Do extreme returns matter in emerging markets? Evidence from the
Chinese stock market
Gilbert V. Nartea
Senior Lecturer in Finance
Department of Accounting, Economics and Finance
Faculty of commerce
Lincoln University
Lincoln 7647
Christchurch, NEW ZEALAND
Email: [email protected]
Phone: 64-3-325-3627
Fax: 64-3-325-3847
Ji Wu*
Assistant Professor of Finance
Institute for Financial & Accounting Studies
Xiamen University
Fujian, 361005 P.R.CHINA
Email:[email protected]
Mobile: 86 18650431259
Fax: 86-592-2181787
*Corresponding author
Do extreme returns matter in emerging markets? Evidence from the
Chinese stock market
Abstract
Recent evidence in the U.S. indicates that stocks with high maximum daily returns in
the previous month, perform poorly in the current month. We investigate the presence
of a similar effect in the Chinese stock markets with portfolio-level analysis and firm-
level Fama-MacBeth cross-sectional regressions. Unlike Bali et al’s (2011) findings,
we find no evidence of a MAX effect if we hold portfolios for just one month.
However, we find evidence of a negative relationship between extreme positive
returns (MAX) and stock returns in the Chinese stock market if we extend the holding
period to three and six months. Interpreted together with the strong evidence of risk-
seeking behaviour among Chinese investors, our results are consistent with the
suggestion that the negative MAX effect is driven by investor preference for stocks
with lottery-like features. Our results underscore the importance of in-country
verification, especially in emerging markets, of apparent anomalies initially
discovered in developed stock markets.
JEL classification: F39; G12
Keywords: Cross-section of stock returns; Extreme returns; Predictability; China.
Conference category: Economics and Finance
Do extreme returns matter in emerging markets? Evidence from the
Chinese stock market
1. Introduction
Motivated by recent evidence from Kumar (2009) that investors in the U.S. stock
markets exhibit a preference for stocks with lottery-like characteristics, Bali, Cakici,
and Whitelaw (2011) investigate the role of extreme positive returns in the cross-
sectional pricing of stocks in the U.S. Considering stocks that exhibit extreme positive
returns to be lottery-like, Bali, et al. (2011) find that stocks with the highest maximum
daily returns in the previous month (MAX), tend to perform poorly in the following
month. For their decile portfolios, they report negative raw and risk-adjusted return
spreads between portfolios with the highest and lowest maximum daily returns
exceeding 1% per month. The negative relationship is robust even as they control for
size, book-to-market, momentum, short-term reversals, liquidity, and skewness. Bali
et al. (2011) explain the apparent negative MAX effect as a result of investor
preference for stocks with lottery-like characteristics, in particular those with the
potential to produce high maximum daily returns albeit with low probability. A
preference for these stocks leads to overpayment which eventually results to
underperformance in the following month. Such behaviour is consistent with two
descriptive models of decision making under uncertainty -- Tversky and Kahneman’s
(1992) cumulative prospect theory (CPT), as recently extended by Barberis and
Huang (2008) and Kothiyal et al. (2011) and the optimal expectations framework of
Brunnermeier and Parker (2005) and Brunnermeier et al. (2007). Cumulative prospect
theory is a non-expected utility model that accommodates overweighting of tails of
distributions as a modelling device that captures the common preference for lottery-
like (positively skewed) wealth distributions (Barberis and Huang, 2008). In the
optimal expectations framework of Brunnermeier and Parker (2005) and
Brunnermeier et al. (2007) decision makers deliberately choose to distort their beliefs
by overestimating the probabilities of events in which their investments pay off well.
Brunnermeier et al. (2005) show that this model leads to a) portfolios that are
underdiversified, b) investors exhibiting a preference for lottery-like assets and c) that
these lottery-like assets tend to have lower returns.
Empirical evidence of the MAX effect in other markets is still very sparse. Apart from
Nartea, Wu, and Liu (2011) who also document a negative MAX effect in the South
Korean stock market, we are not aware of studies done in other emerging stock
markets. By focussing on China we not only examine the world’s largest emerging
market but we are also presented with a unique opportunity to test Bali et al.’s (2011)
“preference for lottery stocks” explanation of the MAX effect, in as much as Chinese
investors have been shown to exhibit risk-seeking behaviour (see for example Ma,
1996; Ng and Wu, 2006; Lee and Wong, 2009; Fong, Wong, and Yong, 2010).
Studies have also shown that social gambling is considered an acceptable form of
entertainment within the Chinese culture (Raylu, and Oei, 2004; Loo, Raylu, and Oei,
2008) which leads to a predisposition for lottery-like stocks. If Bali et al.’s (2011)
explanation is valid, we expect to document a negative MAX effect in the Chinese
stock markets.
In less than twenty years China’s two stock markets, located in Shanghai and
Shenzhen, have grown from a handful of listed stocks to collectively become the
second largest stock market in the world by the end of 2011 behind only the US stock
markets and the largest emerging stock market. Inspite of this remarkable progress
there is still enormous potential for growth since the proportion of their market
capitalisation to China’s GDP is only 49% as of the end of 2009 compared with 86%
for the U.S. stock markets. As China continues to open its markets to foreign investors,
understanding the factors that drive stock price movements in its stock markets has
become an important issue.
For comparability we follow Bali, et al’s (2011) portfolio sorting approach. We sort
stocks according to maximum daily returns in the previous month, form portfolios on
this basis and track the returns of these portfolios in the succeeding month, reforming
portfolios monthly. We also vary the portfolio holding period to three and six months
and confirm the robustness of our results with a double-sort procedure to control for
various cross-sectional effects including size, book-to-market, momentum, short-term
reversal, illiquidity, idiosyncratic volatility, and skewness. In addition to portfolio
analysis, we perform firm-level Fama-MacBeth cross-sectional regressions as further
robustness tests.
Unlike Bali et al. (2011), we find no evidence of a MAX effect if we hold portfolios
for just one month. However, we find evidence of a negative relationship between
extreme positive returns (MAX) and stock returns if we extend the holding period to
three and six months. Our results support the findings of Bali, et al. (2011) and Nartea
et al. (2011) of similar negative MAX effects in the U.S. and Korean stock markets,
respectively. Interpreted together with the strong evidence of risk-seeking behaviour
among Chinese investors documented in the extant literature, our results are also
consistent with Bali, et al.’s (2011) suggestion that the negative MAX effect is driven
by investor preference for stocks with lottery-like features. Insofar as the negative
MAX effect is driven by overpayment of investors for high MAX stocks which then
leads to underperformance, we suggest that this could be due to price trading limits
which cause lags in the adjustment of prices back to fundamental levels, by up to six
months. We further suggest that the negative MAX effect persists due to short-selling
constraints that limit the opportunity to trade this effect away.
The rest of the paper is organized as follows: Section 2 describes our data and
discusses our estimation procedures. It describes the single-sort method of portfolio
analysis and the double-sort procedure that is used to control for various known
effects. Section 3 reports the empirical results with section 3.1 dealing with results of
portfolio-level analysis while section 3.2 reports results of firm-level Fama-MacBeth
cross-sectional regressions. We use an alternative measure of extreme returns in
section 3.3 and extend the portfolio holding period to three and six months in section
3.4. Section 4 concludes.
2. Data and Methods
Daily and monthly stock returns and accounting data for individual firms were
obtained from DataStream. We use A-shares listed in both the Shanghai and Shenzhen
stock exchanges. The data set covered the period from August 1993 with 83 firms, to
November 2007 with 1,201 firms with an average of 741 firms per month resulting in
a total of 128,325 firm-month observations. The risk-free rate which is defined as the
demand deposit rate was also obtained from DataStream. Market returns are the
value-weighted returns of all firms used in the study.
Following common practice in the existing literature we eliminated investment trusts,
closed-end funds, exchange traded funds, and preferred shares from the sample. We
also deleted stocks with daily returns less than -100% and stocks with monthly returns
greater than 200% to avoid the influence of extreme returns and possible data
recording errors.
At the end of each month, we form tertile portfolios according to MAX, defined as the
maximum daily return in the past calendar month-- high MAX (HMAX), medium
MAX (MMAX), and low MAX (LMAX). We apply holding periods of one, three and
six months and determine the raw and risk-adjusted returns (alpha) of each portfolio.
Portfolios are reformed every month. The risk-adjusted return refers to the Fama-
French (1993) three-factor model alpha (FF-3 alpha) estimated using the full sample
of the time-series of value- or equal-weighted returns for each portfolio. We relate
MAX in t with raw and risk-adjusted returns in month t+1, the three-month return
ending in month t+3, and the six-month return ending in month t+6.
We control for several variables including size, book-to-market (BM), intermediate-
term momentum, short-term reversals, illiquidity, skewness, and idiosyncratic
volatility using dependent bi-variate sorts similar to that employed by Bali et al.
