NET SHARE ISSUANCE, INSTITUTIONAL TRADING, AND STOCK MARKET RETURNS YINFEI CHEN

Transcription

1 NET SHARE ISSUANCE, INSTITUTIONAL TRADING, AND STOCK MARKET RETURNS By YINFEI CHEN A dissertation submitted in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY WASHINGTON STATE UNIVERSITY Carson College of Business MAY 2016 Copyright by YINFEI CHEN., 2016 All Rights Reserved

2 Copyright by YINFEI CHEN., 2016 All Rights Reserved

3 To the Faculty of Washington State University: The members of the Committee appointed to examine the dissertation of YINFEI CHEN find it satisfactory and recommend that it be accepted. George J. Jiang, Ph.D. Chair Jarl G. Kallberg, Ph.D. Sheen Liu, Ph.D. ii

4 ACKNOWLEDGEMENTS I am thankful to my doctoral committee members Dr. George Jiang, Dr. Jarl Kallberg, and Dr. Sheen Liu for their priceless guidance along my way. Foremost, I would like to express my special appreciation to my committee chair Dr. George Jiang, who paid endless patience and effort in my development, and help me grow as a researcher. You have been tremendous mentor for me. I am grateful to all faculty and staff members at the Department of Finance and Management Science, especially Dr. Gene Lai, Dr. Jarl Kallberg, Dr. DJ Fairhurst, Dr. Dallin Alldredge, Dr. David Whidbee and Dr. Michael McNamara, for their support and inspiration throughout my journey. Thank you to all my lovely colleagues: Chang, Hart, Angelina, Wei, Yoshi, Jeff, Bo, Guanzhong, Leila, Sanyong, Xun and Victor, among many others. I am thankful for our friendship. I will never forget the moments we had together. You guys let my doctoral study be an enjoyable experience. Lastly, a special thanks to my family. My parents, Xichong and Lina, supported me in all my pursuits. I thank my dear wife Vicky, who is always loving, caring, understanding, encouraging, and patient. Thank you. iii

5 NET SHARE ISSUANCE, INSTITUTIONAL TRADING, AND STOCK MARKET RETURNS Abstract by Yinfei Chen, Ph.D. Washington State University May 2016 Chair: George J. Jiang This dissertation contains two essays that study net share issuance, institutional trading and stock market returns. The first essay examines whether short-term institutions and short sellers exploit the net share issuance (NSI) effect. Existing literature documents that net share issuance significantly predicts cross-sectional stock returns, and yet institutional investors in aggregate trade in the wrong direction of the net share issuance effect. Motivated by the findings in the existing literature that short-term institutions and short sellers are better informed and more skilled, in this study I examine whether these investors exploit the net share issuance effect. I provide evidence that short-term institutions and short sellers indeed trade in the right direction of the net share issuance effect. On the other hand, firms exploit long-term institutions as liquidity provision in share issuance or repurchase. In addition, the results show that short sellers trade against institutional investors, especially long-term institutions, when exploiting the net share issuance. iv

6 In the second essay, I examine whether the cross-sectional NSI effect extends to the time-series level. I find that aggregate NSI negatively predicts future stock market returns, which is consistent with the cross-sectional NSI effect. The predicting power is statistically strong and economically large both in and out-of-sample. In addition, I provide evidence that the market-level net share issuance effect is stronger during periods with high investor sentiment. Moreover, aggregate net share issuance is not only positively related to contemporaneous and lagged stock market returns but also negatively related to analyst forecast errors. I interpret these findings as evidence that mispricing is more likely the source of the NSI effect. v

7 TABLE OF CONTENTS Page ACKNOWLEDGEMENTS... iii ABSTRACT... iv TABLE OF CONTENTS... vi LIST OF TABLES... ix LIST OF FIGURES... xi CHAPTER ONE: DO SHORT-TERM INSTITUTIONS AND SHORT SELLERS EXPLOIT THE NET SHARE ISSUENCE EFFECT? Introduction Data Descriptive Main Empirical Analysis The NSI Effect and Institutional Trading Trading of Different Institutions Trading of Short-Sellers Fama-MacBeth Regressions Performance of Institutional Trading Further Analysis The Effect of Reg FD Granger-Causality: NSI and Trading among Institutions and Short Sellers vi

8 1.4.3 Trading among Institutions and Short Sellers Conclusion Appendix Table A1: High/Low Spread of Average Return across NSI Quintiles: Size Subsamples Table A2: Fama-MacBeth Regressions of Stock Returns on NSI Table B: Average Changes in Institutions across NSI Deciles Table C: High-low Spread of Change in Institutional Ownership and Change in Institutional Portfolio Weight Table D: Vector Auto-Regressions of Stock Returns, Change of Institutional Ownership, Net Share Issuance and Change of Relative Short Interest References CHAPTER TWO: AGGREGATE NET SHARE ISSUANCE AND STOCK MARKET RETURNS: A TIME-SERIES ANALYSIS Introduction Data and Variable Construction Predictability of Aggregate NSI Preliminary Analysis Univariate Regression Multivariate Regression The NSI Effect during Different Investor Sentiment Periods Out-of-sample Results vii

9 2.6 Further Analysis on the Explanation of the Aggregate NSI Effect Contemporaneous Market Returns, Lagged Market Returns and Aggregate NSI Analysts Forecasts Error and Aggregate NSI Conclusions References viii

10 LIST OF TABLES Table 1-1 Summary Statistics of Firm Characteristics Table 1-3 NSI Effect and Institutional Trading Different Types of Institutions Table 1-4 NSI Effect and Trading of Short Sellers Table 1-5 NSI Effect and Trading of Different Types of Institutions and Short Sellers Fama-MacBeth Regression Table 1-6 Performance of Institutional and Short Seller Trading Table 1-7 NSI Effect and Trading of Different Types of Institutions and Short Sellers Effect of Reg FD Table 1-8 Granger-Causality Analysis Table 1-9 Trading among Different Parties Table 2-1 Summary Statistics Table 2-2 Univariate Regression of Market Returns on Net Share Issuance Table 2-3 Multivariate Regressions of Market Returns Table 2-4 Univariate Regression of Market Returns on Net Share Issuance during Low and High Investor Sentiment Periods Table 2-5 Multivariate Regression of Market Returns on Net Share Issuance during Low and High Investor Sentiment Periods Table 2-6 Out-of-Sample Forecasts of the Univariate Model ix

11 Table 2-7 Out-of-Sample Forecasts from Multivariate Regression Table 2-8 Univariate Regressions of Aggregate NSI on Stock Market Returns Table 2-9 Net Share Issuance, Contemporaneous Market Returns, and Lagged Market Returns Table 2-10 Net Share Issuance and Analyst Forecasts Error Table 2-11 Net Share Issuance and Analyst Forecasts Error during Low and High Investor Sentiment Table 2-12 Correlations between Market Excess Returns, NSI and Other Variables during Low and High Investor Sentiment Periods Table 2-13 Regression of Market Returns on Net Share Issuance and Interaction Term Table 2-14 Multivariate Regressions of Market Returns and Interaction Term x

12 LIST OF FIGURES Figure 2-1 Aggregate Net Share Issuance Figure 2-2 Investor Sentiment Index Figure 2-3 Market Excess Returns of Different Net Share Issuance Groups Figure 2-4 Market Excess Returns of Different Net Share Issuance Groups during Low and High Investors Sentiment Periods Figure 2-5 Equal-weighted and Value- weighted Market Excess Return xi

13 Dedication To Vicky L. Chen, my love xii

14 CHAPTER ONE DO SHORT-TERM INSTITUTIONS AND SHORT SELLERS EXPLOIT THE NET SHARE ISSUENCE EFFECT? 1.1 Introduction Net share issuance (NSI) is defined as the net change in shares outstanding over a given time period. Existing literature documents that net share issuance is a strong predictor for crosssectional stock returns (Daniel and Titman, 2006; Pontiff and Woodgate, 2008; and Fama and French, 2008a). For instance, Pontiff and Woodgate (2008) shows that a one-standard deviation increase in net share issuance is associated with a 0.33% decrease in monthly cross-sectional stock return. That is, the firms with low net share issuance significantly outperform the firms with high net share issuance over both long and short horizons. This anomaly is referred to as the net share issuance effect. Yet, in a recent comprehensive empirical study, Edelen, Ince and Kadlece (2014) document that institutional investors, in aggregate, do not exploit the NSI effect. They show that institutions actually trade in the opposite direction of the NSI effect. Namely, institutions increase their holdings on stocks with high NSI and decrease their holdings on stocks with low NSI. Further, they show that portfolios formed with stocks where institutional trading is opposite to the NSI anomaly significantly outperform portfolios formed with the stocks where institutional trading is consistent with the NSI anomaly. The main research question of our study is whether short-term institutions and short sellers exploit the NSI effect. Our study is motivated along the following dimensions. Several recent studies examine the robustness of NSI anomaly and show that NSI is one of the most robust stock 1

15 return predictors. Fama and French (2008b) find that the NSI effect is pervasive across all size groups. Jiang and Zhang (2013) find that the NSI effect is significant in both long and short sides of hedge portfolios. Drechsler and Drechsler (2014) show that the NSI effect is prevalent among all short-sale fee (both high and low) sub-groups of stocks. In addition, compared to the predictability of equity issuance events such as SEOs, repurchases and stock mergers, the NSI effect reveals a more general association between equity issuance and stocks future returns. Pontiff and Woodgate (2008) conducts a comprehensive analysis on the relation between future stock returns and NSI. They remove returns related to SEOs, repurchases and stock mergers and still find the significant predictability of NSI. Second, NSI contains important information about firm s fundamentals, corporate policy and stock valuation. Following the rational explanation of the NSI effect, NSI contains information related to the future risk of firm. Daniel and Titman (2006) argue that the exercise of real option following equity issues reduces firm s uncertainty and marginal product of current investment. Thus, NSI captures the information about the change in firm s risk. It s also widely believed that issuing firms tend to have larger investments relative to earnings while the opposite is true for repurchasing firms (Fama and French, 2005; Li, Livdan and Zhang, 2009). Therefore, NSI also conveys crucial information about firms investment policy to the investors. Moreover, Loughran and Ritter (1995), Ikenberry, Lakonishok and Vermaelen (1995), and Daniel and Titman (2006) argue that managers tend to issue when stocks are overvalued and tend to repurchase when stocks are undervalued. Such behavioral explanations of the NSI effect suggest that NSI contains information about the valuation of stocks. Given that NSI contains firms information related to 2

16 investment policy, future growth options, and valuation, it is expected that informed investors have incentive to trade on such information accordingly. Third, while institutions are generally considered as the informed investors (Gompers and Metrick, 2001; Campbell, Ramadorai and Schwartz, 2009), some recent literature provides evidence that the ability of exploiting information varies among different types of institutions. Bushee (2001) shows that the short-term (transient) institutions change their holding according to firms disclosure more efficiently than the long-term institutions. The given explanation is that the transient institutions have stronger incentives to gather private information because they engage in strategies to profit short-term price appreciation. Several other studies also find that the shortterm institutions are more sophisticated in terms of exploiting the anomaly information. For example, Ali, Hwang and Trombley (2000) find that, institutions, in aggregate, do not exploit the accruals information, however, Balsam, Bartov and Marquardt (2002), Collins, Gong and Hribar (2003) and Lev and Nissim (2006) show that the short-term institutions are able to exploit the accruals information. Barone and Magilke (2009) argue that the degree of sophistication plays a role in exploiting the mispricing information implied in accruals. Further, Yan and Zhang (2009) provide the evidence that short-term institutions trading is significantly correlated to future earnings surprise while long-term institutions trading is not, and predictive power by institutional ownership on stock returns is driven by the short-term institutions trading. Motivated by the above studies, we examine whether the short-term institutional investors are able to exploit the NSI effect. Finally, short sellers are widely believed to be informed and sophisticated. Extant studies show that short sellers trading correctly predicts future negative abnormal returns (see e.g. Diether, Lee and Werner, 2009; Boehmer, Jones and Zhang, 2008). Moreover, Karpoff and Lou (2010), 3

17 Christophe, Ferri and Angel (2004), and Boehmer, Jones and Zhang (2008) provide evidence that short sellers trade before the public information is released and earn significant abnormal returns. In the meantime, Asquith, Pathak and Ritter (2005), Engelberg, Reed and Ringgenberg (2012), Boehmer and Wu (2013) s work infers that short sellers do not possess private information and their trading advantage comes from their superior ability to analyze public information. More shorting flow accelerates the incorporation of public information into stock price. Even though the two sets of studies show different information sets that short sellers exploit, both indicate that short sellers are sophisticated in information processing. We develop our hypotheses based on two important roles for institutional investors and short sellers: the information production role suggested by Chemmanur and Jiao (2005) and the manipulative role suggested by Gerard and Nanda (1993). Institutions engage in the information production role if they consistently trade in the same direction of the information they obtain and earn positive abnormal returns. Chemmanur and Jiao (2005) show that institutions increase their holdings on stocks both before and after share offerings if they obtain favorable information. Gibson, Safieddine and Sonti (2004) look at the institutional trading in the pre-seo market. They find that institutions buy shares in pre-seo market when they have favorable private information about the firm. Also, Chemmanur, He and Hu (2009) state that institutions trade in the same direction as their private information in both pre- and post SEO and earn significant higher return than the naïve buy-and hold trading strategy. These findings all show that institutional investors play the information production role through their trading. On the other hand, Gerad and Nanda (1993) introduce a model showing the potential manipulative role of the institutional investors, that is, institutional investors manipulate the offering price by trading in the opposite direction as 4

18 the information before the share offering and subsequently profit by trading in the same direction as the information after the share offering. For short sellers, existing literature shows evidence that short sellers sometimes play a manipulative role in exploiting the certain types of information. Henry and Koski (2010) show that short sellers manipulate the offering price around SEOs and profit from the manipulative trading. In this study, we investigate whether institutional investors and short sellers play a manipulative role or an information product role when they trade on the NSI effect. We also examine the implications of agency problem on the role of institutional investors in the information production. We obtain the institutional holding data from Thompson-Reuters 13F database. The sample period is from 1980 to In addition, we obtain information of security from the Center for Research in Security Prices (CRSP) Monthly Stocks File for NYSE, Amex, and Nasdaq stocks. We perform the analysis on institutional investors classified by the measure used in Bushee (2001). The institutional investors are classified into three categories: transient, quasi-indexing and dedicated institutions. Transient institutions are defined as short-term investors who focus on maximizing the short-term profit. Quasi-indexing institutions are the ones who focus on long-term investment with diversified portfolio. Dedicated institutions are the long-term investors with concentrated portfolio. We measure the trading of short sellers using firm s change in relative short interest. Our data of short interest is obtained from COMPUSTAT database. To examine how institutional investor trade on the net share issuance effect, we sort stocks into deciles based on their net share issuance, and we calculate the average changes in institutional ownership and the average changes in institutional investors portfolio weight for each decile. Our results are in line with Edelen, Ince and Kadlece (2014) that NSI is negatively related with the 5

19 change in institutions ownership. We further show that this relation is mainly driven by long-term investors, namely, quasi-indexing and dedicated institutions. Quasi-indexing and dedicated institutions trade in the opposite direction of the NSI effect. More importantly, we find that shortterm (transient) institutions are able to exploit the NSI effect and adjust holdings to take advantage of the NSI effect. Further, our results show that the changes of relative short interest in stocks with the lowest NSI are significantly lower than that of the stocks with the highest NSI, suggesting that short sellers take advantage of the NSI effect. Existing studies (Gompers and Metrick, 2001; Bennett, Sias and Stark, 2003) show that institutional demand is related to the certain firm characteristics. We perform the Fama-Macbeth regressions of the changes in institutional ownership and the changes in short interest on NSI, controlling for the firm s characteristics. The results show that even after controlling for the firm characteristics, there is still a significantly positive relation between NSI and the changes in institutional ownership for quasi-indexing and dedicated institutions. The results for transient institutions show an insignificant relation between NSI and the changes in institutional ownership since quarter t. Thus, short-term institutions exploit the NSI effect starting from the quarter that NSI information is available. The positive relations between NSI and short interest are significant in both quarter t+1 and t+2. That is consistent with our earlier finding that short sellers exploit the NSI effect starting from one quarter after NSI information is available. As noted earlier, institutions and short sellers may manipulate information in their trading. To address this question, we examine the performance of trading for three types of institutional investors and short sellers. Our analysis is under the premise that if investors play a manipulative 6

20 role, they may trade against the information but realize positive abnormal return. Our results show that transient institutions and short sellers earn significant positive abnormal returns in quarter t+1 and t+2. All the significant abnormal returns are from their trading on the short side of the anomaly. Given that transient institutions and short sellers trade on the NSI effect, our findings support the hypothesis that short-term institutions and short sellers play an information production role. The results for quasi-indexing insitutionsshow that they don t earn significantly positive abnormal return. For dedicated institutions, they can earn positive returns but the result is much weaker than transient institutions and short sellers. Thus, there is no strong evidence that long-term institutions play a manipulative role either. While NSI is public information, it s hard to argue that institutions do not exploit the NSI effect because they do not know the information. In other words, institutions intentionally ignore NSI information and sacrifice higher returns for other interest. One possible interest is maintaining the channel of obtaining private information from firms. While institutions, especially for dedicated institutions, are usually provided private information by firms, institutions will not trade against the firms by exploiting the private information, otherwise, firms will turn off the channel of providing private information and institutions will lost the interest of obtaining private information. Thus, institutions should not have incentive to exploit the NSI effect if they are provided private NSI information by firms. If institutions do not trade to earn higher returns for their clients but trade for other interest, by definition, it is the agency problem of institutions. Regarding of this, we argue that better access to private information may deter institutional investors and short sellers from exploiting NSI information, and we expect that institutions and short sellers perform better 7

21 in terms of exploiting the NSI effect while they have worse access to private information about NSI. To further investigate the effect of information environment on the role of information production or information manipulation by institutional investors and short sellers, we use the introduction of Regulation of Fair Disclosure (Reg FD) as a natural experiment. The introduction of Regulation of Fair Disclosure in August 2000 prevents the equity issuer from disclosing any private information about issuance to any certain group of investors and thus provides us different information environment in pre-reg FD and post-reg FD periods. According to the extensive literature (see, Heflin, Subramanyam and Zhang, 2003; Ke, Petroni and Yu, 2008; Li, Radhakrishnan, Shin and Zhang, 2011), Reg FD significantly reduces institutions and short sellers advantage on exploiting private information. In the pre-reg FD period, institutional investors and short sellers may have better access to private information about NSI. According to our previous discussion, we expect that institutional investors and short sellers would perform better in terms of exploiting public information in the post-reg FD period, especially for dedicated institutions who are usually have close relationship to firm s management with relationship investment. In the analysis, we separate our whole sample into two sub-samples: the pre- and the post- Reg FD periods, and we compare the performance of trading of institutional investors and short sellers in the two sub-periods. Our results show that in the pre-reg FD period, quasi-indexing institutions don t earn significant abnormal return after the NSI month. For dedicated institutions, in the pre- Reg FD period, they don t earn significant abnormal returns after the NSI information is available to the public. However, in the post-reg FD, they are able to earn significantly positive abnormal returns. This finding is consistent with our hypothesis that certain institutions perform better in 8

22 terms of exploiting public NSI information in the post-reg FD period. While dedicated institutions may ignore private information on purpose, we consider it as a potential agency problem. Moreover, we find that transient institutions and short sellers have better performance in the pre- Reg FD periods, suggesting that they are more informed in the pre-reg FD period when they have better access to the private NSI information. Note that short-term institutions and short sellers exploit the NSI effect and long-term institutions do not, we are interested in the potential causal effect among NSI, institutions trading, short sellers trading and firms equity activities. We perform the Granger-Causality test to investigate the causality among stock returns, NSI, change in institutional ownership and change in relative short interest. Our results show that firms NSI is caused by stock returns. This finding is consistent with the behavioral explanation of the NSI effect in Loughran and Ritter (1995), Ikenberry, Lakonishok and Vermaelen (1995) and Daniel and Titman (2006): managers tent to issue stocks while stocks are overvalued and tent to buy back stocks while stocks are undervalued. Further, we observe that stock returns cause change in institutional ownership while the opposite is also not true. This results confirm Badrinath and Wahal (2002) s finding that institutions are momentum traders. We also find that institutional ownership is the cause of firms equity decisions. This finding is in line with Alti and Sulaeman (2012) that high institutional demand is a necessary condition for firms to trigger SEOs. Interestingly, the change in short interest is caused by the institutions trading but not NSI. This is the evidence that short sellers may not exploit the NSI effect directly, but simply take advantage of the trades by institutional investors who don t exploit the NSI effect. Lastly, we examine the trading among different types of institutions and short sellers, and we observe that short sellers always trade in the opposite direction to institutional 9

23 investors. Further, we find that long-term institutions always buy shares when firms issue and sell shares when firms buy back. This is the further evidence that short sellers do not exploit the NSI effect directly. Meanwhile, firms exploit the long-term institutions as liquidity providers. The rest of the paper is structured as follows. We describe the data and variable construction in the section 1.2. In the section 1.3, we present our main empirical results. The section 1.4 performs further analysis. We conclude in the section Data Descriptive The main data used in our study is the holding of institutional investors. According to the Securities and Exchange Act, all institutions with greater than $100 million of securities under discretionary management are required to file Form 13F and report their holdings to the Security and Exchange Commission (SEC). In this paper, we define institutional investors as the institutions that file Form 13F. We obtain the institutional holding data from Thompson-Reuters 13F Holding database. Thompson-Reuters 13F collects institutional holding information starting from 1980 on quarter basis. Thus, we extract quarterly institutional holdings starting in the first quarter of 1980 and ending in the last quarter of Besides, we obtain stock information (return, price, date, etc.) from the Center for Research in Security Prices (CRSP) Monthly Stocks File for NYSE, Amex, and Nasdaq stocks. To ensure our data quality, we exclude observations with stock price lower than $5. For matching the quarterly institutional holding data, we convert the monthly data into quarterly data. This leaves us with 462,052 firm-quarter observations. As explanatory variables in regressions, we use stock price (PRC), market capitalization (Size), momentum (MOM), book-to-market (B/M), idiosyncratic 10

24 volatility (IVOL), shares turnover (TURN), relative short interest (RSI), institutional ownership (IO) and net share issuance (NSI). Stock price (PRC) is the stock price at the end of each quarter. Table 1-1 reports the summary statistics for our explanatory variables for selected years. Firm s market capitalization is calculated as firm s shares outstanding times the stock price. Momentum (MOM) is the cumulative return for the past 12 months at the end of each quarter. Book-to-Market ratio (BM) is the log term of the ratio book equity over market equity. The book equity is obtained from COMPUSTAT. Following Fama and French (1993), BM is calculated in the end of June of every year. Idiosyncratic volatility (IVOL) is defined as the quarterly standard deviation of daily stock return residuals from the Fama and French 3-factor model. Shares turnover (TURN) is the number of shares traded as reported on CRSP, divided by shares outstanding each month and averaged within each quarter. Number of shares traded for NASDAQ is adjusted by a factor of.5. Relative short insterest (RSI) is calculated as the number of shares held short as reported on COMPUSTAT, divided by the shares outstanding and averaged within the quarter. We calculate firm s institutional ownership (IO) as dividing the sum of all reported institutional holding shares by the firms shares outstanding in the same quarter. Stocks are assumed to have zero institutional holding if they are reported in CRSP but not in 13F. While the abnormal return is defined as the difference between the raw return and benchmark return, we use the benchmark portfolio returns proposed by Daniel, Grinblatt, Titman and Wermers (1997). In Bushee (2001) s data, institutions are classified into three categories, according to the characteristics of their trading: transient, quasi-indexer and dedicated. Transient institutions are defined as short-term investors who focus on maximizing the short-term profit. Quasi-indexers are 11

