Momentum Strategies with Stock Index Exchange-Traded Funds
Yiuman Tse
Department of Finance
University of Missouri – St. Louis
St. Louis, MO 63121
Email: [email protected]
Tel: +1 314 516 6828
Previously reported momentum profits may not be available to individual investors who have
more trading constraints. Therefore, I examine the profitability of momentum strategies with
international iShares and US sector exchange-traded funds (ETFs) traded on the NYSE. The
index ETFs provide individual investors easy access to international stock markets and US
sectors for asset allocations. Using both cross-sectional and time series momentum strategies, in
contrast to prior research, I find that momentum profits are insignificant for the late 1990s−2013
period. The results indicate that the US and international stock markets are more efficient in
recent years. The results are consistent with Chordia, Roll, and Subrahmanyam (2011), who
show that the increased trading activity of the US market has been accompanied by improved
market efficiency because of lower trading costs and increased sensitivity of arbitrage activity to
past returns.
January 2014
JEL classifications: G11; G14; 15;
Keywords: Momentum; market efficiency; index exchange-traded funds; asset allocation
1
Momentum Strategies with Stock Index Exchange-Traded Funds
The profitability of momentum strategies (buying past winners and selling past losers) over an
intermediate holding period is one of the most resilient market anomalies. Few financial issues
have drawn more controversies and interests than these strategies from both academia and the
financial industry. Using data from 1965 to 1989, Jegadeesh and Titman (1993) show that US
stocks with the best performance over the past 3–12 months continue to outperform the worstperformance stocks over the next few months. Evidence from international stocks and other asset
classes is also extensively provided by many papers.
Financial theories have not yet conclusively explained why momentum exists, although
initial underreaction and delayed overreaction of stock prices are the most cited explanations.
Some papers apply momentum strategies across asset classes. Of particular interest, Chan,
Hameed, and Tong (2000) report significant momentum profits by reallocating funds across
different international stock indices. Doeswijk and van Vliet (2011) find significant momentum
profits using global sector indices.
In this study, I use iShares exchange-traded funds (ETFs) traded on the New York Stock
Exchange (NYSE) of 20 countries of the Asia–Pacific region, Europe, and the Americas as well
as the US index ETF to examine the momentum profitability. International iShares of Barclay
Global Investors have grown in popularity because they allow US investors to diversify their
holdings in international markets. The sample countries are similar to those in Chan et al. (2000),
but they use stock indices. As they point out, individual investors may not be able to implement
their strategies because short selling is restricted and stock index futures are not available in
2
some countries. Indeed, individual investors may not be able to exploit the momentum profits
reported in prior research.
Focusing on domestic sector asset allocation, I also examine momentum profits using US
sector ETFs. The Select Sector SPDRs Funds partition the S&P 500 into nine sector index ETFs.
I use the terms “industry” and “sector” interchangeably in this study, although industry generally
describes a much more specific group of companies. Moskowitz and Grinblatt (1999) show that
once US stock returns are adjusted for industry effects, momentum profits from individual stocks
become insignificant. Buying stocks from past winning industries and selling stocks from past
losing industries offer higher profits than individual stock momentum strategies. Related to the
industry momentum effect, using monthly returns, Hong, Torous, and Valkanov (2007) find that
industries lead stock markets in the US and eight developed countries.
An important benefit ETFs provide to individual investors is ease of entry. ETFs can be
sold short with no constraints and purchased on margin. This is important for investors hoping
for quick entry to capitalize on the market’s upward and, particularly, downward momentum.
ETFs also avoid the rolling over mechanism for futures contracts.1 Nevertheless, using onemonth to twelve-month returns over the periods of January 1997 to June 2013 for the
international ETFs and January 1999 to June 2013 for the US sector ETFs, I find that the
Jegadeesh-Titman momentum strategies do not provide any significant profits either
econometrically or statistically. Replicating the approach of Chan et al. (2000) with weekly data
using iShares ETFs, I still find no significant profits.
The momentum strategies of Jegadeesh and Titman (1993) and all the previously
mentioned papers depend on the relative performance of securities or asset classes in the cross-
3
section. Moskowitz, Ooi, and Pedersen (2012) show that momentum strategies can be used in the
time series based on the past performance of the securities only: buying (selling) if the past 1–12
months offer positive (negative) returns. They report significant time series momentum in the
equity index and other futures markets. Although I find significant excess returns from time
series momentum of iShares for one-month holding returns, momentum returns generally do not
outperform the buy-and-hold returns. The short-term strategies also require high turnover and,
accordingly, profits may be eliminated by high transaction costs.
The overall results indicate that momentum profits are not readily available to individual
investors using stock index ETFs. The results are also consistent with the improved market
efficiency and higher correlation between markets in recent years. In particular, Chordia, Roll,
and Subrahmanyam (2011) show that the increased trading activity of the US market has been
accompanied by improved market efficiency with prices conforming more closely to random
walks and declining cross-sectional predictability of returns.
The article is organized as follows. In the next section, I summarize the literature of
cross-sectional and time series momentum strategies. The subsequent sections describe the ETF
data and examine the results of momentum profits. The final section concludes the study and
discusses the implications for stock market efficiency.
Prior Research
Jegadeesh and Titman (1993) offer one of the first extensive studies on the momentum effect. A
self-financing strategy that buys the winners and sells the losers of US stocks based on the past
3–12 months yields a 1% monthly return over the next few months. Jegadeesh and Titman (2001)
4
show that momentum returns persist in the 1990s and they report return reversals over the long
run.
Several papers report that momentum profits are higher for stocks with certain
characteristics, such as small size and low analyst coverage (Hong et al., 2000), high analyst
forecast dispersion (Zhang, 2006), and high market-to-book ratios (Daniel and Titman, 1999).
However, Bandarchuk and Hilscher (2013) show that such strategies enhance momentum profits
simply by trading in stocks with more extreme past returns or higher volatility. More extreme
past returns (the only effect) are associated with higher momentum. Avramov et al. (2007) find
significant momentum profits in the stocks of low-grade firms, but not in high-grade firms.
Jostova et al. (2013) report similar patterns in the US high-grade and low-grade corporate bonds.
