Decomposing Time Series

Momentum
Kun Yang, Ph.D., CFA

June 2020
Highlights

 In this paper, we connect the two well-documented momentum effects in financial markets,
time-series momentum (TSMOM, aka trend-following) and cross-sectional momentum
(CSMOM), by introducing another momentum concept — asset-class momentum (ACMOM).
 Building an analytical decomposition framework, we show that both ACMOM and CSMOM are
important components of TSMOM. Furthermore, the ACMOM has been the predominant driver
behind the performance of TSMOM over the past half century. The findings explain why
TSMOM is often found to be superior to CSMOM.
 We further show that the crisis alpha of TSMOM consistently came from the ACMOM of com-
modities and currencies, which is in contrast with the conventional wisdom that the crisis al-
pha of trend-following strategies mainly came from riding on the negative equity momentum.
 Examining the performance of Managed Futures/CTA indices, we find that the trend-following
industry has been increasingly exposed to asset-class momentum, which likely diminishes the
diversification benefits from multi-managers investing.

Acknowledgements:

I am grateful to Eric Sorensen, Edward Qian and Bryan Belton for helpful comments and suggestions;
a special thanks goes to Vic Malla for outstanding research assistance.

2
Introduction

Time-series momentum (TSMOM) refers to the on the TSMOM returns, the intercepts became
persistency of an asset’s price behavior relative insignificant and often negative, while the beta
to its own time-series. That is, positive (nega- of TSMOM remained highly significant.
tive) past price changes tend to be followed by
positive (negative) future price changes. In a In this article, we attempt to bridge the gap be-
seminal work by Moskowitz, Ooi, and Pedersen tween TSMOM and CSMOM by incorporating an-
(2012), the authors documented significant other momentum concept: the asset-class mo-
TSMOM effects across major asset classes and mentum (ACMOM). As a natural extension of
markets. Hurst, Ooi, and Pedersen (2013) further TSMOM, ACMOM refers to the performance per-
showed that TSMOM largely explained the per- sistency of an asset class relative to its own his-
formance of Managed Futures (MF) funds and tory. While not studied as comprehensively as
Commodity Trading Advisors (CTA). TSMOM and CSMOM, ACMOM has been well ap-
plied by practitioners in tactic asset allocation
Cross-sectional momentum (CSMOM), on the (TAA) strategies. For instance, Faber (2007) used
other hand, refers to the persistency of an asset a simple momentum signal (10-month moving
relative performance to its peers (typically average) to tactically allocate capital across five
within the same asset class). The CSMOM ef- asset classes (U.S. stocks, foreign stocks, U.S.
fects were first documented in the equity mar- bonds, commodities, and real estates) and
kets (see Jegadeesh and Titman 1993, and found superior risk-adjusted returns compared
Rouwenhorst 1998 for evidence in the U.S. and to buy-and-hold strategies.
global equity markets, respectively). Asness,
Moskowitz, and Pedersen (2013) extended the We argue that both ACMOM and CSMOM are im-
analysis and found similar CSMOM effects portant components of TSMOM. Intuitively, the
across major asset classes and markets. In price of an asset may be influenced by either
practice, CSMOM has been a key element in fac- global factors that affect the whole asset class
tor investing and risk management. it belongs to, or idiosyncratic factors that only
affect the specific asset. By the same token, an
The linkage between TSMOM and CSMOM has asset’s time-series momentum could be driven
been studied in the literature as well. Generally, by either the momentum of global factors
research has shown that TSMOM is superior to (which manifests itself as ACMOM) or the mo-
CSMOM at similar horizons (Moskowitz, Ooi, and mentum of asset-specific factors (which mani-
Pedersen 2012; Kolanovic and Wei 2015; Goyal fests itself as CSMOM). For instance, an upward
and Jegadeesh 2017). For instance, using a re- momentum of oil price may reflect an upward
gression approach, Moskowitz, Ooi, and Peder- momentum of the commodity asset class
sen (2012) showed that TSMOM is related to but caused by global inflation expectation, or its rel-
not fully captured by CSMOM across different ative upward momentum within the asset class
asset classes (equities, bonds, commodities, due to the OPEC decision to cut supply.
and currencies). Specifically, when regressing
the 12-month TSMOM returns on the 12-month We therefore propose an analytical framework
CSMOM returns, the authors found significant to decompose TSMOM into ACMOM and
beta of CSMOM as well as significant intercepts CSMOM. Specifically, we create three diversified
(alpha) across different markets. On the other portfolios to capture the effects of TSMOM, AC-
hand, when the CSMOM returns were regressed MOM, and CSMOM, respectively. They are math-
ematically interconnected since the TSMOM of

3
an asset is determined by the stronger effect of world: 23 commodity futures (CM), 13 equity in-
its ACMOM and CSMOM at any point in time. We dex futures (EQ), 11 sovereign bond futures (BD),
show that over the past half century and 19 developed and emerging markets cur-
(19702019), while both ACMOM and CSMOM are rency forwards (FX). Table 1 lists the instru-
important components of TSMOM, ACMOM has ments by asset class. Specifically, daily futures
been a predominant driver behind the perfor- prices are collected from the Commodity Re-
mance of TSMOM. Besides the ACMOM and search Bureau (CRB) and Bloomberg, while daily
CSMOM components, TSMOM also exhibits a FX spot prices and local short-term interest
significant alpha that presumably captures the rates are collected from DataStream and
diversification benefits from low correlations Bloomberg. Daily futures returns are calculated
between ACMOM and CSMOM. The decomposi- assuming contracts are rolled to the next near-
tion framework is further applied to analyzing est contract on the first day of the expiration
the crisis alpha/risk premia properties of month. For financial futures (equity index and
TSMOM, as well as explaining the MF/CTA in- sovereign bond futures), we backfill the history
dustry performance. using derived returns from cash instruments.1
Daily FX forward returns are calculated as FX
spot returns adjusted for carry. We construct a
Data
price index for each instrument by compound-
Our data consist of 60+ liquid futures/forwards ing daily returns, from which we can calculate
contracts across major asset classes and the returns and momentum signals of different ho-
rizons.

1
Specifically, we backfill the equity index futures returns with the (gross dividends, excess cash) returns of the underlying
indices. We backfill the U.S. Treasury futures returns with the derived returns from zero-coupon constant maturity yields of
comparable duration. We find the real and derived returns for the equity indices and U.S. Treasury futures are almost identical
in the overlapping periods (perfect monthly correlations and close to a 100% R-square).

4
Table 1. Data Coverage

Commodities Equity Indices Sovereign Bonds Currencies

(CM) (EQ) (BD) (FX)

Soybean Oil Brent Crude Copper Australian SPI 200 U.S. 2-year AUD BRL

Corn WTI Crude Aluminum Canadian TSX 60 U.S. 5-year CAD CLP

Soybean Meal Heating Oil Nickel German DAX U.S. 10-year CHF CZK

Soybean Gas Oil Zinc Spain IBEX 35 U.S. 30-year EUR HUF

Wheat Natural Gas Gold France CAC 40 German 2-year GBP KRW

Cocoa Gasoline Silver U.K. FTSE 100 German 5-year JPY MXN

Cotton Hong Kong Hang Seng German 10-year NOK PLN

Coffee Italy MIB German 30-year NZD RUB

Sugar Japan TOPIX Australian 10-year SEK TRY

Lean Hogs Netherlands Amsterdam IDX Canadian 10-year ZAR

Live Cattle Sweden OMX 30 U.K. 10-year

Singapore Index

U.S. S&P 500

Data Source: Commodity Research Bureau (CRB), Bloomberg, and DataStream.

Appendix A reports summary statistics of matically different across asset classes. Com-
monthly returns for all instruments by com- modities, on average, exhibit the highest annual
modity sector/asset class. Monthly futures re- volatility (31%), followed by equity indices (19%)
turns are available as early as August 1959 (pri- and currencies (12%). The volatility of sovereign
marily agriculture commodities). Monthly FX bonds (6%) has been about one fifth that of
forwards returns begin in February 1973 when commodities. Table 2 also shows evidence of
major currencies adopted the floating exchange high correlations within the same asset class
rate regime. The complete coverage of all in- (except commodities) and low/negative corre-
struments begins in 2003. Table 2 reports the lations across asset classes.
average volatilities, Sharpe ratios, and pairwise
correlations by asset class. Consistent with the
findings in the literature, equities and sovereign
bonds tend to yield long-term positive returns
across countries (average Sharpe ratios are 0.30
and 0.64, respectively), while commodities and
currencies may add or detract value over time
(average Sharpe ratios are 0.14 and -0.07, re-
spectively). In addition, asset volatilities are dra-

5
Table 2. Average Volatilities, Sharpe Ratios, and Pairwise Correlations by Asset Class

Pairwise Correlation
Asset Class Annual Volatility Sharpe Ratio
within Asset Class w/other Asset Class

Commodities 31% 0.14 0.18 0.09

Equity Indices 19% 0.30 0.61 0.09

Sovereign Bonds 6% 0.64 0.66 -0.10

Currencies 12% -0.07 0.45 0.13

Note: The statistics are calculated based on monthly returns over the sample period August 1959–December 2019.

