High Idiosyncratic Volatility and Low Returns A Prospect Theory Explanation

High Idiosyncratic Volatility and Low Returns: A Prospect Theory Explanation
Author(s): Ajay Bhootra and Jungshik Hur

Source: Financial Management, Vol. 44, No. 2 (SUMMER 2015), pp. 295-322
Published by: Wiley on behalf of the Financial Management Association International
Stable URL: https://www.jstor.org/stable/24736530
Accessed: 26-04-2020 04:28 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Wiley, Financial Management Association International are collaborating with JSTOR to

digitize, preserve and extend access to Financial Management
This content downloaded from 152.118.24.31 on Sun, 26 Apr 2020 04:28:41 UTC
All use subject to https://about.jstor.org/terms
High Idiosyncratic Volatility and Low
Returns: A Prospect Theory Explanation
Ajay Bhootra and Jungshik Hur*
The well-documented negative relationship between idiosyncratic volatility and stock returns
is puzzling if investors are risk-averse. However, under prospect theory, while investors are risk
averse in the domain of gains, they exhibit risk-seeking behavior in the domain of losses. Consistent
with risk-seeking investors' preference for high-volatility stocks in the loss domain, we find that
the negative relationship between idiosyncratic volatility and stock returns is concentrated in
stocks with unrealized capital losses, but is nonexistent in stocks with unrealized capital gains.
This finding is robust to control for short-term return reversals and maximum daily return, among
other variables.
According to the standard asset pricing theory, only the systematic risk of securities should
be priced and there should be no compensation for diversifiable idiosyncratic risk. However,
according to Merton's (1987) investor recognition hypothesis, if investors invest only in securities
with familiar risk-return characteristics and, consequently, hold underdiversified portfolios, id
iosyncratic risk should be priced in equilibrium. In direct contrast to the implications of Merton's
(1987) hypothesis, Ang et al. (2006) document a puzzling negative cross-sectional relationship
between stocks' idiosyncratic volatility and their returns in the following month.
Merton's (1987) implication of a positive volatility-return relationship with suboptimal di
versification assumes risk-averse investors with a concave utility of wealth function within the
standard expected utility framework. However, Kahneman and Tversky's (1979) prospect theory
(PT) model of decision making under uncertainty postulates an S-shaped utility function that is
concave in the domain of gains, but convex in the domain of losses. The S-shaped utility function
is consistent with risk-aversion over positive prospects, but risk-seeking behavior over negative
prospects.
We posit that investors' divergent attitude toward risk over positive and negative prospects is
the key to understanding Ang et al.'s (2006) idiosyncratic volatility anomaly. Specifically, the
risk-seeking behavior of investors in the domain of losses suggests a preference for stocks with
high idiosyncratic volatility. In conjunction with mental accounting (MA) (Thaler, 1980), such
a tendency would result in lower returns to high idiosyncratic volatility stocks with unrealized
capital losses if the relevant mental accounts are the paper gains and losses associated with
individual stocks.
We thank Marc Lipson (Editor) and an anonymous referee for many suggestions that led to significant improvements.
We are grateful to Bing Han, David Hirshleifer, Greg Kadlec, Vijay Singal, and seminar participants at the 2011 FMA
annual meetings and the 2013 California Corporate Finance Conference at Loyola Marymount University for helpful
suggestions. We thank Ken French and Jeff Wurgler for providing the Fama-French factors and the sentiment index,
respectively, on their websites.
'Ajay Bhootra is an Assistant Professor of Finance in the Mihaylo College of Business and Economics at California State
University in Fullerton, CA. Jungshik Hur is an Assistant Professor of Finance in the Department of Economics and
Finance at Louisiana Tech University in Ruston, LA.
Financial Management • Summer 2015 • pages 295 - 322
296 Financial Management • Summer 2015
Note that in the framework of Grinblatt and Han (2005), the demand distortions induced by the
presence of PT/MA investors result in overvaluation (undervaluation) of stocks with unrealized
capital losses (gains).1 We argue that the investors' affinity for high idiosyncratic volatility stocks
within the loss domain would lead to greater overpricing among these stocks. Therefore, the
PT/MA framework provides a rationale for the existence of a negative volatility-return relationship
among stocks with unrealized capital losses (and not among stocks with unrealized capital gains).
Based on the foregoing discussion, we hypothesize that the negative relationship between
idiosyncratic volatility and subsequent stock returns is concentrated in stocks with unrealized
capital losses. In order to empirically test this hypothesis, we construct a capital gains over
hang measure similar to Grinblatt and Han (2005), employing a proxy for the market's aggre
gate cost basis in a stock as the relevant reference point to determine unrealized gains and
losses.
Comparing the value-weighted and equally weighted returns of five idiosyncratic volatil
ity portfolios separately within the subgroups of stocks segregated based on unrealized cap
ital losses (CL) and unrealized capital gains (CG), we find that for the subgroup of stocks
with the largest unrealized losses, the difference in monthly value-weighted (equally weighted)
raw returns of high and low idiosyncratic volatility portfolios is —1.57% (—1.55%) with a
^-statistic of —5.52 (—7.19). The corresponding alphas from the capital asset pricing model
(CAPM) (Sharpe, 1964; Lintner, 1965; Mossin, 1966) and Fama and French (1993) models
are even larger in magnitude and are also statistically significant. In particular, the returns
to the largest idiosyncratic volatility portfolio within the CL subgroup are consistently neg
ative. On the other hand, for the subgroup of stocks with the largest unrealized gains, the
monthly value-weighted (equally weighted) raw return spread between high and low idiosyn
cratic volatility portfolios is 0.35% (0.42%) with a /^-statistic of 1.42 (1.86). The correspond
ing alphas are also positive, but statistically insignificant. We obtain similar results when us
ing the same idiosyncratic volatility cutoffs for the CL and CG subgroups, suggesting that
this result is not driven by the presence of extreme idiosyncratic volatility stocks in the CL
subgroup.2
We perform several tests to ensure the robustness of the above results. Recent evidence in
Huang et al. (2010) suggests that the idiosyncratic volatility puzzle is attributable to the short-term
reversals in returns documented in Jegadeesh (1990), Lehmann (1990), and Lo and MacKinlay
(1990).3 These authors find that in the cross-sectional regressions of stock returns on idiosyncratic
volatility that control for the previous month's return, the coefficient on idiosyncratic volatility
is no longer statistically significant.
In contrast to the evidence in Huang et al. (2010), we find that when stocks priced below
$5 are excluded from the sample, we still obtain a negative relationship between idiosyncratic
volatility and stock returns in CL stocks for both value-weighted and equally weighted portfolio
returns, as well as in Fama-MacBeth (1973) firm-level cross-sectional regressions that control
' The demand distortions occur because the S-shaped value function from the prospect theory, together with mental
accounting, leads to the disposition effect: the tendency of investors to sell their winning stocks too quickly and hold on
to their losing stocks too long (Shefrin and Statman, 1985).
2 It is noteworthy that in Ang et al. (2006, table VIII, Panel B, p. 291), the Fama-French (1993) alpha associated with
high minus low (HML) idiosyncratic volatility portfolio is -2.25% for the loser stocks, but only -0.48% for the winner
stocks, where winners and losers are identified based on the past 12-month returns. In accordance with Grinblatt and Han
(2005), our focus is on unrealized gains and losses rather than on past returns.
1 Fu (2009) also documents a similar role of return reversals in the negative volatility-return relationship.
Bhootra & Hur ♦ High Idiosyncratic Volatility and Low Returns 297
for the past month's return.4 Given the high transaction costs, as well as the severe short-selling
constraints associated with these penny stocks, we believe their exclusion from the sample is
justified.
Another recent study by Bali, Cakici, and Whitelaw (2011) finds a significant negative re
lationship between stocks' maximum daily returns (MAX) in a month and their returns in the
following month. These authors find that after controlling for MAX in the cross-sectional regres
sions of the returns on idiosyncratic volatility, the coefficient on volatility is insignificant in some
specifications or even significantly positive in others. We find that the negative volatility-return
relationship in CL stocks persists after controlling for MAX in our empirical tests. Moreover,
the reported positive IVOL-return relationship is concentrated in stocks with unrealized capital
gains.
In addition to the prior month return and MAX variables, we also include other common
explanatory variables including size, stock price, book-to-market ratio, illiquidity, past 12-month
return, idiosyncratic skewness, and beta in Fama-MacBeth (1973) cross-sectional regressions,
and continue to find a negative volatility-return relationship in CL stocks.
Our study makes several contributions to the literature. This paper is the first to document
and provide robust supporting evidence that the negative volatility-return relationship is only
observed in stocks with unrealized capital losses. This result lends credence to the significance of
PT-based risk preferences in understanding the asset pricing anomalies. In addition, we link this
key result of our study to several previously documented findings in the literature and provide
some novel empirical results. We show that in contrast to prior evidence, the short-term reversals
and maximum daily returns do not explain the volatility-return relationship in the sample of
capital loss stocks. Furthermore, we report that the previously documented role of penny stocks
(George and Hwang, 2013) and January seasonality (Doran, Jiang, and Peterson, 2012) in the
volatility-return relationship is relevant only for capital loss stocks, and not for capital gains
stocks. In sum, our results suggest that whether a stock has unrealized embedded gains or losses
is a dominant factor in understanding the idiosyncratic volatility anomaly.
The rest of the paper is organized as follows. Section I develops the hypothesis and relates our
work to the relevant literature. Section II describes the data and methodology. Section III presents
our results, and Section IV provides our conclusions.
I. Hypothesis Development and Related Literature
The negative relationship between idiosyncratic volatility and subsequent stock returns is
considered anomalous under the implicit assumption that investors are risk-averse. According to
Ang et al. (2009),"... we do not yet have a theoretical framework to understand why agents have
high demand for high idiosyncratic volatility stocks, causing these stocks to have low expected
returns."
Given that under the expected utility theory, investors are uniformly risk-averse with concave
utility of wealth, it has proven challenging to explain the negative volatility-return relationship
under this framework. We explore the alternative possibility that investors' risk preferences under
Kahneman and Tversky's (1979) PT provide a resolution of this puzzle. The key distinctive
element of PT is the S-shaped utility function that is concave in the domain of gains, but convex
4 Note that while Fink, Fink, and He (2012) find no evidence of a relationship between expected idiosyncratic volatility
and expected returns, they rely, in part, on evidence in Huang et al. (2010) to explain Ang et al.'s (2006) findings of a
negative relationship between lagged idiosyncratic volatility and returns.
Figure 1. Prospect Theory Risk Prefere
The figure depicts S-shaped prospect theory utility,

expected to have an embedded gain of A1 (Bl) or A2
A+, A+ is equal to the expected gain from B+,
Analogously, Stock A— (B—) is expected to have a
equal probability, and A— (B—) is the expected lo
utilities from Stocks A+, B+, A—, and B—, respectiv
in the domain of losses. This S-shaped utility fu