Conrad and Kaul (1998) suggest that the momentum profits are attributable to rational
time variation in expected returns. However, Jegadeesh and Titman (2002) show that the Conrad
and Kaul results are driven by small sample bias. The momentum effect can hardly be explained
by risk-based models either, including the Fama–French three-factor model. In contrast,
behavioral models suggest that investors are slow to adjust to new information (underreaction).
But once a trend is established, investors will over extrapolate recent results and simply follow
the herd (herd behavior and delayed overreaction), resulting in short-term momentum and longterm reversals.2
Lesmond, Schill, and Zhou (2004) report that the stocks that generate large momentum
profits are also the stocks with high trading costs, which will wipe out the profits. Similarly,
Korajczyk and Sadka (2004) show that losers are biased to small firms with low liquidity and
short sale restrictions. Korajczyk and Sadka also indicate that transaction costs, in the form of
5
spreads and price impacts of trades, do not fully explain the return persistence of past winner
stocks.
Moskowitz and Grinblatt (1999) show that the momentum effects are mainly contributed
by industry momentum strategies that buy (sell) stocks from past winning (losing) industries.
Buying past winning stocks and selling past losing stocks within the industry offer insignificant
profits. O’Neal (2000) uses mutual funds to capture industry momentum, but I use sector ETFs.
What causes industry momentum? Moskowitz and Grinblatt argue that investors may exhibit
overconfidence and self-attribution in certain types of industries over time. Investors may also be
more conservative in updating their priors about some industries that they are not familiar with.
These behavior biases cause the temporal relation, and therefore momentum effect, between
industries. Hong et al. (2007) further suggest that the returns of some industry portfolios that are
informative about macroeconomic fundamentals will lead the aggregate market and, accordingly,
other industries.
Chordia et al. (2011) find that the increased trading activity of the US market has been
accompanied by improved market quality (such as lower bid-ask spread) and market efficiency
(such as lower return serial correlation). Turnover and information-based trading have been
increased more for stocks with higher level of institutional holdings. They point out that
increased sensitivity of trading activity to past returns represents arbitrage activity that has
reduced the cross-sectional predictability of equity returns. More specifically, the momentum
effect (proxies by the cumulative return on the stock over the past six months) significantly
influences the current monthly return during the period of 1993-2000, but the effect is
6
insignificant during the more recent period of 2001-2008. This declining momentum effect
reported in Chordia et al. is consistent with the current results also based on recent years.
Many papers have examined and confirmed the momentum effect on international stocks,
commodities, currencies, and bonds.3 Some papers, such as Chan, Hameed, and Tong (2000),
Blitz and Vliet (2008), and Doeswiijk and van Vliet (2011), apply momentum strategies across
asset classes. Using the equity indexes of 23 countries (9 from the Asia–Pacific region, 11 from
Europe, 2 from North America, and 1 from Africa), Chan et al. report significant momentum
profits by the zero-cost strategy of buying the past winning index and selling the losing index for
the period 1980–1995. Doeswiijk and van Vliet find significant momentum profits with 10
global sector indexes for the period 1970–2008.
Data
It is worth noting that significant results of all previous studies may not be readily available to
individual investors because of high transaction costs, product accessibility, short sale constraints,
and other market imperfections. The Morgan Stanley Capital International (MSCI) indexes used
in Chan et al. (2000) for international asset allocation and Doeswijk and van Vliet (2011) for
global sector allocation are not tradable and are costly for investors to replicate all the index
constituents. In this paper, I examine the momentum effect with the international iShares
exchange-traded funds (ETFs) and the US sector index ETFs. The international and US sector
ETFs facilitate investors for international asset allocation and domestic sector allocations,
respectively, to fit their particular investment needs or goals.
7
In particular, international ETFs of Barclay Global Investors provide US investors with
an easy way to diversify their holdings in foreign stocks. US sector ETFs of State Street Bank
and Trust Company allow investors to invest the S&P 500 index with the ability to over-weight
or under-weight particular sectors. All these international and US sector ETFs are traded on the
New York Stock Exchange (NYSE) in US dollars with the same trading hours and mechanisms.
The iShares sample countries are similar to those in Chan et al. (2000), but they use
foreign local stock indices (with different market structures) and assume that there is no
restriction for investors in holding long or short positions in individual markets worldwide.
However, indexes are unmanaged and investors cannot invest directly in the index. ETFs provide
individual investors with easy access to international markets for both short selling and buying
on margin. Investors may favor ETFs for quick entry to capitalize on the market momentum.
Table I lists the 21 international ETFs, 20 iShares plus the US ETF represented by SPY:
seven from the Asia–Pacific region (Australia, Hong Kong, Japan, Malaysia, Singapore, Taiwan,
and South Korea), 10 from Europe (Sweden, Germany, Italy, Belgium, Switzerland, the
Netherlands, Austria, Spain, France, and the U.K.), and four from the Americas (Canada, Mexico,
Brazil, and the U.S.). I collect the daily closing prices and dividends from the NYSE TAQ data.
The sample period spans sixteen and a half years from January 1997 through June 2013 (198
months), except for Taiwan, South Korea, and Brazil from January 2001 through 2013 (150
months). These three iShares started trading in May to July 2000 and the other iShares in April
1996, and the US index ETF in January 1993. The US index ETF is much more active than the
others, distantly followed by Brazil and Japan.
8
All these 21 ETFs offered positive returns with a monthly average of 0.48% during the
sample period. Mexico has the highest monthly return of 1.05% and a Sharpe ratio of 0.095. The
U.S. has a return of 0.54%, close to the group average. All the Sharpe ratios calculated with the
US Treasury bill rates are positive except for Japan. The US market has the lowest volatility with
a standard deviation of 4.67%; therefore, the Sharpe ratio of SPY, 0.069, is higher than most
countries and the average, 0.036.
The S&P categorizes US stocks into nine sectors as listed in the last panel of Table 1. The
Select Sector SPDRs Funds divide the S&P 500 into the following index ETFs: materials (XLB),
energy (XLE), financial (XLF), industrial (XLI), technology (XLK), consumer staples (XLP),
utilities (XLU), health care (XLV), and consumer discretionary (XLY). The sector ETFs started
trading in December 1999 and I use the thirteen-and-a-half-year sample from January 2000
through June 2013. I do not examine the ETFs on global sectors because of their short trading
history since September 2006. Doeswijk and van Vliet (2011) find significant momentum profits
using 10 global sector indexes for the period of 1970–2008.
