Enhanced Momentum Strategies

Enhanced Momentum Strategies
Latest version here
Matthias X. Hanauer Steffen Windmüller

TUM School of Management TUM School of Management
Technical University of Munich Technical University of Munich
First version: January, 2019; This draft: September, 2021
Abstract
This paper compares the performance of three enhanced momentum strategies pro-
posed in the literature — constant volatility-scaled momentum, constant semi-volatility-
scaled momentum, and dynamic-scaled momentum. Using data for individual stocks
from the U.S. and across 48 international countries, we find that all three approaches
decrease momentum crashes and lead to higher risk-adjusted returns. However, in
multiple factor comparison tests, no enhanced momentum strategy emerges as consis-
tently superior. Finally, cross-country analyses relate momentum and the two constant
volatility-scaled momentum returns to market dynamics, whereas dynamic-scaled mo-
mentum is significantly less affected, suggesting a reduced sensitivity to time-varying
overconfidence.
Keywords: Anomalies, Asset pricing, Momentum, International stock markets
JEL Classifications: G12, G14, G15
We thank Pedro Barroso, David Blitz, Winfried Hallerbach, Christoph Kaserer, Martin Martens, Laurens
Swinkels, Milan Vidojevic, Pim van Vliet, Florian Weigert, two anonymous referees and the conference and
seminar participants at the 36th International Conference of the French Finance Association (AFFI), the
Mutual Funds, Hedge Funds and Factor Investing Conference, TUM School of Management, Robeco, and
the University of Münster for helpful comments. Furthermore, we are grateful for the provision of Matlab
code to conduct the model comparison tests using the Sharpe measure by Cesare Robotti. Disclosures:
Hanauer is also employed by Robeco, an asset management firm that, among other strategies, also offers
factor investing strategies. The views expressed in this paper are those of the authors and not necessarily
shared by Robeco. Earlier versions of this paper were circulated under the title “Managing the Risk of
Momentum”. Any remaining errors are our own.
Email: [email protected], [email protected]
Electronic copy available at: https://ssrn.com/abstract=3437919

1 Introduction
The evidence for momentum is pervasive: Jegadeesh and Titman (1993) discover that past
winner (loser) stocks tend to have relatively high (low) future returns. Momentum poses an
explanatory problem on the Capital Asset Pricing model (CAPM) of Sharpe (1964), Lintner
(1965), and Mossin (1966), as well as on the Fama and French (1993, 2015) three- and
five-factor models. Within the U.S., a long-short momentum factor generates an average
raw return of 0.60% (Fama-French three-factor model (FF3FM) alpha of 0.87%) per month
between January 1930 and December 2017. Positive momentum returns have also been
identified for non-U.S. equity markets and other asset classes.1
Besides the relatively high profitability, momentum has occasionally experienced large
drawdowns (crashes), i.e., persistent strings of negative returns. In 1932, the momentum
factor for the U.S. equity market exhibited a drawdown of -68.98%. Also in 2009, the mo-
mentum factor for both the U.S. and international (non-U.S.) equity markets experienced
substantial losses. Grundy and Martin (2001) explain the risks of momentum by time-varying
factor exposures. For instance, after bear markets, the market betas of loser stocks tend to
be higher than those of winner stocks. When the market rebounds after a bear market,
the overall negative market sensitivity of the winner-minus-loser strategy generates negative
strategy returns. Similarly, Asem and Tian (2010) investigate the effect of market dynam-
ics on momentum returns in stock markets and document that there are higher momentum
returns when markets continue in the same state than when they transition to a different
1
Rouwenhorst (1998, 1999) finds that momentum strategies earn high abnormal returns in equity mar-
kets internationally, both in developed and in emerging markets. Moskowitz and Grinblatt (1999) document
momentum for industry portfolios, Asness, Liew, and Stevens (1997) and Chan, Hameed, and Tong (2000)
for country equity indices, Okunev and White (2003) and Menkhoff, Sarno, Schmeling, and Schrimpf (2012)
for currency markets, and Erb and Harvey (2006) for commodity futures. Asness, Moskowitz, and Peder-
sen (2013) confirm these findings and uncover a common factor structure among momentum returns across
asset classes. Chui, Titman, and Wei (2010) show that momentum is persistent worldwide except for Asia,
and propose cross-country differences in individualism as an explanation, while Docherty and Hurst (2018)
document that momentum is stronger in more myopic countries. Griffin, Ji, and Martin (2003) find that mo-
mentum returns cannot be attributed to macroeconomic risk factors, whereas Fama and French (2012) show
that local momentum factors are superior to a global momentum factor in pricing regional size-momentum
portfolios. Furthermore, momentum instruments for systemic exposure to latent risk factors in U.S. and
non-U.S. markets as shown in Windmüller (2021).

state, i.e., market up (down) movements following bull (bear) markets, are associated with
higher momentum returns.2
The recent literature has focused on volatility-scaling of factors. This idea is based on the
empirical observation that factor-return volatility is positively autocorrelated in the short
term and that returns are relatively low when volatility is high, named the leverage ef-
fect.3 Volatility-scaling strategies have been tested for single assets, factors, and asset classes,
mostly in U.S. markets.4 Barroso and Santa-Clara (2015) study momentum strategies that
are deflated by their past realized volatility and scaled to a constant target volatility level.
Realized volatility is calculated from past daily momentum returns and serves as a proxy of
future volatility. They find that the Sharpe ratios of constant volatility-scaled momentum
more than double, while portfolio turnover only marginally increases. Relatedly, Wang and
Yan (2021) document that factors scaled by semi (downside) volatility exhibit significantly
better performance than unscaled factors and factors scaled by total volatility. However, only
for momentum, they do not find a significant difference between semi and total volatility
scaling. Daniel and Moskowitz (2016) extend the constant-volatility scaling approach by ad-
ditionally taking the forecasted momentum return into account, creating the dynamic-scaled
momentum. Their dynamic scaling weights are different from those of constant-volatility
scaling because they can take on negative values. Daniel and Moskowitz (2016) show that
the dynamic strategy exceeds the Sharpe ratio achieved with the constant-volatility scaling
2
Hanauer (2014) confirms this pattern for Japan, Korea, Taiwan, and Turkey, while Li and Galvani (2018)
provide further evidence for the state-dependent performance of momentum in the corporate bond market.
3
See, among others, Engle (1982); Bollerslev (1987) for volatility autocorrelation and Bekaert and Wu
(2000) for asymmetry in the risk to return relation. Wang and Yan (2021) find that for momentum in the
U.S., the performance improvement stems mainly from the leverage effect.
4
Moskowitz, Ooib, and Pedersen (2012) test volatility-scaling at the security, not at the portfolio level.
The goal is to prevent portfolios from being dominated by only a few assets with high volatility. Moreira
and Muir (2017) highlight the advantage for mean-variance investors when scaling different equity long-
short strategies by realized variance. Grobys, Ruotsalainen, and Äijö (2018) compare risk management
strategies for industry momentum in the U.S. They find that industry momentum exhibits no time-varying
beta (as standard momentum does) and that volatility-scaling improves its performance. du Plessis and
Hallerbach (2016) find that the volatility-scaling of U.S. industries both lowers the industry momentum
strategy’s volatility and heightens its returns. Recently, Harvey, Hoyle, Korgaonkar, Rattray, Sargaison, and
Van Hemert (2018) compare volatility-targeting strategies across different asset classes.

strategy.5
This paper contributes to the literature in at least three aspects. First, we compare three
enhanced momentum strategies proposed in the literature — constant volatility-scaled mo-
mentum (cMOM), constant semi-volatility-scaled momentum (sMOM), and dynamic-scaled
momentum (dMOM) — using a uniform data set and methodology. Thereby, we align our
research with the ongoing debate about whether volatility-managed investment strategies
yield higher Sharpe ratios than non-managed factors do.6 We use both a long sample of U.S.
and a broad sample of non-U.S. stocks. We then evaluate the enhanced momentum strate-
gies in different dimensions: return and risk characteristics, including higher moments and
maximum drawdowns, mean variance-spanning tests, ex-post maximum Sharpe ratio tests,
cross-country variation, and portfolio turnover.
Second, we add to the replication literature by a large replication of standard momentum
(MOM) as well as strategies to improve its performance. According to Hou, Xue, and Zhang
(2020), replication provides a contribution when extending existing studies out-of-sample.
In this regard, the majority of asset pricing studies are based solely on U.S. market data.
Karolyi (2016) argues that this implicitly creates a “home bias” for the U.S. market. Harvey
(2017) gives further rise to the replication argument by stating that many published results
would not hold in the future because of unreported tests, testing of multiple hypotheses, and
data snooping. Harvey, Liu, and Zhu (2016) link data-snooping concerns with the pressure
and incentive to publish, which generates a publication bias, and propose higher t-statistic
hurdles. Since we are conducting a comparison study that tests already published enhanced
momentum strategies for both the U.S. and international markets, affirmative results for both
samples alleviate concerns that previously documented patterns are due to chance or data
5
However, Daniel and Moskowitz (2016) exploit ex-post information for calibrating their baseline model,
disregarding the optimization set of a real-world investor. This forward-looking bias is addressed only per-
functory, giving rise to the question of a practical implementation.
6
See Cederburg, O’Doherty, Wang, and Yan (2020) for a literature review of the volatility management of
trading strategies. Cederburg et al. (2020) examine 103 factors and find that volatility-management generates
statistically significant Sharpe ratio improvements for only eight out of 103. Importantly, the authors show
that the eight trading strategies all relate to momentum strategies.

snooping. We use a uniform global dataset, an identical factor construction approach, and
proceed with the same statistical tests, reporting the results homogeneously across countries
and strategies (factors). In this way, we overcome potential concerns of data mining, multiple
hypothesis testing, and Type I error concerns.
Third, we contribute to the literature investigating the drivers of cross-country differ-
ences in anomaly returns. The previous literature has shown that country-level factors have
explanatory power for the asset growth (Titman, John Wei, and Xie, 2013; Watanabe, Xu,
Yao, and Yu, 2013), the Stambaugh, Yu, and Yuan (2015) mispricing (Jacobs, 2016), and the
momentum anomaly (Chui et al., 2010). We add to this literature by extensively investigat-
ing the source of the four momentum strategies since 1990. We employ the comprehensive
country-level predictor set from Jacobs (2016) as well as behavioral predictors from Chui et al.
(2010) and Asem and Tian (2010), including proxies for, among others, limits-to-arbitrage,
market frictions, market efficiency, and investor overconfidence. To the best of our knowl-
edge, our study is the first that tests the explanatory power of a market continuation dummy
across countries. This dummy proxies time-varying overconfidence in the spirit of Daniel,
Hirshleifer, and Subrahmanyam (1998). The country-month panel allows us to investigate
momentum and enhanced momentum strategy predictors across time and countries.
Our main findings can be summarized as follows. First, we show that all enhanced
strategies substantially increase Sharpe ratios by using a long sample of U.S. stocks from
1930 to 2017 and a broad sample of stocks from 48 international markets from 1990 to 2017.
Furthermore, skewness, kurtosis, and maximum drawdowns decrease in magnitude compared
to MOM so that their return distributions become more normally distributed. Comparing the
individual enhanced strategies within samples, we find similar improvements for Sharpe ratios
and t-statistics (both roughly double compared to MOM) across all three approaches, while
returns during bear-up months are highest for dMOM. For the U.S. sample, we document
that dMOM exhibits the highest standalone Sharpe ratio, while this is the case for cMOM
for the non-U.S. sample.

Second, we find that all enhanced strategies generate positive and significant alphas in
mean-variance spanning tests with common asset pricing factors and momentum as bench-
mark assets. On the other hand, MOM is also not redundant in spanning tests against
dMOM for both samples. Similarly, ex-post maximum Sharpe ratio tests from Barillas, Kan,
Robotti, and Shanken (2019) confirm our results that all enhanced momentum strategies
add information in comparison to momentum. Specifically, all enhanced momentum strate-
gies significantly increase the maximum Sharpe ratio as compared to a factor opportunity set
that already includes momentum. For both the U.S. and the non-U.S. sample, the augmented
model with cMOM significantly outperforms the one augmented with sMOM, while only for
the non-U.S. sample, the augmented model with cMOM outperforms the one augmented with
dMOM. In contrast, for the U.S. sample, no statistically significant differences exist between
these two models.
Third, cross-country panel regressions document that the market continuation dummy
has by far the strongest impact to explain differences in momentum returns across countries.
This finding is consistent with the evidence provided in Asem and Tian (2010) and Hanauer
(2014) that momentum returns across time are higher in market continuations for the U.S.
and Japan, respectively. The result is also consistent with behavioral explanations of momen-
tum, i.e., time-varying overconfidence drives momentum returns (Daniel et al., 1998). While
cMOM and sMOM are similarly affected by the market continuation dummy as momentum,
we document a weaker impact on dMOM, suggesting a reduced sensitivity to time-varying
investor overconfidence. We also document that MOM, cMOM, and sMOM are more related
to market continuations in bear than to market continuations in bull markets. This result
is also consistent with the option-like behavior of momentum in bear markets as described
in Daniel and Moskowitz (2016), i.e., expected momentum returns are low in bear markets
when market volatility is high and contemporaneous with market reversals. As the weight of
dMOM in MOM linearly increases with the expected momentum return for the next month,
the sensitivity to market continuations is more reduced in bear markets. Furthermore, we

split up our sample into two sub-periods from 01/1990 to 12/2003 and from 01/2004 to
12/2017. We confirm the positive, significant effect of individualism on momentum returns
for the first sub-period, which aligns with the sample period in Chui et al. (2010). However,
for the second sub-period, a country’s individualism score is not related to momentum re-
turns, suggesting that the results in Chui et al. (2010) do not hold out-of-sample. In contrast,
we identify a robust effect of the market continuation dummy on all momentum strategies
in both sub-samples. In addition, the panel regressions indicate that all three momentum
strategies are positively affected by firm-specific return variation (i.e., negative coefficient on
R2 ), although to a lesser degree. Based on Hou, Peng, and Xiong (2013) and Kelly (2014),
this finding appears to be consistent with noise trader-based interpretations of this variable.
Finally, enhanced momentum strategies should be at least as implementable as standard
momentum as their break-even costs (i.e., transaction costs that theoretically would ren-
der the strategies insignificant) are higher than the ones for standard momentum, and all
momentum strategies generate their performance mainly during non-January months.
The rest of the paper is structured as follows: Section 2 describes the data, factor con-
struction, enhanced momentum strategies, and research methodologies. Section 3 presents
our empirical results for the implemented strategies, mean-variance spanning tests, maxi-
mum ex-post Sharpe ratio tests, cross-country panel regressions, break-even transaction cost
analyses, and the January effect. Section 4 concludes.
2 Data and Methodology
2.1 Data
The data analyzed in this paper is collected from various sources. We use a sample consisting
of 66,905 stocks for 49 equity markets from January 1926 to December 2017. We differentiate
into U.S. and non-U.S. (international) equity data with respect to differences in the sample
period availability. Both monthly and daily returns are measured in USD. The U.S. data

comes from the Center for Research on Security Prices (CRSP) and covers January 1926 to
December 2017. International data is collected from Thomson Reuters for the sample period
January 1987 to December 2017 and the country selection follows the Morgan Stanley Capital
International (MSCI) Developed and Emerging Markets Indices. We include all countries
classified either as a developed or an emerging market at some point during the sample period.
More precisely, the countries are only part of the actual sample in those years in which they are
part of the MSCI Developed and Emerging Markets Indices. Next to the U.S., the following
countries are included: Argentina, Australia, Austria, Belgium, Brazil, Canada, Chile, China,
Colombia, Czech Republic, Denmark, Egypt, Finland, France, Germany, Greece, Hong Kong,
Hungary, India, Indonesia, Ireland, Israel, Italy, Japan, Malaysia, Mexico, Morocco, the
Netherlands, New Zealand, Norway, Pakistan, Peru, Philippines, Poland, Portugal, Qatar,
Russia, Singapore, South Africa, (Republic of) South Korea, Spain, Sweden, Switzerland,
Taiwan, Thailand, Turkey, the United Kingdom (U.K.), and the United Arab Emirates.
Our U.S. sample includes all common equity stocks traded on NYSE, NYSE MKT (for-
merly: AMEX), or NASDAQ within the CRSP universe. We exclude all stocks with a CRSP
share code (SHRCD) different than 10 or 11. If available, we use Fama-French factors from
Kenneth French’s website7 and the monthly updated value factor (HMLd ) from AQR8 for
the long sample, which leaves us with constructing momentum and the enhanced strategies.
The non-U.S. sample comprises market data from Datastream and accounting data from
Worldscope. We process data through static and dynamic screens to ensure data quality.
As a first step, we identify stocks by Thomson Reuters Datastream’s constituent lists. We
use Worldscope lists, research lists, and, to eliminate survivorship bias, dead lists. Following
Ince and Porter (2006), Griffin, Kelly, and Nardari (2010), and Schmidt, Von Arx, Schrimpf,
Wagner, and Ziegler (2017), we apply generic as well as country-specific static screens to
eliminate non-common equity stocks. In addition, we apply dynamic screens for stock return
and price data as recommended in the literature. Appendix A.1 describes the static and
7
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.
8
https://www.aqr.com/Insights/Datasets.

