Institutional Investor Participation and Stock Market Anomalies TAO SHU * May 2012

Transcription

1 Institutional Investor Participation and Stock Market Anomalies TAO SHU * May 2012 * Terry College of Business, University of Georgia. taoshu@uga.edu. Parts of this paper were drawn from the working paper Trader Composition, Price Efficiency, and the Cross-Section of Stock Returns. I acknowledge the helpful comments from John Griffin, Norman Strong (Associate Editor), Sheridan Titman, Julie Wu, and an anonymous referee. I appreciate the financial support from the Terry Sanford Award at the University of Georgia.

2 Institutional Investor Participation and Stock Market Anomalies Abstract This paper investigates the impact of institutional trading volume on the cross-section of stock returns. I construct a measure that evaluates the percentage of total trading volume of a stock accounted for by institutional trades. Using a large sample of firms from , I find strong evidence that the strength of stock market anomalies such as price momentum, post-earnings announcement drift, value premium, and investment anomaly is decreasing in institutional trading volume. Additionally, the effects of institutional trading volume are stronger than those of institutional ownership, the major measure of institutional investor participation in the finance literature. These findings suggest that institutional trading significantly improves stock price efficiency.

3 1. INTRODUCTION The rapid growth of institutional investors has motivated numerous studies on the effects of institutional investor participation on financial markets. Most of these studies measure institutional investor participation using a stock s institutional ownership i.e., percentage of shares outstanding of the stock held by institutional investors. In this paper, I examine a different aspect of institutional investor participation that has been largely ignored by the current finance literature. Specifically, I study the percentage of trading volume of a stock accounted for by institutional investors and the impact of institutional trading volume on stock market anomalies. Institutional trading volume can be vastly different from institutional ownership. Specifically, institutional ownership evaluates institutional holding of a stock relative to individuals but institutional trading volume measures how actively institutions trade a stock relative to individuals Institutional trading volume can deviate substantially from institutional ownership because shareholders are not necessarily traders. If, for example, passive pension funds hold 90% of a stock s shares but rarely trade, then little trading volume of the stock is accounted for by institutions despite high institutional ownership. Similarly, if a stock has a low institutional ownership but by a group of hedge funds or active mutual funds, then the stock can have a high institutional trading volume despite low institutional ownership. Empirically, the correlation between fraction of institutional trading volume (the FITV measure, described in section 2) and institutional ownership is only 0.41 during , far from a perfect positive correlation. Institutional trading volume can have significant impact on stock price efficiency and therefore stock market anomalies. Institutions are generally considered more sophisticated traders than individuals. For example, Nofsinger and Sias (1999) find a strong positive relation between change in institutional ownership and future stock returns, suggesting that institutional investors 1

4 have the ability to predict stock returns. 1 Individual investors, in contrast, have been documented to lose significantly from their trading and suffer a number of behavioral biases when they trade (e.g., Odean, 1998; Barber and Odean, 2000; Grinblatt and Keloharju, 2001). If institutions are more sophisticated than individuals, then institutional investor participation can speed up information diffusion into stock prices and improve stock price efficiency. As a result, institutional investor participation will reduce the magnitude of observed stock market anomalies if these anomalies are associated with price inefficiency. Institutional ownership and institutional trading volume are both important aspects of institutional investor participation. Although institutional trading volume has received little attention from financial researchers, its impact on stock price efficiency and stock market anomalies can be greater than that of institutional ownership. This is because institutional trading directly moves stock prices but institutional holding does not. Let s take post-earnings announcement drift as an example. If market under-react to an earnings announcement, then such under-reaction will lead to a postannouncement drift in the same direction as earnings announcement return when prices gradually adjust to the fundamental value. When institutions observe the under-reaction and trade on earnings announcement return, their trading moves prices towards fundamentals, reducing or eliminating the post-earnings announcement drift. In contrast, if institutions observe the under-reaction but simply act as passive shareholders without trading the stock, then under-reaction or the post-earnings announcement drift will remain intact. In this paper, I empirically investigate the effects of institutional trading volume on major stock market anomalies. I construct the FITV measure (fraction of institutional trading volume) that evaluates for a firm-quarter the percentage of total trading volume accounted for by institutional 1 While many subsequent studies confirm the findings of Nofsinger and Sias (1999), some papers show that institutional trading could also move stock prices away from fundamental values. For example, Brunnermeier and Nagel (2004) and Griffin, Harris, Shu, and Topaloglu (2011) find that institutional trading contributed to the high tech bubble. 2

