Active Institutional Investors and Stock Return Anomalies

Transcription

1 Active Institutional Investors and Stock Return Anomalies Weike Xu * Rutgers Business School-Newark and New Brunswick September 2015 Abstract This paper explores the role of active institutional investors in correcting market mispricing. I examine how changes in ownership breadth affect the profitability of 22 anomaly-based strategies. I use two measures of ownership breadth: the number of active institutional investors that own a stock and the market value of stocks held by active institutional investors. I find that the profitability of these strategies is stronger following the declines in the ownership breadth in the prior quarter. The return pattern is primarily attributed to the lower return in the short portfolio. In addition, tighter short-sale constraints due to decrease in the ownership breadth can explain the lower return in the short portfolio. The conclusions stay the same after controlling for the common risk factors including the Fama-French three factors and the momentum factor. My results are robust to different size groups, different portfolio weighting methods and crosssectional regression tests. These findings indicate that active institutional investors improve market efficiency. Keywords: Institutional investors, Asset pricing anomalies, Market efficiency JEL code: G23, G12, G14 * Rutgers Business School-Newark and New Brunswick, Rutgers University, weike@scarletmail.rutgers.edu. I am indebted to Yangru Wu, Jin-mo Kim, Ankur Pareek, and Lin Peng for invaluable support and advice. I also thank Enrique Arzac, Danqi Hu (AAA discussant), Phil Davies, Chong Xiao, Yichuan Liu, Xianjue Wang, Minxing Sun, Akiko Watanabe (NFA discussant), Yi Fang and seminar participants at Rutgers Business School for helpful discussion. 1

2 Asset pricing literature documents that firm characteristics have significant power to predict future stock returns. For instance, future stock returns are positively related to past 6-12 months returns (Jegadeesh and Titman, 1993), and to book-to-market ratios (Fama and French, 1992). In addition, future stock returns are negatively related to past one month return (Jegadeesh, 1990; and Lehmann, 1990), and idiosyncratic volatility (Ang et al., 2006). These patterns are considered as anomalies because they are not explained by standard asset pricing models, such as the CAPM and the Fama and French three-factor model. Institutional investors play a growing role in the US equity market. Whether institutional investors add value to correct market mispricing is a longstanding debate in the finance literature. 1 On one hand, Lewellen (2011) argues that institutional investors hold the market portfolio and fail to take advantage of well-known anomalies. Edelen, Ince and Kadlec (2014) find that institutional investors increase their holdings of overpriced stocks and decrease their holdings of underpriced stocks and therefore can be a possible source of anomalies. On the other hand, several papers find that institutional investors trade on anomalies. Green, Hand, and Soliman (2011) find that the growth of hedge fund investments can partly explain the demise of the accrual anomaly. Cheng et al. (2015) find that short-term reversal is temporally higher following declines in the number of active institutional investors. Hanson and Sunderam (2014) show that short sellers trade on well-known anomalies such as value and momentum. One reason why researchers reach different conclusions on the role of institutional investors is that they study all institutional investors as a whole. However, institutional investors have different trading behaviors. Compared with passive institutional investors such as pension 1 Other papers that study the relationship between institutional investors and anomalies: mutual funds and accruals (Ali et al., 2008), short-term trading and anomalies (Cremers and Pareek, 2014), transient institutional investors and PEAD (Ke and Ramalingegowda, 2005). 1

3 funds, active institutional investors have shorter investment horizons and different investment objectives. Therefore, they are more likely to exploit anomalies. Active institutional investors have grown significantly over the past two decades. According to the Thomson Reuters 13f database, active institutional investors take up 71% of total institutional trading at the end of It is important to separately study the behavior of active institutional investors. This paper empirically tests the role of active institutional investors in correcting stock mispricing. Specifically, I examine how changes in ownership breadth affect the profitability of 22 anomalybased strategies. I use two measures of ownership breadth: the number of active institutional investors that own a stock (institutional number changes) and the market value of stocks held by active institutional investors (institutional value changes). The first measure is motivated by Chen, Hong, and Stein (2002) (hereafter CHS) who argue that the breadth of ownership, as defined by the number of institutions that long a stock, can be a proxy for how tightly the shortsale constraints bind. Decreases in breadth of ownership forecast low future returns as the shortsale constrains bind more tightly. The second method is proposed by Choi, Jin, and Yan (2012) (hereafter CJY) who find that changes in the wealth-weighted ownership breadth is positively correlated with future stock returns in the Chinese market. I examine four hypotheses in this paper. First, I hypothesize that the profitability of anomaly-based strategies is higher following a decline in the ownership breadth in the prior quarter. Stocks whose changes in ownership breadth are in the bottom tertile (LOW group) are expected to be more overpriced than those in the top tertile (HIGH group), as short-sale constraints are binding more tightly for the former stocks and less capital can be devoted to these 2 I calculate the ratio by using the market value held by active institutional investors divided by the total market value held by all institutional investors. Following Abarbanell, Bushee and Raedy (2003), I define active institutional investors as investment companies and independent advisors. 2

