stock duration, analyst recommendations, … · stock duration, analyst recommendations, and ......
TRANSCRIPT
Stock Duration, Analyst Recommendations, and Overvaluation
Martijn Cremers
University of Notre Dame
Ankur Pareek
Rutgers University
Zacharias Sautner
Frankfurt School of Finance & Management
June 2016
Abstract
This paper documents how the interaction between short‐term investors and analyst recommendations relates
to a speculative component in stock prices that results in temporary overvaluation with predictable, large price
reversals. In particular, stocks held by short‐term institutions with optimistic analyst recommendations have
large past outperformance, followed by large negative future alphas. Our results are robust to using Russell
2000 index reconstitutions to capture exogenous changes in institutional ownership, short‐term trading and
analyst coverage, and are stronger among stocks that are harder to short, consistent with limited arbitrage.
________________
We would like to thank Lauren Cohen, Jesse Fried, William Goetzmann, David Hirshleifer, Harald Hau, Zhiguo He, Pedro
Matos (EFA discussant), Florian Peters, Breno Schmidt (AFA discussant), David Sraer, Russ Wermers, and Ramona
Westermann as well as participants at the American Finance Association Meetings, European Finance Association
Meetings, Society of Financial Studies Finance Cavalcade, INQUIRE Europe Munich Conference, Geneva Summit on
Sustainable Finance, and the Roundtable on Long‐Term Value in the Corporation at Harvard Law School for their
comments. This paper was previously titled ‘Stock Duration and Misvaluation’. Comments are very welcome. All errors are
our own.
2
1. Introduction
This study examines the interaction between short‐term institutional trading and analyst
recommendations. Significant separate literatures have considered the questions of whether short‐term
institutional trading is associated with a speculative component in stock prices, and whether security analysts
improve market efficiency. These literatures provide mixed evidence on both questions. On institutional
trading, there is evidence that short‐term institutions are associated with more efficient markets (Collins, Gong,
and Hribar, 2003; Ke and Ramalingegowda, 2005; Bartov, Radhakrishnan, and Krinsky, 2000; and Boehmer and
Kelley, 2009), but also with more anomalous pricing (Hou, Xiong, and Peng, 2009; and Cremers and Pareek,
2015). Similarly, previous literature has documented that analysts provide significant value to investors
(Brennan, Jegadeesh, and Swaminathan, 1993; Womack, 1996; Barber et al., 2001; Jegadeesh et al., 2004), but
also that analysts themselves may be biased (DeBondt and Thaler, 1990; Ertimur, Muslu, and Zhang, 2011).
Combining these literatures, our main contribution is to document that short‐term institutional trading
in the presence of ‘strong buy’ recommendations by analysts leads to overvaluation, followed by significant
price reversals, both in the three‐day windows around subsequent announcements of analyst recommendation
revisions and in calendar‐time returns measured over the next year. For our identification, we use additions to
the Russell 2000 index to capture exogenous changes in institutional ownership, short‐term institutional
trading and analyst coverage, and show that our results are robust when we only consider such index
additions.
Our empirical results are generally consistent with the theoretical models of Harrison and Kreps (1978)
and Scheinkman and Xiong (2003), which predict that heterogeneous investor beliefs combined with short‐
sales constraints lead to high trading volume or shorter holding durations accompanied by overvaluation.
Intuitively, if the marginal investors in some stocks are relatively optimistic, then short‐sales constraints
prevent less optimistic investors from correcting any over‐optimism, allowing temporary overvaluation. In
particular, we test a specific prediction following from these theories, namely that positive public news that
3
confirms and perhaps strengthens the optimistic investors’ beliefs could exacerbate the price impact of
disagreement, leading to even greater overvaluation.1 Moreover, the theoretical model in Froot, Scharfstein,
and Stein (1992) helps explain the role of analyst recommendations, whose public disclosure can act as a
coordination device for short‐term traders. Their model predicts that short‐term investors coordinate around
public news announcements with relatively short‐term investment value (such as analyst recommendations,
see Womack, 1996 and Barber et al., 2001), which they predict negatively impacts the informational quality of
stock prices.
Consistent with these theoretical predictions, we find that short‐term institutional trading combined
with strongly positive consensus analyst recommendations leads to significant stock price overvaluation and
subsequent price reversals. The predictable price declines among stocks with short‐term investors and ‘strong
buy’ recommendations coincide with subsequent mean‐reverting analyst recommendations, which tend
become gradually less optimistic. On the one hand, these results imply significant over‐optimism among short‐
term investors in stocks with ‘strong buy’ recommendations. As our results are stronger among stocks that are
harder to short sell, arbitrage constraints appear to allow temporary overvaluation. On the other hand, analyst
activity seems to exacerbate overvaluation, rather than reducing any speculative component in stock prices
and making markets more efficient, as analysts seem slow in reversing their most optimistic recommendations.
Our main proxy for the presence of speculative trading is a novel measure of the holding durations of
institutional investors, introduced in Cremers and Pareek (2015). This measure, called Stock Duration, is
calculated as the weighted‐average length of time that institutional investors have held a stock in their
1 Various mechanisms can explain the origins of differences of opinion, investor optimism, and effects of public news (see Hong and Stein, 2007, for a review). Scheinkman and Xiong (2003) explain differences of opinion through investor overconfidence, which leads them to overestimate the precision of their private signals, causing them to overreact. As arbitrageurs can arbitrage away overreaction to negative news by taking long positions, only overreaction to positive news by over‐optimistic investors has a price impact, as short‐sales constraints limit arbitrage. Daniel, Hirshleifer, and Subrahmanyam (1998) argue that due to self‐attribution bias, investor overconfidence increases in the presence of positive public news that confirms optimistic private valuations. Hong, Scheinkman, and Xiong (2008) propose an alternative model to explain stock market overreaction, without recourse to investor overconfidence and analyst bias. In their model, some analysts rationally inflate their forecasts to signal that they are well‐informed, but naive investors mistakenly take their forecasts at face value and overreact, causing overvaluation.
4
portfolios. It is based on quarterly 13F holding reports and weighted by the dollar amount invested across all
institutions currently holding a stock. Stocks with short holding durations experience the arrival of many new
and possibly speculative investors.2 Our proxy for extreme analyst recommendations is the top and bottom
quintile of consensus (or mean) analyst recommendations. We show that both extremes in analyst
recommendations and extremes in Stock Duration (either very short or very long) are mean reverting, but not
in ways that are related.
We first document that stocks held by more short‐term institutions tend to overreact to analyst
recommendations when these are made public. In particular, for stocks with the most optimistic (pessimistic)
analyst recommendations, the three‐day abnormal returns around analysts’ recommendation announcements
are significantly higher for stocks with more short‐term institutions. For example, for stocks with short holding
durations and optimistic (pessimistic) consensus analyst recommendations, the average three‐day cumulative
abnormal return around all recommendation announcements during the past one year is 1.51% (‐3.12%); the
difference of 4.63% is highly significant with a t‐stat of 13.68. The corresponding difference for stocks with long
holding durations equals only 1.52%, which is significantly lower.
In the subsequent year, the most optimistic and pessimistic analyst recommendations both tend to
revert back to the mean. Over this period, and conditioning on past information only, we find stock price
reversals in the three‐day window around subsequent analyst recommendations. These abnormal
announcement returns are significantly stronger for stocks with more short‐term institutions, suggesting that
these were most likely to overreact to previous analyst announcements. For example, for stocks with short
holding durations that currently have the most optimistic (pessimistic) consensus analyst recommendations,
2 Stock Duration allows any given institutional investor to be short‐term in some stocks and long‐term in others; e.g. because investments are made by different portfolio managers with heterogeneous horizons and trading motives. Accounting for this is important as one can only observe institutional holdings at an aggregate institution level using 13F filings. The limitation of Stock Duration is that the quarter‐end holdings that are used to compute it ignore roundtrip trades within the quarter. Our results are generally robust to using share turnover instead of Stock Duration. Share turnover captures all trading activity, but in recent years is strongly impacted by high frequency trading, which is arguably different from the speculative and relatively uninformed trading we aim to capture.
5
the average three‐day cumulative abnormal return around subsequent recommendation announcements over
the next one year equals ‐1.36% (1.10%); the difference of 2.47% is highly significant with a t‐stat of 10.75. The
corresponding difference in future abnormal returns is only 0.50% among stocks with long holding durations.
We next examine longer‐term mispricing through calendar‐time portfolios of stocks sorted on past
analyst recommendations and past holding durations. Specifically, each quarter we independently double‐sort
stocks into quintile portfolios based on Stock Duration and consensus recommendations, and then we calculate
portfolio returns over the next twelve months. We find that stocks with short‐term institutions and optimistic
analysts have large negative future alphas over next year, which follow large outperformance over the previous
year. For example, the value‐weighted portfolio of stocks with short (first quintile) holding durations and the
most optimistic (first quintile) analyst recommendations has an annualized 5‐factor alpha of ‐8.8% (t‐stat of
3.71), while the portfolio with short holding durations and the most pessimistic analyst recommendations has a
5‐factor alpha of 5% per year (t‐stat of 1.79). As a result, the long‐short portfolio that buys stocks in the latter
and sells stocks in the former group has a 5‐factor alpha of 13.8% per year (t‐stat of 3.67).3 As stocks with the
most (least) optimistic analyst recommendations have large positive (negative) lagged abnormal returns, these
results indicate large price reversals.4
These price reversals only occur for stocks that have both much short‐term trading (i.e., Stock Duration
in the bottom quintile) and the most optimistic analyst recommendations (i.e., top quintile), indicating that it is
their interaction that matters. Moreover, we find that these price reversals can be explained by limited
3 Consistent with slow mean reversion of both analyst recommendations and holding durations, lagging portfolio construction by up to four quarters hardly diminishes the future alphas of the constructed portfolios. Results are also similar when we use share turnover as a robustness check. 4 For large stocks, price reversals are stronger for stocks with more optimistic recommendations, which helps rule out an explanation based on information trading by short‐term investors. If short‐term investors are generally well‐informed, then we would expect asymmetric results, i.e., that the short‐term traders only hold stocks with positive (rather than negative) future alpha. However, for value‐weighted portfolios the negative alpha (of the portfolio with stocks with the most optimistic analyst recommendations and short holding durations) is higher in absolute value compared to the positive alpha (of the portfolio with stocks with the most pessimistic analyst recommendations and short holding durations). More generally, on average short‐term investors do not appear to be well‐informed, as the portfolio with short holding durations (bottom quintile) has an insignificant alpha for both equally‐ and value‐weighted portfolios.
6
arbitrage (e.g., Duffie, 2010), as the abnormal returns are driven by stocks that are more difficult to short and
for stocks that are more risky to short.5 Our proxies for the difficulty to short are the short ratio (Asquith,
Pathak, and Ritter, 2005) and the percentage of the stock held by DFA (Nagel, 2005), while our proxy for the
riskiness of shorting is a stock’s idiosyncratic volatility (Pontiff, 1996).
Empirically, we do not find any correlation between stocks having extreme analyst recommendations
and having either more speculative investors or more frequent trading. This suggests that extreme analyst
recommendations are not directly attracting short‐term traders, but rather that, among stocks with many
short‐term traders, extreme analyst recommendations serve as a coordination mechanism. This is similar to the
interpretation of the results in a related paper by Balakrishnan, Schrand, and Vashishtha (2014), who find
stronger ‘bubble continuation’ among stocks with strong and concentrated analyst buy recommendations in
the 2000 technology bubble.
The endogeneity of both analyst recommendations and the presence of short‐term traders may
complicate the interpretation of our results. For example, it is possible that exposure to unidentified risk
factors or systematic news could explain this predictability, if such differences in exposure are related to
differences in analyst recommendations. To mitigate endogeneity concerns, we identify plausibly exogenous
changes in both analyst coverage and institutional ownership arising from the annual reconstitution of the
Russell 2000 index.6 Specifically, we focus on stocks newly added to the Russell 2000 ‘from below’, i.e., stocks
5 Duffie (2010) considers how impediments to institutional trading may allow extended periods of asset price distortions. Limits to arbitrage may give rise to temporary mispricing with subsequent price reversals that occur only slowly over time. Duffie (2010) mentions several specific examples of the long‐standing price impact of supply and demand shocks in the presence of slow moving arbitrage capital, such as trading around index reconstitution (e.g., Shleifer, 1986; Chen, Noronha, and Singhal, 2004; Greenwood, 2005; and Petajisto, 2011), trading around large supply shocks induced by extreme mutual fund flows (e.g., Coval and Stafford, 2007; Edmans, Goldstein, and Jiang, 2012), and trading around temporary market dislocations due to the slow moving withdrawal of arbitrage capital in, for example, the convertible bond market (e.g., Mitchell, Pedersen, and Pulvino, 2007). 6 Russell index reconstitutions are events that do not provide new information to the market (e.g., because they are quite predictable, as they are based on market capitalizations at the end of each June), but are accompanied with significant buying and selling, respectively, by funds tracking the index (Lynch and Mendenhall, 1997; Chang, Hong, and Liskovich, 2015). We focus on the Russell 2000, where the price impact of its reconstitutions is particularly significant (Petajisto, 2011; Cremers, Petajisto, and Zitzewitz, 2012).
7
that were previously not included in the Russell 1000 but whose strong recent positive abnormal performance
and increase in market value lead to Russell 2000 inclusions.7
Stocks that are added to the Russell 2000 from below and had previously short holding durations
experience major changes at the time of the index reconstitution. First, institutional ownership in these stocks
sharply increases from a pre‐inclusion average of 20% to over 40% soon after inclusion. Second, both their
holding durations and analyst coverage increase sharply (Stock Duration jumps from about two quarters before
to about four quarters afterwards; the average number of analysts increases from about one analyst before to
over three analysts afterwards). These large increases in institutional ownership, holding durations, and analyst
coverage constitute important and (arguably) largely exogenous shocks to these stocks, which should diminish
the role of the short‐term speculators by improving the informational environment and relieving short sales
constraints, thereby facilitating price reversals of any previous overreactions. There are no correspondingly
large changes for stocks added to the Russell 2000 from below that previously had long holding durations.
Our main result of predictable price reversals for stocks with short holding durations and optimistic
analysts continue to hold when we only consider stocks that are added to the Russell 2000 from below. Stocks
with strongly optimistic analyst recommendations and short durations appear to be significantly overvalued at
the time when they are added to the Russell 2000. After addition to the index, this overvaluation slowly
reverses over the subsequent two years, as the role of speculative, short‐term investors is much reduced due
to the major increase in institutional ownership and analyst coverage.8 As these latter changes are plausibly
7 We focus on index inclusions ‘from below’ as there are about 300 inclusion events per year, on average; this is different for index inclusion ‘from above’ (i.e., Russell 2000 additions of firms previously included in the Russell 1000, but whose market capitalization ranking declined), which occur with a frequency of only about 50 per year. We also cannot find significant changes in institutional ownership, Stock Duration or analyst coverage for stocks added to the Russell 2000 ‘from above’. While our main results rely only on Russell 2000 additions from below, we verify that results are robust once we also include Russell 2000 additions from above. Further, as typical in the literature, we do not look at index exclusions, as these are usually due to information‐related events, such as bankruptcies, mergers, or acquisitions (e.g., Shleifer, 1986; Harris and Gurel, 1986). 8 These price reversals thus appear over longer horizons than considered in the previous literature. Boehmer and Kelley (2009), for example, find that short‐term trading is associated with increased price efficiency, but they only consider periods of up to three months.
8
exogenous to firm‐specific events, we conclude that short‐term speculation can significantly contribute to
overvaluation, especially for stocks that are hard to arbitrage and lack significant analyst coverage.
In conclusion, our main result is that stocks held by short‐horizon institutional investors with optimistic
analyst recommendations have large and predictable negative future alphas that reverse large positive past
outperformance. This result is closely related to several papers. Ertimur, Muslu, and Zhang (2011) document
that stocks that currently have ‘strong buy’ recommendations underperform in the future. We find that these
price reversals only occur for stocks with short holding durations. Accordingly, optimistic analyst
recommendations seem to provide a coordination mechanism that allows speculative investors to synchronize
their trading, consistent with Froot, Scharfstein, and Stein (1992), exacerbating their collective price impact and
leading to substantial temporary mispricing. We find no evidence that optimistic analyst recommendations
themselves attract speculative investors, though our results do suggest that analysts with strong
recommendations may be overreacting and may be too slow in subsequently updating (i.e., moderating) their
views. We thus conclude that strong analyst recommendations aggravate the price impact of speculative
trading, while we find no evidence that the number or the dispersion of recommendations help to mitigate
such price impact.
Our paper is also related to the literature studying how institutions trade upon the release of analyst
information, such as Mikhail, Walther, and Willis (2007), Busse, Green, and Jegadeesh (2012), and Brown, Wei,
and Wermers (2014). These studies focus on how institutions such as mutual funds behave, while our focus is
on the effects thereof on stock prices. For example, Brown, Wei, and Wermers (2014) document that analyst
recommendation revisions induce herding by career‐concerned fund managers. They also link such herding to
an overreaction in stock prices, which is consistent with our findings, though their results are economically
significantly smaller. However, herding is one potential explanation for the behavior of short‐term institutions.
