Do Fund Managers Identify and Share Profitable Ideas?

1
by
Wesley R. Gray2 and Andrew E. Kern3

ABSTRACT

We study data from an organization where fundamentals-based fund managers privately share investment
ideas. Evidence suggests the professional investors in our sample have stock-picking skills. A strategy of
going long (short) buy (sell) recommendations earns monthly calendar-time abnormal returns of 1.38% (2.91%) over the January 1, 2000 to December 31, 2008 sample period. Interestingly, these skilled
investors share their profitable ideas. We test predictions from private information sharing theories and
determine that our sample investors share ideas to receive constructive feedback, gain access to a broader
set of actionable ideas, and to attract additional arbitrager capital to their asset market.

JEL Classification: G10, G11, G14

Key words: Networks, Hedge Funds, and Market Efficiency.

1
We would like to thank Daniel Bergstresser, Dave Carlson, Hui Chen, John Cochrane, Lauren Cohen, Cliff Gray,
Eugene Fama, Ron Howren, Carl Luft, Stavros Panageas, Shastri Sandy, Gil Sadka, Amir Sufi, Pietro Veronesi, and
Rob Vishny. We would also like to thank seminar participants at the University of Chicago Booth School of
Business, Texas A&M, Southern Methodist Univesity, Drexel, Ohio State, University of Virginia, University of
Washington, and Boston College.
2
Drexel University, Lebow College of Business, Philadelphia, PA 19104, Phone: 773-230-4727, Fax: 888-5175529, E-mail: [email protected].
3
Empirical Finance, LLC, 17304 Preston Rd., Suite 1230, Dallas, TX 75252, Phone: 573-356-8993, Fax: 888-5175529, E-mail: [email protected].

Do Fund Managers Identify and Share Profitable Ideas?

ABSTRACT

We study data from an organization in which fund managers privately share investment ideas.
Evidence suggests the investors in our sample have stock-picking skills. A strategy of going
long (short) buy (sell) recommendations earns monthly calendar-time abnormal returns of 1.38%
(-2.91%) over the January 1, 2000 to December 31, 2008 sample period. Interestingly, these
skilled investors share their profitable ideas. We test predictions from private information
sharing theories and determine that the managers in our sample share ideas to receive
constructive feedback, gain access to a broader set of actionable ideas, and to attract additional
arbitrager capital to their asset market.

Fundamentals-based money managers, or value investors, play a key role in the price
discovery process. In equity markets specifically, a value managers job is to research a firms
management, business, and future prospects and determine if the company is selling for less than
its intrinsic value. If the value manager believes a security to be inexpensive relative to its
intrinsic value, he will buy the security, driving its price towards intrinsic value. If he believes it
to be expensive, he will either sell the security or sell the security short, thereby putting
downward pressure on the price and driving its price to intrinsic value. This logic is the basis for
the market efficiency hypothesis (Freidman, 1953). And yet, Grossman and Stiglitz (1980) argue
that market prices can never be perfectly efficient: If prices were always efficient, skilled
investors who acquire private information1 would never be rewarded.
In the first part of this paper, we test the Grossman and Stiglitz prediction that price
discovery agents compensation comes in the form of abnormal returns generated by inefficient
market prices. Specifically, we study a group of specialized market participants (predominantly
hedge fund managers focused exclusively on fundamental analysis) who share detailed
investment recommendations on the private website Valueinvestorsclub.com (VIC). We find
evidence of stock-picking skill among VIC members. Abnormal returns are economically large,
statistically significant, and robust to a variety of controls. For example, the average monthly
calendar-time portfolio alpha estimate (using 3-factor WLS regressions) is +1.38% for buy
recommendations and -2.91% for sell recommendations.
The empirical evidence suggesting that VIC members are talented stock-pickers is
interesting on its own merits. However, the unique organizational structure of VIC, which is

Private information in our context refers to information derived from the effective and efficient collection and
processing of publicly available information.

explicitly designed to facilitate private information exchange among professional investors,

allows us to contribute to the literature by empirically addressing a question about investor
behavior that has largely been ignored: Why would any investor with valuable private
information share such information with others? Traditional theories suggest the process of how
information flows into asset prices is straight forward. First an arbitrageur identifies a temporary
mispricing, then acts on the price discrepancy with all available resources causing asset prices
move to fundamental value (e.g., Friedman, 1953). However, bearing in mind this efficient
pricing process, it remains unclear why a fund manager would share information about a
profitable trading opportunity with other investors.
Recently, three theories have emerged that address the question of why an investor would
share valuable private information. Stein (2008) proposes that fund managers may share private
information because they gain valuable feedback from the person with whom they are sharing
(collaboration argument). Gray (2010) suggests that another reason for information sharing is
the promotion of managers undervalued portfolio positions in order to attract additional capital
to the market from other arbitrageurs (awareness argument). Gray also argues that a resourceconstrained arbitrageur may share profitable ideas with the competition because doing so allows
the arbitrageur to diversify his portfolio among a group of arbitrage trades, rather than allocate
all his capital into his limited set of good ideas (diversification argument).
In this paper we present the first empirical evidence to address the predictions of various
information sharing theories. We determine that all three information exchange theories play a
role in the sharing decision. That is, we find investors share ideas to receive constructive
feedback, gain access to a broader set of profitable ideas, and to attract additional arbitragers to

their asset market. These findings suggest that the mechanisms through which information flows
into security prices are not as simple as traditional asset pricing models suggest.
The remainder of the paper is organized as follows. Section 1 discusses relevant research.
Section 2 describes the data. Section 3 tests for stock-picking skill. Section 4 examines the
relation between ex-ante VIC idea ratings and ex-post abnormal returns. Section 5 addresses why
skilled fund managers share profitable trading opportunities and Section 6 concludes.

1. Related Literature on the Stock-Picking Hypothesis

Research on the collective performance of professional money managers indicates that
outperforming a passive risk-adjusted index is extremely difficult. Specifically, studies of mutual
fund managers have found that mutual funds, on average, do not outperform their benchmarks
(e.g., Carhart, 1997; Malkiel, 1995; and Daniel, Grinblatt, Titman, and Wermers, 1997). A more
recent analysis by Fama and French (2010) suggests that the aggregate portfolio of U.S. equity
mutual funds roughly approximates the market portfolio and that there is little evidence for
stock-picking skill among managers.
Another method of testing the stock-picking skill hypothesis is to study the performance
alternative asset managers such as hedge fund managers. This research often involves analysis of
hedge fund return databases. However, data pitfalls plague this research. First, hedge fund return
databases suffer from survivorship bias and self-selected reporting (Fung and Hsieh, 2000).
Second, hedge fund managers sometimes hold illiquid assets or engage in return smoothing,
which causes the reported hedge fund returns to exhibit large autocorrelations (Asness, Krail,
and Liew, 2001; and Getmansky, Lo, and Makarov, 2004). Third, researchers such as Liang

(2003) find that hedge fund database returns may be unreliable because the same hedge funds
sometimes report different returns to different database operators. Finally, hedge fund managers
often hold assets that have option-like, non-linear payoffs. This payoff profile makes it difficult
for researchers to assess performance using traditional linear factor models (Fung and Hsieh,
2001).
Griffin and Xu (2009) address the aforementioned issues with hedge fund return database
biases by analyzing hedge fund performance via their required 13F equity filings. Nonetheless,
their analysis has its own shortcomings because it can only examine long equity positions and
ignores intra-quarter trading.
Much of the work in this area focuses on the returns of broad portfolio returns to test for
the presence of stock-picking skill. However, Cohen, Polk, and Silli (2009) argue that analyzing
portfolio returns is not a test of stock-picking skill because portfolio returns may disguise a fund
managers true stock-picking ability. They argue that managers have incentives to hold
diversified portfolios that consist of their best ideas along with other positions to round out
their portfolios. In the end, a skilled manager may appear unskilled because of the perverse
incentive structure in the fund management industry.
An alternative approach to testing the stock-picking hypothesis, which does not suffer
from the inference problems associated with studying portfolio returns, is to analyze individual
recommendations from superstar managers or stock analysts. These studies show little evidence
in support of the stock-picking-skill hypothesis. Desai and Jain (1995) examine the performance
of recommendations made by superstar money managers at the Barrons Annual Roundtable
and find little evidence of superior stock-picking skill.

Barber, Lehavy, McNichols, and

Trueman (2001) confirm this result and find that excess returns to the recommendations of stock
analysts are not reliably positive after transaction costs.
We are able to avoid the inference problems associated with studying portfolio returns
and directly test manager stock-picking skill by studying detailed individual stock
recommendations shared on VIC. Moreover, VIC is a unique setting in which managers have
incentives to share profitable ideas (see the discussion in section 5). Further, the detailed
investment recommendations submitted by VIC members can be verified by the clubs
sophisticated membership, thereby eliminating the incentive for the promotion of efficiently
priced recommendations.
The unique environment we study is different than the environment for analyst
recommendations, where incentives to share good ideas may be perversely influenced by
investment banking relationships, or the environment for superstar recommendations at the
Barrons Annual Roundtable, where it is unclear that the managers sharing ideas have any
incentive to share valuable ideas with the general public. Also, because the research and analysis
behind superstar recommendations are never made fully available, it is unclear whether superstar
recommendations are really meant to be bonafied stock-pick recommendations, or simply reflect
an opportunity for the superstar manager to market his firms brand to the general public.
Our database is not a panacea. The ideas under analysis are the simplest, most
straightforward common equity recommendations submitted to VIC and we are further limited
by the data available on CRSP/Compustat. The exclusion of the many complicated arbitrage
trades and special situation scenarios submitted to VIC, but not analyzed due to data and analysis
constraints, may bias the evidence. These sophisticated trades require advanced knowledge and

understanding of niche securities and/or access to expensive resources such as lawyers, industry
specialists, or tax experts. In a Grossman and Stiglitz (1980) equilibrium in which arbitrageurs
are compensated for their information discovery efforts, one may hypothesize that these high
information cost investments would have better gross returns than situations requiring less effort
and fewer resources. If this story is to be believed, the data under analysis will likely be biased in
favor of the null hypothesis that VIC members have no stock-picking skill. In general our data
offer a rare opportunity to test a group of specialized managers for stock-picking skill in a
relatively clean setting.

2. Data
2.1. Value Investors Club
The data in this study are collected from a private internet community called
Valueinvestorsclub.com, an exclusive online investment club in which top investors share their
best ideas.2 Many business publications have heralded the site as a top-quality resource for
those who can attain membership (e.g., Financial Times, Barrons, BusinessWeek, and Forbes).3
Joel Greenblatt and John Petry, managers of the large fundamentals-based hedge fund Gotham
Capital, founded the site with $400,000 of start-up capital. Their goal was for VIC to be a place
for the best-quality ideas on the Web (Barker, 2001). The investment ideas submitted on the
clubs site are broad, but are best described as fundamentals-based. VIC states that it is open to
any well-thought-out investment recommendation, but that it has particular focus on long or
short equity or bond-based plays, traditional asset undervaluation plays such as high book-to-

2
3

http://www.valueinvestorsclub.com/Value2/Guests/Info.aspx
Ibid.

market, low price-to-earnings, liquidations, etc., and investment ideas based on the notion of
value as articulated by Warren Buffett (firms selling at a discount to their intrinsic value
irrespective of common valuation ratios).
Membership in the club is capped at 250 and admittance is highly selective with an
approximate acceptance rate of 6%. Admittance is based solely on a detailed write-up of an
investment idea (typically 1000 to 2000 words). Firm background and prior portfolio returns are
not part of the application process. If the quality of the independent research is satisfactory and
the aspiring member deemed a credible contributor to the club, he is admitted. Once admitted,
members are required to submit two ideas per year with a maximum of six ideas per year. This
maximum exists to encourage the submission of only the members best ideas.
In addition to allowing members to comment on and rate other members ideas, a weekly
prize of $5,000 is awarded to the best idea submitted (prize is determined by VIC management).
Members are monitored to ensure they submit at least two acceptable ideas per year and
members failing to meet the high standards of the club are dismissed through a community-wide
policing mechanism. Members are allowed to submit a thesis on a security that has been
submitted in the past if the write-up is substantially unique.

Otherwise, the members are

required to submit their ideas in the feedback section associated with the original idea posting.
An important aspect of VIC is that members identities are not disclosed to the general
public or to the other members of the club. The intent of this policy is to keep individual VIC
members from forming outside sharing syndicates with other members, who could then take their
valuable research and comments away from the broader VIC community. The anonymity
requirement also ensures the message board does not become a venue for hedge fund managers

to signal to potential investors or market their services to the general public.4 Finally, by keeping
identifying information private, members can speak truthfully and without consequence about
conversations with management, proxy situations, and other sensitive situations in which identity
disclosure could lead to legal or relationship repercussions.
Because membership of VIC is strictly confidential, we are unable to publish the limited
statistics on VIC members profiles. However, the management of VIC agreed to disclose that
VIC members are predominantly long-focused fundamentals-based hedge fund managers who
typically have assets under management of between $50 million and $250 million.

2.2. Data Description

We analyze all investment reports submitted to VIC from the time of the clubs founding
on January 1, 2000 through December 31, 2008. These reports represent all reports submitted to
VIC over the entire time period the club has existed. That is, reports containing what ultimately
prove to be poor recommendations are not deleted from the website and therefore our database
does not suffer from an ex-post selection bias.

