Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

Rev Quant Finan Acc

DOI 10.1007/s11156-014-0469-x

ORIGINAL RESEARCH

Reexamining momentum profits: Underreaction


or overreaction to firm-specific information?

Jungshik Hur • Vivek Singh

Ó Springer Science+Business Media New York 2014

Abstract We design a new measure and find that the predictability of past returns on future
returns increases as stocks respond with delay to firm-specific information. Our results suggest
that momentum is caused by both investors’ underreaction and overreaction to information.
However, underreaction to information seems to be the primary cause, particularly during the
more recent period. Our findings are robust for recent explanations of momentum profits and
alternative methods for computing our measure. We also find that stocks respond with delay to
firm-specific information, partly due to certain firm characteristics, and partly because they
escape investor attention due to their low visibility. Our paper extends and refines Jegadeesh
and Titman’s (J Financ 56(2):699–720, 2001) finding that momentum profits are consistent
with behavioral models’ predictions regarding investors’ overreaction.

Keywords Momentum  Cointegration  Underreaction  Overreaction

JEL Classification G12  G14  G20

1 Introduction

Stock price momentum is a robust anomaly present in the stocks of all market capitalizations
and has continued unabated since its discovery by Jegadeesh and Titman (1993). Using more
recent time periods, Jegadeesh and Titman (2001) confirm that momentum strategy stays
profitable out-of-sample. Haugen and Baker (1996), Rouwenhorst (1998), and Griffin et al.
(2003) find that momentum strategies are reliably profitable in a number of international

J. Hur
Department of Economics and Finance, College of Business, Louisiana Tech University,
Ruston, LA, USA
e-mail: jhur@latech.edu

V. Singh (&)
Department of Accounting and Finance, College of Business, University of Michigan, Dearborn,
Dearborn, MI, USA
e-mail: vatsmala@umich.edu

123
J. Hur, V. Singh

markets. In an interesting paper, Lesmond et al. (2004) contend that momentum profits are
actually negligible when attempting to exploit this anomaly due to the high trading costs.
Similarly in another paper Wu (2011) finds medium-term reversal in Chinese stock markets.
Overall most of these studies authenticate that the momentum phenomenon is not the product
of data-mining or data-snooping bias. Momentum is also shown to be present in commodities
futures [see Miffre and Rallis (2007)] and foreign exchange markets [see Serban (2010)].
However, the debate regarding the origin and interpretation of evidence of it remains una-
bated. Conrad and Kaul (1998) were the first to argue that momentum could be due to the
higher risk of momentum portfolios. Subsequently many others, among them Berk et al.
(1999), Chordia and Shivakumar (2002), Cochrane (1996), Johnson (2002), Antoniou et al.
(2007), Liu and Zhang (2008), Xiafei et al. (2008), and Wang et al. (2012), have offered
rationales and/or have provided evidence consistent with momentum profits as a compen-
sation for risk. At the same time, Fama and French (1996), Jegadeesh and Titman (2001),
Grundy and Martin (2001), Asem (2009), Wang and Wu (2011), Yeh and Li (2011), Jiang
et al. (2012), and Yu (2012) demonstrate that risk-based explanations cannot account for all of
the profits of momentum strategies. Jegadeesh and Titman (2005) review the momentum
literature and conclude that the ‘‘momentum effect is quite pervasive and very unlikely to be
explained by risk.’’ Fama (1998) and Barberis and Thaler (2003) mirror the general agree-
ment in the literature that momentum is worthy of behavioral rationalization.
While interpreting the main findings of their study, Jegadeesh and Titman (1993) comment
that the ‘‘evidence is consistent with delayed price reactions to firm-specific information.’’
They further add that ‘‘transactions by investors who buy past winners and sell past losers
move prices away from their long-run values temporarily and thereby cause prices to over-
react.’’ In fact, researchers have developed several behavioral models the predictions of
which are in accordance with the conjectures of Jegadeesh and Titman (1993). Respective
behavioral models can be divided into two groups, depending on whether initial investor
behavior should be characterized as one of overreaction or underreaction.
In Barberis et al. (1998) model, investors initially exhibit underreaction to information
followed by a delayed overreaction, due to the model’s conservatism bias. In Hong and
Stein’s (1999) model, investors display similar behavior due to a gradual diffusion of
private information. In contrast, in Daniel et al. (1998) model, investors suffer from a self-
attribution bias, as a consequence of which they overreact to information and continue to
overreact for a period of time. A similar overreaction is the outcome of the positive
feedback trading by rational speculators found in DeLong et al. (1990) model. Summing up
these models leads us to predict an initial underreaction and an overreaction to new
information, one that eventually shows up in the form of the continuation of stock returns.
More importantly, the common theme in these models of a delayed overreaction also leads
to the central prediction of a reversal of momentum profit over the long-run, as docu-
mented by Jegadeesh and Titman (2001).
Whereas the above arguments illustrate that momentum may arise as a consequence of
investors’ overreaction or underreaction, most recent studies explicitly or implicitly cor-
roborate the underreaction hypothesis. For example, Grinblatt and Han (2005) and Frazzini
(2006) use capital gains overhang as a proxy for investors’ underreaction to news, and
show that it has explanatory power with respect to momentum. Lee and Swaminathan
(2000) find that stocks with higher trading volume exhibit stronger momentum. Verardo
(2009) finds that stocks with a high dispersion in analysts’ earnings forecasts show greater
momentum. Both studies argue that a high turnover and high dispersion in analyst earnings
forecasts are associated with disagreement amongst investors, which in turn leads prices to
underreact to information.

123
Reexamining momentum profits

In this paper, we provide the first direct evidence in support of the conjectures of
Jegadeesh and Titman (1993) and behavioral models’ predictions that the momentum in
stock prices is due to a combination of underreaction and overreaction to firm-specific
information. We propose that the speed with which pricing errors are corrected in the
ranking period plays an important role in the return behavior for both the holding and post-
holding periods of momentum strategy. In order to examine this, we design a new measure
that directly captures how fast pricing errors due to firm-specific information are corrected.
Our measure, constructed using a cointegration technique, estimates the speed with which
stocks correct for potential mispricing due to firm-specific information during the ranking
period We refer to our measure as the speed of correction. We argue that stocks with a
slower correction of pricing errors during the ranking period are relatively more under-
priced for winners and more overpriced for losers. Consequently, if momentum profits are
caused by an underreaction to firm-specific information, then stocks with a slower cor-
rection of pricing errors (i.e., with more underpriced winners and more overpriced losers)
should generate greater momentum profits during the holding period. If momentum is
caused by delayed overreaction as predicted by some behavioral models, then it is rea-
sonable to expect that more underpriced winners and overpriced losers (i.e., stocks with a
slower correction of pricing errors during the ranking period) will also be susceptible to
overreaction. Therefore, stocks with a slower correction of pricing errors during the
ranking period should show greater price reversals during the post-holding period.
We find that the profit from the momentum strategy using stocks from the top (bottom) 20 %
of the slowest (fastest) correction generates 1.54 % (0.32 %) a month. This implies that, for
momentum portfolios, the predictability of past returns on future returns increases as stocks
respond with delay to firm-specific information. Our examination of post-holding period
returns reveals that momentum profits could be attributed to both underreaction and delayed
overreaction by investors to information during the holding period. However, analysis of the
more recent period shows that the role of overreaction in explaining momentum has diminished
substantially. Our results are robust with respect to recent explanations of momentum profits
and alternative methods for computing the measure of the speed of correction, whether based on
the market model or the three-factor model. Additionally, we find that the impact of the speed of
correction on pricing errors is partly related to firm characteristics and the degree of investors’
inattention. Small, low-priced, growth and high-market beta stocks correct pricing errors
slowly, especially if these stocks had low past returns. Similarly, stocks that escape investors’
attention correct pricing errors more slowly, particularly if these stocks had high past returns.
Overall, our study supports the idea that behavioral models explain momentum.
The remainder of the paper is organized as follows. Section 2 describes the method-
ology used in the paper. Section 3 describes the data, and documents the results based on
calendar-time portfolio average returns and Fama–MacBeth regressions. Section 4
explores the connection between the speed of correction, firm characteristics and the level
of investors’ attention. Section 5 examines the relationship between the speed of correction
and post-holding period returns. Section 6 concludes.

2 Methodology

2.1 The speed of correction and underreaction to information

According to our first hypothesis, we expect that those stocks that reacted to firm-specific
information relatively slowly during prior periods will exhibit greater momentum in the

123
J. Hur, V. Singh

future. In order to test our hypothesis, we require a technique that will provide us with a
parsimonious measure capturing the speed with which stocks incorporate with delay firm-
specific information. We implement the following methodology based on the speed of
correction of pricing errors derived from the cointegration of stock prices with the level of
the market index.
Our rationale for using a cointegrating relationship between stock prices and the
level of the market index is as follows: Since both stock prices and the level of the
market index reflect the overall growth in the economy over the long run (and hence,
tend to move together), it is reasonable to assume that stock prices are cointegrated
with the level of the market index over the long run [see Alexander (2008, p. 225)].
We agree that the change of stock price does not only follow the change of market
index but also size, book-to-market, momentum, and liquidity related information. Our
results are robust to inclusion of factors other than market index in cointegration
relationship with stock price. It is completely plausible that stocks may underreact to
all information that is related to the stock though. This assumption is implicit in the
model. However, it is plausible that this cointegration relationship becomes perturbed
over the short-run, something consistent with Jegadeesh and Titman’s (1993) comment
that ‘‘transactions by investors who buy past winners and sell past losers move prices
away from their long-run values temporarily.’’ We demonstrate below that short-term
perturbation occurs due to an incomplete reaction to firm-specific information. The
speed of correction in the error correction model captures how quickly pricing errors or
short-term perturbations are corrected. We estimate the speed of correction for each
firm each month using daily returns in the past 6 months. We use the daily data in the
past 6 months rather than monthly observations to obtain econometrically robust esti-
mates of speed of correction. We use the following 4 steps to estimate the speed of
correction using the error correction model.
1. We first generate the price of stock i at time t, pit , by
pit ¼ lnðPit Þ  ln½BðT  tÞ
which is the natural log of its market price of stock i ðPit Þ from CRSP minus the log of the
price of a pure discount, risk-free bond BðT  tÞ that pays $1.00 T periods in the future:
The stock price, pit is adjusted for stock splits and dividend payments. The price of a risk-
free bond, BðT  tÞ is generated from T-bill rate downloaded from French Website.
2. In the next step 2, similar to pit in step 1, we generate the level of the value-weighted
market index at time t by
mt ¼ lnðMt Þ  ln½BðT  tÞ
where ln(Mt) is the natural log of the value-weighted market index from CRSP. Note the
use of lower-case letters to denote the natural log of these variables in step 1 and step 2.
3. We run the following regression
pit ¼ ai þ hi mt þ eit
where the dependent variable pit is obtained from Step 1 and the independent variable mt is
obtained from step 2.
4. Finally, we use the lagged error term from step 3 in the following error correction
model and run the ordinary least square regression of