5 investors. Specifically, for each stock-quarter I use quarterly institutional holdings from Thomson Reuters 13f database and calculate an institution s trading volume as the absolute value of its change in holdings over the quarter. The absolute value aims to capture trading volume of the institution whether it buys or sells the stock over the quarter. I then sum up trading volumes of all institutions and divide by total trading volume of the stock-quarter (from CRSP) to obtain the FITV measure. Section 2 describes the construction of the FITV measure. It is worth noticing that the FITV measure excludes intraquarter round-trip institutional trades. For example, if an institution purchases and then sells 1% of a stock s shares within a quarter, then the FITV measure will not include these two trades because they are not reflected in quarterend holdings. Elton, Gruber, Blake, Krasny, and Ozelge (2010) and Puckett and Yan (2011) examine two institution samples comprised of mutual funds and pension funds, and find that intraquarter round-trip trades account for approximately 20 percent of all institutional trades. 2 Therefore, the FITV measure likely captures majority of trading volume by mutual funds, pension funds, banks, and investment advisors but exclude trading volume by day-traders or hedge funds that adopt highfrequency strategies. 3 Since the FITV measure is used for only cross-sectional comparison in this paper, the missing round-trip trades will not introduce substantial noise to the results unless the amounts of intraquarter round-trip trades systematically vary across different stocks. I calculated the FITV measures for a large panel of 177,613 firm-quarters from , with average 1,741 firms in each cross-section. The sample is restricted to NYSE/AMEX firms because trading volumes of NASDAQ stocks are inflated relative to those of NYSE/AMEX stocks by different trading mechanisms. Stocks priced below $5 or with market capitalizations below NYSE 2 Elton et al. (2010) examine 214 actively managed mutual funds during Puckett and Yan (2011) examine institutional trades provided by ANcerno that account for approximately 10 percent of institutional trading volume during 1999 to My sample includes all institutional investors during My sample also excludes small stocks, penny stocks, and NASDAQ stocks. 3 Section 2 presents details about how FITV captures institutional trades with various investment horizons. For example, it can be shown that FITV captures 100% of institutional trades with horizon of over three months, 67% of institutional trades with horizon of two months, but only 8.3% of institutional trades with horizon of one week. 3

6 10 percent breakpoint were dropped to control for microstructure effects. The mean of the FITV measure during is 0.493, suggesting that institutions account for at least 49.3 percent of the trading volume of an average sample firm during this period. It is difficult to assess the exact amount of missing intraquarter round-trip trades for the sample. If intraquarter round-trip trades are assumed to account for 20 percent of all institutional trades as suggested by previous studies (Elton et al., 2010; Puckett and Yan, 2011), then after including round-trip trades institutions account for approximately 61.6 percent (49.3/(1-0.2)) of total trading volume for an average sample firm during More importantly, the cross-sectional distribution of the FITV measure is quite dispersed, suggesting that different stocks indeed receive very different amounts of institutional trading volumes. Next, I examine the effects of the FITV measure on four major stock market anomalies. I first examine price momentum (past winners outperform past losers) and post-earnings announcement drift (PEAD) because they are among the most investigated and robust anomalies in stock markets (Fama, 1998). I also study value premium (value stocks outperform growth stocks) because there is a big literature exploring the risk- and behavioral-based explanations of value premium. The analyses of value premium across levels of institutional trading volume can shed some light on this debate. Finally, I choose investment anomaly (firms with low corporate investments outperform firms with high corporate investments) because it has attracted a lot of attention and that researcher use it to explain other anomalies such as post-seo underperformance and the accrual anomaly. 4 The empirical results provide strong evidence that the strength of all four stock market anomalies is decreasing in the percentage of institutional trading volume. For example, in two- 4 A number of papers provide evidence of investment anomaly and study its relations with other stock market anomalies. An incomplete list of these papers includes Titman, Wei, and Xie (2004), Polk and Sapienza (2007), Xing (2007), Liu, Whited, and Zhang (2007), Wu, Zhang, and Zhang (2007), and Lyandres, Sun, and Zhang (2007). 4

7 dimensional sorting analysis, momentum is 1.52 percent per month (t-stat 6.86) in low FITV stocks but only 1.06 percent (t-stat 4.87) in high FITV stocks. Similarly, post-earnings announcement drift is 0.85 percent per month (t-stat 7.01) in low FITV stocks but only 0.40 percent (t-stat 4.00) in high FITV stocks. The spreads between high and low FITV stocks in momentum (0.46 percent, t-stat 2.94) and post-earnings announcement drift (0.45 percent, t-stat 2.78) are both statistically and economically significant. These findings are robust with the multivariate regressions that control for firm characteristics and other variables documented to affect momentum and post-earnings announcement drift. More strikingly, I find that value premium and investment anomaly only exist in stocks with low institutional trading volume. Specifically, value premium is 1.26 percent (t-stat 4.58) per month in low FITV stocks but an insignificant 0.29 percent (t-stat 1.46) in high FITV stocks. The difference in value premium between low and high FITV stocks is 0.97 percent (t-stat 3.77). Additionally, investment anomaly is 0.83 percent per month (t-stat 2.90) in low FITV stocks but only 0.23 percent (t-stat 1.01) in high FITV stocks. The spread in investment anomaly between low and high FITV stocks is also statistically and economically significant (0.60 percent, t-stat 2.80). These findings are also robust with the multivariate regression analyses that control for firm characteristics and other variables documented to affect value premium and investment anomaly. Overall, the empirical results consistently suggest that the strength of stock market anomalies is strongly decreasing in institutional trading volume. A number of papers find that stocks with higher institutional ownership exhibit weaker anomalous stock returns (e.g., Alangar, Bathala, and Rao, 1999; Bartov, Radhakrishnan, and Krinsky, 2000; Collins, Gong and Haribar, 2003; Ke and Ramalingegowda, 2005; Nagel, 2005). I use two approaches to carefully control for the effects of institutional ownership. First, the tests are repeated using a residual FITV measure constructed as the residual from cross-sectional regression of the 5

