informed trading, earnings surprises, and stock -
TRANSCRIPT
Informed Trading, Earnings Surprises, and Stock Returns
Hui Guo Buhui Qiu
University of Cincinnati Erasmus University
Initial Version: September 2008 This version: March 2012
Contact information: Hui Guo, Department of Finance, College of Business, University of Cincinnati, P.O. Box 210195, Cincinnati, OH 45221-0195, E-mail: [email protected]; Buhui Qiu, Department of Finance, Rotterdam School of Management, Erasmus University, Burgemeester Oudlaan 50, P.O. Box 1738, 3062 PA, Rotterdam, the Netherlands, E-mail: [email protected]. This paper is a substantial revision of an earlier draft circulated under the title “What Drives the Return Predictive Power of Institutional Ownership: Information or Speculation?” We benefit from helpful discussion with Abhay Abhyankar, Ingolf Dittman, Mike Ferguson, John Griffin, Brian Hatch, Haim Kassa, Yong Kim, Brian Kluger, Emre Konukoglu, Bochen Li, Mark Liu, Melissa Prado, P.K. Sen, Richard Sias, Steve Slezak, Sheridan Titman, Albert (Yan) Wang, Yu Wang, and Xuemin (Sterling) Yan. We thank conference and seminar participants at the 2008 Chicago Quantitative Alliance annual meeting in Chicago, the 2010 CRSP Forum in Chicago, the 2010 China International Conference in Finance in Beijing, XVII Finance Forum at IESE Business School, University of Cincinnati, Erasmus University Rotterdam, and Tinbergen Institute for comments and suggestions. Qiu acknowledges financial support from the Chicago Quantitative Alliance (in the form of its 15th Annual Academic Competition Finalist Award) and the Erasmus Trustfund. The usual disclaimer applies.
1
Informed Trading, Earnings Surprises, and Stock Returns
Abstract
Entries and exits are often triggered by substantive private information, and we propose PC_NII, the
percentage change in the number of a stock’s institutional investors, as a measure of informed trading.
Over the 1982 to 2010 period, the top PC_NII decile outperforms the bottom PC_NII decile by 10% and
14% per annum for equal- and value-weighted portfolios, respectively. PC_NII subsumes the
information content of standard institutional trading or herding measures in the forecast of stock returns,
and the predictive power of PC_NII reflects its information content of future earnings surprises. The
relationship between PC_NII and short-run future stock returns has attenuated noticeably for the
equal-weighted portfolio since the Securities and Exchange Commission introduced Regulation Fair
Disclosure in October 2000; however, we do not observe such a change for the value-weighted portfolio.
2
1. Introduction
There is an ongoing debate about the conjecture that institutional investors possess better skills or
information than do individual investors. A widely investigated implication of this hypothesis is that
institutional trading contains significant information about future stock returns; and the main idea is as
follows. Suppose institutions know that a company will have a better-than-expected earnings
announcement in the coming quarter. Anticipating a jump in the company’s stock price at the
announcement, institutions increase their holdings immediately by purchasing shares from uninformed
individual investors. As a result, the increase in institutional ownership is associated with a positive
stock return in the following quarter. Many authors have empirically examined this implication using
the change in the fraction of shares owned by institutional investors, ∆IO, as a proxy for institutional
trading and documented mixed results. For example, the relationship between ∆IO and future stock
returns is found to be positive in Nofsinger and Sias (1999), insignificant in Gompers and Metrick (2001),
and negative in Cai and Zheng (2004). In this paper, we revisit the issue by proposing a novel measure
of institutional trading, PC_NII, the percentage change in the number of institutional investors. We
show that this simple measure has substantially stronger forecasting power for future stock returns than
do the commonly used measures of institutional trading.
Entries and exits are often triggered by substantive private information. Because investing in new
stocks requires sizeable information costs (e.g., Merton (1987)), an institution is more likely to enter into
a stock when it has strong positive private information about the stock’s fundamentals, ceteris paribus.
On the other hand, the routine portfolio rebalance normally does not require liquidating a stock due to
trading costs. Moreover, an institution usually has multiple funds managed by different traders with
different investment strategies; therefore, a synchronized exit from a stock likely reflects significant
negative private information. Consistent with this conjecture, Chen, Hong, and Stein (2002) and Sias,
Starks, and Titman (2006) document a positive relationship between the change in the number of
3
institutional investors, ∆NII, and future stock returns.1 Entries and exits, however, are also triggered by
liquidity reasons: An institution may enter into a stock when it has some extra funds and exit from a stock
when it has liquidity need. Liquidity-based entries or exits, which convey little private information
about future fundamentals, make ∆NII a noisy measure of informed trading. Ceteris paribus, trading
costs and information costs deter liquidity-based entries and exits; therefore, the frequency of
information-based relative to that of the liquidity-based entries or exits increases with these costs.
Specifically, we find that ∆NII is less informative for big stocks with many institutional investors than for
small stocks with few institutional investors; this is because the former have lower trading costs and
information costs (e.g., Diamond and Verrecchia (1991)) and hence more liquidity-based entries and exits.
For example, 5 entries are a strong positive signal for a stock with 40 institutional investors but may
reflect merely liquidity trading for a stock with 200 institutional investors. Thus, PC_NII, which equals
∆NII normalized by the number of institutional investors in the previous quarter, is arguably a better
measure of informed trading than is ∆NII because PC_NII takes into account the fact that the information
content of entries and exits varies across stocks with different number of institutional investors.
Unlike ∆IO, PC_NII takes into account the fact that institutions are heterogeneous in their skills and
information; for example, Chen, Jegadeesh, and Wermers (2000), Wermers (2000), and Kosowski,
Timmermann, Wermers, and White (2006) show that high-turnover mutual funds have superior
stock-picking skills. When only a small fraction of institutions have substantive private information
about a stock, their entries or exits may have a relatively small impact on institutional ownership due to
their trades with uninformed institutions. In contrast, because an exit is usually triggered by more
substantive private information than are adjustments of holdings, the uninformed institutions are unlikely
to cater informed buy orders by selling off their shares. Moreover, when informed institutions try to
unload a stock, the uninformed institutional buyers are more likely to be the current owners of that stock
rather than new investors because the former have lower information costs (e.g., Merton (1987)) and are
1 Baker, Litov, Wachter, and Wurgler (2010) also find that mutual funds’ entries and exits forecast stock returns.
4
more optimistic about the stock (e.g., Miller (1977)).2 Therefore, an informed entry (exit) usually leads
to an increase (decrease) in PC_NII. Under these circumstances, as we confirm in this paper, PC_NII is
more informative about future stock returns than is ∆IO. Moreover, consistent with the existent studies,
we find that the predictive power of PC_NII is mainly due to the trading of high-turnover institutions.
Our empirical findings over the 1982 to 2010 period lend strong support for the conjecture that
PC_NII is a proxy for institutions’ private information. First, PC_NII has strong predictive power for
earnings surprises. Stocks with high PC_NII have substantially more positive earnings surprises in the
following quarter than do stocks with low PC_NII, and the difference is statistically highly significant
even when we control for past earnings surprises and accounting-related fundamental signals. Second,
we document a strong positive relationship between PC_NII and one-quarter-ahead stock returns. The
trading strategy of longing the top and shorting the bottom PC_NII deciles generates an average
annualized equal-weighted return of 10%, with a Carhart (1997) 4-factor alpha of 7%. For the
value-weighed portfolio, the raw return and alpha are 14% and 11%, respectively. Because the return
difference between the top and bottom PC_NII deciles is almost always positive or close to 0 except for
only 2 years, the predictive power of PC_NII is difficult to be reconciled with risk-based explanations.
Moreover, it cannot be explained by well-known asset pricing anomalies documented in the existent
literature either. Last, when we control for quarter t+1 earnings surprises, the predictive power of
quarter t PC_NII for quarter t+1 stock returns becomes negligible. Therefore, PC_NII correlates with
future stock returns mainly because of its information content for future earnings.
PC_NII relates closely to Chen, Hong, and Stein’s (2002) ∆BREADTH measure—∆NII normalized
by the number of all institutions in the market—because cross-sectionally ∆BREADTH and ∆NII are
perfectly correlated with each other. Chen, Hong, and Stein (2002) argue that ∆BREADTH correlates
2 These conjectures are consistent with recent results of Reca, Sias, and Turtle (2011), circulated after the first version of this paper. While these authors find that the number of institutional investors who enter into (exit from) a stock correlates positively (negatively) with the stock’ future returns, they document an inverse relationship for the number of institutional investors who increases or decreases their holdings. Reca, Sias, and Turtle (2011) suggest, “…the other side of an informed entry (exit) trade is an uninformed institutional decrease (increase) trade or an individual investor selling (buying).”
5
positively with future stock returns because it is a proxy for changes in short-sale constraints. In contrast
with this view, we find that, similar to PC_NII, ∆BREADTH forecasts stock returns mainly because of its
information content for future earnings. Moreover, we show in 2 ways that PC_NII is a better measure
of informed trading than is ∆BREADTH because PC_NII recognizes that the information content of
entries and exits varies across stocks due to their different levels of liquidity-based trading. First,
∆BREADTH has significantly stronger predictive power for stocks with higher trading costs and
information costs, while we do not observe such an asymmetry for PC_NII. Second, PC_NII subsumes
completely the information content of ∆BREADTH in the forecast of stock returns.
As a robustness check, we investigate whether our findings are consistent with 2 other major
alternative hypotheses proposed in the existent literature. First, an increase or decrease in the number of
institutional investors may be due to institutional herding—institutions tend to follow each other into and
out of same stocks. Many authors (e.g., Nofsinger and Sias (1999), Wermers (1999), and Sias (2004))
find a positive relationship between herding and short-term future stock returns. We find that standard
herding measures do not subsume the information content of PC_NII for future stock returns. Second,
Gompers and Metrick (2001) and others document a positive relationship between institutional ownership
and one-quarter-ahead stock returns, and argue that an increase in institutional ownership causes
persistent demand “shocks” in stocks preferred by institutions. Again, we find that institutional
ownership does not subsume the information content of PC_NII for stock returns. In contrast, when we
control for PC_NII, the predictive power of institutional herding or ownership becomes negligible.
On October 23, 2000, the Securities and Exchange Commission (SEC) introduced Regulation Fair
Disclosure (FD) in an effort to reduce selective disclosure of material information by publicly-traded
companies to analysts and other investment professionals. Because PC_NII is arguably a better measure
of institutions’ private information, it enables us to shed new light on the ongoing debate about the
effectiveness of Regulation FD (e.g., see Beyer, Cohen, Lys, and Walther (2010) for a survey of this
literature). Consistent with the view that Regulation FD effectively restricts informed trading, the
difference in the equal-weighted returns between the top and bottom PC_NII deciles has attenuated
6
significantly since 2001. However, we find that such a change is rather negligible for the
value-weighted portfolio returns. Our findings are consistent with the results reported by Ahmed and
Schneible (2007), Eleswarapu, Thompson, and Venkataraman (2004), Gomes, Gorton, and Madureira
(2007), and others, who document a larger reduction in information asymmetry for small stocks than for
big stocks following the introduction of Regulation FD.
Recent studies (e.g., Ke and Ramalingegowda (2005), Bushee and Goodman (2007), Yan and Zhang
(2009), and Jia and Liu (2011)) note that some institutions’ trading is more informative about future
fundamentals and stock returns than is the trading of other institutions. Specifically, Yan and Zhang
(2009) follow Gaspar, Massa, and Matos (2005) and construct an investment horizon measure based on
institutions’ portfolio turnover. They find that short-term institutions’ ∆IO forecasts both earnings
surprises and stock returns, while the predictive power is negligible for long-term institutions’ trading.
In a similar vein, we argue that PC_NII has stronger predictive power than ∆IO because PC_NII takes
into account the heterogeneity in institutions’ skills and information. Moreover, consistent with the
results by Ke and Ramalingegowda (2005), Bushee and Goodman (2007), Yan and Zhang (2009), and Jia
and Liu (2011), we find that short-term PC_NII forecasts the cross section of stock returns, while the
predictive power is weak for long-term PC_NII. Nevertheless, PC_NII is potentially a better measure of
informed trading than is short-term ∆IO because (1) only a small fraction of short-term institutions are
informed and (2) some long-term institutions have superior information. As conjectured, we find that
PC_NII drives out the short-term ∆IO in the forecast of stock returns; and our results also differ from
those in Yan and Zhang (2009) along 2 important dimensions. First, as we elaborate below, the trading
strategy formed on PC_NII exhibits a significant long-term price reversal, while Yan and Zhang (2009)
find no reversal for the portfolios formed on ∆IO. Second, unlike Yan and Zhang (2009), we show that
Regulation FD has important effects on the information content of institutions’ trading.
We document a long-run price reversal: Stocks with high PC_NII are outperformed by stocks with
low PC_NII a few quarters after the portfolio formation. This finding seems to be puzzling because it
contradicts the conventional wisdom that informed trading facilitates the incorporation of information into
7
stock prices. It is, however, consistent with the notion of “buying on rumor and selling on news” for a
positive rumor (news), as Brunnermeier (2005) highlights in his rational expectations model of informed
trading.3 In the model, the investor with noisy private information about an impending public
announcement trades aggressively on his information before the announcement. The exaggerated
trading leads uninformed investors to overshoot at the announcement, and the informed investor can
profit from his information again by reverting partially his position following the announcement.
Consistent with this prediction, we find that, coinciding with the price reversal, short-term institutions
reduce the holding of stocks in the top PC_NII quintile and increase the holding of stocks in the bottom
PC_NII quintile. Interestingly, long-term institutions trade in exactly the opposite directions of
short-term institutions, while we do not observe any obvious link between the trading of all institutions
and long-term price reversals.4 These results are consistent with the maintained hypothesis that
informed (short-term) institutions trade frequently with uninformed (long-term) institutions. To the best
of our knowledge, the empirical finding that informed trading leads to long-run price reversals is novel.
The remainder of the paper is organized as follows. Section 2 discusses the information content of
PC_NII and compares it with the two standard institutional trading measures, i.e., ΔIO and ΔNII.
Section 3 describes data. Section 4 shows that PC_NII has strong predictive power for short-term stock
returns. Section 5 examines whether the relationship between PC_NII and future stock returns is driven
mainly by institutions’ superior information about future earnings. Section 6 discusses the effect of
Regulation FD on the information content of PC_NII. Section 7 investigates the long-run performance
of the portfolios formed on PC_NII. Section 8 offers some concluding remarks. A simple stylized
model and simulation results as well as additional tables are presented in the Appendix. 3 Brunnermeier (2005) provides an intriguing example to highlight the importance of this phenomenon. “The New York Times has the following to say about the price movement of BJ Services, an oil and gas company, prior to a negative public earnings announcement in August 1993. Sell on the rumor, buy on the news. That’s Wall Street’s advice for individual investors. But the pros have a different refrain: sell when company officials tell you the news, buy when they tell everyone else.” Fishman and Hagerty (1992), van Bommel (2003), and others also show in their theoretical models that informed trading may lead to long-run price reversals. 4 Many authors have also investigated the information content of retail investors’ net buying and reached conflicting conclusions. While Barber, Odean, and Zhu (2009) and others find that retail investors’ net buying reflects mainly behavioral-based herding and leads to a long-run price reversal, Kaniel, Liu, Saar, and Titman (2011), Kelley and Tetlock (2012), and others show that it forecasts earnings surprises and does not cause a long-run price reversal.
8
2. Information Content of PC_NII
The economic rationale for the predictive power of ΔNII is similar to that ΔIO, the most commonly
used measure of institutional trading in the literature. That is, institutional investors’ trading reveals
their private information. The 2 institutional trading measures, however, differ along an important
dimension. The predictive power of ΔIO hinges on the premise that institutions as a group are
homogeneous and better informed than individual investors, while for ΔNII we need a weaker assumption
that only a small fraction of institutions have private information. That is, as we elaborate below, unlike
ΔIO, ΔNII takes into account the fact that institutions are heterogeneous in their skills and information
and that informed institutions trade frequently with uninformed institutions.
We provide a simple example to illustrate the difference between ΔNII and ΔIO. Suppose a public
company has 10 institutions and many individual investors. We first consider the effect of an informed
entry on ΔNII and ΔIO. The informed institution can purchase shares from the uninformed institutional
investors; however, the uninformed institutional investors would not sell off their positions without any
substantive negative private information. Therefore, an informed entry leads to a positive ΔNII of 1. If
the informed institution purchases most shares from only institutional investors (because institutional
investors’ trading volume dominates overwhelmingly that of individual investors; see, e.g., Kaniel, Saar
and Titman (2008)), the informed entry has negligible effects on ΔIO. In the case of an informed exit,
for 2 reasons, the existing uninformed investors of the company are more likely to be the buyers than the
other uninformed investors who currently have no stake in the company. First, information costs prevent
the other investors to initiate a position in the company’s shares. Second, Miller (1977) and many others
argue that, in the presence of divergence of opinions and short-sale constraints, the existing investors are
more optimistic about the prospect of the company than are the other investors. Therefore, if the
informed institution sells most shares to the other institutional investors of the company, the informed exit
leads to a negative ΔNII of -1 but again has negligible effects on ΔIO.
There is an important caveat, however. ΔNII is a noisy measure of informed trading due to
liquidity-based entries and exits. To address this issue, we note that the frequency of information-based
9
relative to that of the liquidity-based entries or exits increases with trading costs and information costs,
which deter liquidity-based entries and exits. That is, as we confirm empirically in Section 4, ΔNII is
more informative for stocks with higher trading costs and information costs. Specifically, ΔNII is more
informative for small stocks than for big stocks because the former have higher trading costs (e.g.,
Amihud and Mendelson (1986)) and information costs (e.g., Merton (1987)). Diamond and Verrecchia
(1991) suggest that big stocks attract more institutional investors than small stocks, and many empirical
studies (e.g., Gompers and Metrick (2001) and Bennett, Sias, and Starks (2003)) document empirical
support for this conjecture. Therefore, if the information content of ΔNII decreases with the number of
institutional investors, PC_NII, a normalization of the former by the latter, is potentially a better measure
of informed trading than is ΔNII. In Appendix A, we illustrate this point using a simple stylized model
and via simulation. Note that we may normalize ΔNII by other measures of trading costs and
information costs; however, PC_NII has the advantage of being simple and easy to construct. More
importantly, in Section 4, we show that commonly used measures of trading costs and information costs
provide little information beyond PC_NII in the forecast of stock returns.
To address heterogeneity in institutional investors, Yan and Zhang (2009) rank institutions into
terciles by their past turnover, and classify those ranked in the top tercile as short-term institutions and
those ranked in the bottom tercile as long-term institutions. Consistent with the conjecture that
short-term institutions are better informed than are long-term institutions, Yan and Zhang (2009) find that
short-term institutions’ ΔIO has significant forecasting power for stock returns, while long-term
institutions’ ΔIO does not. Partitioning institutions into terciles is arguably a coarse way to address the
heterogeneity. Specifically, a substantial fraction of short-term institutions are arguably uninformed,
while some long-term institutions have superior private information. Moreover, by construction,
short-term institutions have large turnover, many of which may reflect portfolio rebalances and liquidity
trading rather than private information. For these reasons, PC_NII, which does not require any ad hoc
partition of institutions, is potentially a better measure of informed trading than is the short-term ΔIO.
10
3. Data
We obtained quarterly institutional ownership data from the Thomson Financial 13f database. The
Thomson financial 13f database on Wharton Research Data Services (WRDS) underwent an important
change in July 2008. For each MGRNO-RDATE (i.e., institution-report date) combinations, there might
be multiple records in Thomson data (multiple FDATE or vintage dates). Before July 2008, WRDS
filtered out the redundant records. After July 2008, WRDS provides all data as it is from Thomson
without any filters. In the first version of this paper, we used the filtered pre-July 2008 13f database.
In this version, we used post-July 2008 13f database and filtered out redundant records in order to
accurately calculate the institutional variables. We find qualitatively similar results for the two datasets,
and the pre-July 2008 results are available on request.
Data of stock returns, stock prices, shares outstanding, and trading volume are from the Center for
Research on Security Prices (CRSP). We included only common stocks (CRSP codes 10 or 11) listed on
NYSE, AMEX, or NASDAQ. To mitigate market microstructure-related issues, we followed Jegadeesh
and Titman (2001) and many others by including only stocks with a price of no less than $5 at the end of
the portfolio-formation period. Compustat provides earnings and other accounting data such as the book
value of equity, and we obtained analyst earnings forecast data from I/B/E/S. The sample spans the
January 1982 to December 2010 period.
