SSRN Id271458
SSRN Id271458
SSRN Id271458
ABSTRACT
In the presence of sentiment fluctuations, arbitrageurs may engage in different strategies
leading to dispersed sentiment exposures. We find that hedge funds in the top decile ranked
by sentiment beta outperform those in the bottom decile by 0.59% per month on a risk-
adjusted basis, with the spread being larger among skilled funds. We also find that about
10% of hedge funds have sentiment timing skill that positively correlates with fund
sentiment beta and contributes to fund performance. Our findings show that skilled hedge
funds can earn high returns by predicting and exploiting sentiment changes rather than
betting against mispricing.
*
Chen is with Mays Business School at Texas A&M University. Han is with the Rotman School
of Management at the University of Toronto. Pan is with the Cox School of Business at Southern
Methodist University. We thank Stefan Nagel (the Editor) and two anonymous referees for many
constructive comments and suggestions. We are grateful to Vikas Agarwal, Malcolm Baker, Turan
Bali, Charles Cao, Bernard Dumas, Wayne Ferson, Andrew Karolyi, Bing Liang, Jeffrey Pontiff,
Sorin Sorescu, Zheng Sun, Jeffrey Wurgler, Jianfeng Yu, and seminar and conference participants
at the Chinese University of Hong Kong, Georgetown University, Nanyang Technological
University, Peking University, Shanghai University of Finance and Economics, Singapore
Management University, Southwestern University of Finance and Economics, Texas A&M
University, Virginia Tech, Wilfrid Laurier University, Xiamen University, the China International
Conference in Finance, the Annual Hedge Fund Research Conference, HKUST Finance
Symposium on Asset Pricing, and the American Finance Association meeting for helpful
comments. We also thank Zhi Da, Ken French, David Hsieh, Lubos Pastor, Jeffrey Wurgler, and
Jianfeng Yu for making a large amount of data available on their websites. The authors have read
The Journal of Finance disclosure policy and have no conflicts of interest to disclose.
Correspondence: Yong Chen, Department of Finance, Mays Business School, Texas A&M
University, College Station, TX 77843; e-mail: ychen@mays.tamu.edu.
Hedge funds’ strategies are not directly observable to researchers. However, differences in
arbitrageurs’ strategies in the presence of sentiment fluctuations give rise to dispersion in sentiment
exposures, and thus their sentiment exposures can provide insights into how they respond to
changes in sentiment. Uncertainty about sentiment fluctuations reduces arbitrageurs’ willingness
to bet against noise traders, leading prices to diverge from fundamental values (DeLong et al.
(1990a)). Arbitrageurs who bet against mispricing should have negative sentiment exposure, since
overvalued securities tend to be more sensitive to changes in sentiment than undervalued
securities.3 Consistent with this expectation, we find that a betting-against-mispricing strategy that
goes long undervalued stocks and short overvalued stocks has a significantly negative sentiment
beta.4
However, skilled arbitrageurs may not always bet against mispricing. In the dynamic model
of Dumas, Kurshev, and Uppal (2009), the optimal strategy of rational traders facing sentiment
fluctuations is based not only on current stock mispricing but also on their expectations about
future changes in sentiment. The resulting sentiment exposure of arbitrageurs can therefore be
1
See, for example, DeLong et al. (1990a, 1990b), Lee, Shleifer, and Thaler (1991), Barberis, Shleifer, and Vishny
(1998), Brunnermeier and Nagel (2004), Baker and Wurgler (2006, 2007), Dumas, Kurshev, and Uppal (2009),
Stambaugh, Yu and Yuan (2012, 2015), and Kozak, Nagel, and Santosh (2018).
2
Brunnermeier and Nagel (2004) posit that hedge funds are “probably closer to the ideal of ‘rational arbitrageurs’
than any other class of investors.” Akbas et al. (2015) and Chen, Da, and Huang (2019) find that hedge funds function
as arbitrageurs whereas other types of institutional investors, such as mutual funds, do not.
3
Baker and Wurgler (2007) find that speculative stocks (such as smaller, younger, unprofitable, high-volatility, non-
dividend-paying, or growth companies and firms in financial distress) are particularly sensitive to investor sentiment.
4
Specifically, we find that the decile portfolios of undervalued and overvalued stocks, based on the mispricing score
measure of Stambaugh, Yu, and Yuan (2015), have sentiment betas of 0.89 and 1.27, respectively. The portfolio that
goes long undervalued stocks and short overvalued stocks has a sentiment beta of -0.38 (t-statistic = -2.18).
To test the sentiment risk premium explanation, we construct a tradable sentiment factor
as the return spread between the top and bottom decile portfolios of stocks sorted by sentiment
beta. This tradable sentiment factor has a significantly positive mean over our sample period. After
controlling for the sentiment factor, the difference in the risk-adjusted returns of high sentiment
beta hedge funds and low sentiment beta hedge funds is still significantly positive, with little
change in the magnitude. Thus, the outperformance of high sentiment beta hedge funds is largely
unexplained by the sentiment risk premium and is distinct from the relation between sentiment
beta and stock returns. High sentiment beta hedge funds perform better not simply because they
overweight high sentiment beta stocks.
5
See Section III.A for a summary of the theoretical and empirical research on the sentiment risk premium.
To better understand the skill that improves fund performance in the presence of sentiment
fluctuations, we examine the ability of hedge funds to predict sentiment changes and front-run
irrational traders. Specifically, we test three conjectures. First, skilled fund managers are able to
time investor sentiment, increasing sentiment exposure in anticipation of more bullish sentiment
ahead and outperformance of high sentiment beta investments. This is akin to riding the bubble,
except that it applies more generally outside bubble episodes. Second, hedge funds that can time
sentiment changes tend to have higher, rather than lower, sentiment beta, because there are fewer
opportunities to front-run sentiment traders on the downside. Sentiment traders are more likely to
buy stocks when they become more bullish than sell stocks when they turn more bearish, for
example, due to a reluctance to realize losses and short-sale constraints. Thus, front-running bullish
sentiment traders gives rise to high sentiment beta. Third, hedge funds with positive sentiment
timing skill observe better performance.
All three of these conjectures are supported by the data. To evaluate the ability of hedge
funds to exploit investor sentiment fluctuations, we develop a test of sentiment timing that builds
on the classic work on market timing by Henriksson and Merton (1981). Specifically, for each
hedge fund, we examine whether it increases (decreases) the loading on the tradable sentiment
factor when the factor return is higher (lower). We find that about 10% of hedge funds exhibit a
Our comprehensive analysis of how sentiment-related trading affects hedge fund returns
contributes to several strands of literature. First, we extend the seminal work of Brunnermeier and
Nagel (2004), who focus on aggregate hedge fund trading during the tech bubble, by providing
evidence on heterogeneous sentiment trading strategies across individual hedge funds over the
1994 to 2018 period, which covers more general market states. Consistent with the bubble-riding
behavior documented in Brunnermeier and Nagel (2004), our evidence shows that some skilled
hedge funds are able to time investor sentiment and realize larger returns. Importantly, we find
that hedge funds with high sentiment betas outperform those with low sentiment betas and that
such outperformance reflects managerial skill. Our results therefore suggest that some skilled
arbitrageurs are able to benefit from sentiment fluctuations, which are commonly perceived as a
form of limits-to-arbitrage.
Our study also contributes to the literature by uncovering a new source of hedge fund
performance.6 Existing research documents timing skill with respect to market returns, volatility,
liquidity, and macroeconomic uncertainty (e.g., Chen (2007), Chen and Liang (2007), Cao et al.
(2013), Bali, Brown, and Caglayan (2014)). We provide strong evidence of sentiment timing and
its contribution to hedge fund performance. Our results suggest that, beyond simply performing
6
For studies on the risk and performance of hedge funds, see Fung and Hsieh (1997, 2004), Ackermann, McEnally,
and Ravenscraft (1999), Brown, Goetzmann, and Ibbotson (1999), Agarwal and Naik (2004), Getmansky, Lo, and
Makarov (2004), Chen (2007), Chen and Liang (2007), Kosowski, Naik, and Teo (2007), Fung et al. (2008), Griffin
and Xu (2009), Jagannathan, Malakhov, and Novikov (2010), Sadka (2010), Chen (2011), Teo (2011), Titman and
Tiu (2011), Bali, Brown, and Caglayan (2011, 2012, 2014), Sun, Wang, and Zheng (2012), Cao et al. (2013), Chen,
Cliff, and Zhao (2017), and Chen, Kelly, and Wu (2020), among others.
Finally, our paper adds to the growing literature on the impact of investor sentiment in
financial markets. With few exceptions, most studies use the level of investor sentiment as a
conditioning variable in asset pricing tests (Baker and Wurgler (2006, 2007), Stambaugh, Yu, and
Yuan (2012, 2015), among others). While our focus is on the relation between exposure to
sentiment fluctuations and hedge fund returns, we find that the level of sentiment also plays an
important role: the positive relation between sentiment beta and hedge fund returns is stronger
following high sentiment periods than following low sentiment periods. This finding is at odds
with conventional wisdom that hedge funds generate alpha by betting against mispricing and that
their main strength lies in shorting overpriced securities. That view implies that hedge funds with
negative sentiment beta should outperform when the initial sentiment level is high, which is the
opposite of what we find in the data. Our result instead suggests that the bubble-riding type of
sentiment trading that generates a positive sentiment beta can contribute to fund performance,
especially for more skilled hedge funds.
The paper proceeds as follows. Section I describes our data on hedge funds, sentiment
fluctuations, and risk factors. Section II presents the baseline results and various robustness checks
on the relation between sentiment beta and hedge fund returns. Section III investigates alternative
explanations for the results based on a sentiment risk premium and fund manager skill. Section IV
concludes. Auxiliary tests and results are provided in the Internet Appendix.7
I. Data
A. Hedge Funds
7
The Internet Appendix is available in the online version of this article on the Journal of Finance website.
The sample is free of survivorship bias, as TASS covers both live and defunct hedge funds
since 1994 and we examine the period from 1994 onward. Following prior research, we apply
several screens to the fund data. To address the concern that hedge funds may backfill returns
when newly added to the database, we exclude the first 12 months of returns for each fund. We
only include funds that report monthly net-of-fee returns in U.S. dollars and allow for redemption
at a monthly or higher frequency.9 We also delete duplicate funds and funds with assets under
management below $5 million. 10 Finally, we require each fund to have at least 30 return
observations. After these screens, our sample contains 4,073 hedge funds over the period 1994 to
2018.