In total, we examine 3,273 investment

submissions. Report length can range from several hundred to a few thousand words. Investment
ideas are wide-ranging with respect to the security type, trading location of the asset and the
complexity of the strategy employed.
For each investment report analyzed, we record various data: date and time of
submission, symbol, price at time of submission, market(s) traded, security(s) traded, strategy
recommended (long, short, or long/short), and the reasons for investing. All data collected are

This would create a legal predicament for hedge fund managers who rely on Rule 506 of Regulation D in the
Securities Act of 1933 to exempt them from registering their security offerings with the SEC.

unambiguous except for the reasons for investing. We compile a list of sixteen investment
criteria that are frequently cited in VIC submissions. Criteria were judged to be sufficiently
common if at least ten investment submissions acknowledged the use of the category. We then
match the firms associated with a VIC recommendation to accounting and stock return data from
CRSP/Compustat.
For the purposes of this study, we only analyze U.S. exchange-traded long and short
common stock recommendations. We do not analyze U.S. common equity investment
recommendations that have payoffs one may consider non-linear or inappropriate to analyze with
linear factor asset pricing models because they may bias our results (Fung and Hsieh, 2001).
Specifically, we eliminate all recommendations classified as merger arbitrage, stub arbitrage,
pair-trade, liquidation, long/short pair-trade recommendations, and non-common-equity ideas
such as options or preferred stock. We also eliminate foreign-traded or ADR recommendations.
Of the 3,273 observations in the original sample, 2,832 refer to U.S. securities. Of these
2,832 observations, 2,698 are recommendations on U.S. common stock securities. After the
restrictions described above, we are left with 1,956 U.S.-equity long recommendations and 242
U.S.-equity short recommendations with at least one monthly return observation. Tables 1 and 2
present summary statistics of the sample.
[Insert Table 1]
[Insert Table 2]

3. Performance Analysis
In this section, we examine the performance of VIC recommendations. VIC members

10

often state that their ideas should be considered long-term investments and not short-term
trades. To capture this notion of long-term performance, we perform detailed calculations on
holding periods ranging from one-month to three-years. We calculate abnormal returns in both
event-time and calendar-time because of the considerable debate in the literature about the
preferable technique for determining long-run abnormal performance. As Barber and Lyon
(1997) argue, the traditional event-time buy-and-hold abnormal returns (BHAR) precisely
measure investor experience of buy-and-hold investors, the contingent most common in the
value investing community as well as among VIC membership. However, Mitchell and Stafford
(2000) find that BHAR methods fail to account for cross-sectional dependence among firm
abnormal returns in event-time and advocate a calendar-time approach instead. Loughran and
Ritter (2000) further the debate, claiming that the calendar-time approach has low power to
detect abnormal performance associated with events that are clustered across time. In this paper,
we focus the discussion on results generated from the more statistically appealing calendar-time
approaches.
We incorporate CRSP delisting return data using the technique of Beaver, McNichols,
and Price (2007). These authors argue that the appropriate return to use for a delisted firm when
no delisting return is available is the mean delisting return of those that are available among
firms with the same three-digit delisting code. For instance, firms that delist as the result of
merger or acquisition have a much higher mean delisting return than those that delist as the result
of bankruptcy. After accounting for delisting returns, our abnormal return analysis accounts for
delisted firms in a similar fashion to Lyons, Barber, and Tsai (1999). If a firm is delisted, we
assume the proceeds of the delisted firms are invested in the control firm or benchmark-portfolio.

11

We also perform the analysis by assuming delisted firms proceeds are invested in the CRSP
value-weighted index, as well as by assuming delisted firms are eliminated from the database.
The results are similar under all techniques.

3.1.

Control-firm BHAR
The control-firm event-time BHAR methodology we use follows that of Lyon, Barber,

and Tsai (1999). The model is represented as

is the buy-and-hold abnormal return to firm i in period T,

where
month t, and

(1)

is firm is return in

, is the appropriate expected monthly return for firm i in month t. This

method allows our statistical inference to effectively control for the skewness bias in long-run
abnormal returns identified by Barber and Lyon (1997).
Following the methodology of Speiss and Affleck-Graves (1995), we assign each sample
firm a control firm based on size and book-to-market ratio. All firms in the CRSP/Compustat
universe are considered potential matches and from this universe we select as the control firm
that firm for which the sum of the absolute value of the percentage difference in size and the
absolute value of the percentage difference in book-to-market ratio is minimized. We define size
as the market value of equity on December 31 of the prior year and book-to-market ratio as book
value of equity at the end of the last fiscal quarter of the prior calendar year divided by size.
We calculate BHARs for each recommendation using monthly CRSP data, following the
advice of Brown and Warner (1985), who espouse the benefits of using monthly data rather than
12

daily data. The event period return data begin on the first of the month following the date the
recommendation was posted to the VIC community. For example, if an idea is posted on January
15, we start calculating BHARs on February 1. Because return data begins at the first of the
month following the date of the recommendation, which leaves up to 30 days for VIC members
to take positions, the abnormal returns presented likely underestimate the true returns earned by
VIC members and may bias our tests in favor of the null hypothesis that fund managers have no
stock-picking skills.
Table 3 presents the results of the control-firm BHAR analysis. Abnormal returns to long
recommendations are economically large and statistically significant. We find one-year BHAR
of 7.21%, two-year BHARs of 14.91% and three-year BHARs of 18.04%. The evidence from
the short recommendation sample, although directionally correct, suggests we cannot reject the
hypothesis that VIC members have no skill when shorting stocks. However, because the short
recommendation samples are small, we should not expect a rejection of the null hypothesis, since
any long-term abnormal return test lacks power in small samples (Ang and Zhang, 2004).
As a robustness test, we perform an alternate control-firm BHAR analysis. In these tests,
we further require that neither the size nor book-to-market ratio of the control-firm deviates from
that of the sample firm by more than 10%. This method ensures that sample firms examined are
assigned a control firm with very similar characteristics. The results from this analysis (not
shown) are similar to those presented in Table 3, which is consistent with the evidence presented
by Nekrasov, Shroff and Singh (2009) that the specific matching technology is immaterial to the
power of a control-firm test. We also perform the control-firm BHAR analysis after eliminating
the observations with the highest 1% and lowest 1% of abnormal returns to control for extreme

13

outliers seen in the BHAR scatter plots (see Figs. 1 and 2). The results (not shown) are similar to
those presented in Table 3.
[Insert Table 3]
[Insert Figure 1]
[Insert Figure 2]
In addition to standard t-test values, we also present results in Table 3 from a sign test as
per the recommendation by Ang and Zhang (2004), who conclude that the sign test coupled with
a control-firm approach is well specified and has the highest power for detecting long-term
abnormal returns among competing long-term event study methods. The conclusions from this
analysis are similar.

3.2.

Characteristics-based Benchmark-Portfolio BHAR

A shortcoming of the control-firm BHAR approach is that, in small samples such as our

sample of short recommendations, abnormal returns are very sensitive to mismatches between
sample and control firms. Savor and Lu (2009) observe that in small samples only a few control
firms need experience very large returns to make the mean abnormal return of the sample
negative, even if the majority of sample firms experience positive abnormal returns. A remedy to
this problem is the characteristics-based benchmark-portfolio BHAR approach, in which the
benchmark return is the return to a portfolio of stocks with characteristics similar to those of the
sample stock. Nonetheless, the use of benchmark-portfolios reintroduces the skewness bias
Barber and Lyon (1997) identify, which is mitigated under the control-firm BHAR approach.
Therefore, in the analysis of statistical significance for the benchmark-portfolio BHAR approach,

14

we account for event-time skewness bias by using the bootstrapping method Lyon, Barber, and
Tsai (1999) advocate.
To construct the benchmark-portfolios, we follow the characteristics-based benchmark
methodology of Daniel, Grinblatt, Titman, and Wermers (1997). We assign each stock in the
CRSP universe to one of 125 portfolios containing securities with similar size book-to-market
and momentum characteristics. We then define benchmark-portfolio adjusted BHAR as the
difference between the sample stock return and the benchmark-portfolio return, as in Eq. (1)
above.
The results of this analysis are presented in Table 4. The results are consistent with the
findings from the control-firm BHAR analysis. Using the benchmark-portfolio approach, we find
that the investors in our sample generate statistically significant one-year BHARs of 9.52%, twoyear BHARs of 19.03%, and three-year BHARs of 23.60%.
For short recommendations, both the control-firm BHAR approach and the benchmarkportfolio approach lead us to the conclusion that we cannot reject the null hypothesis after
adjusting for skewness in the test statistics. However, unlike the control-firm BHAR analysis for
short recommendations, the benchmark-portfolio BHARs are economically impressive: the oneyear BHAR is 5.15%, two-year BHAR is 18.02%, and three-year BHAR is 21.47%. Taken as a
whole, the analysis of the short recommendations using the various BHAR approaches provides
little statistical evidence that the investors in our sample are successful short sellers.
For robustness, we also perform the benchmark-portfolio BHAR analysis after
eliminating the observations with the highest 1% and lowest 1% of abnormal returns to control
for extreme outliers. The results (not shown) are similar to those presented in Table 4.

15

[Insert Table 4]

3.3.

Calendar-Time Portfolio Abnormal Returns

To assess the robustness of the results from the BHAR analyses, we analyze the data

using the calendar-time portfolio approach advocated by Mitchell and Stafford (2000) and Fama
(1998). Each month, the portfolios consist of all firms that were recommended in the current
month t, and within the last x months (where x is the length of the holding period). We then
calculate the monthly returns to the event-firm portfolio after adjusting for control-firm returns
or benchmark-portfolio returns. We next take the time series of monthly portfolio returns to
calculate a variety of relevant statistics.
We perform the analysis on equal-weight portfolios using both standard parametric and
non-parametric techniques. We also perform the analysis on value-weighted portfolios and find
less convincing evidence of stock-picking skill (results available upon request). However, the
value-weighted portfolio construction effectively decreases the sample size and statistical power
of our tests because of the bimodal distribution of the market capitalization of VIC
recommendations. Fig. 3 is a histogram of market capitalization. The figure shows that the vast
majority of observations are in the small-cap universe, but there is a spike in observations for
very large companies (>$9.5 billion). Thus, the value-weighted portfolio construction can create
portfolios that are essentially one observation. For example, in the long recommendation
portfolio event month of May 2008, General Electrica company with a $375 billion market
capitalization at the timewas an event firm along with eight other companies that had an
average market capitalization of $510mm, with a range of $117 million to $1.27 billion. For the

16

remaining time General Electric was included in the portfolio it was effectively the entire
portfolio. Because of this value-weighted portfolio construction issue we believe the equalweighted constructed portfolios are a more appropriate tool to assess the stock-picking skill
hypothesis in the context of our data. To address size related robustness concerns we use other
standard analytical tools, which maintain the statistical power of the analysis.
The results of the calendar-time portfolio abnormal return analyis are presented in Table
5. The estimates in Table 5 represent the mean (median) monthly abnormal return over the
calendar-time horizon for VIC recommendations. The estimates are consistent with the results of
the BHAR analysis and suggest that the investors in our sample have stock-picking skills. The
abnormal returns diminish the longer a recommendation is included in the portfolio, which is
evidence that the stock market is slowly incorporating the information identified earlier by VIC
members in to stock prices.
Panel B of Table 5 presents the results of portfolios formed from short recommendations.
The evidence weakly suggests that VIC members are successful short sellers, contrasting with
the results from the BHAR analysis which were inconclusive based on statistical inference.
Because the various abnormal return methods provide conflicting statistical evidence, we draw
no definitive conclusions regarding the short-selling ability of the investors in our sample. We
attribute the conflicting signals to the well-known properties of small sample long-term abnormal
return tests, which have low power to reject a false null hypothesis. More observations are
needed to test the hypothesis for stock-picking skill on the short side of the market.
[Insert Table 5]
Table 6 presents a series of robustness checks designed to evaluate how particular

17

characteristics affect the results. In general, the results are robust within a variety of subsamples
including those formed on the basis of market capitalization, liquidity/turnover, event-time
portfolio size, and those in which outliers have been eliminated. The one area in which this
analysis shows weaker robustness of our calendar-time abnormal return results is within the sizeconstrained sample in which market capitalization is required to be greater than $1 billion. The
benchmark-portfolio adjusted calendar-time abnormal returns are persistent, but control-firm
calendar-time abnormal returns vanish. This preliminary evidence suggests that VIC members
stock-picking skill may be limited to smaller companies in which the market is presumably less
efficient and their research efforts are more richly rewarded.
[Insert Table 6]

3.4.

Calendar-Time Portfolio Regressions

To assess the robustness of the results from the BHAR and calendar-time portfolio

abnormal return methodologies, we analyze the data using the calendar-time portfolio regression
approach. In a similar manner to the calendar-time abnormal return approach, we form portfolios
consisting of all firms that were recommended in the current month t, and within the last x
months (where x is the length of the holding period). We then calculate the monthly returns to
the event-firm portfolio in excess of the risk-free rate and regress this variable on a variety of
linear asset pricing models which include the following variables: MKT (excess value-weighted
market index return), SMB (small minus big), HML (high book-to-market minus low book-tomarket), MOM (high momentum minus low momentum), INV (low investment minus high
investment), ROA (high return on assets minus low return on assets), and LQD (traded liquidity

18

factor) (Fama and French, 1993; Carhart, 1997; Pastor and Stambaugh, 2003; and Chen, NovyMarx, and Zhang, 2010).5 Coefficients for the various linear factor asset pricing models are
presented in Table 7. The beta estimates suggest that VIC members typically recommend stocks
that load positively on size and value factors and negatively on the momentum factor.
[Insert Table 7]
We perform all analyses using the single factor market model (MKT), Fama French 3factor model (MKT, SMB and HML), Chen, Novy-Marx, and Zhang 3-factor model (MKT,
INV, and ROA), Carhart (MKT, SMB, HML, and MOM), and the 5-factor model (MKT, SMB,
HML, MOM, and LQD).