PC_NII is the percentage change in the number of a stock’s institutional investors:
i,t
# of inst. holding stock at time # of inst. holding stock at time -1PC_NII = *100
# of inst. holding stock at time -1
i t i t
i t
.
Because the denominator of PC_NII is a stock’s beginning number of institutional investors in a quarter,
we winsorize PC_NII at the 1 and 99 percentiles each quarter to remove potential outliers. For firms
that have no institutional investors at time t-1 but have a positive number of institutional investors at time
t, PC_NII is assigned as a missing observation for time t. In each quarter, we calculate the number of
institutional investors for each stock, and Figure 1 shows a steady increase in its cross-sectional median
(dashed line) across time from 10 institutions in 1982 to 110 institutions in 2010. The pattern is
11
qualitatively similar for the cross-sectional mean (solid line), although it is substantially larger than the
cross-sectional median, indicating that the number of institutional investors is positively skewed. Figure
2 plots the cross-sectional 25 (solid line) and 75 (dashed line) percentiles of changes in institutional
investors across time. Because the median change in institutional investors is close to 0, the absolute
value of the 25 percentile is a proxy for the median number of exits, while the 75 percentile is a proxy for
the median number of entries. The 25 percentile varies across time with a tight range between 0 and -4,
with the exception of -7 during the 2008 financial crisis. Similarly, the 75 percentile has a range
between 2 and 8, with the exception of 9 during the 2009 rebound following the 2008 financial crisis.
Moreover, Table 1 shows that a substantial fraction of stocks (an average of 15% over the 1982 to 2010
period) has zero change in the number of institutional investors. These results suggest that, every
quarter, only a relatively small fraction of institutions exit from or enter into a stock possibly because an
entry or exit is often triggered by substantive private information. Table 1 also shows that the
cross-sectional mean is always positive and the cross-sectional median is mostly positive for PC_NII
across time because of the steady increase of institutional investors in our sample, as shown in Figure 1.
Moreover, the mean is substantially greater than the median, indicating that PC_NII is positively skewed.
For comparison, we also consider several commonly used measures of institutional trading. First is
the change in institutional ownership adopted by Nofsinger and Sias (1999), Gompers and Metrick (2001),
Cai and Zheng (2004), Yan and Zhang (2009), and many others:
i, tΔIO = % of stock held by inst. at time % of stock held by inst. at time -1i t i t .
Yan and Zhang (2009) construct an investment horizon measure based on institutions’ portfolio turnover,
and find that short-term institutions’ ∆IO has significant predictive power for earnings surprises and stock
returns. For comparison, we also follow Yan and Zhang’s (2009) procedures and construct the
short-term ∆IO. Second is the change in the breadth of institutional ownership used by Chen, Hong, and
Stein (2002) and Lehavy and Sloan (2008):
i, t
# of inst. holding stock at time # of inst. holding stock at time -1ΔBREADTH = *100
Total # of 13f filers in the market at time -1
i t i t
t
.
12
Following Chen, Hong, and Stein (2002) and Lehavy and Sloan (2008), when calculating ΔBREADTH
we include only 13f filers that are in the sample in both time t and time t-1 to calculate (1) the number of
institutions that hold the stock and (2) the number of all institutions in the market. Last is the change in
the number of institutional investors considered by Sias, Starks, and Titman (2006):
i, tΔNII = # of inst. holding stock at time # of inst. holding stock at time -1i t i t .
Note that, at any time t, ΔNII is proportional to ΔBREADTH because the denominator for ΔBREADTH is
identical for all stocks. In particular, consistent with findings by Chen, Hong, and Stein (2002), Sias,
Starks, and Titman (2006) document a significantly positive correlation of ΔNII with future stock returns.
For brevity, we report only ΔBREADTH in the empirical analysis of this paper because the results are
virtually identical for ΔNII (the results based on ΔNII are available on request). By contrast, as
mentioned earlier, PC_NII is substantially different from ΔBREADTH because the denominator for
PC_NII is stock-specific to recognize the fact that the information content of ΔNII varies across stocks.
Many authors, e.g., Nofsinger and Sias (1999), Wermers (1999), and Sias (2004), find a positive
relation between institutional herding and short-term future stock returns. For comparison, we construct
standard institutional herding measures advocated by (1) Lakonishok, Shleifer, and Vishny (1992),
LSV_HM, (2) Sias (2004), S_HM, and (3) Brown, Wei, and Wermers (2009), BWW_HM. Lastly, in the
empirical analysis, we conjecture that the private information conveyed by PC_NII is mainly related to
future earnings, a closely watched measure of fundamentals. we define the earnings surprise,
E_SURPRISE, as the difference between actual earnings per share and the median analyst forecast
divided by the end-of-quarter stock price. The seasonal earnings growth, E_CHANGE, is the change in
quarterly earnings before extraordinary items from a year ago scaled by average total assets.
In Table B1 of Appendix B, we report the time-series average of the cross-sectional correlation
coefficients for measures of institutional trading, institutional herding, and earnings. PC_NII correlates
closely with ΔBREADTH, with a correlation coefficient of 45%. Nevertheless, the correlation is far
from being perfect because of substantial variation in the number of institutional investors across stocks.
13
For example, over the 1982 to 2010 period, its 25 and 75 percentiles are 13 and 104 institutions,
respectively (untabulated). Therefore, we may observe quite different values of PC_NII for stocks with
the same value of ΔNII or ΔBREADTH. Because (as we show in Section 4) the number of institutional
investors correlates negatively with the informativeness of entries and exits, its’ large cross-sectional
variation highlights the importance of using PC_NII instead of ΔBREADTH as a measure of informed
trading. PC_NII also correlates positively with ΔIO, with a correlation coefficient of 32%. The
correlation is relatively weak partly because, as we mentioned in Section 2, institutional investors’ trading
volume dominates overwhelmingly that of individual investors and informed institutions trade frequently
with uninformed institutions. Moreover, entry and exit trades account for only a fraction of institutional
investors trading activity. PC_NII also correlates positively with the sort-term ΔIO and the institutional
herding measures, LSV_HM, S_HM, and BWW_HM; and again, the correlations are not particularly
strong. Lastly, we find a positive correlation of PC_NII with the most recently disclosed E_CHANGE
and E_SURPRISE.
4. PC_NII and Short-Term Future Stock Returns
In this section, we show that the proposed measure of informed trading, PC_NII, strongly forecasts
one-quarter-ahead stock returns. For robustness, we use both portfolio sorts and the Fama and MacBeth
(1973) cross-sectional regression, and find qualitatively similar results.
4.1 Single Sort on PC_NII
At the end of each quarter, we sort stocks equally into 10 portfolios by PC_NII, and hold them over
the next 3 months. Panel A of Table 2 reports selected characteristics for each decile portfolio. D1 is
the decile with the lowest PC_NII, and D10 is the decile with the highest PC_NII. We observe
substantial variation in PC_NII across the decile portfolios: The time-series mean of average PC_NII
14
ranges from -22% for D1 to 59% for D10.5 Both ΔBREADTH and ΔIO increase monotonically from D1
to D10. PC_NII is an inverted U-shaped function of market capitalization (SIZE). We are more likely
to observe extreme PC_NII for small stocks than for big stocks because the former have substantially less
institutional investors. Note that we filter out stocks with a price less than $5; and our main results are
qualitatively similar or stronger when we remove stocks that have less than 5 institutional investors
(untabulated). Moreover, as we show in subsection 4.3, the information content of PC_NII does not
concentrate in only ultra-small stocks. Because small stocks tend to have high idiosyncratic volatility
and to be illiquid, PC_NII is also approximately a U-shaped function of idiosyncratic volatility (IV) and
the Amihud (2002) illiquidity measure (AMIHUD). PC_NII has an inverse relationship with the
book-to-market equity ratio (BM) and a positive relationship with the stock return in the past 6 months
(MOM). Similarly, PC_NII increases monotonically with contemporaneous (i.e., the most recently
disclosed) seasonal earnings growth (E_CHANGE) and earnings surprises (E_SURPRISE).6 These
results, which suggest that institutions prefer growth stocks and tend to be momentum traders, are
consistent with those reported in earlier studies (e.g., see Wermers (1999, 2000), Cai and Zheng (2004),
Sias (2004, 2007), and Ke and Ramalingegowda (2005)).
In Panel B of Table 2, we investigate the relationship between PC_NII and future portfolio returns.
As conjectured, we document a strong positive correlation of PC_NII with one-quarter-ahead
equal-weighted portfolio returns. Among the 10 portfolios, the bottom PC_NII decile has the lowest raw
return, while the top PC_NII decile has the highest raw return. The hedge portfolio of longing the top
and shorting the bottom PC_NII deciles has an average raw return of 2.4% per quarter, with a t-statistic of
5 As we show in Table 1, PC_NII equals 0 for a substantial fraction of stocks. To sort stocks evenly into deciles, we add a small uniform-distributed random variable (i.e., 10-9*Uniform) to PC_NII as a tie breaker. Because PC_NII has a large cross-sectional variation, such a small perturbation has no effect on the top and bottom deciles, which are the main concern of our investigation. In subsection 4.5, we show that results are qualitatively similar for the Fama and MacBeth regression, in which we use the raw data for PC_NII. We also apply a similar perturbation for ΔBREADTH in our analysis. 6 When we require analyst forecast data, our dataset reduces from 373498 observations to only 175653 observations (a 53% reduction). To address this issue, we calculate the decile portfolio median E_SURPRISE using the subsample with valid earnings surprise observations.
15
4.5.7 To ensure that our results do not simply reflect the fact that many institutions adopt certain
investment styles, we follow Daniel, Grinblatt, Titman, and Wermers (DGTW hereafter; 1997) and
Wermers (2004) and use the DGTW characteristics-adjusted returns, and find qualitatively similar results
(2.4% per quarter with a t-statistic of 5.3).8 Similarly, the return difference cannot be explained by
standard empirical risk factor models either. Its loadings on excess market returns and the size premium
are statistically insignificant at the 5% level (untabulated). Because the top PC_NII decile has a lower
book-to-market equity ratio than does the bottom PC_NII, the hedge portfolio has negative loadings on
the value premium. As a result, the Fama and French (1996) 3-factor alpha of the hedge portfolio is
2.9%, which is actually larger than the raw return. On the other hand, because institutions tend to buy
past winners and sell past losers, the loadings on the momentum factor are positive. Nevertheless, when
we include the control for the momentum factor, the Carhart (1997) 4-factor alpha is 1.8% with a
t-statistic of 4.8. Moreover, further controlling for loadings on the illiquidity risk factor proposed by
Pastor and Stambaugh (2003) in a 5-factor model does not affect our results in any qualitative manner
(the 5-factor alpha is 1.8% with a t-statistic of 5.3). Results are also similar if we use DGTW
characteristics-adjusted portfolio returns to calculate hedge portfolio alphas (untabulated). As a
robustness check, in Appendix B, Table B2 shows that the results are qualitatively similar or somewhat
stronger for value-weighted portfolio returns.
PC_NII forecasts stock returns partly because of its information content about future earnings
surprises. To investigate this conjecture, we follow Ali, Durtschi, and Trombley (2004) and construct
the raw return and the DGTW characteristics-adjusted return over a 3-day event window around the
earnings announcement date. Panel B of Table 2 shows that, as hypothesized, the difference between the
7 We use a one-quarter lag throughout to compute the New-West (1987) corrected t-statistics. However, results of the paper are insensitive to the lag length used in the Newey-West adjustment (e.g., we have verified that the results do not change in any qualitative manner if we use an eight-quarter lag instead to compute the Newey-West (1987) corrected t-statistics throughout). 8 The DGTW characteristics-adjusted return of a stock is the difference between the stock’s raw return and the return of the stock’s benchmark portfolio matched by size, book-to-market, and momentum. DGTW benchmarks are available from http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverpage.htm. We thank Prof. Russ Wermers for generously sharing the data on his website.
16
top and bottom deciles is significantly positive at the 1% level for both the raw return and the DGTW
characteristics-adjusted return. The magnitude, however, is somewhat small. For example, the raw
return difference is 0.3% over the 3-day window around the earnings announcement date, compared with
2.4% over the whole quarter (i.e., these 3 days, which account for about 4.5% of trading days of a quarter,
explain 12.8% of the quarterly hedge portfolio return). This pattern is consistent with the finding by
Ball and Shivakumar (2008) that returns around earnings announcements account for a relatively small
fraction of variation in quarterly stock returns.9 In the next section, we show that PC_NII has significant
forecasting power for future earnings and that the predictive power of PC_NII for stock returns reflects
mainly its information about future earnings.
In Figure 3, we plot the annual hedge portfolio returns across time, which we constructed by
compounding quarterly returns on the hedge portfolio of longing the top and shorting the bottom PC_NII
deciles. Strikingly, over the sample period of almost three decades, the equal-weighted return (solid line)
is always positive or close to 0 except for the years 2001 (the tech bubble burst) and 2009 (the aftermath
of the subprime mortgage loan crisis), during which the hedge portfolio has a substantial negative return
of -19% and -9%, respectively. The dashed line in Figure 3 shows that results are qualitatively similar
for the value-weighted hedge portfolio return.
It is tempting to suggest that the hedge portfolio’s poor performance in the years 2001 and 2009
reflects mainly systematic risk; therefore, the superior performance of the hedge portfolio formed on
PC_NII is the compensation for disastrous risk that cripples the financial market by draining its liquidity
quickly. A closer look at the data indicates that this is not the case. Specifically, the poor performance
in 2001 is primarily due to the substantial negative hedge portfolio return in the 4th quarter, while the
excess market return is 12% for that quarter. Similarly, the poor performance in 2009 is mainly due to
9 Existing studies (e.g., Ali, Durtschi, and Trombley (2004), Yan and Zhang (2009), and Baker, Litov, Wachter, and Wurgler (2010)) use either equal-weighted or median returns around earnings announcement dates. Interestingly, Table B2 in Appendix B shows that the relationship between PC_NII and future returns around earnings announcement dates is positive albeit statistically insignificant when we use value-weighted returns, possibly because big stocks disseminate information faster than do small stocks (Diamond and Verrecchia (1991)). For example, for those relatively large stocks that have analyst following, substantial amount of information is incorporated in analyst forecast revisions prior to earnings announcements (Ball and Shivakumar (2008)).
17
the substantial negative hedge portfolio return in the 2nd quarter, during which the excess market return is
18%. Moreover, Table 2 shows that the liquidity risk proposed by Pastor and Stambaugh (2003) has
negligible explanatory power for the return difference between the top and bottom PC_NII deciles.
There are 2 possible alternative explanations. First, existent literature suggests that institutions’
trading is significantly affected by liquidity demand during financial market meltdowns (e.g., Anand,
Irvine, Puckett, and Venkataraman (2011), Ben-David, Franzoni, and Moussawi (2011), Griffin, Harris,
Shu, and Topaloglu (2011), and He, Khang, and Krishnamurthy (2010)). Therefore, during market
meltdowns, changes in the number of institutional investors of a stock are more likely to reflect liquidity
demand than to reflect the private information about the stock’s fundamentals. Consistent with this
liquidity-based explanation, Figure 2 documents an unusually low level for the 25 percentile of changes in
institutional investors in 2008, indicating that institutions rushed for exits during the financial crisis.
Moreover, Figure A1 in Appendix A shows that the increase in exits is especially pronounced for liquid
stocks, i.e., stocks with a large number of institutional investors. Therefore, PC_NII is a less reliable
measure of private information during these episodes than during normal market conditions.
Second, the noticeable poor performance of the trading strategy based on PC_NII in 2001 and 2009
also reflects the fact that institutions tend to be momentum traders. The momentum portfolio of buying
past winners and selling past losers has substantial negative returns of -16% and -40% for the 4th quarter
of 2001 and the 2nd quarter of 2009, respectively, during which the hedge portfolio formed on PC_NII
performs poorly. That is, when PC_NII is a poor proxy of institutions’ private information during
market turmoil, the performance of the hedge portfolio formed on PC_NII is primarily influenced by the
performance of the momentum portfolio. Nevertheless, PC_NII provides additional information about
future stocks returns during normal market conditions. Table 2 shows that the hedge portfolio has a
significant DGTW characteristics-adjusted return (that controls for momentum) and a significant alpha
after we control for the momentum factor. Similarly, as we show in subsections 4.3 and 4.4, PC_NII
remains a significant predictor of cross-sectional stock returns when we control for past returns.
To summarize, with a negative return in only 2 out of 29 years, the superior performance of the
18
hedge portfolio formed on PC_NII cannot be easily reconciled with risk-based asset pricing theories.
4.2 PC_NII versus ΔBREADTH
We conjecture that PC_NII is a better measure of informed trading than is ΔBREADTH because
PC_NII explicitly takes into account the fact that an institution’s entry or exit is more informative for the
stocks with higher trading costs and information costs. In this subsection, we test 3 hypotheses that we
develop in Section 2 and Appendix A. First, ΔBREADTH is more informative of future stock returns for
those stocks with higher trading costs and information costs. Second, such a relation is much weaker for
PC_NII. Last, PC_NII is more informative about future stock returns than is ΔBREADTH.
We measure a stock’s trading costs and/or information costs using market capitalization, the number
of institutional investors, Amihud’s (2002) illiquidity measure, analyst earnings forecast dispersion, the
number of analysts following the stock, turnover, and the probability of informed trading (PIN) advocated
by Hvidkjaer, Easley, and O’Hara (2002). For each of the 7 measures, we sort stocks equally into 2
portfolios each quarter and construct a dummy variable that equals 1 for the stocks with low trading
and/or information costs and equals 0 otherwise. In Fama and MacBeth (1973) cross-sectional
regressions, we use (1) the measure of institutional trading, i.e., ΔBREADTH or PC_NII, (2) the dummy
variable for trading and/or information costs, and (3) their interaction term, to forecast one-quarter-ahead
cross-section of stock returns. We argue that changes in the number of institutional investors (ΔNII or
ΔBREADTH) are more likely to reflect liquidity-based trading than information-based trading for stocks
with lower trading and/or information costs. Therefore, we anticipate that the interaction term is
significantly negative for ΔBREADTH. On the other hand, we argue that the asymmetric effect is
specifically taken into account by PC_NII and thus expect that the interaction term is statistically
indifferent from 0 for PC_NII. As a robustness check, we control for standard cross-sectional stock
return predictors, including market beta, market capitalization, the book-to-market equity ratio, the return
in the previous 6 months, Amihud’s (2002) illiquidity measure, and past seasonal earnings growth. For
brevity, we omit the regression results for these control variables but they are available on request.
19
Panel A of Table 3 reports the Fama and MacBeth (1973) cross-sectional regression results for
ΔBREADTH. The 1st column lists the variables that we use to construct the dummy variable. For
example, for “Large Size”, the dummy variable is set equal to 1 if the market capitalization of the stock is
larger than the sample median in a quarter and set equal to 0 otherwise. Consistent with the finding by
Chen, Hong, and Stein (2002) and Sias, Starks, and Titman (2006), we find a significantly positive
correlation of ΔBREADTH with future stock returns. As conjectured, ΔBREADTH is less informative
for stocks with lower trading costs and information costs. Specifically, its interaction terms with the
dummy variables are always significantly negative at least at the 10% level. In contrast, Panel B shows
that the interaction terms of PC_NII with these dummy variables are statistically insignificant at the 10%
level, while PC_NII is always significantly positive at the 1% level.
In Table 4, we directly test the hypothesis that PC_NII is a better measure of informed trading than is
ΔBREADTH by forming portfolios using the two institutional trading measures. In Panel A, we
investigate whether PC_NII forecasts stock returns when we control for ΔBREADTH. Specifically, each
quarter we first sort stocks equally into 5 portfolios by ΔBREADTH; and then within each ΔBREADTH
quintile we sort stocks equally into 5 portfolios by PC_NII. The Carhart (1997) 4-factor alpha of the
equal-weighted return difference between the top and bottom PC_NII quintiles is always positive for all
ΔBREADTH quintiles; and it is statistically significant at least at the 5% level for 3 out of 5 ΔBREADTH
quintiles. In the last row of Panel A, we aggregate each PC_NII quintile returns across 5 ΔBREADTH
quintiles using the equal weight, and find that the return difference between the top and bottom aggregate
PC_NII quintiles is about 1% per quarter, with a t-statistic of 5.5. In contrast, in Panel B, we show that
ΔBREADTH has negligible predictive power for stock returns when we control for PC_NII.