(2011). First we sort on the control factor (i.e., size, value, momentum, and so on) into
tertiles. Within each tertile we sort further into tertiles based on MAX. Then we
average within each MAX category resulting in three portfolios with variation in
MAX but similar levels in the control variable. For example, to control for size,
stocks are first sorted into tertiles according to market capitalisation – Big, Medium,
Small. Within each size category, stocks are sorted again into tertiles according to
MAX – high (HMAX), medium (MMAX), and low MAX (LMAX). Therefore, nine
size-MAX portfolios are formed, namely Big-HMAX, Big-MMAX, Big-LMAX,
Middle-HMAX, Middle-MMAX, Middle-LMAX, Small-HMAX, Small-MMAX, and
Small-LMAX. To control for size, we construct a size-neutral portfolio by averaging
the return (or alpha) spreads within each MAX category. To illustrate, a size-neutral
HMAX portfolio is constructed by averaging the alpha spreads of the three HMAX
portfolios within each size tertile, i.e., Big-HMAX, Middle-HMAX and Small-
HMAX so that we have a high MAX portfolio which contains all sizes. We do the
same for the MMAX, and LMAX portfolios. This process results in three portfolios
with variation in MAX but similar levels in the control variable -- size. We replicate
this procedure for the other control variables.
The size variable at the beginning of month t is defined as the log of the firm’s market
capitalization at the end of month t-1, book-to-market is the firm’s book-to-market
ratio six months prior, i.e. at the end of t-6.1 Following Jegadeesh and Titman (1993),
the momentum variable at time t is the stock’s 11-month past return lagged one month,
i.e. return from month t-12 to month t-2. The short-term reversal variable is defined
following Jegadeesh (1990) as the stock’s one month past return, i.e. return in month
t-1. Skewness of stock i as of the beginning of month tis computed using daily returns
in the past 22 trading days. Idiosyncratic volatility (IV) of stock i at the beginning of
month t is defined as the standard deviation of daily residuals from the Fama-French
three factor model (1) estimated using daily returns in month t-1. Ri,t and MKT are
excess returns of firm i and the market, respectively, over the risk-free rate. We
generate daily values of SMB by sorting stocks at the beginning of every month t into
three groups according to size (Small, Medium, Big). SMB is the difference in daily
returns between the small- and large-stock portfolios. Similarly we generate daily
values of HML by sorting stocks into three groups according to their book-to-market
(BM) ratio in month t-6 (High-, Medium-, and Low-BM). HML is the difference in
daily returns between the High- and Low-BM stock portfolios. Portfolios are
reformed every month.
Ri,t = α + βMKT, i, mMKTt + βSMB, i ,mSMBt + βHML, i ,mHMLt + εi,t (1)
3. Empirical Results
1Clubb and Naffi (2007) also show that expected book-to-market ratios also explain UK stock returns.
However, in this study we only deal with past book-to-market ratios. Michou (2009) also reports that
the predictive power of the book-to-market spread depends on portfolio formation strategies and the
relative proportion of large-caps, small-caps, value, and growth stocks in the portfolio.
3.1 Portfolio Analysis
3.1.1 Univariate sorting
Table 1 shows the returns and FF-3 alpha of portfolios sorted on the maximum daily
returns in the past month (MAX). We report results for both value- and equal-
weighted portfolios. Columns 1 and 3 indicate the absence of MAX effect based on
both equal- (EW) and value-weighted (VW) returns since while the difference
between the returns of the high and low MAX portfolios is negative for both EW and
VW returns, these return spreads are statistically insignificant. However, column 2
indicates a negative MAX effect based on the FF-3 alpha of EW portfolios. A
statistically significant alpha spread of -0.74% suggests that on average, low MAX
portfolios outperform high MAX portfolios. However, this alpha spread is
significantly lower compared with the alpha spread documented by Bali, et al. (2011)
for the U.S. markets as well as the alpha spread documented by Nartea, et al. (2011)
for Korea. The VW alpha spread also suggests a negative MAX effect, but at -0.51%,
it is only marginally significant.
(Insert Table 1 about here)
Table 2 presents the characteristics of the MAX-sorted portfolios. Table 2 indicates
that there appears to be no difference between high and low MAX stocks in terms of
average size, BM ratio, momentum, and illiquidity. However, Table 2 also shows that
high MAX stocks tend to be winners in the previous month, have positively skewed
return distributions and have higher IV than low MAX stocks. These characteristics
all point toward lower returns in the succeeding month based on the short-term
reversal effect (Jegadeesh, 1990; Lehman, 1990), negative IV effect (Ang et al., 2006,
2009), and investor preference for positive skewness (Golec and Tamarkin, 1998;
Mitton and Vorkink, 2007) which implies a negative skewness effect. Therefore, these
variables could potentially explain the negative MAX effect. We test this formally
using dependent bivariate sorts and cross-sectional regressions and report the results
in later sections.
(Insert Table 2 about here)
3.1.2 Bivariate sorting
Now we control for size, BM, momentum, short-term reversal, illiquidity ratio,
skewness and IV to test the robustness of the apparent negative MAX effect for EW
portfolios using a battery of bivariate sorts and report the results in Table 3. Following
Bali, et al. (2011) we focus our attention on the alphas since they control for the
standard set of systematic factors.
(Insert Table 3 about here)
Our results show that the apparent negative MAX effect we document earlier is not
robust. Though the negative MAX effect survives as we control for size, momentum,
illiquidity, and skewness, and IV, the bivariate sorts show that the MAX effect could
potentially be explained short-term return reversals. Panels A, C, E, F and G show
that controlling for size, momentum, illiquidity and skewness, and IV does not
eliminate the significantly negative average alpha spread at -0.73%, -0.77%, -0.51%,
-0.60%, and -0.45% per month respectively. Controlling for BM by averaging across
BM categories, reduces alpha spread to -0.47% and but it is still marginally
significant. However, if we control for return reversal (panel D), the negative MAX
effect disappears, as the average alpha spread becomes insignificant. It is interesting
to note that in panel D the alpha spread in the current month for last month’s winners
(WNR) is positive while the alpha spread for last month’s losers (LSR) is negative
with both being marginally significant. This result suggests a positive MAX effect for
last month’s winners and a negative MAX effect for last month’s losers. This is
consistent with Bali et al. (2011). However, if we control for the past month’s return
(REV) by averaging the alpha spreads within each MAX category, the average alpha
spread turns very close to zero and becomes statistically insignificant, as the alpha
spreads of the WNR and LSR categories cancel each other out. Therefore our results
show that once we control for past returns (REV), the apparent MAX effect also
disappears. The result of this bivariate sort is consistent with our earlier observation
from Table 2 (column 4) that the negative MAX effect could be due to short-term
reversals given that high MAX stocks also tend to be winners (WNRs) while low
MAX stocks tend to be losers (LSRs).2 In sum, the bivariate sorts appear to indicate
that the negative MAX effect could potentially be explained by short-term return
reversals.
3.2 Firm-level cross-sectional regressions
Since dependent bi-variate sorts cannot be used to control for multiple effects
simultaneously we also conduct firm-level Fama-MacBeth regressions. The portfolio
analysis conducted earlier also loses too much information through aggregation. We
estimate the following model and its nested versions:
Ri,t = β0,t-1 + β1,t-1 MAXi, t-1 + β2,t-1 SIZEi, t-1 + β3,t-1 BMi, t-1 + β4,t-1 MOMi, t-1
2 Crucial to this interpretation is evidence of the existence of a short-term reversal effect in the Chinese
stock market which we investigate by sorting stocks into tertiles each month over the sample period
according their returns in the previous month. We then determine the returns of these portfolios in the
current month. The results ?not reported here clearly indicate evidence of a short-term (one-month)
reversal effect. Winner stocks in the previous month have an average EW (VW) monthly return of
10.67% (10.61%), but the return of this portfolio drops to 0.27% (0.05%) in the current month.
Likewise, loser stocks in the previous month have an average EW (VW) monthly return of -8.74% (-
9.01%), but the return of this portfolio rises to 0.93% (0.65%) in the current month.
+ β5,t-1 REVi, t-1 + β6,t-1 ILLIQi, t-1 + β7,t-1 SKEWi, t-1 + β8,t-1 IVi, t-1 + εi,t-1 (2)
Realized stock return in month t, Ri,t , is regressed on one-month lagged values of the
maximum daily return in the previous month (MAX), log of market capitalization
(SIZE), book-to-market ratio (BM), momentum (MOM), short-term reversal (REV),
illiquidity ratio (ILLIQ), and skewness (SKEW), and realized idiosyncratic volatility
(IV). The variables are as defined earlier. Table 4 reports the time series averages of
the slope coefficients over the 116 months from 1993:04-2007:11 for univariate
regressions. The Newey-West t-statistics are given in parenthesis. The univariate
regression shows a statistically significant negative relation between MAX and the
cross-section of one-month ahead stock returns. The results also show a significant
negative size effect, negative IV, negative short-term reversal, a negative skewness
effect, and a marginally significant positive BM effect consistent with expectations.