dynamic screens in detail. The emerging market data is restricted to start only in July 1994.
We require stocks to have a valid market capitalization for June y and December y − 1 as
well a positive book equity value for December y − 1 to be regarded for the market, size, and
value portfolios within the non-U.S. sample. For both the U.S. and the non-U.S. sample, we
calculate momentum, constant volatility-scaled momentum, constant semi-volatility-scaled
momentum, and dynamic-scaled momentum and, therefore, additionally require valid stock
returns from months t − 12 to t − 2.
Finally, a country is only part of the final sample in those months for which at least
30 stock-month observations are available after filters.9 For the construction of dMOM,
24 months of return data for the value-weighted market index are used to calculate the bear
market indicator. We require an additional 24 months for both momentum returns and the
bear-market indicator to calibrate Equation (7). Eventually, our momentum strategy sample
starts in January 1930 (January 1990) for the U.S. (non-U.S.). We end up with a total of
8,550,324 firm-month observations. Table 1 shows the descriptive statistics for the stocks in
the final sample.
[Table 1 about here.]
2.2 Factor construction
Our approach for constructing the factor portfolios follows Fama and French (1993, 2012). We
calculate the portfolio breakpoints for each country separately to ensure that country effects
do not drive our results. The market factor, RMRF, consists of value-weighted returns of
all available (and valid) securities less the risk-free rate. Since we measure returns in USD,
we calculate excess returns based on the one-month U.S. Treasury bill rate. The size and
value factors are constructed by independent double sorts and six value-weighted portfolios.
9
Following Jacobs (2016), we thereby ensure that the six size-momentum portfolios contain at least five
stocks on average. As a consequence, some countries (such as India, Hong Kong, and Spain) are excluded
from the sample for certain months. Jordan, Sri Lanka, Slovakia, and Venezuela are excluded from the whole
sample.

For every end-of-June of year y, we assign stocks among two size-sorted and three book-
to-market sorted portfolios based on their market capitalization and book-to-market ratio,
respectively. Market capitalization is from end-of-June in year y and the book-to-market
ratio is calculated with book value and market capitalization from the fiscal year end in year
y − 1. For the U.S. sample, the size breakpoints are based on the NYSE median market
capitalization and stocks are classified as big or small, indicated by B and S. For the non-
U.S. sample, we follow Fama and French (2012, 2017) and define size breakpoints so that the
largest (smallest) stocks cover 90% (10%) of a country’s market capitalization. Subsequently,
stocks are independently sorted into three portfolios (Growth, Neutral, and Value, indicated
by G, N , and V ) based on the country-specific 70% and 30% percentile breakpoints of book-
to-market ratios. The breakpoints on book-to-market are calculated from NYSE (big) stocks
only for the U.S. (non-U.S.) sample. For the resulting six portfolios (BV , BN , BG, SV ,
SN , and SG), we calculate monthly value-weighted returns and construct the size (SMB)
and value (HML) factor as zero-investment long-short portfolios from July y to end-of-June
y + 1,
SM B = (SV + SN + SG) /3 − (BV + BN + BG) /3,
(1)
HM L = (BV + SV ) /2 − (BG + SG) /2.
Analogous to the value factor, we categorize stocks based on the 70% and 30% percentiles
of the past 12-2 month cumulative returns as Winner, Neutral and Loser (W , N , and L)
stocks to form the momentum factor, where only NYSE (big) stocks are considered for
the breakpoint calculation. We then calculate the monthly value-weighted returns for the
2x3 portfolios (BW , BN , BL, SW , SN , SL). Momentum is constructed by the long-
short portfolio as M OM = (BW + SW ) /2 − (BL + SL) /2 and - in contrast to SMB and
HML - rebalanced every month.10 Asness and Frazzini (2013) introduce the so-called HML-
devil factor (denoted as HMLd ). HMLd is comparable to HML but updates the market
10
As stated above, we use the Fama-French factors from Kenneth French’s website for the U.S. sample.
The momentum factor and enhanced momentum strategies, however, are constructed by ourselves in order
to ensure an identical stock universe for construction.
10

capitalization for the sorting criteria every month. Thus, stocks are sorted into portfolios
based on their monthly updated book-to-market ratio.
2.3 Enhanced momentum strategies
Volatility scaling aims to manage the realized volatility of an investment strategy. For cross-
sectional (our benchmark strategy) momentum, realized volatility has been shown to have
a positive (negative) correlation with future volatility (returns) and to be relatively high
as compared to other factors.11 In this study, we identify two potential channels for Sharpe
ratio improvements by volatility scaling: volatility smoothing refers to the overall lowered ex-
post volatility, and volatility timing to the heightened strategy returns due to the negative
correlation between volatility and returns. We compute the realized total and downside
volatility of the momentum strategy to control for volatility.12
Importantly, volatility scaling does not directly consider the positive autocorrelation of
monthly momentum strategy returns. Thus, one can additionally control for the realized
return of the momentum strategy to capture the positive autocorrelation. Combining these
insights, forecasted returns and variances (or volatilities) at the strategy-level can gener-
ate scaling weights that increase the Sharpe ratio of momentum compared to a non-scaled
strategy. Technically, as a net-zero investment long-short strategy, momentum can be scaled
without assuming leverage costs and the scaling can be interpreted as having a time-varying
weight in the long and short legs. We explicitly distinguish between constant-volatility scaling
and dynamic scaling.
Constant volatility-scaled momentum (cMOM), as proposed in Barroso and Santa-Clara
(2015), adjusts the momentum portfolio to a constant target volatility level. The correspond-
11
See, among other, Bekaert and Wu (2000), Barroso and Santa-Clara (2015) or Moreira and Muir (2017).
12
Alternatively, one could use the volatility of individual constituents of the momentum long-short port-
folios. However, using volatility at the strategy level is preferable for momentum due to the possibility of
volatility timing. Individual volatilities are rather useful to control for realized volatility (volatility smooth-
ing) of time-series momentum strategies (see, e.g., also du Plessis and Hallerbach, 2016, for U.S. industry
portfolios). Time-series momentum strategies are scaled upward by volatility (not the inverse, as for cross-
sectional momentum) due to the positive relation between volatility and returns.
11

ing scaling weight for momentum in month t is defined as:
σtarget
wcM OM,t = , (2)
σ̂t
where σtarget is chosen that the full sample volatilities of momentum and cMOM are identical,
and σ̂t = Et−1 [σt ] is the forecasted respective expected volatility.13 Since the forecasted
volatility varies over time, the weights for the constant volatility-scaled momentum portfolio
can take values between 0 (for σ̂t = ∞) and infinity (for σ̂t = 0). Following Barroso and
Santa-Clara (2015), we calculate the monthly volatility forecast for month t from past daily
returns of momentum in the previous six months (126 trading days),
126 2
RM OM,d−j,t
OM,t = 21 · (3)
2
X
σ̂M .
j=1 126
where RM
2
OM,d−j,t is the squared realized daily return of momentum returns summed over the
last 126 trading days.14 The correspondingly weighted momentum strategy is cMOM, where
the return in month t is calculated by weighting with the inverse of the realized volatility,
RcM OM,t = RM OM,t · wcM OM,t . (4)
Constant semi-volatility-scaled momentum (sMOM) is constructed analogously to cMOM,

but substitutes σ̂t in Equation (2) by the semi- or downside volatility of momentum. Specif-
ically, as in Wang and Yan (2021), the monthly downside volatility for month t is calculated
using the past 126 trading days:
126 2
RM OM,d−j,t I|RM OM,d−j <0|
2
= 21 · (5)
X
σ̂M .
OM,t,semi
j=1 126
13
Doing so, we (i) ensure that the strategy targets a constant risk level over time and (ii) make the returns
of the scaled and unscaled strategy comparable.
14
For robustness, we also use a one-month look-back window (21 trading days) as in Moreira and Muir
(2017). Our results remain unchanged.
12

The return of sMOM is the momentum return in month t weighted with the inverse of
the realized downside volatility, scaled with a static scalar to the full sample volatility of
momentum.
Dynamic-scaled momentum (dMOM) enhances the constant volatility-scaled momentum
by additionally incorporating the expected strategy return. In this regard, a mean-variance
optimizing investor optimizes momentum as her investment asset according to the dynamic
scaling weight that refers to their expected Sharpe ratio.15 We apply the dynamic approach
from Daniel and Moskowitz (2016) and define the dynamic scaling weight for momentum in
month t as
1 µ̂t

wdM OM,t = · 2, (6)
2λ σ̂t
where µ̂t = Et−1 [µt ] (σ̂t2 = Et−1 [σt2 ]) is the forecasted respective the conditional expected
return (variance) of momentum, and λ is a static scalar scaling the dynamic strategy to the
full sample volatility of momentum. The estimation of µ̂t and σ̂t2 can be conducted either
in-sample or out-of-sample. We apply only the out-of-sample approach because the in-sample
estimation suffers from a look-ahead bias. The return of momentum is forecasted with the
following time-series regression:
RM OM,t = γ0 + γint · IBear,t−1 · σRM

2
RF,t−1 + t (7)
where IBear,t−1 is a bear-market indicator that equals one if the cumulative past two year
market return is negative (and zero otherwise), σRM
2
RF,t−1 is the realized variance of RMRF
over the past 126 days, γint is the regression coefficient on the interaction term of the two
independent variables, and γ0 is the regression intercept. The expected return (µ̂t ) is defined
as the fitted values from the regression. We estimate Equation 7 on a monthly updated
15
It is, however, an underlying assumption of how investors end up with their return and variances forecasts,
respectively how their expectations align for the weight components. Daniel and Moskowitz (2016) mention
that investors optimize their objective function in-sample and unconditionally, which implies a forward-
looking bias. In this regard, we estimate return and variance out-of-sample.
13

expanding-window basis.16 For the variance forecast of the dynamic strategy, we rely on
the same approach as for the constant-volatility scaling strategy and use Equation 3 with a
126-day look-back window for σ̂t2 . We eventually derive dMOM as the dynamically weighted
momentum strategies with their return in month t given by
RdM OM,t = RM OM,t · wdM OM,t . (8)
Comparing the constant-volatility and the dynamic strategy, cMOM, and dMOM, the weights
differ by Et−1 [µt ] = µ̂t . Hence, the scaling weights of the dynamic strategies can take on
negative values when µ̂t < 0, e.g., (i) in bear market states and (ii) when market volatility
is higher as γint from Equation 7 is negative for most months. This possibility implies that
the dynamic weights might vary more strongly than the constant-volatility scaling weights,
capturing the dynamics of momentum returns.
2.4 Methodology
We investigate the performance of the different momentum strategies based on a compre-

hensive set of methodologies. First, we evaluate the different momentum strategies by mean
returns, t-statistics, higher moments (skewness and kurtosis) as well as maximum draw-
downs. Second, we compare risk-adjusted strategy returns with respect to the Fama-French
three-factor model and conduct pairwise mean-variance spanning tests. Third, we conduct
ex-post maximum Sharpe ratio tests of factor models, including different momentum strate-
gies. Last, we contrast the profitability of the different momentum strategies with their
portfolio turnover to assess their capacity for potential transaction costs. In this section, the
testing procedures are presented:
16
In contrast to Daniel and Moskowitz (2016), we estimate an out-of-sample regression in an expanding
fashion, starting with an initial 24 months for each sample.
14

Spanning tests To evaluate the strategies’ risk-adjusted performance, we calculate span-
ning alphas for momentum and the three alternative momentum strategies. Therefore, we
regress returns of the test asset (e.g., momentum) on returns of the benchmark assets within
a linear time-series regression model.17 The null hypothesis states that the test asset returns
are spanned, i.e., that the intercept equals zero. If the null hypothesis is rejected and the
intercept is statistically significantly different from zero, the test asset is not spanned by the
respective benchmark asset pricing model and extends the efficient portfolio frontier. The
intercept quantifies the magnitude of the abnormal return the test asset generates.
Maximum Sharpe ratio tests To quantify the differences in maximum Sharpe ratios for
different sets of factors, we apply the methodology from Barillas et al. (2019). In the mean-
variance efficient portfolio optimization, the economic significance of factors is quantified by
comparing how much an investor could gain from adding a certain factor to her investment
opportunity set. We start by calculating the differences between the bias-adjusted sample
squared Sharpe ratios for various pairs of factor models. The squared Sharpe ratio for each
model is modified to be unbiased for small samples under joint normality by multiplying it
by (T-K-2)/T and subtracting K/T, where T is the number of return observations and K is
the number of factors.
However, comparing mean-variance efficient tangency portfolios, as described above, does
not provide statistical evidence. Thus, we compute p-values for the test of equality of the
squared Sharpe ratios, also in a pairwise manner. For nested models (i.e., all of the factors
in one model are contained in the other model), we test whether the squared Sharpe ratio for
the model with more factors is higher than the squared Sharpe ratio of the model with fewer
factors. This becomes equivalent to testing whether the factors in the larger model that are
not also in the smaller model have significant alphas when regressed on the smaller model.
For example, because the CAPM is nested in the Fama-French three-factor model, one would
17
See, among others, Huberman and Kandel (1987).
15

test whether the CAPM alphas of SMB and HML are zero.18 For a single excluded factor,
the corresponding test is whether the intercept in the regression of the excluded factor on
the nested model is significantly different from zero.
Transaction Costs As in Grundy and Martin (2001) and Barroso and Santa-Clara (2015),
we calculate the round-trip costs that would render the profits of the different momentum
strategies insignificant at a certain α-significance level as
1.96 µ¯s 19