Although these nine ETFs together comprise all the companies included in the S&P 500,
some important sectors may be missed in this study. A notable sector is the real estates, and the
first ETF that tracks the U.S. equities in the real estate sector, IYR, is traded in June 2000.
However, including IYR does not change the results qualitatively.
Among the nine US sectors, energy has the highest monthly return (0.815%) and Sharpe
ratio (0.098). The next six sectors have similar return characteristics, while the technology and
financial sectors give the lowest returns (0.033% and 0.048%), highest volatility (7.61% and
9
6.81%), and negative Sharpe ratios. Technology is also the most active sector, closely followed
by energy. In the following, I describe the results of international ETFs before the sector ETFs.
Relative-Strength Momentum Strategy
I examine 25 combinations of five ranking periods denoted by K (one, three, six, nine, and
twelve months) and five holding periods denoted by H. For example, with K = 6 and H = 3, I
rank the ETFs based on their past six-month returns and form a self-financing investment
strategy by buying the winners and selling the losers for the next three months. Previous studies
have used two different sets of weights to form the zero-cost portfolios. Jegadeesh and Titman
(1993, 2001) consider the top 10% of the winners and the bottom 10% of the losers without
using the stocks in the middle range. Some papers also consider the top and bottom 20% or 30%.
Another approach employed by Lehmann (1990), Chan et al. (2000), and others holds assets in
proportion to their market-adjusted returns. This approach invests in all assets (not just the top
and the bottom) and, therefore, incurs more transaction costs.
Specifically, for the first approach, based on their past K-month returns, I buy the top
three iShares with an equal weight of 1/3 and sell equally in the bottom three iShares, holding
this portfolio for H months.4 For the US sector ETFs, I buy the top ETF and sell the bottom ETF.
According to the second approach, the weight of each ETF i in month t equals
wit = (ri,t−1 − rm,t−1)/N
10
(1)
where ri,t−1 is the ETF’s past K-month return at time t−1, rm,t−1 is the corresponding return on the
equal-weighted average of all the ETFs, and N is the total number of ETFs (21 for iShares and 9
for sector ETFs).
I use the same procedures as Jegadeesh and Titman (1993, 2001) and many others to
increase the power of test statistics that are based on overlapping holding returns. In any given
month t, the strategies hold a series of portfolios that are chosen in the current month as well as
the previous H−1 months. These procedures create a time series of monthly returns where each
month’s return is the average of H strategies with an equal weight of 1/H.
I report the results of international ETFs of both weighting approaches in Panel A of
Table 2. These two approaches give similar results and show that the momentum profits are
insignificant with all the t-statistics being less than 1.40 in magnitude. As in Chan et al. (2000),
the t-statistics are calculated using the Newey-West (1987) correction of standard errors for
heteroscedasticity and autocorrelation, although the results are almost the same with the
conventional t-statistics. All the abnormal monthly returns are also less than 0.5%, except the
strategy of K (ranking period) = 6 and H (holding period) = 1 for the first weighting approach,
which gives a return of 0.55% with a t-value of 1.09. Moreover, all the strategies with K = 12
provide insignificantly negative returns. These results are in sharp contrast with the significant
results – both statistically and economically – reported in Jegadeesh and Titman (1993, 2001).
The use of the 12-minus-1 month momentum strategy (i.e., a trailing 12-month period,
excluding the most recent month, and a one-month holding period) has become popular among
practitioners and academics. Novy-Marx (2012) further shows that momentum is primarily
driven by the past performance from 12 to seven months prior to portfolio formation and he
11
suggests excluding the recent six months for the ranking period. Nevertheless, I find that results
using both of these types of momentum strategies are also insignificant. The monthly momentum
profits are 0.015% (t = 0.03) and 0.084% (t = 0.19).
To compare the current results with those of Chan et al. (2000), who also use
international stock indices, on a similar basis, I closely follow their way of forming weekly
returns. For the ranking period, weekly returns are taken on Wednesdays. For the holding period,
weekly returns are taken on Thursdays. Since the ranking period ends before the holding period
starts, the investment strategies do not use price information from the holding period. Similar to
Chan et al., I report the results of the strategies of 1, 2, 4, 12, and 26 weeks with the same values
of K and H in Panel B. I also include the results of Chan et al. (Table 2 of their paper) with the
same weighting approaches of Eq. (1) on the right-hand side of Panel B. Again, all my results are
statistically and economically insignificant. For example, the strategy of K = H = 2 weeks yields
0.004% (t = 0.06) per week. In contrast, Chan et al. find significant weekly returns of 0.48% (t =
6.31).
As expected, the results of Table 2 are still insignificant if I use a regression analysis with
the three well-known factors of Fama–French (1993) and the momentum factor of Carhart
(1997). The data are obtained from the Kenneth French data library. The results remain the same
if I include a dummy variable for the recent financial crisis from October 2007 through March
2009.
I present the results of US sectors in Panel C of Table 2. Similar to Panels A and B, all
the results are statistically insignificant. Only two cases, [K = 12, H = 1] and [K = 6 and H = 6],
for the first weighting approach yield a momentum profit greater than 0.5% (0.790% and 0.714%,
12
respectively), but the t-statistics are only 1.58 and 1.53. Neither do the 12-minus-1 month
momentum strategy and the Novy-Marx modification provide significant results. All the
momentum profits using the proportional weights are insignificant and smaller than 0.3%.
In the US stock markets, Bandarachuk and Hilscher (2013) show that firm characteristics
(such as size, analyst coverage, and market/book ratios) are irrelevant to momentum profits. The
profits are only associated with past returns. Any characteristic that is correlated with volatility
will produce more extreme past returns, and, consequently, higher momentum profits. Therefore,
I limit the samples to the ETFs with higher volatilities: 12 for the iShares sample and 5 for the
US sector sample. The results not reported, nevertheless, are still insignificant.