Portfolio Construction

We create three diversified portfolios to capture Following Moskowitz, Ooi, and Pedersen (2012),
time-series momentum, asset-class momen- we use the sign of an asset’s past 12-month re-
tum and cross-sectional momentum. Following turn to determine if the asset is in an upward or
the convention used in both the time-series and downward momentum. A long (short) position
cross-sectional momentum literature, we focus of an asset is entered if its past 12-month return
on the 12-month momentum formation win- is positive (negative). To account for volatility
dow with a one-month holding period. The re- differences across markets, we size each posi-
sults on other momentum formation windows tion to have the same ex-ante volatility.2 Math-
such as one-month and three-month are qual- ematically, an asset i’s weight in the portfolio at
itatively similar and discussed in Appendix C. time t is

Time-series Momentum (TSMOM)

10%
𝑡𝑠
𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛(12𝑚_𝑟𝑒𝑡𝑖,𝑡 ) (1)
𝑁𝑡 × 𝑣𝑜𝑙𝑖,𝑡

where 12𝑚_𝑟𝑒𝑡𝑖,𝑡 is the past 12-month return of risk and is not essential in evaluating empirical
asset i, 𝑣𝑜𝑙𝑖,𝑡 is the annual volatility of asset I, results. Easily proven, each position in the port-
10%
and 𝑁𝑡 is the number of assets at time t. The folio has an ex-ante annual volatility .
𝑁𝑡
scalar of 10% is used to equalize the position

2
The equal volatility approach is a common method to size a trend-following portfolio and is used here for illustration. Baltas
(2015), and Yang, Qian, and Belton (2019) show that the risk parity approach provides better diversification than the equal vol-
atility approach.

6
Asset-Class Momentum (ACMOM)

The ACMOM portfolio is a natural extension of class is in an upward or downward momentum.

the TSMOM portfolio. Instead of holding long or To gauge the momentum of an asset class, we
short positions of individual assets, the ACMOM first construct the asset class return using an
portfolio holds long or short positions of an en- equal-weighting scheme. Mathematically, the
tire asset class, depending on whether the asset return of an asset class I at time t is

1
𝑟𝑒𝑡𝐼,𝑡 = ∑ 𝑟𝑒𝑡𝑖,𝑡 (2)
𝑁𝑡𝐼 𝑖∈𝐼

where 𝑟𝑒𝑡𝑖,𝑡 is the return of an asset i which returns and construct the ACMOM portfolio.
belongs to the asset class I, 𝑁𝑡𝐼 is the number Mathematically, for asset i in the asset class I,
of assets within the asset class I at time t. Sov- its weight in the portfolio is where 12𝑚_𝑟𝑒𝑡𝐼,𝑡 is
ereign bond returns are duration-adjusted to be the past 12-month return of asset class I.
their 10-year equivalent before averaging so that
the bond asset class return is not overly influ-
enced by longer-duration bonds. 3 We then ap-
ply the 12-month return rule to the asset class

𝑎𝑐
10%
𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛(12𝑚_𝑟𝑒𝑡𝐼,𝑡 ) (3)
𝑁𝑡 × 𝑣𝑜𝑙𝑖,𝑡

Cross-Sectional Momentum (CSMOM)

relative performance of an asset to its asset
The CSMOM portfolio holds long or short posi- class. Based on a similar 12-month return rule,
tions of individual assets, depending on the an asset i’s weight in the CSMOM portfolio is

3
We also use an equal-volatility weighting approach to calculate asset class returns and find similar results.

7
 𝑐𝑠
10%
𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛(12𝑚_𝑟𝑒𝑡
⏟ 𝑖,𝑡 − 12𝑚_𝑟𝑒𝑡𝐼,𝑡 )
𝑁𝑡 × 𝑣𝑜𝑙𝑖,𝑡 (4)
12−𝑚𝑜𝑛𝑡ℎ 𝑟𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑟𝑒𝑡𝑢𝑟𝑛

Comparing Equations (1), (3), and (4), we can see position signs (long or short) of TSMOM, AC-
that the TSMOM, ACMOM, CSMOM portfolios are MOM, and CSMOM are determined by the signs
designed to have the same sizing method such of trailing 12-month returns of an asset, an as-
that each position in the portfolio has an ex- set class, and an asset relative to its asset class,
ante annual volatility
10%
. The only difference respectively. 4
𝑁𝑡
lies in the position sign (momentum signal): the 𝑡𝑠 𝑎𝑐 𝑐𝑠
Clearly, 𝑤𝑖,𝑡 , 𝑤𝑖,𝑡 , and 𝑤𝑖,𝑡 are closely related. To
see this, we re-write Equation (1) as

𝑡𝑠
10%
𝑤𝑖,𝑡 = 𝑠𝑖𝑔𝑛[12𝑚_𝑟𝑒𝑡𝐼,𝑡 + (12𝑚_𝑟𝑒𝑡𝑖,𝑡 − 12𝑚_𝑟𝑒𝑡𝐼,𝑡 )]
𝑁𝑡 × 𝑣𝑜𝑙𝑖,𝑡
𝑎𝑐
(5)
𝑤𝑖,𝑡 , if |12𝑚_𝑟𝑒𝑡𝐼,𝑡 | > |12𝑚_𝑟𝑒𝑡𝑖,𝑡 − 12𝑚_𝑟𝑒𝑡𝐼,𝑡 |
= { 𝑐𝑠
𝑤𝑖,𝑡 , if |12𝑚_𝑟𝑒𝑡𝐼,𝑡 | < |12𝑚_𝑟𝑒𝑡𝑖,𝑡 − 12𝑚_𝑟𝑒𝑡𝐼,𝑡 |

In words, an asset TSMOM is determined by the France, and the U.S., and short positions in
stronger effect of its ACMOM and CSMOM countries with negative relative returns, such as
(measured by the magnitude of the 12-month Australia, Canada, and Singapore. The TSMOM
asset class return and relative return, respec- portfolio held the same equity positions as in
tively) at any point in time. the ACMOM portfolio since the magnitude of the
12-month asset class return dominated that of
Table 3 illustrates the relations among the the 12-month relative returns for all equity mar-
TSMOM, ACMOM, and CSMOM portfolios using a kets.
snapshot as of December 2019. Seen from the
table, global equities as an asset class exhibited As regards the currency sleeve, the ACMOM
a strong upward momentum (the 12-month as- portfolio held short positions in all currencies
set class return was around 21%), resulting in due to a negative trailing 12-month return for
long positions of all equity futures in the AC- the asset class (-0.41%). The CSMOM portfolio
MOM portfolio. The CSMOM portfolio held long held long positions in currencies with positive
positions in countries that had positive 12- relative returns such as AUD, CHF and JPY, and
month relative returns such as Germany, short in those with negative relative returns,

4
Note the CSMOM portfolio is not necessarily dollar neutral due to the volatility weighting method. The results are qualita-
tively similar if we impose dollar neutrality. We also experiment with a more concentrated CSMOM portfolio that holds long
(short) positions only within the top (bottom) quantile ranked by relative returns. The results are slightly better.

8
such as EUR, BRL, and CLP. The TSMOM portfo- since the magnitude of their 12-month relative
lio held the same positions as the CSMOM port- returns outweighed that of the asset class re-
folio for all the above currencies except AUD, turn.

Table 3. A Snapshot of TSMOM, ACMOM, and CSMOM Portfolios, December 2019

As- 12-month 12-month

set 12-month Re- TSMOM ACMOM CSMOM
Asset Asset Class Re- Relative Re-
turn Weight Weight Weight
Class turn turn

EQ Australian SPI 200 20.16% 20.86% -0.70% 1.39% 1.39% -1.39%

Canadian TSX 60 18.68% 20.86% -2.18% 1.57% 1.57% -1.57%

German DAX 23.67% 20.86% 2.81% 1.02% 1.02% 1.02%

France CAC 40 27.02% 20.86% 6.15% 1.11% 1.11% 1.11%

Singapore Index 11.53% 20.86% -9.33% 1.08% 1.08% -1.08%

U.S. S&P 500 26.00% 20.86% 5.14% 1.20% 1.20% 1.20%

FX AUD -0.14% -0.41% 0.27% -1.60% -1.60% 1.60%

CHF 1.60% -0.41% 2.01% 1.96% -1.96% 1.96%

EUR -2.12% -0.41% -1.71% -2.10% -2.10% -2.10%

JPY 1.03% -0.41% 1.43% 1.86% -1.86% 1.86%

BRL -2.43% -0.41% -2.03% -0.99% -0.99% -0.99%

CLP -6.99% -0.41% -6.58% -1.25% -1.25% -1.25%

Note: For illustrative purpose, only 12 out of 66 positions are shown in the table.

