s

THE JOURNAL OF PORTFOLIO MANAGEMENT

ard
Aw
zi/ A ng

evy
ob al
Jac nnu
oz th nci

sL
ab e 20 ou
n F h nn
tei T A
rns
Be

JPM.iprjournals.com
  QUANTITATIVE STRATEGIES: FACTOR INVESTING |

QUANTITATI VE STR ATEG IES:

2019 FAC TOR INVESTING
FOURTH EDITION

CONTRIBUTORS: DEMIR BEKTIĆ | JENNIFER BENDER | THOMAS BLACKBURN | DAVID BLITZ | JACK DAVIES | KRISTIN FERGIS
KATELYN GALLAGHER | DAVE GIBBON | TARUN GUPTA | HAN HAO | PHILIP HODGES | KED HOGAN | JOOP HUIJ | BRYAN KELLY
GEORGI KYOSEV | ASHLEY LESTER | FEIFEI LI | NONGCHAO LI | LIONEL MARTELLINI | JOHN M. MULVEY | DIRK SCHIERECK
JOSEPH (YOSEOP) SHIM | SARA SHORES | JOSEPH SIMONIAN | JOSEPHINE SMITH | TIMO SPIELMANN | XIAOLE SUN
FOURTH EDITION

EDUARD VAN GELDEREN | MILAN VIDOJEVIC | MICHAEL WEGENER | JOSEF-STEFAN WENZLER | CHENWEI WU

SPONSORED BY
| MARCH 2019
 INTRODUCTION
Quantitative Special Issue
S
ince 2016, five quantitative special
issues have been published by The
Journal of Portfolio Management focusing
on two themes: factor investing and
multiasset strategies. This issue is the fourth
special issue on factor investing. There are 11
articles in this issue.
The article by Tarun Gupta and
Bryan Kelly presents the robust momentum
behavior in 65 widely studied characteristic-
based equity factors around the world and
explains how this behavior can be exploited
by a time-series momentum trading strategy
that scales factor exposures based on recent
behavior. Their time-series momentum
strategy involves aggregating individual factor
timing strategies and dominates all individual
timing strategies. Relative to untimed fac-
tors, factor timing employing their proposed
strategy generates economically and statisti-
cally large excess performance. Their results
indicate that the momentum phenomenon is
driven in large part by persistence in common
return factors and not solely by persistence in
idiosyncratic stock performance. The Gupta–
Kelly strategy complements stock momentum
because both enter optimized multifactor
portfolios with significant positive weights.
Markowitz mean–variance portfolio
construction can be used to create efficient,
cost-effective portfolios that capture
factor exposure. Jennifer Bender, Thomas
Blackburn, and Xiaole Sun accomplish this
Quantitative Special Issue 2019
and then evaluate the merits of mean–variance
factor portfolios against alternative weighting
schemes. In the opinion of the authors, alter-
native weighting schemes have arguably weak
theoretical foundations. They then empiri-
cally show that a large part of the outper-
formance of alternative weighting schemes
is attributable to a few well-known factors.
Bender, Blackburn, and Sun argue that factor
portfolios that are sensibly constructed gen-
erate a similar or higher information ratio by
explicitly harnessing the factors in an efficient
risk- and transaction cost–aware manner.
Three possible approaches portfolio
managers can employ in response to time-
varying factor returns are to (1) ignore the
short-term variation, (2) develop a short-term
prediction model for generating excess
returns by tactically changing positions, and
(3) pursue defensive factor timing. In their
article, Kristin Fergis, Katelyn Gallagher,
Philip Hodges, and Ked Hogan describe the
third approach, defensive factor timing, as
proactively using market signals to reduce
exposures to individual factors or a portfolio
of factors when performance outcomes are
unattractive. The signals they use are mea-
sures of aggregate risk tolerance, the effec-
tiveness of diversification, and valuation
measures. In contrast to well-known market
timing applications, defensive factor timing
is done infrequently, seeking to preserve cap-
ital during bad times. The authors provide
The Journal of Portfolio Management   1
an illustration of how these signals can be applied in a
macro factor setting.
An understanding of how factors perform indepen-
dently and how they are related to one another is neces-
sary for the efficient implementation of factor investing.
Investing in stocks with high scores on one factor (i.e.,
generic single-factor portfolio) is shown by David Blitz
and Milan Vidojevic to be suboptimal owing to the
failure to recognize the possibility that these stocks may
be unattractive from the perspective of other factors.
Their model is able to attribute performance differences
between mixed-sleeve and integrated multifactor portfo-
lios to differences in their factor characteristics. Because
generic single-factor sleeves are an inefficient way to
obtain factor exposure, mixing them into a portfolio also
results in suboptimal multifactor portfolios. However,
the authors demonstrate that a mixed-sleeve portfolio of
enhanced single factors that do not invest in stocks with
implied underperformance can match the performance
of a bottom-up, integrated, multifactor portfolio with
similar factor exposures. Finally, performance differ-
ences between factor portfolios in the small-cap and the
large-cap space, as well as equal- and value-weighted
portfolios, can be explained by their model.
The article by Ashley Lester also considers the
investment choice between a set of single-factor sub-
portfolios and a single integrated portfolio. Defining
factor exposures so that they precisely relate relative
factor exposures to the number and correlation of fac-
tors, Lester predicts analytically consequent changes
in risk and return. He finds that integrated portfolios
theoretically outperform segregated portfolios to an
increasing extent when the number of factors is large
and average correlation is low. In contrast to studies that
test predictions by testing the performance on a handful
of portfolios using particular versions of factors, in this
article the author generates more than 1,000 matched
pairs of historical portfolios using a wide range of factor
definitions to test performance. He finds that an inte-
grated portfolio of four orthogonal factors generates (1)
twice the factor exposure of the corresponding set of
single-factor portfolios, (2) twice the outperformance,
and (3) a 40% higher information ratio.
Investors seeking factor exposures may have a
choice of different investment vehicles in which to imple-
ment factor strategies. The decision as to which invest-
ment vehicle and the resulting constraints may lead to
meaningful differences in the returns of factor strategies.
2   Introduction: Quantitative Special Issue Quantitative Special Issue 2019
Constraints include leverage, trading frequency, and risk
levels. The article by Jack Davies, Dave Gibbon, Sara
Shores, and Josephine Smith investigates how imple-
mentation matters for two hypothetical equity factor
strategies: momentum and value. They find that as con-
straints on the factor strategies are relaxed, risk-adjusted
potential returns may improve. Conversely, they find
that the Sharpe ratios for the two factor strategies when
moving from an unconstrained implementation to a
low-turnover, long-only implementation may decrease
by as much as 60%.
In their article, Feifei Li and Joseph (Yoseop)
Shim focus on the implementation issues of multifactor
investing. These issues, which are all crucial elements in
real-world product design, include portfolio concentra-
tion, turnover, trading cost, and capacity. The authors
address two questions in the presence of implementation
costs: (1) which factors should be included in a multi-
factor portfolio, and (2) what is the best way to construct
a single-factor portfolio? They treat portfolio design as a
choice based on the trade-off between the most effective
procedure for capturing factor premium and low-cost
implementation.
Compared to traditional actively managed mutual
funds, Eduard van Gelderen, Joop Huij, and Georgi
Kyosev find that significantly higher alphas can be
earned by mutual funds that pursue factor investing
strategies based on equity asset pricing anomalies such
as the small-cap, value, momentum, profitability, and
investments effects. However, the authors show that the
actual returns that investors earn by investing in factor
mutual funds appear to be significantly lower because
investors dynamically reallocate their funds across both
factors and factor managers. They argue that rather than
timing factors and factor managers, investors would be
better off by using a buy-and-hold strategy and selecting
a multifactor manager.
Although there have been numerous studies on
the explanatory power of size, value, profitability, and
investment for equity markets, the relevance of these fac-
tors for global credit markets has been far less explored.
The expectation is that equities and bonds should be
related according to structural credit risk models. In
this article Demir Bektić, Josef-Stefan Wenzler, Michael
Wegener, Dirk Schiereck, and Timo Spielmann inves-
tigate the impact of the four Fama–French factors in
the US and European credit markets. Although they
find that all factors exhibit economically and statistically
significant excess returns in the US high-yield market,
they report mixed evidence for US and European invest-
ment-grade markets. Nevertheless, the authors find that
on a risk-adjusted basis, investable multifactor portfolios
outperform the corresponding corporate bond bench-
marks. Their results also suggest that corporate bond
returns cannot be fully subsumed by traditional equity
risk factors because corporate bond factor risk premiums
are not the same across the two markets.
Hedge fund replication is one application of factor-
tracking portfolio construction. Because of their sim-
plicity and familiarity, factor models based on standard
multiple regression analysis are among the commonly
used hedge fund replication frameworks. However,
despite its popularity, a portfolio manager is exposed
to several risks when using the standard regression-
based framework to construct factor-tracking portfo-
lios. These risks include overfitting to a single sample
(severely undercutting predictive effectiveness); multi-
collinearity, wherein predictors are linearly correlated
to a degree (rendering the model problematic); and use
of a single sample that may fail to capture the evolu-
tion of hedge fund exposures over time (owing to the
dynamic nature of hedge funds). Dealing with these risks
calls for a better statistical model than standard multiple
regression analysis. Joseph Simonian and Chenwei Wu
Quantitative Special Issue 2019
propose the application of a regularization method
known as “ridge regression” in pursuing a hedge fund
replication strategy to deal with these risks. The authors
show how ridge regression can help generate generaliz-
able models that are useful in both the ex post risk analysis
and ex ante replication of hedge fund strategies.
Many defined pension plans are seriously under-
funded and unlikely to support their future long-term
liabilities without massive contributions and possible
bailouts. To address the mismatch of contributions and
liabilities, John Mulvey, Lionel Martellini, Han Hao,
and Nongchao Li develop a factor and goal-driven
framework for assessing asset allocation and contribu-
tion decisions within defined benefit pension plans.
Their framework is more realistic than traditional
approaches for setting pension benefits and combines
a micro-study of a representative cohort of individuals
with an aggregation across a target population to esti-
mate consistency between the micro and macro envi-
ronments. The proposed approach employs simulation
to assess whether the factor/asset risks of the planned
contributions and investment returns achieve the target
promised benefits.
Frank J. Fabozzi
Editor
The Journal of Portfolio Management   3
 Frank J. Fabozzi
Editor
Peter Bernstein Barbara S. Bernstein
Founding Editor Founding Assistant Editor

Rosie Instance Marketing Manager David Rowe Reprints Manager

Mitchell Gang Production Editor
Deborah Brouwer Production & Design Manager
Ryan C. Meyers Account Manager Mark Lee Advertising Director
Arielle Whitney Audience Development Manager
Mark Adelson Content Director Albina Brady Agent Sales Manager
Dave Blide Publisher

AMBASSADOR BOARD
Clifford S. Asness Mohamed A. El-Erian William N. Goetzmann Bruce I. Jacobs
AQR Capital Management Allianz and Chair of President’s Yale University Jacobs Levy Equity Management, Inc.
Global Development Council
Turan Bali Campbell R. Harvey Daniel Kahneman
McDonough School of Business Robert Engle Duke University Princeton University
Georgetown University New York University, and Man Group plc
Stern School of Business Andrew W. Lo
MIT

ADVISORY BOARD
Michele Aghassi Lev Dynkin Petter N. Kolm Karthik Ramanathan
AQR Capital Management Barclays Courant Institute Consultant
Sergio M. Focardi of Mathematical Sciences, Marc R. Reinganum
Carol Alexander New York University
University of Sussex Pôle Universitaire Léonard de Vinci Senior Investment
and The Intertek Group Mark Kritzman Management Executive
Noël Amenc
EDHEC-Risk Institute Russell J. Fuller Windham Capital Management Vincenzo Russo
Fuller & Thaler Asset Management Assicurazioni Generali S.p.A.
Mark J.P. Anson Wai Lee
Commonfund Gary L. Gastineau Wells Fargo Asset Management Laurence B. Siegel
ETF Consultants.com, Inc. Research Foundation
Stan Beckers Jim Kyung-Soo Liew of CFA Institute
London Business School Chris Gowlland Johns Hopkins
Delaware Investments Carey Business School António Baldaque da Silva
Jennifer Bender BlackRock
Steven P. Greiner Marcos López de Prado
State Street Global Advisors Charles Schwab & Co., Inc. Eric Sorensen
AQR Capital Management PanAgora Asset Management
Richard Bernstein Ronald Hua and Cornell University
Richard Bernstein Advisors LLC Goldman Sachs Asset Management Meir Statman
Lionel Martellini Santa Clara University
Michele Leonardo Bianchi Brian J. Jacobsen EDHEC-Risk Institute
Bank of Italy Wells Fargo Funds Management Stoyan V. Stoyanov
Dimitris Melas Stony Brook University
David C. Blitz Robert Jones
Robeco Asset Management MSCI Radu S. Tunaru
System Two Advisors,
Arwen Advisors John Mulvey University of Kent
Gerald W. Buetow, Jr.
BFRC Services, LLC Ronald N. Kahn Princeton University M. Barton Waring
BlackRock Anna Obizhaeva Barclays Global Investors (retired)
Peter Carr
New York University Jang Ho Kim New Economic School Andrew B. Weisman
Kyung Hee University Windham Capital Management
Joseph Cerniglia Wesley Phoa
University of Pennsylvania The Capital Group Companies Robert E. Whaley
Woo Chang Kim Vanderbilt University
Denis Chaves KAIST Edward E. Qian
The Vanguard Group and Princeton University PanAgora Asset Management Jarrod Wilcox
Wilcox Investment
Jim Clayton Will Kinlaw Xiao Qiao
Cornerstone Real Estate Advisers LLC State Street Global Exchange Paraconic Technologies US Inc. Frank Zhang
and University of Connecticut Yale School of Management
C.G. (Kees) Koedijk Svetlozar T. Rachev
Kevin Dowd Tilburg School of Economics Texas Tech University Guofu Zhou
Durham University Business School and Management and FinAnalytica Washington University in St. Louis

The Journal of Portfolio Management ® (ISSN# 0095-4918) (USPS# 060-830) is published six
times a year (quarterly, plus an extra issue in the Spring and Fall) by Pageant Media, Ltd.,
41 Madison Avenue, 20th Floor, New York, NY 10010. Phone (646) 891-2110. Copyright
© 2019 Pageant Media, Ltd. POSTMASTER: Send address changes for The Journal of Port-
folio Management to Pageant Media, Ltd., 41 Madison Avenue, 20th Floor, New York, NY
10010 USA, Att: Journals Department. Reproduction in whole or in part without written
permission is prohibited.
Manuscripts: The Journal of Portfolio Management considers but assumes no responsibility for
unsolicited manuscripts. Unsolicited manuscripts will not be returned. Receipt of all manuscripts
will be acknowledged within two weeks. Instructions for submission of manuscripts are printed
at the end of the journal.
Subscriptions: For subscription enquiries please contact a member of the team at
+1 646 931 9045/+44 20 7139 1600 or at iprjournal@pageantmedia.com. Printed by The
Manson Group, St Albans, UK.
The opinions expressed are those of the individual authors. IPR Journals is not responsible for
the accuracy, completeness, or timeliness of the information contained in the articles herein. If you
require expert advice, you should seek the services of a competent professional. No statement in
this journal is to be construed as a recommendation to buy or sell securities. Product names men-
tioned in the articles in this journal may be trademarks or service marks of their respective owners.

4   The Journal of Portfolio Management Quantitative Special Issue 2019

 THE 20th ANNUAL
BERNSTEIN FABOZZI/JACOBS LEVY AWARDS
The Bernstein Fabozzi/Jacobs Levy Awards were established in 1999, on the 25th
anni­versary of The Journal of Portfolio Management, to honor Editors Peter Bernstein and
Frank Fabozzi for their extraordinary contributions and to promote research excellence
in the theory and practice of portfolio manage­ment. The annual awards, co-founded
and generously funded by Jacobs Levy Equity Management, consist of a $2,500 prize
for the Best Article and $1,000 prizes for each of three Outstanding Articles.
n n n

The Journal of Portfolio Management is pleased to announce the recipients of the

20th Annual Bernstein Fabozzi/Jacobs Levy Awards for articles appearing in four regular
issues beginning with Winter 2018 through Fall 2018, as well as the Multi-Asset
Strategies issue from December 2017, the Quantitative Strategies issue from March 2018,
and the Stephen A. Ross issue from June 2018. On the basis of voting by subscribers,*
the 20th Annual Bernstein Fabozzi/Jacobs Levy Awards are presented to:

BEST ARTICLE
Behavioral Efficient Markets
Meir Statman
Winter 2018

OUTSTANDING ARTICLES
Proverbial Baskets Are Uncorrelated Risk Factors! A Factor-Based Framework
for Measuring and Managing Diversification in Multi-Asset Investment Solutions
Lionel Martellini and Vincent Milhau
Multi-Asset Strategies Issue, December 2017
Buyback Derangement Syndrome
Clifford Asness, Todd Hazelkorn, and Scott Richardson
Spring 2018
The Impact of Volatility Targeting
Campbell R. Harvey, Edward Hoyle, Russell Korgaonkar, Sandy Rattray,
Matthew Sargaison, and Otto Van Hemert
Fall 2018

*Articles authored by Frank Fabozzi were not eligible for an award. Authors were not permitted to vote for their own articles.
The ballots were tallied by Institutional Portfolio Research Journals.
QUANTITATIVE SPECIAL ISSUE 2019
Editors Introduction:
Quantitative Special Issue 1
Frank J. Fabozzi
Factor Momentum
Everywhere 13
Tarun Gupta and Bryan Kelly
In this article, the authors document robust momentum
behavior in a large collection of 65 widely studied char-
acteristic-based equity factors around the globe. They
show that, in general, individual factors can be reliably
timed based on their own recent performance. A time-
series factor momentum portfolio that combines timing
strategies of all factors earns an annual Sharpe ratio of
0.84. Factor momentum adds significant incremental
performance to investment strategies that employ tra-
ditional momentum, industry momentum, value, and
other commonly studied factors. The results demon-
strate that the momentum phenomenon is driven in large
part by persistence in common return factors and not
solely by persistence in idiosyncratic stock performance.
Clash of the Titans: Factor
Portfolios versus Alternative
Weighting Schemes 38
Jennifer Bender, Thomas Blackburn,
and Xiaole Sun
In this article, the authors (re) introduce mean–variance
portfolio construction for factor portfolios. These
models, first popular with quants in the 1990s, are being
resurrected today in a different context for transparent
factor portfolios. The authors then evaluate the merits
of these mean–variance factor portfolios against alter-
native weighting schemes. They point out that alterna-
tive weighting schemes have arguably weak theoretical
foundations, and their supporters rationalize them with
6   The Journal of Portfolio Management Quantitative Special Issue 2019
VOLUME 45 NO. 3
a range of (very different) reasons, most of them dissat-
isfying in the view of the authors. They then show that
alternative weighting schemes derive a large part of their
outperformance from a handful of well-known factors.
The authors argue that sensibly built factor portfolios
deliver a similar or higher information ratio by explicitly
harnessing the factors and doing so in an efficient risk-
and transaction cost-aware way.
Defensive Factor Timing 50
Kristin Fergis, Katelyn Gallagher,
Philip Hodges, and Ked Hogan
In this article, the authors define defensive factor timing
as proactively using market signals—measures of aggre-
gate risk tolerance, the effectiveness of diversification,
and valuation measures—to reduce exposures to indi-
vidual factors or a portfolio of factors when outlooks are
unattractive. Unlike other market timing applications,
defensive factor timing is done infrequently with an aim
to preserve capital during bad times. The authors illustrate
how these signals can be applied in a macro factor setting.
The Characteristics
of Factor I nvesting 69
David Blitz and Milan Vidojevic
In this article, the authors dissect the realized perfor-
mance of factor-based equity portfolios using a charac-
teristics-based multifactor return model. They show that
generic single-factor portfolios, which invest in stocks
with high scores on one particular factor, are suboptimal
because they ignore the possibility that these stocks may
be unattractive from the perspective of other factors.
The authors also show that differences in performance
between (1) integrated and mixed-sleeve multifactor
portfolios, (2) small-cap and large-cap factor portfolios,
and (3) equal and value-weighted factor portfolios can
be fully attributed to the differences in their factor char-
acteristics. They conclude that efficient factor investing
A Single Solution
for All of Your Factor
Investing Needs
Power your investment process with unique insights,
test valuable strategies, and communicate portfolio
performance—all in one essential solution. Discover the
strength of S&P Global Market Intelligence with streamlined
factor-based investing tools, including portfolio analytics
and strategy development, powered by ClariFI®.

Request a demo today:

fuel@spglobal.com

Copyright © 2019 by S&P Global Market Intelligence,

a division of S&P Global Inc. All rights reserved.
QUANTITATIVE SPECIAL ISSUE 2019
requires an understanding of how factor characteristics
drive portfolio returns.
On the Theory and Practice
of Multifactor Portfolios 87
Ashley Lester
Investors interested in factor investing often seek expo-
sure to several factors, not just one or two. The decision
on how to implement multiple exposures may have a
considerable effect on performance. In this article, the
author considers the investment choice between a set of
single-factor subportfolios and a single integrated port-
folio. He defines factor exposures in a way that precisely
relates relative factor exposures to the number and cor-
relation of factors and predicts analytically consequent
changes in risk and return. Theoretically, integrated
portfolios increasingly outperform segregated portfo-
lios when the number of factors is large and average
correlation is low. He tests his predictions by generating
over 1,000 matched pairs of historical portfolios using a
wide range of factor definitions. This contrasts with the
existing literature, which typically focuses on a handful
of portfolios using particular versions of factors. The
empirical results strongly support the theoretical predic-
tions. In practical terms, an integrated portfolio of four
orthogonal factors generates twice the factor exposure of
the corresponding set of single-factor portfolios, twice
the outperformance, and a 40% higher information ratio.
Implementation M atters: Relaxing
Constraints Can Improve the Potential
Returns of Factor Strategies 101
Jack Davies, Dave Gibbon, Sara Shores,
and Josephine Smith
Eligible investors seeking factor exposures may have a
choice of different investment vehicles to implement
an investment strategy, ranging from fully transparent,
8   The Journal of Portfolio Management Quantitative Special Issue 2019
VOLUME 45 NO. 3
index-based exchange-traded funds to private funds or
undertakings for collective investment in transferable
securities. In this article, the authors assess the impact
of constraints common to these investment vehicles
through the lens of hypothetical equity momentum and
value factor strategies. As constraints—leverage, trading
frequency, and risk levels—on the factor strategies are
relaxed, risk-adjusted potential returns may improve.
Conversely, moving from an unconstrained implemen-
tation to a low-turnover, long-only implementation may
decrease the Sharpe ratios of momentum and value strat-
egies by as much as 60%.
Trade-Off in Multifactor
Smart Beta Investing: Factor
Premium and Implementation Cost 115
Feifei Li and Joseph (Yoseop) Shim
The authors study the impact of the inclusion of the
momentum and size factors, and the selectiveness in
stock screening, on the performance and implementa-
tion cost of a multifactor strategy. Whereas the cost of
implementing a stand-alone momentum strategy is quite
high because of its extraordinarily high turnover, adding
momentum to a mix of the value, low beta, profitability,
investment, and size factors helps lower tracking error
and improves the information ratio without a signifi-
cant increase in implementation cost because offsetting
trades cancel each other out. Contrary to conventional
wisdom, the inclusion of the size factor does not nega-
tively affect the multifactor strategy’s trading costs
in light of its relatively low turnover. Additionally,
including the size factor improves the performance and
coverage of the multifactor strategy, which otherwise
would be constructed with only large and mid-size
companies. As expected, by using factor portfolios with
more concentrated holdings, investors can improve the
performance of a multifactor strategy, but these benefits
come at the expense of high turnover and high trading
costs. The authors specify portfolio design as a conscious
JPM 2019

QUANT OF THE YEAR AWARD

MARCOS LÓPEZ DE PRADO

THIS NEW ANNUAL AWARD presented by The Journal of Portfolio Management,

recognizes the outstanding contributions made by a single individual to our
field of research.

Machine learning (ML) has a growing importance in modern society. Today,

many areas of scientific research rely on the use of ML algorithms to build new
theories. Dr. Lopez de Prado has an extensive body of academic work that
has fostered the adoption of machine learning (ML) techniques in finance
and has been a vocal advocate for the responsible use of ML in finance.

“For many years, Marcos has led the way towards the adoption of machine
learning techniques in finance,” said Frank J. Fabozzi, Editor of JPM. “His many
publications have introduced innovative ways of thinking about financial
problems and solving them in practice. Our ‘Quant of the Year Award’
recognizes the totality of work by a researcher, and I think Marcos’ name was
in everyone’s mind from the onset of the selection process.”
QUANTITATIVE SPECIAL ISSUE 2019
choice made on the trade-off between the effective har-
vesting of the factor premium and low-cost implementa-
tion. They highlight the importance of the thoughtful
portfolio construction work required to deliver a smart
beta multifactor strategy that serves investors’ needs.
Factor Investing from
Concept to Implementation 125
Eduard van Gelderen, Joop Huij,
and Georgi Kyosev
Mutual funds following factor investing strategies based
on equity asset pricing anomalies, such as the small-cap,
value, and momentum effects, earn significantly higher
alphas than traditional actively managed mutual funds.
The authors report that a buy-and-hold strategy for a
random factor fund yields 110 basis points per annum
in excess of the return earned by the average traditional
actively managed mutual fund. However, they find
that the actual returns that investors earn by investing
in factor mutual funds are significantly lower because
investors dynamically reallocate their funds both across
factors and factor managers. Although factor funds have
attracted significant fund f lows over their sample period,
it appears that fund f lows have been driven by factor
funds earning high past returns and not by the funds
providing factor exposures. The authors argue that,
rather than timing factors and factor managers, inves-
tors would be better off by using a buy-and-hold strategy
and selecting a multifactor manager.
Extending Fama–French Factors
to Corporate Bond M arkets 141
Demir Bektić, Josef-Stefan Wenzler,
Michael Wegener, Dirk Schiereck,
and Timo Spielmann
The explanatory power of size, value, profitability,
and investment has been extensively studied for equity
10   The Journal of Portfolio Management Quantitative Special Issue 2019
VOLUME 45 NO. 3
markets. Yet, the relevance of these factors in global
credit markets is less explored, although equities and
bonds should be related according to structural credit
risk models. In this article, the authors investigate the
impact of the four Fama–French factors in the US and
European credit space. Although all factors exhibit eco-
nomically and statistically significant excess returns in
the US high-yield market, the authors find mixed evi-
dence for US and European investment-grade markets.
Nevertheless, they show that investable multifactor
portfolios outperform the corresponding corporate
bond benchmarks on a risk-adjusted basis. Finally, their
results highlight the impact of company-level charac-
teristics on the joint return dynamics of equities and
corporate bonds.
Factors in Time: Fine-Tuning
Hedge Fund Replication 159
Joseph Simonian and Chenwei Wu
Hedge fund replication has become a cottage industry
in investing. Among the most popular hedge fund rep-
lication frameworks are factor models based on ordi-
nary least squares (OLS) regression, a development that
is no doubt due to its simplicity and familiarity among
investment practitioners. Despite their widespread use,
the OLS regression-based factor models that form the
basis for many hedge fund replication programs are often
overfitted to a single sample, severely undercutting their
predictive effectiveness. As a remedy to the latter short-
coming, in this article the authors apply the regularization
method known as “ridge regression” to the replica-
tion of hedge fund strategies. Ridge regression works
by formally imbuing a regression with additional bias
in exchange for a reduction in the variance between
training and test samples. Using a simple yet robust meth-
odology, the authors show how to dynamically calibrate
the predictively optimal level of bias without significantly
reducing the backward-looking explanatory power of a
given model. In doing so, the authors demonstrate that
–

ACTIONABLE RESEARCH—
SMARTER INVESTING
THE IPR JOURNALS PLATFORM is the go-to resource for all
independent research for those within the investment
management community. Delve further than ever
before into our content on our new global platform.

UNRESTRICTED
ACCESS
11 JOURNALS
400 ARTICLES ANNUALLY
COVERING 60+ TOPICS

A RCHIVE OF OVER
10,000 ARTICLES

CONTACT US (646) 931-9045 | iprjournals@pageantmedia.com | www.iprjournals.com

QUANTITATIVE SPECIAL ISSUE 2019 VOLUME 45 NO. 3

ridge regression can help produce generalizable models

that are useful in both the ex post risk analysis and ex ante
replication of hedge fund strategies.

A Factor- and Goal-Driven

Model for Defined Benefit
Pensions: Setting Realistic Benefits 165
John M. Mulvey, Lionel Martellini,
Han Hao, and Nongchao Li
A factor and goal-driven framework for assessing asset
allocation and contribution decisions within defined-
benefit pension plans is developed in this article. A crit-
ical element is setting future benefits with reference to
the ability of the pension sponsors to support liabilities
under reasonable investment expectations. The approach
suggested by the authors combines a micro study of a
representative cohort of individuals with an aggrega-
tion across a target population to estimate consistency
between the micro and macro environments. A sto-
chastic inf lation risk factor affects both contribution and
spending cash f lows. This agent-based model suggested
by the authors provides a more realistic framework than
traditional approaches for setting pension benefits.

Patron Sponsor Section 178

12   The Journal of Portfolio Management Quantitative Special Issue 2019


Tarun Gupta
is a managing director at
AQR Capital Management
LLC in Greenwich, CT.
tarun.gupta@aqr.com
Bryan K elly
is a vice president at AQR
Capital Management LLC
in Greenwich, CT, and
professor of finance at Yale
School of Management in
New Haven, CT.
bryan.kelly@aqr.com
Quantitative Special Issue 2019
Factor Momentum Everywhere
Tarun Gupta and Bryan Kelly
P
rice momentum is most com-
monly understood as a phenom-
enon in which assets that recently
enjoyed high (low) returns relative
to others are more likely to experience high
(low) returns in the future. It is customarily
implemented as a cross-sectional trading
strategy among individual stocks ( Jegadeesh
and Titman 1993; Asness 1994) or long-only
equity portfolios (Moskowitz and Grinblatt
1999; Lewellen 2002). It has an impressive
and robust track record of risk-adjusted per-
formance (Asness et al. 2014; Geczy and
Samonov 2016).
Grouping stocks based on relative cross-
section performance has led many to inter-
pret momentum as a strategy that isolates
predominantly idiosyncratic momentum
(e.g., Grundy and Martin 2001; Chaves
2016). In this article, we document robust
momentum behavior among the common
factors that are responsible for a large fraction
of the covariation among stocks. A portfolio
strategy that buys the recent top-performing
factors and sells poor-performing factors (i.e.,
that exploits factor momentum) achieves sig-
nificant investment performance above and
beyond traditional stock momentum. On
a standalone basis, our factor momentum
strategy outperforms stock momentum,
industry momentum, value, and other com-
monly studied investment factors in terms
of Sharpe ratio. Furthermore, although
factor momentum and stock momentum
are correlated, they are also complementary.
Factor momentum earns an economically
large and statistically significant alpha after
controlling for stock momentum. Nor does
factor momentum displace stock momentum.
Because of stock momentum’s especially
strong hedging benefit with respect to value,
we find a significant benefit to combining
factor momentum, stock momentum, and
value in the same portfolio. 1 
In recent decades, academic literature
and industry practice have accumulated
dozens of factors that help explain the co-
movement and average returns among indi-
vidual stocks. We build and analyze a large
collection of 65 such characteristic-based fac-
tors that are widely studied in the academic
literature. From this dataset, we establish
factor momentum as a robust and pervasive
phenomenon based on the following facts.
Serial correlation in returns is the
basic statistical phenomenon underlying
momentum and is thus the launching point
for our analysis. First, we show that indi-
vidual factors exhibit robust time-series
momentum (Moskowitz, Ooi, and Pedersen
2012), a performance persistence phenom-
enon by which an asset’s own recent return
1
 This is especially true when value is con-
structed following the HML-Devil refinement of
Asness and Frazzini (2013).
The Journal of Portfolio Management   13
(in an absolute sense rather than relative to a peer group)
predicts its future returns. Persistence in factor returns
is strong and ubiquitous. The average monthly AR(1)
coefficient across all factors is 0.11; it is positive for 59
of our 65 factors and is significantly positive in 49 cases.
Second, we demonstrate that individual factors can
indeed be successfully timed based on their own past
performance. A time-series momentum trading strategy
that scales exposure to a given factor in proportion with
its own past one-month return generates excess perfor-
mance over and above the raw factor. This individual
time-series momentum alpha (i.e., after controlling for a
passive investment in the factor) is positive for 61 of the
65 factors and is statistically significant for 47 of them.
The annualized information ratio of this strategy is 0.33
on average over all 65 factors.
Third, a combined strategy that averages one-
month time-series momentum of all factors earns an
annual Sharpe ratio of 0.84, exceeding the performance
of any individual factor’s time-series momentum. We
refer to this combined portfolio of individual factor
timing strategies as time-series factor momentum (TSFM).
It performs similarly well with longer formation
windows. For example, the strategy’s Sharpe ratio is
0.70 when based on previous 12-month factor perfor-
mance and remains at 0.72 with a five-year look-back
window. TSFM is strongest with a one-month look
back, although we continue to find positive and signifi-
cant performance with longer nonoverlapping windows
as well (i.e., based on momentum over 2-12 or 13-60
months prior to formation).
The TSFM strategy is largely unexplained by
other well-known sources of excess returns. It has two
natural benchmarks for comparison. One is the equal-
weighted average of the 65 raw factors, which itself has
an impressive annual Sharpe ratio of 1.07. TSFM earns
large and significant alphas relative to this, indicating
that the performance of TSFM arises from beneficial
timing and is not simply picking up static factor per-
formance. The second natural benchmark is the tradi-
tional stock-level momentum strategy using the 2-12
formation strategy of Asness (1994), which we refer to
as UMD (up minus down) henceforth. UMD has an
annual Sharpe ratio of 0.56 in our sample. In spanning
regressions, UMD partially explains the performance of
TSFM, particularly when TSFM is based on a matched
2-12 formation period (i.e., excluding the most recent
month). Factor momentum, however, is strongest at the
14   Factor Momentum Everywhere Quantitative Special Issue 2019
one-month horizon, and this short-horizon persistence
is unexplained by UMD. We find that there are benefits
to longer formation periods as well, though the perfor-
mance of TSFM becomes more similar to UMD when
the formation window is extended to include the most
recent 12 months.
An important differentiating feature of TSFM is its
behavioral stability with respect to look-back window.
TSFM exhibits positive momentum whether it is based
on prior one-month, one-year, or even five-year per-
formance. This contrasts starkly with stock-based
momentum strategies. For both short (one month) and
long (beyond two years) formation windows, stocks
in fact exhibit reversals as opposed to momentum (De
Bondt and Thaler 1985; Jegadeesh 1990).
TSFM is an average of time-series momentum
strategies on individual factors. A natural alternative
strategy is to construct factor momentum relative to the
performance of the other factors in the cross section, as
in the Jegadeesh and Titman (1993) approach. We refer
to this as cross-section factor momentum (CSFM). We find
that CSFM and TSFM share a correlation above 0.90
for any formation window, and the standalone average
returns and Sharpe ratios of CSFM and TSFM are very
similar. However, when we regress TSFM returns on
CSFM, we find positive and highly significant TSFM
alphas, yet CSFM has generally negative (and signifi-
cant) alphas when controlling for TSFM. Their high
correlation and opposing alphas reveal that TSFM and
CSFM are fundamentally the same phenomenon but
that the time-series approach provides a purer measure
of expected factor returns than does the cross-sectional
method.
We also investigate the turnover and transaction
costs of factor momentum. Our conclusions regarding its
outperformance are unchanged when we look at Sharpe
ratios net of transaction costs. The net standalone Sharpe
ratios of TSFM and CSFM continue to exceed those
of stock momentum, industry momentum, short-term
reversal, and the Fama–French factors.
Our last empirical finding is that factor momentum
is a global phenomenon. We demonstrate its robustness
in international equity markets with magnitudes on par
with our US findings. We find similar outperformance
of TSFM over international versions of UMD, industry
momentum, and CSFM.
Each of our 65 factors represents a large, diversi-
fied long–short portfolio. These portfolios are (to close
approximation) devoid of idiosyncratic stock-level
returns, which are washed out by the law of large
numbers. Yet the TSFM strategy that buys (sells)
factors with high (low) past returns outperforms the
traditional stock momentum strategy. In other words,
factor momentum captures variation in expected factor
returns of roughly similar magnitude to that in stock-
level momentum, despite being purged of idiosyncratic
returns by construction. Factor momentum thus iso-
lates persistence in the common factors and shows that
momentum is a more general phenomenon, existing
alongside idiosyncratic stock return momentum.
We build upon recent work by Avramov et al.
(2016) and Arnott et al. (2018) in analyzing momentum
among factors. Their work focuses only on cross-section
factor momentum and only in the United States. Our
findings differ from previous work in establishing that
factor momentum is best understood and implemented
with a time-series strategy rather than a relative cross-
sectional approach. Our finding that TSFM explains
the performance of stock momentum is likewise a new
contribution to the literature. We also provide a more
expansive view of factor momentum, studying a more
comprehensive collection of US equity factors, and we
are the first to document factor momentum in interna-
tional equity markets.
The behavior of factor momentum is distinctly
reminiscent of work by Moskowitz, Ooi, and Pedersen
(2012), who demonstrated that asset class indexes may
be timed based on their recent past performance. Aggre-
gate commodity, bond, and currency indexes are rightly
viewed as factors within those asset classes, and as such
the results of Moskowitz, Ooi, and Pedersen (2012) can
be understood as a manifestation of factor momentum.
Taken together with the ubiquity of our equity-based
findings of factor momentum in the time series, in the
cross section, and around the world, we conclude that
there is indeed factor momentum everywhere.
FACTOR SAMPLE
We construct 65 characteristic-based factor portfo-
lios. Our aim is to cover the expanse of factors proposed
in the academic literature that studies the cross section
of stock returns, subject to constraints. We cover the
most well-cited and robust factors and have a substan-
tial overlap with recent research on high-dimensional
Quantitative Special Issue 2019
factor analysis.2  We focus on factors that can be con-
structed beginning in the 1960s. This excludes, for
example, Institutional Brokers’ Estimate System–based
analyst research factors, which only become available
in the late 1980s.
We for m factors as fol lows. First, we
cross-sectionally winsorize the top and bottom 1% of
raw characteristic values each period. Next, we split the
universe into large and small stocks with a cutoff equal
to median NYSE market capitalization (or the 80th per-
centile of market capitalization for international stocks).
Within size bins, we divide further into low/medium/
high characteristic values according to a 30/40/30 per-
centile split. Breakpoints are taken over NYSE stocks for
the US sample or all stocks in the international sample.
Within these six bins, we form value-weighted portfo-
lios and then combine these into an ultimate long–short
factor portfolio according to 0.5 × (Large High + Small
High) - 0.5 × (Large Low + Small Low), reconstituting
portfolios each month.
Our factor list includes, among others, a variety
of valuation ratios (e.g., earnings/price, book/market);
factor exposures (e.g., betting against beta); size, invest-
ment, and profitability metrics (e.g., market equity, sales
growth, return on equity); idiosyncratic risk measures
(e.g., stock volatility and skewness); and liquidity mea-
sures (e.g., Amihud illiquidity, share volume, and bid–
ask spread).
FACTORS AT A GLANCE
Exhibit 1 lists the variables and basic performance
characteristics. We report each factor’s average return,
Sharpe ratio, and Fama and French (2016) five-factor
alpha (returns and alphas are in percent per annum).
The Appendix provides additional details, including a
factor description and the original articles that analyzed
each factor (we follow these articles as closely as possible
when constructing our factor dataset). We orient the
long–short legs of each factor such that the predicted
sign of the factor’s expected return is positive (according
to the paper originally proposing each factor). Note that
this does not mean that all factors have positive average
returns in our sample—we find that 3 of 65 have nega-
tive average returns when extended through 2017, and
2
 See Harvey, Liu, and Zhu (2016); McLean and Pontiff
(2016); Gu, Kelly, and Xiu (2018); and Kelly, Pruitt, and Su (2018).
The Journal of Portfolio Management   15
Exhibit 1
Factor Sample Summary Statistics

)DFWRU (>5@ 6KDUSH ))α 3DQHO$(>5@


$%1,19   
$&&    
$'9(57&+* ± ± ±
$'09   ± 
$,0   
$72    
$72&+*   
%$%    

&$372   


&(*7+   
&$6+   

$66(7*   
&3    
'&$3   
'(/$< ± ± ± ±
'3   ±
,668(   ± ±

5(96853
'&$3

'3

63
&3
(3
;5'$7

,92/

6($621$/

2/
6+257,17

0$;5(7

(1708/7
'(/$<

$'09

$%1,19

;5'09
,6.(:

52$

12$
$'9(57&+*
30
(&216

/9*%0

(36020
&$6+
30&+*
*6
)6&25(
;5'6

$72&+*
52(
60%
=6&25(
$,0
92/0751
&(*7+
/9*
26&25(
&$372
$&&
1:&&+*

*5352)
:+
/12$&+*
33$&+*
$66(7*
%0
,668(
,668(
(17%0
$72
1&2$&+*
,9*
40-
)5((&)
;),1

%$%
,1'020
020
/75

785129(5
5(5

686*5

675
,668(   
(17%0   ±
(1708/7   
(3   
(36020    3DQHO%6KDUSH5DWLR
)5((&)    

)6&25(   


*5352)   
*6   ±

%0   ±
,9*    
,6.(:   ±
,92/    
/9*   ±
/12$&+*    
/75   ±
/9*%0    

0$;5(7   

1&2$&+*    ±

12$   

±
1:&&+*   
(&216 ± ± ±
±
2/   
'&$3
5(96853

'3

(3

63

&3
,92/
;5'$7

6($621$/

2/
0$;5(7

6+257,17

(1708/7
'(/$<

$'09
,6.(:

$%1,19
;5'09

52$

12$
$'9(57&+*
30
(&216

/9*%0

(36020
;5'6

*6
30&+*
60%
/9*
92/0751
:+
=6&25(
&$6+
52(
$,0
)6&25(
$72&+*

&$372
&(*7+
26&25(
*5352)
(17%0
%0
,668(
$72
$&&
,668(

$66(7*
,1'020
1:&&+*
40-
33$&+*

020
)5((&)

%$%
;),1
/12$&+*
,9*
1&2$&+*
/75

785129(5

686*5

5(5
675

26&25(   

30 ± ± ±
30&+*   
33$&+*   
3DQHO&))$OSKD
40-    
5(5   
675   
5(96853   

52$   
52(   
6($621$/   
6+257,17    

60%   ±

63   ±
686*5   ± 
785129(5   
020   
,1'020   
±
92/0751   ±
:+   
;),1   
;5'$7    ±
'3

63

(3

5(96853

&3

'&$3
(1708/7
,92/
2/

6($621$/
0$;5(7

6+257,17

;5'$7
$'09
'(/$<

52$

$%1,19

12$

;5'09
,6.(:
/9*
$'9(57&+*

*6
(17%0
(&216
92/0751
%0
60%
,668(
30
&(*7+
$66(7*
$,0
33$&+*
&$6+
52(
)6&25(
30&+*
&$372

$72&+*
(36020
26&25(
1&2$&+*
,668(
/12$&+*
,9*

=6&25(
;),1
$&&

$72
)5((&)
*5352)
1:&&+*
:+

40-
/9*%0
%$%
;5'6
,1'020
020
/75

686*5

785129(5

5(5

675

;5'09   

;5'6   
=6&25(   

Note: * and ** signify that a weight estimate is significantly different from zero at the 5% or 1% level, respectively.

16   Factor Momentum Everywhere Quantitative Special Issue 2019

25 are statistically indistinguishable from zero. Mean-
while, the factors with the strongest and most statistically
reliable performance are the best-known usual suspects,
such as betting against beta, stock momentum, industry
momentum, and valuation ratios (cash f low/price, sales/
price, and earnings/price).
Although all factors represent large and diversi-
fied portfolios, they nonetheless possess rather distinct
return behavior. More than half of all factor pairs have
a correlation below 0.25 in absolute value. Principal
component (PC) analysis also supports the view that an
unusually large amount of the portfolio return variation
is factor specific: It takes 19 PCs to explain 90% of the
65-factor correlation structure, 28 to explain 95%, and
46 to explain 99%.
FACTOR MOMENTUM
Factor Persistence
We begin our analysis by investigating the pri-
mary statistical phenomenon underlying momentum:
serial correlation in returns. In Exhibit 2, we report
monthly first-order autoregressive coefficients (denoted
as AR(1)) for each factor portfolio along with 95% con-
fidence intervals. When zero lies outside the confidence
interval, it indicates that the estimate is statistically sig-
nificant at the 5% level (or, equivalently, the t-statistic
is greater than 1.96 in absolute value).
The strength and pervasiveness of one-month
own-factor serial correlation is stunning. Of our 65 fac-
tors, 59 have a positive monthly AR(1) coefficient, and
the coefficient is statistically significant for 49 of these.
For comparison, the monthly AR(1) coefficient for the
excess market return is 0.07 during our sample, which
Moskowitz, Ooi, and Pedersen (2012) demonstrated
is powerful enough for implementing a time-series
momentum strategy. The average AR(1) coefficient of
our factors is 0.11, and 50 of them have an AR(1) coef-
ficient larger than that of the market. 3  This is a first
indication that it may be possible to time factors based
on their own past performance.
3
 We believe that own-factor persistence may be even stronger
than these results portray because any illiquidity imbalance in a
factor will tend to create some negative serial correlation, and we
are not directly accounting for that here.
Quantitative Special Issue 2019
Time-Series Factor Momentum
The strong autoregressive structure in factor
returns suggests that it may be possible to time each
factor individually based on its own recent performance.
The idea of portfolio timing based on the portfolio’s
own past return underlies the time-series momentum
methodology of Moskowitz, Ooi, and Pedersen (2012).
We begin by exploring the benefits of portfolio
timing by applying a time-series momentum strategy
one factor at a time. We focus on one-month holding
periods and consider various formation windows of one
month up to five years. Our strategy dynamically scales
one-month returns of the ith factor, f i,t+1, according to
its performance over the past j months
j ,t +1 = si , j ,t × f i ,t +1 ,
f i ,TSFM
  1 j
 
si , j ,t = min  max 
 σ i , j ,t
∑f i ,t −τ+1 , −2 ,2 (1)
 
 τ=1
Unpacking Equation 1, we use the scaling term si,j,t
to time positions in factor i based on the factor’s return
over the formation period (t − j to t). If formation returns
are positive, it buys the factor; if negative, it sells the
factor. We convert recent returns to z-scores by dividing
by σi,j,t , which is the annualized factor volatility, over
the previous three years (for short formation windows,
j < 12) or over the previous 10 years (if j ≥ 12), and we
cap z-scores at ±2.4
The benefits of factor timing can be assessed in
terms of alpha by regressing the scaled factor on the
raw factor
f i ,TSFM
j ,t = α i , j + βi , j f i ,t + ei , j ,t
Exhibit 3, Panel A, reports the annualized per-
centage alphas from the time-series strategy with a one-
month formation period for each factor, as well as 95%
confidence intervals. The performance of time-series
momentum in individual factors is extraordinarily per-
vasive. It is positive for 61 out of 65 factors and statistically
significant for 47 of these. To provide a clearer inter-
pretation in terms of risk–return trade-off, Exhibit 3,
 Our findings are robust to other estimation choices for σi,j,t ,
4
including shorter windows and exponentially weighted moving
averages, and to other caps such as ±1.
The Journal of Portfolio Management   17
Exhibit 2
Factor Return Monthly AR(1) Coefficients












±

±
,668(

*6
,668(
60%

;5'6

%$%

(&216
020

*5352)

)5((&)

$72
&$372
'&$3

&3

(3
5(96853

$&&

63
'3
(1708/7
6+257,17
,1'020

;5'$7

26&25(
:+

0$;5(7

)6&25(
$,0

&(*7+

(36020

=6&25(

52(
&$6+

,92/

6($621$/
30

%0
2/
/9*%0

(17%0
30&+*

;),1
$'9(57&+*

'(/$<
$72&+*

/12$&+*

33$&+*
;5'09

92/0751
$%1,19

,9*

/9*

$66(7*

1:&&+*
$'09
12$

1&2$&+*

52$
,6.(:

40-
/75
5(5
675

785129(5

686*5
Panel B, shows Sharpe ratios for each individual factor
momentum strategy: It exceeds 0.20 for 56 factors and
is statistically significant for 48 of them.
Our overall TSFM strategy combines all individual
factor time-series momentum strategies into a single
portfolio. In particular, TSFM aggregates timed factors
(with formation window j) according to
TSFM j ,t = TSFM Long
j ,t − TSFM Short
j ,t
where
TSFM earns an annualized average return of 12.0%.
The last bar in Panel A reports the alpha from regressing
the one-month TSFM return on the equal-weighted
average of raw factor returns. The equal-weighted
portfolio of raw factors is itself an impressive strategy,
earning an annualized Sharpe ratio of 1.07. Nevertheless,
the portfolio of individual factor momentum strategies
generates a highly significant 10.3% alpha (t-statistic of
4.6) after controlling for the average of untimed factors.
The last bar in Panel B reports the annual Sharpe ratio
of the combined factor momentum portfolio. It is 0.84,
exceeding the Sharpe ratio of every individual factor

TSFM Long =
∑1 f TSFM
i { si , j ,t >0} i , j ,t +1
and
momentum strategy.
Exhibit 4 explores how TSFM performance
j ,t
∑1 s
i { si , j ,t >0} i , j ,t changes with alternative implementations. We form

TSFM Short
=
∑1 f TSFM
i { si , j ,t ≤0} i , j ,t +1
the momentum signal using look-back windows of one
month (1-1) up to five years (1-60). We also split out
j ,t
∑1 s
i { si , j ,t ≤0} i , j ,t the 11 months excluding the most recent month (2-12)
for more direct comparability with UMD and the four
That is, the long and short legs are rescaled to form years excluding the most recent year (13-60) to compare
a unit leverage ($1 long and $1 short) TSFM portfolio. the role of long-term versus short-term return trends.

18   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 3
Time-Series Momentum for Individual Factors (one-month formation)
3DQHO$2ZQ)DFWRU$OSKD








±

±
$,0
$'9(57&+*
$%1,19
'(/$<
,6.(:
6+257,17
&$6+
&(*7+
)6&25(
/12$&+*
12$
52$
92/0751

26&25(

30
(1708/7
*6
(36020
;),1
&3
*5352)
$66(7*
,668(
;5'6
&$372
)5((&)
(3
:+
$72
;5'$7
/9*%0
2/
,1'020
$72&+*
30&+*
1&2$&+*
$&&
5(96853
1:&&+*
,668(
;5'09
,9*
'&$3
6($621$/
33$&+*
(&216

40-
60%
$'09
020
52(
%0
/9*
(17%0
63
=6&25(
'3
0$;5(7
%$%
,92/
76)0
5(5

686*5
675

785129(5

/75
3DQHO%$QQXDOL]HG6KDUSH5DWLR










±

±

±
$,0
$'9(57&+*
$%1,19
'(/$<
,6.(:
6+257,17
;5'09
'&$3
92/0751
&(*7+
,1'020
)6&25(
52$
&$6+
:+
;5'6
6($621$/
5(96853
/12$&+*
;5'$7
12$
$72&+*
30&+*
,668(
26&25(
/9*%0
020
$&&
,668(
*5352)
33$&+*
30
$'09
&3
(3
(&216
1&2$&+*
60%
'3
*6
(1708/7
,92/
/9*
1:&&+*
)5((&)
0$;5(7
40-
$72
;),1
$66(7*
&$372
,9*
(36020
%0
52(
63
(17%0
=6&25(
2/
%$%
76)0
5(5

675

785129(5

/75

686*5

Panel A reports the raw TSFM average return for With a look-back as long as five years, TSFM earns
each formation period, and Panel B reports annualized 7.1% per annum with a Sharpe ratio of 0.72. Panel A
Sharpe ratios. The 12-month TSFM strategy achieves shows that although one-month factor momentum is the
an expected return of 9.5% and a Sharpe ratio of 0.70. overall performance driver, large positive and significant

Quantitative Special Issue 2019 The Journal of Portfolio Management   19

Exhibit 4
Risk-Adjusted TSFM Performance
3DQHO$5DZ(>5@ 3DQHO%6KDUSH5DWLR
 





 





 
               
0RQWKVLQ)RUPDWLRQ3HULRG 0RQWKVLQ)RUPDWLRQ3HULRG

3DQHO&αRQ(:)DFWRU 3DQHO'αRQ))
 
 
 
 
 
 
 
 
 
± 
               
0RQWKVLQ)RUPDWLRQ3HULRG 0RQWKVLQ)RUPDWLRQ3HULRG

contributions from longer (nonoverlapping) 2-12 and
13-60 formation windows remain as well, with Sharpe
ratios of 0.54 and 0.53, respectively.
When we benchmark TSFM against the equal-
weighted average factor (EW, shown in Panel C) or
against the Fama–French f ive-factor model (FF5,
Panel D), the excess performance of TSFM is little
affected. For a one-month formation, EW explains less
than one-sixth of the TSFM average return, and at one
year it explains less than one-third. The Fama–French
model explains less than one-tenth of TSFM’s average
return for all formation windows.
Cross-Section Factor Momentum
section of all factors. CSFM buys (sells) factors that have
recently outperformed (underperformed) peers, rather
than sizing factor exposures based on their own recent
performance. For example, if all factors recently appre-
ciated, TSFM will take long positions in all of them.
CSFM, on the other hand, will be long only the relative
outperformers and will short those with below-median
recent returns (despite their recent positivity).5 
Exhibit 5 explores the performance of CSFM with
various look-back windows for portfolio formation (in
analogy with Exhibit 4). The results show that CSFM
and TSFM have similar behavior. The Sharpe ratios of
CSFM are slightly inferior to TSFM, and it has slightly
smaller alphas with respect to the equal-weighted

An alternative approach to forming a factor 5

 CSFM cross-sectionally de-means factors’ formation-
momentum strategy is to take positions in factors based window returns but otherwise follows the same construction as
on the recent performance of factors relative to the cross TSFM.