Table I reports descriptive statistics for the hedge fund sample based on fund-month
observations. All variables are winsorized at the 1% and 99% levels. The average fund return is
0.59% per month. The mean (median) assets under management is $181 million ($56 million), and
the average fund age is 81 months. The mean (median) management fee is 1.35% (1.50%), while
8
The choice of U.S. equity-oriented hedge fund categories follows Cao et al. (2013). Nonetheless, our inference is
robust to including all strategy categories in the sample (see the Internet Appendix).
9
Hedge funds allowing for redemption at a monthly or higher frequency face relatively frequent capital withdrawals,
which is consistent with the crucial assumption of DeLong et al. (1990a) that arbitrageurs often have short horizons.
10
Our results are robust to alternative data filters, such as the exclusion of funds with assets under management below
$10 million and exclusion of the first 24 months of fund returns. See the Internet Appendix for details.
B. Sentiment Changes
We adopt the Baker–Wurgler sentiment changes index as the main measure of sentiment
fluctuations in our tests. Baker and Wurgler (2006) capture market-wide sentiment as a composite
of six proxies: closed-end fund premium, New York Stock Exchange (NYSE) share turnover,
number and average first-day returns of initial public offerings, equity share in new issues, and
dividend premium. The sentiment level is obtained by first using principal component analysis on
these proxies and then orthogonalizing against macroeconomic variables to remove the impact of
business cycles. 11 To capture time-variation in investor sentiment, Baker and Wurgler (2007)
construct the monthly sentiment changes index from the first principal component of changes in
these sentiment proxies. The Baker-Wurgler index significantly expands the traditional measure
based on the closed-end fund premium (Lee, Shleifer, and Thaler (1991)) and has spurred a
growing body of research on the effects of investor sentiment on asset prices and corporate
decisions. As reported in Panel A of Table II, the monthly sentiment changes index has a mean of
-0.50 and a standard deviation of 1.39 over our sample period, with 25th and 75th percentiles of -
1.38 and 0.47, respectively.12
11
In addition, we explicitly control for the macroeconomic variables that Bali, Brown, and Caglayan (2011) find to
have significant explanatory power for hedge fund returns. This ensures that our results are not driven by exposures
to macroeconomic factors.
12
By construction, the Baker–Wurgler sentiment changes index has a mean of zero after orthogonalization against
macroeconomic variables. However, since the original index is constructed using data going back to 1965 and our
sample starts in 1994, the mean value of the index over our sample period is not exactly zero.
C. Risk Factors
To measure risk-adjusted returns (i.e., alpha), we control for exposures to standard risk
factors identified in the hedge fund literature. We start with Fung and Hsieh’s (2004) seven factors:
an equity market factor, a small-minus-big size factor, the change in the constant-maturity yield of
the 10-year Treasury, the change in the yield spread between Moody’s Baa bond and the 10-year
Treasury bond, and three trend-following factors for bonds, currencies, and commodities.13 These
factors are commonly used to evaluate hedge fund performance (e.g., Kosowski, Naik, and Teo
(2007), Fung et al. (2008), Jagannathan, Malakhov, and Novikov (2010), Sadka (2010), and Cao
et al. (2013)). We also control for the inflation rate and default spread, as Bali, Brown, and
Caglayan (2011) find that exposures to these two factors are significantly related to hedge fund
13
Since the two bond factors capture yield changes rather than excess returns, we follow Sadka (2010) to replace them
with returns on factor-mimicking portfolios so that the regression intercept can be interpreted as a risk-adjusted return.
In particular, the yield change of the 10-year Treasury is replaced with the return spread between the 10-year Treasury
index and the one-month T-bill rate, while the change in the spread between Moody’s Baa yield and the 10-year
Treasury is replaced with the return spread between the Corporate Bond Baa index and the 10-year Treasury index.
The index return data come from Barclays Capital.
A. Portfolio Sorts
We first use portfolio sorts to examine the relation between sentiment beta and hedge fund
returns. Each month starting in December 1996, we form 10 equal-weighted portfolios of hedge
funds based on the fund sentiment beta (i.e., the loading on the sentiment changes index) estimated
from a rolling window of the most recent 36 months (including the current month), with the first
rolling window spanning the period from January 1994 to December 1996.14 We then track the
portfolio returns over the next month (starting in January 1997). These portfolios are rebalanced
each month to generate a time series of returns from January 1997 to December 2018.
Specifically, each fund’s sentiment beta is estimated by regressing fund excess returns on
the sentiment changes index controlling for standard risk factors. In month t, for each fund with at
least 30 return observations during the 36-month rolling window, we perform the time-series
regression
where ri,t is the excess return (in excess of the one-month T-bill rate) on fund i in month t,
Δsentiment is the sentiment changes index, βS is sentiment beta, and the vector f contains the Fung–
14
The average fund size is similar across decile portfolios sorted by the sentiment beta (see the Internet Appendix for
details). Moreover, the regression analyses below control for fund size explicitly.
10
We track the returns for the decile portfolios over the next month after portfolio formation.
These portfolios are rebalanced each month. Finally, we estimate alpha (i.e., the risk-adjusted
return) by regressing the time series of the excess returns of each decile portfolio on the Fung–
Hsieh seven factors, the momentum factor, and the liquidity factors.15 Accordingly, the spread in
alpha between the two extreme decile portfolios (i.e., portfolios 10 and 1) reveals performance
dispersion attributed to sentiment beta. In the test, we calculate t-statistics using Newey–West
(1987) standard errors with two lags, where the number of lags is based on autocorrelations in
monthly hedge fund returns.16
Table III reports results for both hedge fund excess returns and alpha across the sentiment
beta-sorted portfolios. The decile portfolio with the highest sentiment beta (i.e., portfolio 10)
delivers an average excess return of 0.58% (t-statistic = 3.48) per month and an alpha of 0.51% (t-
statistic = 2.98) per month, indicating significantly positive abnormal performance, while the
decile portfolio with the lowest sentiment beta (i.e., portfolio 1) shows an average excess return of
0.27% (t-statistic = 1.77) per month and an alpha of -0.08 (t-statistic = -0.51) per month. The return
spread between the two extreme decile portfolios is 0.31% (t-statistic = 3.16) per month and both
economically and statistically significant. On a risk-adjusted basis, the spread in alpha between the
extreme decile portfolios becomes even larger, at 0.59% (t-statistic = 3.55) per month. While the
portfolio return does not increase strictly monotonically with sentiment beta, the top three
portfolios with the highest sentiment betas are also the top three portfolios with the highest average
excess returns and alphas, while the portfolio with the lowest sentiment beta also has the lowest
average excess return and alpha. Thus, the results from portfolio sorts indicate that sentiment beta
15
The two macroeconomic variables (namely, the inflation rate and default spread) are not included in the regression
when estimating the risk-adjusted return, since they are not tradable factors.
16
The average first-order (second-order) autocorrelation in hedge fund excess returns is 0.158 (0.063), and higher-
order autocorrelations are generally smaller.
11
Figure 1 plots the spread in one-month-ahead risk-adjusted returns (i.e., alpha) between the
two extreme decile portfolios with high and low sentiment betas. The series start in January 1997,
since we use a 36-month formation period. As shown in Panel A, the spread in alpha is positive
for more than two-thirds of the months over our sample period. Panel B plots the cumulative spread
in alpha between the two extreme decile portfolios sorted by sentiment beta. Again, the plot shows
a persistent difference in risk-adjusted returns between hedge funds with high and low sentiment
betas.
We perform a battery of sensitivity tests. First, instead of tracking returns from the month
immediately following portfolio formation, we skip one month. Second, to address concerns about
the precision of sentiment beta estimates, we use different combinations of risk factors as control
variables in regression (1). Our inferences continue to hold. We also examine the effect of
sentiment beta on hedge fund returns over holding periods longer than one month. The spread in
monthly alpha is 0.59%, 0.51%, 0.44%, and 0.31% over the subsequent three, six, nine, and 12
months, respectively. Finally, we find that hedge fund sentiment beta displays a fair amount of
persistence in the short run. For example, about 59% (45%) of the hedge funds placed in the top
sentiment beta decile in a given month will continue to be in the top decile six months (one year)
later. These additional findings are reported in the Internet Appendix.
12
where ri,t+1 is the fund excess return in month t + 1, and ̂ i,St is fund i's sentiment beta estimated
from regression model (1) using fund returns in the 36-month rolling window from month t – 35
to month t. That is, the key independent variable—sentiment beta—is estimated from a backward-
looking window prior to the return evaluation period for the dependent variable of the regression.
The control variables x are pre-determined fund characteristics including fund size, fund age,
management fee, incentive fee, high-water mark dummy, lockup period, redemption notice period,
and fund style dummies.
We also run the test in regression (2) using fund alpha instead of fund excess return as the
dependent variable. For each fund in month t, we remove the sentiment changes index from the
set of independent variables in regression (1) and reestimate the factor loadings using the past 36
months of data. The alpha for each fund in month t + 1 is then obtained as the difference between
the fund excess return in month t + 1 and the products of its factor loadings estimated in month t
and factor realizations in month t + 1, that is,
where the vector f denotes realized returns of factors including the Fung–Hsieh seven factors, the
momentum factor, and the liquidity factor (but excluding the sentiment changes index).
Table IV reports the results from the cross-sectional regressions, with the dependent
variable being either the hedge fund excess return or alpha. From the univariate regression, the
13
In sum, our results from both portfolio sorts and cross-sectional regressions show a
significant and positive relation between sentiment beta and hedge fund performance, even after
adjusting for common risk exposures and controlling for fund characteristics.
14
In our main analysis, we employ the Baker–Wurgler sentiment changes index, which is
widely used in the literature. In this subsection we check whether our inference is sensitive to the
choice of sentiment measure. As an alternative measure, we first consider the University of
Michigan consumer sentiment index. The results, reported in Table V, confirm a significantly
positive relation between sentiment beta and hedge fund performance. For example, on average,
the top sentiment beta hedge funds outperform the bottom sentiment beta hedge funds by 0.26%
(t-statistic = 1.88) per month and by 0.58% (t-statistic = 2.17) per month on a risk-adjusted basis.
The spreads in excess returns and alpha are very close in magnitude to those obtained from the
Baker–Wurgler sentiment measure. In addition, similar to the results based on the Baker–Wurgler
measure, the top sentiment beta portfolio delivers the largest alpha (positive and significant) while
the bottom sentiment beta portfolio has the lowest alpha (negative and insignificant).