of gains, but risk-seeking behavior in the doma
Figure 1 depicts a simple setting to illustrate
high-volatility stocks in the domain of losses. T
the following function of gains or losses, x:
U(x) =x05, x>0,
U(x) = - 2(-xf5, x< 0.
Bhootra & Hur« High Idiosyncratic Volatility and Low Returns 299
Note that this utility function captures the key features of PT. It is concave over gains and
convex over losses, and is steeper in the loss domain (due to multiplication by —2), consistent
with investors' loss-aversion.
Consider two stocks, A+ and B+, in which an investor has accumulated capital gains with A+
being more volatile than B+. Without loss of generality, assume that the shares of A+ and B+
were purchased at an identical price, do not pay dividends, and have identical embedded gains.
We make these assumptions so that we can focus on the difference in volatility of stocks while
controlling for other relevant characteristics. In the next period, the investor expects the price of
Stock A+ to go up or down such that A+ will have an embedded gain of A1 or A2 with equal
probability. Similarly, the investor expects the price of B+ to go up or down such that B+ will
have an embedded gain of B1 or B2 with equal probability. The higher volatility of A-f- relative
to B+ is implied by the fact that Al< B1 and A2 > B2, as shown in the figure. To focus on
the volatility differences, assume that the expected gain from A+, A+, equals the expected gain
from B+, B+, where A+ = 0.5(A1 + A2) and B+ = 0.5(B1 + B2).
Given the choice between A+ and B+ in the domain of gains, which stock does the investor
prefer? Since A+ is expected to have a gain of A1 or A2 with equal likelihood, the investor's
expected utility from A+, Ua+, is given by 0.5[t/(A 1) + t/(A2)]. Similarly, the investor's expected
utility from B+, Ub+, is 0.5[L'(B1) + £/(B2)]. As illustrated in Figure 1, the concavity of the
utility function in the domain of gains implies that 0b+ > Ua+. Consistent with risk-aversion,
the investor derives greater utility from holding the less volatile stock in the domain of gains.
Similar reasoning implies a preference of high-volatility stock in the domain of losses due to
the convexity of the utility function. In this case, Stocks A— and B— have identical embedded
capital losses, with A— being more volatile than B—. In the next period, the investor holding A—
(B—) expects the stock to go up or down such that the embedded losses in A— (B—) would be
—A1 (—Bl) or — A2 (—B2) with equal probability. Also, — A1 > — Bl, —A2 < —B2, and the
expected level of embedded losses is equal; that is, A— = B—. As is evident from the figure,
due to the convexity of the utility function in the loss domain, the expected utility of A—, Ua-,
is greater than the expected utility of B—, Ub-, where Ua- = 0.5[t/(—Al) + U(—A2)] and UB_
= 0.5[L/(—Bl) + £/(—B2)]. In this scenario, the investor would prefer the more volatile Stock
A— over Stock B—, consistent with risk-seeking behavior. The higher demand for high-volatility
stocks would lead to higher equilibrium prices and lower expected returns in the domain of losses.
It is also clear that ceteris paribus, the more volatile the stock, the greater the investor's utility and,
therefore, the investor's preference for stocks will increase monotonically with their volatility in
the loss domain.
Note that for ease of illustration and interpretation, the above discussion focuses on total
volatility rather than the idiosyncratic component. However, similar reasoning would apply to
idiosyncratic volatility as a risk measure. Moreover, Ang et al. (2006) find a similar negative
relationship for total volatility as for idiosyncratic volatility (see table VI, p. 285).
It is clear that an empirical investigation of the role of PT preferences in explaining the negative
volatility-return relationship hinges on defining the domain of gains and losses with respect to
a reference point. If investors segregate gains and losses on their stocks into separate mental
accounts, then the relevant reference point is the cost basis of a stock (Grinblatt and Han, 2005).
Furthermore, since PT preferences are relevant when investors face the prospect of a gain or loss,
a measure of unrealized gains and losses is appropriate for empirical analysis.
In the model of Grinblatt and Han (2005), the tendency of disposition prone investors to hold
on to their loser stocks too long results in overvaluation of these stocks with unrealized capital
losses. We expect that due to their risk-seeking behavior, the tendency to hold on to high-volatility
stocks would be particularly strong, resulting in greater overpricing and subsequent lower returns
of these stocks. Alternatively, the tendency of these investors to sell their winners too quickly
leads to the undervaluation of stocks with unrealized capital gains. However, within the domain
of gains, the relationship between volatility and subsequent returns is not clear. On one hand, it is
plausible that risk-averse investors may have a greater inclination to liquidate the high-volatility
stocks, resulting in greater underpricing and higher returns in future. Alternatively, it is also
conceivable that risk-averse investors eliminate the idiosyncratic risk of their portfolios through
diversification, as is assumed in standard asset pricing models. In the latter case, we would expect
no relationship between idiosyncratic risk and returns for stocks with unrealized gains.
The foregoing discussion forms the basis of our primary hypothesis:
concentrated in stocks with unrealized capital losses.
Note that in the model of Barberis, Huang, and Santos (2001), investors become risk-averse
after the realization of a loss. We are interested in PT risk preferences that are relevant when
investors are sitting on paper gains/losses, but have not realized them. This distinction is lucidly
illustrated in Barberis, Huang, and Santos (1999) using the casino/horse-race betting example. In
that example, a gambler at a casino who is sitting on losses (unrealized losses) exhibits risk-seeking
behavior in an effort to break-even, consistent with the PT preferences. However, subsequent to
the realization of the loss, the gambler becomes risk-averse due to fear of accumulating further
losses.
In another recent study, Barberis and Xiong (2012) develop a model of realization utility that
provides a potential explanation of the negative volatility-return relationship. In their model,
investors derive positive (negative) utility from realizing gains (losses) on their investments. This
realization utility leads to a preference for buying high-volatility stocks due to their greater upside
potential. The greater downside risk of these stocks is not particularly worrisome for investors
since the realization of a loss can be postponed to avoid the derivation of negative utility. Thus,
the risk-seeking behavior of investors with realization utility leads to a greater demand for and
subsequently lower returns of high-volatility stocks.
Since our hypothesis is based on PT preferences, it differs from the framework of Barberis and
Xiong (2012) in two important aspects. First, unlike the case of PT preferences, in the framework
of Barberis and Xiong (2012), the preference for high-volatility stocks does not hinge on whether
the stocks have unrealized gains or losses. In addition, as Barberis and Xiong (2012) point out,
the idea of realization utility is more relevant for individual rather than institutional investors,
suggesting that the negative volatility-return relationship should be observed in stocks with
relatively higher individual investor ownership. In contrast, our hypothesis postulates a negative
volatility-return relationship among stocks with unrealized losses, without conditioning on the
magnitude of individual investor ownership.
II. Data and Methodology
Our base sample is comprised of common stocks listed on NYSE, AMEX, and NASDAQ
(share codes 10 and 11) from July 1963 to December 2007. We obtain daily and monthly stock
returns from the Centre for Research in Security Prices (CRSP) files. Except where necessary
to illustrate the impact of low-priced stocks on inferences, we exclude stocks priced below $5
from the sample in the portfolio formation month to ensure that the results are not driven by
these illiquid securities. We also exclude firms with missing data required for the computation of
Bhootra & Hur • High Idiosyncratic Volatility and Low Returns 301
capital gains overhang and the estimation of monthly idiosyncratic volatility. The final sample is
comprised of a total of 19,736 unique firms, with an average of 3,567 firms per month.
Following Ang et al. (2006), the idiosyncratic volatility of stock i is obtained from the following
time-series regression of excess daily stock returns during the month on the contemporaneous
Fama-French (1993) market, size, and book-to-market factors:5
Rj d — Rfd = ai + PiMKTRFd + SiSMBd + hjHMLd + (1)
where j is the return of stock i on day d, R/d is the daily risk-free rate, and MKTRFd, SMBd, and
HMLd are the daily Fama-French (1993) factors. The daily idiosyncratic volatility is the standard
deviation of the residuals from this regression and the monthly idiosyncratic volatility (IVOL) is
obtained by multiplying the daily volatility by the square root of the number of trading days in
the month.
We construct the capital gains overhang variable using CRSP daily returns following the
methodology employed in Grinblatt and Han (2005) and Hur, Pritamani, and Sharma (2010). The
capital gains overhang of a stock represents the percentage deviation of its price from its current
reference price, where the reference price is assumed to be the market's aggregate cost basis for
the stock. Specifically, for each stock i at the end of each month t, the capital gains overhang
(CGO«v) is obtained as:
CGO,-, = (Pu - RPu)/Pi,t, (2)
where Pi4 is the price of stock i at the end of month t and R P: J is the reference price for each
stock i at the end of month t. The reference price, RP:J, is estimated as follows:
(3)
where Vi t is the turnover in stock i on day t, T is the number of trading days in the previous three
years with available daily price and volume information, and Pu-„ is the price of security i on day
t — n. The term in the parentheses multiplying Pu-„ is a weight that reflects the probability that
the shares purchased on date t — n have not been traded since, the probability being a function of
the turnover from t — ntot— 1. Intuitively, a higher turnover in any given month (relative to other
months) over the past three years would imply a greater likelihood that the stock was purchased
by the current shareholders in that particular month. The denominator k is a constant that makes
the weights sum up to one. In computing RPifl, the share price and share turnover variables are
adjusted for stock splits and stock dividends.6
For empirical tests, we segregate the stocks into groups with unrealized capital losses and
unrealized capital gains every month (negative and nonnegative values of CGO in Equation (2),
respectively). We label the stocks with unrealized losses (gains) as CL (CG) stocks and, for
5 The Fama-French factors are obtained from Ken French's data library available at: http://mba.tuck.dartmouth.edu/
pages/faculty/ken.french/data_library.html.
6 In the Grinblatt and Han (2005) model, the reference price for a stock is obtained as the turnover weighted average of the
past stock price and the past reference price. That is, RPU = VU-\PU.\ +(1 — The recursive substitution
of past reference prices on the right-hand side of the above equation leads to Equation (3) with the exception that the
summation is over infinite periods. For empirical implementation, we sum over the number of trading days in the three
most recent years. Grinblatt and Han (2005), however, find robust results using three, five, or seven years of data.
portfolio level tests, we further group them into two loss portfolios (CL1 and CL2) and two
gain portfolios (CGI and CG2) based on the median month-end values. We examine the returns
to resulting portfolios by employing univariate and multivariate sorts on key variables, such as
idiosyncratic volatility and capital gains overhang. To control for multiple explanatory variables,
we employ the monthly firm-level Fama-MacBeth (1973) cross-sectional regressions. To account
for any potential impact of autocorrelations on the statistical inferences, we use the Newey and
West (1987) adjusted standard errors when computing the ^-statistics in all our tests.
III. Results
A. Idiosyncratic Volatility and Stock Returns
We begin our analysis by verifying the presence of a negative relationship between idiosyn
cratic volatility and the subsequent returns in our sample. Each month, we sort the stocks into
five portfolios based on their monthly idiosyncratic volatility and examine the returns of these
portfolios in the following month. Table 1 reports the value-weighted and equally weighted returns
of these volatility sorted portfolios. To highlight the impact of low-priced stocks on the returns,
we report the returns for samples that include sub-$5 stocks (in Panel A) and that exclude sub-$5
stocks (in Panel B).
Panel A of Table I demonstrates that the difference in the value-weighted returns of high and low
volatility portfolios is negative, —0.91%, and statistically significant (/-statistic = —3.15). The
corresponding CAPM and Fama-French (1993) alphas are even larger in magnitude at — 1.27%
and — 1.23%, respectively, and are also statistically significant. These results are consistent with
Ang et al. (2006), who find the raw return differential to be —1.06% per month from July 1963
to December 2000. However, in the sample that includes the sub-$5 stocks, the difference in
the equally weighted returns of the extreme volatility portfolios is a statistically insignificant
—0.05%. This finding is consistent with Bali and Cakici (2008) and Huang et al. (2010), who
find an insignificant difference of 0.02% and —0.005% per month, respectively, for the equally
weighted portfolios from July 1963 to December 2004.7
However, when we exclude the sub-$5 penny stocks from the sample in Panel B, we find that
the difference in the equally weighted returns of the high and low-volatility portfolios is —0.63%,
which is statistically significant (^-statistic = —2.66). The corresponding CAPM and Fama-French
(1993) alphas are —0.98% (^-statistic = —4.75) and —0.81% (f-statistic = —6.03), respectively.
The difference in the value-weighted returns and alphas continues to be significant after the
exclusion of the penny stocks, although the magnitudes are relatively smaller when compared
with those in Panel A. These findings suggest that the absence of the negative volatility-return
relationship in the equally weighted portfolios in Bali and Cakici (2008) and Huang et al. (2010)
is driven by the presence of the sub-$5 stocks in their sample. The significant impact of the
penny stocks on the equally weighted portfolio returns coupled with the fact that these stocks
are typically associated with high illiquidity, high transaction costs, and severe short-selling
restrictions, justifies their exclusion from the sample.8
7 Note that the inferences based on the value-weighted portfolio returns are likely to be more reliable due to the potential
biases from the noisy prices in the equally weighted portfolio returns (Asparouhova, Bessembinder, and Kalcheva, 2011).
8 For comparison purposes, we continue to illustrate the results for the sample that includes penny stocks in some of the
subsequent tables.
Table I. Idiosyncratic Volatility and Stock Returns
During each month, we run time-series regressions of excess daily stock returns on the c
daily Fama-French (1993) factors. Daily idiosyncratic volatility is the standard deviation o
the regressions. Monthly idiosyncratic volatility (IVOL) is the daily idiosyncratic volatility
root of the number of trading days in the month. We sort the stocks into quintile portfo
IVOL each month and compute the value-weighted (VW) and equally weighted (EW) re
portfolios in the following month. VI (V5) is the portfolio of stocks in the bottom (top) q
Panel A (Panel B) reports the returns of IVOL portfolios for the sample that includes
priced below $5 at the end of portfolio formation month. The table also presents the differ
extreme IVOL portfolios, V5 — VI, along with the corresponding CAPM and Fama-French
returns are in percent per month. The ?-statistics, reported in parentheses, are adjusted fo
using the Newey-West (1987) method. The sample covers the period from July 1963 to De
VW Return EW Return
Panel A. Average Returns of IVOL Portfolios (No Price Filter)
VI (Low) 0.93 1.16