A complimentary hypothesis is considered by Cella, Ellul, and Giannetti (2013), who show that selling
pressure of stocks held by short‐term institutional investors is amplifying market‐wide negative shocks. They
9
document that such selling pressure leads to temporary undervaluation, which they explain by the demand of
liquidity at times when arbitrage capital is scarce. While they focus on particular periods of market turmoil with
systematic negative shocks, our paper does not condition on ‘abnormal’ times, considers both overvaluation
and undervaluation (we find mostly evidence for overvaluation), and focuses on the role of analysts.
Our paper is also related to Sias, Starks, and Titman (2006), who consider the relation between stock
returns and ownership changes to interpret the strong contemporaneous positive association between
quarterly changes in aggregate institutional ownership and returns (e.g., Nofsinger and Sias, 1999; Wermers,
1999). As they argue, “the demand for shares from one group of investors must be offset by the supply of
shares from another group of investors. Hence, if we believe that aggregate institutional buying causes returns
to increase, we are implicitly assuming that selling by non‐institutional investors does not have a countervailing
effect.” However, Sias, Starks, and Titman (2006) find no subsequent price reversals after institutional
turnover—consistent with our unconditional results—and conclude that the positive correlation is due to
information rather than any price impact. The main innovation of our paper that explains our different results
is that we distinguish among institutional investors those who are generally short‐term (and thus may be more
likely to demand liquidity), versus those that are more patient. Moreover, we focus on the interaction of short‐
term investors with analysts, which can serve as a coordination device for speculative investors.
2. Data and Summary Statistics
2.1 Data
We use institutional investor holdings data from the Thomson Financial CDA/Spectrum database of SEC
13F filings to create Stock Duration.9 Stock returns are from CRSP and accounting data is from COMPUSTAT. We
obtain the most recent consensus analyst recommendations at the end of each quarter from IBES. We focus on
9 All institutional investors with more than $100 million of securities under management are required to report their holdings to the SEC. Holdings are reported quarterly and all common stock positions greater than 10,000 shares or $200,000 must be disclosed.
10
US common stocks from December 1993 to December 2013 because IBES recommendations are not available
before November 1993. We eliminate stocks without analyst recommendations, with missing market
capitalization data, and with prices below $1. Further, we require stocks to be in CRSP for at least two years
before they are included in the sample to ensure that IPO‐related anomalies do not affect our results. To
eliminate a sample bias, we further require institutional investors to be present for two years before being
included. We do this as new institutions have short past holding durations for stocks in their portfolios by
construction. Our resulting main sample consists of 55.7% of all common stocks in CRSP (on average 2,752
stocks per year) and covers 84.3% of the CRSP market capitalization.
Using the methodology introduced in Cremers and Pareek (2015), we calculate the holding duration of
each stock for every institutional investor by calculating a weighted‐measure of buys and sells by an
institutional investor, weighted by the duration for which the stock was held. For each stock in a given
institution’s portfolio, the measure is calculated by looking back over the time period since that particular stock
has been held continuously in the portfolio. Intuitively, our variable measures how long a $1 investment in a
stock has on average been in an institution’s portfolio at a particular point in time.
The calculation of the duration for stock i that is included in the institutional portfolio j at time T‐1, for
all stocks i = 1 … I and all institutional investors j = 1 … J, is given by:
jiji
jiT
WTt jiji
tjiTji BH
HW
BH
tTDuration
,,
,1
,,
,,1,,
)1()1(
where
Bi,j = total percentage of shares of stock i bought by institution j between t = T‐W and t = T‐1 (t, T
are in quarters).
Hi,j = percentage of total shares outstanding of stock i held by institution j at time t = T‐W.
αi,j,t = percentage of total shares outstanding of stock i bought or sold by institution j between
time t‐1 and t, where αi,j,t > 0 for buys and <0 for sells.
11
We choose W = 20 quarters, as very few stocks are held continuously for longer than five years. If stock
i is not included in institutional portfolio j at time T‐1, then Durationi,j,T‐1 = 0. Our measure takes into account
tax selling and other temporary adjustments in portfolios because intermediate sells are cancelled by
immediate buybacks, with only a small effect on the duration of current holdings. The limitation of our
measure is that any round‐trip trades within a quarter are ignored, as we only observe end‐of‐quarter holdings.
Next, we compute Stock Duration at the individual stock level by averaging stock level Durationi,j,T‐1 across all
institutions currently holding the stock, using each institution’s total current holdings in the stock as weights.
2.2 Summary Statistics
Table 1, Panel A provides summary statistics across the sample. Stock Duration has a mean of 1.4 years,
and the mean of Analyst Recommendation equals 2.2, where a recommendation of 1 (5) corresponds to ‘strong
buy’ (‘strong sell’). Therefore, on average analysts are more likely to issue optimistic than pessimistic
recommendations. Table 1, Panel B reports correlations between Stock Duration, Analyst Recommendation,
and other stock characteristics. The correlation between Stock Duration and Analyst Recommendation is only
16%, showing that analyst recommendations are mostly unrelated to the presence of short‐term investors.
We employ Share Turnover as an alternative measure of trading frequency. As expected, Stock
Duration and Share Turnover are negatively correlated with a correlation of ‐40%. One advantages of Share
Turnover is that it covers all investors rather than only institutional trading, as it also includes roundtrip trades
within the quarter. However, the disadvantage of Share Turnover is that it is heavily influenced by high‐
frequency trading, especially in the second half of our sample. We think high‐frequency traders are very
different from the relatively uninformed and short‐term (but not high frequency) speculators whose trading we
try to capture. Accordingly, we present our main results using Stock Duration, and show results using Share
Turnover as robustness checks only.
12
Table 1, Panel C, presents summary statistics for the subsample of stocks that are newly added to the
Russell 2000. We include these statistics for the twelve months after an addition to the index. We only include
stocks that are added to the index ‘from below’, i.e., stocks that were not part of Russell 1000 index of large
stocks before, but whose relative increase in market cap warranted inclusion in the Russell 2000. On average
143 stocks in our sample are added to the Russell 2000 each year. Table 1, Panel C shows that these newly
included stocks have lower Stock Duration (mean of 1.1 years) and higher turnover (mean of 0.9%) compared
to stocks in the full sample. As expected, newly included stocks are also smaller and covered by fewer analysts.
3. Stock Duration, Analyst Recommendations, and Overvaluation: Main Results
Our main goal is to consider how the presence of short‐term investors in stocks with extreme analyst
recommendations is related to temporary stock price distortions. To provide direct evidence that stock return
predictability, conditional on holding duration and extreme analyst recommendations, is driven by
overreaction by short‐term speculators, we first examine event‐time cumulative abnormal returns (CARs)
around past and future analyst recommendation announcements. To this end, we create stock portfolios based
on quarterly, independent 5x5 double sorts based on Analyst Recommendation and Stock Duration.
3.1 Changes in Analyst Recommendations and Stock Duration
Before looking at the resulting returns, we plot in Figure 1 mean values of Analyst Recommendation (in
Figure 1A) and Stock Duration (in Figure 1B) of stocks in the four ‘extreme’ portfolios of the 5x5 double sorts,
from eight quarters before to eight quarters after portfolio construction. The figures show that both Analyst
Recommendation and Stock Duration are strongly mean reverting over a similar period of about two years.
Importantly, there is no significant difference between long and short holding duration across the two
extreme Analyst Recommendation quintiles in Figure 1A, and there is also no significant difference in between
buy and sell recommendations across the two extreme Stock Duration quintiles in Figure 1B. This suggests that
13
investors do not substantially change their holding durations in reaction to analyst recommendations, nor do
analysts seem to condition their recommendations based on holding durations.
Figure 1A also implies that stocks that currently have the most optimistic (pessimistic) consensus
analyst recommendations typically had recommendation announcements in the past eight quarters that grew
increasingly optimistic (pessimistic) over time, while the strong mean reversion in analyst recommendations
implies predictable downward (upward) changes in future analyst recommendations.
3.2 Results from Portfolio Sorts: Analyst Recommendation Announcement Returns
If announcements of analyst recommendations contain value‐relevant information (as suggested by
Womack, 1996; Barber et al., 2001; and Loh and Stulz, 2011), for the stocks that currently have the most
optimistic (pessimistic) consensus recommendation we would expect to find positive (negative) abnormal
returns around the days of past recommendations announcements. Further, if short‐term investors are more
likely to focus and perhaps more likely to overreact to analyst recommendations (as suggested by Froot,
Scharfstein and Stein, 1992), then we may find that such abnormal returns are stronger for stocks with short
holding duration relative to stocks with long durations, followed by stronger subsequent price reversals. Note
that as value‐relevant information contained in analyst recommendations could potentially vary across stocks
with different holding‐duration levels, evidence for overreaction does not necessarily require a stronger
reaction upon past recommendation announcements, but any systematic overreaction does require stronger
subsequent price reversals. Furthermore, given the tendency of analyst recommendations to revert back to the
mean (Figure 1A), if there is overreaction to recommendations, then we would expect significant predictable
price reversals that are concentrated around the days when analysts downgrade these stocks in the future.
Furthermore, if short‐term investors systematically overreact to analyst recommendations, then it is
possible that other investors try to step in and arbitrage away any price impact of such overreaction. As short‐
sales constraints generally constrain such arbitrage (e.g., Scheinkman and Xiong, 2003), overreactions are more
likely to have price impact for stocks with optimistic rather than pessimistic analyst recommendations, as
14
arbitrage constraints may be less binding for stocks that are likely to be undervalued compared to the stocks
that are overvalued. In that case, we would expect that price reversals around future recommendation
announcements are weaker for stocks that currently have the most pessimistic recommendations.
To test these predictions, we consider the performance of stocks in quintile portfolios created from
independent double‐sort based on current Analyst Recommendations and current Stock Duration. Specifically,
we calculate cumulative three‐day size‐adjusted abnormal stock returns (CARs) around all analyst
recommendation announcements in the previous four quarters as well as in the next four quarters for each
stock in each of the resulting 25 portfolios.10 We give equal weight to each recommendation announcement for
stocks in each portfolios.11 We then calculate the average of these mean returns in each portfolio across time,
resulting in a quarterly average of past and future CARs for each of the 25 portfolios.
The corresponding results are presented in Table 2, Panel A. In the first half of the panel, we report
CARs around previous analyst recommendation announcements (i.e., averaged over the past four quarters).
The first row and the first column report CARs for quintile portfolios sorted unconditionally based on either
current Analyst Recommendation or Stock Duration. As expected, we find that past returns are negative
(positive) for stocks with currently pessimistic (optimistic) consensus recommendations. The difference across
the first and fifth quintile analyst recommendation portfolio equals ‐3.03% and is highly statistically significant.
Next, we find that these past abnormal returns are stronger for stocks with shorter holding durations.
Specifically, stocks in quintile 1 (5) with currently the most optimistic (pessimistic) analyst recommendations
have higher (lower) past CARs around previous recommendation announcements when they are held for
10 The CAR for a stock is calculated as the sum of daily size‐adjusted abnormal returns over a three‐day window centered on the recommendation‐announcement date, where the daily size‐adjusted abnormal stock returns are calculated as the difference between a stock’s return and a value‐weighted return for a portfolio of stocks in the same size‐quintile. 11 These event‐time results equally‐weight each event (i.e., analyst recommendations), but can also be interpreted as giving more weight to stocks with more analyst recommendations. As a robustness check, we calculate abnormal returns first across all events for each stock separately, and then compute the average (i.e., equally‐weight) of these mean abnormal returns across all stocks. We find that these results are similar to the results we report.
15
shorter durations compared to when they are held for longer durations. For example, for the stocks with the
most optimistic consensus recommendations, the difference in abnormal returns around past
recommendations for stocks with short versus long durations equals 0.67% and is highly significant with a t‐stat
of 6.03. Similarly, for the most pessimistic consensus recommendations, the difference in CARs between stocks
with low and high holding durations equals 2.44%, which is also highly significant.
Having documented that investors in general—and short‐term investors in particular—react to the
announcement of analyst recommendations, we turn to considering whether we can predict any subsequent
price reversals. To link any price reversals as closely as possible to prior analyst recommendations and firm
fundamentals, we consider the predictability of abnormal stock returns around subsequent announcements of
both new analyst recommendations and quarterly earnings. If short‐term investors follow analyst
recommendations naively, i.e., without paying close attention to changes in firm fundamentals (such as, e.g.,
those contained in earnings announcements), then we should expect much of the documented reaction to past
recommendation announcements to reverse around future recommendations. This may especially be the case
as both the most optimistic and the most pessimistic analyst recommendations tend to be subsequently
adjusted in the opposite direction (see above).
Another possibility is that short‐term investors may overreact to optimistic or pessimistic
recommendations due to having limited attention in situations of significant information uncertainty. In that
case, a large proportion of any price reversal may be concentrated around future earnings announcements, as
a large proportion of uncertainty about fundamentals is likely to be resolved around such announcements (Ball
and Kothari, 1991), while even attention‐constrained investors are more likely to pay attention to salient
events such as earnings announcements. These two distinct cases are not mutually exclusive, however.
In the second half of Table 2, Panel A, we present future abnormal returns (CARs) for the first case,
namely around subsequent recommendation announcements, separately for stocks in quintile portfolios based
on stocks’ current analyst recommendations and holding durations. For the next four quarters, we find strong
16
evidence for price reversals around future analyst announcements. In the first row and column we again
present returns for stocks in quintile portfolios that are unconditionally sorted either on Analyst
Recommendation or Stock Duration, respectively. Unconditionally, we find strong stock price reversals around
future analyst recommendation announcements, which can be explained by a mean reversion of the most
optimistic and pessimistic consensus analyst recommendations. In the first row, the difference in the three‐day
CARs around these announcements for stocks with the most pessimistic versus optimistic analyst
recommendations equals 1.51%, with a highly significant t‐stat of 11.27.
In rest of the table, we confirm that these stock price reversals are driven by stocks with short holding
durations. For example, for stocks in the lowest duration quintile, the difference in future abnormal three‐day
returns between stocks with the most pessimistic versus optimistic analyst recommendations equals 2.47%,
with a highly significant t‐stat of 10.75 (the CARs are significant in each of the four subsequent quarters). The
price reversals tend to be much weaker for long‐duration stocks, for which the difference in CARs for stocks
with the most pessimistic versus optimistic consensus recommendations equals only 0.50% (t‐stat of 4.58); this
difference becomes statistically insignificant after two quarters.
For stocks with short holding durations (i.e., in bottom quintile), the CAR for stocks with the most
optimistic analyst recommendation was 1.51% for the past four quarters, and it equals a negative ‐1.36% over
the next four quarters. Similarly, for stocks with short holding durations and the most pessimistic analyst
recommendations, the CAR over the past four quarters was ‐3.12%, and the return over next four quarters
equals a positive 1.1%. Therefore, for optimistic recommendations, most of the reaction to past
recommendations is reversed around future recommendations, whereas the reversal is limited for pessimistic
recommendations. This is consistent with the hypothesis that overreaction by short‐term investors is due to
17
short‐sales constraints: overreaction to optimistic recommendations leads to overvaluation, which is more
subject to arbitrage constraints than undervaluation following overreaction to pessimistic recommendations.12
Next, we consider whether there is any price reversal around subsequent earnings announcements. In
particular, Table 2, Panel B presents average three‐day abnormal returns for stocks around earnings
announcements in the previous and next four quarters. If short‐term investors overreact to analyst
recommendations due large information uncertainty and limited attention, then we expect a significant portion
of the previous returns to be reversed around future earnings announcements, which should reduce
information uncertainty and which constitute salient events that investors are likely to pay attention to.
We find in Table 2, Panel B some but more limited evidence for this hypothesis. The magnitude of the
price reversal around earnings announcements is small compared to the returns around analyst
recommendation announcements, suggesting that only a small proportion of overvaluation is reversed around
future earnings announcement. For example, for short‐duration stocks, the difference in earnings
announcement returns for stocks with optimistic and pessimistic consensus recommendations equals ‐1.87%
over the last four quarters, and 0.42% over next four quarters.13
3.3 Results from Portfolio Sorts: Calendar‐Time Returns
In this subsection, we consider predictable price reversals of stocks conditional on current analyst
recommendations and holding durations over longer periods of time, rather than around recommendation or
earnings announcements as in the previous subsection. In particular, we calculate abnormal returns of the 25
portfolios resulting from a 5x5 independent double sort on Analyst Recommendations and Stock Duration in
12 Appendix Table A‐2, Panel A presents the same returns for portfolios sorted by Analyst Recommendations and Share Turnover. The table shows that we can find similar results. For example, for high‐turnover stocks, the difference in CARs over the next four quarters between stocks with the most pessimistic and optimistic recommendations equals 1.91% (t‐stat of 8.19), while the corresponding difference amongst low‐turnover stocks is only 0.60% (t‐stat of 4.50). 13 Earnings announcement returns for portfolios sorted based on Share Turnover and Analyst Recommendation are in Appendix Table A‐2, Panel B and are similar in magnitude to the results for Stock Duration.