4 stocks to correct mispricing. I show that 15 out of 22 anomalies are significantly stronger in the LOW group than in the HIGH group. For institutional number changes, the combination strategy (the strategy that takes equal positions across all 22 anomalies) yields a risk-adjusted return of 83 bps per month in the LOW group, and 50 bps per month in the HIGH group. The difference between the LOW and the HIGH groups is 33 bps per month, which is statistically significant at the 1% level. The second hypothesis is that the higher profits of anomaly-based strategies in the LOW group are primarily attributed to the lower returns in the short portfolios. For instance, the combination strategy yields a risk-adjusted return of -19 bps per month in the LOW group for the long side while it yields a risk-adjusted return of -102 bps per month in the LOW group for the short side. Thus, the combination strategy yields a risk-adjusted return of 83 bps per month in the LOW group. The third hypothesis is that the lower returns from the short side in the LOW group are due to tigher short-sale constraints. I find that returns from the short side are significantly lower (more negative) for each anomaly in the LOW group. When averaged across all anomalies, the risk-adjusted return of the short side increases by 82% in the LOW group and decreases by 46% in the HIGH group compared with that from pure anomaly-based strategies. Two factors can account for this result: (1) reduction in short-sale constraint; (2) superior managerial skills of active institutional investors. I find evidence to support the short-sale constraint hypothesis. I use two other short-sale constraints proxies (size and total institutional ownership, Nagel, 2005) to divide stocks into short-sale constrained and unconstrained stocks. I find that the return pattern that the anomalies are stronger in the LOW group is more significant for short-sale constrained stocks and is less significant or insignificant for unconstrained stocks. 3

5 The fourth hypothesis is that strategies that long the stocks with high-performing characteristics and highest changes in ownership breadth, and short the stocks with the lowperforming characteristics and lowest changes in ownership breadth outperform pure anomalybased strategies. For instance, compared with the value-weighted risk adjusted return of pure anomaly-based strategies, when averaged across all anomalies, it increases by 38 bps per month for conditional strategies on institutional number changes and anomalies. I conduct a series of robustness check. I find that the return pattern is not driven by small stocks. I show that the results are robust after controlling for the Fama-French three factors and the momentum factor. I provide evidence that the return pattern is hold by forming valueweighted portfolios. Finally, I show that the results are similar in cross-section regression tests. This paper makes several contributions. First, I show evidence that active institutional investors add value to correct market mispricing. I find that the 22 anomalies documented in the literature are weaker for the stocks following the growth in the ownership breadth. Second, I study the role of institutional investors on a broad set of anomalies from the angle of short-sale constraints. Cheng et al. (2015) consider movements of active institutional investors as a measure of effectiveness of liquidity provision to investigate short-term reversal. In this paper, I argue that movements of active institutional investors would affect the short-sale constraints, and therefore would affect the profitability of anomalies. Third, combining ownership breadth changes and well-known anomalies, I develop new trading strategies that outperform these pure anomaly-based strategies. The remainder of the paper is organized as follows. Section I describes the data and methodology. Section II presents returns of 22 pure anomaly-based strategies and the return 4

6 pattern of ownership breadth changes. Section III reports sources of the return pattern. Section IV reports the returns of new strategies. Section V conducts some robustness tests and Section VI concludes. I. Data and Methodology My data are from four sources. Accounting data come from the Annual and Quarter CRSP/Compustat Merged database. Monthly data on stock return, stock price, and shares outstanding are from Center for Research in Security Prices (CRSP) monthly files. Returns on Fama-French s three factors and the momentum factor are from Kenneth French s website. Finally, the institutional investor holdings data come from Thomson Reuters Institutional (13f) holdings S34 files. The SEC requires all institutional investors with greater than $100 million of securities under management to report their quarter-end holdings within 45 days after each calendar quarter. All common stock positions greater than 10,000 shares or $200,000 must be disclosed. Thomson Reuters classifies institutions as five types: (1) banks, (2) insurance companies, (3) investment companies, (4) independent advisors, and (5) others. Unfortunately, the type code is not reliable after 1998, as Thomson Reuters improperly classified many institutions as endowment and others. 3 So I use the institutional investors classification data from Brian Bushee s personal website. 4 Following Abarbanell, Bushee and Raedy (2003), I define active institutional investors as investment companies and independent investment advisors. This measure excludes passive 3 According to User s Guide to Thomson Reuters Mutual Fund and Investment Company Common Stock Holdings Databases on WRDS, Thomson Reuters has no plan to fix the mapping errors. 4 I thank Brian Bushee for providing the institutional investor classifications data at this website: 5

7 institutions such as bank trusts, insurance companies, university and foundation endowments, and pension funds. I use two measures of changes in ownership breadth: changes in the number of active institutional investors that own a stock (CHS, 2002), and the market value held by active institutional investors (CJY, 2012). As institutions grow over time, to control for the growth effect, in constructing the measures of changes in ownership breadth, I require institutions to hold at least one stock in both quarter q and quarter q-1. My sample consists of all NYSE/AMEX/Nasdaq common stocks (share codes 10 and 11) from January 1981 to March The sample begins in 1981 due to the availability of institutional investors type data. I exclude ADRs, REITs, financials, closed-end funds, foreign shares, all firms in the financial sectors (SIC ), and stocks with share prices less than $5 at the beginning of portfolio formation date. Following Lewellen (2011), I reverse Thomson Reuters 13f s split adjustment using CRSP cumulative shares and price adjustment factor whenever there is a difference between the filing date and the reporting date. To ensure that the results are not driven by small stocks, following Fama and French (2008), I group all stocks into two subsamples based on market equity at the beginning of each month: all-but-tiny stocks (those larger than NYSE 20 percentile), and large stocks (those larger than NYSE 50 percentile). My tests include 22 well-documented anomalies. Following Kogan and Tian (2014), I separate them into 7 groups below (the detailed construction of these anomalies is described in Appendix 1): (1) Prior returns: short-term reversal (STREV), long-term reversal (LTR), momentum (MOM); 6