Round-trip costsα=5% = 1 − . (9)
t-stats T ¯Os
where µ¯s is the average monthly return, T ¯Os is the average monthly turnover, and t-stats is
the t-statistic of strategy s. Round-trip costs are advantageous since they define an upper
bound for the potential transaction costs instead of quantifying them directly. Only the
strategy returns, as well as the associated portfolio turnover, are necessary input factors.20
Moreover, statistical significance can be incorporated within the round-trip costs measure.
Holding turnover fixed, round-trip costs with reliance on the α-significance level increase for
strategies with higher t-statistics and, therefore, lead to a higher upper bound.
3 Empirical results
3.1 Comparison of Momentum Strategies
In this section, we compare the improvement of enhanced momentum factors as compared

to standard momentum in global markets. Table 2 depicts the return characteristics for the
four momentum strategies.
18
The statistical test for multiple excluded factors, as in this example, is the application of the (GRS) test
in Gibbons et al. (1989).
19
We choose the Z-value of 2.58 for the 1% significance level (instead of 1.96 for the 5% level).
20
We calculate portfolio turnover as the sum of changes in the securities’ weights within assigned long-short
factor portfolios. The details of the turnover calculation are presented in Appendix A.2.
16

For the long U.S. sample, the average monthly return of the standard momentum (MOM)
factor amounts to 0.60%, with a highly significant t-statistic of 4.40. However, as Barroso
and Santa-Clara (2015) and Daniel and Moskowitz (2016) already show, momentum also has
a dark side. The high returns come with a very high kurtosis and negative skewness, implying
large drawdowns (fat left tails) such as the maximum drawdown for the momentum factor in
1932 of -69.09%. The returns of the standard momentum factor for the non-U.S. sample show
similar features. The average monthly return is 0.54%, with a t-statistic of 2.81. The higher t-
statistics for the U.S. sample can be attributed to the larger number of monthly observations,
as indicated by similar annualized Sharpe ratios (0.47 vs. 0.53). The momentum returns for
the non-U.S. sample also exhibit a high kurtosis and a negative skewness but are more
normally distributed compared to their counterparts for the long sample. This result is also
reflected in the lower maximum drawdown in 2009 of -41.36%.
Comparing enhanced momentum strategies with MOM, we show that all of them exhibit
higher t-statistics and Sharpe ratios (both nearly double compared to standard momentum).
Furthermore, skewness, kurtosis, and maximum drawdowns decrease in magnitude as com-
pared to MOM so that their distributions become more normally distributed. As momentum
crashes typically happen following a bear market and when the contemporaneous market
return is positive (cf. Asem and Tian, 2010, ‘bear-up month’), we also compute the average
return (and t-statistic) for these months. We document statistically significant negative mean
returns returns during these bear-up states for both samples and all momentum strategies
except dMOM. The reason for this exception seems that solely dMOM accounts for the lower
expected momentum returns during bear market states, i.e., that it can also take a negative
weight in momentum.
To shed light on the co-movement of factor returns, Table 3 shows pairwise factor corre-
lation coefficients for both the U.S. and non-U.S. samples.
17

MOM and cMOM or sMOM are highly correlated with correlation coefficients above 94%.
The correlation between MOM and dMOM is lower, which can be traced back to the larger
amendments of the momentum factor by the strategy. Across the enhanced strategies, cMOM
and sMOM co-move nearly perfectly with each other (99% and 100% for U.S. and non-U.S.,
respectively) and less with dMOM (only up to 95%). In addition to the FF3FM factors,
we include a monthly updated HML-devil factor (HMLd ) in our correlation analysis. The
negative correlation between MOM and HMLd as compared to HML is evident for both our
U.S. and non-U.S. samples confirming the results in Asness and Frazzini (2013) and Hanauer
(2020). cMOM and sMOM are similarly negativly correlated with HMLd , while dMOM is
slightly less negatively correlated. However, all three enhanced momentum strategies are
more negatively correlated with HMLd than with HML, which have a similar mean return.
Thus, in non-categorized tests (where only either HML or HMLd is included), we henceforth
benchmark our enhanced momentum strategy returns against the FF3FM model where we
substitute HML by HMLd , denoted as FFd .
Figure 1 displays the buy-and-hold returns of momentum and the proposed enhanced
strategies for the U.S. sample. All strategies are scaled to the average in-sample volatility
of standard momentum for comparability. For the U.S., all strategies increase in returns
as compared to momentum, with dMOM as the highest return strategy. dMOM manages
to hedge the momentum downturns in 2001 and 2008 and generates stable returns from the
1950s on. All strategies show a similar performance by construction, differing solely by either
the numerator or the denominator in the scaling weights. In the non-U.S. sample, cMOM
slightly outperforms the other strategies, as shown in Figure 2.
Next, we analyze the improvement of our enhanced momentum strategies in comparison
to momentum by answering the following two questions: How well can enhanced momentum
strategies explain standard momentum? And are enhanced momentum strategies able to
generate risk-adjusted returns when controlling for momentum? Therefore, we conduct mean-
variance spanning tests by varying the benchmark and test assets. In case the spanning alpha
18

is economically meaningful and statistically significant, the test asset is not spanned by the
respective benchmark asset pricing model and therefore contains unexplained information.
We start by regressing momentum on the FFd factors and subsequently add single enhanced
momentum factors. Following Barillas and Shanken (2017), we consider the FFd factors as
alternative factors when conducting mean-variance tests.
The first row in both panels of Table 4 depicts the results from factor spanning tests of
momentum. The FFd alpha of MOM amounts to 0.88% and 0.97% monthly for the U.S.
and non-U.S. samples, respectively. However, when adding enhanced momentum strategies
as benchmark assets, the alphas substantially decrease in value (up to 0.04% and -0.04%),
remaining only statistically significant with respect to dMOM for both the U.S. and the non-
U.S. sample. If standard momentum was regressed on dMOM standalone, the alpha would
be insignificant. This difference stems from the interaction with FFd factors. As we saw in
Table 3, HMLd is most negatively correlated with momentum, followed by cMOM, sMOM,
and dMOM. This highly negative correlation makes momentum relatively more attractive
when controlling for HMLd than when not.
To get further insights into the relative improvement of enhanced strategies, we employ
them as test assets. In the second to the fourth row of each panel, we regress enhanced
momentum strategies on the FFd factors plus momentum in the first column and separately
add another enhanced momentum strategy as benchmark asset in columns 2 to 4. We find
that the alphas of all three enhanced strategies with respect to the FFd factors plus momen-
tum are economically and statistically significant (first column). For the U.S. in Panel A,
sMOM is subsumed by cMOM and dMOM, while cMOM and dMOM yield significant span-
ning alphas when controlling for sMOM, as depicted in the third column. These results are
confirmed for the non-U.S. sample in Panel B, with the sole difference that dMOM is sub-
sumed by both cMOM and sMOM. In sum, we conclude that momentum is not subsumed
by dMOM and that the volatility-scaled strategies add to the investment opportunity set
19

including momentum.
The spanning tests in Table 4 reveal that none enhanced momentum strategy consistently
subsumes all other momentum strategies. However, a particular momentum strategy might
still span a higher Sharpe ratio than another momentum strategy when added to an existing
investment opportunity set, such as the FF3FM or FFd factors. To test for the statistical
significance, we follow Barillas et al. (2019) and conduct pairwise equality tests of the models’
(unbiased) maximum squared Sharpe ratios (cf. also Section 2.4).
Panel A in Table 5 shows the differences between the (bias-adjusted) sample squared
Sharpe ratios (column minus row for the upper triangle; vice versa for the lower triangle)
for different sets of factors and the U.S. (non-U.S.) sample in the upper (lower) triangles.
Panel B reports p-values for the tests of equality of the squared Sharpe ratios. The p-values
are computed differently depending on whether the models to be compared are nested or
non-nested.
The main findings can be summarized as follows: First, the results show that the FF3FM
plus momentum (Carhart, 1997) is dominated by the FFd model plus momentum with signif-
icance at the 1% level for both samples. This result primarily stems from the more negative
correlation of momentum and HMLd as identified in Table 3 and confirms the results in Asness
and Frazzini (2013) and Hanauer (2020). When we augment the FFd model plus momentum
with enhanced strategies, the resulting five-factor models outperform the non-augmented
model, irrespective of the enhanced momentum strategy (cMOM, sMOM, or dMOM), at the
1% level. As already suggested by the factor spanning tests, enhanced momentum strate-
gies incorporate significant information, even beyond standard momentum. When comparing
the models augmented with enhanced strategies among each other, the emerging picture is
less clear: in the non-U.S. sample, the augmented model with cMOM outperforms the one
augmented with sMOM (-0.024) at the 5% level (p=0.037), and the augmented model with
dMOM is outperformed by the one with cMOM (-0.034) at the 10% level (p=0.089). In con-
20

trast, the difference in the sample squared Sharpe ratios between the augmented model with
sMOM and dMOM is not significant. For the U.S. sample, the model with cMOM outper-
forms the one augmented with sMOM (-0.010) at the 5% level (p=0.013), while there is no
statistically significant improvement among the other combinations of enhanced strategies,
as indicated by the associated p-values that are 0.144 and 0.810, respectively.
3.2 Cross-country analysis
In the previous subsection, we demonstrated that the enhanced momentum strategies exhibit
higher risk-adjusted performance than momentum and that they are distinct phenomena for
the U.S. and the non-U.S. sample. Within this subsection, we investigate for which countries
enhanced momentum strategies yield an improvement. Moreover, we utilize diverse cross-
country characteristics to investigate potential drivers of momentum premia.
In Table 6, we present mean returns and t-statistics for momentum and the enhanced
versions. We include only countries for which more than 120 monthly observations of the
strategies are available. The values printed in bold indicate statistically significant (enhanced)
momentum strategy returns. The enhanced versions have both higher returns and risk-return
ratios (t-statistics) across countries. For 19 out of 23 developed market countries, momentum
has significant returns.
For Japan, none of the enhanced strategies, for the Ireland cMOM, for Spain cMOM and
sMOM, and for Singapore all enhanced strategies generate significant premia, respectively.
In contrast, momentum has an insignificant premium in each of the countries. 14 out of
20 countries in the emerging markets have insignificant momentum returns, of which seven
show significant returns for at least one enhanced momentum strategy. Special attention is
paid to Asia, for which momentum is found to deliver insignificant returns (see, e.g., Chui
et al., 2010). The enhanced strategies generate statistically significant and positive returns
21

in Asian countries where momentum does not, e.g., Indonesia, Malaysia, South Korea, or
Taiwan.
To further investigate the effectiveness of the enhanced momentum strategies, we present
the maximum drawdown for each country in Figure 3. For most countries, the maximum
drawdowns are reduced for the enhanced strategies as compared to momentum, providing
further evidence for the lowered crash risk. This is particularly true for cMOM and sMOM,
while some more exceptions exist for dMOM (e.g., Japan, Argentina, Philippines, and In-
donesia). In sum, we see a robust improvement across countries.
The results above reveal that standard momentum and enhanced momentum have cross-
sectional stock return predictability around the globe. However, the observed momentum
premia vary across countries. Therefore, we next analyze the effect of cross-country predictors
on (enhanced) momentum returns. Following Jacobs (2016), we run Panel regressions of
monthly momentum strategy returns as dependent variables on contemporaneous country-
month averages or time-invariant country characteristics in a fixed effects model.21 For this
purpose, we utilize proxies for overconfidence that were found to explain momentum from
Chui et al. (2010) and Asem and Tian (2010). Additionally, we compute the comprehensive
country-level predictor set from Jacobs (2016) as control variables.
We proxy for time-invariant and time-varying investor overconfidence with countries’ indi-
vidualism scores (Chui et al., 2010) and market state continuation dummies (Asem and Tian,
2010), respectively. Chui et al. (2010) argue that individualism is positively associated with
country-level trading volume and volatility, and is found to be positively associated with mo-
mentum premia across countries. The market state continuation dummy distinguishes market
conditions into continuations and transitions. Following Asem and Tian (2010) and Hanauer
(2014), we classify a country-month as market state continuation, depending on whether the
past 12 months and the current month market return have the same sign. Asem and Tian
(2010) and Hanauer (2014) find that momentum returns across time are much higher in
21
The results remain unchanged when using FFd alphas instead of raw returns.
22

market continuations than in market transitions for the U.S. and Japan, respectively. They
argue that this pattern is consistent with the model of Daniel et al. (1998), which assumes
that time-varying investor overconfidence (induced by confirming market movements) and
self-attribution drive momentum profits.22 To the best of our knowledge, our study is the
first that tests the explanatory power of market continuations across countries.
As positive and negative proxies for limits-to-arbitrage and market frictions, we construct
a country’s average firm size (neg.), average firm-level idiosyncratic volatility (pos.), average
number of analysts that provide one-year earnings estimates (neg.) as well as a dummy
variable that equals one if short selling is legally allowed (neg.). Idiosyncratic volatility
is the standard deviation of the regression residual from the stock-specific estimation of a
country-specific FFd model over the past 12 months on a rolling-window basis. We proxy
for information uncertainty and opinion dispersion with average analyst forecast dispersion,
calculated as the standard deviation of one-year earnings estimates scaled by the absolute
value of the mean forecast. Additionally, we use return R2 (inversely related to firm-specific
return variation) and the average fraction of zero return days (inversely related to trading
activity) as proxies for market efficiency. As discussed in Jacobs (2016), competing evidence
exists on the direction of these two predictors, i.e., if they proxy for market efficiency or
inefficiency.
The average fraction of zero returns can be interpreted as a proxy for trading activity.
Higher trading activity can represent arbitrage trading and therefore result in market effi-
ciency (McLean and Pontiff, 2016), or reflect noise trading and result in market inefficiency
(Daniel et al., 1998). The interpretation of R2 depends on whether the information that trig-
gers investors’ non-systematic trading behavior either is valuable firm-specific information or
22
In the model of Daniel et al. (1998), traders receive public signals after trading a stock based on a
private signal. If the public signal confirms their private signal, the investors attribute the success to their
skills. However, they attribute non-confirming signals to bad luck. Because of this self-attribution bias,
traders become overconfident about their stock selection skills, and this overconfidence drives momentum.
Asem and Tian (2010) argue that investors, on average, traded more based on positive (negative) private
signals when the past market was positive (negative). Consequently, subsequent positive months should drive
overconfidence more than subsequent negative months and vice versa. Therefore, momentum returns should
also be higher for market continuations than for market reversals.
23

noise. In the former (latter) case, a lower R2 is proxying for market efficiency (inefficiency),
as argued in Morck, Yeung, and Yu (2000) (Hou et al., 2013).
We start by analyzing momentum in a fixed effects model and estimate the following
equation:
Reti,t = Xi,t + controls + ei,t , (10)
where Xi,t is a vector of country characteristics. We include time effects to account for
country-invariant time effects, e.g., global macroeconomic shocks that affect countries across
the globe, and subsequently add country fixed effects to control for time-invariant country
characteristics that affect momentum, such as the accessibility of the local stock market.
Standard errors are double-clustered by country and month. All explanatory variables (ex-
cept dummy variables) are standardized to have a mean of zero and a variance of one.
Furthermore, we also estimate a fixed-effects model to investigate the differences in expo-
sure to the cross-country predictors between momentum and enhanced momentum strategies.
Econometrically, we follow Ozdagli (2017), who compares the point estimates between rated
and unrated firms within a panel regression framework, including variation across firms and
time. The results of this specification are presented in Table A.8 of the appendix.
Table 7 shows the results for the cross-country analysis. We find the by far strongest
impact (t-statistic of 6.96!) for the market continuation dummy. For countries in a market
continuation month (e.g., positive market return following a positive year for the market),
momentum returns are, on average, 1.86% higher than for countries in a market reversal
month (e.g., positive market return following a negative year for the market). This result
expands the findings of Asem and Tian (2010) and Hanauer (2014), who provide evidence
for the time-series momentum predictability of the market continuation dummy in single
markets. The individualism score is statistically significant but economically and statistically
weaker (t-statistic of 2.07), suggesting that time-invariant overconfidence proxied by the
24