18
calendar time. We also present results from unconditional quintile sorts on Stock Duration or Analyst
Recommendation only.
We sort stocks into quintiles based on Stock Duration and/or Analyst Recommendation at the end of
each quarter, calculating returns for the next four quarters. Our portfolio sorts are intended to capture any
reversal of longer‐term mispricing (i.e., over the next four quarters). We account for overlapping portfolios by
following Jegadeesh and Titman (1993), such that stocks ranked in each of the last four quarters form one‐
fourth of the portfolio. Returns from each of the four sub‐portfolios are equally weighted to calculate the
monthly portfolio returns. We report equally‐weighted and value‐weighted returns using a 5‐factor model that
includes the Fama‐French 3‐factors, Carhart’s momentum factor, and the Pastor‐Stambaugh liquidity factor.
Table 3, Panel A first reports (in the first row) monthly alphas for portfolios conditional on analyst
recommendations only. On average, stocks with pessimistic recommendations earn a higher return relative to
stocks with optimistic recommendations, but the difference is statistically significant only for equal‐weighted
portfolios (with an annualized 5‐factor alpha of 12*0.36%= 4.32% and a t‐stat of 2.93; consistent with Ertimur,
Muslu, and Zhang, 2011). However, these results are insignificant for value‐weighted 5‐factor alphas (alpha of
12*0.18%= 2.16% per year, t‐stat of 1.19).
Most importantly, we next consider whether analyst recommendations mitigate or aggravate any price
impact of speculative trading by short‐term investors. In particular, we test the joint importance of Stock
Duration and Analyst Recommendations for stock return predictability. If stocks held by short‐term investors
are more likely to be overvalued and the most optimistic analyst recommendations provide a coordination
mechanism allowing speculative investors to synchronize their trading, then the above return
reversal/predictability should be stronger conditional on a stock’s holding duration. We examine this prediction
by independently double sorting stocks into quintiles based on both Stock Duration and Analyst
Recommendation at the end of each quarter. As we observed earlier in Figure 1, both sorting variables are
strongly mean reverting over a similar period of about two years, and investors do not substantially change
19
their holding durations depending on analyst recommendations, nor do analysts seem to condition their
recommendations based on holding durations. Further, the correlation between Analyst Recommendation and
Stock Duration equals only 16% (see Table 1, Panel B), such that all 25 portfolios in the independent double
sort contain a comparable number of stocks.
Table 3, Panel A also contains the monthly future performance of stocks for these independent double
sorts. We find that stocks with the most optimistic recommendations (quintile 1) only underperform stocks
with the most pessimistic recommendations (quintile 5)—and thus appear overvalued—when held by short‐
term investors. For example, using equally‐weighted portfolios, the long‐short portfolio that buys (sells) stocks
in the most pessimistic (optimistic) recommendation quintile among stocks that are also in the short‐duration
quintile has a strongly positive abnormal return of (12*0.93%=) 11.16% per year (t‐stat of 4.76). To the
contrary, the analogous long‐short portfolio among stocks that are in the long‐duration quintile has a 5‐factor
alpha of only (12*0.01%=) 0.12% per year, which is statistically insignificant (t‐stat of 0.08). The difference of
(12*‐0.92%=) 11.04% alpha per year between the two stock‐duration groups is highly significant (t‐stat of 4.30).
The results using value‐weighting also show a large positive alpha of the long‐short portfolio that buys
(sells) the portfolio with most pessimistic (optimistic) recommendations among stocks in the short‐duration
quintile, though this alpha is driven by the future underperformance of stocks with optimistic
recommendations and short holding durations. This portfolio has an alpha of (12*‐0.73%=) ‐8.76% per year
with a t‐stat of 3.71 if value‐weighted, as compared to (12*‐0.25%=) ‐3.00% per year with a t‐stat of 1.72% if
equal‐weighted. This return difference remains statistically significant for value‐weighted 3‐factor alphas
(9.60% yearly, t‐stat of 3.04) as shown in Appendix Table A‐3, Panel A. The finding that the results for value‐
weighted portfolios are driven by negative alphas seems consistent with short‐sales constraints.
More generally, our results are asymmetric for both value‐weighted and equal‐weighted portfolios in a
way that is consistent with the limits‐to‐arbitrage argument in Scheinkman and Xiong (2003) and Shleifer and
Vishny (1997). While our predictability results are only apparent for short‐duration stocks, abnormal returns
20
across stocks with short versus long holding durations are only significantly different among stocks with
optimistic recommendations and not among stocks with pessimistic ones. For example, the long‐short portfolio
that buys (sells) stocks with long (short) holding durations among stocks with the strongest buy
recommendations has a future annualized alpha of (12*0.69%=) 8.28% with a t‐stat of 4.98 for equal‐weighted
portfolios, and of (12*0.87%=) 10.44% with a t‐stat of 3.26 for value‐weighted portfolios. However, the
analogous long‐short portfolio among stocks with the most pessimistic recommendations has an insignificant
abnormal return for both equal‐weighted and value‐weighted portfolios. Therefore, our results indicate little
evidence of undervaluation, as the return predictability conditional on Stock Duration is driven by overreaction
to optimistic recommendations leading to temporary overvaluation followed by predictable negative alphas.
In Table 3, Panel B, we find that the return predictability conditional on Stock Duration and Analyst
Recommendation is remarkably persistent over each of the next four quarters. For example, if we postpone
portfolio construction by one quarter (i.e., use the Stock Duration and Analyst Recommendation three months
prior to portfolio construction) and hold the stocks for only one quarter, the long‐short portfolio that buys
(sells) stocks with the most pessimistic (optimistic) analyst recommendations among stocks with the shortest
holding durations generates an annualized 5‐factor alpha of 14.52% (t‐stat of 3.07). Lagging by two, three, or
four quarters, respectively, gives annualized alphas of 13.56% (t‐stat of 2.82), 14.86% (t‐stat of 3.14), and
12.39% (t‐stat of 2.56), respectively.14
14 Appendix Table A‐3, Panel B presents 5‐factor alphas for portfolios formed by sorting based on Share Turnover and Analyst Recommendation. The returns for these portfolios are weaker compared to the portfolios based on Stock Duration and Analyst Recommendation (Table 3, Panel A). For example, the value‐weighted return for stocks with optimistic recommendations and high turnover is only ‐1.32% per year with a t‐stat of 0.44, compared to ‐8.76% per year (t‐stat of 3.71) for stocks with low holding duration and optimistic recommendations. The return predictability conditional on turnover is probably weaker as the years since the early 2000s have become dominated by high‐frequency traders, who are unlikely to trade on analyst recommendations. To test this argument, we divide in Appendix Table A‐4 the sample into two time‐periods (1993‐2003 versus 2004‐2013) and show that results conditional on Stock Duration are similar in both periods whereas those for turnover are significantly stronger in the first period when turnover is more likely to proxy for trading activity of fundamentals‐based investors. Results in the second period are weaker as turnover is then more likely dominated by high‐frequency trading.
21
To complement this analysis, Figure 2A provides cumulative alphas of the four ‘extreme’ portfolios in
the 5x5 independent double sorts in event time. Unsurprisingly, stocks with the most optimistic (pessimistic)
analyst forecasts previously had large positive (negative) abnormal returns. For long‐duration stocks, there is
no price reversal of this past abnormal performance, but for short‐duration stocks the figure shows a strong
reversal which is consistent with the results discussed above.
In Figure 2B, we plot the performance of two different long‐short portfolios based on Stock Duration
and Analyst Recommendation in calendar time. For both long and short holding‐duration quintiles, we
construct long‐short portfolios that are long in stocks with the most pessimistic recommendations (fifth
quintile) and short in stocks with the most optimistic recommendations (first quintile). The figure shows that
one dollar invested in the long‐short portfolio with long holding durations significantly outperforms the long‐
short portfolio with short holding durations over the period 1993 to 2013.
4. Exploiting Exogenous Variation: Evidence from Russell 2000 Inclusions
4.1 Russell 2000 Inclusions as an Exogenous Shock
A potential endogeneity concern with our analysis is that some unobservable variables (e.g.,
information or news) could affect overvaluation through the decision by short‐term investors to invest in
certain stocks (‘selection effect’). Similarly, the same unobservable variables could also affect analysts and their
recommendations. In other words, short‐term investors and analysts may simply be bystanders, rather than
variables that directly cause any of the overvaluation and price reversals that we documented.
To mitigate this concern and to corroborate that the documented return predictability is caused by
short‐term investor overreaction due to their coordination through extreme analyst recommendations, we
exploit annual Russell 2000 index reconstitutions as exogenous shocks to analyst coverage and institutional
ownership (and, as a result, to Stock Duration). We focus on index inclusions as these are events that do not
provide new information to the market. In fact, membership in the Russell 2000 is quite predictable, as it is
22
based only on a firm’s market cap rank on May 31 each year: firms whose market cap ranks between 1,000 and
3,001 are included in the index, and whether stocks are just above or below these cutoffs is likely to be random
(e.g., Chang, Hong, and Liskovich, 2015). As motivated below, we focus on stocks newly added to the Russell
2000 ‘from below’, i.e., stocks that were previously not in the Russell 1000 but whose strong recent abnormal
performance and increase in market value lead to Russell 2000 inclusions. Stocks added to the Russell 2000
enter our sample at the end of June when they are first included. We keep these stocks in our tests for four
quarters (i.e., from the second quarter of a given year to the first quarter of the next year).
Figure 3 shows that, in general, extremes in analyst recommendation and holding durations are
similarly mean reverting in the Russell 2000 subsample as in the full sample (see Figure 1). There is one
exception though, namely that stocks that have long holding durations at the time they are added to the index
exhibit no subsequent mean reversion in their holding durations. This suggests that these stocks are
significantly less impacted by index inclusion than stocks with low holding durations.
We verify this difference in Figure 4, where we show that Russell 2000 reconstitutions ‘from below’
generate substantial changes in institutional ownership and analyst coverage for short‐duration stocks (Figure
4A), but not for long‐duration stocks (Figure 4B). In particular, firms with low holding durations experience a
sharp increase in institutional ownership from a pre‐inclusion average of around 20% to over 40% soon after
inclusion. Moreover, the number of analysts covering these stocks triples; it increases from about one analyst
before to over three analysts afterwards. There are no such large changes for stocks added to the index that, at
that time, have had long holding durations. As a result, the large increases in institutional ownership and
analyst coverage constitute an important and (arguably) largely exogenous shock for stocks with mostly short‐
term institutional ownership, which diminishes the role of these short‐term speculators after index inclusion.
As a result, Stock Duration increases substantially after index inclusion, from about two quarters before to
about four quarters thereafter.
23
Figure 5 shows that we cannot find significant changes in institutional ownership, analyst coverage, or
Stock Duration for stocks added to Russell 2000 ‘from above’ (stocks that were part of the Russell 1000).
Therefore, we do not include such stocks in our main analysis as we have no evidence for significant changes in
the information environment or holding durations. Furthermore, the majority of inclusions into the Russell
2000 index are from below due to market cap increases (rather than from above due to drops in market cap).15
4.2 Russell 2000 Inclusions: Results from Portfolio Sorts
Table 4 shows that our result of price reversals for stocks with short holding durations and extreme
analyst recommendations in Table 3 continue to hold when we only consider stocks added from below to the
Russell 2000. Specifically, we find strong evidence for a positive relation between Stock Duration and future
returns, supporting the hypothesis that stocks held by short‐term investors are more likely to be overvalued at
the time of the index inclusion. For example, as shown in the second and third columns of Table 4, Panel A, a
long‐short equal‐weighted portfolio with long (short) positions in stocks in the highest (lowest) duration
quintile earns a 3‐factor alpha of (12*1.16%=) 13.92% per year (t‐stat of 4.85), and a 5‐factor alpha of
(12*1.07%=) 12.84% (t‐stat of 4.52). The magnitude and significance of the abnormal returns is similar for
value‐weighted portfolios; for example, the long‐short 5‐factor alpha equals 1.25% per month (t‐stat of 4.17).16
These findings are strongly asymmetric, as all results are driven by overvalued stocks with short durations, with
little evidence of undervaluation of stocks with long durations; this is again consistent with limits to arbitrage.
We next consider in Table 4, Panel B whether analyst recommendations mitigate or aggravate any
price impact of speculative trading by short‐term investors before Russell 2000 inclusions. Similar to our results
in the full sample, we find that the return reversal for Stock Duration is stronger conditional on extreme analyst
15 Nevertheless, we show in Appendix Table A‐5 that our results are robust when we use stocks added to Russell 2000 from both above and below. 16 In Appendix Table A‐6 we consider for robustness Share Turnover as an alternative measure of trading frequency and find results that are similar. Most notably, we find that stocks with high turnover strongly underperform the stocks with low turnover after Russell 2000 inclusion. For example, as shown in the last column of Appendix Table A‐6, Panel A, a long‐short value‐weighted portfolio with long positions in stocks in the lowest turnover quintile and short positions in stocks in the highest turnover quintile earns a 5‐factor alpha of (12*0.85%=) 10.20% per year (t‐stat of 2.44).
24
recommendations, which seem to exacerbate misvaluation due to the presence of short‐term investors. For
example, in Table 4, Panel B the value‐weighted 5‐factor alpha for stocks in the most optimistic analyst
recommendation quintile and the lowest duration group equals (12*‐1,45%=) ‐17.4% per year (t‐stat of 4.17),
whereas the 5‐factor alpha for stocks with similarly optimistic analyst recommendations but long holding
durations is positive (3.84% per year), with a t‐stat of 1.06. The difference of 21.24% alpha per year between
the two stock‐duration groups is highly significant (t‐stat of 4.05).
To be in line with the results for the full sample and the theoretical predictions from Scheinkman and
Xiong (2003), we expect weaker return predictability conditional on pessimistic recommendations and short
holding durations in our Russell 2000 sample. Indeed, we find that the portfolio with short‐duration stocks
(tercile 1) and the most pessimistic analyst recommendations (quintile 5) has an insignificant future value‐
weighted 5‐factor alpha of (12*‐0.28%=) 3.36% per year (t‐stat of 0.64), while the value‐weighted portfolio
with long‐duration stocks (tercile 3) and similarly pessimistic analyst recommendations (quintile 5) has an alpha
of (12*0.15%=) 1.80% per year that is statistically insignificant (t‐stat of 0.69). The difference in annualized
alphas of 5.16% is also insignificant (t‐stat of 0.89).
4.3 Russell 2000 Inclusions: Results from Analyst Recommendation Announcement Returns
The previous subsection shows that short‐duration stocks added to Russell 2000 are overvalued and
this overvaluation is exacerbated in the presence of optimistic consensus recommendations. To provide direct
evidence that this negative return predictability—conditional on short holding duration and optimistic
recommendations—is driven by overreaction to analysts, we examine the reaction of short‐term investors to
analyst recommendation and earnings announcements in the four quarters before and after portfolio
formation. If overvaluation is exacerbated in presence of optimistic analysts, similar to results for the full
sample in Table 2, we expect short‐term investors to overreact to past recommendations which are optimistic.
We further expect this overreaction to correct when there is a reversal in recommendations in the future.
25
To test this prediction in the Russell 2000 sample, we consider the performance of stocks in portfolios
created by independently double sorting stocks into 3x5 portfolios based on Stock Duration terciles and Analyst
Recommendation quintiles. As before, we calculate three‐day CARs around all analyst recommendation
announcements in the previous (next) four quarters for each stock in each of the resulting 15 portfolios.17
The results are presented in Table 5, Panel A. In the first half of the panel, we find that past abnormal
returns are stronger for stocks with short holding durations. Specifically, stocks with currently the most
optimistic analysts have higher past CARs around previous recommendations when they are held for shorter
durations compared to when they are held for longer durations. For example, for the stocks with the most
optimistic recommendations, the difference in CARs for stocks with short versus long durations equals 1.87%
and is highly significant with a t‐stat of 2.50. For the most pessimistic recommendations, the difference in CARs
between stocks with low and high holding durations is insignificant.
We next consider whether we can predict any subsequent price reversals around future
recommendation announcements. In the second half of Table 5, Panel A, we present future CARs around
subsequent recommendation announcements, for stocks in portfolios formed conditional on current Analyst
Recommendation and Stock Duration. We confirm that stock price reversals are driven by stocks held by
investors with short holding durations and, as predicted, reversals are strongest for stocks with short holding
duration and optimistic recommendations. For example, for stocks in the lowest duration tercile, the difference
in future CARs across stocks with the most pessimistic versus optimistic analyst recommendations equals
2.26% with a highly significant t‐stat of 5.24. Price reversals tend to be much weaker for stocks with long
17 As before, we give equal weight to each recommendation announcement for stocks in each portfolio. We then calculate the average of these mean abnormal returns in each portfolio across time, resulting in a quarterly average of past and future recommendation announcement returns for each of the 15 portfolios. As discussed above, these event‐time results can be interpreted as giving more weight to stocks with more analyst recommendations. Results calculating abnormal returns first across all events for each stock separately, and then computing the average (i.e., equally‐weight) of these mean abnormal returns across all stocks (available upon request) are again similar to the ones we report.
