Short-term Momentum

Mamdouh Medhat
Cass Business School, City, University of London

Maik Schmeling
Goethe University Frankfurt and Centre for Economic Policy Research
(CEPR)

We document a striking pattern in U.S. and international stock returns: double

sorting on the previous month’s return and share turnover reveals significant
short-term reversal among low-turnover stocks, whereas high-turnover stocks
exhibit short-term momentum. Short-term momentum is as profitable and as
persistent as conventional price momentum. It survives transaction costs and is
strongest among the largest, most liquid, and most extensively covered stocks.
Our results are difficult to reconcile with models imposing strict rationality
but are suggestive of an explanation based on some traders underappreciating
the information conveyed by prices. (JEL G12, G14)

Received November 1, 2019; editorial decision February 26, 2021 by Editor

Ralph Koijen.

A key stylized fact in the asset pricing literature is that stock returns
exhibit reversal at short horizons of 1 month (Jegadeesh 1990) but
continuation—or momentum—at longer horizons between 2 and 12
months (Jegadeesh and Titman 1993, 2001). In this paper, we show
that reversal and momentum coexist with striking magnitudes at the 1-
month horizon. While the previous month’s thinly traded stocks exhibit
a strong short-term reversal e↵ect, the previous month’s heavily traded
stocks exhibit an almost equally strong continuation e↵ect, which we

We appreciate helpful comments from Ralph Koijen and two anonymous referees. We also
thank Cli↵ Asness, Pedro Barroso, John Campbell, Giovanni Cespa, Zhi Da, Alexander
Hillert, Alexandre Jeanneret, Christian Julliard, Tim Krönke, Albert Menkveld, Markus
Nöth, Richard Payne, Lasse Pedersen, Chris Polk, Angelo Ranaldo, Savina Rizova, Ioanid
Roşu, Lucio Sarno, Julian Thimme, Gyuri Venter, Michela Verardo, Christian Wagner,
Josef Zechner, and Irina Zviadadze; seminar participants at various business schools and
asset managers; and participants at the Chicago Quantitative Alliance 2018 fall conference,
the Research in Behavioral Finance 2018 conference, the London Empirical Asset Pricing
2019 fall workshop, the Swiss Society for Financial Market Research 2019 annual meeting,
the French Finance Association 2019 annual meeting, the London Business School 2019
summer symposium, and the 12th annual Hedge Fund Research Conference. We are
grateful for the Best Paper Award at the Chicago Quantitative Alliance (CQA) 2018
Academic Competition. All errors are our own. Send correspondence to Maik Schmeling,
[email protected].

Published by Oxford University Press on behalf of The Society for Financial Studies 2021.
doi:10.1093/rfs/Sample Advance Access publication XXXX XX, 2021

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Coexistence
Coexistenceofofreversal
reversaland
andmomentum
momentumininone-month returns
1-month returns
20%
16.44%

10%
Average excess return (annualized)

0%

-3.39%

-10%

-16.92%
-20%
Short-term reversal* Short-term reversal Short-term momentum
[Lowest turnover decile] [All stocks] [Highest turnover decile]

Figure 1
Coexistence of reversal and momentum in 1-month returns
This figure shows the average excess returns to three long-short strategies that buy
the previous month’s winners and short the losers among U.S. stocks. The conventional
short-term reversal strategy (Jegadeesh, 1990) is from a univariate decile sort on the
previous month’s return using NYSE breakpoints. The short-term reversal* and short-
term momentum strategies trade the corner portfolios from double decile sorts on the
previous month’s return and the previous month’s share turnover using NYSE breakpoints;
short-term reversal* trades only in the lowest-turnover decile while short-term momentum
trades only in the highest-turnover decile. The portfolios underlying all three strategies
are value weighted and rebalanced at the end of each month. The performance of the
portfolios underlying the short-term reversal* and short-term momentum strategies is
provided in Table 1. The sample is all common nonfinancial shares on the NYSE, AMEX,
and NASDAQ exchanges and covers July 1963 to December 2018.

dub short-term momentum. Figure 1 illustrates our main results for the
United States.
To obtain our main results, we double decile sort stocks on the
previous month’s return and the previous month’s share turnover using
NYSE breakpoints and form value-weighted portfolios. We find, first,
significant short-term reversal among stocks with low share turnover.
The short-term reversal* strategy, which buys the previous month’s
winners and shorts the losers within the lowest turnover decile, generates
a negative and significant average return of 16.9% per annum (Figure
1, left bar). Second, our key finding is that short-term reversal is reversed
among stocks with high share turnover. The short-term momentum

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

strategy, which buys the previous month’s winners and shorts the losers
within the highest turnover decile, generates a positive and significant
average return of +16.4% per annum (Figure 1, right bar). We show
that both strategies generate significant abnormal returns relative to
the standard factor models currently applied in the literature. We also
show that short-term momentum persists for 12 months and is strongest
among the largest and most liquid stocks. Finally, we show that our main
findings extend to 22 developed markets outside the United States.
We provide additional results and robustness tests that shed light
on the economic drivers of our findings. First, skipping the last few
days of the formation month implies stronger short-term momentum (on
average 22.0% per annum) because this mitigates end-of-month liquidity
trading (Etula, Rinne, Suominen, and Vaittinen 2020). Second, short-
term momentum survives conservative estimates of transaction costs
(Novy-Marx and Velikov 2015). Third, while short-term momentum
is naturally related to conventional price- and earnings-momentum
strategies, its correlations with such strategies are moderate and do
not appear to fully capture its average return, at least when judged
by spanning tests. Fourth, short-term momentum does not appear to
be driven by momentum in industry returns (Moskowitz and Grinblatt
1999) or by momentum in factor returns (Ehsani and Linnainmaa 2020).
Fifth, it is not a result of any correlation between share turnover and
size, liquidity, or volatility. Lastly, it exhibits far less crash risk than
does conventional price momentum (Daniel and Moskowitz 2016).
Short-term momentum is difficult to reconcile with rational expec-
tations equilibrium (REE) models of the volume-return relation. In
Campbell, Grossman, and Wang (1993), noninformational trading due
to liquidity demand causes temporary price pressure when absorbed
by liquidity suppliers. As a result, returns coupled with high volume
will subsequently reverse. This runs opposite to short-term momentum.
In Wang (1994) and Llorente et al. (2002), informed trading due to
private information causes persistent price movements that counteract
temporary price pressure. Hence, among stocks with a high degree of
information asymmetry, returns coupled with high volume will reverse
less and may even continue. This would be an explanation for short-term
momentum if it were driven by high-information-asymmetry stocks.
However, while Wang (1994) and Llorente et al. (2002) argue that such
stocks should be small, illiquid, and have low analyst coverage, short-
term momentum is strongest among the largest, most liquid, and most
extensively covered stocks.1

1
Wang (1994) argues that contemporaneous evidence “is consistent with our model if we
assume that there is more information asymmetry in the market for small-size firms than
for large-size firms” (p. 151). Llorente et al. (2002) proxy for higher information asymmetry
using smaller size, lower liquidity, and lower analyst coverage.

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

The literature on “boundedly rational” traders suggests an alternative

mechanism that potentially explains short-term momentum: under-
inference from prices. When some traders fail to fully infer others’
information from prices, expected volume is higher and prices underreact
to the available information relative to the case with solely rational
traders. Underreaction in turn causes persistent price movements that
counteract temporary price pressure.2 Hence, an explanation based on
bounded rationality suggests that short-term momentum should be
stronger among stocks where any underinference from prices is not
overwhelmed by noninformational trading. The cross-sectional variation
in short-term momentum is suggestive of such an explanation: It is
strongest among the largest and most liquid stocks, whose returns tend
to be less a↵ected by temporary price pressure (Avramov, Chordia, and
Goyal 2006; Nagel 2012; Hendershott and Menkveld 2014), and it is also
stronger among stocks with greater dispersion in analysts’ forecasts, a
common proxy for disagreement among traders (Diether, Malloy, and
Scherbina 2002; Verardo 2009; Banerjee 2011).
Lastly, to assist in di↵erentiating between fully and boundedly
rational explanations, we investigate how volume a↵ects the ability
of realized returns to predict firms’ fundamentals. Intuitively, “fully
rational” volume ultimately emanates from noninformational trading.
By contrast, “boundedly rational” volume reflects both underreaction
to the available information and any noninformational trading. Hence,
higher volume should decrease the ability of realized returns to predict
fundamentals if all traders are fully rational, but should increase
it if underinference has a detectable e↵ect. Empirically, we find in
cross-sectional regressions that the interaction of 1-month returns and
turnover predicts the following year’s growth in gross profits and
earnings with a positive and significant slope coefficient.
Because our findings are at the 1-month horizon, they di↵er from
those in the extant literature on the volume-return relation, which has
primarily focused on the weekly horizon. Conrad, Hameed, and Niden
(1994) find that, among NASDAQ stocks, higher growth in the number
of transactions is associated with more reversal in weekly returns. In
contrast, Cooper (1999) finds that, among the largest NYSE/AMEX
stocks, higher growth in trading volume is associated with less reversal
in weekly returns.3 Despite their opposing findings, both papers find

2
Models of boundedly rational traders have been invoked to explain why volume greatly
exceeds what is expected under REE and why patterns in returns and volume are tightly
linked (see, e.g., Hong and Stein 2007; French 2008). Underreaction is often invoked as
the mechanism underlying return continuation in models that relax the strict rationality
assumption (see, e.g., Fama 1998; Daniel and Hirshleifer 2015).
3
In his explanation of the opposing findings, Cooper (1999) conjectures that “in the context
of Wang’s (1994) model, it may be that in periods of large price movements, high volume

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

only limited evidence of continuation. We conjecture that this is because

temporary price pressure dominates other e↵ects at the weekly horizon.
Indeed, Avramov, Chordia, and Goyal (2006) find that, controlling
for illiquidity, higher turnover implies more reversal at the weekly
horizon but less reversal at the monthly horizon. They do not, however,
document monthly continuation among high-turnover stocks, as we do,
nor do they study their findings in the context of momentum e↵ects.
Moreover, we find that the returns to weekly short-term momentum
strategies are on average negative (in line with our price-pressure
conjecture) but become more negative with illiquidity, which shows
that our evidence against the high-information-asymmetry mechanism
also holds at the weekly horizon. We contribute to this literature by
documenting the coexistence of significant reversal and momentum
e↵ects in 1-month returns separated by the level of turnover.4,5
Our paper is also related to a recent literature that sheds new light
on the economic sources of momentum. Goyal and Jegadeesh (2018)
find that time-series momentum, once stripped of its implicit time-
varying investment in the market, has little incremental power beyond
cross-sectional momentum in pricing tests (see also Huang et al., 2020).
In turn, Ehsani and Linnainmaa (2020) find that conventional price
momentum as well as industry momentum are spanned by momentum in
factor returns. Theoretically, Luo, Subrahmanyam, and Titman (2021)
propose an explanation for momentum based on sequential learning with
overconfidence and skepticism, that is, a form of bounded rationality. We
contribute to this literature by reconciling the strands of the literature
that document reversal at the 1-month horizon but momentum at the
12-2 months horizon. In particular, we show that momentum coexists

for smaller (larger) stocks represents a higher percentage of liquidity (informed) traders,
resulting in greater subsequent reversals (continuations)” (p. 921). This contrasts with
Wang’s (1994) argument that high-information-asymmetry stocks should be smaller, not
larger (see our footnote 1). While our evidence in principle supports Cooper’s conjecture,
we will argue in Section 3 that it is for reasons outside of the Wang model.
4
Our asset pricing tests employ value-weighted portfolios from sorts based on NYSE
breakpoints as well as weighted least squares (WLS) cross-sectional regressions with
market capitalization as the weight. Llorente et al. (2002) use stock-by-stock, full-sample
regressions of daily returns on the previous day’s return and its interaction with de-trended
log-turnover in their tests. Based on ex post cross-sectional variation in the interaction
coefficients, they argue that high-volume days are followed by return continuation among
the smallest and most illiquid stocks. Since their horizon is daily and their findings are not
based on investable portfolios, their tests are inherently di↵erent from ours. Furthermore,
the fact that any such daily return continuation is limited to the smallest and most illiquid
stocks raises questions about its economic significance.
5
Asness (1995) finds stronger 1-month reversal among low-volume stocks but does not
document continuation among high-volume ones. Lee and Swaminathan (2000) study how
volume a↵ects the relation between the value and momentum e↵ects, but consider longer
formation periods of at least 3 months. Gervais, Kaniel, and Mingelgrin (2001) find that
stocks with unusually high volume outperform those with unusually low volume. Cespa,
Gargano, Riddiough, and Sarno (2020) find stronger reversal in daily returns among low-
volume currencies but do not find continuation among high-volume currencies.

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

with reversal at the 1-month horizon, albeit confined to the stocks with
the highest trading activity. In addition, we provide guidance as to
which broader class of models (fully vs. boundedly rational) that is most
plausible as an explanation for short-term momentum.

1. Main Results
This section presents our main results. We first show that double sorting
on the previous month’s return and turnover reveals strong reversal
among low-turnover stocks but almost equally strong momentum among
high-turnover stocks. We then show that standard factors cannot
account for either e↵ect; that the returns to the short-term momentum
strategy persist for 12 months after formation; and that short-term
momentum is strongest among the largest stocks. Finally, we show that
our main results also hold internationally.

1.1 Data and key variables

Our main sample is all NYSE/AMEX/NASDAQ stocks on both CRSP
and Compustat. We keep only ordinary common shares (share codes
10 or 11). Following Fama and French (1993, 2015), we impose a six
month lag between annual accounting data and subsequent returns to
avoid look-ahead bias. Hence, if a firm’s fiscal year ends in December
of calendar year t 1, we assume that these data are publicly available
at the end of June of calendar year t. Following Hou, Xue, and Zhang
(2015, 2020), we exclude financial firms, although retaining these does
not a↵ect our results. Our main sample covers July 1963 to December
2018. We will later show that our main results extend back to 1926.
Our asset pricing tests relate 1-month returns to the previous month’s
return (r1,0 ) and trading volume. A 1-month holding period is standard
in the literature and the previous month’s return is the signal underlying
Jegadeesh’s (1990) short-term reversal strategy. To be consistent, we use
the same 1-month horizon when measuring trading volume. Following Lo
and Wang (2000) and Avramov, Chordia, and Goyal (2006), we deflate
the total volume of trades by the number of shares outstanding to obtain
share turnover (TO 1,0 ). Following Gao and Ritter (2010), we adjust the
trading volume of NASDAQ stocks prior to 2004 to ensure comparability
with NYSE and AMEX stocks, although our results are the same with
or without this adjustment.6 We will later show that our findings are
robust to using longer formation periods of up to 6 months, but that
the 1-month formation period produces the strongest results.

6
Prior to February 2001, we divide NASDAQ volume by 2.0. From February 2001 to
December 2001, we divide by 1.8. From January 2002 to December 2003, we divide by
1.6. From January 2004 and onward, NASDAQ volume no longer di↵ers from NYSE and
AMEX volume, and we apply no adjustment.

Electronic copy available at: https://ssrn.com/abstract=3150525

 Table 1
Double sorts on the previous month’s return and turnover

r1,0 deciles r1,0 strategies

Low 2 3 4 5 6 7 8 9 High E[re ] ↵FF6 ↵q

TO 1,0 deciles
Portfolio excess return
Low 1.28 1.23 0.99 0.85 0.70 0.80 0.59 0.74 0.26 0.14 1.41 1.45 1.43
( 7.13) ( 6.19) ( 5.95)
2 1.54 1.22 0.98 0.99 1.05 0.98 0.70 0.69 0.57 0.35 1.19 1.31 1.34
( 4.61) ( 4.04) ( 4.21)
3 1.71 1.53 0.96 1.11 0.99 0.94 0.75 0.60 0.64 0.36 1.34 1.62 1.66
( 5.02) ( 5.61) ( 4.87)
Short-term Momentum

4 1.51 1.35 1.43 0.98 1.10 1.07 0.81 0.83 0.64 0.65 0.85 1.02 0.91
( 3.63) ( 4.15) ( 2.92)
5 1.11 1.10 1.26 1.17 1.10 1.00 0.60 0.92 0.90 0.66 0.45 0.63 0.51
( 1.94) ( 2.29) ( 1.54)
6 1.26 1.38 1.40 1.14 1.00 1.14 1.12 1.19 0.78 0.67 0.59 0.60 0.41
( 2.50) ( 2.35) ( 1.20)
7 1.39 1.06 1.12 1.22 0.84 1.05 0.83 0.98 0.96 0.73 0.67 0.85 0.96
( 2.52) ( 2.55) ( 2.37)
8 0.92 1.17 1.25 0.99 1.12 1.02 1.02 0.85 0.82 1.15 0.23 0.13 0.21
(0.85) (0.47) (0.60)
9 0.71 1.37 1.29 1.24 1.21 1.22 1.00 1.12 1.02 0.75 0.05 0.00 0.19
(0.21) (0.01) (0.55)
High 0.00 0.83 1.14 1.08 1.03 0.78 1.01 1.16 0.99 1.36 1.37 1.37 1.65
(4.74) (4.22) (4.47)
TO 1,0 strategies

E[re ] 1.28 0.41 0.15 0.23 0.33 0.01 0.42 0.42 0.73 1.50
( 5.04) ( 1.75) (0.59) (0.80) (1.34) ( 0.05) (1.58) (1.54) (2.76) (5.46)

Electronic copy available at: https://ssrn.com/abstract=3150525

↵FF6 1.26 0.31 0.18 0.17 0.29 0.02 0.34 0.57 0.81 1.56
( 4.71) ( 1.33) (0.79) (0.73) (1.32) ( 0.09) (1.48) (2.12) (3.29) (5.34)
↵q 1.39 0.14 0.21 0.05 0.33 0.02 0.28 0.42 0.91 1.70
( 4.98) ( 0.57) (0.82) (0.18) (1.39) (0.07) (1.17) (1.41) (3.15) (5.07)
This table shows portfolios double sorted on the previous month’s return (r1,0 ) and turnover (TO 1,0 ). We use conditional sorts into deciles based on

7
NYSE breakpoints, first on r1,0 and then on TO 1,0 . Portfolios are value weighted and rebalanced at the end of each month. The table also shows
the performance of long-short strategies across the deciles. Test statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation.
Time-series averages of the portfolios’ characteristics are provided in Table A.1 in the appendix. The sample excludes financial firms. Data are at the
monthly frequency and cover July 1963 to December 2018, except for the q-factors, which are available from January 1967.
 The Review of Financial Studies / v 00 n 0 2021

1.2 Double sorts on the previous month’s return and

turnover
Table 1 shows the average excess returns to portfolios double sorted
on the previous month’s return (r1,0 ) and turnover (TO 1,0 ). We use
conditional decile sorts based on NYSE breakpoints, first on r1,0 and
then on TO 1,0 within r1,0 deciles. Hence, we first separate winners
from losers and then separate heavily traded stocks from thinly traded
ones among winners and losers. Portfolios are value weighted and
rebalanced at the end of each month. We use deciles to account for
any nonlinearities and conditional sorts because independent ones give
a few empty portfolios before July 1969.7 The table also shows the
performance of high-minus-low strategies within each decile. Abnormal
returns are relative to Fama and French’s (2015) five-factor model
plus the momentum factor (FF6) and Hou, Xue, and Zhang’s (2015)
q-factor model. Test statistics are adjusted for heteroscedasticity and
autocorrelation (Newey and West, 1987).
The double sorts reveal a striking pattern in 1-month returns. The
strategy that buys the previous month’s winners and shorts the losers
within the lowest turnover decile yields 1.41% per month with t =
7.13, which is evidence of strong reversal in 1-month returns for low-
turnover stocks. We label this e↵ect “short-term reversal*” (STREV*)
to distinguish it from the conventional short-term reversal e↵ect. In stark
contrast, the winner-minus-loser strategy within the highest turnover
decile yields +1.37% per month with t = 4.74, which is evidence of strong
continuation in 1-month returns for high-turnover stocks. Consequently,
we label this e↵ect “short-term momentum” (STMOM).8 The abnormal
returns of the STMOM and STREV* strategies are as large and as
strong as the strategies’ average returns. On average, the STMOM
strategy generates all its profits on the long side.
The double sorts also reveal significant 1-month reversal in the
bottom-seven turnover deciles, although the magnitude of the reversal
e↵ect generally weakens with turnover before switching signs and

7
The models of Campbell, Grossman, and Wang (1993), Wang (1994), and Llorente et al.
(2002) suggest a nonlinear relation between expected and realized returns for a given level
of volume. Conditional sorts allow us to use the full sample period, but Table A.2 in the
appendix shows similar results for independent sorts starting from July 1969. Sorting first
on returns and then on turnover produces the largest spreads in holding-period returns,
but the results are similar for the reverse sorting order. The average cross-sectional rank
correlation between 1-month returns and turnover is 9.99% (t = 8.29), so the interpretation
of the double sorts is largely una↵ected by the sorting order. Our use of conditional decile
sorts based on NYSE breakpoints follows Fama and French (1992). Fama and French
(2015, 2016) also use conditional sorts based on NYSE breakpoints in some of their tests
and Ken French maintains several decile double sorts on his website.
8
For comparison, the conventional STREV strategy (from a univariate decile sort on r1,0 )
yields 0.28% with t = 1.68 over our sample; see also Hou, Xue, and Zhang (2015, 2020).
The corresponding conventional momentum strategy (from a univariate decile sort on the
prior 12-2 month returns) yields 1.21% per month with t = 4.77.