Empirical Results

We perform historical simulation of the TSMOM, based on an expanding window with exponen-
ACMOM, and CSMOM portfolios from 1970 to tial decay.5 A two-year learning period is re-
2019 during which the majority of asset classes quired before a new asset is included in a port-
are available. Each portfolio is rebalanced folio. Transaction (rollover and rebalancing)
monthly, based on the month-end momentum costs are incorporated in our empirical evalua-
signals and volatility estimates. The asset vola- tion, as explained in Appendix B. The three port-
tilities are estimated using monthly returns folios have different realized volatilities, with

5
Results are robust to different decay rates, such as 1-, 2-, and 5-year half-life. The results reported in this paper are based
on a 2-year half-life decay rate.

9
ACMOM being the most volatile, followed by returns to have the same ex-post volatility of
TSMOM, and CSMOM being the least. For an 10% per annum. Scalars are 2.96, 2.61, and 4.81
equal foot comparison, we scale the portfolio for TSMOM, ACMOM, and CSMOM, respectively.

Returns, Volatilities, and Correlations

Figure 1 plots the (log) cumulative returns of and Yang, Qian, and Bryan (2019). Table 4 re-
each portfolio. All three portfolios exhibited ports summary statistics of monthly returns
positive returns over time, although with much and correlations. TSMOM has a Sharpe ratio of
weaker performance since 2009 (in particular, 0.82, superior to both ACMOM and CSMOM (0.54
the CSMOM portfolio). The pattern of TSMOM and 0.28), likely benefiting from a moderate cor-
performance is broadly consistent with the relation between the two components (0.34).
MF/CTA industry performance, and has been TSMOM is more correlated with ACMOM (0.83)
discussed in details in Moskowitz, Ooi, and than with CSMOM (0.62), indicating that the as-
Pedersen (2012), Hurst, Ooi, and Pedersen (2013), set-class momentum is the more important
driver of an asset’s time-series momentum.

Figure 1. (Log) Cumulative Returns of Momentum Portfolios

1.8

1.3

0.8

0.3

-0.2

TSMOM ACMOM CSMOM

Note: TSMOM, ACMOM, and CSMOM are the returns (net of transaction costs) of the time-series momentum, asset-class momentum,
and cross-sectional momentum portfolios, all standardized to 10% volatility per annum. Sample period: 1970–2019.

10
Table 4. Summary Statistics

Panel A. Returns, Volatilities and Sharpe Ratios

TSMOM ACMOM CSMOM

Annual Excess Return (Ex. Ret.) 8.23% 5.39% 2.84%

Annual Standard Deviation (Std. Dev.) 10% 10% 10%

Sharpe Ratio 0.82 0.54 0.28

Panel B. Monthly Correlations

TSMOM ACMOM CSMOM

TSMOM 1 0.83 0.62

ACMOM 1 0.34

CSMOM 1

Note: Sample period: 1970–2019.

We further perform monthly TSMOM, ACMOM The Sharpe ratios by asset class in the ACMOM
and CSMOM return attributions by asset class portfolio are generally lower, but still reasona-
by simply aggregating the returns within the bly good, ranging from 0.25 for equities to 0.36
same asset class for each portfolio. From Panel for bonds. The CSMOM portfolio, on the other
A of Table 5, all asset classes in the TSMOM hand, has weaker Sharpe ratios (around 0.2 for
portfolio have delivered decent Sharpe ratios, currencies and bonds, and below 0.2 for com-
ranging from 0.39 for equities, to 0.58 for bonds. modities and equities). The superiority of the
TSMOM portfolio over the CSMOM portfolio is
consistent with the evidence documented in
Moskowitz, Ooi, and Pedersen (2012).

11
TABLE 5. Momentum Return Attributions by Asset Class

Panel A. Returns, Volatilities and Sharpe Ratios by Asset Class

TSMOM ACMOM CSMOM

CM EQ FX BD CM EQ FX BD CM EQ FX BD

Annual Ex. Ret. 2.98% 1.43% 1.89% 2.08% 1.94% 0.98% 1.21% 1.36% 0.59% 0.55% 1.00% 0.77%

Annual Std. Dev. 5.31% 3.67% 4.47% 3.61% 6.08% 3.96% 4.86% 3.82% 6.60% 3.35% 4.63% 3.67%

Sharpe Ratio 0.56 0.39 0.42 0.58 0.32 0.25 0.25 0.36 0.09 0.16 0.22 0.21

Panel B. Monthly Correlations

TSMOM ACMOM CSMOM

CM EQ FX BD CM EQ FX BD CM EQ FX BD

TSMOM_CM 1.00 0.14 0.25 0.04 0.72 0.07 0.20 -0.08 0.62 0.18 0.16 0.09

TSMOM_EQ 1.00 0.20 0.10 0.06 0.87 0.15 0.08 0.12 0.41 0.19 0.01

TSMOM_FX 1.00 0.13 0.15 0.15 0.83 0.11 0.14 0.23 0.57 0.09

TSMOM_BD 1.00 -0.01 0.07 0.13 0.85 -0.01 0.07 0.14 0.10

ACMOM_CM 1.00 0.03 0.11 -0.10 0.33 0.13 0.05 0.09

ACMOM_EQ 1.00 0.11 0.08 0.09 0.27 0.19 -0.02

ACMOM_FX 1.00 0.10 0.11 0.18 0.41 0.06

ACMOM_BD 1.00 -0.02 0.05 0.14 -0.01

CSMOM_CM 1.00 0.07 0.10 0.00

CSMOM_EQ 1.00 0.09 0.04

CSMOM_FX 1.00 0.02

CSMOM_BD 1.00

Note: The TSMOM, ACMOM, and CSMOM returns by asset class are calculated by simply aggregating the returns within the same asset
class for each portfolio. Bold numbers in Panel B indicate correlations higher than 0.3. Sample period: 1970–2019.

12
Panel B reports monthly correlations among Note that ACMOM and CSMOM are moderately
momentum returns across and within different correlated within commodities (0.33) and cur-
asset classes. For the same momentum style rencies (0.41), even though by intuition the two
(TSMOM, ACMOM, or CSMOM), the correlations momentum portfolios should not be correlated
are generally low across asset classes, suggest- within the same asset class (absolute momen-
ing diversification benefits from multi-asset tum vs. relative momentum). This is due to pe-
momentum investing. Specifically, the maxi- riodical directional exposure in the CSMOM of
mum correlations across asset classes are 0.25, commodities (for example, energy vs. precious
0.11, and 0.09 for TSMOM, ACMOM, and CSMOM, metal) and currencies (e.g., EMFX vs. DMFX). For
respectively. Within the same asset class, example, during a bullish commodity cycle,
TSMOM is correlated with both ACMOM and high-beta energy sector generally outperform
CSMOM, but more so with ACMOM. For exam- low-beta precious metal sector. As a result, the
ple, for commodities, TSMOM has a correlation CSMOM portfolio likely has net long positions in
of 0.72 with ACMOM, and a correlation of 0.62 the energy sector and net short positions in the
with CSMOM. Similar patterns are observed for precious metal sector, creating a directional
other asset classes. This observation is con- long exposure in the commodity beta which co-
sistent with Panel B of Table 4, suggesting that incides with positive commodity ACMOM expo-
TSMOM is mainly driven by ACMOM. sure.

Regression Analysis

We now turn to the regression approach to Specifically, we estimate the following multi-
quantify the contribution of momentum com- variate regression model:
ponents to the TSMOM performance.