individualism score affects momentum returns for the full sample period.23
Among the cross-country predictor variables for momentum, only the estimated coeffi-
cients on firm size, return R2 , and idiosyncratic volatility are significant. R2 is negatively
associated with momentum returns, a finding consistent with Jacobs (2016) for the mispric-
ing factor. In this regard, R2 is proxying for market inefficiency, so that countries with low
R2 experience more noise trading and have higher momentum returns. The estimated coef-
ficients on firm size and idiosyncratic volatility are negative. A negative coefficient for size
is proxies for limits-to-arbitrage, whereas the negative coefficient for idiosyncratic volatility
differs from the predicted sign. The remaining predictors have no statistically meaningful
effect on momentum returns.
The results for the market continuation dummy remain unchanged when including coun-
try fixed effects to address unobserved time-invariant country characteristics. However, the
estimated coefficient on return R2 is no longer significant. Therefore, we conclude that
the market continuation dummy (proxying time-varying investor overconfidence) has the
strongest impact on differences in country momentum returns.
The returns of cMOM and sMOM have a similar cross-country behavior as momentum,
but more variation across countries can be attributed to R2 (more noise trading). Impor-
tantly, both enhanced strategies are comparably affected by the market continuation dummy
as momentum. The point estimates of 1.74% and 1.71% are both highly statistically signif-
icant, suggesting an almost unaltered market-state dependence of cMOM and sMOM (see
also Table A.8). Additionally, the individualism dummy has a marginally significant effect
on cMOM and sMOM.
dMOM is significantly less affected by the market continuation dummy than momentum,
with a point estimate amounting to 0.70%.24 This finding highlights that dMOM hedges
out a substantial part of the market state dependence of momentum. Moreover, we find
23
Our results remain unchanged when including the long-term orientation proxy from Docherty and Hurst
(2018) as an additional predictor.
24
Table A.8 shows that the difference in exposure toward the market continuation dummy between mo-
mentum and dMOM is statistically significant.
25

that the returns of dMOM across countries are significantly higher related to idiosyncratic
volatility than standard momentum, leading to an insignificant point estimate of -0.07. The
coefficients for firm size and R2 do not differ significantly as compared to momentum. Finally,
the individualism dummy has no effect on dMOM. Controlling for country-fixed effects leads
to similar results.
We further conduct a robustness test in Table A.9 and split up our sample into two sub-
periods. We confirm the positive, significant effect of individualism on momentum returns for
the first sub-period from January 1990 to December 2003, which aligns with the sample period
in Chui et al. (2010). Moreover, we show that enhanced momentum returns are significantly
affected by individualism across countries as well. However, for the second sub-period from
January 2004 to December 2017, a country’s individualism score is not related to either
momentum or enhanced momentum returns, suggesting that the results in Chui et al. (2010)
do not hold out-of-sample. In contrast, we identify a robust effect of the market continuation
dummy on all momentum strategies in both sub-samples. Our study thereby provides out-
of-sample evidence for the explanatory power of individualism scores for momentum returns
and suggests that time-varying investor overconfidence, proxied by the market continuation
dummy, is superior to time-invariant investor overconfidence, as proxied by individualism.
In sum, our results suggest that MOM, cMOM, and sMOM show the strongest relationship
with the proxy for time-varying investor overconfidence, while dMOM is significantly less
affected by these proxies. Furthermore, we also find some evidence that all momentum
strategies are related to proxies for the presence of noise trading.
To further investigate the lower sensitivity of dMOM toward the market continuation
dummy, we analyze the market continuation dummy separately for bull or bear markets.
Specifically, we apply the same cross-country panel analysis design as in Table 7 and regress
momentum returns on a dummy that is one for market continuations after a bear (bull)
market while controlling for bull (bear) market states with another dummy. The results are
shown in Table 8.
26

Table 8 shows that MOM, cMOM, and sMOM are more related to market continuations
in bear than to market continuations in bull markets. Specifically, we find that during
bear markets, a market continuation, defined as a contemporaneous down-movement of the
market, generates significantly higher returns (2.96% to 3.73%) as compared to a market
reversal, defined as a contemporaneous market up-movement. In contrast, the difference is
statistically insignificant (t=1.34) for dMOM. During bull markets, a market continuation
generates significantly higher returns than a reversal across all four strategies, although the
difference is economically weaker for dMOM. The cross-country results for MOM, cMOM,
and sMOM are still consistent with Daniel et al. (1998) and Asem and Tian (2010). However,
for dMOM, the sensitivity toward the market continuation dummy is only existent for bull,
but not for bear markets, so that the behavior of dMOM is not entirely consistent with Daniel
et al. (1998).
The higher sensitivity of momentum to market continuations in bear markets is also con-
sistent with the option-like behavior of momentum during bear markets described in Daniel
and Moskowitz (2016), i.e., momentum in bear markets behaves like a written (short) call
option on the market. This behavior can be seen as consistent with the theory of Merton
(1974) that a common stock is a call option on the value of the firm. As a consequence,
expected momentum returns are low in bear markets with high volatility and a contempora-
neous market reversal. As the weight of dMOM in MOM linearly increases with the expected
momentum return for the next month (see Equation 6), it is smallest in these moments.25
These mechanics explain why the sensitivity of dMOM to market continuations is reduced
more in bear markets than in bull markets.
25
The estimated value for γint in Equation 6 is negative for nearly all months
27

3.3 Turnover and Transaction Costs
All momentum strategies are constructed as zero-cost long-short strategies. The returns
reported in Table 2, however, ignore transaction costs for implementing the strategies. As
mentioned in Barroso and Santa-Clara (2015), “[o]ne relevant issue is whether time-varying
weights induce such an increase in turnover that eventually offsets the benefits of the strategy
after transaction costs.” Table 9 shows the average (over time) one-way portfolio turnover of
the long leg plus the short leg. We find, on average, a turnover of more than 50% per month
when using value-weighted returns for all strategies.26 For the U.S. sample in Panel A,
turnover increases for all enhanced strategies, especially for dMOM, reaching a maximum
of 85.43% monthly and an increase in turnover of 30.09 (85.43-55.34) percentage points as
compared to momentum. Similar results hold for the shorter global sample (from January
1990 to December 2017) in Panel B, where dMOM generates a maximum average monthly
turnover of 74.24%.27 Next, we aim to investigate whether the increase in turnover offsets
the benefits of volatility scaling.
Round-trip costs describe transaction cost levels (in percent) that would render the strate-
gies’ returns statistically insignificant at confidence levels of 5% and 1%. Panel A of Table 9
shows that momentum investors within the U.S. are only 5% sure that their strategy will have
positive net profits when transaction costs do not exceed 0.60%. Comparing the enhanced
strategies, the transaction costs that would remove the statistical significance of profits (at
the 5% level) are higher than for conventional momentum and highest for dynamic-scaled mo-
mentum (1.03%). When increasing the confidence level, a similar picture emerges. Panel B
shows that for the global sample, constant volatility-scaled momentum clearly gives the high-
26
Our monthly turnover for momentum, however, is lower than the 74% in Barroso and Santa-Clara (2015).
We trace back the difference to not using decile momentum portfolios but forming HML-style portfolios based
on 70/30% percentile breakpoints and double-sorts including size. We are able to validate the turnover for
the equivalent sub-sample period using the more extreme decile breakpoints and single sorts.
27
Daniel and Moskowitz (2016) do not report any turnover or break-even cost statistics for their dynamic
strategies.
28

est bounds for round-trip costs.
Our approach does not explicitly test the after-trading cost performance of the differ-
ent momentum strategies, nor does it analyze the effectiveness of transaction cost mitigation
techniques.28 Instead, this break-even cost study reveals how profitable each strategy remains
when assuming a certain level of transaction costs. Stated differently, it merely defines an
upper transaction cost bound for momentum investors. In this regard, Frazzini, Israel, and
Moskowitz (2014) and Novy-Marx and Velikov (2016) show for the U.S. that even standard
momentum, which reveals the lowest break-even round-trip costs within our analysis, is a
profitable and thus implementable trading strategy. Notably, we are aware of the following
caveats with respect to our argumentation: the potential interaction effects between stock
market volatility (and thus also realized volatility of momentum) and transaction costs (esp.
bid-ask spreads) due to liquidity reasons. For the non-U.S. sample, we argue that all en-
hanced strategies are — as indicated by higher break-even round-trip costs — at least as
implementable as standard momentum.
3.4 Momentum Strategies and the January Effect
Jegadeesh and Titman (1993) document that momentum returns are positive in all months
except January, but losers significantly outperform winners in January. This result has been
confirmed by Jegadeesh and Titman (2001), Grundy and Martin (2001), and Blitz, Huij, and
Martens (2011), among others. Importantly, Grundy and Martin (2001) point out that in
January, the momentum strategy “[...] goes short in prior losers, and prior losers tend to
have become extremely small firms”, which is traced back to fund managers selling small-cap
loser stocks in December that reverse in January. In Table 10, we split up our sample into
January and non-January months.
Panel A shows the mean percentage returns and t-statistics for the U.S. sample. Even
28
See Novy-Marx and Velikov (2016) for details on these two subjects.
29

though all enhanced momentum strategies exhibit higher returns than conventional momen-
tum in January, none of the returns are positive on average. Furthermore, the difference
between non-January and January month returns is statistically significant for momentum
and all enhanced strategies. Panel B shows results for the non-U.S. sample: all strategies
have higher returns in non-January than in January months, but the difference between Jan-
uary and non-January returns is not statistically significant. In sum, all enhanced strategies
exhibit their performance mainly during non-January months.
4 Conclusion
This paper studies the performance of three enhanced momentum strategies proposed in
the literature — constant volatility-scaled momentum (cMOM), constant semi-volatility-
scaled momentum (sMOM), and dynamic-scaled momentum (dMOM) — for a total of 49
developed and emerging market countries and a sample period of about 28 years (88 years
for the U.S.). We document that all three enhanced momentum strategies substantially
increase in Sharpe ratios, both skewness and kurtosis decrease in magnitude, and suffer
from less pronounced drawdowns. Furthermore, enhanced momentum strategies generate
significantly positive alphas in mean-variance spanning tests on asset pricing factors that
include standard momentum. Ex-post maximum Sharpe ratio tests, as in Barillas et al.
(2019), confirm that all enhanced momentum strategies significantly increase the maximum
Sharpe ratio as compared to the factor opportunity set that already includes momentum for
both the U.S. and the non-U.S. sample. However, no clear picture emerges when we compare
the three enhanced approaches with this methodology.
We then exploit the dispersion in both country characteristics and momentum returns of
our international dataset and conduct panel regressions to explain the variation of momen-
tum returns across countries. Our results show that momentum, constant volatility-scaled
momentum, and constant semi-volatility-scaled momentum are most affected by proxies for
30

time-varying investor overconfidence, while dynamic-scaled momentum is significantly less
affected by these proxies. This result is robust when we split up our sample into two sub-
periods from 01/1990 to 12/2003 and from 01/2004 to 12/2017. In contrast, a country’s
individualism score is only significantly related to momentum returns for the first sub-period,
suggesting that the results in Chui et al. (2010) do not hold out-of-sample. Furthermore, we
find some evidence that all momentum strategy returns are positively related to proxies for
the presence of noise trading.
Finally, enhanced momentum strategies should be at least as implementable as standard
momentum, as their break-even costs (i.e., transaction costs that theoretically would render
the strategies insignificant) are higher than the ones for standard momentum. Moreover, both
momentum and enhanced momentum strategies generate their performance mainly during
non-January months.
31

Table 1: Descriptive statistics
The table presents summary statistics for the 49 countries of our CRSP, Datastream and Worldscope sample. Column 2 states
the market affiliation according to MSCI, with DM as Developed Markets and EM as Emerging Markets. Columns 3, 4, and 5
report the total, minimum, and maximum number of firms per country, respectively. Columns 6 states the average mean size
per country-month. Column 7 shows the average total size per country-month and column 8 reports these values in percentage
of the respective total across countries. Size is measured as market capitalization in million USD. The last two columns report
the start and end date of data availability for each country. We require that firms have non-missing values for the following
items: market value of equity, book-to-market, current month return, and lagged month market capitalization.
Total no. Min no. Max no. Mean Average Average

Country Market Start date End date
firms firms firms size total size total size in %
Argentina EM 89 63 75 451 30835 0.09 2001-07-01 2017-12-31
Australia DM 2645 113 1574 762 646804 1.96 1990-01-01 2017-12-31
Austria DM 163 45 87 1024 70348 0.21 1991-07-01 2017-12-31
Belgium DM 234 80 136 1829 204069 0.62 1991-07-01 2017-12-31
Brazil EM 244 77 179 2574 364046 1.10 2000-07-01 2017-12-31
32
Canada DM 3715 223 2183 757 789063 2.39 1990-01-01 2017-12-31
Chile EM 242 73 170 918 141856 0.43 1996-07-01 2017-12-31
China EM 3036 101 2943 1320 2573417 7.78 1999-07-01 2017-12-31
Colombia EM 73 30 55 2531 117104 0.35 2005-07-01 2017-12-31
Czech Republic EM 57 30 48 529 18235 0.06 2001-07-01 2006-01-31
Denmark DM 319 124 205 945 144378 0.44 1993-07-01 2017-12-31
Egypt EM 157 116 135 400 50502 0.15 2008-07-01 2017-12-31
Finland DM 220 44 136 1614 191458 0.58 1994-07-01 2017-12-31
France DM 1616 176 793 1896 1232386 3.73 1990-01-01 2017-12-31
Germany DM 1459 271 867 1640 1033357 3.13 1990-01-01 2017-12-31

Greece DM/EM 381 88 282 344 75558 0.23 1995-07-01 2017-12-31
Hong Kong DM 1529 76 1348 541 439253 1.33 1995-07-01 2017-12-31
Hungary EM 63 30 41 740 25666 0.08 2003-07-01 2016-12-31
India EM 3250 277 2499 540 778699 2.36 1997-07-01 2017-12-31
Indonesia EM 596 125 474 521 186989 0.57 1996-07-01 2017-12-31
Ireland DM 80 30 50 1307 49043 0.15 1993-07-01 2012-05-31
[Continued on next page]

Total no. Min no. Max no. Mean Average Average
Country Market Start date End date
firms firms firms size total size total size in %
Israel DM/EM 497 69 403 481 124644 0.38 2002-07-01 2017-12-31
Italy DM 552 160 261 1958 434512 1.31 1990-01-01 2017-12-31
Japan DM 5207 1065 3748 1184 3297967 9.97 1990-01-01 2017-12-31
Malaysia EM 1261 202 928 380 239358 0.72 1994-01-01 2017-12-31
Mexico EM 192 65 117 1869 195116 0.59 1997-07-01 2017-12-31
Morocco EM 76 69 73 818 58102 0.18 2010-07-01 2017-12-31
Netherlands DM 250 70 161 2728 279561 0.85 1990-01-01 2017-12-31
New Zealand DM 212 50 122 381 39284 0.12 1999-07-01 2017-12-31
Norway DM 415 60 180 942 137922 0.42 1992-07-01 2017-12-31
Pakistan EM 298 56 251 155 21701 0.07 1997-07-01 2017-12-31
Peru EM 128 46 99 550 46179 0.14 2004-08-01 2017-12-31
Philippines EM 299 95 231 493 99193 0.30 1996-07-01 2017-12-31
Poland EM 680 72 495 406 113879 0.34 2000-10-01 2017-12-31
Portugal DM/EM 133 40 92 1086 56786 0.17 1994-07-01 2017-12-31
Qatar EM 43 41 43 3714 155183 0.47 2014-07-01 2017-12-31
33
Russia EM 468 163 347 2412 590562 1.79 2008-07-01 2017-12-31
Singapore DM 834 48 563 598 212403 0.64 1990-07-01 2017-12-31
South Africa EM 746 133 433 1091 276045 0.83 1995-07-01 2017-12-31
South Korea EM 2478 109 1930 488 530532 1.60 1994-01-01 2017-12-31
Spain DM 318 100 160 3550 474230 1.43 1992-07-01 2017-12-31
Sweden DM 861 96 498 1001 305963 0.93 1992-07-01 2017-12-31
Switzerland DM 358 108 238 3655 769711 2.33 1990-01-01 2017-12-31
Taiwan EM 2097 231 1740 552 588967 1.78 1998-07-01 2017-12-31
Thailand EM 790 225 614 389 179716 0.54 1995-07-01 2017-12-31
Turkey EM 435 65 349 587 149831 0.45 1997-07-01 2017-12-31