In sum, our evidence on the relation between sentiment beta and hedge fund returns is
robust to the choice of sentiment measure. We use the Baker–Wurgler sentiment changes index as
15
Our analysis so far relies on hedge funds’ net-of-fee returns, which represent fund investors’
payoffs. However, arbitrage profits could be more closely related to gross fund returns. This should
not matter if all funds charge the same fees, including management and incentive fees. In reality,
however, fee arrangements are heterogeneous across hedge funds. If fees charged by fund
managers are systematically correlated with their sentiment betas, our inference based on net
returns could be biased. To address this concern, we repeat the tests of portfolio sorts using gross
fund returns. Since most hedge funds report net-of-fee returns in the data, we follow Teo (2009)
and Chen (2011) to compute gross returns using a simple algorithm to add back management and
incentive fees.
Table VI shows a significantly positive relation between hedge funds’ gross returns and
their sentiment betas. Based on gross fund returns, the top sentiment beta fund portfolio has an
average excess return of 1.07% (t-statistic = 6.00) and an alpha of 1.11% (t-statistic = 5.95) per
month, while the bottom sentiment beta fund portfolio has an average excess return of 0.72% (t-
statistic = 4.57) and an alpha of 0.37% (t-statistic = 2.27) per month. Between the top and bottom
portfolios, the spread in monthly fund returns is 0.35% (t-statistic = 3.04) while the spread in
monthly alpha is 0.74% (t-statistic = 4.14). The analysis using gross fund returns therefore leads
to the same inference about the relation between sentiment beta and subsequent hedge fund
performance.
16
In this section, we investigate two potential explanations for our results. The first, a risk-
based explanation, suggests that the outperformance of high sentiment beta hedge funds comes
from holding high sentiment beta stocks that have higher expected returns due to a positive
sentiment risk premium. The second explanation, a skill-based story, attributes the outperformance
of high sentiment beta funds to managerial skill. The two explanations are not mutually exclusive.
A. Risk-Based Explanation
One potential explanation for the results is that high sentiment beta hedge funds hold stocks
with large sentiment exposures and their outperformance reflects a positive sentiment risk
premium. Theoretically, the extent to which sentiment risk is priced depends on the model setup:
there can be a positive, negative, or no relation between stock sentiment beta and expected returns.
In the model of DeLong et al. (1990a), the price of an asset that is subject to the influence
of unpredictable sentiment is lower than the fair value to compensate arbitrageurs for the risk that
sentiment traders can become bearish. However, the expected price change in their model is zero
(see their equation (12)). Dumas, Kurshev, and Uppal (2009) develop an equilibrium model of
investor sentiment in which sentiment traders are intertemporal optimizers with overconfident
beliefs. As the difference in beliefs between rational and overconfident investors, sentiment
follows a driftless diffusion process with stochastic volatility proportional to investor disagreement.
When there is a difference of opinion about the fundamental, rational investors realize that
sentiment will fluctuate randomly in response to dividend shocks. As a result, a risk premium
arises from the unpredictable fluctuations of other investors. In their model, the pricing kernel is a
concave (convex) function of the sentiment level, if the coefficient of investor risk-aversion is
greater (less) than one.17 However, the instantaneous risk premium for sentiment innovations that
17
Han (2008) presents evidence based on stock market index option prices that the asset pricing kernel depends on
various proxies for market sentiment.
17
Kozak, Nagel, and Santosh (2018) show that sentiment risk can be priced when there are
common components of sentiment-driven asset demand because it is risky for arbitrageurs to take
the other side. In their model, sentiment or belief distortion leads to systematic excess demand for
risky assets, giving rise to time-varying investment opportunities and an ICAPM-like stochastic
discount factor as a function of sentiment.18 Specifically, stock prices are linear in sentiment, and
the log value function of an arbitrageur is a decreasing quadratic function of sentiment. The sign
of the sentiment risk premium is ambiguous, however, because more extreme sentiment (both
positive and negative) leads to greater stock mispricing and investment opportunities, which
translate to greater wealth and lower marginal utility for the arbitrageur. If the arbitrageur’s
portfolio is long-biased (e.g., induced by a high equity premium), the arbitrageur faces better
investment opportunities in negative sentiment states. Thus, the marginal utility of the arbitrageur
increases with sentiment and the sentiment risk premium is negative. Under alternative
assumptions, however, the arbitrageur may have better investment opportunities in positive
sentiment states and thus sentiment risk could be positively priced. To summarize, Kozak, Nagel,
and Santosh (2018) suggest that market-wide sentiment risk could be priced but the sign is an
empirical question.
Empirically, only a few papers examine the price of sentiment risk. These studies provide
mixed evidence. Glushkov (2006) finds that the relation between sentiment beta and stock returns
has an inverse U-shape, that is, stocks with extreme values of sentiment beta earn lower, not higher,
future returns relative to those with near-zero sentiment beta. In contrast, Ho and Hung (2012)
18
In a related paper, Campbell and Kyle (1993) present an equilibrium model of smart money and noise trading where
the risk premium on stocks is perfectly negatively correlated with noise trader demand. Their Theorem 3.2 shows that
the value function of the arbitrageur depends on noise trader demand and its square.
18
Nonetheless, if a positive sentiment risk premium exists, one may wonder whether the
outperformance of high sentiment beta hedge funds is just a manifestation of the positive relation
between sentiment beta and stock returns. We test this possibility in Table VII. Specifically, we
run a time-series regression of the hedge fund return spread between high and low sentiment beta
decile portfolios on a tradable sentiment factor constructed from the stock market, which is the
return spread between the top decile and the bottom decile of stocks sorted by the Baker–Wurgler
sentiment beta. Over our sample period, this tradable sentiment factor has a positive mean of 0.28%
(t-statistic = 2.06).
As shown in Table VII, after controlling for the exposures to the sentiment factor and
standard risk factors, the spread in the risk-adjusted returns between the two extreme deciles of
hedge funds is 0.56% (t-statistic = 3.54) per month. Comparing this number with the corresponding
value of 0.59% (t-statistic = 3.55) per month in Table III (where we examine the same hedge fund
portfolios but do not adjust for the sentiment factor when estimating fund alpha), the
outperformance of high sentiment beta hedge funds is only 0.03% lower on average after
controlling for the sentiment factor. Therefore, our result for hedge funds is largely unexplained
by and distinct from the relation between sentiment beta and stock returns. The result also suggests
that high sentiment beta hedge funds do not simply hold positive sentiment beta stocks. Their high
sentiment exposure comes mainly from complex and dynamic trading strategies.
19
B. Skill-Based Explanation
Next, we consider the skill-based explanation, which holds that the outperformance of high
sentiment beta hedge funds is due to managerial skill. We find initial support for this possibility in
Table III, which shows that after controlling for a comprehensive set of risk factors, the top three
deciles of hedge funds (especially the top decile) ranked by sentiment beta exhibit positive alphas
that are economically and statistically significant, while the other hedge funds have insignificant
alpha. Below we examine the skill-based explanation in detail.
B.1. Hedge Fund Skill and the Sentiment Beta-Fund Performance Relation
The literature has identified several hedge fund characteristics that are related to
managerial skill. For example, skilled hedge funds tend to charge higher management and
incentive fees, adopt a high-water mark, and impose longer lockup and notice periods (e.g.,
Ackermann, McEnally, and Ravenscraft (1999), Agarwal, Daniel, and Naik (2009)). Furthermore,
Titman and Tiu (2011) propose a hedge fund skill measure based on fund returns’ exposure to
factor risks. According to this measure, low-skill managers, who are less confident in their ability
to generate alpha from active strategies, choose greater exposure to systematic factors. This implies
that low-skill managers’ funds will have higher R2 with respect to systematic factors. In contrast,
high-skill managers will have lower R2.
Consistent with the skill explanation, we find that the outperformance of high sentiment
beta hedge funds is much stronger among high-skill hedge funds. Panel A of Table VIII reports
results for two subsamples of hedge funds sorted by the Titman–Tiu fund skill measure. Here, the
spread between high and low sentiment beta funds has a mean return of 0.41% (t-statistic = 4.07)
20
To shed light on the specific managerial skill that helps improve fund performance in the
presence of sentiment fluctuations, we examine whether hedge funds can time changes in investor
sentiment and position accordingly. Specifically, we test three conjectures. First, skilled fund
managers can time their exposures to investor sentiment, increasing sentiment beta of their funds
if sentiment change is forecast to be positive. This strategy is akin to bubble-riding, except that it
applies more generally than just during bubble episodes. Our second conjecture is that hedge funds
able to time sentiment fluctuations tend to have higher sentiment beta. Although skilled managers
may reduce sentiment exposure when sentiment is expected to decrease, this effect should be
relatively weak because there are fewer opportunities to front-run sentiment traders on the
21
We begin by examining sentiment timing among hedge funds. In general, timing ability
refers to the ability of fund managers to adjust factor exposures at opportune times as market
conditions change.19 Theoretical justification for sentiment timing can be found in DeLong et al.
(1990b), who argue that arbitrageurs may jump on the bandwagon and purchase ahead of sentiment
traders if the latter follow positive-feedback strategies. Brunnermeier and Nagel (2004) provide
evidence that hedge funds in aggregate were able to predict investor sentiment during the tech
bubble. Here, we test sentiment timing at the individual hedge fund level during more general
market states over a longer period spanning 1994 to 2018.
We propose a sentiment timing model based on the classic market timing test of Henriksson
and Merton (1981) in which fund managers have higher (lower) exposure to the stock market when
market returns are expected to be higher (lower). Since the sentiment changes index itself does not
capture investment returns, we examine the dynamics of hedge fund exposure to the tradable
sentiment factor, as proxied by the return spread between the portfolio of stocks with high
sentiment beta (top decile) and the portfolio of stocks with low sentiment beta (bottom decile).
We perform the analysis at the individual fund level rather than at the hedge fund index
level as not all hedge funds are expected to be able to time sentiment. More importantly, we are
interested in whether heterogeneity in hedge fund sentiment timing skill is related to the cross-
sectional dispersion in hedge fund sentiment beta and performance. Specifically, for each hedge
fund with at least 30 monthly return observations, we perform the sentiment timing regression
19
Prior hedge fund studies find evidence of timing skill with respect to market returns, volatility, liquidity, and macro
uncertainty (e.g., Chen (2007), Chen and Liang (2007), Cao et al. (2013), Bali, Brown, and Caglayan (2014)).