𝑇𝑆𝑀𝑂𝑀𝑡 = 𝛼 + ∑ 𝛽𝑖 𝐶𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡𝑖,𝑡 + 𝜀𝑡 (6)

𝑖

where 𝑇𝑆𝑀𝑂𝑀𝑡 and 𝐶𝑜𝑚𝑝𝑜𝑛𝑒𝑛𝑡𝑖,𝑡 are the Panel A of Table 6 reports the results from the
monthly returns of TSMOM and its components first-level regression. Both ACMOM and CSMOM
at time t. We perform two levels of regressions. are significant drivers of TSMOM, with ACMOM
The first level is to regress TSMOM on ACMOM exhibiting much greater influence. Specifically,
and CSMOM, and the second level expands the the coefficients (t-statistics) of ACMOM and
independent variables to the ACMOM and CSMOM are 0.68 (15.93) and 0.32 (12.38), respec-
CSMOM returns by asset class. tively. Combined, the two momentum compo-
nents explain around 76% of the variance of
TSMOM returns (56% and 20% from ACMOM
and CSMOM, respectively). The results also in-

13
dicate that TSMOM is not fully captured by AC- above 12. The coefficient of commodity ACMOM
MOM and CSMOM, showing a positive alpha of is lower at 0.37, but still quite significant with a
0.3% per month with a t-statistic of 4.66. t-statistic of 6.29. The CSMOM components, on
the other hand, generally have significant but
Panel B of Table 6 reports the results from the weaker explanatory power (except the com-
more granular level regression. The adjusted R- modity CSMOM), as shown by smaller coeffi-
square slightly increases to 82% (59% and 23% cients and t-statistics. The bond CSMOM com-
from ACMOM and CSMOM, respectively). Within ponent, in particular, has the lowest coefficient
the ACMOM components, equities, currencies, of 0.13 and t-statistic of 2.35. The alpha compo-
and bonds contributed comparably with their nent remains significant at 0.22% per month
coefficients around 0.7-0.9 and t-statistics with a t-statistic of 3.99.

Table 6. Regression Analysis

Panel A. Decomposing TSMOM into ACMOM and CSMOM

Dependent Variable Independent Variables

Adjusted
R-square
TSMOM Intercept ACMOM CSMOM

Coefficient 0.30%** 0.68** 0.32** 76%

T-statistic 4.66 15.93 12.38

Risk Contribution 56% 20%

Panel B. Decomposing TSMOM into ACMOM and CSMOM by Asset Class

Dependent Variable Independent Variables

ACMOM CSMOM
Adjusted
TSMOM Intercept
R-square
CM EQ FX BD CM EQ FX BD

Coefficient 0.22** 0.37** 0.82** 0.73** 0.83** 0.41** 0.43** 0.31** 0.13* 82%

T-statistic 3.99 6.29 16.25 12.14 12.22 11.62 5.97 6.41 2.35

Risk Contribution 7% 17% 22% 13% 11% 4% 7% 1%

Note: * and ** indicate statistical significance at the 5% and 1% level, respectively. Risk contribution is defined as the contribution of
each component to the variance of TSMOM returns. Sample period: 1970–2019.

14
Combining the results from the correlation and 50% of the variance of TSMOM returns. There is
regression analysis, we can conclude that while also a significant alpha in TSMOM in excess of
both ACMOM and CSMOM are important drivers ACMOM and CSMOM, which presumably results
of TSMOM, the former has been the most dom- from the low correlations among the ACMOM
inant factor behind TSMOM, explaining over and CSMOM components (the diversification
benefits).

Practical Applications
Crisis Alpha and Risk Premia

One desired property of TSMOM is that it not threshold from peak to trough. For illustration,
only yields positive returns over normal periods we use a threshold of -10% which results in 12
(risk premia), but also provides downside pro- crisis periods over our sample period (1970–
tection during crisis periods (crisis alpha) (Grey- 2019). The results are qualitatively similar to
serman and Kaminski 2014). In this section, we other threshold parameters. Figure 2 plots the
investigate whether the characteristics of crisis (log) cumulative returns of MSCI as well as the
alpha and risk premia are universally shared identified crisis periods (highlighted in grey).
across its ACMOM and CSMOM components. Among the most severe ones are the 2007-
2008 Global Financial Crisis (GFC), the 2000-
Following Greyserman and Kaminski (2014), we 2002 Dot-com Bubble Burst, and the 1973-1974
define crisis periods as when the MSCI World Stagflation.
Total Return Index (MSCI) was down below a

Figure 2. (Log) Cumulative Returns of MSCI

4
3.5
3
2.5
2
1.5
1
0.5
0
-0.5

crisis MSCI

Note: Crisis periods are defined as when the MSCI World Total Return Index was down more than 10% from peak to trough. Sample
period: 1970–2019.

15
Table 7 reports performance statistics of MSCI, 480-month normal periods, TSMOM yielded an
TSMOM, and ACMOM/CSMOM by asset class average of 0.53% per month with a t-statistic of
during crisis and normal periods. Consistent 4.33, compared to 1.60% and 10.14 of MSCI. The
with the literature, we find TSMOM provided sig- hit-ratios (frequency of positive-return months)
nificant returns during both crisis and normal of TSMOM during crisis/normal periods were
periods. Specifically, over the 120-month crisis 63% and 60%, respectively, well above the 50%
periods, TSMOM yielded an average return of threshold.
1.32% per month with a t-statistic of 4.03, com-
pared to -2.83% and -6.23 of MSCI. Over the

Table 7. Crisis Alpha vs. Risk Premia

ACMOM CSMOM

MSCI TSMOM CM EQ FX BD CM EQ FX BD

Crisis Periods: 120 months

Monthly Return -2.83%** 1.32%** 0.50%* -0.09% 0.35%* 0.01% 0.12% -0.08% 0.17% 0.15%

T-Statistic -6.23 4.03 2.28 -0.61 2.27 0.10 0.55 -0.81 1.34 1.44

Hit-Ratio 63% 61% 48% 61% 53% 58% 38% 53% 58%

Normal Periods: 480 months

Monthly Return 1.60%** 0.53%** 0.08% 0.13%** 0.04% 0.14%** 0.03% 0.08% 0.06% 0.04%

T-Statistic 10.14 4.33 1.09 2.85 0.64 2.96 0.40 1.75 0.99 0.91

Hit-Ratio 60% 52% 59% 54% 57% 50% 51% 53% 52%

Note: * and ** indicate statistical significance at the 5% and 1% level, respectively. Hit-ratio is the frequency of positive-return months
out of total months. Sample period: 1970–2019.

The TSMOM components, however, exhibited crisis alpha of momentum/trend-following

heterogeneities in terms of crisis alpha and risk strategies mainly came from riding on the neg-
premia properties. During crisis periods, only ative equity momentum. During normal periods,
two components (the ACMOM of commodities while all ACMOM and CSMOM components
and currencies) provided significant crisis alpha yielded positive returns, the ACMOM of equities
(average monthly returns of 0.50% and 0.35% and bonds appeared to provide the most signif-
with t-statistics of 2.28 and 2.27, respectively). icant risk premia as measured by t-statistics
The crisis alpha of other components were ei- (average monthly returns of 0.13% and 0.14%
ther insignificant or negative. In particular, the with t-statistics of 2.85 and 2.96, respectively).
ACMOM and CSMOM of equities on average de-
livered negative returns during crisis periods To further understand the drivers of crisis alpha,
(average monthly returns of -0.09% and -0.08%, we report in Table 8 the cumulative returns of
respectively). These observations are in con- TSMOM as well as its components during crisis
trast with the beliefs of some investors that the periods. The duration of crisis periods ranged

16
from two months (February 1980–March 1980; com Bubble Burst: April 2000September 2002;
July 1998–August 1998) to 30 months (April and 3. Stagflation: March 1973September 1974),
2000–September 2002), with an average dura- MSCI detracted over 40% in each period. On the
tion of about 10 months. During these crisis pe- other end, TSMOM recorded its best perfor-
riods, MSCI detracted an average of -24.3%, mance during these three periods, up more
while TSMOM yielded an average of 14.7% with than 35% each. During Crisis 1 and 2, a majority
the ACMOM of commodities and currencies of the momentum components contributed
contributed the most (5.4% and 3.1%, respec- positive returns. During Crisis 3, however, the
tively). ACMOM of commodities contributed the major-
ity of returns, thanks to its net long commodity
During the three worst drawdown periods (1. exposure amid the Stagflation environment.
GFC: November 2007February 2009; 2. Dot-

Table 8. Cumulative Performance during Crisis Periods

Crises Periods ACMOM CSMOM

Start Month- MSCI TSMOM

Duration
CM EQ FX BD CM EQ FX BD
End Month (Months)

