Is Momentum an Echo?

Amit Goyal
Swiss Finance Institute
University of Lausanne
Email: Amit.Goyal@unil.ch

Sunil Wahal
WP Carey School of Business
Arizona State University
Email: Sunil.Wahal@asu.edu

April 2013

Abstract

In the U.S., momentum portfolios formed on returns from 12 to seven months prior to the current
month deliver higher returns than momentum portfolios formed from six to two months prior,
suggesting an “echo” in returns (Novy-Marx (2012)). In 37 countries not including the U.S.,
there is no such echo. In portfolios that combine securities in developed and emerging markets,
there is also no echo. Any echo in the U.S. appears to be driven largely by a carryover of short-
term reversals from month −2.

*
We thank Jennifer Conrad, Ken French, Narasimhan Jegadeesh, and seminar participants at Babson College and
the University of Toronto for helpful comments. We especially thank Michael Wolf for help with bootstrap issues.
Zhiyi Qian provided excellent research assistance. Amit Goyal would like to thank Rajna Gibson for her support
through her NCCR-FINRISK project.
1. Introduction
Momentum—the notion that winners continue to win and losers continue to lose—is
robust and pervasive. It exists cross-sectionally in individual stocks (Jegadeesh and Titman
(1993, 2001)), in portfolios (Lewellen (2002)), outside the U.S. (Rouwenhorst (1998), Griffin, Ji,
and Martin (2003), and Chui, Titman, and Wei (2010)), in various asset classes (Asness,
Moskowitz, and Pedersen (2009)), and apparently in the time series (Moskowitz, Ooi, and
Pedersen (2012)). Unsurprisingly, it has spawned theory, both behavioral (Barberis, Shleifer, and
Vishny (1998), and Daniel, Hirshleifer, and Subrahmanyam (1998)), and rational (Berk, Green,
and Naik (1999), Johnson (2002), and Sagi and Seasholes (2007)). In general, the theory fits one
or more of these manifestations of momentum, but not all.
A number of authors have worried about the term structure of predictability based on
returns over the prior 12 months. Specifically, which months contribute more or less to the
profitability of momentum strategies, and why? Jegadeesh (1990) estimates cross-sectional
regressions of current month returns on all prior 12 months’ returns. His results show that there
is a reversal in returns from the prior month (the now well-known one-month reversal effect),
and that the prior 12 month’s return has a large positive influence on the current month
(sometimes referred to as the “12-month” effect). Heston and Sadka (2008, 2010) estimate
similar return-response regressions and find that stocks have high or low returns in every year in
the same calendar month. The most recent contribution to this area comes from Novy-Marx
(2012). His rather striking finding is that portfolios formed from sorting on returns from 12 to
seven months prior (intermediate horizon returns, hereafter denoted as IR) predict average future
returns better than portfolios formed from sorting on returns from the prior six to two months
(recent horizon returns, hereafter denoted as RR). The magnitude of the difference in profits
from IR versus RR strategies is as much as 0.50% per month, which Novy-Marx argues can
make an enormous difference in overcoming transaction costs incurred during implementation of
momentum strategies. More importantly, this “echo”, rather than a continuation in returns,
challenges theory and effectively leaves it flat-footed—no model predicts such a term-structure,
and Novy-Marx himself does not present an explanation for the result.
It is conventional to perform out-of-sample tests to establish the veracity of such
provocative results. One way to do this is to examine different sample periods. For example,
Jegadeesh and Titman (2001) look for momentum between 1990 and 1998 to establish the out-

1
of-sample reliability of their original results (which are based on data from 1965 to 1989).1
Novy-Marx (2012) looks at various subperiods and finds that the echo in returns occurs during
most (but not all) of his sample period. Another approach is to see if the phenomenon occurs in
other markets. Indeed, virtually every major “fact” in cross-sectional asset pricing (such as the
value premium or momentum) is established by both U.S. and international evidence. Novy-
Marx examines the term-structure of momentum returns in commodities, currencies, and
international (country) indices. He argues that while few of these tests are not quite independent,
the fact that the same patterns appear in these indices suggests that the echo is genuine. This
apparent robustness, together with the lack of rational or behavioral intuition for why there might
be an echo, makes the result quite puzzling.
As a true out-of-sample test, we examine the term-structure of momentum across stocks
in 37 countries between 1980 and 2010. We create winner-minus-loser portfolios defined
separately over intermediate and recent returns for each country. Future monthly returns and
three-factor alphas for IR portfolios are largely indistinguishable from RR portfolios for all 37
countries. Since the number of stocks and market capitalization of some countries can be small,
and to address power concerns, we also evaluate the performance of portfolios for all firm-
countries in developed and emerging markets. Here too there is no conclusive evidence that IR
portfolios deliver returns that are different from those of RR portfolios.
We also construct portfolios independently sorted on recent and intermediate returns.
Double-sorting allow us to look at returns on portfolio strategies that focus on intermediate
returns but are neutral with respect to recent returns, thereby avoiding any mechanical covariance
between strategies. For example, within the loser portfolio constructed from recent returns, we
can look at the winner-minus-loser portfolio constructed from intermediate returns. Comparing
the above return to its “mirror” (the winner-minus-loser portfolio constructed from recent returns
for the loser portfolio based on intermediate returns), allows us to hold one source of variation
constant. In the U.S., Novy-Marx (2012) finds that the excess returns of IR portfolios are roughly
twice that of RR portfolios, even after holding recent and intermediate returns constant. We do
not see this in our data. Using equal- and value-weighted double-sorted portfolios, and
examining both portfolio returns and three-factor alphas, there is no systematic evidence that IR

1
Similarly, Heston and Sadka (2008) confirm the cross-sectional serial correlation patterns in Jegadeesh (1990),
extending the sample period by 20 years.

2
portfolios outperform RR portfolios in any of the 37 countries, or in aggregate developed and
emerging markets.
Both single- and double-sorted portfolio strategies do not explicitly control for stock
characteristics. Therefore, we also estimate Fama and MacBeth (1973) regressions that forecast
future monthly returns using intermediate and recent returns, controlling for size, book-to-
market, and the prior month’s return. The regressions show small size effects and large value
effects across these countries. In markets where momentum is known to be strong, the slopes on
both intermediate and recent returns are reliably positive. Most important, however, is the
difference in the slopes on intermediate and recent returns: in most markets, these slope
differences are small and statistically insignificant. There are two markets (Japan and China) in
which the coefficients on intermediate returns are significantly larger than those on recent
returns, but there are also three markets (Belgium, Denmark, and Chile) in which the opposite is
true. These likely random exceptions aside, we do not see any evidence that intermediate returns
are better or worse at forecasting future returns than recent returns.
These results raise the obvious question of why the echo in momentum returns is so
strong in the U.S. and not anywhere else. Chui, Titman, and Wei (2010) suggest that the degree
of individualism across countries (which they argue is correlated with overconfidence and
attribution bias) helps explain cross-country variation in momentum profits. Indeed, the U.S. has
the highest value in Hofstede’s individualism index that they report (91); one might, therefore,
speculate that this is somehow related to the echo in returns. But the next two countries with the
highest individualism index are Australia (90) and the U.K. (89). In both these countries, there is
no reliable evidence that portfolios based on intermediate returns outperform those based on
recent returns.
Theory also offers no guidance what can or should be considered “intermediate” or
“recent.” Therefore, we conduct an in-depth examination of the full term structure of returns in
the U.S. We estimate regressions in the spirit of Jegadeesh (1990) and forecast future monthly
returns using each of the past 12 months’ returns while also controlling for size and book-to-
market ratios. The coefficients from these regressions can be interpreted as standard return
responses, describing the impact of each prior month’s return on future returns, and are quite
informative for the puzzle. Intermediate return portfolios contain the prior 12 month’s return
response, which is higher than all other months’ responses (the “12-month” effect described

3
earlier). Recent return portfolios contain the return response from month −2. On average, the
return response of month −2 is negative, which appears to be a spillover from short-term
reversals associated with the prior month −1. The return response of all other months (months −3
through −11) are reliably positive, consistent with momentum.
To determine if these two return response effects account for the outperformance of
intermediate return portfolios, we aggregate coefficients in ways that correspond to all possible
definitions of intermediate and recent. The definitions are obviously not independent, so we take
seriously Leamer’s (1978, 1983) monition regarding inadvertent data snooping biases. 2 We
employ a bootstrap procedure designed by Romano and Wolf (2005), which is a generalization
of White’s (2000) reality-check method. The procedure effectively asks whether a specific
trading strategy (i.e. a specific aggregation of coefficients from the above regressions) generates
returns that beat a benchmark, while explicitly recognizing that a researcher could examine many
such correlated strategies. We find that many definitions of intermediate and recent returns
(although not the Novy-Marx (2012) definition per se) show that return responses of
intermediate returns are higher than those of recent returns, even when assessing statistical
significance using the bootstrap. But this difference in return responses is driven by the negative
return response of month −2. Indeed, if one excludes coefficients from month −2, then there are
no differences average coefficients using any definition of intermediate versus recent returns.
Aggregated coefficients from regressions are informative but are not trading strategies
per se. Therefore, we also create intermediate return portfolios using months −12 to −8, and
recent return portfolios from −7 to −3 (thereby excluding the month −2 return in portfolio
formation). The intermediate return portfolios have an inherent and deliberate bias because they
still include the 12-month effect. The difference in future portfolio value-weighted three-factor
alphas between the two strategies is only 0.24% (t-statistic = 0.98). Using double-sorted returns,
a similar difference in future portfolio three-factor alphas is 0.23% (t-statistic = 1.20). Thus even
the evidence from portfolio strategies suggests that month −2 has an inordinate influence on the
difference in returns between intermediate and recent return portfolios.

2
Leamer (1983) complains of the “fumes which leak from our computing centers” and argues that it is critical that
the fragility of results from trading strategies be studied in a systematic way to establish or alter beliefs. The danger
of falsely rejecting the null is of course not unique to finance or the social sciences. For an interesting account of
such inadvertent inferences in the physical sciences, see Mlodinow (2008). Leamer (1978, 1983) and Lo and
MacKinlay (1990) also emphasize that the effects of data snooping need not be willful; chance has a role to play
when researchers and investors engage in completely appropriate specification searches.

4
 If, as it seems, there is no real echo in returns and the apparent echo is just a
manifestation of short-term reversals, then there is no real challenge to theories of momentum.
Instead the challenge is passed on to theories of short-horizon reversals. There are two such
candidates. Models such as Campbell, Grossman, and Wang (1993) argue that reversals are due
to liquidity effects (Jegadeesh and Titman (1995a) similarly argue for microstructure biases and
inventory effects). Although not outside the realm of possibility, it is hard to believe that these
liquidity effects last for almost two months. A second candidate explanation is overreaction by
investors (Jegadeesh and Titman (1995b)). It is analytically hard (or impossible) to reject
overreaction stories because the theory does not identify specific horizons – overreaction can
occur over two months just as much as one month. Regardless, there does not appear to be a
difference in returns of portfolios constructed from various combinations of intermediate and
recent returns – the enigma remains short-horizon reversals.
The remainder of the paper is organized as follows. Section 2 describes our data and
explains sampling procedures. Section 3 describes out-of-sample tests outside the U.S. Section 4
pursues explanations for the U.S. results and includes bootstrapping exercises. Section 5
concludes.

2. Data and Sample

We obtain a time series of market and accounting information for a broad cross-section of
firms (excluding the U.S.) from Datastream. We start with an unconstrained universe of all firms
in 45 countries in the MSCI All Country Index between 1980 and 2010. MSCI classifies 23 of
these countries as developed markets and the remaining 22 as emerging markets. The universe
includes live as well as dead stocks, ensuring that the data are free of survivorship bias. We
apply the following sequence of filters that are derived from the extensive data investigations by
Ince and Porter (2006), Griffin, Kelly, and Nardari (2010), and Hou, Karolyi, and Kho (2011):

1. We require that stocks have data from Datastream and Worldscope. The former is the source
for market data whereas the latter contains necessary accounting data.
2. We only retain issues that are listed as equity and require that they be from the firm’s
primary exchange. The latter, along with another Datastream supplied field, serves to remove
duplications.

5
3. We eliminate all non-local firms with a GEOG code different from the local market. This
eliminates foreign listings and therefore doubling counting of securities.
4. When a security dies, the post-death local currency returns are marked as zero, which we
eliminate.
5. We employ a text search algorithm to eliminate securities that are not common stock. This
eliminates preferred stock, trusts, warrants, rights, REITS, closed-end funds, ETFs and
depository receipts (GDRs and ADRs) that are not caught by the above screens.
6. We compute returns using the return index (which includes dividends) supplied by
Datastream. Since both return indexes and market capitalization are provided in local
currency, we convert them to U.S. dollar equivalents using the conversion function built into
Datastream. We perform some simple checks to ensure that local and U.S. dollar returns are
consistent and computed correctly.
7. We set returns to missing for a stock when it rises by 300% or more during one month and
drops by 50% or more the following month (or falls and subsequently rises). We also treat as
missing returns above 1000% per month.
8. Each month and for each country, we winsorize returns, market capitalization and book-to-
market ratios using the first and the 99th percentiles.

For most of our tests, we require that a country-month have at least 100 stocks. This
ensures that country-specific portfolios are reasonably well diversified. The restriction removes
the following seven countries from our single country tests: Ireland, Colombia, Czech Republic,
Egypt, Hungary, Morocco, and Poland.
For the U.S. sample, we use all NYSE, AMEX, and NASDAQ stocks in the CRSP
database. Book values are from Compustat and are calculated following the procedure described
in Fama and French (1993).3 The rest of the data come from CRSP. The book-to-market ratio is
calculated as the ratio of the most recently available (assumed to be available six months after
the fiscal year-end) book value of equity divided by the current market capitalization. Book-to-
market ratio is winsorized each month using the 0.5 and the 99.5 percentiles. We include only

3
Book values from Compustat are supplemented with hand-collected values from Moody’s, whenever available (see
Davis, Fama, and French (2000) for the exact description of these data). These are available on Kenneth French’s
website at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

6
common stocks with share codes 10 or 11 on CRSP. This criterion filters out ADRs, units,
Americus Trust components, closed-end funds, preferred stocks, and REITs.
Table 1 provides a brief overview of the sample. In this and other tables, we separate
developed and emerging markets, and order countries alphabetically within each region.
Although a country’s designation as developed or emerging could have changed over time, we
use current MSCI definitions. Data availability varies across countries. We report the start date
for the data series for which momentum portfolios can be constructed for each country (we
discuss portfolio formation procedure slightly later in Section 2.3). Not surprisingly, the
available time series is longer for developed markets. For major developed markets (e.g.
Australia, Canada, the U.K., Japan, etc.), there is a full time series of 30 years. We report the
average and aggregate market capitalization as of the end of 2010 in millions and billions of U.S.
dollars, respectively. In our sample, Canada, France, Germany, Japan, U.K., China, and India
each have market capitalization in excess of $1 trillion at the end of 2010 (with Switzerland and
Brazil close behind). At the end of 2010, these seven countries have a joint market capitalization
of around $12 trillion, roughly two thirds that of the U.S. These seven countries together
constitute about 23% of the global stock market capitalization. As a check on the coverage of our
data, we compare the aggregate market capitalization for each country at the end of each year
with statistics reported by the World Federation of Exchanges (www.world-exchanges.org). For
the years for which these comparisons are feasible, the firms in our sample cover between 80%
and 99% of the aggregate market capitalization of each country. We also report the average
number of stocks per country, where the average is computed over the entire (monthly) time
series. Despite the fact that we require a minimum of 100 securities per month, there is still
considerable variation in the average number of stocks per country. Throughout, we urge caution
in interpreting results for countries with a small universe of securities.

2.1. Factors
To estimate alphas from three-factor models, we also need factors, which we construct as
follows. For a country to have a non-missing factor for a particular month, we require at least
five stocks per portfolio used to create the factor. For example, typical double-sorted factors that
use a 3×2 sort require at least 30 stocks (six portfolios multiplied by five stocks). This is a
minimal requirement and most factors are well diversified. To establish size breakpoints, we sort

7
all stocks in a country into two groups, small (S) and big (B), based on their market
capitalization as of June of year t. Big stocks are those in the top 75% of market cap, and small
stocks are those in the bottom 25% (the breakpoint for big corresponds to that used by MSCI).
Book-to-market ratios are computed by dividing book value for the fiscal year ending in year t−1
by market capitalization at the end of December of year t−1. These values are used to generate
breakpoints corresponding to low (L), medium (M) and high (H) portfolios. The low and high
portfolios correspond to the 30 and 70 percentiles respectively, with medium being between
those two. For each size and value portfolio, we compute monthly value-weighted returns in U.S.
dollars and local currency from July of year t through June of year t+1. We compute country
specific factor returns as the difference in average returns across size groups.
Some of our tests described below combine securities in developed and emerging markets
into portfolios aggregated in a variety of ways. In constructing developed and emerging market
factors to evaluate these portfolios, we follow approaches that correspond directly to the method
of portfolio construction. For instance, when we value-weight countries to generate a value-
weighted developed markets portfolio, we also value-weight country-specific factors (using each
country’s total market capitalization at the end of the previous month). Similarly, when we
equal-weight countries, we also equal-weight country-specific factors. In specifications where
we pool stocks in developed or emerging markets, we generate breakpoints and factors using all
firms in the entire region. This ensures consistency between the construction of portfolios and
the factors used to evaluate them.
The U.S. factors are obtained directly from Ken French’s website.

2.2. Conventional momentum sorts

Table 1 also shows momentum profits in our sample of countries. One could construct
momentum portfolios in a number of different ways. We start our analysis, though, with the
simple and conventional approach of using prior eleven-month returns, skipping the most recent
month, and holding these portfolios for one month.
For all countries except the U.S., we sort all stocks in a country-month with valid returns
separately into quintiles based on month −12 to −2 returns.4 We require at least 100 stocks per

4
We use quintiles to ensure that portfolios are well populated. We also conduct tests using deciles whenever
possible to ensure the robustness of our results. Details are reported in Section 3.5.

8
month to do the sorts. We calculate both equal- and value-weighted monthly returns for each
quintile (skipping month −1) but report only the value-weighted returns in Table 1 to conserve
space. In addition to average monthly returns, we also estimate CAPM and three-factor alphas.
These are based on time series regressions of these monthly returns on local market, SMB and
HML factors. We calculate alphas only if there are at least 60 months of return data. In countries
where we are unable to calculate these alphas (e.g. Austria and Russia), we only report portfolio
returns. Countries where returns and/or alphas are statistically significant are shaded.
Table 1 shows winner-minus-loser portfolio returns, CAPM alphas, and three-factor
alphas for these momentum portfolios along with t-statistics in parentheses. Momentum portfolio
returns are statistically significant in 12 out of 22 developed markets (excluding the U.S.), and in
two out of 15 emerging markets. Three-factor alphas are positive in 12 out of 22 developed
markets and in five out of 15 emerging markets. As is well known, momentum portfolios
generate no profits in Japan. In general, the cross-country variation in momentum lines up well
with the results in Chui, Titman, and Wei (2010), despite the fact that they use a different
portfolio formation approach.5 This is especially true in what one may consider “major”
countries. Momentum returns are notably high in countries with large market capitalizations such
as Australia, Canada, France, Germany, and the U.K.

3. Results
3.1. Portfolio Returns and Alphas
We now change our procedures to sort on different horizons of past returns. For all ex-
U.S. countries, we require that all stocks in a country month have valid intermediate (months −12
to −7) and recent (months −6 to −2) returns, as well as market capitalization and prices in the
prior month. Using these stocks, we compute return breakpoints for each country-month,
assigning stocks into quintiles. For the U.S, we modify the process slightly by excluding stocks
under $1, using deciles instead of quintiles, and establishing breakpoints using only NYSE
stocks. We report statistics on portfolio returns for the U.S. over the sample period of 1980−2010
to facilitate comparison with the international sample, as well as over the sample period
1927−2010 to facilitate comparison with prior studies.

5
Chui, Titman, and Wei (2010) assign stocks to terciles (not quintiles) based on the prior six-month return and hold
securities for six months (not one month).

9
 We require a minimum of 100 stocks per country-month to construct return quintiles; this
is a minimum so that in major developed markets, the quintiles contain many more securities.
For instance, in the U.K., on average, each quintile contains approximately 250 stocks. Portfolios
returns are equal- or value-weighted based on the prior month-end market capitalization.6

3.1.1. Portfolio Returns

Panel A of Table 2 shows winner-minus-loser (WML) portfolio returns for intermediate
and recent return based momentum portfolios along with t-statistics in parentheses. Thus,
RR(WML) denotes the hedge portfolio based on recent return sorts and IR(WML) denotes the
hedge portfolio based on intermediate return sorts. We also report the difference in returns
between these two hedge portfolios (RR(WML) − IR(WML), labeled “Difference”).
Focusing on value-weighted returns, in the U.S., between 1980 and 2010, momentum
portfolios formed on RR underperform portfolios formed on IR by 0.62% per month with a t-
statistic of 1.71. The equivalent number over the full sample period is 0.51% (t-statistic = 2.11),
which is close to the numbers reported by Novy-Marx (0.54% with a t-statistic of 2.21). There is
no difference in equal-weighted returns over 1980–2010, suggesting that at least over this sample
period, the return differences are not driven by small stocks. The equal-weighted return
difference shows up in the longer sample, albeit weaker at 0.39% with a t-statistic of 1.84.7
Outside the U.S., Table 2 shows there is no statistically significant difference in IR versus
RR based value-weighted portfolio returns in 35 out of 37 countries. The two exceptions are
Australia and Israel, where IR portfolios underperform RR portfolios by 1.33% and 1.27% per
month, respectively. If we use equal-weighted portfolios, there is again no statistically significant
evidence of over- or underperformance of IR based portfolios again in 34 out of 37 countries. In
Brazil and India, IR portfolios have a higher return than RR portfolios and in Chile, the opposite
is true.
In addition to looking at the aforementioned exceptions, it is also informative to focus on
countries with large market capitalizations. In major developed countries, there is no difference
in returns for IR versus RR portfolios for Canada, France, Germany, or the U.K. In Japan, the IR

6
It is conventional to use value-weighted returns. However, in some markets outside the U.S., a few firms can
dominate the local market capitalization. As a result, we employ both equal and value-weighted returns.
7
Novy-Marx (2012) does not report statistics for equal-weighted portfolio returns. However, from tests for size
quintiles (pg. 441–442), it is obvious that the dispersion in future returns from intermediate and recent return
portfolio sorts is bigger for larger firms.