20   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 5
Risk-Adjusted CSFM Performance
3DQHO$5DZ(>5@



 
       
0RQWKVLQ)RUPDWLRQ3HULRG
3DQHO&αRQ(:)DFWRU


 
 
       
0RQWKVLQ)RUPDWLRQ3HULRG
portfolio of raw factors, but their performance patterns
are otherwise closely aligned.
Factor, Stock, and Industry Momentum
Next, we directly compare various incarnations of
the momentum effect against each other, including factor
momentum (TSFM and CSFM), stock-level momentum
(UMD), short-term stock reversal (STR), and industry
momentum (INDMOM, following Moskowitz and
Grinblatt 1999 and Asness, Porter, and Stevens 2000,
for which we use a 1-12 formation strategy). To make a
clearer comparison among average returns, we rescale
all five series to have an ex post annualized volatility
of 10%.
Exhibit 6 provides a preliminary visual comparison
of momentum strategies. It shows the cumulative log
Quantitative Special Issue 2019
3DQHO%6KDUSH5DWLR






±
       
0RQWKVLQ)RUPDWLRQ3HULRG
3DQHO'αRQ))


       
0RQWKVLQ)RUPDWLRQ3HULRG
returns for each momentum variable, including 1-month
and 12-month look-back windows for TSFM and
CSFM, along with the excess market portfolio. Two
features of this plot stand out. First is the comparatively
steep slope of TSFM. This is consistent throughout the
sample rather than being driven by a few good runs.
(One-month CSFM shares a similarly steep slope, but
the 12-month implementation drops off substantially).
Second is the sharp drawdown of UMD, when stock
momentum experienced a loss of 31% from March to
May 2009 (Daniel and Moskowitz 2016). INDMOM also
experienced a drawdown of 24% over this time. In con-
trast, factor momentum entirely avoided the momentum
crash. Over the same months, 12-month TSFM and
CSFM earned 16% and 15%, respectively (one-month
versions of TSFM and CSFM both earned 18%).
The Journal of Portfolio Management   21
Exhibit 6
Cumulative Returns of Momentum Strategies

76)0
 &6)0
76)0
 &6)0
80'
&XPXODWLYH3HUFHQWDJH/RJ5HWXUQ

 ,1'020
5P5I










±
          

Exhibit 7
Momentum Strategy Return Correlations

)RUPDWLRQ &6)0 76)0

:LQGRZ 80' 675 ,1'020 80' 675 ,1'020 &6)0 76)0
  ±   ±  
  ±   ±  
  ±   ±  
  ±   ±  
  ±   ±  
  ±   ±  
  ±   ±  
  ±   ±  

It is well known that stock momentum is con-
centrated in intermediate formation windows of 6 to
12 months. With very short look-backs (one month)
or at long horizons, stocks experience reversals rather
than momentum. To gain a basic understanding of
co-movement in strategies, particularly with respect to
different formation periods, Exhibit 7 reports momentum
correlations. We include UMD, which describes stock
momentum from a 2-12 strategy, as well as STR, which
captures short-term stock reversals that arise in a 1-1
strategy. We compare each of these to TSFM and CSFM
with a range of formation choices ranging from 1 month
to 60 months and again splitting out 2-12 and 13-60.
Exhibit 7 highlights an interesting distinction in the
time series dynamics of different momentum strategies.
When factor momentum is based on an intermediate

22   Factor Momentum Everywhere Quantitative Special Issue 2019

window of 1–12 months, it bears a close correlation with
UMD (0.76 and 0.75 for TSFM and CSFM, respectively,
and similar for 2-12 factor momentum). Exhibit 7 also
illustrates a close similarity between factor momentum
and industry momentum.
In contrast, with a one-month window, factor
momentum behavior is strongly opposite to the stock-
based STR strategy (correlation of −0.80 for both TSFM
and CSFM). If factor momentum were simply capturing
stock-level persistency, we would expect it to also dis-
play a short-term reversal (in contrast to the findings of
Exhibits 4 and 5) and would therefore expect it to be
positively correlated with STR.
The last column shows the extremely high correla-
tion between time-series and cross-section approaches
to factor momentum.
Next, we regress TSFM and CSFM on momentum
alternatives to understand whether these strategies sub-
sume factor momentum. Panel A of Exhibit 8 reports
the average return of TSFM from various look-back
windows as well as the alphas of TSFM relative to
UMD, INDMOM, and STR. All estimates are accom-
panied by their 95% confidence intervals. A confidence
interval that excludes (includes) zero indicates that the
estimate is statistically significant (insignificant) at the
5% level. Bar colors correspond to different look-back
windows for TSFM and are described in the legend
(UMD, INDMOM, and STR look-back windows are
held fixed).
Controlling for UMD only explains the perfor-
mance of the 2-12 TSFM strategy. For all other look-
back windows, TSFM has a significant alpha of at least
2% per year versus UMD. For one-month TSFM in
particular, UMD has no explanatory power because the
alpha and raw average returns are essentially the same.
Alphas relative to INDMOM show a pattern similar to
those relative to UMD but are somewhat larger. Con-
trolling for STR in fact raises TSFM’s alpha above its
raw average return, which is perhaps expected given
their strong negative correlation. The right-most bars in
Panel A show the alpha of TSFM relative to CSFM with
matching formation window. Despite nearly perfect cor-
relations between them, TSFM’s alpha is significantly
positive for all formation windows and becomes stronger
at long horizons.
Panel B of Exhibit 8 performs the same compar-
ison for CSFM. There are two key distinctions between
Panels A and B. First, UMD and INDMOM explain
Quantitative Special Issue 2019
more of CSFM’s performance than they do TSFM’s per-
formance, and CSFM’s alphas on UMD and INDMOM
are insignificant for formation windows of a year or
more. Second, CSFM has negative and significant alpha
relative to TSFM. In other words, although TSFM and
CSFM earn similarly high average returns and are highly
correlated, TSFM harvests factor momentum compensa-
tion more efficiently than CSFM does.
In Exhibit 9, we reverse this analysis to assess the
performance of UMD, INDMOM, and STR after con-
trolling for factor momentum. We report alphas from
regressions of these factors on TSFM and CSFM with
various formation windows. As in Exhibit 8, bar colors
correspond to different look-back windows for TSFM
(holding UMD, INDMOM, and STR fixed).
The 1-12, 1-36, and 1-60 TSFM strategies can
each individually explain most of the performance of
UMD and INDMOM. The average annual return of
UMD is 6.1%, but its alpha versus 12-month TSFM, for
example, drops below 1% and its t-statistic falls below
1.0. The alpha of INDMOM is slightly negative and is
likewise insignificant. CSFM is unable to explain the
performance of UMD, but it does capture a large por-
tion of industry momentum. The central conclusion
from this spanning analysis is that TSFM tends to out-
perform, and to a large extent account for, the returns
to UMD.
Neither TSFM nor CSFM explains short-
term reversal. To the contrary, controlling for factor
momentum boosts the performance of STR from
an unconditional average return of 3.4% per year to
alphas in excess of 5% and as high as 10.1% versus one-
month TSFM. Thus, unlike UMD and TSFM, factor
momentum and short-term reversal seem to capture
distinct patterns in expected stock returns because both
have large unexplained alphas relative to each other.
PORTFOLIO COMBINATIONS
We next investigate the extent to which various
momentum strategies play an incrementally beneficial
role in a broader portfolio that includes other common
investment factors. In particular, we form ex post (i.e.,
full sample) mean–variance-efficient tangency portfolios
of factors. The first column of Exhibit 10 lists the fac-
tors that we consider, which continue to be standard-
ized to have 10% annualized volatility to put all factors
on equal volatility footing. We include nonoverlapping
The Journal of Portfolio Management   23
Exhibit 8
Comparison of Momentum Strategies
3DQHO$76)05HODWLYH3HUIRUPDQFH







$QQXDO

±

±

±
(>5@ αYV80' αYV,1'020 αYV675 αYV&6)0

3DQHO%&6)05HODWLYH3HUIRUPDQFH







$QQXDO

±

±

±
(>5@ αYV80' αYV,1'020 αYV675 αYV76)0

       

24   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 9
Relative Performance of UMD, INDMOM, and STR
3DQHO$80'





$QQXDO






±
αYV76)0 αYV&6)0

3DQHO%,1'020



$QQXDO

±

±
αYV76)0 αYV&6)0

3DQHO&675





$QQXDO


αYV76)0 αYV&6)0

       

Quantitative Special Issue 2019 The Journal of Portfolio Management   25

Exhibit 10
Ex Post Tangency Portfolios

,QGLY 7DQJHQF\3RUWIROLR:HLJKWV
)DFWRU 6KDUSH       
76)0    
76)0    ±
76)0    
76)0    
&6)0  
&6)0  
&6)0  
&6)0  ±
80'     
675   
0.7     
60%     
+0/     
50:     
&0$     
6KDUSH       

Note: * and ** signify that a weight estimate is significantly different from zero at the 5% or 1% level, respectively.

1-1, 2-12, and 13-60 versions of TSFM and CSFM,
as well as the 1-12 versions of each. We also include
UMD, INDMOM, and STR. Finally we investigate
combinations with the Fama–French five-factor model.
Superscripts of single and double asterisks signify that
a weight estimate is significantly different from zero at
the 5% or 1% level, respectively.
The second column reports the standalone Sharpe
ratios for each factor. The remaining columns, labeled 1
to 7, report tangency portfolio weights among various
sets of factors. Column 1 shows that the ex post efficient
combination of 1-1, 2-12, and 13-60 TSFM puts the
heaviest weight (0.47) on 1-1 TSFM but also puts sig-
nificantly positive weight on 2-12 and 13-60 (0.22 and
0.31, respectively). This tangency combination achieves
a Sharpe ratio of 1.07. For CSFM, the tangency portfolio
is dominated by a weight of 0.67 on the 1-1 component.
Column 3 shows that the optimal combination of TSFM
and CSFM takes a highly levered position in TSFM
with a large negative offsetting position in CSFM. This
result restates the fact that TSFM and CSFM are highly
correlated but have alphas of opposite signs with respect
to one another.
Column 4 considers the optimal combination of
TSFM with UMD and the Fama–French factors. In
this case, 2-12 TSFM takes an exact zero weight and
is replaced by a significantly positive weight of 0.10 on
UMD. This combination earns a Sharpe ratio of 1.65
(the five Fama–French factors on their own achieve a
tangency Sharpe ratio of 1.09). The same conclusion
emerges if we simultaneously include UMD and STR
alongside TSFM (Column 6), where all three enter the
tangency portfolio with significantly positive weights.
Among the Fama–French factors, MKT, conserva-
tive minus aggressive (CMA), and robust minus weak
(RMW) are significant contributors to tangency across
the board.
The diversif ication benef its from combining
momentum factors with value factors become more
pronounced when using the HML-Devil refinement
of Asness and Frazzini (2013), which incorporates more
timely price data in its value signal construction and
significantly outperforms the traditional Fama–French
HML. Exhibit 11 shows that the correlation of UMD
and HML-Devil is −0.64, whereas UMD is only −0.18
correlated with Fama–French HML. Likewise, the

26   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 11
Correlation of Momentum and Value Variants
76)0
80' ,1'020   
80' ±   
,1'020 ± ±  
+0/ ± ±  ± ±
+0/'HYLO ± ±  ± ±
correlation of 1-12 TSFM drops from −0.02 with HML
to −0.37 with HML-Devil.
Motivated by the potential for stronger hedging
benefits, Exhibit 12 investigates the impact of replacing
HML with HML-Devil in our tangency portfolio anal-
ysis. Three observations emerge from this table. First,
our central conclusions regarding factor momentum are
unchanged—it remains a strong contributor to optimal
multifactor portfolios. Second, HML-Devil takes a
large and statistically significant portfolio weight in
all cases, in contrast with the general insignificance
of Fama–French HML in Exhibit 10. Third, UMD
becomes one of the most important components of
the tangency portfolio thanks to the added diversifica-
tion benefits of coupling UMD and HML-Devil.6 In
summary, factor momentum and stock momentum are
most effectively used in tandem when devising optimal
portfolios.
IMPLEMENTABILITY
Momentum strategies are high turnover by
nature; thus, trading costs are a first-order consider-
ation for understanding the practical usefulness of factor
momentum in portfolio decisions. Panel A of Exhibit 13
compares the average annualized turnover of factor
6
 Exhibit 12 highlights the benefits of combining value and
momentum strategies (a point previously emphasized by Asness,
Moskowitz, and Pedersen 2013). Our factor momentum findings
naturally call for an investigation into an analogous factor value
strategy that times factors based on factor-level value signals (as
discussed, for example, by Asness 2016a, 2016b and Asness et al.
2000). Although an exploration of factor value, and in particular
the benefits of combining it with factor momentum, is beyond the
scope of this article, it is a fascinating direction for future research.
Quantitative Special Issue 2019
&6)0
    
     
     
±  ± ± ±
± ± ± ± ±
momentum with that of other price trend factors.7 
In terms of formation window, STR is the natural stock-
level benchmark for one-month factor momentum, and
UMD and INDMOM are most natural for comparison
with 12-month factor momentum. In both cases we see
that factor momentum turnover is comparable to, but
slightly lower than, its stock-level counterpart. Panel A
also shows that, like other momentum varieties, factor
momentum involves substantially more trading than
Fama–French factors.
Panel B of Exhibit 13 compares the performance
of strategies net of transaction costs. Our calculations
assume costs of 10 bps per unit of turnover (based on
the estimates of Frazzini, Israel, and Moskowitz 2015).
Red bars represent the net annualized Sharpe ratio for
each strategy, along with the gross Sharpe ratio in blue
for comparison. The main takeaway from the exhibit is
that although trading costs indeed eat into the perfor-
mance of factor momentum, its net performance con-
tinues to exceed that of UMD, INDMOM, STR, and
the Fama–French factors. For example, the net Sharpe
ratio of TSFM 1-12 is 0.63, versus 0.70 gross. The next
best net Sharpe ratio among stock-level price trend fac-
tors is 0.51 for UMD, and the best among Fama–French
factors is 0.45 for RMW.
Lastly, Panel B sheds new light on the findings in
Exhibit 9: It reveals that the strong performance of STR
after controlling for factor momentum is illusory. Even
on a standalone basis, the performance of short-term
reversal is entirely wiped out by transaction costs.
7
 Average annualized turnover is defined as the sum of abso-
lute changes in portfolio weights each month, averaged over all
months and multiplied by 12. This describes total two-sided trading
volume (both entering and exiting positions) as a fraction of gross
asset value.
The Journal of Portfolio Management   27
Exhibit 12
Ex Post Tangency Portfolios Including HML-Devil

,QGLY 7DQJHQF\3RUWIROLR:HLJKWV
)DFWRU 6KDUSH        
76)0    
76)0   ± 
76)0    
76)0    
80'    
,1'020    
+0/'HYLO         
0.7     
60%     
50:     
&0$  ±   
6KDUSH        

Note: * and ** signify that a weight estimate is significantly different from zero at the 5% or 1% level, respectively.

FACTOR MOMENTUM AROUND THE WORLD
In this section, we show that each of our main
factor momentum conclusions from the US sample is
strongly corroborated in international equity markets.
We study three international samples. The Europe
sample includes Austria, Belgium, Switzerland,
Germany, Denmark, Spain, Finland, France, the
United Kingdom, Greece, Ireland, Italy, Netherlands,
Norway, Portugal, Sweden, and Israel. The Pacific
sample includes Australia, Hong Kong, Japan, New
Zealand, and Singapore. The broadest international
sample we consider is global (without the United States)
and combines Europe, Pacific, and Canada. Because of
data limitations, we study only 62 of the original 65
factors in the international sample. 8  Third, international factor momentum demon-
First, individual factor returns are highly per-
sistent. The average AR(1) coefficient is 0.10 (versus
0.11 in the United States), is positive for 51 of 62 fac-
tors, and is signif icant for 30 of these. Exhibit 14
shows that the success of individual factor time-series
momentum strategies (one-month formation) work as
well for international factors as they do for the United
States. The alpha of momentum-timed factors versus
raw factors is positive for 55 of 61 factors and is sig-
nificant for 22 of these (versus 61 of 65 positive in
the United States, 47 of those significant). The TSFM
8
 Excluded are ADVERTCHG, AD2MV, and AIM.
portfolio that aggregates individual time-series factor
momentum strategies has a Sharpe ratio of 0.73 (versus
0.84 in the United States) and earns an alpha of 6.6%
per year after controlling for the equal-weighted port-
folio of raw (untimed) factors.
Second, international factor momentum demon-
strates extraordinarily stable performance regardless
of formation window (shown in the left-most bars of
Exhibit 15). TSFM and CSFM earn essentially the same
average return whether they use a short look-back of
one month, all the way through a long look-back of five
years. As in the US sample, this is a remarkable diver-
gence from stock-level continuation patterns, where a
one-month window gives rise to reversal but a one-year
window captures momentum.
strates large and significant excess performance after con-
trolling for other varieties of international momentum,
including UMD, INDMOM, and STR (Exhibit 15).
The TSFM alpha versus UMD is significantly positive
for all formation windows except 13-60.
Fourth, TSFM and CSFM are more than 0.95 cor-
related for all formation windows. Yet TSFM tends to
possess positive alpha relative to CSFM, and CSFM earns
negative alpha versus TSFM; this finding indicates that,
as in the US sample, TSFM more efficiently captures the
benefits of factor momentum.

28   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 13
Turnover and Net Sharpe Ratio
3DQHO$7XUQRYHU



$YHUDJH$QQXDO7XUQRYHU











$
/
5
'


%



7
0

:
0




&0
67

60
0

.




50




+
0
)0

)0

0
)0

)0

'
76

&6

,1
76

&6

3DQHO%6KDUSH5DWLR










$
/
5
'


%



7
0

:
0




&0
67

60
0

.




50




+
0
0

)0

0
)0

)0
)

'
76

&6

,1
76

&6

*URVV 1HW

Fifth, the performance of UMD and INDMOM
is explained by factor momentum. Exhibit 16 shows that
UMD’s alpha is essentially zero, and INDMOM has a
negative alpha after controlling for either TSFM or CSFM.
Sixth, international tangency portfolio analysis in
Exhibit 17 highlights the additivity of factor momentum
to the broader set of investment factors.9 The conclusions
9
 All momentum variables are based on international equi-
ties. However, because our data begin earlier than Ken French’s
international five-factor data, we use the US Fama–French factors.
from Exhibit 17 are qualitatively similar to the US
analysis in Exhibit 10. US and international TSFM 1-1
share a correlation of 0.62, and the US and international
1-12 versions are 0.64 correlated. The ex post tangency
portfolio that combines US and international TSFM 1-1
earns an annual Sharpe ratio of 0.83; individually, they
each earn 0.73.
In further (unreported) robustness analyses, we
find that the majority of the performance of the factor
momentum strategy arises from dynamically adjusting

Quantitative Special Issue 2019 The Journal of Portfolio Management   29

 ±
±
±







±
±




&)92/ &)92/
5(5 5(5
&$6+ 6($621$/
92/0751 ,9*
Exhibit 14

;5'6 $&&
675 1:&&+*
$72&+* &$6+
1:&&+* 675
$&& ;5'6
,9* 92/0751
'(/$< 5(96853

30   Factor Momentum Everywhere

$72 &$372
5(96853 *5352)
3DQHO$2ZQ)DFWRU$OSKD

*5352) '(/$<
785129(5 $72&+*
30&+* &(*7+
&(*7+ 33$&+*

3DQHO%$QQXDOL]HG6KDUSH5DWLR
'&$3 30&+*
;5'$7 12$
60% '&$3
/12$&+* $72
&$372 2/
*6 /12$&+*
2/ (36020
52( 1&2$&+*
33$&+* *6
;),1 52(
12$ ,6.(:
;5'09 785129(5
$%1,19 60%
26&25( 686*5
30 $%1,19
6($621$/ 26&25(
686*5 30
(36020 $66(7*
1&2$&+* ;),1
(3 ;5'$7
$66(7* (&216
/9*%0 /9*%0
52$ ;5'09
,1'020 52$
(1708/7 )5((&)
0$;5(7 &3
,6.(: (3
(&216 (1708/7
:+ %0
,92/ ,668(
/75 =6&25(
=6&25( /9*
'3 )6&25(
/9* '3
%0 ,668(
)5((&) (17%0
)6&25( 63
63 %$%
(17%0 /75
%$% :+
Time Series Momentum for Individual Factors (global ex. US, one-month formation window)

&3 ,92/
020 40-
,668( ,1'020
40- 0$;5(7
,668( 020
76)0 76)0

Quantitative Special Issue 2019

Exhibit 15
Comparison of Momentum Strategies (global ex. US)
Panel A: TSFM Relative Performance
14

12

10

6
Annual %

–2

–4

–6
E[R] α vs. UMD α vs. INDMOM α vs. STR α vs. CSFM

Panel B: CSFM Relative Performance

14

12

10

6
Annual %

–2

–4

–6
E[R] α vs. UMD α vs. INDMOM α vs. STR α vs. TSFM

1-1 1-3 1-6 1-12 1-36 1-60 2-12 13-60

Quantitative Special Issue 2019 The Journal of Portfolio Management   31

Exhibit 16
Relative Performance of UMD, INDMOM, and STR (global ex. US)
Panel A: UMD
10

6
Annual %

–2

–4
α vs. TSFM α vs. CSFM

Panel B: INDMOM
8

4
Annual %

–2

–4

–6
α vs. TSFM α vs. CSFM

Panel C: STR
8

4
Annual %

–2

–4
α vs. TSFM α vs. CSFM

1-1 1-3 1-6 1-12 1-36 1-60 2-12 13-60

32   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit 17
Ex Post Tangency Portfolios (global ex. US)

7DQJHQF\3RUWIROLR:HLJKWV
,QGLY
)DFWRU 6KDUSH       
76)0    
76)0    
76)0    
76)0    
&6)0  
&6)0  
&6)0  ±
&6)0  ±
80'     
675 ±  
0.7     
60%    ± 
+0/  ± ±  ±
50:     
&0$     
6KDUSH       

Note: * and ** signify that a weight estimate is significantly different from zero at the 5% or 1% level, respectively.

factor weights over time, rather than from taking static
long (short) bets on factors that have higher (lower)
average returns unconditionally. We also find that the
performance of factor momentum is not dependent on
using dozens of fine-grained factors. Instead, with a set
of only six broad theme factors,10  we reproduce the same
basic factor momentum phenomenon found in the 65
factor dataset.
CONCLUSION
We document robust persistence in the returns
of equity factor portfolios. This persistence is exploit-
able with a time-series momentum trading strategy
that scales factor exposures up and down in propor-
tion to their recent performance. Factor timing in this
manner produces economically and statistically large
excess performance relative to untimed factors. We
aggregate individual factor timing strategies into a
combined time-series factor momentum strategy that
dominates all individual timing strategies. TSFM is
10
 The six theme factors are valuation, momentum, earnings
quality, sustainable growth, management, and risk. Each theme
aggregates a set of closely related subfactors—for example, valuation
includes book-to-market, earnings-to-price, and dividend yield.
complementary with stock momentum, as both enter
optimized multifactor portfolios with signif icant
positive weights (particularly when combined with
HML-Devil).
An interesting aspect of factor momentum is its
stability with respect to the definition of recent per-
formance. Whether the look-back window is as short
as one month or as long as five years, our strategy
identifies large positive momentum among factors.
This contrasts sharply with stock momentum, which
exhibits reversal with respect to recent one-month
performance, momentum at intermediate horizons of
around one year and again reversal for windows beyond
two years.
Factor momentum is a truly global phenom-
enon. It manifests equally strongly outside the United
States, both in a large global (excluding the United
States) sample and f iner Europe and Pacif ic region
subsamples.
Taken alongside the evidence of time series
momentum in commodity, bond, and currency factors
(Moskowitz, Ooi, and Pedersen 2012), our findings of
momentum among equity factors—in the time series,
in the cross section, and around the world—support the
conclusion that factor momentum is a pervasive phe-
nomenon in financial markets.

Quantitative Special Issue 2019 The Journal of Portfolio Management   33

Appendix
Exhibit A1
Factor List

$EEUHY 'HVFULSWLRQ $XWKRUV -RXUQDO <HDU

$%1,19 $EQRUPDOFDSLWDOLQYHVWPHQW 7LWPDQ:HLDQG;LH -)4$ 
$&& $FFUXDOV 6ORDQ $5 
$'09 $GYHUWLVHPHQWH[SHQVHGLYLGHGE\PDUNHWYDOXH &KDQ/DNRQLVKRNDQG6RXJLDQQLV -) 
$'9(57&+* &KDQJHLQH[SHQGLWXUHVRQDGYHUWLVLQJ &KHPPDQXUDQG<DQ -)( 
$,0 $PLKXGLOOLTXLGLW\PHDVXUH $PLKXG -)0 
$66(7* <HDURQ\HDUSHUFHQWDJHFKDQJHLQWRWDODVVHWV &RRSHU*XOHQDQG6FKLOO -) 
$72 $VVHWWXUQRYHU 6ROLPDQ $5 
$72&+* &KDQJHLQDVVHWWXUQRYHU 6ROLPDQ $5 
%$% 0DUNHWEHWD )UD]]LQLDQG3HGHUVHQ -)( 
%0 %RRNWRPDUNHW 6WDWPDQ &0%$ 
&$372 &DSLWDOWXUQRYHU +DXJHQDQG%DNHU -)( 
&$6+ &DVKKROGLQJ 3DOD]]R -)( 
&(*7+ &DSLWDOH[SHQGLWXUHJURZWK $QGHUVRQDQG*DUFLD)HLMRR -) 
&3 &DVKIORZWRSULFH /DNRQLVKRN6KOLHIHUDQG9LVKQ\ -) 
'&$3 'HEWFDSDFLW\ +DKQDQG/HH -) 
'(/$< 'HOD\LQDVWRFNSULFH¶VUHVSRQVHWRLQIRUPDWLRQ +RXDQG0RVNRZLW] 5)6 
'3 'LYLGHQG\LHOG /LW]HQEHUJHUDQG5DPDVZDP\ -) 
(&216 (DUQLQJVFRQVLVWHQF\ $OZDWKDLQDQL %$5 
(17%0 (QWHUSULVHFRPSRQHQWRIERRNWRPDUNHW 3HQPDQ5LFKDUGVRQDQG7XQD -$5 
(1708/7 (QWHUSULVHPXOWLSOH /RXJKUDQDQG:HOOPDQ -)4$ 
(3 (DUQLQJVWRSULFH %DVX -)( 
(36020 0RPHQWXPLQHDUQLQJVSHUVKDUH %DOODQG%URZQ -$5 
)5((&) &DVKIORZWRERRNYDOXHRIHTXLW\ )UH\EHUJHU1HXKLHUO:HEHU :3 
)6&25( )VFRUH 3LRWURVNL -$5 
*5352) *URVVSURILWDELOLW\ 1RY\0DU[ -)( 
*6 6DOHVJURZWK /DNRQLVKRN6KOLHIHUDQG9LVKQ\ -) 
,1'020 ,QGXVWU\PRPHQWXP 0RVNRZLW]DQG*ULQEODWW -) 
,6.(: ,GLRV\QFUDWLFVNHZQHVV %R\HU0LWWRQDQG9RUNLQN 5)6 
,668( /RJJURZWKRIDGMXVWHGVKDUHVRYHUPRQWKV 3RQWLIIDQG:RRGJDWH -) 
,668( /RJJURZWKRIDGMXVWHGVKDUHVRYHUPRQWKV 3RQWLIIDQG:RRGJDWH -) 
,9* ,QYHQWRU\JURZWK %HORDQG/LQ 5)6 
,92/ ,GLRV\QFUDWLFYRODWLOLW\ $QJ+RGULFN;LQJDQG=KDQJ -) 
/12$&+* *URZWKLQORQJWHUPQHWRSHUDWLQJDVVHWV )DLUILHOG:KLVHQDQWDQG<RKQ $5 
/75 /RQJWHUPUHWXUQUHYHUVDO 'H%RQGWDQG7KDOHU -) 
/9* /HYHUDJH %KDQGDUL -) 
/9*%0 /HYHUDJHFRPSRQHQWRIERRNWRPDUNHW 3HQPDQ5LFKDUGVRQDQG7XQD -$5 
0$;5(7 ([WUHPHVWRFNUHWXUQV %DOL&DNLFLDQG:KLWHODZ -)( 
020 0RPHQWXP -HJDGHHVKDQG7LWPDQ -) 
1&2$&+* 1RQFXUUHQWRSHUDWLQJDVVHWVFKDQJHV 6ROLPDQ $5 
12$ 1HWRSHUDWLQJDVVHWV +LUVKOHLIHU+RX7HRKDQG=KDQJ -$( 
1:&&+* 1HWZRUNLQJFDSLWDOFKDQJHV 6ROLPDQ $5 
2/ 2SHUDWLQJOHYHUDJH 1RY\0DU[ 52) 
26&25( 2VFRUH *ULIILQDQG/HPPRQ -) 
30 3URILWPDUJLQ 6ROLPDQ $5 
30&+* &KDQJHLQSURILWPDUJLQ 6ROLPDQ $5 

(continued)

34   Factor Momentum Everywhere Quantitative Special Issue 2019

Exhibit A1 (continued)
Factor List

$EEUHY 'HVFULSWLRQ $XWKRUV -RXUQDO <HDU

33$&+* &KDQJHVLQSURSHUW\SODQWDQGHTXLSPHQWGLYLGHGE\DVVHWV /\DQGUHV6XQDQG=KDQJ 5)6 
40- 3URILWDELOLW\ 6ROLPDQ $5 
5(5 5HDOHVWDWHKROGLQJVGLYLGHGE\SURSHUW\SODQWDQGHTXLSPHQW 7X]HO 5)6 
5(96853 5HYHQXHVXUSULVHV -HJDGHHVKDQG/LYQDW -$( 
52$ 5HWXUQRQDVVHWV &RRSHU*XOHQDQG6FKLOO -) 
52( 5HWXUQRQHTXLW\ +DXJHQDQG%DNHU -)( 
6($621$/ 5HWXUQVHDVRQDOLWLHV +HVWRQDQG6DGND -)( 
6+257,17 6KRUWLQWHUHVW $VTXLWK3DWKDNDQG5LWWHU -)( 
60% 0DUNHWHTXLW\ %DQ] -)( 
63 6DOHVWRSULFH /HZHOOHQ &)5 
675 6KRUWWHUPUHYHUVDO -HJDGHHVK -) 
686*5 6XVWDLQDEOHJURZWK /RFNZRRGDQG3URPEXWU -)5 
785129(5 6KDUHYROXPH 3RQWLIIDQG0DFOHDQ -)0 
92/0751 9ROXPHWUHQG +DXJHQDQG%DNHU -)( 
:+ ZHHNKLJK *HRUJHDQG+ZDQJ -) 
;),1 ([WHUQDOILQDQFLQJ %UDGVKDZ5LFKDUGVRQDQG6ORDQ -$( 
;5'$7 5HVHDUFKDQGGHYHORSPHQW5 'H[SHQGLWXUHGLYLGHGE\WRWDODVVHWV /L 5)6 
;5'09 5 'H[SHQGLWXUHGLYLGHGE\PDUNHWYDOXH &KDQ/DNRQLVKRNDQG6RXJLDQQLV -) 
;5'6 5 'H[SHQGLWXUHGLYLGHGE\VDOHV &KDQ/DNRQLVKRNDQG6RXJLDQQLV -) 
=6&25( =VFRUH 'LFKHY -) 

AR = The Accounting Review; BAR = The British Accounting Review; CFR = Critical Finance Review; CMBA = The Chicago MBA: A Journal
of Selected Papers; JAE = Journal of Accounting and Economics; JAR = Journal of Accounting Research; JF = The Journal of Finance; JFE = Journal
of Financial Economics; JFM = Journal of Financial Markets; JFQA = Journal of Financial and Quantitative Analysis; JFR = Journal of Financial
Research; ROF = Review of Finance; RFS = The Review of Financial Studies; WP = unpublished paper.

ACKNOWLEDGMENTS Asness, C., and A. Frazzini. 2013. “The Devil in HML’s

We thank Cliff Asness, Kobi Boudoukh, Andrea
Frazzini, Ronen Israel, and Toby Moskowitz for many helpful
suggestions throughout this project. We thank William Fu
for outstanding research assistance.
Details.” The Journal of Portfolio Management 39 (4): 49.
Asness, C., A. Frazzini, R. Israel, and T. Moskowitz. 2014.
“Fact, Fiction, and Momentum Investing.” The Journal of Port-
folio Management 40 (5): 75.

REFERENCES Asness, C. S., J. A. Friedman, R. J. Krail, and J. M. Liew.

Arnott, R. D., M. Clements, V. Kalesnik, and J. T.
Linnainmaa. “Factor Momentum.” Working paper, SSRN,
2018.
Asness, C. “Variables That Explain Stock Returns.” Ph.D.
thesis, University of Chicago, 1994.
——. “My Factor Philippic.” Working paper, SSRN, 2016a.
——. 2016b. “The Siren Song of Factor Timing aka ‘Smart
Beta Timing’ aka ‘Style Timing.’” The Journal of Portfolio
Management 42 (5): 1–6.
2000. “Style Timing: Value versus Growth.” The Journal of
Portfolio Management 26 (3): 50–60.
Asness, C. S., T. J. Moskowitz, and L. H. Pedersen. 2013.
“Value and Momentum Everywhere.” The Journal of Finance
68 (3): 929–985.
Asness, C. S., R. B. Porter, and R. L. Stevens. “Predicting
Stock Returns Using Industry-Relative Firm Characteris-
tics.” Working paper, SSRN, 2000.
Avramov, D., S. Cheng, A. Schreiber, and K. Shemer. “Scaling
Up Market Anomalies.” Working paper, 2016.

Quantitative Special Issue 2019 The Journal of Portfolio Management   35

Chaves, D. B. 2016. “Idiosyncratic Momentum: US and
International Evidence.” The Journal of Investing 25 (2): 64–76.
Daniel, K., and T.  J. Moskowitz. 2016. “Momentum Crashes.”
Journal of Financial Economics 122 (2): 221–247.
De Bondt, W. F., and R. Thaler. 1985. “Does the Stock
Market Overreact?” The Journal of Finance 40 (3): 793–805.
Fama, E. F., and K. R. French. 2016. “Dissecting Anomalies
with a Five-Factor Model.” The Review of Financial Studies
29 (1): 69–103.
Frazzini, A., R. Israel, and T. Moskowitz. “Trading Costs
of Asset Pricing Anomalies.” Working paper, AQR, 2015.
Geczy, C. C., and M. Samonov. 2016. “Two Centuries of
Price–Return Momentum.” Financial Analysts Journal 72 (5):
32–56.
Grundy, B. D., and J. S. M. Martin. 2001. “Understanding
the Nature of the Risks and the Source of the Rewards to
Momentum Investing.” The Review of Financial Studies 14 (1):
29–78.
Gu, S., B. T. Kelly, and D. Xiu. “Empirical Asset Pricing via
Machine Learning.” Working paper, Yale, 2018.
Harvey, C. R., Y. Liu, and H. Zhu. 2016. “... and the Cross-
Section of Expected Returns.” The Review of Financial Studies
29 (1): 5–68.
Jegadeesh, N. 1990. “Evidence of Predictable Behavior of
Security Returns.” The Journal of Finance 45 (3): 881–898.
36   Factor Momentum Everywhere Quantitative Special Issue 2019
Jegadeesh, N., and S. Titman. 1993. “Returns to Buying
Winners and Selling Losers: Implications for Stock Market
Efficiency.” The Journal of Finance 48 (1): 65–91.
Kelly, B., S. Pruitt, and Y. Su. “Characteristics Are Covari-
ances: A Unified Model of Risk and Return.” Discussion
paper, National Bureau of Economic Research, 2018.
Lewellen,  J. 2002. “Momentum and Autocorrelation in Stock
Returns.” The Review of Financial Studies 15 (2): 533–564.
McLean, R. D., and J. Pontiff. 2016. “Does Academic
Research Destroy Stock Return Predictability?” The Journal
of Finance 71 (1): 5–32.
Moskowitz, T. J., and M. Grinblatt. 1999. “Do Indus-
tries Explain Momentum?” The Journal of Finance 54 (4):
1249–1290.
Moskowitz, T. J., Y. H. Ooi, and L. H. Pedersen. 2012. “Time-
Series Momentum.” Journal of Financial Economics 104 (2):
228–250.
Disclaimer
AQR Capital Management is a global investment management firm, which
may or may not apply similar investment techniques or methods of analysis
as described herein. The views expressed here are those of the authors and
not necessarily those of AQR.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.
 DOES YOUR ENTIRE
i p rj o u r n a l s . c o m

COMPANY or DEPARTMENT
NEED ACCESS to
the JOURNALS?
We offer IP authenticated access to our
content which will allow your team to find
the research and insights they need.

FOR INFORMATION ON HOW TO CUSTOMIZE A PLAN TO MEET YOUR NEEDS AND BUDGET
CONTACT A REPRESENTATIVE at (646) 891-2154 or journals-sales@pageantmedia.com.

FIND US AT OUR NEW ADDRESS

IPRJOURNALS.COM

Jennifer Bender
is a senior managing
director at State Street
Global Advisors in Boston,
MA.
jennifer_bender@ssga.com
T homas Blackburn
is an assistant vice
president at State Street
Global Advisors in Boston,
MA.
tj_blackburn@ssga.com
X iaole Sun
is a vice president at State
Street Global Advisors in
Boston, MA.
xiaole_sun@ssga.com
38   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
Clash of the Titans: Factor
Portfolios versus Alternative
Weighting Schemes
Jennifer Bender, Thomas Blackburn, and Xiaole Sun
N
ew portfolio construction ideas
have proliferated rapidly in
the last decade. These include
numerous contestants, from
fundamental weighting to minimum vari-
ance to risk parity to maximum diversifi-
cation. But too much of a good thing has
created confusion for investors now at risk
of embracing methods without a full under-
standing of their potential limitations.
Fifty years ago, times were much sim-
pler. Investors sought to maximize wealth/
return while minimizing the volatility of
that return stream. The main question was
how to define this utility function. The
mean–variance framework (Markowitz
1959) emerged as the most prevalent model
for capturing investor preferences. These
return-maximizing investors sought sources
of return that could improve return while
reducing risk. Stock pickers looked at qualita-
tive aspects of individual companies, whereas
quantitative investors gravitated toward fac-
tors—for example, security characteristics
that were shown by the academics to be able
to explain the cross section of stock returns
across broad universes.
A curious thing happened in the
mid-2000s. Investors, possibly tired of
boom–bust equity market cycles, started to
challenge the idea of market-cap-weighted
benchmarks. One of the first challengers
was fundamental indexing (Arnott, Hsu, and
Moore 2005), which weights securities by
fundamentals such as book value or earn-
ings instead of by market cap. The intriguing
part was the fact that fundamental indexing
exhibited phenomenal performance in back-
tests against market-cap-weighting schemes.
From there, a cottage industry exploded,
including risk parity or equal risk contribution
(weighting based on risk contribution), not to
be confused with risk weighting (weighting
based on risk), equal weighting (weighting
based on the number of securities), diversity
weighting, and many others. Just to make
matters more confusing, some methods such
as minimum variance and maximum Sharpe
ratio were many times incorrectly labeled as
“new” weighting schemes when in fact they
were variants of the original mean–variance
paradigm.
The time is ripe for sanity to be restored.
It is well known that the theoretical founda-
tions for alternative weighting schemes are
weak. Supporters have pointed instead to the
empirical evidence, mainly how well these
weighting schemes perform in backtests.
Here, we isolate the return drivers behind the
most popular alternative weighting schemes
and show that they come from a handful of
well-known factors.
Beating the market is not easy and never
has been. Identifying robust systematic and
repeatable sources of return has always been
the objective of investors and should continue
to be. Investors should refocus on isolating sources of
return that are rewarded and building portfolios that
capture these sources in the most risk-efficient and cost-
efficient way. Separating the source of return from how
to build the portfolio is a critical distinction. Factor
portfolios do exactly that.
AN OLDIE BUT GOODIE: EXTENDING
THE MEAN–VARIANCE FRAMEWORK
TO FACTOR PORTFOLIOS
Ross (1976) was among the first to note that one
way to understand the returns to stocks was to model
them as a function of exposures to various factors, which
could be thought of as attributes relating to a set of
securities’ returns. Although a wide range of factors
has been proposed (macroeconomic factors, statistical
factors, technical factors, and fundamental factors, to
name a few), the most widely cited today are those pro-
duced by the seminal research of Fama and French (1992,
1993)—value, size—and extended by Carhart (1997) to
momentum.
The term factor portfolios, or factor investing, is rela-
tively newer, born of the growth of smart beta in the last
decade. Smart beta started with simple rules-based ways
to capture factors. In fact, fundamental indexing is often
viewed as one of the first examples of smart beta. First-
generation smart beta used simple rules-based portfolio
construction because investors preferred their simplicity
and ease of understanding. (One of the benefits of the
rules-based approach is a direct link between each secu-
rity’s factor profile and weight. Rules-based portfolios
overweight stocks that rank highly on the factor and
underweight stocks that rank low on the factor.) For our
own research on smart beta rules-based strategies, see
Bender, Brandhorst, and Wang (2014) and Bender and
Wang (2015) for example.
We have long recognized the limitations to
capturing factors in rules-based portfolios. First, one
can take risk into account by adjusting all factor scores
by security volatilities, but that only takes into account
individual security volatility and not correlations
between securities. Second, one can control country
and sector active weights by establishing limits and redis-
tributing excess weights pro rata back to securities such
that the limits are met; however, this involves increased
computational complexity. Third, turnover can only be
mitigated using simple buffer rules, but almost always at
Quantitative Special Issue 2019
the expense of factor exposure. As investors increasingly
recognize these limitations, this opens the door to opti-
mization-based methods that can balance competing
objectives.1
Our preferred algorithmic framework is based on
mean–variance optimization. The framework has been
an asset allocation workhorse for decades. Mean–vari-
ance assumes investors care only about the first two
moments of portfolio return—return and risk. (Mean–
variance itself was a special case of Markowitz’s 1952
original expected utility maximization framework, it
being consistent with the latter under the assumption
that either investors have quadratic utility or returns are
normally distributed).
First-generation quant models, as by Grinold and
Kahn (1994), in fact used a mean–variance framework
with factors, much the same as in factor portfolios today.
Thus, we are not claiming to invent anything new; we
are merely reviving a well-known framework in the
context of harnessing factor premiums.
We define an objective function that maximizes
factor exposure scaled by risk aversion. That is, the
portfolio objective is to maximize exposure to intended
factor(s) in a risk-aware way. As first proposed by Bender,
Sun, and Wang (2016), a factor portfolio should seek to
deliver as much exposure to the targeted factor(s) as pos-
sible while minimizing risk and ensuring the portfolio is
sufficiently liquid and not too expensive to trade.
The traditional return–risk objective function can
be summarized as
Max w ′r − λw ′ ∑ w (1)
in which w is the vector of optimal weights, r is the
vector of expected returns for all securities, l is the risk
aversion, and S is the covariance matrix for all securities.
Applying this framework to factor models is a natural
extension because factor investing rests on the belief
1
 Comparing rules-based portfolios to optimization-based
portfolios is, in our view, comparing apples to oranges. Rules-
based portfolios have the benefit of being more transparent in how
a security’s factor characteristics translate into its weight in the
portfolio. For many investors, that transparency is desirable from a
governance and attribution perspective. Optimization-based port-
folios, however, are more efficient at balancing multiple, often-
competing objectives. To most efficiently maximize factor exposure
while minimizing risk and controlling for other objectives such
as concentration and turnover, the latter is a preferable approach.
The Journal of Portfolio Management   39
that factors are the source of return. The framework
then becomes
Max w ′X − λw ′ ∑ w (2)
in which, instead of expected return, we have X as the
vector of factor exposures for all securities. Written more
compactly, the objective is to
Maximize f –λ ⋅ ω (3)
in which f is the portfolio’s composite factor exposure
(individual exposures to all factors are averaged), l is
a risk aversion parameter, and w is the portfolio’s risk
(which can be active or total risk depending on whether
the investor wants to maximize exposure relative to a
benchmark).
The risk aversion parameter is calibrated based on
a desired risk level. The only constraints we impose are
that the portfolio cannot short securities, that is, must be
long only and must be fully invested (no cash allowed).
The online supplement discusses in more detail how we
define the objective function. In practice (not in this
article), we apply additional investability and risk-miti-
gating constraints such as constraints on sectors, coun-
tries, and turnover. We observe no loss of information
in these fully implementable portfolios.
This framework relates to pure factor premiums in
an important way. In theory, it is well understood how
to capture pure factor premiums (assuming the factors
have been specified). A pure factor portfolio has unit
exposure to the desired factor, zero exposure to all other
factors, and minimum risk. Such a pure factor portfolio
will have the same return as a factor return estimated in
a cross-sectional model:
f = ( X ′∆ −1X )−1 X ′∆ −1r (4)
in which stock returns are modeled as r = Xf + e, X is
the exposure matrix of securities to all factors, and D is
the diagonal-specific (idiosyncratic) risk matrix.
It turns out that the solution to a specific instance
of the quadratic objective function solved using opti-
mization (specifically, we minimize risk while holding
exposure constant at 1) under certain conditions is math-
ematically equivalent to the pure factor risk premiums as
defined by the practitioners (estimated using generalized
40   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
least squares cross-sectional regression). (The proof is
available from us upon request.)
The conditions for the pure factors to be equiva-
lent to the solution to our optimization function are as
follows:
• The factors must be orthogonal.
• There is no long-only constraint or any other
constraints.
Of course, these conditions are not practical in the
real world. To create an investable long-only version
of the pure factor portfolio, the goal is to build a port-
folio that keeps high exposure to the desired factor and
to minimize unwanted risk (while keeping transaction
costs, liquidity, and other implementation considerations
in mind). Importantly, the exposures to other factors
will no longer be zero, but they can be set such that, at
a given level of desired factor exposure (for the targeted
factor), the portfolio variance is minimized. By maxi-
mizing factor exposure while minimizing risk, we retain
the premium coming from the factor while minimizing
unrelated sources of risk. Thus, as much as possible of
the portfolio variance will be due to the factor exposure.
One last point to mention is the relationship
between the objective function in Equation 3 and max-
imum Sharpe ratio and minimum variance portfolios.
At any point in time, the solution to an unconstrained
mean–variance optimization problem is a combina-
tion of the highest-Sharpe-ratio portfolio, with either
the risk-free asset or the minimum-variance portfolio
(depending on whether the portfolio is required to be
fully invested). However, this is only the case if con-
straints are convex, and it is only a single point in time.
THE MEDLEY OF ALTERNATIVE WEIGHTING
SCHEMES
Alternatively weighted portfolios are portfolios that
are constructed using some set of rules or algorithm with
diversification or risk as its focus. What they all have in
common is a lack of need for an explicit alpha source.
Previous papers have provided helpful taxonomies.
Amenc et al. (2014) divided schemes based on scien-
tific diversification (i.e., minimum variance, maximum
Sharpe ratio) from those based on naïve diversification
(i.e., equal weighted, equal risk contribution). Chow
et al. (2011) similarly labeled them “optimization-based
Exhibit 1
Various Weighting Schemes Explored in Past Papers

6WUDWHJ\ 2EMHFWLYH $VVXPSWLRQVIRU0HDQ±9DULDQFH2SWLPDOLW\

0LQLPXP9DULDQFH3RUWIROLR 0LQσ  ZƍΩ Z

S
([SHFWHGUHWXUQVDUHHTXDO
0D[LPXP'HFRUUHODWLRQ3RUWIROLR 0LQσS ZƍΩ Z ([SHFWHGUHWXUQVDQGYRODWLOLWLHVDUHHTXDO
Zƍσ
0D[LPXP'LYHUVLILFDWLRQ 0D['S  ([SHFWHGUHWXUQVDUHSURSRUWLRQDOWRYRODWLOLWLHV
√ZƍΩ Z
3RUWIROLR
∂σS ∂σS
5LVN3DULW\3RUWIROLR(TXDO5LVN ZL  ZMIRUDQ\LM ([SHFWHGUHWXUQVDUHSURSRUWLRQDOWRYRODWLOLWLHV
∂ZL ∂ZM
&RQWULEXWLRQ FRUUHODWLRQVPXVWEHFRQVWDQW
0D[LPXP'HFRQFHQWUDWLRQ 0D[LPL]HVZƍZ ([SHFWHGUHWXUQVDQGYRODWLOLWLHVDUHHTXDOFRUUHODWLRQV
DUHFRQVWDQW
ZƍU
0D[LPXP6KDUSH5DWLR 0D[65S  1RDVVXPSWLRQVQHHGHGKRZHYHUH[SHFWHGUHWXUQV
√ZƍΩ Z
PXVWEHVSHFLILHGHLWKHUDVDQDGGLWLRQDOLQSXWRU
DVVXPHGWREHSURSRUWLRQDOWRULVN

Note: *EDHEC’s maximum Sharpe ratio strategy uses semi-deviation or downside risk as estimates for expected returns.

weighting schemes” versus “heuristic-based weighting
schemes.” In that vein, we list popular rules-based
(heuristic) weighting schemes:
• Fundamental indexing: Weighting by fundamen-
tals; see Arnott, Hsu, and Moore (2005)
• Equal weighting: All securities are assigned the
same weight; see DeMiguel, Garlappi, and Uppal
(2009)
• Diversity weighting: Weights of securities are based
on market diversity, measured as the level of con-
centration of capital in the market; see Fernholz,
Garvy, and Hannon (1998)
Apart from equal weighting, the heuristic
weighting schemes require an expected return source
to be a special case of mean–variance optimality.
Optimization-based portfolio construction
schemes include
• Minimum variance: Weights are selected to min-
imize the variance of the portfolio; see Clarke,
de Silva, and Thorley (2011)
• Maximum decorrelation: Weights are selected to
minimize the variance of the portfolio, assuming
securities all have the same volatility (a special case
of minimum variance); see Christoffersen et al.
(2012) and Amenc, Goltz, and Martellini (2013)
• Maximum diversification: Weights of securities
are selected to maximize the ratio of weighted-
average asset volatilities to portfolio volatility; see
Choueifaty and Coignard (2008)
• Risk parity/equal risk contribution: Weights are
based on contribution to portfolio risk such that
each asset has the same contribution; see Qian
(2011); Maillard, Roncalli, and Telietche (2010);
and Carvalho, Lu, and Moulin (2012)
• Maximum deconcentration: A modified form
of equal weighting that maximizes the effective
number of stocks; see DeMiguel, Garlappi, and
Uppal (2009) and Amenc, Goltz, and Martellini
(2013)
• Maximum Sharpe ratio: Weights are selected to
maximize the Sharpe ratio of the portfolio where,
in practice, the forecast returns are often assumed
to be proportional to security volatilities; see Tobin
(1958)
Exhibit 1 summarizes the optimization-based
weighting schemes.
All of the weighting schemes shown in Exhibit 1
can be equivalent to a mean–variance optimal portfolio
given a concrete set of assumptions. These are shown

Quantitative Special Issue 2019 The Journal of Portfolio Management   41

in the far right column of Exhibit 1. For instance, a
minimum variance portfolio is mean–variance optimal
as long as expected returns across all assets are the same.
An equal-weighted portfolio is mean–variance optimal
as long as expected returns and volatilities across all assets
are the same, and correlations between all assets are the
same. A risk parity portfolio is mean–variance optimal
as long as all assets have the same Sharpe ratios and the
correlations are constant. Some of these assumptions
are clearly more restrictive (relative to mean–variance)
than others; minimum variance and maximum Sharpe
ratio have the least restrictive assumptions, and equal
weighting and maximum decorrelation have the most
restrictive.
If these weighting schemes’ assumptions are so
unlikely to hold in real life, what explains their popu-
larity? Although mean–variance is strong on theory,
implementing it correctly is not easy. The most common
criticism of optimization is its sensitivity to inputs;
future returns, volatilities, and correlations all need
to be estimated. Critics of mean–variance optimiza-
tion have pointed out potential errors in estimation
(Michaud and Michaud 2007). Furthermore, parameters
such as risk aversion and various constraints must be
calibrated (Kan and Zhou 2007; Zhou 2008). A second
criticism of optimization is that, as constraints become
more restrictive, the optimality of the optimized solu-
tion becomes less clear cut. Third, proponents of risk-
based weighting schemes such as risk parity point to past
research showing risk estimation to be more robust than
return forecasts (Merton 1980). Alternative weighting
schemes have become popular because they avoid some
or all of the challenges of implementing mean–variance
optimization.
Supporters of alternative weighting schemes also
argue they have strong intuitive appeal. For instance,
not only does equal weighting not require any esti-
mates of returns, volatility, or correlations, it is easy
to understand: One simply weights every asset equally,
resulting in a diverse, nonconcentrated portfolio. Risk
parity advocates point to the attractive appeal of all assets
contributing equally to the risk of the portfolio. What
could be more appealing than that?
We agree that the intuitive appeal is important;
however, intuitive appeal should be complemented by
theoretical robustness. We believe the mean–variance-
based approach is stronger on theory, and its goal of
maximizing risk-adjusted exposure to targeted factor(s)
42   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
is just as intuitive. Moreover, we do not believe the
implementation challenges to optimization obviate the
general value of using optimization. Errors in estimation
can be addressed, and sensible parameters can be cali-
brated. Optimization is, after all, employed in a broad
range of applications in asset management.
AN EMPIRICAL COMPARISON
Next we compare a multifactor mean–variance-
based approach to a range of alternative weighting
schemes. The mean–variance-based approach targets
the Fama–French factors—value, momentum, low size
(small cap), investments, and profitability—plus a sixth
factor, low volatility (i.e., 60-month historical vari-
ance). These factors have well-established factor returns
that have existed over time and different investment
universes. The factors also continue to persist since
discovery. In addition to historically producing excess
returns, these factors have low correlations with each
other, benefiting the portfolio during various parts of
the business cycle.
We evaluate a US large cap (S&P 500 Index con-
stituent) universe and assess backtests over the period
January 1995 to December 2016. Our backtests begin in
January 1995 based on the availability of data we have
for fundamental data.
How does the mean–variance-based proposed
portfolio perform compared to other weighting schemes?
The performance of the multifactor strategy is shown
in Exhibit 2 over the backtest period of January 1995 to
December 2016.
The key metric to focus on in the exhibit is the
information ratio, which ref lects how much incremental
return we achieve per unit of risk. Excess returns alone
are misleading because most of these weighting schemes
can be calibrated for different levels of risk appetite. The
mean–variance-based framework delivers the highest
information ratio at 0.48 over this 20-year period. Only
the maximum diversification and maximum decorrela-
tion portfolios underperform the market-cap-weighted
benchmark.
Next, we examine the factor exposures (loadings) of
the various weighting schemes. Our mean–variance-based
framework should de facto have high factor exposures on
average because the objective function maximizes these
exposures per unit of risk. (Depending on the correlation
between factors, some factors may be offsetting, so we do
Exhibit 2
Performance Summary (USD gross returns, January 1995 to December 2016, results using three-factor principal
component analysis model, S&P 500 Universe)

0D[LPXP 5LVN 0HDQ± %PN

0D[LPXP 6KDUSH 3DULW\ 0LQLPXP 0D[LPXP 0D[LPXP 9DULDQFH &DS
'LYHUVLILFDWLRQ 5DWLR (5& 9DULDQFH 'HFRQFHQWUDWLRQ 'HFRUUHODWLRQ %DVHG3RUWIROLR :WG
$QQXDOL]HG5HWXUQ        
$QQXDOL]HG9RODWLOLW\        
6KDUSH5DWLR        
$FWLYH5HWXUQ ±     ±  
7UDFNLQJ(UURU        
,QIRUPDWLRQ5DWLR ±     ±  ±
%HVW0RQWK5HWXUQ        
:RUVW0RQWK5HWXUQ ± ± ± ± ± ± ± ±
8S&DSWXUH        
'RZQ&DSWXUH        
0D[LPXP'UDZGRZQ ± ± ± ± ± ± ± ±

Source: State Street Global Advisors (SSGA).

Exhibit 3
Comparison of Factor Exposures Across Approaches (sample period January 1995 to December 2016; results
using three-factor principal component analysis model, Fama–French Factors plus BAB)

Maximum Risk Mean–

Maximum Sharpe Parity/ Minimum Maximum Maximum Variance-
Diversification Ratio ERC Variance Deconcentration Decorrelation Based Portfolio
Intercept 0.02% 0.07% 0.03% 0.13% 0.02% 0.05% 0.24%
Market 0.70 0.72 0.91 0.59 1.06 0.76 0.92
SMB 0.02 0.06 0.16 –0.06 0.27 0.10 0.27
HML 0.05 0.07 0.07 –0.02 0.10 0.07 0.25
RMW –0.03 0.00 0.10 –0.02 0.10 –0.10 0.06
CMA 0.24 0.25 0.17 0.32 0.13 0.10 0.17
WML –0.04 –0.10 –0.11 –0.02 –0.17 –0.06 –0.09
BAB 0.15 0.15 0.10 0.18 0.05 0.08 0.08

Notes: BAB = betting against beta; CMA = conservative minus aggressive (investment); HML = high minus low (value); RMW = robust minus weak
(profitability); SMB = small minus big; WML = momentum.
Statistically significant loadings adjusted for Newey-West are in bold. Results adjusted for both Newey-West and Bonferroni corrections are qualitatively
similar and available upon request.
Sources: SSGA; Kenneth French and AQR data libraries.

not expect all six factors to have equally high exposures
individually; however, the sum/average of the exposures
should be high.) What is of interest is whether any of the
other weighting schemes, none of which are designed
to have high factor loadings, do in fact. Exhibit 3 shows
factor loadings using the traditional Fama–French defini-
tions plus the betting against beta (BAB) (low volatility)
factor identified by Frazzini and Pedersen (2011). All
Fama–French factor returns are sourced from Kenneth
French’s website (http://mba.tuck.dartmouth.edu/pages/
faculty/ken.french/data_library.html). The BAB factor
returns are sourced from AQR’s website (https://www.
aqr.com/Insights/Datasets). For consistent comparison to
the universe used for the strategies, the market factor we
use is the return to the cap-weighted universe composed
of the S&P 500 Index constituents.

Quantitative Special Issue 2019 The Journal of Portfolio Management   43

Exhibit 4
Information Ratio by Subperiod (USD gross returns, January 1995 to December 2016; results using three-factor
principal component analysis model, S&P 500 Universe)

0HDQ±
0D[LPXP 5LVN 9DULDQFH %PN
0D[LPXP 6KDUSH 3DULW\ 0LQLPXP 0D[LPXP 0D[LPXP %DVHG)DFWRU &DS
'LYHUVLILFDWLRQ 5DWLR (5& 9DULDQFH 'HFRQFHQWUDWLRQ 'HFRUUHODWLRQ 3RUWIROLR :WG
$QQXDOL]HG5HWXUQ
±        
±        ±
±      ± ± 
±        
$QQXDOL]HG([FHVV5HWXUQ
± ± ± ± ± ± ± ±
±       
±      ± ±
± ± ±  ±  ± 
7UDFNLQJ(UURU
±       
±       
±       
±       
,QIRUPDWLRQ5DWLR
± ± ± ± ± ± ± ±
±       
±      ± ±
± ± ±  ±  ± 

Source: SSGA.