25 the institutions who focus on long-term investment with diversified portfolio. Dedicated institutions are the long-term investors with concentrated portfolio. In the analysis on different institution types, we use such classification of institution types in Bushee (2001). Data in Bushee (2001) starts from Thus, our sample period for this analysis is from January 1981 to June Following the previous studies (see, e.g., Daniel and Titman 2006; Pontiff and Woodgate, 2008; Fama and French 2008), we define the NSI as the net change of the log term of shares outstanding over the past 12 months. Table 1-1 also reports the summary statistics of NSI for the selected years. We construct the measure of net share issuance at the end of each month t as ISSUE = Ln (Adjusted Sharest) Ln (Adjusted Sharest-12),where we compute the real number of shares outstanding adjusted for splits and other shares distribution events as the product of cumulative factor to adjust shares outstanding and with the number of shares outstanding: Adjusted Sharest = Shares Outstandingt * the cumulative factor to adjust shares outstanding For matching our quarterly 13F data, we convert the monthly NSI into quarterly data by keeping the NSI in the last month of each quarter. Table 1-1 reports summary statistics of firm characteristics for the last quarter of selected years: 1980, 1990, 2000, 2010 and For each quarter, we report the log term of firm size (SIZE), the log term of book-to-market ratio (B/M), momentum (MOM), idiosyncratic volatility (IVOL), stock price (PRC), shares turnover (TURN), relative short interest (RSI), net share issuance (NSI) and institutional ownership (IO). Some firm s characteristics see increases overtime in the crosssectional mean. Firm size monotonically increases from 1980 to The average relative short 12

26 interest continue increasing from 1% in 1980 to 4.00% in The average institutional ownership significantly increases from 20% in 1980 to 62% in Other firm s characteristics also see variation overtime. For example, the mean of the log term of BM decreases from in 1980 to in 2000 then increases to in Mean IVOL increases from 0.02 in 1980 to 0.04 in 2000, but it drops back to 0.02 in This is consistent with Campbell, Lettau, Malkiel and Xu (2001) s finding that firm-level volatility increases dramatically from 1980 to Mean net share issuance varies overtime. It increases from 3.00% in 1980 to 6.00% in 2000 and drops to 2.00% in Main Empirical Analysis The NSI Effect and Institutional Trading We revisit the anomaly net share issuance. Table 1-2 reports the net share issuance effect. At the end of each quarter t (Qt), stocks are sorted into decile based on NSI. For each decile, we calculate the equal-weighted (Panel A) and value-weighted (Panel B) quarterly average abnormal returns for quarter t-3 (Qt-3) to t+4 (Qt+4) and the cumulative abnormal returns of each decile for year 2 ([Qt+5, Qt+8]). The high-low spreads are differences of the returns between the two extreme deciles. Our value-weighted result shows that in quarter t+1, t+2, t+3 and t+4, the high-low spreads are -2.15%, -1.96%, -2.30% and -2.33%, respectively. The negative and significant spreads indicate that stocks with highest NSI earn significant lower abnormal returns than stocks with lowest NSI. The high-low spread for year 2 shows that the portfolio with highest NSI on average earns 8.48% lower cumulative abnormal return than the portfolio with lowest NSI. Thus, our result provides evidence that the NSI effect is significant and persistent for at least two years. The equalweighted result shows the same pattern as the value-weighted result. Jiang and Zhang (2013) 13

27 shows the NSI effect on portfolio s both long and short side returns, and Drechsler and Drechsler (2014) shows that the NSI effect is robust across short-sale fees groups. In our value-weighted result, for quarter t+1 to t+4 and for year 2, the top decile earns significant higher abnormal returns than the median decile (D5). Thus, we provide evidence that the NSI effect is driven by both sides of the hedge portfolio and cannot be driven by the short-sale constrain on the short side of the portfolio. In regard of this, our results support the findings in Jiang and Zhang (2013) and Drechsler and Drechsler (2014). Fama and French (2008b) shows that the NSI effect is robust across size groups. We further examine the robustness of the NSI effect across the size groups. Our result shown in appendix table A1 suggests that the NSI effect is robust in all size groups. In appendix table A2, we report the coefficients of Fama-MacBeth regressions of future returns on net share issuance and other return-predictive variables. We find the significant negative relation between stock future returns and NSI both before and after we control the firm s characteristics. Thus, our results imply that NSI is a strong, persistent and robust predictor for cross-sectional future stock returns Trading of Different Institutions While the NSI effect is strong, persistent and robust, Edelen, Ince and Kadlec (2014) find that institutional investors in aggregate trade in the opposite direction of the NSI effect. We use a different method to analyze the institutional investors trading on the net share issuance and also find the evidence that institutional investors as a whole trade against the NSI effect. Using the portfolios we constructed before, we calculate the equal-weighted and value-weighted quarterly average change in institutional ownership of each portfolio. Table B in appendix reports the result of the analysis. The high-low spread shows the difference in the change in institutional ownership 14

28 between the short-leg (bottom) portfolio and log-leg (top) portfolios. If the institutional investors exploit the NSI effect, the change in the institutional ownership in the top portfolio should be greater than that in the bottom portfolio, and thus the high-low spread should be negative. The equal-weighted result (Panel A) shows that for quarter t+1 to t+4 and for year 2, the high-low spreads are all positive and significant. Similar to Edelen, Ince and Kadlec (2014), our results provide evidence that institutions in aggregate increase more holding on the stocks with high NSI than on the stocks with low NSI, namely, institutional invests trade in the opposite direction to the NSI effect. For value-weighted results (Panel B), we don t find strong evidence that institutions trade against the NSI effect except for quarter t+1. Thus, institutional investors trade in opposite direction to NSI effect mainly in small firms. Moreover, we calculate the change in institutional investors portfolio weight. Change in holdings is part of the causality of change in weight. The change in institutional investors portfolio weight can be attributed to the change in institutional holdings, or to the change of stock prices, or both. The change in weight driven by the change in holdings could be considered as active change while the change in weight driven by the change in stock prices could be considered as passive change. Thus, analyzing on both change in investors holdings and weight helps us identify whether institutional investors trade to adjust their portfolio weight based on NSI. Our result shows that institutions trade against the NSI effect in quarter t+1, t+2 and t+4. Overall, our finding is consistent with Edelen, Ince and Kadlec (2014) that institutional investors trade in the opposite direction of the NSI effect. It s documented in the existing literature that short-term institutions are more sophisticated in terms of exploiting anomaly information (Bushee, 2001; Ali, Hwang and Trombley, 2000; Balsam, Bartov and Marquardt, 2002; Collins, Gong and Hribar, 2003; Lev and Nissim, 2006; Barone and 15

29 Magilke, 2009; Yan and Zhang, 2009). Based on findings in previous research, we further analyzes whether short-term institutional investors exploit the NSI effect. Using the classification proposed in Bushee (2001), we classify institutional investors into three types: transient, quasiindexers and dedicated institutions. Table 1-3 shows the analysis on the trading of three types of institutions. For each type of institutions, we calculate the equal-weighted and value-weighted quarterly average change in institutional ownership and also the change in institutional investors weight across the NSI portfolios we constructed before. The high-low spread is the difference of the change in institutional ownership between bottom and top decile. Panel A, B and C show the result for quasi-indexers, dedicated and transient institutions, respectively. In Panel A, the equalweighted average high-low spreads of quasi-indexers institutions drops from 0.83% in quarter t+1 to 0.28% in year 2. The spreads are all positive and significant, which means quasi-indexers institutions increase significantly more holding on the stocks in the bottom decile than on the stocks in the top decile. The value-weighted results and the results for institutional investors portfolio weights are similar to the equal-weighted result. Such results show that quasi-indexers institutions trade in the opposite direction of the NSI effect. Specifically, for quarter t+1 to year 2, quasi-indexers institutions decrease their holding in the top decile while increase holding in the bottom decile. The amounts of increase in the bottom decile is much greater than the amounts of decrease in the top decile. This finding indicates that quasi-indexers institutions trade oppositely to the NSI effect at both sides of portfolio while the opposite trading to the NSI effect is mainly driven by buying the stocks on the short side. So, we can t attribute transaction frictions to the opposite trading to the NSI effect. 16

30 Panel B shows the results for dedicated institutions. The equal-weighted results show that from quarter t+1 to t+4, the high-low spreads are positive and significant. Thus, similar to quasi-indexers institutions, dedicated institutions trade in the wrong direction of the NSI effect. However, the significance of the high-low spreads is gone in the value-weighted results. So, for dedicated institutions, opposite trading to the NSI effect only happens in the small firms. The insignificant high-low spreads for the change in institutional investors weight provide further evidence that dedicated institutions actively trade against the NSI effect. Given that quasi-indexers and dedicated institutions are both long-term institutions, we give the argument that long-term institutions trade in the opposite direction to the NSI effect. Panel C shows the results for the trading of transient institutions. The equal-weighted results show that in quarter t, transient institutions switch their trading on the top and the bottom deciles. From quarter t+1 to t+4, transient institutions decrease their holdings on the bottom decile in which stocks have highest NSI. Meanwhile, they increase their holdings on the top decile in which stocks have lowest NSI. Thus, our results show that transient institutions trade accordingly to the NSI effect on both long-side and short-side of the hedge portfolio. Regarding the amount of changes in holdings of the bottom and top decile, the significantly negative high-low spreads are driven the selling on the short side. The significantly negative high-low spreads last for at least two years. The result for value-weighted change in institutional ownership is similar. Given that transient institutions are short-term investors, our results provide evidence that short-term institutional investors exploit the NSI effect. 17

31 1.3.3 Trading of Short-Sellers The extant studies show that short sellers predict future stock returns correctly (Diether, Lee and Werner, 2009; Boehmer, Jones and Zhang, 2008; Karpoff and Lou, 2010; Christophe, Ferri and Angel, 2004; Boehmer, Jones and Zhang, 2008). On the other hand, some other work shows that short sellers have superior ability on processing information (Asquith, Pathak and Ritter, 2005; Engelberg, Reed and Ringgenberg, 2012; Boehmer and Wu, 2013). While short sellers are widely considered as sophisticated investors, we analyze whether short sellers exploit the NSI effect. Table 1-4 shows short sellers trading on the NSI effect. Our results show evidence that short sellers trade in the same direction to the NSI effect. The equal-weighted results (Panel A) show that from quarter t+1 to t+4, short sellers decrease short interest in the portfolio with lowest NSI and increase short interest in the portfolio with highest NSI. The high/low spreads are significantly positive from quarter t+1 to t+4. The results imply that short sellers exploit the NSI effect in one quarter after NSI information is available and keep exploiting in the subsequent three quarters. The value-weighted results are much weaker and we don t see significantly positive high-low spreads from quarter t+1 to year 2. Thus, the short sellers take advantage of the NSI effect only on small firms. The insignificant high-low spreads of the change in short sellers portfolio weight provide further evidence that short sellers actively trade on the NSI effect Fama-MacBeth Regressions Gompers and Metrick (2001), Bennett, Sias and Stark (2003) content that institutional demand is related to certain firm characteristics. They show evidence that institutional investors prefer firms with large size and high liquidity. Motivated by their discussion, we control firm characteristics and investigate the relationship between the trading of institutional investors and 18

32 NSI. Table 1-5 reports the coefficients of Fama-MacBeth regressions of changes in different types of institutions ownership on the net share issuance and other firm s characteristics. The firm s characteristics are firm s size, book-to-market value, momentum, stock price, idiosyncratic volatility, shares turnover, lagged return and relative short interest. NSI is the net share issuance at the end of quarter t. All the control variables are one quarter lagged to the corresponding dependent variable. Panel A shows the result for quasi-indexers institutions. We find the significant positive correlation between NSI and the change in institutional ownership from quarter t-1 to t+2. The result is consistent with our earlier finding that quasi-indexers institutions trade in the opposite direction to the NSI effect. In addition, the results also show that quasi-indexers institutions trade on certain firm s characteristics. For example, quasi-indexers institutions prefer low book-tomarket (growth) stocks. IVOL is negatively related to quasi-indexers institutional demand overtime, which means quasiindexers institutions tend to buy firms with low firm-level risk. Panel B shows the result for dedicated institutions. After controlling the firm s characteristics, the relationship between NSI and the change in dedicated institutional ownership is significantly positive for quarter t+1 and t+2. The results are consistent with our previous finding that dedicated institutions trade in the wrong direction of the NSI effect. Similar to quasi-indexers institutions, dedicated institutions tend to acquire growth stocks, and they are in favor of the stocks with low idiosyncratic risk. Panel C shows the result for transient institutions. After controlling on the firm s characteristics, NSI is significantly related to the change in transient institutions ownership in quarter t+1 and t+2. 19

33 Besides, transient institutions trade on all kinds of information embedded in firm s characteristics such as momentum, stock price, shares turnover and lagged returns. Transient institutions show different preference on the stocks from quasi-indexers and dedicated institutions. For example, transient institutions tend to hold high book-to-market (value) stocks. While transient institutions strongly tend to buy the stocks that perform well in the last quarter. Our finding provide further evidence to Bushee (2001) s argument that transient institutions trade for short-term profits. Overall, our results for regressions confirm our previous finding that long-term institutions trade in the opposite direction of the NSI effect while short-term institutions trade accordingly. Panel D shows the results for short sellers. In quarter t+1 and t+2, the coefficient between NSI and the change in short interest is significantly positive after controlling the firm s characteristics. This result is consistent with our previous finding that short sellers trade in the same direction of the NSI effect in one quarter after NSI information is available. The short sellers also trade on other firm s characteristics. For instance, they are more likely to short the firms with low momentum and high one quarter lagged returns Performance of Institutional Trading In this section, we analyze the roles that different types of institutions and short sellers play in their trading. Gerad and Nanda (1993) introduce a model showing the potential manipulative role that the institutional investors may play around SEOs. In their setting of the model, the informed institutional investors sell their holding prior to SEOs, even though they have the favorable information about the firm. The selling drives down the equity offering price, and thus the informed institutions are able to obtain the SEO share allocations at a lower offering price and profit by selling the allocations subsequently. In contrast, Chemmanur and Jiao (2005) argue that 20

34 institutions increase their holdings on stocks both before and after share offerings if they obtain the favorable information about the firm. Institutions engage in the information production role if they consistently trade in the same direction of the information they obtain. While quasi-indexers and dedicated institutions trade in the different direction to the NSI effect, it s possible that they play a manipulative role if they have private information about the firms. If quasi-indexers and dedicated institutions engage in manipulation, we expect that they earn significantly positive abnormal returns. Based on the above discussion, we examine the trading performance of different types of institutional investors and short sellers. The results are presented in Table 1-6. We use the NSI portfolios constructed before. In each panel, the "Top NSI Deciles" portfolio is long side portfolio consist of the stocks in the top two deciles with lowest NSI. The "Bottom NSI Deciles" portfolio is the short side portfolio consist of the stocks in the bottom three deciles with highest NSI. For each portfolio, from quarter t (Qt) to year 2 ([Qt+5, Qt+8]), we respectively calculate the equal-weighted and value-weighted average abnormal returns of the stocks bought, hold and sold by different types of institutional investors and also the mean buy-sell spreads of returns. Panel A shows the result for quasi-indexers institutions. Both equal-weighted and valueweighted results do not show any significant positive spreads of abnormal returns across our analysis period. Thus, quasi-indexers institutions do not generate significant positive abnormal returns from trading in the opposite side of the NSI effect. Accordingly, we reject the hypothesis that quasi-indexers institutions play a manipulative role in their trading. In the Bottom NSI Deciles, quasi-indexers institutions sell the stocks with negative abnormal returns, however, they buy stocks with even worse performance. Thus, this is consistent with our previous argument that trading in the opposite direction to the NSI effect is not due to transaction frictions. 21

35 Panel B shows the results for dedicated institutions. For the Bottom NSI Deciles, the equal weighted buy-sell spread for quarter t+2 is significantly positive. We don t observe other significant positive buy-sell spread from the results. Thus, we do not find clear evidence that dedicated institutions profit from their buy-sell strategy which is against the NSI effect. Accordingly, we reject the hypothesis that dedicated institutions play a manipulative role. Panel C shows the results for transient institutions. The evidence for transient institutions are clearer. For the equal-weighted result, in the Bottom NSI Deciles, the buy-sell spreads of abnormal returns are positive and significant for quarter t+1 and t+2. That means transient institutions earn significant positive abnormal returns on the short side stocks when their buy-sell strategy are consistent with the NSI effect. Considering transient institutions are short-term institutions, the significant negative spreads of abnormal returns for year 2 is probably because transient institutions realize their profit in year 2. In the value-weighted results, again we find that transient institutions earn significant abnormal returns in quarter t+1 from trading the bottom NSI deciles. Thus, while their trading is consistent with the NSI effect, transient institutions generate significant abnormal returns by trading on the short side. Our finding is consistent with the hypothesis that transient institutions play an information production role. Panel D shows the performance of short sellers trading. In this panel, we calculate the equalweighted and value-weighted average abnormal returns of the stocks with increased, decreased and hold relative short interest positions and also the mean decrease-increase spreads of abnormal returns. For equal-weighted results, in the Bottom NSI Deciles, the decrease-increase spreads of abnormal returns are significantly positive for quarter t+1 and t+2. It s indicative that short sellers earn 0.73%, and 0.74% significant abnormal returns in quarter t+1 and t+2, respectively. However, 22

36 in the Top NSI Deciles, we don t find clear evidence that short sellers earn significant abnormal returns on those portfolios. The value-weighted result is weaker. In the Bottom NSI Deciles, the only significant negative spread is for quarter t+2. In the Top NSI Deciles, there is no evidence that short sellers earn significant abnormal returns. Instead, we find a significant positive spread for quarter t+4, which indicates that short sellers earn significant negative abnormal return in that quarter. Short sellers earn significant abnormal returns mainly on small stocks on the short side. Our finding is consistent with the hypothesis that short sellers play an information production role. 1.4 Further Analysis The Effect of Reg FD Our previous finding shows that Long-term institutions do not exploit the NSI effect while short-term institutions and short sellers do. After NSI information is available to public, it s hard to argue that long-term institutions do not use the information because they don t know it. The only possible explanation is that long-term institutions avoid using NSI information in their trading on purpose. If institutions obtain private information about NSI from firms, they may not trade against the firms by using the information, otherwise they will not be provided private information by firms any more. Thus, better access to private information about NSI may deter institutions from exploiting the NSI effect. If institutions or short sellers do not exploit the NSI effect for higher returns but intentionally ignore NSI information for other interest, by definition, it is potential agency problem in institutions and short sellers. In this section, we investigate whether information environment play a role on exploiting information and whether there is agency problem in institutions and short sellers. We use the introduction of Regulation of Fair Disclosure 23

37 (Reg FD) as a natural experiment. Reg FD implemented in August 2000 prevents equity issuers from disclosing any private information about issuance to any certain group of investors. Namely, after Reg FD is implemented, institutions and short sellers do not access to private information about NSI. Reg FD provides us different settings of information environment. In pre-reg FD period, certain institutions and short sellers have better access to private information about NSI. However, even though they obtain the private information about NSI from the invested firms, there are two possible reasons for them not to exploit the private information. First, their invested firms who are the information providers will not continue providing them private information if they trade against the firms by exploiting the private information. Second, they may lost the business relation with the invested firms and thus lost the business interest shared with the invested firms. The latter reason is more applied to dedicated institutions who usually play a relationship investment role and keep business relation with the invested firms. Even after NSI information is released, the benefit of obtaining privation information would deter institutional investors and short sellers from exploiting public NSI information. While institutions and short sellers do not exploit NSI information on purpose, it s related to the agency problems in institutions and short sellers. On the other hand, in post-reg FD period, it is less likely that institutional investors and short sellers can obtain private information about NSI from the invested firms. Thus, certain institutions and short sellers may play a better role on exploiting public NSI information. Based on the above discussion, we examine the performance of trading of institutions and short sellers before and after the implement of Regulation of Fair Disclosure (Reg FD). We separate our data into two subsamples, pre-reg FD period and post-reg FD period. In each sub-period, we 24

38 examine the trading performance of different types of institutional investors and short sellers. Specifically, we use the portfolio Top NSI Deciles and Bottom NSI Deciles that we constructed previously, and respectively calculate the equal-weighted and value-weighted average abnormal returns of the stocks bought, hold and sold by different types of institutional investors and also the mean buy-sell spreads of returns. The results are reported in Table 1-7. Panel A shows the results for quasi-indexers institutions. In both pre- and post- Reg FD periods, the buy-sell spreads are all significantly positive overtime. It s indicative that the performance of quasi-indexers institutions is consistent in both pre- and post- Reg FD periods. Panel B shows the result for dedicated institutions. In equal-weighted results, we find significantly positive buy-sell spreads for quarter t+2 in post-reg FD period. There is no significant spreads in pre-reg FD results. Thus, we find weak evidence that dedicated institutions perform even better when they are lack of access to private information. Since dedicated institutions usually have business relation with their invested firms, they are more possible to obtain private information from their invested firms in the pre-reg FD period. For maintaining the benefit of getting private information, dedicated institutions have few incentive to trade against the firms who provide them private information by trading along the NSI effect before and after the NSI quarter. Thus, the benefit of receiving private information deters dedicated institutions from exploiting NSI information. This finding is consistent with our hypothesis that certain institutions play a better role on exploiting public NSI information in the post-reg FD period. Also, our results provide further evidence that there is potential agency problems in dedicated institutions. Panel C shows the results for transient institutions. For equal-weighted results, in both top and bottom deciles, transient institutions earn significantly positive buy-sell spreads in pre-reg FD 25

39 period. Oppositely, transient institutions do not earn any significantly positive but-sell spread in post-reg FD period. Our result indicates that while transient institutions have better access to private information about NSI, they exploit the NSI effect better. On the other hand, while transient institutions are lack of access to private information about NSI, they don t exploit the NSI effect well. Value-weighted result is similar. Panel D shows the results for short sellers. For value-weighed result, in both pre- and post-reg FD period, short sellers earn positive decrease-increase spreads in the quarters after NSI information could be observed. However, in pre-reg FD period, the magnitude and the significance of the spreads are clearly greater than that in post-reg FD period. Thus, similar to transient institutions, short sellers perform better in pre-reg FD period when they have better access to private information Granger-Causality: NSI and Trading among Institutions and Short Sellers While our finding indicates that certain types of institutions and short sellers exploit the NSI effect, it s important to know whether NSI drives the trading of institutions and short sellers or the trading of certain group of investors is driven by the trading of other parties. For addressing this issue, we perform a Granger-Causality analysis to investigate the causality relation among stock returns, NSI, change in long-term institutional ownership, change in short-term institutional ownership and change in relative short interest. Table 1-8 shows the results for the Granger-Causality analysis. Horizontal variables are causal variables and vertical variables are result variables. Our null hypothesis is that variable A is not the causality of variable B. The table shows the p-values of the Granger-Causality test. We use 1% significance level as the threshold to reject causality hypothesis. 26