10
portfolio outperforms the RR portfolio by 0.57% per month but the t-statistic is only 1.50. In
contrast, in Australia, the IR portfolio underperforms by 1.33% per month (t-statistic = 2.72). In
the major emerging markets of Brazil, China, and India, there is no robust evidence that
intermediate returns generate any more or less predictability than recent returns. Indeed, given
the multitude of numbers and the role of chance, it is not surprising to find a few IR portfolios
exhibiting statistically different performance than that of RR portfolios. Our reading of this
evidence is that, apart from a few exceptions, IR portfolios do not out (or under) perform RR
portfolios.

3.1.2. Portfolio Alphas

For each country, we also calculate factor model alphas based on local CAPM and three-
factor models. That is, we regress the time-series of winner-minus-loser portfolio returns for a
country on that country’s market factor, SMB and HML. We use the one-month T-bill rate as the
risk free asset. To not overwhelm the reader, we do not report CAPM alphas and loadings on the
market factor, SMB and HML for each country and specification. Instead, Panel B of Table 2
displays 3-factor alphas for equal- and value-weighted portfolios with t-statistics in parentheses.
We calculate these alphas only if we have at least 60 time-series observations. This is not always
possible either because of paucity of data in portfolio returns or the unavailability of factor
returns. Fortunately, we lose only Austria, Portugal, and Russia because of this requirement.
Consistent with the extensive literature on momentum in the U.S., factor models have
very little impact on the profits of momentum strategies; there are only modest differences
between returns and alphas. For instance, in value-weighted portfolios, the difference in three
factor alphas between RR(WML) and IR(WML) based momentum portfolios for the U.S. is
−0.63% per month with a t-statistic of 1.74 (compared to −0.62% for simple portfolio return
differences).
Outside the U.S., there are no statistically significant differences in alphas between IR
and RR value-weighted portfolios in 36 out of the 37 countries (Japan is an exception). Using
equal-weighted portfolios, there are two countries (Japan and India) where alphas of IR
portfolios are higher than those of RR portfolios. However, there are also two countries (the U.K.
and Chile) in which the opposite is true. Focusing again on major developed and emerging
markets, there is no difference in alphas between IR and RR portfolios for Canada, France,

11
Germany, the U.K., Brazil, China, or India. As with portfolio returns, Japan shows some
evidence of an echo, with a difference in three-factor alphas of 0.77% (t-statistic = 1.99), while
Australia shows the opposite (a difference in alphas of 0.86% with a t-statistic of 1.71). Again,
our reading is that, apart from a few exceptions, IR portfolios do not out (or under) perform RR
portfolios even on a risk-adjusted basis.

3.1.3. Regional Returns and Alphas

A reader may legitimately be concerned about the power of country-specific tests. To
mitigate these concerns, we impose stringent restrictions on the number of securities and also
pay special attention to “major” countries. Notwithstanding the above, another way to address
power concerns is to combine securities/portfolios across countries and thereby produce
especially diversified portfolios.
There are three ways to do this, all of which provide interesting inferences. We can
equal- or value-weight portfolios across countries in a region using a country’s aggregate market
capitalization from the prior month. Value-weighting countries has obvious advantages, except
in some circumstances where one or two large countries can dominate the overall portfolio; in
those circumstances, equal-weighted portfolios provide additional information. Both these
weighting schemes ensure that all countries are represented in the aggregate portfolio. This is the
approach followed by Fama and French (1998). A third approach is to pool all securities within a
region. The resulting portfolio need not contain securities from all countries at a point in time but
is still broadly diversified. This is the approach used by Fama and French (2012), who combine
23 developed markets into four regions.8 All three approaches produce diversified portfolios,
albeit in different ways.
The results for the three variants of developed and emerging markets portfolios are
reported in the last three rows for developed (below the U.K.) and emerging market countries
(below Turkey) respectively in Panels A and B of Table 2. By way of notation, EW (VW) refers
to portfolios in which each country’s portfolio is equal-weighted (value-weighted) across
countries. “Pooled” refers to portfolios in which all securities in developed (or emerging)
markets are combined before portfolio construction. We require that a pooled portfolio has at

8
One could imagine pooling other groups of countries as well. For instance, we could pool European countries, or
Asia-Pacific countries, or use other such combinations. We stick to developed and emerging markets mostly because
these represent commonly used investment strategies.

12
least 1,000 securities.9 In evaluating each of these portfolios, factors are built in an analogous
manner (as described above in Section 2.1).
Panel A shows that the differences in returns between RR and IR portfolios for developed
markets vary from −0.36% to 0.16% depending on whether one considers equal-weighted, value-
weighted or pooled portfolios. These differences are not statistically significant in any of the six
variations. In emerging markets, the return differences are all negative, ranging from −0.10% to
−0.52% with high standard errors. Panel B reports differences in three-factor alphas. Once again,
in developed markets, there is considerable variation in the alphas, from 0.02% to −0.60%. In
one specification (using value-weighted returns within countries and value-weighting across
countries), the return difference is −0.60% with a t-statistic of −2.06. However, in the remaining
five portfolios, the differences in alphas are economically small and statistically insignificant. In
emerging markets, none of the differences in portfolio alphas are statistically distinguishable
from zero. Thus, even with well-diversified portfolios across countries, there do not appear to be
robust differences in returns or three-factor alphas between portfolios formed on recent versus
intermediate returns.

3.2. Double-Sorted Portfolios

Another way to analyze the incremental effects of intermediate versus recent returns on
momentum is via double-sorted portfolios. We sort all stocks into winner and loser groups based
on recent returns, and independently sort the same stocks into winner and loser groups based on
intermediate returns. The intersection of these two independent sorts informs us about the
marginal contribution of intermediate versus recent returns. In other words, we can examine the
effects of intermediate returns on momentum portfolios while holding recent returns (roughly)
constant, and vice-versa.
This approach is not without difficulties in implementation, especially in ex-U.S. data.
Given the paucity of stocks in certain countries, we cannot use double-sorted quintiles in many
countries; the portfolios would simply have too few securities to have any economic meaning.
To alleviate this problem, we sort stocks into terciles rather than quintiles.10 These 3×3 sorts
generate nine portfolios, on average containing 45 securities each per country. For the U.S., we
9
Note that these equal-, value-weighted, and pooled approaches refer to methods of aggregation across countries,
not weighting within countries. Within a country, returns can be equal- or value-weighted, as shown by the columns
in Table 2.
10
The tercile approach is also used by Chui, Titman and Wei (2010) in dealing with markets with fewer stocks.

13
use quintiles instead of deciles. All other procedures for establishing breakpoints, assigning
stocks to groups and calculating portfolio returns are identical to those for single-sorted
portfolios.
The table below illustrates the sort procedure and clarifies the exposition of our results.
The top row shows intermediate return terciles (IRL through IRW), where “L” and “W” denote
loser and winner terciles respectively. Analogously, the first column shows recent return terciles
(RRL through RRW). We compute the individual portfolio returns X1 through X9 and then
calculate the return differences for the extreme portfolios. For example, we calculate
IR(WML)/RRL (the winner-minus-loser portfolio formed from intermediate returns, for recent
return losers) as well as its mirror RR(WML)/IRL (the winner-minus-loser portfolio formed from
recent returns, for intermediate return losers). If momentum is driven by intermediate returns,
then we expect IR(WML)/RRL > RR(WML)/IRL, or equivalently that IR(WML)/RRW is
greater than its mirror RR(WML)/IRW.11
IRL IR2 IRW IR(WML)
RRL X1 X2 X3 IR(WML)/RRL
RR2 X4 X5 X6 X6-X4
RRW X7 X8 X9 IR(WML)/RRW
RR(WML) RR(WML)/IRL X8-X2 RR(WML)/IRW

To conduct inferences we compute differences in these return series (i.e. IR(WML)/RRW

− RR(WML)/IRW) and use the time series of return differences to generate t-statistics. An added
advantage is one of exposition – instead of conducting a large number of pairwise comparisons,
one can simply look average differences in returns/alphas. Table 3 shows these results. Focusing
again on value-weighted portfolios, in the U.S. between 1980 and 2010, RR portfolios
underperform IR portfolios by 0.75% per month with a t-statistic of 2.48. The difference in three-
factor alphas is 0.85% (t-statistic = 2.86). As before, equal weighted portfolios do not show such
dispersion. Outside the U.S., there are two countries (Japan and Switzerland), in which RR
portfolios underperform IR portfolios significantly. There are also two countries (Australia and
Israel) in which the opposite appears to be true. In the vast majority of countries, the differences
are not statistically significant. Focusing again on countries with large market capitalizations,

11
IR(WML)/RRL is computed as X3−X1, and RR(WML)/IRL is computed as X7−X1. The difference is therefore
X3−X7. Similarly IR(WML)/RRW is computed as X9−X7 and RR(WML)/IRW is computed as X9−X3. The
difference is X3−X7.

14
there are no differences in value-weighted returns or alphas in Canada, France, Germany, the
U.K., Brazil, China, and India. As before, it appears that IR portfolios deliver higher returns than
RR portfolios in Japan (e.g. difference in value-weighted three-factor alpha is −0.66% with a t-
statistic of −2.38). In Australia, the opposite is true; the difference in three-factor alphas is 0.89%
with a t-statistic of 2.04.
Aggregated developed market portfolio returns and alphas are sensitive to specifications.
For instance, when countries are equally-weighted, there is no difference in double-sorted
portfolio returns/alphas. If we value-weight countries, it appears that IR portfolios deliver higher
returns and alphas than RR portfolios. These return and alpha differences are driven by the
influence of Japan. Given its large market capitalization, Japan has a large weight in the value-
weighted developed market portfolio (but obviously less so in equal-weighed or pooled
portfolios). This, combined with the fact that the future returns of IR portfolios are higher than
those of RR portfolios in Japan, causes the value-weighted developed market portfolio to have
higher IR returns. If we remove Japan from the sample of developed countries, there is no
difference in future returns of IR versus RR portfolios in the value-weighted developed market
portfolio (A fuller discussion of the reason why IR portfolios have higher returns than those of
RR portfolios for Japan appears later in footnote 15). If we pool securities in all developed
countries, then once again, the returns and alphas of IR portfolios are statistically
indistinguishable from RR portfolios.
There is considerable variation in returns in emerging market portfolios as well. In equal-
and value-weighted portfolios, there is no evidence that IR portfolios have higher returns or
alphas compared to RR portfolios. In pooled specifications, some of the return/alpha differences
are economically large but none approach statistical significance.

3.3. Fama-MacBeth Regressions

Another way to determine whether it is recent or intermediate returns that are more
important for momentum is to estimate regressions of future returns on past returns, while
controlling for stock characteristics. We do so by estimating the following monthly cross-
sectional Fama and MacBeth (1973) regressions:

Rit  R ft   0t   1t ln MEit 1   2t ln BM it 1   3t Rit 1   4t Rit 2:t 6   5t Rit 7:t 12  i ,t . (1)

15
The independent variables include the logarithm of market value of equity in the prior month, the
logarithm of the book-to-market ratio in the prior year, the last month’s return (month −1), the
recent return (from month −6 to −2), and the intermediate return (from month −12 to −7). For
each cross-section, we require 100 stocks in a country-month to estimate the regression. Panel A
of Table 4 shows the average  coefficients (t-statistics generated using Newey-West procedure
with three lags are reported in parentheses). We report these statistics for each country where we
have a minimum of 60 months in which we are able to estimate these regressions. In addition to
the above variables, we also report the differences in the slopes for intermediate and recent
returns, the average number of stocks per month-country in the regressions and the number of
months for which the regression is estimated.
In Panel B, we change the dependent variable to Fama and French (1993) three-factor
model adjusted returns. The factors are country-specific, and we follow Brennan, Chordia, and
Subrahmanyam (1998) in computing loadings on the market factor, SMB and HML using the
entire sample period:

Rit  R ft  ˆi Ft   0t   1t ln MEit 1   2t ln BM it 1   3t Rit 1   4t Rit 2:t 6   5t Rit 7:t 12  i ,t . (2)

The slopes on size, book-to-market ratios and prior returns generally follow patterns seen
in other papers. The logarithm of market capitalization is statistically insignificant in many
regressions, including the U.S., indicating that the size effect is noticeably weaker in our sample
period. As shown by the reliably positive slopes on book-to-market ratios, value effects are
strong and robust across most, if not all countries. The one month return reversal is also quite
strong across countries; in most markets, the slope on last month’s return is reliably negative.
Consistent with the results in Tables 1 and 2, the coefficients on recent and intermediate returns
show the importance of momentum, particularly in developed markets.
Our primary interest is in the difference in the intermediate and recent return slopes. In
our sample period, in the U.S., this difference is only 0.02 with a t-statistic of 0.07 (the
difference is slightly larger at 0.27, albeit with a t-statistic of 1.15, in the full sample period of
1927–2010). By way of comparison, Novy-Marx (2012) reports a coefficient of 0.25 with a t-
statistic of 0.69 for the 1990–2010 sample period. In Panel A, there is one developed market
(Japan) in which the coefficient on IR is larger than that for RR; the difference in coefficients is
1.27 with a t-statistic of 2.60. But there are also two markets (Belgium and Denmark) for which

16
the opposite is true. Similarly, there is one emerging market (China) in which the coefficient on
IR is higher than that on RR, but also one emerging market (Chile) in which the coefficient on IR
is lower than that on RR. If we use three-factor model adjusted returns as the dependent variable
(Panel B), the results are largely similar. Indeed, in the vast majority of countries, there is no
evidence that intermediate returns are able to predict future returns any better or worse than
recent returns.
Even when Fama-MacBeth regressions are estimated after pooling securities in all
developed or emerging markets (we require at least 1,000 securities to run the pooled
regression), there is no reliable evidence of an echo. In developed markets, for instance, the
difference in IR and RR coefficients is −0.04 (t-statistic = −0.11) in Panel A, and −0.22 (t-
statistic = −0.71) in Panel B. In emerging markets, the differences in coefficients are larger but
statistically insignificant.

3.4. Robustness
In all of the results thus far, there are differences between equal- and value-weighted
portfolios, particularly for the U.S. Recognizing that momentum effects are especially
concentrated in small stocks, Novy-Marx (2012) also pays special attention to large stocks; he
finds that the echo in returns for the U.S. is in fact stronger in large stocks. Since size effects can
also be important in markets outside the U.S. (Fama and French (2012)), we redo our basic tests
solely for large stocks. We identify large stocks as those above the 75th percentile in market
capitalization for each country. We then calculate returns for portfolios based on intermediate
and recent returns, independent double-sorted portfolios, and also re-estimate Fama-MacBeth
regressions. The results of this exercise are presented in appendix Tables A1, A2, and A3. Even
in large stocks, we see no evidence that intermediate returns are better or worse at predicting
future returns than recent returns.
The majority of our ex-U.S. results are based on sorting stocks into quintiles (for single
sorts) or terciles (for double-sorts). Given the large number of firms in the U.S. (and following
convention), we also use deciles for single sorts in the U.S. To make sure that our ex-U.S. results
are not an artifact of the sorting procedure, we also use decile sorts for countries with an
adequate number of securities, namely, Australia, Canada, France, Germany, Japan, India, and
Korea. In addition, we use decile sorts for developed and emerging markets grouped together.

17
For France, Germany, India, and Korea, there is no difference in the returns or alphas for IR
versus RR portfolios. In Japan, IR portfolio alphas are higher than RR portfolio alphas by 0.90%
per month with a t-statistic of 1.86 (the differences in alphas from quintile sorts in Table 2 was
0.77% with a t-statistic of 1.99). Interestingly, in Australia and Canada, IR portfolio alphas are
lower than RR portfolio alphas by 1.29% (t-statistic = 1.90) and 1.58% (t-statistic = 2.26)
respectively, compared to 0.86% (t-statistic = 1.71) and 0.41% (t-statistic = 0.80) from Table 2.
Thus sorting into deciles leaves the results for Japan remain largely the same, but emphasizes the
importance of recent returns for Australia and Canada.

4. Explanations
The data thus far indicate that there is no robust evidence of an echo in returns in the 37
countries or in regions. In the U.S., however, intermediate returns predict future monthly returns
better than recent returns. The obvious question then remains: what explains the echo in the
U.S.? As discussed in the introduction, we do not believe that cross-country differences
attributed to individualism-collectivism can account for the results. Several countries that have
the highest score on individualism indices (most notably, Australia, Canada, and the U.K.) show
no evidence of an echo. Similarly, other conjectured variables such as those describing variation
in information uncertainty, level of development, or liquidity, face the same hurdle – in countries
where they take on values or ranges similar to the U.S., there is no evidence that IR return based
portfolios outperform RR based portfolios. One could also consider other candidate explanation
for momentum such as differences in growth opportunities (Berk, Green, and Naik (1999) and
Sagi and Seasholes (2007)). But they too would have to fundamentally vary across countries, and
somehow link to intermediate versus recent returns. This seems implausible.
To reconcile the differences between the U.S. and non-U.S. countries, we conduct an in-
depth examination of the full term structure of returns in the U.S. We take this approach because
theory provides no guidance; no extant models of momentum offer any advice on what should be
considered intermediate versus recent.

4.1. Full Term Structure for the U.S.

Since intermediate and recent returns are computed by compounding prior monthly
returns, they obfuscate individual prior months’ effects. Modifying Jegadeesh (1990) slightly, a

18
simple way to see the influence of each prior month’s return on future return is to estimate
monthly cross-sectional regressions of form:

12
Rit  R ft   0t  
k 1
kt Rit k   13t ln MEit 1   14t ln BM it 1   i ,t , (3)

where Rit is the monthly return for stock i in month t. The modification to the original Jegadeesh
(1990) specification is the inclusion of size and book-to-market ratios as regressors. The γ’s have
a straightforward interpretation as return responses at various lags k and can be interpreted as
excess returns on zero-cost hedge portfolios.
Panel A of Table 5 reports the full term structure regressions in equation (3) for the full
sample period from 1927 to 2010. In addition to using simple stock returns (labeled “raw
returns”), we also use three-factor-adjusted returns computed as in Panel B of Table 4.
The pattern of coefficients offers important insights into what drives the difference
between IR and RR portfolio returns in the U.S. Not surprisingly, and consistent with the well-
known short-term reversal effect, the coefficient on the prior month (k = 1) is large and negative.
Since the prior month is skipped in all momentum portfolios’ formation, this has no impact on
the IR versus RR results and we ignore it in the rest of our discussion. In general, the remaining
coefficients are positive, indicating momentum. But they show significant variation. Unlike all
the other prior returns, the coefficient on k = 2 is negative. The t-statistic of the coefficient is
−1.26 using raw returns and −2.29 using three-factor adjusted returns. More importantly, the
economic magnitude of the k = 2 coefficients stands in stark contrast to k = 3 through 12 which
are all reliably positive. Thus the data show that some of the short-term reversal effect typically
associated with the prior month’s return (k = 1) slips into the next month (k = 2) as well. The
second insight that emerges from these regressions is that the largest positive coefficient is for
the prior 12 month return (k = 12). This 12-month effect, first documented in Jegadeesh (1990),
shows that the excess returns of momentum portfolios benefit considerably from the “kick”
generated by month −12 return.
Panel B shows similar regressions for four equal sub-periods. The negative coefficient on
k = 2 appears particularly strong in the first and third subperiods (1927–1947 and 1969–1989). In
the other two subperiods (1948–1968 and 1990–2010), this coefficient is not reliably different
from zero but the contrast to positive coefficients for k = 3 through 12 is still stark. The high

19
return response associated with the coefficient on k = 12 (the 12-month effect) is strong in the
first three subperiods. In the last subperiod (1990–2010), there is a noticeable decline in k = 12
coefficient. These results closely match and help explain the Fama-MacBeth results reported by
Novy-Marx (2012). He finds that the slope difference between IR and RR drops to 0.25 in the
last subperiod with a t-statistic of only 0.69 (compared to 0.58 for the overall sample with a t-
statistic of 2.47). The decline in the 12-month effect in the last subperiod accounts for some of
this degradation in performance between IR and RR strategies.

4.2. Definitions of IR and RR

The data thus suggest that at least some of the difference in performance of IR and RR
strategies may be associated with the return response to month −12 and month –2. The former is
well known. Indeed, Novy-Marx (2012) shows that the 12-month effect does not drive his results
(by excluding it in the construction of IR portfolios as a robustness check). The latter, however,
is not well-known and appears to be a carryover from short-term reversals typically thought of as
occurring from the prior month.
As we have pointed out earlier, theory does not help us determine what is recent or
intermediate. Assuming one uses returns from months –12 to –2, there are in fact 55 different
ways in which one can define intermediate or recent returns. Table 6 describes this
comprehensive set of feasible definitions that can be compiled from the prior 11 months of
returns.
The first two columns show the specific months used to identify intermediate and recent
returns. The third column shows the difference between the number of months used in
definitions of intermediate and recent returns. The table is sorted in ascending order by this
difference; definitions that use more months in computation of intermediate returns appear at the
top of the table. For the months listed in the IR and RR definition columns, we calculate (but do
not report) the sums and averages of the coefficients from the full sample term structure
regression (3) reported in Table 5. The table shows the difference in the sums and averages of
these coefficients (IR−RR) with t-statistics in parentheses.
An example is useful in clarifying Table 6. Consider the very first row in the table. The
intermediate return definition is based on months –12 through –3. The recent return definition is
based only on month –2. The sum of coefficients (from Table 5) from regressions using raw

20
returns for months –12 through –3 is 13.38, while the coefficient for month –2 is –0.34 (these
two numbers are not shown separately in the table). The difference in sums is 13.73 with a t-
statistic of 11.18. Similarly, the average of coefficients from months –12 through –3 is 1.34, and
the average coefficient for month –2 is –0.34 (again, these two numbers are not shown separately
in the table). The difference in averages is shown in the table and is 1.68 (t-statistic = 6.83).
There are good economic reasons to consider both sums and averages of coefficients. The
sum of the coefficients corresponds to the IR and RR coefficients in Novy-Marx’s (2012) Fama-
MacBeth regressions (Table 1 in his paper). The first (perhaps obvious) observation from Table
6 is that the difference in the sums of coefficients follows an almost monotonic pattern. There are
large positive differences when the intermediate return definition uses more months than the
recent return definition (at the top of the table) and large negative differences when the reverse is
true (at the bottom of the table). This pattern is, of course, mechanical. Suppose that all
individual month coefficients have the same impact on future returns so that they contribute
equally to momentum. Under this null of no term structure, if an intermediate return definition
employs more months than a recent return definition, then intermediate return portfolios are
expected to deliver higher future returns than recent return portfolios.
Difference in sums of coefficients would, thus, indicate an echo for definitions at the top
of the table and a reverse echo for definitions at the bottom of the table. These mechanical effects
do not reflect a true echo. One way around this issue is to consider the differences in averages of
coefficients, rather than their sums. The averages control for differences in number of months
and are shown in the last column of the raw and three-factor adjusted return portion of Table 6.
Unsurprisingly, the monotonicity disappears. However, it is notable that averages of coefficients
for all intermediate return definitions are still larger than those of recent return definitions. In the
55 combinations displayed in the table, there are still 49 to 51 definitions (depending on the
specification) in which the average coefficients of intermediate return months are larger than
those of recent return months at 95% significance level.