What are the reasons for the failure of the momentum strategies in this study? The stock
markets could have been more efficient in recent decades. The sample periods of Jegadeesh and
Titman (2001), 1965–1998, and Chan et al. (2000), 1980–1995, for example, are before mine,
the late 1990s–2013. Doeswijk and van Vliet (2011) do not provide subperiod results of recent
years for their 39-year-period (1970–2008) analysis of global sector allocation. Another reason
can be the increased correlations of international stock markets. As discussed in Solnik and
McLeavey (2008), the diversification benefits of international stocks drop substantially in the
recent decade, particularly during market downturns, because of higher correlations between
markets.5
As shown in Sullivan and Xiong (2012), index trading in index mutual funds, particularly
ETFs, has caused an increased trading commonality among index constituents through the
interactions of market participants, resulting in higher return correlations among stocks. The
iShares ETFs are significantly correlated with each other. The average correlation between SPY
13
and other ETFs is 0.742. For highly correlated markets, the relative strength between winners
and losers is small. Consider the momentum strategy with ranking and holding periods of six
months. The monthly return for the top-performance portfolio is 0.682% (t = 1.31) and that for
the bottom-performance portfolio is 0.449% (t = 0.80), while the equal-weighted average index
of all the ETFs is 0.517% (t = 1.14). The differential between the best and the worst performers
(as well as the average) is small, resulting in minor relative strength profits. Also note that
returns of the best and worst performers are not statistically significant.
The US sector ETFs are also highly correlated with SPY with an average coefficient of
0.743. For the momentum strategy of K = H = 6, the monthly return for the top-performance ETF
is 0.656% (t = 1.57) and that for the bottom-performance ETF is −0.059% (t = −0.10), while the
equal-weighted average index of all the ETFs is 0.366% (t = 1.04). Although the differential
between the best and the worst performers is larger than the corresponding results for the iShares
ETFs, the returns of the best and worst performers as well as the average index are still
insignificant.
Time Series Momentum
A recent paper by Moskowitz, Ooi, and Pedersen (2012) shows that momentum profits can be
derived purely from an asset’s own past return, without using the relative performance of other
assets. Moskowitz et al. term this strategy time series momentum, distinguishing it from the
cross-sectional (or relative-strength) momentum examined in the previous section. They also
show that positive auto-covariance in futures returns drives most of the time series and crosssectional momentum effects. This time series momentum strategy is related to other popular
14
trend-following technical analyses, such as the moving average and breakout systems. However,
time series momentum is likely to be the easiest strategy: buying the asset if its past return is
positive and selling if negative (regardless of the ranking of the asset during the past period). It is
worth noting that the momentum effect examined in Chordia et al. (2011) is closely related to
time series momentum, although the authors describe the results as cross-sectional profitability.
Since time series momentum does not depend on the returns of other assets, high crossmarket correlations shown in the iShares and sector ETFs should not weaken the momentum
profits. I therefore test whether this strategy works with ETFs. Moskowitz et al. (2012) use 58
futures contracts of different asset classes, including nine international equity index futures for
the period of 1975 through 2009. Moskowitz et al. aggregate all the scaled futures returns and
only report the t-statistics from a pooled panel regression. I do not scale the returns because I
want to examine the magnitude of momentum profits and I do not consider other classes (such as
commodities and bonds) that have very different volatilities from stock indexes. Moreover,
Moskowitz et al. point out that their results are similar if they use OLS without adjusting for each
security’s volatility.
I calculate the excess return of each ETF by subtracting the return from the one-month
Treasury bill rate. As noted by Moskowitz et al. (2012), this excess return is virtually the same as
the futures return. I report the individual results of each ETF in Table 3. As in the cross-sectional
momentum strategy, I examine evaluation and ranking and holding periods of one, three, six,
nine, and twelve months. For simplicity, I only report the results of K = H, as well as the case of
[K = 3, H = 1]. The last case reported in the last two columns yields the greatest profits both
economically and statistically. Although almost all of the individual results are not statistically
15
significant, the excess returns are generally positive. Indeed, all the ETFs offer positive returns
for the [3, 1] case. The numbers of ETFs that yield positive returns are 20, 19, 19, 12, and 9 for
the cases of [1, 1], [3, 3], [6, 6], [9, 9] and [12, 12], respectively.
When I stack all the ETFs with a sample size of about 4,000, the pooled excess return of
the [3, 1] case is 0.733% (t = 6.10). The cases of [1, 1], [3, 3], and [6, 6] also offer significant
returns, 0.547% (t = 4.44), 0.332% (3.32), and 0.371% (3.50), but the returns of longer-term
strategies of [9, 9] and [12, 12] are insignificant, 0.088% (0.82) and −0.034% (−0.34). In contrast,
Moskowitz et al. (2012) report significant results for all cases up to 12 months, except [1, 1] and
[3, 1], even after being adjusted for the Fama–French and other factors. It will be interesting to
investigate whether the results remain the same for the recent decade. As noted by Handley
(2012), however, Moskowitz et al. do not break down their 25-year-long sample period into
subperiods.
To further investigate the economic significance of the results, I report the pooled returns
in excess of the buy-and-hold returns. The [3, 1] strategy offers higher returns than the buy-andhold by 0.451% (t = 2.27). The next highest return is given by the [1, 1] case with 0.292% (t =
1.47) in excess of buy-and-hold. Both the 12-minus-1 month and the Novy-Marx strategies give
insignificant results, 0.093% (0.43) and −0.056% (−0.26), respectively.
I notice that time series momentum profits reported in this study are only obtained from
strategies with a holding period of one month. This short-term strategy requires more turnover
and, accordingly, transaction costs that may erode profits. Including a one-way transaction cost
of 0.5% as in Jegadeesh and Titman (1993), I find that the [3, 1] strategy gives a pooled return of
16
0.329% (t = 1.65) more than the buy-and-hold. If I consider a 5% significance level or higher
because of a large sample size, the result is not statistically significant.