The intercepts, or unexplained returns, are low,
and none are statistically significant. All weighting
schemes except for maximum decorrelation have sig-
nificant exposures (of the right sign) to at least two of the
nonmarket factors. This indicates that these weighting
schemes are indirectly deriving performance from some
of the factors. Risk parity and maximum deconcentra-
tion stand out as having statistically significant exposures
to many (five or greater) factors. Maximum decorrela-
tion is the only weighting scheme that does not have
any significant exposures, though the magnitude of the
exposures to SMB, CMA, and BAB is positive and not
trivial.
The results overall are striking because none of
the weighting schemes are designed to capture fac-
tors either implicitly or explicitly. Minimum variance
portfolios are not factor portfolios. Their purpose is to
minimize risk. Maximum diversification is a spin on
minimum variance, using again volatilities and corre-
lations, not factor exposures. Maximum Sharpe ratio
portfolios conceptually could be a special case of our
framework, as long as the expected returns (numer-
ator) are factor exposures. However, here we follow
the route of Chow et al. (2011) and Qian, Alonso, and
Barnes (2015) for the maximum Sharpe portfolios, so
the expected returns to securities have nothing to do
with factor exposures but are instead proportional to
their semi-volatilities.
To understand the robustness of these results, it is
important to look at how sensitive they are to
• Different time periods
• Different specifications for the covariance matrix
• Different sets of explanatory factors
Sensitivity of Results to Time Periods
Exhibit 4 shows the performance by subperiod.
There is in fact a sizable amount of variation across
periods. The mean–variance portfolio does not have

44   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
Exhibit 5
Probability of Outperforming the Benchmark in Any Three-Year Period (USD gross returns, January 1995 to
December 2016; results using three-factor principal component analysis model, S&P 500 Universe)
90.0%
79.0%
Probability of Outperforming

80.0% 74.7%
68.1%
in any 3-year Period

70.0%
60.0% 53.7% 56.3%
51.5%
50.0% 46.3%
40.0%
30.0%
20.0%
10.0%
0.0%

Risk Parity/ERC
Maximum Sharpe Ratio

Mean–Variance Based Portfolio

Minimum Variance
Maximum Diversification

Maximum Deconcetration

Maximum Decorrelation
Source: SSGA.

the highest information ratio in all subperiods, but than to build the most robust possible mean–variance
taken across all periods, it is among the most consistent. portfolio.
During the first period (1995–1999), the mean–variance Exhibit 5 plots the probability of outperforming
portfolio has the highest (least negative) information the market-cap-weighted portfolio in any three-year
ratio. During the second period (2000–2004), the equal period, another way to assess consistency in perfor-
risk contribution and maximum deconcentration port- mance. (We note that overlapping periods are used here,
folios exhibited higher information ratios; however, the so we do not assign any statistical significance to the
mean–variance portfolio is not far behind. The third estimates). They are, however, good relative indicators.
period (2005–2009) is the one period in which the The mean–variance-based framework does not in fact
mean–variance portfolio has the weakest information have the highest probability of outperformance; the risk
ratio, and the last period (2010–2016) is again dominated parity portfolio is the highest.
by the mean–variance portfolio. We acknowledge that
there is enough intertemporal variation to warrant some Sensitivity of Results to the Covariance
caution in extrapolating that the mean–variance port- Matrix Specification
folio will always be the best performer. The weakness
in 2005–2009 is in part driven by the mean–variance The covariance matrix used in the results so far is
portfolio having higher concentration (lower effec- constructed using principal component analysis with 60
tive number of names) than the alternative weighting months of returns. Prior papers use different approaches
schemes. That said, none of the alternative weighting in generating the covariance matrix. This is similar to
schemes consistently exhibit the highest information Amenc et al. (2012), though they estimated principal
ratio in all periods. Moreover, our aim in this article component analysis with two years of weekly returns.
is to compare the drivers of returns across representa- Qian, Alonso, and Barnes (2015) employed five years of
tive portfolios using different weighting schemes rather historical monthly returns with exponentially decayed

Quantitative Special Issue 2019 The Journal of Portfolio Management   45

weights. Clarke, de Silva, and Thorley (2013) used a
single-factor risk model in which the market is the risk
factor. Techniques can also be used to mitigate estima-
tion error of the covariance matrix, such as employing
a shrinkage estimator as done by Chow et al. (2011). As
robustness checks, we rerun our results using different
specifications for the covariance matrix, including one
in which we use equally weight historical data and one
in which we more heavily weight recent observations.
We also test a version with a covariance matrix using
a shrinkage estimator. Our results remain qualitatively
the same, though the statistical significance of some of
the factor exposures does change. All results are avail-
able upon request.
Sensitivity of Results
to the Explanatory Factors
How sensitive are the results to the choice of fac-
tors? The explanatory power of all factors taken together,
as well as the magnitude and statistical significance of
the intercepts, are quite sensitive to the factor combina-
tions. The online appendix shows the results without
the BAB factor. There, the intercept terms across all
weighting schemes are somewhat larger and more posi-
tive but remain insignificant statistically. The exposure
results stay the same, however. The alternative weighting
schemes all have meaningful exposures to these factors.
PAST STUDIES COMPARING ALTERNATIVE
WEIGHTING SCHEMES
How do our results compare to past studies? Chow
et al. (2011); Clarke, de Silva, and Thorley (2013); and
Qian, Alonso, and Barnes (2015) conducted empirical
tests similar to ours. They reached varying conclusions.
It is worth putting our results in the context of these
prior findings.
Chow et al. (2011) compared seven portfolio con-
struction techniques, four of them overlapping with
ours: equal weighting, risk-clustered equal weighting,
diversity weighting, fundamental weighting, minimum
variance, maximum Sharpe, and maximum diversifi-
cation.2 They evaluated US and global data; their uni-
verses were (1) the largest 1,000 stocks in the United
2
 In their nomenclature, equal weighting, diversity weighting,
and fundamental weighting are heuristic-based weighting method-
46   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
States over the 1964–2009 time period and (2) the
largest 1,000 global stocks using the Datastream uni-
verse over the 1987–2009 period. They constructed the
portfolios using a risk model based on 60-month secu-
rity returns (with a shrinkage estimator) and measured
absolute and relative performance and four-factor risk-
adjusted performance. They found that all seven strate-
gies outperform because of their exposure to value and
size factors. According to the authors, “Almost entirely
spanned by market, value, and size factors, any one of
these strategies can be mimicked by combinations of
the others.” They found that none of the portfolios had
statistically significant excess returns over the market-
cap-weighted benchmark once they controlled for the
Fama–French factors and momentum. Although we use
a US universe and a shorter time period, and they did
not include the BAB factor, our results are consistent
with their paper.
We do, however, disagree with the findings of
Chow et al. (2011) that there is no meaningful improve-
ment in using optimization-based methods. There,
[the authors find] that mean–variance optimi-
zation, which is required for a number of the
alternative betas, does not appear to result in a
meaningful diversification improvement over
non-optimized portfolios, despite the added
complexity. Investment insights about the rela-
tionship between a stock’s expected excess return
and volatility, or downside semi-volatility, do
not appear to produce performance benefits that
are otherwise not present in other simple port-
folio heuristics, which are derived from equal
weighting.
The mean–variance-based framework is quite dif-
ferent from the optimization schemes analyzed by Chow
et al. (2011) because it explicitly seeks to harness returns
from the factors. With respect to risk efficiency and
liquidity, it is also superior precisely because it balances
factor exposure with risk and liquidity.
A second prior paper, by Clarke, de Silva, and
Thorley (2013), focused on three weighting schemes,
all optimization based: minimum variance, maximum
diversification, and risk parity. They employed US data
ologies. Meanwhile, minimum variance, maximum Sharpe, and
maximum diversification are optimization-based methodologies.
(largest 1,000 stocks) over the 1968–2012 period and
used a single-factor risk model for the covariance matrix.
They found that, although all three outperform market
cap weighting, minimum variance and risk parity are
superior to maximum diversification based on risk-
adjusted returns. This is consistent with our findings.
Lastly, Qian, Alonso, and Barnes (2015) provided
the most recent contribution to the literature. They ana-
lyzed four different weighting schemes: equally weighted,
minimum variance, maximum diversification, and risk
parity. They called these four weighting schemes “naïve
beta” because all four naively assume that at least one
input (expected returns, volatilities, correlations) is the
same across all securities. For their empirical tests, they
used the 10 Global Industry Classification Standard S&P
500 sectors as the assets (in contrast to Chow et al. 2011
and Clarke, de Silva, and Thorley 2013 who used indi-
vidual securities as the assets). Qian, Alonso, and Barnes
(2015) found that, for all four weighting schemes, risk-
adjusted returns are higher than market cap weighting.
However, they argued that the risk parity approach is
the best of the three because the minimum variance and
maximum diversification approaches are highly sensi-
tive to risk inputs (volatilities and correlations) resulting
in concentrated portfolio holdings and high portfolio
turnover.
In contrast to Qian, Alonso, and Barnes (2015),
the framework we present here is not as sensitive to risk
inputs. The main difference is that in the Chow et al.
(2011) and Qian, Alonso, and Barnes (2015) papers, the
security risk estimates have a first-order impact on the
optimal portfolio. In the minimum variance optimiza-
tions, the volatility and correlation estimates are the
only source of information. In the maximum Sharpe
ratio optimizations, both papers assume a Sharpe ratio
at which the expected return is proportional to the risk
estimates; thus, again, the optimal portfolio is very sensi-
tive to the volatility estimates. In our framework, vola-
tilities and correlations are only one input. The other
critical input is the expected factor returns/exposures.
These balance out the risk estimates and overcome the
main criticisms of the two prior papers.
A SPECIAL NOTE ON RISK PARITY
One final point we want to highlight concerns risk
parity and equal risk contribution. Weighting schemes
that allocate to risk equally across assets can pertain to
Quantitative Special Issue 2019
asset classes, securities, or factors. Risk parity, as con-
ceptualized early on at Bridgewater and AQR, focused
on asset classes, specifically bonds versus equities. In this
article, we apply the notion of equal risk contribution
to securities—that is, each stock contributes the same
amount of risk to the final portfolio. This is similar to
Chow et al. (2011) and Badaoui and Lodh (2013). We
acknowledge that risk parity/ERC may be justified at
the asset class level and not at the securities level. For
instance, risk parity for asset classes has been shown to
be less sensitive to input parameters than other risk-based
allocation schemes; see Kaya and Lee (2012).
Separately, requiring equal risk from securities
is different from requiring each factor to contribute
the same amount of risk to the portfolio. We equally
weight factor scores in the objective function that would
follow if expected returns for each factor were equal. We
believe this is an appropriate assumption for an investor
without an explicit factor forecasting model. A port-
folio constructed by weighting factors proportional to
risk contribution is only optimal if factors are uncor-
related and if all factors have the same Sharpe ratio (i.e.,
each factor is rewarded in proportion to its own risk).
Empirically, the historical evidence does not support
this assumption. For instance, for single-factor portfolios
built with similar levels of tracking error, momentum,
and size generally have higher return than low volatility.
One point to note is that proponents of equal risk
weighting factors often cite the fact that factor returns
are hard to forecast, so weighting by risk makes sense.
We do not disagree, but the optimal solution in this
situation is to assume factor returns are equal, not that
factors are proportional to risk.
SUMMARY
Factors can be captured today in eff icient,
conceptually clean, cost-effective portfolios. Modern
computing has led to cheaper and easier ways to access
data, execute robust portfolio construction techniques,
and trade securities quickly and efficiently. We can now
meet the practical requirements of making pure factors
investable. Factor portfolios are very intuitive: They are
based on the belief that factors drive returns and risk and
that we can harness them easily.
Here we resurrect Markowitz’s mean–variance
framework for modern-day factor portfolios, which not
coincidentally are similar to the first-generation quant
The Journal of Portfolio Management   47
models. The benefits are that one targets the intended
sources of return and risk, minimizes the unintended
sources of return and risk, and controls turnover. Our
rationale behind this approach starts with the objective
for smart beta strategies, which is to capture exposure
to the desired factor(s) and to do so in a risk aware way.
Thus, maximizing factor exposure per unit risk with
various investability and risk constraints is precisely what
we seek in creating smart beta strategies. By maximizing
factor exposure while minimizing risk, we retain the
premium coming from the desired factors, while mini-
mizing unrelated sources of risk from other factors or
idiosyncratic risk. In this way, as much of the portfolio
variance will be due to the factor exposure as possible.
We compare the mean–variance framework against
a host of well-known alternative weighting schemes.
The industry has seen a f lood of research around alterna-
tive portfolio construction approaches. However, such
frameworks have arguably weak theoretical foundations
and are rationalized by a range of (very different) rea-
sons, most of them dissatisfying in our view. We show
that many popular alternative weighting schemes derive
most, if not all, of their outperformance from a handful
of well-known factors. A sensibly built factor portfolio
delivers a similar or higher information ratio by explic-
itly harnessing the factors and doing so in an efficient
risk- and transaction-cost-aware way.
ACKNOWLEDGMENTS
We thank Richard Hannam, Michael Feehily, Emiliano
Rabinovich, Ric Thomas, and Alex Rudin for their helpful
comments.
REFERENCES
Amenc, N., F. Goltz, A. Lodh, and L. Martellini. 2012.
“Diversifying the Diversifiers and Tracking the Tracking
Error: Outperforming Cap-Weighted Indices with Lim-
ited Risk of Underperformance.” The Journal of Portfolio
Management 38 (3): 72–88.
——. 2014. “Towards Smart Equity Factor Indices: Har-
vesting Risk Premia without Taking Unrewarded Risks.”
The Journal of Portfolio Management 40 (4): 106–122.
Amenc, N., F. Goltz, and L. Martellini. “Smart Beta 2.0.” June
2013, http://docs.edhec-risk.com/mrk/130515_SciBeta/
ERI_SciBeta_Smart_Beta_2.0.pdf.
48   Clash of the Titans: Factor Portfolios versus Alternative Weighting Schemes Quantitative Special Issue 2019
Arnott, R., J. Hsu, and P. Moore. 2005. “Fundamental Index-
ation.” Financial Analysts Journal 61 (2): 83–99.
Badaoui, S. and A. Lodh. “Scientific Beta Diversified Multi-
Strategy Index.” October 2013, http://docs.edhec-risk.com/
ERI-Days-North-America-2013/documents/SciBeta_
Diversified_Multi-Strategy_Index.pdf.
Bender, J., E. Brandhorst, and T. Wang. 2014. “The Latest
Wave of Advanced Beta.” The Journal of Index Investing 5 (1):
67–76.
Bender, J., X. Sun, and T. Wang. 2016. “A New Metric for
Smart Beta: Factor Exposure per Unit of Tracking Error.”
The Journal of Index Investing 7 (2): 109–118.
Bender, J., and T. Wang. 2015. “Tilted and Anti-Tilted Port-
folios: A Coherent Framework for Advanced Beta Portfolio
Construction.” The Journal of Index Investing 6 (1): 51–64.
Carhart, M. 1997. “On the Persistence in Mutual Fund Per-
formance.” The Journal of Finance 52 (1): 57–82.
Carvalho, R. L., X. Lu, and P. Moulin. 2012. “Demysti-
fying Equity Risk-Based Strategies: A Simple Alpha plus Beta
Description.” The Journal of Portfolio Management 38 (3): 56–70.
Choueifaty, Y., and Y. Coignard. 2008. “Toward Maximum
Diversification.” The Journal of Portfolio Management 35 (1):
40–51.
Chow, T., J. Hsu, V. Kalesnik, and B. Little. 2011. “A Survey
of Alternative Equity Index Strategies.” Financial Analysts
Journal 67 (5): 37–57.
Christoffersen, P., V. Errunza, K. Jacobs, and X. Jin. 2012. “Is
the Potential for International Diversification Disappearing?
A Dynamic Copula Approach.” Review of Financial Studies 25
(12): 3711–3751.
Clarke, R., H. de Silva, and S. Thorley. 2011. “Minimum
Variance Portfolio Decomposition.” The Journal of Portfolio
Management 37 (2): 31–45.
——. 2013. “Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective.” The Journal
of Portfolio Management 39 (3): 39–53.
DeMiguel, V., L. Garlappi, and R. Uppal. 2009. “Optimal
versus Naïve Diversification: How Inefficient Is the 1/N
Portfolio Strategy?” The Review of Financial Studies 22 (5):
1915–1953.
Fama, E. F., and K. R. French. 1992. “The Cross-Section
of Expected Stock Returns.” The Journal of Finance 47 (2):
427–465.
Fama, E. F., and K. R. French. 1993. “Common Risk Fac-
tors in the Returns on Stock and Bonds.” Journal of Financial
Economics 33 (1): 3–56.
Fernholz, R., R. Garvy, and J. Hannon. 1998. “Diversity
Weighted Indexing.” The Journal of Portfolio Management 24
(2): 74–82.
Frazzini, A., and L. Pedersen. 2011. “Betting against Beta.”
Journal of Financial Economics 111 (1): 1–25.
Grinold, R. C., and R. N. Kahn. Active Portfolio Manage-
ment: A Quantitative Approach for Providing Superior Returns and
Controlling Risk. New York: McGraw-Hill, 1994.
Kan, R., and G. Zhou. 2007. “Optimal Portfolio Choice with
Parameter Uncertainty.” The Journal of Financial and Quantita-
tive Analysis 42 (3): 621–656.
Kaya, H., and W. Lee. 2012. “Demystifying Risk
Parity.” March 2012, https://papers.ssrn.com/sol3/papers.
cfm?abstract_id=1987770.
Maillard, S., T. Roncalli, and J. Telietche. 2010. “The Prop-
erties of Equally-Weighted Risk Contribution Portfolios.”
The Journal of Portfolio Management 36 (4): 60–70.
Markowitz, H. 1952. “Portfolio Selection.” The Journal of
Finance 7: 77–91.
Quantitative Special Issue 2019
——. Portfolio Selection: Efficient Diversification of Investments.
New York: Jon Wiley & Sons, 1959.
Merton, R. 1980. “On Estimating the Expected Return on
the Market: An Exploratory Investigation.” Journal of Financial
Economics 8: 323–361.
Michaud, R., and R. Michaud. 2007. “Estimation Error and
Portfolio Optimization: A Resampling Solution.” Journal of
Investment Management 6 (1): 8–28.
Qian, E. 2011. “Risk Parity and Diversification.” The Journal
of Investing 20 (1): 119–127.
Qian, E., N. Alonso, and M. Barnes. 2015. “The Triumph
of Mediocrity: A Case Study of Naïve Beta.” The Journal of
Portfolio Management 41 (4): 19–34.
Ross, S. 1976. “The Arbitrage Theory of Capital Asset
Pricing.” Journal of Economic Theory 13 (3): 341–360.
Tobin, J. 1958. “Liquidity Preference as Behavior towards
Risk.” The Review of Economic Studies 25 (2): 65–86.
Zhou, G. 2008. “On the Fundamental Law of Active Port-
folio Management: What Happens If Our Estimates Are
Wrong?” The Journal of Portfolio Management 34 (4): 26–33.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.
The Journal of Portfolio Management   49

K ristin Fergis
is an associate at
BlackRock, Inc., in
New York, NY.
kristin.fergis@blackrock.com
K atelyn Gallagher
is a director at BlackRock,
Inc., in New York, NY.
katelyn.gallagher@blackrock.com
P hilip Hodges
is a managing director
at BlackRock, Inc., in
San Francisco, CA.
philip.hodges@blackrock.com
K ed Hogan
is a managing director
at BlackRock, Inc., in
San Francisco, CA.
ked.hogan@blackrock.com
50   Defensive Factor Timing Quantitative Special Issue 2019
Defensive Factor Timing
Kristin Fergis, Katelyn Gallagher,
Philip Hodges, and Ked Hogan
A
well-balanced strategic allocation
to factors, or asset classes, is the
foundation of a well-constructed
 long-term portfolio. However,
factor premiums vary over time. There
are three possible approaches that inves-
tors could take in response to time-varying
factor returns. The first approach is to ignore
the short-term variation—since Markowitz
(1952), we have built strategic allocations
to access sources of return that should be
rewarded in the long run. Second, investors
might develop a short-term prediction model
that aims to generate excess returns by tac-
tically changing positions around strategic
factor weights (e.g., Hodges et al. 2017). A
disadvantage of this approach is that it may
lead to short-term underperformance rela-
tive to the strategic allocation, and those risks
may increase with the frequency and size of
the tactical decisions. The third approach is
to practice defensive timing: to recognize
that during certain periods, it may be appro-
priate to reduce risks to certain factors or
across a portfolio with an aim to preserve
capital during bad market regimes.
In this article, we explore this third
approach to defensive factor timing. Defen-
sive factor timing focuses on mitigating the
magnitude of losses by periodically reducing
risk, as opposed to aiming to outper-
form a strategic benchmark by continually
over- or underweighting different factors.
Put another way, defensive timing focuses on
risk mitigation, which manages second and
higher moments, rather than on incremen-
tally increasing returns, which targets the
first moment. The risk reductions are taken
infrequently and only for those factors that
are very expensive or offer few diversifica-
tion benefits. During periods when aggre-
gate market risk tolerance is unusually low,
or correlations unusually high, it may also
be appropriate to reduce risk for the total
portfolio.
To illustrate the concept of defensive
factor timing, we work with a portfolio of
macro factor exposures in a multiasset context:
economic growth, real rates, inf lation, credit,
emerging markets, and liquidity. These fac-
tors are rewarded macro risk premiums, and
we define these factors so that, to the prac-
tical extent possible, they maximize exposure
to the underlying macroeconomic source of
risk. For example, nominal bonds are affected
by changes in both real rates and inf lation.
An inf lation factor portfolio goes long nom-
inal bonds and short real bonds (linkers) to
concentrate on only inf lation-related com-
ponents (or inflation breakevens). Both devel-
oped market and emerging market equity has
tended to perform strongly when the global
economy is growing, but certain risks present
in emerging market equities are either absent
or minimal in developed market equities (see,
among many others, Harvey 1995). Thus, an
emerging market factor would go long emerging market
equity and short developed market equity to focus on
the risks unique to emerging markets.
We use three quantitative measures to defensively
time factors: (1) a measure of aggregate risk tolerance,
which intuitively captures the effect of flight to quality, or
a rotation from risky to safer assets during times of stress;
(2) a diversification ratio, which captures a potential
temporary reduction in diversification benefits; and (3)
valuation ratios of each factor.
We discuss how defensive timing can be imple-
mented using these metrics in a systematic investing
framework and discuss case studies of risk reductions
when these metrics have hit extreme levels. First, we
investigate the European sovereign default crisis begin-
ning in 2010 when Greece, Spain, and Portugal were
sharply downgraded (and Greece eventually defaulted).
After a brief period of recovery, fragility in Europe
returned in 2012 when Greek officials admitted an
exit from the Eurozone was possible. Second, during
the Taper Tantrum in May 2013, the Federal Reserve
announced that it planned to wind down (taper) bond
purchases, which caused a steep fall in government bond
prices that spread to other asset classes. Finally, in the
second half of 2015, there were concerns about slowing
Chinese growth and falling commodity prices. During
each of these episodes, unattractive market conditions
can be captured by the defensive indicators, potentially
allowing portfolio managers to reduce risk allocations.
Our article is related to two streams of literature.
The first is on factor timing and, more generally, market
timing. Ferson and Harvey (1991) wrote a seminal paper
in this literature, with more recent contributions made
by Arnott et al. (2016), Asness (2016), and Hodges et al.
(2017). Our defensive factor timing framework is a form
of factor timing, but it concentrates on mitigating risk
as opposed to generating excess returns. There is also
a voluminous risk management literature. Most of the
papers in risk management focus on using signals and
characteristics to monitor the risk of a given portfolio,
rather than using inputs to make ex ante portfolio deci-
sions. In one stream of the risk management literature,
researchers such as Grossman and Zhou (1993) and
Browne (1997) devised optimal strategies for managing
downside risk—but they only used information from
the return series of the portfolio as opposed to allowing
investors to make defensive timing decisions using other
fundamental information.
Quantitative Special Issue 2019
After describing the macro factor setting, in
the next section we describe a systematic framework
for defensive factor timing using three signals: a risk
tolerance indicator of aggregate market conditions, a
measure capturing the benefits of diversification, and
valuation indicators for each factor. Finally we discuss
historical case studies in which these metrics can be
applied.
MACRO FACTORS
Following the framework of Ross (1976), we sum-
marize the movements of global asset class returns with
common factors. These macro factors have three funda-
mental characteristics: (1) They help explain the majority
of the variability in asset class returns; (2) historically,
they have been rewarded over long-term horizons
because of an undiversifiable risk premium; and (3) they
are economically intuitive.1 Put another way, the macro
factors have broad effects and span several asset classes,
and exposure to these factors has resulted in persistent
risk premiums over the long run.
Macro Factor Mimicking Portfolios
Statistical approaches suggest using a parsimonious
set of factors. We take a global portfolio of 14 major
asset classes from April 2004 to December 2017.2 As
shown in Exhibit 1, the first three principal compo-
nents explain 82% of the variation, and the explanatory
power increases to 92% with six principal components.
Although statistical principal components show that the
1
 See Ang (2014) for a literature summary on factor investing.
2
 The asset classes and representative indexes used are
inf lation-linked debt, Bloomberg Barclays World Government
Inf lation-Linked Bond Index; developed sovereign debt, Citigroup
World Government Bond Index 7–10 Yr; investment-grade credit,
Bloomberg Barclays Global Aggregate Corporate Index; emerging
sovereign debt, JPMorgan Emerging Markets Bond Index Com-
posite; high-yield credit, Bloomberg Barclays US Corporate High
Yield Index; developed equity, MSCI World Index; developed
small-cap equity, MSCI Small Cap World Index; emerging equity,
MSCI World EM Index; private equity, S&P Listed Private Equity
Index; infrastructure, S&P Global Infrastructure Index; prop-
erty, FTSE EPRA/NAREIT Global Real Estate Index; volatility,
Chicago Board Options Exchange Volatility Index; and commodi-
ties, Bloomberg Ex-Energy Subindex Total Return and Bloomberg
Energy Subindex Total Return. Principal components analysis is
done on the correlation matrix.
The Journal of Portfolio Management   51
Exhibit 1
Principal Component Decomposition of Global Asset Classes


&XPXODWLYH9DULDQFH([SODLQHG
E\3ULQFLSDO&RPSRQHQWV




3&
global asset class returns data can be represented by a
small number of factors, it has little to say beyond the
fact that we need only model a few underlying factors.
To move to a tradeable representation of a small number
of factors based on economic rationale, we follow Chen,
Roll, and Ross (1986) and choose to work with macro
factors as shown in Exhibit 2.
These macro factors are intuitive and have been
the focus of many in the literature. For example, eco-
nomic growth is the foundation of consumption-based
asset pricing and real business cycle models (see, among
many others, Lucas 1978). Policymakers treat separately
the effects of real rates and inf lation (Taylor 1993) and
assign a role to credit in channeling monetary policy
(Bernanke and Gertler 1995).
We form each of the macro factors to be a hypo-
thetical long–short portfolio to maximize exposure to
the underlying macro factor. This is especially impor-
tant for emerging markets, which have significant cor-
relations with developed markets, but there are specific
risks for emerging markets in excess of their exposure
to developed equity (see Bekaert and Harvey 2003),
and the liquidity factor because illiquidity risk increases
during market downturns (see Amihud, Mendelson, and
Pedersen 2005).
52   Defensive Factor Timing Quantitative Special Issue 2019
3& 3& 3& 3& 3& 3& 3&
Economic Intuition for Hypothetical
Macro Factor Portfolios
In Exhibit 3, we show that these hypothetical
macro factor portfolios—economic growth, real rates,
inf lation, credit, emerging markets, and liquidity—each
capture a fundamental risk premium and are constructed
so as to provide relatively clean exposure to the under-
lying premium. Exhibit 3 compares the realized returns
of the economic growth, real rates, and inf lation trade-
able factor portfolios with the respective underlying
economic phenomena.
Panel A graphs gross domestic product (GDP) sur-
prises, calculated as the difference between realized GDP
and GDP expectations, sourced from Consensus Economics
survey forecasts. The correlation between the series from
1997 to 2017 is 0.6. The premiums investors expect from
economic growth reflects compensation for bearing the
risk that GDP will slow below expectations. For example,
during the financial crisis of 2008, realized GDP was sig-
nificantly under expectations. During this time, we see
the economic factor track the decline in GDP with losses
in developed equities, commodities, and listed real estate.
In Panel B of Exhibit 3, we show returns of the real
rates factor (which is the difference in returns between real
government bonds and cash rates) and surprises in real yields.
Exhibit 2
Macroeconomic Factor Definitions

)DFWRU (FRQRPLF5DWLRQDOH )DFWRU0LPLFNLQJ3RUWIROLR

(FRQRPLF 5HZDUGIRUWDNLQJH[SRVXUHWRWKHJOREDOHFRQRP\
*URZWK
5HDO5DWHV 5HZDUGIRUWDNLQJH[SRVXUHWRWKHULVNRIPRYHPHQWV
LQLQWHUHVWUDWHV
,QIODWLRQ 5HZDUGIRUWDNLQJH[SRVXUHWRFKDQJHVLQSULFHV
&UHGLW 5HZDUGIRUOHQGLQJWRFRUSRUDWLRQVUDWKHUWKDQ
JRYHUQPHQWV
(PHUJLQJ 5HZDUGIRUWDNLQJH[SRVXUHWRWKHDGGLWLRQDOSROLWLFDO
0DUNHWV ULVNIURPHPHUJLQJPDUNHWV
/LTXLGLW\ 5HZDUGIRUWDNLQJH[SRVXUHWRLOOLTXLGDVVHWV
Notes: For illustrative purposes only using hypothetical factors and not representative of an actual investment or account. As of June 30, 2018.
The surprises in real yields on the left-hand side of the chart
are sorted in reverse order to illustrate that when yields
surprise on the downside, we expect both real bonds and
the real rates factor to deliver a positive return. The cor-
relation between these two series is 0.8. There is a notable
negative surprise in real yields in 2009, which corresponds
to rising real bonds during the recovery from the financial
crisis. There is a large positive real yield surprise in 2013,
when Federal Reserve Chair Ben Bernanke surprised bond
markets by announcing a decrease in large-scale bond pur-
chases, effectively winding down the quantitative easing
program that had been in place since 2008. This coincides
with one of the worst years for the real rates factor, which
falls when real rates rise more than expected.
We construct the inf lation factor as the difference
between a portfolio of nominal government bonds and
inf lation-protected bonds. This factor has tended to
deliver positive returns when realized inf lation falls
below expectations, as Panel C illustrates. For example,
inf lation expectations fall in the financial crisis of 2008,
and traditional safe havens such as US Treasuries rise
faster than their inf lation-linked counterparts; con-
sequently, the inf lation factor rises significantly, cap-
turing the downside inf lation surprise. The correlation
between the inf lation factor and inf lation surprises is
0.9 over 1997 to 2017.
Having demonstrated that these macro factor
portfolios are good proxies for the underlying eco-
nomic risks, we now illustrate how an investor holding
Quantitative Special Issue 2019
/RQJ (TXLW\IXWXUHVOLVWHGUHDOHVWDWHUHDOHVWDWHLQYHVWPHQWWUXVWV
FRPPRGLWLHV
6KRUW &DVK
/RQJ %DVNHWRIVRYHUHLJQLQIODWLRQOLQNHGERQGV
6KRUW &DVK
/RQJ %DVNHWRIQRPLQDOVRYHUHLJQERQGV
6KRUW %DVNHWRILQIODWLRQOLQNHGVRYHUHLJQVRIPDWFKLQJPDWXULW\
/RQJ ,QYHVWPHQWJUDGHERQGVKLJK\LHOGERQGV
6KRUW *RYHUQPHQWERQGV
/RQJ (PHUJLQJPDUNHWHTXLW\HPHUJLQJPDUNHWGHEW
6KRUW 'HYHORSHGPDUNHWHTXLW\GHYHORSHGJRYHUQPHQWERQGV
/RQJ 6PDOOFDSHTXLW\
6KRUW /DUJHFDSHTXLW\YRODWLOLW\IXWXUHV
a portfolio exposed to these macro factors may be able
to defensively time the factor exposures using various
indicators.
INDICATORS FOR DEFENSIVE
FACTOR TIMING
We start with a robust strategic multifactor port-
folio that is deliberately diversified across the underlying
drivers of asset class returns. The simplest approach is to
allocate an equal amount of risk to each factor portfolio.
Investors with unique investment goals, such as higher
returns or reduced risk, can adjust the factor weights to
meet those bespoke requirements, as discussed by Bass,
Gladstone, and Ang (2017).
For the purposes of this article, we use a hypo-
thetical strategic factor allocation constructed such that
30% of the risk is attributable to economic growth and
30% to real rates, and the remaining 40% is driven by
an equally weighted combination of inf lation, credit,
emerging markets, and liquidity. The total portfolio
is scaled to have an unconditional volatility of 10%.
By upweighting economic growth, the investor has a
higher exposure to a factor that historically has had
larger long-term Sharpe ratios than the other factors;
similarly, the investor benefits from higher exposure
to real rates, which have tended to do well when
policymakers seek to lower real rates below expecta-
tions to stimulate economic growth during bad times
The Journal of Portfolio Management   53
(see Exhibit 3). Our systematic approach to defensive Risk Tolerance Indicator
factor timing seeks to deviate from this diversified
strategic factor allocation and targeted volatility when Under normal economic conditions, we expect
we observe extremes in risk aversion, correlations, or assets to be priced such that expected returns are a posi-
valuations. We now describe each of these indicators tive function of asset riskiness: The higher the risk of
separately. an asset, the higher the expected return, and vice versa.

Exhibit 3
Relating Macro Factors to Economic Shocks
3DQHO$(FRQRPLF*URZWK)DFWRUDQG6XUSULVHVLQ(FRQRPLF*URZWK
 

 



(FRQRPLF)DFWRU5HWXUQ


*'36XUSULVH

±
±
±
±
±
±
±

± ±

± ±



















*'36XUSULVH/+6 (FRQRPLF)DFWRU5HWXUQ5+6 

3DQHO%5HDO5DWHV)DFWRUDQG6XUSULVHVLQ5HDO5DWHV
± 

±

5HDO5DWHV)DFWRU5HWXUQ
5HDO<LHOG6XUSULVH

±

 


±


 ±





















5HDO<LHOG6XUSULVH/+6 5HDO5DWHV)DFWRU5HWXUQ5+6

(continued)

54   Defensive Factor Timing Quantitative Special Issue 2019

Exhibit 3 (continued)
Relating Macro Factors to Economic Shocks
3DQHO&,QIODWLRQ)DFWRUDQG6XUSULVHVLQ,QIODWLRQ
± 

±

±

,QIODWLRQ)DFWRU5HWXUQ

,QIODWLRQ6XUSULVH

±

 


±

±


 ±




















,QIODWLRQ6XUSULVH/+6 ,QIODWLRQ)DFWRU5HWXUQ5+6

Notes: Past performance is not indicative of future results. Returns do not account for any fees or transaction costs. If fees were included, returns would be
lower. For illustrative purposes only using hypothetical factors and not representative of an actual investment or account. The modeled performance is cal-
culated with the benefit of hindsight and knowledge of factors that may have positively affected its performance and has inherent limitations. This analysis
cannot account for risk factors that may affect actual portfolio performance.

However, during periods of low risk tolerance, the denote the ranking by risk. The RTI is then calculated
positive correlation between risk and return can turn as the correlation between the two rank vectors:4
negative: When investors panic, they tend to f lee from
risk assets, driving down the price while bidding up the RTI = corr [ q( Rt ), q(σ t )]
price of safer assets. During these periods of extreme
negative sentiment, even well-diversified portfolios Typically, positive RTIs accompany periods of
can suffer sharp negative returns as market dynamics increasingly positive investor sentiment, or risk-on
break down and investors are penalized for holding environments, whereas periods of declining investor
risky assets.3 sentiment, often coinciding with a f light to quality, will
The risk tolerance indicator (RTI) aims to identify see a negative RTI. Investor sentiment tends to per-
such low-frequency, high-loss periods ex ante. Let q( Rti ) sist, so rising (falling) investor sentiment usually leads
denote the ranking of asset i by return, and let q(σ it )
4
 Note that we could also have defined the RTI equal to the
correlations between the quantities themselves. Taking ranks allows
the indicator to be applicable to asset pricing models that have a non-
linear dependence between risk and expected return and prevents
3
 There is a substantial literature following, French, Schwert, the dominance of extreme values. Rank statistics are commonly
and Stambaugh (1987), documenting a pronounced contempora- used in robust statistical inference. For well-behaved, especially
neous negative correlation between realized returns and volatility, normally distributed, data, there is little difference between using
which we capture in the RTI. The relation between expected Pearson correlations on raw series versus Spearman rank correla-
returns and volatility is much harder to pin down empirically but tions—see, for example, Huber (1981). A similar rank correlation
should theoretically be positive over most of the range of volatility measure is computed by Kumar and Persaud (2002) to assess senti-
(see Ang and Liu 2007). ment in the equity market.

Quantitative Special Issue 2019 The Journal of Portfolio Management   55

Exhibit 4
Risk Tolerance Indicator
 

5ROOLQJ<U5HWXUQ06&,:RUOG,QGH[KHGJHG
 

 
5LVN7ROHUDQFH,QGLFDWRU

 

 

± ±

± ±

± ±

± ±
            

5LVN7ROHUDQFH,QGLFDWRU/+6 5ROOLQJ<U5HWXUQ06&,:RUOG,QGH[KHGJHG5+6

Notes: Indexes are unmanaged, and it is not possible to invest directly in an index. Past performance is not a guarantee of future results.

to a further increase (decrease) in investor sentiment. Diversification Ratio

Under these conditions, our strategy of bearing system-
atic risk is susceptible to drawdowns; the size of these
drawdowns depends on the extent of the risk aversion.
When the RTI reaches extreme negative levels, we
can scale down, pro rata, the exposure of our strategic
multifactor portfolio while remaining invested across
asset classes.
We compute the RTI using rolling three-month
periods, with weekly return frequency data, for the
14 asset classes and graph the statistic in Exhibit 4. In
this exhibit we also overlay rolling one-year returns of
hedged MSCI World equities. (The correlation between
the RTI and developed equity returns is 0.6.) When
equity returns fall and investors demand safer assets in a
risk-off environment, such as in 2008, 2011, and, more
recently, the first quarter of 2018, we see a commensu-
rate decline in the RTI. By construction, the RTI falls
most sharply when correlations between risky and safe
assets fall toward -1, when the rotation from risky assets
to safe havens is most severe. Such an episode happens
in 2008.
While the RTI measures the relation between
assets’ risk and returns, during bad times the correlations
between assets can also change. In particular, diversifica-
tion can become less effective. To capture these effects,
we include the diversification ratio as a measure of port-
folio concentration
Diversification   Ratio =
∑w σ
i i
σp
where wi and σi are the weight and risk of asset i, and σp
is the risk of the portfolio. To calculate volatilities, we
use a short-term, 21-day risk model.
A high diversification ratio indicates that the
portfolio’s risk is substantially less than the sum of the
risks of the individual assets and, as such, that the port-
folio benefits from significant diversification. As correla-
tions increase, the diversification ratio decreases toward
one, signifying diminishing diversification benefits in
the portfolio. Periods when the diversification ratio

56   Defensive Factor Timing Quantitative Special Issue 2019

Exhibit 5
Diversification Ratio
 

 

 

5ROOLQJ<U5HWXUQ
'LYHUVLILFDWLRQ5DWLR

 

 

 ±

 ±

 ±
           

'LYHUVLILFDWLRQ5DWLR/+6 5ROOLQJ<U5HWXUQ5+6

Notes: The 20/80 portfolio represented by MSCI World Index (hedged) and Bloomberg Barclays Global Aggregate Index (hedged), respectively. Indexes
are unmanaged, and it is not possible to invest directly in an index. Past performance is not a guarantee of future results.

is low tend to correspond with periods of significant
market stress, when tail risks are elevated.
By construction, our strategic multifactor port-
folio seeks to maximize diversification by balancing
risk across historically rewarded macroeconomic fac-
tors, so it is vulnerable to correlation spikes that reduce
the benefits of diversification. A short-term measure
of portfolio diversification could serve as an indicator
for reducing portfolio risk against a background of
increasing correlations.
Exhibit 5 illustrates historic diversification ratios
against the rolling one-year return of a 20/80 stock/
bond portfolio, consisting of 20% MSCI World Index
(hedged) and 80% Bloomberg Barclays Global Aggregate
Index (hedged). We use a 20/80 portfolio to represent
a simple multiasset portfolio with balanced equity and
interest rate risks, which normally benefits from diver-
sification. The Taper Tantrum of 2013 is an extreme
episode of deteriorating diversification benefits, when
correlations between equities and bonds spike toward
one. In Exhibit 5 we see the diversification ratio and the
20/80 portfolio decline in tandem. It is worth noting
that during 2013 when the diversification ratio decreases,
the RTI remains broadly positive (see Exhibit 4), illus-
trating the benefit of having multiple defensive timing
indicators.
Valuation Indicators
Factors, like all investments, become richer or
cheaper compared to their historical long-term averages.
We construct factor valuation indicators to measure their
prices relative to intrinsic value. Each macro premium is
aggregated and smoothed over time to produce a mea-
sure of how the valuation of the factor compares to its
own history. A positive valuation score indicates cheap-
ness, and a negative score indicates richness. A factor is
fairly valued compared to its own historical long-term
average when its valuation score is zero. The time-
series scoring also adjusts for long-term risk premiums
embedded in each factor. We describe our valuation
methodology for each factor.
Economic growth. We use a variant of Shiller’s
cyclically adjusted price-to-earnings ratio (CAPE)

Quantitative Special Issue 2019 The Journal of Portfolio Management   57

(see Campbell and Shiller 1988) as the valuation indicator
for economic risk because it measures the price of equity
markets relative to real earnings. Specifically, we build
a market-cap-weighted global CAPE using data from
individual countries and then take the inverse to produce
an earnings yield. We also look at the difference between
the earnings yield and bond yields as a measure of the
economic risk premium.
Real rates. Since Solow (1956), many authors
have linked the equilibrium real rate in the economy to
economic growth and other preference parameters of a
representative agent. In these formulations, the real rate
is the rate consistent with economic growth matching
potential without accelerating inf lation. Real rates
embedded in real bonds, however, can differ because of
risk premiums and the action of policymakers. Never-
theless, we use the motivation of these economic anchors
to look across various developed-market economies at
how real interest rates compare to expected GDP growth
to determine the valuation of the real rates factor.
Inf lation. We measure the valuation of the inf la-
tion premium by comparing market-implied breakeven
inf lation, based on five-year government bonds, with
expected inf lation, which we proxy with a combination
of survey inf lation expectations and trailing realized
inf lation.
Credit. Consistent with Merton (1974), who
showed that credit spreads must compensate investors for
expected defaults, we measure the valuation of credit risk
as the difference between the current credit spread and
the loss given default, calculated using a combination of
historical data and economic intuition. We calculate this
premium across both investment-grade and high-yield
credit, for both the United States and Europe.
Emerging markets. We measure the emerging
markets risk premium using both equity and bond valu-
ations. For equities, we compare the Shiller earnings
yields and dividend yields of emerging and developed
market countries. For bonds, we compare the current
spread of emerging market debt over US Treasuries to
its long-term average. Additionally, following Harvey
(1995) and others, we use regression-based analysis on
various political risk indicators to calculate the marginal
price of political risk, which indicates whether polit-
ical risk in the marketplace is being priced to earn a
premium.
Liquidity. The liquidity valuation indicator
accounts for both strategies that compose the factor: the
58   Defensive Factor Timing Quantitative Special Issue 2019
small-minus-big effect (see Banz 1981) and volatility
selling (Bakshi and Kapadia 2003). For the small-stock
component, we again use the Shiller earnings yield,
this time comparing the earnings yields of small and
large cap equities. The volatility selling component
consists of two valuation measures: the carry implied
by the VIX term structure and the ratio of the cur-
rent price, spot VIX, to fundamental value, which
we take as the short-term realized volatility of the
S&P 500 Index.
Exhibit 6 provides the historical valuation levels
for each factor, where the level corresponds to the valu-
ation score, or number of standard deviations away from
fair value. Scores below zero imply a factor is currently
expensive relative to its history, and scores above zero
suggest attractive valuations. Because interest rates have
largely decreased since the 1970s (at least until 2017), the
real rates factor has a prolonged period of being over-
valued (Panel A). Similarly, the inf lation factor becomes
very expensive after the financial crisis as f light to safety
effects push up nominal bond prices (Panel B). Con-
versely, the growth factor becomes cheap during the
financial crisis because equity returns are significantly
negative during that time (Panel D). Since the begin-
ning of the equity bull market in 2009, multiple expan-
sion drives prices higher, outpacing fundamentals, and
the premium associated with economic growth declines
in turn.
EMPIRICAL RESULTS
In this section, we evaluate the three defen-
sive timing indicators—RTI, diversif ication ratio,
and factor valuation indicators—across recent market
environments.
Philosophy of Defensive Factor Timing
The goal of defensive factor timing is to protect
capital against prolonged losses; it is not intended to
increment returns via tactical decisions. Therefore, we
do not reduce the risk in our portfolio every time our
signals indicate a negative outlook. Rather, the signals
must indicate extreme levels of risk aversion, corre-
lation, or valuations, to warrant reducing risk. Some
extreme positions might be represented by the indi-
cators approaching, or breaching, plus or minus two
Exhibit 6
Factor Valuation Scores
3DQHO$5HDO5DWHV)DFWRU



&KHDSHU !


([WUHPHO\&KHDS




±
±
 5LFKHU

±
([WUHPHO\5LFK
±
±
±
±
            

3DQHO%,QIODWLRQ)DFWRU



&KHDSHU !


([WUHPHO\&KHDS




±
±
 5LFKHU

±
± ([WUHPHO\5LFK
±
±
±
            

3DQHO&&UHGLW)DFWRU



&KHDSHU !


 ([WUHPHO\&KHDS



±
±
 5LFKHU

±
([WUHPHO\5LFK
±
±
±
±
            

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   59

Exhibit 6 (continued)
Factor Valuation Scores
3DQHO'(FRQRPLF*URZWK)DFWRU



&KHDSHU !


([WUHPHO\&KHDS




±
±
 5LFKHU

±
([WUHPHO\5LFK
±
±
±
±
            

3DQHO((PHUJLQJ0DUNHWV)DFWRU



&KHDSHU !


 ([WUHPHO\&KHDS



±
±
 5LFKHU

±
([WUHPHO\5LFK
±
±
±
±
            

3DQHO)/LTXLGLW\)DFWRU



&KHDSHU !


([WUHPHO\&KHDS




±
±
 5LFKHU

±
([WUHPHO\5LFK
±
±
±
±
            

Note: For illustrative purposes only using hypothetical factors and not representative of an actual investment or account.

60   Defensive Factor Timing Quantitative Special Issue 2019

Exhibit 7
RTI and Diversification Ratio during Case Study Periods

(XUR (XUR 7DSHU &KLQD

 &ULVLV &ULVLV 7DQWUXP DQG2LO 



5LVN7ROHUDQFH,QGLFDWRU57,



'LYHUVLILFDWLRQ5DWLR'5



 

±

±

±

± 
$SU
-XO
2FW
-DQ
$SU
-XO
2FW
-DQ
$SU
-XO
2FW
-DQ
$SU
-XO
2FW
-DQ
$SU
-XO
2FW
-DQ
$SU
-XO
2FW
3HULRGV+LJKOLJKWHGLQ&DVH6WXGLHV 57, '5

standard deviation thresholds.5 Although the definition consistently experiences a drawdown in each scenario,
of infrequent action is subjective, we estimate that defen- the metrics that serve as our forward-looking indicators
sive actions might take place, on average, once every exhibit dissimilar behavior across the scenarios. For the
12–18 months. first and second scenarios, the RTI f lags the imminent
Exhibit 7 plots the RTI and diversification ratio downturn, whereas in the third the diversification ratio
from 2006 through the second quarter of 2018. We does so. This underscores the importance of having a
overlay the four time periods on which the following mosaic of indicators to make defensive timing deci-
case studies are based: European Crisis 2010, European sions robust. Over the sample from January 2006 to
Crisis 2012, Taper Tantrum of 2013, and China and Oil June 2018, the correlation between the RTI and the
Concerns of 2015. In the case studies that follow we see diversification ratio is -0.2, suggesting our two indica-
that although our hypothetical macro factor portfolio tors are complementary.
Our analysis is hypothetical and represents actions
5
that could be taken on the diversified macro factor
 In terms of hypothesis testing, we are focused on Type I portfolio in response to elevated risks highlighted by
errors. In practicing defensive timing, on some occasions we will
reject the null (no regime change of good times) in favor of the
one or more of the indicators. In the case studies that
alternative (regime change to bad times). Thus, some false posi- follow, we describe the market backdrop, comment on
tives are inevitable. The use of extreme thresholds means that we how it affects the defensive timing indicators, and show
will fail to reject with some small probability, corresponding to the how the indicators could be used to inform changes
p-value of a classical test. In practice, we can reduce the frequency to the portfolio in an effort to mitigate portfolio loss.
of Type I errors by contextualizing the information we receive
We compare the simulated drawdowns of our de-risked
from the signal. For example, in September 2016, the diversification
ratio sharply declines to its lowest level since the Taper Tantrum in hypothetical macro factor portfolio against the simu-
2013. This indicates a breakdown in cross-asset correlations and an lated performance of our baseline hypothetical macro
environment that would warrant reducing risk. However, a further factor portfolio from the second section to examine
assessment of market conditions indicates a benign environment, the efficacy of our risk-reducing portfolio changes.
and correlations increase as all asset classes move higher.

Quantitative Special Issue 2019 The Journal of Portfolio Management   61

For simplicity, we assume any reduction in total portfolio
risk is done on a pro rata basis by 20%. When a factor
valuation indicator signals a reduction in one or more
specific factors, we assume the reduction is 5%.
Euro Crisis 2010 and 2012
After improvement in the global economy fol-
lowing the global financial crisis, the European sover-
eign debt crisis began to drive a sharp decline in investor
sentiment in the first half of 2010. Within a two-day
period in April 2010, Spain and Portugal’s investment-
grade ratings were cut, and Greek government bonds
were downgraded to junk status. Financial reform
debates in Washington, DC and in European capitals,
fraud charges in the financial sector, the oil spill disaster
in the Gulf of Mexico, and uncertainties around global
economic growth further contributed to the increasingly
negative investor sentiment.
Exhibit 7 shows that, at the end of April 2010, the
RTI deteriorates rapidly, from very positive levels to
below -30% in a matter of weeks. This rapid deteriora-
tion ref lects an extreme risk environment—one that
we believe warrants a reduction in total portfolio risk.
In September 2010, risks of a double-dip recession sub-
side, and we see the RTI rebound to positive territory.
(Exhibit 7 shows that during this period, the diversifi-
cation ratio is within normal tolerance, and Exhibit 8,
Panel A shows that valuations are not unusually high
or low.)
What is the effect if an investor uses the sudden
decrease in the RTI, ref lecting the sudden spike in
aggregate risk aversion, to help inform when to reduce
risk in the baseline multifactor portfolio? In Exhibit 9,
Panel A, we illustrate the difference in the drawdown
profiles of our hypothetical portfolio, de-risked by 20%
and left at full risk, during the period between May and
September 2010. We see there is some drawdown reduc-
tion: The baseline portfolio experiences a maximum
drawdown of -4.1%, whereas the de-risked portfolio
experiences a maximum drawdown of -3.3%.
Using the RTI as an indicator for when to reduce
risk has a similar effect in the second quarter of 2012.
At this time, fragility in Europe returned to the fore-
front with renewed concerns about Greece and Spain.
The inconclusive elections in Greece on May 6 led offi-
cials to admit for the first time that a Greek exit from
the Eurozone was possible. Shortly thereafter, Spanish
62   Defensive Factor Timing Quantitative Special Issue 2019
authorities announced on May 25 that Bankia, Spain’s
largest real estate lender, required an emergency invest-
ment of €19 billion, driving heightened nervousness
ahead of the European bank recapitalization deadline in
June. Exhibit 7 graphs the second European-driven fall
in RTI, from over 20% at the end of April 2012 to -60%
by mid-June 2012. Exhibit 8, Panel B also shows that, at
this time, the real rates factor exhibits expensive valua-
tions above two standard deviation bounds.
Panel B of Exhibit 9 shows the effect of reducing
the risk of the multifactor portfolio by 20% in response
to the steep drop in the RTI in May 2012. Similar to
2010, reducing risk across the portfolio mitigates some
of the portfolio loss: Between May and August 2012,
the maximum drawdowns of the baseline and de-risked
portfolios are -1.9% and -1.5%, respectively. In August
2012, the RTI returns to approximately zero amid
improving investor sentiment on the announcement of a
possible purchase of government bonds by the European
Central Bank; this improving investor sentiment sug-
gests that taking a normal level of risk in the portfolio
is once again prudent.
Taper Tantrum 2013
In May 2013, Federal Reserve Chairman Ben
Bernanke surprised markets with his declaration to
Congress that the Fed was preparing to end (taper) bond
purchases. The market reaction, nicknamed the Taper
Tantrum, is swift and sharp: Interest rates spike, driving
down bond prices at the same time that equities fall. This
is unusual because often when bond prices rise in a f light
to safety, equity prices fall (as happened in 2008 in the
financial crisis). The Taper Tantrum is an interesting
example in which the RTI is ineffective at f lagging risk
on an ex ante basis, but the diversification ratio picks up
the adverse market environment.
Exhibit 7 shows that the falling diversification
ratio identifies the spike in asset class correlations. In
response to the decreasing diversification ratio, we
reduce the risk in our hypothetical portfolio by 20% in
June 2013. In September, we see correlations return to
pre-May levels, and in response we rescale our portfolio
to full risk.
During this period, our factor valuation indica-
tors also f lag extreme market conditions. Specifically,
Exhibit 8, Panel C shows that the real rates factor appears
to be very expensive—two standard deviations below
fair value; in an environment of rising rate uncertainties In light of this extreme valuation and changing market
this could mean less downside protection from the factor. environment, we alter our factor allocation, reducing the
The valuations of the other factors during this time all exposure to real rates by 5% and increasing the exposure
hover around fair value or the factors’ long-term means. to economic growth accordingly.

Exhibit 8
Valuation Indicators during Case Study Periods
3DQHO$(XUR&ULVLV



&KHDSHU !

±
 5LFKHU

±

±

±
5HDO5DWHV ,QIODWLRQ &UHGLW (FRQRPLF (PHUJLQJ /LTXLGLW\
*URZWK 0DUNHWV

5DQJHRI9DOXDWLRQ6FRUHV ([WUHPHO\5LFK ([WUHPHO\&KHDS $SU

3DQHO%(XUR&ULVLV



&KHDSHU !

±
 5LFKHU

±

±

±
5HDO5DWHV ,QIODWLRQ &UHGLW (FRQRPLF (PHUJLQJ /LTXLGLW\
*URZWK 0DUNHWV

5DQJHRI9DOXDWLRQ6FRUHV ([WUHPHO\5LFK ([WUHPHO\&KHDS 0D\

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   63

Exhibit 8 (continued)
Valuation Indicators during Case Study Periods
3DQHO&7DSHU7DQWUXP



&KHDSHU !

±
 5LFKHU

±

±

±
5HDO5DWHV ,QIODWLRQ &UHGLW (FRQRPLF (PHUJLQJ /LTXLGLW\
*URZWK 0DUNHWV

5DQJHRI9DOXDWLRQ6FRUHV ([WUHPHO\5LFK ([WUHPHO\&KHDS 0D\

3DQHO'&KLQDDQG2LO&RQFHUQV



&KHDSHU !

±
 5LFKHU

±

±

±
5HDO5DWHV ,QIODWLRQ &UHGLW (FRQRPLF (PHUJLQJ /LTXLGLW\
*URZWK 0DUNHWV

5DQJHRI9DOXDWLRQ6FRUHV ([WUHPHO\5LFK ([WUHPHO\&KHDS -XO

Note: For illustrative purposes only using hypothetical factors and not representative of an actual investment or account.