8 (2) Valuation: book-to-market (BM), 5 earnings-to-price (EP), sales-to-price (SP); (3) Earnings: returns on assets (ROA), standardized unexpected earnings (SUE), sales growth (SG), growth profitability premium(gp); (4) Distress: O-score (OS), market leverage (LEV); (5) Investment: investment-to-assets (IA), asset growth (AG), accruals (TOTA), net operating assets (NOA), investment-to-capital (IK), investment growth (IG); (6) External financing: net stock issuance (NSI), long-term stock issuance (LSI); (7) Others: turnover (TO), idiosyncratic return volatility (IVOL). Financial statements are often released to the public after the fiscal end period. I use the method below to take into account the financial reporting lag issue. Following Fama and French (1992), I match the annual accounting data at fiscal year ending in calendar year y-1 with the monthly variables at the end of June in calendar year y. Following Chen, Marx and Zhang (2014), I match the monthly return data with the quarterly accounting data using the months immediately after the most recent earnings announcement data (RDQ). For example, if a company announces the financial statements of the fourth quarter of year y-1 on February 20 th of year y, I use the returns data in March of year y to match with the fourth quarter financial information of year y-1. In addition, I merge the institutional investor s holdings data with CRSP, and CRSP/Compustat using CUSIP numbers and report dates. Specifically, the CUSIP number from institutional investor s holding data at the end of quarter q-1 of report year is matched with NCUSIP from CRSP at calendar quarter q. Table 1 reports summary statistics of firm characteristics for all stocks, all-but-tiny stocks and large stocks from January 1981 to March 2014 after matching the institutional investor s 5 I construct BM as book equity in the prior fiscal year divided by market equity in last month. See Asness and Frazzini (2013), Asness, Moscowitz, and Pedersen (2013), and Lewellen (2014). Similarly, I use last month market equity as the denominator for EP, SP, and LEV. See Lewellen (2014). 7

9 holding data with CRSP and CRSP/Compustat database. All-but-tiny stocks take up around 60% of the whole sample and the large stocks take up around 30% of the whole sample. Except for turnover (TO) and standard unexpected earnings (SUE), variations of characteristics are highest in all stocks and lowest in large stocks. For instance, the mean of idiosyncratic volatility for all stocks is 0.15 with the standard deviation of 0.14, while the mean of idiosyncratic volatility for large stocks is 0.08 with the standard deviation of II. Empirical Results: The Return Pattern A. Returns of Pure Anomaly-based Strategies This section examines the significance of the 22 anomalies. Specifically, I sort all stocks into quintile portfolios based on 22 firm characteristics and compute the equal-weighted and value-weighted returns for each portfolio. Then I calculate hedge portfolio return and riskadjusted return for each anomaly. I consider the higher-performing quintile as the long portfolio, and the lower-performing quintile as the short portfolio. For example, quintile 5 in book-tomarket ratio (BM) is the long portfolio; however, quintile 5 in the short-term reversal (STREV) is the short portfolio. In addition, I construct a combination strategy (COMB) by taking equal positions across all 22 anomalies. The portfolio formation and rebalancing frequencies differ across strategies. Specifically, for variables from CRSP, and CRSP/Compustat annual merged data, I form portfolios at the beginning of each month and the portfolios are rebalanced monthly. 6 For variables from Compustat annual data, I form portfolios at the end of each June and the portfolios are 6 They include short-term reversal (STREV), momentum (MOM), long-term reversal (LSR), net stock issuance (NSI), long-term stock issuance (LSI), turnover ratio (TO), idiosyncratic return volatility (IVOL), book-to-market (BM), earnings-to-price (EP), sales-to-price (SP) and market leverage (LEV). 8

10 rebalanced annually. 7 For variables from Compustat quarterly data (ROA, OS, and SUE), I form portfolios at the month after the recent quarterly announcement, and the portfolios are rebalanced monthly. Table 2 reports the profits from all 22 anomaly-based strategies across different size groups from January 1981 to March The risk-adjusted return (α) is computed from the following regression: R t = α + β 1 MKT t + β 2 SMB t + β 3 HML t + ε t, (1) where the R t is the return spread between the long and the short portfolios at month t, MKT t is the value-weighted market excess return at month t, SMB t is return spread between small and large stocks at month t, HML t is the return spread between high and low value stocks at month t. The t-statistics are computed using heteroskedasticity-consistent standard errors of White (1980). Panels A and B report the equal-weighted average hedge portfolio returns and riskadjusted returns of all 22 anomalies for each size group. Consistent with the prior literature, all anomaly-based strategies generate significantly positive average hedge portfolio returns. Similarly, all anomaly-based strategies generate significantly positive risk-adjusted returns except for short-term reversal (STREV), long-term reversal (LTR) and market leverage (LEV). In addition, profits of these strategies are highest in all stocks, and lowest in large stocks, implying that anomalies are strongest in tiny stocks. For instance, the combination strategy yields an average return of 81 basis points (bps) per month (t-statistic=7.90) and a risk-adjusted 7 They include investment-to-assets (IA), assets growth (AG), accruals (TOTA), investment-to-capital (IK), net operating assets (NOA), investment growth (IG), sales growth (SG), and growth profitability (GP). 9