123
Reexamining momentum profits

Dpit ¼ hi Dmt þ ki ðpit1  ai  hi mt1 Þ þ vit


where Dpit is pit  pit1 , Dmt is mt  mt1 , and pit1  ai  hi mt1 is the lagged residual
from the regression in step 3. The estimated coefficient ki is the speed of correction.
It is important to note that in step 3
eit ¼ qi eit1 þ vit ; 0  qi  1 ð1Þ
It is assumed that vit is random white noise, independently and normally distributed with a
mean of zero and a constant variance. If pit and mt are cointegrated in Eq. (3), then the
sequence feit g is stationary, which implies that qi \1.
Hence, using the lag operator L, we can rewrite (1) as
eit ¼ vit =ð1  qi LÞ ð2Þ
Next, we substitute Eqs. (2) into step (3) equation:

pit ¼ ai þ hi mt þ vit =ð1  qi LÞ


ð3Þ
ð1  qi LÞpit ¼ ð1  qi LÞai þ hi ð1  qi LÞmt þ vit

pit ¼ qi pit1 þ ð1  qi Þai þ hi mt  hi qi mt1 þ vit ð4Þ


Subtracting pit1 from both sides of Eq. (4), and then adding and subtracting hi mt1 from
the right hand side of Eq. (4), we get
Dpit ¼ ð1  qi Þai þ ðqi  1Þpit1 þ hi Dmt  hi ðqi  1Þmt1 þ vit ð5Þ
Let ki ¼ qi  1. Then
Dpit ¼ ki ai þ ki pit1 þ hi Dmt  hi ki mt1 þ vit ; ð6Þ
or1
D pit ¼ hi Dmt þ ki ðpit1  ai  hi mt1 Þ þ vit ð7Þ

This is the regression estimated in step 4 above. It is worth noting that since qi \1, ki in
Eq. (7) is negative. Equation (7) is written in the form of the error correction model
suggested by Kremers et al. (1992), where Dpit and Dmt are continuously compounded
excess returns—that is, returns in excess of the risk-free rate of both security i and the
market index at time t, respectively. Additionally, the error correction term is written as:
eit1 ¼ pit1  ai  hi mt1 ð8Þ
We label eit1 in Eq. (8) as pricing error due to firm-specific information, since it captures
how much the level of security i deviates from the level of the market index.
In Eq. (7), the speed of correction, ki measures the sensitivity between returns and
the pricing error of security i due to the firm-specific component. It is important to
recognize that since ki is negative, it corrects for possible undervaluation/overvaluation,

1
It is important note that if price of a security and the level of the market index are (not) cointegrated, then
the returns of individual securities are (not) predictable by the Eq. 8 even if the market return is white noise.
We conducted simulation tests and found that in almost all cases the error correction term has predictability
for cointegration series.

123
J. Hur, V. Singh

namely pricing errors in the price of stock i possibly due to the incomplete incorpo-
ration of firm-specific information—that is, an underreaction to information during prior
periods. The more (less) negative ki is, the faster (slower) is the correction of mis-
pricing [see Greene (2008)]. Therefore, the stocks with a lower absolute value of ki are
expected to have a larger price increase/decrease since they have a lot more infor-
mation to impound during the current period; this is because of the underreaction to
information during the prior period. Since the stocks could have been either overpriced
or underpriced during the prior periods, a lower absolute value for ki is associated with
lower returns for overpriced stocks and higher returns for underpriced stocks. In the
context of momentum portfolios, this implies that the slower the speed of correction
(ki), the greater (smaller) the returns for the winner (loser) portfolios. Therefore, the
momentum profit of a winner minus loser portfolio is greater for stocks with a lower
absolute value for ki.
Since the pricing error in Eq. (7) purges only systematic information impounded in the
market factor, for the sake of robustness, we also estimate the speed of correction for firm-
specific information, ki, using a Fama–French three-factor model. To this end, we first
assume that the price of security i in period t can be modeled as
pit ¼ ai þ h1i mt þ h2i st þ h3i ht þ eit ; ð9Þ
where pit ; mt ; st , and ht are the logs (levels) for security i, market, size, and a book-to-
market factor index in excess of the log (level) of a risk-free asset at month t, respectively.
Second,
Dpit ¼ ci þ h1i Dmt þ h2i Dst þ h3i Dht þ ki eit1 þ vit ð10Þ
where Dpit ; Dmt ; Dst ; and Dht are the continuously compounded monthly excess returns for
security i, the market portfolio, size factor, and a book-to-market factor at month t,
respectively. eit1 is the residual from Eq. (10). We estimate ki for each security i for each
month t using the past 6 months of daily returns observations.2 For the remainder of the
paper, we refer to ki or speed of correction as SPi.

2.2 Speed of correction and momentum

If momentum is caused by underreaction, then stocks with low past returns (losers) will be
overpriced, and stocks with high past returns (winners) will be underpriced during the
ranking period. Therefore, it is expected that the pricing errors in Eq. (3) are, on average,
positive for losers and negative for winners. Since the speed of correction (ki/SPi) in Eqs.
(10) and (13) is negative, during the ranking period, the price for losers will go down, while
for winners it will increase. The slower the speed of correction, the smaller the decrease
(increase) in the price of losers (winners), and the less firm-specific information will be

2
Although we estimate k for each stock every month using past 6 months of daily returns, one may
question the constancy of k parameter due to policy changes or other structural breaks and its impact on
using the standard error correction model. In order to ensure that ks generated from our estimations are
constant, we conducted Chow test and Elliott–Müller (EM) tests and found that most estimations generate ks
that were indeed constant. We also estimated k for each stock every month using past 6 months of monthly
returns for robustness and found results similar to reported here.

123
Reexamining momentum profits

incorporated during the ranking period. The following table summarizes the essence of our
arguments.

Stocks Ranking period Holding period

Pricing Speed of correction (SPD)


error

Losers Positive Losers with more Losers with less negative Losers with less negative
(overpriced) negative SPD ? SPD is more SPD have lower return in
pricing error is overpriced than losers the holding period than
corrected more with greater negative losers with greater
Losers with less SPD negative SPD
negative SPD ?
pricing error is
corrected less
Winners Negative Winners with Winners with less Winners with less negative
(underpriced) more negative negative SPD is more SPD have higher return in
SPD ? pricing underpriced than the holding period than
error is winners with greater winners with greater
corrected more negative SPD negative SPD
Winners with less
negative SPD ?
pricing error is
corrected less

In summary, stocks with a slower correction of pricing errors during the ranking period are
relatively more underpriced for winners and more overpriced for losers during the ranking
period. Consequently, stocks with a slower correction of pricing errors should generate
greater momentum profits during the holding period. Although the momentum profits of
winner-minus-loser portfolios during the holding period can be the result of the correction of
underreaction experienced during the ranking period, it could also be attributed to the delayed
overreaction by investors during the holding period. This can be distinguished through an
examination of return behavior during the post-holding period [e.g., Jegadeesh and Titman
(2001)]. If the momentum profit during the holding period is caused by a delayed overre-
action, then the correction will occur during the post-holding period, thus resulting in neg-
ative returns for winner-minus-loser portfolios. However, if it is caused by a correction
brought about by an underreaction during prior periods, then the returns for winner-minus-
loser portfolios should be statistically no different from zero during the post-holding period.

3 The data and the results

3.1 Sample selection and momentum portfolios

Our initial sample comes from the CRSP monthly files for all NYSE and AMEX common
stocks with share codes of 10 and 11, from January 1965 to December 2009. We also
extract data from the CRSP daily, Thomson-Reuters Institutional (13-F) holdings, and
IBES historical summary files for analyses in section III.C and IV. Institutional (13-F)
holdings and IBES data commence beginning in 1980 and 1976 respectively.

123
J. Hur, V. Singh

For each month t from January 1965 to December 2009, we select stocks from the
NYSE and AMEX, if in the past, they had at least 6 months of returns. Following
Jegadeesh and Titman (2001), we exclude stocks priced \$5 (illiquid stocks) during the
momentum portfolio formation months. We also delete stocks with a market value less
than the lowest size decile of NYSE stocks, inasmuch as Hong et al. (2000) find negative
momentum profits for the smallest size decile of NYSE/AMEX stocks (see Fig. 1,
p. 276). Similar to recent papers examining momentum profits, we employ a 6 9 6
momentum strategy—that is, a ranking and holding period of 6 months [e.g., Ali and
Trombley (2006), Arena et al. (2008), and Verardo (2009)]. Each month t, we form ten
portfolios based on the buy-and-hold returns over the previous 6 months, and hold them
from t ? 2 to t ? 7, skipping a month after the ranking period. We skip a month after
forming portfolios so as to avoid short-term return reversal due to the microstructure
issues of bid-ask bounce, etc. [see Lo and MacKinlay (1990), Jegadeesh (1990), and
Lehmann (1990)]. To account for autocorrelations in the portfolios’ returns and in
monthly regressions, we compute all t statistics in our analyses using the Newey and
West (1987) technique.