4.3. Bootstrapping Correlated Strategies

The above comparisons are complicated by the fact that various definitions are
correlated. This necessitates that statistical inference of differences in coefficients account for
dependencies between rows. White’s (2000) reality check bootstrap recognizes dependencies

21
between different rules to construct trading strategies but does so in the context of the best
trading strategy. Specifically, White’s approach asks whether the best of a multitude of strategies
beats its benchmark, after controlling for dependencies between all possible strategies. Our
interest is not in the best strategy/definition, but in a specific definition (the Novy-Marx (2012)
specification) and more generally with all possible intermediate and recent return definitions.
Romano and Wolf (2005) generalize the intuition of White’s reality check bootstrap and develop
a method that allows one to assess the performance of any specific strategy, while recognizing
dependencies among potential strategies. This is accomplished by controlling the familywise
error rate (FWE), which is the probability of incorrectly identifying at least one strategy as
beating the benchmark when, in truth, it does not.
The details of our implementation of Romano and Wolf (2005) procedure are provided in
Appendix A. Briefly, the procedure starts by defining a test statistic; in our case, the difference in
the sum or average of coefficients of intermediate and recent return definitions or their
studentized versions, the t-statistic of those coefficient differences.12 Various definitions are then
organized in descending order using this test-statistic and critical values for the statistic of each
definition are constructed by bootstrapping the joint confidence interval. White (2000) and
Romano and Wolf recommend a bootstrap procedure that is applicable to time-series data;
accordingly we use Politis and Romano (1994) stationary bootstrap. The one-sided null
hypothesis is that the test statistic is less than zero. Because the statistics are ordered, the
procedure is continued until no further hypotheses are rejected. This allows one to reject (or fail
to reject) the null hypothesis for each specific definition.
We implement the procedure for three different significance levels corresponding to 10%,
5%, and 1%. In Table 6, next to each conventional t-statistic, we insert asterisks to indicate
whether the difference in sum/average of coefficients rejects the one-sided null hypothesis of less
than zero based on a 10% (*), 5% (**) or 1% (***) level of bootstrapped significance.13
If one considers the sum of the coefficients equivalent to the Novy-Marx (2012)
specification (the highlighted row in Table 6), the differences in sums of coefficients is
statistically significant at the 5% level even when assessing statistical significance using the

12
There are power advantages to using studentized versions of the test statistic. See Appendix C in Romano and
Wolf (2005).
13
Since the null hypothesis is one-sided, negative differences of coefficients in Table 6 do not show bootstrapped
statistical significance (in other words, our implementation of the bootstrap does not show statistical significance for
the reverse echo).

22
bootstrap. If we adjust for the mechanical effect of unequal number of months in IR and RR
strategies by using the averages instead of sums, the difference is no longer statistically
significant, even at the 10% level. This is true regardless of whether one assess statistical
significance using conventional t-statistics or the Romano and Wolf (2005) bootstrap. Thus, it
seems that Novy-Marx’s (2012) effect derives from the fact that his intermediate return strategy
uses seven months of portfolio formation returns while the recent return strategy uses only six
months of portfolio formation returns.
At the same time, Table 6 shows that there are many definitions in which even the
difference of average intermediate and recent return coefficients is positive and statistically
significant. For instance, if one looks at intermediate returns constructed from months −12 to −6,
and recent returns constructed from months −5 to −2, the difference in average coefficients using
raw returns is 0.47 which is significant at the 5% level using the bootstrap. Indeed, as we noted
before, difference in the averages of coefficients for intermediate and recent return coefficients
are positive for all the 55 definitions in Table 6. It seems that even though Novy-Marx’s (2012)
definition of IR and RR does not lead to statistically significant evidence of an echo, there are
many other definitions of intermediate and recent return that do deliver this evidence, even after
accounting for data snooping biases.
However, there is an obvious and noticeable commonality in all these definitions. Every
recent return definition includes month −2, which as we show in Table 5, has a return response
coefficient that is either negative or indistinguishable from zero. Symmetrically, every
intermediate return definition includes month −12, which we know has a return response
coefficient that is large and positive. This means that averages of coefficients from recent returns
are pushed down by the inclusion of month −2, and averages of coefficients from intermediate
returns are pushed up by the inclusion of month −12.
We formally investigate the influence of these two months in Table 7. This table is
constructed and formatted exactly the same as Table 6, but excludes month −2 in averaging
recent return coefficients. There are 45 possible ways in which one can define intermediate or
recent returns using months −3 to −12. We still include month −12 in computing averages of
intermediate return coefficients, deliberately biasing inferences in favor of intermediate return
definitions. Excluding month −2, virtually none of the intermediate return definitions have higher
average coefficients than those of recent return definitions, either using conventional t-statistics

23
or using bootstrapped levels of significance. A prime example is when intermediate returns are
constructed using months −12 to −8, and recent returns use months −7 to −3 (this row is
highlighted in the table). Despite the fact that the intermediate return coefficients include month
−12, the difference in averages of coefficients is −0.12 using raw returns and −0.16 using three-
factor adjusted returns. The corresponding conventional t-statistics are only −0.68 and −1.18, and
the bootstrap fails to reject the null hypothesis of equality. The few exceptions in Table 7, where
difference in averages of coefficients are statistically significant, all share the common theme of
including month −12 in intermediate return definitions and, therefore, benefit from the 12-month
effect.

4.4. Portfolio Strategies and Reconciling the Evidence

The above analysis of regression coefficients thus strongly suggests that the superior
performance of intermediate return strategies in the U.S. is likely driven by the
underperformance of recent return strategies, which is really due to the inclusion of month −2 in
recent return portfolio formation (the inclusion of month −12 in intermediate return portfolio
formation also has a role to play). If this is indeed the case, then if one constructs recent return
strategies excluding month −2, the difference in future portfolio returns between intermediate
and recent return portfolios should not be different from zero. To check this, we reconstruct the
single- and double-sorted portfolios in Tables 2 and 3 but define months −12 to −8 as
intermediate, and months −7 to −3 as recent. Excluding month −2 also means that both
intermediate and recent return strategies now contain five months in portfolio formation period.
There are two added advantages of portfolio analysis over averaging regression coefficients.
First, portfolios readily deliver economic significance which is not straightforward to infer from
difference in averages of regression coefficients. Second, portfolio analysis allows us to check
the influence of equal- and value-weighting whereas the impact of weighting is not apparent
from regression coefficients.14
Excluding month −2, in single-sorted portfolios, the future return difference between
value-weighted intermediate and recent return portfolios is 0.34% per month with a t-statistic of
1.40. The difference in three-factor alphas is only 0.23% (t-statistic = 0.98). For equal-weighted

14
One can, of course, also calculate portfolio returns based on various definitions of intermediate and recent returns
in Tables 6 and 7. However, it is not possible to control for the mechanical effects of different number of formation
months in intermediate and recent return definitions in portfolio analysis.

24
portfolios, the return difference is 0.13% with a t-statistic of 0.62 and three-factor alpha
difference is −0.02% with t-statistic of −0.07. In double-sorted portfolios, the difference in
returns of value-weighted portfolios is 0.37% with a t-statistic of 1.91 and the difference in three
factor alpha is 0.23% with a t-statistic of 1.20. For equal-weighted double-sorted portfolios, the
return difference is 0.07% (t-statistic = 0.42) and three-factor alpha difference is −0.04% (t-
statistic = −0.14). Clearly, even with the experiment being biased in favor of IR strategies, there
is no statistically significant difference in returns between intermediate and recent return
portfolios, implying that the influence of month −2 is economically large.15
Finally, in Section 3.5 (and Appendix Tables A1–A3), we replicate our ex-U.S. tests for
large stocks because of the especially strong evidence of term structure in large U.S. stocks. We
also conduct an analysis of the sum and averages of coefficients mirroring Tables 6 and 7 for
U.S. large stocks. We use the top quintile of stocks based on NYSE breakpoints to identify large
stocks, estimate the full terms structure regressions, and conduct a similar bootstrap. The results
are reported in Table A4 and parallel those in Tables 6 and 7. Similar to the full sample of
stocks, there are many specifications of intermediate return strategies that have higher average
coefficients than for recent return strategies. However, as before, if one excludes month −2, the
vast majority of differences disappear. Thus, even in large stocks in the U.S., the difference in
intermediate return versus recent return strategies appears to be driven by month −2.

5. Conclusions
Novy-Marx (2012) provides startling evidence that in the U.S., momentum portfolios
formed on the basis of returns from 12 to seven months prior to portfolio formation have
substantially higher profits than momentum portfolios formed based on returns from six to two
months prior. These results, reminiscent of an “echo”, appear remarkably robust, present in
different subperiods, in different groups of stocks, in industries, in size and book-to-market
portfolios. Similar effects appear to exist in currencies, commodities and international equity
indices. What does one make of this? It is hard to imagine a story that could generate such an
15
We also look at the influence of month −2 outside the U.S. by estimating equation (3) for all 37 countries. We do
not report these coefficients to conserve space. Most interesting is the case of Japan, where the coefficient on k = 2 is
−2.65 with a t-statistic of −4.14. As with the U.S., this spillage of short-term reversals from month −1 to month −2
has an important influence on the difference in performance of IR and RR portfolios. In Tables 2 and 3, the future
returns of IR portfolios for Japan were significantly higher than those of RR portfolios. If we exclude month −2
from the recent return portfolio, the return and alpha differences between intermediate and recent return portfolios
are no longer statistically significant for Japan.

25
effect on prices; even the age-old “relative strength” trading strategies of Wall-Street lore having
nothing to say about such an effect. For financial economists, the challenge to theory is
enormous. No existing theory, whether behavioral or rational, predicts an echo in returns.
We conduct out-of-sample and in-sample tests to assess whether this predictability is
genuine. We look for the same echo in 37 countries around the world, and in regions. We do not
find convincing evidence of an echo outside the U.S. Within the U.S., an examination of the full
term structure of returns suggests that the superior performance of strategies based on
intermediate returns is driven by the presence of reversal (or at least no continuation) from
month −2 returns. The true puzzle is not why intermediate returns forecast future returns better
than recent returns, but why return reversals from month −1 also extend (somewhat) to month
−2.

26
Appendix A: Romano and Wolf (2005)

Denote the returns on M strategies by the matrix Rt,m where t = 1, ..., T, and m = 1, ..., M.
The ‘returns’ in our setting are differences of sums or averages of Fama-MacBeth coefficients on
intermediate and recent returns.16 Consider average return, Rm  Tt 1 Rt ,m / T as the basic test
statistic. This test statistic tests a univariate parameter θm. This parameter is defined in such a
way that θm≤ 0 under the null hypothesis (for example, IR does not beat RR under strategy m
using sums of coefficients as in Table 6). Thus, for a given strategy m, the individual testing
problem is:
H m : m  0 vs. H m : m  0
A multiple testing method yields a decision concerning each individual testing problem
by either rejecting Hm or not. A commonly used method for controlling for the Familywise Error
Rate (FWE) is the Bonferroni method. The Bonferroni method, however, is conservative and
results in low power because it does not account for the dependence structure in correlated
strategies. White (2000) provides the first solution to this problem using bootstrap reality check
(BRC). The BRC estimates the asymptotic distribution of max1mM Rm accounting for the
dependence structure of the individual test statistics. It addresses the question of whether the
strategy that appears best in the observed data really beats the benchmark. However, it does not
attempt to identify specific strategies as beating the benchmark or as many outperforming
strategies as possible. The solution to this problem is proposed by Romano and Wolf (2005). The
algorithm to conduct the test is described below.

0. Re-label the strategies according to the size of the individual test statistics (in this case, the
average return), from largest to smallest. Thus, R1  R2   RM .
1. Construct a rectangular joint confidence region for the vector 1 , , M  with nominal joint
coverage probability 1−. The confidence region is of the form [ R1  c1, )   [ RM  c1, ) ,
with a common critical value c1.
a. Estimate the critical value c1 using bootstrap (procedure described below).
b. Reject all the null hypotheses Hm where Rm is bigger than c1.
c. If no null hypothesis is rejected, stop. Else count the number M1 of strategies where
the null is rejected in this step.
2. M−M1 strategies remain. Repeat step 1. Construct a rectangular joint confidence region for
the vector  M1 1 , , M  with nominal joint coverage probability 1−. The confidence region
is of the form [ RM1 1  c2 , )   [ RM  c2 , ) , with a common critical value c2.
a. Estimate the critical value c2 using bootstrap (procedure described below).
b. Reject all the null hypotheses Hm where Rm is bigger than c2.
c. If no null hypothesis is rejected, stop. Else count the number M2 of strategies where
the null is rejected in this step.
3. Repeat the stepwise process until no further hypotheses are rejected.

16
Since we use estimated parameters in the construction of our test statistic, it is not obvious whether one has to re-
estimate these parameters in each simulation draw. However, White (2000) and Romano and Wolf (2005) show that
it is not necessary to re-estimate the parameters for first-order consistency.

27
 The critical values in the above steps are computed using the Politis and Romano (1994)
stationary bootstrap with smoothing parameter of 0.2 to generate B = 10,000 bootstrap data
matrices Rt*,bm , where the star in superscript denotes simulated data and b in the superscript
denotes the bth bootstrap iteration. For each bootstrap iteration b, we compute the test statistics,
R1*b , *b
, RM  *b

and compute max j  max m{Remaining strategies} Rm  Rm for the jth step described in the
*b

previous paragraph. Finally, cj is computed as the 1− empirical percentile of the B values
max*1j , , max*jB .
The algorithm is extended for a studentized test statistic defined as zm  Rm / ˆ m where ˆ m
is the standard error of Rm . For example, null hypothesis m in step j is rejected if Rm  ˆ m  d j  0
where dj is the critical value in the jth step. We modify the computation of critical values as
follows. For each bootstrap iteration b, we compute the test statistics R1*b , , RM*b and the
corresponding standard errors ˆ1*b , ,ˆ M*b . Then we compute the maximum test-statistic
max*jb  max m{Remaining strategies} ( Rm*b  Rm ) / ˆ m*b for the j step. Finally, dj is computed as the 1−
th

empirical percentile of the B values max*1j , , max*jB . The main advantage of a studentized test
statistic is that of improved power (see Appendix C of Romano and Wolf (2005)).
Finally, we can use alphas and their z-statistics, in place of returns and their z-statistics, as
the objects of interest. As mentioned before, White’s (2000) procedure can be viewed as just the
first step of the above algorithm.17 By continuing after the first step in the Romano and Wolf
(2005), more false hypotheses can be rejected.

17
Another approach to multiple testing is that of false discovery rate (FDR). For example, Barras, Scaillet, and
Wermers (2010) use this approach to identify the fraction of skilled mutual fund managers. FDR is defined as an
error rate, or more precisely the expected value of the fraction of falsely rejected null hypotheses, that the researcher
is willing to tolerate. However, our main interest is in precisely identifying which null hypotheses are rejected,
which is not possible with FDR methods.

28
 References

Asness, Clifford S., Tobias J. Moskowitz, and Lasse H. Pedersen, 2009, “Value and Momentum
Everywhere,” Working paper.

Barberis, Nicholas, Andrei Shleifer, Robert Vishny, 1998, “A Model of Investor Sentiment,”
Journal of Financial Economics 49, 307–343.

Barras, Laurent, Olivier Scaillet, and Russ Wermers, 2010, “False Discoveries in Mutual Fund
Performance: Measuring Luck in Estimated Alphas,” Journal of Finance 55, 179–216.

Berk, Jonathan, Richard Green, and Vasant Naik, 1999, “Optimal Investment, Growth Options
and Security Returns,” Journal of Finance 54, 1553–1608.

Brennan, Michael, Tarun Chordia, and Avanidhar Subrahmanyam, 1998, “Alternative Factor
Specifications, Security Characteristics and the Cross-Section of Expected Stock Returns,”
Journal of Financial Economics 49, 345–373.

Campbell, John Y., Sanford J. Grossman, and Jiang Wang, 1993, “Trading Volume and Serial
Correlation in Stock Returns,” Quarterly Journal of Economics 108, 905–939.

Chui, Andy C.W., Sheridan Titman, and K.C. John Wei, 2010, “Individualism and Momentum
Around the World,” Journal of Finance 65, 361–392.

Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam, 1998, “Investor Psychology and
Security Market Under- and Overreactions,” Journal of Finance 53, 1839–1885.

Griffin, John M., Susan Ji, and Spencer Martin, 2003, “Momentum Investing and Business Cycle
Risk: Evidence from Pole to Pole,” Journal of Finance 58, 2515–2547.

Griffin, John M., Patrick J. Kelly, and Federico Nardari, 2010, “Do Market Efficiency Measures
Yield Correct Inferences? A Comparison of Developed and Emerging Markets,” Review of
Financial Studies 23, 3225–3277.

Fama, Eugene F., and Kenneth R. French, 1993, “Common Risk Factors in the Returns on
Stocks and Bonds,” Journal of Financial Economics 33, 3–56.

Fama, Eugene F., and Kenneth R. French, 2012, “Size, Value, and Momentum in International
Stock Returns,” Journal of Financial Economics 105, 457–472.

Fama, Eugene F., and James D. MacBeth, 1973, “Risk, Return, and Equilibrium: Empirical
Tests,” Journal of Political Economy 81, 607–636.

Heston, Steven L., and Ronnie Sadka, 2008, “Seasonality in the Cross-Section of Stock
Returns,” Journal of Financial Economics 87, 418–445.

29
Heston, Steven L., and Ronnie Sadka, 2010, “Seasonality in the Cross Section of Stock Returns:
The International Evidence,” Journal of Financial and Quantitative Analysis 45, 1133–
1160.

Hou, Kewei, G. Andrew Karolyi, and Bong Chan Kho, 2011, “What Factors Drive Global Stock
Returns?” Review of Financial Studies 24, 2527–2574.

Ince, Ozgur S., and R. Burt Porter, 2006, “Individual Equity Return Data from Thomson
Datastream: Handle with Care!” Journal of Financial Research 29, 463–479.

Jegadeesh, Narasimhan, 1990, “Evidence of Predictable Behavior of Security Returns,” Journal

of Finance 45, 881–898.

Jegadeesh, Narasimhan, and Sheridan Titman, 1993, “Returns to Buying Winners and Selling
Losers: Implications for Stock Market Efficiency,” Journal of Finance 48, 65–91.

Jegadeesh, Narasimham, and Sheridan Titman, 1995a, “Short-Horizon Return Reversals and the
Bid-Ask Spread,” Journal of Financial Intermediation 4, 116–132.

Jegadeesh, Narasimham, and Sheridan Titman, 1995b, “Overreaction, Delayed Reaction, and
Contrarian Profits,” Review of Financial Studies 4, 973–993.

Jegadeesh, Narasimhan, and Sheridan Titman, 2001, “Profitability of Momentum Strategies: An

Evaluation of Alternative Explanations,” Journal of Finance 56, 699–720.

Johnson, Timothy C., 2002, “Rational Momentum Effects,” Journal of Finance 57, 585–608.

Leamer, Edward E., 1978, Specification Searches: Ad hoc Inference with Non-Experimental
Data, New York: Wiley.

Leamer, Edward E., 1983, “Let’s Take the Con out of Econometrics,” American Economic
Review 73, 31–43.

Lewellen, Jonathan, 2002, “Momentum and Autocorrelation in Stock Returns,” Review of

Financial Studies 15, 533–563.

Lo, Andrew W., and A. Craig MacKinlay, 1990, “Data-Snooping Biases in Tests of Financial
Asset Pricing Models,” Review of Financial Studies 3, 431–468.

Moskowitz, Tobias J., Yao Hua Ooi, and Lasse H. Pedersen, 2012, “Time Series Momentum,”
Journal of Financial Economics 104, 228–250.

Mlodinow, Leonard, 2008, The Drunkard’s Walk: How Randomness Rules Our Lives, Pantheon
Books.

Novy-Marx, Robert, 2012, “Is Momentum Really Momentum?” Journal of Financial Economics
103, 429–453.

30
Politis, Dimitris, and Joseph P. Romano, 1994, “The Stationary Bootstrap,” Journal of the
American Statistical Association 89, 1303–1313.

Romano, Joseph P., and Michael Wolf, 2005, “Stepwise Multiple Testing as Formalized Data
Snooping,” Econometrica 73, 1237–1282.

Rouwenhorst, Geert K., 1998, “International Momentum Strategies,” Journal of Finance 53,
267–284.

Sagi, Jacob S., and Mark S. Seasholes, 2007, “Firm-Specific Attributes and the Cross-Section of
Momentum,” Journal of Financial Economics 84, 389–434.

White, Halbert L., 2000, “A Reality Check for Data Snooping,” Econometrica 68, 1097–1126.