11 return of 76 bps per month (t-statistic=13.28) for all stocks while it earns an average return of 45 bps per month (t-statistic=3.73) and a risk-adjusted return of 35 bps per month (t-statistic=5.12) for large stocks. Panels C and D report the value-weighted average hedge portfolio returns and riskadjusted returns of all 22 anomalies for each size group. Consistent with prior literature, anomalies are usually weak by forming value-weighted portfolios. I find that some anomalies become statistically insignificant in Panels C and D. For average hedge portfolio returns, short term reversal (STREV), net stock issue (NSI), turnover (TO), market leverage (LEV), sales growth (SG), and investment-to-capital (IK) are statistically insignificant in each size group. For risk-adjusted returns, short term reversal (STREV), long-term reversal (LTR), turnover (TO), market leverage (LEV), book-market ratio (BM), sales-price ratio (SP) and sales growth (SG) are statistically insignificant in each size group. The combination strategy yields an average return of 46 bps per month (t-statistic=4.14) and a risk-adjusted return of 40 bps per month (tstatistic=6.73) for all stocks, compared with an average return of 81 bps per month (tstatistic=7.90) and a risk-adjusted return of 76 bps per month (t-statistic=13.28) in Panels A and B. Overall, the results in Table 2 show that these firm characteristics have predict power for future stock returns. B. The Return Pattern This section tests how changes in ownership breadth affect the profitability of anomalybased strategies. First, I sort stocks into tertile portfolios: low (T1), medium (T2), and high (T3), according to quarterly changes in ownership breadth, and quintile portfolios based on the lagged 22 firm characteristics. The stocks in the T1 portfolio (LOW group) are those with lowest 10

12 changes in ownership breadth in the prior quarter. The stocks in the T3 portfolio (HIGH group) are those with highest changes in ownership breadth in the prior quarter. Second, I independently sort stocks into 3 5 porfolios based on quarterly changes in ownership breadth and firm characteristics. 8 Finally, I compute average hedge portfolio returns and risk-adjusted returns of all anomalies across the LOW and HIGH groups. In addition, I calculate the mean difference of average hedge portfolio returns and risk-adjusted returns of all anomalies between LOW and HIGH groups. To obtain the difference of risk-adjusted returns, I compute the difference of hedge portfolio returns between the LOW and the HIGH groups each month, and then run the following regression: R d,t R i,t = α dif + β 1 MKT t + β 2 SMB t + β 3 HML t + ε it, (2) where R d,t is the hedge portfolio return in the LOW group at month t, R i,t is the hedge portfolio return in the HIGH group at month t. The t-statistics are computed using heteroskedasticityconsistent standard errors of White (1980). Table 3 presents the performance of anomaly-based strategies that are conditioned on quarterly institutional number changes and institutional value changes in Panels A and B, respectively. Each panel reports the average hedge portfolio returns and risk-adjusted returns of all anomalies for the LOW and HIGH groups, and the mean difference between the LOW and HIGH groups. In Panel A, for the average hedge portfolio returns, 13 out of 22 anomalies are significantly stronger in the LOW group than in the HIGH group. When averaged across all anomalies, the profit in the HIGH group decreases by 28% compared with the profit from pure 8 For anomalies constructed from the Compustat annual database, I use the quarterly changes in the number of active institutional investors to form quarterly-balanced portfolios. I get similar results by using the changes in the number of active institutional investors in June as the information for the whole year and form annual-balanced portfolios,. 11

13 anomaly-based strategies reported in Table 2. The combination strategy yields an average hedge portfolio return of 89 bps per month (t-statistic=8.09) in the LOW group, and 58 bps per month (t-statistic=5.07) in the HIGH group. The difference between the LOW and HIGH groups is 31 bps per month (t-statistic=4.60), which is statistically significant at the 1% level. The results for the risk-adjusted returns are even stronger. I find that 15 out of 22 anomalies are significantly stronger in the LOW group than in the HIGH group. When averaged across all anomalies, the profit in the HIGH group decreases by 34%, compared with the equallyweighted profit from pure anomaly-based strategies reported in Table 2. The combination strategy yields a risk-adjusted return of 83 bps per month (t-statistic=12.81) in the LOW group, and 50 bps per month (t-statistic=7.06) in the HIGH group. The difference between the LOW and HIGH groups is 33 bps per month (t-statistic=4.50), which is significant at the 1% level. I find similar results for the anomaly-based strategies conditional on quarterly institutional value changes. In Panel B, for the average hedge portfolio returns, 15 out of 22 anomalies are significantly stronger in the LOW group. When averaged across all anomalies, the profit in the HIGH group decreases by 33% compared with the profit from pure anomaly-based strategies presented in Table 2. In addition, in terms of risk-adjusted returns, 14 out of 22 anomalies are significantly stronger in the LOW group than in the HIGH group. When averaged across all anomalies, the risk-adjusted return of the HIGH group decreases by 38%, compared with the equally-weighted return from pure anomaly-based strategies reported in Table 2. The combination strategy yields a risk-adjusted return of 85 bps per month (t-statistic=11.73) in the LOW group, and 47 bps per month (t-statistic=6.63) in the HIGH group. The difference between the LOW and HIGH group is 37 bps per month (t-statistic=4.38), which is statistically significant at the 1% level. 12