22
Jagannathan and Korajczyk (1986) point out that fund managers without true timing skill
may sometimes appear as successful market timers under the Henriksson–Merton timing model
when they adopt strategies with nonlinear payoffs, even though these strategies do not contribute
to fund performance. This caution applies to our sentiment timing test for hedge funds because
regression (4) follows the spirit of the Henriksson–Merton model and hedge funds tend to use
derivatives and dynamic trading strategies (e.g., Chen (2011)). To verify the validity of our
sentiment timing proxy, we investigate the relation between the sentiment timing coefficient and
hedge fund performance (see Table X). The idea here is that, if the sentiment timing coefficient
captures at least part of true timing skill, it should be positively correlated with fund performance.
Panel A of Table IX reports the cross-sectional distribution of the t-statistic for the
sentiment timing coefficient γ. The panel shows the percentage of t-statistics that exceed the
indicated cutoff values under the assumption of a normal distribution. In our sample, 13.82% of
funds have a t-statistic greater than 1.65 (i.e., 5% significance level in the right tail under
normality), whereas only 2.77% have a t-statistic less than -1.65 (5% significance level in the left
tail under normality). Thus, the right tail is thicker than the left tail, which suggests that the positive
sentiment timing coefficient is more pronounced than the negative sentiment timing coefficient.
20
Equation (4) uses the ex post mean of the sentiment factor return as the reference point for hedge funds. This practice,
which follows the market timing literature, should not cause a concern as we are interested in evaluating whether a
fund’s loading is large when the realized factor return is high, as opposed to proposing an implementable trading
strategy based on available information in real time.
23
The above inference is drawn under the assumption of normality, but hedge fund returns
are not normally distributed (e.g., Fung and Hsieh (1997)). We therefore apply the bootstrap
technique, which imposes no assumption of normality, to analyze statistical significance of the
sentiment timing coefficient. We use three alternative bootstrap approaches. The baseline approach
follows Kosowski et al. (2006) and assumes independently and identically distributed (IID)
residuals from the sentiment timing regression. The second approach accounts for serial correlation
in the residuals, as in Cao et al. (2013). When implementing these two approaches, we require each
fund to have at least 30 observations. The third approach follows Fama and French (2010) and
allows for cross-sectional correlation in the residuals. When implementing this approach, we
follow Harvey and Liu (2020) and require each fund to have at least 60 observations. Details on
these bootstrap approaches are provided in the Appendix.
Panel B of Table IX shows that the results hold across three bootstrap approaches. The
reported empirical p-values correspond to the t-statistics of the sentiment timing coefficients in
both tails. The empirical p-values for the right tail are below the usual threshold for statistical
significance. This holds true across the alternative approaches. Meanwhile, the empirical p-values
for the left tail are all close to one. Thus, the result suggests that the top sentiment timing
coefficients are unlikely due to pure luck. Our evidence of sentiment timing for hedge funds
suggests that the bubble-riding type of sentiment trading first documented in Brunnermeier and
Nagel (2004) during the tech bubble episode holds more generally. Our results also reveal large
heterogeneity in the sentiment timing coefficient across individual hedge funds. The tests in Table
X take advantage of this heterogeneity.
Next, we check whether sentiment timing is related to sentiment beta in the cross section
of hedge funds. We find that the two variables are significantly positively correlated, as shown in
24
Finally, consistent with our third conjecture, we find that the sentiment timing coefficient
does indeed contribute to hedge fund performance. In Panel B of Table X, we perform Fama–
MacBeth regressions of one-month-ahead fund performance (excess return or alpha) on the
sentiment timing coefficient. As in Section II, the sentiment timing coefficient is estimated from
running regression (4) using data from a 36-month backward-looking rolling window. Regardless
of whether we use the excess return or alpha as the dependent variable, the coefficient of the
sentiment timing skill is positive and statistically significant. Meanwhile, sentiment beta continues
to exhibit a strong relation with hedge fund performance. We also confirm the positive relation
between our measure of sentiment timing skill and hedge fund performance using the portfolio
sorting approach.21 These findings again suggest that our inference about sentiment timing skill is
unlikely to be artificial.
To summarize, we find that hedge funds with positive sentiment timing skill have higher
sentiment beta as well as superior performance. However, sentiment timing ability seems to
account for only a small portion of the effect of sentiment beta on hedge fund performance, leaving
a large portion of the effect unexplained. A more comprehensive study of hedge funds’ sentiment
trading strategies would lead to a better understanding of hedge fund performance in general, and
in particular its relation to sentiment beta documented in this paper.
21
We find that the decile portfolio of hedge funds with the highest sentiment timing coefficients significantly
outperforms the decile portfolio of hedge funds with the lowest sentiment timing coefficients. The spread in alpha
between these two decile portfolios is 0.44% (t-statistic = 2.80) per month. See the Internet Appendix for details.
25
Following the literature, we partition the full sample into high and low sentiment periods
based on whether the level of investor sentiment in each month exceeds the time-series median.
We then examine the relation between sentiment beta and fund performance following the different
periods. As shown in Table XI, the spread in excess returns between the two extreme decile
portfolios of hedge funds sorted by sentiment beta is 0.34% (0.29%) per month following high
(low) sentiment periods. After risk adjustment, the contrast in terms of the spread in alpha becomes
starker at 0.73% (t-statistic = 3.59) versus 0.38% (t-statistic = 1.50), with the spread in alpha
following high sentiment periods nearly twice as large as that following low sentiment periods.
This result is at odds with the conventional wisdom that hedge funds generate alpha by
betting against mispricing and that their main strength lies in shorting overpriced securities. That
view implies that hedge funds with negative sentiment beta would do well, especially following
periods of high sentiment, which is opposite of what we find in the data. Our result does lend
support, however, to the view that skilled managers can improve fund performance with the
bubble-riding type of sentiment trading that generates a positive sentiment beta.
22
See Baker and Wurgler (2006), Brown and Cliff (2005), Kumar and Lee (2006), Lemmon and Portniaguina (2006),
Yu and Yuan (2011), Stambaugh, Yu, and Yuan (2012, 2015), Shen, Yu, and Zhao (2017), DeVault, Sias, and Starks
(2019), among others.
26
In this paper, we explore how hedge fund exposure to sentiment fluctuations (i.e., sentiment
beta) is related to fund performance. On the one hand, unpredictable fluctuations in investor
sentiment could deter arbitrage activity. On the other hand, skilled arbitrageurs may be able to
predict and take advantage of changes in investor sentiment. Different sentiment trading strategies
by hedge funds can lead to cross-sectional variation in their sentiment exposures. We show robust
evidence that hedge funds with large positive exposures to changes in investor sentiment
significantly outperform other funds. The return spread between the top and bottom deciles of
hedge funds ranked by sentiment beta is as large as 0.59% (t-statistic = 3.55) per month on a risk-
adjusted basis.
We investigate two distinct but not mutually exclusive economic explanations. The first, a
risk-based explanation, holds that the outperformance of high sentiment beta hedge funds comes
from a sentiment risk premium on the high sentiment beta stocks they hold. The second explanation,
a skill-based story, holds that managerial skill drives the outperformance. Our analyses provide
more support for the skill-based explanation. The relation between sentiment beta and fund returns
is much stronger among skilled hedge funds. Moreover, we find evidence of sentiment timing skill
for a subset of hedge funds, with sentiment timers exhibiting both high sentiment beta and large
alpha. Thus, although sentiment fluctuations can deter arbitrage activity, some skilled arbitrageurs
are able to profit from such fluctuations (e.g., by predicting changes in sentiment). Extending the
existing evidence of Brunnermeier and Nagel (2004), our results show that the bubble-riding type
of sentiment trading that generates a positive sentiment beta can enhance fund performance beyond
the socially useful function of betting against mispricing.
27
The baseline bootstrap approach follows Kosowski et al. (2006), assuming IID residuals.
The basic idea is to resample the data randomly to generate pseudo funds that have no sentiment
timing skill but the same factor exposures as the actual funds. Empirical p-values are then
computed by comparing t-statistics of the estimated timing coefficients of the actual funds at
various cutoff percentiles with the distribution of the t-statistics of the pseudo funds at the same
cutoff percentiles. We perform the bootstrap for the t-statistic, because, as a pivotal statistic, it has
favorable sampling properties in bootstrapping (e.g., Horowitz (2001)). The bootstrap procedure
is as follows.
Step 1: Perform the sentiment timing regression (4) for fund i and store the estimated coefficients
{𝛼̂, 𝛽̂ 𝑆 , 𝛾̂, 𝛽̂ } as well as the time series of the residuals {𝜀̂𝑖,𝑡 , t = 1, …, Ti}, where Ti is the
number of monthly observations for the fund.
Step 2: Resample the regression residuals with replacement to obtain a randomly resampled time
𝑏
series {𝜀̂𝑖,𝑡 }, where b is the index of bootstrap iteration (b = 1, 2, …, B). Next, generate
𝑏
monthly excess returns {𝑟̂𝑖,𝑡 } for a pseudo fund that has no sentiment timing (i.e., γ = 0)
by construction. That is, set the sentiment timing coefficient to zero:
𝑏
𝑟𝑖,𝑡 = 𝛼̂ + 𝛽̂ 𝑆 𝛥sentiment 𝑡 + 𝛽̂ ′𝒇𝑡 + 𝜀̂𝑖,𝑡
𝑏
. (5)
Step 3: Run the sentiment timing regression (4) using the pseudo fund returns from Step 2 and
store the t-statistic of the estimated timing coefficient. Since the pseudo fund has a true γ
of zero, any nonzero timing coefficient comes from randomness.
Step 4: Complete Steps 1 to 3 for all funds so that we can observe the cross-sectional statistics
(e.g., the top 10th percentile) of the t-statistic for the pseudo funds.
28
We consider two extensions to the baseline bootstrap to account for non-IID residuals. Both
extensions modify Step 2; the other steps remain unchanged.
The first extension accounts for serial correlation in the residuals using the sieve bootstrap
(e.g., Bühlmann (1997)), as in Cao et al. (2013). This approach assumes that the residuals {𝜀̂𝑖,𝑡 } of
fund i follow a p-order autoregressive process, that is, AR(p), with the value p (up to 6) chosen by
the Akaike information criterion for each fund separately. Next, we resample error terms from the
𝑏
AR model and generate bootstrap residuals {𝜀̂𝑖,𝑡 } by plugging the resampled error terms into the
model. With the bootstrap residuals, we perform the remaining steps as before.
The second extension accounts for cross-sectional correlation in the residuals, using the
approach of Fama and French (2010). This approach resamples factors and residuals jointly across
all funds. Recently, Harvey and Liu (2020) show that the Fama–French approach, when including
funds with short histories, has low power to detect skilled funds. They recommend requiring at
least 60 observations for each fund when using the approach.
29
Abreu, Dilip, and Markus Brunnermeier, 2002, Synchronization risk and delayed arbitrage,
Journal of Financial Economics 66, 341–360.