Specifically, the return difference between the top and bottom ΔBREADTH quintiles is always
statistically insignificant for each PC_NII quintile. When we aggregate across PC_NII quintiles, the
difference is economically negligible (0.03%) and statistically insignificant. In Tables B3 and B4, we
show that results are qualitatively similar using value-weighted portfolio returns. In subsection 4.5, we
also find qualitatively similar results using the Fama and MacBeth (1973) cross-sectional regression. To
20
summarize, as conjectured, PC_NII is a better measure of informed trading than is ΔBREADTH.
Chen, Hong, and Stein (2002) argue that ΔBREADTH is a proxy for short-sale constraints. In
particular, stocks with an increase in institutional investors are less likely to have a binding short-sale
constraint than are stocks with a decrease in institutional investors. Similar to Miller (1977), Harrison
and Kreps (1978), Scheinkman and Xiong (2003), and others, Chen, Hong, and Stein (2002) show that in
the presence of heterogeneous beliefs, because short-sale constraints temporarily suppress pessimistic
investors’ view, stocks with binding short-sale constraints tend to have lower future returns than do stocks
with nonbinding short-sale constraints, ceteris paribus. Because stocks with higher trading costs and
information costs arguably are more vulnerable to short-sale constraints and have stronger divergence of
opinion, results reported in Panel A of Table 3 are potentially consistent with Chen, Hong, and Stein’s
(2002) hypothesis. Nevertheless, their hypothesis does not explain why the predictive power of
ΔBREADTH becomes negligible when we control for PC_NII. More importantly, it does not explain
why, as we show in Section 5, the predictive power of PC_NII and ΔBREADTH for stock returns reflects
mainly its information content about future earnings.10
4.3 Controlling for Determinants of Cross-Sectional Stock Returns
In this subsection, by forming portfolios using various sequential double sorts, we show that the
predictive power of PC_NII for stock returns is not a manifestation of well-known asset pricing anomalies
documented in the existent literature. For brevity, we report only equal-weighted portfolio returns in
Panel A of Table 5; in Appendix B, Panel A of Table B3 shows that results are qualitatively similar for
value-weighted portfolio returns.
The top and bottom PC_NII deciles have relatively small market capitalization (Panel A of Table 2).
10 Chen, Hong, and Stein’s (2002) short-sale constraint hypothesis implies a negative contemporaneous correlation of ΔBREADTH or PC_NII with stock valuations. In contrast with this view, Panel A of Table 2 shows an inverse relationship between PC_NII and the book-to-market equity ratio. Similarly, Fama-Macbeth regressions show that both ΔBREADTH and PC_NII correlate significantly and positively with contemporaneous Tobin’s Q at the 1% level even when we control for size, momentum, lag value of Tobin’s Q, and industry dummies. For brevity, we do not report these results but they are available on request.
21
Many authors (e.g., Wermers (1999), Sias (2004), and Yan and Zhang (2009)) argue that institutions have
a greater informational advantage for small stocks than for big stocks because the former have more
uncertainty and are more difficult to evaluate. To investigate this issue, we form portfolios using
sequential double sorts first by market capitalization and then by PC_NII. Specifically, at the end of
each quarter, we first sort stocks equally into 5 portfolios by market capitalization; within each size
quintile, we then sort stocks equally into 5 portfolios by PC_NII. We hold the portfolios over the next
quarter and rebalance them at the end of the holding period. In Panel A of Table 5, we report the Carhart
(1997) 4-factor alphas of excess portfolio returns for the 25 portfolios. Within each size quintile, the
return difference between the top and bottom PC_NII quintiles is positive, and the difference is
statistically significant at the 1% level except for the quintile of biggest stocks. When we aggregate the
top and bottom PC_NII quintile portfolio returns across the size quintiles using the equal weight, the
return difference between the top and bottom aggregate PC_NII quintiles has an alpha of 1.3% per quarter,
with a t-statistic of 4.5. Therefore, as conjectured, the predictive power of PC_NII is weaker among the
biggest stocks; nevertheless, PC_NII has strong return forecasting power for most of the stocks in the
market and its predictive power does not concentrate in only ultra-small stocks.
The top PC_NII decile has a smaller book-to-market equity ratio than does the bottom PC_NII decile.
Such a relation does not help explain our main finding because the book-to-market equity ratio correlates
positively with future stock returns (e.g., Fama and French (1992)). We illustrate this point in the
manner similar to that for the size effect. For brevity, in Panel A of Table 5, we report only the Carhart
(1997) 4-factor alphas of PC_NII quintiles aggregated across all book-to-market quintiles.11 The top
PC_NII quintile has the highest alpha, while the bottom PC_NII quintile has the lowest alpha. Moreover,
the return difference between the two extreme PC_NII quintiles has an alpha of 1.5% per quarter, with a
t-statistic of 5.7. Therefore, as expected, the book-to-market effect does not explain the positive
11 Similar to the evidence documented by Yan and Zhang (2009), we find that the relationship between PC_NII and future stock returns is somewhat stronger for stocks with lower book-to-market equity ratios. Nevertheless, for the quintile of stocks with the highest book-to-market equity ratio, the return difference between the top and bottom PC_NII quintiles still has a significantly positive Carhart (1997) 4-factor alpha for both equal-weighted and value-weighted portfolio returns. We do not tabulate these results for brevity but they are available on request.
22
correlation of PC_NII with future stock returns.
Panel A of Table 2 shows that the top PC_NII decile has a substantially higher average return in the
past 6 months than does the bottom PC_NII decile. Many authors (e.g., Jegadeesh and Titman (1993))
find that past winners continue to outperform past losers in the next a few quarters. Therefore, it is
possible that the predictive power of PC_NII for stock returns is simply a manifestation of the momentum
effect. However, in contrast with this conjecture, Panel B of Table 2 shows that the top PC_NII decile
outperforms the bottom PC_NII decile even when we explicitly control for the momentum effect using
either DGTW characteristics-adjusted returns or Carhart (1997) 4-factor alphas. Panel A of Table 5
further confirms that PC_NII remains a significant predictor of cross-sectional stock returns when we
control for the return in the past 6 months—the return difference between the top and bottom aggregate
PC_NII quintiles have a 4-factor alpha of 1.1% with a t-statistic of 4.9 when controlling for momentum.12
In a similar vein, Panel A of Table 2 documents a positive correlation of PC_NII with contemporaneous
seasonal earnings changes and earnings surprises. The predictive power of PC_NII for stock returns
thus may reflect the post-earnings announcement drift, as documented by Bernard and Thomas (1989)
and many subsequent studies. Again, we find that the post-earnings announcement drift does not fully
explain the predictive power of PC_NII. Panel A of Table 5 shows that the return difference between the
top and bottom aggregate PC_NII quintiles has a Carhart (1997) 4-factor alpha of 1.1% with a t-statistic
of 2.7 when controlling for the most recently disclosed earnings surprise. Similarly, when we control for
the most recently disclosed seasonal earnings growth, the alpha is 1.1%, with a t-statistic of 4.3.
Lastly, Amihud and Mendelson (1986) and many subsequent studies find a positive relationship
between illiquidity and expected stock returns, and Ang, Hodrick, Xing, and Zhang (2006, 2009) find a
negative relationship between idiosyncratic volatility and future stock returns. We construct the
illiquidity measure as in Amihud (2002) and the idiosyncratic volatility as in Ang, Hodrick, Xing, and
Zhang (2006). Panel A of Table 5 shows that controlling for these two variables does not change our
12 Results are qualitatively similar if we measure the momentum using the return over the past 12 months or the return from the past 12th month to the past 4th month (untabulated).
23
main finding in any qualitative manner. To summarize, the predictive power of PC_NII for stock returns
is distinct from the well-known anomalies documented in the existent literature.
4.4 Controlling for Standard Measures of Institutional Trading and Herding
In this subsection, by forming portfolios using sequential double sorts first on a measure of
institutional trading or herding and then on PC_NII, we show that PC_NII has stronger forecasting power
for stock returns than commonly used measures of institutional trading and herding. Again, for brevity,
we report only equal-weighted PC_NII quintile portfolio returns aggregated across the quintile portfolios
formed on a measure of institutional trading or herding in Panel B of Table 5; in Appendix B, Panel B of
Table B3 shows that results are qualitatively similar for value-weighted portfolio returns. The detailed
double-sort results of the paper are available from us on request.
Gompers and Metrick (2001) document a positive relationship between institutional ownership, IO,
and future stock returns. These authors interpret IO as a measure of institutions’ persistent demand
shocks. Many authors (e.g., Asquith, Pathak, and Ritter (2005) and Nagel (2005)), however, argue that
IO is a proxy for short-sale constraints. This is because increases in institutional ownership increase the
supply of equity shares available for lending and thus alleviates short-sale constraints. Consistent with
this view, D’Avolio (2002) finds that stocks with a large fraction of institutional ownership tend to have
low lending fees on shorted shares. In contrast, Yan and Zhang (2009) show that the predictive power of
IO for stock returns reflects mainly its correlation with changes in ownership of short-term institutions,
short-term ΔIO, a measure of institutions’ private information. Panel B of Table 5 shows that, when we
control for IO, the return difference between the top and bottom aggregate PC_NII quintiles has a Carhart
(1997) 4-factor alpha of 1.3% per quarter, with a t-statistic of 4.9. Similarly, we find that PC_NII
remains a significant stock return predictor when we control for ΔIO, the change in institutional
ownership. Moreover, Yan and Zhang (2009) show that the short-term institutions’ ΔIO has strong
predictive power for stock returns, while the predictive power of long-term institutions’ ΔIO is negligible.
However, Panel B of Table 5 shows that controlling for the short-term ΔIO has negligible effect on our
24
main findings—the return difference between the top and bottom aggregate PC_NII quintiles has a
Carhart (1997) 4-factor alpha of 1.3% per quarter, with a t-statistic of 4.7.13 By contrast, in Appendix B,
Table B4 shows that IO, ΔIO, or the short-term ΔIO has negligible predictive power for stock returns
when we control for PC_NII.
Lastly, many authors (e.g., Nofsinger and Sias (1999), Wermers (1999), and Sias (2004)) find a
positive relationship between herding and short-term future stock returns. Again, Panel B of Table 5
shows that PC_NII continues to have significant forecasting power for stock returns when we control for
the institutional herding measure proposed by Lakonishok, Shleifer, and Vishny (1992), LSV_HM, Sias
(2004), S_HM, or Brown, Wei, and Wermers (2009), BWW_HM. By contrast, in Appendix B, Table B4
shows that LSV_HM, S_HM, or BWW_HM has negligible predictive power for stock returns when we
control for PC_NII.
To summarize, PC_NII has stronger predictive power for stock returns than the commonly used
measures of institutional trading and herding.
4.5 Fama and MacBeth Regressions
As a robustness check, in this subsection, we investigate the predictive power of PC_NII for stock
returns using Fama and Macbeth (1973) cross-sectional regressions. To remove potential outliers and
data errors, in each quarter we winsorize PC_NII, seasonal earnings growth, and earnings surprises at the
1 and 99 percentiles. In a similar vein, we also use the log transformation for size and the
book-to-market equity ratio. Table 6 shows that the Fama and MacBeth regression results are consistent
with those obtained from the portfolio formation analysis. First, PC_NII has a strong positive
correlation with one-quarter-ahead stock returns in the univariate regression (model 1). Second, the
13 Similar to Yan and Zhang (2009), we decompose PC_NII into a short-term component and a long-term component. The short-term (long-term) PC_NII is the change in the number of short-term (long-term) institutional investors divided by the number of all institutional investors in the previous quarter. In Appendix B, Table B5 shows that, consistent with Yan and Zhang (2009), the short-term PC_NII has significant predictive power for cross-sectional stock returns, while the predictive power is negligible for the long-term PC_NII when we use equal-weighted portfolios.
25
predictive power of PC_NII is not subsumed by commonly used cross-sectional stock return predictors.
Specifically, model 2 shows that the predictive power of PC_NII remains statistically significant at the 1%
level when we control for market beta, size, book-to-market equity ratio, momentum, illiquidity, and
idiosyncratic volatility. The results are qualitatively similar when we also include (the most recently
disclosed) seasonal earnings growth (model 3) or earnings surprises (model 4) in the forecasting
regression to control for the post-earnings announcement drift. Last, the predictive power of PC_NII is
not subsumed by commonly used measures of institutional trading. Consistent with the findings of Yan
and Zhang (2009), model 5 shows that short-term ∆IO correlates positively and significantly with
one-quarter-ahead stock returns in our sample. In model 6, we include PC_NII along with ∆BREADTH,
∆IO, short-term ∆IO, IO, and Brown, Wei, and Wermers’ (2009) signed herding measure, BWW_HM, as
independent variables, and find that PC_NII is the only variable that is statistically significant at the 1%
level.14 In contrast, the other institutional trading measures are all insignificant at the 10% level.
Moreover, PC_NII remains a significant predictor when we include both cross-sectional stock return
predictors and other institutional trading and herding measures as explanatory variables (models 7, 8, and
9).15 PC_NII also drives out the other institutional trading and herding measures when we include only
one of these measures at a time along with PC_NII in cross-sectional regressions (untabulated).
5. The Information Content of PC_NII for Future Earnings
In the preceding section, we show that PC_NII has strong predictive power for stock returns, and the
14 The results are qualitatively similar when we control for the herding measure by Lakonishok, Shleifer, and Vishny (1992), LSV_HM, or by Sias (2004), S_HM. For brevity, we do not report these results but they are available on request. As a robustness check, we follow Yan and Zhang (2009) and use weighted-least-square regressions (with log market capitalization as weights) for all the regression models analyzed in this section. The results, which we omit for brevity, are qualitatively similar to those reported in Table 6. 15 Although the coefficient of PC_NII in model 9 is comparable to that in model 8, it is only marginally significant at the 10% level in model 9, where we include both E_SURPRISE and ΔBREATH as explanatory variables. There are 2 possible explanations. First, the I/B/E/S data used for constructing E_SURPRISE include only a fraction of stocks with PC_NII observations (as mentioned earlier, there is a 53% reduction in observations when we require analyst earnings forecast data), and stocks included in I/B/E/S tend to have large market capitalization. Table 5 shows that the predictive power of PC_NII tends to be weaker for big stocks than for small stocks. Second, the predictive power of PC_NII for stock returns gets similar to that of ΔBREATH when we use only big stocks. That said, PC_NII still completely subsume the information content of ΔBREATH in model 9.
26
predictability cannot be explained by (1) risk-based asset pricing models, (2) well-documented asset
pricing anomalies, or (3) commonly used measures of institutional trading and herding. In this section,
we explore an alternative hypothesis that PC_NII forecasts stock returns because of its information
content of fundamentals, i.e., future earnings surprises. The mandatory quarterly earnings
announcement is arguably the most important channel through which publicly-traded companies regularly
disseminate cash flow information that has a direct and immediate impact on their stock prices.
Moreover, because firm-level stock price volatility reflects mainly news about fundamentals (e.g.,
Vuolteenaho (2002)), a good earnings forecast is arguably a crucially important component of
stock-picking skills. Our conjecture is also consistent with empirical results of existent studies, for
example, Cohen, Gompers, and Vuolteenaho (2002), Ali, Durtschi, Lev, and Trombley (2004), Campbell,
Ramadorai, and Schwartz (2009), Yan and Zhang (2009), and Baker, Litov, Wachter, and Wurgler (2010),
who find that institutional trading correlates with future earnings surprises or with abnormal returns
around earnings announcement dates.
In Panel A of Table 7, we investigate whether PC_NII forecasts one-quarter-ahead seasonal earnings
growth, E_CHANGE, using the Fama and MacBeth (1973) cross-sectional regression. Model 1 shows
that PC_NII correlates with future E_CHANGE (which is disclosed in the quarter following the PC_NII
measurement quarter) positively and significantly at the 1% level in the univariate regression. To
control for autocorrelation in E_CHANGE, in model 2, we include the most recently disclosed
E_CHANGE as an additional independent variable in the cross-sectional regression and find qualitatively
similar results. Ou and Penman (1989), Lev and Thiagarajan (1993), Abarbanell and Bushee (1997,
1998) and others, find that several accounting-based fundamental signals forecast earnings. As a
robustness check, we add these fundamental signals to the cross-sectional regression. We follow
Abarbanell and Bushee (1998) to construct fundamental signals, including INV (inventory), AR (accounts
receivable), CAPX (capital expenditures), GM (gross margin), S&A (selling and administrative expenses),
AQ (audit qualification), LF (labor force), ETR (effective tax rate), and EQ (earnings quality). To utilize
the most recently disclosed accounting information, we obtain the data used to calculate INV, AR, GM,
27
S&A, and LF from Compustat quarterly file up to the fiscal quarter ended in the quarter before the
PC_NII measurement quarter. We obtain the data used to calculate CAPX, AQ, ETR, and EQ from
Compustat annual file up to the fiscal year before the fiscal year of the fiscal quarter ended in the quarter
before the PC_NII measurement quarter (to avoid look-ahead bias). As with E_CHANGE and PC_NII,
we winsorize all fundamental signals except for the dummy variable AQ and EQ at 1 and 99 percentiles
each quarter. ETR and EQ have many missing observations; for robustness, we exclude them from the
regression in model 3 but include them in model 4. Again, we find a significantly positive correlation of
PC_NII with one-quarter-ahead E_CHANGE at the 1% level after controlling for fundamental signals.
Consistent with the existent studies, we find that 7 out of the 9 fundamental signals have significant
forecasting power for E_CHANGE. Panel B shows that results are qualitatively similar for forecasting
the one-quarter-ahead E_SURPRISE. As a further robustness check, in Table B6 of Appendix B, we
form portfolios using sequential double sorts first on (the most recently disclosed) E_CHANGE or
E_SURPRISE then on PC_NII. We show that PC_NII has significant predictive power for future
E_CHANGE (E_SURPRISE) even when we control for past E_CHANGE (E_SURPRISE).
Bernard and Thomas (1989) and many subsequent studies document a strong positive relationship
between stock returns and contemporaneous earnings surprises. In Table 8, we confirm this stylized fact
using either seasonal earnings growth (model 1) or earnings surprises (model 5). Because PC_NII has a
strong correlation with one-quarter-ahead earnings surprises, as we show in Table 7, its predictive power
for one-quarter-ahead stock returns may reflect mainly its information content of the one-quarter-ahead
earnings. To address formally this issue, we first run the regression of PC_NII on one-quarter-ahead
seasonal earnings growth and then use the regression residual or orthogonalized PC_NII to forecast stock
returns. Note that the results are identical if we use raw PC_NII along with the one-quarter-ahead
seasonal earnings growth in multivariate regressions (e.g., see Wooldridge (2006)). Model 2 shows that,
when we control for its correlation with future seasonal earnings growth, the predictive power of PC_NII
for stock returns becomes statistically insignificant at the 10% level in the univariate regression. The
results are qualitatively similar when we control for commonly used stock return predictors (models 3 and
28
4). When we use PC_NII orthogonalized by future earnings surprises, its effects on future stock returns
are positive and statistically significant at the 5% level (model 6). The significant relation reflects the
fact that institutions tend to be momentum traders. When we control for momentum and past seasonal
earnings changes or past earnings surprises, the orthogonalized PC_NII has negligible predictive power
for stock returns (models 7 and 8). These results are consistent with the finding that PC_NII forecast
stock returns around earnings announcement dates, as we report in Table 2.
As a robustness check, we further investigate the hypothesis that the predictive power of PC_NII for
stock returns reflects its information content of future fundamentals by performing portfolio double sorts
first by future seasonal earnings growth and then by PC_NII. Specifically, at the end of each quarter, we
sort stocks equally into 5 portfolios by one-quarter-ahead seasonal earnings growth; within each seasonal
earnings growth quintile, we then sort stocks equally into 5 portfolios by PC_NII. Panel C of Table 5
shows that, consistent with the Fama-MacBeth cross-sectional regression results reported in Table 8,
when we control for future seasonal earnings growth, the positive relationship between PC_NII and future
stock returns disappears completely. In particular, the Carhart (1997) 4-factor alpha of the return
difference between the top and bottom aggregate PC_NII quintiles is 0.04%, with a t-statistic of 0.2.
The results are also qualitatively similar when we use future earnings surprises as a control: The Carhart
4-factor alpha is -0.13%, with a t-statistic of -0.3.