31
 Table 1
Market Statistics, Returns, and Alphas from Monthly Sorts on
Returns over Prior 11 Months (−12 to −2)

We sort all stocks with valid data in a country into quintiles based on prior returns over months −12 to −2 (we
require at least 100 stocks per month to do the sorts). We calculate value-weighted returns of each quintile, and
report average monthly returns of winner minus loser quintiles. We calculate alphas only if there are at least 60
months of return data. All returns are in US$ and are reported in percent per month (numbers in parenthesis are their
t-statistics). Countries where the returns and/or alphas are statistically significant are shaded. The table also shows
the start date for which portfolios can be constructed for each country, the average number of stocks in each
country-month with valid returns, average market capitalization (at the end of 2010 in US$ million), and aggregate
market capitalization (at the end of 2010 in US$ billion). The sample period is 1980–2010.

Start Average Average Aggregate Portfolio CAPM 3-factor

Date # stocks market cap market cap returns alpha Alpha

Developed Markets
Australia 1980-01 742 503 844 2.43 (4.50) 2.51 (4.63) 2.27 (4.31)
Austria 1980-01 70 1,399 116 2.59 (1.93) –– ––
Belgium 1980-01 108 958 172 1.29 (1.89) 1.48 (2.50) 1.55 (2.55)
Canada 1980-01 831 1,390 1,404 2.17 (4.80) 2.36 (5.31) 2.48 (5.02)
Denmark 1980-01 150 938 163 1.41 (3.26) 1.50 (3.44) 1.43 (3.20)
Finland 1987-01 72 1,567 191 0.97 (1.05) 1.11 (1.21) 0.63 (0.67)
France 1980-01 527 1,729 1,442 0.98 (2.13) 1.14 (2.53) 1.40 (3.08)
Germany 1980-01 506 1,140 1,120 1.46 (3.50) 1.64 (3.99) 1.75 (4.19)
Greece 1980-01 158 220 56 1.35 (2.09) 1.42 (2.22) 1.50 (2.33)
Hong Kong 1980-01 139 3,961 784 0.68 (1.28) 0.92 (1.75) 0.98 (1.84)
Israel 1980-01 351 228 126 1.36 (2.47) 1.50 (2.65) 1.97 (2.04)
Italy 1980-01 175 1,688 467 0.54 (1.24) 0.63 (1.48) 0.52 (1.23)
Japan 1980-01 1,781 1,440 3,316 0.21 (0.59) 0.23 (0.64) 0.10 (0.28)
Netherlands 1980-01 136 3,198 352 0.41 (0.75) 0.74 (1.42) 0.81 (1.48)
New Zealand 1980-01 73 317 38 1.51 (1.65) 1.60 (1.79) 1.53 (1.68)
Norway 1980-01 115 1,024 191 1.36 (2.23) 1.53 (2.54) 1.45 (2.40)
Portugal 1988-01 63 1,269 75 0.80 (0.85) 0.57 (0.61) 0.37 (0.39)
Singapore 1980-01 236 734 354 0.14 (0.29) 0.40 (0.85) 0.45 (0.96)
Spain 1980-01 96 4,688 609 0.65 (1.34) 0.77 (1.59) 0.80 (1.68)
Sweden 1982-01 172 980 459 1.06 (0.94) 1.52 (1.42) 1.59 (1.46)
Switzerland 1980-01 183 4,228 994 1.06 (2.54) 1.21 (2.88) 1.15 (2.64)
UK 1980-01 1,305 1,684 2,278 1.48 (3.68) 1.61 (4.06) 1.91 (4.30)

Emerging Markets
Brazil 1990-01 150 3,127 935 0.85 (0.95) 1.21 (1.35) 1.21 (1.35)
Chile 1989-07 130 1,783 312 0.83 (1.79) 0.87 (1.85) 0.99 (2.12)
China 1991-01 326 2,234 1,934 -0.07 (-0.16) -0.06 (-0.14) 0.53 (1.35)
India 1981-01 1,609 338 1,112 1.36 (2.21) 1.52 (2.27) 1.56 (2.25)
Indonesia 1990-04 176 896 339 -0.13 (-0.16) 0.20 (0.26) 0.90 (1.19)
Korea 1980-01 850 452 804 0.52 (0.93) 0.57 (0.95) 0.97 (1.57)
Malaysia 1980-01 447 422 334 0.02 (0.04) 0.30 (0.55) 0.47 (0.87)
Mexico 1988-01 77 3,393 383 0.97 (1.81) 1.11 (2.09) 1.15 (2.04)
Peru 1991-01 77 606 69 0.26 (0.34) 0.26 (0.33) 0.78 (0.95)
Philippines 1980-01 123 658 144 -0.28 (-0.31) -0.08 (-0.10) -0.04 (-0.05)
Russia 1995-09 59 2,165 794 -2.14 (-1.11) –– ––
South Africa 1980-01 280 1,641 509 1.78 (3.30) 1.71 (2.93) 1.84 (3.06)
Taiwan 1987-09 351 935 712 0.01 (0.03) 0.04 (0.08) 0.15 (0.29)
Thailand 1987-01 277 457 232 0.79 (1.25) 1.00 (1.63) 1.46 (2.24)
Turkey 1988-01 162 898 287 -0.55 (-0.75) -0.39 (-0.53) -0.34 (-0.50)

32
 Table 2
Returns and Three-Factor Alphas from Monthly Sorts on
Recent (−2 to −6) Or Intermediate (−7 to −12) Returns, 1980–2010

We sort all stocks with valid data in a country into quintiles based on prior recent months (−2 to −6, RR) or
intermediate months (−7 to −12, IR) returns. We require at least 100 stocks per month to do the sorts. We calculate
equal- and value-weighted returns of each quintile. Panel A shows average monthly returns and Panel B shows
monthly three-factor alphas of the winner minus loser (WML) quintile for each country. The three-factor alphas are
based on time-series regressions of monthly returns on local country-specific market, SMB and HML factors. We
calculate alphas only if there are at least 60 months of return data. The difference column in each panel shows the
differences in portfolio returns or alphas of the WML quintile based on RR minus those based on IR returns.
Aggregate results for developed and emerging markets are provided in the last three rows for each region. EW refers
to equal-weighting of country returns; VW refers to value-weighting (using total country market capitalization of
last month) of country returns; and Pooled refers to pooling all securities within a region. In each case, factors for
alpha calculation are constructed analogously. All returns are in US$ and are reported in percent per month
(numbers in parenthesis are their t-statistics). The overall sample period is 1980–2010. We follow the same
procedure for the U.S. with the following exceptions. We use deciles, the return breakpoints are based only on
NYSE stocks, and we include stocks with a share price greater than $1 at the end of the portfolio formation period.
Statistics for U.S. are reported for the full sample period of 1927–2010 as well as the international sample period of
1980–2010.

33
 Equal-weighted Value-weighted
RR(WML) IR(WML) Difference RR(WML) IR (WML) Difference
Panel A: Returns
U.S. (1927-2010) 0.66 (3.03) 1.05 (5.80) -0.39 (-1.84) 0.62 (2.66) 1.13 (5.38) -0.51 (-2.11)
U.S. (1980-2010) 1.19 (3.95) 1.20 (5.40) -0.01 (-0.04) 0.56 (1.56) 1.18 (4.08) -0.62 (-1.71)

Developed Markets
Australia 0.69 (1.95) 0.10 (0.34) 0.59 (1.56) 2.72 (5.54) 1.40 (3.53) 1.33 (2.72)
Austria 1.06 (0.99) 1.19 (1.22) -0.13 (-0.12) 0.20 (0.23) 2.40 (2.35) -2.20 (-1.76)
Belgium 1.50 (3.71) 1.22 (3.79) 0.28 (0.87) 1.35 (2.30) 0.68 (1.20) 0.67 (1.12)
Canada 0.83 (2.91) 0.65 (2.42) 0.18 (0.54) 1.68 (4.02) 1.45 (3.90) 0.23 (0.50)
Denmark 1.48 (5.75) 1.05 (3.99) 0.44 (1.67) 1.70 (4.02) 0.91 (2.27) 0.80 (1.45)
Finland 1.34 (2.48) 1.11 (2.27) 0.23 (0.39) 1.45 (1.63) 1.60 (1.92) -0.15 (-0.16)
France 0.96 (3.01) 1.08 (4.21) -0.12 (-0.34) 0.60 (1.42) 1.03 (2.87) -0.43 (-0.87)
Germany 0.91 (3.17) 0.94 (3.67) -0.03 (-0.12) 0.85 (2.05) 1.18 (3.23) -0.33 (-0.70)
Greece 0.84 (1.46) 0.15 (0.29) 0.69 (1.13) 0.86 (1.24) 0.34 (0.59) 0.52 (0.63)
Hong Kong 0.17 (0.35) -0.52 (-1.13) 0.70 (0.99) 0.45 (0.73) 0.33 (0.60) 0.12 (0.14)
Israel -0.18 (-0.61) -0.17 (-0.72) -0.01 (-0.04) 1.45 (2.88) 0.18 (0.40) 1.27 (2.10)
Italy 0.86 (2.47) 0.87 (3.38) -0.01 (-0.02) 0.61 (1.63) 0.76 (2.09) -0.15 (-0.37)
Japan -0.13 (-0.45) 0.36 (1.57) -0.49 (-1.63) -0.02 (-0.04) 0.55 (1.90) -0.57 (-1.50)
Netherlands 1.29 (3.88) 1.53 (4.99) -0.24 (-0.73) 0.64 (1.19) 0.79 (1.68) -0.15 (-0.24)
New Zealand 0.81 (1.53) 1.21 (2.09) -0.40 (-0.56) 0.51 (0.67) 1.06 (1.39) -0.54 (-0.60)
Norway 1.68 (3.65) 1.69 (3.55) -0.01 (-0.02) 1.26 (1.99) 1.74 (2.88) -0.49 (-0.66)
Portugal -1.38 (-1.57) -0.43 (-0.52) -0.95 (-0.74) 0.26 (0.28) 0.69 (0.81) -0.43 (-0.38)
Singapore 0.26 (0.61) -0.04 (-0.11) 0.31 (0.66) 0.23 (0.49) 0.25 (0.61) -0.01 (-0.02)
Spain 0.82 (2.34) 1.01 (3.18) -0.19 (-0.52) 0.62 (1.34) 0.63 (1.54) -0.01 (-0.02)
Sweden 2.11 (3.04) 0.91 (1.59) 1.20 (1.58) 1.14 (1.07) 0.46 (0.64) 0.68 (0.61)
Switzerland 1.05 (3.56) 1.13 (4.27) -0.08 (-0.29) 0.54 (1.35) 0.91 (2.39) -0.37 (-0.89)
UK 1.40 (6.32) 1.10 (6.10) 0.30 (1.55) 0.75 (2.02) 1.00 (3.14) -0.25 (-0.68)
Dev. (EW) 0.79 (5.02) 0.65 (4.79) 0.16 (1.03) 0.88 (4.01) 0.79 (4.37) 0.09 (0.42)
Dev. (VW) 0.52 (2.64) 0.64 (4.09) -0.11 (-0.55) 0.43 (1.57) 0.78 (3.70) -0.35 (-1.28)
Dev. (Pooled) 0.82 (3.40) 0.79 (3.71) 0.03 (0.11) 0.55 (1.59) 0.91 (3.13) -0.36 (-0.94)

Emerging Markets
Brazil -1.06 (-2.32) 0.25 (0.54) -1.31 (-2.02) -0.45 (-0.89) 0.28 (0.52) -0.72 (-1.06)
Chile 1.38 (4.63) 0.56 (2.06) 0.82 (2.51) 1.19 (2.60) 0.21 (0.45) 0.98 (1.66)
China -0.04 (-0.12) 0.25 (0.82) -0.29 (-0.80) 0.11 (0.27) 0.56 (1.45) -0.45 (-0.92)
India -1.21 (-2.52) -0.01 (-0.02) -1.21 (-2.24) 1.59 (2.82) 1.16 (2.16) 0.43 (0.75)
Indonesia -1.11 (-1.60) -1.18 (-1.81) 0.06 (0.09) 0.11 (0.12) 0.19 (0.30) -0.08 (-0.08)
Korea -0.41 (-0.65) 0.03 (0.06) -0.44 (-0.77) -0.01 (-0.03) 0.42 (0.84) -0.43 (-0.70)
Malaysia 0.18 (0.42) 0.03 (0.09) 0.15 (0.31) 0.27 (0.55) 0.12 (0.30) 0.15 (0.29)
Mexico 0.72 (1.60) 0.41 (0.93) 0.31 (0.59) 0.19 (0.36) 1.07 (2.20) -0.89 (-1.34)
Peru -0.40 (-0.69) -0.44 (-0.72) 0.04 (0.05) 0.45 (0.59) -0.04 (-0.05) 0.49 (0.51)
Philippines -1.17 (-1.35) -0.64 (-1.05) -0.53 (-0.66) -0.32 (-0.35) -0.19 (-0.30) -0.14 (-0.14)
Russia -0.66 (-0.57) -2.58 (-2.17) 1.92 (1.85) 0.44 (0.36) -1.65 (-1.20) 2.09 (1.21)
South Africa 1.41 (4.36) 1.27 (4.88) 0.14 (0.36) 2.00 (4.06) 1.62 (3.39) 0.37 (0.63)
Taiwan -0.72 (-1.55) 0.36 (0.95) -1.08 (-1.97) -0.44 (-0.91) 0.55 (1.16) -1.00 (-1.42)
Thailand -0.14 (-0.20) 0.00 (0.00) -0.14 (-0.21) 0.43 (0.63) 0.42 (0.77) 0.01 (0.01)
Turkey -1.22 (-2.46) -0.52 (-1.15) -0.70 (-1.08) -0.97 (-1.29) 0.86 (1.24) -1.83 (-1.71)
Emg. (EW) -0.30 (1.26) 0.08 (0.48) -0.37 (1.41) 0.28 (1.08) 0.43 (2.16) -0.11 (-0.35)
Emg. (VW) -0.13 (-0.61) 0.28 (1.73) -0.40 (-1.56) 0.28 (1.16) 0.54 (2.76) -0.22 (-0.76)
Emg.(Pooled) -0.05 (-0.10) 0.47 (1.00) -0.52 (-0.86) 0.48 (0.88) 0.59 (1.17) -0.10 (-0.16)

34
 Equal-weighted Value-weighted
RR(WML) IR(WML) Difference RR(WML) IR (WML) Difference
Panel B: Three-factor alphas
U.S.(1927-2010) 1.17 (6.14) 1.39 (8.37) -0.22 (-1.06) 1.11 (5.25) 1.51 (7.78) -0.40 (-1.67)
U.S.(1980-2010) 1.45 (4.80) 1.38 (6.19) 0.07 (0.27) 0.84 (2.35) 1.47 (5.19) -0.63 (-1.74)

Developed Markets
Australia 0.79 (2.15) 0.50 (1.72) 0.29 (0.73) 2.51 (5.13) 1.65 (4.04) 0.86 (1.71)
Austria –– –– –– –– –– ––
Belgium 1.45 (3.71) 1.21 (3.84) 0.24 (0.74) 1.58 (2.96) 0.87 (1.64) 0.71 (1.17)
Canada 0.94 (3.03) 0.92 (3.26) 0.02 (0.05) 2.02 (4.37) 1.61 (3.99) 0.41 (0.80)
Denmark 1.53 (5.85) 1.12 (4.11) 0.41 (1.55) 1.64 (3.79) 0.93 (2.25) 0.71 (1.29)
Finland 1.27 (2.29) 1.11 (2.29) 0.17 (0.28) 1.34 (1.49) 1.29 (1.54) 0.05 (0.05)
France 1.15 (3.59) 1.20 (4.73) -0.05 (-0.14) 0.81 (1.96) 1.13 (3.13) -0.33 (-0.66)
Germany 1.00 (3.59) 1.05 (4.19) -0.05 (-0.17) 1.03 (2.44) 1.39 (3.77) -0.36 (-0.75)
Greece 0.99 (1.72) 0.34 (0.68) 0.64 (1.05) 1.04 (1.52) 0.55 (0.99) 0.48 (0.59)
Hong Kong 0.24 (0.49) -0.13 (-0.29) 0.37 (0.55) 0.67 (1.10) 0.65 (1.17) 0.02 (0.02)
Israel -0.36 (-0.54) 0.69 (1.28) -1.05 (-1.82) 1.43 (1.68) 1.40 (1.79) 0.03 (0.04)
Italy 0.74 (2.24) 0.81 (3.27) -0.06 (-0.20) 0.56 (1.53) 0.74 (2.08) -0.18 (-0.45)
Japan -0.23 (-0.79) 0.46 (2.05) -0.68 (-2.21) -0.20 (-0.54) 0.58 (2.01) -0.77 (-1.99)
Netherlands 1.39 (4.22) 1.70 (5.58) -0.31 (-0.92) 0.89 (1.67) 1.08 (2.25) -0.19 (-0.29)
New Zealand 0.89 (1.66) 1.11 (1.99) -0.23 (-0.32) 0.45 (0.57) 0.98 (1.27) -0.53 (-0.58)
Norway 1.79 (3.91) 1.88 (3.95) -0.09 (-0.17) 1.29 (2.01) 1.93 (3.19) -0.64 (-0.86)
Portugal –– –– –– –– –– ––
Singapore 0.48 (1.15) 0.25 (0.68) 0.23 (0.49) 0.34 (0.73) 0.57 (1.46) -0.23 (-0.40)
Spain 0.99 (2.92) 1.08 (3.47) -0.10 (-0.27) 0.82 (1.80) 0.75 (1.86) 0.07 (0.12)
Sweden 2.16 (3.20) 0.91 (1.78) 1.25 (1.63) 1.51 (1.45) 0.64 (0.90) 0.86 (0.77)
Switzerland 1.09 (3.55) 1.22 (4.46) -0.13 (-0.42) 0.71 (1.69) 1.04 (2.61) -0.33 (-0.75)
UK 1.61 (6.42) 1.13 (5.55) 0.48 (2.25) 1.10 (2.66) 1.23 (3.33) -0.13 (-0.31)
Dev. (EW) 0.83 (5.03) 0.81 (5.71) 0.02 (0.14) 0.91 (4.00) 1.08 (5.33) -0.09 (-0.39)
Dev. (VW) 0.51 (2.48) 0.79 (5.01) -0.28 (-1.36) 0.33 (1.14) 0.93 (4.22) -0.60 (-2.06)
Dev. (Pooled) 0.69 (2.73) 0.69 (3.12) 0.00 (0.01) 0.47 (1.28) 0.77 (2.51) -0.30 (-0.74)

Emerging Markets
Brazil -0.86 (-1.87) 0.20 (0.42) -1.06 (-1.60) -0.22 (-0.44) 0.36 (0.67) -0.57 (-0.82)
Chile 1.37 (4.64) 0.57 (2.11) 0.80 (2.42) 1.07 (2.49) 0.65 (1.58) 0.41 (0.81)
China 0.22 (0.67) 0.81 (3.01) -0.59 (-1.67) 0.28 (0.67) 1.08 (3.02) -0.80 (-1.64)
India -1.12 (-2.09) 0.35 (0.72) -1.47 (-2.49) 1.74 (2.73) 1.27 (2.11) 0.48 (0.76)
Indonesia -0.34 (-0.53) -0.11 (-0.25) -0.22 (-0.33) 1.20 (1.51) 0.76 (1.26) 0.43 (0.47)
Korea 0.22 (0.35) 0.69 (1.32) -0.47 (-0.78) 0.33 (0.57) 0.95 (1.73) -0.62 (-0.93)
Malaysia 0.49 (1.12) 0.22 (0.63) 0.28 (0.57) 0.70 (1.41) 0.35 (0.91) 0.34 (0.64)
Mexico 0.77 (1.63) 0.55 (1.25) 0.22 (0.38) 0.19 (0.35) 1.32 (2.57) -1.13 (-1.59)
Peru 0.01 (0.02) 0.45 (0.71) -0.44 (-0.47) 0.04 (0.06) 0.41 (0.52) -0.36 (-0.36)
Philippines -1.08 (-1.34) -0.34 (-0.61) -0.74 (-0.92) -0.08 (-0.09) 0.01 (0.02) -0.09 (-0.09)
Russia –– –– –– –– –– ––
South Africa 1.17 (3.27) 1.08 (3.93) 0.08 (0.19) 2.01 (3.68) 1.65 (3.11) 0.35 (0.52)
Taiwan -0.44 (-0.93) 0.36 (0.89) -0.80 (-1.36) -0.25 (-0.47) 0.57 (1.20) -0.81 (-1.16)
Thailand 0.91 (1.37) 0.20 (0.45) 0.71 (1.12) 1.20 (1.79) 0.94 (1.62) 0.26 (0.36)
Turkey -0.99 (-2.20) -0.26 (-0.59) -0.73 (-1.18) -0.96 (-1.27) 0.86 (1.33) -1.83 (-1.70)
Emg. (EW) 0.08 (0.34) 0.20 (1.09) -0.12 (-0.48) 0.57 (2.25) 0.60 (2.83) -0.03 (-0.10)
Emg. (VW) 0.07 (0.32) 0.27 (1.53) -0.20 (-0.82) 0.42 (1.82) 0.61 (2.90) -0.19 (-0.68)
Emg.(Pooled) -0.26 (-0.45) 0.66 (1.39) -0.92 (-1.49) 0.60 (1.08) 0.59 (1.19) 0.01 (0.01)

35
 Table 3
Differences in Returns and Three-Factor Alphas from Monthly Double Sorts on
Recent (−2 to −6) and Intermediate (−7 to −12) Returns, 1980–2010

We sort all stocks with valid data in a country independently into terciles based on prior recent months (−2 to −6,
RR) or intermediate months (−7 to −12, IR) returns. We require at least 100 stocks per month to do the sorts. We
calculate equal- and value-weighted returns of each of these nine (3×3) portfolios. Recent return losers are labeled
RRL and recent return winners are labeled RRW; Intermediate return losers are labeled IRL and intermediate return
losers are labeled IRW. Winner minus loser portfolios are denoted by WML. RR(WML)/IRW denotes winner
recent-return minus loser recent-return portfolio of stocks in the winner intermediate-return tercile portfolio.
Similarly, IR(WML)/RRL denotes the winner intermediate-return minus loser intermediate-return portfolio of stocks
in the winner recent-return tercile portfolio. The table finally reports statistics on the difference in portfolio returns
of RR(WML)/IRW and IR(WML)/RRW. We report average monthly returns and monthly three-factor alphas of
these portfolios. The three-factor alphas are based on time series regressions of monthly returns on local country-
specific market, SMB and HML factors. We calculate alphas only if there are at least 60 months of return data.
Aggregate results for developed and emerging markets are provided in the last three rows for each region. EW refers
to equal-weighting of country returns; VW refers to value-weighting (using total country market capitalization of
last month) of country returns; and Pooled refers to pooling all securities within a region. All returns are in US$ and
are reported in percent per month (numbers in parenthesis are their t-statistics). The overall sample period is 1980–
2010. We follow the same procedure for the U.S. with the following exceptions. We use quintiles, return
breakpoints are based only on NYSE stocks, and we include stocks with a share price greater than $1 at the end of
the portfolio formation period. Statistics for U.S. are reported for the full sample period of 1927–2010 as well as the
international sample period of 1980–2010.