To obtain monthly alpha estimates we perform the regression

procedures using both portfolios constructed on an equal-weighted basis using OLS and
portfolios constructed on an equal-weighted basis using WLS to control for heteroskedasticity
issues (weights are the number of stocks in the portfolio in a given month).
The estimated alphas of these monthly regressions are presented in Table 8. The
estimates in Panel A of Table 8 represent the mean monthly abnormal return over the calendartime horizon for long recommendations. These statistically significant estimates range from
1.17% to 2.32% under the single-factor model specification and are consistent with the results of
our previous analyses, providing further evidence suggesting that the investors in our sample
have stock-picking skills.
[Insert Table 8]
Panel B in Table 8 presents the results of portfolios formed from short recommendations.
The evidence suggests that VIC members are successful short sellers, in contrast to the results
5

Factors obtained from Ken Frenchs website http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_librar

y.html, Lubos Pastors website http://faculty.chicagobooth.edu/lubos.pastor/research/liq_data_1962_2008.txt and
Long Chens website http://apps.olin.wustl.edu/faculty/chenl/linkfiles/data_equity.html.

19

from our earlier analyses which were inconclusive based on statistical inference. Because the
various abnormal return methods provide conflicting statistical evidence, we conclude the data
provide weak evidence for the short-selling ability of the investors in our sample.

3.5.

Robustness Tests
To better understand where VIC members are generating alpha, we divide the sample into

size, book-to-market and turnover quintiles and perform the calendar-time portfolio regression
analysis. For the sake of conservatism, we focus on the Fama-French three-factor model results
because simulation evidence suggests four-factor calendar time regression results too often overreject the null hypothesis of market efficiency (Ang and Zhang, 2004). For robustness, we also
perform our tests using the alternative asset pricing models and find that the results are more
compelling, but qualitatively similar to the three-factor results.
Tables 9 through 11 present the quintile analysis results. These results provide a more
transparent view of how and from where VIC members derive benefits from private information
collection. Not surprisingly, VIC members find the most alpha in smaller and more illiquid
stocks. Alpha estimates for both the size and turnover quintiles generally decrease
monotonically. In contrast to size and turnover, there is no clear indication that VIC members
find value in a particular book-to-market quintile. The evidence suggests the members certainly
have skill in value stocks, but there is also evidence that they have skill in the stocks tilted more
towards growth.
[Insert Table 9]
[Insert Table 10]

20

[Insert Table 11]

Figs. 1 and 2 show the scatter plots of BHARs for long and short recommendations
plotted over time. The plots suggest recommendations tend to cluster in December. According to
VIC management, the reason we see more recommendations in December is because members
must submit at least two recommendations per calendar year in order to fulfill their membership
duties. Often, members procrastinate until the end of the year to fulfill their requirement.
Because many of these recommendations in December may be submitted due to time
constraints and may be less thorough, a reasonable hypothesis is that the abnormal returns should
be stronger once ideas in December are eliminated. We perform this analysis using the controlfirm BHAR, benchmark-portfolio BHAR, calendar-time abnormal returns, and calendar-time
portfolio regression approaches, and we find that the results are essentially the same with or
without the December recommendations. Fig. 4 shows this result graphically for the set of long
recommendations.
[Insert Figure 4]

3.6.

Performance Analysis Discussion

Regardless of a researchers preference for BHAR methods, calendar-time portfolio

abnormal returns, or calendar-time portfolio regression approaches, all three methods presented
in this study provide robust evidence that VIC members are successful long-only investors. The
results of the various methods we use are statistically mixed with respect to VIC members
ability to successfully short sell stocks. The calendar-time portfolio regressions hint that VIC
members have stock-shorting skills, however the BHAR and calendar-time abnormal return

21

analysis is indeterminate. Our overall conclusion is that VIC members appear to have stockpicking skills for buy recommendations, but the evidence for short-selling skill is weak.
A potential criticism of VIC members recommendations is that these ideas are not
implementable. This concern is likely unwarranted. First, the quantitative evidence in this paper
suggests that the results are robust to the addition of liquidity-related factors. Second, VIC is not
designed as a means for fund managers to showcase their ability to write research reports on
opportunities that cannot be implemented. Because reports are submitted under pseudonyms,
reputation incentives are presumably negligible, making reports for extremely illiquid names a
waste of both the authors and the membership's time. In fact, VIC has specific guidelines
pertaining to the liquidity of investment recommendations submitted: Small market
capitalization ideas are fine, but as a general guideline, at least $250,000 worth of securities
should trade on an average week. We understand that it is much more difficult to identify a
compelling idea with $1billion of market capitalization, than one with $10mm of market
capitalization and we take that into consideration when reviewing applications.
A critic may also be concerned that the full sample of VIC recommendations, without
controlling for the quality of the recommendations, may bias the results in favor of the null
hypothesis that investors have no stock-picking skill. For example, if a member submits a very
low-quality idea because he was under time constraints, made mistakes in his analysis, or simply
had no good ideas at the time, this idea may bias the results even though the VIC member
submitting the idea, and the broader VIC community, can recognize the idea is not high-quality.
Eliminating the procrastination ideas in December was an initial attempt to address this
concern. The following section explores the relation between quality and performance in detail.

22

4. The Relation between VIC Ratings and Abnormal Returns

All VIC recommendations are not created equal. On September 14, 2008, the member
agape1095 posted a buy recommendation for Lehman Brothers. The reports thesis was based
on a serious mistake in the writers analysis. The idea was given a rating of 1.3 by the VIC
communitythe worst rating in the history of VIC and more than five standard deviations below
the mean for the entire sample. Then, on September 15, 2008, a VIC member posted a comment
pointing out that the company had already filed bankruptcy. Agape1095 quickly replied, I
didnt know Lehman was already bankrupted when I posted this. And this report totally deserves
the low rating.
The example above highlights why analyzing the full sample of VIC recommendations
may misrepresent the skill of the majority of the investors in our sample. Although VIC
membership is difficult to attain, it remains difficult for the organization to completely screen out
poor performers with certainty. To address this concern, we analyze how recommendations
perform after controlling for the quality of the idea, as measured by the VIC community rating
on individual investment theses.
When a report is posted to VIC, members are given the opportunity to rate it on a scale of
1 (bad) to 10 (good). Ratings are recorded if five or more members rate the idea, and the rating
period is open for only two weeks to ensure members do not rate ideas based on ex-post
performance. Since 2007, which is when data on the time of rating became available, 60% of
ratings were submitted within 72 hours of posting. The clubs guidance for ratings is that they
should be objective and based purely on the quality of the investment thesis. Moreover, to

23

encourage active participation, the club requires members to rate at least 20 ideas per year. The
club also requests that extremely high (9 or 10) or extremely low (1 or 2) ratings be accompanied
by some specific commentary about the investment thesis.
With the data on VIC ratings, we can perform additional tests for stock-picking skill
among VIC members. In this analysis, we assume ratings approximate how favorably (or
unfavorably) the VIC community believes the stock will perform in the future. To test whether
VIC members can identify the best and worst recommendations within their universe of ideas,
we estimate a simple model such that a linear relation exists between abnormal returns and the
VIC community rating. The model is represented as
,
where

(2)

is the abnormal return to stock i from t=2 to t=h (h is holding period), and
is the VIC members rating of the particular stock i. As the dependent variable we use

both the benchmark-portfolio adjusted BHAR as well as the control-firm adjusted BHAR. We
calculate the dependent variable from t=2 to t=h to avoid the endogeneity that may arise in a
model relating ratings with BHARs that encompass the two-week rating period. That is, the
rating may be endogenenously determined should an idea perform exceptionally well during the
two-week rating period inducing members to rate it very highly. For example, if stock XYZ is
recommended on June 20 and performs exceptionally well through July 3, members may rate the
idea extremely favorably on July 3 not because they believe it will outperform in the future, but
because it has performed well thus far.
We run Fama-Macbeth regressions to test the predictive content of ratings for abnormal
returns. For each month t, we estimate a cross-sectional regression of the abnormal returns on
24

the rating of the recommendation as well as controls for size, book-to-market ratio, prior sixmonth return, and turnover. We repeat this process for all periods to produce a set of T
coefficient estimates (subject to the constraint that there is at least 10 observations in a given
month). We then average the T estimates to get Fama-MacBeth coefficient estimates and tstatistics, which are robust to cross-sectional correlation.
The results from the Fama-MacBeth regressions presented in Table 12 provide weak
evidence that VIC members have an ability to identify the best recommendations posted to the
website as measured by ex-post BHAR. Point estimates for

are positive and statistically

significant for the regression specifications of the benchmark-portfolio BHAR +2 to +6 on

rating. The effect diminishes for control-firm BHARs and longer dated BHARs.
[Insert Table 12]
Because of the limited power of the Fama-MacBeth regression analysis, we further
investigate how ratings are related to abnormal returns by analyzing the abnormal returns within
quintiles formed on the basis of rating. In this analysis, we create calendar-time portfolios for
each rating quintile. We then regress the calendar time portfolio returns for a given rating
quintile on the returns to the mimicking portfolios of Fama and French (1993), where the
calendar time portfolios are formed by assigning all firms that were recommended in month t and
within the last x months (where x is the length of the holding period) to the calendar time
portfolio. We present the alpha estimates of this procedure in Table 13. Alpha estimates for the
highest rating quintile are large and statistically significant at the 5% level. Over one-year
holding periods, the firms in the highest rating quintile have a 3.42% (t=3.14) average monthly
alpha, whereas firms in the lowest rating quintile have a 0.49% (t=0.85) average monthly alpha.
25

This evidence suggests that VIC members are able to identify ex-ante which VIC
recommendations will perform the best.
Next we investigate the ability of ratings to predict returns by examining the difference in
mean BHAR between the highest and lowest rating quintile. Inference is based on both a
students t test and sign test. We present the results in Table 14. They provide strong evidence
of a significant difference between the mean control-firm adjusted BHAR of the highest and
lowest rated submissions. The mean one-year BHAR among the highest rating quintile is
21.69%, whereas the mean one-year BHAR to the lowest rating quintile is -0.16% (see Fig. 5).
This is further evidence that VIC members can distinguish between good and bad
recommendations.
[Insert Figure 5]
To summarize, we find strong evidence from both the calendar-time and the BHAR
analysis that VIC members have an ability to distinguish between good ideas and bad ideas.
This is further evidence in support of our stock-picking skill hypothesis; however, this analysis
reveals that VIC members are successful stock pickers not only with regard to their own ideas,
but in their evaluation of other members ideas as well. The ability to distinguish between good
and bad ideas is clearly a manifestation of stock-picking skill and suggests that at least some
skilled managers exist in the investment management industry.
[Insert Table 13]
[Insert Table 14]

5. Why do Managers Share Profitable Ideas?

26

VIC is an organization explicitly designed to facilitate the sharing of private information

among fund managers. However, the mere fact that VIC members are sharing this valuable
private information at all is puzzling. Traditional theories (Friedman, 1953) suggest that
arbitrageurs with valuable private information should take full advantage of the information
advantage until prices reflect fundamental values. Moreover, in a market with efficient funds
allocation, competing arbitrageurs should keep their valued information private so they can
outperform their competition and thus attract more investor capital (Stein, 2008).
Such theories compellingly suggest that rational agents will not share private information,
but few theories explain why rational agents do share private information in the asset
management industry. Stein (2008) suggests managers might share information because they can
get valuable feedback that improves their ideas (collaboration argument). Gray (2010) shows
that a resource-constrained arbitrageur will share profitable ideas with his competition because
doing so allows him to diversify his portfolio among a group of arbitrage trades. The benefits of
sharing come from the fact that diversification lowers the probability the arbitrageur will
experience a large negative noise trader shock in the short run and have his funds withdrawn by
his investors (diversification argument). Finally, Dow and Gorton (1994) suggest arbitrageurs
will only make investments if they believe subsequent arbitrageur demand will push the asset
price higher (arbitrage chains). In the Dow and Gorton model, arbitrageurs are unable to
reliably expect another arbitrageur to push asset prices further, and market prices end up being
inefficiently priced. Grays theory abstracts from the Dow and Gorton model and suggests that
one obvious way arbitrageurs can help ensure other arbitrageurs will take a position in an asset is
by sharing private information (awareness argument). Practitioners refer to this practice as

27

talking up your own book.

5.1.