26
holding durations, for which the difference in CARs for stocks with the most pessimistic versus optimistic
recommendations equals only ‐0.07% with a t‐stat of 0.08.18
Finally, we consider whether there is any price reversal around subsequent earnings announcements.
Similar to the results for recommendation announcement, we find in Table 5, Panel B some (but weaker)
evidence of overreaction and reversal for stocks with the lowest holding duration and the most optimistic
analysts. For these stocks, the average three‐day CAR around earnings announcements in past four quarters is
1.60% (t‐stat of 6.31), and it is ‐0.57% (t‐stat of 3.53) over next four quarters. We cannot find evidence of
return reversals around earnings announcements for stocks with long durations or pessimistic analysts.
4.4 Russell 2000 Inclusions: Results for Stocks around the Market Cap Cutoff
One potential concern with examining stocks recently added to the Russell 2000 from below is that
these stocks tend to be different compared to rest of the stocks in our sample. For example, Table 1 shows that
newly included firms are smaller (median market cap of $235 million versus $477 million for the full sample)
and have significantly outperformed over the past year (median returns of 34.7% in the inclusion sample, see
Panel C, versus 7.3% in the full sample, see Panel A). To address this concern, we compare the returns of the
hundred smallest stocks that were added to the Russell 2000 from below (and thus just made the cutoff), with
the returns of the hundred largest stocks just outside of the Russell 2000 and 1000 (i.e., stocks that just missed
the cutoff and thus were not added to the index). The comparison of these two groups is useful as their market
cap and past returns are more similar.
We first show that the hundred stocks that just made it into the index experienced substantial changes
to their information environment, compared to similar firms that just missed the inclusion. Figure 6 shows that,
four quarters before inclusion, institutional ownership in the hundred stocks that were subsequently added to
the index is very similar to the one of the stocks that later just missed the cutoff. However, four quarters after
18 Appendix Table A‐6, Panel C reports for robustness results using Share Turnover instead of Stock Duration.
27
the inclusion, their institutional ownership is substantially higher. We estimate the statistical significance of
these changes using a difference‐in‐difference (DiD) approach and calculate the change in institutional
ownership from four quarters before to four quarters after the index inclusion quarter, for the two groups of
just included and non‐included stocks, with results presented in in Table 6, Panel A. For the hundred smallest
included stocks, we find that institutional ownership increases by 9.2%, while for stocks that just missed the
cutoff, institutional ownership increases by only 5.45%. The difference in these changes is significant with a t‐
stat of 6.15. We find similar results for analyst coverage where the DiD equals 0.34 (t‐stat of 4.82), which
indicates that stocks that were ‘just added’ to the index see a greater increase in analyst coverage. Finally, for
Stock Duration, the DiD estimate equals ‐0.11 years (t‐stat of 3.72), implying that four quarters after the
inclusion, stocks added to the Russell 2000 are held by more new institutions.
We next compare returns of both sets of firms. In Table 5, we documented negative future returns
after index inclusion for stocks with short holding durations. To show that this result is driven by exogenous
changes in holding duration after firms are added to the Russell 2000, we compare the return predictability for
‘just included’ stocks with that for ‘just not‐included’ stocks. If the price reversal is driven by overly optimistic
short‐term investors becoming less important as more long‐term institutional investors buy stocks because
they are added to the index, then we should expect the return reversals to be stronger for stocks that were just
added to the index. On the other hand, return reversals should be weaker for stocks that just missed the index.
The corresponding results are presented in Table 6, Panel B. The comparison that is of most relevance
for our analysis is the future return predictability conditional on Stock Duration, for which the results confirm
the above prediction of the role of holding durations. For example, stocks with a short holding duration earn a
negative value‐weighted 5‐factor alpha of (12*‐1.02%=) ‐12.24% in the year after the index inclusion, which is
highly significant with a t‐stat of 3.06. In contrast, short‐duration stocks that just missed the Russell 2000 cutoff
do not experience any return reversals (with an insignificant 5‐factor alpha of only 12*‐0.02%= ‐0.24% per year;
the return difference is highly significant with a t‐stat of 2.11). This provides additional evidence supporting the
28
prediction that stocks owned by short‐term investors are more likely to be overvalued, where the
overvaluation declines as their importance diminishes.
5. Limited Arbitrage and Robustness
5.1 Role of Limits to Arbitrage
In the previous two sections, we present evidence of strong price reversals for stocks with short
holding durations and extreme analyst recommendations, suggestive of temporary distortions in stock prices. If
so, it is important to uncover what frictions prevent arbitrageurs from trading these away, or to show evidence
of limits to arbitrage as in Shleifer and Vishny (1990). We consider three stock characteristics that previous
literature has identified as being associated with greater limitations to arbitrage: (i) higher short‐interest ratios;
(ii) lower ownership by a major stock lender for small cap stocks, namely Dimensional Fund Advisor (DFA); and
(iii) higher idiosyncratic volatility. The first two characteristics relate to stocks that are more likely to be subject
to short‐sales constraints (e.g., Asquith, Pathak, and Ritter, 2005; Nagel, 2005), while the third one more
generally relates to stocks that are more difficult to trade, implying also greater limits to arbitrage (e.g., Shleifer
and Vishny, 1997).19 We form 5x5x2 independent triple sorts on Stock Duration, Analyst Recommendation, and
the respective proxy for arbitrage costs, with the results reported in Table 7. We use in Panel A the short‐
interest ratio, in Panel B DFA ownership, and in Panel C idiosyncratic volatility as proxies for arbitrage costs. For
all three panels, we report monthly value‐weighted 5‐factor alphas.
Table 7 shows that the patterns documented in Table 3, Panel A, namely that stocks with extreme
analyst recommendations exhibit significant price reversals only if held by short‐term investors, are driven by
the subset of stocks with high arbitrage costs. For example, we find in Table 7, Panel A that the value‐weighted
annualized 5‐factor alpha for stocks with optimistic recommendations that are also held by short‐term
19 The short‐interest ratio captures the shorting demand: the higher the demand, the more expensive shorting is likely to be. Ownership by DFA captures the supply of shares available for shorting. The lower the ownership by DFA, the more expensive shorting is likely to be, particularly for small cap stocks.
29
investors equals (12*‐0.85%=) ‐10.16% (t‐stat of 3.65) if the short‐interest ratio is high. However, the analogous
5‐factor alpha is neither statistically nor economically significant if the short‐interest ratio is low (12*‐0.18%= ‐
2.20% per year, t‐stat of 0.77). The difference in alphas of ‐7.97% is significant with a t‐stat of 2.39. Note that
shorting costs as proxied by the short‐interest ratio do not seem to matter for long‐duration stocks. Consistent
with limited arbitrage and the short‐interest ratio being a proxy for shorting cost, conditioning on the short‐
interest ratio hence matters mostly on the short‐side or for negative return predictability.
We find similar results when we use the DFA ownership. In Table 7, Panel B, the value‐weighted 5‐
factor alpha for stocks with optimistic recommendations that are held by short‐term investors equals (12*‐
0.81%=) ‐9.67% per year (t‐stat of 3.56) if DFA ownership is low. The 5‐factor alpha is significantly lower in
magnitude if DFA ownership is high (12*‐0.35%=‐4.21%, t‐stat of 2.12). The difference in alphas of 5.46% per
year is again significant (t‐stat of 1.98). Finally, in Table 7, Panel C we find that the return predictability is
stronger for high idiosyncratic risk stocks, and especially so on the long‐side or for positive abnormal returns,
compared to the previous two proxies of arbitrage costs. For example, the value‐weighted annualized alpha for
stocks with the most pessimistic recommendations that are also held by short‐term investors equals (12*0.63
%=) 7.60% (t‐stat of 2.29) if idiosyncratic risk is high. However, the alpha is neither statistically nor economically
significant among stocks with low idiosyncratic volatility (2.23% per year, t‐stat of 0.80).
5.2 Fama‐MacBeth Regressions
As an additional robustness test, we use the Fama‐MacBeth (1973) methodology and estimate
predictive cross‐sectional regressions of next one‐year raw returns on Stock Duration and Analyst
Recommendation while controlling for other stock characteristics. We present the results for the full sample in
Table 8, Panel A, and the corresponding results for the Russell 2000 inclusion sample in Table 8, Panel B.
We find estimates that are generally consistent with the results in Table 3. For example in Table 8,
Panel A, the coefficient on Stock Duration is positive and highly significant in most of the specifications. In
30
column 1, it equals 3.3%, with a t‐stat of 3.05, which confirms the results in Table 3, Panel A, namely that short‐
duration stocks earn lower returns compared to long‐duration stocks. In column 2, we replace the continuous
measure with dummy variables corresponding to top and bottom holding‐duration quintiles. We find that the
coefficient corresponding to the short (or bottom) duration quintile is negative and highly significant (t‐stat of
2.74), whereas the coefficient of the longest (or top) quintile is positive but insignificant (t‐stat of 0.88). This
confirms that among short‐duration stocks, results are driven by overvaluation rather than undervaluation,
consistent with the presence of short‐sales constraints.
Looking at the interplay of holding durations and analyst recommendations, in column 3 we find that
the coefficient of the interaction between Stock Duration and Analyst Recommendation is negative and highly
statistically significant (t‐stat of 4.19). This confirms that return predictability based on analyst
recommendations is stronger for stocks with short holding durations (as shown in Table 3, Panel A). In column
4, using the extreme‐duration dummies, we confirm that returns conditional on analyst recommendations are
significant only for short‐duration stocks. Column 5 also includes interaction terms of Stock Duration with
analyst forecast dispersion and analyst coverage. The interaction between Analyst Recommendation and Stock
Duration remains significant once these other interactions are added, whereas the two additional interactions
terms have insignificant coefficients. Finally, in column 6 we include turnover as an alternative proxy for trading
frequency. We find that the interaction between analyst recommendation and turnover is positive and
significant, consistent with the results in Appendix Table A‐2, Panel B.
We verify the robustness of our results for Russell 2000 inclusion stocks in Table, 8 Panel B. In columns
1 and 2, we estimate regressions over the full sample and include a dummy variable that equals one if a stock
has been added to Russell 2000 from below in the current quarter or in any of the previous three quarters. We
also include the interaction of this dummy with Stock Duration, as well as other stock characteristics. In column
1, the interaction term between Stock Duration and the Russell 2000 dummy equals 3.6% and is significant at
the 10% level. The coefficient on Stock Duration is 3.0% with a t‐stat of 3.23. Therefore, the coefficient on Stock
31
Duration is approximately two‐times larger (0.030+0.036=0.066) conditional on a recent Russell 2000 inclusion;
this shows that return reversals due to holding‐duration reversals are even stronger after Russell 2000
inclusions. This is consistent with the hypothesis that these events provide an exogenous shock to the
institutional‐investor base, leading to less speculative and more long‐term investors holding the stock. In
column 2, we find similar results conditional on share turnover. The interaction term between turnover and the
inclusion dummy is negative and highly significant with a t‐stat of 4.67. Therefore, return reversals conditional
on high turnover are stronger for stocks recently added to the Russell 2000.
In columns 3 to 5, we focus only on the Russell 2000 subsample and show results analogous to the
portfolios results in Table 4. The regression estimates are generally consistent with our portfolio results. For
example, in column 3, the coefficient of Stock Duration is 5.1% and significant with a t‐stat of 1.97. This
confirms that short‐duration stocks earn lower returns post Russell 2000 inclusion compared to long‐duration
stocks. In column 4, the interaction between Stock Duration and Analyst Recommendation is negative and
highly statistically significant (t‐stat of 2.26). This confirms that return predictability based on Stock Duration is
stronger for stocks with optimistic recommendations. Finally, when we include in column 5 share turnover as
an alternative proxy, we find that the interaction term between analyst recommendation and turnover is
positive as predicted, though we note that it is insignificant with a t‐stat of only 1.01.
6. Conclusion
We document novel evidence that the presence of short‐term investors, when combined with extreme
consensus analyst recommendations, is strongly associated with future price reversals. In particular, stocks
held by short‐term institutional investors with the most optimistic analyst recommendations tend to be
overvalued: these stocks generally have large positive past abnormal returns, which are followed by large
negative subsequent abnormal stock returns. This finding is consistent with and expands the results of Ertimur,
Muslu, and Zhang (2011). We further show that our economically large abnormal returns are driven by stocks
that are harder to arbitrage, such as stocks with higher short‐interest ratios and higher idiosyncratic volatility.
32
Our main interpretation of these results is that extreme analyst recommendations serve as a coordination
device for short‐term traders (Froot, Scharfstein, and Stein, 1992), assisting herding among speculators and
exacerbating their collective price effect. This interpretation is also broadly consistent with both Shiller (2002,
2003) and the theory of Hong, Scheinkman, and Xiong (2008).
Our main identification to corroborate that short‐term traders indeed cause the documented
misvaluation is obtained from stocks added ‘from below’ to the Russell 2000 index. This identification strategy
has the advantage that it allows us to rule out other explanations, for example that short‐term traders are
trading on the misvaluation or that the documented price reversals are driven by news events that attract
short‐term investors (or other selection effects). Stocks added to the Russell 2000 have generally performed
very well in the past year, such that their significant increase in market capitalization now newly qualifies them
for inclusion in this popular small cap benchmark. If such newly included stocks have short holding durations,
i.e., have institutional owners who are generally short‐term, we find that the index inclusion causes substantial
changes to analyst coverage, institutional ownership, and holding durations. As argued by, for example, Chang,
Hong, and Liskovich (2015), these are changes that are arguably mostly exogenous to firms and cause a
diminished role of short‐term speculators.
We find that stocks with strongly optimistic analyst recommendations and short holding durations
appear to be significantly overvalued at the time when they are added to the Russell 2000. After their index
inclusion, this overvaluation slowly reverses over the subsequent two years, as the importance of short‐term
investors in these stocks declines due to substantial increases in institutional ownership and analyst coverage.
Our results are remarkably large economically. For example, the value‐weighted 5‐factor alpha for newly‐
added stocks with the most optimistic analyst recommendations and the shortest holding durations equals a
negative ‐17.4% per year (t‐stat of 4.17) in the year after they are first included. Our study thus suggests that
short‐term traders seem to coordinate around extreme analyst recommendations causing significant
overvaluation.
33
References
Asquith, Paul, Parag A. Pathak, and Jay R. Ritter, 2005, Short interest, institutional ownership, and stock returns, Journal of Financial Economics 78, 243–276.
Ball, Ray, and S. P. Kothari, 1991, Security returns around earnings announcements, The Accounting Review 66, 718–738.
Barber, Brad, Reuven Lehavy, Maureen McNichols, and Brett Trueman, 2001, Can investors profit from the prophets? Security analyst recommendations and stock returns, Journal of Finance 56, 531–563.
Bartov, Eli, Suresh Radhakrishnan, and Itzhak Krinsky, 2000, Investor sophistication and patterns in stock returns after earnings announcements, The Accounting Review 75, 43–63.
Balakrishnan, Karthik, Catherine Schrand, and Rahul Vashishtha, 2014, Public information, traders’ beliefs, and rational speculation, Working paper, London Business School.
Boehmer, Ekkehart, and Eric K. Kelley, 2009, Institutional investors and the informational efficiency of prices, Review of Financial Studies 22, 3563–3594.
Brennan, Michael J., Narasimhan Jegadeesh, and Bhaskaran Swaminathan, 1993, Investment analysis and the adjustment of stock prices to common information, Review of Financial Studies 6, 799–824.
Brown, Nerissa C., Kelsey D. Wei, and Russ Wermers, 2014, Analyst recommendations, mutual fund herding, and overreaction in stock prices, Management Science 60, 1–20.
Busse, Jeffrey, T. Clifton Green, and Narasimhan Jegadees, 2012, Buy‐side trades and sell‐side recommendations: Interactions and information content, Journal of Financial Markets 15, 207–232.
Cella, Cristina, Andrew Ellul, and Mariassunta Giannetti, 2013, Investors' horizons and the amplification of market shocks, Review of Financial Studies 26, 1607–1648.
Chang, Yen‐Cheng, Harrison G. Hong, and Inessa Liskovich, 2015, Regression discontinuity and the price effects of stock market indexing, Review of Financial Studies 28, 212–246.
Chen, Honghui, Gregory Noronha, and Vijay Singal, 2004, The price response to S&P 500 index additions and deletions: Evidence of asymmetry and a new explanation, Journal of Finance 59, 1901–1930.
Chen, Joseph, Harrison Hong, and Jeremy C. Stein, 2002, Breadth of ownership and stock returns, Journal of Financial Economics 66, 171–205.