Similarly, as conjectured, we also find that the predictive power of ∆BREADTH for stock returns
reflects mainly its correlation with future earnings. ∆BREADTH has significant predictive power for
earnings; and interestingly, its predictive power is stronger for stocks with higher trading costs and
information costs. More importantly, when we control for ∆BREADTH’s correlation with future
seasonal earnings growth or future earnings surprises, its predictive power for stock returns becomes
negligible or even negative. These results, which we omit for brevity, suggest that the short-sale
constraint hypothesis proposed by Chen, Hong, and Stein (2002) cannot fully explain the positive
relationship between ∆BREADTH and future stock returns.
To summarize, consistent with the view that some institutions have an informational advantage, the
29
predictive power of PC_NII or ∆BREADTH for stock returns comes mainly from its information content
of future earnings. Our results cast doubts on the alternative explanations that PC_NII forecasts stock
returns because of its correlation with momentum, the post-earnings announcement drift, or herding.
6. Regulation FD
On October 23, 2000, the SEC introduced Regulation FD in an effort to reduce selective disclosure
of material information by publicly-traded companies to analysts and other investment professionals.
Specifically, Regulation FD requires companies to make significant information public simultaneously to
all investors. If information leaks unintentionally, it should be announced to all investors within 24
hours. One main objective of Regulation FD is to reduce the informational advantage of institutional
investors relative to individual investors, and our new measure of institutions’ private information,
PC_NII, allows us to provide a direct estimate of the influence of Regulation FD on informed trading.
That is, if Regulation FD is effective, we should observe a weaker relationship between PC_NII and
future stock returns following its introduction by the SEC.
Figure 3 shows that the equal-weighted return on the hedge portfolio of longing the top and shorting
the bottom PC_NII deciles appears to have attenuated substantially since 2001. Consistent with this
visual impression, the average quarterly return is 3.5% over the 1983Q2 to 2000Q4 period and drops
sharply to 0.7% for the 2001Q1 to 2010Q4 period. Moreover, the difference in Carhart (1997) 4-factor
alphas between the two sample periods is statistically significant at the 1% level. As we discussed above,
the hedge portfolio suffers a substantial loss during the 2001 and 2009 market crashes possibly because
PC_NII is a noisy measure of institutions’ private information during the market turmoil. To ensure that
these particular episodes do not drive our finding, we exclude the observations from the years 2001 and
2009, and find that the difference in Carhart (1997) 4-factor alphas between the two sample periods
remains statistically significant at the 5% level.
The weaker predictive power of PC_NII for equal-weighted portfolio returns in the post Regulation
FD sample possibly reflects the diminishing active management skills, as documented by Barras, Scaillet,
30
and Wermers (2010). However, we show that this explanation has rather limited success. In particular,
while Regulation FD might have substantially reduced PC_NII’s information content of future
equal-weighted portfolio returns, Figure 3 shows that such an effect, however, is much weaker for the
value-weighted return on the hedge portfolio of longing the top and shorting the bottom PC_NII deciles.
Despite its poor performance during the 2001 and 2009 market crashes, the hedge portfolio still achieved
an admirable average return of 2.8% per quarter over the 2001 to 2010 period, compared with the average
of 4.1% for the pre-Regulation FD period. Moreover, the difference in returns between the two sample
periods is statistically insignificant at the 10% level.
Because the value-weighted portfolio return gives more weight to big stocks and less weight to small
stocks than does the equal-weighted portfolio return, our results suggest that Regulation FD is more
successful in reducing information asymmetry for small stocks than for big stocks.16 This interpretation
is consistent with the empirical findings by Ahmed and Schneible (2007), Eleswarapu, Thompson, and
Venkataraman (2004), Gomes, Gordon, and Madureira (2007), and others, that Regulation FD is more
effective for small stocks than for big stocks. These authors offer two possible explanations for why
Regulation FD has a different impact on companies of different market capitalization. First, large firms
have relatively rich information environments. Second, small firms have more limited resources to
provide information than do large firms. The empirical findings by Jorion, Liu, and Shi (2005) offer an
intriguing explanation as well. Regulation FD has a number of exclusions, including the disclosure of
nonpublic information to credit rating agencies. Interestingly, Jorion, Liu, and Shi (2005) examine the
effect of credit rating changes on stock prices and find that the informational effect of downgrades and
upgrades is much greater in the post-FD period. To the extent that companies with access to the bond
16 Using mutual fund stock holding data over the 1980 to 2005 period, Baker, Litov, Wachter, and Wurgler (2010) find that the predictive power of mutual fund trading for equal-weighted abnormal stock returns around earnings announcements have completely vanished after SEC introduced Regulation FD. In contrast, using a longer sample ending in 2010, we find that, although it becomes substantially weaker post Regulation FD, the equal-weighted return on the hedge portfolio of longing the top and shorting the bottom PC_NII deciles has a significantly positive Carhart (1997) 4-factor alpha at the 5% level over the 2002 to 2010 period (untabulated). More importantly, unlike Baker, Litov, Wachter, and Wurgler (2010), we show that Regulation FD has negligible effects on the predictive power of institutional trading for value-weighted portfolio returns.
31
market tend to have large market capitalization, the disclosure of nonpublic information to credit rating
agencies might help explain the limited success of Regulation FD on large stocks.
7. Long-Run Portfolio Returns
In Section 5, we find support for the conjecture that PC_NII forecasts stock returns mainly because
of its information content of short-run future earnings. An important implication of this conjecture is
that the predictive power of PC_NII becomes substantially weaker or negligible after earnings
announcements. To test this implication, we examine the long-run performance of the portfolios formed
on PC_NII. At the end of each quarter, we sort stocks equally into 5 portfolios by PC_NII. Table 9
reports equal-weighted quarterly portfolio returns over the portfolio formation period, Q0, and the
subsequent 16 quarters from Q1 (the 1st holding quarter following the portfolio formation) to Q16 (the
16th holding quarter following the portfolio formation). Consistent with the result reported in Table 2, in
the first holding quarter, the top PC_NII quintile outperforms the bottom PC_NII quintile, with a raw
return difference of 1.9% per quarter and a t-statistic of 4.3. The results are qualitatively similar for the
DGTW characteristics-adjusted return or the Carhart (1997) 4-factor alpha. Moreover, as conjectured,
the return difference between the top and bottom PC_NII quintiles becomes both statistically and
economically insignificant in Q2 and Q3, especially when we adjust for loadings on systematic risks.
Interestingly, Table 9 reveals a significant price reversal over longer horizons. The raw return
difference between the top and bottom PC_NII quintiles turns negative from Q4 through Q13, and the
negative return difference is statistically significant at least at the 10% level in a few quarters starting
from Q4 before it eventually becomes negligible. Results are qualitatively similar for the DGTW
characteristics-adjusted return or the Carhart (1997) 4-factor alpha. Because institutions tend to sell
stocks that have performed poorly in the near past, these stocks are potentially susceptible to delisting risk.
To ensure that the long-run price reversal is not due to the survivorship bias, we calculate the long-run
portfolio returns using CRSP delisting returns and follow the procedure proposed by Beaver, McNichols,
and Price (2007). In Appendix B, Table B7 shows that using delisting returns does not affect our main
32
findings in any qualitative manner. This result should not be too surprising because (1) we exclude
stocks with a price of less than $5 and (2) institutional investors tend to hold relatively big stocks that
have a small probability of being delisted.
The long-term price reversal is in general consistent with the recent empirical findings by Dasgupta,
Prat, and Verardo (2011) and Brown, Wei, and Wermers (2009), who find that institutional herding
correlates negatively with future stock returns in the long run, although many earlier authors (e.g.,
Wermers (1999) and Sias (2004)) document only a positive relationship between herding and short-run
future stock returns. In a similar vein, Puckett and Yan (2011) find that, while institutional demand
correlates positively with intra-quarter stock returns, there is a negative relationship between institutional
demand and future stock returns at the quarterly frequency. The reversal might also reflect the fact that
institutional investors tend to be momentum traders, and Jegadeesh and Titman (2001) and others find a
long-run price reversal for the momentum strategy. Moreover, DeBondt and Thaler (1985) and many
subsequent studies document that past winners underperform past losers over long holding horizons.
Below, we first show that the documented long-run price reversal associated with PC_NII is consistent
with informed trading and then investigate how it is related to other alternative explanations.
Several theoretical studies (e.g., Fishman and Hagerty (1992), van Bommel (2003), and
Brunnermeier (2005)) suggest that informed trading itself can lead to long-run price reversal.
Specifically, Brunnermeier (2005) emphasizes the long-term price reversal as a novel and important
prediction of his variant of Kyle’s (1985) rational expectations model of informed trading. In the model,
there are informed traders, liquidity traders, and a market maker. Informed traders are heterogeneous:
There are a short-run informed trader and many long-run informed traders. The short-run informed
trader receives a private informative albeit noisy signal regarding a firm’s forthcoming public news
announcement. Brunnermeier (2005) shows that it is optimal for the short-run informed trader to trade
aggressively on his signal so that other investors will on average overreact to the news announcement.
Anticipating the overreaction, the short-run informed trader can profit from his private information again
by partially unwinding his positions after the news announcement.
33
In Table 10, we investigate an important implication of Brunnermeier’s (2005) model that informed
institutional investors partially reverse their positions after the release of the news to the public.
Following Yan and Zhang (2009), we classify the short-term and long-term institutions as informed and
uninformed institutions, respectively. This categorization, which is consistent with our empirical finding
that only short-term PC_NII forecasts stock returns (see Table B5 in Appendix B), allows us to investigate
whether informed and uninformed institutions trade differently following the portfolio formation. In
Panel A, we report changes in short-term institutions’ holdings of stocks in the quintile portfolios formed
on PC_NII for the portfolio formation quarter (Q0) and the following 8 holding quarters (Q1 to Q8).
Quintile 5 has the largest PC_NII and quintile 1 has the smallest PC_NII at the portfolio formation period.
Consistent with the Brunnermeier’s (2005) prediction, short-term institutions initially increase their
holdings in stocks of quintile 5 and then reduce their holdings starting from Q4. Similarly, they reverse
the holdings in stocks of quintile 1 starting from Q2. Overall, the difference between changes in
holdings of quintiles 5 and 1 becomes significantly negative starting from Q4, coinciding with the price
reversal, as reported in Table 9. In contrast, Panel B shows that there is no reversal in holdings for
long-term institutional investors. Interestingly, starting from Q4, long-term institutions trade in the
exactly opposite directions of short-term institutions; that is, long-term institutions reduce their holdings
of stocks in quintile 1 and increase their holdings of stocks in quintile 5. This result is consistent with
the conjecture that long-term institutions are in general uninformed. Moreover, Panel C shows that there
is little evidence of reversal in the holding of all institutional investors, indicating that institutional
investors as a group appear to have no information disadvantage relative to individual investors. These
results confirm that informed institutions trade intensively with uninformed institutions and thus provide
further support for the argument that PC_NII is a better measure of informed trading than is ΔIO. Our
findings also cast doubts to the herding, overreaction, and price pressure hypotheses of the long-run price
reversal because none of these hypotheses can explain why short-term institutions and long-term
institutions trade in opposite directions after the portfolio formation.
As a robustness check, in Table B8 of Appendix B, we further investigate the long-run return reversal
34
using Fama-Macbeth regressions. Based on the portfolio results in Table 9, we use the return of the 4th
quarter, the buy-and-hold return of the 4th and 5th quarters, the buy-and-hold return of the 4th to 6th quarters,
or the buy-and-hold return of the 4th to 7th quarters after the portfolio formation quarter as the dependent
variable. We control for commonly used determinants of stock returns and report the results in Models 1,
2, 3 and 4 in Table B8. Consistent with the portfolio results, the coefficient of PC_NII is significantly
negative in all 4 models. Next, in Models 5, 6, 7 and 8, we further control for (1) the persistence herding
measure proposed by Dasgupta, Prat, and Verardo (2011), Persist, (2) the signed herding measure
proposed by Brown, Wei, and Wermers (2009), BWW_HM, and (3) the past three-year’s buy-and-hold
return advocated by Conrad and Kaul (1993), Past3yr_Return, which is an improved measure of DeBondt
and Thaler (1985). PC_NII is still negatively and significantly related to the 4th quarter return (Model 5)
and the buy-and-hold return over the 4th and 5th quarters (Model 6). However, it becomes insignificant
in Models 7 and 8 (where the dependent variable is, respectively, the buy-and-hold return of the 4th to 6th
quarters and the 4th to 7th quarters after the portfolio formation quarter). Consistent with the long-run
price reversals documented by Dasgupta, Prat, and Verardo (2011) and Brown, Wei, and Wermers (2009),
the coefficients of both Persist and BWW_HM are significantly negative in all 4 models; by contrast, the
coefficient of Past3yr_Return is negative but insignificant in these regressions.17
This finding that the predictive power of PC_NII for the long-run reversal gets weaker when we
control for herding measures is intriguing. A negative relationship between institutional herding and the
long-run price reversal can arise in several ways, including behavioral bias, reputational concerns or
career concerns, and informed trading (see, e.g., Dasgupta, Prat, and Verardo (2011) and Brown, Wei, and
Wermers (2009)). We have shown that PC_NII forecasts stock returns mainly because of its information
content about future earnings; therefore, the long-run reversal regression results in Table B8 of Appendix
B suggest that institutional herding partly reflects informed trading. Indeed, Brunnermeier (2005)
17 The long-run predictive power of PC_NII becomes weaker in models 5 to 8 of Table B8 mainly because of its correlation with herding measures by Dasgupta, Prat, and Verardo (2011) and Brown, Wei, and Wermers (2009). Specifically, if we control for only Past3yr_Return, the coefficient of PC_NII is always statistically significant at least at the 5% level in all regressions. For brevity, these results are omitted but are available on request.
35
emphasizes that this aggressive trading is a novel form of stock price manipulation by an informed trader
because it worsens other market participants’ ability to conduct technical analysis. Similarly, Dasgupta,
Prat, and Verardo (2011) emphasize that it is possible that the patterns of return predictability arise
because institutions trade against insiders with superior knowledge of future cash flows. On the other
hand, because herding measures remain significant predictors of long-run stock returns when we control
for PC_NII, informed trading is unlikely to be the only explanation for institutional herding. For
example, Brown, Wei, and Wermers (2009) find that career concerns are an important driver of
institutional herding. It is worth reiterating that PC_NII is a proxy for informed trading and differs from
herding measures in a fundamental way: Dasgupta, Prat, and Verardo’s (2011) herding measure correlates
negatively with one-quarter-ahead stock returns, while the correlation is weak for Brown, Wei, and
Wermers’ (2009) herding measure (untabulated). Moreover, while the price pressure can lead to
short-run returns and long-run reversals, it does not explain the predictive power of PC_NII for stock
returns for at least 2 reasons. First, PC_NII forecasts short-run stock returns mainly because of its
information content about future earnings. Second, the long-run reversal coincides with a reversal in
short-term (informed) institutions’ holdings while we do not observe a similar reversal for long-run
(uninformed) institutions.
Lastly, in Brunnermeier’s (2005) model, the market is semi-strong-form efficient but not strong-form
efficient. That is, while public information does not forecast stock returns, private information does.
Note that, even if an institution’s orders are observable by market makers, its information-driven entry or
exit is unobservable when it submits the orders. Moreover, institutions do not need to disclose their
positions to the SEC until 45 days from the end of a calendar quarter (Rule 13f-1 under the Securities
Exchange Act of 1934). Therefore, our main finding that PC_NII forecasts short-term stock returns
because of its information content of future earnings is consistent with the implication of the
Brunnermeier’s (2005) model. In a similar vein, Brunnermeier’s (2005) model sheds light on why ΔIO
has weaker predictive power for stock returns than does PC_NII because (1) the former is potentially
observable by market makers and (2) institutions have heterogeneous information (i.e., only some, but not
36
all, institutional traders have information about short-run earnings).
8. Conclusion
We propose a novel measure of institutional informed trading, PC_NII, which differs from commonly
used institutional trading measures in two important ways. First, it allows for heterogeneous
information across institutions. An institution’s entry or exit is likely to be triggered by its substantive
private information, and only a small fraction of institutions have access to such information. Because
of intensive trades between informed and uninformed institutions, PC_NII is a better measure of informed
institutions’ private information than is ∆IO. Second, PC_NII is also a better measure of institutional
informed trading than is ∆NII or ∆BREADTH because it explicitly takes into account the fact that the
information content of entries and exits varies across stocks with different number of institutional
investors. Specifically, trading costs and information costs deter liquidity-based entries and exits, and
small stocks with few institutional investors tend to have higher trading costs and information costs than
do big stocks with many institutional investors. Therefore, an institution’s entry or exit is more
informative (i.e., is more likely to be triggered by private information than by liquidity reasons) for stocks
with few institutional investors than for stocks with many institutional investors. As conjectured, we
find that PC_NII has strong predictive power for the cross-section of stock returns because of its
information about future earnings; moreover, it subsumes the information content of standard institutional
trading and herding measures about future stock returns. PC_NII is also easy to construct and does not
require any potentially ad hoc partition of institutions. Therefore, PC_NII is potentially an attractive
alternative measure of institutional informed trading that can be easily adopted in empirical studies.
Existent literature has intensively investigated the impact of Regulation FD on financial markets from
various perspectives. Few studies, however, have examined directly whether the SEC has achieved its
primary goal of reducing informed trading mainly because the existent evidence of informed trading is
rather elusive. In this paper, we argue that PC_NII is a good measure of institutions’ private information
and show that it indeed has a strong positive correlation with future earnings surprises and stock returns.
37
PC_NII thus allows us to provide a formal albeit preliminary evaluation of the effectiveness of Regulation
FD on informed trading. We find that Regulation FD has exerted some noticeable effects but the success
is rather limited. Specifically, it has significantly reduced informed trading for small stocks; however,
its effects on big stocks are rather elusive. To the extent that Regulation FD has a rather limited success
in reducing informed trading, the SEC might need to develop alternative regulatory policies.
Last but not the least, we document a significant long-term price reversal for the trading strategy
formed on PC_NII. This finding casts doubts on the conventional wisdom that informed trading always
facilitates incorporating available information into stock prices. It is, however, consistent with the
prediction of Brunnermeier’s (2005) rational expectations model of informed trading, which cautions
about potential adverse effects of informed trading on market efficiency. Specifically, Brunnermeier
(2005) shows that the informed investor strategically leads uninformed investors to overreact to a
forthcoming news announcement and then partially reverses his/her holding positions after the
announcement. Interestingly, consistent with this model prediction, we find that the long-term price
reversal for the trading strategy formed on PC_NII coincides with the reversal in short-term (informed)
institutions’ holding positions. Moreover, by contrast, the long-term (uninformed) institutions trade in
the exactly opposite directions of short-term (informed) institutions. Our results suggest that
Brunnermeier’s (2005) theoretical concern is potentially important economically and warrants further
investigations.
38
REFERENCES
Abarbanell, J., and B. Bushee, 1997, Fundamental analysis, future earnings, and stock prices, Journal of
Accounting Research, 35, 1-24.
Abarbanell, J., and B. Bushee, 1998, Abnormal stock returns to a fundamental analysis strategy,
Accounting Review, 73, 19-45.
Admati, A., and P. Peiderer, 1988, A theory of intraday patterns: Volume and price variability, Review of
Financial Studies, 1, 3-40.
Ahmed, A., and R. Schneible, 2007, The impact of regulation fair disclosure on investors’ prior information
quality: Evidence from an analysis of changes in trading volume and stock price reactions to earnings
announcements, Journal of Corporate Finance, 13, 282–299.
Ali, A., C. Durtschi, B. Lev, and M. Trombley, 2004, Changes in institutional ownership and subsequent
earnings announcement abnormal returns, Journal of Accounting, Auditing, and Finance, 3, 221–248.
Amihud, 2002, Illiquidity and stock returns, Journal of Financial Markets, 5, 31–56.
Amihud, Y., and H. Mendelson, 1986, Asset pricing and the bid-ask spread, Journal of Financial
Economics, 17, 223–249.
Anand, A., P. Irvine, A. Puckett, and K. Venkataraman, 2011, Market crashes and institutional trading,
Unpublished Working Paper, University of Georgia.
Ang, A., R. Hodrick, Y. Xing, and X. Zhang, 2006, The cross-section of volatility and expected returns,
Journal of Finance, 51, 259–299.
Ang, A., R. Hodrick, Y. Xing, and X. Zhang, 2009, High idiosyncratic volatility and low returns:
International and further us evidence, Journal of Financial Economics, 91, 1–23.
Aslan, H., D. Easley, S. Hvidkjaer, and M. O'Hara, 2008, Firm characteristics and informed trading:
Implications for asset pricing, Unpublished Working Paper, Cornell University.