Collaboration Argument
Steins theory of information exchange between competitors suggests that an asset

manager will share his idea if it gives him access to constructive feedback that will make his idea
more valuable. For example, fund manager X has developed a promising investment thesis, but
his information set is incomplete so his idea is not worth much; however, by sharing his thesis
with fund manager Y and receiving feedback, his investment thesis will become more valuable.
As long as this give-and-take relationship is valuable for both parties involved, information
exchange will occur between competitors. Steins theory provides three basic predictions: (i)
managers will share ideas in situations in which they receive constructive feedback, (ii) lower
value ideas will be shared among a larger group of collaborators, and (iii) the most valuable
ideas will remain localized within a small group.
Anecdotal evidence from VIC supports Steins prediction that managers will share ideas
when they can expect to receive constructive feedback. For example, on October 7, 2009,
Seahawk Drilling was recommended as a long by user ronmexico. Over the next two days,
eight VIC members posted various comments relating to the investment thesis. On October 8,
2009, a detailed comment of more than 3,000 words entitled disagree with some of the
analysis by user ruby831 outlined the detailed short thesis for Seahawk Drilling. After some
heated discussion between ronmexico and the VIC community, user ad188 came to the
following conclusion on October 9, 2009: Excellent writeup, better Q&Aproves that VIC is
worth the effort, as this would have taken me a week on my ownmy conclusion [after] reading

28

this is that HAWK [Seahawk Drilling] is not a long, at any pricehowever, with no debt, it
doesnt seem that it is a short either. This vignette certainly suggests that one reason VIC
members are sharing information is to receive valuable feedback to help develop their own ideas.
To quantitatively assess Steins primary hypothesis in more depth we analyze the more
than 40,000 comments attached to VIC recommendations. VIC has a robust infrastructure to
facilitate collaboration and comments on individual ideas. Whenever an idea is posted to VIC,
members receive an idea alert and are able to share their comments and thoughts on the
investment thesis. Another feature of VIC is the private comment function. These comments
are only visible to the VIC community and are not accessible by the general public (anyone can
sign up for guest access to VIC, but access comes with a 45 day delay). For example, if VIC
member stockpicker posts an idea on January 1, 2008 and another VIC member makes a
comment on the idea that he designates as private, then after February 14, 2008, all VIC
members will still be able to view the private message, but anyone from the general public who
is reading stockpickers investment thesis and following the comments will not have access to
the comments designated as private.
Table 15 provides a more detailed description of the comments from VIC. We analyze
the comments for the sample of observations with the necessary data to perform the control-firm
analysis (results are very similar for other samples). In total we examine the comments on 1869
observations: 1671 long recommendations, and 198 short recommendations. We tabulate the
total number of comments submitted, the number of unique VIC members involved in a
particular conversation, the number of comments that are designated as private, the number of
comments that are author submitted, and the number of comments that are submitted within 45

29

days of the recommendations posting.

Summary statistics certainly suggest that ideas submitted to VIC receive plenty of
feedback. Over 91% of the recommendations receive at least 1 comment, and recommendation
receives 12.03 comments on average. Author comments represent 43% of the total comments
submitted for a particular idea, which suggests that the conversational, give-and-take nature of
the comments between author and VIC members, fits the primary prediction of Steins
collaboration theory, that managers share their ideas to receive feedback.
We next test Steins other hypotheses: (i) less valuable ideas will be shared among a
larger group of agents, and (ii) more valuable ideas will be shared among a smaller group of
agents. To assess these hypotheses we use the percentage of total comments identified as
private, as a proxy for the size of the collaboration group. For example, if idea XYZ has 20
comments and 15 are private, the feedback information for idea XYZ will be primarily limited to
VIC members, whereas, if idea ABC has 20 comments and 0 are private, the feedback
information is available to VIC members and the general public after 45 days.
We use the rating assigned to an investment recommendation as a proxy for the perceived
value of an idea. We then divide the sample into quintiles formed on the percentage of total
comments marked private. Table 16 presents the summary statistics and tests for differences in
means and medians between the quintile of ideas with the lowest percentage of comments, and
the quintile with the highest percentage of comments. The p-values associated with the t-test for
differences in means and the Wilcoxson rank-sum test for differences in medians are significant
at the 1% level. The evidence supports Steins hypotheses that highly valued ideas will be shared
with fewer people than lower valued ideas; the mean (median) rating for the quintile of ideas

30

with the lowest percentage of private comments is 4.89 (5.00) versus 5.14 (5.20) for the ideas
with the highest percentage of private comments. Admittedly, while the evidence from this test
generally supports Steins prediction, it is not entirely persuasive.
To further test Steins hypothesis, we use the rating assigned to an investment
recommendation as a proxy for the perceived value of an idea. We then estimate the parameters
for the model
%

where %

is the percentage of total comments marked private and

(3)
is the rating of

stock i. We estimate this model using data from January 1, 2004 to December 31, 2008 because
the option to label comments private was rarely used prior to January 1, 2004 (10.01% of ideas
had at least one private comment prior to 2004 versus 74.64% after January 1, 2004). Panel B in
Table 16 shows ordinary least squares coefficient estimates as well as the maximum likelihood
coefficient estimates from a logit regression model.
The regression estimates are mixed in their support for Steins hypotheses. The estimated
slopes from the OLS regressions suggest there is a positive linear relation between the quality of
an idea (as proxied by rating) and the distribution of the idea (as proxied by the percentage of
private comments); however, the logit regression estimates, while directionally correct, are not
statistically significant from zero.

5.2.

Diversification Argument
If an arbitrageur is endowed with only a few great ideas in each time period, he will face

difficult decisions: Does he invest all his assets under management in his handful of ideas and
31

expose his business and investors to extreme noise trader risk? Or should he couple his few good
ideas with a diversified index of efficiently priced assets and dilute his performance? Grays
theory shows that a third option is possible for arbitrageurs. Specifically, Gray finds that in a
world in which investors simply focus on past returns as a rough proxy for arbitrageur skill
(Shleifer and Vishny, 1997), arbitrageurs can share profitable ideas with the competition because
doing so allows the arbitrageurs to diversify their portfolios among a group of arbitrage trades,
which allows them to decrease their portfolio volatility, while at the same time, keeps them from
diluting their performance. In addition to the basic prediction that constrained arbitrageurs will
share private information, Grays model is specific about the situations in which information
exchange will occur. His model predicts that managers will share profitable ideas when (i) they
have limited research resources and they are capital constrained, (ii) noise trader risk is high (i.e.,
market participants can drive prices from fundamentals), and/or (iii) arbitrage fund investors
have a high propensity to withdraw funds following poor performance.
To test Grays hypotheses that sharing will occur when managers have limited research
resources and are capital constrained, we use a firms assets under management as a proxy for
their research and capital constraints. That is, we assume smaller firms have more constraints
and larger firms have fewer constraints. To address this hypothesis, we analyze basic data from
VIC and more detailed data from Sumzero.com. Sumzero.com is another exclusive buy-sideonly information sharing network similar to VIC, yet it releases more detailed information on the
characteristics of the 815 unique buy-side funds that make up the membership of the
organization (as of September, 2009).
We find evidence that the funds sharing ideas on both VIC and Sumzero.com are

32

predominately small. Specific data on the investor profiles of VIC members is confidential and
cannot be disclosed. However, VIC management agreed to disclose that VIC members are
almost exclusively small- to mid-size hedge funds ($50 million to $250 million assets under
management). For more concrete data, we analyze the profile of asset managers who share ideas
on Sumzero.com. Similar to VIC, Sumzero.com members are affiliated with funds that are
overwhelmingly small (over 53% have less than $250mm assets under management). Fig. 6
shows the distribution of assets under management (AUM) for managers who share ideas
through Sumzero.com.
[Insert Figure 6]
We next test the hypothesis that managers will share if they hold assets with high noise
trader risk: we find preliminary evidence in support of this prediction. VIC recommendations are
concentrated among investments thought to have higher noise trader risk such as small
capitalization stocks, merger arbitrages, stub arbitrages (Mitchell, Pulvino, and Stafford, 2002),
and pairs/twin arbitrage (Froot and Dabora, 1999). Specifically, we find that the long ideas
submitted to VIC are recommendations for small capitalization stocks (median market
capitalization is $393mm) or special situations such as stub and pair arbitrages, liquidations, and
spin-offs in relatively illiquid markets (10.33% of ideas submitted). We find similar results for
the submissions on Sumzero.com. Of the ideas submitted to the site, 15.5% are categorized as
event-driven or special situations, and the median market cap for long equity recommendations
is $559mm.
For more evidence that sharing managers trade high noise trader risk assets, we analyze
the institutional holdings of VIC stocks. If VIC stocks are dominated by individual investors,

33

who are presumably noise traders, institutional holdings for VIC stocks should be small. We
examine institutional ownership data from the Thomson Reuters Institutional Holdings database.
This data source compiles the number of outstanding shares held by institutions for individual
firms. The data are compiled from all SEC form 13(f) filings and are reported quarterly (March,
June, September, and December). We then use CRSP price and shares outstanding data to
calculate the percentage of shares outstanding held by institutions for a given firm. Similar to
Chung and Zhang (2009), we exclude observations with missing variables or obvious data errors
(i.e., institutional ownership greater than 100% of shares outstanding) and winsorize percent
holdings at the 1st and 99th percentile to reduce the influence of extreme observations and
possible data errors. Finally, we perform a paired t-test for unequal variances to test for
differences in means and the Wilcoxson signed rank test to test for differences in medians
between the lowest quintiles and the highest quintiles.
Table 18 summarizes institutional holding for the nearest quarter for the sample of
investment recommendations submitted to VIC. In total, we have are 1546 observations with
institutional data. In order to assess how institutional ownership is related to key characteristics
of VIC recommendations, we present results for various quintiles related to size, B/M, and 12-,
24-, and 36-month control-firm BHARs. Average institutional ownership in the nearest quarter
for VIC recommendations averages 53.13% of outstanding shares and varies widely by quintile.
Chung and Zhang (2009) report that over the 2001 to 2006 period institutions held, on average,
56.31% of the shares outstanding of all firms in the Thomson Reuters Institutional Holdings
database. They also find that the largest 25% of stocks have average institutional holdings of
79.48%, whereas the average within the quintile of the largest VIC recommendations is 70.47%.

34

Thus, relative to institutional holdings of stocks in general, and large stocks in particular,
institutional ownership of VIC ideas is relatively small. If the level of noise traders and
institutional ownership are inversely related, the evidence weakly supports the notion that
arbitrageurs will only share ideas when there is high noise trader risk.
The initial evidence suggests that managers are sharing ideas within high-noise trader
asset classes. Empirical evidence shows that the ideas submitted to both VIC and Sumzero.com
are concentrated in smaller stocks and special situation investments, which are typically
thought to have higher noise trader risk than other asset categories. Moreover, the empirical
analysis of institutional holdings weakly suggests that VIC firms are sharing ideas with a higher
proportion of noise traders than the typical stock in the investment universe.

5.3.

Awareness Argument
A key insight of the Dow and Gorton (1994) analysis of arbitrage chains is that short-

horizon arbitrageurs will only make investments if the probability of another arbitrageur ( )
subsequently entering the market is high enough. If

is too low, arbitrageurs will not take an

immediate position in a long-horizon arbitrage because the price will not be supported in
subsequent periods and the arbitrageur will be exposed to various transaction costs. Although
is fundamental to the analysis of arbitrage chains, there is little discussion about the origins of
and it is assumed to be exogenous. However, Gray suggests arbitrageurs might endogenously
increase the chances of future arbitrageurs coming into the market. One way arbitrageurs can
help ensure other arbitrageurs take a position in an asset is by providing awareness of their
investment thesis. Promotion on the basis of no information is unlikely to convince other smart
35

investors to take a position in a particular asset; however, if investors share their private
information, which can subsequently be verified by another arbitrageur, they can likely convince
other arbitrageurs the idea is profitable. A distinguishing aspect of awareness sharing is that the
arbitrageur shares his private information after he has already taken a full position in an asset.
Awareness sharing is likely one of the reasons investors share ideas on both VIC and
Sumzero.com. In fact, at the 45th day after posting, VIC releases all their investment
recommendations and analysis (except for comments marked as private) to the general public,
which is an explicit attempt to awareness share. Moreover, anecdotal evidence from a few of the
write-ups submitted to VIC suggests the member is sharing after he has taken a full position. For
example, a VIC member who recommended purchasing Aavid Thermal Technologies 12.75%
Senior Subordinate Notes states in his December 31, 2002 write-up, Self-interest precluded me
from posting the idea [earlier] because the bonds are fairly illiquid and it takes a few months to
build a position.
While there certainly awareness sharing occurring after the 45th day after posting, it is
unclear whether awareness sharing is the only dimension of the sharing decision during the 45day period that VIC keeps recommendations private. Fortunately, one prediction from the
awareness sharing theory is that a manager who awareness shares will exchange his private
information with as many arbitrageurs as possible, as long as the transactional costs of sharing
his private information are negligible. This prediction contrasts with the predictions of the
collaboration and diversification theories of information exchange, which suggest managers will
keep their private information sharing limited to smaller groups. Therefore, if managers are
engaging in awareness sharing within the 45-day period that VIC keeps ideas private, as opposed

36

to sharing for collaboration or diversification reasons, we should see a significant overlap in

ideas submitted to both VIC and Sumzero.com (the null hypothesis would be 100% of ideas
would be overlapping, because it is costless to post an idea to Sumzero.com after posting the idea
to VIC). That is, if the sharing arbitrageur is trying to share his private information to the largest
possible audience of sophisticated investors, rather than limiting his idea to an exclusive venue
like VIC, he will seek to share his idea on Sumzero.com in addition to VIC.
We find that during the ten-month overlap period between the Sumzero.com and the VIC
database (March 1, 2008 through December 31, 2008), 4.17% of the 456 ideas submitted on VIC
are also submitted on Sumzero.com within fifteen days. Of the nineteen overlapping idea
submissions to both VIC and Sumzero.com only seven are actually submitted simultaneously.
This evidence rejects the awareness sharing null hypothesis during the forty-five day period in
which the VIC recommendations are closed to the general public and suggest that VIC members
are also using the site for collaboration and diversification benefits.
Another unique prediction of the awareness sharing theory is that large arbitrageurs will
join sharing networks, but will not share ideas. The role of the large arbitrageur in the awareness
sharing framework is simply to provide capital for arbitrage opportunities revealed by capitalconstrained arbitrageurs. The situation is a win-win for all parties involved: capital constrained
arbitrageurs win because they attract additional capital to their arbitrage situation, thus lowering
the probability of a liquidation in the event of a noise trader shock, and large arbitrageurs win
because they get access to arbitrage opportunities.
There is evidence to support the hypothesis that large funds will be members of private
information groups, but will not share. Fig. 6 shows that just under 5% of the fund population for

37

Sumzero.com have over 20 billion in assets under management. Smaller funds submit 2.04 ideas
per fund on average, whereas the largest funds submit 1.19 ideas on average; however, because
Sumzero.com requires that members submit at least one idea a year, the marginal contribution of
ideas above the mandate for small funds is 1.04 a year versus .19, or approximately zero, for the
largest funds. The evidence in support of the hypothesis that large funds will not share is thin,
but generally consistent with the idea that smaller funds will be the primary information sharers,
and large funds will only provide arbitrage capital.
Overall, it is difficult to make an overarching statement with respect to the prevalence of
awareness sharing and how it is used in practice by investors. Intuitively, one would suspect that
awareness sharing is the only benefit investors care about when they share their private
information with other sophisticated investors. Nonetheless, the empirical evidence shows only a
small percentage of ideas submitted to VIC are actually shared with a broader audience, which
suggests VIC members engage in limited awareness sharing. However, the empirical evidence
from Sumzero.com does support the awareness sharing prediction that large funds will join
information sharing groups, but their participation will be limited.
It appears that VIC takes a hybrid approach to awareness sharing: within the 45-day
window, before the organization releases their recommendations to the public, members refrain
from broad awareness sharing and likely capture the benefits from collaboration and
diversification sharing. However, after the 45-day window VIC explicitly engages in awareness
sharing by giving broad public access to its research. Sumzero.com appears to take a different
approach. Membership of Sumzero.com is much broader in nature, so it is likely that the
investors involved in this organization are primarily sharing their research to gain the benefits

38

from awareness sharing.