V2 1.02 1.39
V3 1.06 1.40
V4 0.74 1.26
V5(High) 0.02 1.11

V5 - VI -0.91 -0.05
(-3.15) (-0.15)
CAPM alpha -1.27 -0.38
(-4.62) (-1.40)
FF alpha -1.23 -0.40
(-6.39) (-1.89)
Panel B. Average Returns of IVOL Portfolios (Price > $5)
VI (Low) 0.93 1.15
V2 0.98 1.36
V3 1.06 1.43
V4 0.94 1.26
V5(High) 0.36 0.51

V5 - VI -0.57 -0.63
(-2.12) (-2.66)
CAPM alpha -0.92 -0.98
(-3.91) (-4.75)
FF alpha -0.75 -0.81
(-4.58) (-6.03)
B. Preliminary Tests of the Hypothesis
We next begin the preliminary investigation of our main hypothesis that the negative relationship
between idiosyncratic volatility and stock returns is concentrated among stocks with unrealized
capital losses. Within each of the four CGO portfolios, CL1, CL2, CGI, and CG2, we sort the
stocks into quintile portfolios based on idiosyncratic volatility. Panel A of Table II reports the
returns in the following month of the portfolios thus obtained, along with the difference in the
returns of the extreme volatility portfolios and the corresponding CAPM and Fama-French (1993)
alphas.
Table II. Idiosyncratic Volatility and Stock Returns Conditional on Capital Gains
Overhang
At the end of each month, we divide the sample stocks into subgroups based on their capital gains overhang.
CL1 and CL2 are portfolios of capital loss (CL) stocks with below and above median capital losses,
respectively, at the end of each month. Similarly, CGI and CG2 are portfolios of capital gain (CG) stocks
with below and above median capital gains at the end of each month. For the results in Panel A, the
stocks within the CL and CG subgroups are further sorted into quintile portfolios based on their monthly
idiosyncratic volatility (IVOL). For the results in Panel B, the sample stocks are independently sorted
into quintile portfolios based on their IVOL. The table provides the value-weighted (VW) and equally
weighted (EW) returns of the quintile IVOL portfolios. VI (V5) is the portfolio of stocks in the bottom (top)
quintile of IVOL. The table also reports the difference in returns of extreme IVOL portfolios, V5 — VI, the
corresponding CAPM and Fama-French (FF) alphas, and the difference of the IVOL spread between the
CL1 and CG2 subgroups. The returns are in percent per month. The ^-statistics, reported in parentheses, are
adjusted for autocorrelation using the Newey-West (1987) method. Stocks with prices less than $5 at the
end of the portfolio formation month are excluded from the sample. The sample covers the period from July
1963 to December 2007.
VW Return EW Return
CL CG CL CG
CL1 CL2 CG1 CG2 CL1-CG2 CL1 CL2 CG1 CG2 CL1-CG2
Panel A. Sequential Sorting
VI (Low) 1.08 0.82 0.93 0.97 1.24 0.99 1.05 1.18

V2 0.98 0.82 0.86 1.15 1.37 1.15 1.29 1.58
V3 0.91 0.77 1.01 1.49 1.23 1.17 1.40 1.80
V4 0.52 0.75 0.86 1.64 0.78 0.94 1.36 1.91
V5(High) -0.49 -0.53 0.57 1.33 -0.31 -0.17 0.85 1.59
V5-V1 -1.57 -1.35 -0.36 0.35 -1.93 -1.55 -1.15 -0.19 0.42 -1.97
(-5.52) (-5.15) (-1.32) (1.42) (-8.24) (-7.19) (-5.50) (-0.81) (1.86) (-10.82)
CAPM alpha -1.90 -1.67 -0.66 0.15 -2.04 -1.81 -1.45 -0.47 0.15 -1.97
(-7.20) (-6.23) (-2.41) (0.67) (-8.80) (-8.38) (-6.91) (-1.89) (0.72) (-9.58)
FF alpha -1.85 -1.47 -0.46 0.18 -2.03 -1.68 -1.31 -0.31 0.24 -1.92
(-8.54) (-8.18) (—2.15) (0.93) (-8.35) (-10.84) (-9.65) (-1.56) (1.21) (-8.46)
Panel B. Independent Sorting