36
 Equal-weighted Value-weighted
Portfolio returns Three-factor alphas Portfolio returns Three-factor alphas
U.S. (1927-2010) -0.29 (-1.60) -0.19 (-1.06) -0.48 (-2.25) -0.36 (-1.73)
U.S. (1980-2010) -0.13 (-0.59) -0.09 (-0.43) -0.75 (-2.48) -0.85 (-2.86)

Developed Markets
Australia 0.30 (0.89) 0.22 (0.68) 0.78 (1.79) 0.89 (2.04)
Austria -0.71 (-0.46) –– -1.47 (-1.05) ––
Belgium -0.34 (-0.88) -0.26 (-0.67) -0.47 (-1.06) -0.35 (-0.79)
Canada 0.23 (0.85) 0.08 (0.29) 0.42 (1.11) 0.35 (0.86)
Denmark 0.30 (1.07) 0.35 (1.28) 0.35 (0.78) 0.32 (0.71)
Finland -0.57 (-0.95) -0.53 (-0.85) -0.53 (-0.77) -0.61 (-0.86)
France -0.35 (-1.24) -0.33 (-1.18) -0.45 (-1.24) -0.41 (-1.12)
Germany -0.07 (-0.31) -0.06 (-0.25) -0.60 (-1.52) -0.60 (-1.48)
Greece 0.76 (1.44) 0.67 (1.29) 0.88 (1.21) 0.77 (1.07)
Hong Kong 0.80 (1.45) 0.48 (0.92) 0.38 (0.60) 0.21 (0.34)
Israel 0.04 (0.15) -1.05 (-1.89) 1.10 (2.21) 0.46 (0.65)
Italy 0.11 (0.42) 0.10 (0.35) -0.67 (-1.80) -0.73 (-1.99)
Japan -0.45 (-1.95) -0.57 (-2.44) -0.54 (-1.98) -0.66 (-2.38)
Netherlands -0.46 (-1.43) -0.37 (-1.09) 0.01 (0.03) 0.03 (0.05)
New Zealand -0.23 (-0.28) 0.00 (0.00) -0.16 (-0.19) -0.20 (-0.23)
Norway 0.00 (-0.01) -0.05 (-0.10) -0.94 (-1.57) -0.93 (-1.52)
Portugal -1.37 (-1.53) –– -0.10 (-0.11) ––
Singapore 0.09 (0.24) 0.15 (0.40) -0.37 (-0.84) -0.37 (-0.83)
Spain -0.44 (-1.26) -0.41 (-1.16) -0.47 (-1.07) -0.53 (-1.18)
Sweden 0.58 (0.89) 0.78 (1.19) 0.22 (0.27) 0.25 (0.31)
Switzerland -0.48 (-1.73) -0.50 (-1.74) -1.07 (-2.85) -0.90 (-2.29)
UK 0.24 (1.33) 0.42 (2.11) -0.23 (-0.81) -0.06 (-0.20)
Dev. (EW) 0.00 (0.03) -0.10 (-0.72) -0.06 (-0.37) -0.18 (-1.05)
Dev. (VW) -0.18 (-1.22) -0.31 (-1.96) -0.42 (-2.12) -0.58 (-2.77)
Dev. (Pooled) 0.04 (0.19) 0.01 (0.05) -0.40 (-1.45) -0.37 (-1.28)

Emerging Markets
Brazil -1.10 (-1.44) -0.88 (-1.14) -0.66 (-0.90) -0.56 (-0.75)
Chile 0.54 (1.51) 0.62 (1.84) 1.14 (2.19) 0.79 (1.73)
China -0.28 (-0.92) -0.51 (-1.65) -0.51 (-1.33) -0.80 (-2.05)
India -0.95 (-2.44) -1.18 (-2.74) 0.06 (0.13) 0.00 (-0.00)
Indonesia -0.26 (-0.46) -0.51 (-0.92) -0.27 (-0.38) 0.18 (0.27)
Korea -0.32 (-0.75) -0.38 (-0.79) -0.07 (-0.15) -0.07 (-0.14)
Malaysia 0.00 (0.00) 0.11 (0.26) 0.21 (0.45) 0.34 (0.70)
Mexico 0.16 (0.32) 0.10 (0.18) -0.41 (-0.68) -0.67 (-1.08)
Peru 0.12 (0.15) -0.24 (-0.27) -0.03 (-0.04) -0.17 (-0.19)
Philippines -0.66 (-0.89) -0.85 (-1.16) -0.54 (-0.72) -0.58 (-0.76)
Russia 0.87 (0.84) –– 1.11 (0.71) ––
South Africa 0.11 (0.30) -0.04 (-0.10) -0.02 (-0.04) -0.14 (-0.25)
Taiwan -0.60 (-1.46) -0.32 (-0.72) -0.40 (-0.75) -0.09 (-0.15)
Thailand -0.24 (-0.45) 0.30 (0.61) -0.11 (-0.20) 0.21 (0.36)
Turkey -0.31 (-0.60) -0.40 (-0.84) -1.71 (-2.06) -1.81 (-2.16)
Emg. (EW) -0.26 (-1.27) -0.15 (-0.73) -0.17 (-0.73) -0.03 (-0.14)
Emg. (VW) -0.32 (-1.64) -0.24 (-1.26) -0.21 (-0.92) -0.08 (-0.38)
Emg. (Pooled) -0.57 (-1.22) -0.84 (-1.77) -0.53 (-1.21) -0.34 (-0.79)

37
 Table 4
Monthly Fama-MacBeth Regressions, 1980–2010

We run Fama and MacBeth (1973) regressions each month for each country. ME is market capitalization, BM is
book-to-market, R(−1) is prior month return, RR(−2 to −6) is cumulative return over recent months (−2 to −6), and
IR(−7 to −12) is cumulative return over intermediate months (−7 to −12). The difference column in each panel
shows the differences in coefficients on RR and IR. In Panel A, we use excess returns (raw returns minus the risk
free rate) as the dependent variable:

In Panel B, the dependent variable is Fama and French (1993) three-factor model adjusted returns (the factors are
country-specific, and the betas on factors are calculated over the entire sample period):

The table also shows the average number of stocks as well as the number of months used in the estimation of
regressions for each country. Regressions are not estimated if the number of stocks in a month is less than 100. If
there are less than 5 years of regression coefficients for a country, estimates are not reported. Aggregate results for
developed and emerging markets are provided in the last three rows for each region. EW refers to equal-weighting
of country coefficients; VW refers to value-weighting (using total country market capitalization of last month) of
country coefficients; and Pooled refers to pooling all securities within a region. All returns and independent
variables are in US$. All coefficients are multiplied by 100 and t-statistics, reported in parenthesis, are based on
Newey-West procedures with three lags. The overall sample period is 1980–2010. Statistics for U.S. are reported for
the full sample period of 1927–2010 as well as the international sample period of 1980–2010.

38
 Ln ME Ln BM R(−1) RR(−2 to −6) IR(−7 to −12) Difference # #
stocks months
Panel A: Excess returns
U.S. (1927-2010) -0.08 (-2.18) 0.29 (4.62) -6.31 (-16.4) 0.81 (3.63) 1.08 (6.57) -0.27 (-1.15) 2,369 1,007
U.S. (1980-2010) 0.01 (0.14) 0.42 (4.23) -2.95 (-7.82) 0.85 (3.30) 0.86 (4.52) -0.02 (-0.07) 4,340 371

Developed Markets
Australia -0.29 (-3.88) 0.28 (2.17) -4.04 (-4.27) 1.93 (4.15) 1.13 (2.89) 0.80 (1.48) 646 246
Austria – – – – – – – – – – – – – –
Belgium 0.03 (0.50) 0.36 (3.30) -6.03 (-3.86) 3.03 (3.00) 0.74 (0.78) 2.29 (3.93) 125 132
Canada -0.21 (-3.65) 0.13 (1.15) -1.86 (-2.60) 1.72 (3.99) 0.88 (2.52) 0.84 (1.90) 456 306
Denmark 0.02 (0.41) 0.27 (1.76) -2.72 (-3.02) 2.56 (5.27) 1.43 (2.65) 1.14 (2.06) 162 246
Finland -0.09 (-0.87) 0.35 (1.61) -2.75 (-1.82) 1.53 (2.22) 1.15 (1.49) 0.38 (0.43) 122 138
France -0.01 (-0.13) 0.40 (3.73) -8.22 (-10.1) 1.23 (2.48) 1.32 (3.65) -0.09 (-0.22) 542 258
Germany 0.01 (0.23) 0.33 (3.17) -4.04 (-5.56) 1.37 (3.31) 1.73 (4.23) -0.36 (-0.73) 476 270
Greece -0.41 (-1.49) 0.17 (0.94) -3.58 (-2.72) 0.22 (0.25) 0.37 (0.57) -0.15 (-0.20) 232 174
Hong Kong -0.22 (-1.56) 0.18 (0.84) -2.18 (-1.33) 0.36 (0.44) -0.65 (-0.60) 1.01 (0.92) 166 174
Israel – – – – – – – – – – – – – –
Italy 0.04 (0.73) 0.33 (2.88) -2.81 (-2.63) 1.19 (1.91) 1.69 (4.03) -0.50 (-0.81) 192 258
Japan -0.05 (-0.72) 0.44 (5.56) -6.29 (-7.81) -0.94 (-2.03) 0.34 (0.89) -1.27 (-2.60) 1,551 354
Netherlands 0.06 (1.15) 0.22 (1.85) -0.93 (-0.84) 2.01 (3.62) 2.25 (4.09) -0.24 (-0.39) 135 234
New Zealand 0.05 (0.39) 0.04 (0.22) -5.65 (-2.52) -0.29 (-0.26) 0.72 (0.68) -1.00 (-0.68) 111 64
Norway -0.04 (-0.51) 0.07 (0.37) -2.54 (-2.31) 2.67 (4.53) 1.28 (1.67) 1.39 (1.73) 154 162
Portugal – – – – – – – – – – – – – –
Singapore -0.16 (-1.40) 0.33 (2.32) -7.38 (-5.98) -0.40 (-0.31) -0.33 (-0.36) -0.07 (-0.05) 303 186
Spain 0.00 (0.05) 0.25 (2.08) -1.29 (-0.90) 1.65 (2.76) 1.41 (2.09) 0.24 (0.34) 128 162
Sweden 0.02 (0.16) 0.30 (1.13) -4.82 (-3.53) 1.63 (2.22) 1.12 (1.94) 0.51 (0.70) 297 138
Switzerland 0.02 (0.51) 0.27 (3.43) -2.81 (-2.97) 2.26 (3.84) 1.61 (3.07) 0.65 (1.08) 184 251
UK -0.04 (-0.81) 0.23 (2.85) -1.88 (-2.99) 1.41 (3.86) 1.05 (3.82) 0.36 (1.11) 960 354
Dev. (EW) -0.09 (-2.95) 0.29 (5.04) -4.20 (-9.40) 1.13 (3.77) 1.08 (4.07) 0.05 (0.18) – 354
Dev. (VW) -0.07 (-1.65) 0.37 (6.00) -5.04 (-10.14) 0.26 (0.78) 0.76 (2.66) -0.50 (-1.51) – 354
Dev. (Pooled) -0.11 (-2.89) 0.31 (3.53) -1.88 (-2.99) 0.90 (2.63) 0.94 (3.05) -0.04 (-0.11) 6,340 318

Emerging Markets
Brazil -0.21 (-1.84) -0.03 (-0.31) -5.49 (-3.53) 0.57 (0.77) 0.32 (0.60) 0.25 (0.23) 131 68
Chile 0.06 (0.89) 0.38 (2.67) 2.10 (1.55) 1.98 (2.92) 0.07 (0.13) 1.91 (2.33) 167 138
China -0.26 (-1.29) 0.24 (1.68) -6.23 (-5.73) -1.12 (-1.46) 0.59 (1.07) -1.72 (-2.35) 594 126
India -0.28 (-2.37) 0.14 (0.83) -6.06 (-5.70) 0.80 (1.82) 0.71 (1.36) 0.09 (0.22) 680 198
Indonesia -0.36 (-2.30) 0.59 (2.81) -2.87 (-2.40) -1.77 (-1.59) -0.81 (-0.86) -0.96 (-0.93) 236 174
Korea -0.09 (-0.55) 0.65 (4.35) -6.39 (-5.80) -0.47 (-0.95) -0.16 (-0.29) -0.31 (-0.48) 723 210
Malaysia -0.06 (-0.42) 0.54 (4.51) -5.83 (-4.57) -0.48 (-0.37) -0.16 (-0.30) -0.31 (-0.27) 504 222
Mexico 0.14 (0.89) 0.40 (1.39) -0.58 (-0.24) 0.43 (0.27) -1.04 (-0.92) 1.47 (1.03) 104 76
Peru – – – – – – – – – – – – – –
Philippines -0.11 (-1.06) 0.64 (3.29) -10.60 (-6.83) -4.15 (-2.24) -0.50 (-0.45) -3.65 (-1.87) 163 160
Russia – – – – – – – – – – – – – –
South Africa -0.23 (-3.42) 0.57 (4.51) -7.97 (-6.78) 0.96 (2.24) 0.55 (1.42) 0.41 (0.75) 244 213
Taiwan -0.03 (-0.26) 0.28 (0.94) -1.85 (-2.00) -0.75 (-0.88) -0.17 (-0.26) -0.58 (-0.61) 488 186
Thailand -0.13 (-1.08) 0.54 (3.05) -1.61 (-1.33) 0.21 (0.22) 0.42 (0.85) -0.21 (-0.22) 313 210
Turkey -0.09 (-0.65) 0.35 (2.11) -2.94 (-1.82) -1.63 (-1.85) -0.72 (-1.35) -0.91 (-1.24) 206 138
Emg. (EW) -0.19 (-3.40) 0.44 (6.12) -4.18 (-6.32) -0.25 (-0.42) -0.17 (-0.44) -0.09 (-0.15) – 222
Emg. (VW) -0.20 (-3.32) 0.40 (4.66) -3.56 (-5.62) 0.09 (0.23) 0.01 (0.03) 0.08 (0.17) – 222
Emg. (Pooled) -0.18 (-3.48) 0.45 (3.66) -1.57 (-1.92) -0.47 (-0.72) 0.33 (0.70) -0.80 (-1.38) 4,551 198

39
 Ln ME Ln BM R(−1) RR(−2 to −6) IR(−7 to −12) Difference # #
stocks months
Panel B: Three-factor adjusted returns
U.S. (1927-2010) -0.05 (-2.41) 0.20 (5.57) -6.84 (-19.6) 0.75 (4.71) 0.98 (8.36) -0.23 (-1.33) 2,344 1,007
U.S. (1980-2010) -0.02 (-0.50) 0.34 (6.32) -3.76 (-10.8) 0.72 (3.72) 0.73 (5.30) -0.01 (-0.08) 4,254 371

Developed Markets
Australia -0.19 (-3.11) 0.40 (3.85) -4.51 (-5.13) 1.48 (3.70) 1.06 (3.23) 0.42 (0.84) 667 234
Austria – – – – – – – – – – – – – –
Belgium -0.03 (-0.68) 0.21 (1.86) -6.71 (-4.63) 2.72 (3.29) 0.71 (1.04) 2.01 (3.69) 124 132
Canada -0.13 (-2.87) 0.19 (1.67) -2.27 (-3.29) 1.68 (4.32) 0.98 (3.37) 0.69 (1.72) 455 306
Denmark -0.01 (-0.34) 0.12 (0.91) -2.62 (-3.00) 2.44 (5.55) 1.42 (3.01) 1.02 (1.97) 161 246
Finland -0.11 (-1.14) 0.17 (1.12) -3.62 (-2.45) 1.61 (2.51) 0.80 (1.12) 0.81 (0.96) 122 137
France -0.06 (-2.41) 0.27 (3.52) -8.63 (-11.7) 1.06 (2.40) 1.32 (4.91) -0.25 (-0.69) 539 258
Germany -0.03 (-1.08) 0.23 (2.84) -4.59 (-6.66) 1.08 (3.02) 1.58 (4.86) -0.50 (-1.12) 475 270
Greece -0.12 (-1.20) 0.13 (0.90) -4.20 (-3.66) 0.02 (0.02) 0.43 (0.93) -0.41 (-0.63) 231 174
Hong Kong -0.14 (-1.58) 0.15 (0.78) -3.57 (-2.40) 0.22 (0.33) -0.23 (-0.29) 0.45 (0.55) 166 174
Israel – – – – – – – – – – – – – –
Italy -0.04 (-1.31) 0.20 (2.14) -3.18 (-3.27) 1.23 (2.21) 1.55 (4.28) -0.32 (-0.59) 191 258
Japan 0.05 (1.78) 0.34 (4.78) -6.90 (-9.22) -0.99 (-2.46) 0.11 (0.36) -1.11 (-2.86) 1,550 354
Netherlands 0.02 (0.54) 0.11 (1.21) -1.59 (-1.47) 1.68 (3.25) 2.07 (4.37) -0.39 (-0.62) 135 234
New Zealand – – – – – – – – – – – – – –
Norway -0.08 (-1.85) 0.04 (0.24) -3.00 (-2.79) 2.31 (4.71) 1.45 (2.34) 0.86 (1.17) 150 162
Portugal – – – – – – – – – – – – – –
Singapore -0.04 (-1.03) 0.28 (2.16) -8.43 (-7.15) -0.42 (-0.43) -0.06 (-0.10) -0.36 (-0.40) 301 186
Spain 0.00 (-0.06) 0.24 (2.13) -1.74 (-1.35) 1.26 (2.28) 1.35 (2.33) -0.09 (-0.12) 128 162
Sweden -0.04 (-0.55) 0.31 (1.85) -5.84 (-4.78) 1.16 (1.77) 1.06 (2.31) 0.10 (0.14) 292 137
Switzerland -0.01 (-0.24) 0.23 (3.06) -3.14 (-3.59) 2.10 (4.00) 1.61 (3.80) 0.48 (0.89) 183 251
UK -0.07 (-2.27) 0.11 (1.89) -2.02 (-3.34) 1.06 (2.98) 0.98 (4.35) 0.08 (0.27) 1,075 306
Dev. (EW) -0.03 (-1.53) 0.29 (5.21) -4.48 (-10.37) 0.92 (3.22) 0.96 (4.36) -0.05 (-0.17) – 354
Dev. (VW) 0.01 (0.23) 0.34 (5.70) -5.53 (-11.75) 0.09 (0.29) 0.61 (2.55) -0.52 (-1.73) – 354
Dev. (Pooled) -0.05 (-2.74) 0.23 (3.54) -2.91 (-5.51) 0.78 (2.61) 1.00 (4.39) -0.22 (-0.71) 6,309 318

Emerging Markets
Brazil -0.28 (-2.83) -0.02 (-0.22) -5.71 (-3.35) 0.38 (0.58) 0.16 (0.31) 0.22 (0.22) 129 68
Chile -0.04 (-0.85) 0.26 (2.51) 1.37 (1.20) 1.62 (2.53) -0.03 (-0.07) 1.65 (2.09) 167 138
China -0.03 (-0.36) 0.12 (1.26) -6.75 (-6.32) -1.48 (-1.91) 0.43 (0.83) -1.91 (-2.73) 594 126
India -0.18 (-2.24) 0.08 (0.53) -6.24 (-6.18) 0.67 (1.81) 0.89 (2.23) -0.22 (-0.63) 675 198
Indonesia -0.24 (-2.32) 0.41 (3.03) -4.21 (-3.85) -1.40 (-1.75) 0.11 (0.20) -1.51 (-1.69) 233 174
Korea -0.04 (-0.46) 0.62 (4.94) -6.56 (-6.26) -0.20 (-0.37) -0.02 (-0.05) -0.17 (-0.27) 717 210
Malaysia 0.01 (0.15) 0.43 (4.10) -6.68 (-5.52) -0.40 (-0.39) -0.47 (-1.03) 0.06 (0.07) 503 222
Mexico 0.16 (1.64) 0.38 (1.65) -2.67 (-1.41) 0.46 (0.35) -0.96 (-1.17) 1.42 (1.15) 103 74
Peru – – – – – – – – – – – – – –
Philippines -0.07 (-1.09) 0.60 (3.78) -11.60 (-8.45) -4.05 (-3.20) -0.80 (-1.18) -3.25 (-2.64) 163 160
Russia – – – – – – – – – – – – – –
South Africa -0.24 (-4.51) 0.41 (3.59) -8.27 (-9.55) 0.88 (2.08) 0.71 (1.93) 0.17 (0.30) 243 210
Taiwan 0.01 (0.09) 0.15 (0.83) -2.63 (-2.76) -0.38 (-0.55) -0.31 (-0.61) -0.07 (-0.10) 486 186
Thailand -0.11 (-1.73) 0.35 (3.04) -3.03 (-3.06) 0.66 (1.06) 0.49 (1.32) 0.17 (0.27) 312 210
Turkey -0.11 (-1.02) 0.24 (1.60) -2.30 (-1.55) -1.22 (-1.61) -0.76 (-1.77) -0.46 (-0.77) 205 138
Emg. (EW) -0.11 (-3.83) 0.34 (6.22) -5.20 (-9.33) -0.14 (-0.31) -0.12 (-0.46) -0.02 (-0.04) – 222
Emg. (VW) -0.10 (-3.25) 0.31 (4.90) -4.66 (-8.54) 0.12 (0.35) 0.02 (0.09) 0.09 (0.25) – 222
Emg. (Pooled) -0.14 (-3.48) 0.34 (3.12) -2.29 (-2.98) -0.24 (-0.41) 0.41 (1.14) -0.65 (-1.34) 4,518 198

40
 Table 5
Average Coefficients of Cross-Sectional Regressions of Monthly Returns on prior 12 Monthly Returns, U.S. Sample, 1927–2010

We estimate the following cross-sectional regressions of the form:

̂ ∑
For raw returns, the dependent variable is the simple monthly stock return in excess of the risk-free rate. For three-factor adjusted returns, the dependent variable is the stock
return, minus loading on the market factor, SMB and HML, multiplied by their respective factor realizations. Loadings are estimated over the entire sample period. The
regressions are estimated for each month between January 1927 and December 2010. The table shows average  coefficients on lagged returns and t-statistics corrected for serial
correlation using three lags.