5.4.

Conclusions
The empirical and anecdotal evidence from VIC and Sumzero.com generally support the

predictions of the collaboration, diversification, and awareness theories of private information

exchange. We cannot reject that members of VIC and Sumzero.com are using these networking
sites to develop their own theses, create awareness of opportunities in which they have a
position, and to get access to a pool of ideas that allows them to invest in a broader set of alphaproducing opportunities.
The next step in the research process would be to develop sharing models that incorporate
all three sharing theories and determines how organizations will optimally behave. A good start
for this research is the VIC model, which appears to be an organization that has made a first
attempt at maximizing the benefits of sharing private information. VICs approach can be
summarized as follows: (1) an individual constrained agent identifies private information, (2) the
agent takes an appropriate position such that internal risk management and investment mandates
are satisfied, (3) the agent promotes the position to other arbitrageurs in VIC, generating the
benefits of awareness, (4) the agent collaborates with other agents to receive constructive
feedback on the idea, then adds to or subtracts from his current position accordingly, (5) the
agent can also diversify his portfolio among the good ideas of other investment managers in VIC,
and (6) after forty-five days the ideas are released to the general public to capture additional
awareness sharing gains.

39

6. Conclusion
With our database, which is free from many of the biases found in databases other
researchers analyze, we address two basic economic questions: (1) Do professional money
managers have stock-picking skill? And, (2) why do they share their good ideas with their
competition?
With respect to question (1), the evidence suggests the fund managers in our sample have
stock-picking skills for long recommendations although the results for short recommendations
are less conclusive. These results should not be completely surprising. The recommendations we
analyze are well researched and required costly resources to develop. In equilibrium, skilled
investors should be compensated for their efforts in accurately analyzing firms and driving assets
to fundamental value as follows from the theory of Grossman and Stiglitz (1980).
To address question (2), we test the various predictions from the collaboration,
diversification, and awareness theories of information exchange. We find that the investors in
our sample appear to be sharing profitable ideas in order to increase awareness, which may
mitigate inefficiencies in the securities pricing. The investors also appear to use the VIC
platform to receive constructive feedback on their analyses and gain insight on a more diverse
collection of securities than they would be able to analyze if working alone. Overall, these
findings suggest that the mechanisms through which information flows into security prices are
not as simple as traditional asset pricing models would suggest.
In conclusion, this study brings into question the broader concepts of market efficiency in
the stock market and the asset manager market. It provides evidence that some investors have
skill to identify undervalued securities and are willing to share their valuable insights with their

40

competition.

7. References
Ang, James and Shaojun Zhang, 2004. An evaluation of testing procedures for long horizon
event studies. Review of Quantitative Finance and Accounting 23, 251-274.
Asness Cliff, Robert Krail, and John Liew, 2001. Do hedge funds hedge? Be cautious in
analyzing monthly returns. Journal of Portfolio Management 28, 6-19.
Barber, Brad, and John Lyons, 1997. Detecting long-run abnormal stock returns: the empirical
power and specification of test statistics. The Journal of Finance 43, 341-372.
Barber, Brad, Reuven Lehavy, Maureen McNichols and Brett Trueman, 2001. Can investors
profit from the prophets? Security analyst recommendations and stock returns, The
Journal of Finance 56, 531-563.
Barker, Robert, 2001. This message board may really light up. Businessweek,
http://www.businessweek.com/archives/2001/b3722161.arc.htm, accessed October 30,
2008.
Beaver, William, Maureen McNichols, and Richard Price, 2007. Delisting returns and their
effect on accounting-based market anomalies. Journal of Accounting and Economics 43,
341-368.
Berk, Jonathan and Richard Green, 2004. Mutual fund flows and performance in rational
markets. Journal of Political Economy 112, 1269-1295.
Brown, Stephen and Jerold Warner, 1985. Using daily stock returns: the case of event studies.
Journal of Financial Economics 14, 3-31.
Carhart, Mark, 1997. On persistence in mutual fund performance. The Journal of Finance 52, 5782.
Chen, Long, Robert Novy-Marx, and Lu Zhang, 2010. An Alternative Three-Factor Model.
Unpublished working paper. University of Chicago.
Cohen, Randolph, Christopher Polk, and Bernhard Silli, 2009. Best Ideas. Unpublished working
paper. Harvard University.
Chung, Kee and Hao Zhang, 2009. Corporate Governance and institutional Ownership. Journal
of Financial and Quantitative Analysis, forthcoming.
41

Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997. Measuring mutual
fund performance with characteristics based benchmarks. The Journal of Finance 52,
1257-1274.
Desai, Hemang and Prem C. Jain, 1995. An analysis of the recommendations of the superstar
money managers at Barrons annual roundtable. The Journal of Finance 50, 1257-1273.
Dow, James and Gorton, Gary, 1994. Arbitrage chains. The Journal of Finance 49, 819-849.
Fama, Eugene, 1998. Market efficiency, long-term returns, and behavioral finance. The Journal
of Financial Economics 49, 283-306.
Fama, Eugene and Kenneth French, 1993, Common risk factors in the returns on stocks and
bonds, The Journal of Financial Economics 33, 3-56.
Fama, Eugene and Kenneth French, 2010. Luck versus skill in the cross section of mutual fund
alpha estimates. The Journal of Finance 65, 1915-1947.
Friedman, Milton, 1953. The case for flexible exchange rates, in: Essays in Positive Economics.
University of Chicago Press, 157203.
Froot, Kenneth and Emil Dabora, 1999. How are stock prices affected by the location of the
trade. The Journal of Financial Economics 53, 189-216.
Fung, William and David Hsieh, 2000. Performance characteristics of hedge funds and
commodity funds: Natural vs. spurious biases. Journal of Financial and Quantitative
Analysis 35, 291307.
Fung, William and David Hsieh, 2001. The risk in hedge fund strategies: Theory and evidence
from trend followers. Review of Financial Studies 14, 313341.
Getmansky, Mila, Andrew Lo, and Igor Makarov, 2004. An econometric model of serial
correlation and illiquidity in hedge fund returns. Journal of Financial Economics 74, 529
609.
Gray, Wesley, 2010, Why do fund managers share valuable information with their competition?
Unpublished working paper. University of Chicago.
Griffin, John and Jin Xu, 2009. How smart are the smart guys? A unique view from hedge fund
stock holdings. The Review of Financial Studies 22, 2531-2570.
Grossman and Stiglitz, 1980. On the impossibility of informationally efficient markets.
American Economic Review 70, 393-408.

42

Liang, Bing, 2003. The accuracy of hedge fund returns. The Journal of Portfolio
Management 29, 111-122.
Loughran, Tim and Jay R. Ritter, 2000. Uniformly least powerful tests of market efficiency.
Journal of Financial Economics 55, 361-389.
Lyon, John, Brad Barber, and Chih-Ling Tsai, 1999. Improved methods for tests of long-run
abnormal stock returns. The Journal of Finance 54, 165-201.
Malkiel, Burton, 1995. Returns from investing in mutual funds 1971-1991. The Journal of
Finance 50, 549-572.
Mitchell, Mark and Erik Stafford, 2000. Managerial decisions and long-term stock performance.
Journal of Business 73, 287-329.
Mitchell, Mark, Todd Pulvino, and Erik Stafford, 2002. Limited arbitrage in equity markets. The
Journal of Finance 57, 551-584.
Nekrasov, Alexander, Pervin Shroff, and Rajdeep Singh, 2009. Tests of long-term abnormal
performance. Unpublished working paper. University of California at Irvine.
Pastor, Lubos and Robert Stambaugh, 2003. Liquidity risk and expected stock returns. Journal of
Political Economy 111, 642-685.
Savor, Nejat and Qi Lu, 2009. Do stock mergers create value for acquirers? The Journal of
Finance 64, 1061-1098.
Shleifer, Andrei and Robert Vishny, 1997. The limits of arbitrage. The Journal of Finance 52,
35-55.
Spiess, D.K., Affleck-Graves, J., 1995. Underperformance in long-run stock returns following
seasoned equity offerings. Journal of Financial Economics 38, 243-267.
Stein, Jeremy, 2008. Conversations among competitors. American Economic Review 98, 21502162.

43

Figure 1

Figure 1: Scatter plot of long recommendation one-year control-firm and benchmark-portfolio BHAR. This figure represents a scatter plot of sample firm
BHAR estimates. The Y-axis represents the abnormal return. The X-axis represents time.

44

Figure 2

Figure 2: Scatter plot of short recommendation one-year control-firm and benchmark-portfolio BHAR. This figure represents a scatter plot of individual
sample firm BHAR estimates. The Y-axis represents the abnormal return. The X-axis represents time.

45

Figure 3
Recommendations by Market Capitalization
Long Recommendations

Short Recommendations

25.00%

Probability

20.00%

15.00%

10.00%

5.00%

more

4550

4400

4250

4100

3950

3800

3650

3500

3350

3200

3050

2900

2750

2600

2450

2300

2150

2000

1850

1700

1550

1400

1250

950

1100

800

650

500

350

50

200

0.00%

Market Capitalization (Millions)

Figure 3: VIC recommendations by market capitalization. This figure represents the histogram of market capitalizations for the sample of firms with at least
one monthly return observation. The Y-axis represents the probability. The X-axis represents market capitalizations. There are 1959 long recommendations and
242 short recommendations.

46

Figure 4

Figure 4: BHAR estimates for +1 to +36 months with and without December observations. This figure represents BHAR over time. The Y-axis represents
the BHAR. The X-axis represents the holding period in months.

47

Figure 5

Figure 5: BHAR estimates for +1 to +36 months by rating (1=high, 5=low). This figure represents BHAR over time. The Y-axis represents the BHAR. The
X-axis represents the holding period in months.

48

Figure 6
Sumzero.com Fund Manager AUM Profile
2.1

55%
50%
45%

1.6

35%
1.1
30%
25%
0.6
20%
15%
0.1

10%
5%
0%

-0.4

Figure 6: Sumzero.com Fund Manager AUM Profile. The left axis is the percentage of funds that fit into a
given asset under management (AUM) category from Sumzero.com (there are a total of 815 unique funds, but
only 679 have AUM data). The right axis is the average idea submissions per fund for a given AUM category
(there are 1211 ideas submissions by those funds with AUM data). The X-axis represents AUM categories.
Data as of September 20, 2009.