40 Our results show that NSI causes stock return while stock return is also a causality of NSI. The causality relationship between NSI and return is consistent with the NSI effect that NSI is negatively related to expected stock return. Literature provides rational and behavioral explanations for the negative relationship. For rational explanations, Daniel and Titman (2006) content that exercising firm s real option funded by issuing equity will reduce firm s uncertainty and lower expected stock returns. Moreover, Li, Livdan and Zhang (2009) use Q-theory to show that new investments funded by issuing equity reduce firm s marginal product of capital, and thus cause the drop of expected stock returns. On the other hand, for behavioral explanation, Loughran and Ritter (1995), Ikenberry, Lakonishok and Vermaelen (1995) and Daniel and Titman (2006) argue that managers tend to issue stocks while stocks are overpriced and tend to repurchase stocks while stocks are underpriced. Their arguments help us explain why stock return causes NSI. Equity issuance and repurchase convey the information about stock pricing to the market and the adjustment on the price drives the negative relationship between NSI and future stock return. Our results also show that stock return causes change in long-term and short-term institutional ownership while the trading of long-term and short-term institutions is not the causality of future stock performance. This is the evidence that institutional investors are momentum traders. Our results confirm Badrinath and Wahal (2002) s finding that institutions adjust their ongoing holding according to momentum. We further find that both changes in long-term and short-term institutional ownership can cause NSI positively. This finding confirms Alti and Sulaeman (2012) s finding that SEOs follow periods of high stocks returns only when institutional investor demand is strong. 27

41 Our results also show that NSI causes change in long-term and short-term institutional ownership. Our results of systematic VAR regression show that NSI is related to future change in long-term institutional ownership positively while NSI is related to future change in short-term institutional ownership negatively. These results further confirm our previous finding that longterm institutions trade in the wrong direction of the NSI effect while short-term institutions trade in the right direction of the NSI effect. Moreover, we find that the trading of short-term institutions can cause the trading of long-term institutions but not the other way around. The systematic VAR regression shows that change in short-term institutional ownership is negatively related to future long-term institutional ownership and the negative relationship is driven by the quarter t-1. Thus, while short-term institutions buy (sell), long-term institutions sell (buy) in the next quarter. For the trading of short sellers, it can cause NSI and the change in long-term institutional ownership. However, it s not a cause of ret and short-term institutional ownership. Based on the above results, in general, short-term institutions trade by using stock information merely other than following the trading of other groups. Long-term institutions use information about stock and the trading of other groups. Previously we find that short sellers take advantage of the NSI effect. However, our results for Granger-Causality analysis show that NSI is not a cause of the change in short interest. The contradiction could be explained by our finding on the causality relationship between short sellers and institutions. We find that both changes in long-term and short-term institutional ownership can cause the change in short interest. Specifically, the causality from long-term institutions is more significant. Regarding Edelen, Ince and Kadlece (2014) s finding that institutions in aggregate do 28

42 not exploit the NSI effect, we conjecture that while short sellers exploit the NSI effect, they may trade to take advantage of the trading of institutional investors, other than trading on the information about NSI directly. For further addressing this conjecture, in the next section, we investigate the trading among different types of institutions and short sellers Trading among Institutions and Short Sellers In this section, we examine the trading among different types of institutions and short sellers to investigate whether short sellers take advantage of institutions trading in terms of exploiting NSI effect. Table 1-9 reports the result for the analysis. Each panel presents one party s average trading (bought or sold) of the stocks and the other parties trading on the same stocks. For example, for the block Stock with Positive NSI in Panel A, Row B presents the trading of quasi-indexers institutions on the stocks bought in quarter t. For the same group of stocks, we calculate the change in dedicated and transient institutional ownership from quarter t-3 to year 2. First of all, we observe different patterns of trading among three types of institutions and short sellers. In Panel A, the result for stocks bought or sold by quasi-indexers institutions in quarter t. The first block shows the results for stocks with positive NSI. For the stocks bought in quarter t, there is a net buying in following quarters. Similarly, for the stocks sold in quarter t, there is a gradual net buying in following quarters. The second block shows the results for stocks with negative NSI bought or sold by quasi-indexers institutions in quarter t. In quarter t+1, there is a net selling for those stock bought in quarter t. In other quarters, quasi-indexers consistently buy the stocks. Panel B shows the result for stocks bought or sold by dedicated institutions in quarter t. The first block shows the results for stocks with positive NSI. For stocks bought by the dedicated institutions in quarter t, there is a net buying in the next two quarters. Afterwards, from quarter t+3 29

43 to year 2, dedicated institutions unload the stocks. For stocks sold in quarter t, there is virtually no trading in the following two quarters followed by a net buying from quarter t+3 to year 2. The second block shows the result for stocks with negative NSI. For stocks bought in quarter t, dedicated institutions buy the stocks in quarter t+1, then shift the trading from quarter t+2 and gradually sell the stocks until year 2. For stocks sold in quarter t, there is a consistent selling in following three quarters. Panel C shows the result for stocks bought or sold by transient institutions in quarter t. The first block shows the result for stocks with positive NSI. For stocks bought in quarter t, transient institutions have almost no trading in quarter t and have gradual selling in following quarters. For stocks sold in quarter t, there is a selling in quarter t+1 and then gradual buying follows. The second block shows the result for stocks with negative NSI. For stocks bought in quarter t, there is consistent buying from quarter t+1 to year 2. The trading is same for stocks sold in quarter t. Panel D shows the result for stocks with increased or decreased short interest. The first block shows the result for stocks with positive NSI. Stocks with increased short interest in quarter t have consistent reduce in short interest in future quarters. On the other hand, stocks that experience decreased short interest in quarter t continue unloading short positions in the next quarter, then short interest increases gradually. The second block shows the result for stocks with negative NSI. The trading pattern for stocks with negative NSI is the same as that for stocks with positive NSI. Second, we observe that in quarter t, all three types of institutions tend to buy stocks together while short sellers always trade in the opposite direction. On the other hand, in quarter t, when quasi-indexers, dedicated or transient institution unload shares, the other two types of institutions are buying. Short sellers trade in the opposite direction to the institutions who are selling. 30

44 Third, in quarter t+1 or quarter t+2 or both, if long-term institutions trade in the opposite direction to the NSI effect, transient institutions and short sellers always do the opposite trading. For example, in panel A, for quarter t+1, while quasi-indexers and dedicated institutions buy stocks with positive NSI, transient institutions decrease their holding on the same stocks. While quasiindexers and dedicated institutions sell stocks with negative NSI, transient institutions are buy the same stocks. Further, short sellers also always trade in the opposite direction to long-term institutions and thus in the same direction to the NSI effect. This finding is consistent with our previous finding that long-term institutions do not exploit the NSI effect while short-term institutions and short sellers can exploit in short-term. Thus, our results provide evidence that transient instructions and short sellers take advantage of trading of long-term institutional investors in short-term when long-term institutional investors do not exploit the NSI effect. 1.5 Conclusion Edelen, Ince and Kadlece (2014) document that institutional investors in aggregate do not exploit the NSI anomaly. By classifying the institutional investors into three categories, quasiindexers, and dedicated institutions, our study shows that the empirical result shown by Edelen, Ince and Kadlece (2014) is driven by quasi-indexers and dedicated institutions who are defined as long-term institutions. We show short-term institutions trade accordingly to the NSI information and exploit the NSI effect, while long-term institutions trade in the opposite direction of NSI trading strategy. Our finding contributes to the recent comprehensive literature by showing that sophistication and investment horizon play a role in exploiting anomaly information. Also, we support Yan and Zhang (2009) s argument that short-term institutions are more sophisticated than 31

45 long-term institutions. Moreover, our study shows that short sellers exploit the NSI effect. In regard of this, our finding support the hypothesis that short sellers are sophisticated. We also investigate the roles that different types of institutional investors and short sellers play in their trading on NSI information. Transient and short sellers play an information production role by trading consistently with NSI trading strategy and earn subsequent significant abnormal return. Long-term institutions play neither an information production role nor manipulative role by their trading. We also investigate whether private information about NSI is used in the trade of institutional investors and short sellers. We find that dedicated institutions and short sellers do a better job in exploiting public information even when they are lack of access to private information. Thus, we provide evidence that dedicated institutions and short sellers ignore private information intentionally and there are potential agency problems when they obtain private NSI information. Finally, our study show that short sellers take the advantage of long-term institutional investors on the short side of the NSI effect. 32

46 Appendix Table A1: High/Low Spread of Average Return across NSI Quintiles: Size Subsamples Table A1 reports the net share issuance effect in different size groups. At the end of each quarter t from January 1980 to June 2013, stocks are ranked by market capitalization (SIZE) as of quarter t-1. Stocks are sorted into five groups by Size, using quintile breakpoints. Each Size group are then subsequently sorted into deciles based on net share issuance (NSI) at the end of quarter t. The table presents the mean of return spreads between bottom and top deciles for each size group over quarter t-3 (Q t-3) to t+4 (Q t+4) and the period from quarter t+5 to t+8 [Q t+5,q t+8]. 33

47 Size 1 Size 2 Size 3 Size 4 Size 5 Q t ** 1.64** 2.38** 2.36** 1.66** (2.65) (6.13) (7.89) (7.42) (4.80) Q t ** 1.71** 2.09** 1.41** (0.46) (2.94) (5.89) (6.00) (4.44) Q t * ** 1.44** 1.04** (-1.98) (0.33) (2.68) (4.30) (3.60) Q t -2.25** -1.29** * 0.44 (-6.68) (-4.77) (-0.89) (2.06) (1.80) Q t ** -0.77** -0.76** * (-3.96) (-2.75) (-2.81) (-1.30) (-2.24) Q t * -1.04** -1.00** -0.4* -0.74** (-2.37) (-4.14) (-3.94) (-1.99) (-3.18) Q t ** -0.87** -1.02** -0.58** -0.85** (-3.03) (-3.79) (-3.98) (-2.91) (-3.63) Q t ** -1.05** -0.52* -0.80** (-1.28) (-2.89) (-4.52) (-2.43) (-3.53) [Q t+5, Q t+8] * -1.44* -1.52** -2.00** (0.08) (-2.35) (-2.41) (-2.62) (-3.30) 34

48 Table A2: Fama-MacBeth Regressions of Stock Returns on NSI Table A2 reports the coefficients from Fama-MacBeth regressions of future returns on net share issuance and other return-predictive variables. The dependent variables are the quarterly raw returns for quarter t+1 (Q t+1) to t+4 (Q t+4) and the cumulative future return for the period from quarter t+5 to t+8 [Q t+5, Q t+8], respectively. In univariate regressions, the independent variable is NSI. The control variables in multivariate regressions are change of institutional ownership in quarter t-1 ( IO_lag1), market capitalization (Size), book-to-market ratio (BM), momentum (MOM), stock price (PRC), idiosyncratic volatility (IVOL), shares turnover (TURN) and relative short interest (RSI). Size, BM, MOM, PRC, IVOL, TURN and RSI are constructed in quarter t. 35

49 36 RET (Qt+1) RET (Qt+2) RET (Qt+3) RET (Qt+4) RET [Qt+5, Qt+8] NSI -1.27** -0.58** -1.49** -0.93** -1.89** -1.14** -1.67** -0.78** -4.21** (-3.30) (-2.60) (-4.12) (-3.53) (-4.91) (-4.00) (-4.42) (-2.70) (-3.48) (-1.86) IO_lag1 2.23** ** (4.51) (1.93) (3.16) (-0.44) (-0.35) ln(size) -0.11* -0.1* ** (-2.27) (-2.06) (-1.85) (-1.52) (-4.34) ln(b/m) (0.92) (1.42) (1.78) (1.79) (1.48) MOM 2.37** 2.09** 1.21** 0.97* -1.65* (5.56) (5.86) (3.28) (2.53) (-2.43) PRC (1.18) (0.76) (1.17) (0.8) (0.92) IVOL ** ** (-4.14) (-2.62) (-1.42) (-0.93) (1.46) TURN -4.81** -5.66** -6.33** -6.01** -21.2** (-3.88) (-4.41) (-4.87) (-4.83) (-6.45) RSI 3.13** 3.99** 4.32** 4.61** 16.99** (4.28) (4.95) (5.13) (5.73) (9.48) Intercept 1.25** 3.94** 1.24** 4.03** 1.24** 4.01** 1.26** 3.74** 5.62** 16.1** (3.53) (4.02) (3.52) (4.07) (3.48) (4.10) (3.52) (3.96) (6.27) (5.30) Adj R2 0.25% 4.02% 0.22% 3.69% 0.24% 3.62% 0.23% 3.47% 0.29% 4.48% N

50 Table B: Average Changes in Institutions across NSI Deciles This table reports the quarterly average change in institutions of the stocks across the NSI deciles. At the end of quarter t (Q t), stocks are sorted into deciles based on the net share issuance at the end of quarter t. Panel A and Panel B reports the equal-weighted and value-weighted average change in institutional ownership ( IO) for quarter t-3 (Q t-3) to t+4 (Q t+4) and the cumulative change in institutional ownership for period from quarter t+5 to quarter t+8 [Q t+5, Q t+8]. Panel C presents the quarterly average change in institutions portfolio weight and the cumulative change in institutions portfolio weight ( IW) for the same period. In each panel, the average spread between bottom and top deciles are reported. The t-statistics reported in parentheses are adjusted using Newey-West correction for heteroskedasticity and serial correlation. The numbers are reported in percentage term per quarter. The last row in each panel is the average number of stocks in each decile portfolio. Panel A: Equal-Weighted Average Change in Firms' Institutional Ownership ( IO) Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Low High High-Low 1.16** 1.30** 0.94** ** 0.37** 0.32** 0.28** 0.14 (6.74) (9.60) (7.47) (1.34) (8.11) (4.26) (3.37) (2.91) (1.82) N

51 Panel B: Value-Weighted Average Change in Firms' Institutional Ownership ( IO) Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Low High High-Low ** ** * (1.02) (2.80) (1.86) (-1.84) (4.38) (1.89) (2.00) (1.82) (-0.29) N Panel C: Average Change in Institutions' Portfolio Weight ( IW) Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Low High High-Low 0.86** 0.84** 0.74** 0.64** 0.15** 0.14** (12.18) (12.30) (10.61) (10.00) (2.74) (2.76) (1.57) (1.16) (1.34) 38

52 Table C: High-low Spread of Change in Institutional Ownership and Change in Institutional Portfolio Weight In quarter t (Q t), stocks are sorted into deciles based on the net share issuance (NSI). For the three types of institution, this table presents equal-weighted (EW) and value-weighted (VW) average high/low spreads of change in institutional ownership and the average high/low spreads of the change in institutional portfolio weight between the bottom and top NSI deciles in pre-rfd and post-rfd period. Panel A, B and C show the spreads of the change in quasi-indexers, dedicated and transient institutions, respectively. Panel D presents the high/low spread of the change in relative short interest and the high/low spread of the change in short sellers portfolio between the bottom and top NSI decile in pre-rfd and post RFD period. The pre-rfd period is from the first quarter in 1981 to the fourth quarter in The post-rfd period is from the first quarter in 2001 to the second quarter in

53 Panel A: Trades of Quasi-indexing Institutions Holding Period Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] High-low Spread of Change in Institutional Ownership EW Pre 0.13* 0.3** 0.24** ** 0.49** 0.42** 0.32** 0.21** (2.12) (4.98) (3.71) (-0.89) (9.44) (8.09) (6.79) (5.64) (6.59) Post ** 0.49** 0.35** 1.05** 0.82** 0.74** 0.63** 0.40** (-0.16) (4.52) (5.81) (4.35) (11.74) (8.80) (6.98) (6.37) (8.97) VW Pre ** 0.65** 0.55** 0.42** 0.11** (-0.93) (1.68) (1.39) (-0.73) (8.68) (5.28) (5.02) (4.62) (3.12) Post ** 0.26* 0.37** 0.31** 0.21** (-1.02) (1.19) (1.56) (-0.82) (5.72) (2.54) (4.03) (4.48) (3.35) High-low Spread of Change in Institutional Portfolio Weight 40 Pre 0.72** 0.74** 0.69** 0.62** 0.23** 0.18** 0.14* (9.02) (8.62) (7.71) (7.93) (3.8) (2.85) (2.09) (1.75) (1.78) Post 0.69** 0.74** 0.67** 0.59** 0.20* 0.23** (6.27) (7.08) (6.79) (6.15) (2.24) (2.84) (1.66) (0.94) (1.23)

54 Panel B: Trades of Dedicated Institutions Holding Period Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] High-low Spread of Change in Institutional Ownership EW Pre ** 0.07* -0.06* 0.11** 0.07* 0.08** 0.09** 0.07** (1.39) (2.94) (2.15) (-2.06) (3.44) (2.14) (2.63) (3.31) (4.23) Post 0.22** 0.25** 0.25** 0.20** 0.19** 0.12** 0.09* (3.81) (4.35) (5.24) (4.60) (4.62) (2.87) (2.01) (1.58) (1.03) VW Pre * * ** (-0.86) (-0.17) (-0.56) (-2.41) (1.56) (1.64) (2.24) (1.92) (2.80) Post ** (1.10) (1.65) (1.72) (-0.24) (-0.87) (-1.63) (-1.93) (-1.46) (-4.45) High-low Spread of Change in Institutional Portfolio Weight 41 Pre 0.82** 0.76** 0.73** 0.60** ** (7.38) (6.81) (7.06) (5.66) (1.70) (1.75) (1.34) (1.75) (3.79) Post 1.08** 0.99** 0.82** 0.65** (7.56) (7.65) (5.56) (4.86) (0.38) (0.84) (-0.71) (-0.52) (-1.39)

55 Panel C: Trades of Transient Institutions Holding Period Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] High-low Spread of Change in Institutional Ownership EW Pre 0.92** 0.76** 0.39** -0.2** -0.24** -0.35** -0.38** -0.32** -0.21** (11.05) (9.28) (5.42) (-3.08) (-3.90) (-5.74) (-6.63) (-6.42) (-6.84) Post 1.03** 0.83** 0.57** ** -0.3** -0.2* ** (11.49) (8.34) (6.42) (-0.15) (-3.20) (-3.86) (-2.51) (-1.73) (-3.71) VW Pre 0.48** 0.29** ** -0.27** -0.28** -0.35** -0.35** -0.31** (5.70) (3.46) (0.89) (-4.58) (-3.12) (-3.09) (-4.1) (-4.68) (-8.4) Post ** -0.26** -0.22** -0.14* ** (1.42) (0.84) (-1.29) (-4.00) (-3.57) (-2.58) (-2.04) (-1.63) (-2.97) High-low Spread of Change in Institutional Portfolio Weight 42 Pre 1.53** 1.36** 1.11** 0.71** * (11.70) (10.17) (8.21) (5.68) (-1.37) (-1.46) (-1.68) (-1.81) (-2.22) Post 1.21** 1.08** 0.86** 0.68** (8.80) (8.43) (5.63) (5.16) (-0.65) (0.76) (-0.48) (-0.21) (-0.43)

56 Panel D: Trades of Short Sellers Holding Period Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] High-low Spread of Change in Relative Short Interest EW Pre ** -0.13** * (0.53) (1.62) (2.63) (4.48) (-1.66) (-1.83) (-2.41) (-1.08) (-1.60) Post ** 0.27** 0.19** 0.13** 0.03 (0.28) (0.21) (0.30) (0.34) (5.39) (5.35) (4.54) (3.02) (0.89) VW Pre ** -0.37** (-1.98) (1.30) (3.70) (5.7) (1.19) (-1.01) (-0.65) (-0.83) (-1.28) Post ** -0.4** -0.39** ** (-1.22) (-3.46) (-5.18) (-6.17) (-1.28) (1.28) (-0.37) (-1.46) (-3.82) High-low Spread of Change in Short Sellers Portfolio Weight 43 Pre 1.13** (4.09) (1.41) (-0.44) (-1.97) (-1.74) (1.16) (1.61) (1.36) (0.35) Post 0.96** 0.57** * (4.88) (2.62) (0.46) (0.28) (-0.03) (2.49) (0.03) (-0.44) (-1.32)

57 Table D: Vector Auto-Regressions of Stock Returns, Change of Institutional Ownership, Net Share Issuance and Change of Relative Short Interest This table reports the estimation results of the vector auto-regressions (VAR) of stock returns (RET), change in institutional ownership ( IO), net share issuance (NSI) and change in relative short interest ( RSI). Regressions are based on quarterly observations of each variable as well as the lagged terms. The t-statistics reported in parentheses are adjusted using Newey-West (1987) correction for heteroskedasticity and serial correlation. Model 1 Model 2 Model 3 Model 4 Model 5 RET LT ST NSI RSI RET_lag ** 2.20** 0.96** -0.25** (0.31) (-5.77) (17.1) (8.12) (-2.76) RET_lag2 2.58** 0.21* 0.60** 0.71** (2.69) (2.09) (8.15) (5.59) (-1.86) RET_lag3 3.2** 0.52** 0.18* 1.04** 0.00 (3.08) (5.34) (2.08) (8.37) (0.05) RET_lag ** -0.27** 0.26* 0.00 (1.79) (6.66) (-3.51) (2.5) (0.03) RET_lag ** -0.43** (-0.92) (4.29) (-6.24) (0.31) (1.78) ST_lag ** ** 10.03** 3.03** (0.58) (-7.13) (-11.02) (11.88) (3.77) ST_lag ** 5.78** (-0.76) (0.29) (-11.56) (7.70) (-0.33) ST_lag3-5.47** 2.06** -6.99** 4.77** 0.87* (-2.8) (3.71) (-12.61) (8.64) (2.21) ST_lag ** -5.82** 7.28** 0.96** (1.46) (4.79) (-12.42) (8.41) (2.95) ST_lag ** -6.45** * (-0.28) (3.93) (-14.25) (-1.41) (2.44) 44

58 Model 1 Model 2 Model 3 Model 4 Model 5 RET LT ST NSI RSI LT_lag ** ** 2.22** (-1.67) (-8.83) (-1.70) (8.24) (6.72) LT_lag ** ** 0.56* (-0.07) (-4.09) (1.74) (4.87) (2.47) LT_lag ** (-0.76) (-5.15) (0.71) (1.89) (1.00) LT_lag4-2.22* -1.91** ** 0.33 (-2.07) (-4.52) (1.50) (13.54) (1.48) LT_lag ** ** 0.18 (-1.40) (-4.58) (-0.52) (-2.83) (0.97) NSI_lag1-3.35** 2.10** ** 0.31 (-4.01) (8.06) (1.76) (171.96) (1.76) NSI_lag (-0.04) (-1.55) (-1.47) (0.67) (-1.36) NSI_lag (-1.78) (-0.72) (-1.41) (-0.74) (0.35) NSI_lag ** (-0.16) (-0.73) (0.13) (-30.27) (-0.36) NSI_lag5-2.25** 0.84** -0.34** 31.31** 0.09 (-3.55) (5.43) (-3.44) (27.42) (1.05) RSI_lag ** ** 1.31 (-1.8) (3.17) (1.17) (4.41) (0.39) RSI_lag * ** ** (-0.42) (-2.31) (-1.87) (3.07) (-6.59) RSI_lag * ** (-0.79) (-2.33) (0.58) (1.48) (-3.87) RSI_lag * ** -6.87** (-0.51) (-2.18) (0.69) (8.63) (-2.70) RSI_lag * ** -6.77** (-1.98) (-2.51) (-0.50) (-4.74) (-2.66) Intercept 2.99** ** 0.05** (4.16) (-0.66) (-1.28) (3.92) (2.70) Adj R % 38.24% 45.78% 48.67% 53.42% 45