8 FITV measure on institutional ownership. The residual measure captures the component of FITV that is orthogonal to institutional ownership. Second, I estimate multivariate regressions that control for the effects of institutional ownership on the examined anomalies. The results of both approaches show that the results on the FITV measure are robust after controlling for institutional ownership, indicating that the effects of institutional trading volume is not driven by institutional ownership. Finally, I perform an interesting horse race between the FITV measure and institutional ownership in the multivariate regression framework. Consistent with previous studies, I find negative relations between institutional ownership and stock market anomalies. However, these relations largely disappear after I control for the FITV measure, suggesting that the effects of institutional ownership may be due to its correlation with institutional trading volume. This paper makes important contribution to the literature of institutional investors. It is the first study to show that fraction of institutional trading volume, which has been largely ignored by the current literature, has significant impact on stock market anomalies. While this paper focuses on stock market anomalies, future studies can explore the effects of institutional trading volume on other stock market phenomena. This paper also makes two contributions to the literature of stock market efficiency. First, the empirical results provide supporting evidence of institutional investor participation improving stock price efficiency (e.g., Alangar et al., 1999; Bartov et al., 2000; Gibson, Safieddine, and Sonti, 2004; Nagel, 2005; Boehmer and Kelley, 2009). Second, the decreased strength of all anomalies examined in stocks with low institutional trading volumes is also consistent with these anomalies being associated with price inefficiencies. The findings in this paper therefore also shed light on the large literature that explores the risk- and behavioral-based explanations of stock market anomalies. 6

9 The rest of the paper is organized as follows. Section 2 introduces the construction of the FITV measure. Section 3 describes data and sample construction. Section 4 examines the effects of institutional trading volume on stock market anomalies, and Section 5 concludes. 2. MEASURING FRACTION OF INSTITUTIONAL TRADING VOLUME I construct the FITV measure (fraction of institutional trading volume), which evaluates the percentage of trading volume of a firm-quarter accounted for by institutional trading, using the equation below. FITV iq = n IOijq IOijq 1 j= 1 Vol ijq (1) where FITV iq is the FITV measure of stock i in quarter q. IO ijq is institution j s share ownership of stock i at the end of quarter q, and IO ijq-1 is institution j s share ownership of stock i at the end of quarter q-1. Vol iq is total share volume of stock i in quarter q, calculated as the sum of monthly share volumes during quarter q. Specifically, to calculate FITV for a firm-quarter, I first sum up the absolute values of quarterly changes in ownership of the stock across all institutions, and then divide by total share volume of the stock during the quarter. 5 A caveat of the FITV measure is that it double counts the trades between two institutions. For example, if institution A sells 100 shares to institution B, then this trade will be counted twice as it is incorporated into the changes in holdings of both A and B. While it is difficult to evaluate the magnitude of double counting, this issue does not seem to have a big impact on my inferences. Conceptually, double counting assigns greater weights to the trades with intensive participation of institutional traders, which is consistent with the goal of the measure. Empirically, since my tests 5 I carefully adjust for stock split during the quarter when constructing the FITV measure. Specifically, I calculate the numerator using change in shares holdings adjusted for stock split, expressed as a percentage of shares outstanding at the end of the quarter. I then calculate quarterly share volume in the denominator as sum of monthly share volumes, expressed as percentages of shares outstanding of the months. To control for outliers, both numerator and denominator are winsorized at 99% cutoff points. 7

10 focus on cross-sectional comparison of the FITV measure, the inferences will be largely unaffected unless there is evidence that double counting varies significantly across individual stocks. It is worth noticing that the FITV measure excludes round-trip institutional trades during a quarter. Specifically, FITV includes a trade with investment horizon longer than three months and excludes a trade with investment horizon shorter than one day (day-trading). For a trade with investment horizon between one day to three months, its probability of being captured by FITV is an increasing function of investment horizon. For example, assuming that trades occur randomly in a quarter, then a trade with investment horizon of two months will be included in FITV as long as it occurs in the second or the third month of the quarter (included with 2/3 probability). In contrast, a trade with investment horizon of one week will be included in FITV only when it occurs in the last week of the quarter (included with a probability of one-twelfth, or 8.3 percent). The chart below lists the corresponding probabilities for trades with different investment horizons. Investment Horizon Percentage of Trades Captured by FITV 3 months 100.0% 2 months 66.6% 1 month 33.3% 2 weeks 16.7% 1 week 8.3% 2 days 3.0% <=1 day 0.0% Elton et al. (2010) and Puckett and Yan (2011) both find that intraquarter round-trip trades account for approximately 20 percent of all institutional trades. Their findings indicate that the FITV measure captures the vast majority of institutional trades. An alternative approach to examine missing round-trip trades is through portfolio turnovers of institutional investors. Carhart (1997), for example, documents that mutual funds have an average turnover of 77.3% per year. Reca, Sias, and Turtle (2012) find that annualized turnover rate is 109% for hedge funds and 41% for non-hedge fund institutions. Since the turnover ratios in both studies are based on quarterly data, after 8