49

Average Submissions per Fund

Percentage of Population

40%

Table 1: Recommendation Summary Data

This table reports summary statistics for the sample of investment recommendations submitted to
Valueinvestorsclub.com. The sample includes all recommendations shared with the VIC community from the time
of the communitys launch on January 1, 2000, through December 31, 2008. Panel A reports where assets are traded
and the asset type recommended. Panel B reports the number of each long, short, and long/short recommendation
by the type of asset. Panel C reports the number of each long, short, and long/short recommendation by trading
location.
Panel A: Asset type and trading location (n=3273)
Market

Common
Stock

Bonds

Preferred
Stock

Convertible
Securities

Warrants

US

2698

46

32

12

Canada

156

UK/Europe

149

Japan

15

Hong Kong

Options

Other

Total

30

2832

161

153

16

19

19

Korea

14

14

Other

77

78

Total

3128

50

34

12

35

3273

Panel B: Recommendation by asset type (n=3273)

Bonds

Long

Common
Stock
2816

Short
Long/Short
Total

Convertible
Securities
12

Warrants

Options

Other

Total

44

Preferred
Stock
25

11

2922

274

283

38

19

68

2798

40

26

30

3273

Japan

Korea

Other

Total

15

Hong
Kong
17

13

72

2922

Panel C: Recommendation and market location (n=3273)

US

Canada

Long

2508

158

UK/
Europe
139

Short

273

283

Long/Short

51

68

2832

161

153

16

19

14

78

3273

Total

50

Table 2: Recommendation Descriptive Statistics

This table reports summary statistics for VIC recommendations. The sample consists of all firms that have at least one monthly return observation.
Panels A and B show the characteristics of investment ideas. Panel C shows the frequency of recommendations by calendar year. B/M is the ratio of the
LTM book value of equity to the market value of equity measured at the recommendation date. E/M is the ratio of LTM trailing earnings to the market
value of equity measured at the recommendation date. ROA is the LTM return on assets. ME is the market value of equity measured at the
recommendation date.
Panel A: Long recommendation fundamental characteristics (n=1959)
ME (millions)

B/M

E/M

ROA

ROE

Mean

3806

1.26

0.00

0.03

-0.05

25 Percentile
Median

112

0.33

-0.02

0.00

-0.01

393

0.62

0.05

0.04

0.09

1536

1.06

0.08

0.09

0.18

ME (millions)

B/M

E/M

ROA

ROE

Mean

2010

0.28

-0.11

0.09

0.41

25 Percentile

251

0.17

0.00

0.00

0.01

Median

641

0.34

0.04

0.05

0.12

75th Percentile

1672

0.66

0.07

0.11

0.22

th

th

75 Percentile

Panel B: Short recommendation fundamental characteristics (n=242)

th

Panel C: Time-series distribution of recommendations

Year
2000
2001
2002
2003
2004
2005
2006
2007
2008

Long Recommendations

Short Recommendations

110
191
204
211
226
210
228
310
269

1
2
11
34
31
44
34
35
50

51

Table 3: Control-Firm Buy-and-Hold Abnormal Returns

Returns to sample firms and control firms from January 1, 2000 to December 31, 2008. Control firms are selected by choosing the firm for which the sum of the
absolute value of the percentage difference in size and the absolute value of the percentage difference in book-to-market ratio is minimized. The mean samplefirm returns and mean control-firm returns in panel B are returns to a short position in the security. P-values associated with a two-tailed paired t-test and a signtest are presented. The sample consists of all firms that have the necessary data to conduct the control-firm BHAR analysis.
Panel A: Long recommendations
N

Mean sample
firm return

Mean control
firm return

Difference
(BHAR)

P-value of t-test
for difference

P-value of sign-test for

difference

One-year

1429

17.28%

10.07%

7.21%

0.0015***

0.1010

Two-year

1152

43.34%

28.43%

14.91%

0.0003***

0.0087***

Three-year

945

72.34%

54.30%

18.04%

0.0066***

0.0007***

Mean sample
firm (short)

Mean control
firm (short)

Difference
(BHAR)

P-value of t-test
for difference

P-value of sign-test for

difference

One-year

156

-4.16%

-7.05%

2.88%

0.6421

0.0924*

Two-year

128

-9.06%

-18.37%

9.32%

0.3275

0.2504

Three-year

97

-22.73%

-24.62%

1.90%

0.8784

0.1548

Panel B: Short recommendations

*, ** and *** denote two-tailed statistical significance at the 10%, 5% and 1% levels respectively.

52

Table 4: Benchmark-Portfolio Buy-and-Hold Abnormal Returns

Returns to sample firms and benchmark-portfolios from January 1, 2000 to December 31, 2008. Benchmark-portfolio abnormal returns are calculated by
assigning each stock to one of 125 benchmark-portfolios based on size, book-to-market ratio, and momentum characteristics, then subtracting the benchmarkportfolio return from the sample firm return. Mean sample returns and mean benchmark-portfolio returns in panel B represent the return to a short position in the
security or portfolio. P-values associated with a paired t-test and the Lyon, Barber, and Tsai (1999) bootstrapped skewness-adjusted t-statistics are also presented
(1000 resamples of size=n/4). The sample consists of all firms that have the necessary data to conduct the benchmark-portfolio BHAR analysis.
Panel A: Long Recommendations
n

Mean sample
firm return

Mean benchmark-portfolio
return

Difference
(BHAR)

P-value of
paired t-test
for difference

P-value of
skewness-adjusted
t-test for difference

One-year

1327

17.11%

7.59%

9.52%

0.0000***

0.0000***

Two-year

988

45.02%

25.99%

19.03%

0.0000***

0.0000***

Three-year

777

74.39%

50.80%

23.60%

0.0000***

0.0013***

Mean sample
firm return (short)

Mean sample
firm return (short)

Difference
(BHAR)

P-value of
paired t-test
for difference

P-value of
skewness-adjusted
t-test for difference

One-year

148

-2.02%

-7.17%

5.15%

0.0840*

0.4717

Two-year

115

-3.35%

-21.37%

18.02%

0.0014***

0.1877

Three-year

88

-12.74%

-34.21%

21.47%

0.0008**

0.4906

Panel B: Short Recommendations

*, ** and *** denote two-tailed statistical significance at the 10%, 5% and 1% levels respectively.

53

Table 5: Calendar-Time Portfolio Abnormal Returns

This table reports calendar-time abnormal returns to portfolios of VIC recommended stocks. The long-recommendations sample contains stocks recommended as
a buy. The short-recommendations sample contains stocks recommended as a sell. The samples consists of all firms that have at least one monthly return
observation. Each month, the portfolios consist of all firms that were recommended in the current month t, and within the last x months (where x is the length of
the holding period). Portfolios are rebalanced monthly. N represents the number of event months used in the calculations. The time period under analysis is from
January 1, 2000, to December 31, 2008, using event observations from January 1, 2000, to December 31, 2008. There are 1959 (242) long (short) recommended
firms with at least one monthly return observation. Average returns are in monthly percent, t-statistics are shown below the return estimates, and 5% statistical
significance is indicated in bold. Non-parametric results are in median monthly percent, z-statistics from a Wilcoxon signed rank test for zero median are shown
below the median return estimates, and 5% statistical significance is indicated in bold.

Onemonth
Panel A: Long recommendations
Average/Median returns
2.16%
[2.94]
Control-firm adjusted
2.06%
[3.98]
Benchmark-portfolio Adjusted 0.81%
[2.02]
N
107
Panel B: Short recommendations
Average/Median returns
-4.03%
Control-firm adjusted
Benchmark-portfolio Adjusted
N

[-2.93]
-0.69%
[-0.42]
-3.37%
-1.33
76

Equal-weight portfolio (parametric)

ThreeSixOneTwomonth
month
year
year

Threeyear

Onemonth

Equal-weight portfolio (non-parametric)

ThreeSixOneTwomonth
month
year
year

Threeyear

1.61%
[2.41]
0.95%
[2.86]
0.62%
[2.81]
107

1.34%
[2.12]
0.48%
[1.83]
0.53%
[2.45]
107

1.23%
[1.97]
0.31%
[1.34]
0.51%
[2.65]
107

1.09%
[1.78]
0.27%
[1.43]
0.39%
[2.12]
107

1.03%
[1.67]
0.20%
[1.14]
0.33%
[1.82]
107

1.84%
[3.15]
2.54%
[3.99]
0.37%
[1.66]
107

1.45%
[2.75]
0.93%
[3.03]
0.50%
[2.51]
107

1.65%
[2.66]
0.49%
[2.00]
0.52%
[2.69]
107

1.47%
[2.64]
0.55%
[1.68]
0.20%
[2.14]
107

1.37%
[2.51]
0.27%
[1.63]
0.27%
[1.77]
107

1.37%
[2.44]
0.13%
[1.22]
0.12%
[1.41]
107

-2.10%
[-2.01]
-1.56%
[-1.28]
-1.99%
-2.07
76

-0.82%
[-0.91]
-0.69%
[-0.87]
-1.39%
-1.99
76

-1.09%
[-1.34]
-0.57%
[-0.77]
-1.19%
-2.04
76

-1.06%
[-1.45]
-0.54%
[-0.78]
-1.24%
-2.28
76

-1.05%
[-1.6]
-0.47%
[-0.68]
-1.28%
-2.38
76

-2.94%
[-2.71]
0.69%
[-0.09]
-1.92%
[-1.09]
76

-2.15%
[-1.94]
-0.72%
[-0.8]
-1.31%
[-1.87]
76

-0.61%
[-0.76]
0.25%
[-0.2]
-0.35%
[-1.49]
76

-0.88%
[-0.94]
0.11%
[-0.02]
-0.66%
[-1.97]
76

-0.37%
[-0.74]
-0.06%
[-0.54]
-0.93%
[-2.75]
76

-0.68%
[-1.31]
0.06%
[-0.25]
-0.80%
[-2.88]
76

54

Table 6: Robustness: Calendar-Time Abnormal Returns (Long Recommendations)

This table reports calendar-time abnormal returns for VIC long recommended stocks. The sample consists of all
firms that have at least one monthly return observation. Each month, the portfolios consist of all firms that were
recommended in the current month t, and within the last x months (where x is the length of the holding period).
Portfolios are rebalanced monthly. N represents the number of event months used in the calculations. The time
period under analysis is from January 1, 2000, to December 31, 2008, using event observations from January 1,
2000, to December 31, 2008. Average returns are in monthly percent, t-statistics are shown below the return
estimates, and 5% statistical significance is indicated in bold. TO is the past 3 months average daily trading volume
divided by shares outstanding measured at the recommendation date.

Panel A: ME>$1B
Control-firm adjusted
Benchmark-portfolio Adjusted
N
Panel B: Price >$5 & TO>1%
Control-firm adjusted
Benchmark-portfolio Adjusted
N
Panel C: Minimum 10 stocks
Control-firm adjusted
Benchmark-portfolio Adjusted
N
Panel D: Windsorized Sample (5%)
Control-firm adjusted
Benchmark-portfolio Adjusted
N

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

Threeyear

-0.19%
[-0.19]
-0.78%
[-1.01]
74

0.61%
[1.11]
0.76%
[2.04]
103

0.11%
[0.25]
0.51%
[1.83]
103

0.19%
[0.73]
0.60%
[2.75]
103

0.12%
[0.57]
0.55%
[2.75]
103

0.06%
[0.33]
0.51%
[2.65]
103

1.35%
[2.57]
0.93%
[2.23]
101

0.81%
[2.19]
0.79%
[3.13]
105

0.26%
[0.79]
0.53%
[2.51]
105

0.25%
[0.89]
0.57%
[2.88]
105

0.22%
[0.95]
0.45%
[2.48]
105

0.16%
[0.72]
0.41%
[2.37]
105

1.62%
[2.89]
1.08%
[2.43]
65

0.98%
[2.91]
0.54%
[2.4]
104

0.50%
[1.9]
0.51%
[2.35]
104

0.33%
[1.42]
0.50%
[2.54]
104

0.29%
[1.53]
0.37%
[2.01]
104

0.22%
[1.24]
0.30%
[1.7]
104

2.06%
[3.98]
0.81%
[2.01]
105

0.97%
[2.9]
0.62%
[2.79]
105

0.53%
[2.05]
0.47%
[2.19]
105

0.37%
[1.71]
0.39%
[2.03]
105

0.31%
[1.73]
0.27%
[1.47]
105

0.25%
[1.51]
0.20%
[1.08]
105

55

Table 7: Calendar-Time Portfolio Regressions, Factor Loadings 2000-2008

This table reports calendar-time abnormal returns and factor loadings for VIC recommended stocks. The longrecommendations sample contains stocks recommended as a buy. The short-recommendations sample contains
stocks recommended as a sell. The samples consist of all firms that have at least one monthly return observation.
Each month, the portfolios consist of all firms that were recommended in month t, and within the last x months
(where x is the length of the holding period). Portfolios are rebalanced monthly. The time period under analysis is
from January 1, 2000, to December 31, 2008, using event observations from January 1, 2000, to December 31, 2008.
There are 1959 (242) long (short) recommended firms with at least one monthly return observation. Alpha is the
intercept on a regression of monthly excess return from the rebalanced strategy. The explanatory variables are the
monthly returns from the Fama and French (1993) and Chen, Novy-Marx, and Zhang (2010)(CNZ) mimicking
portfolios, Carhart (1997) momentum factor, and Pastor and Stambaugh (2003) traded liquidity factor. Alphas are in
monthly percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in
bold.
Panel A: Long recommendations (one-year portfolios)
Alpha

MKT

Market Model

1.37%

1.18

Fama-French

[4.03]
0.68%

[15.86]
1.13

[2.69]
1.54%
[4.38]
0.73%
[3.00]
0.74%
[2.99]

[20.26]
1.09
[11.48]
1.04
[17.78]
1.05
[16.19]

CNZ
Carhart alpha
5-factor

SMB

HML

0.68

0.49

[10.13]

[6.29]

0.75
[11.11]
0.75
[10.83]

MOM

LQD

0.48
[6.56]
0.49
[6.22]

-0.14
[-3.32]
-0.14
[-3.30]

-0.02
[-0.26]

SMB

HML

MOM

LQD

0.80

1.23

[3.00]

[4.38]

INV

ROA

0.28
[1.82]

-0.12
[-1.71]

MOM

LQD

0.63
[1.75]

0.26
[1.19]

Panel B: Short recommendations (one-year portfolios)

MKT

Market Model

Alpha
-1.18%

Fama-French

[-1.67]
-1.75%

[4.96]
0.69

[-2.86]
-1.45%
[-1.92]
-1.71%
[-2.75]
-1.69%
[-2.62]

[4.49]
1.01
[4.92]
0.66
[3.90]
0.69
[3.23]