3.2 Descriptive analysis and summary portfolio results

In our first step, we document the existence of momentum in our sample of stocks. To do
this, stocks in our sample are assigned to ten portfolios on the basis of their previous
6 months buy-and-hold returns. We then track the subsequent 6 months average monthly
returns of these portfolios. We provide some of the salient characteristics of our
momentum portfolios in Table 1.
The average monthly return of M10–M1 portfolios during the holding period is 0.99 %.
We designate M1 as the loser portfolio and M10 as the winner portfolio. In the second
column, we report the average buy-and-hold returns for the previous 6 months for each of
the ten momentum portfolios. We observe that the loser portfolios decreased on average by
28 %, whilst winner portfolios increased by almost 68 % in the previous 6 months. We
also report the average size (the log of market capitalization) and the average price during
the month of portfolio construction for each of the 10 portfolios. Consistent with previous
research, we find that firm size is similar for each portfolio. The average price is lower for
the extreme portfolios, particularly for loser portfolios, suggesting that low-priced stocks
show greater momentum. The next column shows the average speed of correction (SPi) for
the momentum portfolios. We observe that the speed of correction is not linearly related to
past returns. More interestingly, the speed of correction displays a U-shaped pattern, since
the extreme portfolios (M1 and M10) are associated with a slower SPi relative to other
portfolios. This is perhaps indicative of a greater overvaluation/undervaluation for these
portfolios. Thus, we expect to see larger absolute returns for winner and loser portfolios
during the holding period. Next, we report the average values of dispersion in analysts’
earnings forecasts for each momentum portfolio. Although loser portfolios have higher
dispersions in analyst earnings forecasts, we do not see any pattern in the forecasts across
other portfolios. Winner and loser portfolios seem to have a higher measure of idiosyn-
cratic volatility, possibly indicating higher arbitrage cost along the lines of Lesmond et al.
(2004) argument. The trading turnover is also higher for winner and loser portfolios
relative to other portfolios, indicative perhaps of more information-induced trading.
Winner and loser portfolios also appear to have less analyst coverage relative to other
portfolios. Lastly, we observe a monotonic relationship between capital gains overhang,

123
Reexamining momentum profits

Cumulative Momentum Profits-SP1


0.06

0.05
Culative Returns

0.04

0.03

0.02

0.01

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month

Cumulative Momentum Profits-SP2


0.06

0.05
Culative Returns

0.04

0.03

0.02

0.01

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month
Cumulative Momentum Profits-SP3
0.08

0.07

0.06
Culative Returns

0.05

0.04

0.03

0.02

0.01

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month

Fig. 1 Cumulative momentum profits. It represents the cumulative returns of the winner minus the loser in
the momentum portfolio of NYSE and AMEX stocks. SP1 (SP5) is the portfolio reflective of the fastest
(slowest) speed of correction

123
J. Hur, V. Singh

Cumulative Momentum Profits-SP4


0.1

0.09

0.08
Culative Returns

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month
Cumulative Momentum Profits-SP5
0.18

0.16

0.14
Culative Returns

0.12

0.1

0.08

0.06

0.04

0.02

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month
Cumulative Momentum Profits- All Stocks
0.09

0.08

0.07
Culative Returns

0.06

0.05

0.04

0.03

0.02

0.01

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

Event Month

Fig. 1 continued

past returns, and momentum portfolios’ future returns, something we elaborate on in


section III.
In Table 2, we investigate the validity of our hypothesis that those stocks that correct
pricing errors due to firm-specific information with greater delay display greater

123
Table 1 Descriptive statistics of 10 momentum portfolios

Rtþ2;tþ7 Ri,t-5,t Sizei,t Pricei,t SPi,t DISPi,t IVOLi,t TURN COVERi,t CGOi,t
OVERi,t

M1 0.54 -28.25 14.03 17.27 -0.050 0.39 10.60 3.39 8.44 -25.96
(1.79) (-23.24) (7.12) (38.50) (-27.14) (12.52) (45.63) (14.51) (48.45) (-31.69)
M2 0.94 -14.05 14.57 32.78 -0.068 0.21 8.34 2.37 9.37 -10.94
(3.18) (-12.02) (7.76) (8.93) (-34.58) (15.25) (43.20) (15.85) (51.25) (-15.37)
M3 1.04 -7.15 14.77 57.13 -0.080 0.16 7.58 2.15 9.84 -4.36
(3.85) (-6.25) (8.11) (6.12) (-43.24) (15.89) (45.52) (16.28) (47.80) (-7.21)
Reexamining momentum profits

M4 1.12 -1.74 14.85 50.03 -0.088 0.14 7.25 2.05 10.02 0.13
(4.45) (-1.52) (8.49) (8.49) (-55.89) (14.83) (49.65) (16.81) (44.75) (0.97)
M5 1.15 3.21 14.92 52.68 -0.092 0.12 7.10 2.05 10.22 3.93
(4.80) (2.75) (8.63) (9.00) (-64.27) (13.02) (52.85) (16.95) (41.74) (2.35)
M6 1.20 8.22 14.93 51.01 -0.092 0.12 7.05 2.08 10.25 7.64
(5.13) (6.80) (8.89) (9.09) (-66.88) (13.80) (55.68) (16.81) (37.99) (11.54)
M7 1.20 13.81 14.93 56.45 -0.086 0.12 7.20 2.20 10.18 11.27
(5.17) (10.78) (9.23) (7.50) (-56.04) (14.33) (56.74) (17.03) (35.90) (18.39)
M8 1.28 20.77 14.91 44.24 -0.077 0.12 7.51 2.45 9.89 15.18
(5.33) (14.71) (8.74) (10.59) (-46.42) (15.02) (55.54) (17.07) (33.46) (22.47)
M9 1.34 31.38 14.78 47.29 -0.064 0.13 8.26 2.95 9.39 19.97
(5.21) (18.61) (8.24) (8.33) (-39.23) (14.51) (54.26) (17.89) (31.77) (26.74)
M10 1.53 67.73 14.27 39.28 -0.045 0.18 10.78 4.69 7.50 28.65
(4.95) (22.07) (7.41) (6.26) (-32.21) (12.78) (55.44) (21.44) (27.86) (38.61)
M10–M1 0.99 95.98 0.24 22.01 0.005 -0.21 0.18 1.30 -0.94 54.61
(4.81) (38.63) (2.97) (3.52) (2.66) (-6.55) (0.58) (7.81) (-3.21) (54.67)

This table reports the summary statistics of the 10 momentum portfolios created based on the buy-and-hold returns over the past 6 months. Stocks from the NYSE and AMEX are
grouped into 10 momentum portfolios for each month t from January 1965 until December 2009. After portfolios are formed for each month t, the respective stocks are held from month
t ? 2 to t ? 7. M1 (M10) is the loser (winner) portfolio. Rtþ2;tþ7 is the average monthly return during the holding period from month t ? 2 to t ? 7. Ri,t-n,t is the buy-and-hold returns
from month t - n to t. Size is the log (market capitalization). SP is the speed of correction derived from the error correction model, the estimation of which is described in the text. DISP
is the standard deviation of the analysts’ earnings forecasts divided by the absolute mean earnings forecasts. IVOL is the monthly idiosyncratic volatility, which is the standard deviation
of the residuals from a four-factor model using the daily returns times the square root of the number of trading days in the month. TURNOVER is the average of the daily turnover ratio
for the month. Cover is the number of analysts. CGO is the capital gains overhang. The t statistics, reported in parentheses, are adjusted for autocorrelation using the Newey and West

123
(1987) method. COVER and DISP is from January 1976. TOVER is measured in thousands
J. Hur, V. Singh

momentum.3 In order to do this, we segment the stocks into groups of 5 speeds of cor-
rection, on the basis of the SPi’s estimated using the prior 6 months of daily returns as
described in section II.A. We then further sort the stocks within each speed of correction
group into 10 momentum portfolios on the basis of the past 6 months of buy-and-hold
returns. If stock price continuation during the holding period is the result of prices slowing
adjusting to firm-specific information, then the stocks with the slowest speed of correction
should display the greatest momentum.
In Panel A of Table 2, we show the summary statistics of the speed of correction
measures.4 The time-series mean (median) of the cross-sectional average of the speed of
correction is -0.16 (-0.14) for the group of stocks (designated as SP1) that correct for the
possible ranking period’s overvaluation/undervaluation the quickest. The maximum and
minimum value for SP1 ranges from -0.11 to -0.32. In contrast, the corresponding mean
(median) is -0.02 (-0.02) for the group of stocks (designated as SP5) that correct for the
possible ranking period’s overvaluation/undervaluation the slowest. The cross-sectional
variation for SP5 is much smaller as it ranges from -0.01 to -0.03. The different statistics
for our measure of speed of correction for the two extreme groups is statistically significant
as well, as indicated in the last column. It is worth noting that the values of the speeds of
correction (SPi) are negative and statistically different from zero. We did not impose a non-
positive restriction on the speed of correction in the regression. The negative speed of
correction is consistent with the theory of error correction model. (See ‘‘Unit Roots,
Cointegration, and Structural Change’’ by Maddala and Kim.) This implies that prices
deviate from equilibrium in the short-term, but eventually adjust to their long-term equi-
librium values.
In Panel B, we scrutinize the average monthly returns of stocks for different momentum
portfolios conditioned on our measure of speed of correction. A few patterns emerge that
are worth noting. The average monthly returns for momentum portfolios M1 through M5
correlate negatively with the speed of correction with respect to firm-specific information.
Since the speed of correction is negative, moving from SP1 to SP5 implies that the
magnitude of the speed of correction moves from more negative to less negative. There-
fore, the change of speed of correction is positive as we move from SP1 to SP5. These
portfolios had relatively lower returns during the ranking period. For example, for the M1
portfolio, the average monthly return varied from 0.21 % per month for the stock group
with the slowest speed of correction, to 1.01 % per month for the group of stocks with the
highest speed of correction. The difference in the average monthly return is statistically
significant as well between SP1 and SP5. We observe similar patterns for portfolios M2
through M5, though the magnitude of the difference between the two extreme groups of
stocks conditioned on the speed of correction diminishes as we move from M2 to M5. This
pattern is consistent with the idea that stocks that were relatively more overpriced during
the ranking period due to the slower incorporation of possibly negative firm-specific

3
In order to reliably infer the results of this section one requires that momentum portfolios’ returns are
stationary since we employ error correction model. We investigated this for each of the momentum port-
folios in this section by evaluating the time-series plots and more rigorously by Dickey–Fuller unit root tests.
Our investigation confirmed that momentum portfolio returns are indeed stationary. We thank an anony-
mous referee for this suggestion.
4
The cointegration between the stock and the market index may not be I(1) for some stocks. We found that
about 17 % of stocks in our sample are not cointegrated with the market index. There are, on average, 1,709
firms in our sample. So, about 290 firms, on average, are not cointegrated with the market index, which is
about 5.8 firms in each portfolio in Table 2. We verified that our results shown in the paper are not affected
by the exclusion of the non-cointegrated firms. We thank an anonymous referee for this robustness check.