VI (Low) 1.03 0.85 0.90 0.99 1.08 1.03 1.10 1.24
V2 1.18 0.82 0.89 1.14 1.42 1.19 1.30 1.47
V3 1.03 0.82 0.90 1.50 1.35 1.18 1.36 1.78
V4 0.78 0.57 0.80 1.55 1.04 0.84 1.29 1.93
V5(High) -0.49 -0.33 0.63 1.37 -0.25 -0.09 0.68 1.58
> in 1
> -1.52 -1.18 -0.27 0.38 -1.90 -1.33 -1.12 -0.42 0.34 -1.67
(-5.89) (-5.35) (-0.98) (1.45) (-7.95) (-6.93) (-6.19) (-1.76) (1.51) (-7.75)
CAPM alpha -1.83 -1.47 -0.60 0.14 -1.97 -1.62 -1.44 -0.71 0.07 -1.69
(-6.78) (-5.82) (-2.15) (0.59) (-8.12) (-8.25) (-6.48) (-2.61) (0.35) (-7.43)
FF alpha -1.75 -1.35 -0.39 0.18 -1.93 -1.49 -1.29 -0.55 0.15 -1.64
(-6.51) (-6.43) (-1.71) (0.85) (-8.01) (-9.24) (-8.43) (-2.50) (0.76) (-7.05)
Bhootra & Hur • High Idiosyncratic Volatility arid Low Returns 305
Within the CL1 and CL2 portfolios, we find that the differences in the value-weighted and
equally weighted raw returns of the high and low idiosyncratic volatility portfolios are negative
and statistically significant at conventional levels, and the differences are larger in magnitude
for the CL1 portfolio. Specifically, the value-weighted (equally weighted) raw return difference
is —1.57% (—1.55%) with a ^-statistic of —5.52 (—7.19) for the CL1 portfolio, and —1.35%
(—1.15%) with a /-statistic of —5.15 (—5.50) for the CL2 portfolio. The corresponding alphas
from the CAPM and the three-factor models are also negative and statistically significant for the
CL1 and CL2 portfolios. On the other hand, the differences in both the raw value-weighted and the
equally weighted returns are negative, but insignificant for the CGI portfolio, and positive (and
significant at the 10% level in the case of the equally weighted returns) for the CG2 portfolio. The
alphas based on both the market and the three-factor models generally present a similar picture,
except that in some instances, the alphas are negative and significant for the CGI portfolio. In
all of the cases, the difference in the return spreads of the extreme IVOL portfolios between the
CL 1 and the CG2 stocks is negative and statistically significant.
The pattern of returns across the five IVOL portfolios within the CL and CG groups is also
noteworthy. For the CL1 and CL2 value-weighted portfolios, the returns decline monotonically
from the low to high-volatility portfolios. In particular, for the CL group, the high IVOL portfolio,
V5, consistently earns negative raw returns and is primarily responsible for the observed negative
IVOL-return relationship. This evidence is consistent with the high demand and low subsequent
returns for the high IVOL stocks within the loss domain. On the other hand, we do not observe
a similar systematic return variation across IVOL for the CG portfolios. For CGI in particular,
although the extreme portfolios' return difference (V5 — VI) is negative, the value-weighted
returns first tend to increase and then decline as we move from the VI to the V5 portfolio. The
returns to the V5 portfolio are also relatively much larger for the CG group as compared with the
CL group.
In our sample, the average idiosyncratic volatility of stocks in the CL portfolio (9.17%) is
relatively higher when compared with the stocks in the CG portfolio (7.98%). Since the results
in Panel A of Table II employ sequential sorting, a potential explanation of the findings in Panel
A is that the cutoffs for the high and low-volatility portfolios within the CL portfolios are more
extreme as compared to the CG portfolios and, therefore, the return predictability is stronger. To
ensure that these more extreme cutoffs are not driving the results, we repeat the analysis with
independent sorting instead of sequential sorting, so that the same IVOL cutoffs are employed
for the CL and CG portfolios.
The returns using independent sorting are presented in Panel B of Table II. The results are
qualitatively similar to those in Panel A. Both the value-weighted and the equally weighted raw
return spreads between the high and low-volatility stocks are large and statistically significant
for the CL1 and CL2 portfolios, but generally insignificant for the CG portfolios. In particular,
the V5 — VI value-weighted (equally weighted) raw return is —1.52% (—1.33%) for the CL1
group and —1.18% (— 1.12%) for the CL2 group. The alphas also exhibit a pattern similar to the
case of sequential sorting. Thus, the concentration of the negative volatility-return relationship in
the CL portfolio does not seem to be attributable to the presence of stocks with relatively higher
volatility in this portfolio.
Overall, the results in Table II offer strong support to our hypothesis that the negative IVOL
return relationship is predominantly concentrated in those stocks with unrealized losses. In
unreported results, we also examine the market capitalization of the quintile volatility portfolios
across the CL and CG groups. We find a monotonic decline in the market cap as the volatility
increases for both the CL and CG groups. Given that the negative volatility-return relationship
is only observed among CL stocks, it does not appear that the results can be attributable to the
small size or illiquidity of the high-volatility portfolios. Note that through the exclusion of sub-$5
stocks, we have already eliminated the smallest, most illiquid stocks. Furthermore, we explicitly
control for size and illiquidity in our cross-sectional regression tests later in the paper.
C. Are the Results Robust to Control for MAX and REV Variables?
Having documented the preliminary evidence supporting our hypothesis, we next turn o
attention to the examination of the robustness of our findings. To begin with, we focus on
variables that have been found to explain away the negative volatility-return relationship in
recent literature: 1) the prior month's maximum daily return (MAX) and 2) the prior mon
stock return (REV). As previously mentioned, Bali et al. (2011) find that the maximum dai
return of a stock in a given month is negatively related to the subsequent month's stock ret
Of particular interest, these authors find that after controlling for MAX, the relationship betw
IVOL and subsequent returns is insignificant in some regression specifications and even beco
positive in other specifications. Huang et al. (2010) determine that the negative volatility-re
relationship is due to an omitted variable bias. The omitted variable is the previous month's s
return (REV). Owing to the short-term return reversals, the inclusion of the previous mont
return in the regression specifications renders the relationship between volatility and subseq
returns insignificant.
Using firm-level Fama-MacBeth (1973) cross-sectional regressions, we first verify that t
results of Bali et al. (2011) and Huang et al. (2010) are obtained in our sample. The res
from the first set of regressions, with the stock return as the dependent variable and vari
combinations of IVOL, MAX, and REV as the explanatory variables, are reported in Panel A
of Table III.9 Panel A. 1 reports the results for the sample that includes the sub-$5 stocks
be consistent with Bali et al., 2011 and Huang et al., 2010), while the results in Panel A.2 a
obtained after excluding these penny stocks.
For the sample that includes penny stocks, the mean coefficient on IVOL in the univari
regression is negative (—0.012), but is statistically insignificant with a f-statistic of —0.92.
coefficient on the MAX variable in the univariate regression is negative (—0.037) and statisti
significant with a /^-statistic of —2.50. Moreover, controlling for MAX in Specification (
Panel A.l reverses the sign of the slope coefficient on IVOL, which is now positive (0.074) w
a /-statistic of 4.74. These findings are consistent with those reported in Bali et al. (2011)
table 10, p. 441). Also, consistent with Huang et al. (2010), we find that the average coeffici
on the REV variable in the univariate regression is negative (—0.051) and statistically signif
(^-statistic = —10.03). After controlling for REV, the coefficient on IVOL turns positive (0.0
but is still insignificant. When we control for both MAX and REV in Specification (6), th
coefficients on IVOL and MAX become statistically insignificant, but the coefficient on R
continues to be negative and statistically significant.
Turning to the results for the sample without penny stocks in Panel A.2, the coefficient on IV
in the univariate regression is now negative (—0.057) and statistically significant (/-statis
—4.20). This result mirrors the results obtained in Table I for the equally weighted portf
without the inclusion of the penny stocks. The MAX variable continues to be negative and
statistically significant in the univariate regression. However, the coefficient on IVOL, af
controlling for MAX, is no longer positive as in Panel A.l, but is negative (—0.022), a
statistically insignificant (/-statistic = —1.25). On the other hand, the REV variable does n
9 In addition to the Fama-MacBeth (1973) regressions, we also examine the returns of portfolios obtained using biv
sorts on IVOL and MAX, as well as on 1VOL and REV Our conclusions are similar to those with the regression
such, we report only the regression results for brevity. The results from the sorts are available upon request.
Table III. Role of MAX and REV Variables in IVOL-Return Relationship
We run monthly firm-level Fama-MacBeth (1973) cross-sectional regressions of month t individual stock
returns on the lagged explanatory variables, computed in month t — 1. The various specifications include
subsets of a stock's monthly idiosyncratic volatility (IVOL), maximum daily return (MAX), and monthly
stock return (REV) as explanatory variables. The table reports the time-series means of coefficient estimates
from the cross-sectional regressions. The results with and without the January returns are provided in Panels
A and B, respectively. For the results in Panels A.l and B.l (A.2 and B.2), the stocks priced below $5 at the
end of month t — 1 are included in (excluded from) the sample. The return and volatility numbers are in
percentages. The f-statistics, reported in parentheses, are adjusted for autocorrelation using the Newey-West
(1987) method. The sample covers the period from July 1963 to December 2007.
Panel A. All Months
Panel A. I. No Price Filter Panel A.2. Price > $5
IVOL MAX REV IVOL MAX REV
(1) -0.01 -0.06
(-0.92) (-4.20)
(2) -0.04 -0.08
(-2.50) (-5.00)
(3) 0.07 -0.13 -0.02 -0.05
(4.12) (-8.40) (-1.25) (-4.18)
(4) -0.05 -0.03
(-10.03) (-6.25)
(5) 0.00 -0.06 -0.05 -0.03
(0.06) (-11.15) (-3.09) (-6.02)
(6) 0.01 -0.02 -0.06 -0.05 0.01 -0.03
(0.65) (-1.26) (-11.26) (-3.20) (0.95) (-6.20)
Panel B. Non-January Months

Panel B. 1. No Price Filter Panel B.2. Price > $5
IVOL MAX REV IVOL MAX REV
(7) -0.04 -0.07
(-3.01) (-4.99)
(8) -0.07 -0.09
(-3.99) (-5.35)
(9) 0.02 -0.90 -0.05 -0.03
(1.00) (-6.38) (-3.04) (-2.51)
(10) -0.04 -0.02
(-8.64) (-5.00)
(11) -0.03 -0.05 -0.07 -0.02
(-1.99) (-9.49) (-4.29) (-4.55)

(12) -0.03 0.01 -0.05 -0.08 0.02 -0.02
(-1.92) (0.39) (-9.68) (-4.44) (1.18) (-4.77)
explain away the negative volatility-return relationship. The coefficient on IVOL, after controlling
for REV, is —0.046 with a ^-statistic of —3.09. In contrast to the results in Panel A. 1, the IVOL
is negative and significant when controlling for both MAX and REV, although, similar to Panel
A. 1, the coefficient on the MAX variable is not statistically significant.
In addition to the exclusion of the penny stocks, the well-documented January seasonality in
stock returns provides motivation for examining the IVOL-return relationship after excluding the
January returns as well. Peterson and Smedema (2011) document a robust negative volatility
return relationship outside of January In particular, these authors find that once the January
months are excluded, the REV variable is no longer able to explain the Ang et al. (2006) result.
Doran et al. (2012) suggest that the investors' preference for lottery-like stocks at the turn of the
year due to their gambling mentality results in superior performance of these stocks in January
and subsequent underperformance during the rest of the year. Note that the evidence of investors'
preference for lottery-like stocks was the primary motivation behind examining the relationship
between MAX and subsequent returns in Bali et al. (2011).
In light of the above evidence, we repeat the analysis in Panel A of Table III after excluding
the January returns from the sample. The results are reported in Panel B. In contrast to the results
in Panel A.l, Panel B.l demonstrates that the coefficient on IVOL is negative (—0.041) and
statistically significant (/-statistic = —3.01) in the univariate regressions, even with the penny
stocks included in the sample. Consistent with Doran et al. (2012), this finding suggests that
the high IVOL stocks earn positive returns in January and exhibit poor performance over the
remainder of the year. Also in contrast with Panel A.l's results, we find that 1) IVOL is no
longer significantly positive after controlling for MAX, and 2) IVOL is negative and statistically
significant at the 5% (10%) level after controlling for REV (MAX and REV). In Panel B.2,
we exclude the penny stocks in addition to excluding the January returns. The most noticeable
conclusion that we can draw from Panel B.2 is that the coefficient on IVOL is negative and
significant in the univariate regression, as well as in the regressions that control for MAX and
REV, individually or both at the same time.
The preceding evidence suggests that after accounting for the presence of low-priced stocks
and January seasonality, the negative relationship between IVOL and subsequent returns is robust
to control for MAX and REV Thus, while the Ang et al. (2006) result cannot yet be dismissed,
we still need to rule out the possibility that the concentration of the negative volatility-return
relationship in the CL stocks is attributable to the MAX and/or REV measures. Therefore, we
reexamine the results in Table II after controlling for MAX and REV
We employ a three-way sorting procedure for this test. As before, the stocks are first segregated
into CL1, CL2, CGI, and CG2 subgroups each month. Then, within each of the CL and CG
subgroups, five portfolios are formed by sorting the stocks on MAX (REV) in Table IV (Table V).
Then, within each of these five portfolios, five additional portfolios are formed by sorting the
stocks on their IVOL. For each of the CL and CG portfolios, we average the returns across the five
MAX (REV) portfolios within each of the IVOL portfolios. This procedure ensures similar levels
of the MAX (REV) variable across each of the IVOL portfolios, thereby providing a convenient
method to control for MAX (REV). Bali et al. (2011) adopt a similar scheme in their tests.
Tables IV and V report both the value-weighted and the equally weighted returns. We exclude the
penny stocks from the sample in these tests and report the results with and without the January
returns in Panels A and B, respectively.
Focusing first on the results in Panel A of Table IV, we find that after controlling for MAX,
the difference in the value-weighted raw returns of the high and low IVOL portfolios is negative
(—0.79%) and statistically significant (^-statistic = —4.27) for the CL1 stocks. On the other
hand, the value-weighted return difference is positive (0.18%), but statistically insignificant
(f-statistic = 1.09) for the CG2 stocks. Similar patterns are observed in the differences of the
CAPM and Fama-French (1993) alphas as well. The difference in the equally weighted raw
returns of the extreme IVOL portfolios for the CL1 (CG2) stocks is —0.24% (0.36%), which
is statistically significant with a /-statistic of —1.93 (2.57). The differences in the CAPM and
-0.59
(-4.50) -0.60 (-4.93) -0.63 (-5.19)
CL1-CG2 (Continued)
1.36 1.53 1.64 1.77 1.71 0.36