CNZ
Carhart alpha
5-factor

0.84

0.81
[3.00]
0.81
[2.97]

1.23
[4.36]
1.24
[4.27]

56

-0.06
[-0.40]
-0.06
[-0.33]

-0.04
[-0.19]

Table 8: Calendar-Time Portfolio Regressions

This table reports calendar-time abnormal returns for VIC recommended stocks. The long-recommendations sample contains stocks recommended as a buy. The
short-recommendations sample contains stocks recommended as a sell. The samples consist of all firms that have at least one monthly return observation. Each
month, the portfolios consist of all firms that were recommended in month t, and within the last x months (where x is the length of the holding period). Portfolios
are rebalanced monthly. The time period under analysis is from January 1, 2000, to December 31, 2008, using event observations from January 1, 2000, to
December 31, 2008. There are 1959 (242) long (short) recommended firms with at least one monthly return observation. Alpha is the intercept on a regression of
monthly excess return from the rebalanced strategy. The explanatory variables are the monthly returns from the Fama and French (1993) and Chen, Novy-Marx,
and Zhang (2010) mimicking portfolios, Carhart (1997) momentum factor, and Pastor and Stambaugh (2003) traded liquidity factor. Alphas are in monthly
percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

Threemonth

WLS
SixOnemonth
year

Threeyear

Onemonth

Twoyear

Threeyear

2.32%

1.77%

1.49%

1.37%

1.23%

1.17%

1.95%

1.43%

1.10%

0.85%

0.57%

0.42%

[4.75]
1.59%
[3.6]
2.42%
[4.76]
1.64%
[3.73]
1.71%
[3.85]

[4.84]
1.11%
[3.81]
1.96%
[5.21]
1.16%
[4.23]
1.19%
[4.24]

[4.31]
0.80%
[3.00]
1.62%
[4.51]
0.85%
[3.26]
0.84%
[3.19]

[4.03]
0.68%
[2.69]
1.54%
[4.38]
0.73%
[3.00]
0.74%
[2.99]

[3.73]
0.57%
[2.29]
1.40%
[4.09]
0.61%
[2.56]
0.62%
[2.58]

[3.56]
0.51%
[2.08]
1.34%
[3.95]
0.55%
[2.34]
0.56%
[2.35]

[4.94]
1.38%
[3.79]
2.17%
[5.23]
1.43%
[4.14]
1.44%
[4.11]

[4.54]
0.78%
[3.14]
1.69%
[5.07]
0.88%
[3.88]
0.90%
[3.86]

[3.71]
0.47%
[2.1]
1.25%
[4.01]
0.55%
[2.63]
0.51%
[2.37]

[2.99]
0.25%
[1.32]
0.93%
[3.14]
0.30%
[1.71]
0.26%
[1.44]

[2.22]
0.06%
[0.38]
0.65%
[2.42]
0.11%
[0.7]
0.09%
[0.53]

[1.75]
-0.04%
[-0.26]
0.52%
[2.09]
0.00%
[0.05]
-0.02%
[-0.11]

Panel A: Long recommendations

Market Model alpha
Fama-French alpha
CNZ alpha
Carhart alpha
5-factor alpha

57

Table 8: Calendar-Time Portfolio Regressions (Cont.)

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

-4.05%

-2.15%

-0.88%

-1.18%

[-3.29]
-4.76%
[-4.24]
-4.08%
[-3.01]
-4.82%
[-4.21]
-5.05%
[-4.29]

[-2.36]
-2.93%
[-3.9]
-2.21%
[-2.22]
-2.97%
[-3.88]
-2.96%
[-3.73]

[-1.17]
-1.46%
[-2.21]
-1.10%
[-1.37]
-1.41%
[-2.09]
-1.34%
[-1.93]

[-1.67]
-1.75%
[-2.86]
-1.45%
[-1.92]
-1.71%
[-2.75]
-1.69%
[-2.62]

WLS
Threeyear

Onemonth

Threemonth

Sixmonth

Oneyear

Twoyear

Threeyear

-1.18%

-1.19%

-2.19%

-1.50%

-0.64%

-0.95%

-0.95%

-1.10%

[-1.78]
-1.74%
[-3.1]
-1.41%
[-1.98]
-1.66%
[-2.91]
-1.61%
[-2.74]

[-1.93]
-1.69%
[-3.17]
-1.33%
[-1.99]
-1.64%
[-3.01]
-1.59%
[-2.83]

[-2.03]
-2.91%
[-2.78]
-1.91%
[-1.6]
-2.97%
[-2.76]
-3.05%
[-2.83]

[-2.04]
-2.46%
[-3.91]
-1.32%
[-1.61]
-2.49%
[-3.86]
-2.54%
[-3.89]

[-1]
-1.22%
[-2.15]
-0.61%
[-0.85]
-1.09%
[-1.9]
-1.16%
[-1.98]

[-1.76]
-1.42%
[-2.96]
-0.85%
[-1.39]
-1.29%
[-2.65]
-1.37%
[-2.8]

[-2.06]
-1.35%
[-3.48]
-0.75%
[-1.4]
-1.15%
[-2.94]
-1.22%
[-3.11]

[-2.75]
-1.40%
[-4.16]
-0.84%
[-1.78]
-1.32%
[-3.79]
-1.38%
[-3.95]

Panel B: Short recommendations

Market Model alpha
Fama-French alpha
CNZ alpha
Carhart alpha
5-factor alpha

58

Table 9: Calendar-Time Portfolio Regressions by Market Equity (Long Recommendations)

This table reports calendar-time abnormal returns for VIC recommended stocks. The samples consist of all firms that have at least one monthly return
observation. At the beginning of every calendar month, all event firms are assigned to one of 5 quintiles based on their market capitalization at the beginning of
the month. Each month, the quintile portfolios consist of all firms that were recommended in month t, and within the last x months (where x is the length of the
holding period). Portfolios are rebalanced monthly. The time period under analysis is from January 1, 2000, to December 31, 2008, using event observations
from January 1, 2000, to December 31, 2008. There are 1959 long recommended firms with at least one monthly return observation. Alpha is the intercept on a
regression of monthly excess return from the rebalanced strategy. The explanatory variables are the monthly returns from the Fama and French (1993) mimicking
portfolios. Alphas are in monthly percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

Threeyear

Onemonth

4.89%
[3.19]
2.44%
[2.54]
1.78%
[1.9]
2.28%
[2.27]
0.92%
[1.16]

2.22%
[2.69]
2.45%
[3.6]
1.77%
[2.59]
1.69%
[2.65]
1.32%
[2.22]

1.77%
[3.4]
1.95%
[3.8]
1.19%
[2.64]
0.91%
[1.94]
0.54%
[1.41]

4.17%
[3.62]
2.15%
[2.58]
2.36%
[2.92]
1.85%
[2.11]
0.35%
[0.51]

Threemonth

WLS
SixOnemonth
year

Twoyear

Threeyear

2.09%
[2.87]
2.12%
[3.53]
1.84%
[3.07]
1.39%
[2.38]
0.71%
[1.48]

1.98%
[3.62]
1.24%
[2.3]
1.71%
[3.66]
0.64%
[1.34]
0.64%
[1.48]

0.91%
[2.4]
0.75%
[2.01]
0.96%
[3.17]
0.41%
[1.42]
0.32%
[1.38]

0.64%
[1.77]
0.85%
[2.48]
0.78%
[2.57]
0.47%
[1.85]
0.20%
[1.03]

Panel A: Fama-French model

1
(Small)
2
3
4
5
(Large)

2.60%
[4.11]
1.86%
[3.02]
1.88%
[3.4]
0.80%
[1.45]
0.90%
[1.72]

2.01%
[3.81]
2.17%
[3.93]
1.39%
[2.85]
0.95%
[1.86]
0.70%
[1.61]

1.95%
[3.75]
1.83%
[3.49]
1.28%
[2.86]
0.82%
[1.73]
0.63%
[1.61]

59

1.09%
[2.68]
1.26%
[2.78]
1.19%
[3.19]
0.71%
[1.89]
0.47%
[1.52]

Table 10: Calendar-Time Portfolio Regressions by Book-to-Market Equity (Long Recommendations)

This table reports calendar-time abnormal returns for VIC recommended stocks. The samples consist of all firms that have at least one monthly return
observation. At the beginning of every calendar month, all event firms are assigned to one of 5 quintiles based on their book-to-market at the beginning of the
month. Each month, the quintile portfolios consist of all firms that were recommended in month t, and within the last x months (where x is the length of the
holding period). Portfolios are rebalanced monthly. The time period under analysis is from January 1, 2000, to December 31, 2008, using event observations
from January 1, 2000, to December 31, 2008. There are 1959 long recommended firms with at least one monthly return observation. Alpha is the intercept on a
regression of monthly excess return from the rebalanced strategy. The explanatory variables are the monthly returns from the Fama and French (1993) mimicking
portfolios. Alphas are in monthly percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

1.44%
[1.39]
0.72%
[0.66]
1.05%
[0.9]
3.98%
[3.92]
0.46%
[0.33]

3.07%
[3.73]
0.96%
[1.51]
1.28%
[1.48]
1.82%
[2.43]
0.72%
[1.1]

Threeyear

Onemonth

Threemonth

WLS
SixOnemonth
year

1.29%
[2.1]
0.75%
[1.78]
-0.05%
[-0.09]
0.93%
[1.9]
0.50%
[1.17]

1.11%
[1.14]
1.24%
[1.3]
0.34%
[0.34]
2.72%
[2.83]
-0.19%
[-0.19]

2.44%
[2.97]
0.96%
[1.65]
0.21%
[0.29]
1.33%
[1.89]
0.66%
[1.06]

1.34%
[1.79]
0.88%
[1.61]
0.41%
[0.67]
0.80%
[1.3]
0.46%
[0.82]

Twoyear

Threeyear

0.81%
[1.6]
0.63%
[2.12]
-0.91%
[-1.81]
0.47%
[1.35]
0.22%
[0.74]

0.73%
[1.68]
0.26%
[0.91]
-0.82%
[-1.97]
0.20%
[0.61]
0.36%
[1.17]

Panel A: Fama-French model

1
(Value)
2
3
4
5
(Growth)

1.49%
[1.91]
1.14%
[1.9]
0.84%
[1.23]
1.27%
[1.92]
0.44%
[0.73]

1.69%
[2.36]
1.23%
[2.68]
0.28%
[0.47]
1.25%
[2.16]
0.59%
[1.34]

1.40%
[2.17]
1.01%
[2.4]
-0.11%
[-0.17]
1.12%
[2.27]
0.40%
[0.99]

60

1.32%
[2.04]
0.96%
[2.47]
-0.36%
[-0.7]
0.69%
[1.42]
0.44%
[1.17]

Table 11: Calendar-Time Portfolio Regressions by Turnover (Long Recommendations)

This table reports calendar-time abnormal returns for VIC recommended stocks. The samples consist of all firms that have at least one monthly return
observation. At the beginning of every calendar month, all event firms are assigned to one of 5 quintiles based on their turnover at the beginning of the month,
where turnover is the past 3 months average daily trading volume divided by shares outstanding measured at the recommendation date. Each month, the quintile
portfolios consist of all firms that were recommended in month t, and within the last x months (where x is the length of the holding period). Portfolios are
rebalanced monthly. The time period under analysis is from January 1, 2000, to December 31, 2008, using event observations from January 1, 2000, to December
31, 2008. There are 1959 long recommended firms with at least one monthly return observation. Alpha is the intercept on a regression of monthly excess return
from the rebalanced strategy. The explanatory variables are the monthly returns from the Fama and French (1993) mimicking portfolios. Alphas are in monthly
percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

WLS
Threeyear

Onemonth

Threemonth

Sixmonth

Oneyear

Twoyear

Threeyear

2.65%
[3.26]
1.90%
[2.12]
4.62%
[3.56]
0.32%
[0.36]
0.44%
[0.38]

3.12%
[4.82]
1.85%
[2.7]
2.13%
[2.98]
1.02%
[1.39]
0.83%
[1.09]

1.13%
[2.68]
0.91%
[1.85]
0.48%
[1.06]
0.65%
[1.35]
1.27%
[1.99]

2.34%
[3.08]
1.95%
[2.64]
3.32%
[3.45]
-0.13%
[-0.15]
0.15%
[0.15]

2.52%
[3.99]
1.88%
[2.88]
1.52%
[2.5]
0.22%
[0.31]
0.53%
[0.77]

0.93%
[2.1]
1.53%
[2.8]
0.24%
[0.45]
0.28%
[0.45]
0.93%
[1.57]

1.03%
[2.56]
1.54%
[3.56]
-0.06%
[-0.16]
0.45%
[1.01]
1.06%
[2.09]

0.24%
[0.62]
0.80%
[2.53]
-0.06%
[-0.23]
0.38%
[1.21]
0.91%
[2.25]

0.00
[0.35]
1.95%
[2.64]
0.00
[-0.99]
0.00
[1.2]
0.01
[2.21]

Panel A: Fama-French model

1
(Illiquid)
2
3
4
5
(Liquid)

1.49%
[3.11]
1.62%
[2.52]
0.63%
[1.00]
0.36%
[0.53]
1.03%
[1.46]

1.72%
[4.01]
1.64%
[3.03]
0.55%
[1.07]
0.67%
[1.19]
1.21%
[1.8]

1.19%
[2.7]
0.92%
[1.94]
0.65%
[1.49]
0.66%
[1.31]
1.26%
[1.95]