We plot the drawdowns of the theoretical de-risked with the scaling down of real rates exposure, mitigates
and the baseline portfolios during the period from May losses in our de-risked factor portfolio: The maximum
to September 2013 in Exhibit 9, Panel C. As both equi- drawdown is -7.3% versus -9.5% in the baseline port-
ties and bonds fall, our pro rata reduction in risk, coupled folio. This perfect-storm environment of heightened

64   Defensive Factor Timing Quantitative Special Issue 2019

Exhibit 9
Drawdowns versus Baseline Portfolio during Case Study Periods
3DQHO$(XUR&ULVLV


±

±

±

±

±
0D\ -XQ -XO $XJ 6HS
%DVHOLQH)DFWRU3RUWIROLR 'H5LVNHG)DFWRU3RUWIROLR

3DQHO%(XUR&ULVLV


±

±

±

±
0D\ -XQ -XO $XJ
%DVHOLQH)DFWRU3RUWIROLR 'H5LVNHG)DFWRU3RUWIROLR

3DQHO&7DSHU7DQWUXP


±

±

±

±

±
-XQ -XO $XJ 6HS
%DVHOLQH)DFWRU3RUWIROLR 'H5LVNHG)DFWRU3RUWIROLR

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   65

Exhibit 9 (continued)
Drawdowns versus Baseline Portfolio during Case Study Periods
3DQHO'&KLQDDQG2LO&RQFHUQV


±

±

±

±

±

±
-XO $XJ 6HS 2FW
%DVHOLQH)DFWRU3RUWIROLR 'H5LVNHG)DFWRU3RUWIROLR

Notes: Past performance is not indicative of future results. Returns do not account for any fees or transaction costs. If fees were included, returns would be
lower. For illustrative purposes only using a hypothetical macro factor portfolio and not representative of an actual investment or account. The modeled per-
formance is calculated with the benefit of hindsight and knowledge of factors that may have positively affected its performance and has inherent limitations.
This analysis cannot account for risk factors that may affect actual portfolio performance.

cross-asset correlations and broadly unchanged investor
sentiment demonstrates the need for multiple, uncor-
related indicators to effectively implement defensive
factor timing.
China and Oil Concerns 2015
Financial markets respond to three main sources
of uncertainty during the summer of 2015: China,
Greece, and commodities. Throughout the month of
July, Chinese equity markets remained highly volatile
despite measures taken by authorities to stabilize prices,
and commodities suffered intense liquidation pressure,
led by oil. Greece dominated headlines after a negative
vote on the terms of the bailout referendum on July 5.
We see in Exhibit 7 the deterioration of investor senti-
ment throughout July, as the RTI falls to -43% by the
end of the month. This rapid drop in the RTI again sig-
nals a defensive timing opportunity, and we reduce risk
in our macro factor portfolio by 20%. (Note that at this
time, there are no warning signs from the diversification
ratio in Exhibit 7 or the factor valuations in Exhibit 8).
The RTI effectively signals in July what will
become the worst month in financial markets since the
financial crisis. During the month of August, with the
economic slowdown in China at the center of the market
correction, commodity prices reach their lowest point
since 2008, and the VIX Index spikes to more than
50, its highest level since the financial crisis. The RTI
continues to spiral downward and ends the month of
August at -60%. In October 2015, markets begin to nor-
malize as investors start to accept the reality of a Chinese
slowdown; upon this normalization we feel comfortable
increasing portfolio risk back to normal levels.
During the period between July and October
2015, the de-risked hypothetical portfolio has a subtly
improved drawdown profile compared to the baseline
portfolio, -4.5% versus -5.5%, respectively, as illus-
trated in Exhibit 9, Panel D. Although the hypothetical
portfolios have similar drawdown profiles at the start
of the period, the improvement becomes more pro-
nounced in late August and September at the peak of
the market correction, when the RTI further deterio-
rates to -80%.
CONCLUSION
Defensive factor timing periodically, but infre-
quently, takes down risk exposures to a given factor or
the full portfolio during adverse market conditions. This

66   Defensive Factor Timing Quantitative Special Issue 2019

is in contrast to opportunistic factor timing, which tries
to generate excess returns by making frequent tactical
changes to the portfolio; defensive timing tries to miti-
gate the effects of risk. We show how defensive factor
timing may be accomplished by several indicators: an
indicator measuring the risk tolerance of financial mar-
kets; a statistic measuring the effectiveness of diversi-
fication at a point in time; and valuation indicators of
each factor. An investor may take down risk if these
indicators hit high enough thresholds, signaling large,
adverse market conditions.
There are many extensions to our analysis,
including refinement of the measures we use or addi-
tional defensive predictors. We have concentrated on
macro factors, but defensive factor timing could also
be done on style factors—especially with certain fac-
tors such as momentum exhibiting pronounced nega-
tive skewness. We have taken discrete risk reductions
in our examples of defensive factor timing, but other
applications could exploit more gradual de-risking and
re-risking decisions. This would be especially important
in total portfolio applications involving hard-to-trade
illiquid asset classes. Private market assets, such as private
equity, infrastructure, and real estate, have significant
exposures to macro factors, and taking defensive total
portfolio decisions would require taking more defensive
positions in liquid, public portfolios to offset the macro
risk that cannot be traded in private markets.
ACKNOWLEDGMENTS
We thank Andrew Ang, Ted Daverman, Justin
Peterson, He Ren, Paolo Miranda, Trey Heiskell, and
Kaitlyn Piper. The views expressed here are those of the
authors alone and not of BlackRock, Inc.
REFERENCES
Amihud, Y., H. Mendelson, and L. H. Pedersen. 2005.
“Liquidity and Asset Prices.” Foundations and Trends in Finance
1 (4): 269–364.
Ang, A. Asset Management: A Systematic Approach to Factor Based
Investing. New York, NY: Oxford University Press, 2014.
Ang, A., and J. Liu. 2007. “Risk, Return, and Dividends.”
Journal of Financial Economics 85 (1): 1–38.
Quantitative Special Issue 2019
Arnott, R., N. Beck, V. Kalesnik, and J. West. “How Can
‘Smart Beta’ Go Horribly Wrong?” Fundamentals, Research
Affiliates, 2016.
Asness, C. S. “My Factor Philippic.” Working paper, AQR,
2016.
Bakshi, G., and N. Kapadia. 2003. “Delta-Hedged Gains and
the Negative Market Volatility Risk Premium.” The Review
of Financial Studies 16 (2): 527–566.
Banz, R. W. 1981. “The Relationship between Return
and Market Value of Common Stocks.” Journal of Financial
Economics 9 (1): 3–18.
Bass, R., S. Gladstone, and A. Ang. 2017. “Total Portfolio
Factor, Not Just Asset, Allocation.” The Journal of Portfolio
Management 43 (5): 38–53.
Bekaert, G., and C. Harvey. 2003. “Emerging Markets
Finance.” Journal of Empirical Finance 10 (1–2): 3–55.
Bernanke, B., and M. Gertler. 1995. “Inside the Black Box:
The Credit Channel of Monetary Policy Transmission.”
Journal of Economic Perspectives 9 (4): 27–48.
Browne, S. 1997. “Survival and Growth with a Liability:
Optimal Portfolio Strategies in Continuous Time.”
Mathematics of Operations Research 22 (2): 257–512.
Campbell, J., and R. Shiller. 1988. “Stock Prices, Earnings, and
Expected Dividends.” The Journal of Finance 43 (3): 661–676.
Chen, N., R. Roll, and S. Ross. 1986. “Economic Forces and
the Stock Market.” The Journal of Business 59 (3): 383–403.
Ferson, W., and C. Harvey. 1991. “The Variation of Eco-
nomic Risk Premiums.” Journal of Political Economy 99 (2):
385–415.
French, K., G. Schwert, and R. Stambaugh. 1987. “Expected
Stock Returns and Volatility.” Journal of Financial Economics
19 (1): 3–29.
Grossman, S., and Z. Zhou. 1993. “Optimal Investment
Strategies for Controlling Drawdowns.” Mathematical Finance
3 (3): 241–276.
Harvey, C. 1995. “The Risk Exposure of Emerging Equity
Markets.” World Bank Economic Review 9: 19–50.
The Journal of Portfolio Management   67
Hodges, P., K. Hogan, J. Peterson, and A. Ang. 2017. “Factor
Timing with Cross-Sectional and Time-Series Predictors.”
The Journal of Portfolio Management 44 (1): 30–43.

Huber, P. Robust Statistics. New York, NY: Wiley, 1981.

Lucas, R. 1978. “Asset Prices in an Exchange Economy.”

Econometrica 46 (6): 1429–1445.

Kumar, M., and A. Persaud. 2002. “Pure Contagion and

Investors’ Shifting Risk Appetite: Analytical Issues and
Empirical Evidence.” International Finance 5 (3): 401–436.

Markowitz, H. 1952. “Portfolio Selection.” The Journal of

Finance 7 (1): 77–91.

Merton, R. C. 1974. “On the Pricing of Corporate Debt:

The Risk Structure of Interest Rates.” The Journal of Finance
29 (2): 449–470.

Ross, S. 1976. “The Arbitrage Theory of Capital Asset

Pricing.” Journal of Economic Theory 13 (3): 341–360.

Solow, R. 1956. “A Contribution to the Theory of Economic

Growth.” The Quarterly Journal of Economics 70 (1): 65–94.

Taylor, J. B. Macroeconomic Policy in a World Economy: From

Econometric Design to Practical Operation. New York, NY: W.W.
Norton, 1993.

To order reprints of this article, please contact David Rowe at

d.rowe@pageantmedia.com or 646-891-2157.

68   Defensive Factor Timing Quantitative Special Issue 2019


T
David Blitz
is the head of quantitative
equity research at Robeco
Asset Management
in Rotterdam, the
Netherlands.
d.c.blitz@robeco.nl
M ilan Vidojevic
is a quantitative
researcher at Robeco
Asset Management in
New York, NY, and
holds a position at Vrije
Universiteit Amsterdam
in Amsterdam, the
Netherlands.
m.vidojevic@robeco.com
Quantitative Special Issue 2019
The Characteristics
of Factor Investing
David Blitz and Milan Vidojevic
raditional asset allocation models
are centered around capturing
risk premiums offered by distinct
asset classes such as public and
private equities, corporate and government
bonds, and commodities. Decades of empir-
ical research on asset prices have uncovered
a set of factors that drive expected returns
within and across asset classes. Explicit, top-
down allocation to factor premiums (e.g.,
size, value, momentum, quality, and low
volatility) is now becoming mainstream. Yet
many questions remain about how to effi-
ciently gain exposures to these premiums.
Factors are systematic drivers of stock
returns—they explain why certain stocks go
up or down at the same time (co-move) and
why certain stocks command higher expected
returns than others. In practice, investment
portfolios that provide exposures to these
factors are constructed by sorting stocks
on certain characteristics and buying those
that score favorably and, if implemented in
a long–short setting, short selling those that
score unfavorably. Over the years, researchers
have identified a number of factor character-
istics that are key determinants of expected
stock returns. Some examples include market
capitalization (size), book-to-market (value),
gross or operating profitability, asset growth
(investment), and past return (momentum).
In particular, holding all other character-
istics constant, stocks with lower market
capitalizations (small-caps), higher book-to-
market (value), higher profitability and lower
asset growth (quality), and higher recent
returns (winners) have delivered high returns
relative to their peers with lower scores on
these characteristics.
Widely used asset pricing models take
the capital asset pricing model (CAPM) as
a starting point and then augment it with
other factors (e.g., the Fama and French
1993, 2015 three- and five-factor models).
These models inspire a top-down approach
to factor investing, wherein investors allo-
cate a certain amount to each of these fac-
tors. For instance, an investor who wishes
to capture the value premium could allo-
cate a part of the portfolio to the theoret-
ical Fama–French high minus low (HML)
value factor or, more likely, a long-only
value index in practice. A pitfall of the top-
down approach, however, is that each factor
portfolio is constructed in isolation, with a
complete disregard for all other factors. To
address this problem, we propose a bottom-
up approach instead, wherein the expected
return of each stock is a function of all its
factor characteristics.
To illustrate, take two hypothetical
value stocks. These stocks, by definition,
have a high book-to-market characteristic,
which contributes positively to their long-
term expected returns. However, these stocks
can be completely different in terms of other
The Journal of Portfolio Management   69
characteristics that also inf luence their expected returns.
If one stock scores favorably on other factors and the
other has very poor scores on these same factors, these
two stocks will have vastly different expected returns,
even though they are very similar in the value (book-to-
market) dimension. In fact, some value stocks may even
have a lower expected return than the market portfolio
because their favorable book-to-market score is entirely
offset by poor scores on the other factors.
Simple, cookie-cutter approaches to factor investing
that construct factor portfolios to deliver exposure to
one select factor often ignore the impact of other factor
characteristics on expected stock returns and conse-
quently on expected portfolio returns. Many generic
factor products, often labeled smart beta, completely
disregard the impact of other factors when constructing
portfolios with high exposures to any single factor.
A value strategy in its simplest form would invest in
all stocks with the ratio of book-to-market equity
in the top quintile, quartile, or tercile of the invest-
ment universe. However, this clearly ignores the fact
that many value stocks could have, for instance, poor
momentum or profitability characteristics (which is
often the case empirically) that completely dominate
the positive expected return contribution of the high
value characteristic.
In this article, we show that generic factor strat-
egies experience a significant return drag because of
their disregard for other factor exposures. To this end,
we use a characteristics-based multi-factor model that
consists of the aforementioned six well-established stock-
level factor characteristics to estimate the historical pre-
miums that these factor characteristics have delivered.
Furthermore, at each point in time, for each stock in
the universe, we calculate the model-implied return
by multiplying the characteristics of the stock with the
estimated factor premiums. Once we have the implied
returns on each stock in the universe, we analyze the
model-implied and realized returns of five prominent
factor investing styles: size, value, profitability, invest-
ment, and momentum.
Panel A of Exhibit 1 shows the average weight of
each of the five generic single-factor portfolios invested
in the stocks with model-implied returns higher (posi-
tive) and lower (negative) than the return on the market
portfolio. The effect is astonishing: Around 20% of the
generic single-factor portfolio weight is invested in
stocks that have negative implied market-relative returns.
70   The Characteristics of Factor Investing Quantitative Special Issue 2019
In the case of the profitability style, this number is closer
to 50%. This has big implications for the performance
of these investment portfolios.
Panel B shows the average realized returns on
these portfolios. It also shows the returns on the
single-factor portfolios that, at each portfolio rebal-
ancing moment, simply exclude the stocks with such
poor characteristics on other factors that their implied
returns end up being below those of the market bench-
mark. The effect on the portfolio performance is sub-
stantial, with the largest impact visible for the small-cap
and profitability styles.
Our characteristics-based model allows us to
address many other important issues. For instance, the
market portfolio itself is nothing but a collection of
stocks, and at each point in time, each stock has its
own model-implied return. This means that we can
calculate the weight of the market portfolio that is
invested in stocks with such poor factor characteristics
that their model-implied returns are even lower than
those of government bonds. We find that, on average,
around 10% of the market portfolio is invested in stocks
with dismal model-implied returns and that removing
these stocks from the market each month would result
in an increase in the realized return of about 16%, in
relative terms.
Recently, there has been a heated debate about the
best way to design a strategy that provides exposures to
multiple factor premiums simultaneously. Some advo-
cate an integrated approach whereby stocks with the
highest combined rank on the selected factor character-
istics are chosen; others prefer a mixed-sleeve approach
whereby single-factor strategies (sleeves) are used as
building blocks for a fund-of-funds–like multi-factor
portfolio. Our characteristics-based model allows us to
resolve this debate as well. We show that mixing generic
single-factor strategies is clearly suboptimal as around
20% or more of each portfolio is invested in stocks that
have negative implied market-relative returns. Inte-
grated multi-factor portfolios do not suffer from this
issue because they, by construction, select stocks by
looking at all of their factor characteristics. However,
there is also an in-between approach whereby one first
constructs single-factor strategies by selecting stocks
that have high characteristics on the targeted factor but
do not have extremely poor other factor characteris-
tics; in the second step, these strategies are mixed into
one multi-factor portfolio. We show that, so long as
Exhibit 1
Generic and Enhanced Single-Factor Strategies
3DQHO$:HLJKWRI*HQHULF)DFWRU6WUDWHJLHVLQ6WRFNZLWK
3RVLWLYHDQG1HJDWLYH,PSOLHG0DUNHW5HODWLYH5HWXUQV











6PDOO 9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

3RVLWLYH,PSOLHG5HWXUQV 1HJDWLYH,PSOLHG5HWXUQV

3DQHO%5HDOL]HG([FHVV5HWXUQVPRQWK
RI*HQHULFDQG(QKDQFHG6LQJOH)DFWRUV





















6PDOO 9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

*HQHULF (QKDQFHG

the integrated and enhanced single-factor strategies are
designed to provide the same level of factor exposure
(i.e., have comparable overall factor characteristics), their
model-implied and historically realized returns are in
fact not statistically different.
Lastly, through the lens of our model, we also
address some longstanding questions about why dif-
ferent weighting schemes have delivered different
results or why many factors appear to be stronger in
the small-cap segment of the market. We show that the
differences in performance among all these portfolios
can be fully explained by the differences in their factor
characteristics.
Although the characteristics-based model that we
apply in this article can be used to forecast out-of-sample
stock and portfolio-level expected returns, this is not

Quantitative Special Issue 2019 The Journal of Portfolio Management   71

how we apply it in this article. Here we use the model
to ex-post explain why certain factor strategies per-
formed as they did. That is, the fact that we use the full-
sample estimates of factor premiums restricts us from
making any statements about future expected returns,
but because we are applying the model to understand
what happened in the past data, using the full-sample
estimates of factor premiums is not an unreasonable
choice. We further elaborate on this choice in the data
and methodology section of the article.
We conclude that to implement factor investing
efficiently, investors should always consider all relevant
drivers of expected stock returns (i.e., multiple factor
characteristics). A characteristics-based multi-factor
return model can be used to provide insights into the
return drivers of equity portfolios.
DATA AND METHODOLOGY
To estimate the return premiums associated with
various stock-level characteristics, we make use of cross-
section regressions, building on the Fama and MacBeth
(1973) procedure. This means that at the end of each
month, we regress stock returns in excess of the risk-
free rate on characteristics measured at the end of the
previous month. In this way, we obtain a time series
of estimated premiums per unit of each characteristic,
holding all other characteristics constant. We then
average these coefficients over the whole length of the
sample to obtain the average premiums per unit of expo-
sure to a particular characteristic (i.e., factor).
The decision to use the full sample estimates of
factor premiums can be justified by the fact that the pur-
pose of our model is not to forecast the future; instead,
we want to quantitatively describe ex-post the returns
that various factor-based investing strategies have deliv-
ered. It is well-known that empirical estimates of the
magnitude of factor premiums are associated with a wide
margin of error—in fact, some claim that one already
needs more than 100 years of data to adequately estimate
the expected equity risk premium. Because our sample
consists of less than 60 independent cross-section years,
we use full-sample estimates of the premiums as proxies
for their expected values.
One could use our characteristics model to forecast
stock returns out of sample, in which case rolling or
expanding-window Fama and MacBeth regressions could
be used to estimate the factor premiums. However, the
72   The Characteristics of Factor Investing Quantitative Special Issue 2019
issue with the use of rolling regressions is that, following
a period during which a certain factor did not do well
compared to what is expected over longer periods of
time, the estimated premium on this factor could even
end up being negative, which would lead us to errone-
ously conclude that there is a premium of the opposite
sign. On the other hand, the expanding-window regres-
sions are very sensitive to the performance of factors
during the beginning of the expanding-window period;
if certain factors have exhibited returns far different
from their long-term expectations, this would signifi-
cantly affect the estimates for many subsequent periods.
Because our model is not used to forecast factor
portfolio returns, but instead to describe the historical
performance of the factor portfolios, the fact that we
use the ex-post full-sample estimates of the factor pre-
miums is justified. Another way to circumvent the issue
of having to estimate the expected factor premiums in
the first place is to impose a very conservative assump-
tion that all these factor premiums are equal to one
another and set at a certain level (e.g., 15bps a month
per unit of standard deviation in the cross-section). In
this case, the fit of the model would be somewhat worse
than when using the actual premiums estimated from
the data; however, it would still lead to qualitatively
similar results. In such a model, similar to the model that
we use with the full-sample estimates of premiums, all
variation in implied returns over time would be driven
by the variation in the portfolio characteristics.
To calculate implied returns on each stock in the
universe, at each point in time, we multiply a stock’s
current characteristics by the estimated long-term pre-
miums and add them up. We derive one-month-ahead
implied returns for portfolios by taking the weighted
average of the implied returns for the individual stocks
in the portfolio. Our approach is related to the work
of Haugen and Baker (1996) and Lewellen (2015), who
showed that expected return forecasts obtained from
cross-sectional Fama–MacBeth regressions have strong
predictive power for realized stock returns. We apply
a similar model, albeit somewhat simpler because we
include fewer stock-level characteristics to preserve the
parsimony of the model and use static, as opposed to
dynamic, estimates of factor premiums. We find that this
specification is well suited to the purpose of our study;
however, we do acknowledge that one can potentially
build a richer model of expected stock returns. The
characteristics that we consider are market capitalization
(size), book-to-market (value), operating profitability
(profitability), investment, momentum, and market
beta. Although beta is not a priced characteristic, as
we will show later, it is an important control variable
in these regressions primarily because of its theoretical
underpinnings. The profitability and investment fac-
tors are the two quality factors that Fama and French
(2015) recently added to their classic three-factor model,
turning it into a five-factor model. By also including
momentum, for the simple reason that it is too impor-
tant to ignore, our model uses the characteristics that
form the basis of what Fama and French (2018) refer to
as the six-factor model.
We obtain security-level data from the Center
for Research in Security Prices (CRSP) database.
The sample consists of common stocks (share codes
10 and 11) traded on NYSE, AMEX, and NASDAQ
exchanges (exchange codes 1, 2, and 3) from June 1963
to December 2017. We exclude stocks with beginning-
of-month prices less than $1 and stocks with market
capitalization below the 20th percentile market capital-
ization of NYSE-listed stocks. These stocks are known
as micro-caps; according to Fama and French (2008),
they represent around 60% of stocks in the universe but
account for less than 3% of the total market capitalization
of the market portfolio. Therefore, micro-caps have a
big impact on estimation procedures that equally weigh
observations, although their importance in the overall
market is small. The exclusion of micro-caps from asset
pricing tests has been motivated by, among others, Hou,
Xue, and Zhang (2017).
The ex-ante estimates of market betas are obtained
by running univariate regressions of excess stock returns
on the market factor over the 60-month window
(minimum 24 months). The market capitalization of a
stock is its price times the number of shares outstanding.
The balance sheet and income statement information
is from Compustat’s annual files. Book value is the
sum of the book value of stockholders’ equity, bal-
ance sheet deferred taxes, and investment tax credit (if
available), minus the book value of preferred stock. If
available, we use the redemption, liquidation, or par
value to calculate the book value of preferred stock.
Stockholders’ equity is obtained either from Moody’s
industrial manuals or Compustat. If unavailable, we
measure stockholders’ equity preferably as the sum of
the book value of common equity and the par value of
preferred stock, or the book value of assets minus total
Quantitative Special Issue 2019
liabilities if the former is not available. For the fiscal
years ending in 1993 or later, we do not add deferred
taxes to book equity in light of changes in their treat-
ment (Financial Accounting Standards Board 109). The
book value of equity is then divided by the market
capitalization calculated at the end of the previous cal-
endar year to obtain the book-to-market ratio. Oper-
ating profitability is defined as annual revenues minus
cost of goods sold, interest expense, and selling, general,
and administrative expenses divided by book equity
for the last fiscal year end in t - 1, and investment is
the percentage change in firms’ total assets from year
t - 2 to t - 1. Accounting data for a given fiscal year are
updated once a year at the end of June of the following
calendar year. The 12-2 month total return momentum
is the total return from month t - 12 to t - 2.
As a proxy for the risk-free rate, we use the one-
month US Treasury bill rate obtained from the website
of Professor Kenneth French, and we obtain returns on
a constant maturity portfolio of 10-year government
bonds from CRSP. The market portfolio is the value-
weighted portfolio of all stocks in our universe.
We assume that the stock-level characteristics are
linearly related to average stock returns and estimate the
Fama–MacBeth regressions using ordinary least squares.
All characteristics are standardized by subtracting their
cross-sectional mean and dividing by their standard
deviation. Although this does not have a significant
impact on the results of the estimation, it does help
with the interpretation of the estimated parameters and,
more importantly, gives a meaningful interpretation to
the constant. In this setting, the constant represents the
excess return on a stock with characteristics equal to the
universe average (i.e., z-score of zero).
In Exhibit 2, we report the factor premiums esti-
mated over the full sample period. Consistent with the
existing literature, we find statistically significant, nega-
tive premiums for size and investment (i.e., firms with
smaller market capitalization and lower investment, all
else constant, have higher average returns) and statisti-
cally significant, positive premiums for value, profit-
ability, and momentum. In particular, we find that a
stock with a market capitalization that is one standard
deviation lower than the capitalization of the average
stock in the universe is expected to earn a return that
is 10 bps a month higher than that of the average stock,
all else constant. Similarly, the reward to a one stan-
dard deviation higher book-to-market is 15 bps, the
The Journal of Portfolio Management   73
Exhibit 2
Estimated Factor Premiums

Constant Beta ln(Mcap) ln(BtM) OP INV MOM

Coefficient 0.73** 0.00 –0.10** 0.15** 0.14** –0.14** 0.26**
t-Stat 3.39 0.04 –2.71 3.01 4.12 –6.39 4.55

Note: Significant at the ** 1% level and * 5% level.

Exhibit 3
Portfolios Sorted on Implied Return





5HDOL]HG5HWXUQ












        
0RGHO,PSOLHG5HWXUQ

reward to a one standard deviation higher profitability
is 14 bps, the reward to a one standard deviation higher
momentum is 26 bps, and the reward to a one standard
deviation lower investment is 14 bps, all per month.
These results are fully in line with the existing asset
pricing literature, which has previously established the
existence of these factor premiums.
Contrary to the predictions of the CAPM and the
Fama–French asset pricing models, market beta is not
priced in the cross-section of stock returns, whereas the
constant is positive and significant, instead of zero. This
is a manifestation of the well-known low-beta anomaly,
which is also fully consistent with the existing litera-
ture; see, for instance, Blitz (2014), Clarke et al. (2014),
and Blitz and Vidojevic (2017). The positive constant
effectively ref lects the equity risk premium (i.e., the fact
that average stock returns are higher than the risk-free
return), which market beta fails to capture. Our esti-
mate of the constant, 0.73% a month, which translates
to 8.76% a year, is in line with the realized excess return
across all stocks over this sample period.
MODEL VALIDATION
Using our multi-factor characteristics-based return
model, each month we calculate the implied return of
each stock over the following month, based on its cur-
rent factor characteristics. To validate the predictions
of our model, we first sort stocks every month into five
portfolios, based on their model-implied returns. We use
value weighting (i.e., the weight of a stock in a portfolio
is proportional to its market capitalization). We further
calculate the implied and realized returns of these five
portfolios over the subsequent month. In Exhibit 3, we
show that the average implied and realized returns line
up closely over the full sample period.
A more formal way to evaluate the fit of our model
is to rerun Fama–MacBeth regressions with realized

74   The Characteristics of Factor Investing Quantitative Special Issue 2019

Exhibit 4
Generic Factor Portfolios: Implied versus Realized Return
0.40%
0.35%
0.30%
Excess Return Per Month
0.25%
0.20%
0.15%
0.10%
0.05%
0.00%
Small
–0.05%
excess stock return on the left-hand side and the implied
stock returns on the right-hand side. If the model fit is
good, we expect to see a slope coefficient equal to 1.
We estimate this model and find a coefficient of 1.05
(t-statistic of 7.11). We further test if this coefficient
is statistically different from 1 and find that it is not
(t-statistic of 0.34). We conclude that our simple model
does a good job matching the first moment of the return
distribution.
We further examine the accuracy of our model’s
predictions by considering various generic factor port-
folios. Specifically, we select the most attractive 20%
of stocks every month based on their size, value, prof-
itability, investment, or momentum characteristics.
Exhibit 4 shows how the average implied returns relate
to the average realized returns of these generic factor
portfolios. Again we observe that realized and implied
returns line up nicely, with the difference between them
being statistically indistinguishable from zero in all five
cases. The size, value, momentum, and investment port-
folios show the best fit. Interestingly, the average implied
market-relative return for the profitability portfolio
Quantitative Special Issue 2019
Value Winners Profitable Low Investment
Implied Realized
is close to zero; this is because better profitability
characteristics tend to go hand in hand with worse size
and value characteristics, which, on balance, implies
market-relative returns close to zero according to our
model. We elaborate on this issue in the next section.
Realized returns of the profitability portfolio do show a
positive excess return, but much lower than that for the
other factors. Although our characteristics-based model
seems to have some trouble explaining the performance
of the profitability portfolio, the difference between the
realized and predicted return is also statistically insig-
nificant for this factor portfolio.
THE CHARACTERISTICS OF GENERIC
FACTOR STRATEGIES
Generic factor strategies select stocks with the
best scores on one particular factor, regardless of their
other factor characteristics. A generic value strategy, for
instance, may buy a stock that has strong value charac-
teristics but, at the same time, poor size, momentum,
profitability, or investment characteristics. On balance,
The Journal of Portfolio Management   75
Exhibit 5
Generic Factor Portfolios: Weight of (Un)Attractive Stocks
100%
90%
80%
70%
Portfolio Weight

60%
50%
40%
30%
20%
10%
0%
Q3

Q3

Q3

Q3

Q3
Q2

Q2

Q2

Q2

Q2
Q4

Q4

Losers

Q4
Winners

Q4

Q4
Small

Value
Big

Growth

HighInv
LowInv
Unprof

Prof
Negative Implied Return (Relative to Market)
Positive Implied Return (Relative to Market)

taking all this information into account, this may imply a
negative, instead of a positive, return for this stock, rela-
tive to the market. With our multi-factor characteristics-
based return model we are able to assess, at each point in
time, how much of a generic factor portfolio is invested
in stocks with negative implied returns relative to the
market. Panel A of Exhibit 1 shows the average results
for the top quintile factor portfolios, and Exhibit 5
presents results for all five quintiles of the five factor
strategies. Returning to the example of a generic value
portfolio, the graph shows that this strategy invests about
20% in stocks that have a negative predicted market-
relative return. This is clearly a non-negligible part of
the portfolio. Similar weights are found for the generic
small-cap strategy (about 20%), the generic momentum
strategy (about 15%), and the generic investment strategy
(about 30%). We also see that going from the top to the
bottom quintile for each factor, the weight in stocks
with implied outperformance declines monotonically.
We next focus on the small, value, winners, profit-
able, and low investment portfolios (i.e., the top port-
folios) and in Exhibit 6 present the result of the factor
contribution decomposition implied by our model.
For instance, the small-cap portfolio has a full-sample
implied outperformance over the market of 30 bps a
month (3.6% a year), the vast majority of which is driven
by the size (ln(Mcap)) characteristics. However, this
portfolio also gets a positive boost from having had a
positive exposure to the value factor and a negative con-
tribution from a negative exposure to the profitability
factor. The value strategy has also experienced a negative
contribution from the profitability factor, on average.
The low-investment portfolio has primarily benefited
from a high value of the (inverse of the) investment
characteristic; however, a substantial part of its outper-
formance comes from a high value exposure. This is not
surprising given that Fama and French (2015) showed
that their value (HML) and investment (conservative
minus aggressive [CMA]) factors are closely related and
that, in fact, the CMA factor subsumes HML in the
spanning regressions, making it obsolete.
Exhibit 7 shows the percentage of months in our
sample when each of the six characteristics of these factor
portfolios were lower (unfavorable) than that of the
market (i.e., the targeted characteristic is going against
other characteristics). For instance, the small-cap port-
folio never has a worse size (ln(Mcap)) characteristic than
the (capitalization-weighted) market; in 12% of months
it has a lower value (ln(BtM)) characteristic, and in 51%
of months it has a worse momentum score. However, the
small-cap portfolio systematically goes against the two
quality factors—in every single month the profitability

76   The Characteristics of Factor Investing Quantitative Special Issue 2019

Exhibit 6
Implied Return Decomposition
0.5%

0.4%

0.3%
Implied Return

0.2%

0.1%

0.0%

–0.1%

–0.2%
Small Value Winners Profitable Low Investment

ln(Mcap) ln(BtM) OP INV MOM

Exhibit 7 is quite weak. Our model allows us to look at the

Percentage of Months with Unfavorable performance of the small-cap stocks from a bottom-
Characteristics up characteristics perspective. We confirm that small
stocks indeed tend to have poor quality characteristics
ln(Mcap) ln(BtM) MOM OP INV and conclude that our results are consistent with those
Small 0 12 51 100 78 of Asness et al. (2018).
Value 9 0 46 100 14 Clarke, de Silva, and Thorley (2017) constructed a
Winners 29 56 0 52 73 so-called pure size factor that is intended to be orthog-
Profitable 99 100 39 0 65 onal to other factors and showed that the difference in
Low Inv. 15 10 45 75 0
performance between this pure factor and the generic
factor is very small. This finding contradicts that of our
characteristic is worse than that of the market, and in the article and that of Asness et al. (2018), who also reported
case of investment, this occurs 78% of the time. substantial improvements in the performance of a size
Asness et al. (2018) showed that the average small- factor that eliminates strong negative exposures to other
cap stock is junky (i.e., the opposite of high quality), factors—quality in particular. The reason is that the
and this detracts from the realized return on the small multivariate regressions used by Clarke, de Silva, and
minus big (SMB) factor. Using ex-post spanning return Thorley (2017) ensure that the estimated factor pre-
regressions, they show that if one corrects the SMB miums are pure ex-ante but not necessarily ex-post. We
factor for its negative exposures to other factors, the conjecture that a pure size premium estimated using
small-cap premium emerges as highly economically their approach will still have substantial exposures to
and statistically significant, although its raw return other factors—negative quality exposure in particular.

Quantitative Special Issue 2019 The Journal of Portfolio Management   77

 The prof itability factor merits a closer look,
especially because it is a relatively newly discovered
factor that has gained quite some traction over the recent
period. Novy-Marx (2013) showed that stocks with a
high ratio of gross profits to assets generate abnormally
high returns from the perspective of the, at the time,
prominent asset pricing models and, conversely, unprof-
itable companies generate abnormally low returns.
Hou, Xue, and Zhang (2015) used another version of
the profitability factor (return on equity, or ROE) in
their investment-based Q-factor asset pricing model, and
Fama and French (2015) augmented their three-factor
model (Fama and French 1993) with a profitability and
an investment factors, wherein they use a slightly dif-
ferent proxy for the profitability factor based on firm’s
operating profitability, which we also use in this article.
As a motivation for this new factor, Fama and French
used a variant of the dividend discount model (DDM)
that implies high expected returns on profitable stocks,
holding the level of investment (change in assets) and
book-to-market constant. This is an important condi-
tion that is often overlooked; namely, the DDM gives
the Fama–French factors a conditional interpretation.
In this model, a stock that has a higher profitability also
has higher expected returns than a peer that trades at
the same price multiple (book-to-market) and does the
same level of investment. The model does not predict
that a stock with higher profitability will outperform
unconditionally, for instance, if a stock is more profitable
than another but also more expensive.
The generic high-profitability portfolio delivers
a fairly small excess return of 0.08% a month in our
sample, which is three times smaller than the return
of the momentum portfolio. Our characteristics-based
model can explain this discrepancy: Profitable stocks
tend to score poorly on value and size, which con-
tributed negatively to their model-implied returns.
Exhibit 7 shows that there is a systematically negative
relationship between the operating profitability (OP)
and the book-to-market (ln(BtM)) characteristics—that
is, highly profitable stocks also tend to be quite expen-
sive. This is also why Novy-Marx (2013) called (gross)
profitability the other side of value. The same goes for size
because profitable stock also tend to be larger in terms
of market capitalization.
For the generic profitability strategy, we find that
the part of the portfolio invested in stocks with a nega-
tive implied market-relative return amounts to 55%, so
78   The Characteristics of Factor Investing Quantitative Special Issue 2019
from this perspective it is not surprising that this factor
strategy shows a much weaker raw performance than
the other factor strategies. These results have important
implications for the design of strategies that harvest the
profitability premium. In particular, generic strategies
that target solely the profitability premium are expected
to have suboptimal performance unless they take other
factors and, crucially, the price paid per unit of funda-
mental (book) value into account.
ENHANCING FACTOR STRATEGIES
We next examine what happens to performance if,
each month, we simply remove stocks that have negative
implied market-relative returns from generic factor port-
folios. The performance of such enhanced factor strate-
gies is shown in Panel B of Exhibit 1. Compared with the
generic factor strategies from which they are derived, the
performance improvements are about 20% for the value,
momentum, and investment strategies and about 50%
for the small-cap strategy. For the profitability strategy,
performance more than triples, from 0.08% to 0.29% per
month. This large improvement is not surprising because
a much bigger adjustment is made to this portfolio than
to the other factor portfolios. These results imply that
generic factor strategies are suboptimal and that, even
when targeting one particular factor premium, investors
should not ignore other factor premiums.
Excluding stocks with implied underperfor-
mance helps to enhance a single-factor strategy, but the
resulting portfolios can still have stocks with negative
exposures to other factors that detract from their per-
formance. We next examine how performance changes
if, in addition to removing stocks with implied under-
performance, we also require stocks to have a non-neg-
ative exposure (z-score) to at least one, two, three, or
four other, non-targeted premiums. Exhibit 8 shows
that realized, full-sample returns of each single-factor
strategy tend to increase as we require stocks in the
portfolios to have non-negative exposures to more fac-
tors. For instance, the raw value strategy has a return of
0.23% a month, which increases to 0.28% per month if
we ex-ante exclude stocks with negative implied excess
returns. If in addition, at the time of portfolio formation,
we require that stocks have non-negative exposures to
at least two, three, four, and all five factor premiums,
the strategy returns increase to 0.30%, 0.37%, 0.54%,
and 0.69%, respectively. As we impose more constraints,
Exhibit 8
Further Enhanced Factor Strategies







5HDOL]HG5HWXUQ












6PDOO 9DOXH :LQQHUV 3URILWDEOH /RZ
,QYHVWPHQW
$OO6WRFNV ,5! $W/HDVW $W/HDVW $W/HDVW $OO

Exhibit 9
Realized Versus Implied Returns on Enhanced Single-Factor Strategies
0.9%

0.7%
Implied Return

0.5%

0.3%

0.1%

–0.1%
–0.1% 0.1% 0.3% 0.5% 0.7% 0.9%
Realized Return

the number of stocks in the portfolio decreases from,
on average, 302 with no constraints to 276, 273, 223,
96, and only 13. A similar pattern is observed for other
factors, albeit not always monotonic, because very con-
centrated portfolios can be subject to a fair amount of
stock-specific risk that is in general diversified away in
broader portfolios.
Exhibit 9 shows that the realized returns on these
portfolios align with their model-implied values; that
is, the returns of these portfolios can be attributed to
their factor characteristics. None of the differences
between the realized and implied returns are statistically
significant, although we do observe a couple of more
notable deviations from the 45-degree line. The largest

Quantitative Special Issue 2019 The Journal of Portfolio Management   79

Exhibit 10
Market Portfolio Weight in Stocks with Implied Return below Bonds





3RUWIROLR:HLJKW






































XJ
XJ
XJ

XJ

XJ

XJ
XJ

XJ

XJ
XJ

$
$
$

$
$

$
$

deviations are for the size and investment portfolios
that are constrained to have non-negative exposures
to all five factors; these are also the most concentrated
portfolios, on average containing just over 10 stocks, so
naturally the fit of our return model is somewhat lower.
Nevertheless, even for these portfolios, the difference
between the average implied and realized returns is sta-
tistically indistinguishable from zero.
A similar analysis can be conducted for the equity
market portfolio, which many investors track passively
to earn the equity risk premium. The existence of fac-
tors that are priced in the cross-section implies that the
market portfolio is not mean–variance efficient; that is,
it is suboptimal to hold the market in the presence of
factor portfolios. Using our multi-factor characteristics-
based return model, we can calculate at each point in
time which stocks have an implied excess return that
is not merely below average but is even less than the
excess return on 10-year US Treasury bonds. Over
our July 1963 to December 2017 sample period, the
average excess return on a portfolio of 10-year US bonds
amounted to 0.18% per month.
Exhibit 10 plots the weight of the market port-
folio that is invested in stocks that have a lower model-
implied excess return than that of bonds at each point
in time. This weight varies over time between less than
5% and more than 25%, with the peaks occurring in the
mid-1960s, late 1970s, and early 2000s. On average, 9.8%
of the market portfolio consists of stocks with implied
returns lower than bonds. Systematically removing
such stocks from the equity market portfolio increases
its excess return over 10-year US bonds from 0.36%
to 0.42% per month, which amounts to a 16% increase
in realized returns for investors, in relative terms. The
difference in returns is statistically significant at the
most conservative levels (t-statistic of 3.22). Thus, our
multi-factor characteristics-based return model can also
be used to enhance the capitalization-weighted market
portfolio.
MIXED-SLEEVE VERSUS INTEGRATED
MULTI-FACTOR PORTFOLIOS
We next compare the performance of the inte-
grated and mixed-sleeve multi-factor portfolios. Clarke,
de Silva, and Thorley (2016); Bender and Wang (2016);
and Fitzgibbons et al. (2017) all argued that integrated
portfolios, which are constructed by investing in
stocks with a high combined rank on multiple factors,
deliver higher absolute and risk-adjusted returns than
multi-factor portfolios constructed by mixing single-
factor sleeves. On the other hand, Amenc et al. (2017)
emphasized some of the shortcomings and risks of inte-
grated portfolios, such as inefficiency, instability, and

80   The Characteristics of Factor Investing Quantitative Special Issue 2019

Exhibit 11
Performance of Integrated Multi-Factor Portfolios
7RS 7RS 7RS 7RS
5HDOL]HG 



single-factor strategies that exclude stocks with negative
,PSOLHG 



implied market-relative returns (IR), and the further
'LIIHUHQFH ± ± ± ±
W6WDW ± ± ± ±
Note: Significant at the ** 1% level and * 5% level.
inability to control factor exposures and non-factor risks.
Although both streams of literature focus on keeping
a certain risk-related metric, such as tracking error or
total volatility, constant across these portfolios, none of
these studies explicitly aims to control for the differ-
ences in factor exposures between the mixed-sleeve and
integrated portfolios. This makes the comparisons inap-
propriate: An integrated portfolio may simply exhibit
superior performance characteristics because it comes
with better factor characteristics (exposures) than a
mixed-sleeve portfolio.
The main challenge here is to make the comparison
fair (i.e., to compare the performance of multi-factor
portfolios that offer a similar degree of factor expo-
sure). An integrated portfolio, which buys the top 20%
of stocks with the highest combined rank on the five
factors in our model, contains around 300 names on
average over our sample period. On the other hand,
a mix of the five generic single-factor portfolios that
each contain 300 names provides much lower factor
exposures, implying a lower implied return. We there-
fore look at integrated portfolios with various degrees
of concentration and mixes of not only generic single-
factor strategies but also of the enhanced single-factor
strategies discussed in the previous section.
The integrated portfolios are based on a joint
ranking on the five factors in our model, where each
characteristic receives an equal weight. Based on this
multi-factor ranking, we create value-weighted port-
folios consisting of the top 50%, top 33.3%, top 20%,
or top 10% of stocks. The performance of these inte-
grated portfolios is reported in Exhibit 11. The implied
and realized returns increase steadily as the integrated
portfolio becomes more concentrated, and, once again,
realized returns are close to the implied returns, with
none of the differences being statistically significant.
Quantitative Special Issue 2019
We construct mixed single-factor portfolios by
calculating an equally-weighted average of the five sepa-
rate single-factor portfolios. We do this not only for the
generic single-factor strategies but also for the enhanced
enhanced single-factor strategies from which stocks with
negative exposures to other factors are removed. The
performance of these mixed-sleeve portfolios is reported
in Exhibit 12. The implied and realized returns increase
steadily as one moves from generic to progressively more
enhanced single-factor portfolios. Once more, all real-
ized returns are close to the implied returns, with none
of the differences being statistically significant.
Several conclusions can be drawn from these
results. First, all realized returns are in line with the
implied returns, indicating that the factor characteristics
of a portfolio are the key determinant for its return.
Second, a top 20% integrated multi-factor portfolio
achieves significantly higher returns than a mix of var-
ious top 20% generic single-factor portfolios, but that is
to be expected because it is much more concentrated and
therefore provides more factor exposure. Third, an inte-
grated portfolio diluted to such an extent that its factor
exposures are similar to those of a mix of generic single-
factor portfolios also ends up with a similar return. Or,
the other way around, a mix of enhanced single-factor
portfolios that offer more pronounced factor exposures is
able to match the performance of an integrated approach.
Thus, we conclude that there is no evidence of a factor
integration premium; that is, there is no evidence that
integrating factors gives more return than what may
be expected simply from the resulting exposures to the
individual underlying factors.
Ghayur, Heaney, and Platt (2018) also empha-
sized the importance of making an apples-to-apples
comparison between mixed and integrated solutions
by matching their factor characteristics. However, their
approach is different from ours in a two important ways.
First, when constructing multi-factor portfolios using
the two approaches, they only matched the charac-
teristics of the targeted factors, ignoring the fact that
these portfolios may be different with regard to the
non-targeted factors. Granted, the authors were aware
of this shortcoming, and they did disclose it in footnote
9 (“The portfolios were exposure-matched for the tar-
geted factors, but they may have had differences in other,
non-targeted factor exposures. We did not attempt to
The Journal of Portfolio Management   81
Exhibit 12
Performance of Mixed-Sleeve Multi-Factor Portfolios
$OO6WRFNV ,5! ,5!DQG
5HDOL]HG 





,PSOLHG 





'LIIHUHQFH   
W6WDW   
Note: Significant at the ** 1% level and * 5% level.
control for these ancillary exposures.”). In our article,
we emphasize the importance of thinking about port-
folio characteristics from a holistic standpoint because
both the targeted, as well as the non-targeted, factor
exposures drive expected returns.
Another difference in our approaches is in how
the exposures are matched: Ghayur, Heaney, and Platt
(2018) matched factor exposures over the full sample
period and not at each point in time (footnote 8 in
their paper). This means that their approach could lead
to substantial deviations in any given time period. In
our article, we account for both the targeted and the
non-targeted factor characteristics at each point in time
because the implied stock and portfolio returns are cal-
culated monthly. Our results are consistent with those
of Leippold and Rueegg (2017), who concluded that the
two approaches yield similar results if the portfolios are
constructed in comparable ways.
EXPLAINING OTHER FACTOR PORTFOLIO
RETURN DIFFERENCES
It is well-known that small-cap factor portfo-
lios tend to generate higher returns than large-cap
factor portfolios (Fama and French 2008), but it is not
entirely clear what drives this spread. One possibility
is that the outperformance of the factors in the small-
cap universe is driven by a higher exposure of these
portfolios to the size premium. Another possibility is
that the outperformance comes from a higher expo-
sure to the targeted factor—for instance, if the small-
cap value strategy has higher value characteristics than
the large-cap value strategy. A third possibility is that
the outperformance comes from different exposures to
the non-targeted factor premiums (e.g., if the small-
cap value strategy has better momentum, investment,
or profitability characteristics than the large-cap value
82   The Characteristics of Factor Investing Quantitative Special Issue 2019
,5!DQG ,5!DQG ,5!DQG
  ±
  ±
strategy). A fourth possibility is that the small-cap and
large-cap portfolios have comparable factor characteris-
tics but that the factor premiums are larger in the small-
cap segment, in which case a linear expected return
model is not appropriate. Lastly, it could be the case that
factors that are not accounted for drive these differences.
Another well-known phenomenon is that the
equally-weighted factor portfolios tend to generate
higher returns than their value-weighted counterparts.
Because equally-weighted factor portfolios invest more
in smaller stocks, all of the aforementioned explanations
may also apply here. In addition, equally-weighted factor
portfolios might profit from the periodic rebalancing
that is required to maintain equal weights over time
(i.e., a potential rebalancing premium).
In this section, we decompose the implied returns
of these portfolio pairs using our characteristics-
based multi-factor model and pinpoint the reason for
the observed return differences between small-cap
and large-cap factor portfolios and between equally-
weighted and value-weighted factor portfolios.
Exhibit 13 shows the implied and realized perfor-
mance difference between top quintile value-weighted
factor portfolios in the small-cap space and top quintile
value-weighted factor portfolios in the large-cap
space, where stocks are split into two size groups at
the median market capitalization value of our stock
universe. Exhibit 14 shows the implied and realized
performance difference between equally-weighted and
value-weighted factor portfolios. Both graphs show
a close match between implied and realized returns,
which indicates that our model is able to explain the
observed return differences quite well. The tables below
the graphs present these numbers in a tabular format
and confirm that the difference between realized and
implied returns is not statistically significant. In other
words, the superior performance of the small-cap factor
Exhibit 13
Small-Cap versus Large-Cap Factor Portfolios



5HDOL]HG5HWXUQ








     
,PSOLHG5HWXUQ

9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

5HDOL]HG    
,PSOLHG    
'LIIHUHQFH    ±
W6WDW  ±  ±

Note: Significant at the ** 1% level and * 5% level.

portfolios compared to the large-cap factor portfolios,
and the superior performance of equally-weighted factor
portfolios compared value-weighted factor portfolios,
appears to be fully in line with the implied returns that
can be derived from the factor characteristics of these
portfolios. This means that there is no need to assume
that the same factor exposure is rewarded more in the
small-cap space than in the large-cap space, nor is there
a need to assume that equally-weighted factor portfolios
benefit from some kind of rebalancing premium.
Exhibit 15 shows the breakdown of the implied
return contributions across the factor characteristics
for the small-big factor portfolios. Clearly, the reason
why small-cap factor portfolios outperform their
large-cap counterparts can be mostly attributed to a
higher exposure to the size factor (i.e., lower market
capitalization). All small-cap factor legs tend to have
better value characteristics but worse prof itability
scores. In the case of the past winners, it appears that
the small-cap winners portfolio not only benefits from
a higher exposure to the small-cap factor but also from
a higher exposure to the momentum factor itself; that
is, small stocks have higher momentum scores than
large stocks, on average.
Exhibit 16 shows these breakdowns for the equal
versus value-weighted portfolios. We find that the dif-
ference in performance across factors constructed using
these two weighting schemes can to a large extent be
attributed to the small-cap and value tilts that equal-
weighting and rebalancing back to the equal weight
introduce.
CONCLUSION
Eff icient implementation of factor investing
requires an understanding of how factors perform inde-
pendently and how they are related to one another. In
this article, we apply a characteristics-based multi-factor

Quantitative Special Issue 2019 The Journal of Portfolio Management   83

Exhibit 14
Equal versus Value-Weighted Factor Portfolios


5HDOL]HG5HWXUQ 










      
,PSOLHG5HWXUQ

9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

5HDOL]HG    
,PSOLHG    
'LIIHUHQFH    ±
W6WDW ± ±  ±

Note: Significant at the ** 1% level and * 5% level.

Exhibit 15
Implied Return Decomposition for Factor Portfolios in Small- versus Large-Cap Universes





,PSOLHG5HWXUQ







±
9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

OQ0FDS OQ%W0 23 ,19 020

84   The Characteristics of Factor Investing Quantitative Special Issue 2019

Exhibit 16
Implied Return Decomposition for Equal- versus Value-Weighted Factor Portfolios







,PSOLHG5HWXUQ







±

±
9DOXH :LQQHUV 3URILWDEOH /RZ,QYHVWPHQW

OQ0FDS OQ%W0 23
return model to show that investing in generic single-
factor portfolios leads to suboptimal investment results
because stocks that score highly on one factor can go
strongly against other factors and, on net, have nega-
tive market-relative model-implied and realized returns.
We show that removing such stocks from the portfolio
yields substantially higher returns. We further show that
if, in addition to excluding stocks with implied under-
performance relative to the market, we also exclude
stocks that have negative exposures to one, two, three,
or four non-targeted factors, we can enhance these
single-factor strategies even more. Our model can also
attribute performance differences between mixed-sleeve
and integrated multi-factor portfolios to differences
in their factor characteristics. As generic single-factor
sleeves are an inefficient way to obtain factor exposure,
mixing them into a portfolio also results in suboptimal
multi-factor portfolios. However, a mixed-sleeve port-
folio of enhanced single factors that do not invest in
stocks with implied underperformance is able to match
the performance of a bottom-up integrated multi-factor
portfolio with similar factor exposures. Finally, we show
Quantitative Special Issue 2019
,19 020
that our model can explain performance differences
between factor portfolios in the small-cap and in the
large-cap space, as well as equal- and value-weighted
portfolios. Therefore, an efficient implementation of
factor-based, or any kind of equity strategies requires
an understanding of the underlying portfolio factor
characteristics.
The discussion in this article is confined to the
question of how investors should approach building
an expected return model for equity portfolios. In the
real world, investors’ preferences and constraints cause
portfolios to have a significant exposure to factors that
are not rewarded with higher returns but certainly do
contribute to the risk. An example is industry expo-
sures—an investor should not expect to be compen-
sated for taking industry bets unless they are a result
of the rewarded factor characteristics of the underlying
industry. Although an effective management of the
unrewarded sources of risk is of paramount importance,
one has to be cognizant of the fact that, in the long run,
expected portfolio returns depend on the priced port-
folio characteristics.
The Journal of Portfolio Management   85
ACKNOWLEDGMENTS
The views expressed in this article are solely those of the
authors and are not necessarily shared by Robeco or its sub-
sidiaries. We thank Matthias Hanauer for valuable comments.
REFERENCES
Amenc, N., F. Ducoulombier, M. Esakia, F. Goltz, and
S. Sivasubramanian. 2017. “Accounting for Cross-Factor
Interactions in Multi-Factor Portfolios without Sacrificing
Diversification and Risk Control.” The Journal of Portfolio
Management 43 (5): 99–114.
Asness, C., A. Frazzini, R. Israel, T. Moskowitz, and
L. Pedersen. 2018. “Size Matters, If You Control Your Junk.”
Journal of Financial Economics 129 (3): 479–509.
Bender, J., and T. Wang. 2016. “Can the Whole Be More
Than the Sum of the Parts? Bottom-Up versus Top-Down
Multi-Factor Portfolio Construction.” The Journal of Portfolio
Management 42 (5): 39–50.
Blitz, D. 2014. “Agency-Based Asset Pricing and the Beta
Anomaly.” European Financial Management 20: 770–801.
Blitz, D., and M. Vidojevic. 2017. “The Profitability of Low
Volatility.” Journal of Empirical Finance 43: 33–42.
Clarke, R., H. de Silva, and S. Thorley. 2014. “The Not-So-
Well-Known Three-And-A-Half-Factor Model.” Financial
Analysts Journal 70: 13–23.
——. 2016. “Fundamentals of Efficient Factor Investing.”
Financial Analysts Journal 72 (6): 9–26.
——. 2017. “Pure Factor Portfolios and Multivariate Regres-
sion Analysis.” The Journal of Portfolio Management 43 (3):
16–31.
Fama, E., and K. French. 1993. “Common Risk Factors in
the Returns on Stocks and Bonds.” Journal of Financial Eco-
nomics 33: 3–56.
——. 2008. “Dissecting Anomalies.” The Journal of Finance
63 (4): 1653–1678.
86   The Characteristics of Factor Investing Quantitative Special Issue 2019
——. 2015. “A Five-Factor Asset Pricing Model.” Journal of
Financial Economics 116 (1): 1–22.
——. 2018. “Choosing Factors.” Journal of Financial Economics
128 (2): 234–252.
Fama, E., and J. MacBeth. 1973. “Risk, Return, and Equi-
librium: Empirical Tests.” Journal of Political Economy 81:
607–636.
Fitzgibbons, S., J. Friedman, L. Pomorski, and L. Serban.
2017. “Long-Only Style Investing: Don’t Just Mix, Inte-
grate.” The Journal of Investing 26 (4): 153–164.
Ghayur, K., R. Heaney, and S. Platt. 2018. “Constructing
Long-Only Multi-Factor Strategies: Portfolio Blending vs.
Signal Blending.” Financial Analysts Journal 74 (3): 70–85.
Haugen, R., and N. Baker. 1996. “Commonality in the
Determinants of Expected Stock Returns.” Journal of Finan-
cial Economics 41 (3): 401–439.
Hou, K., C. Xue, and L. Zhang. 2015. “Digesting Anomalies:
An Investment Approach.” The Review of Financial Studies 28
(3): 650–705.
——. “Replicating Anomalies.” Working paper, 2017.
Leippold, M., and R. Rueegg. 2017. “The Mixed vs the
Integrated Approach to Style Investing: Much Ado about
Nothing?” European Financial Management 24 (5): 1–27.
Lewellen, J. 2015. “The Cross-Section of Expected Stock
Returns.” Critical Finance Review 4: 1–44.
Novy-Marx, R. 2013. “The Other Side of Value: The Gross
Profitability Premium.” Journal of Financial Economics 108 (1):
1–28.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

Ashley L ester
is global head of
multi-asset research
and systematic investment
at Schroders in London,
UK.
ashley.lester@schroders.com
Quantitative Special Issue 2019
On the Theory and Practice
of Multifactor Portfolios
Ashley Lester
A
 s factor investing strategies have
 moved from academia to main-
 stream practical application over
 recent years, the question of
how best to build portfolios consisting of
more than one factor has become increas-
ingly pressing. Proponents of traditional
smart-beta–style approaches argue in favor
of building factor portfolios through a set of
single-factor subportfolios (e.g., a value port-
folio, a momentum portfolio, and so on). We
will follow the literature in referring to this
approach as top down. A growing literature
argues instead that combining information
about each factor at the stock level and then
building a single, combined portfolio—an
approach known as bottom up—is likely to
produce superior outcomes.1
The persistence of this discussion
ref lects two shortcomings in the existing
literature. First, although there is a general
intuitive sense that the advantages of com-
bination are larger when the factors are less
correlated, there is (at this stage) no explicit
theory of why this is the case or of whether
there are other empirically testable propo-
sitions that might differentiate the two
1
 The main author who explicitly argues in favor
of single-factor portfolios (albeit with some modifi-
cations) is Amenc et al. (2017), although others also
hedge their bets (e.g., Ghayur, Heaney, and Platt 2016).
The combination side was taken by Bender and Wang
(2016) and Fitzgibbons et al. (2017).
approaches.2 Second, existing empirical work
is typically relatively small in scale, involving
the comparison of a limited number of
portfolios, sometimes under quite special
circumstances.3
This article makes two contributions
to the debate. First, it makes an explicit and
quantitative set of theoretical predictions
relating to simple bottom-up and top-down
multifactor portfolios. These predictions are
highly tractable because they use the sim-
plest possible building blocks. We derive
analytic expressions for the point-in-time
factor exposure of portfolios formed from
stocks with factor characteristics above some
percentile cut-off. As Amenc et al. (2017)
2
 A notable exception that develops a theoretical
framework addressing a similar question was under-
taken by Clarke, de Silva, and Thorley (2016). Their
framework focused on the relative information ratios
of various portfolio construction choices. Although
this is a powerful framework, the machinery is less
well suited to highlighting the simple test of the linear
factor model we highlight here, in particular its appli-
cation over a very large number of empirical tests.
3
 Fitzgibbons et al. (2017) reported the results
of experiments on simulated data but did not make
entirely clear how their results were generated, nor did
they explicitly link their results to a theory of expo-
sure. Ghayur, Heaney, and Platt (2016) ran a larger
number of experiments than most but focused on
the question of relative performance for given factor
exposure. This contrasts with most practically avail-
able approaches to factor investing.
The Journal of Portfolio Management   87
correctly noted, a theory of exposure is not the same
as a theory of returns. They even went so far as to sug-
gest that there is no clear connection between the two.4
We therefore link our analytic expressions for factor
exposure to the simplest possible linear factor models
of risk and return. Explicitly linking factor exposure to
portfolio risk and return generates strong predictions
about the returns, risk, and risk-adjusted returns of top-
down and bottom-up portfolios as the number of factors
and their correlation vary.
The second contribution of this article is the testing
of our quantitative predictions—not on a handful of
carefully chosen portfolios but on over 1,000 matched
top-down and bottom-up portfolios. Previous work
has invariably focused on a small number of portfo-
lios, ignoring the randomness inherent in such results.
In contrast, when examined en masse, the results are
overwhelmingly consistent with the theoretical predic-
tions generated from our theoretical framework. On
average, factor returns and risk scale linearly with factor
exposure.
The results are of strong practical value. A striking
outcome of the approach taken to portfolio construction
is that a bottom-up portfolio of four orthogonal fac-
tors has twice the factor exposure of the corresponding
top-down portfolio. More importantly, the empirical
results show that the integrated portfolio also generates
(in expectation) twice the outperformance of the seg-
regated portfolios. Although there is no similar quan-
titative analytical result for risk-adjusted performance,
empirically, the four-factor bottom-up portfolios gen-
erate an information ratio 40% greater than the top-
down portfolios.
Our results suggest that examining larger sets of
simple portfolios may be a fruitful path for future research.
For example, the strong support our article provides for
simple linear factor models obviates, or at least renders
second order, the increasingly tortured discussion of factor
4
 They argued (e.g., on page 101) that “Although there is
ample evidence that portfolios sorted on a single characteristic
are related to robust patterns in expected returns, such patterns
may break down when incorporating many different exposures at
the same time.” Later (p. 106), they state that “a naïve attempt to
increase factor intensity may be justified from a mechanical engi-
neering perspective in which one tries to tweak calculations to
obtain a high value for some metric; however, it is entirely incon-
sistent with financial concepts.” Ultimately, as they note on page
103, “the relevant question is … whether stronger factor tilts do lead
to improvements in risk-adjusted performance.”
88   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
interactions and nonlinearities.5 Although such effects
may remain important in particular investment situa-
tions, in general the hurdle for identifying such effects
ought to be lifted relative to current practice, which rou-
tinely relies on purely empirical examination of small sets
of portfolios. We suggest that our framework provides
a useful baseline against which to test for more exotic
effects. Finally, in this article, we use exposures and char-
acteristics interchangeably, rather than appealing to factor
betas. Again, the strength of our empirical results lends
support to the characteristics literature rather than the beta
literature (see Kelly, Pruitt, and Su 2018 for an up-to-date
discussion of this literature and Liu 2017 for a related
recent application). Whether similarly strong results can
be generated using factor betas as the measure of exposure
we leave to future research.
A THEORY OF FACTOR EXPOSURES
In this section, we derive a set of analytic statements
relating top-down to bottom-up portfolio construction
for a particular, simple, portfolio construction technique.
These are statements rather than theories because they
relate portfolio construction to ex ante factor exposure
rather than ex post performance. However, combining
these statements with simple linear factor models yields
a precise set of predictions for how factor exposure
translates into expected portfolio performance. If the
empirical results diverge from the theoretical predic-
tions, we can isolate whether the mechanism driving
the failure is exposures, the linear model of returns, or
the linear model of risk.
We define factor scores such that the distribution
of scores for each factor is standard normal, and we form
factor portfolios by selecting and equal-weighting the
top x-percent of stocks ranked on their factor score. For
multifactor portfolios, we add the individual scores to
form a combined score. The combined factor score for a
stock ranked on K factors will therefore be distributed:
 