For the US sector ETFs, the excess profits presented in Panel B of Table 3 are generally
lower than the results of iShares presented in Panel A. Similar to the iShares, most of the sector
ETFs yield positive, albeit insignificant, excess returns. All sector ETFs offer positive returns for
the 12-minus-1 month strategy as reported in the last two columns of Panel B. This strategy also
gives higher returns than other cases. When I stack all the sectors, the pooled excess return of
this strategy is 0.421 (t = 2.72), but the return is only 0.249% (t = 0.97) more than the buy-andhold. The next highest pooled return in excess of buy-and-hold is given by the [12, 1] case with
0.233% (t = 0.89). The Novy-Marx strategy gives much smaller returns, 0.031% (t = 0.12). For
completeness, I also include the 0.5% transaction cost in the 12-minus-1 month strategy, giving a
pooled return of 0.186% (t = 0.73) more than the buy-and-hold.
In sum, all the profits from cross-sectional momentum are insignificant for both the
iShares and sector ETFs. The time series momentum strategy works better than the crosssectional momentum strategy. However, excess profits from time series momentum of the
iShares only happen for short-term holding periods with higher transaction costs. The excess
profits are even smaller for the sector ETFs.
Conclusions
Momentum investing remains one of the most puzzling market anomalies since the extensive
analysis by Jegadeesh and Titman (1993). Although institutional investors may be able to use
momentum strategies to earn excess profits, individual investors have more trading constraints
17
and higher transaction costs. I examine the profitability of momentum strategies by using
international iShares exchange-traded funds (ETFs) and US sector ETFs traded on the NYSE.
The 21 ETFs across the Asia–Pacific region, Europe, and America and the nine US sector ETFs
provide individual investors with easy access to international and domestic asset allocations. The
ETFs allow individual investors quick entry to capitalize on the market momentum in both
directions with short selling and buying on margin.
I use more recent periods, namely January 1997–June 2013 for iShares and January
1999–June 2013 for US sectors, than prior studies. I do not find significant momentum profits
using the relative-strength performance strategies across these ETFs for all the combinations of
ranking and holding periods of one to 12 months. Relative-strength momentum focuses on the
relative returns of the ETFs in the cross section.
Following the time series momentum strategy of Moskowitz et al. (2012), I investigate
the momentum profits based only on the past returns of an ETF. The results are more significant
than the cross-sectional momentum profits. Nevertheless, the time series momentum profits are
only obtained from strategies with short-term one-month holding periods. This short-term
strategy incurs higher transaction costs because of higher turnover. My results are different from
prior studies using stock indexes in earlier decades, including Chan et al. (2000), Moskowitz et
al., and Doeswijk and van Vliet (2011).
Although a detailed analysis will shed more light on the impact of transaction costs on
momentum strategy, my results show that momentum profits are not readily available to
individual investors with the ETFs. The results based on more recent years suggest that the US
and international stock markets are efficient. The results are consistent with Chordia et al. (2011),
18
who show that the US stock market has been more efficient with higher turnover because of
lower trading costs and increased sensitivity of arbitrage activity to past returns. They show that
the momentum effect (proxies by the impact of the past six-month returns on the current monthly
returns) is significant in the 1990s, but not during the recent period of 2001−2008.
With improved market efficiency over the years, a simple momentum investing rule
should not produce abnormal returns. Future research may examine how the momentum strategy
interacts with the value strategy and other trading rules in the ETFs.6
19
Notes
1. Because futures contracts expire in a year and only the nearby contracts are liquid,
investors must rollover to the nearby contracts every three months for index futures.
2. See the behavioral models of Barberis, Shleifer, and Vishny (1998), Daniel, Hirshleifer,
and Subrahmanyam (1998), and Hong and Stein (1999).
3. See Rouwenhorst (1998, 1999) for international stocks and Asness, Moskowitz, and
Pedersen (2013) and Novy-Marx (2012) for various asset classes such as commodities
and currencies.
4. Moskowitz and Grinblatt (1999) choose the top three and the bottom three among 20
industries. I also use the top two and four and the bottom two and four ETFs: the results
are similar. These and other robust testing results are not reported but are available upon
request.
5. Asness, Israelov, and Liew (2011) find that international diversification can disappoint
over the shorter term because international stock markets have a tendency to crash
together. However, they show that international diversification is still beneficial over the
long term.
6. Asness (1997) published one of the first studies on the interaction of value and
momentum strategies in the US stocks. Asness, Moskowitz, and Pedersen (2013)
examine the returns to value and momentum strategies jointly across different asset
classes.
20
References
Asness, Cliff S. 1997. “The Interaction of Value and Momentum Strategies.” Financial Analyst
Journal, vol. 53, no. 2 (April): 29-36.
Asness, Cliff S., Roni Israelov, and John M. Liew. 2011. “International diversification works
(eventually).” Financial Analysts Journal, vol. 67, no. 3 (June): 24–38.
Asness, Cliff S., Tobias J. Moskowitz, and Lasse Heje Pederson. 2013. “Value and Momentum
Everywhere.” Journal of Finance, vol. 68, no. 3 (June): 929-985.
Avramov, Doron, Tarun Chordia, Gergana Jostova, and Alexander Philipov. “Momentum and
Credit Rating.” Journal of Finance, vol. 62, no. 5 (October): 2503-2520.
Bandarchuk, Pavel, and Jens Hilscher. 2013. “Sources of Momentum Profits: Evidence on the
Irrelevance of Characteristics.” Review of Finance, vol. 17, no. 2 (April): 809-845.
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. 1998. “A Model of Investor Sentiment.”
Journal of Financial Economics, vol. 49: 307-343.
Blitz, David C., and Pim van Vliet. 2008. “Global Tactical Cross-Asset Allocation: Applying
Value and Momentum across Asset Classes.” Journal of Portfolio Management, vol. 34, no. 1:
102-113.
Carhart, Mark M. 1997. “On Persistence in Mutual Fund Performance.” Journal of Finance, vol.
51, no. 5: 1681-1713.
Chan, Kalok, Hameed Allaudeen, and Wilson Tong. 2000. “Profitability of Momentum
Strategies in the International Equity Markets.” Journal of Financial and Quantitative Analysis,
vol. 35, no. 3: 153-172.
Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam. 2011. “Recent Trends in Trading
Activity and Market Quality.” Journal of Financial Economics, vol. 101, no. 3: 243-263.
Conrad, Jennifer, and G. Kaul. 1998. “An Anatomy of Trading Strategies.” Review of Financial
Studies, vol. 11, no.3: 489-519.
Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam. 1998. “Investor Psychology and
Security Market Under- and Overreaction.” Journal of Finance, vol. 53: 1839-1886.
Daniel, Kent, and Sheridan Titman. 1999. “Market Efficiency in an Irrational World.” Financial
Analysis Journal, 55, 28-40.
Doeswijk, Ronald, and Pim van Vliet. 2011. “Global Tactical Sector Allocation: A Quantitative
Approach.” Journal of Portfolio Management, vol. 38, no. 1: 29-47.
21
Fama, Eugene F., and Kenneth R. French. 1993. “Common Risk Factors in the Returns on
Stocks and Bonds.” Journal of Financial Economics, vol. 22, no. 1: 3-56.
Handley, Nicholas J. 2012. “Time Series Momentum (Digest Summary).” CFA Digest, vol. 42,
no.3 (August): 179-181.
Hong, Harrison, and Jeremy C. Stein. 1999. “A Unified Theory of Underreaction, Momentum
Trading and Overreaction in Asset Markets.” Journal of Finance, vol. 54: 2143-2184.
Hong, Harrison, T. Lim, and Jeremy C. Stein. 2000. “Bad News Travels Slowly: Size, Analyst
Coverage, and the Profitability of Momentum Strategies.” Journal of Finance, 55: 265-295.
Hong, Harrison, Walter Torous, and Rossen Valkanov. 2007. “Do Industries Lead Stock
Markets?” Journal of Financial Economics, vol. 83: 367-396.
Jegadeesh, Narasimhan, and Sheridan Titman. 1993. “Returns to Buying Winners and Selling
Losers: Implications for Stock Market Efficiency.” Journal of Finance, vol. 48, no. 1: 65-91.
Jegadeesh, Narasimhan, and Sheridan Titman. 2001. “Profitability of Momentum Strategies: An
Evaluation of Alternative Explanations.” Journal of Finance, vol. 56, no. 2: 699-720.
Jegadeesh, Narashimhan, and Sheridan Titman. 2002. “Cross-Sectional and Time-Series
Determinants of Momentum Returns.” Review of Financial Studies, vol. 15, no. 1: 143-157.
Jostova, Gergana, Stanislava Nikolova, Alexander Philipov, and Christof W. Stahel. 2013.
“Momentum in Corporate Bond Returns.” Review of Financial Studies, vol. 26, no. 7 (July):
1649-1693.
Korajczyk, Robert A., and Ronnie Sadka. 2004. “Are Momentum Profits Robust to Trading
Costs?” Journal of Finance, vol. 59: 1039-1082.
Lehmann, Bruce N. 1990. “Fads, Martingales, and Market Efficiency.” Quarterly Journal of
Economics, vol. 105, no. 1: 1-28.
Lesmond, David A., Michael J. Schill, and Chunsheng Zhou. 2004. “The Illusory Nature of
Momentum Profits.” Journal of Financial Economics, vol. 71: 349-380.
Moskowitz, Tobias J., and Mark Grinblatt. 1999. “Do Industries Explain Momentum?” Journal
of Finance, vol. 54: 1249-1290.
Moskowitz, Tobias J., Yao Hua Ooi, and Lasse Heje Pedersen. 2012. “Time Series Momentum.”
Journal of Fiancial Economics, vol. 104: 228-250.
Newey, Whitney K, and Kenneth D. West. 1987. “A Simple Positive Definite, Heterosdasticity
and Autocorrelation Consistent Covariance Matrix.” Econometrica, vol. 55: 703-705.
22
Novy-Marx, Robert. 2012. “Is Momentum Really Momentum?” Journal of Financial Economics,
vol. 103: 429-453.
O’Neal, Edward S. 2000. “Industry Momentum and Sector Mutual Funds.” Financial Analysts
Journal, vol. 56, no. 4: 37-49.
Rouwenhorst, K. Geert. 1998. “International Momentum Strategies.” Journal of Finance, vol. 53,
no. 1: 267-284.
Rouwenhorst, K. Geert, 1999. “Local Return Factors and Turnover in Emerging Market Stocks.”
Journal of Finance, vol. 54, no. 4 (August): 1439-1464.
Solnik, Bruno, and McLeavey, Dennis. 2008. International Investments. Pearson Addison
Wesley, Boston.
Sullivan, Rodney N., and James X. Xiong. 2012. “How Index Trading Increases Market
Vulnerability.” Financial Analysts Journal, vol. 68, no. 2: 70-84.
Zhang, Frank X. 2006. “Information Uncertainty and Stock Returns.” Journal of Finance, vol.
61: 105-136.
23
Table 1. Exchange-Traded Funds: Countries and US Sectors
Country/
Industry
Asia-Pacific
Australia
Hong Kong
Japan
Malaysia
Singapore
Taiwan
S. Korea
Europe
Sweden
Germany
Italy
Belgium
Switzerland
Netherlands
Austria
Spain
France
U.K.
America
Canada
Mexico
Brazil
U.S.
US Sector
Materials
Energy
Financial
Industrial
Technology
Cons. staples
Utilities
Health care
Cons. discretion.
Symbol
Sample period
Mean
(%)
Std dev
(%)
Sharpe Volume
ratio
(m $)
EWA
EWH
EWJ
EWM
EWS
EWT
EWY
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 2001 - Jun 2013
Jan 2001 - Jun 2013
0.679
0.335
0.019
0.269
0.258
0.297
1.024
6.99
7.93
5.82
10.73
8.51
7.99
9.33
0.067
0.015
−0.033
0.005
0.005
0.019
0.094
644
814
3001
254
329
1575
1801
EWD
EWG
EWI
EWK
EWL
EWN
EWO
EWP
EWQ
EWU
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
0.623
0.443
0.232
0.302
0.584
0.267
0.415
0.501
0.440
0.320
8.12
7.67
7.48
7.03
5.42
6.68
7.92
7.56
6.77
5.13
0.051
0.030
0.003
0.013
0.069
0.008
0.026
0.038
0.034
0.021
72
560
76
24
58
28
50
133
100
220
EWC
EWW
EWZ
SPY
Jan 1997 - Jun 2013
Jan 1997 - Jun 2013
Jan 2001 - Jun 2013
Jan 1997 - Jun 2013
0.666
1.045
0.870
0.536
6.62
8.77
10.75
4.67
XLB
XLE
XLF
XLI
XLK
XLP
XLU
XLV
XLY
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
Jan 1999 - Jun 2013
0.453
0.815
0.048
0.445
0.033
0.375
0.397
0.449
0.524
6.56
6.45
6.81
5.69
7.61
3.60
4.51
4.14
5.60
0.069
532
0.095
1,528
0.067 11,023
0.069 249,227
0.046
0.098
−0.020
0.046
−0.020
0.053
0.047
0.064
0.061
3,741
15,696
16,779
3,940
2,184
1,812
2,590
2,063
2,234
Note: The table reports the summary monthly statistics of 20 iShares ETFs, the S&P500 index
ETF, and nine US sector ETFs. All these ETFs are traded on the New York Stock Exchange
(NYSE) in US dollars with the same trading hours and mechanisms.