14 Table 3 supports the first hypothesis that the profitability of anomaly-based strategies is significantly stronger following a decline in ownership breadth in the prior quarter. In addition, the anomalies are not fully eliminated after the growth in the number of active institutional investors and the market value of stocks held by active institutional investors. III. Sources of the Return Pattern To determine sources of the return pattern, I examine the risk-adjusted returns to the long and short sides for each anomaly across the LOW and HIGH groups in Table 4. A. Short Side Shorting plays a critical role in generating profits for anomaly-based strategies. The profits from anomaly-based strategies are mainly attributed to short leg portfolio. For instance, from Table A1 which describes returns to long and short portfolios for pure anomaly-based strategies, we can see that the profit of the combination strategy is from the short portfolio. The risk-adjusted return of the long portfolio is insignificant while the return of the short portfolio is significantly negative. I expect that the return pattern is primarily attributed to lower returns from short sides of the anomalies. From Panel A of Table 4, I find that the returns from the short side for each strategy are statistically significantly lower in the LOW group. The return differences between the LOW and HIGH groups from the short sides are statistically negative for all anomalies except momentum (MOM). Moreover, when averaged across all strategies, the profit from short side increases (decreases) by 82% (46%) in the LOW (HIGH) group compared with the equally-weighted profit 13

15 from short side of pure anomaly-based strategies reported in Table A1. 9 For example, for the short side, the combination strategy yields a risk-adjusted return of -102 bps per month (tstatistic=-8.30) for the LOW group, and -30 bps per month (t-statistic=-3.76) for the HIGH group, the difference between the two groups being -71 bps per month (t-statistic=-5.05), which is statistically significant at the 1% level. The results in Panel B of Table 4 deliver the same message as that in Panel A. As the market value held of stocks by active institutional investors decreases, the return from the short side is significantly lower. The differences of returns from the short side between the LOW and HIGH groups are statistically negative for all anomalies except momentum (MOM). In addition, when averaged across all anomalies, the profit from the short side increases by 80% in the LOW group and decreases by 32% in the HIGH group compared with the equally-weighted profit from the short side of pure anomaly-based strategies reported in Table 1A. For instance, for the short side, the combination strategy yields a spread of -101 bps per month (t-statistic=-6.81) for the LOW group, -38 bps per month (t-statistic=-3.42) for the HIGH group, the difference between two groups being -63 bps per month (t-statistic=-2.96), which is statistically significant at 1% level. B. Long side The profits of the anomaly-based strategies can come from the long side. It is possible that the higher returns from the long sides of anomalies contribute to higher returns in the low group. However, I do not find evidence to support it. In Panel A of Table 4, the returns from the long side of anomalies are significantly lower in the low group. The return differences between LOW and HIGH groups are statistically negative for each anomaly except short-term reversal 9 I present the returns to long and short portfolio of unconditional strategies in Table A1. 14

16 (STREV). For instance, the return from the long side of the combination strategy is -19 bps per month (t-statistic=-2.11) in the LOW group, and 20 bps per month (t-statistic=3.24) in the HIGH group. The difference between the LOW and HIGH groups is -39 bps per month (t-statistic=- 3.90), which is statistically significant at 1% level. The evidence show that the long sides of anomalies do not contribute to higher returns in the LOW group. However, I find that long sides contribute to profits of anomaly-based strategies in the HIGH group. 12 out of 22 anomalies earn statistically positive returns in the long side. For instance, the spread from the long side of the combination strategy is 20 bps per month (tstatistic=3.24). The positive spreads in the long side can partially explain that anomalies are not fully eliminated as the number of active institutional investors increases. From Panel B of Table 4, I find that the return differences between the LOW and HIGH groups are insignificant. Specifically, the spread from long side of the combination strategy is - 16 bps per month (t-statistic=-1.39) in the LOW group, 10 bps per month (t-statistic=1.18) in the HIGH group. The return difference between the LOW and HIGH groups is -26 bps per month (tstatistic=-1.63), which is insignificant. There are two explanations for the lower returns to the long side of anomalies in the LOW group. First, active institutional investors have skills to identify underpriced and overpriced stocks. They buy the underpriced stocks and sell the overpriced stocks. Second, changes in ownership breadth are positively correlated with future stock return (CHS (2002)). To confirm whether the ownership breadth has predicted power for future returns, I test the performance of strategies based on changes in the number of active institutional investors and the market value of stocks held by active institutional investors. I find that the changes in the number of active institutional investors have a strong positive predict power for future returns. The 15

17 strategy that long highest quintile and short lowest quintile yields a risk-adjusted equal-weighted return of 80 bps per month (t-statistic=4.20), and a risk-adjusted value-weighted return of 51 bps per month (t-statistic=3.51). However, the changes in market value of stocks held by active institutional investors have a weaker predict power for future returns. The strategy that long highest quintile and short lowest quintile only yields a risk-adjusted equal-weighted return of 66 bps per month (t-statistic=2.51), and a risk-adjusted value-weighted return of 20 bps per month (t-statistic=1.27). C. Short-sale Constraints and Managerial Skills Table 4 shows that the lower returns from short sides contribute to higher profits of anomalies for the low group. Two factors can explain the lower returns from short sides in the low group: (1) relaxation in short-sale constraints; and (2) superior managerial skills of active institutional investors. On one hand, CHS (2002) argue that reduction in ownership breadth would tighten the short-sale constraints. Stocks from the short side of anomalies in the low group should be more overpriced as the short-sale constraints bind more tightly and shorting is more difficult. Therefore, the future returns of anomalies would be lower in the low group. On the other hand, active institutional investors may have skill to identify mispriced stocks. They buy the stocks in the short side of anomalies because the stocks are underpriced. For instance, a stock was overpriced and the price went down by 20% in January. Active institutional investors realized that the price of this stock was below its fundamental value and bought this stock in February and then the price went up. Therefore, the stocks with increase in the number of active institutional investors have higher returns than those with decrease in the number of active 16