59 References Ali, A., Hwang, L., Trombley, M.A., Accruals and future stock returns: tests of the naïve investor hypothesis. Journal of Accounting, Auditing & Finance 15, Alti, A., Sulaeman, J., When do high stock returns trigger equity issues? Journal of Financial Economics 103, Asquith, P., P. Pathak, and J. Ritter, 2005, Short Interest, Institutional Ownership, and Stock Returns, Journal of Financial Economics 78, Badrinath, S.G. Wahal, S.,2002, Momentum trading by institutions. Journal of Finance, 57.6, Balsam, S., Bartov, E., Marquardt, C., Accruals management, investor sophistication, and equity valuation: evidence from 10-Q filings. Journal of Accounting Research 40, Barone, G.J., Magilke, M.J., An examination of the effects of investor sophistication on the pricing of accruals and cash flows. Journal of Accounting, Auditing & Finance 24, Bennett, J. A., Sias, R. W., and Starks, L. T. (2003) Greener Pastures and the Impact of Dynamic Institutional Preferences, Review of Financial Studies 16, Boehmer, E., Jones, C.M., Zhang, X., Which shorts are informed? Journal of Finance 63, Boehmer, E., Wu, J., Short selling and the price discovery process. Review of Financial Studies 26, Bushee, B.J., Do institutional investors prefer near-term earnings over long-run value? Contemporary Accounting Research 18,

60 Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu, 2001, Have individual stocks become more volatile? An empirical exploration of idiosyncratic risk, Journal of Finance 56, Campbell, J.Y., Ramadorai, T., Schwartz, A., Caught on tape: institutional trading, stock returns, and earnings announcements. Journal of Financial Economics 92, Chemmanur, T.J., He, S., Hu, G., The role of institutional investors in seasoned equity offerings. Journal of Financial Economics 94, Chemmanur, T.J., Jiao, Y., Seasoned Equity Issues With 'Soft' Information: Theory and Empirical Evidence. Working paper. Boston College. Christophe, S.E., Ferri, M.G., Angel, J.J., Short selling prior to earnings announcements. Journal of Finance 59, Christophe, S.E., Ferri, M.G., Hsieh, J., Informed trading before analyst downgrades: evidence from short sellers. Journal of Financial Economics 95, Collins, D.W., Gong, G., Hribar, P., Investor sophistication and the mispricing of accruals. Review of Accounting Studies 8, Daniel, K., Grinblatt, M., Titman, S., Wermers, R., 1997, Measuring mutual fund performance with characteristic-based benchmarks, Journal of Finance 52, Daniel, K., Titman, S., Market reactions to tangible and intangible information. Journal of Finance 61, Desai, H., Ramesh, K., Thiagarajan S.R., Balachandran B.V., An Investigation of the Informational Role of Short Interest in the Nasdaq Market. Journal of Finance 57, Diether, Karl B., K. Lee and I. M. Werner, 2009, Short-Sale Strategies and Return Predictability, Review of Financial Studies, Society for Financial Studies, vol. 22(2), pages , February. 47

61 Drechsler, I., Drechsler, Q., The Shorting Premium and Asset Pricing Anomalies. Working paper, New York University and Wharton Research Data Services. Edelen, R.M., Ince, O.S., Kadlec, G.B., Institutional investors and stock return anomalies. Working paper, UC Davis. Engelberg, J.E., Reed, A.V., Ringgenberg, M.C., How are shorts informed? Short sellers, news, and information processing. Journal of Financial Economics 105, Fama, E.F., French, K.R., Financing decisions: who issues stock? Journal of Financial Economics 76, Fama, E.F., French, K.R., Average Returns, B/M, and share issues. Journal of Finance 63, Fama, E.F., French, K.R., Dissecting Anomalies. Journal of Finance 63, Gerad, B., Nanda, V., Trading and Manipulation around Seasoned Equity Offerings. Journal of Finance 48, Gibson, S., Safieddine, A., Sonti, R., Smart investments by smart money: evidence from seasoned equity offerings. Journal of Financial Economics 72, Gompers, P. A., and A. Metrick, Institutional Investors and Equity Prices, Quarterly Journal of Economics, CXIV (2001), Heflin, F., Subramanyam, K.R., Zhang, Y., Regulation FD and the Financial Information Environment: Early Evidence. Accounting Review 78, Henry, T.R., Koski, J., Short Selling Around Seasoned Equity Offerings. Review of Financial Studies 23,

62 Hertzel, M.G., Li, Z., Behavioral And Rational Explanations of Stock Price Performance Around Seos: Evidence From A Decomposition of Market-To-Book Ratios. Journal of Financial and Quantitative Analysis 45, Ikenberry, D., Lakonishok, J., Vermaelen, T., Market Underreaction to Open Market Share Repurchases. Journal of Financial Economics 39, Jiang, G.J., Zhang, A.J., The Shrinking Space for Anomalies. Journal of Financial Research 36, Karpoff, J.M., Lou, X., Short sellers and financial misconduct. Journal of Finance 65, Ke, B., Petroni, K.R.., Yu, Y., The Effect of Regulation FD on Transient Institutional Investors' Trading Behavior. Journal of Accounting Research 46, Lev, B., Nissim, D., The persistence of the accruals anomaly. Contemporary Accounting Research 23, Li, E.X.N., Livdan, D., Zhang, L., Anomalies. Review of Financial Studies 22, Li, X., Radhakrishnan, S., Shin, H., Zhang, J., Regulation FD, accounting restatements and transient institutional investors trading behavior. J. Account. Public Policy 30, Loughran, T., Ritter, J.R., The New Issues Puzzle. Journal of Finance 50, Lyandres, E., Sun, L., Zhang, L., The new issues puzzle: testing the investment-based explanation. Review of Financial Studies 21, Pontiff, J., Woodgate, A., Share issuance and gross-sectional returns. Journal of Finance 63, Yan, X., Zhang, Z., Institutional investors and equity returns: are short-term institutions better informed? Review of Financial Studies 22,

63 Table 1-1 Summary Statistics of Firm Characteristics This table presents the firm statistics of the variables that we use in this paper. Sample period is from the first quarter in 1980 to the second quarter in Firm size (Size) is the log term of market capitalization calculated at the end of each quarter. Net share issuance (NSI) is defined as the net change of the log term of shares outstanding over the past 12 months at the end of each quarter. Momentum (MOM) is the cumulative return for the past 12 months at the end of each quarter. Book-to-Market ratio (BM) is the log term of the ratio book equity over market equity. BM is calculated in the end of June of every year. Idiosyncratic volatility (IVOL) is the quarterly standard deviation of the residuals of the Fama and French 3-factor model estimated with stock daily return in each quarter. Stock price (PRC) is the stock price at the end of each quarter. Shares turnover (TURN) is the trading volume divided by shares outstanding each month and averaged over the quarter. Trading volume for NASDAQ is adjusted by a factor of.5. Relative short interest (RSI) is calculated each month as number of shares held short divided by the shares outstanding and averaged over the quarter. Institutional ownership (IO) is the portion of shares outstanding held by institutions at the end of each quarter. We also report the number of stocks (N) with valid observations for each variable, the cross-sectional mean, median, standard deviation and 5th, 25th, 75th, and 95th percentiles for the last quarter of five selected during our sample period: 1980, 1990, 2000, 2010, and Year Variable N Mean StDev 25% Median 75% 1980 Log(Size) NSI MOM Log(BM) IVOL PRC TURN RSI IO Log(Size) NSI MOM Log(BM) IVOL PRC TURN RSI IO

64 Year Variable N Mean StDev 25% Median 75% 2000 Log(Size) NSI MOM Log(BM) IVOL PRC TURN RSI IO Log(Size) NSI MOM Log(BM) IVOL PRC TURN RSI IO Log(Size) NSI MOM Log(BM) IVOL PRC TURN RSI IO

65 Table 1-2 The Net Share Issuance Effect This table reports the equal-weighted and value-weighted average abnormal returns of decile formed on net share issuance (NSI). NSI is the difference between the log term of shares outstanding for month t and the long term of shares outstanding for month t-12. Stocks are sorted into decile based on NSI at the end of each quarter t (Q t). The table reports both equal-weighted (Panel A) and value-weighted (Panel B) average abnormal returns of each decile for quarter t-3 (Q t-3) to t+4 (Q t+4) and also the cumulative returns of each decile for the period from quarter t+5 to t+8 ([Q t+5,q t+8]). The mean of return spreads between the bottom and top deciles (High-Low) are presented. Newey-West (1987) t-statistics for return spreads (in parentheses) are adjusted for heteroskedasticity and serial correlation. The average number of stocks in each decile portfolio (Obs) is also presented. Panel A: Equal-Weighted Abnormal Returns across NSI Deciles Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Low High High - Low 6.75** 4.84** 2.28** -1.41* -2.56** -2.70** -2.82** -2.53** -4.80** (9.66) (7.29) (3.57) (-2.19) (-4.06) (-4.79) (-5.94) (-6.00) (-4.82) Obs

66 Panel B: Value-Weighted Abnormal Returns across NSI Deciles Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Low High High - Low 2.21** 1.28* ** -1.96** -2.30** -2.33** -8.48** (3.84) (2.04) (0.68) (-1.55) (-3.79) (-3.16) (-3.79) (-4.07) (-6.91) Obs

67 Table 1-3 NSI Effect and Institutional Trading Different Types of Institutions This table reports the change in three different types of institutions' ownership and the change in three different types of institutions' portfolio weights across the stocks of the net share issuance (NSI) deciles. Institutions are divided into three groups, transient, quasi-indexer and dedicated, based on classifications proposed in Bushee (2001). NSI is defined as the net change of shares outstanding in log term over the previous 12 months. At the end of quarter t (Q t), stocks are sorted into deciles based on the NSI. This table presents the equal-weighted change in quasi-indexer institutional ownership (Panel A), the value-weighted change in quasi-indexer institutional ownership (Panel B), the change in quasi-indexer institutions' weight (Panel C), the equal-weighted change in dedicated institutional ownership (Panel D), the value-weighted change in dedicated institutional ownership (Panel E), the change in dedicated institutions' weight (Panel F), the equal-weighted change in transient institutional ownership (Panel G), the value-weighted change in transient institutional ownership (Panel H) and the change in transient institutions' weight (Panel I). The reported period is from quarter t-3 (Q t-3) to t+4 (Q t+4) and the period of quarter t+5 to t+8 [Q t+5, Q t+8]. In each panel, it also reports the spreads between the bottom and top decile with their Newey-West (1987) t- statistics (in parentheses) adjusted for heteroskedasticity and serial correlation. The numbers are reported in percent per quarter. 54

68 Panel A: Trades of Quasi-indexing Investors Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Change in Institutional Ownership Low High High-Low ** 0.34** ** 0.62** 0.54** 0.44** 0.28** (1.43) (6.68) (6.41) (1.84) (13.76) (11.15) (9.21) (7.97) (9.83) Obs High/Low Spread of Value-Weighted Change in Institutional Ownership High-Low ** 0.49** 0.48** 0.38** 0.14** (-0.93) (1.68) (1.39) (-0.73) (8.68) (5.28) (5.02) (4.62) (3.12) High/Low Spread of Change in Institutional Investors Weight High-Low 0.71** 0.74** 0.68** 0.61** 0.22** 0.20** 0.13** 0.10* 0.06* (10.67) (11.19) (10.24) (10.08) (4.27) (4.01) (2.66) (1.98) (2.16) 55

69 Panel B: Trades of Dedicated Investors Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Change in Institutional Ownership Low High High-Low 0.12** 0.18** 0.14** 0.05** 0.13** 0.09** 0.09** 0.08** 0.06** (3.55) (5.08) (4.84) (1.57) (5.53) (3.51) (3.33) (3.50) (3.70) Obs High/Low Spread of Value-Weighted Change in Institutional Ownership High-Low * (0.06) (0.99) (0.89) (-2.09) (0.84) (0.31) (0.61) (0.59) (-0.24) High/Low Spread of Change in Institutional Investors Weight High-Low 0.93** 0.86** 0.77** 0.62** (10.13) (9.93) (9.03) (7.44) (1.46) (1.84) (0.61) (0.94) (1.77) 56

70 Panel C: Trades of Transient Institutions Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Change in Institutional Ownership Low High High-Low 0.97** 0.79** 0.47** -0.12* -0.25** -0.33** -0.31** -0.25** -0.19** (15.34) (12.72) (8.10) (-2.24) (-5.01) (-6.87) (-6.46) (-5.69) (-7.63) Obs High/Low Spread of Value-Weighted Change in Institutional Ownership High-Low 0.33** 0.19** ** -0.27** -0.26** -0.27** -0.25** -0.24** (5.23) (3.27) (0.03) (-5.93) (-4.45) (-3.97) (-4.48) (-4.71) (-7.99) High/Low Spread of Change in Institutional Investors Weight High-Low 1.39** 1.24** 1.01** 0.70** * (13.96) (12.95) (9.88) (7.55) (-1.47) (-0.80) (-1.69) (-1.61) (-2.07) 57

71 Table 1-4 NSI Effect and Trading of Short Sellers This table reports average change in short sellers across the net share issuance (NSI) deciles. NSI is defined as the net change of shares outstanding in log term over the previous 12 months at the end of each quarter. At the end of quarter t (Qt), firms are sorted into deciles based on the NSI. Panel A reports the equalweighted quarterly average change in firms' relative short interest ( RSI) for the quarters from t-3 (Q t-3) to t+4 (Q t+4) and the cumulative change in firms' relative short interest for quarter t+5 to t+8 ([Q t+5, Q t+8]). Panel B reports the value weighted quarterly average change in firms' relative short interest for the same periods. Panel C reports quarterly average change in short sellers' portfolio weight for quarters from t-3 (Q t- 3) to t+4 (Q t+4), and the cumulative change in short sellers' portfolio weight for the period of quarter t+5 to t+8 ([Q t+5, Q t+8]). The numbers are reported in percent per quarter. In each panel, the mean of spreads between bottom and top deciles are reported with the t-statistics (in parentheses) adjusted using Newey- West (1987) correction for heteroskedasticity and serial correlation. The last row of each panel is the average number of stocks (Obs) in each decile portfolio. Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Panel A: Equal-Weighted Change in Relative Short Interest Low High High-Low * -0.09** 0.13** 0.15** 0.12** 0.08** 0.02 (-0.41) (-1.62) (-2.18) (-2.64) (4.04) (4.83) (4.02) (2.56) (0.93) Obs Panel B: Value-Weighted Change in Relative Short Interest High-Low ** -0.42** -0.09* (0.76) (-3.65) (-6.37) (-8.42) (-2.22) (1.59) (1.35) (0.38) (-1.42) Panel C: Change in Short Sellers' Portfolio Weight High-Low 1.12** 0.61** (5.72) (3.28) (-0.18) (-1.59) (-1.69) (1.58) (1.51) (1.33) (0.38) 58

72 Table 1-5 NSI Effect and Trading of Different Types of Institutions and Short Sellers Fama- MacBeth Regression Panel A, B and C reports Fama-MacBeth regressions of change in quasi-indexers, dedicated and transient institutional ownership on the net share issuance (NSI), respectively. Panel D reports the regression of change in relative short interest on the NSI. NSI is constructed as the net change of shares outstanding in log term over the previous 12 months at the end of each quarter t (Q t). In panel A, B and C, the dependent variables are the quarterly average change in quasi-indexers, dedicated and transient institutional ownership, respectively, for quarters t-1 (Q t-1) to t+2 (Q t+2). In panel D, the dependent variable is the quarterly average change in relative short interest. In each panel, the independent variable NSI is the NSI in quarter t. All the control variables are one quarter lagged to the corresponding dependent variable. The control variables in multivariate regressions are market capitalization (Size), book-to-market ratio (BM), momentum (MOM), stock price (PRC), idiosyncratic volatility (IVOL), shares turnover (TURN), stock return (RET) and relative short interest (RSI). All the control variables are defined following the way introduced in Table 1-1. Panel A: Quasi-Indexing Institutions QIX (t-1) QIX (t) QIX (t+1) QIX( t+2) NSI 0.38** ** 1.45** 1.06** 1.02** (3.45) (0.27) (0.96) (-0.99) (13.44) (14.31) (10.45) (10.86) ln(size) -0.04* -0.04* -0.04* ** (-2.43) (-2.28) (-2.43) (-2.61) ln(b/m) -0.19** -0.23** -0.21** -0.21** (-10.67) (-12.52) (-11.67) (-11.65) MOM 0.43** 0.42** 0.39** 0.41** (14.51) (11.7) (11.21) (10.68) PRC -0.00** -0.00** -0.00** -0.00** (-2.71) (-2.99) (-2.81) (-2.83) IVOL ** ** ** (-1.09) (-5.13) (-6.03) (-6.01) TURN ** -2.32** -2.28** (-1.6) (-6.06) (-6.77) (-6.43) RET ** -0.76** -0.82** (0.86) (-3.63) (-3.78) (-4.06) RSI -1.23** (-4.62) (-1.64) (-1.44) (-0.97) Intercept 0.34** ** * * (4.45) (-1.98) (1.17) (-3.10) (0.25) (-2.41) (0.36) (-2.21) Adj R % 2.13% 0.12% 1.99% 0.26% 2.12% 0.14% 2.00% 59

73 Panel B: Dedicated Institutions DED (t-1) DED (t) DED (t+1) DED (t+2) NSI 0.19** 0.19** ** 0.20** 0.14** 0.10* (3.08) (2.76) (1.31) (1.22) (4.48) (4.2) (2.77) (2.17) ln(size) -0.02** -0.02** -0.02** -0.02** (-2.88) (-3.06) (-3.15) (-3.07) ln(b/m) -0.04** -0.05** -0.05** -0.05** (-4.41) (-5.45) (-5.41) (-5.20) MOM * (-0.97) (-1.39) (-1.79) (-2.24) PRC * -0.00* -0.00* (-1.66) (-2.3) (-2.36) (-2.49) IVOL ** -2.54** -2.83** (-1.06) (-3.71) (-3.88) (-4.27) TURN (1.09) (-0.59) (-0.95) (-0.60) RET -0.18** -0.35** -0.37** -0.42** (-3.75) (-5.83) (-6.14) (-6.74) RSI -0.25* (-2.4) (-0.8) (-0.42) (-0.6) Intercept (1.72) (-0.39) (-1.26) (-0.56) (-1.75) (-0.34) (-1.80) (-0.44) Adj R % 0.66% 0.08% 0.64% 0.06% 0.61% 0.05% 0.64% 60

74 Panel C: Transient Institutions TRA (t-1) TRA (t) TRA (t+1) TRA (t+2) NSI 0.70** 0.72** * ** -0.29** (5.51) (6.44) (-1.70) (-0.68) (-2.48) (-1.21) (-5.20) (-4.12) ln(size) (0.08) (0.40) (0.40) (0.43) ln(b/m) 0.14** 0.11** 0.12** 0.12** (6.98) (5.67) (5.80) (5.82) MOM 0.17** 0.14** 0.14** 0.11** (5.37) (4.15) (4.03) (2.96) PRC * (-1.23) (-2.05) (-1.86) (-1.86) IVOL 4.81** (3.44) (0.07) (0.23) (0.10) TURN -2.41** -3.32** -3.46** -3.54** (-7.53) (-9.90) (-10.47) (-10.65) RET 3.27** 2.63** 2.63** 2.67** (20.60) (14.28) (14.09) (14.05) RSI * (-0.61) (1.42) (1.44) (2.30) Intercept ** ** ** ** (0.86) (6.62) (-0.88) (5.72) (-0.90) (5.79) (-0.85) (6.08) Adj R % 3.34% 0.17% 3.01% 0.11% 2.98% 0.12% 2.99% 61

75 Panel D: Short Sellers RSI (t-1) RSI (t) RSI (t+1) RSI (t+2) NSI -0.19* -0.31** -0.24** -0.33** 0.14** 0.13* 0.15** 0.16** (-2.03) (-3.24) (-3.35) (-4.63) (2.61) (2.24) (2.86) (2.82) ln(size) (0.20) (0.18) (0.24) (0.00) ln(b/m) (-0.96) (-1.00) (-0.53) (-0.16) MOM (1.20) (1.59) (0.97) (0.49) PRC (-0.75) (-0.95) (-0.74) (-0.68) IVOL 3.96* (2.12) (1.95) (1.62) (1.79) TURN (1.53) (1.46) (1.33) (1.20) RET 0.25* (2.02) (1.83) (1.75) (1.47) RSI (-1.52) (-1.12) (-1.06) (-0.59) Intercept (1.43) (-1.18) (1.38) (-1.14) (0.81) (-0.82) (0.87) (-0.32) Adj R % 1.80% 0.08% 1.71% 0.05% 1.65% 0.03% 1.74% 62

76 Table 1-6 Performance of Institutional and Short Seller Trading This table reports the abnormal returns earned by three different types of institutions and short sellers. Institutions are divided into three groups, transient, quasi-indexer and dedicated, based on classifications proposed in Bushee (2001). At the end of quarter t (Qt), stocks are sorted into deciles based on the net share issuance (NSI). The "Top NSI Deciles" portfolio is consist of the stocks in the top two deciles with lowest NSI, while the "Bottom NSI Deciles" portfolio is consist of the stocks in the bottom three deciles with highest NSI. In quarter t, stocks in both portfolios are classified into three groups by the change of institutional ownership ( IO), using buy ( IO>0), sell ( IO<0) and no change ( IO=0) as breakpoints. The table reports the average returns earned by three types of institutions' trading for quarter t (Qt) to t+4 (Qt+4) and the cumulative return for the period from quarter t+5 to t+8 ([Q t+5, Q t+8]). Panel A, B and C present average abnormal return of quasi-indexers, dedicated and transient institutions' trading, respectively. Both equal-weighted and value-weighted results are presented. The table also presents the mean of buy-sell spreads of returns. The buy-sell spreads is defined as the difference between the buy group ( IO>0) and the sell group ( IO<0). Panel D presents average abnormal return of short sellers trading. The Newey-West (1987) t-statistics in parentheses is adjusted for heteroskedasticity and serial correlation. The numbers are reported in percent per quarter. 63

77 Top NSI Deciles Panel A: Quasi-Indexing Institutions Bottom NSI Deciles IO Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Result Buy Hold Sell Buy-Sell (-0.76) (-0.91) (1.06) (-1.57) (-0.28) (0.72) (0.08) (-0.60) (-0.61) (0.03) (-0.93) (-1.04) Value-Weighted Result 64 Buy Hold Sell Buy-Sell (0.57) (-0.50) (-0.56) (0.59) (-1.88) (-0.35) (-1.11) (1.26) (-1.70) (-0.55) (0.62) (-0.11)