123
Reexamining momentum profits

Table 2 The average speed of correction and holding period monthly returns
SP1 SP2 SP3 SP4 SP5 SP1–SP5

Panel A: Speed of correction


Minimum -0.32 -0.11 -0.07 -0.05 -0.03 -0.29
(-70.86) (-62.90) (-54.16) (-45.66) (-35.45) (-67.87)
Mean -0.16 -0.09 -0.06 -0.04 -0.02 -0.14
(-74.40) (-58.99) (-50.43) (-41.54) (-23.84) (-84.16)
Median -0.14 -0.09 -0.06 -0.04 -0.02 -0.12
(-70.86) (-58.80) (-50.23) (-41.42) (-26.19) (-80.64)
Maximum -0.11 -0.08 -0.05 -0.03 -0.01 -0.10
(-63.73) (-54.73) (-46.26) (-36.30) (-20.52) (-76.21)
Panel B: Average holding period monthly returns
M1 1.01 0.81 0.72 0.56 0.21 0.81
(6.33)
M2 1.11 1.04 1.03 0.80 0.63 0.47
(4.34)
M3 1.24 1.13 1.08 0.97 0.83 0.42
(4.81)
M4 1.28 1.11 1.08 0.99 0.93 0.35
(3.92)
M5 1.23 1.18 1.11 1.03 1.06 0.17
(2.34)
M6 1.27 1.16 1.16 1.08 1.13 0.14
(1.88)
M7 1.27 1.20 1.19 1.16 1.21 0.06
(0.64)
M8 1.28 1.28 1.25 1.19 1.32 -0.03
(-0.32)
M9 1.34 1.37 1.35 1.36 1.49 -0.15
(-1.31)
M10 1.34 1.37 1.43 1.54 1.74 -0.41
(-2.75)
M10–M1 0.32 0.56 0.71 0.99 1.54 1.22
(1.88) (3.22) (3.76) (4.45) (5.70) (5.84)
Panel C: Average holding period monthly returns
All Stocks 1.24 1.17 1.14 1.06 1.04 0.24
(3.33)
This table reports the summary statistics for the speed of correction and the average monthly returns during
the holding period from month t ? 2 to t ? 7. For each month t from January 1965 until December 2009,
we group stocks from the NYSE and AMEX into five portfolios based on the speed of correction and then
group each of these into 10 momentum portfolios based on the buy-and-hold returns from t - 5 to t. The
estimation of the speed of correction (SP) is described in the text. SP1 (SP5) is the portfolio with the fastest
(slowest) speed of correction. M1 (M10) is the loser (winner) portfolio. The t statistics, reported in
parentheses, are adjusted for autocorrelation using the Newey and West (1987) method

information show lower returns during the holding period in order to correct for under-
reaction. This phenomenon contributes to the fact that stocks performing poorly during the
ranking period continue to show a downward drift.
We observe a reverse pattern for portfolio M10. This portfolio had the highest return
during the ranking period. The average monthly return tends to correlate positively with

123
J. Hur, V. Singh

our measure of the delay of information. The difference in profit between the groups with
the fastest and the slowest speeds of correction is -0.41 % a month. This is consistent with
the idea that those stocks relatively more underpriced during the ranking period due to a
slower incorporation of possibly positive firm-specific information show higher returns
during the holding period in order to correct for underreaction during the ranking period.
This phenomenon contributes to the fact that stocks performing well during the ranking
period continue to show an upward drift.
The aforementioned patterns directly impact the average monthly returns of winner-
minus-loser portfolios (M10–M1) across different groups based on the speed of correction.
For example, the average monthly return of the winner-minus-loser portfolios in the SP5
group (with the slowest speed of correction) is 1.54 % compared to 0.32 % for those in the
SP1 group (with the fastest speed of correction). This difference of more than 1 % a month
is statistically significant at 1 %. Thus, stocks with the slowest speeds of correction earn
the momentum profits almost five times as much as stocks with the fastest speeds of
correction. This is consistent with the notion that since the speed of correction corrects for
the degree of underreaction during the ranking period, the winner-minus-loser portfolios
with the slowest speeds of correction show the greatest momentum.
In Panel C of the Table 2, we computed monthly average return on stocks sorted by
speed of correction only. This shows that that firms with greater negative speed of cor-
rection (SP1, less underreaction) have higher return than firms with less negative speed of
correction (SP5, more underreaction).
We conduct additional analyses to verify the robustness of our results shown in Table 2.
We considered sorting stocks into 10 momentum portfolios, followed by sorting each
momentum portfolio further based on speeds of correction. Sorting this way, our results are
qualitatively similar to those reported here. We also considered forming portfolios sorted
independently based on past returns and the speed of correction and then we match the
portfolios. In comparison to these alternative approaches, our sorting scheme ensures
identical speeds of correction across different momentum portfolios. Additionally, it does
not suffer from problem of unbalanced portfolios. We measure speed of correction (lamda)
using past 6 months of daily returns. Although in Table 1 we observed that the speed of
correction and past returns do not appear to be related, we orthogonalize the speed of
correction by regressing the SPi’s on the past 6-months’ buy-and-hold returns cross-sec-
tionally for each month, and use the residuals to form 5 speed of correction portfolios. As
in Table 2, we further examine the momentum portfolios’ average monthly returns for the
subsequent 6 months within each speed of correction group.
The results, shown in Panel A of Table 3, indicate that the average monthly returns
across the different speed of correction groups from the various momentum portfolios are
similar to those in Table 2. Similarly, the average monthly returns for the winner-minus-
loser portfolios are highest for the slowest speed of correction group. Therefore, our results
in Table 2 are driven by the speed of correction, uncontaminated by past returns.
We also estimate our measure of the speed of correction using the Fama–French three
factor model, as described in section II.A, and repeat the analysis done in Table 2. Our
results, shown in Panel B of Table 3, suggest that the model of the estimation for the speed
of correction has little impact on our main results.

3.3 Multiple regression results

The results presented so far support our hypothesis that those stocks that incorporate with
delay firm-specific information are associated with greater momentum profits during the

123
Reexamining momentum profits

Table 3 The average monthly


SP1 SP2 SP3 SP4 SP5 SP1–SP5
returns for 10 momentum
portfolios
Panel A: Based on orthogonalization
M1 1.04 0.85 0.77 0.57 0.13 0.91
(7.19)
M2 1.14 1.06 0.99 0.84 0.57 0.57
(5.47)
M3 1.24 1.09 1.04 1.00 0.75 0.49
(6.14)
M4 1.29 1.18 1.11 0.96 0.91 0.38
(4.86)
M5 1.23 1.17 1.11 1.05 0.97 0.25
(3.49)
M6 1.26 1.20 1.18 1.07 1.04 0.22
(3.12)
M7 1.33 1.23 1.20 1.17 1.17 0.16
(1.76)
M8 1.28 1.31 1.25 1.24 1.25 0.03
(0.32)
M9 1.35 1.39 1.33 1.39 1.47 -0.11
(-1.13)
M10 1.35 1.39 1.51 1.62 1.62 -0.27
This table reports the average (-2.10)
monthly returns during the
holding period from month t ? 2 M10–M1 0.31 0.53 0.74 1.05 1.49 1.18
to t ? 7. For each month t from (1.75) (3.02) (3.81) (4.65) (5.67) (6.04)
January 1965 until December Panel B: Based on three factor model
2009, we group stocks from the M1 0.85 0.70 0.58 0.46 0.32 0.52
NYSE and AMEX into five (4.63)
portfolios based on the speed of
M2 1.05 1.00 1.02 0.85 0.71 0.35
correction, and then group each
(4.25)
of these into 10 momentum
portfolios based on the buy-and- M3 1.18 1.10 1.04 0.99 0.90 0.28
hold returns from t - 5 to t. The (3.72)
estimation of the speed of M4 1.24 1.18 1.08 1.06 0.93 0.31
correction (SP) using the market (4.57)
model (the Fama–French three M5 1.20 1.18 1.13 1.08 1.05 0.16
factor model), based on an error (2.45)
correction model for panel A (B),
is described in the text. For Panel M6 1.19 1.25 1.20 1.14 1.14 0.05
A, we regress the speeds of (0.71)
correction for the past 6-month M7 1.28 1.23 1.24 1.17 1.16 0.12
returns cross-sectionally for each (1.56)
month, and use the residuals to M8 1.25 1.28 1.27 1.28 1.26 -0.01
form 5 speed of correction (-0.08)
portfolios. SP1 (SP5) is the
fastest (slowest) speed of M9 1.38 1.37 1.35 1.35 1.42 -0.04
correction portfolio. M1 (M10) is (-0.36)
the loser (winner) portfolio. The M10 1.35 1.47 1.51 1.58 1.64 -0.29
t statistics, reported in (-2.15)
parentheses, are adjusted for M10–M1 0.50 0.77 0.93 1.12 1.32 0.82
autocorrelation using the Newey (2.47) (3.77) (4.88) (4.98) (5.23) (4.36)
and West (1987) method

holding period. Nevertheless, our analysis in the preceding section is fundamentally uni-
variate in nature. Correspondingly, our results may lend themselves to alternative expla-
nations. It is quite probable that some of the recent explanations for momentum profits may