Rt-1 Rt-2 Rt-3 Rt-4 Rt-5 Rt-6 Rt-7 Rt-8 Rt-9 Rt-10 Rt-11 Rt-12
Panel A: Full Sample
1927-2010 Raw -6.78 -0.34 1.64 0.97 1.27 1.83 127 0.63 1.40 1.08 1.41 1.88
(-17.47) (-1.26) (6.09) (3.71) (4.95) (6.79) (5.54) (2.74) (6.34) (4.55) (6.28) (7.73)
3-Factor -7.27 -0.55 1.53 0.94 1.20 1.77 1.14 0.58 1.25 1.07 1.24 1.67
(-19.75) (-2.29) (6.57) (4.47) (5.97) (7.93) (6.02) (3.03) (6.63) (5.26) (6.38) (7.98)
Panel B: Subperiods
1927-1947 Raw -10.96 -2.29 0.62 -0.53 1.07 1.44 1.34 0.42 1.64 1.28 2.01 1.79
(-11.45) (-3.21) (0.77) (-0.69) (1.49) (1.77) (1.93) (0.63) (2.68) (1.89) (3.06) (2.41)
3-Factor -11.51 -2.40 0.81 -0.32 1.24 1.60 1.37 0.39 1.22 1.59 1.54 1.66
(-12.74) (-4.01) (1.24) (-0.58) (2.25) (2.43) (2.43) (0.74) (2.31) (2.89) (2.80) (2.56)
1948-1968 Raw -6.42 0.88 2.93 2.16 2.07 2.93 1.21 -0.16 1.03 1.17 1.50 2.58
(-11.41) (1.83) (6.01) (4.93) (4.39) (6.84) (3.18) (-0.39) (2.69) (2.63) (3.39) (6.00)
3-Factor -7.04 0.71 2.44 2.05 1.85 2.54 1.21 0.01 0.88 1.14 1.38 2.19
(-13.31) (1.78) (5.38) (5.87) (5.23) (6.68) (3.75) (0.02) (2.54) (2.93) (3.67) (6.12)
1969-1989 Raw -3.41 -0.26 0.78 0.87 0.55 0.78 0.68 0.63 0.93 0.67 0.62 1.06
(-7.64) (-1.27) (4.28) (4.53) (2.56) (3.98) (3.73) (3.36) (4.70) (3.89) (4.45) (6.03)
3-Factor -3.60 -0.35 0.70 0.74 0.44 0.64 0.69 0.77 0.97 0.58 0.56 0.79
(-8.23) (-2.02) (4.74) (4.67) (2.77) (4.03) (4.63) (4.74) (5.63) (3.90) (4.05) (5.53)
1990-2010 Raw -0.84 0.15 0.41 0.11 0.23 0.40 0.36 0.32 0.32 0.15 0.26 0.28
(-4.96) (1.11) (3.44) (1.03) (2.04) (2.91) (3.23) (3.02) (2.78) (1.26) (2.75) (2.71)
3-Factor -1.12 -0.00 0.33 0.14 0.19 0.41 0.20 0.15 0.19 0.08 0.17 0.20
(-6.11) (-0.04) (3.07) (1.39) (2.25) (4.39) (2.67) (1.93) (2.44) (0.82) (2.16) (2.58)

41
 Table 6
Differences in Sums and Averages of Coefficients from Fama-MacBeth Return Regressions, U.S. Sample, 1927–2010
Columns IR and RR show the months contained in intermediate and recent return definitions. The adjacent column shows the
difference in the number of months between IR and RR. For each definition, we calculate the sum and average of the coefficients
from Fama-MacBeth regression contained in Table 5. The table shows the differences in the sum (or averages) of these
coefficients. Conventional t-statistics for a two-sided test of difference in coefficients appear in parentheses. The returns and
conventional t-statistics are bootstrapped using the stepwise multiple testing method of Romano and Wolf (2005). Definitions
that are statistically significant at the 10%, 5% or 1% level based on the bootstrap are marked *, **, and *** respectively.
Difference in Sum of Coefficients Difference in Average of Coefficients
Months Raw 3-factor Raw 3-factor
IR. RR Diff Diff. t-stat Diff. t-stat Diff. t-stat Diff. t-stat

[12,3] [2] -9 13.73*** (11.18)*** 12.90*** (12.78)*** 1.68*** (6.83)*** 1.78*** (8.25)***
[12,4] [2] -8 12.09*** (11.02)*** 11.37*** (12.48)*** 1.65*** (6.53)*** 1.75*** (7.90)***
[12,4] [3,2] -7 10.45*** (10.27)*** 9.84*** (11.44)*** 0.66** (3.35)** 0.71*** (4.15)***
[12,5] [2] -7 11.12*** (11.54)*** 10.42*** (12.87)*** 1.69*** (6.62)*** 1.78*** (7.94)***
[12,5] [3,2] -6 9.48*** (10.41)*** 8.89*** (11.41)*** 0.70** (3.45)*** 0.74*** (4.20)***
[12,6] [2] -6 9.85*** (11.78)*** 9.21*** (12.87)*** 1.70*** (6.64)*** 1.78*** (7.93)***
[12,5] [4,2] -5 8.51*** (9.79)*** 7.94*** (10.55)*** 0.59** (3.38)** 0.59*** (3.89)***
[12,6] [3,2] -5 8.21*** (10.22)*** 7.68*** (11.01)*** 0.71** (3.45)*** 0.75*** (4.17)***
[12,7] [2] -5 8.02*** (11.05)*** 7.46*** (11.98)*** 1.62*** (6.22)*** 1.70*** (7.42)***
[12,6] [4,2] -4 7.24*** (9.10)*** 6.74*** (9.71)*** 0.60** (3.32)** 0.59*** (3.82)***
[12,7] [3,2] -4 6.38*** (8.87)*** 5.92*** (9.55)*** 0.63** (2.98)** 0.66*** (3.61)***
[12,8] [2] -4 6.75*** (10.49)*** 6.32*** (11.41)*** 1.62*** (6.09)*** 1.70*** (7.28)***
[12,6] [5,2] -3 5.97*** (7.47)*** 5.53*** (8.02)*** 0.47** (2.92)** 0.45** (3.34)**
[12,7] [4,2] -3 5.41*** (7.28)*** 4.98*** (7.81)*** 0.52** (2.77)** 0.51** (3.17)**
[12,8] [3,2] -3 5.11*** (7.67)*** 4.79*** (8.37)*** 0.63** (2.86)** 0.66*** (3.50)***
[12,9] [2] -3 6.11*** (11.10)*** 5.75*** (11.91)*** 1.79*** (6.57)*** 1.85*** (7.70)***
[12,7] [5,2] -2 4.14*** (5.29)*** 3.77*** (5.74)*** 0.39* (2.30)* 0.37* (2.59)**
[12,8] [4,2] -2 4.14*** (5.77)*** 3.84*** (6.32)*** 0.52** (2.64)** 0.51** (3.05)**
[12,9] [3,2] -2 4.47*** (7.32)*** 4.22*** (8.00)*** 0.79*** (3.44)** 0.81*** (4.09)***
[12,10] [2] -2 4.72*** (9.78)*** 4.50*** (10.65)*** 1.80*** (6.37)*** 1.86*** (7.47)***
[12,7] [6,2] -1 2.31* (2.73)** 2.01* (2.89)** 0.21 (1.27) 0.17 (1.31)
[12,8] [5,2] -1 2.86** (3.69)*** 2.63** (4.09)*** 0.40* (2.18)* 0.37* (2.47)**
[12,9] [4,2] -1 3.50** (5.04)*** 3.27*** (5.57)*** 0.69** (3.25)** 0.66*** (3.71)***
[12,10] [3,2] -1 3.08** (5.42)*** 2.96*** (6.04)*** 0.81*** (3.33)** 0.82*** (3.93)***
[12,11] [2] -1 3.63*** (8.81)*** 3.43*** (9.47)*** 1.99*** (6.60)*** 1.99*** (7.54)***
[12,8] [6,2] 0 1.04 (1.19) 0.87 (1.24) 0.21 (1.19) 0.18 (1.24)
[12,9] [5,2] 0 2.23* (2.87)** 2.06* (3.24)** 0.56** (2.87)** 0.52** (3.24)**
[12,10] [4,2] 0 2.11 (3.13)** 2.02* (3.55)*** 0.70** (3.13)** 0.67*** (3.55)***
[12,11] [3,2] 0 1.99 (3.78)*** 1.90* (4.22)*** 1.00*** (3.78)*** 0.95*** (4.22)***
[12] [2] 0 2.23* (6.79)*** 2.20** (7.56)*** 2.23*** (6.79)*** 2.20*** (7.56)***
[12,8] [7,2] 1 -0.24 (-0.25) -0.26 (-0.34) 0.17 (1.05) 0.15 (1.10)
[12,9] [6,2] 1 0.41 (0.45) 0.31 (0.43) 0.37* (1.97)* 0.32* (2.12)*
[12,10] [5,2] 1 0.84 (1.08) 0.81 (1.28) 0.57** (2.75)** 0.53** (3.08)**
[12,11] [4,2] 1 1.02 (1.57) 0.95 (1.76) 0.89*** (3.63)*** 0.80*** (3.88)***
[12] [3,2] 1 0.59 (1.24) 0.67 (1.66) 1.23*** (4.21)*** 1.17*** (4.58)***
[12,9] [7,2] 2 -0.87 (-0.87) -0.83 (-1.04) 0.34* (1.85) 0.29* (2.01)*
[12,10] [6,2] 2 -0.99 (-1.10) -0.95 (-1.31) 0.38* (1.90)* 0.34* (2.04)*
[12,11] [5,2] 2 -0.25 (-0.32) -0.25 (-0.41) 0.76*** (3.31)** 0.66*** (3.47)***
[12] [4,2] 2 -0.39 (-0.62) -0.27 (-0.53) 1.13*** (4.04)*** 1.01*** (4.25)***
[12,9] [8,2] 3 -1.50 (-1.38) -1.40 (-1.62) 0.40* (2.33)* 0.36* (2.55)**
[12,10] [7,2] 3 -2.27 (-2.24) -2.09 (-2.56) 0.35* (1.78) 0.31* (1.92)*
[12,11] [6,2] 3 -2.07 (-2.29) -2.01 (-2.79) 0.57** (2.55)** 0.46** (2.52)**
[12] [5,2] 3 -1.66 (-2.19) -1.48 (-2.42) 1.00*** (3.74)*** 0.87*** (3.86)***
[12,10] [8,2] 4 -2.90 (-2.59) -2.65 (-2.96) 0.42* (2.20)* 0.37* (2.38)**
[12,11] [7,2] 4 -3.35 (-3.28) -3.15 (-3.86) 0.54** (2.46)* 0.44** (2.42)**
[12] [6,2] 4 -3.48 (-3.88) -3.24 (-4.51) 0.81*** (3.10)** 0.68*** (3.08)**
[12,10] [9,2] 5 -4.29 (-3.62) -3.91 (-4.07) 0.37* (2.07)* 0.33* (2.24)*
[12,11] [8,2] 5 -3.98 (-3.50) -3.72 (-4.10) 0.61** (2.83)** 0.50** (2.81)**
[12] [7,2] 5 -4.76 (-4.67) -4.38 (-5.37) 0.77*** (3.03)** 0.65*** (3.01)**
[12,11] [9,2] 6 -5.38 (-4.43) -4.97 (-5.08) 0.56** (2.72)** 0.46** (2.68)**
[12] [8,2] 6 -5.39 (-4.72) -4.95 (-5.43) 0.84*** (3.33)** 0.72*** (3.33)**
[12,11] [10,2] 7 -6.46 (-5.07) -6.03 (-5.87) 0.56** (2.84)** 0.45** (2.72)**
[12] [9,2] 7 -6.79 (-5.55) -6.20 (-6.27) 0.80*** (3.23)** 0.68*** (3.20)**
[12] [10,2] 8 -7.87 (-6.08) -7.26 (-6.94) 0.80*** (3.31)** 0.67*** (3.22)**
[12] [11,2] 9 -9.28*** (-6.90)*** -8.49*** (-7.75)*** 0.77*** (3.31)** 0.64*** (3.21)**

42
 Table 7
Differences in Sums and Averages of Coefficients from Fama-MacBeth Return Regressions Excluding Month -2, U.S.
Sample, 1927–2010

Columns labeled IR and RR show the months contained in intermediate and recent return strategies. The adjacent column shows
the difference in the number of months between IR and RR strategies. For each strategy, we calculate the sum and average of the
coefficients from Fama-MacBeth regression contained in Table 5. The table shows the differences in the sum (or averages) of
these coefficients, calculated as IR minus RR. Conventional t-statistics for a two-sided test of IR coefficients are greater than RR
coefficients appear in parentheses. The returns and conventional t-statistics are bootstrapped using the stepwise multiple testing
method of Romano and Wolf (2005). Strategies that are statistically significant (for a one-sided test) at the 10%, 5% or 1% level
based on the bootstrap are market *, **, and *** respectively.

# Difference in Sum of Coefficients Difference in Average of Coefficients

Months Raw 3-factor Raw 3-factor
IR. RR Diff Diff. t-stat Diff. t-stat Diff. t-stat Diff. t-stat

[12,4] [3,3] -8 10.10*** (9.61)*** 9.29*** (10.62)*** -0.34 (-1.46) -0.33 (-1.63)
[12,5] [3,3] -7 9.13*** (9.85)*** 8.34*** (10.70)*** -0.29 (-1.23) -0.30 (-1.43)
[12,5] [4,3] -6 8.16*** (9.44)*** 7.40*** (10.03)*** 0.04 (0.22) 0.00 (-0.03)
[12,6] [3,3] -6 7.86*** (9.74)*** 7.14*** (10.36)*** -0.28 (-1.16) -0.29 (-1.39)
[12,6] [4,3] -5 6.89*** (8.89)*** 6.19*** (9.28)*** 0.05 (0.26) 0.00 (-0.01)
[12,7] [3,3] -5 6.04*** (8.62)*** 5.38*** (9.08)*** -0.36 (-1.45) -0.38 (-1.78)
[12,6] [5,3] -4 5.62*** (7.37)*** 4.98*** (7.64)*** 0.06 (0.36) 0.01 (0.07)
[12,7] [4,3] -4 5.07*** (7.27)*** 4.43*** (7.51)*** -0.03 (-0.12) -0.09 (-0.50)
[12,8] [3,3] -4 4.76*** (7.62)*** 4.24*** (8.04)*** -0.36 (-1.39) -0.38 (-1.72)
[12,7] [5,3] -3 3.79*** (5.26)*** 3.23*** (5.37)*** -0.01 (-0.08) -0.08 (-0.51)
[12,8] [4,3] -3 3.79*** (5.83)*** 3.29*** (6.05)*** -0.02 (-0.11) -0.08 (-0.47)
[12,9] [3,3] -3 4.13*** (7.56)*** 3.67*** (7.94)*** -0.20 (-0.73) -0.23 (-1.01)
[12,7] [6,3] -2 1.97 (2.55)** 1.47 (2.35)* -0.15 (-0.87) -0.21 (-1.54)
[12,8] [5,3] -2 2.52** (3.61)*** 2.09** (3.64)*** -0.01 (-0.07) -0.07 (-0.47)
[12,9] [4,3] -2 3.16** (5.20)*** 2.72*** (5.39)*** 0.14 (0.59) 0.06 (0.32)
[12,10] [3,3] -2 2.73** (5.71)*** 2.42** (5.98)*** -0.18 (-0.65) -0.21 (-0.90)
[12,8] [6,3] -1 0.69 (0.89) 0.33 (0.53) -0.15 (-0.80) -0.21 (-1.42)
[12,9] [5,3] -1 1.89 (2.76)** 1.52 (2.75)** 0.15 (0.71) 0.07 (0.44)
[12,10] [4,3] -1 1.76 (3.12)** 1.47 (3.12)** 0.15 (0.63) 0.08 (0.38)
[12,11] [3,3] -1 1.65 (4.06)*** 1.36 (3.93)*** 0.01 (0.02) -0.09 (-0.35)
[12,8] [7,3] 0 -0.58 (-0.68) -0.81 (-1.18) -0.12 (-0.68) -0.16 (-1.18)
[12,9] [6,3] 0 0.06 (0.08) -0.24 (-0.39) 0.02 (0.08) -0.06 (-0.39)
[12,10] [5,3] 0 0.49 (0.74) 0.27 (0.50) 0.16 (0.74) 0.09 (0.50)
[12,11] [4,3] 0 0.68 (1.30) 0.41 (0.94) 0.34 (1.30) 0.20 (0.94)
[12] [3,3] 0 0.24 (0.75) 0.13 (0.46) 0.24 (0.75) 0.13 (0.46)
[12,]9 [7,3] 1 -1.21 (-1.36) -1.38 (-1.95) 0.05 (0.24) -0.02 (-0.10)
[12,1]0 [6,3] 1 -1.34 (-1.70) -1.49 (-2.40) 0.03 (0.14) -0.04 (-0.26)
[12,11] [5,3] 1 -0.59 (-0.92) -0.80 (-1.55) 0.35 (1.46) 0.22 (1.11)
[12] [4,3] 1 -0.73 (-1.52) -0.82 (-2.07) 0.58* (1.96) 0.42 (1.70)
[12,]9 [8,3] 2 -1.84 (-1.87) -1.95 (-2.52) 0.17 (0.97) 0.11 (0.77)
[12,1]0 [7,3] 2 -2.61 (-2.90) -2.63 (-3.68) 0.06 (0.30) 0.00 (0.00)
[12,11] [6,3] 2 -2.42 (-3.10) -2.56 (-4.17) 0.22 (0.94) 0.08 (0.44)
[12] [5,3] 2 -2.00 (-3.21) -2.03 (-4.07) 0.59* (2.12) 0.43 (1.87)
[12,10] [8,3] 3 -3.24 (-3.22) -3.20 (-4.01) 0.19 (0.96) 0.13 (0.79)
[12,11] [7,3] 3 -3.69 (-4.11) -3.69 (-5.18) 0.25 (1.11) 0.13 (0.70)
[12] [6,3] 3 -3.83 (-4.99) -3.78 (-6.24) 0.45 (1.70) 0.30 (1.34)
[12,10] [9,3] 4 -4.64 (-4.33) -4.45 (-5.17) 0.17 (0.93) 0.12 (0.77)
[12,11] [8,3] 4 -4.32 (-4.24) -4.26 (-5.29) 0.38 (1.72) 0.25 (1.41)
[12] [7,3] 4 -5.10 (-5.72) -4.92 (-6.95) 0.48 (1.86) 0.34 (1.57)
[12,11] [9,3] 5 -5.72 (-5.23) -5.52 (-6.28) 0.36 (1.71) 0.24 (1.41)
[12] [8,3] 5 -5.73 (-5.63) -5.49 (-6.81) 0.61* (2.39)* 0.47 (2.17)
[12,11] [10,3] 6 -6.80 (-5.90) -6.58 (-7.08) 0.38 (1.93) 0.26 (1.57)
[12] [9,3] 6 -7.13 (-6.48) -6.74 (-7.60) 0.59* (2.38)* 0.46 (2.16)
[[12] [10,3] 7 -8.21 (-7.01) -7.81 (-8.24) 0.62* (2.56)* 0.48 (2.30)*
[12] [11,3] 8 -9.62*** (-7.86)*** -9.04*** (-9.05)*** 0.60* (2.60)** 0.47 (2.35)*

43
 Table A1
Returns and Three-Factor Alphas from Monthly Single Sorts on Recent (−2 to −6)
and Intermediate (−7 to −12) Returns, Large-cap Stocks, 1980–2010

The methods and statistics in the table are identical to those in Table 2 but the sample includes only large-cap stocks.
Large-cap stocks are those above the 75th percentile of market capitalization.