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

Table 2
Short-term momentum’s factor exposures and abnormal returns

Intercepts, slopes, and test statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]

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

variable
Intercept 1.37 1.56 1.37 2.21 1.41 1.29 1.45 0.93
(4.74) (5.54) (4.22) (7.54) ( 7.13) ( 6.57) ( 6.19) ( 4.43)
MKT 0.38 0.35 0.14 0.25 0.16 0.01
( 4.53) ( 3.97) ( 1.96) ( 3.67) ( 2.87) ( 0.17)
SMB 0.00 0.04 0.30 0.19
( 0.03) (0.44) ( 2.59) ( 2.40)
HML 0.03 0.08 0.00 0.11
(0.12) (0.45) ( 0.04) (0.91)
RMW 0.40 0.34 0.09 0.14
( 1.56) ( 2.80) ( 0.86) ( 1.29)
CMA 0.18 0.20 0.05 0.02
(0.66) ( 0.98) (0.24) (0.07)
MOM 0.33 0.05 0.30 0.13
(1.89) (0.54) (4.05) (2.80)
STREV 1.51 0.97
( 14.38) ( 11.11)
LTREV 0.34 0.16
(2.68) ( 1.12)
PSLIQ 0.08 0.06
(0.97) ( 0.99)
Adj. R2 3.8% 6.8% 34.0% 3.6% 9.8% 35.1%
This table shows time-series regressions for the short-term momentum (STMOM) and
short-term reversal* (STREV*) strategies from Table 1. The explanatory variables are
the factors from Fama and French’s (2015) five-factor model in addition to the momentum
factor (MOM), the two reversal factors (STREV and LTREV), and the traded liquidity
factor (PSLIQ). Test statistics (in parentheses) are adjusted for heteroscedasticity and
autocorrelation. Data are at the monthly frequency and cover July 1963 to December
2018, except for specifications employing PSLIQ, which is available from January 1968.

becoming significantly positive in the highest turnover decile. Hence,

momentum coexists with reversal at the 1-month horizon, although it
is confined to the highest turnover decile in these benchmark results.
Nonetheless, we will later show that the stocks that tend to exhibit
short-term momentum are among the largest and most liquid, which
emphasizes the economic importance of the e↵ect. In addition, we will
show that skipping the end of the formation month implies not only
stronger short-term momentum but also significant continuation in the
ninth turnover decile. As a result, (independent) quintile double sorts are
in fact sufficient to document short-term momentum when skipping the
end of the formation month. Lastly, we will show that the results of Table
1 are not limited to the United States but extend to 22 international
developed markets.

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

1.3 Factor exposures and abnormal returns

Table 2 shows spanning tests for the STMOM and STREV* strategies.
The explanatory variables are the FF6 factors as well as the conventional
1-month short-term reversal factor (STREV); the 60-13 month long-
term reversal factor (LTREV; see De Bondt and Thaler, 1985); and
Pástor and Stambaugh’s (2003) traded liquidity factor (PSLIQ).
The STMOM strategy has a significantly negative market loading and
a marginally significant, positive loading on the conventional momentum
factor. It does not load significantly on the remaining FF6 factors.
Controlling for the conventional reversal factors increases STMOM’s
abnormal return to 2.21% per month with t = 7.54. The STREV*
strategy has only modest loadings on the FF6 factors, and controlling
for the conventional reversal factors does not fully capture its average
return. Neither strategy loads significantly on PSLIQ.9

1.4 Persistence and historical performance

Panel A of Figure 2 shows the average cumulative sums of post-
formation returns to the STMOM and STREV* strategies along with
95% confidence bands. The average cumulative performance of STREV*
is indistinguishable from zero just 3 months after formation. The short-
lived nature of the average returns to STREV* suggest it is capturing
the easing of strong but temporary price pressure among low-turnover
stocks. In contrast, there is a much stronger drift in the average returns
to STMOM, which on average persist for 12 months after formation.
This is similar to conventional momentum strategies (see, e.g., Jegadeesh
and Titman, 2001) and suggests that STMOM is capturing strong and
persistent price movements among high-turnover stocks. In practical
terms, it means that traders can build up positions in STMOM relatively
slowly and reduce trading costs by rebalancing less frequently.
Panel B shows a time-series plot of cumulative sums of excess
returns to the two strategies. STREV* returns are remarkably consistent
until 2011, but have slightly tapered o↵ more recently. We suspect
that generally increasing market liquidity is responsible for this trend.
STMOM earned low returns during 1975-79, but has otherwise delivered
consistently positive returns with no other subsample significantly
a↵ecting its performance.

9
The STMOM strategy is not within the univariate span of the MOM factor (abnormal
return of 1.13% with t = 3.36). The converse is also true. The STREV* strategy is not
within the univariate span of the standard STREV factor (abnormal return of 0.91%
with t = 4.78), but the standard STREV factor is, in fact, within the univariate span of
the STREV* strategy (abnormal return of 0.03% with t = 0.31).

10

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

A
8
Short-term momentum

Average cumulative sum of excess returns (%)

Short-term reversal*

-2

1 6 12 18 24

Months after portfolio formation

B
1000
Short-term momentum
Short-term reversal*
Short-term reversal
Cumulative sum of excess returns (%)

600

200

-200

-600

-1000

1970 1980 1990 2000 2010 2020

Figure 2
Short-term momentum’s persistence and historical performance Panel A shows
the average cumulative sums of post-formation excess returns to each of the short-term
momentum and short-term reversal* strategies along with 95% confidence bands. Panel
B shows a time-series plot of cumulative sums of excess returns to the two strategies as
well as a conventional short-term reversal strategy. Data are at the monthly frequency
and cover July 1963 to December 2018.

1.5 Short-term momentum and size

Table 3 shows the performance of size-conditional STMOM and
STREV* strategies. In panel A, we use 2⇥3⇥3 conditional sorts on

11

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Table 3
Short-term momentum controlling for size

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]
controlling for size controlling for size

Size group E[re ] ↵FF6 ↵q E[re ] ↵FF6 ↵q

A. Size breakpoint is NYSE median

Small-cap 0.20 0.03 0.06 1.41 1.51 1.53
(0.89) (0.12) (0.17) ( 9.03) ( 7.59) ( 7.33)
Large-cap 0.62 0.52 0.64 0.84 0.92 0.83
(3.01) (2.25) (2.12) ( 5.45) ( 6.38) ( 5.00)
B. Size breakpoints are NYSE quintiles
Microcap 0.34 0.49 0.53 2.04 2.17 2.17
( 1.21) ( 1.35) ( 1.26) ( 10.28) ( 9.97) ( 8.56)
2 0.20 0.07 0.03 1.19 1.20 1.26
(0.77) (0.26) (0.07) ( 7.12) ( 7.01) ( 7.03)
3 0.40 0.29 0.44 1.20 1.28 1.19
(1.66) (1.09) (1.33) ( 6.93) ( 5.97) ( 5.32)
4 0.34 0.27 0.38 1.23 1.15 1.01
(1.40) (1.08) (1.22) ( 7.80) ( 6.71) ( 5.09)
Megacap 0.53 0.43 0.48 0.74 0.80 0.72
(2.53) (2.21) (2.01) ( 4.27) ( 4.77) ( 3.82)
C. 500 largest stocks
Largest 500 0.42 0.43 0.41 0.02 0.02 0.02
(4.91) (4.99) (4.19) ( 0.19) (0.18) ( 0.10)
This table shows the performance of short-term momentum (STMOM) and short-term
reversal* (STREV*) strategies constructed with a control for size (market capitalization).
In panels A and B, the strategies are from N ⇥3⇥3 conditional sorts on size and the
previous month’s return and turnover, in that order, where the breakpoints for returns
and turnover are the 20th and 80th percentiles for NYSE stocks. In panel A, N = 2 and
the size breakpoint is the NYSE median; in panel B, N = 5 and the size breakpoints are
NYSE quintiles. In panel C, the strategies are from 2⇥2 independent sorts on returns
and turnover among the 500 largest stocks by monthly market capitalization, where the
breakpoints for returns and turnover are the 250th rank. All portfolios are value weighted
and rebalanced at the end of each month. Test statistics (in parentheses) are adjusted for
heteroscedasticity and autocorrelation. The sample excludes financial firms. Data are at
the monthly frequency and cover July 1963 to December 2018, except when applying the
q-factors, which are available from January 1967.

the previous month’s size (market capitalization), return, and turnover,

in that order. The breakpoint for size is the median for NYSE stocks,
while the breakpoints for returns and turnover are the 20th and
80th percentiles for NYSE stocks. Portfolios are value weighted and
rebalanced at the end of each month. STREV* yields 0.84% per
month among large-caps (t = 5.45) but a considerably larger 1.41%
per month among small-caps (t = 9.03). STMOM, on the other hand,
yields a significant 0.62% per month among large-caps (t = 3.01) but an
insignificant 20 basis points per month among small-caps. The strategies’
abnormal returns tell a similar story.

12

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

In panel B, we use a finer size sort according to NYSE quintiles

but maintain the breakpoints for returns and turnover as in panel A.
STREV* yields an extremely large 2.04% per month among microcaps
(t = 10.28) but a comparably much smaller 0.74% per month among
megacaps (t = 4.27). In contrast, the average return to STMOM is
almost monotonically increasing with size: 0.34% per month among
microcaps (t = 1.21), around 0.30% per month for the intermediate
quintiles (t-statistics between 0.77 and 1.66), and a statistically and
economically significant 0.53% per month with t = 2.53 among megacaps.
Lastly, we consider the strategies’ performance when constructed
exclusively from the 500 largest stocks by monthly market capitalization.
At the end of each month, we rank these stocks on the previous month’s
return and form two portfolios using the 250th rank as the breakpoint.
Independently, we use the same procedure to form two portfolios based
on the previous month’s share turnover. The intersection produces four
portfolios, which are value weighted and rebalanced at the end of each
month. Panel C shows that the largest-500 STMOM strategy yields
0.42% per month with t = 4.91, along with equally large and strong
abnormal returns. The largest-500 STREV* strategy, however, yields
just 2 basis points per month.10
Table 3 shows that the lion’s share of STREV* returns come from
microcaps. This is not surprising, since short-term reversal is often
attributed to temporary price pressure (as a result of liquidity demand
or other microstructure issues), which tend to have a greater impact on
the returns of smaller stocks (Avramov, Chordia, and Goyal 2006; Nagel
2012; Hendershott and Menkveld 2014). By the same logic, it is not
surprising that STMOM derives the majority of its performance from
megacaps, whose returns tend to be much less a↵ected by temporary
price pressure.11
However, we stress that our results are not mechanically driven by
any correlation between turnover and size. While the two are positively
correlated in the cross section (average cross-sectional rank correlation
of 28.8% with t = 4.43), Table IA.2 in the Internet Appendix shows that
using turnover orthogonal to size does not a↵ect our main results from
Table 1 and that replacing turnover with size does not lead to any

10
The largest-500 STMOM strategy trades in an average of 129 and 121 stocks on its long
and short sides with average market capitalizations of $14.0 billion and $13.7 billion.
In December 2018 (the end of our sample), the top-five holdings on the long side were
Alphabet, Exxon, Walmart, AT&T, and Home Depot, while the top-five holdings on the
short side were Microsoft, Johnson & Johnson, Pfizer, Verizon, and Procter & Gamble.
11
Hendershott and Menkveld (2014), for instance, study price pressure across NYSE size
quintiles and find that “[p]rice pressure for the quintile of largest stocks is 17 basis points
with a half life of 0.54 days. For the smallest-stocks quintile, it is 118 basis points with a
half life of 2.11 days. These price pressures are roughly the size of the (e↵ective) bid-ask
spread” (p. 406).

13

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

continuation. We will later show that our results are also not driven by
any correlation between turnover and each of liquidity and volatility.
As we will discuss in Section 3, the results of Table 3 are at odds
with the “high-information-asymmetry” mechanism in the models of
Wang (1994) and Llorente et al. (2002), which counterfactually predicts
stronger STMOM performance among smaller stocks. In practical terms,
the results of Table 3 mean that STMOM is considerably cheaper
to implement and more scalable than is STREV*. In Section 2.2, we
will study the performance of the large-caps and megacaps STMOM
strategies net of transaction costs and with an implementation lag.

1.6 Cross-sectional regressions

Tables 1–3 illustrate our main results using strategies from corner
portfolios. Here, we use regressions to show that similar results hold
more broadly in the cross section.
Table 4 shows the average slopes from Fama and MacBeth (1973)
cross-sectional regressions of monthly returns on the previous month’s
return, turnover, and their interaction (r1,0 ⇥TO 1,0 ). As such, these
regressions mimic the approximate functional form of the theoretical
volume-return relation in the models of Campbell, Grossman, and
Wang (1993), Wang (1994), and Llorente et al. (2002). We control
for prior 12-2 months return (r12,2 ), book-to-market (log(B/M )), Size
(log of market capitalization as of prior June), cash-based operating
profitability (COP/A 1 ), and asset growth (dA/A 1 ). Independent
variables are trimmed at the 1st and 99th percentiles, then standardized
by their cross-sectional average and standard deviation. The interaction
is the product of the standardized variables. We show results for all
stocks as well as separately within NYSE size quintiles (Fama and French
2010) and we estimate the regressions using weighted least squares
(WLS) with market capitalization as the weight (Hou, Xue, and Zhang
2020).12

12
Following Fama and French (2015), we define book equity, B, as shareholder’s equity plus
deferred taxes minus preferred stock. In the definition of B, shareholder’s equity is SEQ.
If SEQ is missing, we substitute it by common equity (CEQ) plus preferred stock (defined
below), or else by total assets minus total liabilities (AT LT). Deferred taxes is deferred
taxes and investment tax credits (TXDITC) or else deferred taxes and/or investment tax
credit (TXDB and/or ITCB). Finally, preferred stock is redemption value (PSTKR) or
else liquidating value (PSTKL) or else carrying value (PSTK). B/M is book equity divided
by market capitalization (CRSP’s PRC times SHROUT) as of prior December, where the
lagging is to avoid taking unintentional positions in conventional momentum. dA/A 1 is
the year-over-year change in total assets divided by 1-year-lagged total assets. Following
Ball et al. (2016), COP is total revenue (REVT) minus cost of goods sold (COGS), minus
selling, general, and administrative expenses (XSGA), plus R&D expenditures (XRD,
zero if missing), minus the change in accounts receivable (RECT), minus the change
in inventory (INVT), minus the change in prepaid expenses (XPP), plus the change in
deferred revenue (DRC + DRLT), plus the change in trade accounts payable (AP), plus
the change in accrued expenses (XACC). All changes are annual, and missing changes are
set to zero.

14

Electronic copy available at: https://ssrn.com/abstract=3150525

 Table 4
Cross-sectional regressions to predict returns

Average slopes (⇥100) and test statistics (in parentheses)

from monthly WLS regressions of the form rit = 0t Xit +✏it

All Microcaps Size quintile 2 Size quintile 3 Size quintile 4 Megacaps

Independent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
variables

0.32 0.59 0.50 0.63 0.42 0.62 0.46 0.65 0.47 0.73 0.25 0.52
Short-term Momentum

r1,0
( 4.16) ( 8.05) ( 6.76) ( 8.21) ( 5.62) ( 8.29) ( 5.88) ( 8.82) ( 5.15) ( 7.28) ( 2.36) ( 5.08)
TO 1,0 0.06 0.05 0.10 0.10 0.02 0.01 0.01 0.01 0.06 0.02 0.09 0.01
(0.64) ( 0.82) (1.39) (1.31) ( 0.33) ( 0.17) ( 0.13) ( 0.21) (0.91) (0.25) (0.67) (0.12)
r1,0 ⇥TO 1,0 0.19 0.29 0.20 0.19 0.16 0.18 0.18 0.23 0.26 0.33 0.35 0.57
(5.45) (9.06) (6.21) (6.04) (5.23) (5.06) (5.66) (5.17) (5.10) (5.26) (4.01) (5.74)

r12,2 0.36 0.39 0.35 0.48 0.41 0.28

(3.24) (6.28) (3.88) (4.57) (3.50) (2.17)
log(B/M ) 0.21 0.30 0.21 0.24 0.17 0.18
(3.06) (4.22) (2.71) (2.97) (2.12) (2.24)
Size 0.09 0.27 0.32 0.07 0.30 0.02
( 1.30) ( 2.17) ( 1.91) (0.29) (1.64) ( 0.20)
COP/A 1 0.35 0.33 0.29 0.31 0.33 0.31
(5.17) (7.63) (4.26) (4.12) (2.87) (3.65)
dA/A 1 0.20 0.24 0.25 0.28 0.22 0.14
( 3.57) ( 6.58) ( 4.29) ( 3.97) ( 2.15) ( 1.24)
Adj. R2 4.8% 12.7% 1.9% 4.3% 2.4% 5.6% 3.5% 7.0% 3.9% 8.6% 6.9% 16.5%
This table shows Fama and MacBeth (1973) cross-sectional regressions of monthly returns on the previous month’s return (r1,0 ), the previous month’s

Electronic copy available at: https://ssrn.com/abstract=3150525

turnover (TO 1,0 ), and their interaction (r1,0 ⇥TO 1,0 ). Regressions are estimated using weighted least squares (WLS) with market capitalization as
the weight. Independent variables are trimmed at the 1st and 99th percentiles and then standardized by their cross-sectional average and standard
deviation. The interaction is the product of the standardized variables. The controls are prior 12-2 month performance (r12,2 ), cash-based operating
profits-to-lagged assets (COP/A 1 ), asset growth (dA/A 1 ), book-to-market equity (log(B/M ), where M is market equity as of prior December),
and Size (log of market capitalization as of prior June). Specifications 1 and 2 are for all stocks, while the remaining specifications are within NYSE
size quintiles. Test statistics are adjusted for heteroscedasticity and autocorrelation. The sample excludes financial firms and firms with negative book

15
equity. Data are at the monthly frequency and cover July 1963 to December 2018.
 The Review of Financial Studies / v 00 n 0 2021

The all-stock regressions show that r1,0 predicts the cross section
of monthly returns with a negative and significant slope (t-statistic of
4.16 without controls and 8.05 with controls). They also show that
r1,0 ⇥TO 1,0 has roughly the same predictive power as r1,0 itself, albeit
with a positive slope (t-statistics of 5.45 and 9.06). Similar results hold
within each size quintile, although the slope on r1,0 generally decreases
in magnitude with size while that on r1,0 ⇥TO 1,0 tends to increase
with size. This suggests that low-turnover reversal becomes weaker with
size and, conversely, that high-turnover continuation becomes stronger
with size, consistent with the size-conditional strategy performance in
Table 3. Taken literally, the regressions suggest that 1-month returns
are positively autocorrelated when turnover is 0.50/0.20 = 2.5 standard
deviations above average among microcaps, but that 0.25/0.35 = 0.7
standard deviations are enough for the same relation among megacaps.
We caution, however, against reading too much into the magnitudes
of the slope estimates or taking the parametric relations literally. Cross-
sectional regressions are useful for studying return predictability in a
multivariate setting but, unlike portfolio sorts, they impose a potentially
misspecified functional form and are sensitive to outliers (Fama and
French, 2010). Furthermore, the R2 values in Table 4 imply that
over 80% of the variation in monthly returns is not captured by the
parametric relations. A more careful interpretation (cf. Fama, 1976)
is that each slope is the average return to a long-short strategy that
trades on a unit spread in the part of the variation in the independent
variable that is orthogonal to all other independent variables. Under
that interpretation, each t-statistic is proportional to the Sharpe ratio
of the implied long-short strategy. The regressions in Table 4 thus
show that the long-short strategy that trades on orthogonal variation in
r1,0 ⇥TO 1,0 generates a positive Sharpe ratio that is roughly equal in
magnitude to the negative Sharpe ratio on the strategy that trades on
orthogonal variation in r1,0 .13

1.7 International evidence

We conclude this section with out-of-sample evidence using international
stock market data. The sample is from the Compustat Global Securities
database and comprises the 22 developed markets considered by Fama
and French (2017).14

13
This interpretation is applicable here because the interaction is far from perfectly
correlated with the straight variables: if we let z(·) denote standardization, then the
average cross-sectional rank correlation between z(r1,0 )⇥z(TO 1,0 ) and each of z(r1,0 )
and z(TO 1,0 ) is 17.34% (t = 35.22) and 4.00% (t = 12.10), respectively.
14
The countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France,
Germany, Great Britain (the United Kingdom), Greece, Hong Kong, Italy, Ireland,
Japan, the Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, and
Switzerland.