U.K. DM 3822 1072 1526 1626 2011021 6.08 1990-01-01 2017-12-31
U.S. DM 23182 536 6651 1057 4294423 37.24 1930-01-01 2017-12-31
United Arab Emirates EM 105 98 105 2002 200589 0.61 2014-07-01 2017-12-31
Table 2: Summary statistics for Momentum Strategies
The table presents the following summary statistics for MOM, cMOM, sMOM, and dMOM:
(1) average monthly returns (in %), (2) corresponding t-statistics, (3) annualized Sharpe
ratios, (4) skewness, (5) kurtosis, (6) maximum drawdown (in %), (7) average monthly
returns (in %) during bear-up market states, and (8) corresponding t-statistics, defined as
the maximum cumulative loss between a peak and subsequent downturn during the buy-and-
hold resp. sample period. cMOM is momentum scaled to a constant volatility level by its
six month past volatility. sMOM is momentum scaled to a constant volatility level by its six
month past semi-volatility. dMOM is momentum scaled by its forecasted return relative to
its expected variance. The analysis is performed from 01/1930 (01/1990) to 12/2017 for the
U.S. (non-U.S.) sample.
MOM cMOM sMOM dMOM

Panel A: U.S. (01/1930 - 12/2017)
Avg. Returns (in %) 0.60 1.10 1.05 1.16
(4.40) (8.14) (7.76) (8.57)
Sharpe (annualized) 0.47 0.87 0.83 0.91
Skewness -2.18 -0.32 -0.27 0.10
Kurtosis 21.33 2.33 3.60 7.58
Max. Drawdown (in %) -69.09 -38.59 -38.27 -40.05
Bear-Up Returns (in %) -0.39 -0.24 -0.24 -0.09
(-4.31) (-3.79) (-3.77) (-1.62)
Panel B: Non-U.S. (07/1990 - 12/2017)
Avg. Returns (in %) 0.54 1.00 0.96 0.97
(2.81) (5.19) (4.95) (5.01)
Sharpe (annualized) 0.53 0.98 0.94 0.95
Skewness -0.91 -0.11 0.16 0.38
Kurtosis 6.93 0.56 0.88 2.20
Max. Drawdown (in %) -41.36 -25.25 -22.88 -16.98
Bear-Up Returns (in %) -0.43 -0.30 -0.29 0.02
(-3.74) (-3.76) (-3.96) (0.25)
34

Table 3: Correlation coefficients
The table reports the time-series averages of the cross-sectional spearman correlation coeffi-
cients between the following variables: RMRF (market factor), SMB, HML, HMLd , MOM,
cMOM, sMOM, and dMOM. For details regarding variable construction, see Section 2.2. The
analysis is performed from 01/1930 (01/1990) to 12/2017 for the U.S. (non-U.S.) sample and
depicted within the upper (lower) triangle.
RMRF SMB HML HMLd MOM cMOM sMOM dMOM

RMRF 0.27 0.03 0.07 -0.05 -0.03 -0.03 0.01
SMB 0.10 0.01 0.08 -0.04 -0.04 -0.04 -0.02
HML -0.20 -0.11 0.84 -0.17 -0.15 -0.14 -0.12
HMLd -0.07 -0.08 0.80 -0.45 -0.41 -0.41 -0.34
MOM -0.13 0.04 -0.12 -0.53 0.96 0.95 0.84
cMOM -0.08 0.02 -0.12 -0.53 0.96 1.00 0.94
sMOM -0.08 0.01 -0.12 -0.52 0.94 0.99 0.95
dMOM 0.07 0.01 -0.17 -0.44 0.68 0.79 0.79
35

Table 4: Factor Spanning Tests of Momentum Strategies
The table presents alphas and corresponding Newey and West (1987) t-statistics (6 monthly
lags) from mean-variance spanning tests for the U.S. and non-U.S. sample. The dependent
variables are momentum and the enhanced momentum strategies. The independent variables
are: the FFd factors (RMRF (market factor), SMB, and HMLd ) as well as MOM, cMOM,
sMOM, or dMOM. When MOM is considered the test asset, we omit MOM as independent
variable. The independent factor set for each spanning test is shown above the respective
results. For details regarding variable construction, see Section 2.2. The analysis is performed
at monthly frequency over the time-series from 01/1930 (01/1990) to 12/2017 for the U.S.
(non-U.S.) sample in Panel A (Panel B).
Panel A: U.S. (01/1930 - 12/2017)

Ind. var. FFd (+MOM) FFd (+MOM) FFd (+MOM) FFd (+MOM)
+cMOM +sMOM +dMOM
MOM 0.88 0.04 0.12 0.35
(9.34) (0.69) (1.89) (4.78)
cMOM 0.43 0.08 0.05
(4.97) (3.32) (1.44)
sMOM 0.39 -0.05 0.00
(4.26) (-2.12) (-0.08)
dMOM 0.63 0.06 0.15
(5.00) (1.29) (3.12)
Panel B: Non-U.S. (01/1990 - 12/2017)
Ind. var. FFd (+MOM) FFd (+MOM) FFd (+MOM) FFd (+MOM)
+cMOM +sMOM +dMOM
MOM 0.97 -0.04 0.07 0.66
(6.67) (-0.38) (0.63) (4.77)
cMOM 0.47 0.10 0.18
(4.21) (3.41) (1.87)
sMOM 0.44 -0.06 0.13
(3.36) (-1.69) (1.03)
dMOM 0.67 -0.05 0.09
(3.50) (-0.40) (0.70)
36

Table 5: Tests of Equality of Squared Sharpe Ratios
The table presents pairwise tests of the equality of models’ squared Sharpe ratios. The following models are considered: FF
(RMRF (market factor), SMB, and HML) plus MOM, FFd (RMRF (market factor), SMB, and HMLd ) plus MOM, as well
as each combination of FFd plus MOM with either cMOM, sMOM, or dMOM. Panel A reports the difference between the
(bias-adjusted) sample squared Sharpe ratios (column minus row for the upper triangle; vice versa for the lower triangle) and
Panel B reports the associated p-values. The analysis is performed from 01/1930 (01/1990) to 12/2017 for the U.S. (non-U.S.)
sample and depicted within the upper (lower) triangle.
FF+MOM FFd +MOM FFd +MOM+cMOM FFd +MOM+sMOM FFd +MOM+dMOM

Panel A: Differences in Sample Squared Sharpe Ratios
FF+MOM 0.023 0.056 0.047 0.055
37
FFd +MOM 0.065 0.033 0.024 0.032
FFd +MOM+cMOM 0.142 0.077 -0.010 -0.001
FFd +MOM+sMOM 0.119 0.053 -0.024 0.008
FFd +MOM+dMOM 0.108 0.042 -0.034 -0.011
Panel B: p-Values
FF+MOM 0.002 0.000 0.000 0.000
FFd +MOM 0.000 0.000 0.000 0.000
FFd +MOM+cMOM 0.000 0.000 0.013 0.810

FFd +MOM+sMOM 0.000 0.000 0.037 0.144
FFd +MOM+dMOM 0.001 0.000 0.089 0.557
Table 6: Momentum Strategies per Country
The table presents the mean returns (in %) and corresponding t-statistics of momentum
and the enhanced strategies for the 49 countries in our sample. The following strategies are
included: MOM, cMOM, sMOM, and dMOM. For details regarding variable construction,
see Section 2.2. The analyses are performed at monthly frequency from 01/1930 (01/1990)
to 12/2017 for the U.S. (non-U.S. counries). For a country to be regarded, a minimum of
120 observations are required. The third column shows the number of observations for the
strategies. The values printed in bold indicate statistically significant (enhanced) strategy
returns at the 5% level.
MOM cMOM sMOM dMOM
Country Market months Rt t(Rt ) Rt t(Rt ) Rt t(Rt ) Rt t(Rt )

Australia DM 336 1.39 ( 7.17) 1.64 ( 8.48) 1.62 ( 8.37) 1.58 ( 8.16)
Austria DM 318 0.98 ( 3.56) 1.35 ( 4.91) 1.36 ( 4.92) 0.94 ( 3.41)
Belgium DM 318 1.16 ( 4.33) 1.66 ( 6.21) 1.64 ( 6.14) 1.30 ( 4.87)
Canada DM 336 1.12 ( 4.10) 1.74 ( 6.38) 1.75 ( 6.40) 1.65 ( 6.04)
Denmark DM 294 0.92 ( 3.47) 1.32 ( 4.98) 1.34 ( 5.04) 1.25 ( 4.71)
Finland DM 282 1.05 ( 3.03) 1.38 ( 3.98) 1.40 ( 4.04) 0.85 ( 2.46)
France DM 336 0.61 ( 2.37) 1.14 ( 4.45) 1.10 ( 4.31) 0.83 ( 3.24)
Germany DM 336 0.89 ( 3.46) 1.51 ( 5.85) 1.52 ( 5.89) 1.16 ( 4.52)
Hong Kong DM 270 0.88 ( 2.59) 1.67 ( 4.88) 1.73 ( 5.06) 2.09 ( 6.12)
Ireland DM 212 0.79 ( 1.13) 1.46 ( 2.01) 1.41 ( 1.94) 0.89 ( 1.21)
Israel DM 186 1.06 ( 2.44) 1.44 ( 3.30) 1.46 ( 3.34) 0.74 ( 1.69)
Italy DM 336 0.84 ( 3.20) 1.20 ( 4.57) 1.16 ( 4.42) 0.92 ( 3.52)
Japan DM 336 0.03 ( 0.13) 0.21 ( 0.86) 0.19 ( 0.78) 0.19 ( 0.79)
Netherlands DM 336 0.70 ( 2.16) 1.29 ( 4.01) 1.28 ( 3.97) 1.23 ( 3.83)
New Zealand DM 222 1.13 ( 5.12) 1.23 ( 5.61) 1.24 ( 5.63) 1.07 ( 4.88)
Norway DM 306 0.93 ( 2.94) 1.18 ( 3.74) 1.15 ( 3.64) 0.84 ( 2.66)
Portugal DM 272 1.57 ( 3.84) 1.84 ( 4.56) 1.79 ( 4.41) 1.55 ( 3.82)
Singapore DM 330 0.27 ( 0.81) 1.06 ( 3.20) 1.06 ( 3.19) 1.12 ( 3.37)
Spain DM 306 0.51 ( 1.83) 0.79 ( 2.82) 0.76 ( 2.70) 0.54 ( 1.94)
Sweden DM 306 0.90 ( 2.77) 1.59 ( 4.91) 1.58 ( 4.88) 1.52 ( 4.69)
Switzerland DM 336 0.70 ( 2.68) 1.13 ( 4.32) 1.10 ( 4.22) 0.90 ( 3.44)
U.K. DM 336 0.97 ( 4.32) 1.61 ( 7.15) 1.60 ( 7.09) 1.58 ( 7.04)
U.S. DM 1056 0.60 ( 4.40) 1.10 ( 8.14) 1.05 ( 7.76) 1.16 ( 8.57)
Argentina EM 195 0.72 ( 1.39) 0.80 ( 1.48) 0.72 ( 1.33) -0.69 (-1.35)
Brazil EM 210 0.30 ( 0.73) 0.62 ( 1.52) 0.66 ( 1.62) 0.73 ( 1.80)
Chile EM 258 0.83 ( 3.62) 0.96 ( 4.21) 1.00 ( 4.38) 0.94 ( 4.11)
China EM 222 -0.47 (-1.58) -0.18 (-0.62) -0.14 (-0.46) -0.05 (-0.18)
Colombia EM 150 0.35 ( 1.09) 0.46 ( 1.43) 0.38 ( 1.18) 0.01 ( 0.02)
Greece EM 270 0.67 ( 1.31) 1.36 ( 2.66) 1.35 ( 2.63) 1.78 ( 3.46)
Hungary EM 129 0.12 ( 0.28) 0.06 ( 0.14) 0.00 ( 0.00) 0.09 ( 0.20)
India EM 246 1.22 ( 2.51) 2.00 ( 4.10) 1.94 ( 3.98) 2.41 ( 4.95)
Indonesia EM 258 0.00 ( 0.00) 1.39 ( 2.27) 1.37 ( 2.23) 0.77 ( 1.25)
Malaysia EM 288 0.32 ( 0.79) 1.76 ( 4.32) 1.87 ( 4.59) 1.73 ( 4.25)
Mexico EM 246 0.63 ( 2.01) 0.95 ( 3.04) 0.97 ( 3.12) 1.22 ( 3.92)
Pakistan EM 246 1.01 ( 2.33) 1.48 ( 3.41) 1.61 ( 3.72) 1.37 ( 3.16)
Peru EM 158 0.02 ( 0.03) -0.03 (-0.05) 0.18 ( 0.33) -0.14 (-0.26)
Philippines EM 258 0.07 ( 0.12) 0.93 ( 1.53) 0.89 ( 1.47) 0.60 ( 0.99)
Poland EM 207 1.07 ( 3.16) 1.61 ( 4.73) 1.66 ( 4.88) 1.74 ( 5.11)
South Africa EM 270 1.46 ( 5.21) 1.82 ( 6.48) 1.82 ( 6.47) 1.82 ( 6.46)
South Korea EM 286 0.31 ( 0.75) 1.01 ( 2.47) 1.05 ( 2.56) -0.38 (-0.94)
Taiwan EM 234 0.30 ( 0.87) 0.79 ( 2.27) 0.81 ( 2.32) 1.24 ( 3.56)
Thailand EM 270 0.64 ( 1.31) 1.59 ( 3.26) 1.62 ( 3.33) 1.68 ( 3.45)
Turkey EM 246 0.06 ( 0.18) 0.35 ( 1.04) 0.32 ( 0.96) 0.79 ( 2.36)
38

Table 7: Panel Regressions of Momentum Returns and Cross-Country Characteristics
The table presents results from multivariate panel regressios of momentum strategy returns and country-month averages of firm-
level as well as other country characteristics. The dependent variable is the monthly return for each country of the following
momentum strategies: MOM, cMOM, sMOM, and dMOM. We estimate the following regression for each strategy stratgey:
Reti,t = Xi,t + controls + ei,t , where Xi,t is a vector of country characteristics. The country-level control variables are described
in the text and summarized in Table A.7. All regressions include month fixed effects, whereas the last four columns additionally
include country fixed effects. Standard errors are double-clustered by country and month. All covariates (except dummy
variables) are standardized to have a mean of zero and a variance of one. Corresponding t-statistics are reported in parantheses.
The analysis is performed at monthly frequency over the time-series from 01/1990 to 12/2017.
(1) (2) (3) (4) (5) (6) (7) (8)