Asquith, P., P. Pathak, J. Ritter, 2005, Short Interest, institutional ownership, and stock returns, Journal of
Financial Economics, 78, 243–276.
Baker, M., L. Litov, J. Wachter, and J. Wurgler, 2010, Can mutual fund managers pick stocks? Evidence
39
from their trades prior to earnings announcements, Journal of Financial and Quantitative Analysis, 45,
1111–1131.
Ball, R., and P. Brown, 1968, An empirical evaluation of accounting income numbers, Journal of
Accounting Research, 6, 159–178.
Ball, R., and L. Shivakumar, 2008, How much new information is there in earnings? Journal of Accounting
Research, 46, 975-1016.
Barber, B., T. Odean, N. Zhu, 2009, Do retail trades move markets? Review of Financial Studies, 2009, 22,
151–186.
Barras, L., O. Scaillet and R. Wermers, 2010, False discoveries in mutual fund performance: Measuring
luck in estimated alphas, Journal of Finance, 65, 179–216.
Beaver, W., M. McNichols, and R. Price, 2007, Delisting returns and their effect on accounting-based
market anomalies, Journal of Accounting and Economics, 43, 341–368.
Ben-David, I., F. Franzoni, and R. Moussawi, 2011, Hedge Fund Stock Trading in the Financial Crisis of
2007-2009, Unpublished Working Paper, Ohio State University.
Bennett, J., L. Starks, and R. Sias, 2003, Greener pastures and the impact of dynamic institutional
preferences, Review of Financial Studies, 16, 1203–1239.
Bernard, V., and J. Thomas, 1989, Post-earnings-announcement drift: Delayed price response or risk
premium? Journal of Accounting Research, 27, 1–36.
Beyer, A., D. Cohen, T. Lys, and B. Walther, 2010, The Financial Reporting Environment: Review of the
Recent Literature, Journal of Accounting and Economics, 50, 296–343.
Brown, S., S. Hillegeist, and K. Lo, 2004, Conference calls and information asymmetry, Journal of
Accounting and Economics, 37, 343-366.
Brown, N., K. Wei, and R. Wermers, 2009, Analyst recommendations, mutual fund herding, and
overreaction in stock prices, Unpublished Working paper, University of Southern California.
Brunnermeier, M., 2005, Information leakage and market efficiency, Review of Financial Studies, 2005, 18,
417–457.
40
Bushee, B., and T. Goodman, 2007, Which institutional investors trade on private information about
earnings and returns? Journal of Accounting Research, 45, 289–321.
Cai, F., and L. Zheng, 2004, Institutional trading and stock returns, Finance Research Letters, 1, 178–189.
Campbell, J., T. Ramadorai, and A. Schwartz, 2009, Caught on tape: Institutional trading, stock returns, and
earnings announcements, Journal of Financial Economics, 92, 66–91.
Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance, 52, 57–82.
Chen, J., H. Hong, and J. Stein, 2002, Breadth of ownership and stock returns, Journal of Financial
Economics, 66, 171–205.
Chen, H., N. Jegadeesh, and R. Wermers, 2000, An examination of the stockholdings and trades of fund
managers, Journal of Financial and Quantitative Analysis 35, 343–368.
Choi, N., and R. Sias, 2009, Institutional industry herding, Journal of Financial Economics, 94, 469–491.
Cohen, R., P. Gompers, and T. Vuolteenaho, 2002, Who underreacts to cash-flow news? Evidence from
trading between individuals and institutions, Journal of Financial Economics, 66, 409–462.
Conrad, J., and G. Kaul, 1993, Long-term market overreaction or biases in computed returns? Journal of
Finance, 48, 39-63.
Daniel, K., M. Grinblatt, S., Titman, and R. Wermers, 1997, Measuring mutual fund performance with
characteristic-based benchmarks, Journal of Finance, 52, 1035-1058.
Dasgupta, A., A. Prat, and M. Verardo, 2011, Institutional trade persistence and long-term equity returns,
Journal of Finance, 635–653.
D’Avolio, G., 2002, The market for borrowing stock, Journal of Financial Economics, 66, 271–306.
De Bondt, W., and R. Thaler, 1985, Does the stock market overreact? Journal of Finance, 40, 793-805.
Diamond, D., and R. Verrecchia, 1991, Disclosure, liquidity, and the cost of capital, Journal of Finance, 46,
1325-1359.
Easley, D., S. Hvidkjaer, and M. O'Hara, 2002, Is Information risk a determinant of asset returns? Journal of
Finance, 57, 2185–2221.
Eleswarapu, V., R. Thompson, and K. Venkataraman, 2004, The impact of regulation fair disclosure,
41
Journal of Financial and Quantitative Analysis, 39, 209–225.
Fama, E., and K. French, 1992, The cross-section of expected stock returns, Journal of Finance, 47, 427–
466.
Fama, E., and K. French, 1996, Multifactor explanations of asset pricing anomalies, Journal of Finance, 51,
55–84.
Fama, E., and J. MacBeth, 1973, Risk, return and equilibrium: Empirical tests, Journal of Political
Economy, 81, 607–636.
Fishman, M., and K. Hagerty, 1992, Insider trading and the efficiency of stock prices, RAND Journal of
Economics, 23 106-122.
Gomes, A., G. Gorton, and L. Madureira, 2007, SEC regulation fair disclosure, information, and the cost of
capital, Journal of Corporate Finance, 13, 300–334.
Gompers, P., and A. Metrick, 2001, Institutional investors and equity prices, Quarterly Journal of
Economics, 116, 229–259.
Griffin, J., J. Harris, T. Shu, and S. Topaloglu, 2011, Who drove and burst the tech bubble? Journal of
Finance, 66, 1251–1290.
Harrison, M., and D. Kreps, 1978, Speculative Investor Behavior in a Stock Market with Heterogeneous
Expectations, Quarterly Journal of Economics, 92, 323–336.
He, Z., I. Khang, and A. Krishnamurthy, 2010, Balance Sheet Adjustments in the 2008 Crisis, Unpublished
Working Paper, Northwestern University.
Hermalin, B., and M. Weisbach, 2011, Information disclosure and corporate governance, Journal of
Finance, forthcoming.
Jegadeesh, N., and S. Titman, 1993, Returns to buying winners and selling losers: implications for stock
market efficiency, Journal of Finance, 48, 65–91.
Jegadeesh, N., and S. Titman, 2001, Profitability of momentum strategies: An evaluation of alternative
explanations, Journal of Finance, 56, 699–720.
Jia, Y., and M. Liu, 2011, Independent institutional investors and equity returns, Unpublished Working
42
Paper, University of Kentucky.
Jorion, P., Liu, Z., and Shi, C., 2005, Informational effects of Regulation FD: Evidence from rating
agencies, Journal of Financial Economics, 76, 309–330.
Kaniel, R., S. Liu, G. Saar, and S. Titman, 2011, Individual investor trading and return patterns around
earnings announcements, Journal of Finance, forthcoming.
Ke, B., and S. Ramalingegowda, 2005, Do institutional investors exploit the post-earnings announcement
drift? Journal Of Accounting and Economics, 39, 25–53.
Kosowski, R., A. Timmermann, R. Wermers, and H. White, 2006, Can mutual fund “stars” really pick
stocks? new evidence from a bootstrap analysis, Journal of Finance, 61, 2551–2595.
Kyle, A., 1985, Continuous auctions and insider trading, Econometrica, 53, 1315–1336.
Lai, S., L. Ng, and B. Zhang, 2009, Informed trading around the world, Unpublished Working Paper,
Singapore Management University.
Lakonishok, J., A. Shleifer, and R. Vishny, 1992, The impact of institutional trading on stock prices, Journal
of Financial Economics, 32, 23–43.
Lehavy, R., and R. Sloan, 2008. Investor recognition and stock returns. Review of Accounting Studies 13,
327–361.
Lev, B., and S. Thiagarajan, 1993, Fundamental information analysis, Journal of Accounting Research, 31,
190-215.
Merton, R., 1987, A simple model of capital market equilibrium with incomplete information. Journal of
Finance 42, 483–510.
Newey, W., and K. West, 1987, A simple, positive, semi-definite heteroscedasticity and autocorrelation
consistent covariance matrix, Econometrica, 55, 703-708.
Miller, E., 1977, Risk, uncertainty, and divergence of opinion, the Journal of Finance, 32, 1151–1168.
Nagel, S., 2005, Short Sales, Institutional Investors, and the Cross-Section of Stock Returns, Journal of
Financial Economics, 78, 277–309.
Nofsinger, J., and R. Sias, 1999, Herding and feedback trading by institutional and individual investors,
43
Journal of Finance, 54, 2263–2295.
Ou, J., and S. Penman, 1989, Financial statement analysis and the prediction of stock returns, Journal of
Accounting and Economics, 11, 295-330.
Puckett, A., and X. Yan, 2011, The interim trading skills of institutional investors, Journal of Finance, 66,
601–633.
Reca, B., R. Sias, and H. Turtle, 2011, Anatomy of Institutional Positions, Unpublished Working Paper,
University of Arizona.
Scheinkman, J., and W. Xiong, 2003, Overconfidence and speculative bubbles, Journal of Political
Economy, 111, 1183–1219.
Sias, R., 2004, Institutional herding, Review of Financial Studies, 17, 165–206.
Sias, R., 2007, Reconcilable differences: Momentum trading by institutions, Financial Review, 42, 1–22.
Sias, R., L. Starks, and S. Titman, 2006, Changes in institutional ownership and stock returns: Assessment
and methodology, Journal of Business, 79, 2869–2910.
van Bommel, J., 2003, Rumors, Journal of Finance, 58, l499–l520.
Vega, C., 2005, Stock price reaction to public and private Information, Journal of Financial Economics, 82,
103–133.
Vuolteenaho, T., 2002, What drives firm-level stock returns? Journal of Finance, 57, 233–264.
Wermers, R., 1999, Mutual fund herding and the impact on stock prices, Journal of Finance, 54, 581–622.
Wermers, R., 2000, Mutual fund performance: An empirical decomposition into stock-picking talent, style,
transaction costs, and expenses, Journal of Finance, 55, 581–622.
Wermers, R., 2004, Is money really 'smart'? New evidence on the relation between mutual fund flows,
manager behavior, and performance persistence, Unpublished Working Paper, University of Maryland.
Wooldridge, J., 2006, Introductory econometrics: A modern approach, 3rd Edition, South-Western,
Cincinnati, OH.
Yan, X., and Z. Zhang, 2009, Institutional investors and equity returns: Are short-term institutions better
informed? Review of Financial Studies, 22, 893–924.
44
Figure 1 Mean (Solid Line) and Median (Dashed Line) Number of Institutional Investors
Figure 2 25 (Solid Line) and 75 (Dashed Line) Percentiles of Changes in Institutional Investors
Figure 3 Annual Equal-Weighted (Solid Line) and Value-Weighted (Dashed Line) Returns on the Long-Short
Portfolio Formed on PC_NII
0
30
60
90
120
150
180
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
‐10
‐5
0
5
10
‐0.4
‐0.2
0
0.2
0.4
0.6
0.8
45
Table 1 Summary Statistics of PC_NII
Notes: The table reports the summary statistics of PC_NII, which is the percentage change in a stock’s number of institutional investors. For each stock-quarter observation included in our sample, we require (1) common stocks (CRSP codes 10 and 11), (2) traded on NYSE, AMEX or NASDAQ, (3) with end-of-quarter close price no less than $5 per share, and (4) non-missing PC_NII value. Our data spans the 1982Q2 to 2010Q4 period. NOB is the number of observation. Mean is the cross-sectional mean. Median is the cross-sectional median. Zero is the percentage of stocks that have no change in the number of institutional investors. 25 Percentile is the cross-sectional 25 percentile. 75 percentile is the cross-sectional 75 percentile. Except for NOB, all variables are reported in percentage points.
PC_NII
Year NOB Mean Median Zero 25
Percentile75
Percentile
1982 6698 13.94 0 26.19 0 20.00
1983 10391 16.36 0 24.54 -2.49 20.00
1984 11001 22.03 0 27.48 -5.00 11.11
1985 10948 21.95 3.63 22.79 0 20.00
1986 11427 14.88 0 20.57 -6.29 14.29
1987 11874 8.60 0 20.83 -4.76 14.29
1988 11313 11.46 0 20.03 -3.85 12.50
1989 11084 7.15 0 19.21 -6.25 13.04
1990 9879 4.43 0 21.29 -5.88 7.14
1991 10304 6.32 0 19.83 -4.13 9.80
1992 11633 11.45 1.25 16.99 -3.88 11.11
1993 13255 6.78 0 15.61 -8.33 9.43
1994 14663 15.16 0 16.72 -5.45 12.50
1995 15566 10.01 2.41 18.18 -2.22 12.12
1996 16948 12.06 0 15.34 -9.47 9.09
1997 17634 24.01 3.24 14.12 -3.13 14.29
1998 17158 28.73 1.97 11.08 -6.67 14.46
1999 16084 27.12 0.86 12.81 -6.25 12.50
2000 15301 17.94 1.06 12.37 -5.86 11.46
2001 13455 27.45 0 12.09 -4.59 10.00
2002 12555 18.72 1.19 9.90 -4.55 10.00
2003 12987 14.35 3.24 10.42 -1.33 11.43
2004 14073 18.83 3.33 8.48 -2.41 11.67
2005 14080 17.03 0 9.01 -5.71 6.33
2006 14435 23.36 2.78 8.70 -2.17 10.75
2007 14252 20.97 1.28 7.86 -4.76 9.09
2008 12165 8.46 -0.45 7.06 -6.90 6.12
2009 10757 15.74 1.22 6.85 -3.70 7.37
2010 11578 12.37 0.66 7.29 -4.14 6.55
1982-2010 373498 16.37 0.32 14.71 -4.76 11.11
46
Table 2 Decile Portfolios Sorted on PC_NII
Notes: The table reports the averages of selected characteristics (Panel A) and equal-weighted stock returns (Panel B) of portfolios formed on PC_NII across the 1982 to 2010 period. PC_NII is the percentage change in a stock’s number of institutional investors in the portfolio formation quarter, which is winsorized at 1% and 99% each quarter. SIZE is market capitalization in million dollars at the end of the portfolio formation quarter. BM is the book-to-market equity ratio, for which the book value of equity is from the most recently reported fiscal quarter with a four-month reporting lag and the market value of equity is the market capitalization at the end of the portfolio formation quarter. MOM is the stock return in the previous 6 months up to the end of the portfolio formation quarter. IV is the idiosyncratic volatility of the portfolio formation quarter constructed as in Ang, Hodrick, Xing, and Zhang (2006) with at least 44 daily return observations. AMIHUD is the Amihud (2002) illiquidity measure constructed using the daily return data of the portfolio formation quarter. E_CHANGE is the seasonal change in earnings before extraordinary items from the one-year-ago quarterly value scaled by average total assets for the fiscal quarter ended in the quarter before the portfolio formation quarter. E_SURPRISE is the most recently disclosed earnings surprise up to the end of the portfolio formation quarter, which is the difference between actual earnings per share and median analyst forecast scaled by the end-of-last quarter stock price. We use the portfolio median for E_CHANGE and E_SURPRISE and the portfolio mean for the other variables. Raw Return is the stock return in the quarter following the portfolio formation quarter. DGTW-adj. Return is the Daniel, Grinblatt, Titman, and Wermers (DGTW, 1997) characteristic-adjusted stock return in the quarter following the portfolio formation quarter. RAW (-1,1) EAR and DGTW-adj. (-1,1) EAR are, respectively, the stock return and DGTW-adjusted return in the (-1,1) earnings announcement window in the quarter following the portfolio formation quarter. We calculate the 3-factor alpha using the Fama and French (1993) 3-factor model, the 4-factor alpha using the Carhart (1997) 4-factor model, and the five-factor alpha using the Pastor and Stambaugh (2003) 5-factor model. Returns and alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Stock Characteristics
Deciles PC_NII ∆BREADTH ∆IO SIZE BM MOM IV AMIHUD E_CHANGE E_SURPRISE
D1 -22.15 -0.52 -2.72 362.66 0.76 -1.98 3.12 2.12 -0.01 -0.27
D2 -9.25 -0.38 -0.94 978.58 0.72 -1.25 2.59 0.86 0.01 -0.18
D3 -4.35 -0.25 -0.28 2078.55 0.68 1.51 2.33 0.87 0.06 -0.16
D4 -1.52 -0.10 0.02 3103.15 0.67 5.47 2.31 1.35 0.07 -0.08
D5 0.54 0.03 0.28 3464.72 0.66 7.78 2.27 1.47 0.10 -0.06
D6 3.06 0.19 0.67 3926.02 0.61 9.65 2.19 1.02 0.13 -0.04
D7 6.39 0.34 0.96 2890.56 0.59 12.96 2.26 0.77 0.16 -0.01
D8 11.28 0.44 1.44 1788.34 0.59 17.69 2.38 0.58 0.21 -0.01
D9 20.06 0.50 2.31 876.33 0.57 24.94 2.66 1.00 0.27 0.01
D10 58.88 0.71 5.33 414.07 0.50 43.75 3.20 1.61 0.42 0.05
D10-D1 81.03 1.23 8.05 51.41 -0.26 45.72 0.08 -0.51 0.43 0.33
t-Statistic (25.90) (54.15) (33.20) (1.41) (-14.12) (12.84) (1.43) (-2.13) (15.94) (17.46)
47
Panel B: One-Quarter-ahead Stock Returns
Deciles Raw
Return DGTW Return
3-Factor Alpha
4-Factor Alpha
5-Factor Alpha
Raw (-1,1) EAR
DGTW (-1,1) EAR
D1 2.25 -1.25 -1.50 (-3.81) -1.21 (-3.60) -1.20 (-3.40) 0.20 0.01
D2 2.91 -0.79 -0.89 (-3.01) -0.41 (-2.01) -0.38 (-1.89) 0.32 0.13
D3 3.39 -0.22 -0.28 (-0.98) -0.02 (-0.09) -0.02 (-0.07) 0.42 0.23
D4 3.81 -0.02 0.07 (0.25) 0.25 (1.18) 0.32 (1.43) 0.50 0.26
D5 3.64 -0.26 -0.02 (-0.06) 0.03 (0.16) 0.06 (0.31) 0.38 0.21
D6 3.22 -0.19 -0.28 (-0.98) -0.27 (-1.01) -0.22 (-0.83) 0.40 0.23
D7 3.52 -0.00 -0.03 (-0.13) -0.10 (-0.47) -0.04 (-0.20) 0.30 0.14
D8 4.15 0.31 0.67 (3.42) 0.42 (2.34) 0.45 (2.51) 0.38 0.20
D9 4.37 0.62 0.85 (3.78) 0.37 (1.89) 0.38 (1.99) 0.46 0.29
D10 4.67 1.12 1.39 (3.66) 0.57 (1.97) 0.65 (2.06) 0.51 0.33
D10-D1 2.42 2.37 2.89 1.78 1.84 0.31 0.32
t-Statistic (4.47) (5.27) (5.39) (4.83) (5.28) (3.64) (3.96)
48
Table 3 Forecasting One-Quarter-Ahead Returns Using Interaction Terms
Note: The table reports the results of Fama-MacBeth regressions of forecasting one-quarter-ahead stock returns. The dependent variable is one-quarter-ahead stock return. Each row in Panel A represents a F-M regression model using ∆BREADTH, a dummy variable, and the interaction term between ∆BREADTH and the dummy variable as the independent variables. Similarly, each row in Panel B represents a F-M regression model using PC_NII, a control variable, and the interaction term between PC_NII and the control variable. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market, as adopted in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). All of the control variables are dummy variables and their constructions are straightforwardly shown by their names. For example, Large Size is equal to 1 if the market capitalization of the stock is larger than the sample median in a quarter and equal to 0 otherwise; Large N_INST is equal to 1 if the number of institutional shareholders of the stock at the beginning of a quarter is larger than the sample median in that quarter and equal to 0 otherwise; etc.. Dividend Payer is equal to 0 if the stock has never paid a dividend up to the end of the ∆BREADTH/PC_NII measurement quarter and equal to 1 otherwise. Beta, Size, BM, Mom, Amihud, IV and E_CHANGE are added as control variables in each regression model of the table, but their coefficients are omitted from the table for brevity. Intercept is also included in each regression model but is omitted from the table for brevity. The coefficients of PC_NII and its interaction terms with control variables are multiplied by 100 for the ease of presentation. The adjusted R2 is the time-series mean of adjusted R-squares obtained from cross-sectional regressions. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: ∆BREADTH Panel B: PC_NII