18 institutional investors. Unfortunately, I am unable to directly test the managerial skills hypothesis because the 13f data do not capture the intra-quarter institutional trading activities and the exact execution time of institutional trading. In addition, it is difficult to isolate the effect of relaxing short-sale constraints and superior managerial skills of active institutional investors. But I am be able to examine the short-sale constraints hypothesis. I use two proxies for short-sale constraints: size and total institutional ownership (Nagel, 2005) to divide stocks into short-sale constrained stocks and unconstrained stocks. If the return pattern is driven by tighter short-sale constraints, it is expected to be stronger for the short-sale constrained stocks and weaker or even nonexistent for the short-sale unconstrained stocks. I independently sort stocks into portfolios based on short-sale constraints proxies, institutional number changes and anomalies and then redo the tests conducted in Table 3. Table 6 reports the results for size and total institutional ownership in Panels A and B, respectively. The findings in Table 6 support the short-sale constraints hypothesis. First, Panel A shows that the return pattern is strong for the small stocks and disappears for the large stocks. For instance, for the combination strategy, the risk-adjusted return difference between the LOW and HIGH groups is 56 bps per month (t-statistics=4.66) for small stocks and 13 bps per month (t-statistic=1.30) for the large stocks. Second, Panel B shows that the return pattern is strong for the low institutional ownership stocks and weak for the high institutional ownership stocks. For instance, for the combination strategy, the risk-adjusted return difference between the LOW and HIGH groups is 65 bps per month (t-statistics=5.98) for the low institutional ownership stocks and 16 bps per month (t-statistic=1.72) for the high institutional ownership stocks. IV. New Strategies 17

19 I develop new trading strategies by combining changes in ownership breadth and 22 anomalies. Specifically, I long the stocks in the high-performing quintile (quintile 5) and the top ownership breadth changes tertile (tertile 3), and short the stocks in the low-performing quintile (quintile 1) and bottom ownership breadth changes tertile portfolio (tertile 1). As the SEC requires all institutional investors with greater than $100 million of securities under management to report their quarter-end holdings within 45 days after each calendar quarter, I left a two-month gap between the variables from institutional holding data and returns to ensure that these strategies are tradable. To directly compare the performance of strategies, I report the performance of pure anomaly based strategies in Table A2. I present both equal-weighted and value-weighted average hedge portfolio returns and risk-adjusted returns of the new strategies in Table 7. The evidence shows that these strategies outperform pure anomaly-based strategies. In Panel A of Table 7, when averaged across all anomalies, the equal-weighted average hedge portfolio return increases by 17 bps per month compared with the mean return from the pure anomaly-based strategies reported in Table A2; the equal-weighted risk-adjusted return increases by 20 bps per month compared with the mean risk-adjusted return from the pure anomaly-based strategies presented in Table A2. I find stronger results by forming value-weighted portfolios. When averaged across all 22 anomalies, the value-weighted average hedge portfolio return increases by 27 bps per month compared with the mean return from the pure anomaly-based strategies reported in Table A2; the value-weighted risk-adjusted return increases by 38 bps per month compared with the mean risk-adjusted return from the pure anomaly-based strategies presented in Table A2. For instance, the combination strategy yields a value-weighted risk- 18

20 adjusted return of 71 bps per month (t-statistic=5.88), compared with 33 bps per month (tstatistic=5.52) in Table A2. I find similar results in Panel B of Table 7. When averaged across all 22 anomalies, the equal-weighted average hedge portfolio return increases by 35 bps per month compared with the mean return from the pure anomaly-based strategies reported in Table A2; the equal-weighted risk-adjusted return increases by 40 bps per month, compared with the mean risk-adjusted return from the pure anomaly-based strategies presented in Table A2. In addition, when averaged across all anomalies, the value-weighted average hedge portfolio return increases by 26 bps per month compared with the mean return from the pure anomaly-based strategies reported in Table A2; the value-weighted risk-adjusted return increases by 30 bps per month compared with the mean risk-adjusted return from the pure anomaly-based strategies presented in Table A2. For instance, the combination strategy yields an equal-weighted mean return of 107 bps per month (tstatistic=6.64), compared with 72 bps per month (t-statistic=7.68) in Table A2. V. Robustness Check A. Value-weighted Portfolios To test whether the results are robust in value-weighted portfolios, I repeat the tests in Table 3. Table 8 reports the value-weighted risk-adjusted returns of 22 conditional strategies based on changes in ownership breadth and anomalies. The results are robust. For institutional number changes, 10 out of 22 anomalies are significantly stronger in the low group and the riskadjusted return differences between the LOW and HIGH groups are significantly positive at 5% or 1% level. Specifically, the combination strategy yields 63 bps per month (t-statistic=8.84) in the LOW group, and 23 bps per month (t-statistic=3.11) in the HIGH group. The difference 19