11/2007-02/2009 16 -53.6% 35.4% -3.4% 8.9% 7.6% 7.3% 8.2% -1.3% 1.8% -0.6%

04/2000-9/2002 30 -46.2% 46.2% -2.5% 11.2% 9.4% 9.2% 4.6% -0.3% 13.3% 6.3%

03/1973-09/1974 19 -43.4% 54.1% 39.9% -0.7% -3.6% -12.0% 3.3% -6.7% -6.0% 6.4%

01/1990-09/1990 9 -25.4% 11.0% 4.3% 0.5% 2.2% -0.2% 10.0% 2.0% 3.9% 3.0%

12/1980-07/1982 20 -23.8% 15.4% 14.0% -0.4% 8.5% -5.5% -3.7% -2.1% 0.0% 2.6%

09/1987/-11/1987 3 -20.8% -1.1% 3.8% -8.0% 2.6% -0.8% -0.9% 0.7% 0.4% -0.1%

04/1970-06/1970 3 -17.8% 7.1% 0.9% -4.9% 0.0% -2.8% -5.5% 3.6% 0.0% 1.5%

07/1998-08/1998 2 -13.5% 3.9% 4.8% -4.3% 2.5% 2.3% 0.0% -0.5% 2.6% 1.3%

10/2018-12/2018 3 -13.3% -4.3% 1.6% -1.0% 0.7% -2.4% -6.6% -0.7% -3.7% -2.1%

06/2015-02/2016 9 -11.6% 10.0% 7.1% -9.1% 5.5% 3.2% 3.9% -1.4% 1.2% -1.7%

02/1980-03/1980 2 -11.3% -3.1% -6.8% -1.9% -3.9% 4.0% -3.8% -1.7% 1.0% 2.2%

04/1984-07/1984 4 -11.3% 2.2% 0.6% -1.1% 6.1% -0.2% 3.1% -0.8% 4.2% -1.2%

Mean 10 -24.3% 14.7% 5.4% -0.9% 3.1% 0.2% -1.5% -0.3% 1.1% 0.5%

17
TSMOM experienced (moderate) losses during feature of the three periods is their short length
three out of the 12 crisis periods: September (typically twothree months), which made it
1987November 1987 (TSMOM -1.1% vs. MSCI - challenging to capitalize on the 12-month mo-
20.8%), October 2018December 2018 (TSMOM mentum.
-4.3% vs. MSCI-13.3%) and February 1980March
1980 (TSMOM -3.1% vs. MSCI -11.3%). A common

Explaining MF/CTA Performance

As shown in Hurst, Ooi, and Pedersen (2013), Table 9 reports summary statistics of indices’
TSMOM accounts for a significant portion of the performance, as well as correlations among the
MF/CTA industry performance as shown by sig- indices and our momentum strategies. All indi-
nificant regression coefficients and high R- ces have yielded positive returns over their live
squares. We investigate in this section how each periods with Sharpe ratios from 0.24 to 0.41.
momentum component has contributed to the SGTR exhibited the highest annualized volatility
industry performance, as well its evolution over of around 14%, while the annualized volatilities
time. of the other two were around 910%. These in-
dices are highly correlated among themselves
We use three commonly referenced MF/CTA in- (correlations from 0.85 to 0.97), likely due to
dices to measure the industry performance: two reasons. First, there is an overlap of man-
BTOP50 Index from BarclayHedge (BTOP), SG agers selected into these indices (typically the
Trend Index from Societe Generale (SGTR), and largest MF/CTA managers). Second, managers
HFR Systematic Macro/CTA Index from Hedge may follow similar momentum strategies, as
Fund Research (HFRM). All three indices are shown by the high correlations between the in-
manager-based return indices that seek to rep- dices and our momentum strategies (correla-
resent the MF/CTA industry performance. BTOP tions from 0.34 to 0.56).
has the longest history from 1987, followed by
SGTR from 2000, and HFRM from 2005.

18
Table 9. Summary Statistics of MF/CTA Indices

Panel A. Returns, Volatilities, and Sharpe Ratios

BTOP SGTR HFRM

Inception 01/1987 01/2000 01/2005

Annual Excess Return 3.90% 4.02% 2.18%

Annual Standard Deviation 9.54% 13.90% 9.20%

Sharpe Ratio 0.41 0.29 0.24

Panel B. Monthly Correlations

BTOP SGTR HFRM TSMOM ACMOM CSMOM

BTOP 1.00 0.97 0.85 0.50 0.44 0.34

SGTR 1.00 0.87 0.56 0.50 0.37

HFRM 1.00 0.54 0.46 0.41

Note: Sample periods are 1987-2019, 2000-2019, and 2005-2019 for BTOP, SGTR, and HFRM, respectively.

Similar to the previous regression analysis, we other hand, consistently appear to be important
regress monthly returns of each index on the performance drivers among the CSMOM com-
returns of ACMOM and CSMOM components. 6 ponents. This may be due to managers capital-
As shown in Table 10, the majority of the AC- izing on (directional) sector momentum (for ex-
MOM components have significant coefficients. ample, energy vs. precious metal; EMFX vs.
In particular, the ACMOM of bonds consistently DMFX) within commodities and currencies. The
appears to be the most significant driver behind CSMOM of equities and bonds have contributed
these indices’ performance with t-statistics insignificant or even negative returns to the in-
equal to 5.63, 4.56, and 4.06 for BTOP, SGTR, and dex performance. The intercept terms are insig-
HFRM, respectively. It suggests that MF/CTA nificant across indices, and the adjusted R-
managers have been riding the bullish momen- square ranges from 24% to 41%.
tum of global bonds over the last 30+ years. The
CSMOM of commodities and currencies, on the

6
Given the high correlations among indices, the differences in estimation results across indices are more likely due to differ-
ent sample periods rather than underlying data generating processes.

19
Table 10. Decomposing MF/CTA Index Performance

Independent Variables
Dependent Adjusted
ACMOM CSMOM
Variable R-square
Intercept
CM EQ FX BD CM EQ FX BD

0.10 0.40** 0.24* 0.12 0.82** 0.21** -0.05 0.47** 0.08

BTOP
(0.91) (3.17) (2.36) (1.31) (5.63) (2.68) (-0.22) (2.94) (0.79) 25%

-0.01 0.50** 0.91** 0.23 1.21** 0.52** 0.26 0.43 0.21

SGTR
(-0.03) (2.87) (5.13) (1.73) (4.56) (3.08) (0.73) (1.23) (1.12) 38%

0.08 0.27 0.52** 0.19* 0.83** 0.55** 0.25 0.46* -0.07

HFRM
(0.60) (1.48) (3.63) (2.09) (5.06) (4.09) (0.86) (2.42) (-0.52) 40%

Note: The numbers in parenthesis are t-statistics. *and ** indicate statistical significance at the 5% and 1% level, respectively. Regression
periods are 1987-2019, 2000-2019, and 2005-2019 for BTOP, SGTR, and HFRM, respectively.

One important characteristic of these indices is rolling regressions of the indices on the mo-
their time-varying constituents. Generally they mentum components, allowing regression co-
are reconstituted on a fixed (annual) frequency, efficients to change over time. Figure 3 plots the
with managers being included or removed time-varying risk contributions of each momen-
based on preset criteria (AUM, performance, tum component, as well as of the residual to
etc.). The reconstitution could potentially lead the BTOP index, based on a five-year rolling
to significant changes in the momentum strat- window. Risk contribution is defined as the con-
egies underneath these indices. To see the dy- tribution of each component to the variance of
namics of the underlying drivers, we also run the index returns. Results are similar to other
rolling window lengths, as well as for SGTR and
HFRM.

20
Figure 3. Explaining BTOP: Risk Contribution from a 5-Year Rolling Regression

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018

ACMOM_CM ACMOM_EQ ACMOM_FX ACMOM_BD CSMOM_CM

CSMOM_EQ CSMOM_FX CSMOM_BD Residual

Note: Risk contribution is defined as the contribution to the variance of the BTOP index returns.

A couple of observations can be made from Fig- Second, the ACMOM of bonds has consistently
ure 3. First, while the residual (accounting for been the dominant risk contributor among the
omitted factors such as different momentum momentum components, consistent with the
horizons, complex momentum signals, etc.) has results in Table 10. The risk contributions from
contributed the majority of the index risk, its risk other ACMOM components have increased
contribution has exhibited dwindling magnitude steadily since the mid-2000s. On the other
over time. Specifically, the risk contribution hand, the risk contributions from the CSMOM
from the residual was in the range of 60%-80% components (except commodities) have de-
in the 1990s, but has declined to about 40% by clined lately, suggesting that managers have re-
2019. Correspondingly, the sum of the risk con- duced exposure to CSMOM possibly due to its
tributions from our momentum components poor (even negative) performance of CSMOM
(equivalently, the R-square) has increased from since the late 2000s (as shown in Figure 1). In
around 20%-40% in the 1990s to 60% by 2019. particular, the CSMOM of currencies used to ex-
This indicates MF/CTA managers may have be- plain a meaningful portion of the index returns
come more exposed to common momentum (specifically, during the 1990s and 2000s), but
factors, and hence the diversification benefits has become marginal during the 2010s.
from investing in multi-managers may have di-
minished.