Collins, Daniel W., Guojin Gong, and Paul Hribar, 2003, Investor sophistication and the mispricing of accruals, Review of Accounting Studies 8, 251–276.
Coval, Joshua, and Erik Stafford, 2007, Asset fire sales (and purchases) in equity markets, Journal of Financial Economics 86, 479–512.
Cremers, K. J. Martijn, Antti Petajisto, and Eric Zitzewitz, 2012, Should benchmark indices have alpha? Revisiting performance evaluation, Critical Finance Review 2, 1–48.
Cremers, K. J. Martijn, and Ankur Pareek, 2015, Short‐term trading and stock return anomalies: Momentum, reversal, and share issuance, Review of Finance 19 1649–1701.
34
Daniel, Kent, David Hirshleifer, and Avanidhar Subrahnanyam, 1998, Investor psychology and security market under‐ and overreactions, Journal of Finance 53, 18–39.
DeBondt, Werner F.M., and Richard Thaler, 1990, Do security analysts overreact? American Economic Review Papers and Proceedings 80, 52–57.
Diether, Karl B., Christopher J. Malloy, and Anna Scherbina, 2002, Differences of opinion and the cross section of stock returns, Journal of Finance 57, 2113–2141.
Duffie, Darrell, 2010, Presidential address: Asset price dynamics with slow‐moving capital, Journal of Finance 65, 1237–1267.
Edmans, Alex, Itay Goldstein, and Wei Jiang, 2012, The real effects of financial markets: The impact of prices on takeovers, Journal of Finance 67, 933–971.
Ertimur, Yonca, Volkan Muslu, and Frank Zhang, 2011, Why are recommendations optimistic? Evidence from analysts’ coverage initiations, Review of Accounting Studies 16, 679–718.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of Political Economy 81, 607–636.
Froot, Kenneth, David S. Scharfstein, and Jeremy Stein, 1992, Herd on the street: Informational inefficiencies in a market with short‐term speculation, Journal of Finance 47, 1461–1484.
Greenwood, Robin, 2005, Short‐ and long‐term demand curves for stocks: Theory and evidence on the dynamics of arbitrage, Journal of Financial Economics 75, 607–649.
Harris, Lawrence, and Eitan Gurel, 1986, Price and volume effects associated with changes in the S&P 500 list: New evidence for the existence of price pressures, Journal of Finance 41, 815–829.
Harrison, J. Michael, and David Kreps, 1978, Speculative investor behavior in a stock market with heterogeneous expectations, Quarterly Journal of Economics 92, 323–336.
Hong, Harrison, Jose Scheinkman, and Wei Xiong, 2008, Advisors and asset prices: A model of the origins of bubbles, Journal of Financial Economics 89, 268–287.
Hong, Harrison, and Jeremy C Stein, 2007, Disagreement and the stock market, Journal of Economic Perspectives 21, 109–128.
Hou, Kewei, Wei Xiong, and Lin Peng, 2009, A tale of two anomalies: The implications of investor attention for price and earnings momentum, Working Paper.
Jegadeesh, Narasimhan, Joonghyuk Kim, Susan D. Krische, and Charles M. C. Lee, 2004, Analyzing the analysts: When do recommendations add value? Journal of Finance 59, 1083–1124.
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implications for stock market efficiency, Journal of Finance 48, 65–91.
Ke, Bin and Santhosh Ramalingegowda, 2005, Do institutional investors exploit the post‐earnings announcement drift? Journal of Accounting and Economics 39, 25–53.
Loh, Roger K., and René M. Stulz, 2011, When are analyst recommendation changes influential? Review of Financial Studies 24, 593–627.
35
Lynch, Anthony W., and Richard R. Mendenhall, 1997, New evidence on stock price effects associated with changes in the S&P 500 Index, Journal of Business 70, 351–83.
Mikhail, Michael B., Beverly R. Walther, and Richard H. Willis, 2007, When security analysts talk, who listens? The Accounting Review 82, 1227–1253.
Mitchell, Mark, Lasse Pedersen, and Todd Pulvino, 2007, Slow moving capital, American Economic Review 97, 215–220.
Nagel, Stefan, 2005, Short sales, institutional investors and the cross‐section of stock returns, Journal of Financial Economics 78, 277–309.
Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi‐definite, heteroskedasticity and autocorrelation consistent covariance matrix, Econometrica 55, 703–708.
Nofsinger, John R., and Richard W. Sias, 1999, Herding and feedback trading by institutional and individual investors, Journal of Finance 54, 2263–2295.
Ofek, Eli, and Matthew Richardson, 2003, DotCom mania: The rise and fall of internet stock prices, Journal of Finance 58, 1113–1138.
Petajisto, Antti, 2011, The index premium and its hidden cost for index funds, Journal of Empirical Finance 18, 271‐288.
Pontiff, Jeffrey, 1996, Costly arbitrage: Evidence from closed‐end funds, Quarterly Journal of Economics 111, 1135–1151.
Sias, Richard W., Laura T. Starks, and Sheridan Titman, 2006, Changes in institutional ownership and stock returns: Assessment and methodology, Journal of Business 79, 2869–2910.
Scheinkman, Jose, and Wei Xiong, 2003, Overconfidence and speculative bubbles, Journal of Political Economy 111, 1183–1219.
Shiller, Robert J., 2002, Bubbles, human judgment, and expert opinion, Financial Analysts Journal 58, 18–26.
Shiller, Robert J., 2003, From efficient market theory to behavioral finance, Journal of Economic Perspectives 17, 83–104.
Shleifer, Andrei, 1986, Do demand curves for stocks slope down? Journal of Finance 41, 579–590.
Shleifer, Andrei, and Robert W. Vishny, 1990, Equilibrium short horizons of investors and firms, American Economic Review P&P 80, 148–153.
Shleifer, Andrei, and Robert W. Vishny, 1997, The limits of arbitrage, Journal of Finance 52, 35–55.
Wermers, Russ, 1999, Mutual fund herding and the impact on stock prices, Journal of Finance 54, 581–622.
Womack, Kent L., 1996, Do brokerage analysts’ recommendations have investment value? Journal of Finance 51, 137–167.
Table 1: Summary Statistics
This table provides summary statistics. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. Panel A reports summary statistics across the full sample and Panel B reports Spearman rank correlations. Panel C reports summary statistics only for stocks that are added to the Russell 2000 from below. Variables are defined in Appendix Table A‐1.
Panel A: Full Sample
Variable Mean Std. Dev. 25% Median 75%
Stock Duration (years) 1.4 0.6 1.0 1.4 1.8Share Turnover (%) 0.7 0.8 0.3 0.5 0.9Analyst Recommendation 2.2 0.6 1.9 2.2 2.7Market Cap ($ million) 3741 15552 153 477 1825
MB Ratio 5.3 55.0 1.4 2.2 3.8Past 12 Months Return (%) 17.9 67.8 ‐16.1 7.3 34.8Institutional Ownership (%) 53.2 24.3 34.4 56.1 73.2Analyst Coverage 7.1 6.5 2.3 5.0 10.0
Idiosyncratic Risk (%) 2.7 1.6 1.6 2.3 3.4Short Ratio 0.03 0.04 0.01 0.02 0.04# Stocks 2752
Panel B: Rank Correlations
(1) (2) (3) (4) (5) (6) (7) (8) (9)
(1) Stock Duration 1.00 (2) Share Turnover ‐0.40 1.00 (3) Analyst Recommendation 0.16 ‐0.08 1.00 (4) Idiosyncratic Risk ‐0.36 0.27 ‐0.06 1.00
(5) Analyst Coverage 0.03 0.41 0.05 ‐0.33 1.00 (6) Market Cap 0.19 0.25 0.01 ‐0.56 0.77 1.00 (7) MB Ratio ‐0.14 0.26 ‐0.30 ‐0.05 0.28 0.39 1.00 (8) Past 12 Months Return ‐0.03 0.00 ‐0.20 ‐0.17 0.03 0.23 0.40 1.00(9) Institutional Ownership 0.05 0.42 ‐0.09 ‐0.25 0.52 0.54 0.17 0.11 1.00
Panel C: Russell 2000 Inclusion Stocks Sample
Variable Mean Std. Dev. 25% Median 75%
Stock Duration (years) 1.1 0.6 0.6 1.0 1.4Share Turnover (%) 0.9 1.0 0.3 0.6 1.1Analyst Recommendation 2.0 0.6 1.5 1.9 2.4Market Cap ($ million) 324 278 171 235 362
MB Ratio 12.4 67.8 1.9 3.2 6.3Past 12 Months Return (%) 63.8 114.9 1.3 34.7 91.0Institutional Ownership (%) 48.6 21.8 31.4 48.8 65.0Analyst Coverage 3.8 2.5 1.9 3.2 5.0
Idiosyncratic Risk (%) 3.2 1.4 2.3 3.0 3.8Short Ratio 0.04 0.05 0.01 0.02 0.04# Stocks 143
37
Table 2: Cumulative Abnormal Returns around Analyst Recommendation and Earnings Announcements
Panel A reports average three‐day cumulative abnormal stock returns around analyst recommendation announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. Panel B reports average three‐day cumulative abnormal stock returns around earnings announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. We first independently double sort stocks into quintiles at the end of each quarter based on Analyst Recommendation and Stock Duration. We then calculate average three‐day size‐adjusted cumulative abnormal stock returns (CARs) around all analyst recommendation announcements/earnings announcements in the previous four quarter (the next four quarters) for each stock in each of these 25 portfolios. We report the average of these mean CARs. We report results for the extreme Stock Duration groups only. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Cumulative Abnormal Returns (CAR) around Analyst Recommendation Announcements
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Stock Duration Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1
Uncond. 1.18 0.44 ‐0.12 ‐0.89 ‐1.85 ‐3.03 ‐0.80 ‐0.45 ‐0.05 0.27 0.70 1.51 (15.94) (11.87) (‐1.32) (‐8.76) (‐11.09) (‐16.41) (‐6.74) (‐6.24) (‐1.05) (5.14) (10.87) (11.27)
1 (Short) ‐0.43 1.51 0.58 ‐0.46 ‐2.07 ‐3.12 ‐4.63 ‐0.59 ‐1.36 ‐0.91 ‐0.28 0.16 1.10 2.47 (‐4.20) (16.80) (8.42) (‐3.89) (‐9.53) (‐9.22) (‐13.68) (‐5.17) (‐7.33) (‐6.51) (‐2.61) (1.50) (7.41) (10.75)
5 (Long) ‐0.04 0.84 0.33 0.07 ‐0.23 ‐0.68 ‐1.52 0.12 0.02 ‐0.22 0.09 0.20 0.52 0.50 (‐1.14) (10.78) (8.18) (1.70) (‐4.56) (‐8.88) (‐12.94) (3.51) (0.26) (‐4.30) (1.39) (4.08) (8.23) (4.58)
5‐1 0.39 ‐0.67 ‐0.25 0.53 1.85 2.44 3.11 0.71 1.39 0.69 0.37 0.04 ‐0.58 ‐1.97 (4.59) (‐6.03) (‐3.70) (4.77) (9.37) (7.84) (10.36) (6.94) (7.61) (4.96) (3.52) (0.36) (‐3.94) (‐8.47)
Panel B: Cumulative Abnormal Returns (CAR) around Earnings Announcements
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Stock Duration Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1(Buy) 2 3 4 5 (Sell) 5‐1
Uncond. 0.81 0.64 0.32 ‐0.02 ‐0.52 ‐1.33 0.02 0.17 0.22 0.22 0.25 0.23 (19.41) (18.45) (12.23) (‐6.32) (‐18.27) (‐28.15) (0.83) (6.65) (10.83) (8.70) (7.52) (5.54)
1 (Short) 0.31 1.03 0.86 0.31 ‐0.61 ‐0.84 ‐1.87 ‐0.09 ‐0.24 ‐0.11 0.03 ‐0.10 0.18 0.42 (7.77) (14.16) (12.38) (4.54) (‐6.42) (‐11.52) (‐17.65) (‐2.55) (‐4.57) (‐1.52) (0.42) (‐1.22) (2.80) (5.10)
5 (Long) 0.15 0.56 0.44 0.21 0.01 ‐0.19 ‐0.75 0.26 0.26 0.22 0.18 0.20 0.35 0.09 (5.16) (8.74) (10.36) (5.70) (0.23) (‐5.26) (‐11.99) (10.05) (4.48) (4.93) (4.24) (5.15) (8.81) (1.35)
5‐1 ‐0.16 ‐0.47 ‐0.42 ‐0.10 0.62 0.64 1.11 0.35 0.51 0.33 0.16 0.29 0.18 ‐0.33 (‐3.30) (‐5.13) (‐5.92) (‐1.13) (5.93) (7.97) (9.80) (8.38) (6.18) (3.46) (2.06) (3.30) (2.50) (‐2.84)
38
Table 3: Return Predictability: Portfolio Results
Panel A reports monthly equal‐weighted and value‐weighted alphas for portfolios formed by independent double sorts based on Stock Duration and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into five groups based on Stock Duration and then independently divided into five groups based on Analyst Recommendation. We then report returns for these 25 portfolios which are calculated over next four quarters. Panel B reports results once we lag the portfolio construction by one, two, three, and four quarters. We report results for the extreme groups only. To account for overlapping portfolios when calculating returns, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All reported returns are in monthly percentages. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Double Sorts on Stock Duration and Analyst Recommendation
Monthly Equal‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Stock Duration Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1
Uncond. 0.05 0.16 0.32 0.38 0.41 0.36 ‐0.09 0.13 0.14 0.14 0.09 0.18 (0.51) (1.85) (4.21) (5.21) (3.61) (2.93) (‐0.79) (2.16) (2.01) (1.94) (1.08) (1.19)
1 (Short) 0.06 ‐0.25 ‐0.03 0.31 0.47 0.67 0.93 ‐0.17 ‐0.73 ‐0.20 0.29 0.26 0.42 1.16 (0.49) (‐1.72) (‐0.19) (2.26) (2.92) (3.33) (4.76) (‐1.11) (‐3.71) (‐1.06) (1.37) (1.39) (1.79) (3.67)2 0.23 0.11 0.21 0.37 0.37 0.40 0.30 ‐0.12 ‐0.28 ‐0.07 0.09 0.09 0.17 0.45 (2.46) (0.84) (2.06) (3.43) (3.13) (2.28) (1.49) (‐1.15) (‐1.67) (‐0.44) (0.63) (0.64) (1.10) (1.95)3 0.29 0.13 0.24 0.37 0.44 0.53 0.40 0.07 0.00 0.11 0.14 0.21 0.19 0.19 (4.03) (1.09) (2.14) (4.33) (4.70) (4.19) (2.51) (0.87) (‐0.03) (0.89) (1.08) (1.86) (1.41) (0.88)4 0.33 0.40 0.29 0.34 0.36 0.36 ‐0.04 0.09 0.02 0.20 0.05 0.15 0.18 0.16 (4.96) (4.03) (2.86) (4.44) (4.28) (3.21) (‐0.33) (1.29) (0.14) (1.68) (0.53) (1.62) (1.48) (0.80)
