpe (1964), Lintner (1965), and Fama French (1993), Pastor and Stambaugh (2003) for details). However, more recent studies document the firm characteristics that are apparently not from systematic risk such as individual stock liquidity (Amihud, 1986), idiosyncratic volatility (Ang, Hodrick, Xing, and Zhang (2006), Ang, Hodrick, Xing, and Zhang (2009)), maximum of past returns are also cross-sectionally related with stock expected returns beyond factor loadings. Furthermore, empirically, Chordia, Goyal, and Shanken (2015) find stock characteristics explain cross-sectional variation of stock returns more than factor loadings. Hou, Xue, and Zhang (2015) document tens of them. However, many fundamental questions about financial market anomalies are still unanswered. One of them is whether there is commonality across different asset pricing anomalies? Avramov et. al. (2013) indicate that credit risk may explain many anomalies at monthly time interval. They found that most of the anomalies are strong for those high credit risk firms, although the underlying economic reason is still unrevealed. Stambaugh, Yu and Yuan (2012) indicate anomalies are stronger during the time when market-wide investor sentiment is high. Birru (2016) find the day of the week effect of people s mood drives the time-series up and down of daily anomalies. In this paper, we provide evidence over the role of investor attention on financial market anomalies. Using data from East Money Forum, we first generate time-varying investor s abnormal attention for each stock in Chinese stock market each day. information in short term with cross-sectional variation could help explain the daily variation of anomalies. We find there is clear seasonality and weekday effect of investor attention. The patterns coincide with the day-of-the-week effect of financial market anomalies. Furthermore, we do a comprehensive analysis over 9 financial market anomalies, and find that most of the anomalies previously found in the U.S. market are still important in China. Furthermore, the anomalies are stronger for the subsample of stocks which capture relatively higher investor attention. For the stocks that have

3 lower attention, the anomalies are much weaker or even disappear. For example, for size effect, the average return of daily small minus big portfolio is about 31 basis points for stocks with relative high attention, while the return is only 4 basis points for stocks with lowest attention. Investors could generate much more return in stock market if they enhance trading strategies based on anomalies by attention information. In order to disentangle the alternative explanation of investor sentiment (Baker and Wurgler, 2006 and 2007) which endogenously accompany with attention, we first make use of the price limit rule which is unique to Chinese stock market as a repeated natural experiment. Specifically, we focus on stocks that have reached daily upper price limit and group them based on their investor attentions. Stocks with highest returns catch much higher attention than other because of exogenous reasons, even if the change in fundamentals in the two groups of stocks is the same. After purifying other impacts and focusing on the effect of attention, we find that our conclusion still holds. Our evidence supports the theoretical model about casual effect of investor attention on short term financial market anomalies compared with sentiment. Given the observed dramatic cross-sectional variation of attention between different stocks and the clustering feather of investor attention, there could be shocks changing attention and stocks who response to the shocks differently have different expected returns. We generate portfolios based on abnormal attention and come up with the factor mimicking portfolios based on the return difference of stocks with high attention and low attention. Using industrial portfolios as test assets suggested by Lewellen, Nagel, and Shanken (2010), we find that the risk premium for the mimicking portfolio is statistically significant with expected sign. Finally, we try to see the economic reason behind the empirical findings. We evaluate the daily anomaly return in future days up to two months. We found that for groups of stocks with high attention, the anomaly return persists, while for low attention groups, the return shifts around even changing signs. The finding is consistent with Barber and Odean (2008) who find that attention-grabbing stocks attract more noise traders, which makes the market hard to coverage to efficiency.

4 The rest of the paper is organized as follows. We first present a model to motivate the empirical research. In section 3, we provide the data and market information. Section 4 provides main results about financial market anomalies and attention. We explore the economic causality between the two. Section 5 concludes. 2. Data In this paper, we use the investor attention data from the online stock forum, the East Fortune Forum, which is considered as the most popular, influential and actively operated stock forum specializing information exchange among investors, especially individual investors in China. 2 When we search for the keywords stock forum on the most popular search engines in China (Baidu or Google (Hong Kong)), the East Money Stock Forum always ranks among the top search results. Moreover, the forum is fully compatible with a trading software, the East Money Trading Software, which is widely used by investors in China for placing orders in trading. Investors can thus easily access the information posted on the stock forum when they use the software to trade. We regard the discussion among investors in the forum as representative not only for its popularity among investors, but also for its easy connection to the trading software that enables investors to trade after browsing the information in the forum. By using the forum data, we gain the advantage over Google Trends data by removing the ambiguity on identifying stocks and restricting the sample among the individuals with the trading purposes. Moreover, compared with the information releasing websites on which individuals may well simply take a glance, the forum provides platform for investors to post their ideas and communicate before trading, which provides better proxy for concrete investor attention. Google Trends, in essence, is not as reprehensive and comprehensive as our dataset because it doesn t cover many listed firms in the Chinese stock market, probably attributable to the continued government blocking of the search engine. The number of posts rather than views/searches data in the Google trends dataset help avoid the double counting of 2 Recent studies (e.g., Hong, Jiang, Wang, and Zhao (2014) and Chang, Hong, Tiedens, Wang, and Zhao (2015)) use data from the East Money Stock Forum in their analyses.