Ackermann, Carl, Richard McEnally, and David Ravenscraft, 1999, The performance of hedge
funds: Risk, return, and incentives, Journal of Finance 54, 833–874.
Agarwal, Vikas, Naveen Daniel, and Narayan Naik, 2009, Role of managerial incentives and
discretion in hedge fund performance, Journal of Finance 64, 2221–2256.
Agarwal, Vikas, and Narayan Naik, 2004, Risks and portfolio decisions involving hedge funds,
Review of Financial Studies 17, 63–98.
Akbas, Ferhat, William Armstrong, Sorin Sorescu, and Avanidhar Subrahmanyam, 2015, Smart
money, dumb money, and equity return anomalies, Journal of Financial Economics 118, 355–382.
Baker, Malcolm, and Jeffrey Wurgler, 2006, Investor sentiment and the cross-section of stock
returns, Journal of Finance 61, 1645–1680.
Baker, Malcolm, and Jeffrey Wurgler, 2007, Investor sentiment in the stock market, Journal of
Economic Perspectives 21, 129–151.
Bali, Turan, Stephen Brown, and Mustafa Caglayan, 2011, Do hedge funds’ exposures to risk
factors predict their future returns? Journal of Financial Economics 101, 36–68.
Bali, Turan, Stephen Brown, and Mustafa Caglayan, 2012, Systematic risk and the cross-section
of hedge fund returns, Journal of Financial Economics 106, 114–131.
Bali, Turan, Stephen Brown, and Mustafa Caglayan, 2014, Macroeconomic risk and hedge fund
returns, Journal of Financial Economics 114, 1–19.
Barberis Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor sentiment,
Journal of Financial Economics 49, 307–343.
Brown, Gregory, and Michael Cliff, 2005, Investor sentiment and asset valuation, Journal of
Business 78, 405–440.
Brown, Stephen, William Goetzmann, and Roger Ibbotson, 1999, Offshore hedge funds: Survival
and performance, 1989-95, Journal of Business 72, 91–117.
30
Bühlmann, Peter, 1997, Sieve bootstrap for time series, Bernoulli 3,123–148.
Campbell, John, and Albert Kyle, 1993, Smart money, noise trading and stock price behavior,
Review of Economic Studies 60, 1–34.
Cao, Charles, Yong Chen, Bing Liang, and Andrew Lo, 2013, Can hedge funds time market
liquidity? Journal of Financial Economics 109, 493–516.
Chen, Yong, 2007, Timing ability in the focus market of hedge funds, Journal of Investment
Management 5, 66–98.
Chen, Yong, 2011, Derivatives use and risk taking: Evidence from the hedge fund industry,
Journal of Financial and Quantitative Analysis 46, 1073–1106.
Chen, Yong, Michael Cliff, and Haibei Zhao, 2017, Hedge funds: The good, the bad, and the lucky,
Journal of Financial and Quantitative Analysis 52, 1081–1109.
Chen, Yong, Zhi Da, and Dayong Huang, 2019, Arbitrage trading: The long and the short of it,
Review of Financial Studies 32, 1608–1646.
Chen, Yong, Bryan Kelly, and Wei Wu, 2020, Sophisticated investors and market efficiency:
Evidence from a natural experiment, Journal of Financial Economics 138, 316–341.
Chen, Yong, and Bing Liang, 2007, Do market timing hedge funds time the market? Journal of
Financial and Quantitative Analysis 42, 827–856.
Da, Zhi, Joey Engelberg, and Pengjie Gao, 2015, The sum of all FEARS: Investor sentiment and
asset prices, Review of Financial Studies 28, 1–32.
DeLong, Bradford, Andrei Shleifer, Lawrence Summers, and Robert Waldmann, 1990a, Noise
trader risk in financial market, Journal of Political Economy 98, 703–738.
DeLong, Bradford, Andrei Shleifer, Lawrence Summers, and Robert Waldmann, 1990b, Positive
feedback investment strategies and destabilizing rational speculation, Journal of Finance 45, 375–
395.
DeVault, Luke, Richard Sias, and Laura Starks, 2019, Sentiment metrics and investor demand,
Journal of Finance 74, 985–1024.
31
Fama, Eugene, and James MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal
of Political Economy 81, 607–636.
Fama, Eugene, and Kenneth French, 2010, Luck versus skill in the cross section of mutual fund
returns, Journal of Finance 65, 1915–1947.
Fung, William, and David Hsieh, 1997, Empirical characteristics of dynamic trading strategies:
The case of hedge funds, Review of Financial Studies 10, 275–302.
Fung, William, and David Hsieh, 2004, Hedge fund benchmarks: A risk-based approach, Financial
Analysts Journal 60, 65–80.
Fung, William, David Hsieh, Narayan Naik, and Tarun Ramadorai, 2008, Hedge funds:
Performance, risk, and capital formation, Journal of Finance 63, 1777–1803.
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, 529–610.
Glushkov, Denys, 2006, Sentiment beta, Working paper, University of Texas at Austin.
Griffin, John, Jeffrey Harris, Tao Shu, and Selim Topaloglu, 2011, Who drove and burst the tech
bubble? Journal of Finance 66, 1251–1290.
Griffin, John, and Jin Xu, 2009, How smart are the smart guys? A unique view from hedge fund
stock holdings, Review of Financial Studies 22, 2531–2570.
Han, Bing, 2008, Investor sentiment and option prices, Review of Financial Studies 21, 387–414.
Harvey, Campbell, and Yan Liu, 2020, Luck versus skill in the cross section of mutual fund returns:
Reexamining the evidence, Working paper, Duke University and Purdue University.
Henriksson, Roy, and Robert Merton, 1981, On market timing and investment performance II:
Statistical procedures for evaluating forecasting skills, Journal of Business 54, 513–533.
Ho, Jerry, and Daniel Hung, 2012, Predicting stock market returns and volatility with investor
sentiment: Evidence from eight developed countries, Working paper, Queensland University of
Technology and University of Glasgow.
32
Jagannathan, Ravi, and Robert Korajczyk, 1986, Assessing the market timing performance of
managed portfolios, Journal of Business 59, 217–235.
Jagannathan, Ravi, Alexey Malakhov, and Dmitry Novikov, 2010, Do hot hands exist among
hedge fund managers? An empirical evaluation, Journal of Finance 65, 217–255.
Keynes, John M., 1936, The General Theory of Employment, Interest and Money, (Harcourt Brace
and Company, NY).
Kosowski, Robert, Narayan Naik, and Melvyn Teo, 2007, Do hedge funds deliver alpha? A
Bayesian and bootstrap analysis, Journal of Financial Economics 84, 229–264.
Kosowski, Robert, Allan Timmermann, Russ Wermers, and Hal White, 2006, Can mutual fund
“stars” really pick stocks? New evidence from a bootstrap analysis, Journal of Finance 61, 2551–
2595.
Kozak, Serhiy, Stefan Nagel, and Shrihari Santosh, 2018, Interpreting factor models, Journal of
Finance 73, 1183–1223.
Kumar, Alok, and Charles Lee, 2006, Retail investor sentiment and return comovements, Journal
of Finance 61, 2451–2486.
Lee, Charles, Andrei Shleifer, and Richard Thaler, 1991, Investor sentiment and the closed-end
fund puzzle, Journal of Finance 46, 75–109.
Lemmon, Michael, and Evgenia Portniaguina, 2006, Consumer confidence and asset prices: Some
empirical evidence, Review of Financial Studies 19, 1499–1529.
Liang, Samuel, 2018, The systematic pricing of market sentiment shock, European Journal of
Finance 24, 1835–1860.
Newey, Whitney, and Kenneth West, 1987, A simple, positive semi-definite, heteroskedasticity
and autocorrelation consistent covariance matrix, Econometrica 55, 703–708.
Pastor, Lubos, and Robert Stambaugh, 2003, Liquidity risk and expected stock returns, Journal of
Political Economy 111, 642–685.
33
Shen, Junyan, Jianfeng Yu, and Shen Zhao, 2017, Investor sentiment and economic forces, Journal
of Monetary Economics 86, 1–21.
Stambaugh, Robert, Jianfeng Yu, and Yu Yuan, 2012, The short of it: Investor sentiment and
anomalies, Journal of Financial Economics 104, 288–302.
Stambaugh, Robert, Jianfeng Yu, and Yu Yuan, 2015, Arbitrage asymmetry and the idiosyncratic
volatility puzzle, Journal of Finance 70, 1903–1948.
Sun, Zheng, Ashley Wang, and Lu Zheng, 2012, The road less traveled: Strategy distinctiveness
and hedge fund performance, Review of Financial Studies 25, 96–143.
Teo, Melvyn, 2009, The geography of hedge funds, Review of Financial Studies 22, 3531–3561.
Teo, Melvyn, 2011, The liquidity risk of liquid hedge funds, Journal of Financial Economics 100,
24–44.
Titman, Sheridan, and Christian Tiu, 2011, Do the best hedge funds hedge? Review of Financial
Studies 24, 123–168.
Wurgler, Jeffrey, and Ekaterina Zhuravskaya, 2002, Does arbitrage flatten demand curves for
stocks? Journal of Business 75, 583–608.
Yu, Jianfeng, and Yu Yuan, 2011, Investor sentiment and the mean-variance relation, Journal of
Financial Economics 100, 367–381.
34
This table summarizes the hedge fund sample that covers both active and defunct hedge funds. For each fund, the first
12 months of returns are excluded to mitigate backfill bias. The sample includes equity-oriented hedge funds that
report monthly net-of-fee returns in U.S. dollars, allow for redemption at a monthly or higher frequency, and have
assets under management of at least $5 million. Each hedge fund is required to have at least 30 return observations.
The sample contains 4,073 hedge funds. The summary statistics are based on fund-month observations. All variables
are winsorized at the 1% and 99% levels. The sample period is from January 1994 to December 2018.
35
Panel A describes the measures of sentiment fluctuations, including the Baker–Wurgler (2007) sentiment changes index, changes in the University of Michigan
consumer sentiment index, and the FEARS index. Panel B reports the correlation coefficients between these indexes and risk factors. The risk factors include the
Fung–Hsieh (2004) seven factors (market excess returns (Mktrf), a size factor (SMB), a tradable factor mimicking the change in the constant-maturity yield of the
10-year Treasury (∆Term), a tradable factor mimicking the change in the yield spread between Moody’s Baa bond and the 10-year Treasury bond (∆Credit), and
three trend-following factors on bonds (Ptfsbd), currencies (Ptfsfx), and commodities (Ptfscom)), as well as the momentum factor (UMD), the Pastor–Stambaugh
(2003) liquidity factor (LIQ), the inflation rate (INF), and the default spread between the yields on Baa-rated and Aaa-rated corporate bonds (DEF).