123
J. Hur, V. Singh

be related to stocks identified by our measure that react to information with relative delay.
Therefore, unless we explicitly control for alternative explanations, our results will not be
credible.
Arena et al. (2008) find that momentum investing is more profitable for stocks with
high idiosyncratic volatility. They argue that idiosyncratic volatility is related to firm-
specific risk and the cost of arbitrage. Hou and Moskowitz (2005) also find that those
stocks with high residual volatility face difficulty in incorporating information speedily.
Similarly, the high cost of arbitrage limits the ability of arbitrageurs to exploit market
inefficiency and leads to momentum, as per Shleifer and Vishny (1997). Since our
measure of the speed of correction attempts to capture firms that react with delay to
information, it might be biased towards those firms that have high idiosyncratic vola-
tility. Therefore, we control for idiosyncratic volatility using multiple regressions. Idi-
osyncratic volatility is computed for each stock for each month as the standard deviation
of the residuals using a four-factor model based on the daily returns times the square root
of the number of trading days in the month. In a recent paper, Verardo (2009) finds that
heterogeneity of beliefs, as measured by the dispersion of analyst earnings forecasts, and
momentum are related. The link between the dispersion of beliefs and momentum
originates from several rational and behavioral theoretical models. In a rational setting,
Allen et al. (2006) show that higher order beliefs lead to price drifts. In a behavioral
setting, Hong and Stein (1999, 2007), employing the framework of gradual information
diffusion, imply that investor heterogeneity leads to return continuation in stocks. In a
related stream of literature, Lee and Swaminathan (2000) find that those stocks with
higher trading volume exhibit stronger momentum. Hong and Stein (2007) show that the
trading volume proxies for the degree of disagreement, and causes momentum. Since
high dispersion of opinions/beliefs may cause information to be reflected with delay in
the prices, our results with respect to the speed of correction may be a manifestation of
these alternative explanations of momentum.
We control for heterogeneous beliefs using turnover and dispersions in analysts’
earnings forecasts. Turnover is computed for each stock as the average of the daily
turnover ratio for the month. To proxy dispersion, we use the standard deviation of ana-
lysts’ earnings forecasts divided by the mean earnings forecasts as per Verardo (2009).
Specifically, the following formula computes the dispersion for a stock i at the end of
month t:
Stdðfi;t Þ
Dispi;t ¼  ; ð11Þ
Avgðfi;t Þ

where Stdðfi;t Þ—refers to the


 standard deviation of analysts’ earnings forecasts for stock i
at the end of month t, and Avgðfi;t Þ is the absolute value of the analysts’ mean earnings
forecasts for stock i at the end of month t. These variables are extracted from the IBES
Historical Summary files for the period 1976–2009.
Recent research has explored yet another explanation for momentum in stock returns
arising out of the behavioral bias of investors. This is the tendency of investors to hold onto
loser stocks for too long, and to sell winner stocks too soon—a widely documented
regularity in experimental and financial markets in many countries. Shefrin and Statman
(1985) label this phenomenon the disposition effect. Grinblatt and Han (2005) find that a
variable proxying for the aggregate of unrealized capital gains (or capital gains overhang)
is the primary driver of momentum, and that past returns do not predict future returns after
controlling for this variable. The economic reasoning behind capital gains overhang’s

123
Reexamining momentum profits

predictive power for future returns is that it essentially captures the degree of mispricing in
stocks due to the incomplete incorporation of information. Specifically, those stocks with
good information are underpriced, since investors’ eagerness to book profits creates an
excess supply in the market, thus causing prices to be lower. Stocks with negative infor-
mation remain overpriced, as investors’ reluctance to book losses restricts the supply in the
market, thus causing prices to be higher. It is possible that our measure, which directly
captures the speed with which stocks incorporate information with delay, may be related to
capital gains overhang. Therefore, we control for the disposition effect as an explanation
for momentum by including capital gains overhang in our regressions. We estimate it using
the technique provided in Grinblatt and Han (2005). Our proxy for capital gains overhang,
CGOi,t, for each stock i at the end of each month t is
Pi;t  RPi;t
CGOi;t ¼ ð12Þ
Pi;t
Pi,t is the price of stock i at the end of month t. We compute the reference price (cost basis)
RPi,t for each stock i at the end of every month from the end of 1964 to 2009 using up to
the previous 3 years of daily data. Our estimate of the reference price is as follows:
!
1X T Y
n1  
RPi;t ¼ Vi;tn 1  Vi;tnþs Pi;tn ð13Þ
k n¼1 s¼1

Vi,t is date t’s turnover in stock i. T refers to the number of trading days in the previous
3 years with available daily price and volume information. The term in the parentheses
multiplying Pi,t-n is weights, and k is a constant that makes the entire weights equal to one.
The weight on Pi,t-n reflects the probability that the shares purchased on date t - n have
not been traded since. We make appropriate adjustments for stock-splits, stock-dividends,
etc., in the share turnover and share price variables when computing RPi,t and CGOi,t.
Underreaction to firm-specific information drives momentum in the models of Barberis
et al. (1998), and Hong and Stein (1999). If analysts’ research efforts help to speed up the
incorporation of firm-specific information, we should expect to observe greater momentum
in stocks with less analyst coverage. In fact, Hong et al. (2000) document that stocks with
less analyst coverage show greater momentum. Our measure of the speed of correction,
which also captures stock prices’ response with delay to firm-specific information, can
reasonably be related to analyst coverage. Therefore, we also control for analyst coverage
using multiple regressions. We proxy analyst coverage with the log of the number of valid
earnings estimates used to compute the mean earnings forecasts in the IBES Historical
Summary file.
Consequently, we control for idiosyncratic volatility, dispersion in analysts’ earnings
forecasts, turnover, the capital gains overhang, and analyst coverage using multiple
regressions. We use versions of the following specification to estimate predictive cross-
sectional regressions in a Fama–MacBeth framework5:

5
We do not use the interaction variable between past returns and the capital gains overhang, since Grinblatt
and Han (2005) show that, for momentum, past returns is a noisy proxy of capital gains overhang.

123
J. Hur, V. Singh

Ri;tþ2;tþ7 ¼ k0t þ k1t Ri;t5;t þ k2t Speedi;t þ k3t Ri;t5;t  Speedi;t þ k4t Ri;t5;t  Idio: Vol:i;t
þ k5t Ri;t5;t  Turnoveri;t þ k6t Ri;t5;t  Dispi;t þ k7t Ri;t5;t  Cover:i;t
þ k8t Idio: Vol:i;t þ k9t Turnoveri;t þ k10t Disp:i;t þ k11t Cover:i;t þ k12t CGOi;t þ ei;t
ð14Þ

Ri;tþ2;tþ7 is the average monthly holding period’s return for security i from t ? 2 to
t ? 7. Ri,t-5,t is the buy-and-hold return for security i during the ranking period, from t–5
to t. Speedi,t is the speed of correction estimated using Eq. (10) in models (1)–(7) and using
Eq. (13) in model (8), as described in Sect. 2. For each security i, Idio.Vol.i,t is the
idiosyncratic volatility computed for month t, Turnoveri,t is the average daily turnover for
month t, Disp.i,t is the dispersion in the analysts’ earnings forecasts for month t, Cover.i,t is
the analyst coverage for month t, CGOi,t represents the capital gains overhang as of month
t, and 9 represents the interaction between the two variables. To enhance the power of our
tests, we confine our analysis to stocks in winner and loser portfolios. In the Fama–
MacBeth regressions, we control for positive autocorrelation using the Newey and West
(1987) technique. We hypothesize that those stocks that react relatively with delay to firm-
specific information should display a greater momentum effect. Since the speed of cor-
rection is negative, moving from SP1 (the fastest speed of correction) to SP5 (the slowest
speed of correction) implies an increase in the value of the speed of correction. This
suggests that we expect the interaction term between the variable Speedi,t and the past
return (Ri,t-5,t) to be positive and statistically significant for multiple regressions.
The results of these regressions are reported in Panel B of Table 4. In Panel A, we
examine the time-series average of the cross-sectional Pearson correlation coefficients
between the speed of correction and some of the important variables identified in prior
research that have explanatory power with respect to momentum returns. We find that the
speed of correction is positively correlated with idiosyncratic volatility, turnover, disper-
sion and capital gains overhang. The speed of correction is negatively correlated with
analyst coverage. This is not surprising, since stocks with a lower speed of correction, high
idiosyncratic volatility, turnover, dispersion, capital gains overhang and low analyst cov-
erage are associated with a slower incorporation of information into stock prices.
Panel B of Table 4 shows the incremental explanatory power of speed of correction
with respect to momentum profit.6 We find that the interaction term between the speed of
correction and past returns is positive and statistically significant after controlling for other
independent variables in different versions of regression Eq. (14).
The marginal effect of past returns on future returns in momentum strategy in Eq. (14),
oE½Ri;tþ2;tþ7 jRi;t5;t ;Speedi;t 
oRi;t5;t ¼ k1t þ k3t Speedi;t , is a function of Speedi,t, and increases with
Speedi,t if k3t is positive. Again, it is important to note that since the speed of correction is
negative, an increase in the value of the speed of correction implies a slower speed of
oR
i;tþ2;tþ7
correction. For example, the regression (2) in Table 4 reports that oSpeed i;t
¼ 1:468þ
4:447Ri;t5;t . In case of loser portfolio (M1), Ri;t5;t is -0.2825 in Table 1.
oRi;tþ2;tþ7
o Speedi;t ¼ 1:468 þ 4:447  ð0:2825Þ ¼ 2:724. In case of winner portfolio (M10),
oRi;tþ2;tþ7
Ri;t5;t is -0.6773 in Table 1. o Speedi;t ¼ 1:468 þ 4:447  ð0:6773Þ ¼ 1:544. Thus the

6
To the extent the speed of correction (lamda) suffers from errors in variables problems, the statistical
significance of lambda and its interaction with other explanatory variables is understated thus making our
results more conservative.