∆BREADTH Dummy ∆BREADTH*
Dummy Adj. R2 PC_NII Dummy PC_NII* Dummy
Adjusted R2
Large Size 1.43 -0.02 -1.29 6.39% 1.97 -0.09 -0.40 6.36%
(3.48) (-0.10) (-3.05) (3.96) (-0.42) (-0.46)
Large N_INST 1.26 0.43 -1.19 6.41% 2.00 0.35 -0.10 6.40%
(3.32) (1.78) (-3.13) (4.15) (1.48) (-0.14)
Low Amihud 2.15 -0.05 -2.02 6.44% 2.12 -0.16 -1.20 6.44%
(4.62) (-0.18) (-4.27) (4.11) (-0.60) (-1.29)
Low Dispersion 0.39 0.36 -0.28 7.48% 3.86 0.40 -1.70 7.57%
(2.53) (1.61) (-1.82) (4.31) (1.74) (-1.48)
High Following 0.90 0.44 -0.75 6.97% 2.77 0.31 -0.20 7.04%
(3.72) (2.09) (-3.06) (4.50) (1.44) (-0.20)
High Turnover 0.72 0.15 -0.56 6.47% 1.97 0.13 -0.60 6.49%
(3.72) (0.62) (-2.44) (4.35) (0.56) (-0.74)
Low PIN 0.68 0.08 -0.53 6.02% 2.47 0.05 0.05 6.09%
(3.07) (0.32) (-2.62) (3.02) (0.22) (0.03)
49
Table 4 Double Sort by ΔBREADTH and PC_NII
Notes: In Panel A, we sort stocks each quarter equally into 5 portfolios by ∆BREADTH and then, within each ∆BREADTH quintile, we sort stocks equally into 5 portfolios by PC_NII. We calculate the equal-weighted return over the next three months for each of the 25 portfolios. The portfolios are rebalanced every quarter. We report the Carhart (1997) 4-factor alphas for the 25 portfolios. For each PC_NII quintile, we also average returns across the ∆BREADTH quintiles and report the Carhart alphas for each aggregate PC_NII quintile. The analysis in Panel B is similar to that in Panel A except that we sort stocks each quarter first by PC_NII then by ∆BREADTH into 25 portfolios. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market, as adopted in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). Alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: First by ΔBREADTH then By PC_NII
Control Variable PC_NII
1(L) 2 3 4 5(H) 5-1 t-Statistic
1(L) -1.74 -0.91 -0.12 -0.12 0.24 1.98 3.73
2 -0.50 -0.08 0.29 0.33 0.42 0.92 2.00
ΔBREADTH 3 -0.40 0.02 -0.60 -0.16 0.17 0.57 1.39
4 -0.16 -0.24 0.15 0.06 0.20 0.36 0.77
5(H) 0.05 0.17 0.58 0.53 1.00 0.95 2.31
Avg. -0.55 -0.21 0.06 0.13 0.41 0.96 (5.52)
Panel B: First by PC_NII then by ΔBREADTH
ΔBREADTH
1(L) 2 3 4 5(H) 5-1 t-Statistic
1(L) -1.19 -0.83 -0.86 -0.42 -0.81 0.38 0.63
2 -0.11 -0.05 0.26 0.34 0.09 0.20 0.50
PC_NII 3 0.32 0.19 -0.27 -0.34 -0.31 -0.63 -1.90
4 0.39 0.02 -0.01 0.44 0.04 -0.35 -0.87
5(H) 0.41 0.02 0.41 0.47 0.98 0.57 0.81
Avg. -0.04 -0.13 -0.09 0.10 -0.00 0.03 (0.15)
50
Table 5 Portfolios Sorted on PC_NII with Control for Other Characteristics
Notes: We sort stocks each quarter equally into 5 portfolios by a control variable and then within each quintile we sort stocks equally into 5 portfolios by PC_NII. We calculate the equal-weighted return over the next three months for each of the 25 portfolios. The portfolios are rebalanced every quarter. For the control variable SIZE, we report the Carhart (1997) 4-factor alphas for the 25 portfolios. For each PC_NII quintile, we also average returns across the SIZE quintiles and report the Carhart alphas for each aggregate PC_NII quintile. For the other control variable, we only report the Carhart alphas for aggregate PC_NII quintiles for brevity. Quintile 1 is the stocks with the lowest PC_NII and Quintile 5 the stocks with the highest PC_NII. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. SIZE is the market capitulation. BM is the book-to-market equity ratio. MOM is the stock return in the previous 6 month. E_CHANGE is the (most recently disclosed) seasonal earnings growth. E_SURPRISE is the (most recently disclosed) earnings surprise. AMIHUD is the Amihud (2002) illiquidity measure. IV is the idiosyncratic volatility. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market, as adopted in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). IO is the fraction of shares owned by institutions. ∆IO is the change in IO. Short-term ∆IO is the change in short-term institutions’ IO proposed by Yan and Zhang (2009). LSV_HM is Lakonishok, Shleifer and Vishny’s (1992) herding measure. S_HM is Sias’ (2004) standardized herding measure. BWW_HM is Brown, Wei and Wermers’ (2009) signed herding measure. Future E_CHANGE is the one-quarter lead of E_CHANGE, and Future E_SURPRISE is the one-quarter lead of E_SURPRISE. Alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Controlling for Determinants of Cross-Sectional Stock Returns
Control Variable PC_NII 1(L) 2 3 4 5(H) 5-1 t-Statistic
1(S) -0.95 -0.30 0.04 0.29 0.45 1.40 (4.50)
2 -0.99 -0.30 -0.31 0.13 0.72 1.71 (3.79)
Size Quintile 3 -0.93 0.15 0.14 0.58 0.60 1.53 (3.35)
4 -0.99 0.11 0.01 0.19 0.44 1.43 (3.59)
5(B) -0.22 0.27 0.04 -0.07 0.05 0.27 (0.57)
SIZE Avg. -0.81 -0.01 -0.01 0.22 0.45 1.27 (4.49)
BM Avg. -0.77 0.16 0.02 0.25 0.71 1.48 (5.66)
MOM Avg. -0.66 -0.06 0.10 0.15 0.48 1.13 (4.88)
E_CHANGE Avg. -0.45 0.38 0.03 0.16 0.62 1.07 (4.34)
E_SURPRISE Avg. -0.76 -0.26 -0.58 -0.48 0.30 1.06 (2.67)
AMIHUD Avg. -0.84 -0.10 0.07 0.30 0.43 1.27 (4.63)
IV Avg. -0.53 -0.05 -0.20 0.01 0.59 1.12 (4.71)
Panel B: Controlling for Standard Measures of Institutional Trading and Herding
∆BREADTH Avg. -0.55 -0.21 0.06 0.13 0.41 0.96 (5.52)
IO Avg. -0.77 -0.15 0.09 0.17 0.50 1.27 (4.89)
∆IO Avg. -0.77 -0.01 -0.07 0.18 0.51 1.29 (4.78)
Short-Term ∆IO Avg. -0.78 -0.09 0.03 0.19 0.48 1.26 (4.69)
LSV_HM Avg. -0.79 -0.17 0.12 0.21 0.58 1.37 (4.94)
S_HM Avg. -0.86 -0.05 0.07 0.17 0.63 1.49 (5.09)
BWW_HM Avg. -0.83 -0.15 0.11 0.27 0.55 1.37 (5.04)
Panel C: Controlling for Future Earnings Surprises
Future E_CHANGE Avg. 0.04 0.54 0.02 -0.06 0.09 0.04 (0.19)
Future E_SURPRISE Avg. 0.26 0.36 0.05 -0.19 0.14 -0.13 (-0.30)
51
Table 6 Fama and MacBeth Regressions
Notes: The table reports the results of Fama-MacBeth regressions of forecasting one-quarter-ahead stock returns. PC_NII is the percentage change in a stock’s number of institutional investors. SIZE is the log market capitalization. BM is the log book-to-market equity ratio. BETA is market beta. MOM is the stock return in the previous 6 month. E_CHANGE is the seasonal earnings growth. E_SURPRISE is the earnings surprise. PC_NII, E_CHANGE and E_SUPRISE are winsorized at 1% and 99% each quarter. AMIHUD is the illiquidity measure. IV is the idiosyncratic volatility. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market. IO is the fraction of shares owned by institutions. ∆IO is the change in IO. Short-term ∆IO is the change in short-term institutions’ IO. BWW_HM is BWW’s (2009) signed herding measure. One-quarter-ahead stock return (the dependent variable), PC_NII, ∆BREADTH, ∆IO, IO, MOM, IV, E_CHANGE and E_SURPRISE are all in percentage. The coefficient of PC_NII is multiplied by 100. The adjusted R2 is the time-series mean of cross-sectional regression adjusted R2. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9
Intercept 3.46 5.33 5.37 5.36 3.57 3.13 5.67 5.75 4.81
(3.67) (4.45) (4.48) (3.85) (3.74) (3.46) (4.27) (4.33) (3.32)
PC_NII (%) 2.12 1.93 1.87 3.04 3.25 3.00 2.95 2.83
(3.10) (4.46) (4.18) (4.00) (5.16) (4.91) (4.69) (1.67)
∆BREADTH (%) 0.22 0.01 -0.01 -0.08
(1.49) (0.05) (-0.08) (-0.30)
∆IO (%) -0.02 -0.05 -0.05 -0.05
(-1.18) (-4.42) (-4.33) (-2.94)
Short-Term ∆IO (%) 0.03 0.01 0.01 0.00 0.03
(2.14) (0.84) (0.48) (0.41) (1.55)
IO (%) 0.01 0.00 0.00 -0.01
(1.19) (0.58) (0.41) (-1.32)
BWW_HM -1.00 0.25 0.18 0.43
(-1.01) (0.64) (0.46) (0.81)
Beta 0.68 0.66 0.44 0.75 0.74 0.60
(1.07) (1.05) (0.65) (1.25) (1.26) (0.91)
Size -0.16 -0.16 -0.15 -0.24 -0.24 -0.02
(-1.57) (-1.59) (-1.41) (-2.19) (-2.21) (-0.17)
BM 0.51 0.57 0.36 0.44 0.48 0.37
(1.89) (2.13) (1.15) (1.63) (1.83) (1.24)
MOM 0.02 0.02 0.00 0.01 0.01 0.00
(2.65) (2.27) (0.60) (2.03) (1.80) (0.56)
Amihud -0.01 -0.01 -0.01 0.07 0.06 0.08
(-0.30) (-0.24) (-0.04) (0.56) (0.47) (0.34)
IV -0.70 -0.69 -0.83 -0.77 -0.77 -0.85
(-3.88) (-3.84) (-3.69) (-3.68) (-3.66) (-3.41)
E_CHANGE 0.22 0.15
(4.52) (2.88)
E_SURPRISE 0.47 0.42
(4.61) (4.14)
Adjusted R2 0.29% 6.03% 6.23% 7.32% 0.10% 1.60% 6.86% 7.05% 8.07%
52
Table 7 PC_NII and Future Earnings Changes
Notes: A firm’s seasonal earnings growth in a quarter is measured as the firm’s seasonal change in earnings before extraordinary items from the one-year-ago quarterly value, scaled by its average total assets. The measure is reported in percentage points. A firm’s earnings surprise for a quarter is measured as the difference between actual earnings and consensus (median) analyst forecast, divided by the end-of-quarter stock price. The measure is reported in percentage points. Panels A and B of the table report the results of Fama-MacBeth regressions of forecasting future E_CHANGE and future E_SURPRISE respectively. The dependent variable of Panel A is the seasonal earnings growth disclosed in the quarter following the PC_NII measurement quarter, and the dependent variable of Panel B is the earnings surprise disclosed in the quarter following the PC_NII measurement quarter. PC_NII is the percentage change in a stock’s number of institutional investors. E_CHANGE is the (most recently disclosed) seasonal earnings growth. E_SURPRISE is the (most recently disclosed) earnings surprise. INV (inventory), AR (accounts receivable), CAPX (capital expenditures), GM (gross margin), S&A (selling and administrative expenses), AQ (audit qualification), LF (labor force), ETR (effective tax rate) and EQ (earnings quality) are the fundamental signals identified in Lev and Thiagarajan (1993) and used by Abarbanell and Bushee (1997, 1998) to forecast future earnings changes. We follow Abarbanell and Bushee (1998) when constructing these variables. To utilize the most recently disclosed accounting information, we obtain the data used to calculate INV, AR, GM, S&A, and LF from Compustat quarterly file up to the fiscal quarter ended in the quarter before the PC_NII measurement quarter. We obtain the data used to calculate CAPX, AQ, ETR and EQ from Compustat annual file up to the fiscal year before the fiscal year of the fiscal quarter ended in the quarter before the PC_NII measurement quarter (to avoid look-ahead-bias). Future E_CHANGE, Future E_SURPRISE, E_CHANGE, E_SURPRISE, PC_NII and all of the fundamental signals except AQ and EQ (which are dummy variables) are winsorized at 1% and 99% each quarter. The coefficient of PC_NII is multiplied by 100. The adjusted R2 is the time-series average of cross-sectional regression adjusted R2. The Newey and West (1987) corrected t-statistics are reported in parentheses in all panels of the table.
53
Panel A: Forecast Future E_CHANGE Panel B: Forecast Future E_SURPRISE
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8
Intercept 0.06 0.03 0.08 0.03 -0.59 -0.29 -0.28 -0.28
(1.35) (0.83) (1.54) (0.50) (-13.45) (11.65) (-5.50) (-3.74)
PC_NII 1.26 0.83 1.16 1.17 1.25 0.54 0.66 0.82
(12.05) (10.03) (11.70) (10.69) (6.45) (5.25) (4.81) (6.02)
E_CHANGE 0.36 0.28 0.28
(34.56) (26.10) (22.14)
E_SURPRISE 0.70 0.69 0.69
(26.45) (24.60) (23.46)
INV 0.25 0.32 0.12 0.14
(11.25) (8.32) (2.76) (2.14)
AR 0.10 0.10 0.09 0.07
(2.85) (2.32) (2.46) (1.33)
CAPX -0.04 -0.02 -0.00 -0.02
(-3.96) (-1.74) (-0.20) (-1.43)
GM 0.18 0.17 0.06 0.08
(7.14) (5.62) (1.63) (1.45)
S&A 0.68 0.68 0.11 0.01
(12.12) (9.62) (1.95) (0.16)
AQ -0.04 -0.01 -0.01 0.02
(-1.22) (-0.27) (-0.15) (0.36)
LF 0.55 0.55 0.06 0.01
(12.49) (8.73) (1.03) (0.21)
ETR 1.13 -1.33
(3.54) (-2.09)
EQ -0.01 0.04
(-0.35) (1.25)
Adjusted R2 1.08% 13.51% 15.69% 17.45% 2.35% 35.48% 39.79% 45.09%
54
Table 8 Fama and MacBeth Regressions with Control for Future Earnings Surprises
Notes: The table reports the results of Fama-MacBeth regressions of forecasting one-quarter-ahead stock returns. PC_NII is the percentage change in a stock’s number of institutional investors. SIZE is the log market capitulation. BM is the log book-to-market equity ratio. BETA is market beta. MOM is the stock return in the previous 6 month. E_CHANGE is the (most recently disclosed) seasonal earnings growth. E_SURPRISE is the (most recently disclosed) earnings surprise. AMIHUD is the illiquidity measure. IV is the idiosyncratic volatility. Future E_CHANGE is the one-quarter-ahead E_CHANGE, and Future E_SURPRISE is the one-quarter-ahead E_SURPRISE. One-quarter-ahead stock return (i.e., the dependent variable), Future E_CHANGE, Future E_SURPRISE, Orthogonalized PC_NII, MOM, IV, E_CHANGE, and E_SURPRISE are all in percentage points. Future E_CHANGE, Future E_SURPRISE, E_CHANGE, E_SURPRISE, and PC_NII are winsorized at 1% and 99% each quarter. The coefficient of Orthogonalized PC_NII is multiplied by 100. The adjusted R2 is the time-series average of cross-sectional regression adjusted R2. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Controlling for Future
E_CHANGE Panel B: Controlling for Future
E_SURPRISE
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8
Intercept 3.61 3.72 5.53 5.50 4.01 3.20 6.57 5.24
(3.81) (3.92) (4.00) (4.60) (4.22) (3.30) (3.92) (3.33)
Future E_CHANGE 1.14
(13.56)
Future E_SURPRISE 1.86
(12.45)
ORTHO PC_NII 1.05 1.11 0.33 2.65 0.60 0.18
(1.45) (1.53) (0.77) (2.56) (0.63) (0.23)
Beta 0.46 0.63 -0.02 0.28
(0.69) (1.01) (-0.02) (0.44)
Size -0.18 -0.17 -0.30 -0.21
(-1.64) (-1.70) (-2.06) (-1.64)
B/M 0.30 0.54 0.24 0.04
(0.97) (2.04) (0.73) (0.12)
Mom 0.01 0.02 0.01 0.02
(1.17) (2.69) (1.47) (2.62)
Amihud -0.02 -0.01 -0.89 0.44
(-0.11) (-0.23) (-1.02) (0.77)
IV -0.83 -0.68 -0.79 -0.58
(-3.63) (-3.76) (-3.21) (-2.35)
E_SURPRISE 0.46 0.55
(4.49) (3.97)
E_CHANGE 0.22 0.07
(4.68) (1.02)
Adjusted R2 1.97% 0.26% 7.22% 6.24% 2.60% 0.08% 7.40% 7.06%
55
Table 9 Long-Run Portfolio Returns and Alphas
Notes: The table reports portfolio returns of the top (5) and bottom (1) PC_NII quintile portfolios, and the raw and DGTW (1997) characteristic-adjusted portfolio returns and the Carhart (1997) 4-factor alphas of the 5-1 strategy in the portfolio formation quarter (Q0) and each of the sixteen portfolio-holding quarters (Q1 to Q16) following the portfolio formation. The equal-weighted quintile portfolios are formed at the end of each portfolio formation quarter based on PC-NII in that quarter. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. Returns and alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Raw Portfolio Return DGTW-adj. Return 4-factor Alpha
Qtr. 5(H) 1(L) 5-1 t-Statistic 5-1 t-Statistic 5-1 t-Statistic
Q0 18.12 -1.83 19.95 (15.42) 19.49 (16.17) 19.16 (14.81)
Q1 4.55 2.60 1.95 (4.27) 1.86 (4.86) 1.28 (4.52)
Q2 3.90 3.22 0.68 (1.60) 0.60 (1.79) 0.03 (0.11)
Q3 3.09 2.98 0.11 (0.25) 0.07 (0.24) -0.24 (-0.81)
Q4 2.71 3.42 -0.71 (-1.77) -0.85 (-2.80) -0.87 (-2.80)
Q5 2.68 3.46 -0.79 (-2.48) -0.60 (-2.56) -0.74 (-2.58)
Q6 2.81 3.57 -0.76 (-2.56) -0.45 (-1.59) -0.69 (-2.39)
Q7 3.25 3.92 -0.67 (-2.43) -0.28 (-1.13) -0.37 (-1.08)
Q8 3.38 3.88 -0.50 (-1.56) -0.13 (-0.52) 0.01 (0.03)
Q9 3.22 4.20 -0.98 (-2.78) -0.80 (-2.47) -0.55 (-1.65)
Q10 3.46 4.01 -0.55 (-1.96) -0.31 (-1.33) -0.19 (-0.84)
Q11 3.72 3.90 -0.18 (-0.49) -0.11 (-0.38) 0.15 (0.34)
Q12 3.47 3.93 -0.46 (-1.61) -0.15 (-0.56) -0.17 (-0.54)
Q13 3.48 3.83 -0.35 (-1.28) -0.22 (-0.84) -0.06 (-0.22)
Q14 4.01 3.98 0.04 (0.12) 0.22 (0.78) 0.07 (0.20)
Q15 3.74 3.52 0.22 (0.80) 0.34 (1.27) 0.34 (1.33)
Q16 3.86 3.71 0.14 (0.56) 0.23 (0.93) 0.19 (0.66)
56
Table 10 Long-Run Portfolio PC_NII
Notes: We rank institutions equally into three terciles each quarter based on their average portfolio turnover in the past 4 quarters. Institutions of the top turnover tercile are classified as short term institutions, and those of the bottom turnover tercile are classified as long term institutions (as in Yan and Zhang (2009)). We then report the portfolio change in institutional ownership, ΔIO, of the short-term institutions, the long-term institutions and all institutions of the top (5) and bottom (1) PC_NII quintile portfolios and the 5-1 difference in the portfolio formation quarter (Q0) and the subsequent 8 portfolio-holding quarters (Q1 to Q8) following the portfolio formation. The equal-weighted quintile portfolios are formed at the end of each portfolio formation quarter based on PC-NII in that quarter. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. Panel A reports ΔIO of short-term institutions; Panel B reports ΔIO of long-term institutions; and Panel C reports ΔIO of all institutions. Short-term ΔIO, long-term ΔIO and ΔIO are demeaned (subtracted by the respective mean across all stocks in that quarter) and are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Short-Term ΔIO Long-Term ΔIO ΔIO
Quarters 5(H) 1(L) 5(H)-1(L) t-Statistic 5(H) 1(L) 5(H)-1(L) t-Statistic 5(H) 1(L) 5(H)-1(L) t-Statistic
Q0 1.94 -1.47 3.40 25.41 0.23 -0.22 0.45 12.26 3.12 -2.53 5.65 32.99
Q1 0.58 -0.25 0.83 13.75 0.10 -0.06 0.16 4.36 0.58 0.08 0.51 5.66
Q2 0.19 0.04 0.15 2.43 0.02 0.03 -0.01 0.31 0.39 -0.06 0.46 6.84
Q3 0.07 0.06 0.01 0.17 0.03 -0.01 0.04 1.16 0.26 0.05 0.20 3.00
Q4 -0.05 0.11 -0.15 -2.29 0.02 -0.04 0.07 1.99 0.15 0.06 0.10 1.26
Q5 -0.02 0.14 -0.16 -2.59 0.03 -0.04 0.07 2.16 0.05 0.15 -0.10 -1.81
Q6 -0.07 0.13 -0.20 -3.90 0.03 0.00 0.03 0.70 0.07 0.12 -0.05 -0.85
Q7 -0.05 0.16 -0.21 -3.73 0.06 -0.05 0.11 3.33 0.10 0.08 0.02 0.30
Q8 -0.01 0.09 -0.11 -1.78 0.03 0.00 0.03 0.91 0.05 0.11 -0.06 -0.95
i
Appendix A
Information Content of PC_NII and ΔNII in a Simple Stylized Model
In this appendix, we use a simple stylized model and simulation to show that PC_NII is a better
measure of informed trading than ΔNII when the frequency of a stock’s informed-based relative to
that of liquidity-based entries and exits decreases with the number of the stock’s institutional investors.