16

Electronic copy available at: https://ssrn.com/abstract=3150525

 Table 5
Double sorts on 1-month returns and turnover: International evidence

r1,0 deciles r1,0 strategies

Low 2 3 4 5 6 7 8 9 High E[re ] ↵DMFF6

TO 1,0 deciles
Portfolio excess return

Low 1.28 0.57 0.42 0.14 0.21 0.43 0.27 0.55 1.06 1.97 3.25 3.06
( 6.36) ( 5.83)
2 1.66 0.69 0.25 0.21 0.17 0.09 0.11 0.02 0.40 0.88 2.53 2.09
( 4.63) ( 2.90)
3 1.02 0.59 0.39 0.24 0.29 0.06 0.31 0.03 0.57 1.37 2.40 2.09
Short-term Momentum

( 5.16) ( 4.50)
4 0.63 0.87 0.44 0.30 0.39 0.41 0.20 0.10 0.18 0.52 1.14 0.74
( 2.73) ( 1.71)
5 1.27 0.75 0.31 0.37 0.27 0.43 0.60 0.33 0.31 0.54 1.81 1.47
( 4.13) ( 2.24)
6 0.46 0.62 0.38 0.47 0.61 0.27 0.46 0.32 0.20 0.15 0.61 0.10
( 1.58) ( 0.22)
7 0.91 0.62 0.04 0.51 0.11 0.42 0.39 0.20 0.21 0.08 0.83 0.93
( 2.01) ( 1.65)
8 0.38 0.46 0.24 0.33 0.33 0.25 0.25 0.81 0.40 0.37 0.01 0.04
( 0.03) ( 0.08)
9 0.06 0.16 0.38 0.36 0.27 0.52 0.26 0.63 0.38 0.26 0.32 0.46
(1.13) (1.41)
High 1.03 1.30 0.63 0.21 1.04 0.11 0.22 0.24 0.10 0.36 1.39 1.36
(3.65) (3.45)

TO 1,0 strategies

E[re ] 2.30 1.87 1.05 0.07 1.25 0.32 0.05 0.31 1.16 2.33

Electronic copy available at: https://ssrn.com/abstract=3150525

( 4.74) ( 4.69) ( 2.08) ( 0.14) ( 3.17) (0.76) (0.10) (0.64) (2.08) (3.77)
↵DMFF6 2.75 2.11 1.42 0.43 1.52 0.07 0.43 0.09 0.78 1.67
( 5.97) ( 4.82) ( 2.74) ( 0.92) ( 3.25) (0.18) ( 0.94) ( 0.19) (1.38) (2.85)

This table shows international portfolios from double sorts on the previous month’s return (r1,0 ) and turnover (TO 1,0 ). We use independent double
sorts to form country-specific portfolios that are value weighted and rebalanced at the end of each month. We then weight each country’s portfolio

17
by the country’s total market capitalization for the previous month to form each international portfolio. All returns and market values are in U.S.
dollars and excess returns are above the monthly U.S. Treasury-bill rate. The table also shows the raw and risk-adjusted returns to long-short strategies
within the deciles, where risk-adjustment is relative to Fama and French’s (2017) developed markets five-factor model including the momentum factor
(DMFF6). Test statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation. Data are at the monthly frequency and cover January
1993 to December 2018.
 The Review of Financial Studies / v 00 n 0 2021

Table 5 shows international portfolios double sorted on the previous

month’s return and share turnover. Following Asness, Frazzini, and
Pedersen (2019), we use independent double sorts to form country-
specific portfolios that are value weighted and rebalanced at the end of
each month, and then weight each country’s portfolio by the country’s
total market capitalization for the most recent month to form each
international portfolio. We consider only common shares and compute
all returns and market values in U.S. dollars. The sample period is
January 1993 to December 2018.
The international STREV* strategy yields an extremely large 3.25%
per month (t = 6.36). This is over twice as large as its U.S. counterpart,
presumably because of the on average less liquid markets outside the
United States. For comparison, the international counterpart of the
conventional STREV strategy yields 0.37% per month with t = 1.29
(untabulated). Despite this extreme reversal among low-turnover stocks,
high-turnover stocks still exhibit strong short-term momentum, as the
international STMOM strategy yields 1.39% per month with t = 3.65,
which is comparable to its U.S. counterpart. The strategies’ abnormal
returns relative to Fama and French’s (2017) developed markets six-
factor model are about as large and strong as the average returns. Figure
IA.1 in the Internet Appendix shows that the returns to the international
STMOM strategy persist for 24 months after formation, while those to
the international STREV* strategy stagnate after 3 months.
Figure 3 shows the annualized Sharpe ratios of the country-specific
STMOM and STREV* strategies. The STMOM strategy yields positive
average returns in all 22 markets. The STREV* strategy yields negative
average returns in all countries but the United Kingdom.
Lastly, Table IA.3 in the Internet Appendix shows that our results on
the size-conditional performance of the STMOM and STREV* strategies
for the United States (Table 3) carry over to the international sample.
The average return to the international STMOM strategy is almost
monotonically increasing across size quintiles: 0.23% per month in
the quintile with the smallest stocks (t = 0.36), around 0.65% in the
intermediate size quintiles (t-statistics between 1.16 and 1.90), and
1.12% per month with t = 4.01 in the quintile with the largest stocks.
Moreover, a value-weighted STMOM strategy based on the largest 100
stocks in each country yields 0.62% per month with t = 3.25.15

15
The long and short sides of the international largest-100-per-country STMOM strategy
trade in an average of 499 and 470 stocks. In December 2018, the top holdings from
the top-five countries on the long side were NTT DoCoMo (Japan), BMW (Germany),
China Mobile (Hong Kong), Banco Santandar (Spain), and AB InBev (Belgium), while
the top holdings from the top-five countries on the short side were Toyota (Japan),
Nestle (Switzerland), Allianz (Germany), Citic (Hong Kong), and AstraZeneca (the United
Kingdom).

18

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

Coexistence
Coexistenceofofreversal
reversaland
andmomentum
momentumininone-month returns:International
1-month returns: Internationalevidence
evidence

1
Sharpe Ratio (annualized)

-1

Short-term momentum
-2 Short-term reversal*

AUS PRT BEL SGP DNK DEU NZL CAN SWE NOR HKG JPN AUT GBR ESP IRL ITA CHE NLD FRA FIN GRC

Figure 3
International short-term momentum returns This figure shows the annualized
Sharpe ratios of the country-specific international short-term momentum and short-term
reversal* strategies. The underlying portfolios are from independent double sorts into
deciles based on the previous month’s return and turnover. Portfolios are value weighted
and rebalanced at the end of each month. All returns and market values are in U.S. dollars.
Data are at the monthly frequency and cover January 1993 to December 2018.

2. Additional Results and Robustness

This section provides additional results and robustness tests. We show
that short-term momentum is stronger when purged of liquidity-related
trades at the end of the formation month; survives transaction costs;
generates abnormal returns relative to standard implementations of
popular stock-level momentum strategies; does not appear to be driven
by earnings announcements, industry momentum, factor momentum, or
volatility; and exhibits far less crash risk than conventional momentum.

2.1 End-of-month liquidity trading and implementation lag

Etula et al. (2020) document that end-of-month trading is dominated
by institutional demand for cash and associated with systematic
price pressure and subsequent reversal. We employ this seasonality
in liquidity demand as an instrument to understand the impact of
noninformational trading on short-term momentum performance. If
short-term momentum is driven by informed trading, then skipping
the last few days of the formation month and purging the signals
of noninformational trading should result in stronger performance.
Skipping the last few days of the formation month also serves as
an implementation lag, facilitating a more realistic assessment of the
strategy’s implementability.
Panel A of Table 6 shows the performance of winner-minus-loser
strategies similar to those in Table 1, except that we skip the sorting
variables’ values for the last 3 trading days in each month. Hence, if
month t 1 has d trading days, the sorting variables at the end of month

19

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Table 6
Short-term momentum and end-of-month e↵ects

A B
Performance of r1,0 EOM strategies Performance of rEOM strategies
within TO 1,0 EOM deciles within TO EOM deciles

TO 1,0 EOM E[re ] ↵FF6 ↵q TO EOM E[re ] ↵FF6 ↵q

Low 0.41 0.49 0.41 Low 2.36 2.45 2.56

( 2.16) ( 2.58) ( 1.75) ( 11.51) ( 11.13) ( 11.59)
2 0.38 0.48 0.48 2 1.72 1.80 1.81
( 1.64) ( 1.98) ( 1.76) ( 7.12) ( 6.61) ( 6.55)
3 0.28 0.47 0.52 3 1.73 1.90 2.04
( 1.22) ( 1.83) ( 1.56) ( 6.17) ( 6.96) ( 6.55)
4 0.21 0.24 0.11 4 1.66 1.75 1.77
( 0.81) ( 0.75) ( 0.29) ( 6.06) ( 5.60) ( 5.64)
5 0.39 0.30 0.46 5 1.72 1.99 1.92
(1.40) (0.90) (1.35) ( 6.78) ( 6.75) ( 6.77)
6 0.22 0.08 0.16 6 1.86 1.97 2.06
(0.89) (0.27) (0.39) ( 7.45) ( 7.00) ( 7.56)
7 0.27 0.33 0.50 7 1.16 1.17 1.20
(1.00) (0.91) (1.29) ( 4.86) ( 4.54) ( 4.42)
8 0.38 0.34 0.53 8 1.67 1.72 1.79
(1.26) (1.11) (1.35) ( 5.60) ( 5.01) ( 4.58)
9 0.72 0.58 0.72 9 1.16 1.29 1.29
(2.89) (2.09) (2.13) ( 4.74) ( 4.91) ( 4.41)
High 1.83 1.80 2.10 High 0.65 0.86 0.87
(5.52) (5.22) (5.16) ( 2.25) ( 2.63) ( 2.41)
This table shows the performance of winner-minus-loser strategies based on the previous
month’s return within deciles of the previous month’s share turnover. In panel A, the
sorting variables skip their end-of-month values (r1,0 EOM and TO 1,0 EOM ) measured
at the month’s last 3 trading days. In panel B, the sorting variables are just the
end-of-month values (rEOM and TO EOM ) measured at the month’s last 3 trading days.
Portfolios are from conditional sorts into deciles based on NYSE breakpoints, first on
returns and then on turnover, and are value weighted and rebalanced at the end of
each month. Test statistics (in parentheses) are adjusted for heteroscedasticity and
autocorrelation. The sample excludes financial firms. Data are at the monthly frequency
and cover July 1963 to December 2018, except when applying the q-factors, which are
available from January 1967.

t 1, used to predict returns over month t, are based on days 1,2,...,d 3

of month t 1. We only consider stocks with at least 15 nonmissing daily
returns and turnover each month. Skipping the end of the formation
month implies a clear shift toward stronger high-turnover continuation
and weaker low-turnover reversal compared to Table 1. The STMOM
strategy (in the highest turnover decile) now yields 1.83% per month
with t = 5.52 (compared to 1.37% with t = 4.74 in the benchmark case).
In addition, there is a significant continuation e↵ect in the ninth turnover
decile: 0.72% per month with t = 2.82. By contrast, the STREV* strategy
(in the lowest turnover decile) yields just 0.41% per month (compared
to 1.41% per month in the benchmark case). The abnormal returns
tell a very similar story. These results suggest that skipping the end of

20

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

the formation month e↵ectively purges the sorting variables of end-of-

month liquidity trades, leading to stronger high-turnover continuation
but weaker low-turnover reversal. In the following, we provide two pieces
of evidence that corroborate this explanation.
First, we consider a complementary version of this exercise where we
sort only on the signals’ end-of-month values, that is, only on returns
and turnover occurring on days d 2,d 1, and d of month t 1. Based
on the same reasoning as above, the last days in a month should be
dominated by noninformational demand for liquidity, and we therefore
expect to see stronger low-turnover reversal but weaker high-turnover
momentum. Panel B of Table 6 shows that this is indeed the case. The
resultant STREV* strategy now yields a very large 2.36% per month
with t = 11.51, while the corresponding STMOM strategy’s average
return is now a negative 0.65% per month.16
Second, we repeat the tests of Table 6 on di↵erent days during
the month. Panel A of Figure 4 shows the performance of STMOM
and STREV* strategies formed on day d of each month, based on
prior return and turnover over d 20 to d 3, and held from d+1 to
d+21. Skipping 3 days only significantly improves the performance
of STMOM when these days lie at the end of the month, and the
performance of the corresponding STREV* strategies is weakest around
the end of the month. Panel B shows the complementary version of this
exercise. Sorting on just the prior 3 days’ signal values is only associated
with stronger STREV* performance when these days lie at the end of
the month, while the corresponding STMOM strategies generate large
negative returns around the end of the month.
Taken together, these results suggest that the performance of
STMOM and STREV* is indeed a↵ected by noninformational demand
for liquidity, but in opposite directions. Practitioners implementing
STMOM should skip the last few days of the formation month to purge
the signals of end-of-month liquidity trades. Conversely, researchers and
practitioners looking into price pressure and the returns to liquidity
provision should pay attention to STREV* near the end of the month.
Lastly, the fact that skipping the end of the formation month implies
a significant continuation e↵ect in the ninth turnover decile (panel A
of Table 6) suggests that quintile double sorts should be sufficient to
document significant short-term momentum. Table A.3 in the appendix
confirms this: using independent quintile double sorts and skipping the
last 3 days of the formation month yields an STMOM strategy with

16
Table IA.4 in the Internet Appendix shows that the results from Table 6 also hold in
our international sample. In untabulated tests, we find very similar results when focusing
on the last 5 (instead of 3) trading days for month t 1 (the formation period) and,
furthermore, that these results are robust to also excluding the first 1–3 trading days
from the return for month t (the holding period).

21

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

A Sorts on day d based on prior return and turnover over days d −20 to d −3
(i.e., skipping days d , d −1, and d −2)

3
Average excess return over days d + 1 to d + 21 (%)
Short-term momentum
Short-term reversal*
2
1
0
-1

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Day of rebalancing during the month (d )

B Sorts on day d based on prior return and turnover over days d , d −1, and d −2
1
Average excess return over days d + 1 to d + 21 (%)

0
-1
-2
-3

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Day of rebalancing during the month (d )

Figure 4
Short-term momentum and end-of-month e↵ects This figure shows the average
excess returns to STMOM and STREV* strategies formed on di↵erent days (d) during
the month. Panel A shows strategies formed on day d based on prior return and turnover
over days d 20 to d 3, skipping the most recent 3 days (i.e., days d 2,d 1, and d).
Panel B shows strategies formed on day d based only on prior return and turnover over the
most recent 3 days (i.e., days d 2,d 1, and d). We use conditional sorts into deciles based
on NYSE breakpoints, first on returns and then on turnover. Portfolios are value weighted
and returns are calculated over days d+1 to d+21. The sample excludes financial firms.
Data are at the daily frequency and cover the end of July 1963 to the end of December
2018.

an average return of 0.74% per month with t = 4.21. The strategy’s

abnormal returns are about as large and as strong as its average return.
It trades in an average of 222 and 180 stocks on its long and short sides
with average market capitalizations of $1.22 billion and $1.01 billion.
As such, short-term momentum exists for a substantially larger subset
of the market than in our benchmark results.

22

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

2.2 Transaction costs

Strategies from sorts on 1-month returns tend to have high turnover
and can be expensive to trade (Da, Liu, and Schaumburg 2012; Novy-
Marx and Velikov 2015). Nonetheless, this section shows that short-term
momentum survives conservative estimates of transaction costs.
Following Koijen et al. (2018) and Bollerslev et al. (2018), we estimate
average transaction costs using average turnover and average “half-
spreads” (one-half of the proportional bid-ask spread).17 For a given
portfolio, let wi,t 1 0 be the value weight of stock i in the sort at the
end of month t 1 and let rit be the stock’s return over month t. The
portfolio’s (one-way) turnover in month t is then
1X
(1+rit )wi,t 1 wit , (1)
2 i

where the sum is taken over all stocks in the portfolio in months t 1 or t
and is divided by 2 to avoid double counting buys and sells. The product
(1+rit )wi,t 1 is the weight of stock i just before rebalancing at the end of
month t. We set wi,t 1 = 0 if stock i is not in the portfolio in month t 1,
and similarly for wit . The portfolio’s half-spread in any given month is
the value-weighted average of 12 (ASK BID)/( 12 (ASK+BID)).
Table 7 shows strategy performance net of transaction costs. The
benchmark STMOM strategy (Table 1) turns over 89% and 90% of its
long and short sides each month on average, yet these incur relatively
modest average half-spreads of 0.17% and 0.24% per month. Table A.4
in the appendix explains why: for every $1 investment, each side puts on
average just 9 cents into microcaps but around 30 cents into megacaps.
The strategy’s average monthly cost is therefore just 0.89⇥0.17+0.90⇥
0.24 = 0.37%, implying a net average return of 1.00% per month with
t = 3.47. Its FF6 net abnormal return is equally large and strong, while
its net information ratio relative to a “FF8” model (FF6 plus the two
conventional reversal factors) is 0.92. The benchmark STREV* strategy
turns over more often and incurs higher half-spreads because it puts
much more weight on microcaps (Table A.4). Its average monthly cost
of 1.94% therefore subsumes its average gross return when judged by
our conservative approach.

17
Half-spreads are simpler than the “e↵ective bid-ask spread” proposed by Hasbrouck (2009)
and used by Novy-Marx and Velikov (2015), but the resultant costs are still conservative,
especially for large institutional traders. Like e↵ective spreads, half-spreads do not account
for the price impact of large trades and should be interpreted as the costs faced by
a small liquidity demander. The resultant costs are nonetheless conservative (an upper
bound) because they assume market orders and immediate liquidity demand, instead
of limit orders, an assumption that likely overstates the actual average costs associated
with implementing a strategy. Frazzini, Israel, and Moskowitz (2015) find that the actual
trading costs faced by a large institutional trader are an order of magnitude smaller than
those estimated for the average trader because a large institutional trader will attempt to
trade within the spread, use limit orders, and supply, rather than demand, liquidity.

23

Electronic copy available at: https://ssrn.com/abstract=3150525

 24
Table 7
Short-term momentum and transaction costs

Long side Short side FF6 (net) FF8 (net)

e e
Strategy E[rgross ] TO HS TO HS Cost E[rnet ] ↵ IR ↵ IR

Short-term momentum 1.37 0.89 0.17 0.90 0.24 0.37 1.00 1.00 0.44 1.78 0.92
(4.74) (3.47) (3.09) (6.11)

Short-term reversal* 1.41 0.96 0.84 0.94 1.22 1.94

( 7.13)

Short-term momentum 1.83 0.89 0.18 0.91 0.24 0.38 1.45 1.42 0.61 2.18 1.11
(excl. EOM) (5.52) (4.38) (4.12) (6.95)

Short-term momentum 0.74 0.83 0.17 0.85 0.21 0.32 0.42 0.45 0.33 1.03 1.07
(5⇥5, excl. EOM) (4.21) (2.40) (2.35) (6.61)
The Review of Financial Studies / v 00 n 0 2021

Short-term momentum 0.77 0.83 0.08 0.84 0.10 0.15 0.62 0.59 0.38 1.22 1.01
(large-caps, excl. EOM) (3.90) (3.14) (2.78) (7.07)

Short-term momentum 0.72 0.84 0.07 0.85 0.08 0.13 0.59 0.59 0.34 1.24 0.90
(megacaps, excl. EOM) (3.76) (3.10) (2.66) (6.46)
This table shows strategy performance net of transaction costs. “TO” is average turnover, computed as the time-series average of one-half of the sum of
absolute monthly changes in the portfolio weights (Eq. (1)). “HS” (in % per month) is the average half-spread, computed as the time-series average of
the portfolios’ monthly value-weighted average half-spreads. “Cost” (in % per month) is the sum of the average turnover times the average half-spread
for the long and short sides. “FF8” is FF6 plus the conventional short-term and long-term reversal factors. “IR” is the information ratio, computed

Electronic copy available at: https://ssrn.com/abstract=3150525

as the abnormal return over the residual standard error. “EOM” is end of month. Test statistics (in parentheses) are adjusted for heteroscedasticity
and autocorrelation. Data are at the monthly frequency and cover July 1963 to December 2018, but the half-spreads are computed starting from 1983
because of data availability in CRSP.
 Short-term Momentum

The STMOM strategy with an implementation lag (panel A of Table

6) earns 1.45% per month with t = 4.38 after costs and has a net FF8
information ratio of 1.11. The coarser version of this strategy from
quintile double sorts (Table A.3) earns 0.42% per month with t = 2.40
after costs, but its net FF8 information ratio is largely undiminished
and its investment capacity is over four times higher (Table A.4).18
Lastly, the large-caps and megacaps STMOM strategies with an
implementation lag (as in Table 3 but excluding end-of-month values)
have almost identical performance at around 0.60% per month after
costs with a t-statistic above 3. The reason is that the large-caps strategy
on average puts over 55 cents of every invested dollar into megacaps. The
megacap strategy’s investment capacity is on average 4.24% of aggregate
market capitalization and totaled $1.24 trillion in December 2018.

2.3 Other stock-level momentum strategies: Price-,

earnings-, and ROE-based momentum
Table 8 analyzes the STMOM strategy in the context of four other stock-
level momentum strategies: A conventional momentum strategy based
on prior 12-2 months performance (r12,2 ), a post-earnings announcement
drift (PEAD) strategy based on cumulative 3-day abnormal return
around quarterly earnings announcements dates (CAR3 ), a PEAD
strategy based on standardized unexpected earnings (SUE), and an
earnings-momentum strategy based on quarterly return on equity
(ROE). The strategies trade value-weighted portfolios from decile sorts
using NYSE breakpoints and are rebalanced at the end of each month.19
Panel A shows the strategies’ average excess returns and characteristic
tilts, defined as the time-series average of the long-short di↵erence
in the portfolios’ monthly value-weighted average characteristics. The
characteristics are the strategies’ main sorting variables. STMOM has
a sizable tilt toward the CAR3 signal but only modest tilts toward
the other momentum signals (negative for ROE). Conversely, the other
momentum strategies have only modest tilts toward the r1,0 signal
(insignificant for conventional momentum and ROE), with the exception

18
For comparison, the conventional momentum strategy incurs an average cost of 0.26% per
month and its net return is 0.95% per month (t = 3.74). Moskowitz and Grinblatt’s (1999)
industry momentum strategy incurs an average cost of 0.34% per month and its net return
is 0.64% per month (t = 3.02). The average cost of the conventional STREV strategy is
0.56% per month, which subsumes its average gross return.
19
We employ quarterly earnings starting from the end of the month of the announcement
date (Compustat’s RDQ). CAR3 is the 3-day cumulative return minus the value-weighted
market return around earnings announcements. SUE is the year-over-year change in split-
adjusted earnings per share (EPSPXQ/AJEXQ) divided by its standard deviation over
the latest eight announcements (minimum of six) excluding the current announcement.
ROE is total quarterly earnings (IBQ) deflated by one-quarter-lagged book equity.

25

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

of the CAR3 strategy.20 Nonetheless, panel B shows that the correlation

between STMOM and the CAR3 strategy is a modest 18% (t = 2.72),
equal in magnitude and strength to the correlation between STMOM
and conventional momentum. For comparison, the correlation between
conventional momentum and each of the PEAD strategies is 36% and
53% with t-statistics exceeding 5.00.
The modest correlations suggest that STMOM’s average return is
not spanned by the other momentum strategies. Panel C confirms this:
STMOM’s abnormal return relative to the other momentum strategies
is 0.96% per month with t = 2.30 and the regression’s adjusted R2 is just
5.3%. In these tests, the only other strategy with an abnormal return
t-statistic exceeding 2.00 is the CAR3 strategy. We caution, however,
that spanning tests are not conclusive about the redundancy of one
e↵ect relative to others: for one, they can be fragile to the choice of
right-hand-side strategies and to the details of the strategy construction.