Equal-weighted Value-weighted
RR(WML). IR(WML) Difference RR(WML) IR(WML) Difference
Panel A: Returns
Developed Markets
Australia 2.26 (6.97) 1.31 (5.31) 0.95 (2.89) 2.42 (5.46) 1.39 (3.93) 1.03 (2.28)
Austria –– –– –– –– –– ––
Belgium 1.64 (3.84) 1.31 (3.74) 0.32 (0.98) 1.38 (2.14) 0.62 (1.08) 0.75 (1.22)
Canada 2.02 (7.40) 1.40 (5.59) 0.62 (2.05) 1.73 (4.44) 1.21 (3.77) 0.52 (1.23)
Denmark 1.58 (5.90) 1.06 (4.00) 0.52 (1.80) 1.57 (3.77) 1.04 (2.66) 0.53 (0.99)
Finland 1.51 (1.15) 1.72 (1.14) -0.21 (-0.16) -0.42 (-0.22) 2.46 (1.21) -2.87 (-1.70)
France 1.46 (4.27) 1.24 (4.64) 0.22 (0.60) 0.55 (1.39) 0.96 (2.96) -0.41 (-0.88)
Germany 1.40 (4.72) 1.34 (5.40) 0.06 (0.19) 1.17 (2.84) 1.28 (3.45) -0.12 (-0.26)
Greece 1.29 (2.16) 0.51 (0.97) 0.77 (1.15) 0.65 (0.87) 0.24 (0.40) 0.41 (0.44)
Hong Kong 0.61 (1.17) -0.05 (-0.10) 0.66 (0.91) 0.57 (1.01) 0.33 (0.61) 0.24 (0.30)
Israel 0.45 (1.49) 0.30 (1.17) 0.16 (0.48) 1.35 (2.81) 0.46 (1.12) 0.89 (1.68)
Italy 0.89 (2.30) 0.78 (2.94) 0.11 (0.30) 0.49 (1.19) 0.74 (1.87) -0.26 (-0.60)
Japan -0.06 (-0.20) 0.41 (1.78) -0.47 (-1.48) -0.04 (-0.11) 0.57 (1.99) -0.61 (-1.62)
Netherlands 1.28 (3.32) 1.14 (3.21) 0.14 (0.34) 1.11 (2.00) 1.06 (2.01) 0.05 (0.07)
New Zealand 1.16 (1.01) 1.18 (1.23) -0.02 (-0.01) -0.39 (-0.26) -0.10 (-0.10) -0.29 (-0.17)
Norway 1.74 (3.64) 1.57 (3.14) 0.17 (0.31) 1.26 (1.89) 1.47 (2.31) -0.21 (-0.27)
Portugal –– –– –– –– –– ––
Singapore 0.71 (1.48) 0.19 (0.48) 0.52 (1.06) 0.19 (0.35) 0.18 (0.40) 0.00 (0.01)
Spain 0.85 (1.80) 0.66 (1.65) 0.19 (0.42) 0.43 (0.70) 0.52 (0.99) -0.09 (-0.11)
Sweden 2.48 (3.24) 1.29 (2.19) 1.19 (1.44) 0.92 (0.88) 0.42 (0.57) 0.50 (0.47)
Switzerland 1.15 (3.82) 1.09 (3.95) 0.07 (0.23) 0.53 (1.27) 1.03 (2.61) -0.50 (-1.12)
UK 1.78 (7.58) 1.28 (6.82) 0.50 (2.48) 0.61 (1.76) 0.84 (2.82) -0.23 (-0.67)
Dev. (EW) 1.16 (6.69) 0.91 (6.40) 0.26 (1.54) 0.84 (3.77) 0.80 (4.52) 0.05 (0.22)
Dev. (VW) 0.77 (3.60) 0.80 (4.84) -0.02 (-0.10) 0.37 (1.38) 0.75 (3.59) -0.37 (-1.36)
Dev. (Pooled) 1.04 (3.91) 1.00 (4.33) 0.04 (0.12) 0.51 (1.52) 0.87 (3.16) -0.36 (-1.01)

Emerging Markets
Brazil -0.81 (-1.43) -0.35 (-0.60) -0.46 (-0.55) -0.78 (-1.23) -0.52 (-0.82) -0.26 (-0.30)
Chile 1.46 (4.93) 0.50 (1.86) 0.96 (2.88) 0.98 (2.22) 0.38 (0.80) 0.59 (1.00)
China 0.22 (0.63) 0.40 (1.30) -0.18 (-0.47) 0.18 (0.42) 0.52 (1.30) -0.34 (-0.66)
India 0.38 (0.83) 0.88 (2.11) -0.50 (-1.00) 1.61 (2.95) 1.23 (2.41) 0.38 (0.69)
Indonesia -0.17 (-0.24) -0.61 (-0.96) 0.44 (0.61) 0.18 (0.20) 0.38 (0.55) -0.20 (-0.21)
Korea 0.14 (0.27) 0.08 (0.18) 0.07 (0.12) -0.03 (-0.07) 0.32 (0.67) -0.35 (-0.59)
Malaysia 0.24 (0.55) 0.24 (0.66) 0.01 (0.02) 0.24 (0.49) 0.13 (0.35) 0.10 (0.20)
Mexico –– –– –– –– –– ––
Peru –– –– –– –– –– ––
Philippines -0.33 (-0.37) -0.41 (-0.64) 0.08 (0.10) 0.09 (0.10) -0.06 (-0.09) 0.15 (0.15)
Russia 0.09 (0.06) -2.39 (-1.61) 2.48 (1.85) 0.57 (0.42) -1.31 (-0.84) 1.88 (1.04)
South Africa 2.68 (8.88) 1.87 (6.55) 0.82 (2.24) 1.98 (4.29) 1.67 (3.62) 0.31 (0.54)
Taiwan -0.48 (-1.03) 0.46 (1.22) -0.94 (-1.62) -0.50 (-1.01) 0.45 (0.98) -0.95 (-1.36)
Thailand 0.15 (0.22) 0.45 (0.90) -0.30 (-0.43) 0.21 (0.32) 0.29 (0.50) -0.08 (-0.12)
Turkey -0.93 (-1.77) -0.44 (-0.92) -0.49 (-0.67) -1.75 (-2.31) 0.59 (0.84) -2.34 (-2.12)
Emg. (EW) 0.23 (0.96) 0.34 (1.84) -0.09 (-0.35) 0.23 (0.90) 0.47 (2.30) -0.19 (-0.66)
Emg. (VW) 0.35 (1.52) 0.52 (3.05) -0.16 (-0.59) 0.23 (0.92) 0.56 (2.74) -0.33 (-1.08)
Emg. (Pooled) 0.38 (0.69) 0.75 (1.52) -0.36 (-0.61) 0.52 (0.93) 0.61 (1.27) -0.09 (-0.16)

44
 Equal-weighted Value-weighted
RR(WML) IR(WML) Difference RR(WML) IR(WML) Difference
Panel B: Three-Factor Alphas
Developed Markets
Australia 2.22 (6.39) 1.42 (5.43) 0.80 (2.29) 2.37 (5.17) 1.46 (3.93) 0.92 (1.98)
Austria –– –– –– –– –– ––
Belgium 1.63 (4.19) 1.38 (4.17) 0.25 (0.76) 1.50 (2.76) 0.73 (1.39) 0.77 (1.27)
Canada 2.16 (7.15) 1.54 (5.68) 0.62 (1.87) 2.02 (4.60) 1.35 (3.86) 0.67 (1.42)
Denmark 1.60 (5.92) 1.14 (4.18) 0.46 (1.58) 1.48 (3.51) 1.09 (2.70) 0.39 (0.73)
Finland –– –– –– –– –– ––
France 1.66 (4.94) 1.37 (5.17) 0.29 (0.78) 0.77 (1.98) 1.07 (3.27) -0.30 (-0.65)
Germany 1.53 (5.27) 1.50 (6.13) 0.03 (0.11) 1.36 (3.31) 1.49 (4.02) -0.12 (-0.27)
Greece 1.43 (2.46) 0.65 (1.28) 0.79 (1.20) 0.79 (1.08) 0.34 (0.55) 0.45 (0.50)
Hong Kong 0.64 (1.27) 0.26 (0.58) 0.38 (0.55) 0.75 (1.35) 0.57 (1.04) 0.18 (0.23)
Israel 0.23 (0.34) 1.12 (2.10) -0.89 (-1.36) 1.37 (1.69) 1.12 (1.37) 0.25 (0.27)
Italy 0.76 (2.03) 0.75 (2.87) 0.02 (0.05) 0.42 (1.03) 0.66 (1.68) -0.24 (-0.56)
Japan -0.19 (-0.65) 0.48 (2.12) -0.68 (-2.10) -0.21 (-0.57) 0.58 (2.03) -0.79 (-2.04)
Netherlands 1.39 (3.56) 1.22 (3.31) 0.17 (0.39) 1.16 (2.02) 1.43 (2.61) -0.27 (-0.38)
New Zealand –– –– –– –– –– ––
Norway 1.75 (3.71) 1.72 (3.42) 0.03 (0.06) 1.17 (1.73) 1.56 (2.43) -0.38 (-0.47)
Portugal –– –– –– –– –– ––
Singapore 1.00 (2.20) 0.47 (1.24) 0.53 (1.08) 0.41 (0.79) 0.43 (0.98) -0.02 (-0.03)
Spain 0.94 (2.13) 0.77 (2.02) 0.16 (0.36) 0.53 (0.87) 0.75 (1.46) -0.23 (-0.30)
Sweden 2.69 (3.60) 1.41 (2.55) 1.28 (1.53) 1.34 (1.32) 0.76 (1.06) 0.58 (0.54)
Switzerland 1.26 (4.06) 1.18 (4.18) 0.08 (0.27) 0.70 (1.61) 1.18 (2.88) -0.48 (-1.04)
UK 2.04 (7.86) 1.32 (6.17) 0.72 (3.19) 1.01 (2.65) 0.99 (2.90) 0.02 (0.06)
Dev. (EW) 1.17 (6.60) 1.05 (7.18) 0.12 (0.66) 0.83 (3.62) 0.95 (5.20) -0.12 (-0.53)
Dev. (VW) 0.72 (3.18) 0.93 (5.45) -0.22 (-0.94) 0.26 (0.91) 0.86 (3.89) 0.86 (3.89)
Dev. (Pooled) 0.87 (3.18) 0.80 (3.34) 0.07 (0.24) 0.40 (1.14) 0.73 (2.51) -0.33 (-0.86)

Emerging Markets
Brazil -0.47 (-0.85) -0.29 (-0.49) -0.18 (-0.21) -0.27 (-0.45) -0.35 (-0.54) 0.08 (0.09)
Chile 1.51 (4.78) 0.63 (2.26) 0.87 (2.51) 0.96 (2.27) 0.63 (1.55) 0.33 (0.63)
China 0.38 (1.10) 0.92 (3.32) -0.54 (-1.51) 0.28 (0.65) 1.04 (2.79) -0.75 (-1.48)
India 0.52 (1.01) 1.15 (2.55) -0.64 (-1.17) 1.79 (2.89) 1.22 (2.15) 0.57 (0.97)
Indonesia 0.30 (0.45) 0.20 (0.40) 0.09 (0.13) 1.20 (1.49) 0.93 (1.45) 0.27 (0.29)
Korea 0.50 (0.88) 0.64 (1.48) -0.14 (-0.23) 0.25 (0.44) 0.80 (1.53) -0.56 (-0.84)
Malaysia 0.57 (1.28) 0.43 (1.28) 0.14 (0.30) 0.66 (1.38) 0.34 (0.90) 0.32 (0.60)
Mexico –– –– –– –– –– ––
Peru –– –– –– –– –– ––
Philippines -0.37 (-0.46) -0.24 (-0.41) -0.13 (-0.16) 0.16 (0.18) 0.01 (0.02) 0.15 (0.15)
Russia –– –– –– –– –– ––
South Africa 2.55 (8.01) 1.70 (5.69) 0.84 (2.06) 1.83 (3.59) 1.67 (3.26) 0.16 (0.26)
Taiwan -0.13 (-0.26) 0.61 (1.44) -0.74 (-1.17) -0.25 (-0.47) 0.45 (0.96) -0.70 (-0.99)
Thailand 1.47 (2.42) 0.56 (1.12) 0.91 (1.42) 1.20 (1.90) 0.85 (1.49) 0.34 (0.51)
Turkey -0.47 (-0.96) -0.31 (-0.70) -0.16 (-0.23) -1.43 (-1.89) 0.59 (0.86) -2.01 (-1.85)
Emg. (EW) 0.53 (2.16) 0.48 (2.41) 0.05 (0.20) 0.54 (2.05) 0.61 (2.78) -0.07 (-0.25)
Emg. (VW) 0.48 (2.14) 0.53 (2.94) -0.05 (-0.18) 0.41 (1.71) 0.61 (2.79) -0.20 (-0.69)
Emg. (Pooled) 0.42 (0.76) 1.01 (2.05) -0.59 (-0.98) 0.58 (1.04) 0.66 (1.37) -0.08 (-0.13)

45
 Table A2
Differences in Returns and Three-Factor Alphas from Monthly Double Sorts on Recent (−2 to −6)
and Intermediate (−7 to −12) Returns, Large-cap Stocks, 1980–2010

The methods and statistics in the table are identical to those in Table 3 but the sample includes only large-cap stocks.
Large-cap stocks are those above the 75th percentile of market capitalization.

Equal-weighted Value-weighted
Portfolio returns Three-factor alphas Portfolio returns Three-factor alphas
Developed Markets
Australia 0.92 (3.15) 0.90 (2.96) 0.66 (1.56) 0.76 (1.77)
Austria –– –– –– ––
Belgium 0.14 (0.36) 0.21 (0.53) -0.01 (-0.01) 0.11 (0.22)
Canada 0.70 (2.72) 0.72 (2.55) 0.27 (0.82) 0.30 (0.83)
Denmark 0.40 (1.39) 0.43 (1.51) 0.25 (0.55) 0.18 (0.39)
Finland -0.36 (-0.29) –– -0.52 (-0.39) ––
France -0.15 (-0.55) -0.11 (-0.40) -0.71 (-2.06) -0.64 (-1.86)
Germany -0.07 (-0.27) -0.08 (-0.30) -0.39 (-1.07) -0.37 (-1.01)
Greece 0.47 (0.87) 0.48 (0.90) 0.90 (1.25) 0.87 (1.21)
Hong Kong 0.88 (1.50) 0.66 (1.17) 0.19 (0.30) 0.10 (0.17)
Israel 0.38 (1.26) -0.77 (-1.22) 1.13 (2.59) 0.93 (1.33)
Italy 0.02 (0.05) 0.00 (0.01) -0.63 (-1.51) -0.65 (-1.57)
Japan -0.34 (-1.42) -0.48 (-2.00) -0.57 (-2.10) -0.67 (-2.40)
Netherlands -0.16 (-0.43) -0.13 (-0.33) 0.12 (0.22) -0.02 (-0.03)
New Zealand 1.78 (1.05) –– 2.49 (1.90) ––
Norway -0.07 (-0.13) -0.14 (-0.26) -0.76 (-1.15) -0.86 (-1.28)
Portugal –– –– –– ––
Singapore 0.24 (0.56) 0.30 (0.70) -0.46 (-0.86) -0.35 (-0.66)
Spain 0.14 (0.31) 0.21 (0.44) -0.69 (-1.24) -0.73 (-1.28)
Sweden 0.45 (0.71) 0.50 (0.79) 0.20 (0.26) 0.21 (0.26)
Switzerland -0.18 (-0.65) -0.19 (-0.66) -0.92 (-2.42) -0.75 (-1.90)
UK 0.30 (1.71) 0.48 (2.41) -0.31 (-1.22) -0.16 (-0.54)
Dev. (EW) 0.19 (1.42) 0.08 (0.59) -0.10 (-0.61) -0.19 (-1.11)
Dev. (VW) -0.06 (-0.40) -0.20 (-1.18) -0.47 (-2.35) -0.60 (-2.84)
Dev (Pooled) -0.39 (-0.84) -0.52 (-1.10) -0.46 (-1.06) -0.39 (-0.89)

Emerging Markets
Brazil 0.04 (0.05) 0.19 (0.23) 0.05 (0.06) 0.16 (0.18)
Chile 0.80 (2.15) 0.74 (2.05) 0.59 (1.21) 0.76 (1.73)
China 0.20 (0.61) -0.16 (-0.49) -0.30 (-0.74) -0.66 (-1.67)
India -0.53 (-1.43) -0.65 (-1.59) -0.06 (-0.13) -0.13 (-0.27)
Indonesia -0.33 (-0.56) -0.49 (-0.84) 0.14 (0.19) 0.55 (0.80)
Korea -0.09 (-0.21) -0.20 (-0.41) 0.10 (0.23) 0.16 (0.32)
Malaysia 0.02 (0.04) 0.18 (0.47) 0.28 (0.64) 0.43 (0.97)
Mexico –– –– –– ––
Peru –– –– –– ––
Philippines 0.02 (0.03) -0.09 (-0.14) -0.26 (-0.36) -0.27 (-0.37)
Russia 0.65 (0.46) –– 1.61 (0.74) ––
South Africa 0.69 (1.99) 0.75 (1.99) 0.23 (0.44) 0.04 (0.07)
Taiwan -0.41 (-0.96) -0.23 (-0.48) -0.22 (-0.42) 0.05 (0.08)
Thailand -0.50 (-0.87) 0.46 (0.88) -0.43 (-0.75) 0.07 (0.13)
Turkey 0.03 (0.06) 0.17 (0.35) -1.73 (-2.13) -1.66 (-2.01)
Emg. (EW) -0.16 (-0.79) -0.05 (-0.23) -0.23 (-0.97) -0.05 (-0.20)
Emg. (VW) -0.17 (-0.83) -0.06 (-0.30) -0.21 (-0.88) -0.07 (-0.32)
Emg. (Pooled) -0.57 (-1.22) -0.84 (-1.77) -0.57 (-1.22) -0.84 (-1.77)

46
 Table A3
Monthly Fama-MacBeth Regressions, Large-cap Stocks, 1980–2010

The methods and statistics in the table are identical to those in Table 4 but the sample includes only large-cap stocks.
Large-cap stocks are those above the 75th percentile of market capitalization.

Ln ME Ln BM R(−1) RR(−2 to −6) IR (−7 to −12) Difference # #

Stock Months
Panel A: Excess returns
Developed Markets
Australia -0.05 (-0.68) 0.25 (1.90) -1.66 (-1.80) 2.43 (5.51) 1.14 (3.01) 1.29 (2.52) 527 246
Austria –– –– –– –– –– –– – –
Belgium 0.00 (0.01) 0.27 (2.58) -2.55 (-1.99) 3.15 (3.43) 0.57 (0.70) 2.58 (3.61) 110 126
Canada -0.09 (-1.87) 0.15 (1.27) -0.60 (-0.87) 1.95 (4.57) 0.96 (2.82) 0.99 (2.24) 405 306
Denmark 0.13 (1.77) 0.27 (1.74) -1.02 (-1.02) 2.55 (5.01) 1.47 (2.49) 1.08 (1.73) 132 230
Finland –– –– –– –– –– –– – –
France 0.06 (1.35) 0.37 (3.20) -6.08 (-7.46) 1.67 (3.38) 1.43 (3.49) 0.23 (0.55) 450 258
Germany 0.09 (2.11) 0.33 (3.05) -3.03 (-3.85) 1.73 (4.01) 1.92 (4.84) -0.19 (-0.36) 404 267
Greece -0.10 (-0.37) 0.26 (1.43) -1.13 (-0.87) 0.58 (0.67) 0.64 (0.94) -0.06 (-0.08) 187 162
Hong Kong -0.02 (-0.13) 0.22 (1.14) 0.92 (0.55) 0.79 (0.95) -0.58 (-0.52) 1.37 (1.18) 131 174
Israel –– –– –– –– –– –– – –
Italy 0.07 (1.38) 0.41 (2.77) -1.90 (-1.66) 1.60 (2.47) 1.53 (3.20) 0.06 (0.10) 158 225
Japan 0.03 (0.45) 0.45 (5.20) -5.50 (-6.95) -0.68 (-1.48) 0.40 (1.06) -1.08 (-2.19) 1,213 354
Netherlands 0.03 (0.38) 0.23 (1.53) -1.09 (-0.76) 2.69 (3.36) 2.09 (2.32) 0.60 (0.65) 117 137
New Zealand –– –– –– –– –– –– – –
Norway 0.00 (0.02) 0.15 (0.72) -1.99 (-1.61) 2.49 (3.65) 0.99 (1.25) 1.51 (1.73) 120 157
Portugal –– –– –– –– –– –– – –
Singapore 0.01 (0.13) 0.28 (1.98) -2.90 (-2.26) 0.11 (0.08) 0.04 (0.04) 0.07 (0.05) 244 174
Spain 0.06 (0.85) 0.17 (1.14) -1.84 (-1.29) 2.64 (3.16) 0.52 (0.64) 2.11 (2.41) 108 121
Sweden 0.15 (2.06) 0.56 (3.30) -2.77 (-1.91) 1.86 (2.24) 1.34 (2.10) 0.53 (0.67) 260 126
Switzerland -0.01 (-0.31) 0.17 (1.51) -1.32 (-1.34) 2.66 (4.44) 1.43 (2.55) 1.22 (1.88) 148 222
UK 0.02 (0.35) 0.23 (2.83) -1.00 (-1.44) 1.80 (4.94) 1.16 (4.11) 0.64 (1.93) 775 354
Dev. (EW) -0.01 (-0.47) 0.28 (4.71) -3.10 (-6.58) 1.41 (4.54) 1.12 (3.94) 0.29 (1.00) – 354
Dev. (VW) 0.02 (0.40) 0.44 (5.53) -5.18 (-7.24) -0.34 (-0.80) 0.46 (1.32) -0.79 (-1.76) – 354
Dev. (Pooled) -0.05 (-1.26) 0.29 (3.18) -0.72 (-1.07) 1.09 (3.13) 0.98 (3.12) 0.11 (0.30) 5,382 318

Emerging Markets
Brazil –– –– –– – – –– –– – –
Chile 0.05 (0.71) 0.47 (3.04) 2.50 (1.63) 2.46 (3.65) 0.14 (0.30) 2.32 (3.00) 133 138
China -0.16 (-0.79) 0.20 (1.24) -5.44 (-4.86) -0.89 (-1.14) 0.73 (1.30) -1.62 (-2.19) 460 126
India -0.14 (-1.19) 0.15 (0.88) -4.83 (-4.78) 0.89 (1.98) 0.61 (1.11) 0.28 (0.63) 649 198
Indonesia -0.07 (-0.35) 0.63 (2.86) -0.03 (-0.03) -1.10 (-1.01) -0.93 (-1.16) -0.17 (-0.18) 183 174
Korea 0.12 (0.77) 0.62 (3.86) -6.47 (-5.67) -0.30 (-0.59) -0.46 (-0.74) 0.17 (0.24) 598 210
Malaysia 0.04 (0.28) 0.62 (4.34) -3.06 (-2.34) -0.26 (-0.20) -0.17 (-0.31) -0.09 (-0.08) 401 222
Mexico –– –– –– – – –– –– – –
Peru –– –– –– – – –– –– – –
Philippines 0.12 (1.04) 0.51 (2.59) -8.59 (-5.47) -1.69 (-1.66) -0.68 (-0.64) -1.01 (-1.13) 139 127
Russia –– –– –– – – –– –– – –
South Africa 0.09 (1.36) 0.61 (5.03) -3.35 (-3.19) 1.73 (3.72) 0.69 (1.73) 1.04 (1.81) 196 213
Taiwan -0.04 (-0.28) 0.32 (1.04) -0.75 (-0.74) -0.09 (-0.10) 0.19 (0.28) -0.28 (-0.27) 400 174
Thailand -0.05 (-0.40) 0.60 (3.37) -0.16 (-0.12) 0.53 (0.53) 0.62 (1.12) -0.09 (-0.09) 255 198
Turkey -0.07 (-0.46) 0.38 (2.30) -2.47 (-1.60) -1.34 (-1.57) -0.66 (-1.20) -0.68 (-0.86) 176 136
Emg. (EW) -0.06 (-1.11) 0.45 (5.91) -2.81 (-4.06) 0.20 (0.40) -0.14 (-0.37) 0.34 (0.72) – 222
Emg (VW) 0.00 (-0.03) 0.33 (3.37) -4.43 (-4.93) -0.65 (-0.98) 0.03 (0.07) -0.68 (-0.93) – 210
Emg. (Pooled) -0.05 (-0.98) 0.43 (3.33) -0.66 (-0.79) -0.38 (-0.61) 0.39 (0.80) -0.77 (-1.31) 4,106 198