CG2 (2.57) 0.23 (2.02) 0.23 (2.22)
CG
1.04 1.19 1.26 1.19 1.26 0.23

CG1 (1.23) 0.11 (0.60) 0.08 (0.59)
EW Return
0.84 0.92 0.84 0.86 0.62

CL2 -0.22
(-1.80) -0.34 (-3.01) -0.35 (-4.28)
CL
0.92 0.91 0.91 0.90 0.67

CL1 -0.24
(-1.93) -0.37 (-3.02) -0.39 (-4.99)
-0.96
(-5.08) -1.00 (-5.63) -1.08 (-5.85)
CL1-CG2
Panel A. Al Months
1.19 1.23 1.32 1.41 1.38 0.18

CG2 (1.09) 0.06 (0.41) 0.07 (0.48)
CG
0.98 0.86 1.01 0.77 0.70

CG1 -0.29
(-1.77) -0.37 (-1.87) -0.40 (-2.33)
VW Return
0.80 0.76 0.69 0.50 0.15

CL2 -0.64
(-4.25) -0.77 (-4.87) -0.76 (-6.25)
CL
0.92 0.87 0.72 0.58 0.13

CL1 -0.79
(-4.27) -0.95 (-5.79) -1.01 (-6.86)
TableIV.diosyncratVolityandSockReturnsCoditnaloCpitalGnsOverhang:CotrlingfoMAX
Athendofacmnth,wedivthsampleockintsubgropaednthircaplgnsoverhag.CL1nd2arepotfliscaptlos(CL) stockwihbelandovemiancptlose,rpctivly,ahendofacmnth.Silary,CGInd2arepotfliscaptgin(CG)stock withbeloandvemiancptlgsahendofacmnth.Wi eCLandGsubgrop,tcksarefuhsotedinquleportfisbaed onmaxiudlyretns(MAX),adwithnecof MAXportflis,anherfivpotlsarefomdbysrtingock theirmonlydiscrat volatiy(IVOL).Wecomputhvale-wightd(VW)anequlywighted(EW)runsoftheIVOLporflisthuobanedithfolwngmth.To contrlfMAX,weavrgtheunsacrofiveMAXportflis eachoftIVOLporflis botheCLandGsubgrop.Thenumbrsithe tablervgsofthevalu-wighted(VW)anqulyweightd(EW)reunsacofiveMAXportflis eachoftIVOLporflis.Therut withand outheJanrytusaeprovidnPaelsAdB,repctivly.VI(5)istheporfli stcknhebotm(p)quintleofIVOL. Thetablsorpthedifrnc etursofhextrmIVOLportflis,V5—IthecorspndigCAPMandFm-rech(F)alps,ndthe diferncothIVOLspreadbtwnheCL1adG2subgrop.Theturnsaipecntrmoh.Te^-staic,repotdinarehs, adjute forautcelionusgtheNwy-Wst(1987)mehod.Stckswihaprceoflsthan$5 edofthprliofmatnohrexcludfomthe sample.Thsamplecovrsthepriodf mJuly1963toDecmbr207. VI (Low) V2 V3 V4 V5(High) V5 - VI CAPM Alpha FF alpha
-0.73 -0.73
(-5.45) (-5.83) -0.72 (-5.78)
1.35 1.49 1.59 1.70 1.57 0.23

(1.53) 0.13 (1.01) 0.17 (1.72)
0.95 1.08 1.13 1.01 1.04 0.10 0.06

(0.43) -0.00 (0.36)
(-0.02)
EW Return
0.65 0.68 0.60 0.55 0.28

-0.37
(-3.16) -0.48 (-4.27) -0.40 (-5.09)
0.61 0.54 0.51 0.43 0.11

-0.50
(-3.87) -0.61 (-4.76) -0.54 (-5.17)
-1.09
(-5.58) -1.10 (-6.36) -1.15 (-6.54)
(Continued)
PanelB.Non—JanuaryMonths
1.20 1.26 1.31 1.38 1.32 0.12
(0.59) 0.00 (0.02) 0.08 (0.56)
0.96 0.82 0.95 0.72 0.60

-0.36
(-2.15) -0.44 (-2.14) -0.40 (-2.20)
VW Return
CLG C
0.67 0.65 0.56 0.32
-0.05 -0.72
(-4.57) -0.84 (-5.22) -0.75 (-6.11)
CL1 2GC L1-G2C L G1C2L-G 0.76 0.65 0.53 0.33

-0.22 -0.98
(-5.17) -1.10 (-6.51) -1.07 (-7.33)
TableIV.diosyncratVolityandSockReturnsCoditnaloCpitalGnsOverhang:CotrlingfoMAX
V1 (Low) V2 V3 V4 V5(High) V5 - VI CAPM Alpha FF alpha
-1.98 -1.96 -1.91

CL1-CG2 (-12.45) (-11.75) (-11.83) (Continued)
1.18 1.66 1.69 1.89 1.72 0.54

CG2 (2.65) 0.31 (1.72) 0.32 (2.11)
CG
1.09 1.35 1.51 1.34 1.16 0.07

CG1 (0.14)
-0.18
(-0.65) -0.14 (-0.63)
EW Return
0.88 1.08 0.95 0.84 0.04

CL2 -0.83
(-4.18) -1.09 (-5.68) -0.97 (-7.28)
CL
1.24 1.21 1.29 0.73

CL1 -0.20 -1.45
(-7.82) -1.66 (-8.45) -1.60
(-11.8)
-1.94 -1.97
(-9.02) (-8.89) -1.98 (-8.91)
CL1-CG2
Panel A. Al Months
1.04 1.38 1.40 1.50 1.51 0.47

CG2 (1.97) 0.24 (1.23) 0.26 (1.44)
CG
1.01 1.09 1.15 1.06 0.78

CG1 -0.23
(-0.99) -0.50 (-1.78) -0.44 (-1.90)
VW Return
0.82 0.93 0.74 0.65

CL2 -0.25 -1.07
(-4.47) -1.35 (-6.15) -1.19 (-7.30)
CL
1.12 1.08 1.01 0.57

CL1 -0.36 -1.48
(-5.98) -1.74 (-6.88) -1.71 (-8.78)
TableV.IdiosyncratVolityandSockReturnsCoditnaloCpitalGnsOverhang-CotrlingforREV
Athendofacmnth,wedivthsampleockintsubgropaednthircaplgnsoverhag.CL1nd2arepotfliscaptlos(CL) stockwihbelandovemiancptlose,rpctivly,ahendofacmnth.Silary,CGInd2arepotfliscaptgin(CG)stock withbeloandvmeiancptlgsahendofacmnth.Wi eCLandGsubgrop,tcksarefuhotedinquleportfisbaedon theirmonlyetur(REV),andwithecof REVprtfolis,anherfvpotlisarefomdbysrtingock theirmonlydiscratvoliy (IVOL).Wecomputhvale-wightd(VW)anequlywighted(EW)runsoftheIVOLporflisthuobanedithfolwngmth.Tocnrlf short-emvsal,wergtheunsacrofiveREVportflis eachoftIVOLprfolis btheCLandGgroups.Thenmbrsitheal arethvgsofthevalu-wigted(VW)anqulyweightd(EW)reunsacofiveREVportflis eachoftIVOLprfolis.Theutwiand withouJanryetusaprovidenPalsAdB,repctivly.VI(5)sportfli scknthebom(p)quintleofIVOL.Thtablesorpt thedifrnc etusofhextrmIVOLportflis,V5—IthecorspndigCAPMandFm-rech(F)alps,ndtheifrncotheIVOL spreadbtwnheCL1adG2subgrop.Theturnsaipecntrmoh.Te^-staic,repotdinarehs, adjutefor celation usingtheNwy-Wst(1987)mehod.Stckswihaprceoflsthan$5 edofthprliofmatnohrexcludfomthesapl.T me coversthepriodfromJuly1963toDecmber207. VI (Low) V2 V3 V4 V5(High) V5 - VI CAPM Alpha FF alpha
-2.08 -2.08 -2.01

CL1-CG2 (-12.73) (-11.40) (-11.38)
1.20 1.65 1.62 1.73 1.56 0.35

CG2 (1.59) 0.16 (0.88) 0.25 (1.69)
CG
1.02 1.24 1.34 1.13 0.84

CG1
(-0.81) -0.39 (-1.38) -0.24 (-1.03)
-0.18
0.71 0.85 0.68 0.47

CL2
-0.33 -1.05
(-4.53) -1.26 (-6.21) -1.06 (-7.70)
CL
0.95 0.85 0.87 0.25

CL1 -0.80 -1.75
(-7.82) -1.93 (-9.43) -1.77
(-12.02)
-1.99
(-8.95) -2.03 (-8.61) -1.96 (-8.55)
CL1-CG2
(Continued)
Panel B. Non-January Months

1.11 1.39 1.37 1.40 1.44 0.33
CG2 (1.31) 0.14 (0.70) 0.24 (1.36)
CG
1.01 1.06 1.11 0.91 0.60
VWReturnEWReturn
CG1
(-1.65) -0.65 (-2.19) -0.48 (-2.00)
-0.42
0.74 0.80 0.60 0.38