5 (Long) 0.40 0.44 0.38 0.37 0.38 0.45 0.01 0.15 0.14 0.23 0.19 0.20 0.10 ‐0.04 (5.54) (3.95) (3.90) (4.05) (4.41) (3.73) (0.08) (2.37) (0.76) (1.99) (1.78) (1.87) (0.76) (‐0.17)
5‐1 0.34 0.69 0.40 0.06 ‐0.09 ‐0.22 ‐0.92 0.32 0.87 0.43 ‐0.11 ‐0.06 ‐0.32 ‐1.20 (2.78) (4.98) (2.55) (0.39) (‐0.50) (‐1.11) (‐4.30) (1.69) (3.26) (1.85) (‐0.43) (‐0.30) (‐1.15) (‐3.21)
39
Table 3 (continued)
Panel B: Double Sorts on Stock Duration and Analyst Recommendation: Lagged Portfolio Construction
Portfolio Construction Lag in Quarters:
One Quarter Two Quarters Three Quarters Four Quarters
Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1
1 (Short) ‐0.92 0.30 1.21 ‐0.63 0.50 1.13 ‐0.80 0.44 1.24 ‐0.59 0.45 1.03 (‐3.97) (0.91) (3.07) (‐2.52) (1.74) (2.82) (‐3.40) (1.36) (3.14) (‐2.40) (1.49) (2.56)
5 (Long) 0.32 0.23 ‐0.09 0.06 ‐0.06 ‐0.12 ‐0.08 0.18 0.26 0.29 0.11 ‐0.18 (1.36) (1.33) (‐0.31) (0.25) (‐0.36) (‐0.39) (‐0.34) (1.10) (0.90) (1.18) (0.73) (‐0.61)
5‐1 1.24 ‐0.07 ‐1.31 0.69 ‐0.56 ‐1.25 0.72 ‐0.26 ‐0.98 0.88 ‐0.34 ‐1.21 (3.75) (‐0.19) (‐2.73) (2.11) (‐1.60) (‐2.52) (2.20) (‐0.69) (‐2.12) (2.40) (‐1.02) (‐2.46)
40
Table 4: Return Predictability: Portfolio Results for Russell 2000 Inclusion Stocks
Panel A presents for stocks added to the Russell 2000 ‘from below’ monthly equal‐weighted and value‐weighted alphas for portfolio strategies from unconditional sorts based on Stock Duration. Panels B reports alphas for independent double sorts based on Stock Duration and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into three groups based on Stock Duration. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 15 portfolios which are calculated over next four quarters. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All the reported returns are in monthly percentages. The sample consists of US common stocks that are added to the Russell 2000 ‘from below’ from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Sorts on Stock Duration for Russell 2000 Inclusion Stocks
Sample: Russell 2000 Inclusion ‘From Below’ Stocks
Monthly Equal‐Weighted Returns Monthly Value‐Weighted Returns
Stock Duration CAPM Alpha 3‐Factor Alpha 5‐Factor Alpha CAPM Alpha 3‐Factor Alpha 5‐Factor Alpha
1 (Short) ‐1.00 ‐1.03 ‐0.92 ‐1.01 ‐1.01 ‐1.08 (‐3.05) (‐4.75) (‐4.45) (‐2.87) (‐4.24) (‐4.44)2 ‐0.34 ‐0.44 ‐0.33 ‐0.26 ‐0.29 ‐0.41 (‐1.19) (‐2.17) (‐1.65) (‐0.72) (‐1.02) (‐1.39)3 0.06 ‐0.11 ‐0.03 ‐0.24 ‐0.35 ‐0.40 (0.24) (‐0.61) (‐0.18) (‐0.78) (‐1.67) (‐1.93)4 0.01 ‐0.16 ‐0.04 0.06 ‐0.10 ‐0.11 (0.06) (‐0.95) (‐0.23) (0.23) (‐0.51) (‐0.56)
5 (Long) 0.39 0.13 0.16 0.48 0.24 0.18 (1.62) (0.78) (0.90) (2.00) (1.35) (0.97)
5‐1 1.38 1.16 1.07 1.50 1.25 1.25 (4.66) (4.85) (4.52) (4.22) (4.26) (4.17)
Panel B: Double Sorts on Stock Duration and Analyst Recommendation for Russell 2000 Inclusion Stocks
Sample: Russell 2000 Inclusion ‘From Below’ Stocks
Monthly Equal‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Stock Duration Uncond. 1 (Buy) 5 (Sell) 5‐1 Uncond. 1(Buy) 5(Sell) 5‐1
Uncond. ‐0.55 ‐0.18 0.37 ‐0.71 ‐0.31 0.41 (‐2.86) (‐0.96) (1.52) (‐3.30) (‐1.62) (1.46)
1 (Short) ‐0.70 ‐1.27 ‐0.20 1.07 ‐0.82 ‐1.45 ‐0.28 1.17 (‐3.87) (‐3.89) (‐0.50) (2.13) (‐3.59) (‐4.17) (‐0.64) (2.09)2 ‐0.11 ‐0.54 ‐0.46 0.08 ‐0.35 ‐0.55 ‐0.92 ‐0.37 (‐0.67) (‐1.72) (‐1.42) (0.16) (‐1.97) (‐1.79) (‐3.21) (‐0.88)
3 (Long) 0.11 0.37 0.05 ‐0.31 0.08 0.32 0.15 ‐0.17 (0.78) (1.35) (0.26) (‐0.95) (0.51) (1.06) (0.69) (‐0.44)
3‐1 0.82 1.64 0.25 ‐1.39 0.90 1.77 0.43 ‐1.33 (4.19) (4.01) (0.62) (‐2.42) (3.36) (4.05) (0.89) (‐2.04)
41
Table 5: Cumulative Abnormal Returns for Russell 2000 Inclusion Stocks
Panel A reports for stocks added to the Russell 2000 ‘from below’ average three‐day cumulative abnormal stock returns (CARs) around analyst recommendation announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. Panel B reports for stocks added to the Russell 2000 ‘from below’ average three‐day cumulative abnormal stock returns around earnings announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. We first independently double sort stocks into terciles/quintiles at the end of each quarter based on Analyst Recommendation and Stock Duration. We then calculate the average cumulative three‐day size‐adjusted abnormal stock returns (CARs) around all analyst recommendation/earnings announcements in the previous four quarter (the next four quarters) for each stock in each of these 15 portfolios. We report the quarterly averages of these mean abnormal stock returns. We report results for the extreme groups only. The sample consists of US common stocks that are added to the Russell 2000 ‘from below’ from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Cumulative Abnormal Returns (CAR) around Analyst Recommendation Announcements for Russell 2000 Inclusion Stocks
Sample: Russell 2000 Inclusion ‘From Below’ Stocks
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 5 (Sell) 5‐1 1(Buy) 5 (Sell) 5‐1
1 (Short) 4.02 ‐1.07 ‐5.09 ‐1.62 0.64 2.26 (5.66) (‐2.34) (‐6.24) (‐4.63) (2.10) (5.24)
3 (Long) 2.15 ‐0.67 ‐2.82 0.87 0.80 ‐0.07 (8.75) (‐1.72) (‐5.55) (1.12) (3.27) (‐0.08)
3‐1 ‐1.87 0.40 2.27 2.49 0.16 ‐2.33 (‐2.50) (0.65) (2.65) (2.67) (0.19) (‐2.46)
Panel B: Cumulative Abnormal Returns (CAR) around Earnings Announcements for Russell 2000 Inclusion Stocks
Sample: Russell 2000 Inclusion ‘From Below’ Stocks
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 5 (Sell) 5‐1 1(Buy) 5 (Sell) 5‐1
1 (Short) 1.60 0.12 ‐1.48 ‐0.57 ‐0.83 ‐0.26 (6.31) (0.46) (‐4.29) (‐3.53) (‐3.49) (‐0.91)
3 (Long) 0.90 0.30 ‐0.60 0.30 0.13 ‐0.17 (5.56) (1.65) (‐3.21) (1.67) (0.76) (‐0.68)
3‐1 ‐0.70 0.18 0.87 0.87 0.96 0.10 (‐2.32) (0.54) (2.20) (3.79) (3.99) (0.30)
42
Table 6: Comparison of Stocks around the Russell 2000 Cutoff
Panel A presents mean values of Institutional Ownership, Analyst Coverage, and Stock Duration of the hundred smallest stocks that were added to the Russell 2000 from below (‘included’). We then compare these mean values with those of the hundred largest stocks that are just below the market cap cutoff for Russell 2000 inclusions from below and hence just missed the Russell 2000 inclusion (‘not included’). We report these mean values four quarters before and four quarters after index inclusion. Panel B then reports stock returns conditional on Stock Duration for the same two groups of stocks. We present monthly value‐weighted 4‐factor alphas for portfolio strategies from unconditional sorts into quintiles based on Stock Duration. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Included and Not Included Stocks: Institutional Ownership, Analyst Coverage, Stock Duration
Institutional Ownership Analyst Coverage Stock Duration
Included Not
Included Diff. Included Not
Included Diff. Included Not
Included Diff.
Quarter t = −4 31.29 28.62 2.67 1.62 1.33 0.29 1.30 1.31 ‐0.01 (50.03) (47.22) (3.06) (27.86) (26.95) (3.76) (61.79) (58.16) (‐0.21)Quarter t = +4 40.49 34.08 6.42 2.33 1.71 0.63 1.32 1.44 ‐0.12 (59.00) (52.43) (6.79) (36.10) (31.42) (7.40) (72.58) (63.55) (‐4.15)
Diff. 9.20 5.45 3.75 0.71 0.37 0.34 0.02 0.13 ‐0.11 (19.73) (13.26) (6.15) (12.12) (6.90) (4.82) (1.06) (6.01) (‐3.72)
Panel B: Included and Not Included Stocks: 4‐Factor Alphas
Monthly Value‐Weighted 4‐Factor Alpha
Return Period: t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
Stock Duration Included Not
Included Diff. Included Not
Included Diff.
All Stocks 1.30 0.15 1.16 ‐0.54 0.33 ‐0.87 (6.65) (0.72) (5.19) (‐3.19) (1.86) (‐3.76)
1 (Short) 2.39 1.10 1.30 ‐1.02 ‐0.02 ‐1.00 (6.30) (3.27) (2.82) (‐3.06) (‐0.07) (‐2.11)2 1.56 0.34 1.22 ‐0.56 0.39 ‐0.94 (6.06) (1.08) (3.45) (‐1.96) (1.57) (‐2.59)3 1.37 0.14 1.23 ‐0.44 0.40 ‐0.84 (5.22) (0.52) (3.66) (‐1.85) (1.66) (‐2.55)4 1.17 ‐0.18 1.34 ‐0.35 0.51 ‐0.86 (4.70) (‐0.78) (4.73) (‐1.55) (2.06) (‐2.70)
5 (Long) 0.76 ‐0.03 0.79 ‐0.08 0.53 ‐0.61 (2.58) (‐0.13) (2.36) (‐0.35) (2.09) (‐1.84)
5‐1 ‐1.64 ‐1.13 ‐0.51 0.94 0.55 0.39 (‐3.82) (‐3.09) (‐0.95) (2.40) (1.50) (0.71)
43
Table 7: Return Predictability: Portfolio Results for Stocks with High versus Low Limits to Arbitrage
This table presents monthly value‐weighted 5‐factor alphas for portfolio strategies based on 5x5x2 independent triple sorts based on Stock Duration, Analyst Recommendation, and three alternative proxies for limits to arbitrage. We report the portfolio alphas separately for stocks with a low or high short‐interest ratio (Panel A), stocks with low or high ownership by DFA (Panel B), and stocks with low or high idiosyncratic volatility (Panel C). At the beginning of each quarter, stocks are first divided into five groups based on Stock Duration. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 25 portfolios which are calculated over next four quarters. We report results for the extreme Stock Duration groups only. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. Returns are in monthly percentages. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Short‐Interest Ratio
Monthly Value‐Weighted 5‐Factor Alpha
Short Ratio = Low Short Ratio = High
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 2 3 4 5 (Sell) 5‐1 1 (Buy) 2 3 4 5 (Sell) 5‐1
1 (Short) ‐0.18 0.23 0.00 0.32 0.33 0.51 ‐0.85 ‐0.04 0.20 ‐0.02 0.28 1.12 0.61 (‐0.77) (1.15) (‐0.00) (1.44) (1.19) (1.48) (‐3.65) (‐0.18) (0.84) (‐0.07) (0.98) (3.37) (1.36)
5 (Long) 0.35 0.12 0.25 0.24 0.24 ‐0.11 0.10 0.15 0.08 0.09 ‐0.03 ‐0.13 ‐0.02 (1.89) (1.11) (1.79) (1.55) (1.42) (‐0.46) (0.41) (0.77) (0.46) (0.60) (‐0.18) (‐0.48) (‐0.05)
5‐1 0.54 ‐0.11 0.25 ‐0.08 ‐0.09 ‐0.63 0.95 0.18 ‐0.13 0.11 ‐0.30 ‐1.25 ‐0.63 (1.83) (‐0.50) (0.89) (‐0.27) (‐0.28) (‐1.56) (3.04) (0.65) (‐0.48) (0.42) (‐1.00) (‐3.13) (‐1.17)
Panel B: DFA Ownership
Monthly Value‐Weighted 5‐Factor Alpha
DFA Ownership = Low DFA Ownership = High
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 2 3 4 5 (Sell) 5‐1 1 (Buy) 2 3 4 5 (Sell) 5‐1 High‐Low
1 (Short) ‐0.81 ‐0.17 0.43 0.30 0.56 1.37 ‐0.35 ‐0.16 ‐0.11 0.25 0.24 0.59 ‐0.78 (‐3.56) (‐0.82) (1.75) (1.39) (1.92) (3.55) (‐2.12) (‐0.78) (‐0.55) (1.29) (1.24) (2.24) (‐1.82)
5 (Long) 0.15 0.21 0.19 0.21 0.12 ‐0.02 0.43 0.14 0.13 0.04 0.22 ‐0.21 ‐0.19 (0.76) (1.85) (1.78) (1.90) (0.86) (‐0.10) (3.38) (0.94) (1.19) (0.33) (1.69) (‐1.32) (‐0.68)
5‐1 0.95 0.38 ‐0.23 ‐0.09 ‐0.44 ‐1.40 0.78 0.30 0.24 ‐0.21 ‐0.01 ‐0.80 0.60 (3.18) (1.54) (‐0.86) (‐0.38) (‐1.31) (‐3.16) (3.90) (1.21) (1.04) (‐0.87) (‐0.07) (‐2.85) (1.11)
44
Table 7 (continued)
Panel C: Idiosyncratic Risk
Monthly Value‐Weighted 5‐Factor Alpha
Idiosyncratic Risk = Low Idiosyncratic Risk = High
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 2 3 4 5 (Sell) 5‐1 1 (Buy) 2 3 4 5 (Sell) 5‐1 High‐Low
1 (Short) ‐0.47 ‐0.15 0.24 0.03 0.19 0.66 ‐0.61 ‐0.31 0.15 0.37 0.63 1.24 0.59 (‐1.79) (‐0.88) (1.28) (0.15) (0.80) (2.01) (‐3.19) (‐1.33) (0.56) (1.56) (2.29) (3.71) (1.53)
5 (Long) 0.17 0.21 0.23 0.16 0.12 ‐0.04 0.25 0.21 0.45 0.06 ‐0.06 ‐0.31 ‐0.26 (0.90) (1.80) (2.05) (1.49) (0.96) (‐0.18) (1.27) (0.67) (1.75) (0.22) (‐0.25) (‐1.02) (‐0.72)
5‐1 0.64 0.36 ‐0.01 0.13 ‐0.06 ‐0.70 0.86 0.52 0.30 ‐0.32 ‐0.69 ‐1.55 ‐0.85 (2.01) (1.63) (‐0.06) (0.56) (‐0.23) (‐1.81) (3.51) (1.36) (0.79) (‐1.10) (‐1.99) (‐3.96) (‐1.67)
45
Table 8: Return Predictability: Fama‐MacBeth Regressions
This table provides in Panel A (full sample) and Panel B (Russell 2000 inclusion sample) quarterly Fama‐MacBeth predictive regressions linking next‐twelve‐months stock returns with Stock Duration and Analyst Recommendation. Newey‐West (1987) adjusted t‐statistics, calculated based on two lags, are reported in parentheses. The sample in Panel A consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. The sample in Panel B consists in columns 3 through 5 of stocks newly added to the Russell 2000 index ‘from below’. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses. Variables are defined in Appendix Table A‐1.