The rest of the variables -- MOM, ILLIQ have the expected positive coefficients but
they are not statistically significant.
(Insert Table 4 about here)
The result of the bivariate regressions with MAX reported in Table 5 shows that the
MAX effect survives when we control for the variables individually, which is
consistent to some extent with the results of the bi-variate portfolio level analysis
reported in Table 3, except when we control for BM and REV. However, if we control
for all nine variables simultaneously, the negative MAX effect disappears. It is
interesting to note that only the factors related to size and illiquidity are statistically
significant in our multi-variate model and their coefficients are consistent with
expectations. This indicates that the apparent negative MAX effect observed in both
the univariate and bivariate cross-sectional regressions could be explained by both the
size and illiquidity effects. SIZE has a negative coefficient indicative of small firms
outperforming big firms while ILLIQ has a positive coefficient which implies that
investors demand compensation for holding on to relatively illiquid stocks. Therefore,
consistent with our results from portfolio analysis, our firm-level cross-sectional
regression results also indicate the absence of a MAX effect if we restrict the holding
period to one month.
(Insert Table 5 about here)
3.3 Alternative measure of extreme returns
Bali et al. (2011) report that the negative MAX effect is stronger if they use the
average of the five highest daily returns in the previous month to sort portfolios. They
call this measure MAX(5). Hence we also test for evidence of a MAX effect using
MAX(5) to sort portfolios. The results reported in Table 6 shows that though both EW
and VW return spreads are negative, they are insignificant. However, the EW alpha
spreads is negative and marginally significant.
(Insert Table 6 about here)
As an additional test we also perform firm-level Fama-MacBeth cross-sectional
regressions as in equation 2 and report the results in Table 7. Similar to our results
with MAX we also find in the univariate regression, a statistically significant negative
relation between MAX(5) and the cross-section of one-month ahead stock returns.
The result of the bivariate regressions with MAX(5) also reported in Table 7 shows
that if we control for the variables individually, the MAX effect remains significantly
negative. However just as in case with MAX, if we control for all nine variables
simultaneously, though the MAX(5) coefficient remains negative, it is no longer
significant. It is interesting to note that SIZE, REV and ILLIQ are statistically
significant in our multi-variate model with coefficient signs consistent with
expectations. This indicates that apparent negative MAX(5) effect observed in both
the univariate and bivariate cross-sectional regressions could be explained by the size,
short-term reversals and illiquidity. Therefore whether we use MAX or MAX(5) we
find no evidence of a MAX effect in the Chinese stock market when MAX portfolios
are held for one-month.
(Insert Table 7 about here)
3.4 Extended holding periods
As the Chinese stock market is subject to trading price limits3, it could be argued that
stock prices may not have fully adjusted within the one holding month period that we
have so far imposed in our analysis. If the negative MAX effect is driven by investors
who overpay for stocks with extreme positive returns which then results in
underperformance when prices settle to fundamental levels, we expect to find
evidence of a stronger MAX effect when we employ a longer holding period. Hence
we also investigate the presence of the MAX effect for extended holding periods of
three and six months. Table 8 shows the three-month returns and FF-3 alpha of
portfolios sorted on the maximum daily returns in the past month (MAX). Columns
one and three of Table 8 show that similar to the results with a one-month holding
period, the equal- and value-weighted raw return spreads are negative but
insignificant even as we increase the holding period to three months. However unlike
in the case of the one-month holding period, both EW and VW alpha spreads are
3 To reduce stock market’s volatility and protect retail investors, the China Securities Regulatory
Commission imposed trading price limits, which only allows stock prices to move up or down on a
single trading day by a maximum of 10% from the last closing price. The policy artificially makes a
barrier for the stock price to efficiently reflect any news impacting stocks, limiting the efficiency of the
Chinese stock market.
significantly negative. This suggests a negative MAX effect in a three-month holding
period and also suggests at least a three-month lag in the adjustment of prices back to
fundamental levels.
(Insert Table 8 about here)
Even when we control for the various effects with bivariate portfolio sorts in Table 9,
the alpha spreads remain significantly negative.
(Insert Table 9 about here)
The results become more interesting when we employ firm-level Fama-MacBeth
cross-sectional regressions. The results of univariate regressions reported in Table 10
show a significantly negative MAX effect. We also observe significant size, reversal
and skewness effects with expected signs. Interestingly we also observe a highly
significant anomalous negative IV effect consistent with the findings of Nartea and
Wu (2012).
(Insert Table 10 about here)
In Table 11 we report the results when we control for various effects individually with
firm-level bivariate regressions. Table 11 shows that the MAX effect remains
significantly negative in bivariate regressions. In fact it is even stronger when
controlled for the size effect. Of particular interest, we find that the negative MAX
effect appears to neutralise the anomalous negative IV effect. Hence our results are
broadly consistent with Bali et al. (2011) who report that MAX reverses the negative
IV effect documented by Ang et al. (2006, 2009). More importantly, the bottom of
Table 11 shows that the negative MAX effect survives even in a multivariate
regression that controls all variables simultaneously. We note that the anomalous
negative IV effect also does not survive in a multivariate setting. We also observe a
size and momentum effect with the expected signs.
(Insert Table 11 about here)
Finally, we also examine the MAX effect in portfolios with a holding period of six
months. The results are similar to the three-month holding period. Table 12 reports the
six-month returns and FF-3 alpha of portfolios sorted on MAX and shows that the EW
and VW return spreads are negative but insignificant. However, the alpha spreads are
negative and highly significant, indicative of a negative MAX effect. The results of
bivariate portfolio sorts reported in Table 13 shows that none of the control variables
can eliminate the negative alpha spreads hence none can explain the negative MAX
effect. But more importantly, the Fama-Mac Beth regressions in Tables 14 and 15
indicate a significantly negative MAX effect in univariate, bivariate and multivariate
regressions. Our results suggest that the negative MAX effect could last as long as six
months.
(Insert Table 12 about here)
(Insert Table 13 about here)
(Insert Table 14 about here)
(Insert Table 15 about here)
Overall, we find evidence of a negative MAX effect in the Chinese stock market but
only in extended holding periods. This MAX effect is also apparently weaker than in
the U.S. markets. The alpha spreads that we document range from 0.49% to 0.74% per
month compared with around 1% per month in the U.S. We suggest that the
persistence of the negative MAX effect stems from constraints to short-selling in the
Chinese stock markets which limit the opportunity for arbitrage.
Our results are consistent with a market driven by investors with a preference for
lottery-like stocks – those that have exhibited extreme positive returns in the past
month. Indeed apart from anecdotal evidence, there is a body of literature that
suggests risk-seeking behaviour among Chinese investors. Ng and Wu (2006) report
that Chinese investors tend to prefer stocks with large betas and high idiosyncratic
risk based on a comprehensive analysis of 64.22 million trades of 6.8 million
institutional and individual investors in mainland China. Lee and Wong (2009) also
suggest that Chinese investors tend to trade more heavily on riskier stocks based on an
analysis of panel data drawn from the Shanghai stock market. This is consistent with
Fong, Wong, and Yong (2010) who find evidence that mainland Chinese investors are
more speculative and have higher risk appetites than Hong Kong and international
investors. In an earlier study, Ma (1996) also documents evidence of risk-seeking
behavior among mainland Chinese investors by establishing a positive relationship
between share prices and domestic beta risk. In the psychology literature, Raylu and
Oei (2004) report that gambling is an acceptable form of social activity in Chinese
communities while in a related study, Loo, Raylu, and Oei (2008) find widespread
social gambling among Chinese communities as it is a preferred form of
entertainment. The results of both studies suggest a predisposition among Chinese
investors to prefer lottery-like stocks. In sum, our results support the suggestion of
Bali et al. (2011) that investor preference for lottery-like stocks drives the apparent
negative MAX effect.
4. Concluding remarks
Motivated by Bali et al.’s (2011) findings of a significant role of extreme returns in
the U.S. stock markets, we investigate the existence of the same in the Chinese stock
markets. Due to its novelty there is an obvious lack of studies investigating the so-
called MAX effect in other markets. Insofar as the Chinese stock markets have been
characterised as highly speculative, with the extant literature indicating a
predisposition among Chinese investors to prefer riskier stocks, there is reason to
believe that the negative MAX effect would also be evident in this market. However,
unlike Bali et al’s (2011) findings, we find no evidence of a MAX effect if we hold
portfolios for just one month. But we find evidence of a negative relationship between
extreme positive returns (MAX) and stock returns in the Chinese stock market if we
extend the holding period to three and six months. We suggest that this could be due
to price trading limits resulting in a lag in the price adjustment back to fundamental
levels. Our results underscore the importance of country verification, especially in
emerging markets, of apparent anomalies initially discovered in developed stock
markets. However, interpreted along with the strong evidence of risk-seeking
behaviour among Chinese investors, our results support the suggestion of Bali et al.