78 Top NSI Deciles Panel B: Dedicated Institutions Bottom NSI Deciles IO Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Result Buy Hold Sell Buy-Sell 0.71** ** ** (3.51) (-0.05) (0.82) (0.21) (-0.11) (-0.28) (4.78) (1.37) (2.72) (0.71) (-0.02) (1.18) Value-Weighted Result 65 Buy Hold Sell Buy-Sell 1.75** ** (4.41) (0.33) (0.75) (0.22) (-0.65) (0.76) (3.39) (1.97) (1.08) (0.11) (0.47) (0.03)

79 Top NSI Deciles Panel C: Transient Institutions Bottom NSI Deciles IO Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Result Buy Hold Sell Buy-Sell 4.59** ** 1.09** 0.50* ** (17.54) (0.47) (1.58) (0.56) (0.28) (0.78) (14.35) (4.48) (2.41) (0.25) (1.11) (-2.73) Value-Weighted Result 66 Buy Hold Sell Buy-Sell 5.34** ** (10.48) (-1.70) (1.27) (0.74) (-1.23) (-0.05) (11.70) (1.84) (0.21) (1.66) (0.92) (-1.06)

80 Top NSI Deciles Panel D: Short Sellers Bottom NSI Deciles IO Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Equal-Weighted Result Buy Hold Sell Buy-Sell -1.82** 0.49* ** 0.73** 0.74** (-8.53) (2.28) (-0.23) (0.24) (-0.93) (-0.11) (-7.12) (3.24) (3.50) (0.50) (1.58) (1.76) Value-Weighted Result 67 Buy Hold Sell Buy-Sell -2.37** ** (-5.14) (-0.50) (-1.48) (-0.51) (-2.98) (0.26) (-3.70) (-0.92) (1.41) (0.94) (-1.91) (-0.43)

81 Table 1-7 NSI Effect and Trading of Different Types of Institutions and Short Sellers Effect of Reg FD In quarter t (Q t), stocks are sorted into deciles based on the net share issuance (NSI). For the three types of institution, this table presents equal-weighted (EW) and value-weighted (VW) average of buy-sell spreads of abnormal returns between the bottom and top NSI deciles in pre-rfd and post-rfd period. Panel A, B and C show the results for quasi-indexers, dedicated and transient institutions, respectively. Panel D presents the results for short sellers. The pre-rfd period is from the last quarter in 1980 to the fourth quarter in The post-rfd period is from the first quarter in 2001 to the second quarter in Panel A: Quasi-Indexing Institutions Top NSI Deciles Bottom NSI Deciles Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] 68 EW Pre * (-1.00) (-0.37) (1.32) (-0.87) (-0.66) (0.02) (2.18) (0.63) (-1.13) (-0.08) (-0.29) (-1.66) Post ** (0.10) (-0.95) (-0.13) (-1.69) (0.21) (1.28) (-2.65) (-1.71) (0.71) (0.18) (-1.15) (0.44) VW Pre (0.06) (-0.22) (-0.78) (0.80) (-1.84) (0.24) (-0.60) (1.33) (-0.69) (0.03) (1.82) (-0.43) Post (0.84) (-0.48) (0.08) (-0.21) (-0.64) (-1.31) (-1.04) (0.36) (-1.79) (-1.00) (-0.86) (0.60)

82 Panel B: Dedicated Institutions Top NSI Deciles Bottom NSI Deciles Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] EW Pre 0.73* ** (2.53) (0.02) (-0.82) (0.45) (-0.86) (1.51) (4.28) (1.96) (1.52) (0.70) (-0.11) (1.45) Post 0.69* ** * 0.79* ** (2.55) (-0.12) (2.59) (-0.36) (0.95) (-2.12) (2.40) (-0.55) (2.74) (0.17) (0.15) (-0.18) VW Pre 0.97* * (2.02) (0.75) (0.52) (0.73) (0.13) (0.55) (2.51) (1.16) (1.40) (0.15) (0.43) (-0.97) Post 2.98** * (5.15) (-0.34) (0.59) (-0.71) (-1.40) (0.53) (2.26) (1.66) (0.12) (-0.02) (0.22) (1.30) 69

83 Panel C: Transient Institutions Top NSI Deciles Bottom NSI Deciles Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] EW Pre 5.06** ** ** 1.48** 0.89** * (14.04) (-0.12) (2.98) (-0.36) (-0.23) (-0.36) (10.95) (4.19) (2.75) (1.21) (1.07) (-2.23) Post 4.40** ** (8.90) (0.33) (-0.52) (1.12) (0.83) (1.54) (10.37) (1.90) (-0.05) (-1.43) (1.72) (-1.09) VW Pre 6.55** ** ** ** (11.81) (-0.02) (2.59) (0.37) (0.02) (-0.45) (11.53) (1.86) (0.31) (2.59) (0.60) (-0.36) Post 3.52** -1.30* ** (4.03) (-2.07) (0.99) (-0.18) (-0.22) (0.47) (5.54) (1.62) (0.01) (-0.68) (0.88) (-0.89) 70

84 Panel D: Short Sellers Top NSI Deciles Bottom NSI Deciles Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] Qt Qt+1 Qt+2 Qt+3 Qt+4 [Qt+5, Qt+8] EW Pre -2.09** ** 1.29** 0.91** (-7.98) (1.94) (-0.82) (0.20) (-0.18) (-0.43) (-6.39) (5.09) (3.42) (1.06) (1.18) (1.60) Post -1.38** ** * (-4.22) (1.20) (0.61) (0.14) (-1.25) (0.51) (-3.09) (0.11) (2.12) (0.10) (1.42) (0.48) VW Pre -2.11** ** (-4.42) (0.21) (-0.93) (-0.65) (-1.74) (-0.03) (-4.56) (0.80) (0.64) (0.36) (-1.45) (-0.33) Post -2.78** ** * (-2.95) (0.53) (-1.51) (0.11) (-2.80) (0.72) (-0.65) (-2.50) (1.85) (0.97) (-1.16) (-0.12) 71

85 Table 1-8 Granger-Causality Analysis This table reports the p-values of the Granger-Causality test among stock returns (RET), change in institutional ownership ( IO), net share issuance (NSI) and change in relative short interest ( RSI). Horizontal variables are causal variables and the vertical variables are result variables. The null hypothesis is variable A is not the causality of variable B. Panel A shows the result of using change in aggregate institutional ownership. Panel B shows the result of using change in long-term institutional ownership where long-term institutional ownership is the total ownership of quasi-indexers and dedicated institutions. Panel C reports the result of using change in short-term institutional ownership where short-term institutional ownership is the ownership of transient institutions. P-value is calculated by using the SSEs from restricted and unrestricted models with nine lagged terms for each variable. Causal Variable Result Variable Ret NSI LT ST RSI (+) (+) (+) Ret % 0.00% 0.01% 0.01% (-) (+) (-) NSI % 0.01% 0.01% 65.19% (+) (+) LT 14.53% % 0.00% 0.01% (+) (-) (+) ST 1.97% % 0.01% 0.02% (+) (+) RSI 35.77% 37.05% % 0.01% 72

86 Table 1-9 Trading among Different Parties This table reports the trading among different types of institutions and the short sellers. It presents one party s average buying and selling of the stocks and the other parties trading on the same buying and selling stocks. For example, in table of Result for All Stocks of Panel A, row B presents quasi-indexers average buying, while row S presents quasi-indexers average selling. For other rows, DED and TRA are the change in dedicated and transient institutional ownership, respectively, on the quasi-indexers buying and selling stocks. RSI is the change in relative short interest, while FIRM is the percentage change of shares outstanding of the stocks. Panel A, B, and C reports stocks bought or sold by quasi-indexers, dedicated, and transient institutions, respectively. Panel D reports the short sell interest increased or decreased by short sellers. The table reports the result for the periods from quarter t-3 (Q t-3) to t+4 (Q t+4) and the period of quarter t+5 to t+8 ([Q t+5, Q t+8]), where quarter t (Q t) is the quarter that net share issuance (NSI) is constructed. Panel A reports the result for the whole sample. Panel B reports the result for the stocks with net share issuance greater than.5%, which implies the firms are issuing new equity. Panel C reports the result for the stocks with net share issuance lower than -0.5%, which implies the firms are buying back equity. 73

87 Panel A: Result for Stocks Bought or Sold by QIX Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Stocks with Positive NSI (NSI>0.5%) B DED TRA RSI Firm S DED TRA RSI Firm Stocks with Negative NSI (NSI< -0.5%) B DED TRA RSI Firm S DED TRA RSI Firm

88 Panel B: Result for Stocks Bought or Sold by DED Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Stocks with Positive NSI (NSI>0.5%) B QIX TRA RSI Firm S QIX TRA RSI Firm Stocks with Negative NSI (NSI< -0.5%) B QIX TRA RSI Firm S QIX TRA RSI Firm

89 Panel C: Result for Stocks Bought or Sold by TRA Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Stocks with Positive NSI (NSI>0.5%) B QIX DED RSI Firm S QIX DED RSI Firm Stocks with Negative NSI (NSI< -0.5%) B QIX DED RSI Firm S QIX DED RSI Firm

90 Panel D: Result for Stocks with Increased or Decreased RSI Holding Period NSI Q t-3 Q t-2 Q t-1 Q t Q t+1 Q t+2 Q t+3 Q t+4 [Q t+5, Q t+8] Stocks with Positive NSI (NSI>0.5%) Increase QIX DED TRA Firm Decrease QIX DED TRA Firm Stocks with Negative NSI (NSI< - 0.5%) Increase QIX DED TRA Firm Decrease QIX DED TRA Firm

91 CHAPTER TWO AGGREGATE NET SHARE ISSUANCE AND STOCK MARKET RETURNS: A TIME-SERIES ANALYSIS 2.1 Introduction Existing literature documents that net share issuance (NSI), defined as the net change in shares outstanding of a firm over a given time period, is a strong and robust predictor of cross-sectional stock returns. (Daniel and Titman, 2006; Pontiff and Woodgate, 2008; Fama and French, 2008a). This anomaly is referred to in the literature as the net share issuance or the NSI effect. Specifically, stocks with high NSI underperform stocks with low NSI. 1 In addition, the literature also shows that the NSI effect is one of the most robust stock return anomalies. 2 In this paper, we examine whether the cross-sectional NSI effect at the firm level extends to the time-series dimension at the market level. Our research is directly motivated by the following arguments. First, in addition to the fact that the NSI effect is one of the most robust stock return anomalies, net share issuance has important and unique information content compared to other stock return predictors. Unlike the price-scaled predictors such as size, B/M and momentum, NSI is solely determined by firm s management. NSI also differentiates from accounting variables, such as accruals, gross profitability and net operating 1 For instance, Pontiff and Woodgate (2008) shows that one-standard deviation change in net share issuance is associated with a 0.33% decrease in the monthly cross-sectional return. 2 There is strong evidence that the NSI effect is pervasive across all size groups (Fama and French, 2008b), in both long and short sides of hedge portfolios (Jiang and Zhang, 2013) and among all short-sale fee sub-groups of stocks (Drechsler and Drechsler, 2014). 78

92 assets. It proxies the financing decision of firm s management, and thus reveals the management s private information about stock s valuation. Second, existing literature has examined whether certain cross-sectional relations between firm characteristics and stock returns extend to the time series aggregate market level. The findings are mixed. Some cross-sectional anomalous effects extend to the time-series level. For instance, Kothari and Shanken (1997), and Pontiff and Schall (1998) find that the B/M effect at the aggregate market level is consistent with the B/M effect at the cross-sectional firm level. 3 Yet, there has been evidence that some of the cross-sectional anomalous effects do not hold at the time-series level. For example, Kothari, Lewellen, and Warner (2006) show that the cross-sectional PEAD effect documented in Bernard and Thomas (2006) significantly decays when extends to the aggregate market level. Hirshleifer, Hou, and Teoh (2009) also find that the accruals effect in aggregate data is in sharp contradiction to the cross-sectional accruals effect as in Sloan (1996). Thus, from the empirical perspective, it is a puzzle that whether the NSI effect holds at the aggregate market level. From the theoretical perspective, predictions are also mixed for whether the firm-level NSI effect should extend to the aggregate level or not. First, when aggregating NSI across the entire market, positive and negative NSI from different firms offsets each other and the NSI effect may be washed away in the aggregate data. Second, information cost and arbitrage cost are significantly lower at the market level than the firm level due to the enormous effort to studying the stock market as a whole by investors and analysts. As a result, anomalous effects at the aggregate level could be arbitraged away by the sophisticated investors. Alternatively, Samuelson (1998), and Jung and 3 Other evidence includes: Baker and Wurgler (2000) find that the poor performance of stock return following higher portion of equity issue in total issues extends to the market level. Wen (2014) shows that the firm-level asset growth effect extends to the aggregate market level. 79

93 Shiller (2005) argue that markets are more efficient in pricing individual stocks than pricing the market as a whole. Under this proof, the anomalous effects are expected to be stronger at the aggregate market level. Underlying both empirical and theoretical perspectives, the relation between aggregate NSI and stock market returns is not clear. Our research provides direct evidence on the NSI effect at the aggregate market level, and adds to the literature about the time-series studies on the anomalous effects. Third and more importantly, one unresolved debate in the literature on the cross-sectional NSI effect is whether it is consistent with the rational asset pricing theories or driven by the mispricing of stocks. Both explanations are proposed in literature. The rational explanations argue that NSI reflects the changes in firm s risk-adjusted discount rate on future cash flows. For example, Carlson, Fisher and Giammarino (2006) argue that exercising firm s real option following equity issues reduces firm s uncertainty and therefore leads to lower expected stock returns. Moreover, Li, Livdan, and Zhang (2009) show that new investments following equity issuance reduce firm s marginal q of capital, and thus cause the decline in future stock returns. Alternatively, the mispricing explanation attributes the cause of the NSI effect to the mispricing of stocks. It argues that firm s manager issues shares when stocks are overvalued and repurchases shares when stocks are undervalued. 4 The negative relation between NSI and future stock returns occurs when the misvaluation is adjusted by investors. From the literature, there is no conclusive evidence in favor of one explanation over the other. By empirical analysis, it is difficult to completely disentangle the rational and mispricing explanations. However, our time-series analysis tries to provide more insight about the source of the NSI effect. 4 See Loughran and Ritter (1995), Ikenberry, Lakonishok, and Vermaelen (1995), Leary and Roberts (2005), Daniel and Titman (2006), Pontiff and Woodgare (2008). 80

94 Our measure of monthly aggregate NSI is calculated as the equal-weighted average of the firmlevel NSI for each month. The firm-level NSI is defined as the net change in the number of shares outstanding over the previous three months. Stock market return is the market excess return, computed as the CRSP indexes market return minus risk-free return. By using the univariate regressions and multivariate regressions of stock market returns on aggregate NSI, we find that from 1980 to 2013, aggregate NSI is a strong and negative predictor of stock market returns. In the univariate regressions, a one standard deviation increase in aggregate NSI relates to 3.17% decline in 2-quarter equal-weighted market excess return. The aggregate NSI effect extends to 3- quarter and 4-quarter stock market returns and becomes stronger as time horizon grows. In the multivariate regressions, we control for other macro-predictors of stock market returns, 5 and find that NSI is negatively and significantly associated with both equal-weighted and value-weighted 2-quarters, 3-quarter and 4-quarter stock market returns. Specifically, a standard deviation increase in aggregate NSI translates into 4.56%, 7.39% and 9.13% decline in 2-quarter, 3-quarter and 4- quarter equal-weighted future stock market returns, respectively. For value-weighted market returns, the NSI effect is weaker but still significant on the 2-quarrer, 3-quarter and 4-quarter stock market returns. We then explore the source of the NSI effect at the aggregate level. Literature about equity market timing provides some evidence that firms time the entire market when issue shares. Loughran, Ritter, and Rydqvist (1994) show that firms equity issues tend to cluster around market peaks. Baker and Wurgler (2000) find that equity share sometimes predicts significant negative 5 The control variables are studied in Goyal and Welch (2008): the earnings-to-price ratio (EP), the dividend-toprice ratio (DP), the book-to-market ratio (BM), the treasury bill rate (TBL), the term spread (TMS), the default spread (DFY), the quity issuance (NTIS), the equity variance (SVAR), the investment-to-capital ratio (IK), and consumptionwealth ratio (CAY). 81

95 market returns, suggesting that firms time the market component in stock returns when issue securities. Also, they note that Correlated investor sentiment implies that firms will be overvalued at the same time and will tend to make similar financing decisions. According the mispricing explanation, the negative relation between aggregate NSI and stock market returns occurs when the market is push back to its efficient value by arbitrage forces. Despite this belief, there has been little direct empirical evidence that mispricing drives the negative relation between aggregate equity issuance and stock market returns. In this paper, we test whether mispricing of stock market drives the negative relation between aggregate NSI and future stock market returns. We examine the NSI effect during different investor sentiment periods. A great amount of literature documents that investors sentiment affects stock pricing and reflects the errors in investor s expectations about future stock returns. 6 During high sentiment periods, the entire market has more participation of naïve sentiment-driven investors. Even sophisticated investors and analysts become overoptimistic. The sentiment-driven trades drive the mispricing of the stock market. Oppositely, during low sentiment periods, there are fewer sentiment-driven participant in the market, and investors trade more rationally. Thus, the price of the stock market is more efficient when investors sentiment is low. Regarding of this, mispricing is considered as more prevalent during high investor sentiment periods. The mispricing explanation for the aggregate NSI effect implies that the effect should be stronger during high investor sentiment periods. Based on Baker and Wurgler (2006) s investor sentiment index, we divide the whole sample into two groups: low and high investor sentiment periods, and examine the aggregate NSI effect within each sub-sample group. We find that the 6 See Delong, et al. (1990), Shleifer and Summers (1990), Lee, Shleifer, and Thaler (1991), Barberis, Shleifer, and Vishny (1997), Shiller (2001), Brown and Cliff (2004, 2005), Yuan (2005), Baker and Wurgler (2006, 2007), Stambaugh, Yu, and Yuan (2012), Hribar and MvInnis (2012) 82

96 aggregate NSI effect is significantly stronger during high sentiment periods than low sentiment periods. Therefore, aggregate NSI has a stronger predictive power when mispricing of the stock market is more extensive. This findings provide the insight that mispricing of stock market is a likely source of the predictive power of aggregate NSI. We also examine the out-of-sample forecasting performance of aggregate NSI. We compare the predicted stock market returns by the univariate forecasting model to the historical mean of actual stock market returns as benchmark. Our results show that aggregate NSI delivers significant positive and significant out-of-sample R 2 on 2-quarter, 3-quarter and 4-quarter equal-weighted stock market returns, suggesting that aggregate NSI is a significant better predictor than the historical mean benchmark. We further compare the predicted stock market returns generated by the multivariate forecasting model including aggregate NSI to the predicted stock market returns generated by the multivariate forecasting model without aggregate NSI but all other predictors. The model with aggregate NSI deliver significant and positive out-of-sample R 2 on 2-quarter, 3- quarter and 4-quarter equal-weighted and value-weighted stock market returns, which ranges from 9% to 52%. In sum, our results suggest that aggregate NSI is a strong and reliable predictor for stock market returns. To further explore whether the mispricing of stock market return and equity market timing is a likely source of the aggregate NSI effect, we examine the relations between aggregate NSI and the stock market returns prior to NSI. If there is an equity market timing, aggregate NSI should be high (low) when stock market returns are also high (low). In other words, the mispricing explanation implies that contemporaneous or lagged stock market returns are positively associated with aggregate NSI. We perform univariate and multivariate regressions of aggregate NSI on 83

97 contemporaneous and 1-quarter lagged stock market returns, and we find that 1-quarter lagged stock market return is positively and significantly associated with aggregate NSI. Our findings provide more evidence on equity market timing. Finally, we examine the relation between aggregate NSI and analyst forecast errors. Analysts forecasts measure investors expectations on firm s future earnings. Since investors widely incorporate analyst forecasts into their trades, analysts forecasts is an important source of information for the valuation of equity. 7 With this regard, analysts forecast error provides a solid measure of mispricing on equity. We test whether past analysts forecasting error is associated with aggregate NSI. We perform univariate regressions of aggregate NSI on our comprehensive measures of past aggregate analyst forecast errors. The results suggest that aggregate NSI is significantly and negatively related to the analysts bias four quarters before NSI month. Our findings are consistent with Bradshaw, Richardson and Sloan (2006) that there is a systematic negative relation between external financing and overoptimism in analysts forecasts. The explanation is either analysts self-select into covering the particular issuing firms that they naively give the best future prospects, or firms self-select into issuing securities during periods in which their inside information indicates the analysts forecast error. No matter which explanation is, our results are in support of that mispricing of the equity market is the source of the aggregate NSI effect. Moreover, prior work documents that investor sentiment affects investors earnings expectation. Hribar and McInnis (2012) find that when sentiment is high, analysts forecasts are more optimistic for certain types of stocks, and analysts forecast errors are significantly associated with temporal variation in investor sentiment. We thus further explore whether the negative 7 See Ball and Brown (1968), Beaver, Clarke, and Wright (1979), Frankel and Lee (1998), Jaganathan, Ma and Silver (2005) 84

98 relation between aggregate NSI and past analysts forecast errors is stronger during high sentiment periods. We find that the significant and negative relation between past analysts forecasts error and the aggregate NSI in whole sample is mainly driven by high investor sentiment periods. During low investor sentiment periods, aggregate NSI is not significantly related to past analysts forecasts error. The finding, again, provides more evidence that the degree of mispricing on equity market is an important determinant of the level of aggregate NSI. Our work adds to the literature about the anomalous effects of equity issues and repurchases. 8 The contribution are two folds. First, we consider the time-series effect of aggregate NSI, rather than the cross-sectional predictability. Second, we provide more insight of the explanations for the predictive power by aggregate NSI from the time-series perspective. In sum, we find that the aggregate NSI is a strong and negative predictor of future stock market returns, which means the NSI effect at the cross-sectional level extends to the aggregate time-series level. The predictability of aggregate NSI is statistically significant and economically large, and holds in- and out-ofsample. Further, the NSI effect at the aggregate market level is significantly stronger during high investor sentiment periods. Finally, aggregate NSI is negatively associated with past stock market returns, as well as past analyst forecast errors. Our findings provide evidence to the mispricing explanation for the aggregate NSI effect. The rest of the paper is organized as follows: Section 2 describes the data that we use and the construction of the variables. Section 3 examines the ability of aggregate NSI to predict stock market returns. Section 4 tests the behavioral hypothesis. Section 5 examines the out-of-sample 8 Ikenberry, Lakonishok, and Vermaelen (1995); Loughran and Ritter (1995); Daniel and Titman (2006); Pontiff and Woodgate (2008), etc. 85