21 between the LOW and HIGH groups is 40 bps per month (t-statistic=4.77), which is statistically significant at 1% level. For institutional value changes, 11 out of 22 anomalies are significantly stronger in the LOW group. However, momentum (MOM) is weaker in the LOW group. The combination strategy yields 59 bps per month (t-statistic=7.14) in the LOW group, and 26 bps per month (t-statistic=3.53) in the HIGH group. The difference between the LOW and HIGH groups is 33 bps per month (t-statistic=3.52), which is statistically significant at 1% level. B. Size Anomalies are usually strongest among small stocks, and weakest among large stocks. It is possible that our results on changes in ownership breadth and anomalies are driven by small stocks. I repeat the analysis conducted in Table 3 using all-but-tiny stocks and large stocks. Table 9 reports the risk-adjusted returns of the 22 conditional strategies based on changes in ownership breadth and anomalies for all-but-tiny stocks and large stocks. In Panel A, I find that 13 out of the 22 anomalies are stronger in the LOW group for the all-but-tiny stocks; 11 out of the 22 anomalies are stronger in the LOW group for the large stocks. For the all-but-tiny stocks, the risk-adjusted return of the combination strategy is 64 bps per month (t-statistic=8.85) in the LOW group, and 29 bps per month (t-statistic=3.93) in the HIGH group. The difference between the LOW and HIGH groups is 34 bps per month (t-statistic=4.32), which is statistically significant at 1% level. For the large stocks, the risk-adjusted return of the combination strategy is 49 bps per month (t-statistic=6.08) in the LOW group, and 19 bps per month (t-statistic=2.23) in the HIGH group. The difference between the LOW and HIGH groups is 30 bps per month (tstatistic=3.38), which is statistically significant at 1% level. 20

22 The results in Panel B of Table 9 are even stronger. I find that 13 out of 22 anomalies are stronger in the LOW group for the all-but-tiny stocks; 11 out of 22 anomalies are stronger in the LOW group for the large stocks. For the all-but-tiny stocks, the risk adjusted return of the combination strategy is 71 bps per month (t-statistic=8.50) in the LOW group, and 28 bps per month (t-statistic=3.50) in the HIGH group. The difference between the LOW and HIGH groups is 44 bps per month (t-statistic=4.42), which is statistically significant at 1% level. For the large stocks, the risk adjusted return of the combination strategy is 55 bps per month (t-statistic=5.94) in the LOW group, and 21 bps per month (t-statistic=2.36) in the HIGH group. The difference between the LOW and HIGH groups is 34 bps per month (t-statistic=3.03), which is statistically significant at 1% level. C. The Carhart Four-factor Alpha I use the Carhart Four-factor model (Carhart, 1997) to calculate the risk-adjusted returns for each anomaly and then I reproduce the tests in Table 8. I run the following regression to obtain alpha for each anomaly: R t = α + β 1 MKT t + β 2 SMB t + β 3 HML t + β 4 UMD t + ε t, (3) Where UMD t is the return spread between winner and loser stocks at month t. I run the following to obtain the alpha difference across LOW and HIGH group for each anomaly: R d,t R i,t = α dif + β 1 MKT t + β 2 SMB t + β 3 HML t + β 4 UMD t + ε it, (4) 21

23 Table 10 reports the Carhart four-factor alphas for the conditional strategies based on quarterly changes in ownership breadth and anomalies. The return pattern is still robust. For the institutional number changes, 8 out of 22 anomalies are significantly stronger in the LOW group. The risk-adjusted return of the combination strategy yield a difference of 32 bps per month (tstatistics=3.51) between the LOW and HIGH groups, compared with 40 bps per month (tstatistic=4.77) in Table 8. For the institutional value changes, 8 out of the 22 anomalies are significantly stronger in the low group. The risk-adjusted return of the combination strategy yield a difference of 21 bps per month (t-statistics=2.16), compared with 33 bps per month (tstatistic=3.52) in Table 8. The lower magnitudes of the return difference imply that the momentum factor can explain the return patterns for some anomalies. For instance, the return patterns of the turnover (TO) and sales-price ratio (SP) disappear for the institutional number changes. D. Passive Institutional Investors I conduct the same tests as in Table 3 using changes in the number of passive institutional investors and changes in the market value of stocks held by passive institutional investors. Table 11 reports the risk-adjusted returns of anomalies for the LOW and HIGH groups and the mean difference between the LOW and HIGH groups. The return pattern is weaker but still statistically significant. For the institutional number changes, 9 out of the 22 anomalies are significantly stronger in the low group. For instance, the difference between the LOW and HIGH groups of the combination strategy is 24 bps (t-statistic=3.70), compared with 33 bps per month (tstatistic=4.50) in Panel A of Table 3. For the institutional value changes, 12 out of the 22 anomalies are significantly stronger in the low group. For instance, the difference between the 22