MOM cMOM sMOM dMOM MOM cMOM sMOM dMOM
Market continuation dummy 1.86 1.74 1.71 0.70 1.87 1.75 1.72 0.70
(6.96) (7.09) (7.06) (3.63) (7.06) (7.16) (7.13) (3.61)
Firm size −0.32 −0.38 −0.39 −0.29 −0.37 −0.48 −0.49 −0.31
(−3.62) (−3.98) (−4.15) (−3.26) (−2.59) (−3.40) (−3.61) (−2.87)
Return R2 −0.32 −0.48 −0.51 −0.29 −0.31 −0.39 −0.42 −0.24
(−2.72) (−4.74) (−4.92) (−2.46) (−1.94) (−3.04) (−3.16) (−2.31)
39
Idiosyncratic volatility −0.44 −0.38 −0.39 −0.07 −0.74 −0.76 −0.78 −0.32
(−2.83) (−3.62) (−3.73) (−0.59) (−3.58) (−5.22) (−5.01) (−1.68)
Fraction zero return days −0.03 −0.09 −0.09 −0.09 0.02 −0.05 −0.05 −0.07
(−0.39) (−1.02) (−1.02) (−0.68) (0.11) (−0.26) (−0.23) (−0.29)
Number of analysts 0.09 0.14 0.14 0.12 0.03 −0.09 −0.11 −0.07
(1.43) (1.40) (1.25) (1.01) (0.25) (−0.71) (−0.74) (−0.31)
Analysts forecast dispersion −0.08 −0.06 −0.06 −0.09 −0.06 −0.05 −0.04 −0.05
(−1.02) (−1.13) (−1.05) (−1.26) (−0.77) (−0.86) (−0.77) (−0.67)
Short selling dummy −0.02 −0.19 −0.21 0.25
(−0.09) (−0.78) (−0.88) (0.82)
Developed market dummy −0.15 −0.26 −0.27 −0.19

(−0.63) (−1.13) (−1.15) (−0.66)
Individualism 0.01 0.01 0.01 0.005
(2.07) (1.75) (1.68) (0.82)
Month FE Yes Yes Yes Yes Yes Yes Yes Yes
Country FE No No No No Yes Yes Yes Yes
Observations 11,217 11,217 11,217 11,217 11,217 11,217 11,217 11,217
No. of Countries 43 43 43 43 43 43 43 43
Adjusted R2 0.25 0.19 0.18 0.06 0.25 0.20 0.19 0.07
Table 8: Panel Regressions of Momentum Returns and Market Continuation Dummies for Different Market
States
The table presents results from multivariate panel regressios of momentum strategy returns and two market continuation
dummies after bull or bear markets. The dependent variable is the monthly return for each country of the following momentum
strategies: MOM, cMOM, sMOM and dMOM. We estimate the following regression: Reti,t = Xi,t + controls + ei,t , where Xi,t is
a market continuation dummy during bear or bull markets, respectively. The country-level control variables are described in the
paper and summarized in Table A.7. All regressions include month fixed effects, country fixed effects and a dummy controlling
for being in the opposite market state as the market continuation dummy.,i.e., in a bull or bear market, respectively. Standard
errors are double-clustered by country and month. All covariates (except dummy variables) are standardized to have a mean
of zero and a variance of one. Corresponding t-statistics are reported in parantheses. The analysis is performed at monthly
frequency over the time-series from 01/1990 to 12/2017.
40
(1) (2) (3) (4) (5) (6) (7) (8)
Market Cont. Dummy (Bear Market) 3.73 3.03 2.96 0.62
(7.50) (7.54) (7.55) (1.34)
Market Cont. Dummy (Bull Market) 1.18 1.33 1.31 0.82
(5.22) (5.42) (5.28) (4.51)
Controls Yes Yes Yes Yes Yes Yes Yes Yes
Country FE Yes Yes Yes Yes Yes Yes Yes Yes

No. of Countries 43 43 43 43 43 43 43 43
Observations 11,217 11,217 11,217 11,217 11,217 11,217 11,217 11,217
Table 9: Turnover and Break-even Round-trip Costs
The table presents the following turnover and trading cost measures for MOM, cMOM,
sMOM, and dMOM: (1) average long-short portfolio turnover (monthly, in %), (2) break-
even round-trip costs significant at the 5% level, stating the upper border for trading costs
so that the strategy is profitable with 5% significance, and (3) break-even round-trip costs
significant at the 1% level. For details regarding the measure construction, see Section 2.4.
The analysis is performed from 01/1930 (01/1990) to 12/2017 for the U.S. (non-U.S.) sample.
MOM cMOM sMOM dMOM

Panel A: U.S. (01/1930 - 12/2017)
Turnover (in %) 55.34 82.88 81.46 85.43
Round-trip costs at 5% sign. level (in %) 0.6 1.01 0.97 1.05
Panel B: Non-U.S. (01/1990 - 12/2017)
Turnover (in %) 49.26 70.77 70.26 74.24
Table 10: Momentum strategies in January versus Non-January months

The table presents the mean returns (in %) and corresponding t-statistics of momentum
and the enhanced momentum strategies in January and Non-January months. The following
strategies are included: MOM, cMOM, sMOM, and dMOM. For details regarding variable
construction, see Section 2.2. The results relate to January (Jan.) months, Non-January
(Non-Jan.) months and the difference between Non-January and January months. The
analyses are performed from 01/1930 (01/1990) to 12/2017 for the U.S. (non-U.S.) sample.
Panel A: U.S. (01/1930 - 12/2017)

MOM cMOM sMOM dMOM
Jan. -1.42 -1.16 -1.13 -0.68

(-3.00) (-2.11) (-1.98) (-1.21)
Non-Jan. 0.78 1.31 1.25 1.33
( 5.56) ( 9.53) ( 9.14) ( 9.66)
Diff. 2.20 2.47 2.38 2.01
( 4.45) ( 4.36) ( 4.06) ( 3.47)
Panel B: Non-U.S. (01/1990 - 12/2017)

MOM cMOM sMOM dMOM
Jan. -0.16 0.31 0.27 1.06

(-0.22) ( 0.35) ( 0.33) ( 1.08)
Non-Jan. 0.61 1.07 1.02 0.96
( 3.04) ( 5.47) ( 5.17) ( 5.00)
Diff. 0.77 0.76 0.75 -0.10
( 1.02) ( 0.84) ( 0.90) (-0.10)
41

Figure 1: Cumulative performance of the momentum strategies: U.S.
This figure displays the cumulated performance of a $1 investment in each of the momentum
strategies (plus the risk-free rate since, all momentum portfolio state zero-cost strategies)
for the U.S. (Long) sample. The following strategies are comprised: MOM, cMOM, sMOM,
and dMOM. cMOM is momentum scaled to a constant volatility level by its six month past
volatility. sMOM is momentum scaled to a constant volatility level by its six month past
semi-volatility. dMOM is momentum scaled by its forecasted return relative to its expected
variance. For details regarding variable construction, see Section 2.2. The sample period
ranges from 01/1930 to 12/2017.
1,000,000
10,000
Cum. Performance
100
1940 1960 1980 2000 2020

Year
variable MOM cMOM sMOM dMOM
42

Figure 2: Cumulative performance of the momentum strategies: Non-U.S.
This figure displays the cumulated performance of a $1 investment in each of the momentum
strategies (plus the risk-free rate since, all momentum portfolio state zero-cost strategies) for
the Global (Broad) sample. The following strategies are comprised: MOM, cMOM, sMOM,
and dMOM. cMOM is momentum scaled to a constant volatility level by its six month past
volatility. sMOM is momentum scaled to a constant volatility level by its six month past
semi-volatility. dMOM is momentum scaled by its forecasted return relative to its expected
variance. For details regarding variable construction, see Section 2.2. The sample period
ranges from 01/1990 to 12/2017.
50
20
Cum. Performance
10
1990 2000 2010

Year
variable MOM cMOM sMOM dMOM
43

Figure 3: Maximum drawdown per country
The figure displays the maximum drawdown of momentum and the enhanced strategies for
the 49 countries in our sample. The following strategies are included: MOM, cMOM, sMOM,
and dMOM. For details regarding variable construction, see Section 2.2. The analyses are
performed at monthly frequency from 01/1930 (01/1990) to 12/2017 for the U.S. (non-U.S.
counries). For a country to be regarded, a minimum of 120 observations are required.
0
Maximum Drawdown (%)
Strategy
−100 WML
cMOM
sMOM
dMOM
−200
New Zealand
South Korea
South Africa
Netherlands
Switzerland
Hong Kong
Philippines
Singapore
Indonesia
Argentina
Colombia
Germany
Denmark
Malaysia
Australia
Thailand
Hungary
Pakistan
Portugal
Belgium
Sweden
Canada
Norway
Finland
Greece
Mexico
France
Austria
Taiwan
Poland
Ireland
Turkey
Japan
China
Spain
Brazil
Israel
Chile
India
Peru
U.K.
U.S.
Italy
44

A Appendix
A.1 Datastream sample definition
Constituent lists
Datastream comprises three types of constituent lists: (1) research lists, (2) Worldscope lists,
and (3) dead lists. By using dead lists, we ensure to obviate any survivorship bias. For every
country we use the union of all available lists and eliminate any duplicates. As a result, we
have one remaining list for every country, which can subsequently be used in the static filter
process. Table A.1 and Table A.2 provide an overview of the constituent lists for developed
markets and emerging markets, respectively, used in our study.
[Table A.1 about here.]
Static screens
We restrict our sample to common equity stocks by applying several static screens, as shown
in Table A.3. Screen (1) to (7) are standard filters as common in the literature.
Screen (8) related to, among others, the following work: Ince and Porter (2006), Campbell,
Cowan, and Salotti (2010), Griffin et al. (2010), Karolyi, Lee, and van Dijk (2012). The
authors provide generic filter rules in order to exclude non-common equity securities from
Thomson Reuters Datastream. We apply the identified keywords and match them with the
security names provided by Datastream. A security is excluded from the sample in case a
keyword coincides with part of the security name. The following three Datastream items
store security names and are applied for the keyword filters: “NAME”, “ENAME”, and
“ECNAME”. Table A.4 gives an overview of the keywords used.
45

In addition, Griffin et al. (2010) introduce specific keywords for individual countries.
Thus, the keywords are applied on the security names of single countries only. Exemplary,
German security names are parsed to contain the word “GENUSSSCHEINE”, which declares
the security to be non-common equity. In Table A.5 we give an overview of country-specific
keyword deletions conducted in our study.
Dynamic screens
For the securities, remaining from the static screens above, we obtain return and market
capitalization data from Datastream and accounting data from Worldscope. Several dynamic
screens that are common in the literature were installed in order to account for data errors
mainly within return characteristics. The dynamic screens are shown in Table A.6.
A.2 Turnover calculation
Specifically, (one-way portfolio) turnover in month t for both the long or short portfolio leg
are calculated as:
Nt
T urnovert,Long(Short) = 0.5 × (11)
X
|xi,t − x̃i,t−1 |
i
where xi,t is the weight of stock i in the respective portfolio leg in month t (i.e., the value
proportion since we use value-weighted portfolio returns), Nt amounts to the total number
of stocks in the portfolio leg at month t, and ri,t is the return of stock i during month t, and
˜ is the weight at the end of month t − 1 resp. at the beginning of month t, right before
xi,t−1
46

trading. We define x̃i,t−1 as:
xi,t−1 (1 + ri,t−1 )
x̃i,t−1 = Nt
(12)
xi,t−1 (1 + ri,t−1 )
P
i
The turnover of the long-short momentum strategies is then the sum of the average turnover
in the long and short legs, i.e., the sum of T urnoverLong and T urnoverShort . For the volatility-
scaled strategies, the turnover is derived from Equation 11 by weighting the turnover in month
t with the corresponding strategy weight:
Nt
T urnovers,t,Long/Short = 0.5 × (13)
X
|wscaled,t xi,t − wscaled,t−1 x̃i,t−1 |
i
47

Table A.1: Constituent lists: Developed markets
The table contains the Research lists, Worldscope lists, and Dead lists of developed markets
countries in our sample.
Country Lists Country Lists

Australia DEADAU Israel DEADIS
FAUS FISRAEL
WSCOPEAU WSCOPEIS
Austria DEADOE Italy DEADIT
FOST FITA
WSCOPEOE WSCOPEIT
Belgium DEADBG Japan DEADJP
FBEL FFUKUOKA
FBELAM FJASDAQ
FBELCM FOSAKA
WSCOPEBG FTOKYO
Canada DEADCN1 JAPOTC
DEADCN2 WSCOPEJP
DEADCN3 Netherlands DEADNL
DEADCN4 FHOL
DEADCN5 WSCOPENL
DEADCN6 New Zealand DEADNZ
FTORO FNWZ
FVANC WSCOPENZ
LTTOCOMP Norway DEADNW
WSCOPECN FNOR
Denmark DEADDK WSCOPENW
FDEN Portugal DEADPT
WSCOPEDK FPOR
Finland DEADFN WSCOPEPT
FFIN Singapore DEADSG
WSCOPEFN FSIN
France DEADFR FSINQ
FFRA WSCOPESG
WSCOPEFR Spain DEADES
Germany DEADBD1 FSPN
DEADBD2 WSCOPEES
DEADBD3 Sweden DEADSD
DEADBD4 FAKTSWD
DEADBD5 FSWD
DEADBD6 WSCOPESD
FGER1 Switzerland DEADSW
FGER2 FSWA
FGERIBIS FSWS
FGKURS FSWUP
WSCOPEBD WSCOPESW
Hong Kong DEADHK U.K. DEADUK
FHKQ FBRIT
WSCOPEHK LSETSCOS
Ireland DEADIR LSETSMM
FIRL LUKPLUSM
WSCOPEIR WSCOPEJE
WSCOPEUK
48

Table A.2: Constituent lists: Emerging markets
The table contains the Research lists, Worldscope lists, and Dead lists of emerging markets
countries in our sample.
Argentina DEADAR Morocco DEADMOR

FPARGA FMOR
WSCOPEAR WSCOPEMC
Brazil DEADBRA Pakistan DEADPA
FBRA FPAK
WSCOPEBR FPAKUP
Chile DEADCHI WSCOPEPK
FCHILE Peru DEADPE
FCHILE10 FPERU
WSCOPECL WSCOPEPE
China DEADCH Philippines DEADPH
FCHINA FPHI
WSCOPECH FPHILA
Colombia DEADCO FPHIMN
FCOL FPHIQ
WSCOPECB WSCOPEPH
Czech Republic DEADCZ Poland DEADPO
FCZECH FPOL
FCZECHUP WSCOPEPO
WSCOPECZ Qatar DEADQT
Egypt DEADEGY FQATAR
EGYPTALL WSCOPEQA
FEGYPT Russia DEADRU
WSCOPEEY FRTSCL
Greece DEADGR FRUS
FGREE FRUSUP
FGRMM WSCOPERS
FGRPM South Africa DEADSAF
FNEXA DEADSAF
WSCOPEGR FSAF
Hungary DEADHU WSCOPESA
FHUN WSCOPESA
WSCOPEHN South Korea DEADKO
India DEADIND FKONEX
FBSE FKOR
FINDIA WSCOPEKO
FINDNW Taiwan DEADTW
FINDUP FTAIQ
FNSE WSCOPETA
WSCOPEIN Thailand DEADTH
Indonesia DEADIDN FTHAQ
FINO WSCOPETH
WSCOPEID Turkey DEADTK
Malaysia DEADMY FTURK
FMAL FTURKUP
FMALQ WSCOPETK
WSCOPEMY United Arab Emirates DEADAB
Mexico DEADME DEADDB
FMEX FABUD
MEX101 FDUBAI
WSCOPEMX WSCOPEAE
49