11 considering the 20 percent missing round-trip trades the observed turnovers can be converted to approximately 51% (41%/0.8) for non-hedge fund institutions, 96% (77%/0.8) for mutual funds, and 136% (109%/0.8) for hedge funds. These turnover ratios correspond to hold periods of 23.5 months, 11.0 months, and 8.8 months, respectively, all well above the one-quarter interval of trading volume calculation. Therefore the evidence of investment horizon also indicates that the FITV measure captures the vast majority of institutional trades. To distinguish the FITV measure from institutional ownership, I also construct a residual FITV measure (ResFITV) that adjusts for institutional ownership. Specifically, in each quarter I estimate a cross-sectional regression of the FITV measure on institutional ownership and take the residuals. The ResFITV measure is therefore the component of the FITV measure that is orthogonal to institutional ownership. I then repeat all the tests using the residual FITV measure to control for institutional ownership. 3. DATA AND SAMPLE CONSTRUCTION I obtain quarterly institutional holdings from Thomson Reuters 13f database, which contains the filings by institutions under Section 13f of the Security and Exchange Act of Section 13f stipulates that all investment managers with discretion over 13f securities worth $100 million or more report their holdings to the SEC each quarter. Common stock, preferred stock, and convertible debt are included in 13f securities. The types of institutions covered by Section 13f are banks, investment advisors, nonprofit institutions, investment companies, pension funds, colleges and foundations, insurance companies, broker-dealers, and investment banks. For each firm-quarter, I obtain the reported net changes in share holdings from the prior report date for each institution to calculate the FITV measure in equation (1). Stock data including return, share volume, price, and shares outstanding are obtained from the monthly CRSP database. The sample is restricted to ordinary common shares (share code 10 or 9

12 11). If a firm-quarter exists in CRSP but not 13f, I do not drop the firm-quarter but follow the literature (e.g., Gompers and Metrick, 2001) and assign zero to institutional holding. Annual accounting data and quarterly earnings announcement data are obtained from annual and quarterly Compustat databases. The data on analyst coverage are collected from I/B/E/S. Finally, daily stock data including return, stock price, and share volume are obtained from the daily CRSP database. NASDAQ stocks are excluded from the sample because their share volumes are inflated relative to those of NYSE/AMEX stocks due to different trading systems. To control for the microstructure effects, I drop the stocks priced below $5 or market capitalizations below the NYSE 10 percent breakpoint. Firms with negative book-to-market ratios are also excluded from the sample as they could be associated with either data errors or extreme operating conditions. My final sample contains 177,613 firm-quarters from the third quarter of 1980 to the last quarter of 2005, with an average of 1,741 firms in each cross-section. 4. THE EFFECTS OF INSTITUTIONAL TRADING VOLUME ON STOCK MARKET ANOMALIES (i) Summary Statistics Table 1 presents summary statistics of sample firms. The average of the FITV measure is 49.3 percent for sample firms. Additionally, the FITV measures vary significantly across individual stocks. For example, the 90 th percentile of FITV is 87.8 percent, about ten times the 10 th percentile of 8.7 percent. These results suggest that individual stocks experience vastly different amounts of institutional trading volume. The equal-weighted average of institutional ownership is 41.1 percent for our sample firms (value-weighted average is 50.2 percent), which is consistent with previous studies. Table 1 also reports other characteristics of sample firms. Size for a firm-quarter is the natural log of a firm s market capitalization at the beginning of the quarter. B/M is the book-to- 10

13 market ratio calculated as the sum of a firm s book equity and deferred tax divided by the firm s market equity. 6 I apply the book-to-market ratio at fiscal year-end in calendar year t to the one-year period starting from the July of year t+1. Ret [-6,-1] for a firm-quarter is the six-month buy-and-hold returns up to the end of previous quarter. Beta is estimated annually with market model using daily stock returns in the previous calendar year. Adjusted analyst coverage for a firm-quarter is analyst coverage at the end of the previous quarter minus the average coverage of the firm s NYSE size quartile (Griffin and Lemmon, 2002). Quarterly stock turnover is calculated by summing up monthly turnovers during the quarter, where monthly turnover is monthly share volume divided by shares outstanding. The Amihud illiquidity measure for a stock is constructed following Amihud (2002) using the following equation Illiq n t= = 1 rt Dvol n t (2) where r t is stock return on day t and Dvol t is dollar volume on day t. The illiquidity measure is calculated annually using all daily returns and dollar volumes in the previous year. 7 The square-root adjustment in Equation (2) is proposed by Hasbrouck (2006) to address skewness of the original Amihud measure. (ii) The Effect of Institutional Trading Volume on Price Momentum I examine the effect of institutional trading volume on price momentum using the rolling momentum strategy proposed by Jegadeesh and Titman (1993). At the beginning of each month of the sample period, an independent sort is used to rank stocks into terciles of the one-quarter lag FITV measures and deciles of past six-month returns. These two-dimensional portfolios are then 6 To eliminate outliers, I winsorize the book-to-market ratios at the 99th percentile as did in the literature. 7 I follow Amihud (2002) and drop the firms with less than 200 valid daily observations in the estimation window. 11