21
Conclusion

Time-series momentum and cross-sectional agers may have reduced. Our analytical frame-
momentum have attracted great attention from work has important implications for practition-
both academia and practitioners. While unique ers. From a risk management perspective, mo-
in their own theoretical underpinnings and mentum investors may use the decomposition
practical applications, they are closely related. framework to identify the key return/risk drivers
Generally, time-series momentum is found to in their portfolios. While a typical momentum
subsume cross-sectional momentum. In this strategy trades a large number of markets to
paper, we incorporate asset-class momentum achieve diversification benefits, our analysis
to bridge the gap between time-series momen- shows that the return/risk drivers can be largely
tum and cross-sectional momentum. Using an reduced to a handful of asset-class and cross-
analytic framework, we show that both asset- sectional momentum components (factors).
class momentum and cross-sectional momen- The reduced-form factor analysis is likely to be
tum are important components, but the former more relevant and efficient now than before as
has played a predominant role in explaining today’s markets are increasingly correlated (and
time-series momentum over the past half cen- so are the momentum effects across markets).
tury.
From a strategy design perspective, by decom-
We further apply the framework to analyzing posing time-series momentum into a reduced
the crisis alpha/risk premia characteristics of number of directly observable components, we
time-series momentum, as well as explaining can model each component separately and
the MF/CTA industry performance. We find that combine them into an optimal momentum
the crisis alpha of time-series momentum over portfolio. Intuitively, different momentum com-
the past half century most evidently came from ponents may have their own drivers and follow
the asset-class momentum of commodities distinct cycles. For example, the bullish asset-
and currencies. This is in contrast with the be- class momentum of bonds reflects the trend of
liefs of some investors that the crisis alpha of global disinflation, and may persist as long as
trend-following strategies mainly came from global inflation continues to drift lower. On the
riding on the negative equity momentum. On other hand, the cross-sectional momentum of
the other hand, the asset-class momentum of country equity indices is mainly driven by cross-
equities and bonds provided the most signifi- country divergence of economic fundamentals.
cant returns (risk premia) during normal peri- Its effect may be weakened in global integration
ods. Our analysis of commonly referenced and strengthened in global disintegration. By
MF/CTA indices shows that the industry perfor- identifying the corresponding drivers, we may
mance has been increasingly exposed to ge- be able to capture (and time) individual mo-
neric time-series momentum (in particular, as- mentum components better and form a more
set-class momentum), implying that the diver- optimal momentum portfolio than a simple
sification benefits from investing in multi-man- time-series momentum portfolio.

22
References

Asness, C.S., T.J. Moskowitz, L.H. Pedersen. 2013. “Value and Momentum Everywhere.” The Journal of
Finance 68(3): 929-985.

Baltas, N. 2015. “Trend-Following, Risk-Parity and the Influence of Correlations.” In Risk-Based and Fac-
tor Investing. Elsevier & ISTE Press.

Baltas, N., and R. Kosowski. 2015. “Demystifying Time-Series Momentum Strategies: Volatility Estima-
tors, Trading Rules, and Pairwise Correlations.” Working paper, Imperial College Business School.

Duke, J., D. Harding, and K. Land. 2013. “Historical Performance of Trend Following.” Working paper,
Winton Capital Management.

Faber, M. 2007. “A Quantitative Approach to Tactical Asset Allocation.” The Journal of Wealth Manage-
ment (Spring ed.).

Goyal, A., and N. Jegadeesh. 2018. “Cross-Sectional and Time-Series Tests of Return Predictability: What
is the Difference?” The Review of Financial Studies 31(5): 1784-1824.

Greyserman, A., and K. Kaminski. 2014. “Trend Following with Managed Futures.” Wiley, Hoboken, NJ.

Hurst, B., Y.H. Ooi, and L.H. Pedersen. 2013. “Demystifying Managed Futures.” The Journal of Investment
Management 11(3): 42–58.

Hurst, B., Y.H. Ooi, and L.H. Pedersen. 2017. “A Century of Evidence on Trend-Following Investing.” The
Journal of Portfolio Management 44(1): 15–29.
Jegadeesh, N., and S. Titman. 1993. “Returns to buying winners and selling losers: Implications for stock
market efficiency.” The Journal of Finance 48: 65–91.

Kolanovic, M., and Z. Wei. 2015. “Momentum Strategies across Asset Classes.” J.P.Morgan.

Moskowitz, T., Y.H. Ooi, and L.H. Pedersen. 2012. “Time Series Momentum.” The Journal of Financial
Economics 104(2): 228–250.
Rouwenhorst, K. 1998. “International Momentum Strategies.” The Journal of Finance 53: 267– 284.

Yang, K., E. Qian, and B. Belton. 2019. “Protecting the Downside of Trend When It’s Not Your Friend.” The
Journal of Portfolio Management 45(5): 99-111.

23
Appendix A. Summary Statistics of Monthly Returns

Futures/Forwards Data Start Ann. Mean Ann. St. Dev. Sharpe Ratio
Agriculture Commodities
Soybean Oil 08/1959 6.29% 29.61% 0.21
Corn 08/1959 -2.64% 24.02% -0.11
Soybean Meal 08/1959 10.51% 31.35% 0.34
Soybean 08/1959 4.98% 25.87% 0.19
Wheat 08/1959 -2.66% 25.53% -0.10
Cocoa 08/1959 4.12% 30.79% 0.13
Cotton 08/1959 2.89% 24.24% 0.12
Coffee 06/1973 4.57% 36.71% 0.12
Sugar 08/1959 -2.74% 44.32% -0.06
Lean Hogs 04/1966 4.84% 27.03% 0.18
Live Cattle 01/1965 4.78% 17.34% 0.28
Energy Commodities
Brent Crude 08/1989 13.30% 33.39% 0.40
WTI Crude 05/1983 9.47% 35.24% 0.27
Heating Oil 12/1978 13.62% 34.88% 0.39
Gas Oil 07/1986 11.91% 32.23% 0.37
Natural Gas 05/1990 -2.03% 55.76% -0.04
Gasoline 03/1976 8.53% 39.32% 0.22
Metal Commodities
Copper 08/1959 9.79% 27.42% 0.36
Aluminum 12/1988 -4.63% 20.14% -0.23
Nickel 02/1989 -0.20% 33.55% -0.01
Zinc 02/1992 -2.33% 24.36% -0.10
Gold 09/1971 4.54% 20.07% 0.23
Silver 07/1967 3.15% 33.06% 0.10
Equity Indices
Australian SPI 200 06/1992 4.16% 12.85% 0.32
Canadian TSX 60 02/1982 3.71% 14.71% 0.25
German DAX 11/1959 3.86% 19.01% 0.20
Spain IBEX 35 02/1987 4.87% 21.50% 0.23
France CAC 40 08/1987 5.70% 19.44% 0.29
U.K. FTSE 100 02/1984 4.53% 15.11% 0.30
Hong Kong Hang Seng 09/1964 11.98% 30.82% 0.39
Italy MIB 02/1998 2.58% 21.40% 0.12
Japan TOPIX 08/1959 4.17% 17.67% 0.24
Netherlands Amsterdam IDX 02/1983 8.12% 19.37% 0.42
Sweden OMX 30 01/1987 9.11% 21.65% 0.42
Singapore Index 02/1988 6.09% 22.45% 0.27

24
U.S. S&P 500 08/1959 5.99% 14.59% 0.41
Sovereign Bonds
U.S. 2-year 07/1961 2.66% 2.81% 0.95
U.S. 5-year 07/1961 3.27% 5.00% 0.65
U.S. 10-year 07/1961 3.47% 6.96% 0.50
U.S. 30-year 09/1971 4.04% 10.48% 0.39
German 2-year 04/1997 0.81% 1.19% 0.68
German 5-year 11/1991 2.78% 3.03% 0.92
German 10-year 12/1990 4.29% 5.07% 0.84
German 30-year 11/2005 7.25% 12.06% 0.60
Australian 10-year 10/1987 3.49% 7.90% 0.44
Canadian 10-year 10/1989 3.58% 5.87% 0.61
U.K. 10-year 12/1982 3.37% 7.33% 0.46

DM Currencies
AUD 02/1984 -0.08% 11.53% -0.01
CAD 02/1973 -0.33% 6.73% -0.05
CHF 02/1973 3.52% 11.91% 0.30
EUR 02/1975 0.12% 10.29% 0.01
GBP 02/1973 -0.76% 9.91% -0.08
JPY 02/1973 2.79% 11.19% 0.25
NOK 02/1973 -0.08% 10.52% -0.01
NZD 02/1986 1.37% 11.71% 0.12
SEK 01/1993 -0.43% 10.93% -0.04
EM Currencies
BRL 02/1999 -1.74% 17.09% -0.10
CLP 02/1988 -3.06% 9.93% -0.31
CZK 07/1993 1.66% 11.51% 0.14
HUF 04/1998 -0.63% 13.10% -0.05
KRW 02/1998 1.96% 11.92% 0.16
MXN 02/1996 -3.41% 10.20% -0.33
PLN 07/1993 -2.12% 12.37% -0.17
RUB 02/1999 -3.92% 12.63% -0.31
TRY 02/2002 -7.18% 15.22% -0.47
ZAR 02/1980 -5.94% 15.07% -0.39

Data Source: Commodity Research Bureau (CRB), Bloomberg, and DataStream.