36
This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based on sentiment beta.
In each month for each hedge fund with at least 30 return observations over the past 36 months, sentiment beta is
estimated by regressing the fund excess returns on the Baker–Wurgler (2007) sentiment changes index, controlling
for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term, ∆Credit, and three
trend-following factors on bonds, currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003)
liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment beta, we form 10 equal-
weighted portfolios and track their returns over the next month. The portfolios are rebalanced each month. Using the
monthly time series of the returns of each portfolio, we estimate alpha by regressing the portfolio excess returns on
the Fung–Hsieh seven factors, the momentum factor, and the Pastor–Stambaugh liquidity factor. Both monthly excess
return and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags.
37
This table reports results from Fama–MacBeth (1973) cross-sectional regressions of hedge fund excess return, as well
as alpha, on sentiment beta, controlling for fund characteristics and style dummies. In each month and for each hedge
fund with at least 30 return observations over the past 36 months, sentiment beta is estimated by regressing the fund
excess returns on the Baker–Wurgler (2007) sentiment changes index with controls for the Fung–Hsieh (2004) seven
factors (including market excess returns, a size factor, ∆Term, ∆Credit, and three trend-following factors on bonds,
currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003) liquidity factor, the inflation rate,
and the default spread. Then, we perform cross-sectional regressions of fund excess return, or alpha, over the next
month on sentiment beta with controls for fund characteristics and style dummies. The fund characteristics include
fund size, fund age, management fee, incentive fee, a high-water mark dummy equal to one if a high-water mark
provision is used and zero otherwise, lockup period, and redemption notice period. Both monthly excess return and
alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags.
Dependent Variable
Excess Return Alpha
Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat
Sentiment beta 0.17 3.03 0.16 2.84 0.14 3.31 0.12 3.16
Log(fund size) 0.01 0.12 0.01 0.21
Log(fund age) -0.03 -0.95 0.06 2.13
Management fee 0.04 1.34 0.02 1.14
Incentive fee -0.01 -0.14 0.01 2.31
High-water mark 0.15 4.76 0.11 4.81
Lockup period 0.06 1.55 -0.06 -2.24
Notice period 0.06 3.80 0.07 5.83
Fund style dummies No Yes No Yes
2
Adjusted R 0.01 0.07 0.01 0.05
38
This table reports results of portfolio sorts based on sentiment beta with respect to changes in the University of
Michigan consumer sentiment index. In each month for each hedge fund with at least 30 return observations over the
past 36 months, sentiment beta is estimated by regressing the fund excess returns on changes in the Michigan consumer
sentiment index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor,
∆Term, ∆Credit, and three trend-following factors on bonds, currencies, and commodities), the momentum factor, the
Pastor–Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment
beta, we form 10 equal-weighted portfolios and track their returns over the next month. The portfolios are rebalanced
each month. Using the monthly time series of the returns of each portfolio, we estimate alpha by regressing portfolio
excess returns on the Fung–Hsieh seven factors, the momentum factor, and the Pastor–Stambaugh liquidity factor.
Both monthly excess return and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard
errors with two lags.
39
This table reports monthly gross (i.e., before-fee) returns of 10 equal-weighted portfolios of hedge funds constructed
based on sentiment beta. In each month for each hedge fund with at least 30 return observations over the past 36
months, sentiment beta is estimated by regressing fund excess returns on the Baker–Wurgler (2007) sentiment changes
index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term,
∆Credit, and three trend-following factors on bonds, currencies, and commodities), the momentum factor, the Pastor–
Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment beta, we
form 10 equal-weighted portfolios and track their returns over the next month. The portfolios are rebalanced each
month. Using the monthly time series of the returns of each portfolio, we estimate alpha by regressing portfolio excess
returns on the Fung–Hsieh seven factors, the momentum factor, and the Pastor–Stambaugh liquidity factor. Both
monthly excess return and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors
with two lags.
40
This table reports the beta on the tradable sentiment factor and the monthly alpha of 10 equal-weighted portfolios of
hedge funds constructed based on sentiment beta. In each month for each hedge fund with at least 30 return
observations over the past 36 months, sentiment beta is estimated by regressing fund excess returns on the Baker–
Wurgler (2007) sentiment changes index, controlling for the Fung–Hsieh (2004) seven factors (including market
excess returns, a size factor, ∆Term, ∆Credit, and three trend-following factors on bonds, currencies, and commodities),
the momentum factor, the Pastor–Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. Based
on the funds’ sentiment beta, we form 10 equal-weighted portfolios and track their returns over the next month. The
portfolios are rebalanced each month. Using the monthly time series of the returns of each portfolio, we estimate
sentiment risk-adjusted alpha by regressing portfolio excess returns on the tradable sentiment factor, in addition to the
Fung–Hsieh seven factors, the momentum factor, and the Pastor–Stambaugh liquidity factor. The column “Alpha”
reports the monthly sentiment risk-adjusted alpha (in percent) for each decile portfolio of hedge funds sorted by
sentiment beta. t-statistics are based on Newey–West (1987) standard errors with two lags.
41
This table reports results of portfolio sorts based on sentiment beta for subsample tests. In Panel A, we partition the
hedge fund sample into high- versus low-skill funds according to the Titman–Tiu (2011) hedge fund skill measure.
Specifically, in each month for each hedge fund with at least 30 return observations over the past 36 months, we
estimate the R2 by regressing fund returns on the Fung–Hsieh (2004) seven factors (including market excess returns,
a size factor, ∆Term, ∆Credit, and three trend-following factors on bonds, currencies, and commodities). A fund is
classified as a high-skill fund if its R2 is below the median level. We then report excess return and alpha for the 10
equal-weighted portfolios of hedge funds constructed based on sentiment beta for each subsample. In Panel B, we
partition the hedge fund sample based on the median level of fund characteristics, including fund size, fund age,
management fee, incentive fee, high-water mark dummy, lockup period, and redemption notice period. We then report
the spreads in excess return and alpha between the top and bottom sentiment beta decile portfolios (i.e., portfolio 10
and portfolio 1) for the subsamples of hedge funds with values of fund characteristics above and below the median
level, separately. In this panel, t-statistics are reported in parentheses. Both monthly excess return and alpha are in
percent. t-statistics are based on Newey–West (1987) standard errors with two lags.
42
43
This tables reports results of sentiment timing. Panel A presents the cross-sectional distribution of t-statistics for the
sentiment timing coefficient across funds. The numbers in parentheses are the significance level under the normality
assumption. Panel B presents bootstrap results. The first row reports ranked t-statistics of the sentiment timing
coefficient, requiring each fund to have at least 30 observations. The second row reports empirical p-values from the
baseline approach assuming IID regression residuals. The third row reports empirical p-values from the approach
accounting for serial correlation in residuals, as in Cao et al. (2013). The fourth row reports ranked t-statistics of the
sentiment timing coefficient, requiring each fund to have at least 60 observations. The last row reports empirical p-
values from the Fama and French (2010) approach, accounting for cross-sectional correlation in residuals. The number
of resampling iterations is 1,000. t-statistics are based on Newey–West (1987) standard errors with two lags.
44
In this table, Panel A reports the cross-sectional correlation between sentiment timing skill and sentiment beta among
hedge funds. Panel B reports results from Fama–MacBeth (1973) regressions of hedge fund excess returns, as well as
alpha, on the sentiment timing coefficient. In each month for each hedge fund with at least 30 return observations over
the past 36 months, sentiment timing is estimated from regression (4). We then perform cross-sectional regressions of
fund excess returns, or alpha, over the next month on the sentiment timing coefficient together with sentiment beta,
various fund characteristics, and style dummies. The fund characteristics include fund size, fund age, management fee,
incentive fee, high-water mark dummy equal to one if a high-water mark provision is used and zero otherwise, lockup
period, and redemption notice period. Both monthly excess return and alpha are expressed in percent. t-statistics are
based on Newey–West (1987) standard errors with two lags.
Panel A: Correlation Between the Sentiment Timing Skill and Sentiment Beta
Sentiment Timing Sentiment Beta
Sentiment timing 1.00
45
This table reports results of portfolio sorts based on sentiment beta following periods of high and low aggregate
investor sentiment. We partition the full sample into high- and low-sentiment periods based on whether the level of
investor sentiment exceeds the time-series median. We then examine the sentiment beta-fund performance relation
following the two periods separately. In each month for each hedge fund with at least 30 return observations over the
past 36 months, sentiment beta is estimated by regressing fund excess returns on the Baker–Wurgler (2007) sentiment
changes index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor,
∆Term, ∆Credit, and three trend-following factors on bonds, currencies, and commodities), the momentum factor, the
Pastor–Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment
beta, we form 10 equal-weighted portfolios and track their returns over the next month. The portfolios are rebalanced
each month. Using the monthly time series of the returns of each portfolio, we estimate alpha by regressing portfolio
excess returns on the Fung–Hsieh seven factors, the momentum factor, and the Pastor–Stambaugh liquidity factor.
Both monthly excess return and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard
errors with two lags.
Portfolio Excess Return t-stat Alpha t-stat Excess Return t-stat Alpha t-stat
46
3.00%
2.00%
1.00%
0.00%
-1.00%
-2.00%
-3.00%
-4.00%
-5.00%
200.00%
150.00%
100.00%
50.00%
0.00%
Figure 1. Spread in risk-adjusted return (i.e., alpha) between top and bottom sentiment beta hedge funds. Each
month starting in December 1996, we form 10 decile portfolios based on hedge fund sentiment beta estimated over the
past 36 months and track their returns over the next month. Panel A plots the time series of the spread in monthly risk-
adjusted returns between the two extreme decile portfolios of top versus bottom sentiment beta hedge funds. Panel B
plots the cumulative spread in risk-adjusted returns between the two extreme decile portfolios.
47
This Internet Appendix provides ancillary tests and results for the main paper. In Section
I, we check the robustness of our inference to the choice of hedge fund categories included in the
sample. Section II examines the robustness to alternative sample filters. Section III presents results
when we skip one month between the portfolio formation period and the portfolio holding period.