5 views/searches by the same person, which is also suggested in the work of Wang (2016). Therefore, the East Money Stock Forum serves as an ideal dataset for observing the attention of investors through communication and discussion among investors, which can be influential for stock trading and prices. The forum allocates a specific sub-forum for each stock where investors can exchange ideas and information. For each stock, we calculate the number of posts in its sub-forum (Att i,t ) on a daily basis. To capture the deviation of investor attention from the normal and time trend parts, the investor attention measure is standardized by the baseline level of investor attention to rule out any seasonality and day of week effects. We define the abnormal investor attention (AbnAtt i,t ) by demeaning the daily attention by its 60-day average attention from one month ago, i.e. from day -90 to -30, scaled by the average attention. We exclude the most recent month in the calculation of the benchmark to avoid potential spillover effects of the investor attention (Liu and Peng, 2015):, =, (,(, ) ) (,(, ) ) In essence, the investor abnormal attention (AbnAtt) indicates the percentage increase in investor attention on a specific day compared with the normal level. High, means relatively high attention compared with stocks with low, at the same time. The online stock forum was formed in 2007 and normally operated in mid Therefore, our dataset spans from June 2008 to December Firms with A-shares listed on the Shanghai Stock Exchange and Shenzhen Stock Exchange are included in the sample. To avoid the effect of outliers in our analysis, the investor attention data is winsorized at the 5% and 95% level 3, and the observations with neither effective posts in the base period nor any post in the specific day is excluded from the sample. Moreover, the observations with infinite abnormal attention attributed to the non-zero post in the specific day and zero posts in all the days of the base period are trimmed to the highest abnormal attention value in the sample. Stock return data and other firm-level data for the calculation of anomalies, including the firm market capitalization, total assets, total debts, operational profits etc. are gathered from RESSET 3 The results are qualitatively similar with minor perturbations.

6 Database, a standard research database of China. [Insert Table 1 about here] Table 1 provides the summary statistics of the sample. The mean and standard deviation of the variables are calculated on a time-series basis and then averaged across all the stocks. Among the 2,877 firms in the sample, the daily abnormal attention of investors averages , varying from -1 to , with a standard deviation as high as , suggesting a high cross-sectional and time-series variation in fluctuation of investors attention. The abnormal attention of -1 refers to the observations with zero posts for the specific day and non-zero average number of posts in the base period. 3. and Anomalies 3.1 Anomalies in China In this section, we use the data of Chinese financial market to provide a general look at the existence of returns of 9 anomalies that are empirically documented in the US financial market. We focus on the cross-sectional patterns in the average stock returns of the portfolios classified by each anomaly. In particular, for each month (or each quarter for SUE effects, or each year for ROA and BM), we divide the stocks into 5 quintiles based on variables, such as the Size, the Amihud Ratio, etc. P1 (P5) denotes the portfolio containing stocks with the lowest (highest) value of the variable. Each strategy buys the extreme quintile portfolios P1 (or P5) and sells the opposite extreme quintile portfolio P5 (or P1) according to the specific strategy elaborated below. Each anomaly return refers to the equally-weighted average returns of the spread in the returns of the two extreme quintile portfolios. Below we describe all the 9 anomalies and trading strategies in detail. Anomaly 1: Size. Banz (1981) and Fama and French (1992) document a higher return of small stocks than large stocks. Therefore, we divide stocks into quintiles according to the market capitalization of the stocks for each month, and calculate the return of the Small-Minus-Big (SMB) portfolio of longing the bottom quantile

7 and shorting the high quantile. The results in Table 2 show the anomaly return of 0.032, significant at 1% level. Anomaly 2: Illiquidity. Amihud (2002) proposes the measure of illiquidity, the Amihud ratio, and find a higher return of illiquid stocks compared with relatively liquid stocks. The Table 2 shows the difference in the returns of illiquid stocks (the top quintile) and the liquid stocks (the bottom quintile) is 0.012, significant at 10% level. Anomaly 3: Idiosyncratic volatility (Idiovol). Ang, Hodrick, Xing, and Zhang (2006) and Ang, Hodrick, Xing, and Zhang (2009) find high idiosyncratic volatility stocks have a lower return than stocks with low idiosyncratic volatility. We sort stocks into 5 quintiles based on idiosyncratic volatility and long the bottom quantile (low idiosyncratic stocks) and short the top quantile (high idiosyncratic stocks). The anomaly return is 0.032, significant at 1% level. Anomalies 4: Lottery (Max). Bali, Cakici, and Whitelaw (2011) find a negative correlation between the maximum return of the previous month and the future expected returns. We capture the lottery-like properties of the stocks by calculating the maximum return of the past calendar month (Max) for each stock. Then we divide stocks based on Max. We could see from Table 2 that the anomaly return is 0.010, significant at 10% level, suggesting lower performance of stocks with lottery-like characteristics. Anomalies 5: Book-to-Market Ratio (BM). The value effects have been documented by Fama and French (1992) and many other papers. Specifically, high book-to-market (value) stocks are proved to have a higher expected return than low book-to-market (growth) stocks. However, the anomaly return doesn t seem to exist in Chinese financial market. As is shown in Table 2, the difference between top and bottom quintile is not significant.

8 Anomalies 6: Profitability (ROA). Profitable stocks are proven to outperform less profitable counterparties in the stock market (Balakrishnan, Bartov, and Faurel, 2010; Ball, Gerakos, Linnainmaa, and Nikolaev, 2015). In particular, we use ROA as proxy for profitability of the firms. By dividing the stocks according to theses proxies, we compute the anomaly returns and find strong profitability effects in Chinese stock market. The strategy is to buy the stocks in the top quintile and sell those in the bottom quintile. The anomaly returns, i.e. the return differences between the top and the bottom quintiles, are significantly positive. Anomaly 7: Short-term Reversal We implement the Jegadeesh (1990) short-term reversal strategy by longing stocks with lower prior-month returns (losers) and shorting stocks with higher prior-month returns (winners). This zero-investment strategy is rebalanced every month. The effect exists in the Chinese capital market, reflected by the significantly positive anomaly return in Table 2. Anomaly 8: Price Momentum Jegadeesh and Titman (1993) advocates the price momentum strategy by constructing portfolios based on the cumulative return over the formation period for month t, i.e. months t-7 to t-2. They derive an anomaly return by buying the winner portfolio (the top quintile) and selling the loser portfolio (the bottom quintile). However, Table 2 shows that the momentum effects are weak in China. Anomaly 9: Earnings momentum Literature documents the excess return generated by price surprise. We calculate the standardized unexpected earnings (SUE) on a quarterly basis, by subtracting the mean earnings four quarters ago e i,i-4 from the earnings of the most recent quarter e i, divided by the standard deviation of the earnings changes over the last eight quarters σ i,(t-8),t (Chan, Jegadeesh and Lakonishok, 1996; Hou,