Panel A: All Stocks
Dependent Variable: Return 12 Months
(1) (2) (3) (4) (5) (6)
Intercept 0.484 0.518 0.309 0.551 0.378 0.089 (1.78) (1.83) (1.23) (1.91) (1.33) (0.31) Log(Stock Duration) 0.033 0.142 0.108 0.031 (3.05) (4.28) (3.69) (2.86) Stock Duration Low ‐0.032 ‐0.188 (‐2.74) (‐4.09) Stock Duration High 0.008 0.019 (0.88) (0.91) Analyst Recommendation × Log(Stock Duration) ‐0.051 ‐0.049 (‐4.19) (‐4.13) Log (Forecast Dispersion) × Log(Stock Duration) ‐0.005 (‐1.12) Log (Analyst Coverage) × Log(Stock Duration) 0.011 (1.07) Analyst Recommendation × Stock Duration Low 0.075 (3.97) Analyst Recommendation × Stock Duration High ‐0.004 (‐0.47) Analyst Recommendation × Log(Share Turnover) 0.036 (4.76) Log (Forecast Dispersion) × Log(Share Turnover) 0.005 (1.30) LOG(Analyst Coverage) × Log(Share Turnover) 0.002 (0.41) Analyst Recommendation 0.006 0.007 0.088 ‐0.007 0.085 0.199 (0.64) (0.72) (4.68) (‐0.59) (4.36) (4.51) Log (Analyst Coverage) 0.012 0.011 0.011 0.011 ‐0.005 0.025 (1.72) (1.67) (1.68) (1.61) (‐0.35) (1.12) Log (Forecast Dispersion) 0.003 0.003 0.003 0.004 0.012 0.030 (0.48) (0.50) (0.48) (0.50) (0.97) (1.25) Log (Market Cap) ‐0.010 ‐0.010 ‐0.011 ‐0.010 ‐0.011 ‐0.010 (‐1.56) (‐1.44) (‐1.62) (‐1.51) (‐1.63) (‐1.60) Log (MB Ratio) ‐0.025 ‐0.024 ‐0.023 ‐0.023 ‐0.024 ‐0.022 (‐2.22) (‐2.18) (‐2.10) (‐2.08) (‐2.12) (‐2.06) Past 12 Months Return ‐0.011 ‐0.011 ‐0.008 ‐0.008 ‐0.008 ‐0.008 (‐0.48) (‐0.50) (‐0.36) (‐0.37) (‐0.34) (‐0.35) Log (Share Turnover) 0.017 0.015 0.017 0.015 0.018 ‐0.056 (1.00) (0.87) (0.97) (0.89) (1.00) (‐1.92) Log (Institutional Ownership) ‐0.022 ‐0.022 ‐0.023 ‐0.022 ‐0.022 ‐0.022 (‐1.21) (‐1.22) (‐1.25) (‐1.23) (‐1.26) (‐1.18) Log (Idiosyncratic Risk) 0.027 0.026 0.025 0.025 0.025 0.027 (1.02) (1.00) (0.95) (0.94) (0.97) (1.00)
Average R‐square (%) 7.4 7.5 7.5 7.7 7.7 7.9 Obs. 148357 148357 148357 148357 148357 148357 Quarters 76 76 76 76 76 76
46
Table 8 (continued)
Panel B: Russell 2000 Inclusion Stocks
Dependent Variable: Return 12 Months
Full Sample R2000 Inclusion Sample
(1) (2) (3) (4) (5)
Intercept 0.551 0.563 0.350 0.155 0.094 (1.99) (2.02) (0.76) (0.35) (0.22) Log(Stock Duration) 0.030 0.030 0.051 0.152 0.053 (3.23) (3.21) (1.97) (2.37) (1.97) Log (Market Cap) ‐0.011 ‐0.011 ‐0.021 ‐0.012 ‐0.009 (‐2.08) (‐2.14) (‐0.79) (‐0.46) (‐0.35) Log (MB Ratio) ‐0.024 ‐0.024 ‐0.014 ‐0.014 ‐0.015 (‐2.12) (‐2.12) (‐0.96) (‐0.91) (‐0.97) Past 12 Months Return ‐0.005 ‐0.005 0.019 0.019 0.019 (‐0.25) (‐0.21) (0.86) (0.93) (0.94) Log (Share Turnover) 0.020 0.021 0.013 0.013 ‐0.017 (1.12) (1.21) (0.67) (0.67) (‐0.51) Log (Institutional Ownership) ‐0.015 ‐0.015 ‐0.015 ‐0.022 ‐0.022 (‐1.05) (‐1.09) (‐0.76) (‐1.12) (‐1.05) Log (Idiosyncratic Risk) 0.042 0.042 ‐0.021 ‐0.021 ‐0.020 (1.30) (1.29) (‐0.68) (‐0.67) (‐0.62) Analyst Recommendation 0.011 0.011 0.064 0.074 (1.55) (1.54) (1.91) (0.85) Analyst Recommendation × Log(Stock Duration) ‐0.051 (‐2.26) Analyst Recommendation × Log(Share Turnover) 0.015 (1.01) Russell 2000 Inclusion Russell 2000 Inclusion × Log(Stock Duration) 0.036 (1.84) Russell 2000 Inclusion × Log(Share Turnover) ‐0.034 (‐4.67)
Average R‐square (%) 6.3 6.3 11.9 13.8 13.8 Obs. 148357 148357 12292 9341 9341 Quarters 76 76 76 76 76
Figure 1: Changes in Analyst Recommendations and Stock Duration
The figure shows for the full sample average values of Analyst Recommendation (Figure 1A) and Stock Duration (Figure 1B) for stocks in four portfolios from eight quarters before to eight quarters after portfolio construction. Analyst recommendations are coded on a scale from 1 to 5. A recommendation of 1 corresponds to a ‘strong buy’ and a recommendation of 5 corresponds to a ‘strong sell’ recommendation. The portfolios are based on annual independent 5x5 sorts into Stock Duration and Analyst Recommendation quintiles. The reported four portfolios include stocks in the interaction of the first (‘Short’) and fifth (‘Long’) Stock Duration quintiles and the first (‘Buy’) and fifth (‘Sell’) Analyst Recommendation quintiles.
Figure 1A: Analyst Recommendations around Portfolio Construction
Figure 1B: Stock Duration around Portfolio Construction
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Analyst Recommendation
Quarter
Short Stock Duration & Buy Rec. Long Stock Duration & Buy Rec.
Short Stock Duration & Sell Rec. Long Stock Duration & Sell Rec.
0.0
0.5
1.0
1.5
2.0
2.5
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Stock Duration (years)
Quarter
Short Stock Duration & Buy Rec. Long Stock Duration & Buy Rec.
Short Stock Duration & Sell Rec. Long Stock Duration & Sell Rec.
48
Figure 2: Abnormal Returns for Stock Duration: Analyst Recommendation Quintile Portfolios
Figure 2A reports for the full sample 5‐factor alphas in event time (i.e., before and after portfolio construction) from eight quarters before to twelve quarters after portfolio construction. We report returns of stocks in four portfolios based on independent 5x5 sorts into Stock Duration and Analyst Recommendation quintiles. The returns are shown for the four portfolios with stocks in the interaction of the first (‘Short’) and fifth (‘Long’) Stock Duration quintiles and the first (‘Buy’) and fifth (‘Sell’) Analyst Recommendation quintiles. Figure 2B reports cumulative returns in calendar time for the long‐short portfolio that buys (sells) stocks in the fifth or ‘Sell’ (first or ‘Buy’) Analyst Recommendation quintile conditional on stocks being in the first (‘Short’) Stock Duration quintile. We also report returns for the long‐short portfolio that buys (sells) stocks in the fifth or ‘Sell’ (first or ‘Buy’) Analyst Recommendation quintile conditional on stocks being in the fifth (‘Long’) Stock Duration quintile.
Figure 2A: Cumulative Abnormal Returns in Event Time
Figure 2B: Cumulative Abnormal Returns in Calendar Time
‐60
‐50
‐40
‐30
‐20
‐10
0
10
20
30
40
‐8 ‐6 ‐4 ‐2 0 2 4 6 8 10 12
Cumulative 5‐Factor Alpha (%
)
Quarter
Short Stock Duration & Buy Rec. Long Stock Duration & Buy Rec.
Short Stock Duration & Sell Rec. Long Stock Duration & Sell Rec.
0
1
2
3
4
5
6
7
8
1‐Dec‐93
1‐Sep
‐94
1‐Jun‐95
1‐M
ar‐96
1‐Dec‐96
1‐Sep
‐97
1‐Jun‐98
1‐M
ar‐99
1‐Dec‐99
1‐Sep
‐00
1‐Jun‐01
1‐M
ar‐02
1‐Dec‐02
1‐Sep
‐03
1‐Jun‐04
1‐M
ar‐05
1‐Dec‐05
1‐Sep
‐06
1‐Jun‐07
1‐M
ar‐08
1‐Dec‐08
1‐Sep
‐09
1‐Jun‐10
1‐M
ar‐11
1‐Dec‐11
1‐Sep
‐12
1‐Jun‐13
Perform
ance of $1 In
itial Investment
Short Stock Duration: Long‐Short Portfolio Long Stock Duration: Long‐Short Portfolio
49
Figure 3: Changes in Analyst Recommendations and Stock Duration around Russell 2000 Inclusions ‘From Below’
The figure shows for the sample of firms that are added to the Russell 2000 ‘from below’ average values of Analyst Recommendation (Figure 3A) and Stock Duration (Figure 3B) for stocks in four portfolios from eight quarters before to eight quarters after portfolio construction. Analyst recommendations are coded on a scale from 1 to 5. A recommendation of 1 corresponds to a ‘strong buy’ and a recommendation of 5 corresponds to a ‘strong sell’ recommendation. The portfolios are based on annual independent 3x5 sorts of all stocks in our sample into Stock Duration terciles and Analyst Recommendation quintiles. The reported four portfolios include stocks in the interaction of the first (‘Short’) and third (‘Long’) Stock Duration tercile and the first (‘Buy’) and fifth (‘Sell’) Analyst Recommendation tercile.
Figure 3A: Analyst Recommendations around Portfolio Construction: Russell 2000 Inclusion Stocks
Figure 3B: Stock Duration around Portfolio Construction: Russell 2000 Inclusion Stocks
0
0.5
1
1.5
2
2.5
3
3.5
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Analyst Recommendation
Quarter
Short Stock Duration & Buy Rec. Long Stock Duration & Buy Rec.
Short Stock Duration & Sell Rec. Long Stock Duration & Sell Rec.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Stock Duration (years)
Quarter
Short Stock Duration & Buy Rec. Long Stock Duration & Buy Rec.
Short Stock Duration & Sell Rec. Long Stock Duration & Sell Rec.
50
Figure 4: Stock Duration, Analyst Coverage, and Institutional Ownership around Russell 2000 Inclusions ‘From Below’
The figure shows for the sample of firms that are added to the Russell 2000 ‘from below’ average values of Stock Duration, Analyst Coverage, and Institutional Ownership from eight quarters before to eight quarters after stocks are being added to the Russell 2000. Figure 4A only considers stocks in the first (‘Short’) Stock Duration tercile, while Figure 4B only considers stocks in the third (‘Long’) Stock Duration tercile. For each sample of stocks, we plot the average values of Stock Duration in quarters and report those on the middle axis. We also report Analyst Coverage on the middle axis. Institutional Ownership is reported on the right axis (in %).
Figure 4A: Stocks in the First Tercile (Short) of Stock Duration: Russell 2000 Inclusion Stocks
Figure 4B: Stocks in the Third Tercile (Long) of Stock Duration: Russell 2000 Inclusion Stocks
0
10
20
30
40
50
60
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Stock Duration Analyst Coverage Institutional Ownership
0
10
20
30
40
50
60
0
1
2
3
4
5
6
7
8
9
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Stock Duration Analyst Coverage Institutional Ownership
51
Figure 5: Stock Duration, Analyst Coverage, and Institutional Ownership around Russell 2000 Inclusion ‘From Above’
The figure shows for the sample of firms that are added to the Russell 2000 ‘from above’ average values of Stock Duration, Analyst Coverage, and Institutional Ownership from eight quarters before to eight quarters after stocks are added to the Russell 2000. Figure 5A only considers stocks in the first (‘Short’) Stock Duration tercile, while Figure 5B only considers stocks in the third (‘Long’) Stock Duration tercile. For each sample of stocks, we plot the average values of Stock Duration in quarters and report those on the middle axis. We also report Analyst Coverage on the middle axis. Institutional Ownership is reported on the right axis (in %).
Figure 5A: Stocks in the First Tercile (Short) of Stock Duration: Russell 2000 Inclusion ‘From Above’ Stocks
Figure 5B: Stocks in the Third Tercile (Long) of Stock Duration: Russell 2000 Inclusion ‘From Above’ Stocks
0
10
20
30
40
50
60
0
1
2
3
4
5
6
7
8
9
10
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Quarter
Stock Duration Analyst Coverage Institutional Ownership
0
10
20
30
40
50
60
0
1
2
3
4
5
6
7
8
9
‐8 ‐6 ‐4 ‐2 0 2 4 6 8
Quarter
Stock Duration Analyst Coverage Institutional Ownership
52
Figure 6: Institutional Ownership, Analyst Coverage, and Stock Duration around the Inclusion Cutoff
This figure shows mean values of Institutional Ownership, Analyst Coverage, and Stock Duration for each of the four quarters before and the four quarters after the Russell 2000 index inclusion date for two groups for stocks: the hundred smallest stocks that were just included in Russell 2000 ‘from below’ (‘included’) and the hundred largest stocks that just missed being included in Russell 2000 ‘from below’ (‘not included’). For each sample of stocks, we plot the average values of Stock Duration in quarters and report those on the middle axis. We also report Analyst Coverage on the middle axis. Institutional Ownership is reported on the right axis (in %).
30
32
34
36
38
40
42
44
0
1
2
3
4
5
6
7
‐4 ‐3 ‐2 ‐1 0 1 2 3 4
Stock Duration (Not Included) Analyst Coverage (Not Included)
Stock Duration (Included) Analyst Coverage (Included)
Institutional Ownership (Not Included) Institutional Ownership (Included)
53
Appendix Table A‐1: Definitions of Variables
This table provides definitions of the variables used in the empirical analysis.
Variable Definition
Stock Duration The weighted average duration a stock has been in the portfolios of institutional investors. This variable is calculated as the holding duration of ownership of each stock for every institutional investor by calculating a weighted‐measure of buys and sells by an institutional investor, weighted by the duration for which the stock was held. For each stock in a given fund manager’s portfolio, the holding duration measure is thus calculated by looking back over the full the time period since that particular stock has been held continuously in that fund’s portfolio. The calculation of the duration for stock i that is included in the institutional portfolio j at time T‐1, for all stocks i = 1 … I and all institutional investors j = 1 … J, is given by:
jiji
jiT
WTt jiji
tjiTji BH
HW
BH
tTDuration
,,
,1
,,
,,1,,
)1()1(
where Bi,j = total percentage of shares of stock i bought by institution j between t = T‐W and t = T‐1; t,T are in quarters; Hi,j = percentage of total shares outstanding of stock i held by institution j at time t = T‐W; αi,j,t = percentage of total shares outstanding of stock i bought or sold by institution j between time t‐1 and t, where αi,j,t > 0 for buys and <0 for sells. We choose W = 20 quarters, as very few stock positions are held continuously for longer than 5 years. If stock i is not included in institutional portfolio j at time T‐1, then Durationi,j,T‐1 = 0. Next, we compute at the individual stock‐level our Stock Duration proxy by averaging the institutional‐stock level duration over all institutions currently holding the stock, using as weights the total current holdings in the stock of each institution.
Analyst Recommendation Mean (consensus) analyst recommendation for a stock according to the IBES database. Analyst recommendations are coded on a scale from 1 to 5. A recommendation of 1 corresponds to a ‘strong buy’ and a recommendation of 5 corresponds to a ‘strong sell’ recommendation.
Analyst Coverage Number of analysts covering a stock according to the IBES database.
Share Turnover Daily number of a firm’s shares that are traded divided by the number of shares outstanding.
Institutional Ownership Percentage ownership of institutional investors.
DFA Ownership Percentage ownership by ownership Dimensional Fund Advisor (DFA), which is a major stock lender for small market capitalization stocks.
MB Ratio Market value of equity over the book value of equity.
Market Cap Market capitalization of the equity of a firm (in million).
Analyst Forecast Dispersion Ratio of the standard deviation of analysts’ next fiscal year earnings forecast divided by the mean forecast.
Short Ratio Short‐interest ratio of a stock.
Idiosyncratic Risk Residual that is obtained from a 3‐factor Fama and French model of stock returns. It is estimated using daily returns over the quarter before the fiscal year end.
Russell 2000 Inclusion Dummy variable that takes the value 1 if a stock is included to the Russell 2000 from below. Such a stock was previously not included in the Russell 1000 but the strong recent positive abnormal performance and increase in market value lead to Russell 2000 inclusion.
Return 12 Months Twelve‐months raw stock return.
Past 12 Months Return Past‐twelve‐months raw stock return.