25
Appendix B. Transaction Costs Assumptions

Trading futures/forwards generally involves two In each type of cost, there are both explicit and
types of transaction costs: rollover costs and implicit components. Explicit costs include
rebalancing costs. Rollover costs refer to the commissions, exchange fees, clearing fees, etc.
costs associated with rolling over an open fu- These are generally small for trading futures /
ture/forward position from the front month forwards. Implicit costs include bid-ask
contract that is expiring to a further-out con- spreads, execution delay, market impact, etc.
tract (typically the next nearest contract). Other These costs tend to vary across markets and
things equal, a portfolio with higher leverage time periods, and can be potentially significant
(more open positions) incurs higher rollover for high-leverage and high turnover portfolios. A
costs. Rebalance costs refer to costs associated full-scope discussion on transaction costs is
with changes in portfolio positions (due to not the focus of this paper. We hereafter follow
changes of momentum signals, volatility and the assumptions made in Baltas and Kosowski
correlation estimates, etc.). A portfolio with (2015) and Hurst, Ooi and Pedersen (2017). Spe-
faster-moving momentum signals or volatil- cifically, we calculate the rollover and the re-
ity/correlation estimates leads to higher re- balancing costs as follows:
balance costs.
𝑁

Rollover costs𝑡 = ∑ |𝑤𝑖,𝑡 | ∙ 𝜃𝑖 (B1)

𝑖=1

Rebalancing costs𝑡 = ∑ |𝑤𝑖,𝑡 − 𝑤𝑖,𝑡−1 | ∙ 𝛾𝑖 (B2)

𝑖=1

where 𝑤𝑖,𝑡 is asset i’s weight in the momentum

portfolio at t, 𝜃𝑖 is the rollover costs for asset i,
𝛾𝑖 is the rebalancing cost for asset i. The exhibit
below shows the rollover and rebalancing cost
assumptions by asset class and time period. In
general, commodity futures have the highest
trading costs, followed by emerging market cur-
rencies and global equity index futures. Sover-
eign bond futures and developed markets cur-
rencies have the lowest trading costs.

26
 Average Rollover Costs (Basis Average 1-Way Rebalancing Costs
Asset Class Time Period
Points Per Annum) (Basis Points Per Transaction)

prior 1993 180 60

Commodities 1993-2002 60 20

post 2003 30 10

prior 1993 72 36

Equity Indices 1993-2002 24 12

post 2003 12 6

prior 1993 12 6

Sovereign Bonds 1993-2002 4 2

post 2003 2 1

prior 1993 36 18

DM Currencies 1993-2002 12 6

post 2003 6 3

prior 1993 90 30

EM Currencies 1993-2002 30 10

post 2003 15 5

27
Appendix C. Robustness to Momentum Horizons

In this Appendix, we check the robustness of returns, the one-month and three-month hori-
our results to different momentum horizons. zons yielded lower Sharper ratios compared to
Specifically, in addition to the 12-month look- the 12-month horizon (0.37 and 0.61 compared
back window, we also use one-month and to 0.82). The declining efficacy of momentum
three-month look-back windows as the mo- signals with shorter look-back window length is
mentum formation horizons. Methodologically, in line with the findings in the literature (Hurst,
we simply replace the 12-month return signals Ooi, and Pedersen 2013; Duke, Harding, and
in Equation (1), (3) and (4) with the one-month Land 2013). Similar to the findings in the 12-
and three-month return signals, while keeping month momentum strategy, the ACMOM com-
other aspects (one-month holding period, vola- ponents in general are more profitable than
tility estimates, etc.) the same. The one-month their CSMOM counterparts. For example, for the
and three-month windows have been used in three-month momentum strategy, the Sharpe
Hurst, Ooi, and Pedersen (2013) to capture ratios of ACMOM range from 0.11 (commodities)
short- and medium-term momentum effects. to 0.28 (bonds), while those of CSMOM ranging
from -0.18 (equities) to 0.08 (bonds). Note that
Appendix C1 reports the returns, volatilities and there appear to be one-month cross-sectional
Sharpe ratios of one-month and three-month reversal effects across asset classes (particular
momentum strategies. In terms of risk-adjusted for equity indices), similar to the short-term re-
versal effects in stocks.

TABLE C1. Returns, Volatilities and Sharpe Ratios

Panel A. 1-Month Momentum

ACMOM CSMOM

TSMOM CM EQ FX BD CM EQ FX BD

Annual Ex. Ret. 3.64% 0.13% 0.59% -0.20% 1.34% -0.86% -1.18% -0.03% -0.64%

Annual Std. Dev. 10% 5.76% 3.69% 4.55% 3.55% 6.87% 3.11% 4.18% 3.52%

Sharpe Ratio 0.37 0.02 0.16 -0.04 0.38 -0.13 -0.38 -0.01 -0.18

Panel B. 3-Month Momentum

ACMOM CSMOM

TSMOM CM EQ FX BD CM EQ FX BD

Annual Ex. Ret. 6.10% 0.60% 0.90% 1.06% 0.99% 0.23% -0.56% 0.30% 0.29%

Annual Std. Dev. 10% 5.69% 3.69% 4.54% 3.58% 6.98% 3.08% 4.32% 3.70%

Sharpe Ratio 0.61 0.11 0.24 0.23 0.28 0.03 -0.18 0.07 0.08

Note: Sample period: 1970–2019.

28
Appendix C2 reports the results from regressing of TSMOM, as seen from their higher t-statistics
monthly TSMOM returns on their components and risk contributions. For the one-month and
for the one-month and three-month momen- the three-month momentum strategies, the
tum strategies. Consistent with the 12-month ACMOM components combined explained 56%
momentum case, all ACMOM and CSMOM com- and 62% of the variance of TSMOM returns,
ponents have significant explanatory power for compared to 24% and 21% for the CSMOM com-
the TSMOM returns, with their t-statistics sig- ponents. The alpha coefficients are significant
nificant at the 1% level and the adjusted R- at the 1% level for both the one-month and
square at least 80%. Again, the ACMOM compo- three-month momentum strategies.
nents appear to be the more important drivers

Table C2. Regression Analysis: Decomposing TSMOM into ACMOM and CSMOM

Panel A. 1-Month Momentum

Dependent
Independent Variables
Variable

ACMOM CSMOM
Adjusted
TSMOM Intercept
R-square
CM EQ FX BD CM EQ FX BD

Coefficient 0.23** 0.49** 0.73** 0.73** 0.96** 0.45** 0.38** 0.45** 0.15** 80%

T-statistic 4.06 7.89 7.57 11.68 12.99 12.52 3.97 6.72 3.09

Risk Contri-
10% 13% 18% 14% 12% 2% 9% 1%
bution

Panel B. 3-Month Momentum

Dependent
Independent Variables
Variable

ACMOM CSMOM
Adjusted
TSMOM Intercept
R-square
CM EQ FX BD CM EQ FX BD

Coefficient 0.22** 0.48** 0.95** 0.65** 0.87** 0.37** 0.28** 0.44** 0.17** 83%

T-statistic 4.55 7.79 12.41 10.00 17.85 9.99 3.56 8.73 3.83

Risk Contri-
12% 21% 16% 13% 10% 2% 8% 1%
bution

Note: * and ** indicate statistical significance at the 5% and 1% level, respectively. Risk contribution is defined as the contribution of
each component to the variance of TSMOM returns. Sample period: 1970–2019.

29
Appendix C3 reports performance statistics of crisis alpha (in particular, equities). During nor-
TSMOM and its ACMOM/CSMOM components mal periods, the three-month TSMOM yielded
during crisis and normal periods. For both the significant risk premia (0.38% per month with a
one-month and three-month momentum hori- t-statistic of 3.42) of which the ACMOM of
zons, TSMOM provided significant crisis alpha bonds and equities contributed the majority.
with a t-statistic of 2.83 and 2.67, respectively. The one-month TSMOM, on the other hand, only
Again, the majority of the crisis alpha came yielded insignificant risk premia (0.14% per
from the ACMOM components (in particular, month with a t-statistic of 1.16). Among its com-
commodities and currencies). The CSMOM ponents, the ACMOM of bonds and equities
components yielded either smaller or negative contributed the most, while the other compo-
nents detracted value.