Table A.3: Static Screens
The table displays the static screens applied in our study, mainly following Ince and Porter
(2006), Schmidt et al. (2017), and Griffin et al. (2010). Column 3 lists the Datastream items
involved (on the left of the equality sign) and the values which we set them to in the filter
process (on the right of the equality sign). Column 4 indicates the source of the screens.
Nr. Description Datastream item(s) Source

involved
(1) For firms with more than one MAJOR = Y Schmidt et al. (2017)
security, only the one with the
biggest market capitalization and
liquidity is used.
(2) The type of security must be eq- TYPE = EQ Ince and Porter (2006)
uity.
(3) Only the primary quotations of a ISINID = P Fong, Holden, and
security are analyzed. Trzcinka (2017)
(4) Firms are located in the respec- GEOGN = country Ince and Porter (2006)
tive domestic country. shortcut
(5) Securities are listed in the respec- GEOLN = country Griffin et al. (2010)
tive domestic country. shortcut
(6) Securities with quoted currency PCUR = currency Griffin et al. (2010)
different from the one of the as- shortcut of the coun-
sociated country are disregarded.a try
(7) Securities with ISIN country code GGISN = country Annaert, Ceuster, and
different from the one of the asso- shortcut Verstegen (2013)
ciated country are disregarded.b
(8) Securities whose name fields indi- NAME, ENAME, Ince and Porter
cate non-common stock affiliation ECNAME (2006), Campbell
are disregarded. et al. (2010), Griffin
et al. (2010) and
Karolyi et al. (2012)
a
In this filter rule, also the respective pre-euro currencies are accepted for countries
within the eurozone. Moreover, in Russia, “USD” is also accepted as currency, besides
“RUB”.
b
In Hong Kong, ISIN country codes equal to “BM” or “KY” and in the Czech Republic
ISIN country codes equal to “CS” are also accepted.
50

Table A.4: Generic Keyword Deletions
The table reports the generic keywords, which are searched for in the names of all stocks of
all countries. If the stock name contains a harmful keyword, the respective stock is removed
from the sample.
Non-common equity Keywords

Duplicates 1000DUPL, DULP, DUP, DUPE, DUPL, DUPLI,
DUPLICATE, XSQ, XETa
Depository Receipts ADR, GDR
Preferred Stock PF, ’PF’, PFD, PREF, PREFERRED, PRF
Warrants WARR, WARRANT, WARRANTS, WARRT, WTS, WTS2
Debt %, DB, DCB, DEB, DEBENTURE, DEBENTURES, DEBT
Unit Trusts .IT, .ITb, TST, INVESTMENT TRUST, RLST IT, TRUST,
TRUST UNIT, TRUST UNITS, TST, TST UNIT, TST
UNITS, UNIT, UNIT TRUST, UNITS, UNT, UNT TST, UT
ETFs AMUNDI, ETF, INAV, ISHARES, JUNGE, LYXOR, X-TR
Expired securities EXPD, EXPIRED, EXPIRY, EXPY
Miscellaneous (mainly taken from ADS, BOND, CAP.SHS, CONV, DEFER, DEP, DEPY,
Ince and Porter (2006)) ELKS, FD, FUND, GW.FD, HI.YIELD, HIGH INCOME,
IDX, INC.&GROWTH, INC.&GW, INDEX, LP, MIPS,
MITS, MITT, MPS, NIKKEI, NOTE, OPCVM, ORTF,
PARTNER, PERQS, PFC, PFCL, PINES, PRTF, PTNS,
PTSHP, QUIBS, QUIDS, RATE, RCPTS, REAL EST,
RECEIPTS, REIT, RESPT, RETUR, RIGHTS, RST,
RTN.INC, RTS, SBVTG, SCORE, SPDR, STRYPES,
TOPRS, UTS, VCT, VTG.SAS, XXXXX, YIELD, YLD
51

Table A.5: Country-Specific Keyword Deletions
The table reports the country-specific keywords, which are searched for in the names of
all stocks of the respective countries. If the stock name contains a harmful keyword, the
respective stock is removed from the sample.
Country Keywords
Australia PART PAID, RTS DEF, DEF SETT, CDI
Austria PC, PARTICIPATION CERTIFICATE, GENUSSSCHEINE,
GENUSSCHEINE
Belgium VVPR, CONVERSION, STRIP
Brazil PN, PNA, PNB, PNC, PND, PNE, PNF, PNG, RCSA, RCTB
Canada EXCHANGEABLE, SPLIT, SPLITSHARE, VTG\\.,
SBVTG\\., VOTING, SUB VTG, SERIES
Denmark \\)CSE\\)
Finland USE
France ADP, CI, SICAV, \\)SICAV\\), SICAV-
Germany GENUSSCHEINE
Greece PR
India FB DEAD, FOREIGN BOARD
Israel P1, 1, 5
Italy RNC, RP, PRIVILEGIES
Korea 1P
Mexico ’L’, ’C’
Malaysia ’A’
Netherlands CERTIFICATE, CERTIFICATES, CERTIFICATES\\),
CERT, CERTS, STK\\.
New Zealand RTS, RIGHTS
Peru INVERSION, INVN, INV
Philippines PDR
South Africa N’, OPTS\\., CPF\\., CUMULATIVE PREFERENCE
Sweden CONVERTED INTO, USE, CONVERTED-,
CONVERTED - SEE
Switzerland CONVERTED INTO, CONVERSION, CONVERSION SEE
United Kingdom PAID, CONVERSION TO, NON VOTING,
CONVERSION ’A’
52

Table A.6: Dynamic Screens
The table displays the dynamic screens applied to the data in our study, following Ince and
Porter (2006), Griffin et al. (2010), Jacobs (2016), and Schmidt et al. (2017). If screens are
adapted solely to monthly (daily) returns, this is indicated by m (d).
No. Description Reference

(1) We delete the zero returns at the end of the return time- Ince and Porter (2006)
series, which exist, because in case of a delisting Datas-
tream displays stale prices from the date of delisting un-
til the end of the respective time-series. We also delete
the associated market capitalizations.
(2) We delete the associated returns and market capitaliza- The screen originally
tions in case of abnormal prices (unadjusted prices > stems from Schmidt
1000000). et al. (2017), whereby
we employ it on the
unadjusted price.
(3m)We delete monthly returns and the associated market Schmidt et al. (2017)
capitalizations in case of return spikes (returns > 990%).
(3d) We delete daily returns and the associated market cap- Griffin et al. (2010)
italizations in case of return spikes (returns > 200%).
(4m)We delete monthly returns and the associated market Ince and Porter (2006)
capitalizations in case of strong return reversals, defined
as follows: Rt−1 or Rt >= 3.0 and (1+Rt−1 )(1+Rt )−1 <
0.5.
(4d) We delete daily returns and the associated market cap- Ince and Porter
italizations in case of strong return reversals, defined as (2006), Griffin et al.
follows: Rt−1 or Rt >= 1.0 and (1 + Rt−1 )(1 + Rt ) − 1 < (2010), Jacobs (2016)
0.2.
53

Table A.7: Descriptive statistics: Cross-country Variables
The table presents summary statistics for 40 countries of our CRSP, Datastream and Worldscope sample. Market states the market affiliation, according
to MSCI, with DM as Developed Markets and EM as Emerging Markets. The last column states the individualism index reported by Hofstede (2001)
for each country. The other columns report time-series averages for the following cross-country variables: Mkt. cont. dummy is a dummy that equals
one if the past 12 month and the current month market return have the same sign, Firm size is the mean market size of all firms in million USD, R2
is the coefficient of determination from the regressions to quantify idiosyncratic volatility, IVOL is the mean firm-level idiosyncratic volatility from
rolling regressions of daily excess returns on a local FF3F model over the previous 12 months, Frac. zero return days is the average fraction of firms
with zero local currency returns, No. analysts is the mean number of analysts that provide firm-level fiscal-year-one earnings estimates, Anal. forecast
disp. is the average standard deviation of analyst earnings forecasts divided by the mean forecast, and Short dummy is a dummy that equals one if
short selling is allowed. We require that countries have non-missing returns for momentum and all enhanced momentum strategies in at least 120
months.
Country Market Mkt. cont. dummy Firm size R2 (%) IVOL Frac. zero return days (%) No. analysts Anal. forecast disp. Short dummy Individualism
Australia DM 0.58 761.56 14.36 3.93 44.47 3.31 28.99 1.00 90
Austria DM 0.55 1023.78 23.86 2.06 38.31 3.78 33.08 1.00 55
Belgium DM 0.58 1829.31 20.28 2.00 30.44 4.23 31.11 1.00 75
Canada DM 0.53 757.41 8.12 4.83 36.53 3.38 47.13 1.00 80
Denmark DM 0.59 944.76 15.41 2.34 44.61 2.87 44.16 1.00 74
Finland DM 0.57 1614.11 18.78 2.32 30.05 6.10 30.20 0.85 63
France DM 0.57 1896.18 14.08 2.62 32.68 4.04 37.07 1.00 71
Germany DM 0.56 1640.07 15.69 2.60 35.62 4.33 44.21 1.00 67
Hong Kong DM 0.61 540.70 10.74 3.60 35.23 2.71 29.88 0.98 25
Ireland DM 0.59 1155.51 16.48 3.21 49.87 3.98 17.89 1.00 70
Israel DM 0.52 481.37 20.98 2.31 30.59 0.35 34.60 1.00 54
Italy DM 0.51 1957.96 25.07 2.03 18.39 5.29 37.15 1.00 76
Japan DM 0.55 1184.32 22.30 2.35 26.37 1.95 30.79 1.00 46
Netherlands DM 0.55 2727.53 22.70 2.14 22.94 9.39 35.74 1.00 80
54
New Zealand DM 0.57 381.43 21.54 2.66 49.63 2.43 15.49 1.00 79
Norway DM 0.56 942.30 16.63 2.93 40.02 4.03 52.16 1.00 69
Portugal DM 0.56 1118.58 21.42 2.61 42.09 3.78 32.28 1.00 27
Singapore DM 0.56 598.20 16.78 3.23 45.58 5.26 26.31 0.00 20
Spain DM 0.53 3550.06 25.75 1.87 29.07 8.99 32.67 1.00 51
Sweden DM 0.56 1001.23 18.58 3.30 26.69 3.78 46.46 1.00 71
Switzerland DM 0.57 3654.92 21.09 1.92 34.53 5.80 28.77 1.00 68
U.K. DM 0.57 1626.10 15.36 2.46 51.41 3.97 25.20 1.00 89
U.S. DM 0.63 2907.11 13.69 3.48 12.44 5.23 3.24 1.00 91
Argentina EM 0.57 452.90 22.55 2.34 50.68 0.74 60.71 1.00 46

Brazil EM 0.54 2574.40 29.66 2.79 44.48 2.95 52.23 1.00 38
Chile EM 0.52 917.97 26.92 1.54 66.10 1.05 24.45 0.88 23
China EM 0.58 1363.22 41.26 2.05 13.97 1.52 22.00 0.00 20

Colombia EM 0.58 2869.86 39.37 1.46 61.32 0.61 22.71 0.00 13
Greece EM 0.57 343.61 26.54 2.92 28.99 2.01 42.36 0.00 35
Hungary EM 0.55 642.91 27.31 3.10 37.62 1.79 36.79 1.00 80
India EM 0.54 540.14 19.89 3.10 15.93 1.96 24.33 0.54 48
Indonesia EM 0.56 521.03 17.65 3.67 55.62 2.27 53.71 0.00 14
Malaysia EM 0.57 379.70 22.05 2.96 36.36 3.07 35.08 0.08 26
Mexico EM 0.56 1869.09 30.00 1.90 49.61 3.71 48.98 1.00 30
Pakistan EM 0.52 150.35 18.63 2.78 46.13 0.57 24.58 0.00 14
Peru EM 0.58 616.01 19.31 1.49 80.66 0.21 24.98 0.00 16
Philippines EM 0.55 492.54 12.17 3.47 56.49 2.31 32.40 0.93 32
Poland EM 0.56 405.77 20.79 3.04 23.99 1.28 45.67 1.00 60
South Africa EM 0.51 1090.83 21.65 3.61 42.63 2.50 28.43 1.00 65
South Korea EM 0.55 485.96 26.49 3.04 13.55 2.40 47.49 0.00 18
Taiwan EM 0.56 551.91 29.03 2.18 18.21 1.56 37.37 1.00 17
Thailand EM 0.55 388.87 17.59 2.80 41.07 2.48 46.95 0.93 20
Turkey EM 0.48 586.83 43.46 2.60 20.42 3.25 38.83 1.00 37
Table A.8: Panel Regressions of Momentum Returns and Cross-Country Characteristics
momentum strategies: MOM, cMOM, sMOM, and dMOM. For columns 2-4 and 6-8, we estimate the following regression
pairwise for MOM and one enhanced momentum stratgey: Reti,t = Xi,t + Xi,t × EnhM om + controls + ei,t , where Xi,t is a
vector of country characteristics, EnhM om is a dummy equal to one if the return is from a enhanced momentum strategy. The
country-level control variables are described in the text and summarized in Table A.7. All regressions include month fixed effects,
columns 2-4 and 6-8 contain month-EnhMom fixed effects, and the last four columns additionally include country fixed effects.
Standard errors are double-clustered by country and month. All covariates (except dummy variables) are standardized to have
a mean of zero and a variance of one. Corresponding t-statistics are reported in parantheses. The analysis is performed at
monthly frequency over the time-series from 01/1990 to 12/2017.
(1) (2) (3) (4) (5) (6) (7) (8)

Market continuation dummy 1.86 −0.12 −0.15 −1.15 1.87 −0.12 −0.15 −1.16
(6.96) (−1.50) (−1.78) (−5.11) (7.06) (−1.50) (−1.77) (−5.14)
Firm size −0.32 −0.06 −0.07 0.02 −0.37 −0.11 −0.13 −0.03
(−3.62) (−1.15) (−1.30) (0.25) (−2.59) (−2.02) (−2.21) (−0.25)
Return R2 −0.32 −0.16 −0.19 0.03 −0.31 −0.10 −0.12 0.08
55
(−2.72) (−2.58) (−2.81) (0.24) (−1.94) (−1.56) (−1.77) (0.78)
Idiosyncratic volatility −0.44 0.06 0.04 0.36 −0.74 0.05 0.03 0.32
(−2.83) (0.73) (0.52) (2.65) (−3.58) (0.57) (0.36) (2.44)
(−0.39) (−1.36) (−1.34) (−0.50) (0.11) (−1.01) (−0.94) (−0.28)
Number of analysts 0.09 0.05 0.04 0.03 0.03 0.01 0.0005 0.005
(1.43) (0.68) (0.51) (0.23) (0.25) (0.19) (0.01) (0.04)
Analysts forecast dispersion −0.08 0.01 0.02 −0.02 −0.06 0.01 0.02 −0.01
(−1.02) (0.32) (0.36) (−0.17) (−0.77) (0.29) (0.33) (−0.06)
Short selling dummy −0.02 −0.17 −0.19 0.27
(−0.09) (−1.17) (−1.25) (0.92)