2.4 Directly controlling for earnings announcement dates

Section 2.3 shows that the STMOM strategy is only moderately
correlated with strategies based on post-earnings announcement drift
(PEAD). Here, we show more generally that our main results are not
mechanically driven by the high trading volume and persistent price
drifts associated with earnings announcements (for a review, see, e.g.,
Hong and Stein 2007; see also Medhat and Schmeling 2018).
Panel A of Table IA.6 in the Internet Appendix shows the performance
of winner-minus-loser strategies similar to those in Table 1, except
that we exclude firms whose most recent announcement date fell in
the previous month. With the exclusion of announcers, STMOM yields
1.31% (t = 2.75) while STREV* yields 1.82% per month (t = 5.99)
over the period from January 1972 for which we have announcement
dates. Panel B shows an alternative version of this exercise where we
keep announcers but exclude the 3 days around announcements from
the return and turnover signals. In this case, STMOM yields 1.16%
(t = 3.30) while STREV* yields 1.80% (t = 7.32) per month. In sum,
the profitability of short-term momentum is largely undiminished when
explicitly excluding earnings announcements.

20
Table IA.5 in the Internet Appendix shows additional characteristic tilts as well as average
portfolio overlap for the five momentum strategies. An interesting observation is that,
while conventional momentum has the well-known strong tilt toward growth stocks, short-
term momentum has a slight but significant value tilt. The two strategies’ average portfolio
overlap is 34% for winners and 20% for losers.

26

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

Table 8
Short-term momentum and other stock-level momentum strategies

A. Excess returns and characteristic tilts

Time-series average of long-short di↵erence
in portfolio characteristics (all in %)

Strategy E[re ] r1,0 r12,2 CAR3 SUE ROE

STMOM 1.37 45.95 3.20 5.85 0.19 0.21

(4.74) (36.21) (2.19) (11.13) (3.31) ( 0.80)
MOM 1.21 0.26 111.20 2.48 1.63 3.14
(4.77) (0.78) (19.38) (12.17) (20.50) (6.17)
CAR3 0.91 7.49 14.45 21.29 0.74 1.62
(6.46) (24.08) (20.62) (10.18) (11.51) (8.38)
SUE 0.53 1.31 24.57 1.73 6.14 8.52
(4.02) (10.88) (15.88) (16.35) (36.12) (9.62)
ROE 0.75 0.34 12.90 1.25 3.52 23.21
(3.18) (1.38) (4.08) (10.04) (20.84) (9.55)
B. Time-series correlations
MOM CAR SUE ROE

STMOM 0.18 0.18 0.06 0.01

(2.50) (2.72) (1.24) ( 0.12)
MOM 0.36 0.53 0.23
(5.18) (9.09) (1.59)
CAR3 0.29 0.13
(4.17) (1.49)
SUE 0.51
(6.06)
C. Spanning tests
Intercepts, slopes, and test statistics (in parentheses)
from time-series regressions of the form yt = ↵+ 0 Xt +✏t

Dependent strategy ↵ STMOM MOM CAR3 SUE ROE Adj. R2

STMOM 0.96 0.25 0.34 0.13 0.07 5.3%

(2.30) (2.44) (2.07) ( 0.86) ( 0.49)
MOM 0.22 0.10 0.41 0.87 0.06 34.4%
(0.94) (2.35) (3.29) (7.55) ( 0.38)
CAR3 0.63 0.04 0.13 0.13 0.00 15.2%
(4.63) (2.18) (2.96) (2.45) ( 0.07)
SUE 0.04 0.01 0.23 0.11 0.29 44.6%
( 0.32) ( 0.87) (7.14) (2.25) (7.37)
ROE 0.44 0.02 0.04 0.01 0.74 26.0%
(1.92) ( 0.47) ( 0.37) ( 0.07) (6.25)

This table studies the benchmark STMOM strategy (Table 1) in the context of four
other stock-level momentum strategies: a conventional momentum strategy (MOM)
based on prior 12-2 months performance (r12,2 ), a post-earnings announcement drift
(PEAD) strategy based on cumulative 3-day abnormal return around quarterly earnings
announcements dates (CAR3 ), a PEAD strategy based on standardized unexpected
earnings (SUE), and an earnings-momentum strategy based on the quarterly return on
equity (ROE). The strategies trade value-weighted portfolios from decile sorts using
NYSE breakpoints and are rebalanced at the end of each month. Test statistics (in
parentheses) are adjusted for heteroscedasticity and autocorrelation. The sample excludes
financial firms, and the sorts on ROE exclude firms with negative book equity. Data are
at the monthly frequency and cover July 1963 to December 2018, but the PEAD and
ROE strategies start from January 1972 because of data availability in Compustat.

27

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

2.5 Industry momentum

Moskowitz and Grinblatt (1999) find that industry momentum is
strongest at the 1-month horizon (see also Asness, Porter, and Stevens
2000). Nonetheless, the profitability of short-term momentum does not
appear to be mechanically driven by industry e↵ects.
Table IA.7 in the Internet Appendix shows spanning tests employing
STMOM and a 1-month industry momentum (IMOM) strategy based on
the 49 industries from Fama and French (1997) but excluding financials.
STMOM is not within the span of the IMOM strategy, with or without
controlling for the FF6 or q-factors. On the other hand, IMOM is in fact
within the combined span of STMOM and the q-factors.
Table A.5 in the appendix shows the performance of the winner-
minus-loser strategies similar to those in Table 1 except that we consider
industry-adjusted signals or performance. Panel A shows strategies from
sorts where the sorting variables are demeaned by their value-weighted
average industry values. With industry-demeaned signals, STMOM
yields 1.18% (t = 4.08) while STREV* yields 2.00% per month (t =
10.25). That is, the profitability of short-term momentum remains
largely undiminished when the strategy is constructed in a way that
explicitly sorts against the 1-month industry-momentum signal.21
Panel B shows a decomposition of Table 1’s benchmark strategies into
an industry-hedged (“within-industry”) component and the industry
hedge. The industry-hedged strategies are from sorts on unadjusted
variables, but each stock’s position is combined with an o↵setting
position of equal size in the corresponding value-weighted industry
portfolio. The industry hedges are these o↵setting positions in the value-
weighted industry portfolios. This follows Novy-Marx (2013) and allows
us to quantify how much of the benchmark strategies’ performance is
due to industry exposure. The benchmark STMOM strategy’s average
monthly return of 1.37% (t = 4.74) decomposes into 0.89% (t = 3.35)
from the within-industry component and 0.48% (t = 5.65) from the
industry hedge. Similar results hold for abnormal returns. That is,
roughly two-thirds of the strategy’s performance is unrelated to industry
exposure and hedging industry exposure does not increase its Sharpe or
information ratios.

21
Moskowitz and Grinblatt (1999) argue that “[o]ne possible explanation for the
discrepancy between short-term (one-month) reversals for individual stocks and short-
term continuations for industries is that the one-month return reversal for individual
stocks is generated by microstructure e↵ects (such as bid-ask bounce and liquidity e↵ects),
which are alleviated by forming industry portfolios” (p. 1274). Our results suggest that
conditioning on high turnover is an alternative and equally e↵ective way of alleviating
microstructure e↵ects.

28

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

2.6 Factor momentum

Ehsani and Linnainmaa (2020) construct “factor momentum” strategies
which are long winners and short losers based on prior 12-1 (not 12-2)
month return from a set of 20 U.S. and international factors. They
find that factor momentum subsumes conventional price momentum
in spanning tests (see also Lewellen, 2002). Table IA.8 in the Internet
Appendix shows that this is not the case for short-term momentum. The
STMOM strategy yields significant abnormal returns relative to both
the time-series factor-momentum strategy and the cross-sectional factor-
momentum strategy, whether employed alone or together and with or
without controlling for the FF6 factors. The table also shows that while
factor momentum explains over 40% of the variation in conventional
momentum, it explains under 5% of that in short-term momentum.

2.7 Longer formation periods

Our main results use past short-term performance and turnover
measured over the previous month. Here, we show that our results are
robust to using longer formation periods of up to 6 months, although
the 1-month horizon produces the strongest results.
Table IA.9 in the Internet Appendix shows the performance of
alternative short-term momentum and short-term reversal* strategies
constructed as in Table 1, except that the sorting is on cumulative
return and average monthly turnover for the previous 2,...,6 months.
The alternative short-term momentum strategies generate significant
average returns between 0.99% and 1.24% per month, although lower
and statistically weaker than the 1.37% for the benchmark strategy. The
average returns to the alternative short-term reversal⇤ strategies are
decreasing with the formation period and insignificant at the 5-month
horizon. Each alternative strategy is within the univariate span of the
corresponding benchmark strategy.

2.8 Volatility and residual turnover

Bandarchuk and Hilscher (2013) argue that “characteristic screens” lead
to elevated profits for conventional price-momentum strategies because
they identify stocks with more volatile past returns. In particular, they
find that the part of share turnover which is unrelated to volatility no
longer has the ability to elevate the profits of conventional momentum
strategies. Here, we show that volatility is not driving our main results.
Table IA.10 in the Internet Appendix shows the performance of
winner-minus-loser strategies similar to those in Table 1, except that we
replace the previous month’s share turnover with “residual turnover”
(RTO 1,0 ; panel A) or volatility ( 1,0 ; panel B). Here, RTO 1,0 is the
residual from cross-sectional regressions of the previous month’s share
turnover on 1,0 , estimated using weighted least squares (WLS) with

29

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

market capitalization as the weight, while 1,0 is the standard deviation

of the previous month’s daily stock returns using a minimum of 15
daily observations. Panel A shows that using residual turnover does
not materially alter our main results. Panel B shows that, while there is
significant 1-month reversal among low-volatility stocks, there is no 1-
month continuation among high-volatility stocks. In untabulated tests,
we find very similar results using idiosyncratic volatility relative to the
FF6 or q-factors.

2.9 Pre-1963 performance and sample splits

As is now standard in the literature, we focus on the post-1963 sample
for our main results. However, the availability of CRSP data going back
to 1926 allows us to verify the robustness of our main results to adding
an additional 37 years to the sample.
Table IA.11 in the Internet Appendix shows spanning tests for
the STMOM and STREV* strategies starting from July 1926. The
explanatory factors are the three Fama-French factors (MKT, SMB,
HML) in addition to the momentum factor (MOM) and the two reversal
factors (STREV and LTREV). Over the full sample period (July 1926 to
December 2018), the average STMOM return is 1.02% per month with
t = 4.23 while the average STREV* return is 2.82% per month with
t = 5.40. The corresponding abnormal returns are 2.16% per month
(t = 9.03) for STMOM and 1.80% per month (t = 6.83) for STREV*.
The table also shows the performance of the strategies in three
nonoverlapping subperiods: July 1926 to June 1963, July 1963 to June
1991, and July 1991 to December 2018. The STMOM strategy’s average
return is monotonically increasing across the subperiods, from 0.60%
per month (t = 1.61) in the pre-1963 era to 2.03% per month (t = 4.80)
in the post-1991 era. The opposite is true for STREV*, for which the
average return decreases monotonically in magnitude from 3.96% to
1.15% per month across the subperiods. We conjecture that generally
increasing market liquidity is responsible for these trends.

2.10 Crash risk

Daniel and Moskowitz (2016) show that conventional price momentum
can experience infrequent but persistent strings of negative returns—
or crashes—that are contemporaneous with market rebounds, in that
the strategy “will have significant negative market exposure following
bear markets precisely when the market swings upward” (p. 229). Here,
we show that short-term momentum exhibits far less crash risk than
conventional momentum. Following Daniel and Moskowitz, we use the
extended sample going back to January 1927.
Table A.6 in the appendix presents the results. Panel A shows that
STMOM exhibits a mild negative skew and a moderate kurtosis, similar

30

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

to those observed for the market but orders of magnitude lower than
those observed for conventional momentum. It also shows that while
conventional momentum exhibits negative and significant coskewness
(Harvey and Siddique 2000; Schneider, Wagner, and Zechner 2020) and
downside beta (Henriksson and Merton 1981), both are positive albeit
small and statistically insignificant for STMOM.
Panel B shows results from Daniel and Moskowitz’s regressions,
rte = (↵0 +↵B IB,t e
1 )+( 0 +( B + B,U IU,t )IB,t 1 )rmt +✏t , (2)
where rte is a strategy’s excess return, rmt e
is the market excess
return, IB,t 1 is an ex ante 24-month “bear market” indicator, and
IU,t is a contemporaneous 1-month “up-market” indicator. Here, B
captures the di↵erence in beta during bear markets, while B,U captures
the di↵erence in beta in rebounds after bear markets. Conventional
momentum has a bear-market beta of b0 + bB = 0.34 when the
contemporaneous market return is negative but b0 + bB + bB,U = 1.06
when the contemporaneous market return is positive. In contrast,
STMOM’s bear-market beta is 0.51 when the contemporaneous market
return is negative but just +0.05 when the contemporaneous market
return is positive. That is, STMOM is a better hedge in bear markets
and has no negative market exposure in rebounds.

2.11 VIX and volatility risk

Drechsler, Moreira, and Savov (2018) find that the CBOE Volatility
Index (VIX) predicts larger returns to daily reversal strategies,
especially among larger stocks (see also Nagel 2012). They also find that
daily reversal strategies are exposed to volatility risk, in that they su↵er
loses when the contemporaneous change in the VIX is higher, again
especially among larger stocks. We show that the monthly strategies we
consider exhibit little predictability by the VIX and only insignificant
exposure to volatility risk after controlling for market exposure.
Table IA.12 in the Internet Appendix shows time-series regression
results for the small- and large-cap STMOM and STREV* strategies
from Table 3. The explanatory variables are the 1-month-lagged VIX,
the contemporaneous monthly change in VIX, and the contemporaneous
market return. The period covered is January 1990 to December 2018,
where the start date is determined by the availability of the VIX. Only
the small-cap STREV* strategy is predictable by the VIX. Its average
return in the post-1990 period is 1.03% per month (t = 4.92) and a
one-point increase in the VIX predicts a 9 basis points more negative
monthly return (t = 2.13) with an R2 of 2.3%. Without controlling for
the market return, all four strategies load positively and significantly
on contemporaneous VIX changes, with 25–30 basis points added to
each strategy’s monthly return for every one-point increase in the

31

Electronic copy available at: https://ssrn.com/abstract=3150525

 The Review of Financial Studies / v 00 n 0 2021

contemporaneous VIX change. That is, ignoring market exposure, a

larger VIX change is associated with losses to the STREV* strategies
but gains to the STMOM strategies. Controlling for the market return,
however, renders the strategies’ volatility-risk exposure insignificant.

3. Testing Theories of the Volume-Return Relation

In this section, we study short-term momentum in the context of
models of the volume-return relation, that is, models of how expected
returns depend on realized returns and volume. We first consider
prominent rational expectations equilibrium (REE) models. Since short-
term momentum proves difficult to reconcile with these models, we
conclude by briefly discussing the potential of models of “boundedly
rational” traders to explain short-term momentum.

3.1 REE models of the volume-return relation

In the model of Campbell, Grossman, and Wang (1993), noninforma-
tional trading due to liquidity demand causes temporary price pressure
when absorbed by liquidity suppliers. As a result, returns coupled with
high trading volume will subsequently reverse. This runs opposite to
short-term momentum and rules out purely noninformational trading
as an explanation.
In the models of Wang (1994) and Llorente et al. (2002), trading due to
private information causes persistent price movements that counteract
temporary price pressure. As a result, returns coupled with high volume
will reverse among stocks with low information asymmetry (similar to
Campbell, Grossman, and Wang 1993) but will reverse less and may
even continue among stocks with high information asymmetry.22,23
Wang (1994) and Llorente et al. (2002) argue that high-information-
asymmetry stocks should be small, illiquid, and have low analyst
coverage (see our footnote 1). As such, high-information-asymmetry is

22
Wang (1994) assumes long-lived private information and that liquidity-related trading is
due to hedging a nontraded asset which is correlated with the risky asset, that is, that
liquidity-related trading is “endogenous.” Inspired by Wang (1994), Llorente et al. (2002)
also assume endogenous liquidity-related trading but drop the assumption of long-lived
private information. As a result, higher volume is always associated with more reversal in
their model, but the e↵ect of volume decreases in magnitude with information asymmetry.
23
We focus on the models’ predictions for return autocorrelation conditional on high volume
because their predictions for return autocorrelation conditional on low volume are less
clear-cut and receive less attention. In Campbell, Grossman, and Wang’s (1993) model,
the sign of return autocorrelation conditional on low volume is ambiguous (equation (16),
p. 929). In Llorente et al.’s (2002) model, which simplifies that of Wang (1994), return
autocorrelation conditional on low volume is captured by ✓1 , about which they write “the
result on ✓1 can be sensitive to the simplifying assumptions of our model. For example,
when motives for hedging trade [...] are persistent, the returns can become positively
serially correlated (even in the absence of trading). In addition, when private information
is long-lived, the behavior of return autocorrelation is more involved, and the impact of
information asymmetry on ✓1 becomes more complex (see, e.g., Wang 1994)” (p. 1015–6).

32

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

an explanation for short-term momentum if it is strongest among small,

illiquid, and low-coverage stocks.
The performance of the size-conditional strategies in Table 3 is at odds
with the high-information-asymmetry mechanism, given that STMOM
is stronger among larger stocks. Since larger stocks tend be more liquid
and have greater analyst coverage, we would a priori expect stronger
STMOM performance among liquid and high-coverage stocks. If so,
this would be further evidence at odds with the mechanism. In the
following, we verify this by constructing STMOM strategies with an
explicit control for illiquidity and analyst coverage.

3.1.1 Short-term momentum and illiquidity. Panel A of Table 9

shows the performance of STMOM and STREV* strategies constructed
with a control for the Amihud (2002) illiquidity measure. The strategies
are from 5⇥3⇥3 conditional sorts on the previous month’s illiquidity,
return, and turnover. The breakpoints for illiquidity are NYSE quintiles
while the breakpoints for the return and turnover signals are the 20th
and 80th percentiles for NYSE stocks. Portfolios are value weighted
and rebalanced at the end of each month. The STMOM strategies’
average returns are monotonically decreasing with illiquidity; from
0.88% per month (t = 3.55) among the most liquid stocks to 0.77%
per month (t = 2.38) among the most illiquid ones. Similar results hold
for abnormal returns. As such, high-turnover continuation is strongest
among the most liquid stocks, a result that directly opposes the high-
information-asymmetry mechanism. The STREV* strategies’ average
returns are all negative, but become monotonically more negative with
illiquidity; from 0.64% to 1.67% per month. Both sets of results are
consistent with those for the size-conditional strategies in Table 3.
Avramov, Chordia, and Goyal (2006) find that, with a control
for illiquidity, higher turnover implies less reversal at the monthly
horizon but more reversal at the weekly horizon.24 Despite the
di↵erent roles of turnover at the monthly and weekly horizons, Table
9 shows that our evidence against the high-information-asymmetry
mechanism also holds at the weekly horizon. In panel B, we consider
strategies constructed similarly to those in panel A, except that we
rebalance the portfolios each Tuesday, skipping signals and holding-
period returns over Tuesday itself to avoid undue influence from bid-ask
bounce. The STMOM strategies’ average returns are negative but again

24
Avramov et al. (2006) argue that the di↵erent roles of turnover “could arise because
demand shocks are attenuated at the monthly frequency as compared to the weekly
frequency, which would suggest that turnover may be a poor proxy for noninformational
trades at the monthly frequency.” (p. 2367). Table IA.13 in the Internet Appendix
shows that using turnover orthogonal to illiquidity does not a↵ect our main results and
that replacing turnover with illiquidity in our double sorts does not lead to significant
continuation.

33

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Table 9
Short-term momentum controlling for illiquidity

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]
controlling for illiquidity controlling for illiquidity

Illiq1,0 quintile E[re ] ↵FF6 E[re ] ↵FF6

A. Monthly sorts
Liquid 0.88 0.67 0.64 0.72
(3.55) (2.51) ( 3.62) ( 3.83)
2 0.38 0.32 1.02 0.97
(1.43) (1.15) ( 5.12) ( 4.42)
3 0.21 0.11 1.24 1.26
(0.85) (0.36) ( 7.05) ( 6.21)
4 0.04 0.15 1.44 1.44
( 0.16) ( 0.53) ( 7.22) ( 4.90)
Illiquid 0.77 1.11 1.67 1.90
( 2.38) ( 2.92) ( 7.32) ( 6.94)
B. Weekly sorts (rebalanced Tuesday, skipping Tuesday)
Liquid 0.18 0.19 0.34 0.32
( 3.15) ( 3.26) ( 8.37) ( 7.94)
2 0.50 0.49 0.45 0.45
( 8.26) ( 7.99) ( 11.02) ( 10.71)
3 0.55 0.54 0.41 0.40
( 9.59) ( 9.11) ( 8.68) ( 8.56)
4 0.79 0.77 0.43 0.41
( 13.64) ( 13.42) ( 9.35) ( 9.06)
Illiquid 1.00 1.00 0.37 0.36
( 14.84) ( 15.15) ( 8.15) ( 7.98)
This table shows the performance of short-term momentum and short-term reversal*
strategies constructed with a control for the Amihud (2002) illiquidity measure. Panel
A shows results for monthly sorts, while panel B shows results for weekly sorts. The
strategies are from 5⇥3⇥3 conditional sorts on illiquidity (Illiq1,0 ), returns, and
turnover, in that order, measured over the previous month or week. Illiq1,0 is the average
absolute daily return relative to the daily dollar trading volume. The breakpoints for
illiquidity are NYSE quintiles, while the breakpoints for returns and turnover are the
20th and 80th percentiles for NYSE stocks. Portfolios are value weighted. Test statistics
(in parentheses) are adjusted for heteroscedasticity and autocorrelation. The sample
excludes financial firms. In panel A, portfolios are rebalanced at the end of each month
and Illiq1,0 is computed using a minimum of 15 daily observations. In panel B, portfolios
are rebalanced each Tuesday, skipping signals and holding-period returns over Tuesday
itself to avoid undue influence from bid-ask bounce. That is, for the sort on Tuesday
of week t, the sorting variables are returns, turnover, and illiquidity measured from
Wednesday of week t 1 to Monday of week t, both inclusive, and the holding-period
return is computed from Wednesday of week t to Monday of week t+1, both inclusive.
For the weekly sorts, Illiq1,0 is computed using a minimum of four daily observations.
Data are at the monthly or weekly frequency and cover the beginning of July 1963 to the
end of December 2018.

monotonically decreasing with illiquidity; from 0.18% per week among

the most liquid stocks to 1.00% per week among the most illiquid
ones, and similarly for the abnormal returns. Hence, even though
there is no high-turnover continuation in any illiquidity quintile at
the weekly horizon, returns coupled with high turnover still reverse

34

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

much less among the most liquid stocks compared to the most illiquid
ones. These weekly-horizon results are consistent with the findings
of Avramov, Chordia, and Goyal (2006) but, again, directly oppose
the high-information-asymmetry mechanism. We conjecture that the
performance of the weekly STMOM strategies is due to temporary price
pressure dominating other e↵ects at the weekly horizon, but less so for
more liquid stocks.