14 held for six months. To control for microstructure effects, I skip a month between portfolio formation and return measurement. I calculate monthly returns of momentum portfolios using the decomposed buy-and-hold method proposed by Liu and Strong (2008). Specifically, portfolio return in an individual month τ of the holding period (τ=2, 3,, 6) is calculated as weighted-average of the month-τ stock returns with the weight being a stock s buy-and-hold return from month 1 to τ-1 in the holding period. This approach assumes that a portfolio, once formed, is not rebalanced in the holding period. In contrast, a commonly used approach that calculates portfolio return in each month as simple average of monthly stock returns ( rebalances method ) actually assumes rebalancing the portfolio every month during the holding period. 8 Panel A of Table 2 reports for each portfolio the average monthly return and t-statistic calculated using Newey-West robust standard error with five lags. 9 The monthly momentum profit (Winner-Loser) is 1.52 percent (t-stat 6.86) for the lowest FITV tercile but only 1.06 percent (t-stat 4.87) for the highest FITV tercile. The difference in momentum profit between the lowest and highest FITV terciles is 0.46 percent (t-stat 2.94), both statistically and economically significant. To separate the effect of the FITV measure from that of institutional ownership, I repeat the sorting analysis using the residual FITV measure which is orthogonal to institutional ownership (described in Section 2). The results in Panel B of Table 2 show that momentum profit of the bottom ResFITV group exceeds the top group by 0.41 percent (t-stat 2.70). This result indicates that the effect of institutional trading volume persists after controlling for institutional ownership. 8 Liu and Strong (2008) demonstrate that the rebalance method can create substantial biases especially in small or lowprice stocks because of the negative return autocorrelations in these stocks. They show that the rebalance method exaggerates size premium but underestimates price momentum. For robustness, I also repeat the tests in this paper with the rebalance method and obtain similar results. 9 For robustness I also calculate Newey-West errors with different numbers of lags and the results are similar. 12

15 I further estimate multivariate Fama-Macbeth regressions to control for other variables that also affect momentum, such as firm size (Jegadeesh and Titman, 2001), book-to-market ratio (Daniel and Titman, 1999), analyst coverage (Hong, Lim, and Stein, 2000), and stock turnover (Lee and Swaminthan, 2000). I estimate cross-sectional regressions of quarterly stock returns and report time-series means of the coefficients and associated t-statistics calculated using Newey-West robust standard errors with five lags. To control for non-linearity of the variables and to ease comparison of economic significances, I follow Chan, Jegadeesh, and Lakonishok (1996) and transform all the independent variables into ranks uniformly distributed between 0 and 1. One month is skipped before return measurement to control for the microstructure effects. Table 3 presents the results on the regressions. The variable of interest is the interaction between FITV and past six-month returns. I expect the coefficient on the interaction to be significantly negative based on the negative association between FITV and momentum found with the sorting analysis (Table 2). The results of Model 1 show that, indeed, the coefficient of the interaction between FITV and past return is a significantly negative (t-stat -4.46). This result persists when I control for firm characteristics including beta, firm size and book-to-market ratio in Model 2. I also repeat the regression but with institutional ownership instead of the FITV measure in Model 3. The interaction between institutional ownership and past return is also significantly negative but much smaller than the FITV interactions in Models 1 and 2. These results suggest that institutional ownership also negatively impacts momentum but the effect is much weaker than that of FITV. For a horse race between the FITV measure and institutional ownership, I include both the FITV interaction and the institutional ownership interaction in Model 4. I further include interactions of past returns with size, book-to-market ratio, turnover, and analyst coverage in Model 13

16 5 to control for the effects of these characteristics on price momentum. Interestingly, I observe that while the coefficient on the FITV interaction remains significantly negative, the coefficient on the ownership interaction becomes insignificantly positive after controlling for FITV. These results indicate that the effect of institutional ownership on momentum is largely due to its correlation with institutional trading volume. The coefficients on the other interaction terms are consistent with the previous studies that momentum is stronger in smaller firms, growth firms, higher turnover firms, and firms with less analyst coverage. The economic significances of coefficients on the FITV interactions are also consistent with the sorting analyses in Table 2. For example, in Model 5, the coefficient on the FITV interaction is (t-stat -2.91), which suggests that quarterly momentum profit is about 1.58 percent stronger in the bottom FITV tercile than in the top FITV tercile. 10 To summarize, the multivariate regression analyses in Table 3 confirm that the strength of price momentum is decreasing in institutional trading volume. (iii) The Effect of Institutional Trading Volume on Post-Earnings Announcement Drift I examine the effect of institutional trading volume on post earnings-announcement drift (PEAD) using a rolling PEAD trading strategy proposed by Chan, Jegadeesh, and Lakonishok (1996). Specifically, I construct an earnings shock measure for each firm-month as the four-day cumulative abnormal return (CAR) during the [-2, 1] window centered on the firm s most recent earnings announcement. Daily abnormal return is calculated using the market model, where the market beta is estimated in the one-year window up to two months before the announcement. 11 To avoid using 10 The 1.58 percent is calculated as follows. Since the independent variables are transformed into ranks uniformly distributed between 0 and 1, the gap of 1.58 percent is calculated as 0.9 (difference between the top and bottom pastreturn deciles) times 0.66 (difference between the top and bottom FITV terciles) times the coefficient of I use CRSP value-weighted index as the market portfolio. Firms with less than six months of stock returns in the estimation window are excluded to avoid estimation errors. 14