Appendix Table A‐2: Robustness Check for CARs around Announcements: Share Turnover instead of Stock Duration
Panel A reports average three‐day cumulative abnormal stock returns around analyst recommendation announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. Panel B reports average three‐day cumulative abnormal stock returns around earnings announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. We first independently double sort stocks into quintiles at the end of each quarter based on Analyst Recommendation and Share Turnover. We then calculate the average three‐day size‐adjusted cumulative abnormal stock returns (CARs) around all analyst recommendation announcements/earnings announcements in the previous four quarter (the next four quarters) for each stock in each of these 25 portfolios. We report the quarterly average of these mean CARs. We report results for the extreme Share Turnover groups only. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Share Turnover and Analyst Recommendation: Recommendation Announcement Returns
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Share Turnover Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1(Buy) 2 3 4 5 (Sell) 5‐1
Uncond. 1.18 0.44 ‐0.12 ‐0.89 ‐1.85 ‐3.03 ‐0.80 ‐0.45 ‐0.05 0.27 0.70 1.50 (16.03) (11.83) (‐2.39) (‐8.77) (‐11.04) (‐16.36) (‐6.70) (‐6.24) (‐1.01) (5.19) (10.88) (11.14)
1 (Low) 0.27 1.02 0.68 0.21 0.07 ‐0.33 ‐1.36 0.47 0.30 0.09 0.07 0.90 0.90 0.60 (6.41) (13.25) (7.65) (2.72) (0.79) (‐4.37) (‐11.96) (8.11) (2.93) (0.72) (0.69) (6.19) (7.61) (4.50)
5 (High) ‐0.63 1.47 0.53 ‐0.44 ‐1.78 ‐3.69 ‐5.16 ‐0.52 ‐1.29 ‐0.88 ‐0.31 0.20 0.62 1.91 (‐6.54) (9.25) (6.84) (‐4.03) (‐8.51) (‐10.10) (‐13.49) (‐5.00) (‐6.68) (‐7.51) (‐3.10) (1.92) (4.58) (8.19)
5‐1 ‐0.91 0.45 ‐0.15 ‐0.64 ‐1.85 ‐3.36 ‐3.80 ‐0.99 ‐1.59 ‐0.96 ‐0.38 ‐0.69 ‐0.28 1.31 (‐8.75) (2.71) (‐1.29) (‐4.60) (‐7.79) (‐9.70) (‐10.37) (‐8.75) (‐7.87) (‐6.35) (‐2.49) (‐3.82) (‐1.46) (4.87)
Panel B: Share Turnover and Analyst Recommendations: Earnings Announcement Returns
Return Period: Quarter t = −3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
CAR(‐1,+1)
Analyst Recommendation Analyst Recommendation
Share Turnover Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1(Buy) 2 3 4 5 (Sell) 5‐1
Uncond. 0.81 0.64 0.31 ‐0.17 ‐0.52 ‐1.33 0.03 0.17 0.22 0.22 0.25 0.22 (19.68) (18.65) (12.18) (‐6.19) (‐18.32) (‐28.40) (1.04) (6.71) (10.74) (8.69) (7.55) (5.42)
1 (Low) 0.29 0.66 0.40 0.27 0.10 0.09 ‐0.57 0.39 0.35 0.36 0.42 0.27 0.47 0.12 (9.25) (11.98) (8.04) (5.79) (2.19) (2.56) (‐10.89) (15.46) (6.15) (7.55) (8.63) (4.65) (12.79) (1.92)
5 (High) 0.05 1.05 0.87 0.26 ‐0.66 ‐1.89 ‐2.94 ‐0.10 ‐0.27 0.04 0.04 0.02 ‐0.44 ‐0.17 (1.09) (13.39) (12.13) (4.03) (‐8.11) (‐14.92) (‐20.85) (‐2.40) (‐3.84) (0.45) (0.59) (0.23) (‐4.29) (‐1.49)
5‐1 ‐0.24 0.39 0.47 ‐0.01 ‐0.77 ‐1.99 ‐2.38 ‐0.49 ‐0.61 ‐0.33 ‐0.38 ‐0.25 ‐0.91 ‐0.30 (‐3.96) (4.62) (5.76) (‐0.10) (‐7.90) (‐13.91) (‐17.01) (‐8.72) (‐6.40) (‐3.27) (‐4.29) (‐2.58) (‐8.46) (‐2.46)
55
Appendix Table A‐3: Robustness Check for Portfolio Sorts: 3‐Factor Alphas and Share Turnover
Panel A reports monthly equal‐weighted and value‐weighted 3‐factor Fama‐French alphas for independent double sorts based on Stock Duration and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into five groups based on Stock Duration. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 25 portfolios which are calculated over next four quarters. Panel B reports monthly equal‐weighted and value‐weighted 5‐factor Fama‐French alphas for independent double sorts based on Share Turnover and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into five groups based on Share Turnover. They are then independently divided into five groups based on Analyst Recommendation. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All the reported returns are in monthly percentages. The sample consists of US common stocks from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Double Sorts on Stock Duration and Analyst Recommendation: 3‐Factor Alphas
Monthly Equal‐Weighted 3‐Factor Alpha Monthly Value‐Weighted 3‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Stock Duration Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1(Buy) 2 3 4 5 (Sell) 5‐1
Uncond. ‐0.04 0.08 0.22 0.22 0.22 0.26 ‐0.05 0.16 0.15 0.08 ‐0.04 0.01 (‐0.37) (0.74) (2.25) (2.05) (1.64) (2.15) (‐0.46) (2.63) (2.11) (1.12) (‐0.38) (0.08)
1 (Short) ‐0.12 ‐0.42 ‐0.20 0.15 0.23 0.45 0.86 ‐0.15 ‐0.69 ‐0.22 0.30 0.20 0.29 0.98 (‐0.80) (‐2.50) (‐1.21) (0.89) (1.09) (1.98) (4.49) (‐1.02) (‐3.55) (‐1.17) (1.45) (1.07) (1.21) (3.09)
5 (Long) 0.32 0.40 0.36 0.30 0.28 0.32 ‐0.08 0.12 0.11 0.23 0.16 0.11 ‐0.06 ‐0.17 (3.91) (3.53) (3.63) (3.11) (2.98) (2.51) (‐0.58) (1.93) (0.62) (2.11) (1.54) (1.00) (‐0.45) (‐0.72)
5‐1 0.444 0.81 0.56 0.15 0.06 ‐0.13 ‐0.94 0.27 0.80 0.45 ‐0.14 ‐0.09 ‐0.35 ‐1.15 (3.29) (5.57) (3.32) (0.97) (0.28) (‐0.62) (‐4.50) (1.48) (3.04) (2.00) (‐0.60) (‐0.45) (‐1.27) (‐3.16)
Panel B: Double Sorts on Share Turnover and Analyst Recommendation: 5‐Factor Alphas
Monthly Equal‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Share Turnover Uncond. 1 (Buy) 2 3 4 5 (Sell) 5‐1 Uncond. 1(Buy) 2 3 4 5 (Sell) 5‐1
1 (Low) 0.40 0.47 0.43 0.54 0.36 0.47 0.00 0.27 0.39 0.46 0.35 0.23 0.05 ‐0.34 (3.70) (3.47) (3.11) (4.03) (2.69) (3.71) (0.01) (1.97) (2.45) (2.30) (1.96) (1.43) (0.33) (‐1.85)
5 (High) 0.10 ‐0.28 ‐0.01 0.25 0.60 0.58 0.86 0.19 ‐0.11 0.29 0.25 0.35 0.65 0.76 (0.69) (‐1.56) (‐0.04) (1.31) (3.37) (2.09) (3.19) (1.00) (‐0.44) (1.25) (1.16) (1.71) (2.46) (2.24)
5‐1 ‐0.30 ‐0.75 ‐0.44 ‐0.29 0.24 0.11 0.86 ‐0.08 ‐0.50 ‐0.17 ‐0.10 0.12 0.60 1.10 (‐1.66) (‐4.13) (‐2.07) (‐1.19) (0.97) (0.37) (2.98) (‐0.30) (‐1.63) (‐0.53) (‐0.29) (0.43) (1.83) (2.98)
Appendix Table A‐4: Results for Sub‐Periods: 1993‐2003 versus 2004 to 2013
Panels A reports alphas for independent double sorts based on Stock Duration and Analyst Recommendation, separately for the periods 1993 to 2003 and 2004 to 2013. At the beginning of each quarter, stocks are first divided into five groups based on Stock Duration. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 25 portfolios which are calculated over next four quarters. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All the reported returns are in monthly percentages. Panels B reports alphas for independent double sorts based on Share Turnover and Analyst Recommendation, separately for the periods 1993 to 2003 and 2004 to 2013. At the beginning of each quarter, stocks are first divided into five groups based on Share Turnover. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 25 portfolios which are calculated over next four quarters. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All the reported returns are in monthly percentages. We report results for the extreme portfolios only. The sample consists of US common stocks. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Double Sorts on Stock Duration and Analyst Recommendation: 1993‐2003 versus 2004 to 2013
1993‐2003 2004‐2013
Monthly Value‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 5 (Sell) 5‐1 1(Buy) 5(Sell) 5‐1
1 (Short) ‐1.04 0.78 1.82 ‐0.47 0.07 0.54 (‐2.92) (2.50) (3.57) (‐2.73) (0.20) (1.43)
5 (Long) ‐0.16 0.12 0.28 0.36 0.00 ‐0.37 (‐0.55) (0.65) (0.75) (1.64) (‐0.02) (‐1.26)
5‐1 0.88 ‐0.66 ‐1.53 0.83 ‐0.08 ‐0.90 (1.88) (‐1.73) (‐2.46) (3.11) (‐0.18) (‐2.14)
Panel B: Double Sorts on Share Turnover and Analyst Recommendation: 1993‐2003 versus 2004 to 2013
1993‐2003 2004‐2013
Monthly Value‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Share Turnover 1 (Buy) 5 (Sell) 5‐1 1(Buy) 5(Sell) 5‐1
1 (Low) 0.21 0.09 ‐0.12 0.53 0.04 ‐0.49 (0.92) (0.39) (‐0.42) (2.83) (0.26) (‐2.24)
5 (High) ‐0.54 1.22 1.76 0.21 0.17 ‐0.04 (‐1.46) (2.78) (3.32) (0.63) (0.62) (‐0.10)
5‐1 ‐0.75 1.13 1.88 ‐0.32 0.13 0.45 (‐1.95) (2.09) (3.26) (‐0.75) (0.36) (1.01)
57
Appendix Table A‐5: Robustness Check for Russell 2000 Tests: Inclusion ‘From Below and Above’ Stocks
Panel A provides different alphas for unconditional sorts on Stock Duration for stocks that were added ‘from below and above’ to the Russell 2000. Panel B reports for stocks added to the Russell 2000 ‘from below and above’ average three‐day cumulative abnormal stock returns (CARs) around analyst recommendation announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. We first independently double sort stocks into terciles/quintiles at the end of each quarter based on Analyst Recommendation and Stock Duration. We then calculate the average cumulative three‐day size‐adjusted CARs around all analyst recommendation announcements in the previous four quarter (the next four quarters) for each stock in each of these 15 portfolios. We report the quarterly averages of these mean abnormal stock returns. Panel C reports for Russell 2000 inclusion stocks ‘from below and above’ monthly equal‐weighted and value‐weighted 3‐factor Fama‐French alphas for independent double sorts based on Share Turnover and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into three groups based on Share Turnover. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 15 portfolios which are calculated over next four quarters. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. We report results for the extreme groups only. All the reported returns are in monthly percentages. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses.
Panel A: Sorts on Stock Duration: Russell 2000 Inclusion ‘From Below and Above’ Stocks
Sample: Russell 2000 Inclusion ‘From Below and Above’ Stocks
Monthly Equal‐Weighted Returns Monthly Value‐Weighted Returns
Stock Duration CAPM Alpha 3‐Factor Alpha 5‐Factor Alpha CAPM Alpha 3‐Factor Alpha 5‐Factor Alpha
1 (Short) ‐0.92 ‐0.97 ‐0.79 ‐0.92 ‐0.93 ‐0.93 (‐3.11) (‐4.88) (‐4.44) (‐3.09) (‐4.94) (‐4.82)
5 (Long) 0.36 0.07 0.10 0.39 0.09 0.10 (1.55) (0.42) (0.64) (1.65) (0.53) (0.55)
5‐1 1.28 1.03 0.89 1.31 1.02 1.03 (4.31) (4.50) (3.99) (3.81) (4.03) (3.95)
Panel B: CARs around Analyst Recommendations: Russell 2000 Inclusion ‘From Below and Above’ Stocks
Return Period: Quarter t = ‐3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
Analyst Recommendation Analyst Recommendation
Stock Duration 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1
1 (Short) 3.46 ‐4.09 ‐7.55 ‐1.51 0.70 2.21 (6.19) (‐7.60) (‐8.80) (‐5.00) (2.54) (5.33)
3 (Long) 2.12 ‐1.17 ‐3.29 0.82 0.91 0.09 (7.63) (‐3.63) (‐7.57) (1.05) (2.79) (0.12)
3‐1 ‐1.34 2.92 4.26 2.33 0.21 ‐2.12 (‐2.26) (5.53) (5.01) (2.52) (0.35) (‐2.15)
Panel C: Sorts on Share Turnover and Analyst Recommendation: Russell 2000 Inclusion ‘From Below and Above’ Stocks
Sample: Russell 2000 Inclusion ‘From Below and Above’ Stocks
Analyst Recommendation Analyst Recommendation
Share Turnover 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1
Uncond. ‐0.52 0.17 0.68 ‐0.66 0.07 0.73 (‐2.87) (0.94) (2.64) (‐3.28) (0.32) (2.48)
1 (Low) ‐1.00 0.53 1.53 ‐1.11 0.21 1.32 (‐3.27) (1.28) (2.76) (‐3.30) (0.50) (2.31)
3 (High) 0.23 0.05 ‐0.18 ‐0.02 ‐0.17 ‐0.14 (0.79) (0.24) (‐0.51) (‐0.08) (‐0.74) (‐0.40)
3‐1 1.23 ‐0.48 ‐1.71 1.09 ‐0.37 ‐1.46 (3.15) (‐1.09) (‐2.85) (2.64) (‐0.78) (‐3.38)
58
Appendix Table A‐6: Robustness Check for Russell 2000 Tests: Share Turnover instead of Stock Duration
Panel A presents for stocks added to the Russell 2000 ‘from below’ monthly equal‐weighted and value‐weighted CAPM alphas, 3‐factor Fama‐French alphas, and 5‐factor alphas for portfolio strategies from unconditional sorts based on Share Turnover. Panels B reports monthly alphas for independent double sorts based on Share Turnover and Analyst Recommendation. At the beginning of each quarter, stocks are first divided into three groups based on Share Turnover. They are then independently divided into five groups based on Analyst Recommendation. We then report returns for these 15 portfolios which are calculated over next four quarters. To account for overlapping portfolios, we follow the methodology in Jegadeesh and Titman (1993) such that stocks ranked in each of the last four quarters form one‐fourth of each portfolio. All the reported returns are in monthly percentages. The sample consists of US common stocks that are added to the Russell 2000 from below from December 1993 to December 2013. We eliminate stocks without analyst recommendations, stocks with missing market capitalization or book value of equity data, and stocks with prices below $1. 5% significance levels are denoted in bold and t‐statistics are reported in parentheses. Panel C reports for stocks added to the Russell 2000 ‘from below’ average three‐day cumulative abnormal stock returns (CARs) around analyst recommendation announcements that were made during the past four quarters (the next four quarters) from the portfolio formation date. We first independently double sort stocks into terciles/quintiles at the end of each quarter based on Analyst Recommendation and Share Turnover. We then calculate the average cumulative three‐day size‐adjusted abnormal stock returns (CARs) around all analyst recommendation announcements in the previous four quarter (the next four quarters) for each stock in each of these 15 portfolios. We report the quarterly averages of these mean abnormal stock returns. We report results for the extreme groups only.
Panel A: Sorts on Share Turnover
Share Turnover EW 3‐Factor
Alpha VW 3‐Factor
Alpha EW 5‐Factor
Alpha VW 5‐Factor
Alpha
1 (Low) 0.02 ‐0.15 0.06 ‐0.13 (0.09) (‐0.82) (0.31) (‐0.72) 2 0.03 0.10 0.04 0.01 (0.18) (0.51) (0.21) (0.05) 3 ‐0.34 ‐0.34 ‐0.22 ‐0.35 (‐1.85) (‐1.84) (‐1.22) (‐1.83) 4 ‐0.41 ‐0.30 ‐0.24 ‐0.39 (‐1.85) (‐1.13) (‐1.12) (‐1.45)
5 (High) ‐0.90 ‐0.92 ‐0.77 ‐0.98 (‐3.59) (‐3.29) (‐3.25) (‐3.44)
5‐1 ‐0.91 ‐0.77 ‐0.83 ‐0.85 (‐2.88) (‐2.26) (‐2.65) (‐2.44)
Panel B: Double Sorts on Share Turnover and Analyst Recommendation
Monthly Equal‐Weighted 5‐Factor Alpha Monthly Value‐Weighted 5‐Factor Alpha
Analyst Recommendation Analyst Recommendation
Share Turnover Uncond. 1 (Buy) 5 (Sell) 5‐1 Uncond. 1 (Buy) 5 (Sell) 5‐1
1 (Low) 0.07 ‐0.17 0.10 0.27 ‐0.09 ‐0.04 ‐0.10 ‐0.06 (0.42) (‐0.64) (0.45) (0.88) (‐0.53) (‐0.13) (‐0.47) (‐0.21)
3 (High) ‐0.58 ‐0.94 ‐0.33 0.61 ‐0.70 ‐1.17 ‐0.44 0.73 (‐2.82) (‐3.01) (‐0.78) (1.24) (‐2.52) (‐3.45) (‐0.97) (1.27)
3‐1 ‐0.65 ‐0.78 ‐0.43 0.35 ‐0.61 ‐1.14 ‐0.34 0.79 (‐2.45) (‐1.87) (‐0.96) (0.60) (‐1.84) (‐2.67) (‐0.70) (1.22)
59
Appendix Table A‐6 (continued)
Panel C: CARs around Analyst Recommendation Announcements
Return Period: Quarter t = ‐3 to Quarter t = 0 Return Period: Quarter t = +1 to Quarter t = +4
Analyst Recommendation Analyst Recommendation
Share Turnover 1 (Buy) 5 (Sell) 5‐1 1 (Buy) 5 (Sell) 5‐1
1 (Low) 1.47 0.37 ‐1.10 1.49 0.63 ‐0.86 (5.43) (0.63) (‐1.74) (2.02) (3.16) (‐1.07)
3 (High) 3.96 ‐2.79 ‐6.75 ‐1.55 0.19 1.74 (4.70) (‐4.35) (‐6.75) (‐4.22) (0.61) (3.68)
3‐1 2.48 ‐3.16 ‐5.65 ‐3.04 ‐0.43 2.61 (2.84) (‐4.01) (‐4.83) (‐3.46) (‐1.24) (2.40)