3.1.2 Short-term momentum and analysts’ forecasts. Table 10

shows the performance of STMOM and STREV* strategies constructed
with a control for analysts’ forecast variables. The strategies are from 3⇥
3⇥3 conditional sorts on the control variable and the previous month’s
return and turnover. The breakpoints are the 20th and 80th percentiles
for NYSE stocks. Portfolios are value weighted and rebalanced at the
end of each month. The sample starts in January 1985 because of the
availability of data on analysts’ forecasts.25
The table’s first specification controls for the number of analysts
covering a stock. The STMOM strategy yields 1.14% per month (t =
3.20) among high-coverage stocks but a comparably much weaker
0.62% (t = 1.52) per month among low-coverage stocks. This again runs
opposite to the high-information-asymmetry mechanism. The second
specification shows that these results are not mechanically driven by
the fact that larger stocks tend to be covered by more analysts: Using
analyst coverage orthogonal to size yields qualitatively similar results.
Since STREV* is stronger among smaller stocks (Table 3) and smaller
stocks tend to have lower analyst coverage, we would expect to see the
strongest STREV* performance among low-coverage stocks. The first
two specifications confirm this.
Following Diether, Malloy, and Scherbina (2002), a large literature
uses measures of the dispersion in analysts’ forecasts as proxies for
heterogeneity or “disagreement” in beliefs among traders.26 While this
interpretation comes with many caveats, there is some consensus that
forecast dispersion is at least positively correlated with the underlying,
unobservable heterogeneity in beliefs (Verardo 2009; Banerjee 2011).
Since short-term momentum is strongest among high-coverage stocks
(presumably those with low information asymmetry), it is interesting
to examine whether it is also strongest among high-dispersion stocks

25
Analysts’ forecast data come from the Institutional Brokers’ Estimate System (I/B/E/S),
but we use only unadjusted forecasts to mitigate the reporting inaccuracies, rounding
errors, and look-ahead-biases identified in previous studies (Diether, Malloy, and Scherbina
2002). Our tests use earnings-per-share forecasts for the month closest to, but preceding,
the month in which a firm announces its quarterly earnings (Compustat’s RDQ).
26
This literature includes Johnson (2004), Verardo (2009), Banerjee (2011), Cen, Wei, and
Yang (2017), Cujean and Hasler (2017), and Loh and Stulz (2018), among many others.

35

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Table 10
Short-term momentum controlling for analysts’ forecast

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]
controlling for controlling for
analysts’ forecasts analysts’ forecasts

Control variable Quintile E[re ] ↵FF6 E[re ] ↵FF6

(1) N Analysts High 1.14 1.12 0.58 0.59

(3.20) (2.84) ( 2.41) ( 2.31)
Low 0.62 0.50 0.88 0.86
(1.52) (1.18) ( 3.69) ( 2.58)

(2) N Analysts ? M High 0.84 0.99 0.41 0.70

(2.50) (2.68) ( 1.27) ( 1.98)
Low 0.47 0.52 0.97 0.94
(1.15) (1.14) ( 3.21) ( 2.71)
d
(3) SD(EPS) High 1.38 1.28 0.28 0.23
(2.69) (2.31) ( 0.99) ( 0.79)
Low 0.32 0.39 1.13 1.10
(1.00) (1.32) ( 4.29) ( 3.20)
d
(4) SD(EPS)/P High 1.08 1.16 0.95 1.24
(2.06) (2.17) ( 2.51) ( 3.08)
Low 0.01 0.21 1.27 1.26
(0.03) (0.71) ( 5.45) ( 5.54)
d
(5) SD(EPS)/|Med( d
EPS)| High 0.97 1.07 0.74 0.91
(2.24) (1.98) ( 1.65) ( 1.93)
Low 0.25 0.39 1.45 1.61
( 0.70) ( 1.07) ( 4.69) ( 4.89)
d
(6) Max(EPS) d
Min(EPS) High 1.49 1.53 0.40 0.38
(3.20) (2.53) ( 1.17) ( 1.19)
Low 0.13 0.03 1.16 1.12
( 0.36) ( 0.08) ( 4.14) ( 2.74)
This table shows the performance of short-term momentum and short-term reversal*
strategies constructed with a control for analysts’ forecast variables. The strategies are
from 3⇥3⇥3 conditional sorts on the control variable and the previous month’s return
and turnover, in that order. The breakpoints are the 20th and 80th percentiles for
NYSE stocks. Portfolios are value weighted and rebalanced at the end of each month.
Test statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation.
All analysts’ forecast variables are for the month closest to, but preceding, the month
in which a firm announces its quarterly earnings (Compustat’s RDQ). N Analysts is
the number of analysts following a stock. N Analysts ? M is the residual from monthly
cross-sectional regressions of log(N Analysts ) on log(M ), where M is market equity from
CRSP, estimated using weighted least squares (WLS) with market capitalization as the
d is analysts’ forecasts of firms’ earnings-per-share and P is price per share
weight. EPS
from CRSP. “SD” denotes standard deviation, and “Med” denotes median. The sample
excludes financial firms. Data are at the monthly frequency and cover January 1985
to December 2018, where the start date is determined by the availability of analysts’
forecast data.

(presumably those with strong disagreement among traders). The table’s

36

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

remaining four specifications show that this is indeed the case. In

contrast, STREV* is strongest among low-dispersion stocks.

3.2 Models of boundedly rational traders

The literature on “boundedly rational” traders suggests an alternative
mechanism that potentially explains short-term momentum: underin-
ference from prices.27 In the following, we relate our results to general
features of models that relax the strict rationality assumption.
Under REE, traders condition on prices so as to infer each others’
information up to any noise introduced by noninformational trading.
However, when boundedly rational traders fail to fully infer others’
information from prices, expected volume is higher and prices underreact
to the available information relative to the case with solely rational
traders. Underreaction in turn induces positive return autocorrelation
that counteracts the negative one induced by noninformational trading.
Hence, if there is underinference from prices but no noninformational
trading, return autocorrelation will be positive and increasing in volume.
Hong and Stein (2007), for instance, show that “the magnitude of the
momentum e↵ect will be increasing with average trading volume” (p.
122) in a simple setup where trading is solely due to private information
and where there is no conditioning on prices. More generally, if there
is some underinference from prices as well as some noninformational
trading (e.g., Banerjee 2011; Eyster, Rabin, and Vayanos 2019),
high volume will be associated with return continuation so long as
underinference is not overwhelmed by noninformational trading.
As such, the cross-sectional variation in short-term momentum is
suggestive of an explanation based on underinference prices: It is
strongest among the largest and most liquid stocks, whose returns tend
to be less a↵ected by temporary price pressure (Avramov, Chordia, and
Goyal 2006; Nagel 2012; Hendershott and Menkveld 2014), and it is also
stronger among stocks with greater dispersion in analysts’ forecasts,
which is a common proxy for disagreement among traders (Diether,
Malloy, and Scherbina 2002; Verardo 2009; Banerjee 2011).
Lastly, to assist in di↵erentiating between fully and boundedly
rational explanations, we investigate how trading volume a↵ects the
ability of realized returns to predict firms’ fundamentals. The idea is

27
Examples include Odean’s (1998) “overconfident” traders who condition on prices
but exaggerate the precision of their own signals (see also Daniel, Hirshleifer, and
Subrahmanyam 1998); Hong and Stein’s (1999) “newswatchers” that trade on their private
information but do not condition on prices (see also Hong and Stein 2007); Hirshleifer, Lim,
and Teoh’s (2011) “inattentive” traders who condition on a subset of publicly available
information, but not on prices; Banerjee’s (2011) “dismissive” traders who condition on
prices but downplay the precision of others’ signals; Eyster, Rabin, and Vayanos’ (2019)
“cursed” traders who do not fully condition on prices; and Mondria, Vives, and Yang’s
(2020) “unsophisticated” traders who condition on a noisy version of the price for inference
purposes.

37

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

that, in models that allow for both underinference and noninformational

trading (e.g., Banerjee 2011; Eyster, Rabin, and Vayanos 2019), volume
will have di↵erent e↵ects on the relation between realized returns
and subsequent fundamental values depending on the relative strength
of underinference versus noninformational trading, just as discussed
above for the e↵ects of volume on return autocorrelation in such
models. Because fully rational traders use prices to infer each others’
information up to any noise introduced by noninformational trading,
“rational volume” ultimately emanates from noninformational trading.
Higher rational volume should therefore decrease the predictive ability
of returns for fundamentals. By contrast, because boundedly rational
traders underappreciate the information conveyed by prices, “boundedly
rational” volume will reflect both underreaction to the available
information and any noninformational trading. Higher boundedly
rational volume should therefore increase the predictive ability of
returns for fundamentals if underinference is not overwhelmed by
noninformational trading.
To test this prediction, we use Fama and MacBeth predictive
regressions of firms’ fundamental growth rates (at horizons of 1, 3,
and 5 years) on 1-month returns, turnover, and their interaction
(r1,0 ⇥TO 1,0 ). We consider the growth in gross profits (REVT
COGS) in Table 11 and the growth in earnings (IB) in Table 12
and control for predictors commonly employed in the literature. All
independent variables are measured at the end of firms’ fiscal years.
We use weighted least squares (WLS) with market capitalization as the
weight. Dependent and independent variables are trimmed at the 1st
and 99th percentiles. Independent variables are standardized by their
cross-sectional average and standard deviation. The interaction is the
product of the standardized variables.
The slope on r1,0 is positive in all specifications and also significant,
except for 5-year earnings growth when employed alongside the controls.
The slope of the interaction should be positive if there is detectable
underinference from prices, but otherwise should be negative. As such,
the regressions are suggestive of detectable underinference. At the 1-year
horizon, the slope on the interaction is positive and significant, with
or without controls, for both gross-profit growth and earnings growth.
At the 3- and 5-year horizons, although the slope on the interaction
loses its significance, its point estimate never becomes negative. The fact
that the slope on the interaction loses its significant at longer horizons
suggests that the interaction reveals information about fundamentals in
the relatively near term, and is broadly consistent with our previous
result that short-term momentum returns on average persist for a year
after portfolio formation (Figure 2).

38

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

Table 11
Cross-sectional regressions to predict the growth in gross profits

Average slopes (⇥100) and test statistics (in parentheses) from

GP i,t+⌧ GP it 0
WLS regressions of the form Ait = t Xit +✏it

1-year 3-year 5-year

Independent (1) (2) (3) (4) (5) (6)

variables

r1,0 0.79 0.70 1.24 0.86 1.51 0.95

(5.15) (9.57) (3.67) (4.98) (2.71) (3.15)
TO 1,0 0.76 0.04 1.90 0.18 2.98 0.52
(4.30) ( 0.35) (2.49) ( 0.39) (2.18) ( 0.57)
r1,0 ⇥TO 1,0 0.19 0.15 0.14 0.12 0.36 0.40
(2.54) (2.96) (0.52) (0.66) (0.96) (1.47)
r12,2 1.54 2.00 2.76
(8.33) (5.90) (5.27)
Illiq1,0 0.19 1.00 1.00
( 1.55) ( 1.74) ( 1.09)
GP/A 1.94 6.38 11.78
(10.97) (8.45) (8.12)
IB/B 1.10 2.50 3.65
( 5.37) ( 3.72) ( 3.17)
D&R/B 0.50 1.91 3.64
( 9.41) ( 9.01) ( 9.40)
dA/A 1 1.49 2.88 5.16
(7.85) (7.36) (5.97)
log(B/M ) 1.75 5.34 9.58
( 16.64) ( 14.67) ( 20.09)
Size 0.26 1.49 3.04
( 2.67) ( 6.70) ( 4.66)
Adj. R2 3.5% 29.1% 2.9% 32.0% 2.6% 35.4%
This table shows Fama and MacBeth cross-sectional regressions of firms’ growth in
gross profits on 1-month returns, turnover, and their interaction (r1,0 ⇥TO 1,0 ). All
independent variables are measured at the end of firms’ fiscal years. Regressions are
estimated using weighted least squares (WLS) with market capitalization as the weight.
Dependent and independent variables are trimmed at the 1st and 99th percentiles.
Independent variables are standardized by their cross-sectional average and standard
deviation. The interaction is the product of the standardized variables. Test statistics (in
parentheses) are adjusted for heteroscedasticity and autocorrelation. Controls are 12-2
months performance (r12,2 ), 1-month Amihud illiquidity (Illiq1,0 ), gross profits-to-assets
(GP/A), earnings-to-book equity (IB/B), dividends and repurchases-to-book equity
(D&R/B, where D&R is DVC + PRSTKCC, both set to zero if missing), asset growth
(dA/A 1 ), book-to-market equity (log(B/M )), and Size (log of market capitalization).
The sample excludes financial firms and firms with negative book equity. Data are at the
annual frequency and cover 1963 to 2018.

39

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Table 12
Cross-sectional regressions to predict the growth in earnings

Average slopes (⇥100) and test statistics (in parentheses) from

IB i,t+⌧ IB it 0
WLS regressions of the form Bit = t Xit +✏it

1-year 3-year 5-year

Independent (1) (2) (3) (4) (5) (6)

variables

r1,0 0.85 0.77 1.00 0.75 0.87 0.37

(4.87) (5.08) (4.79) (3.29) (2.62) (1.21)
TO 1,0 0.06 0.37 0.13 0.65 0.02 0.92
(0.38) ( 2.38) ( 0.27) ( 1.28) (0.04) ( 1.83)
r1,0 ⇥TO 1,0 0.28 0.23 0.00 0.09 0.24 0.25
(2.83) (2.31) (0.00) (0.52) (1.19) (1.74)
r12,2 2.06 1.07 0.98
(9.06) (2.91) (5.07)
Illiq1,0 0.41 0.51 0.95
( 2.13) ( 1.38) ( 2.15)
GP/A 0.87 1.36 1.90
(4.41) (2.90) (3.00)
IB/B 9.02 12.22 14.60
( 4.16) ( 4.04) ( 3.46)
D&R/B 0.42 0.31 0.04
(5.66) (1.34) ( 0.12)
dA/A 1 0.65 0.65 0.03
( 2.81) ( 1.69) ( 0.05)
log(B/M ) 1.67 3.30 4.82
( 7.14) ( 10.33) ( 16.50)
Size 0.98 0.95 0.87
(2.78) (1.03) (0.99)
Adj. R2 2.3% 16.7% 2.6% 14.7% 2.2% 14.5%
This table shows Fama and MacBeth cross-sectional regressions of firms’ growth in
earnings on 1-month returns, turnover, and their interaction (r1,0 ⇥TO 1,0 ). All
independent variables are measured at the end of firms’ fiscal years. Regressions are
estimated using weighted least squares (WLS) with market capitalization as the weight.
Dependent and independent variables are trimmed at the 1st and 99th percentiles.
Independent variables are standardized by their cross-sectional average and standard
deviation. The interaction is the product of the standardized variables. Test statistics (in
parentheses) are adjusted for heteroscedasticity and autocorrelation. Controls are 12-2
months performance (r12,2 ), 1-month Amihud illiquidity (Illiq1,0 ), gross profits-to-assets
(GP/A), earnings-to-book equity (IB/B), dividends and repurchases-to-book equity
(D&R/B, where D&R is DVC + PRSTKCC, both set to zero if missing), asset growth
(dA/A 1 ), book-to-market equity (log(B/M )), and Size (log of market capitalization).
The sample excludes financial firms and firms with negative book equity. Data are at the
annual frequency and cover 1963 to 2018.

40

Electronic copy available at: https://ssrn.com/abstract=3150525

 Short-term Momentum

4. Conclusion
Momentum (the tendency for winners to outperform losers) and reversal
(the tendency for losers to outperform winners) coexist with almost
equal strengths at the 1-month horizon. While the previous month’s
low-turnover stocks exhibit a strong short-term reversal e↵ect ( 16.9%
per annum), the previous month’s high-turnover stocks exhibit an
almost equally strong continuation e↵ect (+16.4% per annum), which
we dub short-term momentum. This finding is not limited to the
United States but extends to 22 international developed markets. We
show that short-term momentum generates significant abnormal returns
relative to standard factors; persists for 12 months; is stronger with an
implementation lag; survives conservative estimates of transaction costs;
and is strongest among the largest, most liquid, and most extensively
covered stocks.
Short-term momentum and conventional price momentum share the
same basic philosophy: both are a bet on recent winners financed
by a bet against recent losers, and both are designed to avoid
diluting the continuation e↵ect with short-term reversal. However, while
conventional momentum does this by skipping the most recent month,
short-term momentum’s trading signal is the performance over the most
recent month, and it instead avoids reversal by conditioning on high
share turnover in the most recent month. The two are alike in terms of
profitability and persistence and are related in terms of characteristic
tilts, portfolio overlap, and time-series correlation. However, there are
also noteworthy di↵erences between them. While short-term momentum
is strongest among the largest and most extensively covered stocks,
Hong, Lim, and Stein (2000) find that “once one moves past the very
smallest capitalization stocks (where thin market making capacity does
indeed appear to be an issue) the profitability of momentum strategies
declines sharply with market capitalization” and “momentum strategies
work particularly well among stocks that have low analyst coverage” (p.
267). In addition, short-term momentum exhibits far less crash risk than
does conventional momentum and, unlike conventional momentum, does
not appear to be spanned or driven by other, well-known momentum
e↵ects, such as those in earnings, industry returns, and factor returns.
As such, an open question for future research is whether short-term
momentum and conventional momentum are the same phenomenon in
di↵erent guises or related phenomena with important di↵erences.
Theoretically, the existence of a high-turnover continuation e↵ect is
postulated by the models of Wang (1994) and Llorente et al. (2002). Still,
these models’ high-information-asymmetry mechanism is not supported
in the data as an explanation for short-term momentum. While high-
information-asymmetry stocks should be small, illiquid, and have low

41

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

analyst coverage, we find the exact opposite for short-term momentum:

It is strongest among the largest, most liquid, and most extensively
covered stocks. Notwithstanding these results, size, liquidity, and analyst
coverage may simply be imperfect proxies for information asymmetry.
Finding better proxies and retesting the mechanism is certainly an
interesting avenue for future research. Alternatively, models that relax
the strict rationality assumption and instead allow for underinference
from prices by some traders (a mild form of bounded rationality)
would predict that short-term momentum should be stronger among
stocks where any underinference from prices is not overwhelmed by
noninformational trading. We argue that the cross-sectional variation in
short-term momentum is broadly consistent with such an explanation:
It is stronger among stocks less likely to be a↵ected by temporary price
pressure and among stocks more likely to cause disagreement among
traders. Refining this argument, including tests of additional predictions
and formally discriminating between alternative classes of models, is also
an interesting avenue for future research.

Appendix. Tables and figures omitted from the main text

This appendix contains tables and figures omitted from the main text. Additional
results can be found in the paper’s Internet Appendix.

42

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

Table A.1
Portfolio characteristics for double sorts on 1-month returns and turnover

r1,0 deciles

TO 1,0 deciles Low 2 3 4 5 6 7 8 9 High

Portfolio r1,0
Low 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.19
2 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.18
3 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.17
4 0.13 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.18
5 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.18
6 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.09 0.18
7 0.14 0.07 0.04 0.02 0.00 0.02 0.04 0.06 0.10 0.19
8 0.14 0.08 0.04 0.02 0.00 0.02 0.04 0.06 0.10 0.20
9 0.16 0.08 0.04 0.02 0.00 0.02 0.04 0.06 0.10 0.22
High 0.18 0.08 0.04 0.02 0.00 0.02 0.04 0.06 0.10 0.28
Portfolio TO 1,0
Low 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02
2 0.04 0.03 0.03 0.03 0.03 0.03 0.03 0.04 0.04 0.05
3 0.06 0.05 0.04 0.04 0.04 0.04 0.04 0.04 0.05 0.06
4 0.07 0.06 0.05 0.05 0.05 0.05 0.05 0.05 0.06 0.08
5 0.09 0.07 0.06 0.06 0.06 0.06 0.06 0.06 0.07 0.10
6 0.11 0.08 0.07 0.07 0.07 0.07 0.07 0.07 0.08 0.12
7 0.14 0.10 0.09 0.08 0.08 0.08 0.08 0.09 0.10 0.14
8 0.17 0.12 0.10 0.10 0.09 0.09 0.10 0.11 0.12 0.18
9 0.23 0.15 0.13 0.12 0.12 0.12 0.13 0.13 0.16 0.24
High 0.43 0.27 0.24 0.22 0.21 0.21 0.22 0.24 0.28 0.47
Average market capitalization ($ mio.)
Low 176 414 803 1,003 985 1,174 1182 1,227 952 364
2 530 1,880 2,957 3,642 4,806 4,167 4,212 3,866 2,789 994
3 711 2,169 3,140 4,219 4,481 4,498 4,439 4,117 3,340 1,290
4 930 2,275 3,262 3,676 3,753 4,009 3,945 3,875 3,116 1,265
5 905 2,376 2,815 3,445 3,753 3,566 3,842 3,550 3,016 1,396
6 934 2,485 2,829 3,485 3,499 3,335 3,391 3,411 2,654 1,284
7 923 2,142 2,488 3,010 3,259 3,347 3,167 2,990 2,389 1,206
8 925 1,886 2,198 2,660 2,811 2,805 2,858 2,565 2,154 1,158
9 855 1,675 1,984 2,079 2,180 2,496 2,364 2,151 1,919 1,132
High 722 1,403 1,589 1,697 1,706 1,756 1,667 1,779 1,511 898
Average number of stocks
Low 177 95 83 74 68 66 63 66 72 124
2 78 41 34 30 28 27 27 29 33 59
3 58 32 27 24 23 23 22 24 27 47
4 49 29 24 22 21 21 21 22 24 42
5 44 26 22 21 20 20 20 21 23 39
6 39 24 21 20 19 19 20 20 22 36
7 37 24 21 20 20 19 19 20 22 36
8 36 24 22 20 20 20 20 21 23 37
9 35 24 22 21 21 21 21 22 24 37
High 41 29 27 26 26 26 26 27 29 47
This table shows the time-series averages of portfolio characteristics for the double sorts
on the previous month’s return (r1,0 ) and turnover (TO 1,0 ) in Table 1. Portfolio r1,0
is the time-series average of each portfolios’ monthly value-weighted r1,0 , and similarly
for portfolio TO 1,0 . Average market capitalization is the time-series average of each
portfolio’s monthly equal-weighted average market capitalization in millions of dollars.
Average number of stocks is the time-series average of each portfolios’ monthly number
of stocks. The sample excludes financial firms. Data are at the monthly frequency and
cover July 1963 to December 2018.