123
Table 4 The average correlation coefficients and the Fama–MacBeth cross-sectional regressions

Idio.Vol. Turnover Disp. Cover. CGO

Panel A: Pearson correlation coefficients


Speed 0.022 0.129 0.014 -0.043 0.032
(4.14) (35.97) (5.42) (-6.30) (3.65)

(1) (2) (3) (4) (5) (6) (7) (8)

Panel B: Fama–MacBeth cross-sectional regressions


Reexamining momentum profits

Ri,t-5,t 1.048 1.528 1.500 1.374 1.205 0.835 1.317 1.255


(5.49) (6.58) (7.96) (5.48) (6.28) (5.17) (5.76) (5.37)
Speedi,t -1.385 -1.468 -1.243 -1.261 -1.369 -1.383 -0.609 -0.454
(-3.91) (-4.79) (-4.26) (-2.58) (-2.90) (-4.19) (-2.01) (-1.98)
Ri,t-5,t 9 Speedi,t 3.463 4.447 4.611 5.682 5.016 3.393 6.189 5.477
(3.31) (3.24) (3.03) (2.83) (2.67) (2.39) (3.28) (3.23)
Ri,t-5,t 9 Idio. Voli,t -2.181 0.992 0.223
(-1.83) (0.60) (0.19)
Ri,t-5,t 9 Turnoveri,t -27.218 -24.315 -23.349
(-1.43) (-0.80) (-0.77)
Ri,t-5,t 9 Dispi,t -0.101 -0.117 -0.104
(-0.61) (-0.73) (-0.65)
Ri,t-5,t 9 Coveri,t 0.079 -0.114 -0.130
(0.60) (-1.32) (-1.25)
Idio. Voli,t -21.080 -20.991 -21.316
(-1.60) (-1.29) (-1.33)
Turnoveri,t -61.978 -13.910 -13.405
(-2.95) (-0.83) (-0.80)
Dispi,t -0.189 -0.189 -0.183
(-1.71) (-1.68) (-1.73)

123
Table 4 continued
(1) (2) (3) (4) (5) (6) (7) (8)

123
Cover. -0.061 -0.103 -0.104
(-1.16) (-1.75) (-1.87)
CGOi,t 0.610 0.631 0.646
(3.78) (2.37) (2.46)

This table reports the time-series averages of the Pearson correlation coefficients and the Fama–MacBeth cross-sectional regression of the average monthly returns during the
holding period, from month t ? 2 to t ? 7 for firm characteristics. For each month t from January 1965 until December 2009, we group the stocks from the NYSE and AMEX
into five portfolios based on the speed of correction and then further divide each portfolio into 10 momentum portfolios based on the buy-and-hold returns from t - 5 to t. The
estimation of the speed of correction (SP) is described in the text. The market model-based error correction model for Eqs. (1)–(7) and a three factor based error correction
model for Eq. (8) are used. Ri,t-5,t is the buy and hold returns during the ranking period from month t - 5 to t. Idio. Vol is the standard deviation of the residuals from the four-
factor model using the daily returns data times the square root of the number of trading days during the given month. Turnover is the average of the daily turnover ratio for the
month. Disp. is the standard deviation of the analysts’ earning forecasts divided by the absolute mean earnings forecasts. Cover. is the log of the number of analysts. CGO is
the capital gains overhang. 9 represents the interaction between the two variables. Stocks in the M1 and M10 portfolios are used in this regression. We control for
autocorrelation using the Newey and West (1987) technique for t statistics
J. Hur, V. Singh
Reexamining momentum profits

future return and speed of correction is negatively related for the loser portfolio, and future
return and speed of correction is positively related for the winner portfolio. Table 2B also
verifies this result. For the loser portfolio (M1), the future return and speed of correction is
negatively related, i.e. as the speed of correction increases, the future return decreases. For
the winner portfolio (M10), the future return and speed of correction is positively related,
i.e. as the speed of correction increases, the future return increases. It is important to note
that the speed of correction is negative number and SP1 is more negative than SP5.7 This is
consistent with our hypothesis that momentum profits are greater for stocks with a delayed
reaction to firm-specific information.8 In the full models 7 and 8, only the interaction term
between past returns and the speed of correction has explanatory power for future returns.9
The interaction terms between idiosyncratic volatility and past returns, and between analyst
coverage and past returns feature correct signs, though both are statistically insignificant.
The signs for the interaction terms between turnover and past returns, and those between
dispersion and past returns are contrary to expectation, but are statistically insignificant.
The past returns and the capital gains overhang retain their explanatory power with respect
to future returns. Overall, the results in this section are consistent with the idea that though
our measure of speed correction is related to idiosyncratic volatility, heterogeneous beliefs
amongst investors, analyst coverage and disposition effect, its ability to explain momentum
profits goes beyond these relationships.

4 The speed of correction, firm characteristics, and investors’ attention

In the previous section, we found that the speed of correction is associated with the proxies
for dispersion of opinion, analyst coverage, disposition effect and information uncertainty,
since all of these are related to the slower incorporation of firm-specific information into
stock prices. In this section, we try to gain more economic intuition behind our measure by
relating it to different firm characteristics and exploring its connection with investors’
attention.
It is conceivable that due to the presence of unsophisticated investors or limited and
hard to interpret information, firms with a small size, a low stock price, and/or a low book-
to-market ratio (growth stocks) may have difficulty incorporating firm-specific informa-
tion. Consequently, it stands to reason that these attributes may be correlated with our
measure of the speed of correction. We investigate this by regressing the speed of cor-
rection for each stock for each month with regards to firm size, stock price and book-to-
market ratio in a Fama–MacBeth setting. Additionally, we also examine whether high

7
As the referee points out, the more negative SP means the increase of the speed of correction. Similar to
findings on momentum strategy in prior studies, M1 (loser) and M10 (winner) have all positive returns and
M10 have higher returns than M1. The more negative SP does not necessarily mean that the future return for
M10 is negative. It implies that the future return for M10 with the more negative SP is greater than the future
return for M10 with the less negative SP because M10 with more negative SP is less underpriced than M10
with less negative SP.
8
The significant negative sign in Table 4 for the speed of correction variable, though confusing, is due to
the negative (positive) relationship between the speed of correction and the loser (winner) portfolios’
returns, as observed in Table 2 in Panel C. We include the speed of correction variable in the regression in
order to avoid model misspecification, rather than due to any theoretical considerations.
9
Contrary to Grinblatt and Han (2005), we find that the past return is statistically significant for explaining
future returns in a momentum strategy, even after controlling for capital gains overhang. We conjecture that
this happens because we follow returns for the subsequent 6 months, in contrast to the weekly returns
focused on in Grinblatt and Han (2005).

123
Table 5 The Fama–MacBeth cross-sectional regression
M1 M10

123
Panel A: Firm characteristics
Size ($ million) -0.0081 0.0051
(-5.34) (0.85)
Average price -0.0007 -0.0003
(-3.38) (-0.25)
BE/ME -0.0177 0.0017
(-6.38) (0.12)
Beta 0.0097 -0.0037
(3.00) (-1.07)
Panel B: Attention variables
Institutional holdings (%) -0.0005 -0.0003
(-0.96) (-3.25)
Number of analysts -0.0015 -0.0012
(-0.43) (-4.25)
Number of shareholders (thousands) -0.0002 -0.0002
(-1.14) (-1.97)
Age -0.002 -0.0003
(-2.52) (-4.23)
This table reports the results of the Fama–MacBeth cross-sectional regression of the speed of correction for size, average price, book-to-market ratio (BE/ME), the CAPM
beta, institutional holdings, the number of analysts, the number of shareholders and age. For each month t from January 1965 until December 2009, we group stocks from the
NYSE and AMEX into five portfolios based on the speed of correction and then group each portfolio into 10 momentum portfolios based on buy-and-hold returns from t - 5
to t. M1 and M10 are the loser and winner portfolios respectively. The estimation of the speed of correction (SP) is described in the text. Size is the market capitalization for
the previous month in $ millions; beta is the slope coefficient of a market model regression using the daily returns for the past 6 months; institutional holdings is the percentage
of institutional ownership; age is the difference between the calendar year and the first year of the firm as per the CRSP data; the number of analysts is the log of the number of
analysts; and BE/ME is the book equity for the fiscal year ending calendar year t - 1, divided by the market equity at the end of December of year t - 1. We control for
autocorrelation using the Newey and West (1987) technique using for t statistics
J. Hur, V. Singh
Reexamining momentum profits

market beta stocks are related to the speed of correction. Although we do not mean to
imply that beta is related to the speed with which firms incorporate information, we include
this as an important firm attribute. We measure firm size by its market capitalization at the
end of the previous month. Beta is measured by the slope coefficient of a market model
regression using the past 6 months of daily returns for a given firm. The average price is
the average daily price for the month, and the book-to-market ratio is the book equity for
the fiscal year ending calendar year t - 1, divided by the market equity at the end of
December of t - 1. We conduct this analysis separately for stocks in the winner and loser
portfolios in order to examine whether past returns interact with firm characteristics to
influence the speed of correction.
In Panel A of Table 5, we report the time-series averages of the coefficients for market
capitalization, the stock price, the book-to-market ratio, and the market beta from monthly
cross-sectional regressions. We find that for the loser stocks, the speed of correction is
negatively associated with firm size, stock price and book-to-market ratio. However, it is
positively correlated with the market beta. This is consistent with the idea that smaller
firms, low-priced stocks, growth stocks, and high market beta stocks possibly react with
delay to negative firm-specific information. For winner stocks, we do not find any asso-
ciation between the speed of correction, firm size, stock price, book-to-market ratio, and
the market beta.
Motivated by Kahneman’s (1973) assertion that attention is a scarce cognitive resource,
several recent papers have examined the role of investors’ inattention with respect to asset
price dynamics [see Barber and Odean (2008); Cohen and Frazzini (2008); Dyck and
Zingales (2003); Hirshleifer and Teoh (2003); Hong and Stein (1999); Huberman and
Regev (2001); Peng and Xiong (2006), Gabaix et al. (2006); Hirshleifer, Lim and Teoh
(2009); and Dellavigna and Pollet (2009)]. These studies generally conclude that because
of cognitive constraints, investors are often distracted and/or have limited attention. This
causes both an underreaction and overreaction to information, leading to asset return
predictability. Therefore, we reason that since the speed of correction captures with delay
investors’ reaction to information, it could be related to investors’ inattention.
Following Hou and Moskowitz (2005), we employ institutional ownership, the number
of analysts, and the number of shareholders as measures of investors’ attention. Institu-
tional ownership and analyst coverage are associated with more recognizable firms, and
thus should catch investors’ attention. The number of shareholders denotes the breadth of
ownership. In addition, we use firm age as a proxy for investors’ attention. We expect that
older firms should be more familiar to investors than younger ones. We aggregate insti-
tutional holding data from Thomson-Reuters Institutional (13-F) files for each stock for
each quarter, and divide it by outstanding shares in order to compute institutional own-
ership. We assume that ownership remains constant during the intervening months, until
next quarter data becomes available. The number of analysts is based on the number of
estimates used each month in order to compute mean annual earnings forecasts in the IBES
historical summary files. The number of shareholders is extracted from the COMPUSTAT
annual files. Age is computed as the difference between the calendar year and the first year
a firm’s return is available in the CRSP data. We regress the speed of correction each
month cross-sectionally within a Fama–MacBeth framework for our proxies of investors’
attention. Similar to firm characteristics, we run separate regressions for the stocks in the
winner and loser portfolios in order to capture any impact of past returns on investors’
attention regarding the speed of correction.
Panel B of Table 5 shows the time-series average of the investor attention coefficients
for winner and loser stocks. Institutional ownership, the number of analysts, the number of