∑s i ∼ ℵ 0, K + ∑ ∑ ρij  (1)

i ∈K i ∈K j ∈{K \i } 
5
 This literature started with the classic paper by Asness (1997)
on interactions between value and momentum stocks. It is discussed
extensively and sympathetically by Ghayur, Heaney, and Platt (2016)
and referenced by Amenc et al. (2017), as noted earlier. Freyberger,
Neuhierl, and Weber (2017) presented an intriguing state-of-the-art
discussion of factor nonlinearities.
because the variance of each single-factor score is unit
by construction. That is, the distribution of total factor
scores is inf luenced by the set of correlations among all
of the factors. The concept of correlation used here (and
throughout) refers to the point-in-time correlation of
the factor scores across stocks, not a time-series measure
of factor return correlation.
Under these assumptions, we can use the inverse
Mills ratio to calculate the average factor score of a
portfolio that consists of the top x-percent of stocks,
equally weighted. The inverse Mills ratio is a useful
property of the normal distribution that allows us to
calculate the expected value of the truncated normal
distribution. Technically,
z − µ
φ
 σ 
E[X X > z ] = µ + σ
z − µ
1− Φ
 σ 
in which f and F are, respectively, the probability dis-
tribution function and cumulative distribution function
of the standard normal distribution. Finally, we need
to know the expected value of another factor, Y, given
that we have formed a portfolio using a single factor, X.
Under our assumptions, this is simply
E[Y X ] = βX
Cov( X ,Y )
= X
Var ( X )
= ρX
Therefore, the expected factor score over all fac-
tors (not just one), Fk, of a single-factor portfolio formed
using scores for factor k, is
 
E(Fk ) = E  ∑s j 
 j ∈K 
 
= E  sk +

∑ ρ jk sk 

j ∈{N /k }
 
= 1+

∑ ρ jk  E(sk ) (2)

factor but also of the extent to which the chosen factor is
correlated with other factors. In particular, if the chosen
factor is negatively correlated with some other factor,
attempts to increase the intensity of exposure to the
chosen factor inevitably result in ever-greater negative
incidental exposure to the secondary factor.
For a portfolio that selects the top x-percent of
stocks, we can substitute in the inverse Mills ratio, and
the resulting score is
φ( Φ −1( x ))
E(Fk ) = (1 + ( K − 1)ρk )
1− x
where ρk is the average correlation between factor k and
the other K - 1 scored factors, and the expression on the
right is the transformed inverse Mills ratio that defines
the expected value of the top x-percent of observations
of a standard normal.
The total factor score for a multifactor portfolio
formed on the top x-percent of stocks can be derived
from Equation 1 and is
φ( Φ −1( x ))
E(F{K } ) = K + ∑ ∑ ρ jk
j ∈K k ∈{K / j } 1− x
φ( Φ −1( x ))
= K + ∑( K − 1)ρ j
j ∈K 1− x
φ( Φ −1( x ))
= K (1 + ( K − 1)ρ ) (3)
1− x
where ρ is the average correlation across all factors.
It will be helpful later in interpreting the empirical results
to note that the factor score of the underlying benchmark
portfolio (with stock weights the same as those used in
forming the factors, here equal weights) is zero.
Finally, we can use Equation 2, the total factor score
of a single-factor portfolio, to derive factor exposure for
the equally weighted set of single-factor portfolios that
forms a top-down portfolio. The total factor score of an
equally weighted set of K single-factor portfolios is
1 φ( Φ −1( x ))
E K ( E(Fj )) =
K
∑(1 + ( K − 1)ρ j ) 1− x
j

j ∈{N /k } φ( Φ −1( x ))
= (1 + ( K − 1)ρ ) (4)
1− x
This expression highlights that the combined
factor score of any single-factor portfolio is a func- Combining Equations 3 and 4 yields an ana-
tion not only of the intensity of exposure to the chosen lytic expression for the ratio of factor exposure in a

Quantitative Special Issue 2019 The Journal of Portfolio Management   89

Exhibit 1
Ratio of Bottom-Up to Top-Down Exposure

1XPEHURI)DFWRUV
   ϰ 







±         
&RUUHODWLRQ

bottom-up portfolio to that in a similarly formed top- the factor exposure ratio is larger when the number of
down portfolio: factors is larger; equivalently, the effect of more factors
is larger when correlation is more negative. The effect
E(F{K } ) K (1 + ( K − 1)ρ ) K is illustrated in Exhibit 1.
= = (5)
E K ( E(Fj )) (1 + ( K − 1)ρ ) 1 + ( K − 1)ρ So far, we have only predicted factor exposures,
but the point is to predict performance. We, there-
which is strictly greater than one for all levels of average fore, make the simplest possible assumptions regarding
correlation less than one. returns and risk. Adopting the workhorse linear model
of expected factor returns (and ignoring time subscripts
throughout), we assume that returns to stock i are driven
Observation 1: Number of Factors
by the underlying factor returns, rj, with an intensity
Suppose that there are K factors and the factors are proportional to stock i’s exposure to the factor, and an
orthogonal. Then from Equation 5, it is apparent that idiosyncratic error term, ei:
the ratio of factor exposures is K .
ri = ∑sij r j + εi (6)
j ∈N
Observation 2: Correlation
The correspondingly simple model of stock risk is
From Equation 5, it is obvious that the ratio of factor
driven by idiosyncratic risk, which for stock i has stan-
exposures is a strictly decreasing function of correlation.
dard deviation hi, and factor risk, which for factor j has
standard deviation sj:
Observation 3: Correlation
and Number of Factors  K

2

σ =  ηi + ∑sij σ j  (7)
2
i
It is possible to show that the cross-partial deriva-  j =1 
tive of Equation 5 is negative, but we omit the proof.
This means that the inf luence of reduced correlation on Equation 6 generates a trivially simple mapping
from factor exposures to expected returns: Any prediction
90   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
for exposures is simply replicated for expected returns.
The linear factor model, therefore, extends observations
1–3 to expected returns, rather than purely exposures.
Under this assumption, the bottom-up approach gen-
erates greater returns than the top-down approach as a
result of the extra factor exposure generated. However,
whereas observations 1–3 describe point-in-time factor
exposure by definition, the real world is not so simple.
In particular, the error term in returns means that only
expected returns are linear in factor exposure: Realized
returns will not be perfectly linear in exposure even if
the model is true.
Generating similarly neat predictions concerning
risk—and by extension risk-adjusted returns—is more
difficult, but assuming that all sources of risk are mutu-
ally orthogonal permits some progress. We then can
derive expressions for bottom-up and top-down risk
(see proofs in the Appendix).
Portfolio risk for the top-down portfolio is
1
 η2
IMR 2 σ 2j  2
σTD = i +  folio is 1 - (1 - x)K . This fraction increases at a rate less
 N TD K 
where ηi2 is the average idiosyncratic variance across all
stocks, σ 2j is the average across all factors of the factor
risk associated with unit factor exposure, IMR is the
transformed inverse Mills ratio defined earlier, and NTD
is the total number of stocks in the top-down portfolio.
We have ignored single-stock concentration risk arising
from overlapping holdings across the underlying single-
factor portfolios.6 Under the same conditions, risk in the
bottom-up portfolio is
1
 η2  2
σ BU =  i + IMR 2 σ 2j 
 N BU  accumulate expected returns proportional to K and
It is easiest to gain intuition for this result if we
temporarily ignore idiosyncratic risk. In that case, risk
in the top-down portfolio decreases in proportion to
6
 In particular, any stock held by more than one underlying
factor portfolio will generate excess concentration risk relative to
stocks held only once. For simplicity, we ignore this channel of
concentration risk from multiple holdings of the same stock. How-
ever, see later in the main text for much more on the effect on risk
of overlapping holdings arising from the number of stocks held.
Quantitative Special Issue 2019
K , whereas risk in the bottom-up portfolio is constant.
This result exactly offsets the previous result on factor
exposure: If the investor cares about risk-adjusted
returns, so far they are indifferent between the two
portfolio construction techniques.
Now consider the effect of idiosyncratic risk. The
result for the bottom-up portfolio is unchanged because
the number of stocks selected is not a function of the
number of factors.
The result for the top-down portfolio, however,
depends on how the number of stocks varies with the
number of factors. Recall that the top x-percent of stocks
are selected for each factor. If there were no overlap
across factor portfolios, the number of stocks would
increase proportionally with the number of factors, and
the previous result would continue to hold exactly (NTD
is equal to a constant times K). Inevitably, however,
some stocks will be chosen in more than one portfolio.
For orthogonal factors, the probability that a particular
stock is chosen in no factor portfolio is (1 - x)K . There-
fore, the fraction of stocks chosen in the top-down port-
than proportional to K, and therefore risk in the top-
down portfolio decreases at a rate less than proportional
to 1/ K .
Observation 4: Portfolio Risk
For orthogonal factors, risk in bottom-up portfo-
lios is constant with respect to the number of underlying
factors, whereas risk in top-down portfolios falls at a rate
less than 1/ K .
Observation 5: Risk-Adjusted Returns
For orthogonal factors, bottom-up portfolios
maintain constant risk. Therefore, risk-adjusted returns
increase with the number of factors at rate K .
For orthogonal factors, top-down portfolios
maintain constant expected returns as the number
of underlying factors increases, and risk falls at a rate
less than proportional to K . Therefore, risk-adjusted
returns increase at a rate less than K . Thus, for
orthogonal factors, risk-adjusted returns are greater
in bottom-up portfolios than in top-down portfolios,
and their relative advantage increases with the number
of factors.
The Journal of Portfolio Management   91
 We now test whether the exposures generated
empirically in quantile portfolios are consistent with
observations 1–5.7
EMPIRICAL RESULTS
Data and Experimental Setup
We create portfolios that follow the construction
process described in the previous section. These are
long-only, equal-weighted portfolios of the top 10% of
stocks for a given factor score using stock-level data for
five of the most common factors: value, momentum,
profitability, size, and quality. These factors are widely
researched and, because the focus of this article is not to
examine the effectiveness of any one factor or particular
combination of factors, we choose three reasonable mea-
sures for each factor. Exhibit 2 lists the definitions for
each factor.
We form all possible combinations of two-, three-,
four-, and five-factor portfolios using a 90th percentile
cutoff, choosing one factor definition from each group
at a time. For example, a single three-factor portfolio
might combine information on earnings yield (value),
12-month price return (momentum), and gross profit
margin (profitability). We form the multifactor portfo-
lios exactly as described in the theoretical section: Scores
for all characteristics are transformed so that they are dis-
tributed unit normal, and the aggregate score for a given
stock is the sum of the individual factor scores. As with
the single-factor portfolios, the multifactor portfolios
are long-only, equal-weighted combinations of the top
10% of stocks measured using the aggregate factor score.
Our investible universe is the MSCI All Country
World Index over the period February 1995 to December
2016. This is a broad, practical universe of stocks, con-
taining both developed and emerging market equities.
We rebalance portfolios monthly. All results are gross
of trading costs because our purpose is not to create
investible portfolios but to investigate the hypothesis that
factor exposure translates linearly into realized returns
and risk.
7
 Results throughout are gross of transaction costs. In this
article, we focus on testing theoretical predictions rather than gener-
ating realistic portfolios. Note that many implementation consider-
ations (particularly, reduced trading costs) favor bottom-up portfolio
construction. See, for example, DeMiguel et al. (2017) on reductions
in transaction costs and Beck et al. (2016) on single factor trading costs.
92   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
Exhibit 2
Factor Definitions Used in Portfolio Construction
)DFWRU 'HILQLWLRQ
9DOXH (DUQLQJV\LHOG
'LYLGHQG\LHOG
2SHUDWLQJFDVKIORZ\LHOG
0RPHQWXP PRQWKSULFHUHWXUQ
6L[PRQWKSULFHUHWXUQ
PRQWKSULFHUHWXUQODJJHGE\RQHPRQWK
3URILWDELOLW\ )LYH\HDUHDUQLQJVSHUVKDUHJURZWKUDWH
*URVVSURILWPDUJLQ
)RUHFDVWORQJWHUPHDUQLQJVSHUVKDUHJURZWK
6L]H 9DOXHWUDGHGLQ86'
0DUNHWFDSLWDOL]DWLRQLQ86'
7RWDODVVHWVLQ86'
4XDOLW\ 'HEWWRHTXLW\
5HWXUQRQHTXLW\
&DVKUHWXUQRQFDSLWDOLQYHVWHG
Our final piece of data is stock-level returns mea-
sured in US dollars. All stock-level data come from
FactSet. As noted earlier, we generate excess returns
relative to an equal-weighted composite of index con-
stituents rather than to the market portfolio because the
equal-weighted composite has zero net factor exposure
by construction.
Results
Our experimental setup yields a rich set of results
to compare bottom-up with top-down portfolios.
Exhibit 3 shows the summary results for two-, three-,
four-, and five-factor portfolios. Returns are excess
returns over the equal-weighted (factor-neutral) bench-
mark already described. Risk is calculated as tracking
error to the same benchmark. On average, there were
around 2,380 stocks in the universe over this period,
so single-factor portfolios contained an average of 238
stocks, and all bottom-up portfolios held a similar num-
ber.8 The top-down portfolios holding more than one
factor held correspondingly more stocks; the five-factor
top-down portfolios held on average 889 stocks.
8
 The slight difference in average number of stocks held across
these portfolios is accounted for by stocks dropped as a result of
missing data.
Exhibit 3
Performance and Risk Measures, February 1995 to December 2016

1XPEHURI)DFWRUV
    
6LQJOH)DFWRU 7RS'RZQ %RWWRP8S 7RS'RZQ %RWWRP8S 7RS'RZQ %RWWRP8S 7RS'RZQ %RWWRP8S
5HWXUQV         
        
5LVN         
± ± ± ± ± ± ± ± ±
,QIRUPDWLRQ5DWLR         
        
$YHUDJH+ROGLQJV         
$YHUDJH&RUUHODWLRQ ±        
([SRVXUH         
1XPEHURI3RUWIROLRV         

Notes: Figures are based on daily total returns in US dollar terms. Returns, risk, and information ratios are relative to an equal-weighted benchmark
based on the stocks in MSCI All Country World Index. Standard errors are in parentheses.
Source: MSCI All Country World Index.

A key feature of Exhibit 3 is that the average cor- Exhibit 4

relation across factors is low, at only 0.03. This for- Exposure and Excess Return Ratios
tunate circumstance allows us to bring observations
1, 4, and 5 to the data.9 Note that this is an average 1XPEHU ([SRVXUH ([FHVV5HWXUQV
result, not a feature of the experiment. Although this RI)DFWRUV 5RRW. 5DWLR 5DWLR
is analytically convenient for interpreting our empirical    
results, nothing in our experimental design guaranteed    
orthogonality.    
   

Observation 1: Number of Factors

Observation 1 predicted that bottom-up portfo-
lios would generate greater outperformance than the
corresponding top-down portfolios and that (for fac-
tors with average correlation zero) the advantage would
grow with the square root of the number of targeted fac-
tors. This outperformance, according to the prediction,
9
 Note that technically this addresses only observation 1,
which requires merely that the average correlation across factors
be zero. Observations 4 and 5 require the stricter assumption of joint
factor orthogonality to hold exactly. Moreover, all of the observa-
tions relate technically to a single period. In practice, the data are
too noisy to make single period tests practical, which is why period
averages are invariably used. However, correlations change through
time. We deal with this by using the time-series average of the
appropriately calculated correlation.
would be generated by an equal amount of additional
factor exposure.
The observation is strongly supported both quali-
tatively and quantitatively. On average, bottom-up
portfolios outperform top-down portfolios. The advan-
tage grows with the number of factors, from 1.9% in
two-factor portfolios to 5.6% in five-factor portfolios.
Exhibit 4 shows how relative performance, averaged
across portfolios, relates to average exposure and the
K prediction. The fit among all three is strikingly
close. The slightly lower-than-predicted exposure ratio
is consistent with the small positive average correlation
across the factors. Intriguingly, excess returns slightly
exceed those predicted by the excess exposure ratio; this
is the subject of ongoing research.
Recall that the prediction related to the ratio of
excess returns on single portfolios. Averaging realized

Quantitative Special Issue 2019 The Journal of Portfolio Management   93

 Exhibit 5
Excess Returns of Top-Down versus Bottom-Up Portfolio Construction
1XPEHURI)DFWRUV  1XPEHURI)DFWRUV 
 

 






%RWWRP8S

%RWWRP8S









± 

± ±
± ±        ±          
7RS'RZQ 7RS'RZQ

1XPEHURI)DFWRUV  1XPEHURI)DFWRUV 
 

 

 

 
%RWWRP8S

%RWWRP8S

 

 

 

 

 
                
7RS'RZQ 7RS'RZQ

returns across portfolios with a given number of factors
is a way to deal with measurement error in expected
returns. However, we can also examine directly whether,
for a given number of factors, relative excess returns are
aligned with those expected. (This is not possible across
all portfolios unconditionally because the predicted slope
of the relationship differs with the number of factors.)
Exhibit 5 shows the results across portfolios
for a given number of factors. In all cases, there is a
tight relationship between top-down and bottom-up
performance, and the relationship is almost exactly as
predicted by observation 1. The theoretical relationship
is shown as a solid line, and the estimated empirical rela-
tionship (without intercept) is shown as a dashed line. In
all cases, the relationship is tight and highly significant.
A notable feature of Exhibit 5 is the scatter around
the predicted relationship. Judged across the many
tested portfolios, the relationship appears very close,

94   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
particularly given the approximations needed to arrive
at the theoretical prediction. However, as noted earlier,
standard practice is to judge bottom-up and top-down
techniques (or alternative portfolio construction tech-
niques more broadly) using just a handful of portfolios.
The scatter highlights that such practices are likely to
discover spurious results purely as a result of the random-
ness in realized returns.
Observations 2 and 3: Correlations
and Number of Factors
The results so far have focused on the role of the
number of factors in generating superior outperfor-
mance. We now consider the predictions relating to
correlation. These predictions are more difficult to test
than the return predictions because correlation is a prop-
erty of each factor at each point in time but may vary
significantly over time. Pragmatically, we deal with this
problem by taking a simple time-series average of the
exposure correlation for each factor targeted in a given
portfolio.
Exhibit 6 shows graphically the extent to which
our time-series average is an effective proxy for the
overall effect of correlation on portfolio factor expo-
sure. The y-axis represents, for each portfolio, the ratio
of average factor exposure of the bottom-up portfolio
to that of the corresponding top-down portfolio. The
results are significantly noisier than those for outper-
formance, as might be expected, but qualitatively they
map well to the theoretical relationship. The advantage
of bottom-up portfolios is greater for those portfolios
built using factors that are more negatively correlated,
on average. Moreover, this effect is larger when the port-
folio is built from more factors (compare with Exhibit 1).
Exhibit 7 reports the same experiment but replaces
exposures with relative outperformance. This relation-
ship is somewhat weaker again because the effect of the
underlying randomness in portfolio returns is magnified
when taking the ratio of outperformance. Nevertheless,
the broad shape of the function, and the relative slope
as the number of factors changes, remains consistent
with the theoretical prediction.
Observation 4: Portfolio Risk
Observation 4 predicted that, for orthogonal fac-
tors, risk in bottom-up portfolios would remain constant
Quantitative Special Issue 2019
as the number of factors increased, whereas risk in
top-down portfolios would decline at a rate less than
1/ K . Exhibit 3 shows that this is indeed the case: Risk
is almost identical on average across the bottom-up port-
folios as the number of factors changes, whereas average
risk for top-down portfolios declines with the number
of factors. Exhibit 8 shows these relationships graphi-
cally, along with the 1/ K line, all indexed to a value
of unity for the average of single-factor portfolios. As
predicted, average risk for top-down portfolios declines
at a slower rate than 1/ K , and the rate of divergence
grows as the number of factors increases.
The prediction of observation 5 was based on the
declining rate at which new stocks are added to the
top-down portfolio as the number of factors increases
because of overlapping holdings across the underlying
single-factor portfolios. It is, therefore, valuable to check
whether the number of stocks held in the portfolios cor-
responds with the predicted number. For bottom-up
portfolios, the number of stocks held should be invariant
to the number of factors targeted, and it is.
For top-down portfolios, Exhibit 3 shows the
average number of stocks held in each set of portfolios.
Using 238 as the base, we can predict how many stocks
the top-down portfolios should hold, on average, if fac-
tors are orthogonal. The results of this exercise, along
with the corresponding actual number of stocks held,
are shown in Exhibit 9. Again, there is a close cor-
respondence. That the portfolios actually hold slightly
fewer stocks than they should is consistent with the small
positive average correlation across the factors.
Observation 5: Risk-Adjusted Returns
Given all the preceding, and in particular the
results concerning returns and risk, observation 5 fol-
lows immediately: Risk-adjusted returns are higher in
bottom-up portfolios, and the difference grows with the
number of targeted factors. The notable feature of this
result is the quantitative importance of the idiosyncratic
risk mechanism that slows the rate of risk reduction in
the top-down portfolios. Risk adjusted, the bottom-up
portfolios outperform by less than 20% in two-factor
combinations but by 47% in five-factor combinations.
Finally, Exhibit 10 shows risk-adjusted returns of
bottom-up portfolios relative to top-down portfolios
grouped by number of factors (analogous to Exhibit 5).
In this case, the theoretical line is simply the 45-degree
The Journal of Portfolio Management   95
 Exhibit 6
Average Correlation and Average Relative Factor Exposure
Empirical Evidence
versus
E(F{K}) K
Theoretical: =
EK(E(Fj)) 1+ (K–1)ρ

Number of Factors = 2 Number of Factors = 3

2.5 3.5

3
2
2.5

1.5
2
Exposures

Exposures
1.5
1

1
0.5
0.5

0 0
–0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 –0.4 –0.3 –0.2 –0.1 0 0.1 0.2 0.3 0.4 0.5
Average Correlation Average Correlation

Number of Factors = 4 Number of Factors = 5

4.5 6

4
5
3.5

3 4
Exposures

Exposures

2.5
3
2

1.5 2

1
1
0.5

0 0
–0.3 –0.2 –0.1 0 0.1 0.2 0.3 –0.25 –0.2 –0.15 –0.1 –0.05 0 0.05 0.1 0.15 0.2
Average Correlation Average Correlation

96   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
 Exhibit 7
Ratio of Excess Returns versus Correlation
1XPEHURI)DFWRUV  1XPEHURI)DFWRUV 
 




5DWLRRI([FHVV5HWXUQV

5DWLRRI([FHVV5HWXUQV










 
± ± ±      ± ± ± ±      
$YHUDJH&RUUHODWLRQ $YHUDJH&RUUHODWLRQ

1XPEHURI)DFWRUV  1XPEHURI)DFWRUV 
 




5DWLRRI([FHVV5HWXUQV

5DWLRRI([FHVV5HWXUQV

 





 





 
± ± ±     ± ± ± ± ±     
$YHUDJH&RUUHODWLRQ $YHUDJH&RUUHODWLRQ

line ref lecting the base case of equal risk-adjusted return
(equivalently, no overlapping holdings). The growing
gap between the empirical result and the base-case result
as the number of factors grows confirms the key insight.
CONCLUSION
We have derived an analytically tractable formula
relating factor exposure in simple bottom-up portfolios
to that in corresponding top-down portfolios. We then
used this expression to examine whether the ex post
performance of these portfolios matched their ex ante
factor exposure. Rather than focusing on just one or
two factor portfolios, we examined data on a total of
1,008 matched portfolios using various factor defini-
tions to abstract from an otherwise inevitable focus on
particular factors, factor definitions, or portfolio con-
struction techniques.
The experimental data strongly support the
contention that factor returns are approximately linear

Quantitative Special Issue 2019 The Journal of Portfolio Management   97

Exhibit 8
Relative Changes in Portfolio Risk








&RQVWUXFWLRQ

%RWWRP8S 7RS'RZQ URRW.


    

Exhibit 9
Top-Down Stock Holdings: Empirical and Theoretical
1400

1200

1000

800

600

400

200
No overlaps Predicted Empirical
0
1 2 3 4 5

in factor exposures. In particular, rather than per-
forming qualitative tests of the relative outperformance
of bottom-up portfolios, we were able to examine quan-
titatively how the outperformance of bottom-up factor
portfolios relative to their top-down counterparts was
affected by the number of factors targeted and the cor-
relation between the factors.
Our findings are highly relevant for all those
engaged in factor investing; in particular, our findings
pose a challenge for practitioners of single-factor smart
betas. Moreover, our investigative framework provides a
novel benchmark against which to judge whether there
may be nonlinearities in expected returns to particular
factors or to particular combinations of factors. We hope
that this framework plays its part in pushing financial
research toward the use of large-scale experiments using
many candidate factors, rather than the current focus on
particular combinations of one or two factors among the
hundreds available.

98   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
Exhibit 10
IR of Top-Down versus Bottom-Up Portfolio Construction
Number of Factors = 2 Number of Factors = 3
2.5 2.5

2
2

1.5
1.5
Bottom–Up

Bottom–Up
1
1
0.5

0.5
0
–1 –0.5 0 0.5 1 1.5 2 2.5

–0.5 0
–0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

–1 –0.5
Top–Down Top–Down

Number of Factors = 4 2.5 Number of Factors = 5

2.5

2
2

1.5
1.5
Bottom–Up

Bottom–Up

1
0.5

0.5
0

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Top–Down Top–Down

TECHNICAL APPENDIX Under these assumptions, tracking error in the top-

down portfolio is
Under the maintained assumptions, single stock vari-
ance (here, as throughout, relative to the benchmark) is 1
 N  ηi  2 K  IMRσ j  2  2
σTD = ∑  + ∑  K  
 i =1  N TD 
2
 K
 j =1 
σ =  ηi + ∑sij σ j 
2
i 1
 j =1   η2 IMR 2σ 2j  2
= i +
N K 
in which we have suppressed time subscripts, and hi is idio- TD

syncratic risk. For tractability, we now assume that all factors

are orthogonal.

Quantitative Special Issue 2019 The Journal of Portfolio Management   99

ACKNOWLEDGMENTS
The author gratefully acknowledges insights from
conversations with colleagues in Schroders’ Multi-Asset
Team. Special thanks are due David Callow, for his
outstanding contributions to statistical analysis, and Pro-
fessor Stephen Satchell, for his generous suggestions and
assistance with notation. Remaining errors are those of
the author alone.
REFERENCES
Amenc, N., F. Ducoulombier, M. Esakia, F. Goltz, and
S. Sivasubramanian. 2017. “Accounting for Cross-Factor
Interactions in Multifactor Portfolios without Sacrificing
Diversification and Risk Control.” The Journal of Portfolio
Management 43 (5): 99–114.
Asness, C. 1997. “The Interaction of Value and Momentum
Strategies.” Financial Analysts Journal 53 (2): 29–36.
Beck, N., J. Hsu, V. Kalesnik, and H. Kostka. 2016. “Will
Your Factor Deliver? An Examination of Factor Robustness
and Implementation Costs.” Financial Analysts Journal 72 (5):
58–82.
Bender, J., and T. Wang. 2016. “Can the Whole Be More
than the Sum of the Parts? Bottom-Up versus Top-Down
Multi-Factor Portfolio Construction.” The Journal of Portfolio
Management 42 (5): 39–50.
100   On the Theory and Practice of Multifactor Portfolios Quantitative Special Issue 2019
Clarke, R., H. de Silva, and S. Thorley. 2016. “Fundamen-
tals of Efficient Factor Investing.” Financial Analysts Journal
72 (6): 9–26.
DeMiguel, V., A. Martin-Utrera, F. Nogales, and R. Uppal.
“A Portfolio Perspective on the Multitude of Firm Charac-
teristics.” CEPR Discussion Paper No. DP12417, 2017.
Fitzgibbons, S., J. Friedman, L. Pomorski, and L. Serban.
2017. “Long-Only Style Investing: Don’t Just Mix, Inte-
grate.” The Journal of Investing 26 (4): 153–164.
Freyberger, J., A. Neuhierl, and M. Weber. “Dissecting Char-
acteristics Nonparametrically.” NBER Working Paper No.
23227, 2017.
Ghayur, K., R. Heaney, and S. Platt. “Constructing Long-
Only Multi-Factor Strategies: Portfolio Blending versus
Signal Blending.” Working paper, Goldman Sachs, 2016.
Kelly, B., S. Pruitt, and Y. Su. “Characteristics Are Covari-
ances: A Unified Model of Risk and Return.” Working paper,
2018.
Liu, D. 2017. “Pure Quintile Portfolios.” The Journal of
Portfolio Management 43 (5): 115–129.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

Jack Davies
is a vice president and
portfolio manager for the
Factor Based Strategies
Group at BlackRock, Inc.
in London, UK.
jack.davies@blackrock.com
Dave Gibbon
is a managing director and
head of investment strategy
in EMEA for the Factor
Based Strategies Group
at BlackRock, Inc. in
London, UK.
david.gibbon@blackrock.com
Sara Shores
is a managing director and
the head of investment
strategy for the Factor
Based Strategies Group at
BlackRock, Inc. in
San Francisco, CA.
sara.shores@blackrock.com
Josephine Smith
is a director and senior
researcher for the Factor
Based Strategies Group at
BlackRock, Inc. in
San Francisco, CA.
josephine.smith@blackrock.com
Quantitative Special Issue 2019
Implementation Matters:
Relaxing Constraints
Can Improve the Potential
Returns of Factor Strategies
Jack Davies, Dave Gibbon, Sara Shores,
and Josephine Smith
M
uch of the asset pricing lit-
erature has examined the
returns of factor strategies
by constructing hypothetical
portfolios formed by ranking or optimizing
securities based on stock attributes (e.g., past
returns, valuation ratios) associated with
excess returns.1 This voluminous academic
literature, however, tends to study return
strategies that are usually not directly invest-
ible. In the real world, investors face a variety
of investment constraints largely governed by
the choice of investment vehicle. Each invest-
ment vehicle carries different restrictions that
affect the returns earned by an investor.
The goal of this article is to quantify
the impact of the investment constraints by
showing how well-known investment strate-
gies could potentially fare in representative
examples of each vehicle type. At one end
of the spectrum of investment vehicles are
exchange-traded funds (ETFs), which are
most often based on a low-turnover, fully
transparent third-party index. ETFs are
traded on an exchange—like a stock—but
most indexes are long only and, because they
are publicly disseminated, rarely ref lect novel
1
 See Ang (2014) for a comprehensive literature
summary on factor investing.
or proprietary investment signals. At the
other end of the spectrum, private funds often
have few constraints. Private funds may take
on leverage, take more frequent rebalancing
decisions, and selectively disclose informa-
tion. Between these extremes, institutional
investors may have access to a host of dif-
ferent vehicles, such as collective investment
funds, separately managed accounts, funds of
one, and mutual funds.
We illustrate the effects of the con-
straints implied by the selection of different
investment vehicles on the potential return
stream available to an investor using two
equity factor strategies—momentum and
value—following Jegadeesh and Titman
(1993) and Fama and French (1993), respec-
tively, among many others. In particular, we
investigate the return patterns of two families
of momentum and value strategies: market
neutral (or absolute return) and benchmark
relative.
We start with a hypothetical long–short,
market-neutral factor portfolio, with a con-
struction that is most suited to a private fund
investment vehicle. These investment vehi-
cles are unregistered under the Investment
Company Act of 1940 (henceforth 40-Act)
and tend to have the fewest constraints on
The Journal of Portfolio Management   101
investments.2 However, the general public is not eligible
for these types of private fund investments because they
are unregistered. Access to private funds is restricted to
institutional investors, individuals who are accredited
investors, or qualified purchasers. In the hypothetical
portfolio representing a private-fund-like vehicle, the
long–short momentum and value strategies have Sharpe
ratios of 0.99 and 1.28, respectively.
Although private funds are widely adopted by
institutions across North America and parts of Asia
Pacific, most European investors prefer undertakings for
collective investment in transferable securities (UCITS)
funds, which typically carry more constraints on risk
and leverage than do private fund vehicles.3 To represent
these UCITS vehicles, we restrict gross leverage of the
investment strategies to four times or lower.4 We find
that, although the realized returns of the momentum and
value strategies implemented via representative UCITS
constructions are lower than the less-constrained pri-
vate fund implementations, the Sharpe ratios are largely
unchanged. This illustrates that the use of leverage can
affect the level of risk that can be taken but may have
little impact on risk-adjusted returns.
Moving away from the market-neutral portfolios,
we also examine benchmark-relative 130/30 and long-
only implementations of the factors that would be typical
of mutual funds and ETFs.5 With a preponderance of
long market exposure, these strategies necessarily take on
a large amount of market risk. Although the majority of
mutual funds are long only, a minority (close to 30%) of
mutual funds may take on limited short positioning up
to 130/30 in leverage terms (see Almazan et al. 2004).
Mutual funds and ETFs are registered 40-Act vehicles
and are thus required to periodically report holdings
2
 Although exempt from the 40-Act, private funds must
comply with other laws and regulations, including the 1933 Securi-
ties Act and other anti-fraud and financial regulations. Some private
funds are called Cayman funds, referring to the jurisdiction of the
formation of these funds (see Logue 2007; Scharfman 2015).
3 trade-offs, leverage, and position holdings.
 In simple terms, a UCITS is a mutual fund based in the
European Union. UCITS funds can be sold to any investor within
the European Union under a harmonized regulatory regime.
4
 The restriction of 4× was selected to be representative of a
leverage cap that would add a meaningful constraint.
5
 An exception is the class of leveraged ETFs, which we do not
examine here (see Cheng and Madhavan 2009). A smaller number
of ETFs, called active transparent ETFs, can take investment deci-
sions that do not follow a third-party index. We do not consider
these variants in this article.
102   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
and other fund information. Mutual funds, however,
typically take more frequent rebalancing decisions and
have higher turnover and higher active risk relative to a
given index than ETFs (see Madhavan 2016). To illus-
trate the effect of the benchmark-relative vehicles on
the momentum and value factors, we construct repre-
sentative versions of these strategies that impose leverage
limits or long-only constraints with monthly or semian-
nual rebalance frequencies.
We find that both the leverage and rebalancing
constraints significantly curtail performance of the
hypothetical 130/30 and long-only strategies when com-
pared to the unconstrained private fund implementation.
Comparing the 130/30 portfolios to the long-only ETF
portfolios, the Sharpe ratio decreases from 0.48 to 0.40
for the momentum strategy and from 0.50 to 0.41 for
the value strategy. On the other hand, the more infre-
quent rebalancing decisions decrease turnover, which
falls by approximately 50% from the 130/30 portfolio
to the long-only ETF portfolio. This shows that there
is a trade-off between performance and trading costs,
although we find the decrease in turnover when moving
to the most constrained, long-only ETF strategy does
not offset the expected degradation in performance.
Our article falls into the literature examining the
effects of constraints on investment performance, such
as Clarke, de Silva, and Sapra (2004); Qian, Hua, and
Sorensen (2007); Brière and Szafarz (2017); and others.
Some notable contributions are by Grinold and Kahn
(2000) and Johnson, Kahn, and Petrich (2007), who pre-
sented theoretical frameworks in which relaxing leverage
may be beneficial. We focus on implementable factor
strategies within a well-defined, liquidity-screened uni-
verse that generally has greater breadth than in these
studies and apply the effects of typical investment con-
straints to different investment vehicles. We present the
implications of these constraints for factor strategies in
historical simulations for a large global stock universe
and show they result in large differences in risk–return
HYPOTHETICAL FACTOR STRATEGIES
In this section, we build theoretical, paper port-
folios of momentum and value factor strategies in a
global equity universe. These provide a baseline case to
illustrate the impact of different investment vehicles on
factor performance.
Exhibit 1
Factor Strategies and Vehicle Types
3RUWIROLR %HQFKPDUN 7\SLFDO,QYHVWRU
3ULYDWH)XQG &DVK ,QVWLWXWLRQDO
4XDOLILHG,QYHVWRUV
8&,76 &DVK (8,QVWLWXWLRQDO
,QGLYLGXDO,QYHVWRUV
 0DUNHW $Q\RQH
/RQJ2QO\ 0DUNHW $Q\RQH
/RQJ2QO\(7) 0DUNHW $Q\RQH
Notes: This exhibit is for illustration purposes only. Characteristics listed are for typical portfolios. The last column reflects the specific constraints applied
to the hypothetical strategies herein; actual constraints may differ significantly depending on the specific investment or account.
Data Sources
The paper portfolios for our analysis rely on data from
several sources. We use fundamental data from Worldscope
and IBES to generate our insights on momentum and
value. For momentum, we also use equity returns sourced
from the MSCI Barra Global Equity Model (GEM2).
GEM2 also provides us with the risk model used for the
simulated portfolio construction exercise. On average,
the trading universe includes 3,000 individual equities
across developed and emerging markets that have been
prescreened along liquidity dimensions such as market cap
and average daily volume. The historical simulation runs
from January 1998 through June 2017 and rebalances on
a monthly basis, save for the most constrained long-only,
ETF strategy, which will rebalance semiannually.
Factor Strategies
Momentum and value are recognized as broad and
persistently rewarded style factors. At a high level, we
construct momentum combining two insights on trends:
a standard price trend over the past 12 months (removing
the most recent month to control for reversal effects,
following Jegadeesh and Titman 1993) and changes (up
or down) in analyst 12-month earnings forecasts, fol-
lowing Ball and Brown (1968), Gleeson and Lee (2003)
and others. The value factor combines three insights.
The first is a backward-looking measure of firm valua-
tion, comparing a firm’s cash f low from operations to its
market capitalization. The other two valuation metrics
are forward-looking, using analyst 12-month earnings
Quantitative Special Issue 2019
7\SLFDO0LQLPXP 1RWHVRQ5LVN
,QYHVWPHQW /HYHUDJH7UDGLQJ
 ([DQWHULVNWDUJHWRI
 ([DQWHULVNWDUJHWRI
JURVVOHYHUDJHOLPLWRIî
9DULHV *URVVOHYHUDJH
OLPLWRIî
9DULHV 1ROHYHUDJH
± 1ROHYHUDJHVHPLDQQXDO
UHEDODQFLQJ
forecasts as an input. We examine both forward earnings
yield (the ratio of forward earnings to price) and the ratio
of forward earnings to enterprise value.
We construct hypothetical portfolios for the
momentum and value factors using our global stock uni-
verse. We construct the portfolios using mean–variance
optimization, without transaction costs, using the Barra
GEM2 risk model. We impose region and sector neu-
trality to focus on taking active risk only along the factor
dimension. These region and sector constraints mean
we only compare like with like—meaning we compare
companies in a certain sector or region with those in the
same sector or region.6 For the active optimizations, we
impose cash neutrality of the final holdings (the capital
allocation of the long side is equal to the capital allo-
cation of the short side). Constraints specific to each
vehicle are described in detail in the following.
Factor Strategy Implementations
and Vehicle Types
Exhibit 1 lists the investment vehicles we consider,
their benchmarks, liquidity and legal characteristics, and
typical leverage and turnover statistics. We divide the
vehicles into two types: market-neutral and benchmark-
relative strategies. Investors can pursue market-neutral
strategies with more f lexible private fund investment
vehicles because they allow offsetting of long–short
6
 The region and sector constraints also mean that perfor-
mance f luctuations are due to factor bets within regions and sectors
rather than across them.
The Journal of Portfolio Management   103
positions to minimize aggregate market exposures.
We define more constrained 130/30 funds, long-only
mutual funds, and long-only ETFs as benchmark-rela-
tive strategies, where active risk is taken relative to the
market benchmark but the final portfolio has a large
market exposure primarily because of the long-only (or
long-biased) constraint.
We emphasize that the characteristics listed in
Exhibit 1 are for typical portfolios; actual constraints
in the real world may differ significantly. Hereafter we
use the terms private fund, UCITS, 130/30, long-only
(implemented as a mutual fund), and long-only ETF to
correspond to these sets of constraints, which are listed
in the last column in Exhibit 1. Our goal is to show the
effect of the constraints on the simulated returns of the
momentum and value strategies and thereby to illustrate
how the risk and return characteristics of implementing
the factor strategies change with the constraints that
typically correspond to different investment vehicles.
Market-neutral strategies. We design the
market-neutral strategies to provide long-term positive
returns irrespective of market moves, without taking
outsized nonfactor bets. We construct two market-neu-
tral factor strategies, each rebalanced monthly:
• Private fund: an optimized long–short portfolio
in the spirit of an unconstrained private fund. We
scale the factor strategies each period to an ex ante
risk target of 10% with no leverage constraints.
• UCITS: an optimized long–short portfolio
imposing leverage constraints typically seen in
UCITS investment vehicles. We scale the portfolio
each period to ex ante risk of 7% and impose a
maximum gross leverage of 4×.
Because both of these strategies are long–short,
the effect of the ex ante risk constraints is to indirectly
impose leverage constraints, on top of the other con-
straints on sector and region neutrality constraints in the
portfolio optimization.7
Benchmark-relative strategies. We build
the benchmark-relative strategies with the aim of
7
 The ex ante risk targets of 10% for the private fund and 7%
for the UCITS were chosen for this analysis because they are both
within the range of risk targets used in the industry and are common
risk targets for these types of strategies. See footnote 4 for discussion
of the leverage choice.
104   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
outperforming a (generally market-capitalization-
weighted) benchmark. These portfolios have ex ante
market betas close to one, given their restriction to take
limited, if any, leverage. We compare three types of
benchmark-relative portfolios:
• 130/30: an optimized portfolio that ref lects a
common leverage limit offered in several fund
types, including mutual funds. The maximum
gross leverage of this type of portfolio is 1.6×, and
we rebalance monthly.
• Long-only: an optimized portfolio that can take
zero leverage (no short positions) and is rebalanced
monthly.
• Long-only ETF: an optimized long-only portfolio,
but we only allow for semiannual rebalancing (in
January and July for the sake of illustration). This
is a typical format observed in ETFs, where the
index construction rules must specify predeter-
mined rebalancing dates.
We can interpret the benchmark-relative strategies
as comprising the market benchmark (a beta of one) plus
a constrained long-short portfolio. The latter ref lects the
active risk produced by taking weights different from
the market portfolio. The market exposure of these
benchmark-relative strategies means that there will be
lower diversification benefits in most investors’ port-
folios holding these types of strategies as compared to
private fund and UCITS strategies, which are designed
to remove market beta exposure.
PERFORMANCE OF HYPOTHETICAL
STRATEGIES IMPLEMENTED
WITH DIFFERENT CONSTRAINTS
We discuss the performance of the hypothetical
factor strategies implemented with constraints typical
of the different investment vehicles by separately con-
sidering the market-neutral strategies (private fund and
UCITS) and the benchmark-relative strategies (130/30,
long-only, and long-only ETF) in the next two subsec-
tions, respectively. We find that implementation choices
matter significantly for performance. In the third sub-
section, we investigate the cross-sectional correlations
of the strategies. Positions have low correlation across
the implementations, even in the rare cases when per-
formance is similar. Varying levels of leverage are an
important feature across the investment vehicles, which Exhibit 2
we separately investigate in the fourth subsection. Factor Strategy Performance in Hypothetical
Market-Neutral Private Fund Strategies
Market-Neutral Factor Strategies
3ULYDWH)XQG 8&,76
Exhibit 2 reports portfolio performance statistics 3DQHO$0RPHQWXP6W\OH)DFWRU
for the hypothetical market-neutral momentum and 5HDOL]HG5HWXUQ  
value strategies. We analyze a variety of metrics: real- 7DUJHW5LVN  
ized return, target and realized risk, Sharpe ratio, rela- 5HDOL]HG5LVN  
6KDUSH5DWLR  
tive turnover (turnover per unit gross leverage), gross
5HODWLYH7XUQRYHU î î
leverage, maximum drawdown, and skewness. All risk $YHUDJH*URVV/HYHUDJH î î
and return numbers are in average annualized percent- 0D['UDZGRZQDW5LVN ± ±
ages and are stated in excess of cash, relative turnover and 0D['UDZGRZQ5DQJH -XQH± -XQH±
gross leverage are averages over the simulation period, 6HSWHPEHU 6HSWHPEHU
and the maximum drawdown is calculated assuming 6NHZQHVVDW5LVN ±


±



returns are scaled to 7% ex post risk.8 We also compare 3DQHO%9DOXH6W\OH)DFWRU
the maximum drawdown statistics of the factor strate- 5HDOL]HG5HWXUQ  
gies scaled to 7% risk with the maximum drawdown of 7DUJHW5LVN  
5HDOL]HG5LVN  
a Monte Carlo simulation of 10,000 normally distrib-
6KDUSH5DWLR  
uted return series over the same sample size, where the
5HODWLYH7XUQRYHU [ [
mean and variance are chosen to match the documented $YHUDJH*URVV/HYHUDJH [ [
Sharpe ratio and 7% risk. This allows us to gauge, along 0D['UDZGRZQDW5LVN ± ±
with the skewness, how close these strategy returns are 0D['UDZGRZQ5DQJH -XO\± $XJXVW±
to normally distributed returns. As an example, for 7% )HEUXDU\ )HEUXDU\
annualized risk and a Sharpe ratio of 0.5, the normal 6NHZQHVVDW5LVN ± ±
distribution has an annualized mean assumption of 3.5%
Notes: The simulation period is January 1998 through June 2017.
and an annualized standard deviation of 7%. Returns and risk are in excess of the cash rate. Return and risk num-
Market-neutral momentum. In Panel A of bers (both realized and drawdown) are reported in average annualized
Exhibit 2, the realized returns and risk of the private percentage. Target risk reflects the optimization constraint of the strategy
achieving a given level of target ex ante risk each period. The Sharpe ratio
fund momentum strategy are 13.9% and 14.1%, respec- is calculated as the ratio of the realized return of the strategy divided by
tively, implying a Sharpe ratio of 0.99. Relative turnover the realized risk of the strategy. The relative turnover is the average turn-
is 3.3×, and gross leverage averaged 5.3×. over per unit of leverage. Average gross leverage is calculated as the average
The observed maximum drawdown of the private sum of the absolute holdings of the strategy. The maximum drawdown
represents the peak-to-trough drawdown and is calculated after scaling each
fund momentum strategy at 7% risk is -24.3%, which strategy to ex post risk of 7%. The skewness statistic is computed on the
occurs from June 2008 through September 2009. This 7% ex post risk-scaled returns for each strategy. Past performance is not
is nearly twice the maximum drawdown of -12.3% for indicative of future results. Returns do not account for any fees or transac-
tion costs. If fees were included, returns would be lower. For illustrative
a simulated normal distribution. This is not surprising; purposes only using hypothetical strategies and not representative of an
momentum strategies have been well documented to actual investment or account. The modeled performance is calculated with
exhibit sharp drawdowns and significant downside risk the benefit of hindsight and knowledge of factors that may have positively
affected its performance and has inherent limitations. This analysis cannot
(see, for example, Daniel and Moskowitz 2016). The account for risk factors that may affect actual portfolio performance.
downside risk of momentum is also ref lected in the *** denotes rejection of the null hypothesis that the risk-scaled returns
skewness statistic of -0.8, which is statistically different have skewness of a corresponding normal distribution at the 1% level.
from zero at the 1% level.
Moving to the more constrained UCITS momen-
8
 We scale to 7% ex-post risk for the drawdown analysis to be tum strategy, still in Panel A, the realized return and
able to compare the private fund and UCITS strategies at similar risk numbers fall to 9.5% and 9.9%, respectively, which
risk levels. We perform a similar exercise for the benchmark-relative
are both lower than the private fund implementation.
strategies.

Quantitative Special Issue 2019 The Journal of Portfolio Management   105

However, the implied Sharpe ratio decreases only
slightly to 0.96, which is almost unchanged from the
private fund Sharpe ratio of 0.99. Thus, the use of
leverage in this case is approximately linear—that is,
leverage changes both returns and risk by the same
factor so that the Sharpe ratio is unchanged. Whereas
relative turnover is roughly the same as the private fund
strategy, the gross leverage falls to 3.4× as a result of the
leverage constraint and the ex ante 7% risk target. The
maximum drawdown, skewness, and normal distribu-
tion statistics of the UCITS strategy is also comparable
to that of the private fund strategy—both -24% and
-0.8, respectively.
Market-neutral value. Turning to value in
Panel B of Exhibit 2, the private fund value strategy
exhibits realized returns and risk of 17.0% and 13.3%,
respectively. The Sharpe ratio of the private fund value
strategy is 1.28. Relative turnover is 1.6×, approximately
half of the momentum strategy, whereas gross leverage
is higher at 8.9×. These results are consistent with most
implementations of value strategies: They generally have
lower turnover than momentum. Momentum strategies
are based on the extrapolation of short-term perfor-
mance (buying recent winners and selling recent losers),
and relative performance can be volatile. On the other
hand, value strategies are based on longer-run measures
of valuations (buying cheap and selling expensive) and
rely on low prices slowly reverting to normal values.
Thus, momentum and value tend to have negative cor-
relations; over the simulation period, the returns of
the momentum and value strategies have a correlation
of -0.23.
Value also tends to have less downside risk than
momentum: At 7% ex post risk, the maximum drawdown
of the value strategy is -20.8%, compared to -24.4%
for momentum. Although the skewness statistic has a
negative point estimate at -0.2, it is insignificantly dif-
ferent from zero. The maximum drawdown period for
value occurs from July 1999 to February 2000. Most of
the regional underperformance came from within the
United States and Europe, and sector underperformance
came from information technology and consumer dis-
cretionary goods.
In the second column in Panel B of Exhibit 2, the
UCITS value strategy has lower return and risk (11.2%
and 8.8%, respectively), although its Sharpe ratio of 1.27
is nearly identical to that of the private strategy. The
UCITS value strategy shows that lower leverage does
106   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
not automatically result in lower turnover. In fact, turn-
over for the UCITS value strategy is higher, at 2.1×,
than the turnover for the private fund strategy, which
is 1.6×. The intuition is that, with more constrained
positioning resulting from tighter leverage constraints,
the optimizer may rebalance more often to maximize
investment opportunities.
Cumulative returns of market-neutral
momentum and value. Exhibit 3 graphs the cumu-
lative returns of the hypothetical momentum and value
strategies on a logarithmic scale. The private fund and
UCITS versions have similar Sharpe ratios (see Exhibit 2),
but because the private fund strategy has a higher real-
ized risk by construction, its cumulative returns are
higher than those of the UCITS implementation.
Starting with momentum in Panel A, the strate-
gies have delivered strong performance over the simu-
lation period of January 1998 to June 2017, but there
are periods of notable underperformance. In particular,
both the global financial crisis and the early part of 2016
saw momentum drawdowns. The strongest periods of
momentum performance were in the expansionary
periods of 2003 through mid-2008 and in 2009 through
2015, after the financial crisis. These results are also con-
sistent with those of Hodges et al. (2017), who showed
that momentum has tended to outperform in both
recovery and expansionary regimes of the economic
cycle and to suffer during contractions.
Panel B of Exhibit 3 graphs cumulative returns
of the market-neutral value strategies. Value has been a
consistent performer with limited losses since the max-
imum drawdown from August 1999 through February
2000. The only regime in which value has tended to
exhibit consistently weak conditional performance is the
slowdown period after an economic expansion.
Benchmark-Relative Factor Strategies
We now turn to the hypothetical benchmark-rel-
ative factor strategies: 130/30, long only, and long-only
ETF, which we report in Exhibit 4. These are similar
statistics to the metrics we report for the market-neutral
strategies (see Exhibit 2), but we look at returns and
risk of the total portfolio in excess of cash as well as the
active portfolio in excess of the cap-weighted bench-
mark of the investment universe. We compute max-
imum drawdown statistics, scaling the returns to 15% ex
post risk. (We use a 15% ex post risk target because this
Exhibit 3
Cumulative Returns of Hypothetical Market-Neutral Private Fund Strategies
3DQHO$0RPHQWXP6W\OH)DFWRU





&XPXODWLYH/RJ5HWXUQV











±
      

3DQHO%9DOXH6W\OH)DFWRU





&XPXODWLYH/RJ5HWXUQV











±
      

3ULYDWH)XQG 8&,76

Notes: Past performance is not indicative of future results. Returns do not account for any fees or transaction costs. If fees were included, returns would be
lower. For illustrative purposes only using hypothetical strategies and not representative of an actual investment or account. The modeled performance is
calculated with the benefit of hindsight and knowledge of factors that may have positively affected its performance and has inherent limitations. This analysis
cannot account for risk factors that may affect actual portfolio performance.

Quantitative Special Issue 2019 The Journal of Portfolio Management   107

Exhibit 4
Factor Strategy Performance in Hypothetical Benchmark-Relative Strategies
3DQHO$0RPHQWXP6W\OH)DFWRU
 /RQJ2QO\ /RQJ2QO\(7)
7RWDO5HWXUQV
5HDOL]HG5HWXUQ   
5HDOL]HG5LVN   
6KDUSH5DWLR   
5HODWLYH7XUQRYHU î î î
0D['UDZGRZQDW5LVN ± ± ±
0D['UDZGRZQ5DQJH 2FWREHU± 2FWREHU± 2FWREHU±
)HEUXDU\ )HEUXDU\ )HEUXDU\
6NHZQHVVDW5LVN ± ± ±
5HWXUQVLQ([FHVVRI&DS:HLJKWHG%HQFKPDUN
5HDOL]HG5HWXUQ   
5HDOL]HG5LVN   
6KDUSH5DWLR   

3DQHO%9DOXH6W\OH)DFWRU
 /RQJ2QO\ /RQJ2QO\(7)
7RWDO5HWXUQV
5HDOL]HG5HWXUQ   
5HDOL]HG5LVN   
6KDUSH5DWLR   
5HODWLYH7XUQRYHU î î î
0D['UDZGRZQDW5LVN ± ± ±
0D['UDZGRZQ5DQJH 2FWREHU± 2FWREHU± 0D\±
)HEUXDU\ )HEUXDU\ )HEUXDU\
6NHZQHVVDW5LVN ± ± ±
5HWXUQVLQ([FHVVRI&DS:HLJKWHG%HQFKPDUN
5HDOL]HG5HWXUQ   
5HDOL]HG5LVN   
6KDUSH5DWLR   

Notes: The simulation period is January 1998 through June 2017. We analyze performance statistics for both the total portfolio in excess of cash and the
active portfolio in excess of returns of a cap-weighted benchmark of the asset universe. Return and risk numbers (both realized and drawdown) are reported
in average annualized percentage. Each optimization has a risk aversion parameter of 1. The Sharpe ratio is calculated as the ratio of the realized return
of the strategy divided by the realized risk of the strategy. The relative turnover is the average turnover per unit of leverage. The maximum drawdown rep-
resents the peak-to-trough drawdown and is calculated after scaling each strategy to ex post risk of 15%. The skewness statistic is computed on the 15%
ex post risk-scaled returns for each strategy. Past performance is not indicative of future results. Returns do not account for any fees or transaction costs. If fees
were included, returns would be lower. For illustrative purposes only using hypothetical strategies and not representative of an actual investment or account.
The modeled performance is calculated with the benefit of hindsight and knowledge of factors that may have positively affected its performance and has
inherent limitations. This analysis cannot account for risk factors that may affect actual portfolio performance.
*** denotes rejection of the null hypothesis that the risk-scaled returns have skewness of a corresponding normal distribution at the 1% level.

is approximately the same volatility as the market port-
folio during the simulation period.) The market-neutral
strategies have insignificant market direction as a result
of the long–short portfolio construction. In contrast,
the major component of the benchmark-relative factor
returns is the market return. We compare the realized
maximum drawdowns with a normal distribution, with
Monte Carlo simulations similar to those examined for
the market-neutral strategies, but with volatility cali-
brated to 15% risk. For the portfolio characteristics in
excess of the cap-weighted benchmark, we report real-
ized return, risk, and Sharpe ratios.