24
Table 2. Returns of Relative-Strength Momentum
Panel A: Momentum returns per month (%): Country ETFs or iShares
Equal weights of top- and bottom-three ETFs
Proportional weights
.
K H=1
3
6
9
12
K H=1
3
6
9
12
1 −0.072 0.387 0.190 0.146 0.025
1 −0.057 0.353 0.204 0.175 0.053
(−0.16) (1.34) (0.80) (0.70) (0.13)
(−0.14) (1.37) (0.92) (0.88) (0.29)
3 0.314 0.189 0.109 0.096 0.041
3 0.497 0.384 0.279 0.177 0.137
(0.71) (0.50) (0.32) (0.34) (0.14)
(1.24) (1.01) (0.83) (0.63) (0.48)
6 0.551 0.376 0.233 0.170 −0.081
6 0.425 0.369 0.166 0.179 −0.062
(1.09) (0.81) (0.58) (0.46) (−0.29)
(0.88) (0.83) (0.46) (0.49) (−0.23)
9 0.208 0.118 0.042 −0.144 −0.307
9 0.329 0.139 0.064 −0.124 −0.316
(0.44) (0.27) (0.09) (−0.41) (−0.90)
(0.76) (0.35) (0.15) (−0.38) (−0.99)
12 −0.243 −0.199 −0.299 −0.368 −0.081
12 −0.233 −0.097 −0.237 −0.363 −0.110
(−0.44) (−0.37) (−0.72) (−0.89) (−0.24)
(−0.46) (−0.20) (−0.64) (−1.00) (−0.37)
Panel B: Momentum returns per week (%): Country ETFs or iShares
Equal weights of topProportional weights
and bottom-three ETFs
K=H=1
−0.055 (−0.69)
−0.068 (−1.00)
Chan et al.
0.277 (2.88)
K=H=2
−0.034 (−0.49)
0.004
0.06
Chan et al.
0.483 (6.31)
K=H=4
0.052 (0.68)
0.075
1.13
Chan et al.
0.254 (3.35)
K = H = 12
0.074 (0.03)
0.088
1.19
Chan et al.
0.094 (1.31)
K = H = 26
0.078 (0.93)
0.089
1.11
Chan et al.
0.116 (2.35)
Panel C: Momentum returns per month (%): US sector ETFs
Equal weights of top and bottom ETFs
K H=1
3
6
9
12
1 0.131 0.083 0.065 0.039 0.049
(0.28) (0.27) (0.27) (0.21) (0.33)
3 0.158 0.058 0.291 0.238 0.020
(0.28) (0.13) (0.85) (0.75) (0.07)
6 0.366 0.456 0.714 0.421 0.383
(0.61) (0.93) (1.58) (1.01) (0.94)
9 0.402 0.484 0.188 0.174 −0.061
(0.69) (0.91) (0.36) (0.32) (−0.11)
12 0.790 0.383 0.457 0.178 0.400
(1.53) (0.72) (0.85) (0.31) (0.87)
25
K H=1
1 −0.034
(−0.12)
3 0.067
(0.214)
6 0.258
(0.70)
9 0.212
(0.57)
12 0.248
(0.72)
Proportional weights
3
6
9
−0.020 0.080 0.061
(−0.11) (0.56) (0.50)
0.104 0.149 0.159
(0.36) (0.65) (0.78)
0.274 0.351 0.175
(0.83) (1.21) (0.62)
0.409 0.188 0.189
(1.20) (0.55) (0.54)
0.204 0.206 0.066
(0.57) (0.56) (0.17)
.
12
0.051
(0.48)
0.077
(0.40)
0.201
(0.71)
0.071
(0.20)
0.232
(0.77)
Note: This table reports the returns of relative-strength momentum. The zero-cost portfolios are
formed based on K-month past returns and held for H months. Panels A and C uses monthly
returns for the country and sector ETFs, respectively. Panel B uses weekly returns as in Chan et al.
(2000) for the country ETFs. On the right-hand side of Panel B, results given by Chan et al. are
included for comparison. For the first weighting approach presented in the left-hand side of Panel
A, based on their past K-month returns, the top three iShares are purchased with an equal weight
of 1/3 and the bottom three iShares are sold equally, holding this portfolio for H months. For the
US sector ETFs in Panel C, the top ETF is purchased and the bottom ETF is sold. According to
the proportional weights reported in the right-hand side of each panel, the weight of each ETF i in
month (week) t equals wit = (ri,t−1 − rm,t−1)/N, where ri,t−1 is the ETF’s past K-month (week) return
at time t−1 and rm,t−1 is the corresponding return on the equal-weighted average of all the ETFs. N
= 21 (9) is the total number of country (US sector) ETFs. The t-statistics calculated by the
Newey-West heteroscedasticity and autocorrelation consistent errors are reported in parentheses.