9 Peng and Xiong, 2014)., =,,,( ), The earnings momentum anomaly return is generated by buying the portfolio with the highest SUE (P5) and selling the portfolio with the lowest SUE (P1). The Table 2 reports a significantly higher return for P5 compared with P1, which suggests the existence of earnings momentum anomaly in Chinese stock market. [Insert Table 2 about here] 3.2 Investor attention and day of week effects The previous section uncovers the existence of most anomalies in China, such as the Illiquidity anomaly, the Size anomaly and Profitability anomaly etc. In this section, we track the cross-sectional variation in these anomalies and attempt to figure out the potential driving forces of these patterns. We point to the possibility that the cross-sectional variation in the anomaly performance is to some extent attributable to the patterns of investor attention. We firstly examine the trend and evolution of investor abnormal attention from Monday through Sunday. Figure 1 graphically displays the average investor attention for all stocks on different days of week. The investor abnormal attention exhibits an increase from Monday to Thursday and drops on Friday. The lower attention on Monday and Friday may be aroused by the darkened mood at the beginning of workdays and the spiritual relaxation on the last work day proceeding weekends. The average investor abnormal attention is generally much lower on weekends, probably because investors don t trade and are more likely to be distracted by other issues on these days. Motivated by the patterns of investor attention on different days of week, we could reasonably expect variations in trading behaviors among the investors with limited cognation, in which case, we trace the fluctuations in anomaly returns from Monday through Friday and compare the pattern with that of investor abnormal attention intensities as illustrated in Figure 1. Table 3 reports the results. We find prominent evidence of different anomaly returns on different workdays: the anomaly returns on Monday and Friday are basically lower than those

10 on Tuesday, Wednesday and Thursday for most anomalies. Moreover, significant anomaly returns appear more on Tuesday through Thursday. Interestingly, the pattern of anomaly performance greatly resembles that of the investor attention, as is shown in Figure 1. For example, the investor attention increases from Monday through Thursday and drops on Friday. Consistent with an investor attention explanation, the anomaly returns for SMB strategy climbs from to from Monday to Wednesday, and declines a bit on Thursday and Friday. Besides, the spikes of anomaly return concentrate on Wednesday and Thursday, such as the SMB, Max, Idiosyncratic Volatility, ROA, short-term reversal, etc. We can therefore infer that a change in investor attention may have a contemporaneous effect on anomaly returns. [Insert Figure 1 and Figure 2 about here] 3.3 Investor attention and anomalies: double-sort results An investor-attention-story predicts a higher anomaly return among those stocks that arouse higher investor abnormal attention. In this section, we use a double-sort methodology to examine the cross-sectional variation of anomaly portfolio performance. Specifically, we divide stocks into 5 quintiles on the basis of the lag-one-day investor abnormal attention (AbnAtt t-1 ), and then in each quintile, we further divide stocks into 5 quintiles according to the strategy-specific conditioning variables, such as the Size, the Amihud Ratio, etc. P1 to P5 denotes the five quintiles in each investor attention group and the conditioning variable increases from P1 to P5 consecutively. Similar to the Section 3.1, we derive the equally-weighted average returns of the opposite extreme quintile portfolio P5 (or P1), i.e. the anomaly returns separately for each investor attention group. Furthermore, we conduct a Difference-in-Difference methodology between the buy-and-sell portfolio returns of top and bottom quintiles of investor attention, which is reported in the last line of Panels in Table 4. T-statistics are reported in parentheses. The set of results in Table 4 are aligned with our predictions. For example, in the SMB anomaly portfolio, the anomaly returns in the lowest investor attention group ATT1 is while that in the highest investor attention group ATT5 is The Diff-in-Diff is 0.266, significantly positive at 1% level, suggesting a weaker anomaly for stocks with lower investor attention. [Insert Table 4 about here]

11 The same pattern exists for all of the anomalies mentioned in Section 3.1 except for Momentum anomaly. The lower attention group ATT1 exhibit much lower or even no anomaly returns; in comparison, the high attention group ATT5 is exactly where the anomaly phenomenon exists. The Diff-in-Diff for the Liquidity anomaly, Idiosyncratic volatility anomaly, and Short-term reversal anomaly are all significantly positive and that for BM anomaly, Max anomaly, and ROA anomaly are all positive, though not significantly. The results are all robust to the exclusion of weekend days in the calculation of investor abnormal attention. Further evidence is borne out in Table 5, which replaces the stock returns with the Sharpe ratio to test the economic significance of the investor attention explanation. The results in Table 5 illustrate similar pattern as that in Table 4. Compared with the low attention group ATT1, the high attention group ATT5 generally have higher anomaly incremental sharp-ratio. These results combine to show effects the increase in noise trading by higher abnormal attention among individual investors on the anomalies. [Insert Table 5 about here] 3.4 Stock price limit and anomalies To address the concern that the investor abnormal attention measured by online stock forum posts are generated endogenously by certain news or attention-grabbing events, in this section we use an exactly exogenous measure of investor attention proposed by Seasholes and Wu (2007) and Wang (2016). In Chinese capital market, there is an ideal repeated natural experiment in which investor attention difference is never affected by any fundamental information but only involves the institutional idiosyncratic reason. There is a strict price limit policy established on December 16, 1996 on the Shanghai Stock Exchange (SSE) and the Shenzhen Stock Exchange (SZSE): Stocks are restricted to 10% increase in the prior-trading-day s close price. However, in the actual situation, the minimum tick size for stocks is RMB 1 cent, whereas the calculated 10% price change is not necessarily an integral number of cents - daily price limit has to be rounded to the nearest cent, giving rise to a variation of daily return movements among those stocks that strike the price limit. Wang (2016) provides an illustrative example: for three stocks with price of 9.99RMB, 10.00RMB, and 10.01RMB, the maximum increases in their