43

Electronic copy available at: https://ssrn.com/abstract=3150525

 Table A.2
Independent double sorts on the previous month’s return and turnover

r1,0 deciles r1,0 strategies

Low 2 3 4 5 6 7 8 9 High E[re ] ↵FF6 ↵q

44
TO 1,0 deciles
Portfolio excess return

Low 1.28 1.28 0.83 0.93 0.63 1.01 0.78 0.74 0.38 0.14 1.42 1.53 1.57
( 6.11) ( 5.95) ( 6.27)
2 1.71 1.06 1.02 0.95 1.05 0.88 0.75 0.73 0.64 0.03 1.74 1.76 1.73
( 5.41) ( 4.61) ( 4.51)
3 1.36 1.27 1.27 1.02 1.16 1.00 0.59 0.75 0.36 0.47 0.89 1.05 1.15
( 2.54) ( 2.75) ( 3.04)
4 1.62 1.46 1.40 1.36 0.90 1.06 0.82 0.59 0.82 0.09 1.53 1.82 1.73
( 5.15) ( 5.88) ( 4.90)
5 1.52 1.55 1.38 1.02 1.21 0.86 0.86 0.98 0.61 0.46 1.06 1.29 1.37
( 3.22) ( 3.00) ( 3.37)
6 1.55 1.10 1.32 1.24 0.82 0.84 0.99 1.07 0.77 0.72 0.83 0.96 0.90
( 2.59) ( 2.63) ( 2.36)
7 1.13 1.01 1.17 0.95 1.04 1.21 0.92 0.98 0.80 0.49 0.64 0.77 0.67
( 3.10) ( 2.73) ( 2.39)
The Review of Financial Studies / v 00 n 0 2021

8 1.33 1.10 1.14 1.06 1.19 1.18 0.96 0.84 0.88 0.79 0.55 0.43 0.40
( 2.07) ( 1.39) ( 1.18)
9 0.94 1.10 1.17 1.24 1.14 0.95 1.05 1.26 1.00 0.69 0.25 0.31 0.23
( 0.91) ( 1.07) ( 0.56)
High 0.12 0.78 0.98 1.00 0.86 0.48 0.89 0.96 0.90 1.12 1.00 0.98 1.06
(4.02) (3.40) (3.31)

TO 1,0 strategies

E[re ] 1.16 0.50 0.15 0.08 0.23 0.53 0.10 0.23 0.52 1.26

Electronic copy available at: https://ssrn.com/abstract=3150525

( 4.63) ( 1.93) (0.46) (0.23) (0.68) ( 1.72) (0.35) (0.75) (1.72) (5.06)
↵FF6 1.15 0.34 0.40 0.07 0.33 0.35 0.28 0.34 0.74 1.36
( 4.47) ( 1.21) (1.33) (0.23) (1.04) ( 1.22) (1.01) (1.17) (2.73) (5.26)
↵q 1.18 0.16 0.38 0.04 0.31 0.38 0.23 0.26 0.78 1.46
( 4.38) ( 0.59) (1.14) (0.11) (0.87) ( 1.18) (0.77) (0.74) (2.45) (5.19)

This table shows portfolios from independent double sorts on the previous month’s return (r1,0 ) and turnover (TO 1,0 ). We use sorts into deciles based
on NYSE breakpoints. Portfolios are value weighted and rebalanced at the end of each month. The table also shows the performance of long-short
strategies across the deciles. Test statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation. Time-series averages of the
portfolios’ characteristics are provided in Table IA.1 in the Internet Appendix. Data are at the monthly frequency and cover July 1969 to December
2018, where the start date ensures nonempty portfolios as a result of the independent sorts.
Short-term Momentum

Table A.3
Independent quintile double sorts on 1-month returns and turnover excluding
end-of-month e↵ects

A. Portfolio excess returns and strategy performance

r1,0 EOM quintiles r1,0 EOM strategies

TO 1,0 EOM Low 2 3 4 High E[re ] ↵FF6 ↵q

quintiles

Portfolio excess return

Low 0.50 0.28 0.47 0.49 0.23 0.26 0.30 0.26
( 1.57) ( 1.83) ( 1.39)
2 0.75 0.73 0.56 0.35 0.33 0.42 0.51 0.45
( 2.26) ( 2.61) ( 2.16)
3 0.62 0.76 0.63 0.59 0.51 0.12 0.23 0.10
( 0.64) ( 1.13) ( 0.41)
4 0.57 0.72 0.68 0.59 0.54 0.04 0.02 0.09
( 0.21) (0.09) (0.34)
High 0.00 0.81 0.72 0.60 0.74 0.74 0.77 0.86
(4.21) (4.03) (3.37)
TO 1,0 EOM
strategies

E[re ] 0.50 0.53 0.25 0.11 0.51

( 2.88) (2.50) (1.07) (0.48) (2.87)
↵FF6 0.52 0.51 0.34 0.10 0.55
( 3.04) (2.89) (1.84) (0.57) (3.92)
↵q 0.52 0.55 0.31 0.02 0.60
( 3.04) (2.85) (1.62) (0.07) (3.29)

B. Portfolio characteristics
r1,0 EOM quintiles r1,0 EOM quintiles

TO 1,0 EOM Low 2 3 4 High Low 2 3 4 High

Portfolio r1,0 EOM Portfolio TO 1,0 EOM

Low 0.10 0.03 0.00 0.04 0.12 0.02 0.02 0.02 0.02 0.02
2 0.09 0.03 0.00 0.04 0.11 0.04 0.04 0.04 0.04 0.04
3 0.09 0.03 0.00 0.04 0.11 0.06 0.06 0.06 0.06 0.06
4 0.10 0.03 0.00 0.04 0.12 0.09 0.09 0.09 0.09 0.09
High 0.12 0.03 0.00 0.04 0.16 0.20 0.18 0.18 0.18 0.21

Average M ($ mio.) Average N

Low 379 1,490 2,082 2,000 732 439 275 246 227 286
2 1,275 3,150 3,850 3,685 1,939 145 106 102 99 116
3 1,367 2,895 3,315 3,261 2,013 126 90 85 89 115
4 1,297 2,334 2,622 2,571 1,719 134 81 74 83 137
High 1,016 1,634 1,800 1,747 1,218 180 70 61 75 222
This table shows portfolios double sorted on the previous month’s return and turnover,
while skipping the sorting variables’ end-of-month values (r1,0 EOM and TO 1,0 EOM )
measured at the month’s last 3 trading days. We use independent sorts into quintiles
based on NYSE breakpoints. Portfolios are value weighted and rebalanced at the
end of each month. Panel A shows portfolio excess returns and the performance of
long-short strategies across the quintiles. Test statistics (in parentheses) are adjusted
for heteroscedasticity and autocorrelation. Panel B shows the time-series averages of the
monthly value-weighted average portfolio characteristics as well as the equal-weighted
average market capitalization and the number of stocks. The sample excludes financial
firms. Data are at the monthly frequency and cover July 1963 to December 2018, except
for the q-factors, which are available from January 1967.

45

Electronic copy available at: https://ssrn.com/abstract=3150525

 46
Table A.4
Short-term momentum’s investment allocation and capacity

A B
Average fraction of each $1 investment
allocated to each NYSE size quintile (%) Investment capacity

Average December
Strategy Leg Micro 2 3 4 Mega fraction (%) 2018 ($B)

Short-term momentum Long 8.5 12.9 19.2 27.1 32.3 0.69 71.4
Short 8.8 14.4 21.3 27.1 28.4 0.57 38.7

Short-term reversal* Long 27.4 14.2 13.8 16.1 28.5 0.88 43.0
Short 39.2 13.9 12.0 13.7 21.1 0.62 12.1

Short-term momentum Long 8.7 13 19.1 27.1 31.9 0.69 96.4

(excl. EOM) Short 8.8 14.1 20.9 27.1 29.1 0.56 53.8
The Review of Financial Studies / v 00 n 0 2021

Short-term momentum Long 5.7 9.1 14.1 23.1 48.1 4.25 387.4
(5⇥5, excl. EOM) Short 6.3 9.9 15.2 23.7 45.0 3.10 863.6

Short-term momentum Long 0.0 0.0 11.4 32.7 56.0 2.31 353.3
(largecaps, excl. EOM) Short 0.0 0.0 10.7 31.9 57.3 2.21 465.7

Short-term momentum Long 0.0 0.0 0.0 0.0 100.0 2.17 587.0
(megacaps, excl. EOM) Short 0.0 0.0 0.0 0.0 100.0 2.07 650.3
This table shows the investment allocation and capacity for the strategies considered in Table 7. Panel A shows, for each strategy, the time-series

Electronic copy available at: https://ssrn.com/abstract=3150525

average of the underlying portfolios’ investment allocation to each NYSE size quintile, that is, how each portfolio on average allocates each invested
dollar to each NYSE size quintile. Panel B shows, for each strategy, the investment capacity of the underlying portfolios measured in two ways: (1)
the time-series average of the portfolio’s market capitalization as a percentage of the aggregate market capitalization and (2) the portfolio’s market
capitalization in billions of dollars at the end of the sample in December 2018.
 Table A.5
Short-term momentum and industry adjustments

A B
Decomposition of benchmark strategy performance into
an industry-hedged (“within-industry”) component and the industry hedge
Strategies from sorts on
industry-demeaned variables Benchmark strategies Industry-hedged strategies The industry hedges

Turnover E[re ] ↵FF6 ↵q E[re ] ↵FF6 ↵q E[re ] ↵FF6 ↵q E[re ] ↵FF6 ↵q

decile
Short-term Momentum

Low 2.00 2.12 2.07 1.41 1.45 1.43 1.73 1.74 1.73 0.32 0.29 0.30
( 10.25) ( 8.51) ( 7.25) ( 7.13) ( 6.19) ( 5.95) ( 8.41) ( 8.26) ( 7.52) (3.68) (3.27) (3.00)
2 1.73 1.94 1.84 1.19 1.31 1.34 1.62 1.78 1.76 0.43 0.47 0.42
( 7.57) ( 6.61) ( 4.87) ( 4.61) ( 4.04) ( 4.21) ( 7.00) ( 5.93) ( 6.62) (3.74) (4.26) (3.04)
3 1.75 1.95 1.93 1.34 1.62 1.66 1.65 1.83 1.87 0.30 0.21 0.22
( 7.79) ( 6.43) ( 4.83) ( 5.02) ( 5.61) ( 4.87) ( 7.32) ( 7.37) ( 6.50) (2.57) (1.57) (1.52)
4 2.08 2.23 2.32 0.85 1.02 0.91 1.22 1.33 1.21 0.37 0.31 0.31
( 7.75) ( 8.63) ( 9.15) ( 3.63) ( 4.15) ( 2.92) ( 5.84) ( 6.08) ( 5.25) (3.21) (2.53) (2.20)
5 1.09 1.28 1.08 0.45 0.63 0.51 0.96 1.08 0.96 0.51 0.46 0.44
( 4.37) ( 4.58) ( 3.72) ( 1.94) ( 2.29) ( 1.54) ( 4.63) ( 4.71) ( 3.57) (4.50) (3.85) (3.14)
6 0.96 0.97 1.02 0.59 0.60 0.41 0.94 1.05 0.97 0.35 0.45 0.56
( 3.95) ( 3.53) ( 3.27) ( 2.50) ( 2.35) ( 1.20) ( 4.88) ( 4.52) ( 3.56) (2.74) (3.35) (3.08)
7 0.89 0.91 0.92 0.67 0.85 0.96 0.94 1.09 1.19 0.27 0.23 0.23
( 3.32) ( 2.85) ( 2.61) ( 2.52) ( 2.55) ( 2.37) ( 4.13) ( 3.77) ( 3.85) (2.31) (1.56) (1.36)
8 0.33 0.44 0.42 0.23 0.13 0.21 0.19 0.22 0.19 0.42 0.35 0.40
( 1.24) ( 1.75) ( 1.38) (0.85) (0.47) (0.60) ( 0.81) ( 1.00) ( 0.69) (4.01) (2.96) (2.98)
9 0.04 0.21 0.10 0.05 0.00 0.19 0.12 0.25 0.12 0.17 0.26 0.31
( 0.17) ( 0.67) ( 0.29) (0.21) (0.01) (0.55) ( 0.58) ( 0.98) ( 0.43) (1.56) (2.31) (2.17)

Electronic copy available at: https://ssrn.com/abstract=3150525

High 1.18 0.95 1.16 1.37 1.37 1.65 0.89 0.86 1.06 0.48 0.51 0.59
(4.08) (2.78) (2.94) (4.74) (4.22) (4.47) (3.35) (2.87) (3.31) (5.65) (5.11) (4.68)
This table shows the performance of winner-minus-loser strategies similar to those in Table 1, except that we consider industry-adjusted signals or
performance. We use conditional sorts into deciles based on NYSE breakpoints, first on returns and then on turnover. Portfolios are value weighted
and rebalanced at the end of each month. Panel A shows strategies from sorts where the sorting variables are demeaned by their value-weighted
average industry values. Panel B shows a decomposition of the benchmark strategies (from sorts on unadjusted variables) into an industry-hedged

47
(“within-industry”) component and the industry hedge. The industry-hedged strategies are from sorts on unadjusted variables, but each stock’s
position is combined with an o↵setting position of equal size in the corresponding value-weighted industry portfolio. The industry hedges are these
o↵setting positions in the value-weighted industry portfolios. The industries are the Fama and French 49 industries. Test statistics (in parentheses) are
adjusted for heteroscedasticity and autocorrelation. The sample excludes financial firms. Data are at the monthly frequency and cover January 1963 to
December 2018.
 Table A.6
Short-term momentum and crash risk

A. Higher-order moments and downside beta

Statistic Market excess return Short-term momentum Conventional momentum

48
Skew 0.55 0.48 5.74
( 7.08) ( 6.24) ( 29.72)
Kurtosis 9.87 7.55 74.05
(12.73) (10.94) (20.65)
Co-skewness 0.47 1.87
(1.60) ( 8.05)
Down-side 0.07 0.71
(0.52) ( 6.35)
B. Regression results
Intercepts, slopes, and test-statistics (in parentheses)
from time-series regressions of strategy returns

Short-term momentum Conventional momentum

Coefficient Variable (1) (2) (3) (4) (5) (6)

↵0 1 1.15 1.50 1.50 1.33 1.22 1.22

(4.33) (5.02) (5.04) (6.20) (5.30) (5.35)
↵B 1B,t 1 1.65 3.29 0.63 1.45
The Review of Financial Studies / v 00 n 0 2021

( 2.55) ( 3.78) ( 1.28) (2.19)

e
0 rmt 0.20 0.22 0.22 0.27 0.19 0.19
( 4.09) ( 3.15) ( 3.16) ( 6.70) (3.58) (3.61)
e
B 1B,t 1 ⇥rmt 0.03 0.29 0.94 0.53
(0.34) ( 1.92) ( 12.53) ( 4.59)
e
B,U 1B,t 1 ⇥1U,t ⇥rmt 0.56 0.72
(2.80) ( 4.66)
Adj. R2 1.4% 1.9% 2.5% 3.8% 16.7% 18.2%
This table shows downside and crash risk measures for the short-term momentum strategy. For comparison, it also shows the corresponding measures
for the market excess return and/or a conventional momentum strategy from decile sorts on prior 12-2 month performance using NYSE breakpoints.

Electronic copy available at: https://ssrn.com/abstract=3150525

Panel A shows, for each strategy, the skewness and kurtosis of log(1+rte +rf t ), where rte is the strategy’s monthly excess return and rf t is the
monthly risk-free rate. The corresponding test statistics (in parentheses) are for the null of normally distributed returns. Panel A also shows the two
e
momentum strategies’ coskewness with the market, computed as the slope coefficient from a univariate regression of "t (r e ) on (rmt )2 , where the
e e e
latter is the squared excess market return and where "t (r ) is the residual from a univariate regression of rt on rmt . Finally, panel A shows the two
e e
momentum strategies’ downside , computed as the coefficient D from the regression rte = ↵+ rmt + D max{0, rmt }+✏t .
Panel B shows results from Daniel and Moskowitz’s (2016) regressions, where the most general form is rte = (↵0 +↵B IB,t 1 )+
e
( 0 +( B + B,U IU,t )IB,t 1 )rmt +✏t . Here, IB,t 1 is an ex ante “bear market” indicator that equals one if the cumulative market excess
return over the previous 24 months is negative and is otherwise zero, and IU,t is a contemporaneous “up-market” indicator that equals one if the
market excess return is positive in month t and is otherwise zero. The intercepts, ↵0 and ↵B , are multiplied by 100 and are thus stated in % per month.
The sample excludes financial firms. Data are at the monthly frequency and cover January 1927 to December 2018.
Short-term Momentum

References

Amihud, Y. 2002. Illiquidity and stock returns: Cross-section and time-series e↵ects.
Journal of Financial Markets 5:31–56.

Asness, C. S. 1995. The Power of past stock returns to explain future stock returns.
Working Paper, AQR Capital Management.

Asness, C. S., A. Frazzini, and L. H. Pedersen. 2019. Quality minus junk. Review of
Accounting Studies 24:34–112.

Asness, C. S., R. B. Porter, and R. Stevens. 2000. Predicting stock returns using industry-
relative firm characteristics. Working Paper, AQR Capital Management.

Avramov, D., T. Chordia, and A. Goyal. 2006. Liquidity and autocorrelations in individual
stock returns. Journal of Finance 61:2365–94.

Ball, R., J. Gerakos, J. T. Linnainmaa, and V. V. Nikolaev. 2016. Accruals, cash flows,
and operating profitability in the cross section of stock returns. Journal of Financial
Economics 121:28–45.

Bandarchuk, P., and J. Hilscher. 2013. Sources of momentum profits: Evidence on the
irrelevance of characteristics. Review of Finance 1:809–45.

Banerjee, S. 2011. Learning from prices and the dispersion in beliefs. Review of Financial
Studies 24:3025–68.

Bollerslev, T., B. Hood, J. Huss, and L. H. Pedersen. 2018. Risk everywhere: Modeling
and managing volatility. Review of Financial Studies 31:2729–73.

Campbell, J. Y., S. J. Grossman, and J. Wang. 1993. Trading volume and serial correlation
in stock returns. Quarterly Journal of Economics 108:905–39.

Cen, L., K. C. J. Wei, and L. Yang. 2017. Disagreement, underreaction, and stock returns.
Management Science 63:1214–31.

Cespa, G., A. Gargano, S. J. Riddiough, and L. Sarno. 2020. Foreign exchange volume.
Working Paper, Cass Business School.

Conrad, J. S., A. Hameed, and C. Niden. 1994. Volume and autocovariances in short-
horizon individual security returns. Journal of Finance 49:1305–29.

Cooper, M. 1999. Filter rules based on price and volume in individual security
overreaction. Review of Financial Studies 12:901–35.

Cujean, J., and M. Hasler. 2017. Why does return predictability concentrate in bad times?
Journal of Finance 72:2717–58.

Da, Z., Q. Liu, and E. Schaumburg. 2012. A closer look at the short-term return reversal.
Management Science 60:658–74.

Daniel, K., and D. Hirshleifer. 2015. Overconfident investors, predictable returns, and
excessive trading. Journal of Economic Perspectives 29:61–88.

Daniel, K., D. Hirshleifer, and A. Subrahmanyam. 1998. Investor psychology and security
market under- and overreactions. Journal of Finance 53:1839–85.

Daniel, K., and T. J. Moskowitz. 2016. Momentum crashes. Journal of Financial

Economics 122:221–47.

De Bondt, W., and R. Thaler. 1985. Does the stock market overreact? Journal of
Finance 40:793–805.

Diether, K. B., C. J. Malloy, and A. Scherbina. 2002. Di↵erences of opinion and the cross
section of stock returns. Journal of Finance 57:2113–41.

49

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Drechsler, I., A. Moreira, and A. Savov. 2018. Liquidity creation as volatility risk.
Working Paper, Wharton School.

Ehsani, S., and J. Linnainmaa. 2020. Factor momentum and the momentum factor.
Working Paper, Northern Illinois University.

Etula, E., K. Rinne, M. Suominen, and L. Vaittinen. 2020. Dash for cash: Monthly market
impact of institutional liquidity needs. Review of Financial Studies 33:76–111.

Eyster, E., M. Rabin, and D. Vayanos. 2019. Financial markets where traders neglect the
informational content of prices. Journal of Finance 71:371–99.

Fama, E. F. 1976. Foundations of finance. New York: Basic Books.

———. 1998. Market efficiency, long-term returns, and behavioral finance. Journal of
Financial Economics 49:283–306.

Fama, E. F., and K. R. French. 1992. The cross-section of expected stock returns. Journal
of Finance 47:427–465.

———. 1993. Common risk factors in the returns on stocks and bonds. Journal of
Financial Economics 33:3–56.

———. 1997. Industry cost of equity. Journal of Financial Economics 43:153–93.

———. 2010. Dissecting anomalies. Journal of Finance 63:1653–78.

———. 2015. A five-factor asset pricing model. Journal of Financial Economics 1:1–22.