In Section IV, we address the concern related to the small degrees of freedom in our regression
estimating sentiment beta. In Section V, we examine the relation between sentiment beta and hedge
fund performance over horizons of three to 12 months after portfolio formation. Section VI checks
the stability of hedge funds’ sentiment beta over time. Finally, in Section VII, we perform the
In this section, we evaluate whether our inference is robust to the choice of hedge fund
categories included in the sample. Lipper TASS classifies hedge funds into 11 categories:
convertible arbitrage, dedicated short bias, emerging markets, event driven, equity market neutral,
fixed income arbitrage, funds of funds, global macro, long/short equity, managed futures, and
multi strategy. In the main paper, we focus on U.S. equity-oriented hedge funds, since the Baker-
Wurgler (2006, 2007) sentiment measure corresponds largely to U.S. stock markets. As a result,
we exclude emerging markets, fixed income arbitrage, and managed futures. This choice of equity-
oriented hedge fund categories follows Cao et al. (2013). In addition, we exclude dedicated short-
Chen, Yong, Bing Han, and Jing Pan, Internet Appendix for “Sentiment Trading and Hedge Fund
Returns,” Journal of Finance [DOI STRING]. Please note: Wiley-Blackwell is not responsible for
the content or functionality of any supporting information supplied by the authors. Any queries
(other than missing material) should be directed to the authors of the article.
includes seven equity-oriented categories. Below we include all 11 hedge fund style categories in
the sample. Funds are included provided they satisfy the sample filter criteria described in Section
Table IA.I reports results from portfolio sorts. We find that the top decile of hedge funds
sorted by sentiment beta has an average excess return of 0.63% (t-statistic = 3.58) per month,
whereas the bottom decile of hedge funds sorted by sentiment beta has an average excess return of
0.24% (t-statistic = 1.60) per month. Thus, hedge funds in the top decile outperform those in the
bottom decile by 0.39% (t-statistic = 3.04) per month on average. After risk adjustment, the spread
in alpha between the two extreme decile portfolios is 0.52% (t-statistic = 2.61). These results are
Therefore, based on the alternative sample containing all hedge fund categories, we
continue to find that sentiment beta significantly predicts hedge fund performance in the cross
section, even after adjusting for exposures to standard risk factors. These results suggest that our
This section examines the robustness of our inference to alternative sample filters. In the
main paper, we delete the first 12 months of returns for each hedge fund to mitigate backfill bias
and exclude hedge funds with assets under management (AUM) less than $5 million from the
sample to avoid possible bias associated with small funds. Here, we apply alternative filters by
deleting the first 24 months of returns and excluding the funds with AUM less than $10 million.
The other screens such as hedge fund categories remain the same as in the main paper.
Table IA.II reports results from portfolio sorts. After applying the alternative sample filters,
we find that the portfolio of top sentiment beta hedge funds has the highest average return and
alpha among the 10 decile portfolios. In particular, the portfolio of top sentiment beta hedge funds
outperforms the portfolio of bottom sentiment beta hedge funds by 0.34% (t-statistic = 3.17) per
finding of a positive relation between sentiment beta and hedge fund returns is robust to the
Our results from portfolio sorts reported in the main paper are based on one-month-ahead
portfolio returns immediately after portfolio formation. In practice, it may not be possible to invest
in these portfolios immediately after formation. Here, we skip one month between the portfolio
formation period and the holding period. This approach is similar to that in Jegadeesh and Titman
(1993), who skip one month between the portfolio formation period and the holding period when
As shown in Table IA.III, the results from skipping one month after portfolio formation
are comparable to those reported in the main paper. For example, the spread in alpha between the
two extreme decile portfolios of hedge funds based on sentiment beta is 0.66% (t-statistic = 3.99)
per month when skipping one month, compared with 0.59% (t-statistic = 3.55) when not skipping
one month.
Given the relatively large number of factors (12 in the baseline model) used in the rolling
regression to estimate sentiment beta with a 36-month window, a concern that arises is that the
degrees of freedom are small and thus the sentiment beta estimate may be of low precision. Such
a problem should not be severe, since we only use the regression coefficient (i.e., the factor
loading), rather than its standard error or t-statistic, in the tests of portfolio sorts and Fama–
MacBeth (1973) regressions. Nonetheless, to further address the concern, we perform the
In the test, we use a two-step procedure to estimate sentiment beta for each fund. In the
first step, we perform the baseline model regression by including all of the factors. In the second
step, we repeat the regression by including the sentiment changes index along with those factors
step. We then measure sentiment beta as the loading on the sentiment changes index from the
second-step regression. The results are reported Table IA.IV. The spread in alpha is 0.50% (t-
statistic = 2.92) per month between the two extreme decile portfolios. Thus, consistent with the
evidence presented in the main paper, we find a significantly positive relation between sentiment
V. The Relation between Sentiment Beta and Hedge Fund Returns over Different Horizons
In the main paper, we find evidence of a positive relation between sentiment beta and hedge
fund returns over the next month. Here, we reexamine this relation by tracking returns of decile
portfolios sorted by sentiment beta over holding horizons ranging from three months to 12 months
Table IA.V shows that, in general, the spreads in average return and alpha decline with the
length of the holding period. For example, the spread in alpha between the top and bottom decile
portfolios sorted by sentiment beta is 0.59% (t-statistic = 3.66), 0.51% (t-statistic = 3.38), 0.44%
(t-statistic = 3.22), and 0.31% (t-statistic = 2.51) per month for holding periods of three, six, nine,
The performance difference related to sentiment beta can be seen more directly in Figure
IA.1. This figure plots the risk-adjusted return for decile portfolios of top versus bottom sentiment
beta hedge funds over different holding periods. The portfolio of top sentiment beta hedge funds
has positive alphas that range from 0.28% to 0.48% per month over the different holding periods,
whereas the portfolio of bottom sentiment beta hedge funds has negative alphas over the same
holding periods.
In this section, we check the stability of hedge funds’ sentiment beta over time based on
transition matrices. Table IA.VI presents transition probabilities across the sentiment beta deciles
over time for individual hedge funds. In each month and for each hedge fund, we estimate its
controlling for the risk factors described in equation (1) of the main paper. We then sort the funds
into 10 decile portfolios based on sentiment beta. In Panels A to C, we report the probabilities of
hedge funds transiting from the initial decile into another decile, or staying in the same decile, over
We find that about 84.2% (84.3%) of the hedge funds in decile portfolio 10 (1) will
continue to stay in the same decile portfolio after one month, but this fraction drops to 59.0%
(58.5%) after six months and to 45.2% (44.7%) after 12 months. This finding suggests that while
sentiment beta changes over time, there is a fair amount of persistence in the short run.
In this section, we present results based on the Financial Economic Attitudes Revealed by
Search (FEARS) index of Da, Engelberg, and Gao (2015). This index measures sentiment changes
based on Internet searches for keywords revealing investors’ bearish attitude toward the economy.
The FEARS index is available from July 2004 to December 2011 at a daily frequency. We convert
the data to a monthly frequency by averaging the daily values in each month. For consistency with
the other measures, we orthogonalize this measure against the macroeconomic variables used in
the construction of the Baker–Wurgler sentiment index. Since the FEARS index captures bearish
(rather than bullish) perceptions about the economy, it negatively correlates with the Baker–
Wurgler (2006, 2007) sentiment changes index. We therefore multiply the index by –1 so that its
regression coefficient has the same interpretation as the two other sentiment measures used in the
paper.
Table IA.VII reports the results. Overall, hedge funds realize relatively poor performance
during the July 2007 to December 2011 test period (the first 36 months are used as the initial
portfolio formation period), since a significant part of this period is the 2008 to 2009 financial
crisis. More importantly, the test based on the FEARS index shows a significant return spread of
0.35% (t-statistic = 1.98) per month between the top and bottom sentiment beta hedge fund
Our portfolio sorting analysis uses equal-weighted portfolios of hedge funds sorted by
sentiment beta. Here, we examine whether fund size is related to sentiment beta. Specifically, we
check the fund size distribution across the 10 equal-weighted portfolios. Since hedge funds’ AUM
has a time trend due to the growth of the hedge fund industry over our sample period, we focus on
the fund size percentile ranking as it is not affected by the time trend.
In Table IA.VII, we report the average fund size percentile for each of the decile portfolios
over time. As shown in the table, there is no apparent relation between sentiment beta and fund
size, in that the average fund size percentile ranking (between 1 and 100) is fairly close to 50 across
all of the decile portfolios. This finding suggests that sentiment beta and fund size are not highly
correlated.
In this section, we use the portfolio sorting analysis to examine the relation between
sentiment timing and hedge fund performance. The procedure is similar to that used in Section
II.A of the main paper. Each month starting in December 1996, we form 10 equal-weighted hedge
fund portfolios based on their sentiment timing ability estimated based on regression (4) for a
backward-looking rolling window over the past 36 months up to the given month, with the first
rolling window spanning the period from January 1994 to December 1996. We then track the
portfolio returns over the next month. These portfolios are rebalanced each month to generate the
time series of returns. Finally, we estimate alpha (i.e., risk-adjusted returns) by regressing the time
series of the excess returns of each decile portfolio on the Fung–Hsieh (2004) seven factors, the
Table IA.IX reports the results. The decile portfolio with the highest sentiment timing
ability (i.e., portfolio 10) delivers an average excess return of 0.47% per month (t-statistic = 2.93)
sentiment timing ability (i.e., portfolio 1) shows an average excess return of 0.26% (t-statistic =
1.86) per month and an alpha of -0.18 (t-statistic = -1.04) per month. Based on both excess returns
and alpha, the portfolio with the highest sentiment timing ability significantly outperforms the
portfolio with the lowest sentiment timing ability. For example, the alpha spread between the two
extreme decile portfolios is 0.44% per month (t-statistic = 2.80) and both economically and
statistically significant. Therefore, the results from portfolio sorts suggest that sentiment timing
Baker, Malcolm, and Jeffrey Wurgler, 2006, Investor sentiment and the cross-section of stock
returns, Journal of Finance 61, 1645–1680.
Baker, Malcolm, and Jeffrey Wurgler, 2007, Investor sentiment in the stock market, Journal of
Economic Perspectives 21, 129–151.
Cao, Charles, Yong Chen, Bing Liang, and Andrew Lo, 2013, Can hedge funds time market
liquidity? Journal of Financial Economics 109, 493–516.
Da, Zhi, Joey Engelberg, and Pengjie Gao, 2015, The sum of all FEARS: Investor sentiment and
asset prices, Review of Financial Studies 28, 1–32.
Fama, Eugene, and James MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal
of Political Economy 81, 607–636.
Fung, William, and David Hsieh, 2004, Hedge fund benchmarks: A risk-based approach, Financial
Analysts Journal 60, 65–80.
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers:
Implications for stock market efficiency, Journal of Finance 48, 65–91.