12 daily prices of the next day are all 1.00RMB while the calculated return limits differ, i.e %, 10.00%, and 9.99%, respectively, as a result of the differences in current price. Although the minor differences are purely aroused exogenously without information about fundamental factors, there is enormous difference in the attention attracted among these stocks. Public media, such as trading software, finance blogs and websites, and newspapers usually list the stocks with top-ranked daily returns. In the example mentioned above, among the three stocks, the one with current price 9.99RMB is more likely to be involved in the Winners List compared with the one with current price 10.01RMB, if they both hit the price limit. Investors gather information from the social media and decide on their trading strategies, thus these stocks with larger returns attract more attention. Moreover, investors tend to refer to the Winners List on the trading software before they submit their offers. To sum up, among the stocks that hit the price limit, those with highest daily returns as a result of the rounding effects are more salient and attract higher level of investor attention. We use this quasi-natural experiment in our empirical design as a treatment to the effects of investor attention on asset pricing. We restrict the sample to the observations of stocks that hit the upper price limit in the sample period. And the observations in those trading days with less than 5 stocks hitting the upper price limit are deleted from our sample as there is less attention differences for these stocks. On each day, we sort stocks into two sub-samples, High attention group and Low attention group, according to the median of their daily returns. Then we calculate the anomaly returns on the days after the pricing hitting day for each investor attention group. Results are shown in Table 6. The evidence substantiates the cross-sectional differences in anomaly returns among stocks with different levels of investor attention. The Amihud ratio anomaly, Max anomaly, Idiosyncratic volatility, ROA anomaly, BM anomaly, and SUE anomaly are all prove to have significantly higher anomaly returns in the high attention group compared with that in the low attention group, and the magnitude is large enough to be regarded as a prominent evidence of the impacts of investor attention. [Insert Table 6 about here] We conduct single-sort tests on the 9 anomalies mentioned in Section 4.1. We divide stocks into 5

13 quintiles based on investor attention and denote the 5 groups as ATT1 to ATT5; meanwhile, we independently divide all stocks into 5 quintiles based on the strategy-specific conditioning variables as ANML1 to ANML5. Other details are similar to that in Section 4.4. Online Appendix Table 1 documents the results by employing the single-sort methodology. The patterns of the anomalies are in line with the results in Table 4, which further supports our hypothesis. 3.5 Mimicking portfolios from Investor attention The previous section has indicated the role played by investor attention in asset pricing. We make a step further by examining whether the shock of investor attention could explain the cross-sectional variation of stock returns since we find dramatic cross-sectional variation of attention between different stocks and the clustering feather of investor attention. By dividing stocks into 5 quintiles based on the lag-one-day investor abnormal attention, we constructing an Investor Abnormal Factor as the difference between returns in the stocks in the top and bottom quintile on daily basis. In order to see the risk premium of the factor, we first estimate the factor loading in time-series regression using portfolios, then we test if the risk premium is significant using Fama-Macbeth (1973) regression. We first use industrial portfolios as test assets since Lewellen, Nagel, and Shanken (2010) indicate that asset pricing test results are sensitive to portfolios. Industrial portfolios which do not have obvious factor structures should generate most robust results. Table 7 tests the explanatory power of the Investor Abnormal Factor on the risk premium of industrial portfolios and the Size and BM portfolios (5*5). The industry category we use here is the China Securities Regulatory Commission (CSRC) Industry Category. We merge the investor attention factor daily returns with the industry returns of a total of 19 industry portfolios and the Size and BM portfolio returns separately. We conduct the time series regression of the industry returns on the investor attention factor for each industry to come up with the factor loading for each test asset. In the Fama-Macbeth (1973) regression, we use estimated factor loadings as independent variables. As is shown in Table 7, the Abnormal Factor has significantly positive explanatory power for the industry portfolio returns and the effects remain even after controlling for market factor. Similarly, we use Size and

14 BM portfolio returns as test asset. As is shown in Panel B of Table 7, the results are significantly positive, which are in line with our predictions. [Insert Table 7 about here] 3.6 The persistence of attention effects on anomaly performance In this part, we try to shed light on the economic reason why attention is related with anomalies based on persistence of anomalies in stocks with different attention. We reevaluate the attention effects on the anomaly performance by double-sorting and report the future 2, 3, 4, 5, 14, 22, and 28 days returns. We propose that noise traders invest in stocks that catch excessive attention and make arbitrage hard and mispricing persistent (Barber and Odean (2008)). Under this assumption, we should see that the anomaly returns for stocks with different attention should be more persistent for the high attention group compared with that in the low attention group. Results are shown in table 8. Interestingly, the results are aligned with our predictions. Noticeably, the return differences from anomalies instead of attention. For example, the size effect comes from sorting portfolios based on firm size of previous month. But in this paper, we find that attention could still affect asset price by affecting the return of financial market anomalies. In previous research, people focus on monthly return of anomalies, thus they cannot see the return difference change within a month. However, here we use daily data and could clearly see the effect of attention on asset price. The results are more clearly detected in Figure 3 and table 8. [Insert Figure 3 about here] [Insert Table 8 about here] 4. Conclusions In this paper, we find that most of the financial market anomalies are stronger for the firms catching abnormal attentions than other firms. Based on repeated natural experiments of price limit rule which is unique to Chinese stock market, we purify other impacts and focus on the effect of attention. We generate portfolios based on abnormal attention and come up with the factor mimicking portfolios based on the return difference of stocks with high attention and low attention and we find that the risk premia for the

15 mimicking portfolios is statistically significant. Noise traders invest in stocks that catch excessive attention and make arbitrage hard and mispricing persistent.