17 stale earnings data, I drop a firm-month if the firm s most recent earnings announcement is more than three months away. The PEAD strategy is similar to the rolling momentum strategy proposed by Jegadeesh and Titman (1993) except that the portfolios are formed on earnings shocks rather than past returns. At the beginning of each month, an independent sort is used to rank stocks into terciles of the onequarter lag FITV measure, and deciles of earnings shocks. The two-dimensional portfolios are held for six months, and PEAD is the difference in returns between the top and the bottom portfolios of earnings shock. Monthly portfolio returns are calculated using the decomposed buy-and-hold method proposed by Liu and Strong (2008). 12 I skip one month before return measurement to control for microstructure effects and calculate t-statistics using Newey-West robust standard errors with five lags. In Panel A of Table 4, I observe that PEAD is 0.85 percent per month (t-stat 7.01) in the bottom FITV tercile but only 0.40 percent (t-stat 4.00) in the top FITV tercile. The gap in PEAD is 0.45 percent per month (t-stat 2.78) between the top and bottom FITV terciles, both statistically and economically significant. To control for the effect of institutional ownership, I repeat the sorting analysis with the residual FITV measure that is orthogonal to institutional ownership. The results in Panel B show that, similar to Panel A, PEAD is 0.27 percent (t-stat 2.01) stronger in the bottom ResFITV tercile than in the top ResFITV tercile. I further perform multivariate regressions to verify the findings from sorting analysis. Specifically, I estimate Fama-Macbeth regressions of quarterly stock returns on the interactions between the FITV measure and earnings shock. Since the sorting analysis in Table 4 shows that PEAD is decreasing in the FITV measure, I expect the coefficient on the interaction term to be significantly negative. The independent variables also include interactions of earnings shock with 12 Section 4(ii) discusses the details of the decomposed buy-and-hold method. I also conduct robustness tests using the rebalance method and obtain similar results. 15

18 institutional ownership and firm size because previous studies find that PEAD is stronger in stocks with lower institutional ownerships and smaller stocks (e.g., Bartov et al., 2000). Table 5 reports the results of return regressions. Models 1 and 2 show that the coefficient on the interaction between FITV and earnings shock is significantly negative, and this result holds when I control for the commonly examined firm characteristics including beta, size, book-to-market ratio, and momentum. These results are consistent with the negative relation between PEAD and institutional trading volume documented in the sorting analysis (Table 4). For a comparison, I also examine the effect of institutional ownership on PEAD in Model 3. Consistent with Bartov et al. (2000), I observe a significantly negative coefficient on the interaction between institutional ownership and earnings shock. However, the coefficient (-2.37) is smaller than that on the FITV interaction (-2.70). Models 4 and 5 include both the FITV and ownership interactions. I further control for the effect of size on PEAD using an interaction term between firm size and earnings shock. Interestingly, the results show that after including the size interaction, the coefficient on the FITV interaction remains significantly negative but that on the ownership interaction becomes insignificantly negative (t-stat -0.69). The economic significance of the regression coefficients is also consistent with sorting analyses in Table 4. For example, in Model 5, the coefficient on the FITV interaction is (t-stat -1.87), which suggests that post-earnings announcement drift is about 0.87 percent per quarter stronger in the bottom FITV tercile than in the top FITV tercile. 13 To summarize, both the sorting analyses (Table 4) and regression analyses (Tables 5) provide strong evidence that institutional trading volume has significantly negative impact on post-earnings announcement drift. 13 The 0.87 percent is calculated as follows. Since the independent variables are transformed into ranks uniformly distributed between 0 and 1, the gap of 0.87 percent is calculated as 0.9 (difference between the top and bottom earnings shock deciles) times 0.66 (difference between the top and bottom FITV terciles) times the coefficient of