We do not attempt to provide a theoretical explanation for the information content of PC_NII and
ΔNII; rather, we illustrate that some well-documented stylized facts help explain the main finding
that PC_NII is a good measure of informed trading. Our model also has some refutable implications.
In each quarter, a company has a pool of active institutional investors. If an active institutional
investor owns some of the company’s shares at the end of the previous quarter, it can choose (1) to
continue to invest in the company or (2) to exit from the company, depending on the signal (discussed
in details later) that it receives this quarter. If an active institutional investor does not own any
shares at the end of the previous quarter, it can choose (1) to continue to stay away from the company
or (2) to enter into the company, depending on the signal that it receives this quarter. The inactive
institutions outside the pool do not own any shares and do not plan to do so in the current quarter.
There are also individual investors; for simplicity, we do not model their trading activity explicitly.
We assume that, for each company, the number of its active institutional investors is constant
across time. This assumption allows us to model the size explicitly in a simple manner: Big stocks
have a larger number of active institutional investors than do small stocks because the former have
lower trading costs and information costs. We ignore the fact that some small stocks will eventually
grow into big stocks, while some big stocks may eventually decline in the long run. Such a long-run
life cycle pattern, however, is likely to have a small effect on our main research question because we
study the relatively high frequency, i.e., quarterly, trading data. Note that the fixed number of active
institutional investors does not imply a fixed composition of active institutional investor pool.
Specifically, a current inactive institution can join the pool in the next quarter by replacing a currently
active institution that does not own any shares at the end of this quarter. For example, suppose that,
in the current quarter, A and B are active institutional investors of IBM and C is an inactive
institutional investor. A owns some IBM shares and B does not. In the next quarter, C may replace
ii
B as an active institutional investor of IBM. In simulation, on average 50% of the active institutions
in the pool own some shares in each quarter; and the other 50% active institutions do not own any
shares and are eligible to be replaced in the next quarter.
During any quarter t, some institutional investors receive information regarding future stock
return of the company in the next quarter and can trade on their private information. Specifically, an
active institutional investor of the company receives a signal with probability p; the institutions
receiving no signal do nothing about their portfolios. The signal is informative about next quarter
t+1’s stock return of the company (e.g., it summarizes fundamental news regarding the company that
will be released in the next quarter) with probability q and is liquidity-based or uninformative about
next quarter’s stock return of the company with probability 1-q. Informative signal is positive
(indicating positive abnormal return) with 30% chance, negative (indicating negative abnormal return)
with 30% chance and neutral (indicating zero abnormal return) with 40% chance. Note that results
are qualitatively similar for different parameterization (e.g., results are qualitatively similar if
informative signal is positive, negative and neutral with 40%, 40% and 20% chance, respectively).
When an active institutional investor receives an informative signal during quarter t, it invests in the
stock for the positive signal (i.e., it will remain a shareholder of the stock if it is a shareholder, and
will enter into the stock if it is not a shareholder, at the beginning of quarter t), does nothing about its
portfolio for the neutral signal, and stays away from the stock for the negative signal (i.e., it will
continue to stay away from the stock if it is not a shareholder, and will exit from the stock if it is a
shareholder, at the beginning of quarter t). Liquidity-based signal is positive with 50% chance and is
negative with 50%. When an institution receives a liquidity-based signal, it invests in the stock for
the positive signal and stay away from the stock for the negative signal. Therefore, we can think of
positive (negative) liquidity-based signal as a positive (negative) liquidity shock to an active
institutional investor. We assume that liquidity signal is idiosyncratic; i.e., unlike the informative
signal, liquidity-based signal does not cause investors to enter into or exit from a stock in a systematic
manner. The parameter q thus determines the informativeness of an entry or exit; for example, when
q equals 0, entries and exits are all liquidity-based and thus provide no information about future stock
returns. Allowing for an infrequent systematic liquidity shock should not affect our main results in
iii
any qualitative matter because it affects PC_NII and ΔNII in a similar manner. However, it is worth
noting that a drain in market-wide liquidity in the 2001 market crash and 2008 financial crisis does
make both PC_NII and ΔNII unreliable measures of informed trading; Indeed, as shown in the main
text (e.g., see Figure 3), the hedge portfolio formed on PC_NII performs rather poorly during both
episodes.
We calibrate the model with 200 companies, and set the number of active institutions for each
company to match the 20, 40, 60, 80, and 100 percentiles of the number of institutional investors in
the actual data (which are 10, 24, 57, 127, and 1683 institutions, respectively). We assume that, for
each company, 50% of its active institutional investors own some shares in the 1st period. We then
generate simulated data recursively for 222 periods, and use the last 122 periods from simulated data
for the regression analysis. We sort (the simulated) stocks equally into 5 portfolios by the number of
active institutional investors at the beginning of quarter t: Q1 (Q5) is the quintile of the fewest (the
most) active institutional investors. In Panel A of Table A1, we assume that the informativeness of
entries and exits decreases with the number of active institutional investors. Specifically, the
parameter q, the probability that a received signal is informative about next quarter t+1’s stock return
of a company, decreases monotonically from 0.50 for Q1 to 0.01 for Q5. We set the parameter p, the
probability of receiving a signal, to 0.25 so that the 25 percentile of PC_NII in simulated data (-4.90%)
matches closely its counterpart in actual data (-4.76%). In actual data, the 75 percentile of PC_NII is
noticeably greater than the 25 percentile of PC_NII in magnitude because the number of institutions
have increased steadily across time. Because our simulation does not incorporate such an increase,
in simulated data the 75 percentile of PC_NII is similar to the 25 percentile PC_NII in magnitude.
To illustrate the information content of PC_NII and ΔNII, we regress the realized value of the
informative signal, which equals 1 for positive future abnormal return , 0 for zero future abnormal
return, and -1 for negative future abnormal return on these two variables. Because by construction,
both PC_NII and Δ_NII correlate positively with the informative signal, their t-statistics go to infinity
asymptotically. To take into account the sampling errors in finite samples, we add an identically,
independently, and normally distributed white noise to the realized value of the informative signal.
The first column of Panel A in Table A1 reports the univariate regression results for ΔNII, whose
iv
coefficient is significantly positive at the 1% level. The second column shows that the coefficient of
PC_NII is also significantly positive in the univariate regression at the 1% level. Interestingly, in the
last two columns, we show that PC_NII drives out ΔNII in the multivariate regression, indicating that
PC_NII is a better measure of informed trading than is ΔNII in finite samples. Panel A shows that
results are qualitatively similar when we set the parameter p to 20% or 30%. As a robustness check,
in Panel B, we decrease the value of the parameter q to 30% for Q1 and find qualitatively similar
results. Because the active investor pool is far larger for the stocks in Q5 than those in the other
quintiles, the stocks in Q5 still have substantially more informed institutions even though it has a
relatively small q. As a robustness check, in untabulated results, we show that results are
qualitatively similar when we increase the parameter q to 0.02 for Q5. For comparison, in panels C
and D of Table A1, we show that ΔNII is more informative about future stock returns than is PC_NII
when the parameter q does not vary across the quintiles formed by the number of active institutional
investors. Overall, our simulation shows that PC_NII is a better measure of informed trading than is
ΔNII when the frequency of informed-based relative to that of liquidity-based trading is higher for
small stocks than for big stocks.
Diamond and Verrecchia (1991), Hermalin and Weisbach (2011), and others argue that large
firms adopt a higher level of information disclosure and thus have a lower level of information
asymmetry than do small firms. Merton (1987), Fishman and Hagerty (1989), and Diamond and
Verrecchia (1991) maintain that an increase in information disclosure reduces investors’ information
costs and trading costs, and hence attracts more liquidity traders. Consistent with this theoretical
implication, existing empirical studies(e.g., Bushee and Goodman (2007), Aslan, Easley, Hvidkjaer
and O'Hara (2008), Vega (2005), Brown, Hillegeist, and Lo (2004), Lai, Ng and Zhang (2009), and
Yan and Zhang (2009)) find that small stocks have substantially higher levels of informed trading than
big stocks. Because institutions prefer to hold large capitalization stocks (e.g., see Gompers and
Metrick (2001)), there is also a negative relationship between the number of institutional shareholders
and the level of informed trading.
There is a caveat, however. Kyle (1985) and Admati and Pfleiderer (1988) show that, if
investors are risk neutral and have non-binding capital constraints, an increase in the expected amount
v
of uninformed trading will be accompanied by a proportional increase in the amount of informed
trading, which renders the relative level of informed trading unchanged. In practice, however,
informed investors are usually risk-averse and have capital constraints; therefore, an increase in
uninformed liquidity trading is likely to attract a less than proportional increase in the amount of
informed trading. Therefore, whether big stocks have a lower level of informed trading than do
small firms is an empirical question. In the main text, we address this issue by investigating our
model’s 3 refutable implications and find strong support for them. First, ΔNII or ΔBREADTH is
more informative for stocks with higher trading costs and information costs. Second, such an
asymmetry is substantially less pronounced for PC_NII. Last, PC_NII is more informative about
future stock returns than is ΔNII or ΔBREADTH.
Financial crises provide a natural experiment to test whether institutional entries and exits are
more likely to be triggered by liquidity reasons for big stocks than for small stocks. To investigate
this issue, each year, we sort stocks in our actual sample equally into 5 portfolios by the number of
institutional investors at the end of the previous year: Q1 (Q5) is the quintile of the fewest (the most)
institutional investors. For our sample period from 1982 to 2010, we plot the 25 percentile of ΔNII
for each quintile in Figure A1 and the 75 percentiles in Figure A2. The 25 percentile is a rough
measure of the median number of exits and the 75 percentile is a rough measure of the median
number of entries. Figure A1 shows that, during the 2001 market crash and the 2008 financial crisis,
there is a big jump in the median exits for Q5—the quintile with the most institutional investors, while
the jump is substantially smaller for the other quintiles. Similarly, Figure A2 shows a more
pronounced decrease in entries for Q5 than the other quintiles. In Figure A3, we plot the ratio of the
median entries of Q5 to the average of the median entries across Q1 to Q3. It shows a substantial
larger decrease in median entries for Q5 than for Q1, Q2, and Q3 in both 2001 and 2008. Because
the median exits equal 0 in many years for the quintiles with few institutional investors, the ratio of
the median exits of Q5 to the average of median exits across Q1 and Q3 is quite volatile across time.
For brevity, we do not plot the ratio here but it is available on request. Thus, the actual data suggests
that institutional entries and exits are indeed more likely to be triggered by liquidity reasons for big
stocks than for small stocks.
vi
Table A1 Simulation Results
Notes: The table investigates the information content of PC_NII, the percentage change in the number of institutional investors, and ΔNII, the change in the number of institutional investors using simulated data from our simple stylized model. We report the results of Fama and MacBeth (1973) cross-sectional regressions of the informative signal about one-period-ahead abnormal stock returns on these variables. We use ΔNII, PC_NII, and both as explanatory variable(s) in column 1, column 2, and columns 3 and 4, respectively. See Appendix A for details.
Panel A: Q1=0.5; Q2=0.3; Q3=0.1; Q4=0.05; Q5=0.01 ΔNII PC_NII ΔBREADTH PC_NII
P=0.20 (PC_NII: 25th percentile=-4.04%; 75th percentile=3.86%) 0.923
(1.433) 1.294
(4.102) 0.251
(0.367) 1.240
(3.704) P=0.25 (PC_NII: 25th percentile=-4.90%; 75th percentile=4.71%)
1.617 (3.121)
1.293 (4.795)
0.935 (1.665)
1.158 (4.003)
P=0.30 (PC_NII: 25th percentile=-5.63%; 75th percentile=5.50%) 1.538
(3.252) 1.134
(4.967) 0.922
(1.848) 0.992
(4.177) Panel B: Q1=0.3; Q2=0.2; Q3=0.1; Q4=0.05; Q5=0.01
ΔNII PC_NII ΔBREADTH PC_NII P=0.20 (PC_NII: 25th percentile=-3.91%; 75th percentile=3.73%)
0.672 (1.032)
0.757 (2.233)
0.346 (0.499)
0.680 (1.893)
P=0.25 (PC_NII: 25th percentile=-4.70%; 75th percentile=4.50%) 1.368
(2.631) 0.954
(3.147) 0.954
(1.652) 0.791
(2.361) P=0.30 (PC_NII: 25th percentile=-5.34%; 75th percentile=5.20%)
1.322 (2.770)
1.066 (4.168)
0.823 (1.615)
0.922 (3.399)
Panel C: Q1=0.15; Q2=0.15; Q3=0.15; Q4=0.15; Q5=0.15 ΔNII PC_NII ΔBREADTH PC_NII
P=0.15 (PC_NII: 25th percentile=-3.69%; 75th percentile=3.60%) 1.368
(3.907) 1.105
(2.555) 1.216
(3.417) 0.781
(1.761) P=0.20 (PC_NII: 25th percentile=-4.70%; 75th percentile=4.73%)
1.212 (4.222)
0.594 (1.711)
1.166 (4.001)
0.267 (0.750)
P=0.25 (PC_NII: 25th percentile=-5.71%; 75t7h percentile=5.75%) 1.096
(4.840) 0.971
(3.088) 0.977
(4.260) 0.634
(1.980) Panel D: Q1=0.20; Q2=0.20; Q3=0.20; Q4=0.20; Q5=0.20
ΔNII PC_NII ΔBREADTH PC_NII P=0.15 (PC_NII: 25th percentile=-4.13%; 75th percentile=4.12%)
1.200 (4.363)
1.525 (3.632)
1.001 (3.512)
1.198 (2.785)
P=0.20 (PC_NII: 25th percentile=-5.32%; 75th percentile=5.42%) 0.988
(4.382) 0.933
(2.747) 0.890
(3.883) 0.623
(1.786) P=0.25 (PC_NII: 25th percentile=-6.59%; 75t7h percentile=6.73%)
0.873 (4.917)
1.334 (4.395)
0.714 (3.952)
1.017 (3.267)
vii
Figure A1 25 Percentile of ΔNII for Quintile Portfolios Sorted by Number of Institution Investors
Figure A2 75 Percentile of ΔNII for Quintile Portfolios Sorted by Number of Institution Investors
Figure A3
Median Entries of Q5 Relative to the Average of Median Entries of Q1, Q2, and Q3
‐25
‐20
‐15
‐10
‐5
0
Q1 Q2 Q3 Q4 Q5
0
5
10
15
20
25
Q1 Q2 Q3 Q4 Q5
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
viii
Appendix B Additional Results
Table B1
Correlations of Institutional Trading, Institutional Herding, and Earnings Surprises Notes: The table provides time-series average of cross-sectional correlations among the institutional and earnings variables. PC_NII is the percentage change in a stock’s number of institutional investors in the quarter, which is winsorized at 1% and 99% each quarter. ∆BREADTH is the change in the number institutional investors in the quarter normalized by the total number of institutions in the market, as adopted in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). IO is the fraction of shares owned by institutions at the end of the quarter. ∆IO is the change in IO in the quarter. Short-term ∆IO is the change in short-term institutions’ IO in the quarter proposed by Yan and Zhang (2009). LSV_HM is Lakonishok, Shleifer and Vishny’s (1992) herding measure. S_HM is Sias’ (2004) standardized (signed) herding measure. BWW_HM is Brown, Wei and Wermers’ (2009) signed herding measure. Persist is Dasgupta, Prat and Verardo’s (2011) persistence herding measure. E_CHANGE is the seasonal change in earnings before extraordinary items from the one-year-ago quarterly value scaled by average total assets for the fiscal quarter ended in the quarter before the PC_NII measurement quarter. E_SURPRISE is the most recently disclosed earnings surprise up to the end of the PC_NII measurement quarter, which is the difference between actual earnings per share and median analyst forecast scaled by the end-of-last quarter stock price. ***, ** and * denote 1%, 5% and 10% statistical significance respectively.