16 Reference Amihud Y Illiquidity and stock returns: Cross-section and time-series effects. Journal of Financial Markets 5: Amihud, Y., and H. Mendelson Asset Pricing and the Bid-Ask Spread. Journal of Financial Economics 17: Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2006, The cross-section of volatility and expected returns, Journal of Finance 51, Andrew Ang, Robert Hodrick, Yuhang Xing and Xiaoyan Zhang, 2009, High Idiosyncratic Volatility and Low Returns: International and Further U.S. Evidence, Journal of Financial Economics 91, Avramov, D., Chordia, T., Jostova, G., & Philipov, A., Anomalies and financial distress. Journal of Financial Economics, 108(1), Baker, Malcolm, and Jeffrey Wurgler, 2006, Investor sentiment and the cross-section of stock returns, Journal of Finance 61, Baker, Malcolm, and Jeffrey Wurgler, 2007, Investor sentiment in the stock market, Journal of Economic Perspectives 21, Balakrishnan, K., E. Bartov, and L. Faurel Post loss/profit announcement drift. Journal of Accounting and Economics 50: Bali, Turan G, Nusret Cakici, and Robert F Whitelaw, 2011, Maxing out: Stocks as lotteries and the cross-section of expected returns, Journal of Financial Economics 99, Ball, Ray, Joseph Gerakos, Juhani Linnainmaa, and Valeri Nikolaev, 2015, Deflating profitability, Journal of Financial Economics, Forthcoming. Banz, R The relationship between return and market value of common stocks. Journal of Financial Economics 9:3 18. Barber, Brad M., and Terrance Odean, 2008, All That Glitters: The Effect of and News on the Buying Behavior of Individual and Institutional Investors, Review of Financial Studies 21 (2): Birru J. Day of the Week and the Cross-Section of Returns[J]. Working Paper, Chan, L., N. Jegadeesh, and J. Lakonishok Momentum strategies. Journal of Finance 51: Chordia, Tarun, Amit Goyal, and Jay Shanken, 2015, Cross-sectional asset pricing with individual stocks: betas versus characteristics, Working Paper, Emory University. Fama, Eugene F., and Kenneth R. French, 1993, Common Risk Factors in the Returns on Stocks and

17 Bonds, Journal of Financial Economics 33, Fama, E., and J. MacBeth Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81: Hou, Kewei, Chen Xue, and Lu Zhang. "Digesting anomalies: An investment approach." Review of Financial Studies (2014): hhu068. Jegadeesh, N Evidence of predictable behavior of security returns. Journal of Finance 45: Jegadeesh, N., and S. Titman Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of Finance 48: Lewellen, Jonathan W., Jay Shanken, and Stefan Nagel, 2010, A Skeptical Appraisal of Asset Pricing Tests, Journal of Financial Economics 96, Lintner, John, 1965, Security prices, risk and maximal gains from diversification, Journal of Finance 20 (6), Pastor, L., Stambaugh, R Liquidity Risk and Expected Stock Returns. Journal of Political Economy 111: Seasholes, Mark, and Guojun Wu, 2007, Predictable Behavior, Profits, and, Journal of Empirical Finance 14 (5): Sharpe, William, 1964, Capital Asset Prices: A Theory of Market Equilibrium Under Conditions of Risk, Journal of Finance 19, Stambaugh, Robert F., Jianfeng Yu, and Yu Yuan. "The short of it: Investor sentiment and anomalies." Journal of Financial Economics (2012): Wang, Baolian The Causal Effect of Investor. Working paper.

18 0.400 Figure 1: Day of week effects of investor abnormal attention Monday Tuesday Wednesday Thursday Friday Saturday Sunday Abn Figure 2: Seasonality effects of investor abnormal attention Abnormal Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec abn_post

19 Figure 3: The persistence of anomaly returns for different attention groups Evolution of Anomaly Returns in the Lowest quintile days 3 days 4 days 5 days 14 days 21 days 22 days 28 days SMB Amihud Momentum Max Idio Volatility Short-term Reversal SUE BM ROA Evolution of Anomaly Returns in the Highest quintile days 3 days 4 days 5 days 14 days 21 days 22 days 28 days SMB Amihud Momentum Max Idio Volatility Short-term Reversal SUE BM ROA

20 Table 1: Summary statistics This table reports the summary statistics of the main variables in the paper. The absolute number of investor discussions (Attn) and the abnormal attention measure (Abn Attn) are calculated by summarizing the mean and standard deviation of all stocks cross-sectionally for each specific day and then generate the time-series mean. Similarly, the anomaly measures are calculated by getting the cross-sectional means and standard deviations for the specific year-month or year-quarter, and then calculate the time-series mean. Variable Obs. Mean Std. Dev. Min Max Median Kurtosis Absolute 6,264, Abn 6,264, Mon-Mkt Value (*E+10) 215, Mon-Return 202, Idiosyncratic Volatility 202, Max 203, Momentum 212, Amihud Ratio 199, Book-to-market Ratio 19, ROA 43, SUE 82,

21 Table 2: Anomalies in China This table reports the anomalies in China stock market. Specifically, for each month (or each quarter for SUE effects, or each year for ROA and Book-to-market ratio), we divide the stocks into 5 quintiles based on the strategy-specific conditioning variables. The anomalies include Small-minus-big (SMB), Book-to-market ratio (BM), Liquidity, Idiosyncratic volatility, Max, Momentum, ROA and Short-term reversal, which are shown in different columns. We divide stocks into five quintiles according to the measures, which increases from Group 1 to Group 5. Trading strategies are designated for each anomaly, in which P1-P5 indicates longing P1 and shorting P5 and P5-P1 indicates longing P5 and shoring P1. The returns of the anomalies based on the trading strategies are reported in the last column. Anomalies P1 P5 Anomaly P2 P3 P4 Strategy (Low) (High) Return Size 0.049*** 0.043*** 0.034** 0.028* P1-P *** (3.32) (2.75) (2.17) (1.74) (1.10) (5.46) Amihud ratio 0.026* 0.030* 0.034** 0.038** 0.038** P5-P * (1.77) (1.92) (2.12) (2.37) (2.40) (1.82) Max 0.034** 0.041*** 0.038** 0.031* P1-P * (2.31) (2.65) (2.41) (1.93) (1.43) (1.95) Idiosyncratic volatility 0.046*** 0.041*** 0.036** 0.028* P1-P *** (2.99) (2.64) (2.32) (1.78) (0.90) (6.29) ROA 0.035** 0.037** 0.038** 0.061*** 0.165*** P5-P *** (2.35) (2.34) (2.39) (3.49) (4.19) (3.52) Short-term reversal 0.043*** 0.040** 0.038** 0.029* P1-P *** (2.76) (2.53) (2.41) (1.88) (0.94) (4.71) SUE *** 0.089*** 0.051*** 0.058*** P5-P *** (1.31) (4.07) (4.18) (3.27) (3.76) (8.69) Momentum 0.036** 0.035** 0.122*** 0.036** 0.026* P5-P ** (2.43) (2.29) (4.68) (2.28) (1.69) (-2.01) BM 0.033** 0.036** 0.034** 0.033** 0.030* P5-P (2.20) (2.30) (2.11) (2.08) (1.93) (-0.41)