———. 2016. Dissecting anomalies with a five-factor model. Review of Financial

Studies 29:69–103.

———. 2017. International tests of a five-factor asset pricing model. Journal of Financial
Economics 123:441–63.

Fama, E. F. and J. D. MacBeth. 1973. Risk, return, and equilibrium: Empirical tests.
Journal of Political Economy 81:607–36.

Frazzini, A., R. Israel, and T. J. Moskowitz. 2015. Trading costs of asset pricing anomalies.
Working Paper, AQR Capital Management.

French, K. R. 2008. Presidential address: The cost of active investing. Journal of

Finance 63:1537–73.

Gao, X., and J. Ritter. 2010. The marketing of seasoned equity o↵erings. Journal of
Financial Economics 97:33–52.

Gervais, S., R. Kaniel, and D. H. Mingelgrin. 2001. The high-volume return premium.
Journal of Finance 56:877–919.

Goyal, A., and N. Jegadeesh. 2018. Cross-sectional and time-series tests of return
predictability: What is the di↵erence? Review of Financial Studies 31:1785–824.

Harvey, C. R., and A. Siddique. 2000. Conditional skewness in asset pricing tests. Journal
of Finance 55:1263–96.

Hasbrouck, J. 2009. Trading costs and returns for U.S. equities: Estimating e↵ective costs
from daily data. Journal of Finance 64:1445–77.

Hendershott, T., and A. J. Menkveld. 2014. Price pressures. Journal of Finanial

Economics 114:405–23.

Henriksson, R. D., and R. C. Merton. 1981. On market timing and investment

performance. II. Statistical procedures for evaluating forecasting skills. Journal of
Business 54:513–33.

50

Electronic copy available at: https://ssrn.com/abstract=3150525

Short-term Momentum

Hirshleifer, D. A., S. S. Lim, and S. H. Teoh. 2011. Limited investor attention and stock
market misreactions to accounting information. Review of Asset Pricing Studies 1:35–73.

Hong, H., T. Lim, and J. C. Stein. 2000. Bad news travels slowly: Size, analyst coverage,
and the profitability of momentum strategies. Journal of Finance 55:265-295.

Hong, H., and J. C. Stein. 1999. A unified theory of underreaction, momentum trading,
and overreaction in asset markets. Journal of Finance 54:2143–84.

———. 2007. Disagreement and the stock market. Journal of Economic Perspec-
tives 21:109–28.

Hou, K., C. Xue, and L. Zhang. 2015. Digesting anomalies: An investment approach.
Review of Financial Studies 28:650–705.

———. 2020. Replicating anomalies. Review of Financial Studies Studies 33:2019–133.

Huang, D., J. Li, L. Wang, and G. Zhou. 2020. Time-series momentum: Is it there?
Journal of Financial Economics 135:774–94.

Jegadeesh, N. 1990. Evidence of predictable behavior of security returns. Journal of

Finance 45:881–98.

Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers:
Implications for stock market efficiency. Journal of Finance 48:65–91.

———. 2001. Profitability of momentum strategies: An evaluation of alternative

explanations. Journal of Finance 56:699–720.

Johnson, T. C. 2004. Forecast dispersion and the cross section of expected returns.
Journal of Finance 59:1957–78.

Koijen, R., T. J. Moskowitz, L. H. Pedersen, and E. B. Vrugt. 2018. Carry. Journal of

Financial Economics 127:197–225.

Lee, C. M. C., and B. Swaminathan. 2000. Price momentum and trading volume. Journal
of Finance 55:2017–69.

Lewellen, J. 2002. Momentum and autocorrelation in stock returns. Review of Financial

Studies 15:533–63.

Llorente, G., R. Michaely, G. Saar, and J. Wang. 2002. Dynamic volume-return relation
of individual stocks. Review of Financial Studies 15:1005–47.

Lo, A. W., and J. Wang. 2000. Trading Volume: Definitions, Data Analysis, and
Implications of Portfolio Theory. Review of Financial Studies 13:257–300.

Loh, R. K., and R. M. Stulz. 2018. Is sell-side research more valuable in bad times?
Journal of Finance 73:959–1013.

Luo, J., A. Subrahmanyam, and S. Titman. 2021. Momentum and reversals

when overconfident investors underestimate their competition. Review of Financial
Studies 34:351–93.

Medhat, M., and M. Schmeling. 2018. Dissecting announcement returns. Working Paper.

Mondria, J., X. Vives, and L. Yang. 2020. Costly interpretation of asset prices. Working
Paper, University of Navarra.

Moskowitz, T. J., and M. Grinblatt. 1999. Do industries explain momentum? Journal of

Finance 54:1249–90.

Nagel, S. 2012. Evaporating liquidity. Review of Financial Studies 25:2005–39.

51

Electronic copy available at: https://ssrn.com/abstract=3150525

The Review of Financial Studies / v 00 n 0 2021

Newey, W. K., and K. D. West. 1987. A simple, positive semi-definite, heteroskedasticity

and autocorrelation consistent covariance matrix. Econometrica 55:703–708.

Novy-Marx, R. 2013. The other side of value: The gross profitability premium. Journal
of Financial Economics 108:1–28.

Novy-Marx, R., and M. Velikov. 2015. A taxonomy of anomalies and their trading costs.
Review of Financial Studies 29:104–47.

Odean, T. 1998. Volume, volatility, price, and profit when all traders are above average.
Journal of Finance 53:1887–934.

Pástor, L., and R. F. Stambaugh. 2003. Liquidity risk and expected stock returns. Journal
of Political Economy 111:642–85.

Schneider, P., C. Wagner, and J. Zechner. 2020. Low-risk anomalies? Journal of

Finance 75:2673–18.

Verardo, M. 2009. Heterogeneous beliefs and momentum profits. Journal of Financial

and Quantitative Analysis 44:795–822.

Wang, J. 1994. A model of competitive stock trading volume. Journal of Political

Economy 102:127–68.

52

Electronic copy available at: https://ssrn.com/abstract=3150525

 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 1 — #1

Short-term Momentum
— Internet Appendix
Mamdouh Medhat
Cass Business School, City, University of London

Maik Schmeling
Goethe University Frankfurt and Centre for Economic Policy Research
(CEPR)

This document is the internet appendix to the paper “Short-term Momentum,”

as published in the Review of Financial Studies. (JEL G12, G14)

Appendix IA. Additional results

This internet appendix contains additional tables and figures omitted from the main
paper.

Send correspondence to Maik Schmeling, [email protected].

doi:10.1093/rfs/Sample

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 2 — #2

The Review of Financial Studies / v 0 n 0 0

Table IA.1
Portfolio characteristics for independent double sorts on the previous month’s
return and turnover

r1,0 deciles

TO 1,0 deciles Low 2 3 4 5 6 7 8 9 High

Portfolio r1,0
Low 0.15 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.20
2 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.19
3 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.19
4 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.18
5 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.18
6 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.18
7 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.18
8 0.14 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.18
9 0.15 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.19
High 0.17 0.08 0.05 0.02 0.00 0.02 0.04 0.06 0.10 0.24
Portfolio TO 1,0
Low 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02
2 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04
3 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05
4 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06 0.06
5 0.07 0.07 0.07 0.07 0.07 0.07 0.07 0.07 0.07 0.07
6 0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08 0.08
7 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10
8 0.13 0.13 0.13 0.13 0.13 0.13 0.13 0.13 0.13 0.13
9 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17
High 0.34 0.30 0.30 0.29 0.30 0.29 0.30 0.29 0.30 0.36
Average market capitalization ($ mio.)
Low 126 352 690 1,058 999 1,160 1,201 1,014 667 276
2 444 1,911 3,381 4,306 5,005 4,795 4,405 4,069 2,919 830
3 627 2,308 3,544 4,099 4,487 4,361 4,337 4,241 3,338 969
4 744 2,493 3,300 3,803 4,023 4,108 4,207 4,098 3,281 1,200
5 842 2,534 3,106 3,716 3,762 3,556 3,748 3,888 3,292 1,288
6 925 2,508 2,782 3,300 3,655 3,714 3,536 3,453 2,858 1,428
7 959 2,355 2,649 2,869 2,963 3,107 3,121 3,110 2,672 1,421
8 988 2,119 2,326 2,481 2,514 2,879 2,746 2,483 2,314 1,353
9 1,032 1,741 1,898 2,070 2,055 2,272 2,224 2,294 2,064 1,232
High 875 1,523 1,772 1,691 1,661 1,716 1,730 1,805 1,693 1,090
Average number of stocks
Low 179 107 96 87 79 75 68 68 67 96
2 68 42 38 35 34 33 32 32 32 46
3 49 32 29 29 28 27 27 27 27 37
4 42 28 26 25 25 25 24 25 25 34
5 39 26 24 23 23 23 23 24 25 34
6 38 26 23 22 21 21 22 23 25 36
7 40 26 22 20 20 20 20 23 26 40
8 44 26 21 19 18 18 19 22 27 48
9 52 27 20 18 16 17 18 21 29 60
High 81 26 18 15 14 14 16 20 29 101
This table shows the time-series averages of the portfolio characteristics for the
independent double sorts on the previous month’s return (r1,0 ) and turnover (TO 1,0 ),
as shown in Table A.2 in the main text. Portfolio r1,0 is the time-series average of each
portfolios’ monthly value weighted r1,0 , and similarly for portfolio TO 1,0 . Average market
capitalization is the time-series average of each portfolio’s monthly equal-weighted average
market capitalization in millions of dollars. Average number of stocks is the time-series
average of each portfolios’ monthly number of stocks. Data are at the monthly frequency
and cover July 1969 to December 2018, where the start date is to ensure non-empty
portfolios as a result of the independent sorts.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 3 — #3

Internet Appendix

Table IA.2
Short-term momentum and size: Robustness checks

A B
Performance of r1,0 strategies Performance of r1,0 strategies
within TO 1,0 ? M deciles within size deciles

TO 1,0 ? M decile E[re ] ↵FF6 ↵q Size decile E[re ] ↵FF6 ↵q

Low 1.88 1.98 1.98 Smallest 2.08 2.37 2.35

( 9.40) ( 9.11) ( 6.82) ( 7.85) ( 7.71) ( 6.31)
2 1.62 1.65 1.55 2 1.04 1.26 1.28
( 7.47) ( 7.34) ( 5.46) ( 4.57) ( 4.48) ( 3.63)
3 1.38 1.44 1.44 3 0.81 1.08 1.06
( 5.57) ( 5.81) ( 4.97) ( 3.48) ( 3.50) ( 2.66)
4 1.21 1.36 1.43 4 0.66 0.85 0.89
( 5.43) ( 4.92) ( 4.17) ( 2.66) ( 2.47) ( 1.98)
5 0.88 1.07 1.00 5 0.82 1.04 1.03
( 3.46) ( 4.06) ( 3.26) ( 3.60) ( 4.19) ( 3.44)
6 1.01 1.12 1.03 6 0.90 0.98 0.90
( 3.58) ( 3.44) ( 2.77) ( 3.95) ( 3.21) ( 2.51)
7 0.11 0.24 0.22 7 0.98 1.04 0.91
( 0.43) ( 0.80) ( 0.68) ( 4.41) ( 4.34) ( 2.87)
8 0.11 0.08 0.04 8 0.61 0.59 0.42
(0.44) (0.29) (0.13) ( 2.40) ( 2.09) ( 1.25)
9 0.22 0.07 0.24 9 0.40 0.40 0.28
(0.87) (0.27) (0.74) ( 1.71) ( 1.42) ( 0.95)
High 1.27 1.26 1.54 Largest 0.03 0.15 0.04
(4.36) (3.78) (3.99) ( 0.17) ( 0.73) ( 0.14)
This table shows the performance of winner-minus-loser strategies based on the previous
month’s return within deciles of the previous month’s ‘residual turnover’ relative to size
(TO 1,0 ? M ; panel A) and size (market capitalization from CRSP, M ; panel B). Portfolios
are from double sorts on the previous month’s return (r1,0 ) and either TO 1,0 ? M or size.
We use conditional sorts into deciles based on NYSE breakpoints, first on r1,0 and then
on either TO 1,0 ? M or size. Portfolios are value weighted and rebalanced at the end
of each month. TO 1,0 ? M is the residual from cross-sectional regression of TO 1,0 on
log(M ), estimated using WLS with market capitalization as weight. Test-statistics (in
parentheses) are adjusted for heteroscedasticity and autocorrelation. The sample excludes
financial firms. Data are at the monthly frequency and cover July 1963 to December 2018,
except when applying the q-factors, which are available from January 1967.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 4 — #4

The Review of Financial Studies / v 0 n 0 0

Short-term momentum

Average cumulative sum of excess returns (%)

10 Short-term reversal*

-5

-10

1 6 12 18 24 30 36

Months after portfolio formation

B
500
Short-term momentum
Short-term reversal*
Short-term reversal
Cumulative excess return (%)

-500

-1000

1995 2000 2005 2010 2015

Figure IA.1
International short-term momentum returns: Persistence and historical
performance Panel A shows the average cumulative sums of post-formation excess
returns to each of the international short-term momentum (STMOM) and international
short-term reversal* (STREV*) strategies along with 95% confidence bands. Panel B shows
a time-series plot of cumulative sums of excess returns to the international STMOM and
STREV* strategies as well as a conventional short-term reversal strategy. Data are at the
monthly frequency and cover January 1993 to December 2018.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 5 — #5

Internet Appendix

Table IA.3
Short-term momentum controlling for size: International evidence

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]
controlling for size controlling for size

Size group E[re ] ↵DMFF6 E[re ] ↵DMFF6

A: Size breakpoint is median

Small 0.52 0.57 3.15 2.84

(1.24) (1.22) ( 8.64) ( 8.31)
Large 0.99 1.03 1.67 1.49
(3.50) (3.55) ( 4.73) ( 4.12)
B: Size breakpoints are quintiles
Small 0.23 0.15 3.45 3.55
( 0.36) ( 0.18) ( 3.54) ( 4.68)
2 0.55 0.69 2.68 2.34
(1.16) (1.38) ( 7.78) ( 6.84)
3 0.71 0.85 2.40 2.11
(1.90) (2.16) ( 6.51) ( 4.41)
4 0.65 0.83 1.85 1.77
(1.45) (1.81) ( 4.00) ( 3.80)
Large 1.12 1.14 1.50 1.06
(4.01) (3.88) ( 3.12) ( 2.31)
C: Largest 100 stocks per country
Largest 100 0.62 0.46 0.44 0.68
per country (3.25) (2.68) (1.45) (1.99)
This table shows the performance of international short-term momentum and short-term
reversal* strategies constructed with a control for size (market capitalization). In panels A
and B, the strategies are from N ⇥3⇥3 conditional sorts on size and the previous month’s
return and turnover, in that order, where the breakpoints for returns and turnover are
the 20th and 80th percentiles. In panel A, N = 2 and the size breakpoint is the median;
in panel B, N = 5 and the size breakpoints quintiles. In panel C, the strategies are from
2⇥2 independent sorts on returns and turnover among the 100 largest stocks by monthly
market capitalization within each country, where the breakpoints for returns and turnover
are the 50th rank. We use independent double sorts to form country-specific portfolios
that are value weighted and rebalanced at the end of each month. We then weight each
country’s portfolio by the country’s total market capitalization for the previous month to
form each international portfolio. All returns and market values are in U.S. dollars and
excess returns are above the monthly U.S. T-bill rate. Abnormal returns are relative to
Fama and French’s (2017) developed markets five-factor model including the momentum
factor (DMFF6). Test-statistics (in parentheses) are adjusted for heteroscedasticity and
autocorrelation. Data are at the monthly frequency and cover January 1993 to December
2018.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 6 — #6

Table IA.4
Short-term momentum and end-of-month e↵ects: International evidence

A B
Performance of r1,0 EOM strategies Performance of rEOM strategies
within TO 1,0 EOM deciles within TO EOM deciles

TO 1,0 EOM decile E[re ] ↵DMFF6 TO EOM deciles E[re ] ↵DMFF6

Low 1.35 0.98 Low 3.29 3.46

( 2.82) ( 1.88) ( 7.97) ( 7.06)
2 0.91 0.16 2 2.11 2.16
( 1.88) (0.28) ( 3.22) ( 3.76)
3 0.78 0.61 3 2.06 2.13
( 1.50) ( 1.06) ( 4.83) ( 5.03)
4 0.30 0.89 4 2.47 2.77
(0.81) (1.93) ( 6.09) ( 6.38)
5 0.19 0.06 5 2.42 2.32
( 0.49) ( 0.12) ( 4.72) ( 4.94)
6 0.04 0.25 6 2.09 1.85
( 0.09) ( 0.63) ( 4.68) ( 3.72)
7 0.62 0.56 7 1.76 2.00
(1.68) (1.48) ( 4.79) ( 4.67)
8 0.48 0.66 8 1.13 1.47
(1.23) (1.69) ( 2.62) ( 3.49)
9 0.98 1.08 9 0.88 0.97
(3.83) (3.43) ( 2.46) ( 2.32)
High 1.66 1.71 High 0.42 0.25
(3.96) (3.78) ( 1.06) ( 0.71)
This table shows the average excess returns and the abnormal returns to international
winner-minus-loser strategies among stocks with di↵erent values for the previous month’s
share turnover. In panel A, the sorting variables exclude their end-of-month values
(r1,0 EOM and TO 1,0 EOM ) measured at the month’s last three trading days. In panel
B, the sorting sorting variables are just the end-of-month values (rEOM and TO EOM )
measured at the month’s last three trading days. The underlying portfolios are from
double sorts on the previous month’s return and turnover. We use independent double
sorts to form country-specific portfolios that are value weighted and rebalanced at the
end of each month. We then weight each country’s portfolio by the country’s total market
capitalization for the previous month to form each international portfolio. All returns and
market values are in U.S. dollars and excess returns are above the monthly U.S. T-bill
rate. Abnormal returns are relative to Fama and French’s (2017) developed markets five-
factor model including the momentum factor (DMFF6). Test-statistics (in parentheses)
are adjusted for heteroscedasticity and autocorrelation. Data are at the monthly frequency
and cover January 1993 to December 2018.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 7 — #7

Internet Appendix

Table IA.5
Short-term momentum and other stock-level momentum strategies: Additional
results

A: Characteristic tilts
Time-series average of long-short di↵erence in monthly
portfolio characteristics and equal-weighted market capitalization

Strategy B/M GP/A 1 COP/A 1 dA/A 1 MC

STMOM 6.38 4.62 1.44 5.42 177

(3.74) ( 4.73) ( 2.53) ( 1.65) (3.32)
MOM 17.40 3.19 2.96 4.52 583
( 4.40) (1.78) (2.21) ( 1.27) (2.38)
CAR3 0.89 1.77 0.70 0.77 203
( 0.91) (2.73) (1.69) ( 0.44) (3.85)
SUE 26.20 9.62 6.15 3.28 1,090
( 10.85) (6.81) (5.90) ( 0.83) (4.02)
ROE 59.69 33.16 21.87 27.67 4,237
( 3.75) (18.77) (19.83) ( 1.67) (0.96)

B: Overlap among winners

MOM CAR3 SUE ROE

STMOM 0.34 0.28 0.16 0.16

(28.94) (42.82) (24.30) (17.08)
MOM 0.24 0.30 0.27
(32.51) (22.52) (21.97)
CAR3 0.22 0.21
(45.82) (36.15)
SUE 0.30
(36.00)

C: Overlap among losers

MOM CAR3 SUE ROE

STMOM 0.20 0.32 0.15 0.26

(15.17) (33.40) (34.06) (16.89)
MOM 0.27 0.33 0.40
(33.01) (32.00) (23.60)
CAR3 0.25 0.34
(41.76) (36.50)
SUE 0.51
(39.85)
This table complements Table 8 in the main text with additional characteristic tilts (panel
A) for the short-term momentum strategy and the four other stock-level momentum
strategies. It also shows time-series averages of monthly overlap of stocks for the strategies’
winner and loser portfolios (panels B and C), where the overlap of two sets, X and Y , is
given by |X \Y |/min{|X|,|Y |}. In panel A, all accounting variables are annual. B/M is
book-to-market equity, where M is as of prior December from CRSP, and where the lagging
is to avoid taking unintentional positions in conventional momentum. GP/A 1 is gross
profits relative to 1-year lagged total asset. COP/A 1 is cash-based operating profits-to-
lagged total assets. dA/A 1 is the year-over-year relative change in total assets. M C is
monthly market capitalization from CRSP in $million. Test-statistics (in parentheses) are
adjusted for heteroscedasticity and autocorrelation. The sample excludes financial firms.
Data are at the monthly frequency and cover July 1963 to December 2018 for short-term
momentum and conventional momentum, but the PEAD and ROE strategies start from
January 1972 due to data availability in Compustat.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 8 — #8

The Review of Financial Studies / v 0 n 0 0

Table IA.6
Short-term momentum excluding earnings announcements

A B

Performance of r1,0 strategies Performance of r[1,0]\EAD strategies

within TO 1,0 deciles within TO [1,0]\EAD deciles
excluding earnings announcers

TO 1,0 decile E[re ] ↵FF6 ↵q TO [1,0]\EAD decile E[re ] ↵FF6 ↵q

Low 1.82 1.89 1.94 Low 1.80 1.85 1.92

( 5.99) ( 5.18) ( 5.64) ( 7.32) ( 6.71) ( 7.87)
2 1.77 1.90 1.90 2 1.44 1.62 1.63
( 5.28) ( 5.70) ( 5.10) ( 5.02) ( 4.58) ( 4.98)
3 1.40 1.50 1.41 3 1.58 1.79 1.76
( 4.08) ( 3.58) ( 3.62) ( 5.51) ( 5.91) ( 5.39)
4 1.05 1.27 1.18 4 1.06 1.16 1.01
( 3.10) ( 3.80) ( 2.87) ( 3.90) ( 4.46) ( 3.23)
5 0.95 0.94 0.83 5 0.84 0.97 0.83
( 2.65) ( 2.26) ( 2.00) ( 3.21) ( 3.22) ( 2.19)
6 0.16 0.13 0.02 6 0.57 0.46 0.34
( 0.40) ( 0.31) ( 0.04) ( 2.27) ( 1.73) ( 0.93)
7 0.40 0.56 0.37 7 0.97 1.04 1.06
( 1.22) ( 1.56) ( 0.76) ( 3.40) ( 3.10) ( 2.89)
8 0.58 0.72 0.72 8 0.49 0.69 0.72
( 1.61) ( 1.71) ( 1.69) ( 1.89) ( 2.16) ( 1.91)
9 0.04 0.13 0.08 9 0.05 0.05 0.12
(0.10) ( 0.30) (0.14) (0.18) ( 0.13) (0.32)
High 1.31 1.30 1.44 High 1.16 1.19 1.40
(2.75) (2.49) (3.05) (3.30) (3.00) (3.46)