123
J. Hur, V. Singh

shareholders, and age are negatively correlated with the speed of correction for winner
portfolios. We find similar correlations for loser portfolio stocks, though here, only age is
statistically significant. This is consistent with the notion that stocks with a slower speed of
correction incorporate information with delay partly because they escape investors’
attention due to their lower visibility, especially if they have high past returns, presumably
due to good information. Overall, the analysis in this section shows that the ability of the
speed of correction to capture the relative delay with which certain stocks react to firm-
specific information is partly related to firm attributes and investor attention. While firm
characteristics are related to the speed of correction when stocks have low past returns,
investor inattention is associated with the speed of correction for stocks with high past
returns.

5 The speed of correction, overreaction and post-holding period returns

Our analysis hitherto has established that momentum profit in prior research has been
primarily a consequence of investors’ delayed reaction to firm-specific information as
captured by the speed of correction. In this section, we investigate this issue more closely.
On the one hand, investors during the holding period may react with delay to information,
and thus correct for possible underreaction during the ranking period. On the other hand, it
is also possible that investors’ actions during the holding period could constitute purely an
overreaction to firm-specific information, thus moving prices away from equilibrium. The
momentum profits observed during the holding period are consistent with either of these
possibilities. Therefore, the speed of correction may be related to momentum profit, either
because it accounts for the correction of the underreaction during the ranking period, or
captures the overreaction during the holding period.
As argued earlier, the common theme of the most prominent behavioral models
employed to explain momentum is that it arises due to investors’ initial underreaction to
information followed by an overreaction, or is entirely due to overreaction [e.g., Barberis
et al. (1998), Hong and Stein (1999), Daniel et al. (1998), and DeLong et al. (1990)]. The
most important studies that document short-term underreaction in stocks are Jegadeesh and
Titman (1993), Conrad and Kaul (1998). The long-term overreaction is well documented in
Debondt and Thaler (1985, 1987, and 1990). Therefore, we can infer that all behavioral
models predict overreaction to information. If stocks do overreact to firm-specific infor-
mation, then we should observe price reversals over the long run. The stocks that overreact
more should also show greater reversals. If stocks associated with slower speeds of cor-
rection display greater momentum because they overreact with delay more to firm-specific
information, then these stocks should also show greater price reversal during the post-
holding period. To this end, we follow Jegadeesh and Titman (2001), and inspect the
average monthly returns of winner-minus-loser portfolios during the post-holding period
conditioned on the speed of correction.
Following Jegadeesh and Titman (2001), we investigate the returns of M10–M1 during
the post-holding period from month t ? 13 to month t ? 60, since the momentum is strong
for up to 12 months after portfolio formation. In Fig. 1, we plot the cumulative monthly
returns of the winner-minus-loser portfolios sorted based on the speed of correction for the
respective post-holding periods extending to 5 years. It is evident that in many cases stock
prices show a reversal. More interestingly, the degree of reversal is related to the speed of
correction. The stocks with the slowest speed of correction (SP5) show the largest reversal.
This group of stocks is also associated with the greatest momentum profit. In contrast, the

123
Reexamining momentum profits

Table 6 Average monthly returns in post holding period months


Months 13–24 Months 25–36 Months 37–48 Months 49–60 Months 13–60

Panel A: 1965–2004
SP1 -0.07 0.02 -0.08 -0.09 -0.06
(-1.26) (0.31) (-1.26) (-1.26) (-1.04)
SP2 -0.10 0.07 -0.06 -0.28 -0.09
(-1.43) (0.96) (-0.85) (-3.90) (-1.33)
SP3 -0.15 0.05 -0.08 -0.30 -0.12
(-2.21) (0.64) (-1.10) (-4.02) (-1.95)
SP4 -0.16 -0.06 -0.09 -0.37 -0.17
(-2.12) (-0.75) (-1.06) (-4.31) (-2.20)
SP5 -0.26 -0.20 -0.19 -0.35 -0.25
(-2.79) (-2.10) (-2.27) (-3.33) (-2.78)
SP5–SP1 -0.19 -0.22 -0.11 -0.26 -0.19
(-2.29) (-2.50) (-1.38) (-2.85) (-3.75)
Panel B: 1965–1984
SP1 -0.21 0.09 -0.00 -0.26 -0.10
(-3.11) (0.89) (-0.03) (-3.36) (-1.48)
SP2 -0.19 0.08 -0.09 -0.29 -0.12
(-2.39) (0.81) (-1.20) (-3.61) (-1.77)
SP3 -0.19 0.02 -0.05 -0.36 -0.15
(-2.77) (0.17) (-0.59) (-4.48) (-2.48)
SP4 -0.19 -0.17 -0.18 -0.43 -0.24
(-2.70) (-1.50) (-2.20) (-5.69) (-4.13)
SP5 -0.14 -0.24 -0.19 -0.49 -0.26
(-1.46) (-1.60) (-2.11) (-5.00) (-3.60)
SP5–SP1 0.07 -0.33 -0.19 -0.23 -0.17
(0.66) (-2.63) (-1.86) (-2.33) (-2.83)
Panel C: 1985–2004
SP1 0.06 -0.04 -0.17 0.07 -0.02
(0.80) (-0.41) (-1.89) (0.75) (-0.39)
SP2 -0.02 0.05 -0.03 -0.28 -0.07
(-0.21) (0.53) (-0.31) (-2.99) (-1.27)
SP3 -0.13 0.05 0.00 -0.25 -0.08
(-1.23) (0.71) (0.19) (-2.21) (-1.25)
SP4 -0.11 0.07 -0.11 -0.25 -0.10
(-1.26) (0.77) (-1.21) (-2.58) (-1.48)
SP5 -0.38 -0.16 -0.19 -0.21 -0.24
(-3.56) (-1.36) (-1.90) (-1.65) (-3.17)
SP5–SP1 -0.45 -0.12 -0.03 -0.28 -0.22
(-3.64) (-0.93) (-0.24) (-2.02) (-3.26)

This table reports the average monthly returns of the M10 (the winner)—the M1 (the loser) portfolios during
the post-holding period from month t ? 13 to month t ? 60. For each month t from January 1965 until
December 2004, we group stocks from the NYSE and AMEX into five portfolios based on the speed of
correction, and then into 10 momentum portfolios within each group of five portfolios. The estimation of the
speed of correction (SP) is described in the text. SP1 (SP5) constitutes the portfolio with the fastest (slowest)
speed of correction. We control for auto-correlation using the Newey and West (1987) technique for t
statistics

123
J. Hur, V. Singh

group with the fastest speed of correction (SP1) shows virtually no reversal. This evidence
is consistent with momentum profits being a consequence of both investors’ underreaction
and overreaction to firm-specific information.
To understand the post-holding period returns more closely, we also show in Panel A of
Table 6 the average monthly returns for the winner-minus-loser stocks sorted based on the
speed of correction for periods ranging from 1 to 5 years after the holding periods.10 The
last column of Panel A of Table 6 shows that the returns monotonically decrease as the
speed of correction declines. For example, over 5 years past the holding period, the
winners-minus-loser portfolios have an average monthly return of negative 25 basis points
for the SP5 group of stocks, which have the slowest speed of correction. The corresponding
figure for the stocks with the highest speed of correction, the SP1 group, is a statistically
insignificant negative 6 basis points. The difference in the monthly average returns
between the SP1 and SP5 groups of stocks is statistically significant as well. We observe a
similar pattern over other post-holding periods as well, though the difference in monthly
average returns between the SP1 and SP5 groups is not statistically significant from month
37 to month 48. These patterns are consistent with the suggestion that the stocks with the
slower speeds of correction overreact beyond the incorporation of firm-specific informa-
tion, leading to reversals in the post-holding period. When we examine specific quintiles of
the speed of correction, the reversal seems to be statistically significant for groups SP3
through SP5. The magnitudes of the reversals for groups SP1 and SP2 are statistically
indistinguishable from zero. This indicates that around 40 % of the stocks in momentum
portfolios (in quintiles SP1 and SP2) with a faster speed of correction simply correct for the
underreaction to firm-specific information. However, the momentum displayed by 60 % of
the stocks in momentum portfolio (in quintiles SP3 through SP5) could be ascribed to an
overreaction by investors during the holding period.
Jegadeesh and Titman (2001) found that while the momentum profits are strong for the
whole sample, the strong reversals are confined to the first period of 1965–1981 only. To
test the robustness of our results, we split our sample into two equal periods of 1965–1984,
and 1985–2004. The results for the two periods are shown in Panels B and C. Consistent
with the findings of Jegadeesh and Titman (2001), the reversals are stronger for the period
1965–1984. In Panel B, we observe statistically significant reversals in all but the SP1
group of stocks. However, Panel C investigates the more recent period 1985–2005; here,
the reversals are confined to only the SP5 group of stocks. Nevertheless, our primary
finding for the overall period is that those stocks with a slower speed of correction
overreacted beyond the incorporation of firm-specific information, leading to reversals
during the post-holding period, continues to hold in the two sub-periods. The average
monthly returns of the stocks with the slowest speed of correction (group SP5) show
statistically significant negative returns, while the average monthly returns of stocks with
the fastest speed of correction (group SP1) are no different from zero during the recent
period. The difference between the average monthly returns of the stocks in group SP5 and
those in group SP1 is statistically significant as well.
One interpretation of our results is that transactions by investors initially underreact to
short-term information such as firms’ short-term prospects such as earnings forecasts/actual
earnings announcement causing the momentum in the short run. However they may
extrapolate the short-term information to assess the long-term prospects of the firm.
However the long-term prospects assessment is more difficult because they require long-run

10
Since our sample ends in 2009, we can only examine 60 months post-holding returns for those
momentum portfolios formed as of December of 2004.