61

Table 12: VIC Ratings and Performance

This table reports Fama-MacBeth predictive regressions of individual BHAR (buy-and-hold-abnormal-returns) on ratings. A minimum of 10 observations are
required to perform a cross-sectional regression. The dependent variable in regressions (1), (2), (5), and (6) is the BHAR from t+2 to t+6. The dependent variable
in regressions (3), (4), (7), and (8) is the BHAR from t+7 to t+12. Size and B/M are the natural logarithms of the firm characteristics of market equity and bookto-market of the given firm. Past 6-month returns are the return of the given firm over the prior sixth month period. Turnover is the average daily volume of the
previous 3 months divided by the shares outstanding. T-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.
Benchmark-portfolio Adjusted
BHAR +2 to +6
BHAR +7 to +12

Control-Firm Adjusted
BHAR +2 to +6
BHAR +7 to +12

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

-0.16
[-1.82]
0.04
[1.96]

-0.23
[-1.12]
0.05
[2.04]
-0.01
[-0.74]
0.01
[0.52]
0.09
[1.52]
0.55
[1.43]

-0.13
[-1.08]
0.02
[1.09]

-0.04
[-0.19]
0.01
[0.43]
0.00
[0.5]
0.00
[0]
-0.09
[-1.01]
-0.45
[-0.74]

-0.19
[-1.37]
0.04
[1.37]

-0.16
[-0.56]
0.06
[1.63]
-0.01
[-0.76]
-0.02
[-0.48]
-0.01
[-0.09]
-0.07
[-0.13]

-0.19
[-1.16]
0.03
[1.01]

-0.16
[-0.46]
0.08
[1.58]
-0.02
[-0.66]
-0.07
[-1.54]
-0.11
[-0.87]
-1.81
[-1.75]

16.70
52

16.70
52

16.40
48

16.40
48

16.70
52

16.70
52

16.40
48

16.40
48

Panel A: Long recommendations

Constant
Rating
Ln(Size)
Ln(B/M)
Past 6-month Returns
Turnover

Avg obs in cross-sec regs

Number of cross-sec regs

62

Table 13: Calendar-Time Portfolio Regressions by Ratings (Long Recommendations)

This table reports calendar-time abnormal returns for VIC recommended stocks. The samples consist of all firms that have at least one monthly return
observation and a rating. At the beginning of every calendar month, all event firms are assigned to one of 5 quintiles based on their rating. Each month, the
quintile portfolios consist of all firms that were recommended in month t, and within the last x months (where x is the length of the holding period). Portfolios are
rebalanced monthly. The time period under analysis is from January 1, 2000, to December 31, 2008, using event observations from January 1, 2000, to December
31, 2008. There are 1959 long recommended firms with at least one monthly return observation. Alpha is the intercept on a regression of monthly excess return
from the rebalanced strategy. The explanatory variables are the monthly returns from the Fama and French (1993) mimicking portfolios. Alphas are in monthly
percent, t-statistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.

Onemonth

Equal-weight portfolio
ThreeSixOneTwomonth
month
year
year

Threeyear

Onemonth

4.54%
[3.13]
0.79%
[0.89]
0.51%
[0.65]
1.79%
[1.96]
0.21%
[0.15]

3.62%
[3.09]
0.67%
[1.16]
2.06%
[2.85]
1.83%
[2.61]
-0.29%
[-0.36]

3.18%
[3.01]
0.84%
[1.85]
0.53%
[1.29]
0.29%
[0.84]
0.44%
[0.86]

2.43%
[2.32]
1.01%
[1.24]
0.95%
[1.31]
1.78%
[2.28]
-0.08%
[-0.06]

Threemonth

WLS
SixOnemonth
year

Twoyear

Threeyear

2.03%
[2.61]
1.13%
[2.11]
1.37%
[2.18]
1.29%
[2.13]
-0.33%
[-0.47]

1.90%
[3.05]
0.49%
[1.14]
0.62%
[1.19]
0.29%
[0.68]
0.47%
[0.86]

0.84%
[2.34]
0.74%
[2.66]
0.10%
[0.31]
0.27%
[1]
0.95%
[2.89]

0.67%
[2.16]
0.70%
[2.93]
0.12%
[0.41]
0.16%
[0.63]
0.79%
[2.79]

Panel A: Fama-French model

1
(High)
2
3
4
5
(Low)

3.78%
[3.42]
0.26%
[0.49]
0.77%
[1.43]
0.35%
[0.73]
0.00%
[0.01]

3.42%
[3.14]
0.89%
[1.83]
0.33%
[0.74]
0.29%
[0.77]
0.49%
[0.85]

3.23%
[3.03]
0.84%
[1.8]
0.54%
[1.27]
0.37%
[1.08]
0.56%
[1.07]

63

1.23%
[2.51]
0.92%
[2.70]
-0.03%
[-0.07]
0.23%
[0.72]
0.88%
[2.11]

Table 14: Top and Bottom Rating Quintile Control-Firm Buy-and-Hold Abnormal Returns for Buy Recommendations
Returns to sample firms and control firms from January 1, 2000, to December 31, 2008. Control firms are selected by choosing the firm for which the sum of the
absolute value of the percentage difference in size and the absolute value of the percentage difference in book-to-market ratio is minimized. The top (bottom)
quintile for rating consists of the highest rated (lowest rated) 20% of the sample. P-values associated with a two-tailed paired t-test are presented. The p-values
for the test for difference in BHAR between the top and bottom quintile are calculated using a two-tailed paired t-test for difference assuming unequal variances.

Top Rating Quintile

Bottom Rating Quintile

Mean sample
firm return

Mean control
firm return

BHAR

Mean sample
firm return

Mean control
firm return

BHAR

One-year

27.76%

6.07%

21.69%

8.56%

8.72%

-0.16%

0.0017***

Two-year

46.59%

20.44%

26.15%

32.26%

28.05%

4.21%

0.0536*

Three-year

86.86%

44.28%

42.58%

46.25%

55.27%

-9.02%

0.0061***

*, ** and *** denote two-tailed statistical significance at the 10%, 5% and 1% levels respectively.

64

P-value of
difference in BHAR

Table 15: Comments Summary Statistics

This table reports summary statistics for the analysis of the comments associated with the sample of investment recommendations submitted to
Valueinvestorsclub.com. The sample includes all recommendations shared with the VIC community from the time of the communitys launch on January 1,
2000 through December 31, 2008. Results are presented for the sample associated with the control-firm BHAR analysis. There are 1869 observations in
total: 1671 long observations and 198 short observations. The full, long-only, and short-only samples have at least 1 comment for 91.55%, 91.02%, and
96.46% of their respective observations.
Panel A: Summary Statistics for full sample (n=1711)
Market
Comments
Members
Private
Mean
12.03
4.84
2.50
Median
8.00
4.00
1.00
Min

1.00

1.00

0.00

% Private
18.55%

Author
5.26

% Author
43.29%

<45 Days
7.83

% < 45 days
74.01%

3.85%

3.00

46.15%

6.00

81.25%

0.00%

0.00

0.00%

0.00

0.00%

154.00
28.00
73.00
Panel B: Summary Statistics for long sample (n=1521)

100.00%

82.00

100.00%

91.00

100.00%

Market
Mean
Median

Max

Min

Comments
11.49

Members
4.71

Private
2.25

% Private
17.58%

Author
5.08

% Author
43.42%

<45 Days
7.65

% < 45 days
74.44%

8.00

4.00

0.00

0.00%

3.00

46.15%

6.00

81.82%

1.00

1.00

0.00

0.00%

0.00

0.00%

0.00

0.00%

138.00
28.00
52.00
Panel C: Summary Statistics for short sample (n=190)

100.00%

57.00

100.00%

91.00

100.00%

Market
Mean
Median

Max

Comments
16.39

Members
5.86

Private
4.47

% Private
26.34%

Author
6.73

% Author
42.32%

<45 Days
9.31

% < 45 days
70.57%

9.00

5.00

2.00

19.09%

4.00

43.88%

7.00

73.33%

Min

1.00

1.00

0.00

0.00%

0.00

0.00%

0.00

0.00%

Max

154.00

24.00

73.00

100.00%

82.00

100.00%

70.00

100.00%

65

Table 16: Relation Between Group Size and Idea Value

Panel A presents ratings summary statistics for sample quintiles formed on the percentage of messages that are
private. P-values for difference in means are calculated using a two-tailed paired t-test assuming unequal variances.
P-values for difference in medians are based on the z-test statistic from a Wilcoxson signed rank test. Panel B
presents OLS estimates and maximum likelihood estimates for a logit regression. The dependent variable is the
percentage of messages that are private. Total comments and number of commenters are the natural logarithms of a
given firms total comments submitted and the unique number of commenters submitting comments. Size and B/M
are the natural logarithms of the firm characteristics of market equity and book-to-market of the given firm. Past 6month returns are the return of the given firm over the prior sixth month period. Turnover is the average daily
volume of the previous 3 months divided by the shares outstanding. We estimate this model using data from January
1, 2004 to December 31, 2008 because the option to label comments private was rarely used prior to January 1,
2004 (10.01% of ideas had at least one private comment prior to 2004 versus 74.64% after January 1, 2004). Tstatistics are shown below the coefficient estimates, and 5% statistical significance is indicated in bold.
Panel A: Summary Statistics for ratings (n=1028)
Total
Mean
Median

5.10

1
(low pvt %)
4.89

1-5

P-value

5.15

5
(high pvt %)
5.14

5.28

5.17

-0.25

0.0000***

5.20

5.00

5.40

5.30

5.20

5.20

-0.20

0.0000***

Min

1.30
3.10
3.50
3.20
1.30
3.20
Max
7.10
6.40
6.40
7.10
7.00
6.70
*, ** and *** denote two-tailed statistical significance at the 10%, 5% and 1% levels respectively.
Panel B: Regression Analysis
Constant
Rating

OLS

OLS

Logit

Logit

0.76
[1.17]
0.04
[3.06]

0.15
[1.71]
0.04
[3.15]
0.00
[-0.9]
0.00
[1.58]
0.00
[-0.72]
0.00
[-0.12]
-0.06
[-2.86]
0.06
[0.50]
909

-1.99
[-3.45]
0.20
[1.78]

-1.59
[-1.99]
0.22
[1.83]
-0.02
[-0.49]
0.01
[0.87]
-0.02
[-0.4]
-0.01
[-0.06]
-0.31
[-1.64]
0.24
[0.24]
909

Ln(Total comments)
Ln(Number of commenters)
Ln(size)
Ln(B/M)
Past 6-month returns
Turnover
Number of observations

909

66

909

Table 17: Institutional Ownership Summary Statistics

This table reports summary statistics for institutional ownership associated with the sample of investment
recommendations submitted to Valueinvestorsclub.com. The sample includes all recommendations shared on the
VIC website from the time of the communitys launch on January 1, 2000, through December 31, 2008. Results are
presented for the sample associated with the control-firm BHAR analysis. In total there are 1514 observations which
have institutional holdings data. P-values for difference in mean institutional ownership are calculated using a twotailed paired t-test assuming unequal variances. P-values for difference in median institutional ownership are based
on the z-test statistic from a Wilcoxson signed rank test.
Panel A: Summary Statistics for full sample (n=1514)
Size
Mean
Median

Total
53.42%
57.47%

1(small)
25.65%
22.78%

2
46.71%
47.49%

3
60.64%
66.83%

4
68.03%
73.20%

5 (big)
70.47%
75.72%

Min

0.16%

0.16%

0.16%

0.22%

0.61%

0.35%

Max

98.36%

98.22%

95.26%

98.36%

98.26%

98.00%

B/M
Mean
Median

Total
53.42%
57.47%

1 (low)
52.53%
58.15%

2
59.17%
64.08%

3
55.91%
61.61%

4
54.62%
57.79%

5 (high)
44.93%
41.99%

Min

0.16%

0.22%

1.57%

0.16%

0.16%

0.27%

Max

98.36%

98.25%

98.00%

97.77%

98.26%

98.36%

CAR 12
Mean
Median

Total
53.42%
57.47%

1 (low)
50.03%
50.84%

2
56.22%
60.07%

3
55.82%
59.56%

4
52.73%
59.81%

5 (high)
45.77%
45.07%

Min

0.16%

0.22%

0.16%

0.60%

0.16%

0.39%

Max

98.36%

97.54%

98.22%

98.00%

97.72%

97.36%

CAR24

1-5
-44.83%
-52.94%

P-value
0.0000***
0.0000***

1-5
7.59%
16.16%

P-value
0.0010***
0.0010***

1-5
4.26%
5.77%

P-value
0.1095
0.0893*

Total

1 (low)

5 (high)

1-5

P-value

Mean

53.42%

51.62%

55.29%

50.71%

51.55%

44.19%

7.43%

0.0074***

Median

57.47%

53.26%

58.70%

55.73%

56.23%

43.01%

10.26%

0.0082***

Min

0.16%

1.19%

0.22%

0.16%

0.39%

0.60%

Max

98.36%

97.77%

97.33%

98.00%

97.72%

97.36%

Total

1 (low)

5 (high)

1-5

P-value

Mean

53.42%

51.66%

52.20%

47.77%

48.15%

47.49%

4.17%

0.1656

Median

57.47%

52.58%

58.40%

47.69%

46.22%

49.69%

2.89%

0.2085

Min

0.16%

0.22%

0.81%

0.16%

0.39%

0.70%

CAR 36

Max
98.36%
97.33%
97.77%
98.00%
96.40%
95.62%
*, ** and *** denote two-tailed statistical significance at the 10%, 5% and 1% levels respectively.

67