This table shows the performance of long-short strategies that buy the previous month’s
winners and sell the previous month’s losers among stocks with di↵erent share turnover in
the previous month. In panel A, the sorts exclude earnings announcers, i.e., excluding
firms whose most recent earnings announcement date (Compustat’s RDQ) fell in the
previous month. In panel B, the sorting variables exclude their values for the three
days around earnings announcement dates (r[1,0]\EAD and TO [1,0]\EAD ). Portfolios are
from conditional sorts into deciles based on NYSE breakpoints, first on returns and then
on turnover, and are value weighted and rebalanced at the end of each month. Test-
statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation. The
sample excludes financial firms. Data are at the monthly frequency and cover January
1972 to December 2018, where the start date is determined by the availability of data on
quarterly earnings announcement dates in Compustat.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 9 — #9

Internet Appendix

Table IA.7
Short-term momentum and industry momentum

Intercepts, slopes, and test-statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t

Short-term momentum Industry momentum

[WML, high turnover]

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

variable
A: Controls are the FF6 factors
Intercept 1.37 0.79 0.92 0.98 0.84 0.63 0.52
(4.74) (2.73) (3.01) (4.59) (3.48) (2.94) (2.32)
IMOM 0.59 0.53
(7.74) (7.89)
STMOM 0.26 0.23
(6.93) (5.57)
MKT 0.31 0.07 0.01
( 3.82) ( 1.05) (0.13)
SMB 0.07 0.14 0.13
(0.51) ( 1.62) ( 1.67)
HML 0.02 0.08 0.07
( 0.09) (0.81) (0.75)
RMW 0.30 0.19 0.09
( 1.52) ( 1.33) ( 0.66)
CMA 0.15 0.04 0.00
(0.62) (0.26) (0.01)
MOM 0.15 0.33 0.25
(1.07) (4.17) (3.59)
Adj. R2 15.1% 18.3% 6.8% 15.1% 18.3%

B: Controls are the q-factors

Intercept 1.42 0.86 1.15 0.95 0.88 0.58 0.46
(4.74) (2.88) (3.67) (4.24) (3.42) (2.60) (1.66)
IMOM 0.59 0.57
(7.40) (7.17)
STMOM 0.26 0.26
(6.65) (5.91)
MKT 0.33 0.10 0.00
( 3.78) ( 1.30) (0.00)
ME 0.07 0.06 0.07
(0.39) ( 0.49) ( 0.91)
ROE 0.30 0.19 0.24
( 1.72) (1.22) (1.91)
I/A 0.13 0.08 0.03
(0.62) (0.35) (0.20)
Adj. R2 15.0% 18.2% 1.8% 15.0% 16.1%
This table shows time-series regression results for the short-term momentum (STMOM)
strategy and for Moskowitz and Grinblatt’s (1999) one-month industry momentum
(IMOM) strategy. We sort the 49 industries from Fama and French (1997), excluding
financials, on their previous month’s value-weighted average return and form a ‘winner’
and a ‘loser’ industry portfolio each containing 5 industries. These portfolios are value
weighted and rebalanced at the end of each month. The IMOM strategy is long the
winner- and short the loser industry portfolio. In panel A, the additional controls are
the FF6 factors. In panel B, the additional controls are the q-factors. Test-statistics (in
parentheses) are adjusted for heteroscedasticity and autocorrelation. The sample excludes
financial firms. Data are at the monthly frequency and cover July 1963 (panel A) or
January 1967 (panel B) to December 2018, where the start date in panel B is determined
by the availability of the q-factors.

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 10 — #10

The Review of Financial Studies / v 0 n 0 0

Table IA.8
Short-term momentum and factor momentum

Intercepts, slopes, and test-statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t

Short-term momentum Conventional momentum

[WML, high turnover]

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

variable

Intercept 0.95 1.01 0.98 1.31 0.18 0.37 0.24 0.27

(2.89) (3.23) (3.00) (4.31) (0.78) (1.76) (1.15) (2.35)
FMOMTS 1.31 0.24 0.18 3.10 1.05 0.27
(3.33) (0.44) ( 0.30) (12.65) (3.48) (1.21)
FMOMXS 1.81 1.55 1.64 4.08 3.00 0.14
(3.53) (2.17) (2.32) (12.93) (7.66) (0.52)
MKT 0.35 0.05
( 4.19) (1.60)
SMB 0.07 0.00
( 0.59) (0.03)
HML 0.05 0.04
( 0.25) (0.59)
RMW 0.34 0.20
( 1.44) ( 2.11)
CMA 0.18 0.18
(0.64) ( 1.75)
MOM 0.07 1.34
(0.32) (19.19)
Adj. R2 3.9% 5.0% 4.9% 8.3% 36.8% 42.2% 43.4% 81.6%
This table shows time-series regression results for the short-term momentum strategy from
Table 1 in the main text. The explanatory variables are Ehsani and Linnainmaa’s (2020)
time-series factor momentum strategy (FMOMTS ) and cross-sectional factor momentum
strategy (FMOMXS ). Additional controls are the factors from Fama and French’s (2015)
five-factor model in addition to the momentum factor (MOM). The table also shows the
corresponding results for a conventional 12-2 month momentum strategy from decile sorts
using NYSE breakpoints.
FMOMTS is long factors with a positive return and short factors with negative
return over the prior 12-1 months. FMOMXS is long factors with an above-median return
and short factors with a below-median return over the prior 12-1 months. The FMOM
strategies’ long and short legs are weighted by the number of factors in each leg relative
to the total number of factors and are rebalanced at the end of each month.
Underlying the FMOM strategies are 14 U.S. factors and 6 develop market factors.
The U.S. factors are size (SMB), value (HML), profitability (RMW), asset growth (CMA),
short-term reversal, long-term reversal, accruals, cash flow-to-price, earnings-to-price, net
share issues, residual variance, Pástor and Stambaugh’s (2003) traded liquidity factor,
Frazzini and Pedersen’s (2014) betting-against-beta (BAB), and Asness, Frazzini, and
Pedersen’s (2019) quality-minus-junk (QMJ). The developed market factors are SMB,
HML, RMW, CMA, BAB, and QMJ. The accruals, cash flow-to-price, earnings-to-price,
net share issues, and residual variance factors are long and short the top and bottom 30%
of stocks based on univariate sorts using data from Ken French’s website. The BAB and
QMJ factors are from AQR’s website.
Test-statistics (in parentheses) are adjusted for heteroscedasticity and autocorrela-
tion. Data are at the monthly frequency and cover July 1964 to December 2018, where
the start date is due to the availability of the FMOM strategies.

10

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i
i

i
i

Table IA.9
Short-term momentum and longer formation periods

Intercepts, slopes, and test-statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t
Internet Appendix

2 months 3 months 4 months 5 months 6 months

formation formation formation formation formation

Independent (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
variable

A: Dependent variable is alternative short-term momentum return

Intercept 0.99 0.02 1.00 0.15 1.06 0.38 1.24 0.64 1.19 0.62
(2.95) (0.08) (3.11) (0.57) (3.16) (1.36) (3.78) (1.93) (3.25) (1.80)
STMOM 0.71 0.62 0.50 0.45 0.42
(15.12) (13.70) (13.38) (9.50) (7.35)
Adj. R2 46.5% 34.0% 21.5% 17.3% 14.9%

B: Dependent variable is alternative short-term reversal* return

Intercept 1.07 0.13 0.88 0.02 0.49 0.33 0.48 0.27 0.49 0.24
( 4.70) ( 0.60) ( 3.78) ( 0.07) ( 2.05) (1.46) ( 1.87) (1.10) ( 1.84) (0.94)
STREV* 0.67 0.61 0.58 0.53 0.51
(12.74) (11.21) (10.02) (9.20) (6.86)
Adj. R2 36.1% 28.4% 23.9% 18.9% 16.1%

This table shows time-series regression results for alternative short-term momentum and short-term reversal* strategies constructed using longer

Electronic copy available at: https://ssrn.com/abstract=3150525

formation periods. The strategies are constructed similar to their one-month counterparts in Table 1 in the main text, except that the sorting variables
are cumulative return and average monthly share turnover for the previous 2,...,6 months. The explanatory variables are the benchmark STMOM
and STREV⇤ strategies from Table 1 in the main text based on a one-month formation period. Test-statistics (in parentheses) are adjusted for
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 11 — #11

heteroscedasticity and autocorrelation. The sample excludes financial firms. Data are at the monthly frequency and cover January 1963 to December
2018.

11

i
i

i
i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 12 — #12

The Review of Financial Studies / v 0 n 0 0

Table IA.10
Short-term momentum and volatility: Robustness checks

A B

Performance of r1,0 strategies Performance of r1,0 strategies

within TO 1,0 ? 1,0 deciles within 1,0 deciles

TO 1,0 ? 1,0 E[re ] ↵FF6 ↵q 1,0 decile E[re ] ↵FF6 ↵q

decile

Low 1.54 1.59 1.57 Low 2.36 2.45 2.56

( 5.55) ( 5.97) ( 5.13) ( 11.51) ( 11.13) ( 11.59)
2 0.77 0.97 0.87 2 1.72 1.80 1.81
( 3.07) ( 3.31) ( 2.21) ( 7.12) ( 6.61) ( 6.55)
3 0.91 0.96 1.02 3 1.73 1.90 2.04
( 3.37) ( 3.28) ( 2.82) ( 6.17) ( 6.96) ( 6.55)
4 1.38 1.36 1.45 4 1.66 1.75 1.77
( 5.13) ( 5.06) ( 4.80) ( 6.06) ( 5.60) ( 5.64)
5 0.63 0.75 0.64 5 1.72 1.99 1.92
( 2.42) ( 2.86) ( 2.18) ( 6.78) ( 6.75) ( 6.77)
6 0.72 0.89 0.75 6 1.86 1.97 2.06
( 3.00) ( 3.09) ( 2.24) ( 7.45) ( 7.00) ( 7.56)
7 0.49 0.59 0.44 7 1.16 1.17 1.20
( 1.95) ( 2.03) ( 1.17) ( 4.86) ( 4.54) ( 4.42)
8 0.15 0.32 0.27 8 1.67 1.72 1.79
( 0.54) ( 0.89) ( 0.69) ( 5.60) ( 5.01) ( 4.58)
9 0.06 0.11 0.06 9 1.16 1.29 1.29
( 0.26) ( 0.46) (0.16) ( 4.74) ( 4.91) ( 4.41)
High 1.06 0.96 1.21 High 0.65 0.86 0.87
(3.86) (3.07) (3.47) ( 2.25) ( 2.63) ( 2.41)
This table shows the performance of winner-minus-loser strategies based on the previous
month’s return within deciles of the previous month’s ‘residual turnover’ relative to
volatility (TO 1,0 ? 1,0 ; panel A) or volatility ( 1,0 ; panel B). Portfolios are from double
sorts on the previous month’s return (r1,0 ) and either TO 1,0 ? 1,0 or 1,0 . We use
conditional sorts into deciles based on NYSE breakpoints, first on r1,0 and then on
either TO 1,0 ? 1,0 or 1,0 . Portfolios are value weighted and rebalanced at the end of
each month. TO 1,0 ? 1,0 is the residual from cross-sectional regression of the previous
month’s share turnover on 1,0 , estimated using WLS with market capitalization as
weight, and 1,0 is the standard deviation of the previous month’s daily stock returns
using a minimum of 15 daily observations. Test-statistics (in parentheses) are adjusted for
heteroscedasticity and autocorrelation. The sample excludes financial firms. Data are at
the monthly frequency and cover July 1963 to December 2018, except when applying the
q-factors, which are available from January 1967.

12

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 13 — #13

Internet Appendix

Table IA.11
Short-term momentum in the pre-1963 era and across sample splits

Intercepts, slopes, and test-statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t

1926/7 - 2018/12 1926/7 - 1963/6 1963/7 - 1991/6 1991/7 - 2018/12

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

variable
A: Dependent variable is short-term momentum
Intercept 1.02 2.16 0.60 2.01 0.71 1.50 2.03 2.31
(4.23) (9.03) (1.61) (5.89) (1.92) (4.73) (4.80) (6.18)
MKT 0.07 0.13 0.27 0.06
( 1.12) (1.94) ( 4.15) (0.55)
SMB 0.24 0.02 0.30 0.11
(1.90) ( 0.13) (3.52) ( 0.67)
HML 0.07 0.15 0.09 0.18
(0.65) ( 0.82) (0.79) ( 0.94)
MOM 0.07 0.09 0.13 0.00
( 0.75) ( 0.87) (1.31) ( 0.02)
STREV 1.47 1.23 1.34 1.62
( 13.73) ( 10.60) ( 9.39) ( 9.23)
LTREV 0.09 0.01 0.15 0.73
( 0.43) ( 0.09) (1.14) (3.25)
Adj. R2 33.6% 26.3% 30.4% 37.2%

B: Dependent variable is short-term reversal*

Intercept 2.82 1.80 3.96 1.99 1.67 0.89 1.15 1.06
( 5.40) ( 6.83) ( 6.28) ( 3.57) ( 7.12) ( 3.30) ( 3.72) ( 3.47)
MKT 0.13 0.07 0.08 0.10
(1.25) ( 0.52) ( 1.33) (1.05)
SMB 0.25 0.52 0.15 0.19
( 1.15) ( 1.99) ( 1.30) ( 2.15)
HML 0.00 0.06 0.11 0.03
( 0.03) (0.23) (0.83) (0.34)
MOM 0.11 0.05 0.05 0.15
(1.02) ( 0.34) (0.61) (2.28)
STREV 1.33 1.55 1.00 0.96
( 7.40) ( 5.64) ( 8.57) ( 8.24)
LTREV 0.01 0.02 0.33 0.06
(0.08) ( 0.05) ( 2.84) (0.46)
Adj. R2 26.3% 31.0% 36.6% 33.2%
This table shows time-series regression results for the short-term momentum strategy
starting from July 1926. The table also shows the corresponding results for the short-
term reversal* strategy. The strategies are constructed as in Table 1 of the main text but
using the extended CRSP sample. The explanatory variables are the three Fama-French
factors (MKT, SMB, and HML) in addition to the momentum factor (MOM) and the two
reversal factors (STREV and LTREV). Test-statistics (in parentheses) are adjusted for
heteroscedasticity and autocorrelation. Data are at the monthly frequency and the sample
periods are indicated above the specification numbers. The MOM factor is available from
January 1927 while the LTREV factor is available from January 1931.

13

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 14 — #14

The Review of Financial Studies / v 0 n 0 0

Table IA.12
Short-term momentum and volatility risk

Intercepts, slopes, and test-statistics (in parentheses)

from time-series regressions of the form yt = ↵+ 0 Xt +✏t

Short-term momentum Short-term reversal*

[WML, high turnover] [WML, low turnover]
controlling for size controlling for size

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

variable

A: Large-caps (above NYSE median)

Intercept 0.91 1.48 0.89 1.05 0.76 0.66 0.75 0.66
(2.86) (1.82) (3.05) (3.56) ( 3.45) (0.80) ( 3.42) ( 2.63)
VIX 1 0.03 0.07
( 0.68) ( 1.61)
VIX 0.30 0.13 0.27 0.17
(3.61) (0.95) (3.25) (1.56)
MKT 0.24 0.15
( 1.48) ( 1.29)
Adj. R2 0.2% 3.6% 4.5% 1.1% 5.9% 6.5%

B: Micro- and small-caps (below NYSE median)

Intercept 0.65 2.21 0.64 0.83 1.03 0.72 1.03 0.86
(1.97) (2.21) (1.99) (2.68) ( 4.92) (0.94) ( 5.34) ( 2.99)
VIX 1 0.08 0.09
( 1.46) ( 2.13)
VIX 0.26 0.05 0.25 0.06
(2.97) (0.42) (5.50) (1.06)
MKT 0.30 0.26
( 2.89) ( 3.75)
Adj. R2 0.6% 2.7% 4.4% 2.3% 5.6% 8.7%
This table shows time-series regression results for short-term momentum (STMOM) and
short-term reversal* (STREV*) strategies constructed among large-caps and all-but-large-
caps. The strategies are from 2⇥2⇥2 conditional sorts on size, the previous month’s
return, and the previous month’s share turnover, in that order. The breakpoint for size is
the median for NYSE stocks, while the breakpoints for returns and turnover are the 20th
and 80th percentiles for NYSE stocks. Portfolios are value weighted and rebalanced at the
end of each month. See Table ?? for more on these strategies. The explanatory variables
are the one-month lagged CBOE Volatility Index (VIX 1 ), the contemporaneous monthly
change in the implied volatility index ( VIX = VIX VIX 1 ), and the contemporaneous
market return (MKT). The slopes on VIX 1 and VIX are multiplied by 100. Test-
statistics (in parentheses) are adjusted for heteroscedasticity and autocorrelation. Data
are at the monthly frequency and cover January 1990 to December 2018, where the start
date is determined by the availability of the VIX.

14

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 15 — #15

Internet Appendix

Table IA.13
Short-term momentum and illiquidity: Robustness checks

A B
Performance of r1,0 strategies Performance of r1,0 strategies
within TO 1,0 ? Illiq1,0 deciles within Illiq1,0 deciles

TO 1,0 ? Illiq1,0 E[re ] ↵FF6 ↵q Illiq1,0 decile E[re ] ↵FF6 ↵q

decile

Low 1.20 1.29 1.25 Low 0.26 0.15 0.27

( 5.28) ( 4.71) ( 4.73) (1.33) (0.73) (1.00)
2 1.23 1.38 1.44 2 0.55 0.51 0.42
( 4.68) ( 5.00) ( 4.27) ( 2.12) ( 1.72) ( 1.22)
3 1.14 1.38 1.36 3 0.62 0.70 0.66
( 4.51) ( 5.06) ( 4.05) ( 2.56) ( 2.39) ( 2.09)
4 0.74 0.91 0.84 4 0.82 0.90 0.84
( 3.32) ( 3.49) ( 2.61) ( 3.63) ( 3.56) ( 2.70)
5 0.68 0.80 0.58 5 0.87 1.02 0.95
( 2.62) ( 2.53) ( 1.61) ( 3.55) ( 3.58) ( 3.07)
6 0.42 0.43 0.56 6 0.78 0.89 0.78
( 1.40) ( 1.37) ( 1.55) ( 3.01) ( 2.74) ( 1.95)
7 0.75 0.77 0.81 7 0.94 1.18 1.14
( 3.06) ( 2.38) ( 2.10) ( 3.82) ( 3.94) ( 2.81)
8 0.12 0.03 0.05 8 1.12 1.23 1.17
(0.41) ( 0.10) (0.14) ( 4.91) ( 4.33) ( 3.29)
9 0.17 0.15 0.31 9 0.92 1.02 0.98
(0.68) (0.57) (0.90) ( 4.14) ( 4.06) ( 3.21)
High 1.37 1.37 1.65 High 2.05 2.33 2.36
(4.49) (4.14) (4.40) ( 7.65) ( 7.59) ( 5.53)
This table shows the performance of winner-minus-loser strategies based on the previous
month’s return within deciles of the previous month’s ‘residual turnover’ relative to
illiquidity (TO 1,0 ? Illiq1,0 ; panel A) and illiquidity (Illiq1,0 ; panel B). Portfolios are from
double sorts on the previous month’s return (r1,0 ) and either TO 1,0 ? Illiq1,0 or Illiq1,0 .
We use conditional sorts into deciles based on NYSE breakpoints, first on r1,0 and then on
either TO 1,0 ? Illiq1,0 or Illiq1,0 . Portfolios are value weighted and rebalanced at the end of
each month. TO 1,0 ? Illiq1,0 is the residual from cross-sectional regression of the previous
month’s share turnover on Illiq1,0 , estimated using WLS with market capitalization as
weight, and Illiq1,0 is the average absolute return relative to the dollar trading volume
using a minimum of 15 daily observations. Test-statistics (in parentheses) are adjusted for
heteroscedasticity and autocorrelation. The sample excludes financial firms. Data are at
the monthly frequency and cover July 1963 to December 2018, except when applying the
q-factors, which are available from January 1967.

15

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i
 i i

i i
“STMOM-InternetAppendix” — 2021/4/26 — 21:54 — page 16 — #16

The Review of Financial Studies / v 0 n 0 0

References

Asness, C. S., A. Frazzini, and L. H. Pedersen. 2019. Quality minus junk. Review of
Accounting Studies 24:34–112.

Ehsani, S., and J. Linnainmaa. 2020. Factor momentum and the momentum factor.
Working Paper, Northern Illinois University.

Fama, E. F. and K. R. French (1997). Industry cost of equity. Journal of Financial

Economics 43 (2), 153–193.

Fama, E. F. and K. R. French (2015). A five-factor asset pricing model. Journal of

Financial Economics 1, 1–22.

Fama, E. F. and K. R. French (2017). International tests of a five-factor asset pricing

model. Journal of Financial Economics 123 (1), 441–463.

Frazzini, A. and L. H. Pedersen (2014). Betting against beta. Journal of Financial

Economics 111, 1–25.

Moskowitz, T. J. and M. Grinblatt (1999). Do industries explain momentum? Journal

of Finance 54 (4), 1249–1290.

Pástor, L. and R. F. Stambaugh (2003). Liquidity risk and expected stock returns.
Journal of Political Economy 111 (3), 642–685.

16

i i

i Electronic copy available at: https://ssrn.com/abstract=3150525

i