47
 Ln ME Ln BM R(−1) RR(−2 to −6) IR (−7 to −12) Difference # #
Stock Months
Panel B: Three-factor adjusted returns
Developed Markets
Australia 0.05 (0.89) 0.39 (3.49) -2.26 (-2.64) 1.97 (5.24) 1.08 (3.62) 0.89 (1.96) 544 234
Austria –– –– – – –– –– –– – –
Belgium -0.07 (-1.74) 0.17 (1.47) -3.49 (-2.94) 2.70 (3.41) 0.48 (0.80) 2.22 (3.42) 109 126
Canada -0.01 (-0.37) 0.22 (1.77) -1.06 (-1.64) 1.87 (4.91) 1.07 (3.76) 0.80 (2.00) 402 306
Denmark 0.12 (2.18) 0.15 (1.12) -0.84 (-0.89) 2.43 (5.29) 1.41 (2.77) 1.02 (1.76) 132 230
Finland –– –– – – –– –– –– – –
France 0.00 (0.11) 0.25 (2.87) -6.50 (-8.88) 1.46 (3.35) 1.39 (4.32) 0.07 (0.18) 448 258
Germany 0.04 (1.48) 0.21 (2.54) -3.66 (-5.00) 1.41 (3.73) 1.75 (5.61) -0.34 (-0.71) 403 267
Greece 0.13 (1.04) 0.20 (1.21) -1.83 (-1.54) 0.23 (0.31) 0.61 (1.13) -0.38 (-0.51) 186 162
Hong Kong 0.06 (0.74) 0.17 (0.98) -0.70 (-0.46) 0.65 (0.92) -0.20 (-0.24) 0.85 (0.97) 130 174
Israel –– –– – – –– –– –– – –
Italy -0.02 (-0.77) 0.29 (2.26) -2.32 (-2.18) 1.69 (3.07) 1.47 (3.54) 0.23 (0.42) 157 225
Japan 0.12 (3.82) 0.35 (4.38) -6.04 (-8.24) -0.71 (-1.72) 0.19 (0.59) -0.89 (-2.24) 1,212 354
Netherlands -0.01 (-0.21) 0.09 (0.69) -2.28 (-1.85) 2.34 (2.95) 1.69 (2.16) 0.65 (0.72) 116 134
New Zealand –– –– – – –– –– –– – –
Norway -0.04 (-0.63) 0.12 (0.68) -2.46 (-1.99) 2.07 (3.62) 1.24 (1.80) 0.83 (0.98) 119 152
Portugal –– –– – – –– –– –– – –
Singapore 0.12 (2.27) 0.23 (1.69) -4.04 (-3.39) 0.11 (0.11) 0.28 (0.45) -0.17 (-0.16) 242 174
Spain 0.05 (0.90) 0.12 (0.78) -2.24 (-1.60) 2.23 (2.67) 0.56 (0.84) 1.66 (1.85) 108 118
Sweden 0.05 (0.95) 0.44 (2.81) -3.12 (-2.37) 1.48 (2.13) 1.02 (2.07) 0.47 (0.65) 257 126
Switzerland 0.00 (-0.14) 0.17 (1.50) -1.83 (-1.95) 2.32 (4.30) 1.44 (3.17) 0.88 (1.50) 147 222
UK -0.02 (-0.77) 0.10 (1.64) -1.09 (-1.65) 1.52 (4.34) 1.12 (4.75) 0.39 (1.25) 863 306
Dev. (EW) 0.04 (1.83) 0.28 (4.61) -3.42 (-7.69) 1.18 (3.96) 1.01 (4.32) 0.17 (0.62) – 354
Dev. (VW) 0.12 (4.00) 0.37 (4.93) -5.74 (-8.71) -0.41 (-1.09) 0.27 (0.92) -0.67 (-1.80) – 354
Dev. (Pooled) 0.01 (0.64) 0.21 (3.09) -1.87 (-3.43) 0.95 (3.14) 1.03 (4.33) -0.08 (-0.25) 5,362 318

Emerging Markets
Brazil –– –– –– – – –– –– – –
Chile -0.03 (-0.55) 0.36 (2.96) 1.62 (1.19) 2.05 (3.34) 0.11 (0.28) 1.94 (2.73) 133 138
China 0.04 (0.43) 0.14 (1.18) -6.01 (-5.61) -1.26 (-1.61) 0.55 (1.06) -1.81 (-2.54) 459 126
India -0.05 (-0.66) 0.09 (0.56) -5.05 (-5.20) 0.76 (1.99) 0.81 (2.02) -0.05 (-0.14) 645 198
Indonesia 0.05 (0.41) 0.40 (2.57) -1.28 (-1.09) -0.83 (-1.06) -0.15 (-0.32) -0.68 (-0.85) 181 174
Korea 0.14 (1.43) 0.58 (4.18) -6.68 (-6.14) -0.09 (-0.17) -0.21 (-0.43) 0.13 (0.20) 594 210
Malaysia 0.12 (1.31) 0.52 (4.07) -4.01 (-3.23) -0.22 (-0.21) -0.46 (-0.96) 0.24 (0.25) 400 222
Mexico –– –– –– – – –– –– – –
Peru –– –– –– – – –– –– – –
Philippines 0.07 (0.88) 0.46 (2.80) -8.97 (-6.12) -1.99 (-2.54) -0.56 (-0.68) -1.43 (-1.94) 139 127
Russia –– –– –– – – –– –– – –
South Africa 0.07 (1.20) 0.47 (4.11) -4.22 (-5.13) 1.61 (3.52) 0.84 (2.24) 0.77 (1.34) 195 210
Taiwan 0.03 (0.41) 0.12 (0.70) -1.95 (-1.97) 0.30 (0.44) 0.01 (0.01) 0.29 (0.36) 195 210
Thailand 0.03 (0.49) 0.37 (2.90) -1.55 (-1.38) 1.02 (1.65) 0.62 (1.44) 0.40 (0.59) 399 174
Turkey -0.07 (-0.60) 0.26 (1.75) -1.85 (-1.31) -0.91 (-1.27) -0.78 (-1.80) -0.13 (-0.20) 254 198
Emg. (EW) 0.02 (0.53) 0.33 (5.75) -3.89 (-6.09) 0.31 (0.86) -0.10 (-0.37) 0.41 (1.05) – 222
Emg (VW) 0.14 (1.93) 0.33 (4.10) -5.18 (-5.73) -0.58 (-1.04) -0.18 (-0.41) -0.40 (-0.60) – 210
Emg. (Pooled) -0.01 (-0.25) 0.32 (2.72) -1.42 (-1.82) -0.15 (-0.27) 0.45 (1.22) -0.60 (-1.21) 4,077 198

48
 Table A4
Differences in Sums and Averages of Coefficients from Fama-MacBeth Return Regressions with Bootstrapped Levels of
Significance for All Strategies, U.S. Large Stocks, 1927–2010

The methods and statistics in Panel A (Panel B) are identical to those in Tables 6 and 7 but the sample includes only large-cap
stocks. Large-cap stocks are those above the median market capitalization of NYSE stocks.

Panel A: Using All Months

# Difference in Sum of Coefficients Difference in Average of Coefficients
Months Returns 3-factor alphas Return 3-factor alphas
IR. RR Diff Diff. t-stat Diff. t-stat Diff. t-stat Diff. t-stat

[12,3] [2] -9 15.95*** (10.51)*** 15.21*** (11.73)*** 1.50*** (4.58)*** 1.68*** (5.78)***
[12,4] [2] -8 13.94*** (10.15)*** 13.38*** (11.31)*** 1.46*** (4.34)*** 1.64*** (5.51)***
[12,4] [3,2] -7 11.93*** (9.15)*** 11.54*** (10.12)*** 0.50* (1.93) 0.64** (2.70)**
[12,5] [2] -7 13.17*** (10.75)*** 12.66*** (12.01)*** 1.56*** (4.60)*** 1.74*** (5.78)***
[12,5] [3,2] -6 11.16*** (9.40)*** 10.83*** (10.38)*** 0.60* (2.24)* 0.73** (3.01)**
[12,6] [2] -6 11.79*** (10.82)*** 11.41*** (11.99)*** 1.60*** (4.68)*** 1.78*** (5.88)***
[12,5] [4,2] -5 10.40*** (8.96)*** 10.12*** (9.87)*** 0.70** (3.01)** 0.77*** (3.69)***
[12,6] [3,2] -5 9.78*** (9.08)*** 9.58*** (9.99)*** 0.64** (2.34)* 0.78*** (3.13)**
[12,7] [2] -5 9.72*** (10.08) ***
9.36*** (11.04)*** 1.53*** (4.41)*** 1.71*** (5.51)***
[12,6] [4,2] -4 9.02*** (8.30)*** 8.87*** (9.15)*** 0.74** (3.06)** 0.82*** (3.76)***
[12,7] [3,2] -4 7.72*** (7.89)***
7.53*** (8.62)*** 0.58* (2.06)* 0.70** (2.76)**
[12,8] [2] -4 8.39*** (9.78)*** 8.24*** (10.86)*** 1.60*** (4.49)*** 1.79*** (5.65)***
[12,6] [5,2] -3 7.64*** (6.89)***
7.61*** (7.66)*** 0.64** (2.91)** 0.70** (3.59)***
[12,7] [4,2] -3 6.95*** (6.83)*** 6.82*** (7.50)*** 0.68** (2.70)** 0.74** (3.30)***
[12,8] [3,2] -3 6.39*** (7.08)***
6.41*** (7.89)*** 0.64** (2.20)* 0.78*** (2.96)**
[12,9] [2] -3 7.40*** (9.84)*** 7.25*** (10.95)*** 1.77*** (4.93)*** 1.94*** (6.07)***
[12,7] [5,2] -2 5.57*** (5.18)***
5.56*** (5.81)*** 0.57* (2.49)** 0.63** (3.07)**
[12,8] [4,2] -2 5.62*** (5.80)*** 5.69*** (6.56)*** 0.74** (2.82)** 0.82*** (3.50)***
[12,9] [3,2] -2 5.39*** (6.53)***
5.41*** (7.36)*** 0.82** (2.71)** 0.94*** (3.47)***
[12,10] [2] -2 5.24*** (8.14)***
5.16*** (8.91)*** 1.68*** (4.52)*** 1.84*** (5.52)***
[12,7] [6,2] -1 3.50** (3.07)**
3.51** (3.46)*** 0.37* (1.73) 0.40** (2.08)*
[12,8] [5,2] -1 4.24** (4.03)***
4.44*** (4.73)*** 0.64** (2.61)** 0.71** (3.27)***
[12,9] [4,2] -1 4.63*** (5.00)***
4.70*** (5.76)*** 0.92*** (3.32)*** 0.98*** (4.02)***
[12,10] [3,2] -1 3.23* (4.31)***
3.33** (4.92)*** 0.72** (2.28)* 0.83*** (2.92)**
[12,11] [2] -1 4.04** (7.40)***
4.07*** (8.18)*** 1.97*** (5.00)*** 2.12*** (5.97)***
[12,8] [6,2] 0 2.17 (1.88) 2.39* (2.33)* 0.43* (1.88) 0.48** (2.33)**
[12,9] [5,2] 0 3.25* (3.17)*** 3.45** (3.85)*** 0.81** (3.17)** 0.86*** (3.85)***
**
[12,10] [4,2] 0 2.46 (2.82) 2.62* (3.37)*** 0.82** (2.82)** 0.87*** (3.37)***
[12,11] [3,2] 0 2.04 (2.97)** 2.24 (3.62)*** 1.02*** (2.97)** 1.12*** (3.62)***
[12] [2] 0 2.20 (5.09)*** 2.42* (6.11)*** 2.20*** (5.09)*** 2.42*** (6.11)***
[12,8] [7,2] 1 0.84 (0.67) 1.27 (1.14) 0.42* (1.93) 0.48** (2.44)**
[12,9] [6,2] 1 1.18 (1.03) 1.40 (1.40) 0.61* (2.48)** 0.63** (2.95)**
[12,10] [5,2] 1 1.08 (1.09) 1.36 (1.57) 0.72** (2.63)** 0.76*** (3.16)**
[12,11] [4,2] 1 1.27 (1.53) 1.53 (2.08)* 1.11*** (3.52)*** 1.16*** (4.10)***
[12] [3,2] 1 0.19 (0.31) 0.59 (1.06) 1.25*** (3.23)** 1.42*** (3.99)***
[12,9] [7,2] 2 -0.15 (-0.12) 0.28 (0.25) 0.60* (2.52)** 0.64** (3.05)**
[12,10] [6,2] 2 -0.98 (-0.87) -0.69 (-0.70) 0.52* (1.95) 0.53** (2.29)**
[12,11] [5,2] 2 -0.11 (-0.11) 0.27 (0.33) 1.01*** (3.40)*** 1.04*** (3.95)***
[12] [4,2] 2 -0.58 (-0.74) -0.12 (-0.18) 1.34*** (3.72)*** 1.46*** (4.41)***
[12,9] [8,2] 3 -1.14 (-0.83) -0.72 (-0.61) 0.64** (2.82)** 0.66** (3.34)***
[12,10] [7,2] 3 -2.32 (-1.82) -1.81 (-1.63) 0.50* (1.96) 0.53** (2.34)**
[12,11] [6,2] 3 -2.18 (-1.94) -1.78 (-1.84) 0.81** (2.78)** 0.81*** (3.17)**
[12] [5,2] 3 -1.95 (-2.08) -1.38 (-1.69) 1.24*** (3.57)*** 1.34*** (4.25)***
[12,10] [8,2] 4 -3.31 (-2.38) -2.80 (-2.36) 0.54* (2.18)* 0.55** (2.53)**
[12,11] [7,2] 4 -3.51 (-2.78) -2.90 (-2.64) 0.80** (2.80)** 0.82*** (3.22)***
[12] [6,2] 4 -4.02 (-3.63) -3.43 (-3.60) 1.04*** (3.03)** 1.11*** (3.59)***
[12,10] [9,2] 5 -5.47 (-3.72) -4.89 (-3.87) 0.43* (1.80) 0.43** (2.05)*
[12,11] [8,2] 5 -4.50 (-3.25) -3.89 (-3.29) 0.84** (3.01)** 0.84*** (3.40)***
[12] [7,2] 5 -5.35 (-4.26) -4.55 (-4.19) 1.03*** (3.05)** 1.11*** (3.64)***
[12,11] [9,2] 6 -6.66 (-4.49) -5.97 (-4.70) 0.72** (2.68)** 0.71** (2.97)**
[12] [8,2] 6 -6.34 (-4.59) -5.54 (-4.72) 1.07*** (3.23)** 1.13*** (3.78)***
[12,11] [10,2] 7 -7.86 (-5.06) -7.06 (-5.30) 0.74** (2.86)** 0.73** (3.15)**
[12] [9,2] 7 -8.51 (-5.70) -7.63 (-6.00) 0.95*** (2.93)** 1.01*** (3.41)***
[12] [10,2] 8 -9.70 (-6.16) -8.71 (-6.48) 0.97*** (3.05)** 1.03*** (3.53)***
[12] [11,2] 9 -11.55*** (-7.01)*** -10.37*** (-7.37)*** 0.92*** (2.99)** 0.98*** (3.48)***
49
 Panel B: Excluding Month -2
# Difference in Sum of Coefficients Difference in Average of Coefficients
Months Returns 3-factor alphas Return 3-factor alphas
IR. RR Diff Diff. t-stat Diff. t-stat Diff. t-stat Diff. t-stat
[12,4] [3,3] -8 12.03*** (9.00)*** 11.37*** (9.81)*** -0.45 (-1.45) -0.37 (-1.29)
[12,5] [3,3] -7 11.27*** (9.35)*** 10.66*** (10.17)*** -0.35 (-1.09) -0.27 (-0.92)
[12,5] [4,3] -6 10.50*** (9.06)*** 9.94*** (9.81)*** 0.27 (1.04) 0.29 (1.22)
[12,6] [3,3] -6 9.89*** (9.17)***
9.40*** (9.90)*** -0.31 (-0.95) -0.23 (-0.76)
[12,6] [4,3] -5 9.12*** (8.56)***
8.69*** (9.21)*** 0.31 (1.15) 0.33 (1.36)
[12,7] [3,3] -5 7.82*** (8.23)***
7.35*** (8.75)*** -0.37 (-1.11) -0.30 (-0.99)
[12,6] [5,3] -4 7.74*** (7.23)***
7.43*** (7.79)*** 0.32 (1.33) 0.34 (1.60)
[12,7] [4,3] -4 7.05*** (7.29)***
6.64*** (7.75)*** 0.25 (0.90) 0.26 (1.03)
[12,8] [3,3] -4 6.49*** (7.65)***
6.23*** (8.24)*** -0.31 (-0.90) -0.22 (-0.70)
[12,7] [5,3] -3 5.68*** (5.63)***
5.38*** (6.02)*** 0.25 (1.02) 0.26 (1.21)
[12,8] [4,3] -3 5.72*** (6.40)***
5.52*** (6.92)*** 0.31 (1.07) 0.34 (1.30)
[12,9] [3,3] -3 5.50*** (7.26)*** 5.24*** (7.86)*** -0.13 (-0.37) -0.07 (-0.20)
[12,7] [6,3] -2 3.61** (3.42)***
3.34** (3.57)*** 0.08 (0.37) 0.07 (0.34)
[12,8] [5,3] -2 4.34*** (4.50)*** 4.26*** (4.96)*** 0.32 (1.21) 0.35 (1.50)
[12,9] [4,3] -2 4.73*** (5.65)***
4.53*** (6.18)*** 0.49 (1.60) 0.50 (1.83)
[12,10] [3,3] -2 3.33** (5.13)*** 3.15** (5.42)*** -0.23 (-0.62) -0.17 (-0.51)
*
[12,8] [6,3] -1 2.28 (2.16) 2.21* (2.37)* 0.14 (0.59) 0.15 (0.70)
[12,9] [5,3] -1 3.35** (3.62)*** 3.27** (4.06)*** 0.49 (1.78) 0.50 (2.09)
[12,10] [4,3] -1 2.57* (3.38)***
2.44* (3.63)*** 0.39 (1.23) 0.39 (1.37)
[12,11] [3,3] -1 2.14 (3.91)*** 2.06 (4.19)*** 0.07 (0.17) 0.12 (0.33)
[12,8] [7,3] 0 0.95 (0.83) 1.09 (1.08) 0.19 (0.83) 0.22 (1.08)
[12,9] [6,3] 0 1.29 (1.24) 1.22 (1.36) 0.32 (1.24) 0.31 (1.36)
[12,10] [5,3] 0 1.19 (1.36) 1.19 (1.56) 0.40 (1.36) 0.40 (1.56)
[12,11] [4,3] 0 1.37 (2.01) 1.35 (2.23)* 0.69 (2.01) 0.68* (2.23)*
[12] [3,3] 0 0.29 (0.69) 0.41 (1.05) 0.29 (0.69) 0.41 (1.05)
[12,]9 [7,3] 1 -0.04 (-0.04) 0.10 (0.10) 0.37 (1.48) 0.37 (1.73)
[12,1]0 [6,3] 1 -0.88 (-0.88) -0.86 (-1.00) 0.22 (0.82) 0.20 (0.84)
[12,11] [5,3] 1 0.00 (-0.01) 0.10 (0.14) 0.69* (2.21)* 0.68* (2.49)*
[12] [4,3] 1 -0.47 (-0.77) -0.30 (-0.55) 0.92** (2.39)* 0.97** (2.81)**
[12,]9 [8,3] 2 -1.04 (-0.83) -0.89 (-0.84) 0.45 (1.92) 0.44 (2.19)*
[12,1]0 [7,3] 2 -2.21 (-1.94) -1.98 (-2.00) 0.27 (1.01) 0.27 (1.15)
[12,11] [6,3] 2 -2.07 (-2.14) -1.95 (-2.34) 0.52 (1.73) 0.49 (1.85)
[12] [5,3] 2 -1.85 (-2.38) -1.55 (-2.30) 0.92** (2.55)* 0.98** (3.02)**
[12,10] [8,3] 3 -3.20 (-2.54) -2.98 (-2.78) 0.36 (1.39) 0.33 (1.51)
[12,11] [7,3] 3 -3.40 (-3.05) -3.07 (-3.18) 0.56 (1.93) 0.55 (2.15)
[12] [6,3] 3 -3.92 (-4.14) -3.60 (-4.45) 0.75* (2.13) 0.78** (2.49)*
[12,10] [9,3] 4 -5.37 (-4.01) -5.06 (-4.42) 0.25 (1.03) 0.23 (1.07)
[12,11] [8,3] 4 -4.39 (-3.53) -4.07 (-3.85) 0.65 (2.29)* 0.62 (2.50)*
[12] [7,3] 4 -5.25 (-4.78) -4.72 (-4.98) 0.79* (2.31)* 0.85** (2.75)**
[12,11] [9,3] 5 -6.56 (-4.88) -6.15 (-5.37) 0.54 (1.99) 0.51 (2.12)
[12] [8,3] 5 -6.24 (-5.06) -5.72 (-5.47) 0.88** (2.62)** 0.92** (3.04)**
[12,11] [10,3] 6 -7.75 (-5.51) -7.24 (-6.01) 0.59 (2.24)* 0.56 (2.39)*
[12] [9,3] 6 -8.41 (-6.25) -7.80 (-6.83) 0.77* (2.35)* 0.81** (2.72)**
[[12] [10,3] 7 -9.60 (-6.74) -8.89 (-7.32) 0.82* (2.55)* 0.85** (2.92)**
[12] [11,3] 8 -11.44*** (-7.64)*** -10.54 *** (-8.26)*** 0.78* (2.52)* 0.82** (2.91)**

50