22 Table 3: Day of the week effect of anomalies This table reports the patterns of the returns of the anomalies on different trading days from Monday to Friday. Trading strategy is shown in the second column for each anomaly. The anomalies in the table include Small-minus-big (SMB), Book-to-market ratio (BM), Liquidity, Momentum, Max, Idiosyncratic volatility, ROA, SUE and Short-term reversal. Anomalies Strategy Monday Tuesday Wednesday Thursday Friday Size P1-P *** 0.215*** 0.166*** 0.173*** (0.93) (2.94) (5.02) (4.13) (3.08) Amihud P5-P * (-0.28) (1.91) (0.48) (0.42) (0.69) Max P1-P *** 0.067* (-1.57) (0.14) (-0.66) (2.62) (1.77) Idio Volatility P1-P * *** 0.140*** (1.75) (1.13) (1.04) (4.89) (3.91) ROA P5-P (0.84) (-1.00) (1.21) (0.64) (0.80) Short-term reversal P1-P *** 0.109** 0.161*** 0.083** (-0.27) (2.86) (2.48) (3.78) (2.12) SUE P5-P *** 0.107*** 0.049** 0.076*** 0.130*** (5.29) (5.45) (2.24) (4.13) (5.89) Momentum P5-P (-0.51) (0.34) (-1.49) (-0.47) (-1.33) BM P5-P (-0.05) (-0.43) (-1.00) (-0.24) (-0.59)

23 Table 4: and anomalies (double-sort) This table reports the effects of investor abnormal attentions on the return of anomalies by double-sorting the stocks. We sort stocks into 5 quintiles based on the lagged daily abnormal attention and within each quintile, we detect the return of the anomalies by further dividing the stocks into 5 quintiles for each anomaly. The aim is to examine whether there are significantly different returns of anomalies for stocks with different level of investor attentions. Panel A-K reports the returns anomalies, i.e., Small-minus-big (SMB), Book-to-market ratio (BM), Liquidity, Momentum, Max, Idiosyncratic volatility, ROA, and Short-term reversal. The columns denote the levels of attention and the rows denote the groups of anomaly measures. Investor attention increases monotonically from ATT1 to ATT5. Anomaly-specific measures, such as Firm size, BM, Amihud Ratio, etc., increase from P1 to P5. The returns of the anomalies in line with the trading strategies are reported for each attention group. Furthermore, we conduct compute the difference in anomaly returns (P1-P5) for the high attention group and the low attention group (ATT5-ATT1). Panel A. Small-minus-big (SMB) Size P *** 0.205*** 0.179*** 0.163*** 0.323*** (Low) (3.11) (3.78) (3.36) (3.09) (5.08) P *** 0.183*** 0.164*** 0.133** 0.095* (3.89) (3.41) (3.09) (2.52) (1.82) P *** 0.160*** 0.127** 0.115** (3.50) (2.98) (2.37) (2.15) (1.36) P *** 0.117** 0.101* 0.093* (2.84) (2.20) (1.92) (1.79) (1.30) P ** (High) (2.21) (1.63) (1.29) (1.11) (0.34) Strategy P1-P ** 0.131*** 0.117*** 0.110*** 0.306*** (2.18) (5.57) (5.22) (4.83) (6.72) Dif -in- Dif 0.266*** (6.02) Observations Panel B. Book-to-market Ratio (BM) P *** 0.131** 0.128** 0.111** 0.097* (Low) (3.07) (2.49) (2.46) (2.15) (1.89) P *** 0.160*** 0.128** 0.126** 0.119** (3.07) (2.98) (2.40) (2.37) (2.26) Book-to-mark P *** 0.153*** 0.124** 0.119** et Ratio (3.27) (2.90) (2.36) (2.27) (1.50) (BM) P *** 0.158*** 0.125** 0.105** (3.12) (2.98) (2.39) (2.03) (1.22) P *** 0.132*** 0.116** 0.107** (High) (2.68) (2.64) (2.35) (2.17) (1.31) Strategy P5-P (-1.48) (-0.16) (-0.59) (-0.19) (-1.25) Dif-in-Dif (-0.22) Observations