Developed market dummy −0.15 −0.11 −0.12 −0.05
(−0.63) (−1.01) (−1.06) (−0.16)
Individualism 0.01 −0.003 −0.003 −0.01
(2.07) (−0.85) (−0.86) (−1.46)
Month-EnhMom FE No Yes Yes Yes No Yes Yes Yes
Country FE No No No No Yes Yes Yes Yes
Observations 11,217 22,434 22,434 22,434 11,217 22,434 22,434 22,434
No. of Countries 43 43 43 43 43 43 43 43
Adjusted R2 0.25 0.22 0.22 0.16 0.25 0.23 0.22 0.16
Table A.9: Panel Regressions of Momentum Returns and Cross-Country Characteristics (Sub-Periods)
momentum strategies: MOM, cMOM, sMOM, and dMOM. We estimate the following regression for each strategy stratgey:
Reti,t = Xi,t + controls + ei,t , where Xi,t is a vector of country characteristics. The country-level control variables are described
in the text and summarized in Table A.7. All regressions include month fixed effects. Standard errors are double-clustered by
country and month. All covariates (except dummy variables) are standardized to have a mean of zero and a variance of one.
Corresponding t-statistics are reported in parantheses. The analysis is performed at monthly frequency over the time-series from
01/1990 to 12/2003 and from 01/2004 to 12/2017.
01/1990 to 12/2003 01/2004 to 12/2017
(1) (2) (3) (4) (5) (6) (7) (8)

Market continuation dummy 2.82 2.41 2.37 0.99 1.08 1.18 1.16 0.46
(5.96) (6.09) (6.01) (2.72) (5.61) (5.93) (5.97) (2.62)
Firm size −0.54 −0.41 −0.35 −0.19 −0.30 −0.36 −0.38 −0.23
(−2.06) (−1.75) (−1.34) (−0.92) (−3.07) (−3.70) (−3.86) (−2.08)
Return R2 −0.62 −0.46 −0.47 −0.16 −0.10 −0.45 −0.47 −0.33
56
(−3.04) (−3.35) (−3.29) (−1.20) (−0.87) (−3.22) (−3.33) (−1.89)
Idiosyncratic volatility −0.78 −0.55 −0.55 −0.24 −0.20 −0.19 −0.20 0.11
(−2.60) (−3.28) (−3.15) (−1.16) (−2.36) (−1.85) (−2.09) (0.94)
(−2.94) (−1.38) (−1.14) (−0.12) (1.49) (−0.67) (−0.71) (−0.97)
Number of analysts 0.11 0.14 0.13 0.16 0.25 0.35 0.34 0.16
(1.57) (1.95) (1.77) (1.48) (1.31) (1.84) (1.77) (0.72)
Analysts forecast dispersion −0.02 −0.01 −0.001 0.003 −0.07 −0.11 −0.12 −0.28
(−0.23) (−0.20) (−0.01) (0.04) (−0.71) (−1.23) (−1.30) (−2.44)
Short selling dummy −0.15 −0.38 −0.40 0.01 0.20 −0.02 −0.06 0.44
(−0.38) (−1.08) (−1.18) (0.02) (1.08) (−0.08) (−0.20) (1.09)

Developed market dummy −1.09 −0.94 −0.94 −0.93 0.25 −0.01 −0.03 0.13
(−3.41) (−3.70) (−3.50) (−3.28) (1.25) (−0.05) (−0.10) (0.32)
Individualism 0.03 0.03 0.03 0.03 −0.001 −0.01 −0.01 −0.01
(4.13) (4.56) (4.61) (4.00) (−0.11) (−1.24) (−1.35) (−1.84)
Country FE No No No No No No No No
Observations 4,527 4,527 4,527 4,527 6,690 6,690 6,690 6,690
No. of Countries 41 41 41 41 43 43 43 43
Adjusted R2 0.21 0.19 0.19 0.05 0.30 0.20 0.19 0.08
References
Annaert, J., Ceuster, M. D., Verstegen, K., 2013. Are extreme returns priced in the stock
market? European evidence. Journal of Banking & Finance 37, 3401–3411.
Asem, E., Tian, G. Y., 2010. Market dynamics and momentum profits. Journal of Financial
and Quantitative Analysis 45, 1549–1562.
Asness, C. S., Frazzini, A., 2013. The devil in HML’s details. Journal of Portfolio Management
39, 49–68.
Asness, C. S., Liew, J. M., Stevens, R. L., 1997. Parallels between the cross-sectional pre-
dictability of stock and country returns. Journal of Portfolio Management 23, 79–87.
Asness, C. S., Moskowitz, T. J., Pedersen, L. H., 2013. Value and momentum everywhere.
Journal of Finance 68, 929–985.
Barillas, F., Kan, R., Robotti, C., Shanken, J., 2019. Model comparison with Sharpe ratios.
Journal of Financial and Quantitative Analysis p. 1–35.
Barillas, F., Shanken, J., 2017. Which Alpha? Review of Financial Studies 30, 1316–1338.
Barroso, P., Santa-Clara, P., 2015. Momentum has its moments. Journal of Financial Eco-
nomics 116, 111–120.
Bekaert, G., Wu, G., 2000. Asymmetric volatility and risk in equity markets. Review of
Financial Studies 13, 1–42.
Blitz, D. C., Huij, J., Martens, M., 2011. Residual momentum. Journal of Empirical Finance
18, 506–521.
Bollerslev, T., 1987. A conditionally heteroskedastic time series model for speculative prices
and rates of return. Review of Economics and Statistics 69, 542–547.
57

Campbell, C. J., Cowan, A. R., Salotti, V., 2010. Multi-country event-study methods. Journal
of Banking & Finance 34, 3078–3090.
Carhart, M. M., 1997. On persistence in mutual fund performance. Journal of Finance 52,
57–82.
Cederburg, S., O’Doherty, M. S., Wang, F., Yan, X. S., 2020. On the performance of volatility-
managed portfolios. Journal of Financial Economics .
Chan, K., Hameed, A., Tong, W., 2000. Profitability of momentum strategies in the interna-
tional equity markets. Journal of Financial and Quantitative Analysis 35, 153–172.
Chui, A. C. W., Titman, S., Wei, K. C. J., 2010. Individualism and momentum around the
world. Journal of Finance 65, 361–392.
Daniel, K., Hirshleifer, D., Subrahmanyam, A., 1998. Investor psychology and security market
under- and overreactions. Journal of Finance 53, 1839–1885.
Daniel, K., Moskowitz, T. J., 2016. Momentum crashes. Journal of Financial Economics 122,
221–247.
Docherty, P., Hurst, G., 2018. Investor myopia and the momentum premium across interna-
tional equity markets. Journal of Financial and Quantitative Analysis 53, 2465–2490.
du Plessis, J., Hallerbach, W. G., 2016. Volatility weighting applied to momentum strategies.
Journal of Alternative Investments 19, 40–58.
Engle, R. F., 1982. Autoregressive conditional heteroscedasticity with estimates of the vari-
ance of United Kingdom inflation. Econometrica 50, 987–1007.
Erb, C. B., Harvey, C. R., 2006. The strategic and tactical value of commodity futures.
Financial Analysts Journal 62, 69–97.
58

Fama, E. F., French, K. R., 1993. Common risk factors in the returns on stocks and bonds.
Journal of Financial Economics 33, 3–56.
Fama, E. F., French, K. R., 2012. Size, value, and momentum in international stock returns.
Fama, E. F., French, K. R., 2015. A five-factor asset pricing model. Journal of Financial
Economics 116, 1 – 22.
Fama, E. F., French, K. R., 2017. International tests of a five-factor asset pricing model.
Fong, K. Y. L., Holden, C. W., Trzcinka, C. A., 2017. What are the best liquidity proxies
for global research? Review of Finance 21, 1355–1401.
Frazzini, A., Israel, R., Moskowitz, T. J., 2014. Trading costs of asset pricing anomalies.
SSRN Working Paper no. 2294498 .
Griffin, J. M., Ji, X., Martin, J. S., 2003. Momentum investing and business cycle risk:
Evidence from pole to pole. Journal of Finance 58, 2515–2547.
Griffin, J. M., Kelly, P. J., Nardari, F., 2010. Do market efficiency measures yield correct
inferences? A comparison of developed and emerging markets. Review of Financial Studies
23, 3225–3277.
Grobys, K., Ruotsalainen, J., Äijö, J., 2018. Risk-managed industry momentum and momen-
tum crashes. Quantitative Finance 7688.
Grundy, B. D., Martin, J. S., 2001. Understanding the nature of the risks and the source of
the rewards to momentum investing. Review of Financial Studies 14, 29–78.
Hanauer, M. X., 2014. Is Japan different? Evidence on momentum and market dynamics.
International Review of Finance 14, 141–160.
59

Hanauer, M. X., 2020. A comparison of global factor models. SSRN Working Paper no.
3546295 .
Harvey, C. R., 2017. Presidential address: The scientific outlook in financial economics.
Harvey, C. R., Hoyle, E., Korgaonkar, R., Rattray, S., Sargaison, M., Van Hemert, O., 2018.
The impact of volatility targeting. Journal of Portfolio Management 45, 14–33.
Harvey, C. R., Liu, Y., Zhu, H., 2016. . . . and the cross-section of expected returns. Review
of Financial Studies 29, 5–68.
Hou, K., Peng, L., Xiong, W., 2013. Is r2 a measure of market inefficiency? Unpublished
working paper .
Hou, K., Xue, C., Zhang, L., 2020. Replicating anomalies. Review of Financial Studies 33,
2019–2133.
Huberman, G., Kandel, S., 1987. Mean-variance spanning. Journal of Finance 42, 873–888.
Ince, O. S., Porter, R. B., 2006. Individual equity return data from Thomson Datastream:
Handle with care! Journal of Financial Research 29, 463–479.
Jacobs, H., 2016. Market maturity and mispricing. Journal of Financial Economics 122, 270–
287.
Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: Implications
for stock market efficiency. Journal of Finance 48, 65–91.
Jegadeesh, N., Titman, S., 2001. Profitability of momentum strategies: An evaluation of

alternative explanations. Journal of Finance 56, 699–720.
Karolyi, A. G., 2016. Home bias, an academic puzzle. Review of Finance 20, 2049–2078.
60

Karolyi, G. A., Lee, K.-H., van Dijk, M. A., 2012. Understanding commonality in liquidity
around the world. Journal of Financial Economics 105, 82–112.
Kelly, P. J., 2014. Information efficiency and firm-specific return variation. Quarterly Journal
of Finance 04, 1450018.
Li, L., Galvani, V., 2018. Market states, sentiment, and momentum in the corporate bond
market. Journal of Banking & Finance 89, 249–265.
Lintner, J., 1965. The valuation of risk assets and the selection of risky investments in stock
portfolios and capital budgets. Review of Economics and Statistics 47, 13–37.
McLean, R. D., Pontiff, J., 2016. Does academic research destroy stock return predictability?
The Journal of Finance 71, 5–32.
Menkhoff, L., Sarno, L., Schmeling, M., Schrimpf, A., 2012. Currency momentum strategies.
Merton, R. C., 1974. On the pricing of corporate debt: The risk structure of interest rates.
Morck, R., Yeung, B., Yu, W., 2000. The information content of stock markets: why do
emerging markets have synchronous stock price movements? Journal of Financial Eco-
nomics 58, 215 – 260, special Issue on International Corporate Governance.
Moreira, A., Muir, T., 2017. Volatility-Managed Portfolios. Journal of Finance 72, 1611–1644.
Moskowitz, T. J., Grinblatt, M., 1999. Do industries explain momentum? Journal of Finance
54, 1249–1290.
Moskowitz, T. J., Ooib, Y. H., Pedersen, L. H., 2012. Time series momentum. Journal of
Financial Economics 104, 228–250.
Mossin, J., 1966. Equilibrium in a capital asset market. Econometrica 34, 768–783.
61

Newey, W. K., West, K. D., 1987. A simple, positive semi-definite, heteroskedasticity and
autocorrelation consistent covariance matrix. Econometrica 55, 703–708.
Novy-Marx, R., Velikov, M., 2016. A Taxonomy of Anomalies and Their Trading Costs.
Review of Financial Studies 29, 104–147.
Okunev, J., White, D., 2003. Do momentum-based strategies still work in foreign currency
markets? Journal of Financial and Quantitative Analysis 38, 425–447.
Ozdagli, A. K., 2017. Financial Frictions and the Stock Price Reaction to Monetary Policy.
Review of Financial Studies 31, 3895–3936.
Rouwenhorst, K. G., 1998. International momentum strategies. Journal of Finance 53, 267–
284.
Rouwenhorst, K. G., 1999. Local return factors and turnover in emerging stock markets.
Schmidt, P. S., Von Arx, U., Schrimpf, A., Wagner, A. F., Ziegler, A., 2017. On the con-
struction of common size, value and momentum factors in international stock markets: A
guide with applications. Swiss Finance Institute Research Paper 10.
Sharpe, W. F., 1964. Capital asset prices: A theory of market equilibrium under conditions
of risk. Journal of Finance 19, 425–442.
Stambaugh, R. F., Yu, J., Yuan, Y., 2015. Arbitrage asymmetry and the idiosyncratic volatil-
ity puzzle. The Journal of Finance 70, 1903–1948.
Titman, S., John Wei, K. C., Xie, F., 2013. Market development and the asset growth effect:
International evidence. Journal of Financial and Quantitative Analysis 48, 1405–1432.
Wang, F., Yan, X. S., 2021. Downside risk and the performance of volatility-managed port-
folios. Journal of Banking & Finance 131, 106198.
62

Watanabe, A., Xu, Y., Yao, T., Yu, T., 2013. The asset growth effect: Insights from inter-
national equity markets. Journal of Financial Economics 108, 529–563.
Windmüller, S., 2021. Firm characteristics and global stock returns: A conditional asset
pricing model. Review of Asset Pricing Studies , forthcoming.
63

Enhanced Momentum Strategies

Uploaded by

Copyright:

Available Formats

Enhanced Momentum Strategies

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Enhanced Momentum Strategies

Uploaded by

Copyright:

Available Formats

Enhanced Momentum Strategies

Latest version here

Matthias X. Hanauer Steffen Windmüller

First version: January, 2019; This draft: September, 2021

posed in the literature — constant volatility-scaled momentum, constant semi-volatility-

decrease momentum crashes and lead to higher risk-adjusted returns. However, in

multiple factor comparison tests, no enhanced momentum strategy emerges as consis-

volatility-scaled momentum returns to market dynamics, whereas dynamic-scaled mo-

mentum is significantly less affected, suggesting a reduced sensitivity to time-varying

Keywords: Anomalies, Asset pricing, Momentum, International stock markets

JEL Classifications: G12, G14, G15

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

2 Data and Methodology

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

[Table 1 about here.]

2.2 Factor construction

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

2.3 Enhanced momentum strategies

Electronic copy available at: https://ssrn.com/abstract=3437919

RcM OM,t = RM OM,t · wcM OM,t . (4)

Constant semi-volatility-scaled momentum (sMOM) is constructed analogously to cMOM,

Electronic copy available at: https://ssrn.com/abstract=3437919

RM OM,t = γ0 + γint · IBear,t−1 · σRM

Electronic copy available at: https://ssrn.com/abstract=3437919

RdM OM,t = RM OM,t · wdM OM,t . (8)

We investigate the performance of the different momentum strategies based on a compre-

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

3.1 Comparison of Momentum Strategies

In this section, we compare the improvement of enhanced momentum factors as compared

Electronic copy available at: https://ssrn.com/abstract=3437919

[Table 3 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

[Table 4 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

[Table 5 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

3.2 Cross-country analysis

[Table 6 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Reti,t = Xi,t + controls + ei,t , (10)

[Table 7 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

Electronic copy available at: https://ssrn.com/abstract=3437919

[Table 9 about here.]

Electronic copy available at: https://ssrn.com/abstract=3437919

3.4 Momentum Strategies and the January Effect

[Table 10 about here.]