26
Table 3. Excess Returns of Time-Series Momentum
Panel A: Country ETFs or iShares
EWA
EWH
EWJ
EWM
EWS
EWT
EWY
EWD
EWG
EWI
EWK
EWL
EWN
EWO
EWP
EWQ
EWU
EWC
EWW
EWZ
SPY
Pooled returns
Pooled returns
in excess of
buy-and-hold
K=H=1
K=H=3
0.144 (0.29)
0.200 (0.49)
1.130 (1.94)
0.047 (0.10)
0.709 (1.63)
0.476 (1.42)
0.608 (0.68)
0.703 (1.10)
0.513 (0.84)
0.322 (0.70)
0.501 (0.74) −0.452 (−0.88)
0.157 (0.21)
0.336 (0.57)
0.740 (1.35)
0.653 (1.30)
0.546 (1.19)
0.283 (0.64)
0.263 (0.52)
0.205 (0.47)
0.902 (1.54) −0.027 (−0.06)
0.537 (1.42)
0.048 (0.13)
−0.009 (−0.02)
0.430 (1.09)
0.845 (1.28)
0.448 (0.89)
0.768 (1.46)
0.020 (0.05)
0.621 (1.45)
0.207 (0.52)
0.449 (1.17)
0.323 (1.00)
0.536 (1.08)
0.436 (1.17)
0.345 (0.57)
0.492 (0.97)
0.924 (0.97)
1.818 (2.74)
0.246 (0.82)
0.174 (0.57)
0.547 (4.44)
0.292 (1.47)
0.332
0.071
(3.32)
(0.38)
K=H=6
0.557 (1.22)
0.571 (1.15)
0.409 (1.13)
0.681 (1.07)
0.693 (1.24)
−0.508 (−0.99)
0.183 (0.26)
0.778 (1.43)
0.451 (0.93)
0.286 (0.60)
0.839 (1.49)
0.271 (0.80)
0.260 (0.60)
0.145 (0.28)
0.145 (0.35)
0.469 (1.12)
0.513 (1.32)
0.116 (0.32)
0.233 (0.54)
−0.226 (−0.32)
0.488 (1.67)
0.371
0.039
(3.50)
(0.20)
K=H=9
−0.037 (−0.09)
−0.343 (−0.77)
0.215 (0.56)
−0.334 (−0.50)
−0.034 (−0.06)
−0.634 (−1.40)
−0.188 (−0.29)
0.263 (0.46)
−0.007 (−0.02)
0.033 (0.07)
0.526 (0.93)
0.125 (0.37)
0.131 (0.33)
0.410 (0.69)
−0.129 (−0.30)
0.182 (0.46)
0.409 (1.08)
−0.097 (−0.24)
0.288 (0.54)
0.402 (0.52)
0.484 (1.48)
K = H = 12
−0.074 (−0.18)
−0.401 (−0.88)
−0.056 (−0.15)
−0.604 (−1.23)
−0.218 (−0.47)
−0.626 (−1.47)
−0.234 (−0.38)
−0.030 (−0.05)
0.008 (0.02)
0.012 (0.03)
0.304 (0.61)
0.219 (0.66)
0.070 (0.18)
0.678 (1.23)
−0.174 (−0.43)
0.123 (0.33)
0.287 (0.85)
−0.332 (−0.73)
−0.216 (−0.48)
−0.021 (−0.02)
0.355 (1.14)
K =3, H = 1
0.384 (0.74)
0.414 (0.79)
1.019 (2.94)
0.387 (0.45)
0.651 (1.22)
0.704 (1.22)
1.258 (1.91)
0.900 (1.54)
0.749 (1.37)
0.485 (0.94)
0.281 (0.54)
0.298 (0.70)
0.430 (0.85)
1.123 (1.80)
0.412 (0.79)
0.712 (1.49)
0.520 (1.35)
0.963 (2.24)
1.180 (2.19)
2.705 (3.16)
0.425 (1.28)
0.088 (0.82)
−0.185 (−0.95)
−0.034 (−0.34)
−0.376 (−2.17)
0.733 (6.10)
0.451 (2.27)
27
Panel B: US Sector ETFs
XLB
XLE
XLF
XLI
XLK
XLP
XLU
XLV
XLY
Pooled returns
Pooled returns
in excess of
buy-and-hold
K=H=1
K=H=3
0.043 (0.08)
0.317 (0.82)
0.004 (0.01)
0.331 (0.94)
0.516 (1.27)
0.421 (0.78)
0.535 (1.46)
0.491 (1.27)
−0.097 (−0.22)
0.344 (0.74)
−0.146 (−0.49)
0.223 (1.05)
0.926 (3.04)
0.251 (0.90)
0.373 (1.29) −0.193 (−0.94)
0.242 (0.65)
0.080 (0.19)
0.266 (1.98)
0.058 (0.26)
0.252
0.068
(2.00)
(0.29)
K=H=6
−0.245 (−0.80)
0.094 (0.28)
0.405 (0.70)
0.271 (0.67)
0.614 (1.36)
0.193 (0.88)
0.674 (2.24)
0.147 (0.60)
0.230 (0.61)
0.265
0.070
(2.12)
(0.29)
K=H=9
−0.123 (−0.36)
0.146 (0.38)
0.391 (0.66)
0.508 (1.19)
0.224 (0.44)
0.266 (1.25)
0.424 (1.37)
0.198 (0.70)
0.312 (0.80)
0.261
0.054
(1.95)
(0.22)
K = H = 12
−0.439 (−1.02)
0.297 (0.61)
0.192 (0.32)
0.048 (0.14)
0.208 (0.45)
0.269 (1.30)
0.461 (1.63)
0.204 (0.75)
−0.039 (−0.11)
K = 12−1, H = 1
0.210 (0.47)
1.002 (2.46)
0.343 (0.52)
0.578 (1.17)
0.251 (0.44)
0.460 (1.67)
0.470 (1.28)
0.304 (0.91)
0.166 (0.34)
0.133 (0.98)
−0.120 (−0.53)
0.421 (2.72)
0.249 (0.97)
Note: This table reports the excess return (of the one-month Treasury bill rate) of the time series momentum. This strategy
requires buying the asset if its past return is positive and selling if negative (regardless of the ranking of the asset during the
past period). Evaluation and holding periods of 1–12 twelve months are examined. For simplicity, only the results of K = H, as
well as the case of [K = 3, H = 1] in Panel A and the case of 12-minus-1 month in Panel B, are reported. The last case
presented in the last two columns of each panel yields the greatest profits both economically and statistically. The pooled
returns are calculated by stacking all ETF returns in each panel. The t-statistics calculated by the Newey-West
heteroscedasticity and autocorrelation consistent errors are reported in parentheses.
28