19 (iv) The Effect of Institutional Trading Volume on Value Premium I first perform sorting analysis to examine the effect of institutional trading volume on value premium. At the beginning of each quarter, stocks are independently sorted into terciles of the onequarter lag FITV measures and deciles of book-to-market ratios. Then the time-series means of the monthly portfolio returns and the associated t-statistics calculated using Newey-West robust standard errors are reported. 14 Panel A of Table 6 reports the results of the sorting analysis. These results are striking because I observe that value premium exists only in stocks with low institutional trading volume and disappears in stocks with high institutional trading volume. Value premium (return difference between the top and the bottom book-to-market deciles) monotonically decreases in the FITV measure, ranging from 1.26 percent per month (t-stat 4.58) in the lowest FITV tercile to 0.29 percent (t-stat 1.46) in the highest FITV tercile. The spread in value premium between the extreme FITV terciles is 0.97 percent per month (t-stat 3.77), both statistically and economically significant. I also repeat the sorting analysis using the residual FITV measure that controls for institutional ownership and find similar results in the Panel B of Table 6. I further perform multivariate Fama-Macbeth regressions to examine the effect of institutional trading volume on value premium. The dependent variables are monthly stock returns and the independent variable of interest is the interaction between FITV and book-to-market ratio. I predict the coefficient on the FITV interaction to be significantly negative because the sorting analysis in Table 6 suggests a negative relation between value premium and institutional trading volume. A cross-sectional regression of stock returns is estimated for each month and time-series 14 Since book-to-market ratio is annual measure and FITV is quarterly measure, the two-dimensional portfolios are formed quarterly and held for three months. I follow Liu and Strong (2008) and calculate monthly portfolio returns using the decomposed buy-and-hold method (discussed in Section 4(ii)). I also use the decomposed buy-and-hold method for the sub-portfolio analysis on investment anomaly in the next subsection. For robustness, I repeat the tests using the rebalance method and observe similar results. 17

20 means of the coefficients are reported. I also report associated t-statistics calculated using Newey- West robust errors with five lags. Table 7 presents the results of the return regressions. In Models 1 and 2, I observe significantly negative coefficients on the FITV interactions and they are robust to the controls of firm characteristics. For a comparison, Model 3 examines the effect of institutional ownership on value premium. Consistent with Nagel (2005), I find significantly negative coefficient on the ownership interaction, which suggests institutional ownership also has a significantly negative effect on value premium. Interestingly, when I include both the FITV interaction and the ownership interaction in Model 4, the coefficient on FITV interaction remains significantly negative (-1.03, t- stat -3.43) but that on ownership interaction becomes insignificant (-0.30, t-stat -0.94). The results are similar after I control for the effect of firm size on value premium in Model While these results confirm the robustness of the FITV effect, they suggest that the effect of institutional ownership on value premium is likely due the correlation between institutional ownership and institutional trading volume. The coefficients of the FITV interactions are also economically significant and in line with the sorting analyses in Table 6. For example, in Model 5, the coefficient on the FITV interaction is , translating into a gap in monthly value premium of 0.58 percent between the lowest and the highest FITV terciles. 16 To summarize, both sorting analyses in Table 6 and multivariate regression analyses in Table 7 suggest that institutional trading volume strongly reduces value premium. 15 Fama and French (1992) find that value premium is larger in smaller stocks. In Model 5, the interaction term is insignificant, probably because I exclude small stocks from the sample by dropping firms priced below $5 or with market capitalization below NYSE 10 percent break point. 16 The 0.58 percent is calculated as follows. Since all independent variables are transformed into the ranks uniformly distributed between 0 and 1, the difference of 0.70 percent is calculated as 0.9 (difference between between the lowest and the highest book-to-market deciles) times 0.66 (difference between the lowest and the highest FITV terciles) times the coefficient of

21 (v) The Effect of Institutional Trading Volume on Investment Anomaly I examine the effect of institutional trading volume on investment anomaly using the corporate investment measure from Titman, Wei, and Xie (2004). Specifically, I define investment of a firm as its capital expenditure (Compustat annual item 128) divided by sales (annual item 12) and apply the investment of a fiscal year ending in calendar year t to the one-year period from July of year t + 1. For robustness tests, I also repeat the tests using the alternative investment measures that scale capital expenditure by net property, plant, and equipment or asset values (Xing, 2008; Wu, Zhang, and Zhang, 2010) and observe similar results. At the beginning of each quarter, firms are independently sorted into terciles of the onequarter lag FITV measures and deciles of investments. I then calculate time-series means of the monthly returns of the two-dimentional portfolios and the associated t-statistics using Newey-West robust standard errors with five lags. Panel A of Table 8 reports the results. Interestingly, investment anomaly concentrates in the firms with low institutional trading volume. Specifically, investment anomaly (return difference between low and high investment firms) is 0.83 percent per month (t-stat 2.90) among the highest FITV tercile, but only 0.23 percent and insignificant (t-stat 1.01) in the lowest FITV tercile. The spread in investment anomaly between the top and bottom FITV terciles is 0.60 percent (t-stat 2.80), both statistically and economically significant. In Panel B of Table 8, I control for the effect of institutional ownership using the residual FITV measure and obtain results similar to those in Panel A. I also perform multivariate Fama-Macbeth regressions of monthly stock returns similar to those in the previous subsections. For brevity these results are not tabulated but they are consistent with the sorting analysis in Table 8 in terms of both statistical and economic significances. To summarize, I find strong empirical results that institutional trading volume has significant effects on the stock market anomalies including momentum, post-earnings announcement drift, 19