99 forecasting performance of aggregate NSI. Section 6 presents the extending analysis on contemporaneous returns, lagged returns and analyst forecasts error. Section 7 concludes. 2.2 Data and Variable Construction Our sample period is from January 1980 to December The measure of stock market return is the market excess return, which is calculated as CRSP market indexes return minus risk-free return. We calculate both equal-weighted and value-weighted stock market excess returns. The equal-weighted and value-weighted CRSP market indexes return are obtained from CRSP Index file. The 1-quarter, 2-quarter, 3-quarter and 4-quarter cumulative stock market returns are calculated on monthly rolling base. Figure 2-5 plots the time-series path of 2-quarter equalweighted and value-weighted stock market returns. Our firm-level security information is obtained from CRSP Monthly Stocks File. We restrict our sample to the stocks listed in NYSE, Amex, and Nasdaq. We construct the measure of monthly aggregate NSI as the equal-weighted average of all firms NSI within that month. 9 The firm-level NSI is defined as a firm s net change in the number of shares outstanding over the previous 3 months: NSIi,t = Ln (Adjusted Sharesi,t) Ln (Adjusted Sharesi,t-3) The number of shares outstanding is adjusted for splits and other shares events by multiplying the cumulative factor to adjust shares outstanding. Figure 2-1 depicts the time-series path of aggregate NSI from January 1980 to December Relative to value-weighted aggregate NSI, equal-weighted aggregate NSI has heavier weight on small-size firms which are more subjected to the issue of mispricing. It provides us the advantage when we test the mispricing explanation of the NSI effect later. 86

100 In multivariate regressions, we control for other predictive variables surveyed in Goyal and Welch (2008): Earning-to-price ratio (EP) is the difference between the log of earnings and the log of CRSP market index. Dividend Price Ratio (D/P) is the difference between the log of dividends and the log of CRSP market index. The Book to Market Ratio (BM) is the ratio of book value to market value for the Dow Jones Industrial Average. Treasury bills are the U.S. yields on 30-days U.S. securities. Term Spreads (TMS) is the difference between the long term yield on government bonds and the T-bill. Default Yield Spread (DFY) is the difference between BAA- and AAA- rated corporate bond yields. Net Equity Expansion (NTIS) is the ratio of twelve-month moving sums of net issues by NYSE listed stocks divided by the total market capitalization of NYSE stocks. Investment to Capital Ratio (IK) is the ratio of aggregate investment to aggregate capital for the whole economy. Consumption, wealth, income ratio (CAY) is the consumption-wealth ratio suggested in Lettau and Ludvigson (2001). Previous studies show that these variables are associated with microeconomic fluctuation and business conditions, and therefore have predictive power on aggregate returns. We use the composite sentiment index introduced in Baker and Wurgler (2006) as the proxy for investor sentiment. The monthly sentiment index is constructed as a first principal component of six measures: closed-end fund discount, NYSE shares turnover, number of IPO, average 1 st -day returns on IPOs, equity share in new issues and dividend premium. A great number of literature studies the components and documents that the market-wide components have potential to influence prices on many securities in the same direction at the same time. 10 Restricted by the 10 Literature about market-wide components include Delong, Shleifer, Summers, and Waldman (1990), Shleifer and Summers (1990), Lee, Shleifer, and Thaler (1991), Barberis, Shleifer, and Vishny (1998), Shiller (2001), Brown and Cliff (2004, 2005), Yuan (2005), Baker and Wurgler (2006, 2007), Kaniel, Saar, and Titman (2008), Kumar and Lee (2006), Lemmon and Portniaguina (2006), Bergman and Roychowdhury (2008), Frazzini and Lamont (2008), Livnat 87

101 availability of sentiment index data, our sub-sample period of different sentiment is from January 1980 to December Figure 2-2 plots the investor sentiment periods over our sample period. We collect consensus (mean) of EPS forecasts from the Institutional Brokers Estimate System (I/B/E/S) summary statistics file, and actual value of EPS from the I/B/E/S actual file. Due to the sparse I/B/E/S data before 1985, our sample period for analyst sample is from 1985 to Analyst forecasts error is computed as the realized EPS minus the corresponding mean of analysts forecasts, all scaled by the stock price at the end of the forecast month: AAAAAAllyyyyyy FFFFFFFFFFFFFFFF EEEEEEEEEE tt = AAAAAAAAAAAA EEEEEE CCCCCCCCCCCCCCCCCC FFFFFFFFFFFFFFFF tt SSSSSSSSSS PPPPPPPPPP tt Negative forecast errors imply optimism. In each month, we calculate firms 1-, 2-, 3- and 4-quarter ahead analysts forecasts errors. The firm-level analysts forecast error is the average of 1-, 2-, 3- and 4-quarter ahead analyst forecast error at each month. Aggregate forecast error is calculated as the equal-weighted average of all firms analysts forecast error. Table 2-1 reports the summary statistics of the stock market returns, aggregate NSI and other variables. The monthly average of equal-weighted stock market excess return (EWRET) is 0.7% and the monthly average of value-weighted stock market excess return (VWRET) is 0.6%, with standard deviation of 5.6% and 4.6%, respectively. The average of aggregate NSI is 1.3%, with standard deviation of 0.5%. Aggregate NSI is highly autocorrelated. The first-order autocorrelation is 0.90, but decays quickly to 0.56 at the fifth order. Stambaugh (1986), Mankiw and Shapiro (1986) show that in time series predictive regressions, the coefficient is subject to an upward small-sample and Petrovic (2008), Yu (2009), Antoniou, Doukas, and Subrahmanyam (2010), Chung, Hung, and Yeh (2010), Gao, Yu, and Yuan (2010), Baker, Wurgler, and Yuan (2011), and Yu and Yuan (2011). 88

102 bias if innovation of independent variables is negatively correlated with contemporaneous returns. The bias is more pronounced when independent variable is highly persistent. In our unreported results, we find significantly positive relation between the innovation of aggregate NSI and contemporaneous market returns. Thus, our regressions don t have the issue of small-sample bias. Baker-Wurgler sentiment index (BW) has an average of 0.29 and a standard deviation of Analysts forecast error (FE) averages at -0.3%, which supports the allegation in literature that analysts routinely generate overoptimistic stock research. 11 Table 2-12 presents the correlations among stock market returns, aggregate NSI, Baker- Wurgler sentiment index and other variables during low and high sentiment periods. Consistent with the firm-level pattern, aggregate NSI is negatively correlated with stock market returns. Particularly, the magnitude of correlations are higher for equal-weighted stock market excess return. Moreover, aggregate NSI is also correlated to other market return predictors. For example, it is 44.8% correlated with dividend to price ratio (DP), and 55.9% correlated with default yield (DFY) during high sentiment periods. It s therefore necessary to control those variables in regression analysis. 2.3 Predictability of Aggregate NSI Preliminary Analysis First, we preliminarily investigate the relation between aggregate NSI and stock market returns. We divide our whole sample (408 months) into low, mid and high NSI groups based on the level of aggregate NSI, and then we calculate the average of 2-quarter stock market returns for each 11 See Dugar and Nathan (1995), Lin and McNichols (1998), Michaely and Womack (1999), Dechow et al., (2000), Agrawal and Chen (2003) 89

103 group. Figure 2-4 plots the average 2-quarter stock market returns of low, mid and high NSI groups. In general, we observe a negative relation between aggregate NSI and stock market returns. The low NSI group earns 8.47% equal-weighted return and 5.35% value-weighted return on average. The high NSI group earns 3.08% equal-weighted return and 3.93% value-weighted return on average. Regardless types of returns, stock market earns higher returns in low NSI months than in high NSI months. If we compare equal-weighted returns to value-weighted returns, the plot of equal-weighted returns shows a monotonic pattern, but the plot of value-weighted returns does not. In addition, the spread between high/low groups is 5.39% for equal-weighted returns, which is 1.42% greater than value-weighted returns. The interpretation for the difference between equalweighted and value-weighted returns is that the aggregate NSI effect is stronger on equal-weighted stock market returns. Relative to value-weighted returns, equal-weighted returns are driven by the returns of small-cap firms who are more subjected to the mispricing problem. The stronger NSI effect on equal-weighted returns presented by the plot implies that mispricing could be a possible source of the aggregate NSI effect Univariate Regression To examine the aggregate NSI effect, we perform univariate regressions of stock market returns on aggregate NSI. Table 2-2 presents the results. The dependent variables are the stock market excess returns with different horizon. We examine the aggregate NSI effect on 1-, 2-, 3- and 4- quarter market excess returns. We use two measures of stock market excess returns: equalweighted (EW) returns and value-weighted (VW) returns. t-statistics reported in parentheses are adjusted by Newey-West correction for heteroskedasticity and serial correlation. Panel A shows the results for EW returns. The coefficients for 2-quarter, 3-quarter and 4-quarter stock market 90

104 returns are (t = -2.51), (t = -3.05) and (t = -3.05), respectively. The results suggest that aggregate NSI negatively predicts 2-quarter, 3-quarter and 4-quarter EW stock market returns. The return predictability becomes stronger as the horizon of market return grows. The negative relation between aggregate NSI and stock market returns is consistent with the crosssectional relation proposed in literature. (Daniel and Titman, 2006; Pontiff and Woodgate, 2008; etc,). So, the NSI effect extends from the cross-sectional level to the time-series level. Panel B shows the results for VW stock market returns as dependent variable. Different from the results for EW returns, the coefficients of aggregate NSI are negative but insignificant. Also, the adjusted RR 2 are much lower for VW results. The comparison between EW and VW results in Table 2-2 is consistent with our observation from Figure 2-4. We find that aggregate NSI has stronger predictability for EW stock market returns, which contains more misvaluation in the pricing Multivariate Regression In this section, we examine whether aggregate NSI provides incremental power to predict market returns when controlling for other aggregate predictors. We control for other predictors of market returns studied in Goyal and Welch (2008). Since aggregate NSI has great correlation with some control variables (refer to Panel B of Table 2-12), controlling these aggregate predictors also helps to disentangle whether the prediction power of NSI differs from the prediction power of these predictors. Table 2-3 reports the results for multivariate regressions. The coefficients of aggregate NSI are negative and significant for both EW and VW returns. When using EW stock market returns as dependent variable (Panel A), a one standard deviation increase in aggregate NSI predicts 4.56% 91

105 (t = -3.49), 7.39% (t = -4.88) and 9.13% (t = -5.04) decline in 2-quarter, 3-quarter and 4-quarter EW stock market returns, respectively. Correspondingly, for VW stock market returns in Panel B, the declines are 1.64% (t = -1.98), 2.85% (t = -3.00) and 2.82% (t = -2.22). Both magnitude and significance of the coefficients are higher than those in univariate regression coefficients due to the negative correlation between aggregate NSI and the control variables. Further, the adjusted RR 2 in multivariate regressions are higher relative to univariate regressions, suggesting that inclusion of other predictors improves the explanatory power of the model. Therefore, adding other predictors does not reduce the ability of aggregate NSI to predict stock market returns. Multivariate regressions show the same pattern as univariate regressions that the predictability of aggregate NSI is stronger for EW returns, which is supportive to the mispricing explanation of the NSI effect. Taking univariate and multivariate regressions together, our results provide strong evidence that aggregate NSI is a significant and negative predictor of stock market returns. 2.4 The NSI Effect during Different Investor Sentiment Periods This section examines the possible source of market return predictability by aggregate NSI. Literature proposes two competing explanations for the NSI effect at the cross-sectional level. The first one is the rational explanation based on the relation between changes in firm s discount rate on future cash flow and stock returns (Carlson, Fisher and Giammarino, 2006; Li, Livdan and Zhang, 2009). The alternative is referred to as the behavioral explanation based on the mispricing story: firm managers tend to issue overvalued stocks and repurchase undervalued stocks. This explanation argues that stock mispricing is the source of the predictability by NSI. However, it s difficult to completely disentangle the two explanations. In this section, we use time-series analysis to test the behavioral explanation at the aggregate market level. 92

106 Stambaugh, Yu and Yuan (2012) argue that sentiment-driven mispricing of stocks is more prevalent during the high investor sentiment periods. They consider impediments to short selling as the cause of mispricing. We argue that mispricing of entire stock market should also be more prevalent during high sentiment periods for two reasons. First, the entire stock market has more participation of naïve sentiment-driven investors. Second, investor sentiment may even play a role in the trades of professional investors. In this regard, the sentiment-driven trades drive the mispricing of the entire market during high sentiment periods. On the other hand, during low sentiment periods, sentiment-driven investors are more passive and investors trade more rationally. In this sense, the mispricing of stock market is expected to be more pervasive when investor sentiment is high. Following Stambaugh, Yu and Yuan (2012), we use Baker-Wurgler Sentiment Index as the measure of investor sentiment. We define the months with the positive sentiment index as the high sentiment periods, and the month with negative sentiment index as the low sentiment periods. Since the behavioral explanation attributes mispricing as the source of the predictability of aggregate NSI, our hypothesis is that the NSI effect is stronger during high (positive) sentiment periods. Figure 2-4 portraits the NSI effect following high/low sentiment periods. Within the high sentiment and low sentiment groups, the observations are divided into low NSI, mid NSI and high NSI groups based on aggregate NSI. The plots portrait the average 2-quarter stock market returns for each group. Figure 2-4a presents the plot for the average EW stock market returns. During high sentiment periods, the average 2-quarter stock market return is 7.49% for low NSI periods, and % for high NSI periods. During low sentiment periods, the average 2-quarter stock market 93

107 return is 10.38% for low NSI periods and 6.20% for high NSI periods. The high-low spread for high sentiment periods is 6.54% greater than the spread for low sentiment periods, suggesting the stronger NSI effect during high sentiment periods. Figure 2-4b presents the plot for VW market returns. It shows the similar pattern of Figure 2-4a. In sum, the preliminary analysis from the plots is consistent with the implication of the mispricing explanation. Table 2-4 presents the results of the univariate regression analysis. We run the regressions of market returns on aggregate NSI during low and high investor sentiment periods, separately. For EW market excess returns (Panel A), during low sentiment periods, the coefficients on NSI are (t = -2.11), (t = -2.35), and (t = -2.15) for 2-, 3- and 4-quarter market excess returns, respectively. During high sentiment periods, the corresponding coefficients are (t = -2.90), (t = -3.40), and (t = -3.39). Both magnitude and significance of the coefficients are greater for high sentiment periods. The results suggest that the aggregate NSI effect on EW market excess returns are stronger during high sentiment periods in which mispricing is more extensive. For the regression of VW market excess returns, we don t observe any significant coefficients of aggregate NSI for both low and high sentiment periods, primarily because VW market excess returns have lower degree of mispricing. We also run the regression of stock market returns on aggregate NSI and interaction term constructed by aggregate NSI and investor sentiment dummy. The results are presented in Table Sentiment dummy is equal to 1 for is equal to 1 when the Baker and Wurgler (2006) s investor sentiment index at month t is positive. Otherwise, the dummy variable is set to 0. The coefficients on the interaction term indicates the additional NSI effect in high sentiment periods than in low sentiment periods. For both EW and VW returns, the coefficients are all negative and 94

108 significant, suggesting that the NSI effect is significantly stronger when investors sentiment is high. Table 2-5 reports the results of the multivariate regressions during low and high sentiment periods. The results show the similar but stronger pattern than the results of univariate regressions. For both EW (Panel A) and VW (Panel B) market returns, it is very clear that the coefficients are higher and more significant during high sentiment periods. In Table 2-14, the coefficients of the interaction term confirms that the aggregate NSI effect is statistically more significant during high sentiment periods than during low sentiment periods at the 1% significance level. Overall, the results for the aggregate NSI effect during low and high investor sentiment periods suggest that the predictive power of aggregate NSI on stock market returns, especially on EW market returns, are significantly stronger during high investor sentiment periods. Because mispricing of stock market returns are more pronounced when investor sentiment periods is high, our finding supports the mispricing explanation for the aggregate NSI effect. Further, the evidence confirms the notion that firms time the market component in stock returns when they make decisions of equity. (Loughran, et al. 1994; Baker and Wurgler, 2000) 2.5 Out-of-sample Results The anomalous effects could be the spurious data mining effects. Fama (1991) firstly addresses this issue. He contends that some clever researchers rummage for forecasting variables and some reliable return predictability are in fact spurious. Empirically, Goyal and Welch (2008) surveyed a broad array of predictors of market returns. They find that most predictors that they surveyed do not have stable predicting power over long period. The out-of-sample forecasting of most 95

109 predictors can t even beat the historical average benchmark. Mclean and Pontiff (2015) also argues that return predictability results solely from statistical biases should disappear out of sample. In this section we examine the out-of-sample performance of aggregate NSI. To evaluate the out-ofsample performance, we use the OOS RR 2 statistics as in Campbell and Thompson (2008), 2 = 1 TT tt=1 (rr tt rr tt ) 2 TT 2 (rr rr tt ) RR OOOO rr tt represents the predictive return by the NSI models. rr is the predictive return by the benchmark 2 models. rr tt is the actual stock market return. A positive RR OOOO indicates that aggregate NSI outperforms the historical average benchmark in forecasting out-of-sample returns: the residual sum of squares by the NSI forecast model TT tt=1 tt=1 (rr tt rr tt ) 2 is smaller than the residual sum of squares by the historical average benchmark model TT 2 (rr rr tt ). Further, we test the significance 2 of RR OOOO using Clark and West (2007) out-of-sample MSPE-adjusted statistic: tt=1 MMMMMMMM aaaaaa = (rr [tt,tt+ττ] rr [tt,tt+ττ] ) 2 [(rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 (rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 ] We test for equal MSPE by regressing MMMMMMMM aaaaaa on a constant and using the resulting t-statistic for a zero coefficient. Since the null hypotheses is that the coefficient is smaller than zero, we apply the one-tailed t-test. The critical value for the one-sided t-statistic is at 10% significant level and at 5% significance level. The in-sample estimation period must contain enough information for a valid out-of-sample test (Goyal and Welch, 2008). Our previous analysis shows that the aggregate NSI effect is varying according to the change in investor sentiment. To assure that our in-sample estimation window contains enough information about the NSI s predictability, we include enough both high and low sentiment periods in our in-sample estimation window. Figure 2-6 shows the time-series plot of 96

110 sentiment index from 1980 to Light and shadowed backgrounds indicate the positive and negative sentiment periods, respectively. Our in-sample estimation window is from January 1980 to June 1996, the period includes 158 months of positive sentiment index and 150 months of negative sentiment index. As a result, the out-of-sample prediction window is from July 1997 to December Table 2-6 presents the results for out-of-sample tests of univariate NSI model. The benchmark 2 is the historical average market return over the in-sample window. Our results show that the RR OOOO for EW 2-quarter, 3-quarter and 4-quarter market returns are 4% (t=1.80), 5% (t=2.12) and 4% 2 (t=2.03), respectively. The positive and significant RR OOOO indicates that aggregate NSI outperforms the historical average benchmark in out-of-sample forecasting. We further examine the out-ofsample performance of the multivariate model. The benchmarks for the multivariate model s prediction is the predictive returns by the multivariate regression model excluding aggregate NSI: RR [tt,tt+ττ] = αα + ββ 1 EEEE tt + ββ 2 DDDD tt + ββ 3 BBBB tt + ββ 4 TTTTTT tt + ββ 5 TTTTTT tt + ββ 6 DDDDDD tt + ββ 7 NNNNNNNN tt + ββ 8 SSSSSSSS tt + ββ 9 IIII tt + ββ 10 CCCCCC tt + μμ tt Table 2-7 presents the out-of-sample tests of multivariate NSI model. Regardless the measure of 2 stock market returns, the statistic RR OOOO is positive and significant for 2-quarter, 3-quarter and 4- quarter stock market returns. The interpretation of the results is that adding aggregate NIS to the model significantly improves the forecasting power of the model. Our out-of-sample tests confirm that the negative relation between aggregate NIS and subsequent stock market returns are not spurious but reliable overtime. We reject the hypothesis that the predictability of aggregate NSI is from statistical bias. 97

111 2.6 Further Analysis on the Explanation of the Aggregate NSI Effect The results in the previous sections show that the predictive power of aggregate NSI is stronger on EW stock market excess returns and during high investor sentiment periods, suggesting that mispricing of stock market is the driver of the predictability. In this section, we provide more insight on the explanation of the aggregate NSI effect by testing different hypotheses Contemporaneous Market Returns, Lagged Market Returns and Aggregate NSI The mispricing explanation of aggregate NSI effect implies the systematical mechanism that firms time the equity market when they make issuance or repurchase decisions, and they do the similar decisions of issuing or repurchasing. Thus, one hypothesis implied by the mispricing explanation is that aggregate NSI holds positive relation with concurrent or past stock market returns. In this subsection, we test the mispricing explanation by examining whether aggregate NSI is positively related to contemporaneous market returns or lagged market returns. We perform the regression of aggregate NSI on contemporaneous stock market returns and 1- quarter lagged stock market returns. Table 2-8 reports the results for the univariate regressions. The dependent variable is aggregate NSI and the independent variables are stock market returns at different time. Ret[t-2, t] is the contemporaneous stock market return, which is calculated as the cumulative stock market return over the previous 3 month before the NSI month t. Ret[t-5, t-3] is the 1-quarter lagged stock market return which is calculated as the cumulative stock market return from month t-5 to t-3. Panel A reports the results for EW market excess returns, and Panel B reports the results for VW market excess returns. We find that the 1-quarter lagged EW and VW lagged stock market returns are positively and significantly related to aggregate NSI. Also the VW contemporaneous return is positively related to aggregate NSI at the 5% significance level. Table 2-9 reports the results for the multivariate regressions, in which we control for other predictors of 98

112 market returns. The results suggest that EW and VW lagged stock market returns are related to aggregate NSI positively and significantly. Our findings support the allegation that firms time the equity market, and make decisions which are similar to their colleagues. So, it is consistent with the mechanism implied in the mispricing explanation Analysts Forecasts Error and Aggregate NSI Next, we employ the measure of aggregate analysts forecasts error to test the mispricing explanation. Analysts forecasts measure the sophisticated prediction on firms future earnings, which is broadly incorporated in investors trading. Regarding of this, analyst forecasts play a critical role in equity pricing, and correspondingly analyst forecasts error is a solid proxy for misvaluation in equity pricing. We construct the measure of aggregate analysts forecasts as equalweighted average of all firm s forecast error. Negative value on our measure of aggregate analyst forecasts errors indicates the overoptimism on future EPS and proxies the overpricing of stock market returns. The mispricing explanation suggests the positive relation between the degree of mispricing of the market and the amount of NSI. Accordingly, we test the mispricing explanation by testing whether past aggregate analyst forecasts errors hold negative relation with aggregate NSI. We perform the regressions of aggregate NSI on the past forecast errors FE [t-τ], where τ is the number of quarters prior to the NSI month. We track the past forecast error up to 4 quarters prior to the NSI month. Table 2-10 presents the results. We find that aggregate NSI is negatively and significantly related to analysts forecasts errors 4 quarters prior to the NSI month, which is consistent with the implication of the mispricing explanation. However, we also find inconsistency that analysts forecasts are insignificantly related to the nearer analysts forecasts in 1, 2, and 3 99