108   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
 Benchmark-relative momentum. Panel A of
Exhibit 4 focuses on the benchmark-relative momentum
strategies. With a modest degree of leverage, the 130/30
strategy has the highest return of all three benchmark-
relative momentum strategies across the sample, posting
a return, risk, and Sharpe ratio of 6.9%, 14.5%, and
0.48, respectively. Relative turnover is 1.9×, and the
maximum drawdown of the strategy scaled to 15%
risk is -54.4%. The maximum drawdown period is
during the global financial crisis from October 2007
to February 2009 and is the same for the long-only
and long-only ETF vehicles. This 130/30 drawdown is
substantially higher than for a comparable normal dis-
tribution, which would have a maximum drawdown
of -35.9%, which is consistent with the large downside
risk of the momentum factor (see Exhibit 2). The skew-
ness of this momentum strategy is also statistically sig-
nificant at -1. Upon removing the benchmark from the
calculations, the active performance exhibits a Sharpe
ratio of 0.79, highlighting both the skill of the 130/30
strategy in allocating around a cap-weighted benchmark
and the benefits of limited leverage.
The two long-only strategies may appear similar at
face value, given that neither takes on leverage, but there
are important differences. We expect that both strategies
should have lower Sharpe ratios (0.43 and 0.40 for the
long-only and long-only ETF, respectively) than the
Sharpe ratio of 0.48 for the 130/30 strategy. Although
Sharpe ratios decline from 130/30 to long-only to the
long-only ETF—consistent with there being more
restrictions on the portfolio—realized risk is approxi-
mately the same. There is a large drop in the turnover of
the long-only ETF strategy, which has a turnover of 0.9×
compared to 1.9× for both the 130/30 and long-only
funds. This is potentially desirable for certain taxable
investors. Panel A shows that the tighter constraints have
not significantly affected the downside risk of imple-
menting the momentum factor in these vehicles, with
similar drawdowns and skewness for all the benchmark-
relative offerings. When moving to active return space,
there is a larger decrease in the Sharpe ratio from the
130/30 strategy at 0.79, to the long-only strategy at 0.50,
and finally to the long-only ETF strategy at 0.38.
Benchmark-relative value. Panel B of Exhibit 4
reports summary statistics for the benchmark-relative
value strategies. Like the momentum strategies in
Panel A, Sharpe ratios decrease as leverage decreases
and rebalancing decisions are taken less frequently.
Quantitative Special Issue 2019
Sharpe ratios range from 0.50 for 130/30, to 0.44 for
long-only, and finally to 0.41 for the long-only ETF.
Turnover falls significantly in the long-only ETF, at
0.6×, compared to the 1.3× turnover for the 130/30 and
long-only vehicles. Consistent with the results for the
market-neutral strategies, turnover is lower for the value
factor (Panel B) than for the momentum factor (Panel
A). As with their momentum counterparts, the active
performance falls precipitously when moving from the
130/30 strategy to the long-only ETF strategy. Although
the 130/30 strategy has an active Sharpe ratio of 1.27,
in line with the market-neutral counterparts, the long-
only and long-only ETF portfolios exhibit sharply lower
active Sharpe ratios of 0.83 and 0.39, respectively.
One interesting result is that the benchmark-relative
value strategies exhibit significantly more downside risk
than the market-neutral value strategies (see Exhibit 2).
The maximum drawdown at 15% risk for the bench-
mark-relative value strategies is around -54%.9 This is
much higher than a normal distribution, which would
have a maximum drawdown of -35.9%. The skewness
of the benchmark-relative value strategies is between
-0.8 and -0.9, and the tests overwhelmingly reject the
null hypothesis of a normal distribution. In contrast,
the market-neutral value strategies in Exhibit 2 exhibit
no pronounced skewness.10 This implies that limited
leverage, in this context, has not decreased risk. That
is, we find that using more unconstrained leverage can,
in the case of a value strategy, result in more consistent
portfolio returns.
Cumulative returns of benchmark-relative
momentum and value. Exhibit 5 graphs the log
cumulative returns of the hypothetical benchmark-
relative momentum and value strategies in excess of the
cap-weighted benchmarks in Panels A and B, respec-
tively. Although the 130/30 portfolios outperformed
their long-only counterparts throughout most of the
sample period, there are periods of drawdown com-
monality. The momentum strategies exhibit their largest
9
 The drawdown range for the long-only ETF value strategy
occurs earlier, in May 2007, compared to October 2007 for the
130/30 and long-only value strategies. This is because of the limited
semiannual rebalancing for the long-only ETF.
10
 Part of this is because the benchmark-relative value strate-
gies have inherited the negative skewness present in the market itself
(see French, Schwert, and Stambaugh 1987). Note that the skewness
of the 130/30 strategy is slightly less in absolute value, at –0.8, than
that of the other two long-only strategies.
The Journal of Portfolio Management   109
Exhibit 5
Cumulative Returns of Hypothetical Benchmark-Relative Strategies in Excess of Cap-Weighted Benchmark
3DQHO$0RPHQWXP6WUDWHJLHV



&XPXODWLYH/RJ5HWXUQV







±
      

3DQHO%9DOXH6WUDWHJLHV



&XPXODWLYH/RJ5HWXUQV







±
      

 /RQJ2QO\ /RQJ2QO\(7)

Notes: Past performance is not indicative of future results. Returns do not account for any fees or transaction costs. If fees were included, returns would be
lower. For illustrative purposes only using hypothetical strategies and not representative of an actual investment or account. The modeled performance is
calculated with the benefit of hindsight and knowledge of factors that may have positively affected its performance and has inherent limitations. This analysis
cannot account for risk factors that may affect actual portfolio performance.

110   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
drawdowns starting in mid-2008 through the end of
2009, whereas the value strategies exhibit their largest
drawdowns through the late 1990s, with the 130/30
strategy exhibiting a significantly shorter drawdown
than the long-only strategies. The long-only portfo-
lios, which rebalance every month, outperformed the
semiannually rebalanced ETFs. The more frequent
rebalancing allows more optimal exposure to the factors.
HOLDINGS AND LEVERAGE
In the first subsection of this part of the article, we
show that although all of the representative implemen-
tations—private fund, UCITS, 130/30, long-only, and
long-only ETF—pursue the same momentum and value
factors, the specific holdings of the hypothetical strate-
gies vary significantly across the different investment
constraints. In the second subsection, we discuss how
leverage varies over time for the factor implementations,
which allow shorting.
Comparing Factor Strategy Holdings
We plot cross-sectional holdings correlations in
Exhibit 6. For each of the factors, we compare the hold-
ings of the two hypothetical market-neutral strategies
and the three hypothetical benchmark-relative strategies.
We compute one correlation series for the market-neu-
tral private fund and UCITS strategies and three correla-
tion series for the benchmark-relative strategies (130/30
and long-only mutual fund, long-only mutual fund and
long-only ETF, and 130/30 and long-only ETF).11
11
 The cross-sectional holdings correlations are computed as
follows. At each month t, there are two portfolios, X = [x(1, t),  market-neutral strategies. The strategies are calibrated to
x(2, t), … , x(n, t), … , x(N, t)] and Y(t) = [y(1, t), y(2, t), … , y(n, t), … , y
(N, t)], each consisting of a set of holdings in n = 1,2, … ,N equities.
Let x (t ) and y (t ) denote the sample means of X(t) and Y(t), and let
sx(t) and sy(t) be the corrected sample standard deviations of X(t)
and Y(t). The means and standard deviations are computed at time
t in the cross section, n = 1,2, … ,N. The cross-sectional holdings
correlation between these two portfolios at time t is calculated as
∑
N
( x(n, t ) −   x (t ))( y(n, t ) −   y (t ))
rXY (t ) =   n =1
. momentum strategies have higher levels of risk than
( N − 1)σ x (t )σ y (t )
For each pair of portfolios (X(t), Y(t)), Exhibit 6 plots the time
series rXY (t). For correlations relative to the long-only ETF portfolio,
we plot a 12-month exponentially weighted moving average of the
correlation to control for the semiannual rebalancing dates. If the
holdings are identical, then the holdings correlations are equal to
one.
Quantitative Special Issue 2019
We start with Panel A of Exhibit 6 and the
momentum strategies. For the market-neutral strate-
gies, the correlation exceeds 0.99 most of the time. The
correlations fall during two periods that coincide with
outperformance of the private fund strategy: November
2005 to January 2008 and October 2014 to the end of
the sample in June 2017.
Although the correlations of the benchmark-relative
momentum strategies are lower than those of the
market-neutral strategies, they are also high, above 0.90.
Overall, the positions in the long-only and the long-
only ETF strategies are most similar. The correlations
for the 130/30 and the long-only ETF are lowest, as
would be expected, given the differences in leverage
and rebalancing frequency.
The cross-sectional correlations of the value strat-
egies, in contrast, can be fairly low. In Panel B, the
market-neutral private fund and UCITS have an average
correlation of 0.92, but there is significant variation,
ranging from a minimum of 0.82 to a maximum of
more than 0.99. The final holdings correlation is 0.93.
Interestingly, correlations tend to fall during bull mar-
kets and rise during bear markets. The low correlation
periods also tend to coincide with times in which the
private fund strategy outperformed the UCITS strategy.
In summary, although the risk and return charac-
teristics of the momentum and value strategies can be
very similar, the actual holdings generating those returns
may sometimes differ markedly.
Time-Varying Leverage
Exhibit 7 examines the time-varying leverage of
the momentum and value strategies for the hypothetical
achieve an ex ante risk of 10% (7%). Note that we place
a constraint on UCITS leverage of 4× per side (gross
leverage of 8×), which is almost always binding for the
value UCITS strategy.
First, the leverage of the value factor is higher
than or equal to the leverage of the momentum strategy
to achieve the same ex ante volatility. This is because
do value strategies. As a result, momentum strategies
require less leverage to hit a given ex ante risk target.
Second, given the constant risk target of these
portfolios, the leverage of each strategy is inversely
related to the volatility of the overall equity market.
The Journal of Portfolio Management   111
Exhibit 6
Cross-Sectional Holdings Correlations by Style Factor
3DQHO$0RPHQWXP6W\OH)DFWRU






+ROGLQJV&RUUHODWLRQ












      

3ULYDWH)XQG8&,76 /RQJ2QO\
/RQJ2QO\/RQJ2QO\(7) /RQJ2QO\(7)

3DQHO%9DOXH6W\OH)DFWRU






+ROGLQJV&RUUHODWLRQ












      

3ULYDWH)XQG8&,76 /RQJ2QO\
/RQJ2QO\/RQJ2QO\(7) /RQJ2QO\(7)

Note: For illustrative purposes only using hypothetical strategies and not representative of an actual investment or account.

112   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
Exhibit 7
Gross Leverage across Style Factors and Hypothetical Market-Neutral Implementation


*URVV/HYHUDJH

   
0RPHQWXP3ULYDWH)XQG
9DOXH3ULYDWH)XQG
Note: For illustrative purposes only using hypothetical strategies and not representative of an actual investment or account.
Times of high market volatility also tend to coincide
with market drawdowns. Given this relation with vola-
tility, the leverage of both momentum and value tend to
be positively correlated; the gross leverage numbers of
the private fund momentum and value strategies have a
correlation of 0.81 over the simulation period.
Third, leverage varies over time, which carries
strong ramifications for risk management capabilities of
factor investors using leverage. The average gross leverage
of the private fund momentum strategy is 5.3×, but it
varies from a minimum of 3.4× to a maximum of 8.6×. Its
leverage also has a negative correlation of -0.72 with the
risk of the overall strategy. The private fund value strategy
exhibits higher average leverage of 8.9×. The minimum
and maximum leverage values are 5.0× in April 2001
and 11.7× in August 2007, respectively. Like momentum,
the private fund value leverage has a negative correlation
(-0.74) with the risk of the strategy itself.
CONCLUSION
In the real world, the practical decisions of invest-
ment vehicle and the resulting constraints may lead to
meaningful differences in the returns of factor strategies.
Quantitative Special Issue 2019
  
0RPHQWXP8&,76
9DOXH8&,76
We examine how implementation matters through
the lens of two well-known factor investment strate-
gies: momentum and value. Eligible investors can choose
to implement these in relatively unconstrained private
funds—which take long and short positions—or the most
constrained ETFs, of which the vast majority are index-
based, fully transparent, and long only. In between,
investment constraints vary across European UCITS,
mutual funds or unit trusts, funds of one, or separately
managed accounts. We take constraints on leverage and
turnover typical for each vehicle and examine the effect
on returns, risk, holdings, and other portfolio charac-
teristics induced by those constraints.
We find that, in many cases, more unconstrained
hypothetical implementations with more f lexible use of
leverage result in higher Sharpe ratios and better perfor-
mance in terms of downside risk statistics, such as max-
imum drawdowns. Higher leverage does not directly
correspond to higher risk or higher turnover. Indeed,
some of the additional f lexibility with higher leverage is
employed by an optimizer to help reduce risk.
We focus only on hypothetical momentum and
value strategies in equities. Implementation consider-
ations also affect, potentially to an even higher degree,
The Journal of Portfolio Management   113
factor strategies in other asset classes—especially in more
illiquid asset classes. We also ignore other real-life impli-
cations of the investment choices, such as taxes, trading
costs, monitoring and reporting requirements, and gov-
ernance considerations, all of which are also important
elements for investors in choosing a practical way to
implement factor strategies.
REFERENCES
Almazan, A., K. C. Brown, M. Carlson, and D. Chapman.
2004. “Why Constrain Your Mutual Fund Manager?” Journal
of Financial Economics 73 (2): 289–321.
Ang, A. Asset Management: A Systematic Approach to Factor Based
Investing. New York: Oxford University Press, 2014.
Ball, R., and P. Brown. 1968. “An Empirical Evaluation of
Accounting Income Numbers.” Journal of Accounting Research
6 (2): 159–178.
Brière, M., and A. Szafarz. “Factor Investing: The Rocky
Road from Long-Only to Long-Short.” CEB Working Paper
No. 17/013, 2017.
Cheng, M., and A. Madhavan. 2009. “The Dynamics of
Leveraged and Inverse-Exchange Traded Funds.” Journal of
Investment Management 7 (4).
Clarke, R., H. de Silva, and S. Sapra. 2004. “Towards More
Information-Efficient Portfolios.” The Journal of Portfolio Man-
agement 31 (1): 34–63.
Daniel, K., and T. J. Moskowitz. 2016. “Momentum Crashes.”
Journal of Financial Economics 122 (2): 221–247.
Fama, E. F., and K. R. French. 1993. “Common Risk Fac-
tors in the Returns on Stocks and Bonds.” Journal of Financial
Economics 33 (1): 3–56.
114   Relaxing Constraints Improves Returns of Factor Strategies Quantitative Special Issue 2019
French, K. R., G. W. Schwert, and R. F. Stambaugh. 1987.
“Expected Stock Returns and Volatility.” Journal of Financial
Economics 19 (1): 3–29.
Gleeson, C. A., and C. M. C. Lee. 2003. “Analyst Fore-
cast Revisions and Market Price Discovery.” The Accounting
Review 78 (1): 193–225.
Grinold, R. C., and R. N. Kahn. 2000. “The Efficiency
Gains of Long-Short Investing.” Financial Analysts Journal
56 (6): 40–53.
Hodges, P., K. Hogan, J. Peterson, and A. Ang. 2017. “Factor
Timing with Cross-Sectional and Time-Series Predictors.”
The Journal of Portfolio Management 44 (1): 30–43.
Jegadeesh, N., and S. Titman. 1993. “Returns to Buying
Winners and Selling Losers: Implications for Stock Market
Efficiency.” The Journal of Finance 48 (1): 65–91.
Johnson, S., R. N. Kahn, and D. Petrich. 2007. “Optimal
Gearing.” The Journal of Portfolio Management 33 (4): 30–43.
Logue, A. C. Hedge Funds for Dummies. New York: Wiley,
2007.
Madhavan, A. Exchange-Traded Funds and the New Dynamics of
Investing. New York: Oxford University Press, 2016.
Qian, E. E., R. H. Hua, and E. H. Sorensen. Quantitative
Equity Portfolio Management: Modern Techniques and Applications.
Boca Raton, FL: Chapman & Hall/CRC, 2007.
Scharfman, J. Hedge Fund Governance: Evaluating Oversight,
Independence, and Conflicts. San Diego, CA: Academic Press,
2015.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

Feifei Li
is director, head of
investment strategy at
Research Affiliates, LLC,
in Newport Beach, CA.
li@rallc.com
Joseph (Yoseop) Shim
is vice president of smart
beta at Research Affiliates,
LLC, in Newport Beach,
CA.
shim@rallc.com
Quantitative Special Issue 2019
Trade-Off in Multifactor Smart
Beta Investing: Factor Premium
and Implementation Cost
Feifei Li and Joseph (Yoseop) Shim
F
actor investing, which requires
choosing securities based on attri-
butes associated with high expected
returns, is becoming increasingly
popular as investors realize they can harvest
the excess return over the market-capital-
ization-weighted benchmark by a simple,
transparent, and rules-based approach. Some
of the most commonly used factors in investor
portfolios are value, momentum, quality, low
risk, and size, although many other factors are
used. These common five factors are based
on the well-established academic literature of
Fama and French (1993, 2012, 2016), Carhart
(1997), and Frazzini and Pedersen (2014).
Because some factors are not highly corre-
lated, combining them can further benefit
investors through the diversification effect.1
A multifactor strategy, which is a one-stop
solution for investors who are seeking the
excess returns associated with various fac-
tors and which can give investors the ability
to weather the storm during adverse market
conditions, seems to be a natural path for
factor-investing product proliferation.
When investors start down the path of
multifactor investing, they typically raise a
number of important questions. Even though
at a conceptual level combining factors makes
1
 Among others, Asness, Moskowitz, and
Pedersen (2013) and Asness et al. (2015) documented
the diversification benefit of combining multiple fac-
tors with low correlation to each other.
sense, a strategy that requires stock selection
based on multiple measures is often viewed
as complex in terms of product features and
the construction process. Brightman et al.
(2017) answered two important questions
for multifactor strategy investors: (1) Which
factors should be included in a multifactor
portfolio, and (2) how to allocate assets across
factors over time. The authors reviewed the
theory and evidence in support of the value,
momentum, profitability, investment, low
volatility, and size factors and then demon-
strated that a combination of those six factor-
based smart betas can outperform the market
with their much lower tracking error (TE)
and downside risk.
In this follow-up article, we focus more
on the implementation issues of multifactor
investing, including portfolio concentra-
tion, turnover, trading cost, and capacity, all
of which are crucial elements in real-world
product design. We address two questions
in the presence of implementation costs:
Which factors should be included in a mul-
tifactor portfolio, and what is the best way to
construct a single-factor portfolio? We pro-
vide the detailed analysis that supports our
recommendations.
FACTOR PERFORMANCE
AND CORRELATION
Consistent with Brightman et al. (2017),
we include six factor-based smart beta
The Journal of Portfolio Management   115
Exhibit 1
Performance of Long-Only Factor-Based Smart Beta Strategies, United States, July 1973–June 2017

$EVROXWH3HUIRUPDQFH 5HODWLYH3HUIRUPDQFH
6WUDWHJ\ 5HWXUQ 9RO 6KDUSH5DWLR 9DOXH$GG 7( ,5
0DUNHW   
9DOXH      
/RZ%HWD      
3URILWDELOLW\      
,QYHVWPHQW      
0RPHQWXP      
6L]H      
$YHUDJHRI6L[)DFWRUV      

Note: TE = tracking error; IR = information ratio.

Source: Research Affiliates, LLC, using CRSP and Compustat data.

strategies in our tests: value, low beta, profitability,
investment, momentum, and size. Each of these factors
is widely deemed robust in the academic literature (Fama
and French 1993, 2012, 2016; Carhart 1997; Frazzini
and Pedersen 2014). To illustrate the opportunities pre-
sented by investing in real-world strategies, we construct
simple long-only, investable portfolios in accordance
with widely accepted academic and investment practice.
We start with the large-cap universe of US stocks,
except for the small-size strategy. For each of these
five factor portfolios, we use the top 30% based on
NYSE breakpoints to select stocks based on the cor-
responding characteristics. For example, we construct
the value portfolio from stocks above the 70th per-
centile on the NYSE by book-to-market ratio. Our
small-size strategy consists of all the available stocks in
the small-cap universe. We then capitalization weight
the selected stocks, except for low beta, which we
weight by beta ranking. The portfolios are rebalanced
annually each July with the exception of momentum
and low beta, which are rebalanced quarterly. Details
about our factor construction method are provided in
the Appendix.
On average, these six factor-based smart beta
strategies deliver economically signif icant returns.
The average annualized excess return over our study
period, July 1973 to June 2017, is 1.96%, as illustrated in
Exhibit 1. As noted by Brightman et al. (2017), profit-
ability, on its own, generates the lowest return. As we
will show, however, profitability’s low or negative corre-
lations with other factors, as well as its low trading cost,
make the strategy a helpful diversifier in the aggregate
portfolio of factor-based smart beta strategies.
We first review the correlations between the fac-
tors to understand how they may interact with each
other. The correlations across a factor’s excess return
over market are mostly low or negative, implying their
performance drivers have orthogonal components.
On average, the pairwise cross-factor correlation,
reported in Exhibit 2, is very close to zero (i.e., 0.01).
Among the factors we study, the only factors that appear
to have a correlation above 0.4 are investment, value,
and low beta. This observation suggests that combining
those factors can create diversification benefits.
In the absence of trading costs, including momentum
and size factors in a multifactor strategy seems a no-
brainer, given the significant diversification benefit these
factors bring to the table. The momentum factor has a
negative correlation with three of the five other factors
(value, low beta, and investment) and a slight positive cor-
relation with the remaining two (profitability and size).
The size factor seems to be a strong diversifier when com-
bined with the profitability factor because the two have a
correlation of −0.60, and it has relatively low correlation
with the other factors. Without considering implementa-
tion issues, both the momentum and size factors seem to
be desirable from a diversification perspective.
IMPLEMENTATION COSTS OF FACTORS
Chow et al. (2017) documented that the cost of
executing certain factor-based smart beta strategies can

116   Trade-off in Multifactor Smart Beta Investing: Factor Premium and Implementation Cost Quantitative Special Issue 2019
Exhibit 2
Cross-Correlation between Factor Excess Returns, United States, July 1973–June 2017
9DOXH /RZ%HWD
9DOXH 
/RZ%HWD  
3URILWDELOLW\ ±  
,QYHVWPHQW  
0RPHQWXP ± ±
6L]H  
Source: Research Affiliates, LLC, using CRSP and Compustat data.
be nontrivial. Following Chow et al., we estimate the
implementation costs of these strategies by applying the
Aked and Moroz (2015) market-impact model. Imple-
mentation cost is the movement in a security’s price as
a result of trading execution: When a large buy order
of a particular security is placed, it typically drives up
the price of the security. The aggregate implementa-
tion cost of a portfolio rebalancing is the summation of
all such costs across all stocks traded to accomplish the
rebalancing.
With simplifying assumptions, we can attribute
the market impact of a rebalancing on the performance
of a smart beta strategy to the following characteristics:
• Portfolio volume—the aggregate of the median daily
trading volume of all stocks in a portfolio. A strat-
egy’s cost is inversely proportional to its portfolio
volume; in other words, a strategy is more costly
to implement if it requires holding illiquid stocks.
• Tilt—a measure of illiquidity that represents the
degree to which portfolio weights deviate from
the weights proportional to trading volume. The
smallest possible tilt value is 1, which is achieved
by a volume-weighted portfolio, theoretically the
most liquid combination of a given set of holdings.
• Turnover—a measure of how frequently assets in
a portfolio are bought and sold by the strategy
over a 12-month period, typically a calendar year.
In general, a strategy that requires a higher rate of
trading incurs higher market-impact costs.
• Turnover concentration—the degree to which trades
are spread across the holdings in a portfolio. Highly
concentrated trades are more costly to execute.
A strategy that demands high liquidity from only
a few names causes a higher market impact than
Quantitative Special Issue 2019
3URILWDELOLW\ ,QYHVWPHQW 0RPHQWXP 6L]H
&URVV)DFWRU
$YHUDJH 
 
 ± 
± ±  
another strategy that spreads out trades across many
holdings.
Intuitively, low portfolio volume, high tilt, high
turnover, and high turnover concentration would lead
to a high market-impact cost. By using these attributes,
we can explain how different aspects of a strategy are
accountable for its implementation costs.
The average turnover across these six strategies
is 62.4%, and the average estimated trading cost of the
strategies is 13 bps, given the assumption of $1 billion in
assets under management (AUM).2 The costs of imple-
menting a strategy with high turnover, strong tilt, or
low volume can be quite large. Using the momentum
strategy as an example, and as reported in Exhibit 3,
annualized one-way turnover is 160.4%, assuming
quarterly rebalancing. This extraordinarily high turn-
over creates a high trading cost (28 bps at $1 billion
AUM) for the strategy even though its portfolio volume
($35.6 billion) and tilt (1.29) are at reasonable levels.
At $5 billion AUM, trading costs erode 138 bps of
return. If we increase AUM to $10 billion, the trading
cost (277 bps) more than offsets the entire estimated
value-add (164 bps) of the strategy reported in Exhibit 1.
For this reason, we do not recommend that investors use
the naively constructed momentum factor as a stand-
alone investment strategy.
2
 We report the recent five-year average of the estimated
trading costs instead of the historical average based on the entire
sample. The trading volume has been increasing exponentially since
the explosion of electronic trading in the late 1990s. Therefore, the
current estimate is a much better representation of the expected
trading cost than the historical average, which would be unreason-
ably high by today’s standards.
The Journal of Portfolio Management   117
Exhibit 3
Implementation Cost of Long-Only Factor-Based Smart Beta Strategies, United States, July 1973–June 2017

Turnover and Trading Cost

Current Portfolio
Trading Cost at
Current WAMC Volume Turnover Capacity
Strategy N ($ billion) Turnover ($ billion) Concentration Tilt ($1 billion) ($5 billion) ($10 billion) ($ billions)
Market 3,154 148.7 4.6% 137.1 58.7% 1.39 0.00% 0.01% 0.02% 248.6
Value 154 136.4 46.0% 18.3 90.7% 1.65 0.06% 0.30% 0.60% 8.0
Low Beta 277 37.9 57.8% 45.6 73.4% 4.51 0.18% 0.89% 1.78% 3.0
Profitability 320 189.2 16.7% 52.2 87.1% 1.27 0.02% 0.10% 0.20% 24.4
Investment 201 89.4 67.3% 26.7 96.0% 1.34 0.13% 0.66% 1.32% 3.7
Momentum 258 112.4 160.4% 35.6 89.6% 1.29 0.28% 1.38% 2.77% 1.9
Size 2,232 1.7 26.0% 12.4 82.9% 2.38 0.10% 0.49% 0.97% 4.9
Average of 574 94.5 62.4% 31.8 86.6% 2.07 0.13% 0.64% 1.27% 7.7
Six Factors

Note: WAMC = weighted average market capitalization.

Source: Research Affiliates, LLC, using CRSP and Compustat data.

In contrast, the profitability strategy is quite inex-
pensive to trade (2 bps at $1 billion AUM) because the
profitability factor tends to favor larger and more-liquid
stocks. This characteristic is ref lected by its high port-
folio volume ($52.2 billion) and low tilt (1.27) and by
the more stable portfolio holdings indicated by its low
turnover (16.7%). The value strategy also incurs rela-
tively low trading costs (6 bps at $1 billion AUM) by
virtue of its relatively low turnover and tilt.
Interestingly, the size strategy, which focuses pri-
marily on stocks with a small market capitalization and
which is often believed to be a high-cost strategy, actu-
ally has a below-average trading cost (10 bps at $1 billion
AUM). Although the size factor does tend to trade small,
illiquid stocks, as indicated by its low portfolio volume
($12.4 billion) and high tilt (2.38) relative to the other
factors, it has low turnover (26%) as a result of its broad
coverage—2,232 names as of June 2017. On balance, the
higher coverage of the size strategy almost offsets the
cost of trading smaller companies.
INCLUSION OF MOMENTUM AND SIZE
IN A MULTIFACTOR STRATEGY
In this section, we study the impact of including
the supposedly high-cost strategies of momentum
and size on the performance and cost of a multifactor
strategy. Momentum and size can be good additions to
a multifactor strategy, however, because of their nega-
tive or low positive correlations with other factors. So,
how should investors measure this diversification benefit
versus the increased costs of implementation?
We consider four portfolios to illustrate how
including the momentum and size factors affects the
performance characteristics and implementation costs of
a multifactor strategy. We combine the factors in each
portfolio by allocating an equal weight to each factor
and rebalancing every quarter:
• Portfolio 1: Value, low beta, profitability, and
investment
• Portfolio 2: Four factors in portfolio 1 plus
momentum
• Portfolio 3: Four factors in portfolio 1 plus size
• Portfolio 4: Four factors in portfolio 1 plus
momentum and size
The performance and implementation cost of
these four portfolios are reported in Exhibits 4 and 5,
respectively.
The implementation cost of the multifactor
strategy is much lower than the simple average of
the implementation costs of each of the underlying
factors. As reported in Exhibit 3, the average cost of
implementing the six single-factor portfolios is 127
bps at $10 billion AUM. By contrast, the trading cost

118   Trade-off in Multifactor Smart Beta Investing: Factor Premium and Implementation Cost Quantitative Special Issue 2019
Exhibit 4
Performance of Multifactor Smart Beta Strategies, United States, July 1973–June 2017

$EVROXWH3HUIRUPDQFH 5HODWLYH3HUIRUPDQFH
1HWRI&RVW65 1HWRI&RVW,5
9DOXH
6WUDWHJ\ 5HWXUQ 9RO 65 ELOOLRQ ELOOLRQ ELOOLRQ $GG 7( ,5 ELOOLRQ ELOOLRQ ELOOLRQ
0DUNHW      
)DFWRUV            
)DFWRUV0RP            
)DFWRUV6L]H            
)DFWRUV            
0RP 6L]H

Note: SR = Sharpe ratio.

Source: Research Affiliates, LLC, using CRSP and Compustat data.

Exhibit 5
Implementation Cost of Multifactor Smart Beta Strategies, United States, July 1973–June 2017

7XUQRYHUDQG7UDGLQJ&RVW
&XUUHQW 3RUWIROLR
:$0& 9ROXPH 7UDGLQJ&RVWDW
&XUUHQW 7XUQRYHU &DSDFLW\
6WUDWHJ\ 1 ELOOLRQ 7XUQRYHU ELOOLRQ &RQFHQWUDWLRQ 7LOW ELOOLRQ ELOOLRQ ELOOLRQ ELOOLRQ
0DUNHW          
)DFWRUV          
)DFWRUV0RP          
)DFWRUV6L]H          
)DFWRUV          
 0RPDQG6L]H

Source: Research Affiliates, LLC, using CRSP and Compustat data.

of a multifactor strategy that equally weights the six
factors and rebalances every quarter (portfolio 4) is
only 25 bps, as Exhibit 5 shows.
The huge reduction in trading costs is partly due
to the cancelling trades initiated by uncorrelated factors
at the headline portfolio level. For example, in Exhibit 5,
the one-way turnover of an equally weighted six-factor
strategy including momentum and size is 55.7%, which
is lower than the simple average turnover (62.4%) of
the six individual factors, as reported in Exhibit 3.
In addition, by including all six factors, the multifactor
strategy has a portfolio volume quite close to that of the
entire market.
Moreover, the turnover concentration of the mul-
tifactor strategy is smaller than the simple average turn-
over concentration of the six underlying factors, which
implies that trades are more spread out across holdings
in the multifactor strategy as a result of diversification.
This observation appears to be informative regarding the
decision an investor pursuing a multifactor strategy must
make: To capture the desired factor premiums, should
the investor hire a different manager for each factor
strategy or hire one manager of a multifactor strategy?
We demonstrate that the latter choice is more cost effec-
tive from a trading perspective.
We now examine portfolio 2, the impact of
adding a momentum sleeve to the four-factor strategy.
As expected, before applying the trading cost model,
because of negative or low positive correlations with
other factors, the addition of momentum to an equally
weighted four-factor strategy (value, low beta, profit-
ability, and investment) reduces TE by 83 bps (3.46%
versus 4.29%) and leads to a much higher information
ratio (IR), 0.60 versus 0.49.

Quantitative Special Issue 2019 The Journal of Portfolio Management   119

 The trading cost of the f ive-factor strategy
including momentum is surprisingly not much higher
than that of the four-factor strategy without momentum.
After taking into consideration its slightly higher trading
cost, 1 bp (0.31% versus 0.30%) at the $10 billion AUM
level, the IR net of trading costs is 0.52, still a major
improvement over 0.42. This modest increase in trading
costs can be explained, yet again, by momentum’s low
or negative correlation with other factors, which leads
to trades that cancel each other out. Adding momentum
to the mix of other factors also increases portfolio
volume and lowers portfolio tilt. In other words, port-
folio 2 trades more liquid stocks, and this additional
liquidity compensates for the increased turnover of the
momentum strategy.
The advantage of adding the size factor (portfolio 3)
to the four-factor portfolio comes from its contribu-
tion to performance and through lower implementation
costs. Size ranks third, after profitability and value, in
terms of the lowest trading costs and also one of the
best performing factors in our sample. Adding size to
the four-factor portfolio improves the return by 20 bps
(13.30% versus 13.10%) while lowering the cost by 6 bps
(0.24% versus 0.30%) at the $10 billion AUM level.
Admittedly, the size strategy has the highest volatility
of the six factors in our sample. Thus, including the size
factor in the mix of other factors increases the volatility
of a multifactor strategy by 0.7%, whereas its low cor-
relation with other factors helps lower TE by 0.27%,
improving the IR.
In summary, because the diversification benefit
outweighs the increased cost of implementation, we
recommend that investors include both the momentum
and size factors in their multifactor strategy. The mul-
tifactor portfolio that includes both momentum and
size (portfolio 4) yields the highest IR, before and
after trading costs: 0.67 and 0.60 (assuming $10 billion
AUM), respectively, with barely any impact on the
Sharpe ratio.
CONCENTRATION OF FACTOR PORTFOLIOS
In this section, we discuss the optimal cutoff point,
or concentration level, for stocks being selected to form
the underlying factor portfolios. The most commonly
used approaches select 20%, 30%, or 50% of stocks from
the starting universe based on the rank of specific signals
of the underlying stocks. Are these cutoffs arbitrary?
120   Trade-off in Multifactor Smart Beta Investing: Factor Premium and Implementation Cost Quantitative Special Issue 2019
For example, is a 20% cutoff better than a 30% cutoff?
What is the optimal level we should use in constructing
a multifactor strategy?
Intuitively, a certain concentration of stocks would
appear necessary to extract the factor premium. Typi-
cally, the academic literature finds that quintile portfo-
lios based on a specific signal show a monotonic return
pattern: The portfolio formed by selecting the top 20%
of stocks based on the underlying signal performs the
best, the portfolio of the next 20% of stocks exhibits the
next-best performance, and so on, so that the bottom
20% of stocks have the weakest signal and the worst
performance. We are therefore reluctant to include too
many stocks in our portfolios because stocks’ attractive-
ness decreases as we go down the roster. Yet, our trading
cost model indicates that having broader coverage, less
tilt toward illiquid stocks, and lower turnover helps
reduce the cost of price impact.
To illustrate the trade-off between performance
and cost, we construct a multifactor strategy that com-
bines the value, low beta, profitability, investment,
momentum, and size factors at different concentration
levels, varying from a 50% cutoff to a 15% cutoff with
a stepwise reduction of 5%. The one exception is the
size strategy, whose cutoff we hold constant. Following
the Fama and French (2016) definition, our small-
size strategy consists of all the available stocks in the
small-cap universe. Thus, no selection is required for
the size strategy. We also estimate the implementation
costs of the multifactor strategies, and we calculate net-
of-trading-cost Sharpe ratios and IRs across AUM of
$1 billion, $5 billion, and $10 billion.
Exhibits 6 and 7 show the net-of-trading-cost
Sharpe ratios and IRs of the six-factor strategy at dif-
ferent concentration levels across different AUM assump-
tions. As the underlying factor portfolios become more
concentrated, the return on the multifactor strategy
improves monotonically, which is consistent with aca-
demic findings and our forward-looking expectations
of those stocks’ future performance.
However, the risks, measured by volatility and TE,
also rise as the concentration level rises because over-
concentration undermines the benefit of diversification
and leads to greater idiosyncratic risk. The risk-adjusted
returns thus do not improve beyond the 25% concentra-
tion level even before taking into account implementa-
tion, or trading, costs. For example, the Sharpe ratio
monotonically increases from 0.53 to 0.57 from the 50%
 Exhibit 6
Net-of-Cost Performance versus Concentration Level, United States, July 1973–June 2017
3DQHO$1HWRI&RVW6KDUSH5DWLRYHUVXV6HOHFWLYHQHVVPXOWLIDFWRU 3DQHO%1HWRI&RVW,QIRUPDWLRQ5DWLRYHUVXV6HOHFWLYHQHVVPXOWLIDFWRU
 

 

$IWHU&RVW,QIRUPDWLRQ5DWLR
$IWHU&RVW6KDUSH5DWLR

 

 

 

 

 
               
6HOHFWLYHQHVV 6HOHFWLYHQHVV

1R&RVW %$80 %$80 %$80

Source: Research Affiliates, LLC, using CRSP and Compustat data.

to the 25% concentration level and remains more or less concentration level and then f lattens out at $1 billion
unchanged after that. The IR also increases from 0.53 to AUM. At $5 billion and $10 billion AUM, however,
0.70 from the 50% to the 25% level but declines there- the net-of-cost Sharpe ratio declines rapidly beyond the
after, even before considering trading costs. 25% concentration level. For example, at $10 billion
The trade-off between performance and cost AUM, the highly concentrated portfolio, with a 15%
becomes much more important at the higher concen- cutoff, yields a lower net-of-cost IR (0.53) than the less-
tration levels because the impact of trading costs on the concentrated portfolio at a 25% cutoff (0.61).
Sharpe ratio and the IR is much larger in more-con- To summarize, the sweet spots for the best per-
centrated multifactor strategies. As we make the under- formance in the presence of implementation costs seem
lying factor portfolios more concentrated, the turnover, to reside around the 25% concentration level, based on
turnover concentration, and tilt of the multifactor our sample. We acknowledge, however, that if we only
strategy all tend to increase, whereas the portfolio focus on empirical evidence, the optimal cutoff level
volume decreases, leading to higher implementation may differ slightly across individual factor portfolios.
costs. The goal of our analysis, however, is to effectively cap-
Additionally, the mix with more-concentrated ture the factor premium at a reasonable cost through a
portfolios has fewer offsetting trades that cancel each simple, transparent, and rules-based approach. Lacking
other out, and thus the turnover is reduced by a lesser a strong reason for the individual factor portfolios to
amount. As a result, the reduction in performance be constructed using different concentration levels, we
from implementation costs is much higher for more- suggest investors avoid overfitting risk and use consistent
concentrated strategies, especially under the assumption coverage across all factors included in the multifactor
of larger AUM. For instance, the net-of-cost Sharpe portfolio.
ratio monotonically increases from the 50% to 25%

Quantitative Special Issue 2019 The Journal of Portfolio Management   121

Exhibit 7
Multifactor Smart Beta Strategies, United States, July 1973–June 2017
3DQHO$3HUIRUPDQFH&KDUDFWHULVWLFV
$EVROXWH3HUIRUPDQFH 5HODWLYH3HUIRUPDQFH
1HWRI&RVW65 1HWRI&RVW,5
9DOXH
6WUDWHJ\ 5HWXUQ 9RO 65 ELOOLRQ ELOOLRQ ELOOLRQ $GG 7( ,5 ELOOLRQ ELOOLRQ ELOOLRQ
0DUNHW      
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            
0XOWLIDFWRU            

3DQHO%,PSOHPHQWDWLRQ&RVWV
7XUQRYHUDQG7UDGLQJ&RVW
&XUUHQW 3RUWIROLR
:$0& 9ROXPH 7UDGLQJ&RVWDW
&XUUHQW 7XUQRYHU &DSDFLW\
6WUDWHJ\ 1 ELOOLRQ 7XUQRYHU ELOOLRQ &RQFHQWUDWLRQ 7LOW ELOOLRQ ELOOLRQ ELOOLRQ ELOOLRQ
0DUNHW          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU          
0XOWLIDFWRU   
Source: Research Affiliates, LLC, using CRSP and Compustat data.
CONCLUSION
In this study, we highlight the trade-off between
gaining excess returns associated with factors and the
impact of implementation costs in constructing a multi-
factor smart beta strategy. We consider portfolio concen-
tration, turnover, trading costs, and capacity as crucial
elements in the implementation of a real-world multi-
factor smart beta product design.
We f ind that the momentum and size factors
are helpful components in a multifactor smart beta
strategy, regardless of the implementation challenges
of each factor. Most of the value added by a stand-
alone momentum strategy would be wiped out by
trading costs from very high turnover under reason-
ably large AUM assumptions, but because of negative
122   Trade-off in Multifactor Smart Beta Investing: Factor Premium and Implementation Cost Quantitative Special Issue 2019
      
or low positive correlations with other factors, when
added to a multifactor strategy, momentum helps lower
TE and improves the IR. These benefits are achieved
without a large increase in implementation costs, how-
ever, because of the offsetting trades across the factor
strategies that cancel each other out and because of the
improved liquidity of the stocks held in a momentum
strategy.
The size factor, contrary to conventional wisdom,
is rather inexpensive to trade because of its relatively
broad coverage and low turnover. Thus, adding the size
factor to the combination of other factors can improve
the performance, and lower the TE, of the multifactor
strategy given the low correlation of size with the other
factors, resulting in a higher IR together with a reduc-
tion in trading costs.
 We find that by using more-concentrated under- Exhibit A1
lying factor portfolios in constructing a multifactor Signals Used to Sort Factor-Based Smart Beta
strategy, investors can improve its performance. These Portfolios
benefits come, however, at the expense of higher vola-
tility, TE, turnover, and trading costs. Our study sug- )DFWRU 6LJQDO 'HILQLWLRQ
gests that the sweet spots for the best risk-adjusted 9DOXH %RRNWR3ULFH5DWLR %RRN9DOXH0DUNHW&DS
performance in the presence of implementation costs /RZ%HWD 0DUNHW%HWD )UD]]LQLDQG3HGHUVRQ>@
reside around the 25% concentration level. GHILQLWLRQLQZKLFK
Neither of the extremes of (1) maximizing paper FRUUHODWLRQLVHVWLPDWHGZLWK
ILYH\HDUVRIGDLO\UHWXUQV
portfolio performance while ignoring the trading
DQGYRODWLOLW\ZLWKRQH\HDU
costs that reduce such performance in practice nor (2) RIGDLO\UHWXUQV
focusing on low-cost implementation while missing
3URILWDELOLW\ 2SHUDWLQJ $QQXDOUHYHQXHVPLQXVFRVW
opportunities for better performance will produce an 3URILWDELOLW\ RIJRRGVVROGLQWHUHVW
optimal result for multifactor smart beta investors. We H[SHQVHDQGVHOOLQJJHQHUDO
strongly advocate the thoughtful design of a multi- DQGDGPLQLVWUDWLYHH[SHQVHV
factor strategy, which requires a conscious and delib- GLYLGHGE\ERRNHTXLW\IRU
erate decision to find the most advantageous balance WKHODVWILVFDO\HDU
between effectively harvesting the factor premium and ,QYHVWPHQW &KDQJHLQ$VVHWV <HDURYHU\HDUSHUFHQWDJH
implementation costs. FKDQJHLQWRWDODVVHWV
0RPHQWXP íWRí0RQWK 3ULRUPRQWKUHWXUQV
5HWXUQ VNLSSLQJPRVWUHFHQWPRQWK
Appendix 6L]H 0DUNHW&DS 0DUNHW&DS

FACTOR-BASED SMART BETA

CONSTRUCTION METHODOLOGY
To construct our portfolios in the United States,
we use the universe of US stocks from the CRSP/Com-
pustat database. We define the US large-cap equity uni-
verse as stocks whose market capitalizations are greater
than the median market capitalization on the NYSE
and the small-cap universe as stocks whose market
capitalizations are smaller than the NYSE median. The
US data extend from July 1973 through June 2017.
The factor-based smart beta portfolios, except the
small size strategy, are constructed from the large-cap
universe. For each of the factor characteristics in these
portfolios, we use the top 30% of the NYSE as a break-
point. For example, we construct the value portfolio
from stocks above the 70th percentile on the NYSE by
book-to-market ratio. Our small-size strategy consists
of all the available stocks in the small-cap universe, fol-
lowing Fama and French (2016). We then capitalization
weight the selected stocks except for low beta, which we
weight by beta ranking, following Frazzini and Pederson
(2014). The value, profitability, and investment portfo-
lios are rebalanced annually each July following Fama
and French (2016), whereas the momentum and low beta
portfolios are rebalanced quarterly according to widely
adopted investment practice.3
The signals used to sort the various factor-based
smart beta portfolios are given in Exhibit A1.
REFERENCES
Aked, M., and M. Moroz. 2015. “The Market Impact of
Passive Trading.” The Journal of Trading 10 (3) Summer: 5–12.
Asness, C. S., A. Ilmanen, R. Israel, and T. J. Moskowitz.
2015. “Investing with Style.” Journal of Investment Management
13 (1) First Quarter: 27–63.
Asness, C. S., T. J. Moskowitz, and L. H. Pedersen. 2013. “Value
and Momentum Everywhere.” The Journal of Finance 68 (3)
June: 929–985.
3
 In academic papers, momentum and low beta portfolios
are often rebalanced monthly. However, monthly rebalancing is
rarely adopted among investment practitioners because of the high
turnover and implementation cost; quarterly or even semiannual
rebalancing is preferred. For instance, the MSCI Momentum Index
and S&P 500 Minimum Volatility Index are rebalanced on a semi-
annual basis, and the AQR Momentum Index and S&P 500 Low
Volatility Index are rebalanced quarterly.

Quantitative Special Issue 2019 The Journal of Portfolio Management   123

Brightman, C., V. Kalesnik, F. Li, and J. Shim. “A Smoother
Path to Outperformance with Multi-Factor Smart Beta
Investing.” Research Affiliates Publications, January 2017.

Carhart, M. M. 1997. “On Persistence in Mutual Fund Per-

formance.” The Journal of Finance 52 (1) March: 57–82.

Chow, T., F. Li, A. Pickard, and Y. Garg. “Cost and Capacity:

Comparing Smart Beta Strategies.” Research Affiliates Pub-
lications, July 2017.

Fama, E. F., and K. R. French. 1993. “Common Risk Fac-

tors in the Returns on Stocks and Bonds.” Journal of Financial
Economics 33 (1) February: 3–56.

——. 2012. “Size, Value, and Momentum in Interna-

tional Stock Returns.” Journal of Financial Economics 105 (3)
September: 457–472.

——. 2016. “Dissecting Anomalies with a Five-Factor

Model.” The Review of Financial Studies 29 (1) January: 69–103.

Frazzini, A., and L. H. Pedersen. 2014. “Betting against

Beta.” Journal of Financial Economics 111 (1) January: 1–25.

To order reprints of this article, please contact David Rowe at

d.rowe@pageantmedia.com or 646-891-2157.

124   Trade-off in Multifactor Smart Beta Investing: Factor Premium and Implementation Cost Quantitative Special Issue 2019

Eduard van
Gelderen
is the chief investment
officer at PSP Investments
in Montreal, QC, Canada.
evangelderen@investpsp.ca
Joop Huij
is an associate professor
at the Rotterdam
School of Management
in Rotterdam, the
Netherlands, and the
head of factor investing at
Robeco in Rotterdam, the
Netherlands.
jhuij@rsm.nl
Georgi Kyosev
is a PhD candidate
at the Rotterdam
School of Management
in Rotterdam, the
Netherlands, and a
researcher at Robeco
in Rotterdam, the
Netherlands.
kyosev@rsm.nl
Quantitative Special Issue 2019
Factor Investing from Concept
to Implementation
Eduard van Gelderen, Joop Huij, and Georgi Kyosev
M
utual funds following factor
investing strategies based on
equity asset pricing anoma-
lies, such as the small cap,
value, and momentum effects, earn signifi-
cantly higher alphas than traditional actively
managed mutual funds. This effect is unre-
lated to other fund characteristics such as age,
expenses, and turnover; is robust to a global
sample of mutual funds and bootstrapped
confidence intervals; and is stronger for funds
that are exposed to multiple factors simul-
taneously. Although excess returns earned
by factor funds net of fees are significantly
smaller than the theoretical premiums of the
asset pricing anomalies, they are still positive
and statistically and economically significant.
For example, if investors would randomly
select a factor fund and would apply a buy-
and-hold strategy, these investors would earn
110 bps per annum in excess of the return that
is earned by the average traditional actively
managed mutual fund.
Nevertheless, the actual returns that
investors earn by investing in factor mutual
funds appear to be significantly lower than
this number because investors do not follow
buy-and-hold strategies, but rather dynami-
cally reallocate their funds both across factors
and factor managers. By attempting to time
across factors, investors lose a large portion of
the return they could earn with a buy-and-
hold strategy.
To better understand how investors
dynamically allocate to factor funds, we
study the f low-performance relation for
these funds. Although factor funds have
attracted signif icant fund f lows over our
sample period, it appears that fund f lows
have been driven by factor funds earning
high past returns and not by the funds pro-
viding factor exposures. Similar to Zheng
(1999), we f ind very little evidence of a
“smart money” effect in the sense that
f lows predict future fund performance. In
fact, consistent with the recent findings of
Cornell, Hsu, and Nanigian (2017), we do
not observe a positive relationship between
fund f lows and future performance.
We argue that rather than timing fac-
tors and factor managers, investors would be
better off by using a buy-and-hold strategy
and selecting a multifactor manager. For
example, if an investor would randomly
select a factor fund that is exposed to two
factors simultaneously and would apply a
buy-and-hold strategy, this investor would
earn 190 bps per annum in excess of the
return that is earned by the average tradi-
tional actively managed mutual fund. This
number would be 240 bps per annum if the
investor would have selected a manager that
is exposed to three factors simultaneously,
and even 270 bps per annum if the manager
would be exposed to four or more factors
simultaneously.
The Journal of Portfolio Management   125
 Our study is closely related to the work of van
Gelderen and Huij (2014), who showed that factor mutual
funds earn significant excess returns, using a large sample
of US equity mutual funds. We also extend the work of
Dichev (2007) and Hsu, Myers, and Whitby (2016), who
show that the actual returns earned by investors in hedge
funds and small cap, value, and growth mutual funds are
significantly lower than the returns they could earn with
buy-and-hold strategies because investors dynamically
reallocate their funds across factors and factor managers.
Our main contributions are the following: First,
the f low-performance analysis we perform helps one
to better understand how investors allocate to factor
funds; and second, our analyses of multifactor strategies
help investors harvest factor premiums more effectively.
Other contributions of our study are the inclusion of the
profitability and investments factors in our analyses; the
use of global equity fund data next to US equity fund
data; and the use of the bootstrap approach put forward
by Fama and French (2010) that has been designed to
help differentiate between skill and luck when assessing
mutual fund performance.
DATA AND METHODOLOGY
Data
For our US sample, we download monthly data
from the CRSP Survivorship Bias Free Mutual Fund
Database. We use monthly returns, total asset values,
quarterly turnover ratio, and expense ratio characteristics.
Fund age is calculated as the number of months with
available observations, and fund size is measured by
total assets. We adjust total net assets for mergers and
acquisitions when calculating the fund inf lows. Next,
we adjust our mutual fund sample to domestic, equity,
long-only funds by selecting the following objective
codes: EI, EIEI, G, GI, I, LCCE, LSGE, LCVE, MC,
MCCE, MCGE, MCVE, MLCE, MLGE, MLVE,
SCCE, SCGE, SCVE, and SG. We use the return of
the longest fund share class throughout our analysis.
For robustness, in an unreported analysis, we also use
value-weighted share classes and the share class with the
highest total asset value—our conclusions remain intact.
The sample period is from January 1990 to December
2015, and we include only funds with more than 36
available monthly return observations and capital asset
pricing model (CAPM) R 2 values higher than 0.6. The
market return is downloaded from Kenneth French’s
126   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
Exhibit 1
Sample Construction
86 *OREDO
6DPSOH3HULRG -DQWR -DQWR
'HF 'HF
$OO)XQGV  
)HZHUWKDQREV ± ±
5 ± ±
5HPDLQLQJ  
'HDG  
$OLYH  
Notes: This exhibit shows summary statistics for our US and global
samples. “All Funds” is the number of all mutual funds in our sample.
“Fewer than 36 obs.” is the number of funds excluded for having less
than the selected minimum number of data points. “R 2 < 0.6” is
the number of funds excluded for having a CAPM R 2 less than 0.6.
“Remaining” is the number of funds used for the analysis. “Dead”
is defined as the number of funds with missing return values during the
last month. Alive are all remaining funds with a valid return observation
in the last month.
data library. To limit incubation bias concerns, we
follow Fama and French (2010) and remove funds with
total assets less than USD 5 million.
Our global sample comprises all global developed
markets equity long-only mutual funds in the Morning-
star database. Similar to our US sample, we restrict funds
to only those with more than 36 return observations and
CAPM R 2 values higher than 0.6. US and global market
factor returns are also downloaded from the Kenneth
French data library. Because of factor return availability,
we start our global sample one year later—from January
1991 to December 2015.
Exhibit 1 shows a detailed summary of our sample
construction process. For the United States, our ini-
tial sample consists of 3,713 mutual funds. We remove
396 funds because they have fewer than 36 observa-
tions available and 208 funds because they have CAPM
R 2 values lower than 0.6. The remaining sample covers
3,109 funds, out of which 1,334 are dead and 1,775 are
alive. In total, we have 493,512 fund-month observa-
tions available. Our global sample starts with 7,334
equity funds. We remove 2,193 funds because they have
fewer than 36 return observations and 282 funds because
they have CAPM R 2 values lower than 0.6; 2,000 dead
funds and 2,859 alive funds remain, for a total sample
of 4,859 funds. In total, we have 670,099 fund-month
observations available.
 Methodology
Our empirical analyses consist of three main
sections: respectively, evaluating the performance of
factor fund managers; computing the actual returns
earned by investors in factor funds; and investigating
the f low-performance relation for factor funds.
Factor fund classification and performance
evaluation. In the first empirical section of our article
we investigate whether mutual funds following factor
investing strategies based on equity asset pricing anoma-
lies, such as the small cap, value, and momentum effects,
earn higher alphas than traditional actively managed
mutual funds. For these analyses we employ three sta-
tistical techniques: return-based style analysis to classify
factor funds; cross-sectional regressions to evaluate the
performance of factor funds; and bootstrap analyses to
test the robustness of our results.
Our fund classification method closely follows
the methodology employed by van Gelderen and
Huij (2014). We download monthly factor returns
from Kenneth French’s data library. For each fund
we run the five-factor Fama and French model, aug-
mented with momentum, using all available return
observations
Ri ,t = α i + βi ⋅ ( RM ,t − R f ,t ) + si ⋅SMBt + hi ⋅ HML t
+ w i ⋅WML t + ri ⋅ RMWt + c i ⋅ CMAt + εi ,t
where R i,t is the excess return of mutual fund i in
month t; R f,t is the risk-free return in period t; ai is
the alpha of fund I; R M,t is the return on the market
portfolio in period t; and SMB, HML, WML, RMW,
and CMA are returns of long–short factor-mimicking
portfolios for the size, value, momentum, profitability,
and investments factors, respectively. b, s, h, w, r, and c
are the estimated fund-specific factor coefficients, and
ei,t is the residual return of fund i in month t, under
the assumption of i.i.d. Similar to van Gelderen and
Huij (2014), we classify a fund as being a factor fund
if the regression coefficient on the respective factor is
positive and statistically significant. For example, if the
SMB beta coefficient of fund i is higher than 2, we
identify fund i as a small cap fund. A fund is consid-
ered to be a low-beta fund when its b is smaller than
0.8. Funds can have multiple factor fund classifications
simultaneously.
Quantitative Special Issue 2019
To measure fund performance, we use the inter-
cept from the following one-factor model:
Ri ,t = α i + βi ⋅ ( RM , t − R f , t ) + εi ,t (2)
We use CAPM alpha instead of the intercept from
Equation 1 as our main performance measure because we
want to measure the excess return coming from exposures
to one of the six factors. We limit the effect of outliers
by calculating the Z-score of fund alphas, winsorizing it
at -2 and 2:
  α − µα  
z _ Alphai = min  2, max  −2, i (3)
  σ α  
where ai is the alpha of fund i from the one-factor
model, ma is the average alpha across all funds in the
sample, and sa is the cross-sectional standard deviation
of all fund alphas.
We use the following cross-sectional regression to
evaluate the performance of factor funds:
z _ Alphai = α i + b1 ⋅ Low _ beta + b2 ⋅ Small _ cap
+ b3 ⋅Value + b4 ⋅ Momentum  
+ b5 ⋅  Profitability + b6 ⋅ Investments + εi (4)
where Low_beta, Small_cap, Value, Momentum, Profit-
(1)
ability, and Investments are 1 if the fund is classified as a
low beta, small cap, value, momentum, profitability, or
investments factor fund, or 0 otherwise.
We also run an augmented version of this
regression:
z _ Alphai = α i + b1 ⋅ Low _ beta + b2 ⋅ Small _ cap
+ b3 ⋅Value + b4 ⋅ Momentum + b5 ⋅ Profitability
+ b6 ⋅ Investments + b7 ⋅ log age + b8 ⋅ log size
+ b19 ⋅ exp _ ratio + b10 ⋅ turn _ ratio + εi
(5)
where log age is the natural logarithm of fund age, cal-
culated as the number of months with available return
observations; log size is the natural logarithm of fund
size, measured as its average total net assets; exp_ratio
is the average total expense ratio; and turn_ratio is the
average turnover ratio. For our global markets sample,
we do not include exp_ratio and turn_ratio in our regres-
sions because the underlying data are unavailable.
The Journal of Portfolio Management   127
 To rigorously test the robustness of our results,
we employ a bootstrap method in the spirit of Fama
and French (2010). In the distribution of active manager
returns, we see that, on average, fund alpha is negative
after cost with approximately the average cost level. This
implies that funds, on average, produce alpha that is
insufficient to cover the fees they charge; however, the
fact that some managers tend to be on the positive side
of the distribution might indicate that they have some
level of skill or that they simply generated high returns
by chance. To control for this effect, we do a bootstrap
analysis in which we simulate mutual fund alpha distri-
bution with a true alpha equal to zero.
To do so, we simulate 5,000 cross-sectional zero-
alpha distributions. First, we subtract the one-factor
alpha from the returns of each fund to force its true alpha
to zero. Second, for each run we select a random number
of months with replacement, similar to Fama and French
(2010). By selecting the same number of months for all
funds, we keep the cross-sectional properties of mutual
fund performance, which is directly related to the alpha
distribution. Third, to control for the difference in the
number of observations for each fund, we compare their
performance based on the t-statistic of alpha (t(a)) and
not on alpha itself. After having 5,000 simulated t(a), we
calculate our bootstrapped distribution by calculating
the average t(a) at each percentile over all 5,000 runs.
The resulting cross-sectional distribution has an implicit
assumption that all managers have enough skills to cover
their fees. Because we know that the true alpha is zero,
that means that all alphas that are different from zero
are observed by luck. As such, to infer that a manager
has skills exceeding its fees, the t(a) of the actual dis-
tribution should be higher than the simulated t(a) at a
certain percentile.
Dollar-weighted returns. In the second empir-
ical section of the article we calculate the actual returns
that investors earn by investing in factor funds. Then
we test whether these returns are different from the
return that a buy-and-hold investor earns by randomly
selecting a factor fund.
To calculate investors’ returns we follow the meth-
odology proposed by Dichev (2007) and estimate fund
distributions (i.e., capital allocations to individual funds)
in the following way:
distributioni,t = (1 + rt ) ⋅ TNAi,t −1 − (TNAi,t + M i,t ) (6)
128   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
where TNAi,t is the total net assets of fund i in month t,
TNAi,t-1 is the total net assets of fund i in month t - 1,
rt is the return of fund i in month t, and Mi,t is the total
growth in assets of fund i in month t due to mergers
and acquisitions.
We then calculate dollar-weighted returns as the
internal rate of return (IRR), with the negative of the
first available TNA as the initial value; the last available
TNA as the terminal value; and the estimated distribu-
tion as the monthly capital f lows. We perform the main
analysis at the aggregate factor level because we first sum
all assets for each factor classification and then calculate
the distribution and the IRR at the total asset level, as
done by Dichev (2007) and Hsu (2016).
To test whether the actual return investors earn
by investing in factor funds is different from the return
earned by randomly selecting a factor fund and applying
a buy-and-hold strategy, we perform a bootstrap analysis
in which we keep the order of capital f lows unchanged
and randomly shuff le fund returns, following Dichev
and Yu (2011).
Flow-performance relation. In the third
empirical section of our study we analyze the f low-
performance relation for factor mutual funds to better
understand how investors dynamically allocate to these
funds. To this end, we regress fund f lows on fund char-
acteristics. Relative fund f lows are calculated as the
negative of the monthly distribution divided by the
beginning-of-month total assets:
TNAi,t − (1 + rt )TNAi,t −1 −   M i,t
  rel _ flow i,t = (7)
TNAi,t −1
We winsorize rel_ flow at 1% to limit the impact of
outliers. We then estimate the f low-performance rela-
tions using the following piecewise linear regression,
following Sirri and Tufano (1998):
rel _ flow i ,t =   α i + b1 ⋅ Low _ beta + b2 ⋅ Small _ cap
+ b3 ⋅Value + b4 ⋅ Momentum + b5 ⋅ Profitability
+ b6 ⋅ Investments + b7 ⋅ log age + b8 ⋅ log size
+ b ⋅ exp ⋅ + b ⋅ turn + b ⋅ Performance
9 ratio 10 ratio 11
+ b12rank bottom
i ,t −1 + b12rank middle
i ,t −1 + b12rankitop,t −1 +   εi
(8)
where Performance refers to the past 12-month
average outperformance over the market portfolio
Exhibit 2
Distribution of Fund Alphas

± ±± ±   !  ! 1RRI)XQGV

$OO)XQGV         
1R([SRVXUH         
/RZ%HWD         
6PDOO&DS         
9DOXH         
0RPHQWXP         
3URILWDELOLW\         
,QYHVWPHQWV         

Notes: This exhibit shows distributions of annualized fund alphas across all US funds in the CRSP Mutual Fund Database with total assets exceeding
$5 million. Alphas are calculated per fund as the intercept from CAPM regressions over all available observations during the sample period of January 1990
to December 2015. “< -4” shows the percentage of funds with annualized alphas less than -4%; “-4:-2” shows the percentage of funds with annualized
alphas between -2% and -4%; and “% > 0” shows the number of funds with positive alphas.