(2011) that the negative MAX effect is driven by investor preference for stocks with
lottery-like features.
References
Ang, A., R.J. Hodrick, Y. Xing, and X. Zhang (2006) The Cross-section of Volatility
and Expected Returns, Journal of Finance, 51, 259-299.
Ang, A., R.J. Hodrick, Y. Xing, and X. Zhang (2009) High Idiosyncratic Volatility and
Low Returns: International and Further U.S. Evidence, Journal of Financial
Economics, 91,1-23.
Bali, T. G., N. Cakici, and R.F. Whitelaw (2011) Maxing out: Stocks as Lotteries and
the Cross-section of Expected Returns, Journal of Financial Economics, 99, 427-
446.
Barberis, N., and Huang (2008) Stocks as Lotteries: The Implications of Probability
Weighting for Security Prices, American Economic Review, 98, 2066-2100.
Brunnermeier, M. K., C. Gollier, and J.A. Parker (2007) Optimal Beliefs, Asset prices,
and the Preference for Skewed Returns, American Economic Review, 97, 159-165.
Brunnermeier, M.K., and J.A. Parker (2005) Optimal Expectations, American
Economic Review, 95, 1092–1118.
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.
Fong, T., A. Wong, and I. Yong, 2010, Share Price Disparity in Chinese Stock Markets,
Journal of financial transformation, 30, 23-31.
Golec, J., andM. Tamarkin, (1998) Bettors Love Skewness, Not Risk, at the Horse
Track’, Journal of Political Economy, 106, 205-225.
Jegadeesh, N. (1990) Evidence of Predictable Behaviour in Security Returns, Journal
of Finance, 45, 881-898.
Jegadeesh, N., and S. Titman (1993) Return to Buying Winners and Selling Losers:
Implications for Stock Market Efficiency, Journal of Finance, 48, 65-91.
Kumar, A. (2009) Who Gambles in the stock market?, Journal of Finance, 64, 1889-
1933.
Lee, J. and A. Wong, 2009, Impact of financial liberalization on stock market liquidity:
Experience of China, Hong Kong Monetary Authority Working Paper 03/2009.
Lehman, B. (1990) Fads, Martingales, and Market efficiency, Quarterly Journal of
Economics,105, 1-28.
Loo, J.M.Y., N. Raylu, and T.P.S. Oei, 2008, Gambling among the Chinese: A
comprehensive review, Clinical Psychology Review 28 1152-1166.
Ma, X. 1996, Capital controls, market segmentation and stock prices: Evidence from
the Chinese market, Pacific Basin Finance Journal 4219-4239.
Mitton, T., andK. Vorkink, (2007) Equilibrium Underdiversification and the
Preference for Skewness, Review of Financial Studies, 20, 1255-1288.
Nartea, G., Wu, J. and Liu, H.T. 2011, Extreme returns in emerging stock markets:
Evidence of a MAX effect in South Korea, Working paper. Lincoln University.
Ng, L. and F. Wu, 2006, Revealed stock preferences of individual investors: Evidence
from Chinese equity markets, Pacific Basin Finance Journal, 14, 175-192.
reform and QFII and QDII schemes, Paper presented at the European Financial
Management Symposium, Renmin University, Beijing China, December 2010.
Raylu, N. and T.P.S., Oei, 2004, Role of Culture in Gambling and Problem Gambling,
Clinical Psychology Review, 23 1087-1114.
Tversky, A., andD. Kahneman, 1992 ‘Advance in Prospect theory: Cumulative
Representation of Uncertainty’, Journal of Risk and Uncertainty, Vol. 5, pp. 297-
323.
Table 1. Returns on portfolios sorted by MAX
At the beginning of every month we sort stocks into tertiles according to their maximum daily
return (MAX) in the past calendar month. We compute each portfolio’s equal- and value-
weighted raw returns for the current month. We also estimate each portfolio’s alpha (α
coefficient) from the FF3-factor model estimated using the full sample of monthly value- or
equal-weighted returns for each portfolio. The last row shows the difference in monthly
returns and differences in alpha between the high and low MAX portfolios. T-statistics are
reported in parenthesis. We conduct the analysis for the full sample period 1993:08-2007:11.
EW portfolios VW portfolios
Average return FF-3 alpha Average return FF-3 alpha
High MAX 0.0025
(0.3174)
-0.0076
(-3.6006) 0.0002
(0.0285)
-0.0049
(-2.5760)
Medium MAX 0.0093
(1.1932)
-0.0015
(-0.7593) 0.0089
(1.1712)
0.0041
(2.9050)
Low MAX 0.0110
(1.4645)
-0.0002
(-0.0943) 0.0065
(0.9420)
0.0002
(0.1015)
High- Low -0.0085
(-0.7892)
-0.0074
(-2.4331) -0.0063
(-0.6242)
-0.0051
(-1.8009)
Table 2. Characteristics of portfolios sorted by MAX
At the beginning of every month we sort stocks into tertiles according to their maximum daily
return in the past calendar month (MAX). The table reports for each tertile the average of the
monthly averages of various characteristics of the MAX-sorted portfolios over the period
1993:08-2007:11. Size at the end of month t is defined as the log of the firm’s market
capitalization at the end of month t, BM is the firm’s book -to-market ratio six months prior,
i.e. at the end of t-6. Following Jegadeesh and Titman (1993), Momentum at time t is the
stock’s 11-month past return lagged one month, i.e. return from month t-12 to month t-2.
REV in month t is short-term reversal defined as the return on the stock in month t-1,
following Jegadeesh (1990) and Lehmann (1990). Illiquidity is measured following Amihud
(2002) as the ratio of the absolute monthly stock return and its dollar trading volume.
Skewness is the third standardized moment of returns of the past 22 trading days. IV is the
standard deviation of the residuals of the FF3-factor model, using daily data for the previous
22 trading days. The last row is the difference between the high and low MAX portfolio. T-
statistics are reported in parenthesis.
Size BM Momentum REV Illiquidity Skewness IV
High MAX 7.9025
(184.2036)
0.3586
(34.1126)
0.0646
(2.4185)
0.0497
(5.3525)
3.70E-04
(3.3204)
0.5138
(11.909)
0.0218
(44.3045)
Med MAX 7.9153
(192.8663)
0.3591
(35.5905)
0.0570
(2.3469)
0.0006
(0.0733)
9.32E-06
(8.7902)
0.0826
(1.9210) 0.0166
(39.0706)
Low MAX 7.9966
(184586)
0.3630
(33.7229)
0.0455
(1.9802)
-0.0312
(-4.8894)
1.36E-04
(2.2541)
-0.1537
(-3.1905)
0.0141
(38.9329)
High- Low -0.0941
(-1.5596)
-0.0044
(-0.2930)
0.0190
(0.5402)
0.0808
(7.1806)
2.35E-04
(1.856)
0.6676
(10.321)
0.0076
(12.505)
Table 3.Alpha of equal-weighted double-sorted portfolios
At the end of each month over 1993:08-2007:11, stocks are double-sorted 3x3, first by the
control factor (size, BM, momentum, REV, illiquidity, skewness, and IV) into three portfolios
and then within each portfolio we sort stocks again by their maximum daily return in the past
calendar month (MAX). The alpha of each portfolio is presented with t-statistics in
parenthesis. Alpha refers to the FF-3model alpha using the full sample of monthly returns for
each portfolio. To control for a particular factor, we average the alpha within each MAX
category ending up with three portfolios with dispersion in idiosyncratic volatility but
containing all values of the factor being controlled. Size, BM, momentum, REV, illiquidity,
skewness, and IV are as defined in Table 2. LMAX, MMAX , HMAX refer to low, medium,
and high MAX portfolio, respectively; BIG, big size; MED, medium size; SMA, small size;
HBM, MBM , LBM refer to high, medium, low book-to-market, respectively; WNR, winner;
MID, middle; LSR, loser. HIR, MIR, LIR refer to high, medium, low illiquidity ratio,
respectively; HSK, MSK, LSK refer to high, medium, and low skewness. LIV, MIV, HIV refer
to low, medium, and high idiosyncratic volatility, respectively.