CL2 -0.53 -1.27
(-5.15) -1.51 (-6.39) -1.27 (-7.76)
CL
1.00 0.88 0.74 0.29

CL1
-0.68 -1.68 -1.89
(-6.58) (-7.23) -1.72 (-8.43)
TableV.IdiosyncratVolityandSockReturnsCoditnaloCpitalGnsOverhag-ContrligfoREV
VI (Low) V2 V3 V4 V5(High) V5 - VI CAPM Alpha FF alpha
Fama-French (1993) alphas have the same signs and are statistically significant as well. Note that
Bali et al. (2011) find that controlling for MAX reverses the sign of the negative IVOL-return
relationship. The results in Table IV suggest that this sign reversal is limited to those stocks with
embedded capital gains, and is not observed for stocks with capital losses.
When the January returns are excluded in Panel B, the results are qualitatively similar for the
value-weighted portfolios. The differences in the returns and the alphas are strongly negative and
statistically significant for the CL1 stocks, and generally positive, but insignificant for the CG2
stocks. The results for the equally weighted CL portfolios are much stronger as compared with
those in Panel A. For the CL1 stocks, the raw return difference between the high and low I VOL
portfolios is —0.50%, which is significant with a ^statistic of—3.87. The corresponding differ
ences in the CAPM and Fama-French (1993) alphas are also statistically significant. In contrast,
for both the CG groups, the equally weighted raw return difference is positive, but statistically
insignificant, and the corresponding differences in the alphas are also mostly insignificant. The
takeaway from these results is that after excluding the penny stocks and January returns, the
negative relationship between IVOL and the returns in stocks with unrealized capital losses is
robust to control for the MAX variable.
Turning to the results in Panel A of Table V, we find a statistically significant return difference
between the extreme IVOL portfolios in the CL1 stocks after controlling for REV, for both the
value-weighted and equally weighted portfolios. The difference in the value-weighted (equally
weighted) raw returns is — 1.48% (— 1.45%) with a ^-statistic of—5.98 (—7.82). The corresponding
difference in the CAPM alphas is —1.74% (—1.66%) with a ^-statistic of —6.88 (—8.45), and in
the Fama-French alphas is — 1.71% (— 1.60%) with a ^-statistic of —8.78 (— 11.82). For the CG2
stocks, the differences in the raw returns are positive and significant, but the alphas are statistically
insignificant for the value-weighted portfolios. The results in Panel B, obtained after excluding
the January returns, are similar. The differences in the raw returns and the alphas are negative
(and larger in magnitude when compared with those in Panel A) and statistically significant for
the CL1 stocks, and positive, but generally insignificant for the CG2 stocks. Thus, the negative
IVOL-return relationship in the capital loss stocks persists after controlling for REV, with and
without the inclusion of the January returns.
D. Fama-MacBeth Cross-Sectional Regressions

The results thus far lend support to our primary hypothesis that the negative volatility-return
relationship is concentrated in stocks with paper losses. In terms of robustness, our focus has been
on the MAX and REV variables, which have been shown to play an important role in explaining
the idiosyncratic volatility puzzle in the recent literature. We now examine the IVOL-return
relationship using firm-level Fama-MacBeth (1973) cross-sectional regressions that allow us to
control for additional variables along with MAX and REV The full cross-sectional regression
specification takes the following form:
Ru+l = at + ftIVOL,r + ftlVOL, ,CGO,,( + ftCGO,., + ftMAX,-,, + ft REV,,
+ ftPrc u + ft Size,,, + ftBTM,-,, + ftllliq,, + ft0Prel2RetlV + ftiSkw,.,
+ ftiBeta,,, + e,,r+i, (4)
where the dependent variable, RiJ+1, is t

variables, computed in month t, includ
(CGO), maximum daily return (MAX), s
(Pre), the natural log of market capitalization (Size), the book-to-market ratio (BTM), illiquidity
(Illiq), the cumulative return over 12-month period ending in month 1 (Prel2Ret), idiosyncratic
skewness (Skw), and the stock's beta (Beta). The details of the computation of these control
variables are included in the Appendix. The role of unrealized gains and losses on the IVOL
return relationship is captured through the interaction term, I VOL, , * CGO,,. The IVOL and all
of the return variables in the regression are in percentages.
We estimate various versions of Equation (4) with different subsets of the explanatory variables.
The average slope coefficients are reported in Table VI. The first specification is simply the
univariate regression on IVOL and confirms the negative IVOL-return relationship.10 Next, we
include the interaction term, IVOL,,, * CGO,,, along with IVOL as the explanatory variables.
We find that while the coefficient on IVOL is negative (—0.059) and statistically significant
(/-statistic = —4.39), the coefficient on the interaction term is positive (0.072) and statistically
significant (/-statistic = 5.91). Given that IVOL is nonnegative, this evidence implies that in
the case of negative CGO (embedded capital loss), the negative relationship between IVOL and
the next month's return is stronger. On the other hand, since the coefficient on interaction term
is larger than that of IVOL (in absolute terms), the relationship between IVOL and subsequent
returns is positive, on average, for a reasonable range of embedded capital gains levels. Even
for a small positive CGO, say 1%, the coefficient on IVOL is positive (—0.059 + 0.072*1 =
0.013). This evidence provides further confirmation that the negative volatility-return relationship
is concentrated among stocks with unrealized losses, but is nonexistent among stocks with
unrealized gains.
We obtain similar results after controlling for MAX or REV in Models (3) and (4) although
the IVOL coefficient is not significant in Model (3) after controlling for MAX. The interaction
term continues to be positive and statistically significant, consistent with a negative IVOL-return
relationship within the loss domain. Controlling for both MAX and REV simultaneously yields
similar results as demonstrated in Specification (5), although the MAX variable loses its statistical
significance.
In the full specification that includes all of the control variables, the coefficient on IVOL is
again negative (—0.052) and statistically significant (/-statistic = —4.22). The interaction term is
once again positive (0.087) and statistically significant (/-statistic = 5.71), with the coefficient
being larger in absolute magnitude than that on IVOL. Most of the control variables have the
expected signs. The coefficients on REV and firm size are negative and statistically significant,
while the coefficients on book-to-market, prior 12-month return, and idiosyncratic skewness are
positive and statistically significant.'1 The coefficient on the MAX variable is insignificant, as in
Specification (5).12 The beta coefficient is also insignificant, perhaps due to the inclusion of size
and the book-to-market ratio as control variables (Fama and French, 1993). Given the previously
documented implications of January seasonality for our results, we also run the regressions
10 This result is same as the one previously reported in Panel A.2 of Table III.
11 The significantly positive coefficient on lagged idiosyncratic skewness is consistent with the evidence in Bali
et al. (2011). Note that Boyer, Mitton, and Vorkink (2010) find a negative relationship between expected (not lagged)
idiosyncratic skewness and returns. As such, our results do not contradict their findings.
12 It is worth emphasizing that our sample differs from Bali et al. (2011) in that we exclude stocks priced below $5. With
this price restriction, the coefficient on MAX is rendered insignificant in the presence of other control variables. Without
the price restriction, we obtain a significantly negative coefficient on MAX, similar to Bali et al. (2011). Similarly, while
the coefficient on Amihud's (2002) illiquidity measure is negative, we obtain a significantly positive coefficient in a
sample that includes penny stocks.
Table VI. Fama-MacBeth Cross-Sectional Regressions
We run monthly firm-level Fama-MacBeth (1973) cross-sectional regressions of mo

stock returns on the lagged explanatory variables computed in month t. The explanatory
specifications include the stock's monthly idiosyncratic volatility (IVOL), the capital ga
the interaction term (IVOL*CGO), the maximum daily return (MAX), the monthly stoc
log of the stock price, the log of the market capitalization (Size), the book-to-mark
illiquidity measure (Illiq), the buy-and-hold return over the previous 12 months (Prel2
skewness (Skw), and the beta of the stock. The table reports the time-series means of c
from the cross-sectional regressions. The results for January and non-January months
in Specifications (7) and (8), respectively. The return and volatility numbers are in
f-statistics, reported in parentheses, are adjusted for autocorrelation using the Newey-W
Stocks with a price of less than $5 at the end month t — 1 are excluded from the sampl
the period from July 1963 to December 2007.
(1) (2) (3) (4) (5) (6) (7) (8)

All Months January Non-January
Months
IVOL -0.06 -0.06 0.00 -0.04 -0.03 -0.05 -0.06 -0.05
(-4.20) (-4.39) ( -0.14) (-2.43) ( -2.04) (-4.22) (-1.38) (-4.36)

IVOL*CGO 0.07 0.08 0.12 0.11 0.09 0.05 0.09
(5.91) (6.61) (7.75) (7.61) (5.71) (0.79) (6.09)
CGO -0.28 -0.30 -0.21 -0.20 -0.77 -1.85 -0.67
(-1-45) ( -1.60) (-1.15) ( -1.08) (-4.24) (-3.67) (-3.79)

MAX -0.08 0.00 -0.01 0.05 -0.01
(•
1
9s oo ( -0.31) (-0.63) (1.01) (-1.03)
REV -0.04 -0.04 -0.04 -0.09 -0.04
(-9.32) (• -9.51) (-9.49) (-8.48) (-8.43)
Log(Price) -0.10 -1.95 0.07
(-1.23) (-6.54) (0.77)

Size -0.11 -0.43 -0.09
(-3.18) (-2.17) (-2.36)
BTM 0.13 0.43 0.10
(2.32) (1.90) (1.79)

Illiq -0.01 -0.01 -0.01
(-2.07) (-0.67) (-2.49)
Pre 12 Ret 0.86 0.18 0.92
(7.34) (0.63) (7.41)
Skw 0.09 0.18 0.08
(3.97) (1.66) (3.54)
Beta -0.01 -0.13 0.00
(-0.47) (-1.49) (-0.11)
separately for January and non-January months.13 Consistent with Doran et al. (2012), the results
indicate that the negative volatility-return relationship is a non-January phenomenon. We do not
find evidence of a significant IVOL-return relationship in January. Similar to the all-month case,
for the non-January months, the coefficient on IVOL is significantly negative, and the coefficient
on the interaction term is significantly positive and larger in absolute magnitude.
13 For January regressions, the dependent variable is the January stock return and the lagged control variables are computed
in December.
E. Time-Series Regressions
In addition to the above cross-sectional tests, we also study the IVOL-return relationship using
monthly time-series regressions. These tests examine the returns to a strategy that takes a long
position in stocks in the highest volatility quintile and a short position in stocks in the lowest
volatility quintile each month, after controlling for various factors. These factors include the
monthly Fama-French (1993) factors (MKTRF, small minus big [SMB], and HML) and the
Carhart's (1997) momentum factor (UMD). In addition, to account for the role of short-term
reversals and maximum daily returns, we include two factors based on REV (the WML factor)
and MAX (the MMM factor) variables. Following Huang et al. (2010), the WML factor represents
the return on a portfolio that takes a long position in stocks in the highest REV decile and a short
position in stocks in the lowest REV decile in the month of portfolio formation. The monthly
MMM factor is constructed similarly as the difference in the return of a portfolio long in stocks
in the top decile of the maximum daily return and short in stocks in the bottom decile of the
maximum daily return.
Finally, we also incorporate a monthly sentiment index (SENT) in these time-series tests.
This sentiment index, based on Baker and Wurgler (2006, 2007), represents the first principal
component of commonly employed proxies for investor sentiment including the closed-end fund
discount, the NYSE share turnover, the number and first-day returns of initial public offerings,
the equity share in new issues of debt and equity, and the dividend premium.14 The motivation
for including the sentiment index in our analysis stems from Baker and Wurgler's (2006) finding
that the high-volatility stocks earn relatively lower (higher) returns when the sentiment index
is positive (negative), suggesting a potential role of investor sentiment in the volatility-return
relationship.15
The results of the time-series regressions are reported in Table VII for the entire sample, as well
as separately for the CL and CG subgroups, CL1, CL2, CGI, and CG2. For the entire sample,
we note that when only the standard Fama-French (1993) and momentum factors are included
as explanatory variables, the intercept from the time-series regression is negative (—0.727) and
statistically significant (/-statistic = —6.23). This result confirms the existence of the negative
IVOL-return relationship. The coefficient on size factor is positive and significant, while that
on the book-to-market factor is negative and significant, consistent with high (low) IVOL firms
being small, growth (large, value) firms. The coefficient on the market factor is positive and
significant, while the coefficient on the momentum factor is insignificant.
When we also include WML, MMM, and SENT factors in the time-series regression, we
find that the intercept is no longer statistically significant. In contrast to Huang et al. (2010),
the coefficient on the WML factor is insignificant, due to the exclusion of penny stocks. The
coefficient on the MMM factor is positive (0.749) and highly significant (/-statistic = 14.70),
confirming a strong positive relation between MAX and IVOL, as documented in Bali et al.
(2011). The MMM factor is responsible for the insignificance of the intercept in these time
series regressions.16 The coefficient on SENT is negative, but not statistically significant. After
14 The sentiment index is obtained from Jeff Wurgler's website at http://pages.stern.nyu.edu/-jwurgler.
15 However, note that Baker and Wurgler's (2006) result is based on total volatility (measured as the standard deviation of
monthly returns over the past year) and not idiosyncratic volatility. Nonetheless, Ang et al. (2006) and others document
a negative relationship between total volatility and subsequent returns as well.
16 We verified this result by including the WML, MMM, and SENT factors, one at a time, along with the Fama-French
(1993) and momentum factors. The intercept is significant with the WML and SENT factors, but insignificant with the
MMM factor.
1.02
(10) (6.33) 0.08 (1.68) 0.33 (4.87) 0.04 (0.61) 0.09 (2.07) 0.15 (3.92) 0.62 0.39 (2.93)
(11.09)
(9) 0.10 (0.57) 0.28 (5.49) 0.91