24 LOW and HIGH groups of the combination strategy is 32 bps (t-statistic=3.80), compared with 37 bps per month (t-statistic=4.38) in Panel B of Table 3. The results for the passive institutional investors support the short-sale constraints hypothesis, as the ownership breadth in principle should include all investors including active and passive institutional investors. E. Regression Analysis In this section, I examine quarterly changes in ownership breadth and anomalies in regression analysis. I run the following monthly cross-section regression: R i,t+1 = α + β 1 A i,t + β 2 D i,num + β 3 (D i,num A i,t ) + ε i,t, (5) where R i,t+1 is the return on stock i at month t+1, A i,t is the lagged firm characteristics (anomaly), D i,num is a dummy variable that takes value 1 if change in the number of active institutional investors is in the lowest tertile and zero otherwise. In addition, I include other firm characteristics that have predicted power for future returns as controls in the above regression: R i,t+1 = α + β 1 A i,t + β 2 D i,num + β 3 (D i,num A i,t ) + (controls i,t ) + ε i,t, (6) The control variables are: the logarithm of Amihud (2002) illiquidity, market capitalization, and book-to-market ratio. The parameter β 3 in the regressions captures the marginal effect of changes in breadth ownership on the relation between firm characteristics and future return. If a firm characteristic 23

25 is negatively correlated with future stock returns, a negative sign of β 3 implies that the predict power of future stock return for this firm characteristic is stronger following the decline in the ownership breadth in the last quarter. Similarly, If a firm characteristic is positively correlated with future stock returns, a positive sign of β 3 implies that the predict power of future stock return for this firm characteristic is stronger following the decline in the ownership breadth in the last quarter. I report the coefficient of interaction term and t-statistic of each anomaly in Table 12. The results from regression analysis are similar to those from portfolio sorts. For tests based on institutional number changes, 12 out of 22 anomalies are stronger following the decline in the number of active institutional investors. For instance, the β 3 of short-term reversal (STREV) is (t-statistic=-7.88) in regression (1). As short-term reversal is negatively related to future stock return, the significantly negative β 3 implies that short-term reversal is stronger following the declines in the number of active institutional investors. For tests based on institutional value changes, 13 out of 22 anomalies are stronger following the decline in the market value held by active institutional investors. For instance, the β 3 of growth profitability (GP) is 0.54 (tstatistic=2.71) in regression (1). As growth profitability is positively related to future return, the significantly positive β 3 implies that growth profitability is stronger following the decline in the number of active institutional investors. In addition, the results are robust after controlling for other firm characteristics that have predicted power for future returns. However, the β 3 of bookto-market (BM) becomes insignificant for the two measures of ownership breadth. VI. Conclusion 24

26 Whether institutional investors add value to correct market mispricing is a longstanding debate in the finance literature. I show that active institutional investors improve market efficiency. Specifically, I find that the profitability of 22 anomaly-based strategies is significantly stronger following a decline in the ownership breadth in the prior quarter. The return pattern remains robust after controlling for common risk factors including the Fama- French three factors and the momentum factor. My results are robust to different size groups, different portfolio weighting methods and cross-sectional regression tests. The return pattern is primarily attributed to the lower returns from the short side portfolio for each anomaly. I find that tighter short-sale constraints due to declines in the ownership breadth can explain the lower return from the short side portfolio. As the ownership breadth declines, the short-sale constraints are binding more tightly and overpricing is more difficult to be corrected. Therefore, the short portfolio is dominated by overpriced stocks which lead to lower returns. In addition to the short-sale constraint hypothesis, the stock picking skills of active institutional investors may also explain the lower returns in the short portfolio. Moreover, I develop new trading strategies that incorporate the changes in ownership breadth with 22 anomalies. The new strategies outperform the pure anomaly-based strategies. My tests have several limitations. First, the measure of active institutional investors includes passive mutual funds. Second, changes in ownership breadth have no significant effect on some anomalies such as momentum. These are left for future research. 25

27 References Abarbanell, Jeffery S., Brian J. Bushee, and Jana Smith Raedy, 2003, Institutional Investor Preferences and Price Pressure: The Case of Corporate Spin Offs, Journal of Business 76, Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2006, The cross section of volatility and expected returns, Journal of Finance 61, Asness, Clifford S., Tobias J. Moskowitz, and Lasse Heje Pedersen, 2013, Value and momentum everywhere, Journal of Finance 68, Asness, Clifford S., and Andrea Frazzini, 2013, The Devil in HML s Details, Journal of Portfolio Management 39, Asquith, Paul, Parag Pathak, and Jay Ritter, 2005, Short interest, institutional ownership, and stock returns, Journal of Financial Economics 78, Avramov, Doron, Tarun Chordia, Gergana Jostova, and Alexander Philipov, 2013, Anomalies and financial distress, Journal of Financial Economics 108, Basu, Sanjoy, 1977, Investment performance of common stocks in relation to their price earnings ratios: A test of the efficient market hypothesis, Journal of Finance 32, Basu, Sanjoy, 1983, The relationship between earnings' yield, market value and return for NYSE common stocks: Further evidence, Journal of Financial Economics 12, Bernard, Victor L., and Jacob K. Thomas, 1989, Post-earnings-announcement drift: delayed price response or risk premium, Journal of Accounting Research 27:1-36. Bhandari, Laxmi Chand, 1988, Debt/equity ratio and expected common stock returns: Empirical evidence, Journal of Finance 43, Brunnermeier, Markus K., and Stefan Nagel, 2004, Hedge funds and the technology bubble, Journal of Finance 59, Cao, C., Liang, B., Lo, A. W., and Petrasek, L., 2014, Hedge fund holdings and stock market efficiency, Board of Governors of the Federal Reserve System (US), Working paper. Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52, Chan, Kalok, and Wai-Ming Fong, 2000, Trade size, order imbalance, and the volatility volume relation, Journal of Financial Economics 57,