113 quarters prior to the NSI month. The possible explanation is that investors take time to incorporate the information of forecasts into their pricing of equity. Moreover, firms take time to realize forecasts errors and make equity decisions. When investors sentiment is high, we expect that analysts forecasts errors are more extensive. When investors sentiment is low, analysts are expected to deliver more rational forecasts. In this sense, the mispricing explanation implies that the negative relation between aggregate NSI and past analysts forecasts errors should be stronger during high sentiment periods. In table 2-11, we regress aggregate NSI on past forecast errors during low and high sentiment periods, separately. During high sentiment periods, aggregate NSI holds the negative and significant relation with the analysts forecast error 4 quarters before. The magnitude and significance level of the coefficients on forecasts error is greater than that in whole sample period. Moreover, after we control the contemporaneous forecast error (FE [t]), aggregate NSI is also negatively and significantly related to analysts forecast error 3 quarters before. Comparably, during low sentiment periods, aggregate NSI is not significantly associated to analysts forecast error at any time point. Collectively, we observe that the negative relation between past analysts forecast errors and aggregate NSI is stronger when there is more misvaluation in analysts forecasts. Since aggregate analysts forecast error measures mispricing of equity market, we interpret the findings as evidence of equity market timing by firms. The evidence supports the mispricing explanation for the aggregate NSI effect. 2.7 Conclusions Our paper examine the NSI effect at the aggregate market level, and investigate whether the mispricing of stock market is a more likely source of the aggregate NSI effect. We find that 100

114 aggregate NSI negatively predicts future stock market returns. For both in and out-of-sample, the predictive power of aggregate NSI is statistically strong and economically large. Furthermore, we find that the aggregate NSI effect on EW stock market returns is stronger than that on VW stock market returns. Moreover, the aggregate NSI effect is significantly stronger during high sentiment periods than during low sentiment periods. These findings suggest that mispricing of stock market play an important role in explaining the NSI effect at the aggregate level. The intuition of the mispricing explanation for the aggregate NSI effect is that firms time equity market and make similar management decisions when the stock market is mispriced. The negative relation between aggregate NSI and stock market returns happens when mispricing is adjusted. We further test the mispricing explanation by examining the timing behavior. Our further analysis shows that aggregate NSI is positively related with 1-quarter lagged stock market returns. Moreover, aggregate NSI is negatively related to past aggregate analysts forecasts errors. The findings suggest that firms take advantage of mispricing of equity market returns by issuing overpriced shares and repurchasing underpriced shares. In sum, our analysis evident that the mispricing of stock market drives the NSI effect at the aggregate level. 101

115 References Alti, A., How Persistent is the Impact of Market Timing on Capital Structure? Journal of Finance 61, Abarbanell, J., Do Analysts' Earnings Forecasts Incorporate Information In Prior Stock Price Changes? Journal of Accounting and Economics 14, Arif, S., Lee, C.M.C., Aggregate Investment and Investor Sentiment. Review of Financial Studies 27, Ali, A., Hwang, L., Trombley, M.A., Accruals and future stock returns: tests of the naïve investor hypothesis. Journal of Accounting, Auditing & Finance 15, Alti, A., Sulaeman, J., When do high stock returns trigger equity issues? Journal of Financial Economics 103, Ball, R., Brown, P., An Empirical Evaluation of Accounting Income Numbers. Journal of Accounting Research 6, Brown, G.W., Cliff, M.T., Investor Sentiment and Asset Valuation. Journal of Business 78, Beaver, W.H., Clarke, R., Wright, W.W., The Association between Unsystematic Security Returns and the Magnitude of Earnings Forecast Errors. Journal of Accounting Research 17, Barberis, N., Shleifer, A., Vishny, R., A Model of Investor Sentiment. Journal of Financial Economic 49, Bernard, V.L., Thomas, J.K., Evidence That Stock Prices Do Not Fully Reflect the Implications of Current Earnings for Future Earnings. Journal of Accounting and Economics 13,

116 Baker, M., Wurgler, J., The Equity Share in New Issues and Aggregate Stock Returns. Journal of Finance 55, Baker, M., Wurgler, J., Investor Sentiment and the Cross-Section of Stock Returns. Journal of Finance 2, Baker, M., Wurgler, J., Investor Sentiment in the Stock Market. Journal of Economic Perspectives 21, Baker, M., Wurgler, J., Yuan, Y., Global, Local, and Contagious Investor Sentiment. Journal of Financial Economic 104, Butler, A.W., Cornaggia, J., Grullon, G., Weston, J.P., Corporate financing decisions, managerial market timing, and real investment. Journal of Financial Economic 101, Cooper, M.J., Gulen, H., Schill, M.J., Asset Growth and the Cross-Section of Stock Returns. Journal of Finance 63, Chung, S.L., Hung, C.H., Yeh, C.Y., When does Investor Sentiment Predict Stock Returns? Journal of Empirical Finance 19, Campbell, J.Y., Thompson, S.B., Predicting Excess Stock Returns Out of Sample: Can Anything Beat the Historical Average. Review of Financial Studies 21, Carlson, M., Fisher, A., and Giammarion, R Corporate Investment and Asset Price Dynamics: Implications for SEO Event Studies and Long-Run Performance. Journal of Finance 61, Clark, T.E., West, K.D., Approximately Normal Tests for Equal Predictive Accuracy in Nested Models. Journal of Econometrics 138,

117 Chemmanur, T.J., Jiao, Y., Seasoned Equity Issues with 'Soft' Information: Theory and Empirical Evidence. Working paper. Boston College. Daniel, K., Titman, S., Market reactions to tangible and intangible information. Journal of Finance 61, Drechsler, I., Drechsler Q.F., 2014, The Shorting Premium and Asset Pricing Anomalies. Working Paper. NBER. Edelen, R.M., Ince, O.S., Kadlec, G.B., Institutional investors and stock return anomalies. Journal of Financial Economic, forth coming. Fama, E.F., French, K.R., The Cross-Section of Expected Stock Return? Journal of Finance 47, Fama, E.F., French, K.R., Size and Book-to-Market Factors in Earnings and Returns. Journal of Finance 50, Fama, E.F., French, K.R., Financing decisions: who issues stock? Journal of Financial Economics 76, Fama, E.F., French, K.R., Average Returns, B/M, and share issues. Journal of Finance 63, Fama, E.F., French, K.R., Dissecting Anomalies. Journal of Finance 63, Gao, P., Lou, D., Cross-Market Time in Security Issuance. Working Paper. Givoly, D., Lakonishok, J., The Information Content of Financial Analysts' Forecasts of Earnings: Some Evidence on Semi-Strong Inefficiency. Journal of Accounting and Economics 1,

118 Givoly, D., Lakonishok, J., Financial Analysts' Forecasts of Earnings: Their Value to Investors. Journal of Banking & Finance 4, Gibson, S., Safieddine, A., Sonti, R., Smart investments by smart money: evidence from seasoned equity offerings. Journal of Financial Economics 72, Greenwood, R., Hanson, S.G., Share Issuance and Factor Timing. Journal of Finance 67, Goyal, A., Welch, I., A Comprehensive Look at the Empirical Performance of Equity Premium Prediction. Review of Financial Studies 21, Ho, C., Hung, C.H., Investor Sentiment as Conditioning Information in Asset Pricing. Journal of Banking & Finance 33, Henry, T.R., Koski, J., Short Selling Around Seasoned Equity Offerings. Review of Financial Studies 23, Hribar, P., Mclnnis, J., Investor Sentiment and Analysts' Earnings Forecast Errors. Management Science 58, Hertzel, M.G., Li, Z., Behavioral and Rational Explanations of Stock Price Performance around SEOs: Evidence from a Decomposition of Market-to-Book Ratios. Journal of Financial and Quantitative Analysis 45, Hirshleifer, D., Hou, K., Teoh, S.H., Accruals, Cash flows, and Aggregate Stock Returns. Journal of Financial Economics 91, Ikenberry, D., Lakonishok, J., Vermaelen, T., Market Underreaction to Open Market Share Repurchases. Journal of Financial Economics 39,

119 Jones, C.M., Lamont, O.A., Short-Sale Constraints and Stock Returns. Journal of Financial Economics 66, Jung, J., Shiller, R., Samuelson s Dictum and the Stock Market. Economic Inquiry 43, Jiang, G.J., Zhang, A.J., The Shrinking Space for Anomalies. Journal of Financial Research 36, Kumar, A., Lee, C., Retail Investor Sentiment and Return Comovements. Journal of Finance 61, Kothari, S.P., Lewellen, J., Warner, J.B., Stock Returns, Aggregate Earnings Surprises, and Behavioral Finance. Journal of Financial Economics 79, Kothari, S.P., Shanken, J., Book-to-market, Dividend Yield, and Expected Market Returns: A Time-Series Analysis. Journal of Financial Economics 44, Kaniel, R., Saar, G., Titman, S., Individual Investor Trading and Stock Returns. Journal of Finance 63, Lamont, O.A., Investment Plans and Stock Returns. Journal of Finance 55, Lamont, O.A., Stein, J.C., Investor Sentiment and Corporate Finance: Micro and Macro. American Economic Review 96, Lakonishok, J., Shleifer, A., Vishny, R.W., Contrarian Investment, Extrapolation and Risk. Journal of Finance 49, Lev, B., Nissim, D., The Persistence of the Accruals Anomaly. Contemporary Accounting Research 23,

120 Ljunggvist, A., Nanda, V., Singh, R., Hot Markets, Investor Sentiment, and IPO Pricing. Journal of Business 79, Li, D., Zhang, L., Does q-theory with Investment Frictions Explain Anomalies in the Cross Section of Returns? Journal of Financial Economics 98, Li, E.X.N., Livdan, D., Zhang, L., Anomalies. Review of Financial Studies 22, Loughran, T., Ritter, J.R., The New Issues Puzzle. Journal of Finance 50, Lyandres, E., Sun, L., Zhang, L., The New Issues Puzzle: Testing the Investment-based Explanation. Review of Financial Studies 21, Miller, E.M., Risk, Uncertainty, and Divergence of Opinion. Journal of Finance 32, McLean, R.D., Pontiff, J., Watanabe, A., Share Issuance and Cross-sectional Returns: International Evidence. Journal of Finance 94, Mankiw, N.G., Shapiro, M.D., Do We Reject Too Often? Small Sample Properties of Tests of Rational Expectations Models. Economic Letters 20, Nagel, S., Short sales, institutional investors and the cross-section of stock returns. Journal of Financial Economics 78, Nelson, W., The Aggregate Change in Shares and the Level of Stock Prices. FEDS Paper. Nelson, C.R., Kim, M.J., Predictable Stock Returns: The Role of Small Sample Bias. Journal of Finance 48, Polk, C., Sapienza, P., The Real Effects of Investor Sentiment. NBER Working paper. Pontiff, J., Schall, L.D., Book-to-Market Ratios as Predictors of Market Returns. Journal of Financial Economics 49,

121 Pontiff, J., Woodgate, A., Share Issuance and Gross-Sectional Returns. Journal of Finance 63, Quan, W., Asset Growth and Stock Market Returns: A Time-Series Analysis. Working Paper. Rangan, S., Earnings Management and the Performance of Seasoned Equity Offerings. Journal of Financial Economics 50, Ritter, J.R., The Long-Run Performance of Initial Public Offerings. Journal of Finance 46, Rapach, D.R., Ringgenberg, M.C., Zhou, G., Short Interest and Aggregate Stock Returns. Working paper. Rapach, D.E., Strauss, J.K., Zhou, G., Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy. Review of Financial Studies 12, Stambaugh, R.F., Predictive Regressions. Journal of Empirical Finance 54, Schmeling, M., Investor Sentiment and Stock Returns: Some International Evidence. Journal of Empirical Finance 16, Sloan, R., Do Stock Prices Fully Reflect Information in Accruals and Cash Flows about Future Earnings? (Digest Summary) Accounting Review 71, Spiess, D.K., Affleck-Graves, J., Underperformance in long-run stock returns following seasoned equity offerings. Journal of Financial Economics 38, Stambaugh, R.F., Yu, J., Yuan, Y., The Short of It: Investor Sentiment and Anomalies. Journal of Financial Economics 104, Shleifer, A., Vishny, R., The Limit of Arbitrage. Journal of Finance 52,

122 Titman, S., Wei, J., Xie, F., Capital Investments and Stock Returns. Journal of Financial and Quantitative Analysis 39, Wen, Q., Asset Growth and Stock Market Returns: A Time-Series Analysis. Working Paper. Yu, J., Yuan, Y., Investor Sentiment and the Mean Variance Relation. Journal of Financial Economics 100,

123 Table 2-1 Summary Statistics This table reports the summary statistics for stock market returns, aggregate net share issuance and other variables. Monthly stock market excess return is calculated as CRSP indexes market return minus risk-free return. EWRET is the equal-weighted market excess return and VWRET is the value-weighted market excess return. NSI is the aggregate net share issuance, which is calculated as the equal-weighted average of firm-level net share issuance defined as firm s net change in shares outstanding over the previous three months. BW is the Baker and Wurgler investor sentiment index as in Baker and Wurgler (2006). FE is the aggregate forecast error, and is calculated as the equal-weighted average of firm-level analysts forecast error on earnings per share. The firm-level analysts forecast error is the average of 1-, 2-, 3- and 4-quarter ahead analyst forecast error at each month. Analysts forecast error is computed as the actual earnings per share minus the corresponding mean of analysts forecasts, all scaled by the stock price at the end of the forecast month. Other predictive variables follow the definitions in Goyal and Welch (2008). EP is the log earning-to-price ratio. DP is the log dividend-to-price ratio. BM is the book-to-market ratio. TBL is the 30- day T-bill rate. TMS is the difference between long-term yield on government bonds and the Treasury-bill. DFY is the difference between BAA and AAA-rated corporate bonds yield. NTIS is the net equity issuance. SVAR is the equity variance. IK is investment-to-capital ratio. CAY is the consumption-wealth ratio. The sample period is from January 1980 to December Variable Mean Std.dev Q1 Median Q3 Autocorrelation EWRET VWRET NSI BW FE EP DP BM TBL TMS DFY NTIS SVAR IK CAY

124 Table 2-2 Univariate Regression of Market Returns on Net Share Issuance This table reports the results for univariate regressions of excess returns on aggregate net share issuance: RR [tt,tt+ττ] = αα + ββnnnnnn tt + μμ tt The dependent variable RR [tt,tt+ττ] is market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. The independent variable NNNNII tt is aggregate net share issuance, computed as the equal-weighted average of firm-level net share issuance at month t. Panel A presents the results for equal-weighted market excess return, and Panel B presents the results for value-weighted market excess return. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December τ = 1 τ= 2 τ= 3 τ= 4 Panel A: EW Market Excess Returns NSI * -9.98** ** (-1.41) (-2.51) (-3.05) (-3.05) Intercept 0.05* 0.14** 0.21** 0.26** (2.20) (3.29) (3.86) (4.02) Adj R % 3.23% 5.23% 5.51% Panel B: VW Market Excess Returns NSI (-0.89) (-1.34) (-1.62) (-1.19) Intercept * 0.11** 0.12** (1.80) (2.45) (2.91) (2.78) Adj R % 0.69% 1.25% 0.64% 111

125 Table 2-3 Multivariate Regressions of Market Returns This table reports the results for multivariate regression of market excess return on aggregate net share issuance and other market return predictors: RR [tt,tt+ττ] = αα + ββ 1 NNNNNN tt + ββ 2 EEEE tt + ββ 3 DDDD tt + ββ 4 BBBB tt + ββ 5 TTTTTT tt + ββ 6 TTTTTT tt + ββ 7 DDDDDD tt + ββ 8 NNNNNNNN tt + ββ 9 SSSSSSSS tt + ββ 10 IIII tt + ββ 11 CCCCCC tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. NNNNNN tt is the aggregate net share issuance, which is calculated as the equal-weighted average of firm-level net share issuance at month t. Other control variables are defined in Table 2-1. Panel A presents the results for equal-weighted market excess return, and Panel B presents the results for value-weighted market excess return. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December NSI EP DP BM TBL TMS DFY NTIS SVAR IK CAY Intercept Adj R 2 Panel A: EW Market Excess Returns 112 τ= 1 Coeff ** -2.71** ** * 0.08 t-stat (-1.76) (-1.06) (0.90) (1.17) (-3.65) (-2.59) (-0.97) (2.75) (-1.16) (1.66) (2.09) (0.28) 11.88% τ= 2 Coeff ** -0.10* ** -7.53** -6.85** ** * 4.69** t-stat (-3.49) (-2.46) (-0.14) (3.76) (-7.07) (-5.15) (1.21) (3.63) (0.47) (2.32) (4.71) (-1.20) 31.26% τ= 3 Coeff ** -0.15** ** ** -7.77** ** ** 6.76** -1.15* t-stat (-4.88) (-3.00) (-0.76) (4.46) (-7.86) (-4.99) (0.67) (4.45) (1.96) (2.84) (5.53) (-2.05) 42.23% τ= 4 Coeff ** -0.19** ** -11.6** -8.41** ** ** 7.90** -1.43* t-stat (-5.04) (-3.40) (-0.87) (4.23) (-7.87) (-4.96) (0.87) (5.00) (1.81) (2.93) (5.76) (-2.23) 46.32%

126 NSI EP DP BM TBL TMS DFY NTIS SVAR IK CAY Intercept Adj R 2 Panel B: VW Market Excess Returns τ= 1 Coeff * ** -2.27** * t-stat (-1.23) (-0.6) (2.16) (1.00) (-4.12) (-2.8) (-0.63) (1.62) (-1.22) (2.49) (1.52) (1.54) 9.75% τ= 2 Coeff * ** -6.15** -5.32** * ** 2.83** 0.22 t-stat (-1.98) (-0.82) (1.55) (2.71) (-7.41) (-5.35) (-0.38) (2.15) (-0.6) (3.29) (3.88) (0.69) 22.69% τ= 3 Coeff ** ** -8.1** -5.94** * ** 4.04** 0.18 t-stat (-3) (-0.59) (1.31) (2.88) (-8.07) (-4.84) (-0.67) (2.55) (0.17) (3.62) (4.22) (0.42) 19.30% τ= 4 Coeff * * -9.47** -6.11** ** ** 4.79** 0.33 t-stat (-2.22) (-0.07) (1.47) (2.54) (-8.23) (-4.6) (-0.16) (2.81) (0.11) (3.74) (4.42) (0.66) 33.20% 113

127 Table 2-4 Univariate Regression of Market Returns on Net Share Issuance during Low and High Investor Sentiment Periods This table reports the results for univariate regressions of market excess return on aggregate net share issuance during low and high investor sentiment periods: RR [tt,tt+ττ] = αα + ββnnnnnn tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. NNNNNN tt is the aggregate net share issuance, which is calculated the equal-weighted average of firm-level net share issuance at month t. Panel A presents the results for equal-weighted market excess return, and Panel B presents the results for value-weighted market excess return. The sample is divided into low and high sentiment periods based on Baker and Wurgler (2006) s investor sentiment index. The low sentiment periods are the months in which the investor sentiment index is negative. The high sentiment periods are the months in which the investor sentiment index is positive. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December

128 Panel A: EW Market Excess Return Low Sentiment Periods High Sentiment Periods τ= 1 NSI Intercept Adj R 2 NSI Intercept Adj R 2 Coeff ** % 0.81% t-stat (-1.90) (2.69) (-1.21) (1.40) τ= 2 τ= 3 Coeff * 0.19** -9.48** 0.14** 5.23% t-stat (-2.11) (3.16) (-2.90) (2.72) Coeff * 0.28** ** 0.23** 7.59% t-stat (-2.35) (3.43) (-3.40) (3.23) 2.77% 9.43% 115 τ= 4 Coeff * 0.32** ** 0.29** 6.36% t-stat (-2.15) (3.40) (-3.39) (3.37) 10.59%

129 Panel B: VW Market Excess Return Low Sentiment Periods High Sentiment Periods τ= 1 NSI Intercept Adj R 2 NSI Intercept Adj R 2 Coeff ** % 3.10% t-stat (-1.90) (3.93) (0.23) (0.46) τ= 2 τ= 3 Coeff ** % t-stat (-1.02) (4.14) (-0.41) (1.07) Coeff ** % t-stat (-1.42) (5.97) (-0.96) (1.58) 0.46% 0.83% 116 τ= 4 Coeff ** % t-stat (-0.23) (-5.90) (-1.19) (1.88) -0.59%

130 Table 2-5 Multivariate Regression of Market Returns on Net Share Issuance during Low and High Investor Sentiment Periods This table reports the results for multivariate regressions of market excess return on aggregate net share issuance and other market return predictors during high and low investor sentiment periods: RR [tt,tt+ττ] = αα + ββ 1 NNNNNN tt + ββ 2 EEEE tt + ββ 3 DDDD tt + ββ 4 BBBB tt + ββ 5 TTTTTT tt + ββ 6 TTTTTT tt + ββ 7 DDDDYY tt + ββ 8 NNNNNNNN tt + ββ 9 SSSSSSSS tt + ββ 10 IIII tt + ββ 11 CCCCCC tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. NNNNNN tt is the aggregate net share issuance, which is calculated as the equal-weighted average of firm-level net share issuance at month t. The regressions control for other market return predictors which are defined in Table 2-1. Panel A presents the results for equal-weighted market excess return, and Panel B presents the results for value-weighted market excess return. The sample is divided into low and high sentiment periods based on Baker and Wurgler (2006) s investor sentiment index. The low sentiment periods are the months in which the investor sentiment index is negative. The high sentiment periods are the months in which the investor sentiment index is positive. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December

131 NSI EP DP BM TBL TMS DFY NTIS SVAR IK CAY Intercept Adj R 2 Panel A: EW Market Excess Return A.1: Low Sentiment Periods (BW < 0) τ= 1 Coeff ** 0.2** ** * ** t-stat (-3.47) (2.58) (-1.64) (1.86) (-3.43) (-1.98) (1.25) (2.52) (0.22) (1.27) (3.33) (-0.93) 13.67% τ= 2 Coeff ** 0.32** -0.5** 0.97** -8.61** -7.17** 11.91* 2.79** 4.02** ** -1.19* t-stat (-2.98) (2.87) (-2.75) (3.86) (-7) (-4.57) (2.09) (3.7) (3.24) (1.53) (6.3) (-2.34) 36.40% τ= 3 Coeff ** 0.36** -0.72** 1.51** ** -7.59** ** 7.04** 18.57* 10.75** -2.22** t-stat (-3.1) (-4.42) (3.09) (-2.94) (4.41) (-7.73) (-4.7) (1.66) (4.22) (3.07) (2.04) (7.03) 47.44% 118 τ= 4 Coeff ** 0.29** -0.8** 1.91** ** -8.17** ** 7.74** 24.62* 11.94** 0.33 t-stat (-4.26) (2.72) (-2.97) (4.45) (-8.78) (-5.52) (1.69) (3.68) (2.73) (2.38) (6.99) (-3.51) A.2 : High Sentiment Periods (BW > 0) 49.26% τ= 1 Coeff ** ** -2.93** 17.63** * 0.55 t-stat (-3.49) (1.44) (1.2) (-0.3) (-3.09) (-2.75) (3.28) (1.56) (-1) (1.62) (2.27) (1.28) 13.67% τ= 2 Coeff ** * -7.97** * 1.95* ** t-stat (-3.52) (-0.2) (0.03) (2.21) (-3.22) (-2.56) (2.49) (2.07) (1.42) (1.9) (3.26) (-0.65) 36.40% τ= 3 Coeff ** * -6.10* ** 29.31** 21.36* 4.86** t-stat (-4.63) (-1.81) (-1.21) (1.99) (-3.88) (-2.17) (0.37) (1.33) (1.84) (0.92) (4.58) (-1.66) 47.44% τ= 4 Coeff ** * 7.03** * t-stat (-4.78) (-0.91) (-1.77) (0.69) (-0.02) (0.71) (2.26) (5.46) (2.56) (1.11) (1.9) (-1.99) 49.26%