Panel A. Double sort on size (market capitalisation) and MAX LMAX MMAX HMAX HMAX-LMAX
BIG -0.0026
(-1.1767)
0.0005
(0.2631)
-0.0060
(-2.6479)
-0.0034
(-1.0683)
MED -0.0019
(-0.8198)
-0.0046
(-1.9500)
-0.0091
(-3.5626)
-0.0072
(-2.074)
SMA 0.0024
(0.9779)
0.0001
(0.0629)
-0.0089
(-3.9272)
-0.0113
(-3.3264)
AVE -0.0007
(-0.5189)
-0.0013
(-2.1050)
-0.0080
(-5.7635)
-0.0073
(-3.7714)
Panel B. Double sort on value (book-to-market) and MAX HBM 0.0011
(0.3112)
-0.0010
(-0.2675)
-0.0103
(-3.0827)
-0.0114
(-2.4060)
MBM -0.0075
(-2.3670)
0.0021
(0.6396)
-0.0028
(-0.8834)
0.0047
(1.0549)
LBM -0.0004
(-0.1107)
-0.0009
(-0.3279)
-0.0077
(-2.1730)
-0.0073
(-1.4748)
AVE -0.0023
(-1.1653)
0.0001
(0.0526)
-0.0069
(-3.6346)
-0.0047
(-1.7100)
Panel C. Double sort on momentum (11/1/1) and MAX WNR 0.0028
(0.9920)
0.0033
(1.1774)
-0.0012
(-0.3795)
-0.0004
(-0.9576)
MID -0.0033
(-1.2111)
-0.0022
(-0.8766)
-0.0103
(-3.5876)
-0.0070
(-1.7700)
LSR -0.0003
(-0.1011)
-0.0080
(-2.7287)
-0.0124
(-3.8524)
-0.0121
(-2.6738)
AVE -0.0003
(-0.1588)
-0.0023
(-1.4547)
-0.0080
(-4.4959)
-0.0077
(-3.1543)
Panel D. Double sort on one-month past return and MAX WNR 0.0908
(26.215)
0.0934
(26.356)
0.1008
(28.170)
0.0100
(1.9920)
MID -0.0089
(-4.4694)
-0.0094
(-4.7834)
-0.0096
(-4.8731)
-0.0007
(-0.2475)
LSR -0.0929
(-27.839)
-0.0923
(-29.007)
-0.1019
(-31.830)
-0.0090
(-1.9600)
AVE -0.0037
(-2.1120)
-0.0028
(-1.6126)
-0.0036
(-2.0520)
0.0001
(0.0407)
Panel E. Double sort on illiquidity and MAX
HIR 0.0009
(0.2790)
0.0050
(1.5746)
-0.0033
(-0.8720)
-0.0042
(-0.8454)
MIR -0.0027
(-1.2024)
-0.0049
(-2.1440)
-0.0102
(-3.8289)
-0.0075
(-2.1150)
LIR -0.0037
(-1.880)
-0.0035
(-1.720)
-0.0074
(-3.9405)
-0.0037
(-1.3413)
AVE -0.0018
(-1.2446)
-0.0011
(-0.7694)
-0.0070
(-4.1519)
-0.0051
(-2.2990)
Panel F. Double sort on skewness and MAX HSK -0.0041
(-1.6234)
-0.0046
(-1.8700)
-0.0097
(-3.9357)
-0.0056
(-1.5839)
MSK 0.0006
(0.2396)
0.0007
(0.3081)
-0.0086
(-3.5049)
-0.0092
(-2.6547)
LSK 2.94E-06
(-0.0012)
-0.0002
(-0.0625)
-0.0033
(-1.2289)
-0.0033
(-0.8976)
AVE -0.0012
(-0.8184)
-0.0014
(-0.9704)
-0.0072
(-4.8555)
-0.0060
(-2.9348)
Panel G. Double sort on IV and MAX HIV -0.0037
(-1.5419)
-0.0084
(-3.3755)
-0.0129
(-4.7170)
-0.0092
(-2.5470)
MIV 0.0007
(0.3147)
-0.0013
(-0.5948)
-0.0021
(-0.8752)
-0.0028
(-0.8600)
LIV 0.0001
(0.0382)
0.0004
(0.1532)
-0.0013
(-0.5245)
-0.0014
(-0.3957)
AVE -0.0010
(-0.6960)
-0.0031
(-3.7200)
-0.0054
(-3.7583)
-0.0045
(-2.2280)
Table 4. Univariate Fama-MacBeth regression results.
Each month from 1993:08 to 2007:11 we run a firm-level univariate Fama-MacBeth cross-
sectional regression of the return on that month with one-month lagged values of the MAX
and other control variables. Each row reports the time-series averages of the slope coefficients
and their associated t-statistics. MAX and the other control variables are defined in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.016
(1.96)
-0.157
(-3.53)
0.044
(2.48)
-0.005
(-2.61)
-0.003
(-0.36)
0.024
(1.78)
0.009
(1.12)
0.007
(1.40)
0.007
(0.85)
-0.040
(-2.92)
0.005
(0.65)
0.004
(1.66)
0.008
(0.95)
-0.004
(-3.32)
0.014
(1.69)
-0.385
(-2.59)
Table 5. Bivariate and multi-variate Fama-MacBeth regression results with
MAX.
Each month from 1993:08 to 2007:11 we run a firm-level bi-variate and multi-variate Fama-
MacBeth cross-sectional regression of the return on that month with one-month lagged values
of the MAX and other control variables. Each row reports the time-series averages of the
slope coefficients and their associated t-statistics. MAX and the other control variables are
defined in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.060
(3.44)
-0.178
(-4.24)
-0.006
(-3.03)
0.010
(1.13)
-0.207
(-3.60)
0.018
(1.60)
0.020
(2.33)
-0.183
(-4.98)
0.007
(1.43)
0.013
(1.60)
-0.167
(-3.60)
-0.018
(-1.17)
0.016
(2.04)
-0.168
(-4.51)
0.005
(1.94)
0.0144
(1.70)
-0.1191
(-2.37)
-0.0017
(-1.11)
0.017
(2.04)
-0.114
(-2.48)
-0.184
(-1.11)
0.041
(1.66)
-0.143
(-1.32)
-0.005
(-2.20)
0.033
(1.15)
0.009
(1.57)
-0.019
(-0.88)
0.006
(3.13)
-0.003
(-0.72)
-0.520
(-0.76)
Table 6. Returns on portfolios sorted by MAX(5)
At the beginning of every month we sort stocks into tertiles according to the average of the
five highest daily returns (MAX5) in the past calendar month. We compute each portfolio’s
equal- and value-weighted raw returns for the current month. We also estimate each
portfolio’s alpha (α coefficient) from the FF3-factor model estimated using the full sample of
monthly value- or equal-weighted returns for each portfolio. The last row shows the
difference in monthly returns and differences in alpha between the high and low MAX
portfolios. T-statistics are reported in parenthesis. We conduct the analysis for the full sample
period 1993:08-2007:11.
EW portfolios VW portfolios
Average return FF-3 alpha Average return FF-3 alpha
High IV 0.0020
(0.2505)
-0.0063
(-2.7024) -0.0007
(-0.0859)
-0.0033
(-1.5381)
Medium IV 0.0083
(1.0450)
-0.0023
(-1.1915) 0.0069
(0.9018)
0.0016
(0.9375)
Low IV 0.0092
(1.2564)
-0.0001
(-0.0382) 0.0053
(0.7862)
0.0006
(0.2501)
High- Low -0.0072
(-0.6624)
-0.0062
(-1.9095) -0.0060
(-0.5775)
-0.0039
(-1.2535)
Table 7. Fama-MacBeth regression results with MAX (5).
Each month from 1993:08 to 2007:11 we run a firm-level univariate Fama-MacBeth cross-
sectional regression of the return on that month with one-month lagged values of the MAX (5)
and bivariate and multivariate cross-sectional regression of the return on that month with one-
month lagged values of the MAX (5) with other control variables. Each row reports the time-
series averages of the slope coefficients and their associated t-statistics. MAX and the other
control variables are defined in Table 2.
Intercept MAX (5) SIZE BM MOM REV ILLIQ SKEW IV
0.0183
(2.18)
-0.2933
(-3.32)
0.0641
(3.55)
-0.3301
(-3.90)
-0.006
(-3.08)
0.0096
(0.96)
-0.2845
(-2.13)
0.018
(1.47)
0.0189
(2.13)
-0.2661
(-2.91)
0.0062
(1.30)
0.0159
(1.85)
-0.3600
(-2.95)
0.0012
(0.06)
0.0161
(2.04)
-0.2721
(-3.14)
0.0004
(1.82)
0.0179
(2.00)
-0.2329
(-2.38)
-0.0032
(-2.45)
0.0189
(2.19)
-0.2579
(-2.44)
-0.0479
(-0.27)
0.236
(1.27)
0.029
(1.09)
-0.013
(-1.70)
-0.329
(-0.96)
0.059
(1.16)
-0.102
(-1.72)
0.005
(1.90)
-0.004
(-1.64)
-0.702
(-1.18)
Table 8. Returns on portfolios sorted by MAX, three-month holding period.