123
Reexamining momentum profits

information much of which is ambiguous in nature causes investors to overreact and causes
long-terms reversals in the returns. Overall, our results indicate that the momentum of the
stock returns is driven by both an underreaction and overreaction to firm-specific infor-
mation, although the role of overreaction in driving momentum profits has dissipated in
more recent times.

6 Conclusion

Using a new measure that directly captures the speed with which stocks react with delay to
firm-specific information, we find that momentum profit could be explained by investors
underreacting/overreacting to firm-specific information. However underreaction seems to be
the primary cause, especially during more recent periods. We also find that the reaction with
delay to firm-specific information is partly related to firm attributes and investor inattention.
Although we find that firm attributes and investor inattention partially explain the
delayed reaction to firm-specific information, it is not entirely clear as to why the reaction
is different for stocks with high and low past returns. Furthermore, momentum seems to be
caused mostly by underreaction, particularly during the most recent period; overreaction
also played a role in it. Our study cannot explain why investors overreact to information for
one set of stocks but underreact to information for another set of stocks. Finally, the
presence of momentum itself remains a big puzzle in efficient markets. We hope future
research will address these issues.

Acknowledgments We are grateful for the valuable suggestions from the Editor, C. F. Lee and an
anonymous referee that have significantly improved the quality of the paper. We also thank discussants and
participants of Eastern Finance Association meetings of 2011, and Midwest Finance Association meetings of
2011 for their helpful comments.

References

Alexander C (2008) Market risk analysis: practical financial econometrics, vol II. Wiley, New York
Ali A, Trombley M (2006) Short sales constraints and momentum in stock returns. J Bus Financ Account
33(3 & 4):587–615
Allen F, Morris S, Shin H (2006) Beauty contests, bubbles and iterated expectations in asset markets. Rev
Financ Stud 19(3):719–752
Antoniou A, Lam H, Paudyal K (2007) Profitability of momentum strategies in international markets: the
role of business cycle variables and behavioral biases. J Bank Financ 31(3):955–972
Arena M, Haggard K, Yan X (2008) Price momentum and idiosyncratic volatility. Financ Rev
43(2):159–190
Asem E (2009) Dividends and price momentum. J Bank Financ 33(3):486–494
Barber B, Odean T (2008) All that glitters: the effect of attention and news on the buying behavior of
individual and institutional investors. Rev Financ Stud 21(2):785–818
Barberis N, Thaler R (2003) A survey of behavioral finance. Handbook of the Economics of Finance In:
Constantinides GM, Harris M, Stulz RM (eds), Handbook of the economics of finance, 1st edn, vol 1.
Elsevier, pp 1053–1128
Barberis N, Shleifer A, Vishny R (1998) A model of investor sentiment. J Financ Econ 49(3):307–343
Berk J, Green R, Naik V (1999) Optimal investment, growth options, and security returns. J Financ
54(5):1553–1607
Chordia T, Shivakumar L (2002) Momentum, business cycle, and time-varying expected returns. J Financ
57(2):985–1019
Cochrane JH (1996) A cross-sectional test of an investment-based asset pricing model. J Polit Econ
104(3):572–621
Cohen L, Frazzini A (2008) Economic links and predictable returns. J Financ 63(4):1977–2011

123
J. Hur, V. Singh

Conrad J, Kaul G (1998) An anatomy of trading strategies. Rev Financ Stud 11(3):489–519
Daniel K, Hirshleifer D, Subrahmanyam A (1998) Investor psychology and security market under- and
overreactions. J Financ 53(6):1839–1886
De Bondt W, Thaler R (1985) Does the stock market overreact? J Financ 40(3):793–805
De Bondt W, Thaler R (1987) Further evidence on investor overreaction and stock market seasonality.
J Financ 42(3):557–581
De Bondt W, Thaler R (1990) Do security analysts overreact? Am Econ Rev 80(2):52–57
DellaVigna S, Pollet J (2009) Investor inattention and Friday earnings announcements. J Financ
64(2):709–749
DeLong B, Shleifer A, Summers L, Waldmann R (1990) Positive feedback investment strategies and
destabilizing rational speculation. J Financ 45(2):379–395
Dyck A, Zingales L (2003). The media and asset prices. Working paper
Fama E (1998) Market efficiency, long-term returns, and behavioral finance. J Financ Econ 49(3):283–306
Fama E, French K (1996) Multifactor explanations of asset pricing anomalies. J Financ 51(1):55–84
Frazzini A (2006) The disposition effect and under-reaction to news. J Financ 61(4):2017–2046
Gabaix X, Laibson D, Moloche G, Weinberg S (2006) The allocation of attention: theory and evidence. Am
Econ Rev 96(4):1043–1068
Greene W (2008) Econometric analysis, 6th edn. Prentice Hall, Upper Saddle River
Griffin J, Ji X, Martin J (2003) Momentum investing and business cycle risk: evidence from pole to pole.
J Financ 58(6):2515–2547
Grinblatt M, Han B (2005) Prospect theory, mental accounting, and momentum. J Financ Econ
78(2):311–339
Grundy B, Martin S (2001) Understanding the nature of the risks and the source of the rewards to
momentum investing. Rev Financ Stud 14(1):29–79
Haugen R, Baker N (1996) Commonality in the determinants of expected stock returns. J Financ Econ
41(3):401–439
Hirshleifer D, Teoh SH (2003) Limited attention, information disclosure, and financial reporting. J Account
Econ 36(1–3):337–386
Hirshleifer D, Lim S, Teoh SH (2009) Driven to distraction: extraneous events and underreaction to earnings
news. J Financ 64(5):2289–2325
Hong H, Stein J (1999) A unified theory of underreaction, momentum trading, and overreaction in asset
markets. J Financ 54(6):2143–2184
Hong H, Stein J (2007) Disagreement and the stock market. J Econ Perspect 21(4):109–128
Hong H, Lim T, Stein J (2000) Bad news travels slowly: size, analyst coverage, and the profitability of
momentum strategies. J Financ 55(1):265–295
Hou K, Moskowitz T (2005) Market frictions, price delay and the cross-section of expected returns. Rev
Financ Stud 18(3):981–1020
Huberman G, Regev T (2001) Contagious speculation and a cure for cancer: a nonevent that made stock
prices soar. J Financ 56(1):387–396
Jegadeesh N (1990) Evidence of predictable behavior of security returns. J Financ 45(3):881–898
Jegadeesh N, Titman S (1993) Returns to buying winners and selling losers: implications for stock market
efficiency. J Financ 48(1):65–91
Jegadeesh N, Titman S (2001) Profitability of momentum strategies: an examination of alternative expla-
nations. J Financ 56(2):699–720
Jegadeesh N, Titman S (2005) Momentum. In: Thaler RM (ed) Advances in behavioral finance, vol II.
Princeton University Press, Princeton
Jiang G, Li D, Li G (2012) Capital investment and momentum strategies. Rev Quant Financ Account
39(2):165–188
Johnson T (2002) Rational momentum effects. J Financ 57(2):585–608
Kahneman D (1973) Attention and effort Englewood and Cliffs. Prentice Hall, NJ
Kremers J, Ericsson N, Dolado J (1992) The power of cointegration tests. Oxf Bull Econ Stat 54(3):325–348
Lee C, Swaminathan B (2000) Price momentum and trading volume. J Financ 55(5):2017–2069
Lehmann B (1990) Fads, martingales, and market efficiency. Q J Econ 105(1):1–28
Lesmond DA, Schill MJ, Zhou C (2004) The illusory nature of momentum profits. J Financ Econ
71(2):349–380
Liu L, Zhang L (2008) Momentum profits, factor pricing, and macroeconomic risk. Rev Financ Stud
21(6):2417–2448
Lo A, MacKinlay A (1990) When are contrarian profits due to stock market overreaction? Rev Financ Stud
3(2):175–205

123
Reexamining momentum profits

Miffre J, Rallis G (2007) Momentum strategies in commodity futures markets. J Bank Financ
31(6):1863–1886
Newey W, West K (1987) A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent
covariance matrix. Econometrica 55(3):703–708
Peng L, Xiong W (2006) Investor attention, overconfidence and category learning. J Financ Econ
80(3):563–602
Rouwenhorst K (1998) International momentum strategies. J Financ 53(1):267–284
Serban A (2010) Combining mean reversion and momentum trading strategies in foreign exchange markets.
J Bank Financ 34(11):2720–2727
Shefrin H, Statman M (1985) The disposition to sell winners too early and ride losers too long: theory and
evidence. J Financ 40(3):777–790
Shleifer A, Vishny R (1997) The limits of arbitrage. J Financ 52(1):35–55
Verardo M (2009) Heterogeneous beliefs and momentum profits. J Financ Quant 44(4):795–822
Wang J, Wu Y (2011) Risk adjustment and momentum sources. J Bank Financ 35(6):1427–1435
Wang Ching-Ping, Huang Hung-Hsi, Tu Kai-Jei (2012) Unsystematic risk explanation to momentum profits
in Taiwan. Rev Pac Basin Financ Mark Polic 15(1):115–143
Wu Y (2011) Momentum trading, mean reversal and overreaction in Chinese stock market. Rev Quant
Financ Account 37(3):301–323
Xiafei L, Miffre J, Brooks C, O’Sullivan N (2008) Momentum profits and time-varying unsystematic risk.
J Bank Financ 32(4):541–558
Yeh C–C, Li C-A (2011) Investor psychological and behavioral bias: Do high sentiment and momentum
exist in the china stock market? Rev Pac Basin Financ Mark Polic 14(3):429–448
Yu S (2012) New empirical evidence on the investment success of momentum strategies based on relative
stock prices. Rev Quant Financ Account 39(1):105–121

123

You might also like