(8.12) 0.14
(1.97)
-0.34
(-3.16)
(8) 0.30
(2.27) 0.29 (5.37) 0.16
(3.77) 0.70
-0.02 -0.06 -0.03 (14.43) -0.39 (-2.39)
(-0.27) (-1.19) (-0.51)
(7) 0.27
(5.32) 0.97 0.16
(2.88)
-0.47 (12.42) -0.48 (-5.18)
(-2.69)
(6) 0.05 (1.83) 0.13 (2.63) 0.11

(4.25) 0.68
-0.77 -0.02 -0.14 (19.37) -0.33 (-3.25)
(-7.78) (-0.43) (-3.50)
0.35 0.06 (1.29)

(5) (8.81) 0.79
-1.37 (16.08) -0.42 (-4.88)
(-10.43)
(4) 0.04
(0.95) 0.84
-0.99
(—7.71) -0.13 (-4.44) -0.06 (-0.95) -0.01
(-0.38) -0.31 (-6.71) (19.02) -0.25 (-2.40)
TableVI.Time-S riesRegresions
(3) 0.27
(5.50) 0.77 (9.79) -0.45
-1.26
(-9.88) (-3.44) -0.12 (-1.59)
(2) 0.02
(0.10) 0.00 (0.11) 0.19 (3.02) 0.00 0.05
(1.39) -0.09 0.75
(-0.06) (-1.51) (14.70) -0.14 (-1.58)
AlStocksCL1 2GC2 (1) 0.32 0.92 0.02

(8.46) (0.37)
-0.73
(-6.23) (12.65) -0.45 (-5.17)
Thetablrpostheulfromthei-sr egions.Athedofachmnt,wedivthsampleockintsubgropaednthircaplgns overhang.CL1d2arepotfliscaptlos(CL)ckwithbeloandvemiancptlose,rpctivlyahendofacmnth.Silary, CG1and2repotfliscaptgin(CG)stockwihbelandovemiancptlgsahendofacmnth.TesockwithneCLad CGsubgropaefuthrsoedintquleportfisbaedonthirm lydiosncratvliy(IVOL).Thedpntvaribleshmontlyreu ofaprtlihakeslongpit heigstIVOLporfliandshortpin helowstIVOLprfoli.Thendp tvariblesncudth Fam-rench(193)market,sizndbok-tmarefctos(MKTRF,SBandHML,respctivly),hemontufacor(UMD),therunoaprtfli thakeslongpit heigstdcleanshortpin helowstdcifheprviousmnth'axiumdlyretn(M),theruno aportflihakeslongpit heigstdcleanshortpinthelowsdciefthpasone-mthrun(WML),adtheBkrand Wurgle's(206)ntime dx(SENT).hertunsaipercnt moh.Te^-staic,repotdinarehs, adjutefor celationusg theNwy-Wst(1987)mehod.Stckswihaprceoflsthan$5 edofthpreviusmontharexcludfomthesapl.T mecovrsthpeiod fromJuly1963toDec mber20 7. Intercept MKTRF SMB HML UMD WML MMM SENT
controlling for these additional factors, the coefficients on market and the HML factors lose their
significance, while the SMB factor continues to be positive and significant.
We now turn to the results for the CL and CG subgroups. The most significant finding for
the CL subgroup is that the intercepts are negative and statistically significant after controlling
for the Fama-French (1993) and the momentum factors, as well as after including the other
three factors as control variables in the regression. The intercept from the seven-factor model is
—0.991 (f-statistic = —7.71) for the CL1 subgroup and —0.774 (/-statistic = —7.78) for the CL2
subgroup. For the CG stocks, the intercept from the four-factor model is negative for CGI and
positive, albeit insignificant, for the CG2 group. The most significant result for the CG stocks is
that when all seven factors are included, the intercepts are positive and statistically significant for
both the CGI and CG2 subgroups.17 Specifically, for stocks in the CGI subgroup, the intercept
from the seven factor model is 0.303 (f-statistic = 2.27), while for stocks in the CG2 subgroup,
the intercept is 1.018 (/-statistic = 6.33).
Overall, the results from the time-series regressions corroborate the evidence from the portfolio
level analyses and the cross-sectional regressions. Our findings support the hypothesis that the at
tractiveness of high idiosyncratic volatility stocks to risk-seeking investors with unrealized capital
losses drives the observed negative relationship between idiosyncratic volatility and subsequent
returns.
IV. Concluding Remarks
considered a significant anomaly in the finance literature. In this paper, we propose a Prospect
Theory-based explanation of this anomalous relationship. Prospect Theory, in conjunction with
mental accounting, suggests that stocks with unrealized capital losses would be overpriced due to
the tendency of disposition-prone investors to hold on to these stocks too long. Furthermore, the
investors in the loss domain are risk seeking, implying a greater affinity for high idiosyncratic
volatility stocks that results in greater overpricing of these stocks in equilibrium. Therefore, we
hypothesize that the observed low average returns to high idiosyncratic volatility stocks would
be concentrated in stocks with unrealized capital losses. Our results support this hypothesis. We
find evidence of a robust negative relationship between idiosyncratic volatility and subsequent
returns among stocks with unrealized losses, but find no such relationship among stocks with
unrealized gains.
Alternative rational explanations, such as those based on limits to arbitrage, are also possible.
For example, it is possible that high idiosyncratic risk deters arbitrage (Pontiff, 2006) causing
greater overpricing and subsequently lower returns among these high-volatility stocks. This
argument by itself, however, does not explain why the anomaly is observed only in stocks with
unrealized losses and not in stocks with unrealized gains. It is predicated on the existence
of overpricing only among unrealized loss stocks. Similar to Grinblatt and Flan (2005), the
overpricing of unrealized loss stocks in our framework stems from risk-seeking behavior in the
loss domain, as postulated in the Prospect Theory. In contrast, high idiosyncratic risk deters
17 The positive intercept for the CG stocks reflects the evidence of sign reversal in the presence of MAX in our cross
sectional tests, as well in Bali et al. (2011). The WML and MMM factors have average monthly returns of —0.99% and
— 1.13%, respectively. The CG stocks have significantly positive loadings on these factors and, as such, the alpha for the
CG stocks turns positive when these factors are included as explanatory variables. Additional tests suggest that consistent
with Bali et al. (2011), it is the MMM factor that drives the sign reversal. Note, however, that this positive IVOL-return
relationship occurs only among CG stocks and not CL stocks.
Bhootra & Hur • High idiosyncratic Volatility and Low Returns 319
arbitrage due to arbitrageurs' risk aversion, as pointed out in Pontiff (2006). Therefore, an
alternative explanation of the overpricing of unrealized loss stocks that does not rely on the
assumption of risk-seeking behavior is necessary for a limits to arbitrage based explanation.
It is also possible that short sale constraints deter the correction of overpricing among high
volatility stocks with unrealized losses. However, the impact of short sale constraints is mitigated
due to the exclusion of stocks priced below $5. Diether, Lee, and Werner (2009) document higher
short selling in stocks priced at or above $5, as compared with those priced below $5, due to the
higher collateral costs associated with stocks priced below $5. Nonetheless, we believe that these
limits to arbitrage-based explanations deserve detailed exploration and offer a fruitful area for
future research.
Appendix
Variable Definitions
IVOL: During month t, we run the following regression of excess daily returns of each stock i
on contemporaneous daily Fama-French (1993) factors:
Rj.d — Rf,d — ai + fiiMKTd + SiSMBd + hjHMLj +
where is the return of stock i on day d, Rfd is the T-bill return on day d, and MKTRFj
SMBj, and HMLj are the daily Fama-French (1993) market, size, and book-to-market factors.
The monthly idiosyncratic volatility of stock i in month t is defined as the standard deviation
of the residuals from this regression times the square root of the number of trading days in th
month:
IVOLi, t = %/var(e, rf) x \/A>
where Dt is the number of trading days for stock i in month t.

CGO: Similar to Grinblatt and Han (2005) and Hur et al. (2010), for each stock i at the end of
each month t, the capital gains overhang (CGO,.,) is obtained as
CGO= (Pi,t - RPi,t)/Pi,„
where Pit is the price of stock i at the end of month t and R P, , is the reference price for each
stock i at the end of month t. The reference price, R P, ,, is estimated as follows:
where Vi t is turnover in stock i on day t, T is the number of trading days in the previous thre
years with available daily price and volume information, and is price of security i on da
t — n.
MAX: MAX is the maximum daily return of a stock in a given month. If Dt is the number of
trading days for stock i in month t, then MAXU is given by
MAXij = max(/?,-,!, ..., Ri.d,),
320 Financial Management - Summer 2015
where Rl (i {d = 1,2, ..., £>,) is the return on stock i on day d.