At the beginning of every month we sort stocks into tertiles according to their maximum daily
return in the past calendar month (MAX). We compute each portfolio’s equal- and value-
weighted raw returns for a three-month holding period. We also estimate each portfolio’s
alpha (α coefficient) from the FF3-factor model estimated using the full sample of value- or
equal-weighted returns for each portfolio. The last row shows the difference in monthly
returns and differences in alpha between the high and low MAX portfolios. T-statistics are
reported in parenthesis. We conduct the analysis for the full sample period 1993:08-2007:11.
EW portfolios VW portfolios
Average return FF-3 alpha Average return FF-3 alpha
High MAX 0.0127
(0.9163)
-0.0173
(-5.0302) 0.0053
(0.4051)
-0.0132
(-4.1172)
Medium MAX 0.0256
(1.8465)
-0.0049
(-1.6545) 0.0210
(1.5401)
0.0044
(1.5804)
Low MAX 0.0302
(2.2396)
0.0006
(0.1474) 0.0226
(1.7980)
0.0049
(1.4245)
High- Low -0.0175
(-0.9096)
-0.0179
(-3.4596) -0.0173
(-0.9541)
-0.0181
(-3.8167)
Table 9.Alpha of double sorted portfolios, three-month holding period.
At the end of each month over 1993:08-2007:11, stocks are double-sorted 3x3, first by the control
factor (size, BM, momentum, REV, illiquidity, skewness, and IV) into three portfolios and then within
each portfolio we sort stocks again by their maximum daily return in the past calendar month (MAX).
The alpha of each portfolio is presented with t-statistics in parenthesis. Alpha refers to the FF-3model
alpha using the full sample of 3-month returns for each portfolio. To control for a particular factor, we
average the alpha within each MAX category ending up with three portfolios with dispersion in
idiosyncratic volatility but containing all values of the factor being controlled. Size, BM, momentum,
REV, illiquidity, skewness, and IV are as defined in Table 2. LMAX, MMAX , HMAX refer to low,
medium, and high MAX portfolio, respectively; BIG, big size; MED, medium size; SMA, small size;
HBM, MBM , LBM refer to high, medium, low book-to-market, respectively; WNR, winner; MID,
middle; LSR, loser. HIR, MIR , LIR refer to high, medium, low-illiquidity ratio, HSK, MSK, LSK refer
to high, medium, and low skewness. LIV, MIV, HIV refer to high idiosyncratic volatility, respectively.
Panel A. Double sort on size (market capitalisation) and MAX
Equal-weighted Value -weighted
LMAX MMAX HMAX HMAX-
LMAX
LMAX MMAX HMAX HMAX-
LMAX
BIG -0.0027
(-0.6509)
-0.0004
(-0.1179)
-0.0165
(-4.4103)
-0.0138
(-2.4686)
0.0035
(0.9448)
0.0071
(2.1445)
-0.0106
(-2.8920)
-0.0141
(-2.6585)
MED -0.0017
(-0.3918)
-0.0076
(-2.0005)
-0.0170
(-4.0553)
-0.0153
(-2.5759)
-0.0015
(-0.3535)
-0.0079
(-2.0892)
-0.0168
(-3.9492)
-0.0153
(-2.5454)
SMA 0.0070
(1.6062)
-0.0051
(-1.6006)
-0.0193
(-4.9042)
-0.0263
(-4.4731)
0.0050
(1.1531)
-0.0060
(-1.8751)
-0.0200
(-4.9744)
-0.0250
(-4.2042)
AVE 0.0009
(0.3544)
-0.0044
(-2.2371)
-0.0176
(-7.6780)
-0.0185
(-5.5098)
0.0023
(0.9760)
-0.0023
(-1.1402)
-0.0158
(-6.8288)
-0.0181
(-5.4504)
Panel B. Double sort on value (book-to-market) and MAX
HBM -0.0079
(-1.3800)
-0.0103
(-1.2279)
-0.0228
(-3.9541)
-0.0149
(-1.8323)
-0.0011
(-0.1593)
-0.0046
(-0.4729)
-0.0154
(-2.5898)
-0.0143
(-1.6018)
MBM -0.0046
(-0.7520)
-0.0035
(-0.6332)
-0.0105
(-1.7939)
-0.0059
(-0.7009)
0.0037
(-0.5359)
-0.0031
(-0.4754)
-0.0092
(-1.3666)
-0.0129
(-1.3316)
LBM -0.0031
(-0.5658)
-0.0089
(-1.9737)
-0.0285
(-4.4054)
-0.0254
(-3.0058)
-0.0041
(-0.5858)
-0.0027
(-0.4462)
-0.0242
(-3.0662)
-0.0201
(-1.9043)
AVE -0.0052
(-1.5690)
-0.0076
(-2.0537)
-0.0206
(-5.9050)
-0.0154
(-3.2004)
-0.0005
(-0.1261)
-0.0035
(-0.7997)
-0.0163
(-4.0743)
-0.0158
(-2.8020)
Panel C. Double sort on momentum (11/1/1) and MAX
WNR 0.0061
(1.2003)
0.0010
(0.1778)
-0.0105
(-1.9678)
-0.0166
(-2.235*)
0.0134
(2.5923)
0.0133
(2.317*)
-0.0016
(-0.2844)
-0.0150
(-1.9635)
MID -0.0037
(-0.7560)
-0.0098
(-2.023*)
-0.0147
(-2.7338)
-0.0110
(-1.5085)
0.0011
(0.2222)
-0.0122
(-2.261*)
-0.0128
(-2.2647)
-0.0139
(-1.8452)
LSR -0.0044
(-0.8376)
-0.0133
(-2.6921)
-0.0300
(-5.5702)
-0.0256
(-3.3834)
-0.0044
(-0.8319)
-0.0097
(-1.5979)
-0.0330
(-5.5746)
-0.0286
(-3.6745)
AVE -0.0007
(-0.2263)
-0.0074
(-2.465*)
-0.0184
(-5.9018)
-0.0177
(-4.1342)
0.0034
(1.1213)
-0.0029
(-0.8600)
-0.0158
(-4.8578)
-0.0192
(-4.3299)
Panel D. Double sort on one-month past return and MAX
WNR -0.0094
(-1.9872)
-0.0134
(-3.3920)
-0.0299
(-6.1979)
-0.0205
(-3.0516)
0.0038
(0.7247)
-0.0025
(-0.5108)
-0.0220
(-4.0101)
-0.0258
(-3.4086)
MID -0.0010
(-0.2328)
-0.0062
(-1.7015)
-0.0127
(-3.3421)
-0.0117
(-2.0136)
0.0016
(0.3151)
0.0013
(0.3266)
-0.0095
(-2.0614)
-0.0111
(-1.6162)
LSR 0.0048
(0.9839)
0.0081
(1.7432)
-0.0052
(-1.1859)
-0.0100
(-1.5185)
0.0056
(1.0727)
0.0111
(2.262*)
-0.0027
(-0.5564)
-0.0083
(-1.1499)
AVE -0.0019
(-0.6921)
-0.0038
(-1.6221)
-0.0159
(-6.3402)
-0.0141
(-3.8159)
0.0037
(1.2212)
0.0033
(1.2203)
-0.0114
(-3.9381)
-0.0151
(-3.6124)
Panel E. Double sort on Illiquidity ratio and MAX
HIR -0.0036
(-0.6943)
-0.0021
(-0.4720)
-0.0131
(-2.6073)
-0.0095
(-1.3169)
-0.0085
(-1.4494)
0.0017
(0.3026)
-0.0143
(-2.5601)
-0.0058
(-0.7149)
MIR 0.0034
(0.8919)
-0.0071
(-1.9863)
-0.0202
(-5.0105)
-0.0236
(-4.2775)
0.0066
(1.7169)
-0.0003
(-0.0685)
-0.0170
(-3.5308)
-0.0236
(-3.8159)
LIR 0.0016
(0.3149)
-0.0048
(-2.0166)
-0.0166
(-3.4636)
-0.0182
(-2.6258)
0.0101
(2.1226)
0.0040
(0.8095)
-0.0089
(-2.0206)
-0.0190
(-2.9512)
AVE 0.0005
(0.1717)
-0.0048
(-2.0166)
-0.0166
(-6.2356)
-0.0171
(-4.4903)
0.0027
(0.9736)
0.0018
(0.6397)
-0.0134
(-4.6808)
-0.0161
(-4.0236)
Panel F. Double sort on skewness and MAX
HSK -0.0079
(-1.8028)
-0.0116
(-3.0869)
-0.0252
(-6.0974)
-0.0173
(-2.8765)
-0.0044
(-1.0138)
-0.0046
(-1.1116)
-0.0193
(-3.9275)
-0.0149
(-2.2625)
MSK 0.0007
(0.1685)
-0.0038
(-1.0791)
-0.0179
(-4.2482)
-0.0186
(-3.1315)
0.0113
(2.1946)
0.0022
(0.4936)
-0.0139
(-3.0779)
-0.0252
(-3.6645)
LSK 0.0054
(1.2019)
0.0019
(0.4377)
-0.0043
(-1.0222)
-0.0097
(-1.5758)
0.0034
(0.7518)
0.0048
(1.0158)
0.0004
(0.0773)
-0.0030
(-0.4229)
AVE -0.0006
(-0.2379)
-0.0045
(-2.0450)
-0.0158
(-6.5675)
-0.0152
(-4.3609)
0.0034
(1.2531)
0.0008
(0.3121)
-0.0109
(-3.8280)
-0.0144
(-3.6299)
Panel G. Double sort on IV and MAX
HIV -0.0030
(-0.7478)
-0.0133
(-3.2537)
-0.0296
(-6.2075)
-0.0266
(-4.2572)
-0.0043
(-1.0657)
-0.0119
(-2.6674)
-0.0257
(-4.7018)
-0.0214
(-3.1467)
MIV -0.0028
(-0.6757)
-0.0053
(-1.5755)
-0.0104
(-2.6871)
-0.0076
(-1.3431)
0.0006
(0.1400)
0.0004
(0.0926)
-0.0036
(-0.8244)
-0.0042
(-0.6905)
LIV 0.0010
(0.1905)
0.0016
(0.3980)
-0.0047
(-1.0911)
-0.0057
(-0.8447)
0.0064
(1.1826)
0.0084
(1.8262)
0.0079
(1.6649)
0.0015
(0.2095)
AVE -0.0016
(-0.6205)
-0.0057
(-2.5752)
-0.0149
(-5.9342)
-0.0133
(-3.6952)
0.0009
(0.3407)
-0.0010
(-0.4091)
-0.0071
(-2.5273)
-0.0080
(-2.0780)
Table 10. Univariate Fama-MacBeth regression results, three-month holding
period.