132 NSI EP DP BM TBL TMS DFY NTIS SVAR IK CAY Intercept Adj R 2 Panel B: VW Market Excess Return B.1: Low Sentiment Periods (BW < 0) τ= 1 Coeff ** 0.24** ** -2.19** 7.67* * ** t-stat (-1.82) (2.76) (2.81) (-0.65) (-3.89) (-3.11) (2.26) (-0.16) (-0.27) (2.32) (1.12) (3.24) 13.67% τ= 2 Coeff ** ** -3.00* 9.2* * 19.8** t-stat (-2.72) (1.39) (1.63) (0.79) (-3.49) (-2.37) (2.1) (0.41) (2.45) (3.28) (1.72) (0.95) 36.40% τ= 3 Coeff ** * * 29.63** 20.01** t-stat (-3.56) (0.75) (0.87) (0.2) (-2.22) (-1.00) (1.7) (2.43) (5.26) (3.42) (0.92) (0.2) 47.44% 119 τ= 4 Coeff ** ** 3.87** 31.61** t-stat (-2.56) (1.67) (-0.02) (-1.96) (0.21) (0.35) (2.65) (3.88) (3.37) (1.75) (0.13) (-0.09) 49.26% B.2 : High Sentiment Periods (BW > 0) τ= 1 Coeff * ** -3.75** * ** t-stat (-0.45) (-1.02) (2.31) (1.20) (-3.61) (-2.78) (-1.59) (2.09) (-0.95) (3.06) (1.74) (1.52) 13.67% τ= 2 Coeff * ** -7.82** ** ** 5.35** 0.17 t-stat (-2.02) (-0.67) (1.59) (1.86) (-6.12) (-5.09) (-0.15) (2.65) (0.2) (4.17) (4.22) (0.35) 36.40% τ= 3 Coeff ** ** -8.52** ** ** 7.74** t-stat (-3.08) (-0.68) (1.41) (1.86) (-6.35) (-4.57) (0.26) (3.23) (1.56) (4.87) (4.94) (-0.18) 47.44% τ= 4 Coeff ** * ** -8.92** ** ** 9.3** t-stat (-2.83) (-1.45) (1.54) (2.15) (-6.21) (-5.05) (-0.09) (3.04) (1.92) (5.32) (5.41) (-0.6) 49.26%

133 Table 2-6 Out-of-Sample Forecasts of the Univariate Model This table reports the out-of-sample forecasts of the univariate model: RR [tt,tt+ττ] = αα + ββnnnnnn tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. NNNNNN tt is the aggregate net share issuance, which is calculated the equal-weighted average of the firm-level net share issuance at month t. The in-sample estimation period is from January 1980 to June The out-of-sample forecast period is from July 1996 to December Panel A presents the out-of-sample forecasts on equalweighted (EW) market returns, and Panel B presents out-of-sample forecasts on value-weighted (VW) market returns. OOS R 2 is the Campbell and Thompson (2008) out-of-sample statistic: 2 = 1 TT (rr tt RR OOOO tt=1 rr tt ) 2 TT 2 tt=1(rr rr tt ) rr tt is the forecast generated by the above univariate model. rr is the benchmark, which is the historical average return. Statistical significance of OOS R 2 is based on Clark and West (2007) out-of-sample MSPEadjusted statistic: MMMMMMMM aaaaaa = (rr [tt,tt+ττ] rr [tt,tt+ττ] ) 2 [(rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 (rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 ] *, and ** indicate significance level at 5% and 1%, respectively. OOS Statistics τ = 1 τ = 2 τ = 3 τ = 4 Panel A: EW Market Excess Returns OOS R 2 (%) * 0.05** 0.04* p-value of MSPE Panel B: VW Market Excess Returns OOS R 2 (%) p-value of MSPE

134 Table 2-7 Out-of-Sample Forecasts from Multivariate Regression This table reports the out-of-sample forecasts of the multivariate model: RR [tt,tt+ττ] = αα + ββ 1 NNNNNN tt + ββ 2 EEEE tt + ββ 3 DDDD tt + ββ 4 BBBB tt + ββ 5 TTTTTT tt + ββ 6 TTTTTT tt + ββ 7 DDDDDD tt + ββ 8 NNNNNNNN tt + ββ 9 SSSSSSSS tt + ββ 10 IIII tt + ββ 11 CCCCCC tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. NNNNNN tt is the aggregate net share issuance, which is calculated the equal-weighted average of the firm-level net share issuance at month t. The models control for other market return predictors which are defined in Table 2-1. The in-sample estimation period is from January 1980 to June The out-of-sample forecast period is from July 1996 to December Panel A presents the out-of-sample forecasts on equal-weighted (EW) market returns, and Panel B presents the out-of-sample forecasts on value-weighted (VW) market returns. OOS R 2 is the Campbell and Thompson (2008) out-of-sample statistic: 2 = 1 TT (rr tt RR OOOO tt=1 rr tt ) 2 TT 2 tt=1(rr rr tt ) rr tt is the forecast generated by the above multivariate model. rr is the benchmark, which is forecast generated by the model with all market return predictors except NSI. Statistical significance of OOS R 2 is based on Clark and West (2007) out-of-sample MSPE-adjusted statistic: MMMMMMMM aaaaaa = (rr [tt,tt+ττ] rr [tt,tt+ττ] ) 2 [(rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 (rr [tt,tt+ττ] rr [tt,tt+ττ]) 2 ] *, and ** indicate significance level at 5% and 1%, respectively. OOS Statistics τ = 1 τ = 2 τ = 3 τ = 4 Panel A: EW Market Return OOS R 2 (%) -0.03** 0.18** 0.39** 0.52** p-value of MSPE <0.01 <0.01 <0.01 <0.01 Panel B: VW Market Return OOS R 2 (%) -0.12** 0.09** 0.22** 0.29** p-value of MSPE <0.01 <0.01 <0.01 <

135 Table 2-8 Univariate Regressions of Aggregate NSI on Stock Market Returns This table reports the time-series univariate regressions of aggregate NSI on stock market returns at different time. The dependent variable is aggregate net share issuance, computed as the equal-weighted average of the firm-level net share issuance at month t. RR [tt 2,tt] is the contemporaneous market excess return, which is calculated as the cumulative market excess return from month t-2 to t. RR [tt 5,tt 3] is the 1-quarter lagged market excess return, which is calculated as the cumulative market excess return from month t-5 to t-3. Panel A reports the results for equal-weighted market excess return. Panel B reports the results for value-weighted market excess return. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December Model 1 Model 2 Model 3 Dependent Variable = NSI t Panel A: EW Market Return Ret [t-2, t] (1.81) (1.81) Ret [t-5, t-3] 1.73** 1.72** (6.77) (6.57) Intercept 1.31** 1.28** 1.27** (34.93) (39.94) (39.45) Adj R % 15.24% 16.22% Panel B: VW Market Return Ret [t-2, t] 0.73* 0.71* (2.06) (2.29) Ret [t-5, t-3] 1.94** 1.93** (6.35) (6.30) Intercept 1.31** 1.29** 1.28** (35.17) (38.86) (38.13) Adj R % 10.88% 12.15% 122

136 Table 2-9 Net Share Issuance, Contemporaneous Market Returns, and Lagged Market Returns This table reports the time-series multivariate regressions of aggregate NSI on stock market returns at different time. The dependent variable is aggregate net share issuance at month t, computed as the monthly equal-weighted average of the firm-level net share issuance. RR [tt 2,tt] is the contemporaneous market excess return, which is calculated as the cumulative market return from month t-2 to t. RR [tt 5,tt 3] is the 1-quarter lagged market excess return, which is calculated as the cumulative market return from month t-5 to t-3. The regressions control for other market return predictors, which are defined in Table 2-1. Panel A reports the results for equal-weighted market excess return, and Panel B reports the results for value-weighted market excess return. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December

137 Model 1 Model 2 Model 3 Dependent Variable = NSI t Panel A: EW Market Return Ret [t-2, t] (0.9) (1.59) Ret [t-5, t-3] 1.27** 1.30** (5.31) (5.31) EP (-0.86) (-0.5) (-0.18) DP (0.33) (1.41) (1.61) BM * (-1.25) (-1.92) (-2.04) TBL (0.49) (-0.33) (-0.53) TMS (0.33) (-0.04) (-0.18) DFY * (-2.30) (-1.41) (-1.35) NTIS 10.24** 10.32** 10.45** (6.90) (7.87) (7.91) SVAR (-0.22) (-0.91) (-0.11) IK (-0.34) (1.03) (1.29) CAY (0.76) (0.89) (0.87) Intercept * 2.54** (1.78) (2.50) (2.64) Adj R % 56.65% 56.98% 124

138 Model 1 Model 2 Model 3 Dependent Variable = NSI t Panel B: VW Market Return Ret [t-2, t] (1.42) (1.90) Ret [t-5, t-3] 1.43** 1.46** (4.64) (4.61) EP (-0.97) (-1.12) (-1.00) DP (0.34) (0.96) (1.13) BM (-1.23) (-1.80) (-1.88) TBL (0.52) (0.17) (0.08) TMS (0.36) (0.28) (0.22) DFY * (-2.31) (-1.33) (-1.29) NTIS 10.34** 10.78** 11.02** (6.91) (7.73) (7.72) SVAR (0.07) (-0.93) (0.05) IK (-0.33) (0.42) (0.61) CAY (0.71) (0.72) (0.63) Intercept * 2.11* (1.77) (2.1) (2.20) Adj R % 54.67% 55.11% 125

139 Table 2-10 Net Share Issuance and Analyst Forecasts Error This table reports the regressions of aggregate NSI on the aggregate analysts forecast error, at different time tt ττ, where ττ = 0, 1, 2, 3 or 4 quarters: NNNNNN tt = αα + ββffff [tt ττ] + μμ tt NNNNNN tt is the aggregate net share issuance at month t, which is calculated as the equal-weighted average of the firm-level net share issuance. FFFF is the aggregate analysts forecast error, which is calculated as the equal-weighted average of the firm-level analysts forecast error on earnings per share. The firm-level analysts forecast error at month t is computed as the average of the 1-, 2-, 3-, and 4-quarter ahead forecast errors at that month. Forecast error is calculated as the realized earing per share minus the corresponding consensus (mean) of analysts forecasts, all scaled by stock price at the end of the forecasts month, winsorized at upper and lower 5 percentile. Newey-West (1987) t- statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January1983 to December

140 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 FE [t] (-0.71) (0.22) (0.81) (1.15) FE [t-1] (1.23) (1.34) FE [t-2] (0.57) (0.15) FE [t-3] (-1.03) (-1.07) 127 FE [t-4] -0.64* -0.79* (-2.26) (-2.50) Intercept 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** (12.87) (12.40) (9.72) (9.53) (8.24) (8.22) (7.30) (10.06) Adj R2 0.64% 0.74% -0.10% -0.34% 0.33% 0.95% 2.54% 3.36%

141 Table 2-11 Net Share Issuance and Analyst Forecasts Error during Low and High Investor Sentiment This table reports the regressions of aggregate NSI on the aggregate analysts forecast error during high and low investor sentiment: NNNNNN tt = αα + ββffff [tt ττ] + μμ tt NNNNNN tt is the aggregate net share issuance at month t, which is calculated as the equal-weighted average of the firm-level net share issuance. FFFF is the aggregate analysts forecast error, which is calculated as the equal-weighted average of the firm-level analysts forecast error on earnings per share. The firm-level analysts forecast error at month t is computed as the average of the 1-, 2-, 3-, and 4-quarter ahead forecast errors at that month. Forecast error is calculated as the realized earing per share minus the corresponding consensus (mean) of analysts forecasts, all scaled by stock price at the end of the forecasts month, winsorized at upper and lower 5 percentile. The sample is divided into low and high sentiment periods based on Baker and Wurgler (2006) s investor sentiment index. The low sentiment periods are the months in which the investor sentiment index is negative. The high sentiment periods are the months in which the investor sentiment index is positive. Panel A presents the results for low investor sentiment periods, and Panel B presents the results for high investor sentiment periods. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January1983 to December

142 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Panel A: Low Investor Sentiment Periods (BW < 0) FE [t] (-0.47) (-0.53) (0.41) (0.66) FE [t-1] (0.59) (1.65) FE [t-2] (0.49) (1.80) FE [t-3] (-0.39) (0.09) 129 FE [t-4] (-1.01) (-0.86) Intercept 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** (12.83) (21.40) (12.53) (9.65) (12.75) (8.17) (11.6) (6.74) Adj R % 8.46% -0.17% 2.47% -0.39% -0.81% 1.13% 1.36%

143 Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Panel B: High Sentiment Periods (BW > 0) FE [t] (0.74) (0.90) (0.99) (0.84) FE [t-1] (0.52) (-0.87) FE [t-2] (-0.10) (-1.45) FE [t-3] * (-1.29) (-2.13) 130 FE [t-4] -1.17* -1.53* (-2.35) (-2.19) Intercept 0.01** 0.01** 0.01** 0.01** 0.01** 0.01** 0.05** 0.01** (5.66) (6.26) (5.14) (4.92) (4.38) (4.27) (3.33) (4.39) Adj R % -0.33% -0.55% 1.49% 1.41% 5.47% 4.73% 7.19%

144 Table 2-12 Correlations between Market Excess Returns, NSI and Other Variables during Low and High Investor Sentiment Periods This table reports the correlation between market excess returns, aggregate net share issuance and other market return predictors during low and high investor sentiment periods. All variables are the same as defined in Table 2-1. The sample is divided into low and high sentiment periods based on Baker and Wurgler (2006) s investor sentiment index. The low sentiment periods are the months in which the investor sentiment index is negative. The high sentiment periods are the months in which the investor sentiment index is positive. Panel A presents the results for low investor sentiment periods, and Panel B presents the results for high investor sentiment periods. 131 Panel A: Correlations between Market Returns, NSI and Other Variables during Low Investor Sentiment Periods EWRET VWRET NSI BW FE EP DP BM TBL TMS DFY NTIS SVAR IK EWRET VWRET NSI BW FE EP DP BM TBL TMS DFY NTIS SVAR IK CAY

145 132 Panel B: Correlations between Market Returns, NSI and Other Variables during High Investor Sentiment Periods EWRET VWRET NSI BW FE EP DP BM TBL TMS DFY NTIS SVAR IK EWRET VWRET NSI BW FE EP DP BM TBL TMS DFY NTIS SVAR IK CAY

146 Table 2-13 Regression of Market Returns on Net Share Issuance and Interaction Term This table reports the results for time-series regressions of market excess return on aggregate net share issuance and interaction term constructed by aggregate NSI and investor sentiment dummy: RR [tt,tt+ττ] = αα + ββ 1 NNNNNN tt + ββ 2 NNNNNN tt SSSSSSSS_DDDDDD tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. Panel A presents the results for equal-weighted market returns, and Panel B presents the results for value-weighted market returns. NNNNNN tt is the aggregate net share issuance, computed as the equal-weighted average of crosssectional net share issuance at month t. SSSSSSSS_DDDDDD is a dummy variable of investor sentiment index. SSSSSSSS_DDDDDD tt is equal to 1 when the Baker and Wurgler (2006) s investor sentiment index at month t is positive. Otherwise, SSSSSSSS_DDDDDD tt is equal to 0. The interaction term is constructed by multiplying aggregate NSI with the corresponding investor sentiment dummy. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December

147 τ = 1 τ = 2 τ = 3 τ = 4 Panel A: EW Excess Returns NSI * -8.08* -9.54* (-1.08) (-2.03) (-2.50) (-2.45) DHS*NSI -2.31* -5.23** -7.56** -9.31** (-2.28) (-3.99) (-4.84) (-5.14) Intercept 0.06* 0.15** 0.23** 0.29** (2.46) (3.79) (4.35) (4.47) Adj R % 7.81% 11.38% 12.45% Panel B: VW Excess Returns NSI (-0.55) (-0.85) (-1.07) (-0.59) DHS*NSI -1.59* -3.5** -5.27** -7.03** (-2.1) (-3.28) (-3.81) (-4.16) Intercept 0.04* 0.08** 0.13** 0.15** (2.05) (2.87) (3.37) (3.28) Adj R % 4.45% 6.72% 7.89% 134

148 Table 2-14 Multivariate Regressions of Market Returns and Interaction Term This table reports the results for time-series multivariate regressions of market excess return on aggregate net share issuance and interaction term constructed by aggregate NSI and investor sentiment dummy: RR [tt,tt+ττ] = αα + ββ 1 NNNNNN tt + ββ 2 NNNNNN tt SSSSSSSS_DDDDDD tt + ββ 3 EEEE tt + ββ 4 DDDD tt + ββ 5 BBBB tt + ββ 6 TTTTTT tt + ββ 7 TTTTTT tt + ββ 8 DDDDDD tt + ββ 9 NNNNNNNN tt + ββ 10 SSSSSSSS tt + ββ 11 IIII tt + ββ 12 CCCCCC tt + μμ tt RR [tt,tt+ττ] is the market excess return for different time horizon, where τ = 1, 2, 3 and 4 quarters. Panel A presents the results for equal-weighted market returns, and Panel B presents the results for value-weighted market returns. NNNNNN tt is aggregate net share issuance, which is the equal-weighted average of firm-level net share issuance at month t. SSSSSSSS_DDDDDD is a dummy variable of investor sentiment index. SSSSSSSS_DDDDDD tt is equal to 1 when the Baker and Wurgler (2006) s investor sentiment index at month t is positive. Otherwise, SSSSSSSS_DDDDDD tt is equal to 0. The regressions control for other market return predictors which are defined in Table 2-1. Newey-West (1987) t-statistics (in parentheses) are adjusted for heteroskedasticity and serial correlation. *, and ** indicate significance level at 5% and 1%, respectively. The sample period is from January 1980 to December

149 τ = 1 τ = 2 τ = 3 τ = 4 Panel A: EW Excess Returns NSI * ** ** (-1.44) (-2.28) (-3.04) (-3.00) DHS*NSI ** -8.35** ** (-0.84) (-2.67) (-3.85) (-4.29) EP * -0.15** -0.18** (-1.03) (-2.42) (-3.06) (-3.41) DP (0.8) (-0.47) (-1.24) (-1.42) BM ** 1.17** 1.36** (1.07) (3.58) (4.4) (4.15) TBL -3** -6.63** -8.55** -9.65** (-3.38) (-6.29) (-6.8) (-6.41) TMS -2.47* -5.84** -6.07** -6.21** (-2.36) (-4.5) (-4.04) (-3.8) DFY * * (1.13) (2.08) (1.98) (2.51) NTIS 1.27** 2.35** 3.03** 3.73** (2.64) (3.25) (3.82) (4.16) SVAR ** 4.05** (-1.04) (0.75) (2.79) (3.02) IK ** 23.96** 28.88** (1.78) (2.86) (3.68) (4.14) CAY 1.32* 4.58** 6.57** 7.67** (2.05) (4.68) (5.54) (5.76) Intercept -0.51* -0.88* -1.74** -2.2** (-1.98) (-2.00) (-3.15) (-3.42) Adj R % 32.86% 45.43% 50.34% 136

150 τ = 1 τ = 2 τ = 3 τ = 4 Panel B: VW Excess Returns NSI * (-1.37) (-1.45) (-2.01) (-1.14) DHS*NSI (0.49) (-0.64) (-1.32) (-1.93) EP (-0.61) (-0.81) (-0.56) (-0.01) DP 0.13* (2.24) (1.49) (1.18) (1.26) BM ** 0.62** 0.62* (1.04) (2.67) (2.82) (2.44) TBL -2.94** -5.98** -7.67** -8.73** (-4.25) (-7.16) (-7.54) (-7.30) TMS -2.38** -5.13** -5.45** -5.28** (-2.9) (-5.05) (-4.38) (-3.94) DFY (-0.73) (-0.12) (-0.13) (0.60) NTIS * 1.66* 1.98* (1.71) (2.05) (2.34) (2.47) SVAR (-1.27) (-0.52) (0.38) (0.51) IK 7.47* 14.63** 20.94** 26.45** (2.24) (3.2) (3.72) (4.16) CAY ** 3.98** 4.70** (1.54) (3.85) (4.18) (4.40) Intercept (1.68) (0.46) (0.03) (0.07) Adj R % 24.76% 31.32% 35.41% 137

151 Figure 2-1: Aggregate Net Share Issuance This figure plots the time path of aggregate net share issuance. Aggregate net share issuance is calculated as the equal-weighted averages of firm-level net share issuance, defined as the firm s net change in shares outstanding over the previous 3 months. The sample period is from January 1980 to December

152 Figure 2-2: Investor Sentiment Index This figure plots the Baker-Wurgler (2006) s investor sentiment index. The sentiment index is the first principal component of six measures: the closed-end fund discount, NYSE share turnover, the number of and the average of first-day returns on initial public offerings, the equity share in new issues, and the dividend premium. The sample period is from January 1980 to December

153 Figure 2-3: Market Excess Returns of Different Net Share Issuance Groups This figure plots the equal-weighted and value-weighted 2-quarter market excess returns across various NSI groups. The monthly 2-quarter market excess returns are divided into high NSI, mid NSI and low NSI groups, based on the aggregate net share issuance at the corresponding month. The plots present the average of the 2-qarter market excess returns for each group. The sample period is from January 1980 to December % 8.00% 7.00% 6.00% 5.00% 4.00% 3.00% 2.00% 1.00% 0.00% Low NSI Mid NSI High NSI EW Market Excess Return VW Market Excess Return 140

154 Figure 2-4: Market Excess Returns of Different Net Share Issuance Groups during Low and High Investors Sentiment Periods This figure plots the 2-quarter market excess returns across various NSI groups during low and high investor sentiment periods. The sample is divided into low and high sentiment periods based on Baker and Wurgler (2006) s investor sentiment index. The low sentiment periods are the months in which the investor sentiment index is negative. The high sentiment periods are the months in which the investor sentiment index is positive. Further, within each sentiment class, the monthly 2-quarter market excess returns are divided into high NSI, mid NSI and low NSI groups, based on the aggregate net share issuance at the corresponding month. The plots present the average of the 2-qarter market excess returns for each group. The sample period is from January 1980 to December a. EW Market Excess Return 2-4b. VW Market Excess Return 12.00% 8.00% 10.00% 7.00% % 6.00% 4.00% 2.00% 0.00% 6.00% 5.00% 4.00% 3.00% 2.00% 1.00% -2.00% 0.00% -4.00% Low NSI Mid NSI High NSI -1.00% Low NSI Mid NSI High NSI High Sentiment Periods Low Sentiment Periods High Sentiment Periods Low Sentiment Periods