(or past 12-month CAPM alpha in some of our regres-
sion specifications) and top, middle, and bottom are cal-
culated as follows:
rankibottom
,t = min(ranki ,t ,0.2)
rankimiddle
,t = min(ranki ,t − rankibottom
,t
rank top
i ,t = min(ranki ,t − rank bottom
i ,t − RANK
where ranki,t is the rank of fund i in month t based on the
measure of past performance, which is past 12-month
outperformance or past 12-month CAPM alpha,
depending on the regression specification.
EMPIRICAL RESULTS
This section describes the results of our empirical
analyses along three research questions: (1) Do factor
premiums exist in mutual fund returns? (2) Do investors
in factor funds successfully harvest factor premiums? and
(3) What drives the allocation decision of investors in
mutual funds?
In the first section of our empirical analysis, we
investigate whether mutual funds following factor
investing strategies earn higher alphas than traditional
actively managed mutual funds. In the second section,
we calculate what returns investors earn by investing in
mutual funds that follow factor investing strategies and
test whether this return is different from the return that
a buy-and-hold investor obtains by randomly selecting
a factor fund. Finally, in the third section, to better
understand how investors dynamically allocate to these
funds, we study the f low-performance relation for factor
mutual funds.
,0.6)
Do Factor Funds Earn Higher Alphas?
middle
i ,t ,0.2) In our first analyses we consider the distribution
of fund alphas for various fund classifications. Exhibit 2
shows that factor funds earn significantly higher alphas
than traditional actively managed mutual funds. Only
17% of the traditional actively managed mutual funds
earn positive alphas after fees in the long run, but this
number is substantially larger for factor funds: 52% for
low-beta funds; 53% for small cap funds; 52% for value
funds; 40% for momentum funds; 57% for profitability
funds; and 60% for investments funds.
To test whether differences in performance are
statistically significant and independent, we perform
a regression analysis in which we regress fund perfor-
mance on fund classifications. The results of this regres-
sion analysis are presented in Panel A of Exhibit 3 and
indicate that factor funds earn significantly higher alphas
than traditional actively managed mutual funds. Specifi-
cally, when we consider the results in our most parsimo-
nious specification (Exhibit 3, Panel A, Regression 7),
we find that funds with exposure to the low-beta, small
cap, value, momentum, profitability, and investments

Quantitative Special Issue 2019 The Journal of Portfolio Management   129

Exhibit 3
Fund Factor Exposures and Outperformance

5HJ 5HJ 5HJ 5HJ 5HJ 5HJ 5HJ

3DQHO$6W\OH3HUIRUPDQFH5HODWLRQVKLS
,QWHUFHSW  ± ±  ± ± ±
WVWDW >@ >±@ >±@ >@ >±@ >±@ >±@
/RZ%HWD  
WVWDW >@ >@
6PDOO&DS  
WVWDW >@ >@
9DOXH  
WVWDW >@ >@
0RPHQWXP  
WVWDW >@ >@
3URILWDELOLW\  
WVWDW >@ >@
,QYHVWPHQWV  
WVWDW >@ >@
5       
3DQHO%6W\OH3HUIRUPDQFH5HODWLRQVKLS&RQWUROOLQJIRU)XQG6SHFLILF&KDUDFWHULVWLFV
,QWHUFHSW ± ± ± ± ± ± ±
WVWDW >±@ >±@ >±@ >±@ >±@ >±@ >±@
/RZ%HWD  
WVWDW >@ >@
6PDOO&DS  
WVWDW >@ >@
9DOXH  
WVWDW >@ >@
0RPHQWXP  
WVWDW >@ >@
3URILWDELOLW\  
WVWDW >@ >@
,QYHVWPHQWV  
WVWDW >@ >@
OQDJH       
WVWDW >@ >@ >@ >@ >@ >@ >@
OQVL]H       
WVWDW >@ >@ >@ >@ >@ >@ >@
H[SBUDWLR ± ± ± ± ± ± ±
WVWDW >±@ >±@ >±@ >±@ >±@ >±@ >±@
WXUQBUDWLR ± ± ± ± ± ± ±
WVWDW >±@ >±@ >±@ >±@ >±@ >±@ >±@
5       

Notes: This exhibit shows univariate and multiple regression results of all US funds during the sample period of January 1990 to December 2015 in the
CRSP Mutual Fund Database with total assets exceeding $5 million. The winsorized (at -2 and 2) Z-score of CAPM alphas is regressed on dummies,
indicating funds belonging to a specific factor group. Ln(age) is the natural logarithm of the fund’s age, calculated as the number of months the fund has been
in our sample. Ln(size) is the natural logarithm of the average fund’s total assets in US dollars. exp_ratio and turn_ratio are the average expense ratio and
turnover ratio per mutual fund in our sample.

130   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
factors have alpha that is, respectively, 0.34, 0.48, 0.20, Exhibit 4
0.12, 0.35, and 0.30 standard deviations higher than Multifactor Exposures and Outperformance
that of traditional actively managed mutual funds. The
t-values of these coefficient estimates are larger than 3 'HS9DULDEOH ]BDOSKD ]BDOSKDFRQWUROV
in all cases, indicating that our results are statistically ,QWHUFHSW ± ±
significant. WVWDW >±@ >±@
In Panel B of Exhibit 3 we extend the analysis by IDFWRU  
controlling our regressions for fund characteristics such WVWDW >@ >@
as fund age, size, total expense ratio, and turnover ratio. IDFWRUV  
WVWDW >@ >@
Our results appear to be robust to controlling for these
IDFWRUV  
fund characteristics because the coefficient estimates and
WVWDW >@ >@
their t-values remain very similar to our first results. •IDFWRUV  
To further understand the effect of factor expo- WVWDW >@ >@
sures on mutual fund performance, we classify funds OQDJH 
according to the number of factors to which they are WVWDW >@
exposed, presented in Exhibit 4. Groups are mutually
exclusive and contain funds with significant loading
to one, two, three, and four or more factors. Results
provide convincing evidence that a larger number of
factor exposures leads to higher risk-adjusted mutual
fund returns even after transaction costs and taxes are
taken into account. The first column shows that funds
with one, two, three, and four or more exposures have
0.37, 0.60, 0.97, and 1.36 standard deviations higher
alpha than funds with no factor exposures.
Results remain intact after controlling for fund-
specific characteristics, as shown in column 2.
Luck versus Skill in Mutual Fund Returns
In this section we take a critical look at our pre-
vious findings. Fama and French (2010) showed that, to
a large extent, the performance of mutual funds can be
attributed to luck. This is a strong and valid argument
against the skill level of outperforming mutual funds;
even if the true alpha is zero in specific periods, it can
be higher or lower than zero just by chance. In the pre-
vious sections we show that funds that incorporate aca-
demic insights in their investment process and provide
exposure to proven factor premiums deliver higher net
alpha relative to the control group. In this section we
take a more conservative approach and test whether the
observed performance exceeds what could have been
generated simply by chance.
Our simulated distribution of mutual fund returns
possesses the important property that true net alpha is
known to be zero, which assumes that all managers have
enough skills to cover for the fees they charge. If the
Quantitative Special Issue 2019
OQVL]H 
WVWDW >@
H[SBUDWLR ±
WVWDW >±@
WXUQBUDWLR ±
WVWDW >±@
5  
Notes: This exhibit shows multiple regression results of all US funds
during the sample period of January 1990 to December 2015 in the
CRSP Mutual Fund Database with total assets exceeding $5 million.
A winsorized (at -2 and 2) Z-score of CAPM alphas (z_alpha) is
regressed on dummies, indicating funds belonging to a specific factor
group. Ln(age) is the natural logarithm of the fund’s age, calculated as the
number of months the fund has been in our sample. Ln(size) is the natural
logarithm of the average fund’s total assets in US dollars. exp_ratio and
turn_ratio are the average expense ratio and turnover ratio per mutual
fund in our sample.
actual distribution of fund returns is skewed to the left,
it shows that fund managers’ skills do not cover for their
expenses and, on average, mutual funds underperform
the market portfolio. If the distribution is skewed to the
right, mutual funds have skills exceeding the fees they
charge and generate added value for their investors.
Exhibit 5 compares the distribution of fund alphas
across all style groups to the simulated distribution.
Consistent with Fama and French (2010), we show
that in the right tail of the distribution, managers have
enough skill to deliver higher returns than their capm
beta predicts. Specifically, at the 90th percentile the
actual distribution has higher t(a) 56% of the time,
compared to the 5,000 simulated runs. At the 99th
percentile this percentage increases to 84 with a t(a) of
2.63, compared to a simulated t(a) of 2.29, indicating
The Journal of Portfolio Management   131
Exhibit 5
Fund Factor Exposures and Outperformance after Controlling for Luck

1R /RZ 6PDOO
6LPXODWHG $OO)XQGV ([SRVXUH %HWD &DS 9DOXH 0RPHQWXP 3URILWDELOLW\ ,QYHVWPHQWV
SHUF  ± ± ± ± ± ± ± ± ±
DFW        
SHUF  ± ± ± ± ± ± ± ± ±
DFW        
SHUF   ± ±    ±  
DFW        
SHUF    ±      
DFW        
SHUF          
DFW        
SHUF          
DFW        
SHUF          
DFW        

Notes: This exhibit shows actual fund performance over simulated performance for different percentiles. Performance is measured by the t-statistic of fund
CAPM alpha t(a). “Simulated” is the average t(a) at the respective percentile over all 5,000 simulated runs. The remaining columns show the actual t(a)
of the respective mutual fund style groups. “% < act” is the percentage of runs out of all 5,000 runs with lower t(a) relative to the actual.

that the returns of top-performing managers are sig-
nificantly higher than the returns that could have been
generated by luck, even after adjusting for the fees they
charge.
This picture significantly changes if we look at
the control group “No exposure.” Even at the 99th
percentile, t(a) is only 2.03, which is higher than a
randomly simulated one in only 24% of the cases. This
result indicates that funds with no factor exposures
systematically fail to deliver positive net alphas that
cannot be attributed to luck. On the contrary, the net
performance of all style groups does not seem to be
attributable to chance. The net alphas of all groups
(except momentum) are significantly higher relative
to those based on our simulated distribution at most of
the percentile levels. Namely, low-beta and value funds
generate positive luck-adjusted net returns in 50% of
the cases and small cap, profitability, and investments
in 75% of the cases. Momentum produces positive luck-
adjusted returns in only the top one percentile, which
might be a result of the higher turnover and total costs
of momentum managers.
These results strengthen the previously docu-
mented positive performance of factor investing funds
because they show that they are much more robust,
surviving even the most conservative tests of our boot-
strapping method. Furthermore, the newly documented
profitability and investments factors seem to be at least
equally robust compared with the well-known value,
size, and low-risk premiums.
In Exhibit 6 we graphically show the cumula-
tive density function of funds with low beta, small
cap, value, momentum, profitability, and investments
exposures. The horizontal axis shows the value of t(a),
and the vertical axis shows the percentile values. The
percentile at which the actual line is below the simu-
lated line indicates that from this percentile onward,
fund managers from the respective group generate
net benchmark-adjusted returns beyond what can be
expected by chance.
Global Markets
In this section we extend the scope of our
research. We conduct our main analysis on a global
markets universe, including long-only equity mutual
funds from all developed countries. In Exhibit 7 we
show the distribution of fund alphas. It strengthens

132   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
the conclusions of our US analysis because all groups
of factor funds have a higher probability of earning a
positive alpha compared to traditionally active global
mutual funds.
In Panel A of Exhibit 8 we show that our US results
spill over to global markets. Funds belonging to all our
style groups—low beta, size, value, momentum, prof-
itability, and investments—deliver higher beta-adjusted
returns compared to the average mutual fund. Results
for the momentum factor stand out; unlike in the US
universe, momentum managers have an alpha that is 0.21
(t-statistic of 6.37) standard deviations higher than that of
non-momentum funds. These results are not explained
by controlling for other style exposures. In our regres-
sion specification 7, in which we include all style dum-
mies simultaneously, we see that with the exception of
value (t-statistic of 1.38), all factor funds have significantly
higher alphas than funds with no factor exposure.

Exhibit 6
Simulated and Actual Cumulative Density Function of CAPM t(a) Factor Funds
/RZ%HWD 6PDOO&DS
 

 

 

 

 

 

 

 

 

 

 
± ± ±     ± ± ±    

9DOXH 0RPHQWXP
 

 

 

 

 

 

 

 

 

 

 
± ± ±     ± ± ±    

6LPXODWHG $FWXDO

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   133

Exhibit 6 (continued)
Simulated and Actual Cumulative Density Function of CAPM t(a) Factor Funds
3URILWDELOLW\ ,QYHVWPHQWV
 

 

 

 

 

 

 

 

 

 

 
± ± ±     ± ± ±    

6LPXODWHG $FWXDO

Notes: This exhibit shows cumulative density function of actual fund performance over simulated performance. Performance is measured by the t-statistic
of the fund CAPM alpha t(a). “Simulated” is the average t(a) at the respective percentile over all 5,000 simulated runs.

Exhibit 7
Distribution of Fund Alphas—Global Markets

± ±± ±   !  ! 1RRI)XQGV

$OO)XQGV         
1R([SRVXUH         
/RZ%HWD         
6PDOO&DS         
9DOXH         
0RPHQWXP         
3URILWDELOLW\         
,QYHVWPHQWV         

Notes: This exhibit shows the distribution of annualized fund alphas across all global developed market funds in the Morningstar Mutual Fund Database
with total assets exceeding $5 million. Alphas are calculated per fund as the intercept from the CAPM regressions over all available observations during
the sample period of January 1991 to December 2015. “< -4” shows the percentage of funds with annualized alphas less than -4%; “-4:-2” shows
the percentage of funds with annualized alphas between -2% and -4%; and “% > 0” shows the number of funds with positive alphas.

Panel B extends the analysis by controlling for
fund-specific characteristics and shows that the supe-
rior performance of factor funds cannot be attributed
to their age or size. Only the coefficient on our value
group becomes negative but insignificant (t-statistic
of -0.77). On the other hand, momentum results remain
robust, indicating that despite the higher turnover, the
momentum premium can be harvested in practice. The
newly documented factors profitability and investments
seem to be some of the strongest factors; their coeffi-
cients remain positive and highly significant in all our
tests.

134   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
Exhibit 8
Factor Exposures and Outperformance—Global Markets

5HJ 5HJ 5HJ 5HJ 5HJ 5HJ 5HJ

3DQHO$6W\OH3HUIRUPDQFH5HODWLRQVKLS
,QWHUFHSW  ±  ± ± ± ±
WVWDW >@ >±@ >@ >±@ >±@ >±@ >±@
/RZ%HWD  
WVWDW >@ >@
6PDOO&DS  
WVWDW >@ >@
9DOXH  
WVWDW >@ >@
0RPHQWXP  
WVWDW >@ >@
3URILWDELOLW\  
WVWDW >@ >@
,QYHVWPHQWV  
WVWDW >@ >@
5       
3DQHO%6W\OH3HUIRUPDQFH5HODWLRQVKLS&RQWUROOLQJIRU)XQG6SHFLILF&KDUDFWHULVWLFV
,QWHUFHSW ± ± ± ± ± ± ±
WVWDW >±@ >±@ >±@ >±@ >±@ >±@ >±@
/RZ%HWD  
WVWDW >@ >@
6PDOO&DS  
WVWDW >@ >@
9DOXH ± ±
WVWDW >±@ >±@
0RPHQWXP  
WVWDW >@ >@
3URILWDELOLW\  
WVWDW >@ >@
,QYHVWPHQWV  
WVWDW >@ >@
OQDJH       
WVWDW >@ >@ >@ >@ >@ >@ >@
OQVL]H       
WVWDW >@ >@ >@ >@ >@ >@ >@
5       

Notes: This exhibit shows univariate and multiple regression results of all global developed market funds during the sample period of January 1991 to
December 2015 in the Morningstar Mutual Fund Database with total assets exceeding $5 million. A winsorized (at -2 and 2) Z-score of CAPM alphas
is regressed on dummies, indicating funds belonging to a specific factor group. Ln(age) is the natural logarithm of the fund’s age, calculated as the number
of months the fund has been in our sample. Ln(size) is the natural logarithm of the average fund’s total assets in US dollars.

Do Investors in Factor Funds Successfully
Harvest Factor Premiums?
In the previous section we provided evidence that
factor premiums survive even the most robust research
specifications and that mutual fund managers seem to
be able to harvest these premiums. However, the actual
returns that investors earn by investing in factor mutual
funds appear to be significantly lower because inves-
tors do not seem to follow buy-and-hold strategies but,
rather, dynamically reallocate their funds across both

Quantitative Special Issue 2019 The Journal of Portfolio Management   135

Exhibit 9
Mutual Fund Returns versus Investor Returns

1R
$OO)XQGV ([SRVXUH /RZ%HWD 6L]H 9DOXH 0RPHQWXP 3URILWDELOLW\ ,QYHVWPHQWV
D7KHRUHWLFDO)DFWRU5HWXUQV       
E0XWXDO)XQGV%X\DQG+ROG(:        
F0XWXDO)XQGV%X\DQG+ROG9:        
G0XWXDO)XQGV'ROODU:HLJKWHGSHU)DFWRU        
'LIIHUHQFHG±E ± ±  ± ± ± ± ±
39DOXH'LIIHUHQFH        

Notes: This exhibit shows returns of all US funds during the sample period of January 1991 to December 2015 in the CRSP Mutual Fund Database with total
assets exceeding $5 million. “EW” is the equally weighted, geometrically calculated, annualized return; “VW” is the total asset-weighted, geometrically calculated,
annualized return; and “Dollar-Weighted” is the geometrically annualized internal rate of return (IRR), or the rate of return that makes the sum of discounted
ending total assets and the present value of monthly distributions equal to the initial total assets. The IRR calculation is done per factor level. It calculates the IRR
on an aggregate level because distributions are calculated based on the sum of all assets per fund style and value-weighted fund returns in the same style group.
Theoretical factor returns are downloaded from Kenneth French Data Library. Value, momentum, profitability, and investments are based on six portfolio sorts as
the average of small attractive and big attractive portfolios. For example, value is the average of small value and big value portfolios. Size is the average of the small
value, small growth, and small middle portfolios based on six size-value sorted portfolios. Low beta is the lowest quintile based on a past market beta sort.

factors and factor managers. By attempting to time
across factors, investors lose a large portion of the return
they could earn with a buy-and-hold strategy, and in this
section we quantify this loss. To do so we calculate the
magnitude of factor premiums in three settings, gradu-
ally reducing the level of abstraction. First, we calculate
the long-only premiums of Fama and French (2015) and
the low beta factors. Second, we calculate the return of
mutual funds with exposure to these factors. Finally, we
estimate the returns realized by investors in these funds.
This analysis extends work by Hsu (2016), who showed
that investors in value and small cap funds underper-
formed the S&P 500, whereas the value and small cap
funds themselves outperformed the benchmark.
Exhibit 9 describes the main results of this section.
For each factor group we show the return of the long-
only academic factor return, equally and value-weighted
mutual fund returns, and dollar-weighted returns at
an aggregation level per factor. Starting with the first
column we see that the market portfolio has earned a
buy-and-hold return of 10.1% per annum, compared to
9.6% for the average mutual fund. Moving from top to
bottom in the exhibit reduces the level of abstraction in
calculating returns and gets closer in approximating the
return to the end investors. A value-weighted return of
9.2% implies that larger mutual funds have generated
lower returns than their small counterparts, consistent
with studies such as that by Chen et al. (2004). Next, we
calculate the dollar-weighted return on an aggregated
level by summing the dollar amount of all assets and
calculating the IRR according to the methodology pro-
posed by Dichev (2007). The resulting return of 7.9%
per annum captures the amount of equity timing or
fund f lowing in and out of our sample of US long-only
domestic equity funds. For example, if a fund has strong
returns in the first year of its existence but a very low
asset base, very few investors benefit from it. If later,
because of good performance, it attracts inf lows, but
subsequent returns are lower, the return of the fund will
be higher than the return of the investors in this fund
over the sample period.
The remaining columns of Exhibit 9 show the
same analysis per different groups of funds. Row (d)
shows that size, value, momentum, profitability, and
investments fund investors lose, respectively, 2.5%,
1.2%, 3.0%, 1.3%, and 0.7% as a result of factor timing.
The highest loss is incurred by investors in momentum
funds. A possible reason is the intrinsic trend following
the nature of momentum, which stimulates investors
to allocate to momentum funds after a period of good
performance and which lowers their subsequent realized
returns compared to the fund returns.
Our results have strong implications for mutual
fund investors. Even though factor funds deliver posi-
tive alpha, their investors have not been able to capture
it because of their allocation decisions. For example, if
investors believe in the value and momentum premiums
but they only invest in value or momentum funds, after

136   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
Exhibit 10
Multifactor Mutual Fund Returns and Investor Returns

1R([SRVXUH )DFWRU )DFWRUV )DFWRUV )DFWRUV

D0XWXDO)XQGV%X\DQG+ROG(:     
E0XWXDO)XQGV%X\DQG+ROG9:     
F0XWXDO)XQGV'ROODU:HLJKWHGSHU)DFWRU     
'LIIHUHQFHF±D ± ± ± ± ±
39DOXH'LIIHUHQFH     

Notes: This exhibit shows returns of all US funds during the sample period of January 1991 to December 2015 in the CRSP Mutual Fund Database
with total assets exceeding $5 million. “EW” is the equally weighted, geometrically calculated, annualized return; “VW” is the total asset-weighted,
geometrically calculated, annualized return; and “Dollar-Weighted” is the geometrically annualized internal rate of return (IRR), or the rate of return
that makes the sum of discounted ending total assets and the present value of the monthly distributions equal to the initial total assets. The IRR calculation
is done per factor level. It calculates the IRR on an aggregate level because distributions are calculated based on the sum of all assets per fund style and
value-weighted fund returns in the same style group.

a period of strong performance they lose a significant
portion of the factor premium because of the cyclicality
in factor returns. A potential solution would be to buy
both funds and hold on to them instead of moving assets
across them.
Exhibit 10 shows the annualized returns of funds
with one, two, three, and four or more factor expo-
sures. The average buy-and-hold investor would have
earned 9.1%, 9.9%, 10.3%, and 10.6%, compared to
8.0% in funds with no factor exposures. Similar to the
single-factor funds, the dollar-weighted returns of 7.4%,
7.6%, 8.9%, and 8.9% are lower than the time-weighted
returns for the same group of funds, indicating that even
if they invest in multifactor funds, investors still make
allocation decisions that cost them a significant portion
of performance.
Exhibit 11 graphically illustrates our main points.
The bars represent the return premium over tradition-
ally actively managed mutual funds, or “no exposure”
funds. The average academic long-only factor of Fama
and French (2015) outperforms our control group with
4.7% per annum. The second bar shows the return
of mutual funds with one-factor exposure. It is based
on net asset values and, as such, includes the negative
effect of taxes and trading costs on returns. This brings
down the premium to 1.1%, which is the return gener-
ated by investors invested in this group of funds at the
beginning of our sample and holding on to the invest-
ment until the end of the sample; however, investors
often make active allocation decisions based on their
views on which factor is going to outperform going
forward. The third bar incorporates the effect of these
decisions on returns. It appears that despite one-factor
mutual funds having a sizable return premium of 1.1%
over traditionally actively managed mutual funds, the
investors in those same funds underperform the control
group with 0.5% as a result of poor timing decisions.
The right three bars of the exhibit present the returns
investors could have had if they allocated to funds with
two, three, or four factors and held on to them instead
of timing across factors. The buy-and-hold premium is
1.9%, 2.4%, and 2.7%, respectively.
What Drives Allocation Decisions
of Mutual Fund Investors?
In this section we unveil the drivers behind
investor allocation decisions. This is crucial in having
a full understanding of why investors consistently lose
returns even when selecting the right funds. The nat-
ural starting point is to follow the insights of Sirri and
Tufano (1998), who show that because of the complexity
of a full understanding of the methodology of each
strategy, investors just buy funds with high past returns,
assuming that past returns are an accurate proxy for
manager skills. We follow their piecewise linear regres-
sion specification to control for the different degree of
sensitivity of f lows to performance in the tails of the
performance distribution.
Our article offers convincing evidence that funds
exposed to factors outperform in the long term, and
investors who strategically allocate to them can benefit
from those premiums. As such, we test the hypothesis of
whether investors allocate to factor funds strategically or

Quantitative Special Issue 2019 The Journal of Portfolio Management   137

Exhibit 11
Outperformance over Traditional Actively Managed Mutual Funds


 













± ±

±

±

±
7KHRUHWLFDO)DFWRU 0XWXDO)XQG ,QYHVWRU5HWXUQV )XQGVZLWK )XQGVZLWK )XQGVZLWK
5HWXUQV 5HWXUQV)DFWRU )DFWRU )DFWRU )DFWRUV )DFWRUV
%X\DQG+ROG %X\DQG+ROG 'ROODU:HLJKWHG %X\DQG+ROG %X\DQG+ROG %X\DQG+ROG

)DFWRU)XQGV 7UDGLWLRQDO$FWLYHO\0DQDJHG0XWXDO)XQGV%X\DQG+ROG

Notes: This exhibit is based on returns of all US funds during the sample period of January 1991 to December 2015 in the CRSP Mutual Fund Data-
base with total assets exceeding $5 million. All returns are calculated in excess of the returns of traditional actively managed mutual funds ( fund with no
positive factor exposures). Mutual fund returns are equally weighted, geometrically calculated, annualized returns. Investor returns are the dollar-weighted,
geometrically annualized internal rate of return (IRR), or the rate of return that makes the sum of discounted ending total assets and the present value of
monthly distributions equal to the initial total assets. The IRR calculation is done per factor level. It calculates the IRR on an aggregate level because the
distributions are calculated based on the sum of all assets per fund style and value-weighted fund returns in the same style group. Styles are funds with one,
two, three, or four or more factor exposures. The theoretical return is the average long-only Fama and French (2015) factor returns and the lowest quantile
based on low beta sorts. Value, momentum, profitability, and investments are based on six portfolio sorts as the average of small attractive and big attractive
portfolios. For example, value is the average of small value and big value portfolios. Size is the average of the small value, small growth, and small middle
portfolios based on six size-value sorted portfolios. Low beta is the lowest quintile based on past market beta sorts.

simply end up investing in factor funds because of their
good past performance. Toward this goal, we extend
the Sirri and Tufano (1998) f low-performance model
with dummies that indicate whether funds belong to a
certain factor group.
Exhibit 12 contains the main results of this sec-
tion. In regression 1 we show that relative f lows are
significantly higher for funds with high past 12-month
outperformance over the market (a coefficient of 0.02,
with a t-statistic of 12.02). This effect is highly non-
linear; the top (bottom) group has a significant coef-
ficient of 0.04 (-0.03), meaning that funds belonging
to this group exhibit abnormally high (low) f lows.
Most importantly, it seems that past performance is
the main driver of investor decisions. The coefficients
on size, value, and momentum dummies are negative,
indicating that investors tend to avoid those funds. The
coefficient on profitability is insignificant, and only the
coefficients on low beta and investments are positive and
significant, showing that investors invest in low beta and
investments funds more than their past 12-month returns
would suggest. In regression 2 we extend the analysis
and test whether the allocation based on past returns is a
good timing decision in terms of future returns. As such,
we include the future 12-month returns in the equation.
The coefficient on future performance is virtually zero

138   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
Exhibit 12
Flow-Performance Relationship

5HJ 5HJ 5HJ 5HJ

0RQWK 0RQWK 0RQWK 0RQWK
2XWSHUIRUPDQFH 2XWSHUIRUPDQFH $OSKD $OSKD
,QWHUFHSW    
>@ >@ >@ >@
/RZ%HWD    ±
>@ >@ >@ >±@
6L]H ± ± ± ±
>±@ >±@ >±@ >±@
9DOXH ± ± ± ±
>±@ >±@ >±@ >±@
0RPHQWXP ± ± ± ±
>±@ >±@ >±@ >±@
3URILWDELOLW\    
>@ >@ >@ >@
,QYHVWPHQWV    
>@ >@ >@ >@
OQDJH ± ± ± ±
>±@ >±@ >±@ >±@
OQVL]H    
>@ >@ >@ >@
H[SBUDWLR ± ± ± ±
>±@ >±@ >±@ >±@
WXUQBUDWLR    
>@ >@ >@ >@
3DVW3HUIRUPDQFH    
>@ >@ >@ >@
%RWWRP ± ± ± ±
>±@ >±@ >±@ >±@
0LGGOH ± ± ± ±
>±@ >±@ >±@ >±@
7RS    
>@ >@ >@ >@
)XWXUH3HUIRUPDQFH  
>@ >@
5    

Notes: This exhibit shows Fama and Macbeth (1973) multiple regression results of all US funds during the sample period of January 1991 to December
2015 in the CRSP Mutual Fund Database. Each month, relative fund flows are regressed on dummies, indicating funds belonging to a specific factor group
and measures on performance. Ln(age) is the natural logarithm of the fund’s age, calculated as the number of months the fund has been in our sample at
each point in time. Ln(size) is the natural logarithm of the most recent fund’s total assets in US dollars, and exp_ratio and turn_ratio are the most recent
expense ratio and turnover ratio per mutual fund at each point in time. Presented coefficients are the average coefficients over the full sample, and t-statistics
are calculated as in Fama and Macbeth (1973).

(0.00, with a t-statistic of 0.09), meaning that investing
in funds with high past performance has no predictive
power for future performance. In regression 3 we sub-
stitute the past return with the past CAPM alpha, and
results remain intact. The only difference is that the
positive coefficient on the low beta dummy becomes
zero, indicating that investors allocate to low beta funds
just as much as their past alpha implies.
This section presents evidence that fund f lows
have been driven by factor funds earning high past

Quantitative Special Issue 2019 The Journal of Portfolio Management   139

returns and not by the funds providing factor exposures.
Consistent with Zheng (1999), we find very little evi-
dence of a “smart money” effect, in the sense that f lows
predict future fund performance. Instead of strategically
allocating to factor premiums, investors seem to avoid
them and allocate to factor funds only if their perfor-
mance has been good. This, combined with the poor
predictive power of f lows to future returns, indicates
that investors indeed time poorly, as proposed by Hsu
(2016). This, in turn, explains the observed effect that
investor returns are lower than fund returns. As such,
investor behavior has important implications for the rea-
sons why factor premiums continue to exist.
CONCLUSIONS
Mutual funds following factor investing strate-
gies based on equity asset pricing anomalies such as the
small cap, value, momentum, profitability, and invest-
ment effects earn significantly higher alphas than tradi-
tional actively managed mutual funds. A buy-and-hold
strategy for a random factor fund would yield 110 bps per
annum in excess of the return earned by the average tra-
ditionally actively managed mutual fund. However, the
actual returns that investors earn by investing in factor
mutual funds appear to be significantly lower than this
number because investors dynamically reallocate their
funds across both factors and factor managers. Although
factor funds have attracted significant fund f lows over
our sample period, it appears that investor fund f lows
have been driven by factor funds earning high past
returns and not by the funds providing factor expo-
sures. We argue that rather than timing factors and factor
managers, investors would be better off by using a buy-
and-hold strategy and selecting a multifactor manager.
REFERENCES
Chen, J., et al. 2004. “Does Fund Size Erode Mutual Fund
Performance? The Role of Liquidity and Organization.”
American Economic Review 94 (5): 1276–1302.
Cornell, B., J. Hsu, and D. Nanigian. 2017. “Does Past Per-
formance Matter in Investment Manager Selection?” The
Journal of Portfolio Management 43 (4): 33–43.
140   Factor Investing from Concept to Implementation Quantitative Special Issue 2019
Dichev, I. D. 2007. “What Are Stock Investors’ Actual His-
torical Returns? Evidence from Dollar-Weighted Returns.”
American Economic Review 97 (1): 386–401.
Dichev, I. D., and G. Yu. 2011. “Higher Risk, Lower Returns:
What Hedge Fund Investors Really Earn.” Journal of Financial
Economics 100 (2): 248–263.
Fama, E. F., and K. R. French. 2010. “Luck versus Skill in
the Cross-Section of Mutual Fund Returns.” The Journal of
Finance 65 (5): 1915–1947.
——. 2015. “A Five-Factor Asset Pricing Model.” Journal of
Financial Economics 116 (1): 1–22.
Fama, E. F., and J. D. MacBeth. 1973. “Risk, Return, and
Equilibrium: Empirical Tests.” Journal of Political Economy 81
(3): 607–636.
Hsu, J., B. W. Myers, and R. Whitby. 2016. “Timing Poorly:
A Guide to Generating Poor Returns While Investing in
Successful Strategies.” The Journal of Portfolio Management 42
(2): 90–98.
Sirri, E. R., and P. Tufano. 1998. “Costly Search and Mutual
Fund Flows.” The Journal of Finance 53 (5): 1589–1622.
Van Gelderen, E., and J. Huij. 2014. “Academic Knowledge
Dissemination in the Mutual Fund Industry: Can Mutual
Funds Successfully Adopt Factor Investing Strategies.” The
Journal of Portfolio Management 40 (114): 157–167.
Zheng, L. 1999. “Is Money Smart? A Study of Mutual Fund
Investors’ Fund Selection Ability.” The Journal of Finance
54 (3): 901–933.
Disclaimer
The views expressed in this article are solely those of the authors and not
necessarily shared by PSP Investments, Robeco, or their subsidiaries.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

Demir Bektić
is a senior portfolio manager,
fixed income at Deka
Investment GmbH, a member
of IQ-KAP, and an assistant
professor at Darmstadt
University of Technology in
Darmstadt, Germany.
demir.bektic@gmail.com
Josef -Stefan Wenzler
is a senior portfolio manager,
fixed income at Deka
Investment GmbH and a
member of IQ-KAP in
Frankfurt, Germany.
josef-stefan.wenzler@deka.de
M ichael Wegener
is a managing director and
head of quant products
equities and fixed income at
Deka Investment GmbH and
a general manager of IQ-KAP
in Frankfurt, Germany.
michael.wegener@deka.de
Dirk Schiereck
is a professor at Darmstadt
University of Technology in
Darmstadt, Germany.
schiereck@bwl.tu-darmstadt.de
T imo Spielmann
is a portfolio manager at
Deka Investment GmbH and
a member of IQ-KAP in
Frankfurt, Germany.
timo.spielmann@deka.de
Quantitative Special Issue 2019
Extending Fama–French Factors
to Corporate Bond Markets
Demir Bektić, Josef-Stefan Wenzler, Michael Wegener,
Dirk Schiereck, and Timo Spielmann
D
o equity market factors explain
corporate bond return dynamics?
This empirical relationship
is of substantial importance to
investors (interested in return and diversifica-
tion characteristics), economists (who want
to know the mechanisms that connect these
markets), and not least policymakers (con-
cerned about the stability of the financial
system).1 Although the relevant empirical
studies can be traced back to the seminal work
by Fama and French (1993), there is limited
direct evidence in the literature concerning
the pricing of risk across stock and bond
markets. Although Merton (1974) provided
an intuitive approach to describe the link
between equities and corporate debt based
on contingent claims, we are still lacking
relevant factors to describe the dynamics of
credit markets. We aim to fill this gap by
examining whether the four risk factors in
the Fama and French (2015) framework—size
(SMB), value (HML), profitability (RMW),
and investment (CMA)—are priced in the
corporate bond market.
1
 As long as some investors have access to both
equity and corporate bond markets, the absence of
arbitrage opportunities imposes cross-market restric-
tions on the stochastic discount factor (SDF). The SDF
is sometimes referred to as the pricing kernel and ref lects
the fact that the price of an asset can be computed by
discounting the future cash f low by a stochastic factor.
Analyzing the link between equity and
corporate bond factors is a tempting proposi-
tion for the following reasons. First, over the
years evidence has mounted that alternative
risk premiums beyond the traditional asset
class risk premiums (e.g., equity and term
premium) do indeed exist. Harvey, Liu, and
Zhu (2016) provided an excellent summary
of factors in the equity space and recount
more than 300 papers on cross-sectional
return patterns published in various journals.
According to structural credit risk models,
both equity and corporate debt are driven by
the fundamentals of the same underlying cor-
poration, implying that stock prices and credit
spread changes must be related to ensure the
absence of arbitrage. Consequently, risk pre-
miums in equity and corporate bond mar-
kets should be related.2 Second, although the
relationship between firms’ default risk and
equity risk premiums has been analyzed in
numerous studies (see Vassalou and Yuhang
2004 or, more recently, Friewald, Wagner,
and Zechner 2014), there is little evidence
2
 Kapadija and Pu (2012) argued that there is
cross-sectional variation in the correlation of equity
and credit markets and that short-horizon pricing
discrepancies across equities and corporate bonds are
common and predominantly anomalous because of
limited arbitrage activity. However, these authors also
suggested that the Merton (1974) model is appropriate
for longer time scales.
The Journal of Portfolio Management   141
on whether corporate bond returns exhibit anomalies
similar to those in stock markets.
To alleviate this shortcoming and to provide a
more comprehensive insight into return dynamics
between equity and debt, we explore the proposition
that size, value, profitability, and investment, as origi-
nally defined by Fama and French (2015) for equity mar-
kets, do indeed extend their explanatory power to US
high yield (HY) and US and European investment-grade
(IG) credit markets.
We contribute to the literature in several ways. We
depart from previous research by employing the original
Fama and French (2015) equity factor definitions of size,
value, profitability, and investment for corporate cash
bonds. Thus, we form portfolios by sorting the cross sec-
tion of bonds into deciles based on corresponding com-
pany characteristics, and then we examine their time
series performance. If these factors are rational pricing
factors (or mispricings caused by behavioral biases), their
factor risk premiums estimated in one market should
arguably be consistent with those estimated in the other.
Given the recent interest in studying equity factors in
the US corporate bond market, analysis of if and how
such factors carry their explanatory power into other
corporate bond markets seems warranted. Finally, the
recent shift toward factor-based investment strategies in
general (sometimes referred to as smart beta) has led to a
revived interest into risk factors.
In the past, empirical research on relevant pricing
factors focused predominantly on equity markets. In the
early 1990s, Fama and French (1992, 1993) introduced
a factor model based on firm-specific factors to explain
cross-sectional stock returns. They demonstrated that
size, value, and beta factors can account for up to 95%
of variability in US stock market returns. A stunning
result that opened the door for many extensive studies
on factor-based investing, leading to a multitude of new
factors (sometimes referred to as the factor zoo)3 and factor
models in the equity space.4 Finally, Fama and French
(2015) enriched their traditional three-factor model by
adding a measure of firm profitability and investment,
showing that the new five-factor model performs better
than their three-factor model.5
3
 See Cochrane (2011).
4
 See Harvey, Liu, and Zhu (2016).
5
 By adding profitability and investment to their model,
the value factor of the three-factor model becomes redundant for
describing average returns in the US stock market. Nevertheless,
142   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
Despite the apparent success of factor-based
investing, the abundance of academic research on factor-
based investing in equity markets and the fact that global
fixed-income markets are bigger than global equity mar-
kets (see Crawford et al. 2015; Goldstein, Jiang, and Ng
2018; or Israel, Palhares, and Richardson 2018), sim-
ilar research for fixed-income securities is less mature.
However, documented corporate bond factors in the
literature include low volatility (Ilmanen et al. 2004),
momentum ( Jostova et al. 2013), value (Correia, Rich-
ardson, and Tuna 2012), and size (Houweling and van
Zundert 2017). Finally, Israel, Palhares, and Richardson
(2018) found that carry, low volatility, momentum, and
value explain nearly 15% of the cross-sectional variation
in US corporate bond excess returns.
In addition to individual factor portfolios, we
investigate the performance of investable equal-weighted
multifactor portfolios by combining size, value, profit-
ability, and investment. Although we concede that a
long–short portfolio might lead to superior risk-adjusted
returns in theory, we consider long–short credit portfo-
lios somewhat impractical because of their operational
difficulties and the high transaction costs associated with
shorting corporate bonds (especially for lower-rated or
illiquid securities). Moreover, most corporate bond
investors are restricted to long-only portfolios. However,
performance and diversification benefits can be achieved
by combining all four factors, which improves investor’s
overall portfolio return. Interestingly, the Sharpe ratios
(SRs) of multifactor portfolios are up to 30% higher
compared to those of the market. These results remain
robust even after accounting for transaction costs.
Finally, our results suggest that corporate bond
returns cannot be fully subsumed by traditional equity
risk factors because corporate bond factor risk premiums
are not the same across the two markets. These find-
ings indicate that the four Fama and French equity risk
factors are priced differently in credit markets, a puzzle
for modern asset-pricing theory that implies market seg-
mentation (see Choi and Kim 2016) and needs further
investigation. Because we conduct an out-of-sample
study only (we use established equity factor defini-
tions), we ensure that our results do not suffer from
in-sample bias.
investors interested in portfolio tilts toward size, value, profitability,
and investment premiums should consider all five factors recom-
mended by Fama and French (2015).
RETURN DYNAMICS BETWEEN
EQUITY AND DEBT
Relating equity and corporate bond returns is not
trivial but represents a natural starting point. Rational
asset-pricing models suggest that risk premiums in the
equity market should be consistent with those of the cor-
porate bond market, assuming that the two markets are
integrated. The earliest formalized structural credit risk
model developed by Merton (1974) provides important
intuition on why changes in equity and corporate bond
returns should be related because they represent con-
tingent claims against the assets of the same company.
The only state variable in the model is the value of
the firm V, and one of the main assumptions is that the
value of a company’s assets follows a geometric Brownian
motion W under the risk-neutral martingale measure Q
in which m and s are the drift and volatility, respectively.
dVt = Vt µdt + Vt σdWtQ (1)
It is assumed that the company issues only a single
zero-coupon bond with face value F payable at T in
which the payoff to the creditors at date T is
D(VT ,T ) = min(VT , F ) = F − (F − VT )+ (2)
To relate equity and debt in the Merton model,
equity is valued as a call option on the value of assets,
and applying the accounting identity or put–call parity
equates the value of debt D and equity E:
E(Vt , t ) = CallBS (Vt , F ,µ,T − t, σ ) (3)
D(Vt , t ) = P (t,T ) − PutBS (Vt , F ,µ,T − t, σ ) (4)
The model makes it clear that the spread between
risky credit debt and risk-free debt is the value of the put
option.6 Consequently, possible determinants of credit
spreads, and hence the key factors that inf luence credit
spreads, are the company’s business risk of the assets s,
maturity of the debt T, and the leverage F.
The theoretical link that equities and corporate
bonds of a company are connected through their expo-
sure to the underlying company value is an important
6
 CallBS (Vt, F, m, T - t, s) denotes the value of a call option
and PutBS (Vt, F, m, T - t, s) is the value of a put option according
to Black and Scholes (1973).
Quantitative Special Issue 2019
insight from the formalized Merton model. Therefore,
equity market factors are relevant for pricing corpo-
rate debt only if they capture changes in firm value or
changes in risk-neutral probabilities.
DATA AND METHODOLOGY
Data
Similar to Israel, Palhares, and Richardson (2018),
we use monthly data from the Bank of America Mer-
rill Lynch (BAML) for this analysis. Prices are provided
by BAML traders and are used as the primary pricing
source. The dataset includes monthly data of all senior
US HY, US IG, and European IG corporate bond issues
rated by at least one of the three major rating agencies
(S&P, Moody’s, and Fitch), issued in euros or US dollars.
The employed BAML indexes only include bonds with
a minimum amount outstanding of 250 million for IG
and 100 million for HY in local currency terms,7 a fixed
coupon schedule, and a minimum remaining time to
maturity of one year. Newly issued bonds must exhibit
a time to maturity of at least 18 months.8
European HY bonds are not included in this anal-
ysis because of the insufficient size of the European HY
market until 2013. As done by Elton et al. (2001), put-
table bonds are excluded. We further eliminate subor-
dinated and contingent capital securities (cocos) as well
as taxable and tax-exempt US municipal, equity-linked,
securitized, dividends-received deduction (DRD)
eligible,9 and legally defaulted securities because these
have distinctly different payout characteristics compared
to standard senior coupon bonds.
The dataset covers the period from December 1996
to December 2016 for US HY and IG bonds and from
December 2000 to December 2016 for European IG
bonds.10 Since the adoption of the euro in 1999, BAML
7
 This is similar to the equity market anomaly literature, in
which too small stocks are typically removed to ensure that results
are not driven by market microstructure or liquidity.
8
 Removing bonds that have less than one year to maturity
is applied to all major corporate bond indexes, such as Citi Fixed
Income Indices and Barclays Capital Corporate Bond Index as well
as the BAML Corporate Master Index. The 18-month cutoff for
newly issued bonds is a standard choice of BAML.
9
 A DRD is a tax deduction received by a corporation on the
dividends paid by companies in which it has an ownership stake.
10
 Trade Reporting and Compliance Engine (TRACE) trans-
action data are available for US bonds only and do not start before
July 2002.
The Journal of Portfolio Management   143
gradually introduced euro-denominated bonds into its
indexes. Euro-denominated IG indexes reached critical
mass of at least 100 unique publicly traded issuers to be
included in this study in December 2000. Because all
four factors are based on financial statement ratios and
equity market data obtained from FactSet Fundamen-
tals, only publicly traded corporations are considered
in this analysis.11 Furthermore, we use a six-month lag
to ensure that financial statement information is com-
pletely priced in by bond market participants and to
avoid a forward-looking bias in our analysis (see Bhojraj
and Swaminathan 2009). In total, our sample contains
1,272,900 unique bond-month observations: 248,820 in
US HY and 849,684 (174,396) in US IG (European IG).
Exhibit 1 reports the descriptive statistics of the time-
series averages. The average number of observations per
month is 1,036 in US HY markets and 3,540 (855) in
US IG (European IG) markets.
Exhibit 1 provides further summary statistics of
our dataset, including the most important bond charac-
teristics, such as duration, spread, and rating. The total
return of corporate bonds is predominantly driven by
two components: interest rates (term premium) and
credit spreads (default premium). Only the latter compo-
nent is relevant in the context of factor-based investing
in credit markets because interest rate changes are usu-
ally independent of credit spreads, and the main purpose
of investing in credit securities is to additionally earn the
default premium. Therefore, to evaluate unbiased factor
returns of credit portfolios, one must consider excess
returns of corporate bonds versus duration-matched sov-
ereign bond returns. These returns, which are provided
by BAML as well, are net of interest rate effects and
thus ref lect the credit premium component of corpo-
rate bonds only. Furthermore, BAML accounts for all
defaults in its monthly excess returns, thereby freeing
the dataset of survivorship bias.
Empirical Methodology
The factors at the heart of this study require equity
data; therefore, only publicly traded issuers are included
in this analysis. The resulting number of issuers in the
universe are further subdivided into three subsamples
11
 Typically between 85% and 90% of the companies consid-
ered for this study publish accounting data, and between 50% and
55% are publicly traded firms.
144   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
according to their ratings and issuance currencies (USD
HY, USD IG, and EUR IG) to accommodate the fact
that bonds with varying credit risks and currencies
exhibit different market behavior (see Merton 1974) and
transaction costs (see Chen, Lesmond, and Wei 2007).
A separation that also prevails in practice because most
investors are looking for either HY or IG bonds denomi-
nated in a particular currency. Moreover, leading index
providers and regulation authorities consequently split
these two market segments. Therefore, our US HY, and
IG subsamples consist of all US dollar–denominated HY
and IG bonds included in the dataset. Similarly, the
European IG universe contains all euro-denominated
IG bonds.
We investigate the existence of factor premiums
for each of the three bond subsamples via decile analysis.
That is, issuers are ranked and grouped into 10 deciles
according to their size, value, profitability, and invest-
ment scores. The weighting scheme for each bond in
each decile is based on the total number of issuers in
the particular bond subsample. Similar to Baker and
Wurgler (2012) and Choi and Kim (2016), we ensure that
decile portfolios are not dominated by single large issuers
and weight each issuer equally rather than employing a
market-capitalization weighting scheme. Accordingly,
we use equal-weighted benchmarks for each market and
segment we study.
Let N be the number of unique issuers in the uni-
verse and let M (n) be the total number of bonds corre-
sponding to issuer n. Then the weight for each issuer is
given by 1/N and the weight for each bond m of issuer n
by 1/N × 1/M (n) = ωnm. The total weight is then given
by Σ n =1 Σ m =1 ω nm = 1, which is distributed equally among
N M (n )
all deciles to ensure that each decile accounts for 10%
of the total weight. The decile portfolios are rebalanced
on a monthly basis (see Israel, Palhares, and Richardson
2018). Given the weighting scheme and monthly excess
returns of each bond, the performance of each decile
for each factor portfolio and bond subsample can be
computed.
Defining Corporate Bond Factors
Size. From an economic point of view, smaller
companies are typically associated with lower liquidity,
higher distress, and more downside risk than larger
firms. Therefore, smaller companies should outper-
form larger firms to compensate investors for taking on
Exhibit 1
Summary of Universe Statistics

Avg. No. of Avg. No. of Avg. No. of Avg. No. of Avg. Modified Avg. Option
Year Firms Public Firms Private Firms Bonds Duration Adjusted Spread Avg. Rating
Panel A: US HY Universe
1997 333 151 182 324 4.18 285 13.65
1998 381 173 208 337 4.38 405 13.65
1999 465 220 245 406 4.37 579 13.73
2000 485 239 246 503 4.2 660 13.81
2001 501 259 242 611 3.89 1,110 14.06
2002 511 293 218 779 3.92 1,369 14.12
2003 615 361 255 980 4.13 802 14.14
2004 719 410 309 1,036 4.28 411 14.04
2005 745 420 325 1,021 4.16 360 14.00
2006 757 424 334 965 4.15 329 13.97
2007 709 404 305 798 4.25 330 13.76
2008 819 460 359 890 4.09 921 13.84
2009 782 477 305 981 3.68 1,639 14.26
2010 876 529 346 1,106 3.94 662 14.12
2011 1,054 619 435 1,278 4.18 605 13.95
2012 1,124 677 447 1,412 3.84 682 14.04
2013 1,231 742 489 1,570 3.89 559 14.24
2014 1,314 799 514 1,722 3.89 494 14.21
2015 1,347 869 478 1,963 3.95 761 14.19
2016 1,292 876 415 2,053 3.74 946 14.29
Panel B: US IG Universe
1997 816 563 253 2,516 5.61 62 6.72
1998 909 619 290 3,026 5.65 101 6.86
1999 860 609 251 3,004 5.61 139 6.90
2000 738 560 178 2,595 5.29 181 6.91
2001 736 578 158 2,785 5.19 201 7.11
2002 756 620 136 2,939 5.21 208 7.25
2003 752 621 131 3,019 5.38 151 7.38
2004 799 647 152 3,186 5.48 100 7.44
2005 700 548 152 2,399 5.68 91 7.22
2006 717 559 158 2,484 5.70 101 7.21
2007 759 594 165 2,267 6.06 124 7.40
2008 809 649 160 2,751 5.64 332 7.46
2009 789 640 149 3,036 5.41 435 7.58
2010 855 700 155 3,396 5.69 192 7.60
2011 928 765 163 3,894 5.86 194 7.67
2012 994 820 175 4,388 5.94 219 7.78
2013 1,117 922 195 5,021 6.08 174 7.86
2014 1,220 1,000 220 5,545 6.00 142 7.85
2015 1,294 1,055 239 6,055 6.00 172 7.85
2016 1,306 1,057 249 6,501 5.89 186 7.87

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   145

Exhibit 1 (continued)
Summary of Universe Statistics

Avg. No. of Avg. No. of Avg. No. of Avg. No. of Avg. Modified Avg. Option
Year Firms Public Firms Private Firms Bonds Duration Adjusted Spread Avg. Rating
Panel C: European IG Universe
2000 221 142 79 289 4.73 104 5.73
2001 266 173 93 402 4.39 113 6.01
2002 320 225 96 551 4.09 128 6.33
2003 319 230 90 591 3.95 90 6.41
2004 340 245 95 624 4.01 56 6.59
2005 325 233 92 569 4.3 46 6.48
2006 324 237 87 602 4.46 51 6.4
2007 350 263 87 567 4.84 64 6.46
2008 388 293 95 725 4.17 199 6.4
2009 383 292 91 918 3.81 288 6.47
2010 385 293 92 1,034 3.98 154 6.63
2011 386 292 94 1,069 3.96 174 6.82
2012 387 295 92 1,125 3.88 181 7.19
2013 395 300 95 1,187 4.32 121 7.35
2014 436 322 113 1,270 4.72 102 7.41
2015 486 362 124 1,415 5.17 109 7.44
2016 545 411 134 1,595 5.18 121 7.54

Notes: Average monthly number of total firms, public firms, private firms, and bonds and the average duration, spread, and rating for each year. Rating
description: AAA = 1, AA+ = 2, AA = 3, AA- = 4, A+ = 5, A = 6, A- = 7, BBB+ = 8, BBB = 9, BBB- = 10, BB+ = 11, BB = 12, BB- =
13, B+ = 14, B = 15, B- = 16, CCC+ = 17, CCC = 18, CCC- = 19.

the additional risk (see Banz 1981). The behavioral bias
argument for a size risk premium is given by limited
investor attention to smaller companies.
Because empirical evidence for size premiums in
credit markets is relatively sparse and its definition has less
consensus than that for the other three factors analyzed in
this article,12 we follow Banz (1981) and Fama and French
(2015) and define size in the corporate bond space as the
market capitalization of the company’s equity:
Sizet = SOt × PPSt (5)
in which SOt denotes the number of shares outstanding
and PPSt the price per share in month t. The same
arguments for a size premium in the equity space apply to
the realm of credit markets; an investor should demand
12
 Houweling and van Zundert (2017) used the index weight
of each company from the Barclays US Corporate IG index and
the Barclays US Corporate HY index, suggesting that the size of
a company’s public debt in a given index represents the size factor
in credit markets.
higher returns for the additional risks inherent to smaller
issuers. As firm size decreases, equity volatility increases
and so does the probability of default (see Merton 1974).
From a bondholder’s perspective, these firms are likely
to be more risky and should therefore offer higher risk
premiums.
To harvest the size premium in credit markets, we
construct the decile portfolio containing the bonds of
the smallest 10% of the companies as measured by equity
market capitalization and rebalance it on a monthly basis.
Value. Fama and French (1992) used the book-
to-market ratio (BE/ME) as a measure of equity value.
A high BE/ME is indicative of a cheap stock in relative
terms, whereas a low BE/ME suggests the opposite.
According to Zhang (2005), a rationale for the value
premium is based on costly reversibility of invest-
ments. Hence, companies with high sensitivity to eco-
nomic shocks in bad times should offer higher returns.
A behavioral-based explanation suggests that investors
overreact to bad news and extrapolate recent price move-
ments into the future, which results in underpricing.