Each month from 1993:08 to 2007:11 we run a firm-level univariate Fama-MacBeth cross-
sectional regression of the three-month return (t+1 to t+3) with one-month lagged values (t) of
the MAX and other control variables. Each row reports the time-series averages of the slope
coefficients and their associated t-statistics. MAX and the other control variables are defined
in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.043
(2.01)
-0.384
(-4.51)
0.116
(2.56)
-0.013
(-2.58)
-0.008
(-0.40)
0.090
(2.14)
0.029
(1.34)
0.014
(1.16)
0.020
(0.96)
-0.066
(-2.63)
0.022
(1.04)
0.002
(0.71)
0.025
(1.17)
-0.009
(-4.22)
0.041
(1.84)
-0.972
(-2.79)
Table 11. Bivariate and multi-variate Fama-MacBeth regression results with
MAX, three-month holding period.
Each month from 1993:08 to 2007:11 we run a firm-level bivariate and multivariate Fama-
MacBeth cross-sectional regression of the three-month return (t+1 to t+3) with one-month
lagged values (t) of the MAX and other control variables. Each row reports the time-series
averages of the slope coefficients and their associated t-statistics. MAX and the other control
variables are defined in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.144
(3.27)
-0.397
(-4.97)
-0.014
(-2.82)
0.008
(0.31)
-0.327
(-2.60)
0.084
(1.97)
0.045
(2.02)
-0.335
(-4.41)
0.015
(1.29)
0.037
(1.73)
-0.399
(-4.60)
-0.018
(-0.59)
0.043
(1.99)
-0.367
(-4.82)
0.002
(0.65)
0.040
(1.83)
-0.312
(-3.00)
-0.005
(-1.67)
0.048
(2.18)
-0.275
(-3.69)
-0.504
(-1.41)
0.202
(2.92)
-0.423
(-2.37)
-0.019
(-2.48)
0.010
(0.23)
0.029
(2.17)
0.055
(0.84)
0.002
(0.73)
-0.001
(-0.10)
-0.888
(-1.84)
Table 12. Returns on portfolios sorted by MAX, six-month holding period.
At the beginning of every month we sort stocks into tertiles according to their maximum daily
return in the past calendar month (MAX). We compute each portfolio’s equal- and value-
weighted raw returns for a six-month holding period. We also estimate each portfolio’s alpha
(α coefficient) from the FF3-factor model estimated using the full sample of value- or equal-
weighted returns for each portfolio. The last row shows the difference in monthly returns and
differences in alpha between the high and low MAX portfolios. T-statistics are reported in
parenthesis. We conduct the analysis for the full sample period 1993:08-2007:11.
EW portfolios VW portfolios
Average return FF-3 alpha Average return FF-3 alpha
High MAX 0.0270
(1.3963)
-0.0286
(-6.0482) 0.0164
(0.8947)
-0.0239
(-5.3212)
Medium MAX 0.0501
(2.4896)
-0.0061
(-1.5079) 0.0392
(1.9980)
0.0108
(2.4499)
Low MAX 0.0578
(2.8446)
0.0010
(0.2029) 0.0433
(2.2453)
0.0082
(1.6078)
High- Low -0.0308
(-1.0996)
-0.0296
(-4.2230) -0.0269
(-1.0129)
-0.0321
(-4.7196)
Table 13.Alpha of double sorted portfolios, six-month holding period.
At the end of each month over 1993:08-2007:11, stocks are double-sorted 3x3, first by the control
factor (size, BM, momentum, REV, illiquidity, skewness, and IV) into three portfolios and then within
each portfolio we sort stocks again by the maximum daily return in the past calendar month (MAX).
The alpha of each portfolio is presented with t-statistics in parenthesis. Alpha refers to the FF-3model
alpha using the full sample of six-month returns for each portfolio. To control for a particular factor,
we average the alpha within each MAX category ending up with three portfolios with dispersion in
idiosyncratic volatility but containing all values of the factor being controlled. Size, BM, momentum,
REV, illiquidity, skewness, and IV are as defined in Table 2. LMAX, MMAX , HMAX refer to low,
medium, and high MAX portfolio, respectively; BIG, big size; MED, medium size; SMA, small size;
HBM, MBM , LBM refer to high, medium, low book-to-market, respectively; WNR, winner; MID,
middle; LSR, loser. HIR, MIR, LIR refer to high, medium, low illiquidity ratio, respectively; HSK,
MSK, LSK refer to high, medium, and low skewness. LIV, MIV, HIV refer to high idiosyncratic
volatility, respectively.