24 Panel C. Liquidity (Amihud Ratio) P *** 0.111** 0.116** 0.102** (Low) (2.61) (2.15) (2.28) (2.02) (1.15) P *** 0.130** 0.113** 0.102* (2.86) (2.42) (2.15) (1.94) (1.25) Amihud P *** 0.157*** 0.119** 0.106** (3.32) (2.92) (2.23) (2.01) (1.38) P *** 0.172*** 0.143*** 0.120** (3.58) (3.25) (2.72) (2.28) (1.63) P *** 0.175*** 0.141*** 0.122** 0.121** (High) (2.83) (3.41) (2.82) (2.44) (2.42) Strategy P5-P ** ** (0.08) (2.53) (1.13) (0.87) (2.34) Dif-in-Dif 0.060*** (2.97) Observations Panel D. Momentum P *** 0.150*** 0.117** 0.124** 0.238*** (Low) (4.07) (2.94) (2.30) (2.44) (3.95) P *** 0.142*** 0.133*** 0.117** 0.124** (2.91) (2.78) (2.63) (2.31) (2.43) Momentum P *** 0.178*** 0.136*** 0.126** 0.150*** (3.88) (3.44) (2.65) (2.47) (2.84) P *** 0.157*** 0.137*** 0.106** 0.126** (3.51) (2.99) (2.64) (2.06) (2.39) P *** 0.138** 0.101* 0.101* (High) (2.96) (2.51) (1.87) (1.91) (0.88) Strategy P5-P * *** (-1.95) (-0.63) (-0.86) (-1.16) (-4.28) Dif-in-Dif *** (-3.25) Observations Panel E. Max P *** 0.151*** 0.128*** 0.116** (Low) (2.89) (3.23) (2.81) (2.54) (1.26) P *** 0.169*** 0.146*** 0.141*** 0.103** (3.56) (3.31) (2.90) (2.80) (2.04) Max P *** 0.160*** 0.138*** 0.131** (3.52) (3.03) (2.64) (2.53) (1.52) P *** 0.132** 0.123** 0.097* (2.98) (2.40) (2.28) (1.81) (1.45) P *** 0.125** (High) (2.62) (2.21) (1.59) (1.28) (1.43) Strategy P1-P ** 0.044** (-1.03) (1.64) (2.04) (2.21) (-0.99) Dif-in-Dif (-0.20) Observations

25 Panel F. Idiosyncratic volatility Idiosyncratic volatility ATT1 (Low) ATT2 ATT3 ATT4 ATT5 (High) P *** 0.172*** 0.167*** 0.150*** 0.176*** (Low) (3.23) (3.46) (3.42) (3.09) (3.62) P *** 0.176*** 0.144*** 0.136*** 0.091* (3.65) (3.45) (2.85) (2.70) (1.82) P *** 0.165*** 0.129** 0.120** (3.35) (3.17) (2.53) (2.38) (1.37) P *** 0.131** 0.112** 0.088* (3.13) (2.46) (2.13) (1.67) (0.88) P ** 0.101* (High) (2.14) (1.81) (1.21) (1.18) (-0.01) Strategy P1-P *** 0.099*** 0.084*** 0.177*** (1.24) (4.12) (5.40) (4.28) (6.96) Dif-in-Dif 0.152*** (6.81) Observations Panel G. ROA P *** 0.144*** 0.125** 0.103** (Low) (2.88) (3.04) (2.47) (2.03) (0.55) P *** 0.138*** 0.106** 0.086* (2.91) (2.81) (2.02) (1.66) (1.15) ROA P *** 0.121** 0.109** 0.096* (3.87) (2.45) (2.07) (1.83) (1.58) P *** 0.095* 0.104** 0.092* 0.100* (5.40) (1.92) (1.96) (1.74) (1.87) P *** 0.100** 0.104** 0.091* 0.224*** (High) (5.26) (2.04) (2.02) (1.80) (2.83) Strategy P5-P *** ** *** (3.22) (-2.33) (-1.10) (-0.65) (2.94) Dif-in-Dif (0.70) Observations Panel H. Return short-term reversal P *** 0.161*** 0.151*** 0.137*** 0.138*** (Low) (3.18) (3.06) (2.90) (2.61) (2.67) P *** 0.154*** 0.140*** 0.135*** 0.096* (2.98) (3.01) (2.75) (2.67) (1.88) Short-term P *** 0.160*** 0.139*** 0.129** reversal (3.29) (3.13) (2.74) (2.54) (1.57) P *** 0.140*** 0.123** 0.106** (2.70) (2.71) (2.41) (2.08) (0.96) P ** 0.095* (High) (2.57) (1.79) (1.61) (1.02) (0.07) Strategy P1-P * 0.064*** 0.067*** 0.082*** 0.134*** (1.83) (2.95) (3.21) (3.67) (4.62) ` Dif-in-Dif 0.087*** (3.52) Observations

26 Panel I. SUE *** 0.117** (4.60) (2.18) (1.63) (1.22) (0.47) *** 0.132** 0.110** ** (5.47) (2.48) (2.07) (1.58) (2.02) SUE *** 0.159*** 0.115** 0.114** 0.157*** (4.39) (3.04) (2.22) (2.23) (2.87) *** 0.145*** 0.141*** 0.133*** 0.162*** (3.52) (2.88) (2.79) (2.66) (3.04) *** 0.174*** 0.165*** 0.164*** 0.196*** (3.78) (3.43) (3.31) (3.28) (3.73) Strategy P5-P *** 0.080*** 0.100*** 0.171*** (1.56) (4.13) (7.39) (8.45) (7.30) Dif-in-Dif 0.136*** (4.75) Observations