PC_NII ∆BREADTH ∆IO IO Short-Term ∆IO LSV_HM S_HM BWW_HM Persist E_CHANGE E_SURPRISE
PC_NII 1.00
∆BREADTH 0.45*** 1.00
∆IO 0.32*** 0.31*** 1.00
IO -0.08*** 0.09*** 0.14*** 1.00
Short-Term ∆IO 0.22*** 0.22*** 0.57*** 0.08*** 1.00
LSV_HM 0.31*** 0.14*** 0.17*** -0.06*** 0.13*** 1.00
S_HM 0.37*** 0.17*** 0.23*** -0.22*** 0.15*** 0.46*** 1.00
BWW_HM 0.33*** 0.18*** 0.22*** -0.19*** 0.14*** 0.38*** 0.93*** 1.00
Persist 0.20*** 0.14*** 0.27*** 0.03*** 0.20*** 0.12*** 0.27*** 0.28*** 1.00
E_CHANGE 0.06*** 0.05*** 0.03*** 0.02*** 0.02*** 0.03*** 0.04*** 0.04*** 0.04*** 1.00
E_SURPRISE 0.03*** 0.03*** 0.00 0.02*** 0.01*** -0.00 0.01 0.01 0.02** 0.22*** 1.00
ix
Table B2 Value-Weighted Decile Portfolios
Notes: The table reports one-quarter-ahead portfolio raw returns and DGTW-adjusted returns, the raw returns and DGTW-adjusted returns in the (-1,1) earnings announcement window, alphas of the Fama-French (1993) 3-factor model, alphas of the Carhart (1997) 4-factor model and alphas of the Pastor and Stambaugh (2003) 5-factor model of the value-weighted decile portfolios formed by sorting on PC_NII. PC_NII is the percentage change in a stock’s number of institutional investors, which is winsorized at 1% and 99% each quarter. We rebalance the decile portfolios at the end of each quarter. Returns and alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses
Deciles Raw
Return DGTW Return
3-Factor Alpha
4-Factor Alpha
5-Factor Alpha
Raw Return around EA
DGTW Return around EA
D1(L) 1.34 -1.74 -2.62 -1.76 -1.82 0.20 -0.07
D2 2.66 -0.38 -0.96 -0.22 -0.07 0.43 0.21
D3 2.67 -0.38 -0.75 -0.32 -0.32 0.48 0.26
D4 3.62 0.36 0.24 0.37 0.40 0.47 0.16
D5 3.18 -0.12 0.08 0.23 0.33 0.34 0.14
D6 2.97 -0.22 -0.26 -0.49 -0.35 0.29 0.09
D7 3.09 -0.11 -0.13 -0.40 -0.27 0.18 0.04
D8 3.99 0.60 0.68 0.12 0.01 0.19 0.06
D9 4.17 0.44 1.09 0.24 0.11 0.29 0.13
D10(H) 4.92 1.13 1.63 0.94 1.03 0.57 0.07
D10-D1 3.59 2.87 4.25 2.71 2.86 0.36 0.14
t-Statistic (4.58) (4.21) (5.34) (3.84) (4.13) (1.54) (0.68)
x
Table B3 Value-Weighted Returns on Portfolios Sorted on PC_NII with Control for Other Characteristics Notes: We sort stocks equally into 5 portfolios by a control variable and then within each quintile we sort stocks equally into 5 portfolios by PC_NII. We calculate the value-weighted return for each of the 25 portfolios. For the control variable SIZE, we report the Carhart (1997) 4-factor alphas for the 25 portfolios. For each PC_NII quintile, we also average returns across the SIZE quintiles and report the Carhart alphas for each aggregate PC_NII quintile. For the other control variable, we only report the Carhart alphas for aggregate PC_NII quintiles for brevity. Quintile 1 is the stocks with the lowest PC_NII and Quintile 5 the stocks with the highest PC_NII. SIZE is the market capitulation. BM is the book-to-market equity ratio. MOM is the stock return in the previous 6 month. E_CHANGE is the seasonal earnings growth. E_SURPRISE is the earnings surprise. AMIHUD is the Amihud (2002) illiquidity measure. IV is the idiosyncratic volatility. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market, as adopted in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). IO is the fraction of shares owned by institutions. ∆IO is the change in IO. Short-term ∆IO is the change in short-term institutions’ IO proposed by Yan and Zhang (2009). LSV_HM is Lakonishok, Shleifer and Vishny’s (1992) herding measure. S_HM is Sias’ (2004) standardized herding measure. BWW_HM is Brown, Wei and Wermers’ (2009) signed herding measure. Future E_CHANGE is the one-quarter lead of E_CHANGE, and Future E_SURPRISE the one-quarter lead of E_SURPRISE. Alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Controlling for Determinants of Cross-Sectional Stock Returns
Control Variable PC_NII 1(L) 2 3 4 5(H) 5-1 t-Statistic
1(S) -1.05 -0.43 0.19 0.18 0.48 1.53 (4.41)
2 -1.00 -0.26 -0.33 0.15 0.66 1.65 (3.55)
SIZE Quintile 3 -0.80 0.13 0.17 0.57 0.57 1.37 (3.03)
4 -1.00 0.14 -0.00 0.19 0.39 1.39 (3.46)
5(B) -0.17 0.11 0.25 -0.20 -0.24 -0.06 (-0.11)
SIZE Avg. -0.80 -0.06 0.05 0.18 0.37 1.18 (4.23)
BM Avg. -0.46 0.47 0.01 -0.10 0.79 1.25 (3.07)
MOM Avg. -0.98 -0.16 -0.28 0.19 0.57 1.55 (4.22)
E_CHANGE Avg. -0.55 0.27 -0.06 -0.22 0.56 1.12 (2.15)
E_SURPRISE Avg. -1.12 -0.52 -0.65 -0.82 0.04 1.16 (2.13)
AMIHUD Avg. -0.63 -0.05 0.07 0.18 0.32 0.95 (3.27)
IV Avg. -1.07 -0.51 -0.53 -0.46 0.33 1.40 (2.74)
Panel B: Controlling for Standard Measures of Institutional Trading and Herding
∆BREADTH Avg. -0.59 -0.45 -0.12 -0.07 0.29 0.88 (3.11)
IO Avg. -1.22 -0.44 0.07 -0.35 0.20 1.42 (2.70)
∆IO Avg. -0.66 -0.10 -0.24 -0.25 0.43 1.09 (2.26)
Short-Term ∆IO Avg. -0.57 -0.42 -0.02 -0.22 0.41 0.98 (1.90)
LSV_HM Avg. -0.72 -0.07 0.09 -0.18 0.41 1.13 (2.26)
S_HM Avg. -1.32 -0.48 -0.08 -0.18 0.48 1.80 (3.46)
BWW_HM Avg. -1.16 -0.60 -0.09 -0.02 0.44 1.60 (3.08)
Panel C: Controlling for Future Earnings Surprises
Future E_CHANGE Avg. -0.24 -0.03 -0.01 -0.52 0.25 0.49 (0.85)
Future E_SURPRISE Avg. -0.44 -0.05 -0.19 -0.86 -0.30 0.15 (0.26)
xi
Table B4 Portfolios Sorted on Measures of Institutional Trading and Herding with Control for PC_NII
Notes: We sort stocks equally into 5 portfolios by PC_NII, and then within each quintile we sort stocks equally into 5 portfolios by a measure of institutional trading or herding. For each of the 25 portfolios, we calculate the equal- and value-weighted returns in Panels A and B, respectively. For each measure of institutional trading or herding, we average returns across PC_NII quintiles, and report the Carhart (1997) 4-factor alphas for each of the aggregated quintiles formed on that institutional trading or herding measure. Quintile 1 (Quintile 5) is the stocks with the lowest (highest) institutional trading or herding. ∆BREADTH is the change in the number institutional investors normalized by the total number of institutions in the market as in Chen, Hong and Stein (2002) and Lehavy and Sloan (2008). IO is the fraction of shares owned by institutions. ∆IO is the change in IO. Short-term ∆IO is the change in short-term institutions’ IO as in Yan and Zhang (2009). LSV_HM is Lakonishok, Shleifer and Vishny’s (1992) herding measure. S_HM is Sias’ (2004) standardized herding measure. BWW_HM is Brown, Wei and Wermers’ (2009) signed herding measure. Alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Equal-Weighted Portfolios
Control Variable 1(L) 2 3 4 5(H) 5-1 t-Statistic ∆Breadth Quintile
PC_NII -0.04 -0.13 -0.09 0.10 -0.00 0.03 (0.15)
∆IO Quintile
PC_NII -0.03 0.16 0.08 -0.08 -0.30 -0.27 (-1.06)
Short-term ∆IO Quintile
PC_NII 0.11 -0.11 -0.04 0.03 -0.15 -0.26 (-1.31)
IO Quintile
PC_NII -0.33 -0.10 0.11 0.07 0.08 0.41 (0.92)
BWW_HM Quintile
PC_NII 0.15 0.22 0.01 0.04 -0.47 -0.62 (-1.80)
S_HM Quintile
PC_NII 0.11 0.25 0.16 -0.15 -0.41 -0.52 (-1.59)
LSV_HM Quintile
PC_NII 0.06 0.07 0.13 0.06 -0.37 -0.43 (-1.84)
Panel B: Value-Weighted Portfolios ∆Breadth Quintile
PC_NII -0.07 -0.35 -0.23 -0.09 -0.10 -0.03 (-0.09)
∆IO Quintile
PC_NII -0.27 -0.12 -0.26 -0.07 -0.27 0.01 (0.02)
Short-term ∆IO Quintile
PC_NII -0.32 -0.23 -0.09 -0.33 -0.03 0.30 (1.02)
IO Quintile
PC_NII -0.70 -0.54 0.13 -0.27 -0.15 0.55 (1.14)
BWW_HM Quintile
PC_NII 0.10 -0.20 -0.11 -0.15 -0.93 -1.03 (-2.02)
S_HM Quintile
PC_NII 0.14 -0.18 0.03 -0.49 -1.00 -1.14 (-2.18)
LSV_HM Quintile
PC_NII 0.15 -0.13 -0.16 -0.17 -0.19 -0.33 (-1.13)
xii
Table B5 Decile Portfolios Formed on Short- and Long-Term PC_NII
Notes: We rank institutions equally into three terciles each quarter based on their average portfolio turnover in the past 4 quarters. Institutions of the top turnover tercile are classified as short term institutions, and those of the bottom turnover tercile are classified as long term institutions. The long-term PC_NII is the change in the number of long-term institutional investors divided by the number of all institutional investors in the previous quarter. The short-term PC_NII is the change in the number of short-term institutional investors divided by the number of all institutional investors in the previous quarter. We sort stocks equally into decile portfolios by short-term PC_NII or long-term PC_NII, and rebalance the portfolios every quarter. Panel A reports the equal-weighted portfolio returns, and Panel B reports the value-weighted portfolio returns. Returns and alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Equal-Weighted Portfolio Returns
Short-Term PC_NII Long-Term PC_NII
Decile Raw
Return DGTW-adj.
Return 3-Factor Alpha
4-Factor Alpha
Raw Return
DGTW-adj. Return
3-Factor Alpha
4-Factor Alpha
D1(L) 2.51 -0.96 -1.15 -0.91 3.51 -0.15 -0.12 -0.15
D2 2.93 -0.62 -0.83 -0.43 3.65 -0.07 -0.12 0.07
D3 3.07 -0.51 -0.60 -0.40 3.42 -0.01 -0.26 -0.11
D4 3.57 -0.09 -0.16 0.08 3.29 -0.22 -0.39 -0.27
D5 3.67 0.00 0.18 0.23 3.65 -0.06 0.05 0.14
D6 3.49 -0.28 -0.07 -0.16 3.44 -0.28 -0.09 -0.16
D7 3.74 0.05 0.21 0.12 3.74 0.06 0.12 0.04
D8 3.83 0.07 0.22 0.12 3.73 0.05 0.15 0.05
D9 4.37 0.61 0.82 0.43 3.72 0.01 0.26 0.10
D10(H) 4.74 1.03 1.39 0.57 3.76 -0.05 0.42 -0.07
D10-D1 2.23 1.99 2.55 1.48 0.25 0.09 0.54 0.08
t-Statistic (4.16) (4.31) (4.76) (4.34) (0.64) (0.29) (1.22) (0.21)
Panel B: Value-Weighted Portfolio Returns
Short-Term PC_NII Long-Term PC_NII
Decile Raw
Return DGTW-adj.
Return 3-Factor Alpha
4-Factor Alpha
Raw Return
DGTW-adj. Return
3-Factor Alpha
4-Factor Alpha
D1(L) 2.02 -1.05 -1.59 -1.09 2.52 -0.67 -1.43 -1.26
D2 2.33 -0.88 -1.35 -0.81 3.43 0.43 -0.25 0.32
D3 2.32 -0.68 -1.04 -0.59 2.75 -0.27 -0.73 -0.41
D4 3.28 0.01 -0.11 0.21 3.29 0.10 0.05 0.04
D5 3.47 0.13 0.28 0.54 3.37 0.00 0.28 0.28
D6 3.20 -0.20 -0.05 -0.45 3.09 -0.23 -0.02 -0.26
D7 3.63 0.16 0.53 0.19 3.17 -0.13 -0.11 -0.16
D8 3.86 0.16 0.51 -0.12 3.25 0.19 0.11 -0.02
D9 4.24 0.68 0.88 0.39 3.51 0.10 0.35 -0.07
D10(H) 4.81 1.04 1.76 0.90 3.84 0.18 0.57 -0.02
D10-D1 2.79 2.09 3.35 1.99 1.32 0.86 2.00 1.24
t-Statistic (3.40) (3.11) (3.95) (2.50) (1.96) (1.50) (2.88) (1.69)
xiii
Table B6 Future Earnings Changes of Portfolios Sorted on Earnings Changes and PC_NII
Notes: In Panel A, we sort stocks equally into 5 portfolios by E_CHANGE, the (most recently disclosed) seasonal earnings growth, and then within each E_CHANGE quintile we sort stocks equally into 5 portfolios by PC_NII. The portfolios are rebalanced every quarter. For each of the 25 portfolios, we calculate the portfolio median future E_CHANGE. We also average across E_CHANGE quintiles and report the average portfolio median future E_CHANGE for each of the aggregated PC_NII quintiles. In Panel B, we sort stocks equally into 5 portfolios by E_SURPRISE, the (most recently disclosed) earnings surprise, and then within each E_SURPRISE quintile we sort stocks equally into 5 portfolios by PC_NII. The portfolios are rebalanced every quarter. For each of the 25 portfolios, we calculate the portfolio median future E_SURPRISE. We also average across E_SURPRISE quintiles and report the average portfolio median future E_SURPRISE for each of the aggregated PC_NII quintiles. The 5-1 differences are also reported. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Panel A: Median Future E_CHANGE (in percentage) for Portfolios Formed on E_CHANGE and PC_NII
Control Variable PC_NII Quintile 1(L) 2 3 4 5(H) 5-1 t-Statistic
1(L) -0.98 -0.74 -0.60 -0.52 -0.62 0.37 (9.28)
2 -0.12 -0.09 -0.06 -0.05 -0.03 0.09 (6.57)
E_CHANGE Quintile 3 0.07 0.09 0.10 0.11 0.11 0.04 (8.33)
4 0.30 0.35 0.40 0.44 0.50 0.20 (14.87)
5(H) 0.77 0.88 1.03 1.20 1.51 0.74 (15.83)
AVERAGE Avg. 0.01 0.10 0.17 0.24 0.30 0.29 (23.01) Panel B: Median Future E_SURPRISE (in percentage) for Portfolios
Formed on E_SURPRISE and PC_NII
Control Variable PC_NII Quintile 1(L) 2 3 4 5(H) 5-1 t-Statistic
1(L) -1.73 -1.47 -1.34 -1.17 -0.95 0.79 (10.18)
2 -0.42 -0.29 -0.24 -0.23 -0.19 0.23 (11.02)
E_SURPRISE Quintile 3 -0.08 -0.05 -0.04 -0.03 -0.02 0.07 (6.31)
4 0.01 0.04 0.05 0.05 0.07 0.05 (6.10)
5(H) 0.08 0.18 0.18 0.24 0.25 0.18 (6.31)
AVERAGE Avg. -0.43 -0.32 -0.28 -0.23 -0.17 0.26 (14.06)
xiv
Table B7 Long-Run Portfolio Returns and Alphas Constructed Using Delisting Returns
Notes: The table reports portfolio returns of the top (5) and bottom (1) PC_NII quintile portfolios, and the raw and DGTW (1997) characteristic-adjusted portfolio returns and the Carhart (1997) 4-factor alphas of the 5-1 strategy in the portfolio formation quarter (Q0) and each of the sixteen portfolio-holding quarters following the portfolio formation (Q1 to Q16). We construct portfolio returns using delisting returns reported by CRSP and following the procedure proposed by Beaver, McNichols, and Price (2007). The equal-weighted quintile portfolios are formed at the end of each portfolio formation quarter based on PC-NII in that quarter. Returns and alphas are in percentage points. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Raw Portfolio Return DGTW-adj. Return 4-factor Alpha
Qtr. 5(H) 1(L) 5-1 t-Statistic 5-1 t-Statistic 5-1 t-Statistic
Q0 18.12 -1.83 19.94 (15.42) 19.50 (16.18) 19.15 (14.80)
Q1 4.59 2.65 1.94 (4.25) 1.87 (4.91) 1.26 (4.46)
Q2 3.87 3.16 0.71 (1.76) 0.65 (2.01) 0.10 (0.33)
Q3 3.02 2.98 0.04 (0.10) -0.01 (-0.02) -0.37 (-1.22)
Q4 2.72 3.39 -0.67 (-1.62) -0.73 (-2.33) -0.84 (-2.65)
Q5 2.65 3.51 -0.86 (-2.51) -0.66 (-2.67) -0.76 (-2.38)
Q6 2.79 3.55 -0.76 (-2.78) -0.42 (-1.57) -0.71 (-2.54)
Q7 3.22 3.98 -0.76 (-2.65) -0.33 (-1.27) -0.47 (-1.23)
Q8 3.34 3.81 -0.46 (-1.43) -0.09 (-0.37) 0.07 (0.20)
Q9 3.16 4.20 -1.04 (-2.88) -0.82 (-2.41) -0.60 (-1.78)
Q10 3.41 3.94 -0.53 (-1.80) -0.33 (-1.33) -0.18 (-0.69)
Q11 3.72 3.90 -0.18 (-0.54) -0.09 (-0.33) 0.20 (0.52)
Q12 3.47 3.78 -0.30 (-1.08) 0.02 (0.09) -0.02 (-0.08)
Q13 3.48 3.74 -0.26 (-0.91) -0.21 (-0.80) 0.04 (0.15)
Q14 3.97 3.83 0.14 (0.45) 0.28 (0.97) 0.17 (0.49)
Q15 3.71 3.48 0.24 (0.87) 0.35 (1.31) 0.34 (1.33)
Q16 3.83 3.68 0.15 (0.57) 0.20 (0.84) 0.20 (0.70)
xv
Table B8 Long-Run Reversal Fama and MacBeth Regressions
Notes: The table reports the results of long-run reversal Fama-MacBeth regressions. The dependent variable in Models 1 and 5 is the return of the fourth quarter after the PC_NII measurement quarter. The dependent variable in Models 2 and 6 is the buy-and-hold return of the fourth and fifth quarters after the PC_NII measurement quarter. The dependent variable in Models 3 and 7 is the buy-and-hold return of the fourth, fifth and sixth quarters after the PC_NII measurement quarter. The dependent variable in Models 4 and 8 is the buy-and-hold return of the fourth, fifth, sixth and seventh quarters after the PC_NII measurement quarter. PC_NII is the percentage change in a stock’s number of institutional investors. SIZE is the log market capitulation. BM is the log book-to-market equity ratio. BETA is market beta. MOM is the stock return in the previous 6 month. AMIHUD is the illiquidity measure. IV is the idiosyncratic volatility. Persist is Dasgupta, Prat and Verardo’s (2011) persistence herding measure. BWW_HM is Brown, Wei and Wermers’ (2009) signed herding measure. Past3yr_Return is the past-three-year’s buy-and-hold return of the stock advocated by Conrad and Kaul (1993). Future stock return (i.e., the dependent variable), PC_NII, Past3yr_Return, MOM and IV are all in percentage points. PC_NII is winsorized at 1% and 99% each quarter. The coefficient of PC_NII is multiplied by 100 for the ease of presentation. The adjusted R2 is the time-series average of cross-sectional regression adjusted R2. The Newey and West (1987) corrected t-statistics are reported in parentheses.
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8
Intercept 2.44 4.44 6.32 8.36 2.58 4.32 6.01 7.76
(2.19) (2.10) (2.03) (2.02) (2.24) (2.05) (1.94) (1.87)
PC_NII -1.10 -1.40 -1.60 -2.40 -1.30 -1.40 -0.90 -1.70
(-2.47) (-2.08) (-2.16) (-2.72) (-2.66) (-1.82) (-1.00) (-1.51)
Persist -0.07 -0.11 -0.16 -0.25
(-2.22) (-2.12) (-2.65) (-3.38)
BWW_HM -0.93 -1.54 -2.43 -2.67
(-2.40) (-2.24) (-2.96) (-2.77)
Past3yr_Return -0.00 -0.00 -0.00 -0.00
(-0.75) (-0.48) (-0.65) (-1.12)
Beta 0.65 1.10 1.64 2.15 0.88 1.66 2.35 2.99
(1.03) (0.92) (0.97) (1.02) (1.41) (1.41) (1.43) (1.42)
Size 0.09 0.30 0.48 0.62 0.03 0.21 0.36 0.49
(0.96) (1.71) (2.10) (2.28) (0.29) (1.17) (1.53) (1.71)
B/M 0.65 1.47 2.10 2.73 0.49 1.14 1.61 1.99
(2.86) (3.44) (3.47) (3.45) (2.09) (2.55) (2.42) (2.26)
Mom 0.01 0.00 -0.01 -0.00 0.01 0.01 -0.00 0.00
(1.63) (0.37) (-0.52) (-0.26) (2.30) (0.72) (-0.30) (0.08)
Amihud 0.09 0.10 0.09 0.06 0.06 0.06 -0.12 -0.23
(1.39) (1.03) (0.80) (0.40) (0.49) (0.32) (-0.49) (-0.78)
IV -0.01 -0.08 0.01 0.24 -0.01 -0.12 0.05 0.33
(-0.07) (-0.22) (0.02) (0.35) (-0.03) (-0.30) (0.08) (0.45)
Adjusted R2 5.11% 5.21% 5.15% 5.18% 5.61% 5.87% 5.81% 5.86%