REV: The REV variable is used to capture short-term reversals in stock returns and equals the
return of stock i in month t; that is, REVlt = Rit .
Skw: Skw is the daily idiosyncratic skewness of a stock, which is the skewness of the residuals
from the following regression:
Ri,d Rf.d = @i ~i~ Rf,d) Yi (Rm,d Rfd) "1" &i,di
where R^, Rfj, and Rmei are the return on stock i on day d, the T-bill return on day d, and the
return on the CRSP value-weighted market index on day d, respectively.
Size: Size is the natural logarithm of the stock's month-end market capitalization (price times
shares outstanding).
BTM: BTM is the firm's book-to-market ratio. Following Fama and French (1993), we compute
the BTM in month t of a year as the ratio of the book value of equity for the fiscal year ending in
the prior calendar year and market equity at the end of December of the prior calendar year. The
book value of equity, computed using Compustat data, is the stockholders' equity (DATA 216),
plus the balance sheet deferred taxes and investment tax credit (DATA 35), minus the book value
of preferred stock (DATA56 or DATA10 or DATA 130, in that order) at the fiscal year end.
Prel2Ret: Prel2Ret is the momentum variable. Following Jegadeesh and Titman (1993), the
momentum variable for each stock in a given month is defined as its buy-and-hold return over
the past 12 months.
Illiq: Illiq is the measure of illiquidity for a stock in a given month. Following Amihud (2002),
Illiq is measured as the ratio of a stock's absolute monthly return to its dollar trading volume:
ILLIQ,, = \Ru\/VOLDiA
where R,, and VOLD, , are the return and dollar volume, respectively, of stock i in month t.
Beta: We use the daily returns within a month to estimate a stocks' beta and employ the
adjustment procedure of Scholes and Williams (1977) and Dimson (1979) to mitigate the impact
of nonsynchronous trading. Beta is estimated using following regression model:
Hi,d Rf.d — @i ~i~ ,i (,^m,d—l Rf.d— l) fi2,i(Rm,d ^i.d)
+P3,i(Rm,d+\ — Rf,d+1) + ^i,d,
where Ri c/, Rj j, and Rmj are the return on stock i on day d, the T-bill return on day d, and the
return on the CRSP value-weighted market index on day d, respectively. The estimate of a stock's
beta is given by = /3U + }2 i + fiXi.
References
Amihud, Y., 2002, "Illiquidity and Stock Returns: Cross-Section and Time-Series Effects," Journal of
Financial Markets 5, 31-56.
Ang, A., R.J. Hodrick, Y. Xing, and X. Zhang, 2006, "The Cross-Section of Volatility and Expected Returns,
Journal of Finance 61, 259-299.
Ang, A., R.J. Hodrick, Y. Xing, and X. Zhang, 2009, "High Idiosyncratic Volatility
International and Further US Evidence," Journal of Financial Economics 91, 1-23.
Asparouhova, E., H. Bessembinder, and I. Kalcheva, 2011, "Noisy Prices and Inference R
Journal of Finance 68, 665-714.
Baker, M. and J. Wurgler, 2006, "Investor Sentiment and the Cross-Section of Stock R
Finance 61, 1645-1680.
Baker, M. and J. Wurgler, 2007, "Investor Sentiment in the Stock Market," Journal of E
21, 129-152.
Bali, T.G. and N. Cakici, 2008, "Idiosyncratic Volatility and the Cross Section of Expected
of Financial and Quantitative Analysis 43, 29-58.
Bali, T.G., N. Cakici, and R.F. Whitelaw, 2011, "Maxing Out: Stocks as Lotteries and t
Expected Returns," Journal of Financial Economics 99, 427^-46.
Barberis, N., M. Huang, and T. Santos, 1999, "Prospect Theory and Asset Prices," NB
No. 7220.
Barberis, N., M. Huang, and T. Santos, 2001, "Prospect Theory and Asset Prices," Quarterly Journal of
Economics 116, 1-53.
Barberis, N. and W. Xiong, 2012, "Realization Utility," Journal of Financial Economics 104, 251-271.
Boyer, B., T. Mitton, and K. Vorkink, 2010, "Expected Idiosyncratic Skewness," Review of Financial Studies
23, 169-202.
Carhart, M.M., 1997, "On Persistence in Mutual Fund Performance," Journal of Finance 52, 57-82.
Diether, K.B., K. Lee, and I.M. Werner, 2009, "Short-Sale Strategies and Return Predictability," Review of
Financial Studies 22, 575-607.
Dimson, E., 1979, "Risk Measurement When Shares are Subject to Infrequent Trading," Journal of Financial
Doran, J.S., D. Jiang, and D.R. Peterson, 2012, "Gambling Preference and the New Year Effect of Assets
with Lottery Features," Review of Finance 16, 685-731.
Fama, E.F. and K.R. French, 1993, "Common Risk Factors in the Returns on Stocks and Bonds "Journal of
Financial Economics 33, 3-56.
Fama, E.F. and J. MacBeth, 1973, "Risk, Return, and Equilibrium: Empirical Tests," Journal of Political
Economy 81, 607-636.
Fink, J.D., K.E. Fink, and H. He, 2012, "Expected Idiosyncratic Volatility Measures and Expected Returns,"
Financial Management 41, 519-553.
Fu, F., 2009, "Idiosyncratic Risk and the Cross-Section of Expected Stock Returns," Journal of Financial
George, T.H. and C. Hwang, 2013, "Analyst Coverage and the Two Volatility Puzzles in the Cross-Section
of Returns," University of Houston Working Paper.
Grinblatt, M. and B. Han, 2005, "Prospect Theory, Mental Accounting, and Momentum," Journal of
Financial Economics 78, 311-339.
Huang, W., Q. Liu, S.G. Rhee, and L. Zhang, 2010, "Return Reversals, Idiosyncratic Risk, and Expected
Returns," Review of Financial Studies 23, 147-168.
Hur, J., M. Pritamani, and V Sharma, 2010, "Momentu
Investors," Financial Management 39, 1155-1176.
Jegadeesh, N., 1990, "Evidence of Predictable Behav

881-898.
Jegadeesh, N. and S. Titman, 1993, "Returns to Buying Winners and Selling Losers: Implications for Marke
Efficiency," Journal of Finance 48, 65-91.
Kahneman, D. and A. T versky, 1979, "Prospect Theory: An Analysis of Decision under Risk," Econometri
47, 263-291.
Lehmann, B.N., 1990, "Fads, Martingales, and Market Efficiency," Quarterly Journal of Economics 105
1-28.
Lintner, J., 1965, "The Valuation of Risky Assets and the Selection of Risky Investments in Stock Portfolios
and Capital Budgets," Review of Economics and Statistics 47, 13-37.
Lo, A.W. and A.C. MacKinlay, 1990, "When are Contrarian Profits Due to Stock Market Overreaction?"
Review of Financial Studies 3, 175-205.
Merton, R.C., 1987, "Presidential Address: A Simple Model of Capital Market Equilibrium with Incomplete
Information," Journal of Finance 42, 483-510.
Mossin, J., 1966, "Equilibrium in a Capital Asset Market," Econometrica 34, 768-783.
Newey, W. and K. West, 1987, "A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation
Consistent Covariance Matrix," Econometrica 55, 703-708.
Peterson, D.R. and A.R. Smedema, 2011, "The Return Impact of Realized and Expected Idiosyncratic
Volatility," Journal of Banking and Finance 35, 2547-2558.
Pontiff, J., 2006, "Costly Arbitrage and the Myth of Idiosyncratic Risk," Journal of Accounting and Eco
nomics 42, 35-52.
Scholes, M. and J. Williams, 1977, "Estimating Betas from Nonsynchronous Data," Journal of Financial
Economics 5, 309—327.
Sharpe, W.F., 1964, "Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk,"
Journal of Finance 19, 425^142.
Shefrin, H. and M. Statman, 1985, "The Disposition to Sell Winners Too Early and Ride Losers Too Long:
Theory and Evidence," Journal of Finance 40, 777-790.
Thaler, R., 1980, "Toward a Positive Theory of Consumer Choice," Journal of Economic Behavior and
Organization 1, 39-60.

High Idiosyncratic Volatility and Low Returns A Prospect Theory Explanation

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

High Idiosyncratic Volatility and Low Returns A Prospect Theory Explanation

Uploaded by

Copyright:

Available Formats

High Idiosyncratic Volatility and Low Returns: A Prospect Theory Explanation

Author(s): Ajay Bhootra and Jungshik Hur

Wiley, Financial Management Association International are collaborating with JSTOR to

I. Hypothesis Development and Related Literature

Figure 1. Prospect Theory Risk Prefere

The figure depicts S-shaped prospect theory utility,

in the domain of losses. This S-shaped utility fu

U(x) =x05, x>0,

U(x) = - 2(-xf5, x< 0.

II. Data and Methodology

Rj d — Rfd = ai + PiMKTRFd + SiSMBd + hjHMLd + (1)

CGO,-, = (Pu - RPu)/Pi,t, (2)

A. Idiosyncratic Volatility and Stock Returns

Panel A. Average Returns of IVOL Portfolios (No Price Filter)

VI (Low) 0.93 1.16

V5(High) 0.02 1.11

V5(High) 0.36 0.51

B. Preliminary Tests of the Hypothesis

Panel A. Sequential Sorting

VI (Low) 1.08 0.82 0.93 0.97 1.24 0.99 1.05 1.18

Panel B. Independent Sorting

Table III. Role of MAX and REV Variables in IVOL-Return Relationship

Panel A. All Months

Panel A. I. No Price Filter Panel A.2. Price > $5

IVOL MAX REV IVOL MAX REV

(1) -0.01 -0.06

(0.65) (-1.26) (-11.26) (-3.20) (0.95) (-6.20)

Panel B. Non-January Months

IVOL MAX REV IVOL MAX REV

(7) -0.04 -0.07

(-1.99) (-9.49) (-4.29) (-4.55)

1.36 1.53 1.64 1.77 1.71 0.36

1.04 1.19 1.26 1.19 1.26 0.23

0.84 0.92 0.84 0.86 0.62

0.92 0.91 0.91 0.90 0.67

1.19 1.23 1.32 1.41 1.38 0.18

0.98 0.86 1.01 0.77 0.70

0.80 0.76 0.69 0.50 0.15

0.92 0.87 0.72 0.58 0.13

1.35 1.49 1.59 1.70 1.57 0.23

0.95 1.08 1.13 1.01 1.04 0.10 0.06

0.65 0.68 0.60 0.55 0.28

0.61 0.54 0.51 0.43 0.11

0.96 0.82 0.95 0.72 0.60

CL1 2GC L1-G2C L G1C2L-G 0.76 0.65 0.53 0.33

V1 (Low) V2 V3 V4 V5(High) V5 - VI CAPM Alpha FF alpha

-1.98 -1.96 -1.91

1.18 1.66 1.69 1.89 1.72 0.54

1.09 1.35 1.51 1.34 1.16 0.07

0.88 1.08 0.95 0.84 0.04

1.24 1.21 1.29 0.73

1.04 1.38 1.40 1.50 1.51 0.47

1.01 1.09 1.15 1.06 0.78

0.82 0.93 0.74 0.65

1.12 1.08 1.01 0.57

-2.08 -2.08 -2.01

1.20 1.65 1.62 1.73 1.56 0.35

1.02 1.24 1.34 1.13 0.84

0.71 0.85 0.68 0.47

0.95 0.85 0.87 0.25

Panel B. Non-January Months

1.01 1.06 1.11 0.91 0.60

0.74 0.80 0.60 0.38