27 Table 5: and anomalies economic significance (double-sort Sharpe ratio) This table reports the effects of investor abnormal attentions on the sharp ratio of anomalies by double-sorting the stocks. We sort stocks into 5 quintiles based on the lagged daily abnormal attention and within each quintile, we detect the differences in sharp-ratio of different anomalies by further dividing the stocks into 5 quintiles for each anomaly. The aim is to examine the economics significance of the effects of investor attention on anomaly returns. Panel A-K reports the anomalies, i.e., Small-minus-big (SMB), Book-to-market ratio (BM), Liquidity, Momentum, Max, Idiosyncratic volatility, ROA, and Short-term reversal. The columns denote the levels of attention and the rows denote the groups of anomaly measures. Investor attention increases monotonically from ATT1 to ATT5. Anomaly-specific measures, such as Firm size, BM, Amihud Ratio, etc., increase from P1 to P5. The Sharpe Ratio of the anomalies in line with the trading strategies are reported for each attention group. Furthermore, we conduct a Diff-in-Diff methodology to examine the gap between the anomaly Sharpe Ratio of top and bottom attention groups (ATT5-ATT1). Panel A. Small-minus-big (SMB) P *** 0.055*** 0.048*** 0.044*** 0.061*** (Low) (2.62) (3.59) (3.12) (2.90) (4.07) P *** 0.052*** 0.046*** 0.037** (3.56) (3.21) (2.89) (2.32) (1.55) Size P *** 0.045*** 0.036** 0.032* (3.27) (2.77) (2.18) (1.96) (1.08) P *** 0.033** 0.028* (2.64) (2.01) (1.71) (1.64) (0.96) P ** (High) (1.97) (1.46) (1.08) (0.92) (0.13) Strategy P1-P *** 0.031*** 0.030*** 0.059*** (1.63) (4.87) (4.49) (4.28) (6.74) Dif-in-Dif 0.050*** (5.86) Observations Panel B. Book-to-market Ratio (BM) P *** 0.035** 0.034** 0.030* (Low) (2.82) (2.24) (2.21) (1.95) (1.56) P *** 0.045*** 0.035** 0.035** 0.029* (2.84) (2.77) (2.17) (2.15) (1.83) Book-to- P *** 0.044*** 0.035** 0.034** market Ratio (3.00) (2.68) (2.12) (2.05) (1.18) (BM) P *** 0.046*** 0.036** 0.030* (2.77) (2.75) (2.18) (1.82) (0.97) P ** 0.038** 0.033** 0.031* (High) (2.35) (2.36) (2.09) (1.94) (1.14) Strategy P5-P (-0.79) (0.31) (-0.10) (0.19) (-0.64) Dif -in- Dif (-0.07) Observations

28 Panel C. Liquidity (Amihud Ratio) P ** 0.030* 0.030** 0.027* (Low) (2.39) (1.92) (1.98) (1.81) (0.90) P *** 0.036** 0.031* 0.028* (2.66) (2.21) (1.94) (1.70) (0.90) Amihud Ratio P *** 0.045*** 0.034** 0.029* (3.08) (2.70) (2.03) (1.79) (1.03) P *** 0.050*** 0.040** 0.034** (3.29) (3.04) (2.46) (2.10) (1.19) P ** 0.051*** 0.041** 0.036** 0.033** (High) (2.39) (3.11) (2.55) (2.20) (2.02) Strategy P5-P *** ** (0.68) (2.79) (1.59) (1.12) (2.29) Dif -in- Dif 0.014** (2.03) Observations Panel D. Momentum P *** 0.042*** 0.031** 0.034** 0.040*** (Low) (3.48) (2.72) (2.03) (2.22) (2.73) P *** 0.040** 0.038** 0.033** 0.031* (2.67) (2.49) (2.41) (2.07) (1.94) Momentum P *** 0.051*** 0.038** 0.035** 0.286*** (3.52) (3.20) (2.40) (2.24) (4.29) P *** 0.044*** 0.039** 0.029* 0.028* (3.14) (2.76) (2.41) (1.85) (1.76) P ** 0.036** 0.027* 0.027* (High) (2.52) (2.30) (1.69) (1.72) (0.44) Strategy P5-P *** (-1.42) (-1.04) (-0.85) (-1.22) (-4.02) Dif -in- Dif *** (-3.26) Observations Panel E. Max P *** 0.045*** 0.038*** 0.035** (Low) (2.60) (3.03) (2.61) (2.34) (1.05) P *** 0.049*** 0.043*** 0.041*** 0.028* (3.29) (3.13) (2.71) (2.62) (1.79) Max P *** 0.045*** 0.039** 0.037** (3.22) (2.80) (2.40) (2.33) (1.33) P *** 0.036** 0.033** (2.68) (2.17) (2.02) (1.55) (1.00) P ** 0.033* (High) (2.28) (1.91) (1.29) (0.99) (0.88) Strategy P1-P ** 0.016*** 0.018*** (-0.08) (2.57) (2.97) (3.06) (0.12) Dif -in- Dif (0.19) Observations

29 Panel F. Idiosyncratic volatility P *** 0.051*** 0.050*** 0.045*** 0.045*** (Low) (2.76) (3.20) (3.17) (2.89) (2.92) P *** 0.051*** 0.041*** 0.040** (3.32) (3.18) (2.59) (2.49) (1.55) Idiosyncratic P *** 0.047*** 0.036** 0.033** Volatility (3.07) (2.94) (2.26) (2.13) (1.08) P *** 0.036** 0.030* (2.84) (2.23) (1.84) (1.42) (0.59) P * (High) (1.93) (1.57) (0.97) (0.92) (-0.27) Strategy P1-P ** 0.027*** 0.034*** 0.030*** 0.050*** (2.01) (5.04) (6.41) (5.29) (7.01) Dif -in- Dif 0.037*** (5.75) Observations Panel G. ROA P *** 0.040*** 0.035** 0.028* (Low) (2.59) (2.81) (2.27) (1.82) (-0.02) P ** 0.038** 0.029* (2.53) (2.52) (1.77) (1.44) (0.96) ROA P *** 0.033** 0.029* * (3.00) (2.17) (1.82) (1.59) (1.68) P *** 0.025* 0.028* *** (3.88) (1.65) (1.73) (1.49) (3.27) P *** 0.026* 0.027* *** (High) (3.50) (1.69) (1.75) (1.46) (3.80) Strategy P5-P ** *** (1.61) (-2.55) (-1.33) (-0.95) (3.99) Dif -in- Dif (-1.05) Observations Panel H. Return short term reversal P *** 0.045*** 0.043*** 0.038** 0.038** (Low) (2.72) (2.88) (2.73) (2.43) (2.43) P ** 0.043*** 0.039** 0.038** (2.50) (2.77) (2.51) (2.46) (1.64) Short-term P *** 0.045*** 0.039** 0.036** Reversal (2.84) (2.85) (2.50) (2.31) (1.33) P ** 0.040** 0.034** 0.029* (2.42) (2.52) (2.16) (1.84) (0.73) P ** (High) (2.38) (1.58) (1.36) (0.81) (-0.17) Strategy P1-P *** 0.021*** 0.025*** 0.041*** (1.40) (3.00) (3.35) (3.74) (4.84) Dif -in- Dif 0.030*** (3.95) Observations