Newey, Whitney, and Kenneth West, 1987, A simple, positive semi-definite, heteroskedasticity
and autocorrelation consistent covariance matrix, Econometrica 55, 703–708.
Pastor, Lubos, and Robert Stambaugh, 2003, Liquidity risk and expected stock returns, Journal of
Political Economy 111, 642–685.
The sample used in the test includes all 11 hedge fund categories. This table reports monthly returns of 10 equal-
weighted portfolios of hedge funds constructed based on sentiment beta. Each month for each hedge fund with at least
30 return observations over the past 36 months, sentiment beta is estimated by regressing the fund excess returns on
the Baker–Wurgler (2007) sentiment changes index, controlling for the Fung–Hsieh (2004) seven factors (including
market excess returns, a size factor, ∆Term, ∆Credit, and three trend-following factors on bonds, currencies, and
commodities), the momentum factor, the Pastor-Stambaugh (2003) liquidity factor, the inflation rate, and the default
spread. Based on funds’ sentiment beta, we form 10 equal-weighted portfolios and track their returns over the next
month. The portfolios are rebalanced each month. Using the monthly time series of the returns of each portfolio, we
estimate alpha by regressing the portfolio excess returns on the Fung–Hsieh (2004) seven factors, the momentum
factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return and alpha are reported in percent.
t-statistics are based on Newey–West (1987) standard errors with two lags.
As alternative sample filters, we delete the first 24 months of fund returns and funds with AUM less than $10 million
from the sample. This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based
on sentiment beta. Each month for each hedge fund with at least 30 return observations over the past 36 months,
sentiment beta is estimated by regressing the fund excess returns on the Baker–Wurgler (2007) sentiment changes
index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term,
∆Credit, and three trend-following factors on bonds, currencies, and commodities), the momentum factor, the Pastor–
Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment beta, we
form 10 equal-weighted portfolios and track their returns over the next month. The portfolios are rebalanced each
month. Using the monthly time series of the returns of each portfolio, we estimate alpha by regressing portfolio excess
returns on the Fung–Hsieh (2004) seven factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity
factor. Both monthly excess return and alpha are reported in percent. t-statistics are based on Newey–West (1987)
standard errors with two lags.
10
This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based on sentiment beta.
Each month for each hedge fund with at least 30 return observations over the past 36 months, sentiment beta is
estimated by regressing the fund excess returns on the Baker–Wurgler (2007) sentiment changes index, controlling
for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term, ∆Credit, and three
trend-following factors on bonds, currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003)
liquidity factor, the inflation rate, and the default spread. Based on the funds’ sentiment beta, we form 10 equal-
weighted portfolios. When tracking the subsequent portfolio returns, we skip one month between the portfolio
formation period and the holding period. The portfolios are rebalanced each month. Using the monthly time series of
the returns of each portfolio, we estimate alpha by regressing the portfolio excess returns on the Fung–Hsieh (2004)
seven factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return
and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags.
11
This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based on the funds’
sentiment beta. Each month for each hedge fund with at least 30 returns observations over the past 36 months,
sentiment beta is estimated by regressing the fund excess returns on the Baker–Wurgler (2007) sentiment changes
index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term,
∆Credit, and three trend-following factors on bonds, foreign exchange, and commodity), the momentum factor, the
Pastor-Stambaugh (2003) liquidity factor, the inflation rate, and the default spread. To address the concern that the
number of factors in the regression model is large relative to the estimation period of 36 months, we employ a two-
step procedure to estimate sentiment beta. In the first step, we include all of the above factors in the regression. In the
second step, we include only the sentiment changes index along with those factors on which the loadings are
statistically significant at the 5% level in the first step, and we measure the fund’s sentiment beta from the second step.
Then, for each portfolio, we estimate alpha based on the monthly time series of the returns of the portfolio relative to
the Fung–Hsieh (2004) seven factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both
monthly excess return and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors
with two lags.
12
This table reports the relation between sentiment beta and hedge fund returns over different holding horizons ranging from three to 12 months. Each month for
each hedge fund with at least 30 return observations over the past 36 months, sentiment beta is estimated by regressing the fund excess returns on the Baker–
Wurgler (2007) sentiment changes index, controlling for the Fung–Hsieh (2004) seven factors (including market excess returns, a size factor, ∆Term, ∆Credit, and
three trend-following factors on bonds, currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003) liquidity factor, the inflation rate, and
the default spread. Hedge funds are then sorted into 10 equal-weighted portfolios based on their sentiment beta, and the portfolios are held for horizons of three,
six, nine, and 12 months. For each portfolio, alpha is estimated based on the monthly time series of the portfolio returns relative to the Fung–Hsieh (2004) seven
factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return and alpha are reported in percent. t-statistics based
on Newey–West (1987) standard errors with two lags are in the parentheses.
1 10 Spread
Portfolio 2 3 4 5 6 7 8 9
(Low) (High) (Port. 10–Port. 1)
Panel A: Holding period = 3 months
Excess return (%/month) 0.30 0.29 0.31 0.31 0.27 0.27 0.32 0.34 0.38 0.56 0.25
t-stat. 1.91 2.14 2.55 2.66 2.28 2.53 3.05 2.79 2.66 3.28 2.39
Alpha (%/month) -0.12 0.06 0.11 0.16 0.09 0.08 0.17 0.23 0.25 0.48 0.59
t-stat. -0.68 0.56 1.21 1.56 0.81 0.72 1.78 2.13 2.15 2.81 3.66
Panel B: Holding period = 6 months
Excess return (%/month) 0.32 0.28 0.31 0.31 0.27 0.28 0.33 0.33 0.36 0.51 0.20
t-stat. 2.02 2.05 2.50 2.57 2.30 2.62 3.05 2.76 2.47 3.08 2.09
Alpha (%/month) -0.09 0.05 0.15 0.17 0.12 0.09 0.19 0.19 0.21 0.42 0.51
t-stat. -0.55 0.37 1.49 1.44 0.99 0.81 1.91 1.65 1.78 2.29 3.38
Panel C: Holding period = 9 months
Excess return (%/month) 0.32 0.29 0.31 0.31 0.28 0.29 0.32 0.33 0.33 0.48 0.16
t-stat. 2.04 2.20 2.43 2.60 2.48 2.69 2.87 2.72 2.27 2.90 1.73
Alpha (%/month) -0.07 0.06 0.15 0.17 0.14 0.11 0.17 0.19 0.16 0.36 0.44
t-stat. -0.48 0.47 1.41 1.48 1.12 0.98 1.57 1.57 1.36 1.99 3.22
Panel D: Holding period = 12 months
Excess return (%/month) 0.32 0.32 0.32 0.32 0.29 0.29 0.31 0.31 0.33 0.45 0.12
t-stat. 2.12 2.48 2.60 2.80 2.58 2.66 2.79 2.65 2.27 2.76 1.31
Alpha (%/month) -0.03 0.11 0.16 0.19 0.14 0.12 0.14 0.16 0.16 0.28 0.31
t-stat. -0.22 0.82 1.49 1.59 1.08 0.97 1.29 1.39 1.33 1.66 2.51
13
This table presents transition matrices of sentiment beta over three-, six-, and 12-month horizons. Each month starting
in December 1996, we assign individual hedge funds into 10 decile portfolios based on their sentiment beta (estimated
from a 36-month rolling window). We then track their future re-assignments based on sentiment beta. The reported
numbers are in percent.
14
This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based on sentiment beta.
Each month for each hedge fund with at least 30 return observations over the past 36 months, sentiment beta is
estimated by regressing the fund excess returns on the FEARS index, controlling for the Fung–Hsieh (2004) seven
factors (including market excess returns, a size factor, ∆Term, ∆Credit, and three trend-following factors on bonds,
currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003) liquidity factor, the inflation rate,
and the default spread. Based on the funds’ sentiment beta, we form 10 equal-weighted portfolios and track their
returns over the next month. The portfolios are rebalanced each month. Using the monthly time series of the returns
of each portfolio, we estimate alpha by regressing the portfolio excess returns on the Fung–Hsieh (2004) seven factors,
the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return and alpha are
reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags. The FEARS index is
available from July 2004 to December 2011.
15
This table is the same as Table III in the main paper, except that we add the last column to report the average fund
size percentile for each of the 10 equal-weighted portfolios of hedge funds sorted by sentiment beta. Each month for
each hedge fund with at least 30 return observations over the past 36 months, sentiment beta is estimated by regressing
the fund excess returns on the Baker–Wurgler (2006, 2007) sentiment changes index, controlling for the Fung–Hsieh
(2004) seven factors (including market excess returns, a size factor, ∆Term, ∆Credit, and three trend-following factors
on bonds, currencies, and commodities), the momentum factor, the Pastor–Stambaugh (2003) liquidity factor, the
inflation rate, and the default spread. Based on the funds’ sentiment beta, we form 10 equal-weighted portfolios and
track their returns over the next month. The portfolios are rebalanced each month. Using the monthly time series of
the returns of each portfolio, we estimate alpha by regressing the portfolio excess returns on the Fung–Hsieh (2004)
seven factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return
and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags. In the
last column, we report the average fund size percentile for each decile portfolio.
16
This table reports monthly returns of 10 equal-weighted portfolios of hedge funds constructed based on sentiment
timing ability. Each month for each hedge fund with at least 30 return observations over the past 36 months, sentiment
timing is estimated from regression (4). Based on the funds’ sentiment timing, we form 10 equal-weighted portfolios
and track their returns over the next month. The portfolios are rebalanced each month. Using the monthly time series
of the returns of each portfolio, we estimate alpha by regressing the portfolio excess returns on the Fung–Hsieh (2004)
seven factors, the momentum factor, and the Pastor–Stambaugh (2003) liquidity factor. Both monthly excess return
and alpha are reported in percent. t-statistics are based on Newey–West (1987) standard errors with two lags.
17
0.6
Risk-adjusted return (%/month)
0.5
0.4
0.3
0.2
0.1
0
3 6 9 12
-0.1
-0.2
Holding period (in months)
Portfolio 10 (Top sentiment beta portfolio) Portfolio 1(Bottom sentiment beta portfolio)
Figure IA.1. Risk-adjusted return (i.e., alpha) of top and bottom sentiment beta hedge funds over different
holding horizons. This figure plots alphas for the portfolios consisting of top versus bottom sentiment beta hedge
funds for holding periods of three, six, nine, and 12 months, respectively. Each month starting in December 1996, we
form the portfolios based on hedge funds’ sentiment beta estimated over the past 36 months (up to the given month)
and then hold the portfolios over different periods.
18