Table C3. Crisis Alpha vs. Risk Premia

Panel A. 1-Month Momentum

ACMOM CSMOM

MSCI TSMOM CM EQ FX BD CM EQ FX BD

Crisis Periods: 120 months

Monthly Return -2.83%** 0.97%** 0.25% 0.02% 0.23% 0.16% -0.07% -0.05% 0.19% -0.01%

T-Statistic -6.23 2.83 1.18 0.15 1.56 1.41 -0.36 -0.69 1.42 -0.06

Hit-Ratio 54% 48% 49% 57% 56% 42% 40% 59% 48%

Normal Periods: 480 months

Monthly Return 1.60%** 0.14% -0.05% 0.06% -0.07% 0.10%* -0.07% -0.11%** -0.05% -0.07%

T-Statistic 10.14 1.16 -0.73 1.35 -1.29 2.27 -0.82 -2.63 -0.88 -1.48

Hit-Ratio 60% 48% 56% 46% 54% 46% 41% 45% 44%

30
 Panel B. 3-Month Momentum

ACMOM CSMOM

MSCI TSMOM CM EQ FX BD CM EQ FX BD

Crisis Periods: 120 months

Monthly Return -2.83%** 1.01%** 0.33% 0.08% 0.28% -0.03% 0.22% -0.22%* 0.04% 0.02%

T-Statistic -6.23 2.67 1.58 0.58 1.91 -0.23 1.00 -2.57 0.31 0.22

Hit-Ratio 57% 60% 48% 55% 52% 48% 31% 52% 49%

Normal Periods: 480 months

Monthly Return 1.60%** 0.38%** -0.02% 0.07% 0.04% 0.11%* -0.03% 0.00% 0.02% 0.02%

T-Statistic 10.14 3.42 -0.30 1.77 0.74 2.49 -0.36 -0.05 0.37 0.51

Hit-Ratio 56% 49% 58% 51% 56% 47% 47% 51% 49%

Note: * and ** indicate statistical significance at the 5% and 1% level, respectively. Hit-ratio is the frequency of positive-return months
out of total months. Sample period: 1970–2019.

In summary, the crisis alpha and risk premia has become less exposed to the one-month
properties of the one-month and three-month short-term momentum over time. Using the
momentum strategies are broadly consistent HFRM index as an example (its regression re-
with those of the 12-month momentum. The sults will reflect the pattern over the most re-
key difference is a weaker risk premia for the cent periods given its shortest history). The ad-
short-term one-month momentum, mainly due justed R-square drops from 40% for the 12-
to some reversal effects in its components month momentum (Table 10), to 30% and 17%
(cross-sectional reversal in equities, in particu- for the three-month and one-month momen-
lar). tum, respectively. The declining industry expo-
sure to short-term momentum is likely a result
Appendix C4 reports the results from regressing from the disappointing momentum perfor-
monthly MF/CTA index returns on the one- mance at the short-term horizon.
month and three-month momentum compo-
nents. Similar to the 12-month momentum In summary, the patterns observed at the one-
case, the ACMOM of bonds was the most con- month and three-month momentum are
sistent contributor among the ACMOM compo- broadly consistent with those of the 12-month
nents to the index returns, with coefficients momentum. In addition, we find weaker effi-
mostly significant at the 1% and 5% level. Among cacy for shorter-term momentum (the one-
the CSMOM components, currencies was the month momentum, in particular), consistent
dominant driver (particularly for the three- with the evidence documented in the literature.
month momentum), followed by commodities. As a result, the MF/CTA industry has become
The evidence also suggests that the industry less exposed to short-term momentum.

31
Table C4. Decomposing MF/CTA Index Performance

Panel A. 1-Month Momentum

Independent Variables
Dependent Adjusted
ACMOM CSMOM
Variable R-square
Intercept
CM EQ FX BD CM EQ FX BD

0.36** 0.25 0.10 0.31** 0.65** 0.32** 0.23 0.68** -0.08

BTOP 26%
(2.95) (1.92) (0.83) (2.76) (3.95) (3.20) (1.10) (4.72) (-0.64)

0.34 0.50* 0.30 0.37* 0.76* 0.44** 0.14 0.65* -0.13

SGTR 20%
(1.56) (2.10) (1.22) (2.26) (2.26) (2.69) (0.25) (2.40) (-0.59)

0.21 0.37 0.21 0.16 0.36 0.51** 0.14 0.29 -0.13

HFRM 17%
(1.25) (1.85) (1.00) (1.29) (1.77) (3.65) (0.34) (1.33) (-0.82)

Panel B. 3-Month Momentum

Independent Variables
Dependent Adjusted
ACMOM CSMOM
Variable R-square
Intercept
CM EQ FX BD CM EQ FX BD

0.18 0.47** 0.19 0.16 0.71** 0.19* -0.10 0.92** -0.10

BTOP 32%
(1.57) (3.47) (1.67) (1.35) (4.63) (2.47) (-0.56) (6.70) (-0.82)

0.19 0.80** 0.42 0.13 1.05** 0.41** 0.00 1.06** -0.05

SGTR 33%
(0.95) (3.96) (1.56) (0.66) (3.39) (2.68) (0.01) (3.82) (-0.23)

0.13 0.31 0.07 0.27 0.57** 0.45** -0.09 0.72** -0.06

HFRM 30%
(0.84) (1.50) (0.39) (1.96) (3.08) (3.52) (-0.21) (3.28) (-0.39)

Note: The numbers in parenthesis are t-statistics. *and ** indicate statistical significance at the 5% and 1% level, respectively. Regression
periods are 1987-2019, 2000-2019, and 2005-2019 for BTOP, SGTR, and HFRM, respectively.

32
Disclosures
This material is solely for informational purposes and shall not constitute an offer to sell or the solicitation to buy securities.
The opinions expressed herein represent the current, good faith views of the author(s) at the time of publication and are
provided for limited purposes, are not definitive investment advice, and should not be relied on as such. The information
presented in this article has been developed internally and/or obtained from sources believed to be reliable; however, PanAgora
Asset Management, Inc. ("PanAgora") does not guarantee the accuracy, adequacy or completeness of such information. Predic-
tions, opinions, and other information contained in this article are subject to change continually and without notice of any kind
and may no longer be true after the date indicated. Any forward‐looking statements speak only as of the date they are made,
and PanAgora assumes no duty to and does not undertake to update forward‐looking statements. Forward‐looking statements
are subject to numerous assumptions, risks and uncertainties, which change over time. Actual results could differ materially
from those anticipated in forward‐looking statements. This material is directed exclusively at investment professionals. Any
investments to which this material relates are available only to or will be engaged in only with investment professionals. There
is no guarantee that any investment strategy will achieve its investment objective or avoid incurring substantial losses.

Hypothetical performance results have many inherent limitations, some of which are described below. No representation is
being made that any account will or is likely to achieve profits or losses similar to those shown. In fact, there are frequently
sharp differences between hypothetical performance results and the actual results subsequently achieved by any particular
investment program. One of the limitations of hypothetical performance results is that they are generally prepared with the
benefit of hindsight. In addition, hypothetical trading does not involve financial risk, and no hypothetical trading record can
completely account for the impact of financial risk in actual trading. For example, the ability to withstand losses or to adhere
to a particular investment program in spite of trading losses are material points which can also adversely affect actual trading
results. There are numerous other factors related to the markets in general or to the implementation of any specific investment
program which cannot be fully accounted for in the preparation of hypothetical performance results and all of which can
adversely affect actual trading results.

The information presented is based upon the hypothetical assumptions discussed in this piece. Specific assumptions: risk is
allocated across and within sectors equally. Certain assumptions have been mad e for modeling purposes and are unlikely to
be realized. No representation or warranty is made as to the reasonableness of the assumptions made or that all assumptions
used in achieving the returns have been stated or fully considered.

Index Disclosures

MSCI World Index: The MSCI World Index is a broad global equity index that represents large and mid-cap equity perfor-
mance across all 23 developed markets countries. It covers approximately 85% of the free float-adjusted market capitaliza-
tion in each country.

Certain information included herein is derived by PanAgora Asset Management, Inc in part from MSCI’s provided Index Data. 
However, MSCI has not reviewed this product or report, and does not endorse or express any opinion regarding this product
or report or any analysis or other information contained herein or the author or source of any such information or analysis. 
Neither MSCI nor any third party involved in or related to the computing or compiling of the Index Data makes any express or
implied warranties, representations or guarantees concerning the Index Data or any information or data derived therefrom,
and in no event will MSCI or any third party have any liability for any direct, indirect, special, punitive, consequential or any
other damages (including lost profits) relating to any use of this information. Any use of MSCI data requires a license from
MSCI.  None of the Index Data is intended to constitute  investment advice or a recommendation to make (or refrain from
making) any kind of investment decision and may not be relied on as such.  

33