22 value premium, and investment anomaly. These findings provide supporting evidence that institutional trading improves stock price efficiency. 5. CONCLUSION This paper is the first study that investigates the impact of institutional trading volume on stock market anomalies. I construct a quarterly measure that evaluates the percentage of total trading volume of a stock accounted for by institutional trading, and examine its effects on four major stock market anomalies. I find that all four anomalies, including price momentum, post-earnings announcement drift, value premium, and investment anomaly, are significantly weaker in stocks with higher fractions of institutional trading volumes. Additionally, value premium and investment anomaly exist only in stocks with low institutional trading volumes. These results are robust with both sorting analyses and multivariate regression analyses that control for firm characteristics and other factors documented to affect these anomalies. This paper makes important contributions to the literature of institutional investors and the literature of stock market efficiency. I present evidence that fraction of institutional trading volume, which has been largely ignored by the current finance literature, has significant impact on stock market anomalies. Additionally, the effects of institutional trading volume are stronger than those of institutional ownership, the most commonly used measure of institutional investor participation in the current finance literature. The findings in this paper also suggest that institutional investor participation can improve stock price efficiency and therefore significantly weaken stock market anomalies associated with price inefficiencies. While this paper focuses on stock market anomalies, future studies can further explore the effects of institutional trading volume on other stock market phenomena or corporate events. 20

23 References Alangar, Sadhana, Chenchuramaiah T. Bathala, and Ramesh P. Rao, 1999, The effect of institutional interest on the information content of dividend-change announcements, Journal of Financial Research 22, Amihud, Yakov, 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal of Financial Markets 5, Barber, Brad M., and Terrance Odean, 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of Finance 55, Bartov, Eli, Suresh Radhakrishnan, and Itzhak Krinsky, 2000, Investor sophistication and patterns in stock returns after earnings announcements, Accounting Review 75, Boehmer, Ekkehart, and Eric Kelley, 2009, Institutional investors and information efficiency of prices, Review of Financial Studies 22, Brunnermeier, Markus K., and Stefan Nagel, 2004, Hedge funds and the technology bubble, Journal of Finance 59, Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52, Chan, Louis K.C, Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, Journal of Finance 51, Collins, Daniel W., Guojin Gong, and Paul Hribar, 2003, Investor sophistication and the mispricing of accruals, Review of Accounting Studies 8, Daniel, Kent, and Sheridan Titman, 1999, Market efficiency in an irrational world, Financial Analysts Journal 55, Elton, Edwin J., Martin J. Gruber, Christopher R. Blake, Yoel Krasny, and Sadi O. Ozelge, 2010, The effect of holdings data frequency on conclusions about mutual fund behavior, Journal of Banking and Finance 64, Fama, Eugene F., and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47, Gibson, Scott, Assem Safieddine, and Ramana Sonti, 2004, Smart investments by smart money: Evidence from seasoned equity offerings, Journal of Financial Economics 72, Gompers, Paul A., and Andrew Metrick, 2001, Institutional investors and equity prices, Quarterly Journal of Economics 116, Grinblatt, Mark and Matti Keloharju, 2000, The Investment behavior and performance of various investor types: A study of Finland s unique data set, Journal of Financial Economics 55,

24 Griffin, John M, Jeffrey H. Harris, Tao Shu, and Selim Topaloglu, 2011, Who drove and burst the tech bubble? Journal of Finance 66, Griffin, John. M., and Mark L. Lemmon, 2002, Book-to-market equity, distress risk, and stock returns, Journal of Finance 57, Hasbrouck, Joel, 2009, Trading costs and returns for US equities: Estimating effective costs from daily data, Journal of Finance 64, Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad news travels slowly: Size, analyst coverage, and the profitability of momentum strategies, Journal of Finance 55, Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implications for stock market efficiency, Journal of Finance 48, Jegadeesh, Narasimhan, and Sheridan Titman, 2001, Profitability of momentum strategies: An evaluation of alternative explanations, Journal of Finance 56, Ke, Bin, and Santhosh Ramalingegowda, 2005, Do institutional investors exploit the post-earnings announcement drift? Journal of Accounting and Economics 39, Lee, Charles M.C., and Bhaskaran Swaminthan, 2000, Price momentum and trading volume, Journal of Finance 55, Liu, Laura Xiaolei, Toni Whited, and Lu Zhang, 2009, Investment-based expected stock returns, Journal of Political Economy 117, Liu, Weimin, and Norman Strong, 2008, Biases in decomposing holding-period portfolio returns, Review of Financial Studies 21, Lyandres, Evgeny, Le Sun, and Lu Zhang, 2008, The new issues puzzle: Testing the investmentbased explanation, Review of Financial Studies 21, Nagel, Stefan, 2005, Short sales, institutional investors and the cross-section of stock returns, Journal of Financial Economics 78, Nofsinger, John R., and Richard W. Sias, 1999, Herding and feedback trading by institutional and individual investors, Journal of Finance 54, Odean, Terrance, 1998, Are investors reluctant to realize their losses? Journal of Finance 53, Polk, Christopher, and Paola Sapienza, 2009, The stock market and corporate investment: A test of catering theory, Review of Financial Studies 22, Puckett, Andy, and Xuemin (Sterling) Yan, 2011, The interim trading skills of institutional investors, Journal of Finance 66,