Panel A. Double sort on size (market capitalisation) and MAX
Equal-weighted Value -weighted
LMAX MMAX HMAX HMAX-
LMAX
LMAX MMAX HMAX HMAX-
LMAX
BIG -0.0062
(-1.1639)
0.0015
(0.3109)
-0.0265
(-4.7534)
-0.0203
(-2.6328)
0.0064
(1.0957)
0.0120
(2.2136)
-0.0151
(-2.7718)
-0.0215
(-2.6881)
MED -0.0046
(-0.8198)
-0.0144
(-2.6579)
-0.0265
(-4.7875)
-0.0219
(-2.7649)
-0.0046
(-0.8198)
-0.0141
(-2.5804)
-0.0266
(-4.8137)
-0.0220
(-2.7775)
SMA 0.0131
(2.1643)
-0.0081
(-1.8153)
-0.0276
(-5.3538)
-0.0407
(-5.0776)
0.0131
(2.1643)
-0.0081
(-1.8153)
-0.0276
(-5.3538)
-0.0407
(-5.0776)
AVE 0.0008
(0.2326)
-0.0070
(-2.4825)
-0.0269
(-8.5604)
-0.0276
(-6.0713)
0.0050
(1.4575)
-0.0034
(-1.1493)
-0.0231
(-7.4533)
-0.0281
(-6.0931)
Panel B. Double sort on value (book-to-market) and MAX
HBM -0.0095
(-1.1638)
-0.0208
(-1.9626)
-0.0304
(-3.6850)
-0.0209
(-1.7913)
0.0059
(0.6212)
-0.0063
(-0.5084)
-0.0230
(-2.6559)
-0.0289
(-2.2564)
MBM 0.0169
(1.9036)
0.0029
(0.3317)
-0.0032
(-0.3374)
-0.0201
(-1.5615)
0.0242
(2.4862)
0.0075
(0.7490)
-0.0022
(-0.2033)
-0.0264
(-1.8280)
LBM -0.0061
(-0.8170)
-0.0132
(-2.0666)
-0.0464
(-5.0724)
-0.0403
(-3.3952)
-0.0180
(-1.7262)
-0.0068
(-0.8766)
-0.0392
(-3.5261)
-0.0212
(-1.3937)
AVE 0.0004
(0.0913)
-0.0104
(-2.0551)
-0.0267
(-5.1638)
-0.0271
(-3.8639)
0.0040
(0.7098)
-0.0019
(-0.3170)
-0.0215
(-3.6378)
-0.0255
(-3.1127)
Panel C. Double sort on momentum (11/1/1) and MAX
WNR -0.0084
(-1.2809)
-0.0185
(-3.1936)
-0.0385
(-6.1144)
-0.0301
(-3.2989)
0.0104
(1.2708)
0.0018
(0.2495)
-0.0338
(-4.8751)
-0.0442
(-4.1244)
MID -0.0025
(-0.4358)
-0.0054
(-1.1508)
-0.0171
(-3.1005)
-0.0146
(-1.8392)
0.0065
(0.9692)
0.0072
(1.3423)
-0.0109
(-1.4147)
-0.0174
(-1.6938)
LSR 0.0027
(0.4149)
0.0038
(0.5988)
-0.0165
(-2.7590)
-0.0192
(-2.1716)
0.0038
(0.4651)
0.0087
(1.2279)
-0.0157
(-2.2701)
-0.0195
(-1.8255)
AVE -0.0027
(-0.7503)
-0.0067
(-2.044*)
-0.0240
(-7.0050)
-0.0213
(-4.2563)
0.0069
(1.5398)
0.0059
(1.5610)
-0.0201
(-4.8591)
-0.0270
(-4.4294)
Panel D. Double sort on one-month past return and MAX
WNR -0.0084
(-1.2809)
-0.0186
(-3.1936)
-0.0385
(-6.1144)
-0.0301
(-3.2989)
0.0104
(1.2708)
0.0018
(0.2495)
-0.0338
(-4.8751)
-0.0442
(-4.1244)
MID -0.0025
(-0.4358)
-0.0054
(-1.1508)
-0.0171
(-3.1005)
-0.0146
(-1.8354)
0.0065
(0.9692)
0.0072
(1.3423)
-0.0109
(-1.4147)
-0.0174
(-1.6938)
LSR 0.0027
(0.4149)
0.0038
(0.5988)
-0.0165
(-2.7590)
-0.0192
(-2.1715)
0.0038
(0.4651)
0.0087
(1.2279)
-0.0157
(-2.2706)
-0.0195
(-1.8289)
AVE -0.0027
(-0.7503)
-0.0067
(-2.0442)
-0.0240
(-7.0050)
-0.0213
(-4.2563)
0.0069
(1.5398)
0.0059
(1.5610)
-0.0201
(-4.8591)
-0.0270
(-4.4294)
Panel E. Double sort on Illiquidity ratio and MAX
HIR -0.0155
(-2.2983)
-0.0081
(-1.2582)
-0.0331
(-4.6521)
-0.0176
(-1.8029)
-0.0267
(-3.7152)
-0.0098
(-1.3395)
-0.0414
(-5.3603)
-0.0147
(-1.3944)
MIR 0.0027
(0.5314)
-0.0105
(-2.2518)
-0.0342
(-6.8792)
-0.0369
(-5.1151)
0.0036
(0.6158)
-0.0085
(-1.6207)
-0.0388
(-6.3623)
-0.0424
(-4.9962)
LIR 0.0143
(2.0230)
0.0042
(0.6806)
-0.0204
(-3.1650)
-0.0347
(-3.6048)
0.0251
(3.0089)
0.0225
(3.1920)
-0.0103
(-1.5677)
-0.0354
(-3.3383)
AVE 0.0005
(0.1356)
-0.0048
(-1.4360)
-0.0292
(-8.8051)
-0.0297
(-5.7578)
0.0007
(0.1604)
0.0014
(0.3659)
-0.0302
(-7.6470)
-0.0308
(-5.3801)
Panel F. Double sort on Skewness and MAX
HSK -0.0087
(-1.3732)
-0.0130
(-2.3826)
-0.0341
(-5.7811)
-0.0254
(-2.9428)
0.0023
(0.3533)
-0.0038
(-0.5400)
-0.0317
(-4.4631)
-0.0340
(-3.5074)
MSK 0.0018
(0.3126)
-0.0014
(-0.2934)
-0.0279
(-4.9577)
-0.0297
(-3.7169)
0.0124
(1.5824)
0.0118
(1.7828)
-0.0220
(-3.5720)
-0.0344
(-3.4525)
LSK 0.0058
(1.0094)
-0.0052
(-0.8953)
-0.0180
(-3.0195)
-0.0238
(-2.8520)
0.0055
(0.7853)
-0.0066
(-0.9446)
-0.0054
(-0.6895)
-0.0109
(-1.0262)
AVE -0.0004
(-0.1069)
-0.0065
(-2.0882)
-0.0267
(-7.9146)
-0.0263
(-5.4709)
0.0067
(1.6235)
0.0005
(0.1182)
-0.0197
(-4.8054)
-0.0264
(-4.5328)
Panel G. Double sort on IV and MAX
HIV -0.0111
(-2.0333)
-0.0252
(-4.7049)
-0.0448
(-7.5052)
-0.0337
(-4.1403)
-0.0129
(-2.3860)
-0.0265
(-4.7736)
-0.0436
(-6.3163)
-0.0307
(-3.5038)
MIV -0.0016
(-0.2853)
-0.0083
(-1.8176)
-0.0190
(-3.4453)
-0.0174
(-2.1967)
0.0013
(0.2025)
0.0045
(0.7400)
-0.0076
(-1.2215)
-0.0089
(-0.9750)
LIV 0.0038
(0.5439)
0.0073
(1.3002)
-0.0017
(-0.2778)
-0.0055
(-0.5972)
0.0137
(1.7035)
0.0232
(3.2014)
0.0192
(2.8784)
0.0055
(0.5271)
AVE -0.0030
(-0.8472)
-0.0087
(-2.8990)
-0.0218
(-6.4395)
-0.0189
(-3.8708)
0.0007
(1.7035)
0.0004
(0.1099)
-0.0107
(-2.7965)
-0.0114
(-2.0792)
HIV -0.0111
(-2.0333)
-0.0252
(-4.7049)
-0.0448
(-7.5052)
-0.0337
(-4.1403)
-0.0129
(-2.3860)
-0.0265
(-4.7736)
-0.0436
(-6.3163)
-0.0307
(-3.5038)
Table 14. Univariate Fama-MacBeth regression results, six-month holding
period.
Each month from 1993:08 to 2007:11 we run a firm-level univariate Fama-MacBeth cross-
sectional regression of the six-month return (t+1 to t+6) with one-month lagged values (t) of
the MAX and other control variables. Each row reports the time-series averages of the slope
coefficients and their associated t-statistics. MAX and the other control variables are defined
in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.082
(2.19)
-0.669
(-6.08)
0.213
(2.94)
-0.023
(-2.98)
-0.013
(-0.36)
0.140
(1.90)
0.055
(1.55)
0.016
(0.82)
0.047
(1.31)
-0.046
(-1.35)
0.045
(1.25)
0.002
(0.50)
0.049
(1.36)
-0.012
(-4.67)
0.077
(2.07)
-1.878
(-4.07)
Table 15. Bivariate and multi-variate Fama-MacBeth regression results with
MAX, six-month holding period.
Each month from 1993:08 to 2007:11 we run a firm-level bivariate and multi-variate Fama-
MacBeth cross-sectional regression of the six-month return (t+1 to t+6) with one-month
lagged values (t) of the MAX and other control variables. Each row reports the time-series
averages of the slope coefficients and their associated t-statistics. MAX and the other control
variables are defined in Table 2.
Intercept MAX SIZE BM MOM REV ILLIQ SKEW IV
0.264
(3.66)
-0.687
(-6.56)
-0.025
(-3.21)
0.009
(0.19)
-0.493
(-2.78)
0.142
(1.85)
0.083
(2.16)
-0.535
(-4.70)
0.018
(0.92)
0.083
(2.23)
-0.742
(-5.62)
0.044
(1.00)
0.081
(2.12)
-0.616
(-6.44)
0.002
(0.45)
0.079
(2.09)
-0.627
(-4.12)
-0.002
(-0.53)
0.092
(2.43)
-0.440
(-4.05)
-1.142
(-2.22)
0.313
(3.35)
-0.748
(-2.54)
-0.030
(-2.68)
0.059
(1.12)
0.043
(2.06)
0.143
(1.49)
0.005
(0.65)
0.005
(0.63)
-0.030
(-1.51)