146   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
 We use the Fama and French (2015) definition
of value and adjust it according to Asness and Frazzini
(2013), using the last available share price:13
BEt −6
Valuet = (6)
MEt
in which BEt−6 and MEt denote book equity and market
equity in month t − 6 and t, respectively. Fama and
French (1995) and Chen and Zhang (1998) showed that
firms with a high book-to-market equity ratio are more
likely to exhibit persistently low earnings, more earn-
ings uncertainty, and high financial leverage and are
more likely to cut their dividend, which in turn leads to
an increased volatility. High equity volatility implies a
higher probability of default in the Merton (1974) model;
thus, value should have the same directional impact on
equity and bond prices.
We construct the decile portfolio containing the
bonds of the most valuable 10% of the issuers as mea-
sured by book-to-market value and rebalance it on a
monthly basis.
Profitability. Fama and French (2015) used the
dividend discount model to show that firms with high
earnings relative to book equity have higher expected
returns. The behavioral explanation suggests that
investors do not differentiate between high and low
profitability in growth firms, and therefore bonds of
companies with high profitability have higher returns.
Chordia et al. (2017) noted that highly profitable compa-
nies generate healthy cash f lows and thus have a smaller
probability of default.
We measure operating profitability, OP, using
accounting data available at end of month t - 6 and
define it as earnings before taxes, EBT (revenues minus
cost of goods sold minus selling, general, and admin-
istrative expenses minus interest expenses), divided by
book equity, BE (see Fama and French 2015).
EBTt −6
OPt = (7)
BEt −6
We construct the decile portfolio containing
the bonds of the most profitable 10% of the firms as
13
 See Correia, Richardson, and Tuna (2012); Houweling and
van Zundert (2017); or Israel, Palhares, and Richardson (2018) for
other definitions of value in credit markets.
Quantitative Special Issue 2019
measured by operating profitability and rebalance it on
a monthly basis.
Investment. Fama and French (2015) suggested
the existence of an investment premium in equity mar-
kets using the dividend discount model and defined the
investment factor as the change in total assets, TA, from
the fiscal year ending in t - 18 to the fiscal year ending
in t - 6, divided by total assets in t - 18. A risk-based
explanation of the investment factor is linked to the fact
that corporations with low levels of investment tend to
have higher costs of capital and therefore only engage
in projects that are most likely to lead to future earnings
and ultimately to higher returns for equity investors.
The behavioral explanation is based on the premise that
investors misprice low-investment companies because of
expectation errors. According to prospect theory, inves-
tors prefer and overprice firms that engage in excessive
investment strategies because of their lottery-like pay-
offs, which could explain the low average returns of such
lotteries. Conversely, they tend to underprice companies
that pick their investments wisely.
Based on the definition of Fama and French (2015),
we define the investment factor, Inv, for credit markets
in month t as
TAt −6 − TAt −18
Invt = (8)
TAt −18
We construct the decile portfolio every month.
The portfolio contains the bonds of the 10% of compa-
nies with the lowest investment and is rebalanced on a
monthly basis.
Comparing Corporate Bond Factor
Portfolio Returns
To compare factor portfolios, we compute risk-
adjusted returns for all portfolios. Furthermore, we
regress factor portfolio returns on corporate bond market
excess returns (DEF) representing the credit default pre-
mium and the equity returns of the Fama–French factors
market (MKT), size (SMB), value (HML), profitability
(RMW), and investment (CMA)14 as well as the bond
factor TERM, representing the term premium that
can be harvested by investing in interest rate securities
14
 The data on MKT, SMB, HML, RMW, and CMA are
obtained from Kenneth French’s website: http://mba.tuck.dart-
mouth.edu/pages/faculty/ken.french/data_library.html.
The Journal of Portfolio Management   147
such as Treasury bonds. For the US HY and IG credit
markets, the TERM factor is constructed as the total
return of the BAML US Treasuries 1-10-year index
minus the one-month T-bill rate, which is also avail-
able on Kenneth French’s website. For the European IG
corporate bond market, this factor is constructed as the
total return of the BAML German Federal Government
1–10-year index minus the one-month Euro Interbank
Offered Rate as the T-bill equivalent for European
money markets. For each currency (euro and US dollar)
and rating segment (HY and IG), we calculate corporate
bond market excess returns by computing the equal-
weighted average of excess returns of each decile to
ensure comparability. Accordingly, we compute three
benchmarks and therefore three DEF factors.
We adjust for risk in three ways to extract the
added value (intercept or alpha) of each factor portfolio
in credit markets, as is commonly done in literature
(see Fama and French 1993; Gebhardt, Hvidkjaer, and
Swaminathan 2005; Bhojraj and Swaminathan 2009;
Bessembinder et al. 2009; or Jostova et al. 2013). First,
we use the SR to measure returns for each factor port-
folio i relative to its total risk:
ri
SRi =   (9)
σi
where r i is the annual average excess return (based on
monthly returns) of factor portfolio i divided by the
annual average standard deviation si of these returns.
Second, we correct for systematic risk of factor portfolio
i by regressing its returns on the default premium:
Rit = α it + βi DEFt + εit (10)
where Rit is the return of factor portfolio i and DEFt is
the default premium in month t. The intercept in this
regression is the equivalent to the capital asset pricing
model (CAPM) alpha for the corporate bond market,
with the default premium being the market factor.
Third, we correct for systematic risk using the
Fama–French five-factor model, the default premium,
and the term premium. We run the following regression:
Rit = α it + βi 1MKTt + βi 2SMBt + βi 3 HML t + βi 4 RMWt
+ βi 5CMAt + βi 6 DEFt + βi 7TERM t + εit maximum drawdowns differ across factors. For instance,
(11)
148   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
where MKT (market), SMB (small minus big), HML
(high minus low), RMW (robust minus weak), and
CMA (conservative minus aggressive) are the equity
market, equity size, equity value, equity profitability,
and equity investment premium, respectively. The idea
behind this approach is to analyze the impact of and
dynamics among traditional equity factors and corre-
sponding factors in corporate bond markets.
EMPIRICAL RESULTS
In Exhibit 2, we report annualized mean returns,
standard deviations (volatilities), SRs, excess returns,
tracking errors, maximum drawdowns, and information
ratios for size (decile 1 represents small and decile 10 large
companies according to equity market capitalization),
value (decile 1 represents low and decile 10 high value
companies according to the BE/ME), profitability (decile
1 represents low and decile 10 high profitable compa-
nies according to operating profitability), and investment
(decile 1 represents low and decile 10 high investment
companies according to the change in total assets) factor
portfolios for all three corporate bond subsamples and the
corresponding market characteristics. Regression inter-
cepts (or alphas) and their t-statistics of the regression anal-
ysis as described earlier for each factor are also provided in
Exhibit 2 for each market segment. It is important to note
that we form our size and investment factor portfolios
with decile 1 (D1) and our value and profitability factor
portfolios with decile 10 (D10), similar to equity markets.
Single-Factor Performance
Panel A of Exhibit 2 reports results for each of the
individual factors across markets and segments. Average
D1 size returns are 6.94% per year in US HY, 1.41% in
US IG, and 1.27% in European IG credit markets. Value
generates average returns of 5.99% (US HY), 1.37%
(US IG), and 1.01% (European IG) compared to 4.28%
(US HY), 0.97% (US IG), and 0.99% (European IG) for
the market. The annualized returns for the profitability
factor are 5.86% (US HY), 0.78% (US IG), and 0.77%
(European IG). Average investment returns are 7.21% (US
HY), 1.19% (US IG), and 1.68% (European IG). Corre-
sponding volatilities (annualized standard deviations) and
size and value exhibit the highest volatilities across mar-
kets and rating segments. Maximum drawdowns are also
Exhibit 2
Performance Summary of Factor Portfolios

6L]H 9DOXH 23 ,QY

0DUNHW 6PDOO' /DUJH' +LJK' /RZ' +LJK' /RZ' /RZ' +LJK'
3DQHO$86+<
$5LVN±5HWXUQ
0HDQ         
9RODWLOLW\         
0D[LPXP'UDZGRZQ ± ± ± ± ± ± ± ± ±
6KDUSH5DWLR         
%([FHVV5HWXUQ
$OSKD      ±  ±
WVWDW        
7UDFNLQJ(UURU        
,QIRUPDWLRQ5DWLR      ±  ±
&&$30$OSKD
$OSKD      ±  ±
WVWDW        
%HWD        
$GMXVWHG5        
')DFWRU$OSKD
$OSKD  ±  
WVWDW    
0.75)    ±
60%    
+0/ ± ± ± ±
50:    
&0$    
7(50   ± 
'()    
3DQHO%86,*
$5LVN±5HWXUQ
0HDQ         
9RODWLOLW\         
0D[LPXP'UDZGRZQ ± ± ± ± ± ± ± ± ±
6KDUSH5DWLR         
%([FHVV5HWXUQ
$OSKD  ±  ± ±   ±
WVWDW        
7UDFNLQJ(UURU        
,QIRUPDWLRQ5DWLR  ±  ± ±   ±
&&$30$OSKD
$OSKD  ±  ± ±   ±
WVWDW        
%HWD        
$GMXVWHG5        

(continued)

Quantitative Special Issue 2019 The Journal of Portfolio Management   149

Exhibit 2 (continued)
Performance Summary of Factor Portfolios

6L]H 9DOXH 23 ,QY

0DUNHW 6PDOO' /DUJH' +LJK' /RZ' +LJK' /RZ' /RZ' +LJK'
')DFWRU$OSKD
$OSKD  ± ± 
WVWDW    
0.75) ±   
60%    
+0/  ±  
50: ±   
&0$ ±  ± 
7(50 ±   ±
'()    
3DQHO&(XURSHDQ,*
$5LVN±5HWXUQ
0HDQ         
9RODWLOLW\         
0D[LPXP'UDZGRZQ ± ± ± ± ± ± ± ± ±
6KDUSH5DWLR         
%([FHVV5HWXUQ
$OSKD  ±  ± ± ±  ±
WVWDW        
7UDFNLQJ(UURU        
,QIRUPDWLRQ5DWLR  ±  ± ± ±  ±
&&$30$OSKD
$OSKD  ± ±   ±  ±
WVWDW        
%HWD        
$GMXVWHG5        
')DFWRU$OSKD
$OSKD ± ± ± 
WVWDW    
0.75)    
60%    ±
+0/    
50:    ±
&0$ ± ±  
7(50 ±   
'()    

Notes: Results for market, size, value, profitability, and investment factor portfolios for the US HY and the US and European IG corporate bond markets.
At the beginning of each calendar month, equal-weighted portfolios are constructed from the 10% issuers with the highest factor exposure to small size, high
value, high profitability, and low investment. Section A summarizes risk and return statistics. Section B shows the excess return statistics. Section C dis-
plays the CAPM alpha and beta. In section D, alphas are estimated from the time-series regression using DEF as the credit default premium, MKT (minus
the risk-free rate, RF) as the equity market premium in addition to the Fama–French factors SMB (size) and HML (value), RMW (profitability), CMA
(investment), and TERM in the seven-factor model. Means, volatilities, SRs, excess returns, tracking errors, maximum drawdowns, information ratios, and
alphas are annualized. We test whether the outperformance of a factor compared to the market is larger than 0 (section B, t-test), and whether the alphas of
a factor portfolio are larger than 0 (t-test in sections C and D).
*, **, and *** denote statistical significance corresponding to the 90%, 95% and 99% confidence levels, respectively.

150   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
largest in size and value for US HY and EUR IG corpo-
rate bonds, whereas size and investment have the largest
drawdowns in US IG markets. SRs are up to 30% higher
compared to market. For instance, in US HY markets,
the SR ranges from 0.46 (value) to 0.62 (profitability)
compared to 0.45 for the market. The SR ranges from
0.24 (profitability) to 0.33 (size) in US IG markets and
from 0.27 (value) to 0.68 (investment) in European IG
markets. For completeness, bottom-decile factor portfolio
returns are also included in Exhibit 2 and are described
in more detail in the robustness section in the context of
long–short portfolios.
Panel B reports the excess returns over benchmark
returns. In the US HY segment, all factors exhibit statis-
tically significant premiums in both US credit segments.
However, for US IG markets, only size is significant at
the 10% level. For European IG credit markets, only
the investment factor generates economically and sta-
tistically significant returns. Interestingly, profitability
is negatively priced in both IG credit markets, and this
is in line with the findings of Chordia et al. (2017) and
Campbell, Chichernea, and Petkevich (2016).15
Panel C reports the excess returns versus DEF
(CAPM alpha), and Panel D reports the excess return
statistics after controlling for Fama–French equity fac-
tors MKT, SMB, HML, RMW, and CMA and the bond
market factors TERM and DEF. Here, for US HY mar-
kets, profitability and investment are still significant fac-
tors, but for US IG markets no factor exhibits statistical
significance. In European IG markets, investment gener-
ates economically and statistically significant returns also
after controlling for DEF and the seven-factor model.
In addition, it is not surprising that DEF carries the
majority of the explanatory power for long-only factor
portfolios considering the common market exposure
incorporated in long-only portfolios.
However, single-factor tracking errors suggest
that investing in factor portfolios can be risky in rela-
tive terms.16 Tracking errors range from 3.45% (high
15
 In contrast, Choi and Kim (2016) reported that profitability
has no explanatory power for the US IG credit market.
16
 Information ratio (IR i) is defined as the active return of
a portfolio i divided by tracking error, where active return is the
difference between the return of the portfolio (Ri) and the return of
a selected benchmark index (Rb), and tracking error is the standard
deviation of the active return (sRi−Rb).
Ri − Rb (12)
IRi =
σ iRi − Rb
Quantitative Special Issue 2019
profitability) to 7% (low size) for US HY, 0.88% (high
profitability) to 2.01% (high value) for US IG, and 0.94%
(high profitability) to 1.87% (high value) for European
IG corporate bond factor portfolios and thus are quite
large compared to the market volatilities of 9.47%, 3.59%,
and 2.36%, respectively. The information ratios range
from 0.28 (high value) to 0.52 (low investment) in the
US HY market, from -0.21 (high profitability) to 0.29
(small size) in the US IG market, and from -0.23 (high
profitability) to 0.58 (low investment) in the European
IG market. The SRs in combination with the relatively
high active risk imply that factor portfolios may under-
perform the benchmark on shorter investment horizons
(see Houweling and van Zundert 2017). Hence, single-
factor portfolios are rather unattractive for portfolio man-
agers and for investors looking for benchmark-oriented
portfolio management. Instead, investors who consider
factor investing with corporate bonds should strategically
allocate to factors to harvest risk premiums on a consistent
basis (see Ang, Goetzmann, and Schaefer 2009).
Multifactor Performance
It is well documented that a portfolio constructed
of different assets (here, factors) will, on average, gen-
erate higher risk-adjusted returns than any individual
security within the investment universe (only true if the
assets or factors in the portfolio are not perfectly corre-
lated). Exhibit 3 shows the outperformance correlations
among our four analyzed factors (small size, high value,
high OP, and low Inv) for US HY and US and European
IG credit markets; the original Fama–French equity fac-
tors of MKT, SMB, HML, RMW, and CMA; and the
bond factors TERM and DEF. For US HY markets, the
corporate bond factor correlations range from -0.15 (OP
and DEF) to 0.56 (size and Inv). For US IG markets, the
corporate bond factor correlations range from -0.44 (OP
and DEF) to 0.58 (size and DEF). Finally, for European
IG markets, the corporate bond factor correlations range
from -0.67 (OP and DEF) to 0.51 (value and DEF).
Low and negative correlations between equity and bond
factors suggest that the single factors capture different
aspects across asset classes.17 However, these low and
17
 Interestingly, CMA and HML exhibit a relatively high
positive correlation of 0.63 for the period from December 1996 to
December 2016, and this is probably why Fama and French argued
that the HML factor of the three-factor model becomes redundant
for describing average returns in the US stock market when incor-
porating CMA and RMW, respectively.
The Journal of Portfolio Management   151
Exhibit 3
Correlation Summary of Factor Portfolios

6L]H 9DOXH 23 ,QY 0.75) 60% +0/ 50: &0$ 7(50 '()
3DQHO$86+<
6L]H 
9DOXH  
23  ± 
,QY  ±  
0.75)   ± ± 
60%      
+0/  ±   ± ± 
50:  ±   ± ±  
&0$  ±   ±    
7(50 ±  ± ± ± ± ±   
'()   ±     ± ± ± 
3DQHO%86,*
6L]H 
9DOXH ± 
23 ± ± 
,QY  ± ± 
0.75)  ± ±  
60%  ± ±   
+0/  ± ±  ± ± 
50: ± ±  ± ± ±  
&0$ ± ± ± ± ±    
7(50 ±   ± ± ± ±   
'()  ± ±     ± ± ± 
3DQHO&(85,*
6L]H 
9DOXH  
23 ± ± 
,QY ± ± ± 
0.75)   ± ± 
60%   ± ±  
+0/       
50: ± ±   ± ±  
&0$  ±   ±    
7(50 ± ±   ± ± ±   
'()   ± ±    ± ± ± 

Notes: Return correlations between the Fama–French factors MKT, SMB, HML, RMW, and CMA and the bond market factors TERM and DEF as
well as the outperformance of the analyzed bond factors size, value, profitability, and investment for the US HY and IG universe over the period December
1996 to December 2016 (December 2000 to December 2016 for European IG bonds).

negative correlations suggest that a combination of two
or more factors offers significant diversification benefits.
Therefore, we also construct equal-weighted long-
only multifactor portfolios by combining small size,
high value, high profitability, and low investment decile
factor portfolios, as described by
rtMultiFactor = 0.25rtSize + 0.25rtValue + 0.25rtOP + 0.25rtInv (13)
where r t denotes the return of each corresponding
single-factor portfolio and the multifactor portfolio in
month t. These equal-weighted portfolios are another
way to examine the efficacy of the four analyzed fac-
tors across markets and segments. Exhibit 4 reports the

152   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
Exhibit 4
Performance Summary of Long-Only Multifactor Portfolios

Long Only Long Only Long Only

US HY BM US HY MF US IG BM US IG MF Eur. IG BM Eur. IG MF
Panel A: Risk–Return
Mean 4.28% 6.61% 0.97% 1.20% 0.99% 1.19%
Volatility 9.47% 11.27% 4.22% 3.88% 2.36% 2.64%
Maximum Drawdown –38.21% –41.76% –21.47% –23.18% –11.91% –13.22%
Sharpe Ratio 0.45 0.59 0.23 0.31 0.42 0.45

Panel B: Excess Return

Alpha 2.33%*** 0.23%* 0.21%*
t-stat 3.12 1.63 1.39
Tracking Error 3.46% 0.66% 0.61%
Information Ratio 0.67 0.35 0.34

Panel C: CAPM Alpha

Alpha 1.74%** 0.17% 0.12%
t-stat 2.43 0.14 0.85
Beta 1.14%*** 1.07%*** 1.09%***
t-stat 52.4 96.77 62.51
Adjusted R2 0.92 0.98 0.95

Panel D: Seven-Factor Alpha

Alpha 1.65%** 0.08% –0.01%
t-stat 1.96 0.52 0.09
MKT-RF 0.01 0.00 0.01
SMB 0.04 0.00 0.00
HML –0.04 0.00 0.01
RMW 0.09 0.00 0.01
CMA 0.07 0.00 0.00
TERM 0.18 0.03 0.01
DEF 1.09 1.07 1.08

Notes: Results for the US HY and the US and European IG benchmarks (BM) as well as long-only multifactor corporate bond portfolios (MF).
At the beginning of each calendar month equal-weighted portfolios are constructed from the 10% issuers with the highest factor exposure to small size, high
value, high profitability, and low investment. Panel A summarizes risk and return statistics. Panel B shows the excess return statistics. Panel C displays
the CAPM alpha and beta. In Panel D, alphas are estimated from the time-series regression using DEF as the credit default premium; MKT (minus the
risk-free rate, RF) as the equity market premium; and the Fama–French factors SMB (size), HML (value), RMW (profitability), CMA (investment),
and TERM in the seven-factor model. Means, volatilities, SRs, excess returns, tracking errors, maximum drawdowns, information ratios, and alphas are
annualized. We test whether the outperformance of a factor compared to the market is larger than 0 (Panel B, t-test) and whether the alphas of a factor
portfolio are larger than 0 (t-test in Panels C and D).
*, **, and *** denote statistical significance corresponding to the 90%, 95%, and 99% confidence levels, respectively.

multifactor portfolio statistics. The multifactor portfolios
deliver annual average excess returns of 2.33% in the US
HY market (t-stat of 3.12), 0.23% in the US IG market
(t-stat of 1.63), and 0.21% in the European IG market (t-stat
of 1.39). Moreover, the equal-weighted combination of
size, value, profitability, and investment within the dif-
ferent markets and segments generates higher SRs than
the equal-weighted market index. SRs are still up to 30%
higher compared to the market. One possible explanation
is that multifactor portfolios are more diversified because
more securities in the cross section have nonzero weights
and the weights are less extreme than in single-factor port-
folios. However, when controlling for the Fama–French
equity factors MKT, SMB, HML, RMW, and CMA as
well as the bond market factors TERM and DEF, the
alphas of multifactor portfolios are still economically and
statistically significant for US HY markets; however, US
and European IG markets cannot withstand the test.

Quantitative Special Issue 2019 The Journal of Portfolio Management   153

Exhibit 5
Cumulative US HY Multifactor Long-Only Portfolio Returns (December 1996–December 2016)

2XWSHUIRUPDQFH
 0XOWL)DFWRU86+<
86+<%HQFKPDUN

3HUIRUPDQFH





'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF
These findings suggest that the combination of
all four factors leads to diversification benefits. The
equal-weighted multifactor portfolios demonstrate an
annualized SR of 0.59% for US HY, 0.31% for US IG,
and 0.45% for European IG corporate bonds; the SRs
of their corresponding markets are 0.45%, 0.23%, and
0.42%, respectively. Maximum drawdowns are compa-
rable to those of the market; thus, similar to the results
of Lettau, Maggiori, and Weber (2014), we find that
downside risk is not able to explain corporate bond
returns. Exhibits 5, 6, and 7 show the multifactor port-
folio performance versus the benchmark and the cumu-
lative outperformance.
Robustness Checks
Corporate bonds are typically traded less frequently
than stocks. Therefore, most academic research focuses
on low turnover strategies to avoid high transaction costs.
Edwards, Harris, and Piwowar (2007) found that bid–
ask spreads for US corporates vary significantly through
time and depend on the bond’s quality and transaction
size. However, most of the literature either ignores costs
completely or assumes transaction costs to be a fixed
amount of 13 bps (Gebhardt, Hvidkjaer, and Swamina-
than 2005) or 12 bps ( Jostova et al. 2013). According
to Amenc et al. (2012), it is possible to achieve lower
154   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019



&XPXODWLYH2XWSHUIRUPDQFH





±
turnover by setting a rebalancing threshold at which
a portfolio is not rebalanced until two-way turnover
reaches 70% in a quarter. The main idea is to avoid
rebalancing when new weights deviate from the cur-
rent weights by a relatively small amount. By applying
this rule, portfolio turnover and corresponding trading
costs can be reduced. Novy-Marx and Velikov (2016)
confirmed that most factor-based strategies continue
to generate statistically significant alphas despite low
turnover. Nevertheless, we estimate transaction costs as
a function of issue rating, maturity and total turnover
associated with each factor portfolio, similar to Chen,
Lesmond, and Wei (2007).
Our results remain unchanged after accounting
for transaction costs (the results are available from the
authors). Thus, the factors studied here are not only
properly motivated and theoretically sound but can also
be implemented. This implies that factor-based investing
in credit markets does indeed offer additional benefit to
corporate bond investors. However, the specific imple-
mentation design18 has an impact on performance and
explains why two portfolios based on the same factor
may perform differently.
18
 Investment universe, rebalancing frequency, weighting
scheme, and definition of portfolio configuration (e.g., decile,
quintile).
Exhibit 6
Cumulative US IG Multifactor Long-Only Portfolio Returns (December 1996–December 2016)
 
2XWSHUIRUPDQFH
 
0XOWL)DFWRU86,*
 86,*%HQFKPDUN


&XPXODWLYH2XWSHUIRUPDQFH



3HUIRUPDQFH








 

 ±

 ±
'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF

Exhibit 7
Cumulative European IG Multifactor Long-Only Portfolio Returns (December 2000–December 2016)
 
2XWSHUIRUPDQFH
 0XOWL)DFWRU(XURSHDQ,* 
(XURSHDQ,*%HQFKPDUN


&XPXODWLYH2XWSHUIRUPDQFH



3HUIRUPDQFH







±

 ±

 ±
'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF 'HF

DISCUSSION AND INTERPRETATION
OF THE RESULTS
Frazier and Liu (2016) showed that risk aversion
is still a fundamental concept in theory and in practice
and stated that, all else equal, investors require additional
reward for bearing additional risk. The relationship
between equities and corporate bonds in general is
explained by Merton (1974), who claimed that a cor-
porate bond, in essence, consists of a default-free bond
and a short put on the issuer’s equity. As such, each
corporate bond is described by a risk-free component

Quantitative Special Issue 2019 The Journal of Portfolio Management   155

(e.g., government bond) and an equity component
(equity put). For higher-rated IG bonds, the probability
of default (PD) is small; hence, the value of the put is
small compared to the risk-free component of the credit
security, leading to a more risk free bond–like behavior
of IG bonds. In contrast, the elevated PD of issuers of
HY debt increases the value of the put and thus leads
to a more equity-like behavior of HY debt. In other
words, just like equity investors, HY bond investors bet
on the future prosperity of the issuer by writing at-the-
money puts. If the issuer’s stock price increases, so does
the value of the put and, hence, the value of the bond
and vice versa.
Our results agree with those of Choi and Kim
(2016), who showed that cross-sectional return pre-
miums in US corporate bonds are not equal to those
in equity markets, implying market segmentation. To
summarize, our results are stronger for HY markets.
We conjecture that this observation is due to the more
equity-like features of HY bond markets compared to
IG bond markets. Because the Fama–French factors
have been shown to hold significant explanatory power
for equity market returns, it is not surprising that these
equity factors perform better in more equity-like bond
markets.
CONCLUSION
In this article, we investigate whether the Fama–
French equity factors can be extended to corporate bond
markets. Although bonds and stocks are driven by the
same firm fundamentals and therefore should react to the
same factors, our empirical results show that these factors
do not fully translate to fixed-income markets, as sug-
gested by structural equity-bond relations (e.g., Merton
1974). On one hand, a possible explanation could be
market segmentation: Bonds occupy a different position
in a firm’s capital structure (see Choi and Kim 2016).
On the other hand, market segmentation could also be
caused by institutional investors who in fact dominate
the corporate bond market and perceive risk differently
compared to individual investors (see Chordia et al.
2017). Therefore, corporate bond markets seem to have
their own unique features.
Nevertheless, after controlling for corresponding
equity factor exposures, some factors do add value to
corporate bond investors. In particular, our results sug-
gest that the examined factors exhibit stronger returns
156   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019
in HY corporate bond markets (especially profitability
and investment), probably because of the equity-like
features of HY bonds, which is in line with theory.
In contrast, no factor is significant for US IG securities,
whereas the investment factor generates economically
and statistically significant returns in the European IG
market. Interestingly, profitability is negatively priced
in both IG markets. Although the original Fama and
French (1993) factors, size and value, do not seem to
add significant value to corporate bond investors, the
two newly proposed Fama and French (2015) factors,
profitability and investment, do offer return potential
to investor portfolios.
Finally, regarding practical considerations, our
results show that an investable (long-only and after
transaction costs) equal-weighted multifactor portfolio
reduces tracking error and drawdown, and higher risk-
adjusted returns are preserved compared to single-factor
portfolios. These advantages are mainly because of the
diversification benefits of multifactor portfolios instead
of single-factor portfolios because factor correlations are
low and, in some cases, even negative. However, tradi-
tional corporate bond investment managers still avoid
certain factors because of the eventually large tracking
errors or low information ratios, whereas those who
incorporate factors in their investment approach usually
employ a multifactor-based approach. When considering
the SR instead of the information ratio, factor investing
in corporate bond markets offers clear benefits, espe-
cially in the long run, and thus should be a strategic
choice for investors and investment managers.
ACKNOWLEDGMENTS
The authors are very grateful for comments from an
anonymous referee, Hsiu-Lang Chen, Victor DeMiguel,
Daniel Giamouridis, Amit Goyal, Patrick Jahnke, Andrew
Karolyi, Harald Lohre, Felix Miebs, Paolo Porchia (dis-
cussant), Tobias Regele, Olivier Scaillet (discussant), Björn
Strauß, Michael Weber, and Andrew Zhang. They also
thanks seminar participants at the 6th International Con-
ference of the Financial Engineering and Banking Society,
13th Citi Global Quant Research Conference, 24th Spanish
Finance Forum, 2016 FMA Annual Meeting, Northfield’s
29th Annual Research Conference, and Frontiers of Factor
Investing Conference. Parts of this research project have
been conducted while the corresponding author was at the
University of Chicago Booth School of Business.
REFERENCES
Amenc, N., F. Goltz, A. Lodh, and L. Martellini. 2012.
“Diversifying the Diversifiers and Tracking the Tracking
Error: Outperforming Cap-Weighted Indices with Limited
Risk of Underperformance.” The Journal of Portfolio Manage-
ment 38: 72–89.
Ang, A., W. Goetzmann, and S. Schaefer. “Evaluation of
Active Management of the Norwegian Government Pension
Fund—Global.” Working paper, Columbia Business School,
Yale School of Management, and London Business School,
2009.
Asness, C., and A. Frazzini. 2013. “The Devil in HML’s
Details.” The Journal of Portfolio Management 114: 49–69.
Baker, M., and J. Wurgler. 2012. “Comovement and Predict-
ability Relationships between Bonds and the Cross-Section
of Stocks.” Review of Asset Pricing Studies 2: 57–87.
Banz, R. 1981. “The Relationship between Return and
Market Value of Common Stocks.” Journal of Financial
Economics 9: 3–18.
Bessembinder, H., K. Kahle, W. Maxwell, and D. Xu. 2009.
“Measuring Abnormal Bond Performance.” The Review of
Financial Studies 22: 4220–4258.
Bhojraj, S., and B. Swaminathan. 2009. “How Does the
Corporate Bond Market Value Capital Investments and
Accruals?” Review of Accounting Studies 14: 31–62.
Black, F., and M. Scholes. 1973. “The Pricing of Options and
Corporate Liabilities.” Journal of Political Economy 81: 637–654.
Campbell, T., D. Chichernea, and A. Petkevich. 2016. “Dis-
secting the Bond Profitability Premium.” Journal of Financial
Markets 27: 102–131.
Chen, L., D. Lesmond, and J. Wei. 2007. “Corporate Yield
Spread and Bond Liquidity.” The Journal of Finance 62: 119–149.
Chen, N. F., and F. Zhang. 1998. “Risk and Return of Value
Stocks.” The Journal of Business 71: 501–535.
Choi, J., and Y. Kim. “Anomalies in the Joint Cross-Section
of Equity and Corporate Bond Returns.” Working paper,
University of Illinois at Urbana-Champaign, 2016.
Quantitative Special Issue 2019
Chordia, T., A. Goyal, Y. Nozawa, A. Subrahmanyam, and
Q. Tong. 2017. “Are Capital Market Anomalies Common to
Equity and Corporate Bond Markets? An Empirical Inves-
tigation.” Journal of Financial and Quantitative Analysis 52:
1301–1342.
Cochrane, J. H. 2011. “Presidential Address: Discount Rates.”
The Journal of Finance 66: 1047–1108.
Correia, M., S. Richardson, and I. Tuna. 2012. “Value
Investing in Credit Markets.” Review of Accounting Studies
17 (3): 572–609.
Crawford, S., P. Perotti, R. Price, and C. J. Skousen.
“Accounting-Based Anomalies in the Bond Market.”
Working paper, University of Houston, University of Bath,
and Utah State University, 2015.
Edwards, A., L. Harris, and M. S. Piwowar. 2007. “Cor-
porate Bond Market Transaction Costs and Transparency.”
The Journal of Finance 62: 1–21.
Elton, E., M. Gruber, D. Agrawal, and C. Mann. 2001.
“Explaining the Rate Spread on Corporate Bonds.”
The Journal of Finance 56: 247–277.
Fama, E., and K. French. 1992. “The Cross-Section of
Expected Stock Returns.” The Journal of Finance 47: 427–465.
——. 1993. “Common Risk Factors in the Returns on Stocks
and Bonds.” Journal of Financial Economics 33: 3–56.
——. 1995. “Size and Book-to-Market Factors in Earnings
and Returns.” The Journal of Finance 50: 131–155.
——. 2015. “A Five-Factor Asset Pricing Model.” Journal of
Financial Economics 116: 1–22.
Frazier, D., and X. Liu. 2016. “A New Approach to Risk–
Return Trade-Off Dynamics via Decomposition.” Journal of
Economic Dynamics & Control 62: 43–55.
Friewald, N., C. Wagner, and J. Zechner. 2014. “The
Cross-Section of Credit Risk Premia and Equity Returns.”
The Journal of Finance 69: 2419–2469.
Gebhardt, W., S. Hvidkjaer, and B. Swaminathan. 2005.
“Stock and Bond Market Interaction: Does Momentum Spill
Over?” Journal of Financial Economics 75: 651–690.
The Journal of Portfolio Management   157
Goldstein, I., H. Jiang, and D. T. Ng. 2018. “Investor Flows Lettau, M., M. Maggiori, and M. Weber. 2014. “Conditional
and Fragility in Corporate Bond Funds.” Journal of Financial Risk Premia in Currency Markets and Other Asset Classes.”
Economics 126: 592–613. Journal of Financial Economics 114: 197–225.

Harvey, C., Y. Liu, and H. Zhu. 2016. “... and the Cross- Merton, R. 1974. “On the Pricing of Corporate Debt: The
Section of Expected Returns.” The Review of Financial Studies Risk Structure of Interest Rates.” The Journal of Finance 29:
29: 5–68. 449–470.

Houweling, P., and J. van Zundert. 2017. “Factor Investing Novy-Marx, R., and M. Velikov. 2016. “A Taxonomy of
in the Corporate Bond Market.” Financial Analysts Journal Anomalies and Their Trading Costs.” The Review of Financial
73: 100–115. Studies 29: 104–147.

Ilmanen, A., R. Byrne, H. Gunasekera, and R. Minikin. Vassalou, M., and X. Yuhang. 2004. “Default Risk in Equity
2004. “Which Risks Have Been Best Rewarded?” The Journal Returns.” The Journal of Finance 59: 831–868.
of Portfolio Management 30: 53–57.
Zhang, L. 2005. “The Value Premium.” The Journal of Finance
Israel, R., D. Palhares, and S. Richardson. 2018. “Common 60: 67–103.
Factors in Corporate Bond Returns.” Journal of Investment
Management 16: 17–46.
Disclaimer
Jostova, G., S. Nikolova, A. Philipov, and C. Stahel. 2013. Views expressed in this article are those of the authors and do not necessarily
ref lect those of Deka Investment or its employees.
“Momentum in Corporate Bond Returns.” The Review of
Financial Studies 26: 1649–1693.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.
Kapadija, N., and X. Pu. 2012. “Limited Arbitrage between
Equity and Credit Markets.” Journal of Financial Economics 105:
542–564.

158   Extending Fama–French Factors to Corporate Bond Markets Quantitative Special Issue 2019

T
Joseph Simonian
is director of quantitative
research at Natixis
Investment Managers in
Boston, MA.
joseph.simonian@natixis.com
Chenwei Wu
is a quantitative analyst
at Natixis Investment
Managers in Boston, MA.
chenwei.wu@natixis.com
Quantitative Special Issue 2019
Factors in Time: Fine-Tuning
Hedge Fund Replication
Joseph Simonian and Chenwei Wu
he primary utility of factor models
is that they can help simplify the
analysis of portfolios consisting of
many assets by explaining their
behavior using a parsimonious set of drivers.
This is true of linear regression–based factor
models and those constructed using prin-
cipal component analysis, as well as non-
linear factor models, including those based on
machine learning algorithms.1 By providing
a more transparent view of the systemic risks
to which a portfolio is exposed, factor models
can be employed in both risk management
and alpha generation. Asset allocators, for
example, use factor models both to track the
amount and type of systemic risk carried by
different asset classes and to build factor-
tracking portfolios in which specific factor
exposures (loadings) are targeted during
portfolio construction to more precisely
express their investment views or to replicate
strategies with specific factor profiles.
One application of factor-tracking port-
folio construction is hedge fund replication,
where liquid investment vehicles are used to
build portfolios that are expected to mimic
the investment behavior of different hedge
fund styles. In regression-based frameworks,
the construction of factor-tracking portfolios
entails a number of risks. Perhaps the major
1
 See Simonian et al. (2019) for an example of
the latter approach.
risk is overfitting, in which the calculation of
beta coefficients may depend so precisely on
the training data that a given model fails to
generalize on test data, producing poor pre-
dictive results. This issue has historically been
somewhat of a secondary concern for those
engaged in building hedge fund replication
models because they have been primarily
interested in building frameworks that maxi-
mize in-sample explanatory power. Never-
theless, predictive power, our focus in this
article, is necessarily the central concern of
practitioners who want to apply hedge fund
replication ex ante in portfolio management.
Another potential risk with replication strate-
gies is multicollinearity, where predictors are
linearly correlated to a degree that renders
them problematic for use in an ordinary least
squares (OLS) regression. The latter risk is
ever-present in hedge fund replication, in
which practitioners generally use publicly
available vehicles, such as exchange-traded
funds (ETFs), as factor proxies. A final risk is
that, because of the dynamic nature of hedge
funds, models built using a single sample may
fail to capture the evolution of hedge fund
exposures over time.
As a remedy to the foregoing shortcom-
ings of the standard approach to factor-based
hedge fund replication, in this article the
authors demonstrate how to use the regu-
larization technique known as ridge regression
(RR) to both mitigate the impact of factor
The Journal of Portfolio Management   159
interdependence and enhance the predictive power
of OLS-type models through time. Accordingly, the
present article can be considered a contribution to the
regression-based hedge fund replication approach previ-
ously explored by Agarwal and Naik (2000), Fung and
Hsieh (1997, 2004), Ennis and Sebastian (2003), Markov
et al. (2006), Hasanhodzic and Lo (2007), Jaeger (2008),
and Amenc et al. (2010), among others.
BUILDING FACTOR-MATCHING PORTFOLIOS
In practice, the standard approach to factor port-
folio construction using a linear factor model is generally
based on a straightforward application of relatively basic
optimization methods such as linear programming. As an
illustration of the standard approach to factor matching,
consider the following simple example, in which we have
a portfolio of three assets—1, 2, and 3—whose behavior
is explained by the following two-factor model:
Ri = β1r1 + β 2r2 (1)
where Ri represents the return on asset i, β1 represents
the sensitivity of asset i to the rate of change in the price
level of the broad equity market, r1 represents the rate of
change in the price level of the broad equity market, β2
represents the sensitivity of asset i to the rate of change
in 10-year Treasury yields, and r 2 represents the rate of
change in 10-year Treasury yields. We list hypothetical
individual betas for each asset in Equation 2:
Asset β1 β2
1 0.75 1.00
2 0.50 0.25
3 1.50 0.75 (2)
Now, suppose we wish to allocate our assets so that
our portfolio has a beta of 1.0 to β1, and a beta of 0.50
to β2. As previously mentioned, a standard approach to
solving the factor-matching problem is to use a linear
program; for example:
Min|1.00 − βT1 xi | + |0.50 − βT2 xi| (3)
subject to
Σ ni =1xi = 1 (No leverage constraint)
x l, x2, x3 ≥ 0 (No shorting constraint)
160   Factors in Time: Fine-Tuning Hedge Fund R eplication Quantitative Special Issue 2019
The solution to Equation 3 is given by the following
portfolio weights: x l = 0%, x2 = 50%, x3 = 50%.
One popular factor-matching approach among
practitioners is returns-based style analysis (RBSA), a
method introduced by Sharpe (1988, 1992). As a species
of factor portfolio construction, it can be considered a
framework tailored to long-only investors who seek to
replicate the performance of particular strategies using
publicly available investment vehicles. RBSA is regres-
sion based and is expressed formally as
Rtm = α + ∑ i =1 βi Rti + εt (4)
I
where Rtm is the return stream for the investment strategy
to be replicated, Rti represents the return streams of a set
of investable factor proxies, I is the number of investable
proxies, and εt is the error term. Formally, RBSA differs
from the factor-matching optimization described above
in that it seeks to match the return of a target portfolio
rather than its specific profile of systemic exposures. In
the standard case, two important constraints are applied
to Equation 4, constraints that are put in place to pro-
duce a combination of investable proxies suitable for a
long-only implementation. First, each beta coefficient
is constrained to be greater than zero; that is, βi > 0, ∀i.
Second, the sum of the betas is constrained to sum to
unity; that is, Σ iI=1βi = 1. As such, each beta is interpreted
as a weight assigned to a particular investable proxy in a
replication portfolio.
In hedge fund replication research (such as the
articles cited earlier), it has been customary to relax
the no-shorting and sum-to-unity constraints, which
transform their frameworks to standard OLS regres-
sions that are differentiated, more or less, by their use
of investable vehicles as factors. We follow that con-
vention here as well. However, it is important to note
that, whereas in the past the relaxation of these con-
straints may have posed implementation challenges to
some investors, given the breadth and depth of modern
derivatives markets, including the advent of options
on ETFs, it is relatively easy for long-only investors
to create net exposures deviating from unity. Thus,
the difference between factor portfolio construction
using standard RBSA and constraint-relaxed RBSA is
minimal today, given the set of investment tools avail-
able to portfolio managers.
APPLYING RIDGE REGRESSION
TO HEDGE FUND REPLICATION
Because hedge fund replication is generally exe-
cuted using publicly available investment vehicles, the
factors in hedge fund replication models are generally
correlated to a sufficient degree to disqualify them as
true factor models, which by definition need to consist
of a (parsimonious) set of uncorrelated predictor vari-
ables. From a statistical standpoint, multicollinearity
negatively affects a regression’s output by inducing
inaccurate estimates of regression coefficients, inf lating
standard errors, generating false p-values, and signifi-
cantly reducing the predictive utility of a given model.
Accordingly, to counter the risk of heightened variance
between training and test samples, it may be desirable
to inject some bias into an OLS regression to generalize
the model to a greater degree.
RR 2 is a regression framework that provides a way
of formally addressing multicollinearity in a standard
OLS regression by adjusting the OLS regression coef-
ficient estimates using a penalty parameter (so called
because it penalizes less inf luential predictors), that
injects varying degrees of bias into the regression. When
the penalty parameter is applied to the diagonal of the
regression matrix, it forces the columns of the matrix
to be linearly independent. The expectation is that by
applying the penalty parameter, the regression standard
errors will be reduced and lead to the generation of
more predictively useful models. Formally, an RR is
described as follows:
βˆ = argmin β∈R p ∑ i =1( yi − xTi β)2
n
+ λ ∑ j =1 β 2j = argmin β∈R p || y − Xβ ||22 + λ||β ||22 (5)
p
  zons. The first is a “tactical” horizon, where we derive
Loss Penalty
where y ∈ Rn is a vector of target observations, X ∈ Rn×p
is a predictor matrix, and l ≥ 0 is the penalty parameter.
HEDGE FUND REPLICATION
To test the efficacy of RR in hedge fund rep-
lication, we set up an investable 3 four-factor model
2
 The ideas underlying ridge regression were independently
discovered by a number of individuals. See Larson (1931), Tikhonov
(1943, 1963), Foster (1961), and Hoerl and Kennard (1970).
3
 ETFs tracking the performance of each of these indexes
are available.
Quantitative Special Issue 2019
and test it on a set of seven hedge fund strategies. Our
factors are the MSCI USA Value and Momentum
indices, the Bloomberg High Yield Total Return
Index, and the ICE Core Treasury Bond Index. The
hedge fund strategies we test are merger arbitrage,
convertible arbitrage, long/short equity, global macro,
equity market neutral, S&P 500 PutWrite, and S&P
500 BuyWrite.4 We show the full-sample correlations
of our factors in Exhibit 1.
Although the matrix in Exhibit 1 is full rank,
and thus valid for a regression, three of the factors
show fairly high correlations to each other (value,
momentum, and high yield). Given this, RR can still
play a useful role in mitigating some of the potentially
deleterious modeling impact of the positive co-move-
ment of these factors.
The central consideration in any application of RR
is what value the penalty parameter l should take. Here
we calibrate (“fine-tune”) the value of l by testing a
range of values through time and choosing the l value
that exhibits the best predictive power over a set of rolling
windows. By evaluating different values for the penalty
parameters across evolving sample sets, we ultimately
arrive at the penalty value that minimizes prediction
error on test data. Given that our primary goal is to
correct for overfitting, it is more prudent to calibrate l
over a multitude of samples to avoid tying the value of
l to a single set of data; doing the latter would risk the
same type of overfitting we are trying to avoid with the
regression coefficients.5 This approach is also adaptive to
new information because the optimal l may change over
time as additional rolling samples are evaluated.
In Exhibit 2, we show the values of  l that provide
the best predictive results over two different test hori-
4
 For merger arbitrage, convertible arbitrage, long/short
equity, global macro, and equity market neutral, we use the corre-
sponding Barclay Hedge index for each. For S&P 500 PutWrite and
S&P 500 BuyWrite, we use the indexes published by the Chicago
Board Options Exchange (CBOE).
5 
Our evaluation methodology can be considered a species
of cross-validation. However our approach is arguably more suited
to time series analysis as we exclude the possibility of using data in
a testing (validation) set that chronologically precedes the data in
a training set, as is the case with many standard cross-validation
methodologies (for a discussion of cross-validation see Hastie
et al. 2009).
The Journal of Portfolio Management   161
Exhibit 1
Full-Sample Correlations of Factors (monthly data, 1/2006–12/2018)

%ORRPEHUJ%DUFOD\V86
06&,86$ 06&,86$ ,&(86 &RUSRUDWH+LJK<LHOG
9DOXH1HW7RWDO 0RPHQWXP86'1HW 7UHDVXU\&RUH 7RWDO5HWXUQ,QGH[
5HWXUQ86',QGH[ 7RWDO5HWXUQ,QGH[ %RQG75,QGH[ 9DOXH8QKHGJHG86'
06&,86$9DOXH1HW   ± 
7RWDO5HWXUQ86',QGH[
06&,86$0RPHQWXP   ± 
86'1HW7RWDO5HWXUQ
,QGH[
,&(867UHDVXU\&RUH ± ±  ±
%RQG75,QGH[
%ORRPEHUJ%DUFOD\V86   ± 
&RUSRUDWH+LJK<LHOG
7RWDO5HWXUQ,QGH[
9DOXH8QKHGJHG86'

Source: Natixis Investment Managers.

Exhibit 2
Hedge Fund Replication—OLS versus Ridge Regression (monthly data, 1/2006–12/2018)

Strategy Optimal λ Average SSE Batting Average In-Sample R2 OLS Average SSE OLS In-Sample R2
Rolling 12-Month Windows; 1-Month-Ahead Replication Horizon
Merger Arbitrage 1.4 6.52E-05 77% 0.35 1.01E-04 0.51
Convertible Arbitrage 0.1 1.95E-04 78% 0.70 2.22E-04 0.73
Long/Short Equity 0.5 7.88E-05 73% 0.70 1.03E-04 0.77
Global Macro 1.4 1.47E-04 90% 0.40 2.51E-04 0.58
Equity Market Neutral 0.2 3.89E-05 85% 0.54 5.21E-05 0.59
S&P 500 PutWrite 0.4 4.26E-04 84% 0.71 6.32E-04 0.76
S&P 500 BuyWrite 0.3 2.65E-04 75% 0.86 4.22E-04 0.89
Rolling 36-Month Windows; 12-Month-Ahead Replication Horizon
Merger Arbitrage 2.4 5.08E-04 71% 0.27 6.03E-04 0.41
Convertible Arbitrage 0.1 7.84E-04 52% 0.68 8.11E-04 0.70
Long/Short Equity 0.2 6.34E-04 63% 0.75 6.81E-04 0.77
Global Macro 0.3 1.21E-03 73% 0.42 1.31E-03 0.45
Equity Market Neutral 1.5 3.62E-04 67% 0.25 4.40E-04 0.46
S&P 500 PutWrite 0.3 2.48E-03 65% 0.73 2.70E-03 0.74
S&P 500 BuyWrite 0.1 1.70E-03 50% 0.87 1.73E-03 0.88

Source: Barclay Hedge, CBOE, and Natixis Investment Managers.

beta coefficients6 over a 12-month window and attempt

to match the performance of a strategy over the following derive re-scaled beta coefficients from the original ridge regression
betas as follows:
k k
xi − xi k
βi xi
y = β 0 + ∑βi xi ,normalized = β 0 + ∑βi = β0 − ∑
6
 We note that it is important to re-scale the raw ridge regres-
T
sion betas so that they are comparable to the original OLS betas. i =1 i =1 x x
i i i =1 xiT xi
x −x k
βi k
Given a training set {x1, … xk} and normalized returns i T i we +∑ xi   = β 0,re − scaled + ∑βi ,re − scaled xi .
xi xi i =1 xiT xi i =1

162   Factors in Time: Fine-Tuning Hedge Fund R eplication Quantitative Special Issue 2019
month. The second test horizon is a “strategic” horizon,
where we derive beta coefficients over a 36-month
window and attempt to match the performance of a
strategy over the following 12 months. Our measure
of predictive error is the sum of squared errors (SSE)
over 1-month and 12-month horizons, respectively. For
example, in the tactical study, given the generation of
beta coefficients at time t (e.g., end-of-month 12/2005),
we multiply the latter coefficients by the factor returns
at the end of time t + 1 (e.g., end-of-month 1/2006)
and compare the weighted sum of the latter values to
the actual hedge fund strategy returns for the month.
The first statistic we show in Exhibit 2 is the
average SSE over the entire set of rolling windows.
Next, we look at what we call the batting average of
the RR replication framework versus a standard OLS
approach. The batting average is simply the percentage
of rolling windows in which the RR SSE is lower than
the OLS SSE. This is an important statistic because it
indicates the generalizability of the ridge framework
over evolving sets of data.
As Exhibit 2 shows, in both the tactical and stra-
tegic studies, the RR-based replication models improve
significantly upon the standard OLS models in terms of
their average SSEs over the set of rolling windows. Per-
haps the most important result, however, is that the bat-
ting average of the ridge approach is generally very high
(the results for the convertible arbitrage and BuyWrite
replication models in the strategic study being the sole
exceptions), indicating that the predictive superiority
of the RR models is consistent through time. Finally,
it is notable that the deterioration in the in-sample R 2
when using RR is generally minimal, indicating that
the significant improvement in predictive power gained
through the use of RR-based models is not made at the
expense of a commensurate reduction in ex post explana-
tory power.
CONCLUSION
Regression-based factor models are often
employed in hedge fund replication because of their
simplicity and familiarity. In practice, however, these
models are often misapplied because the investment
vehicles used to replicate a given strategy are typically
correlated to a sufficient degree to disqualify them
as true factor models. More importantly, OLS-based
models are often overfitted to test data, resulting in
Quantitative Special Issue 2019
poor predictive performance when tested on new
observations. As a remedy to the latter maladies, the
authors employ ridge regression, a technique that is
designed to alleviate multicollinearity and produce
more reliable estimates of predictor variables. Ridge
regression uses a penalty parameter that penalizes less
inf luential predictors, resulting in an increase in the
bias and a decrease in the variance of the regression.
The decrease in variance typically results in enhanced
generalizability of a given model and, accordingly, an
increase in its predictive power. An outstanding ques-
tion in any application of ridge regression is what the
value of the penalty parameter should be. The authors
show how to calibrate the value of the penalty param-
eter dynamically by choosing the value that exhibits
the strongest performance across a set of rolling sam-
ples. By testing different values for the penalty param-
eter across the training data, they are able to arrive at
the parameter value that most consistently minimizes
the prediction error on different sets of test data. The
authors moreover show that the method described pro-
duces superior predictive results without significantly
sacrificing a model’s backward-looking explanatory
power. Thus, ridge regression shows itself to be an
enhancement over the standard OLS-based approach
to hedge fund replication because it can be reliably
employed in both ex post risk decomposition and ex
ante portfolio management.
REFERENCES
Agarwal, V., and Naik, N. 2000. “Generalised Style Analysis
of Hedge Funds.” Journal of Asset Management 1 (1): 93–109.
Amenc, N., L. Martellini, J. Meyfredi, and V. Ziemann. 2010.
“Passive Hedge Fund Replication—Beyond the Linear Case.”
European Financial Management 16 (2): 191–210.
Ennis, R., and Sebastian, M. 2003. “A Critical Look at the
Case for Hedge Funds: Lessons from the Bubble.” The Journal
of Portfolio Management 29 (4): 103–112.
Foster, M. 1961. “An Application of the Wiener-Kolmogorov
Smoothing Theory to Matrix Inversion.” Journal of the Society
for Industrial and Applied Mathematics 9 (3): 387–392.
Fung, W., and D. A. Hsieh. 1997. “Empirical Characteristics
of Dynamic Trading Strategies: The Case of Hedge Funds.”
Review of Financial Studies 10 (2): 275–302.
The Journal of Portfolio Management   163
——. 2004. “Hedge Fund Benchmarks: A Risk-Based
Approach.” Financial Analysts Journal 60 (5): 65–80.
Hasanhodzic, J., and A. W. Lo. 2007. “Can Hedge-Fund
Returns Be Replicated?: The Linear Case.” Journal of
Investment Management 5 (2): 5–45.
Hastie, T., R. Tibshirani, and J. Friedman. The Elements
of Statistical Learning: Data Mining, Inference and Prediction.
Springer, 2009.
Hoerl, A. E., and R. Kennard. 1970. “Ridge Regression:
Biased Estimation for Nonorthogonal Problems.” Technometrics
12 (1): 55–67.
Jaeger, L. Alternative Beta Strategies and Hedge Fund Replication.
Chichester, UK: Wiley Finance, 2008.
Larson, S. C. 1931. “The Shrinkage of the Coefficient of
Multiple Correlation.” Journal of Educational Psychology 22 (1):
45–55.
Markov, M., I. Muchnik, V. Mottl, and O. Krasotkina.
“Dynamic Analysis of Hedge Funds.” In Proceedings of the 3rd
IASTED International Conference on Financial Engineering and
Applications. Cambridge, MA: October 2006.
164   Factors in Time: Fine-Tuning Hedge Fund R eplication Quantitative Special Issue 2019
Sharpe, W. F. 1988. “Determining a Fund’s Effective Asset
Mix.” Investment Management Review 2 (6): 59–69.
——. 1992. “Asset Allocation: Management Style and Per-
formance Measurement.” The Journal of Portfolio Management
18 (2): 7–19.
Simonian, J., C. Wu, D. Itano, and V. Narayanam. 2019.
“A Machine Learning Approach to Risk Factors: A Case
Study Using the Fama–French–Carhart Model.” The Journal
of Financial Data Science 1 (1): 32–44.
Tikhonov, A. N. 1943. “On the Stability of Inverse Prob-
lems.” Doklady Akademii Nauk SSSR 39 (5): 195–198.
——. 1963. “Solution of Incorrectly Formulated Problems
and the Regularization Method.” Soviet Mathematics Doklady
4 (4): 1035–1038.
Disclaimer
This article ref lects the current opinions of the authors, which are not those
of Natixis Investment Managers.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

John M. Mulvey
is a professor at Princeton
University in the
Operations Research and
Financial Engineering
Department in Princeton,
NJ.
mulvey@princeton.edu
Lionel M artellini
is a professor of finance
at EDHEC Business
School and the director of
EDHEC–Risk Institute
in Nice, France.
lionel.martellini@edhec-risk.com
H an H ao
is a doctoral student
in the Operations
Research and Financial
Engineering Department
at Princeton University
in Princeton, NJ.
hhao@princeton.edu
Nongchao Li
is a doctoral student
in the Operations
Research and Financial
Engineering Department
at Princeton University
in Princeton, NJ.
nongchao@princeton.edu
Quantitative Special Issue 2019
A Factor- and Goal-Driven Model
for Defined Benefit Pensions:
Setting Realistic Benefits
John M. Mulvey, Lionel Martellini, Han Hao,
and Nongchao Li
T
hroughout the world, many
defined-pension plans are seri-
ously underfunded and unlikely
to support their future long-term
liabilities without massive contributions and
possible bailouts. This unfortunate situation
has been caused by a convergence of poor
contribution decisions, demographic head-
winds, and inadequate attention to setting
realistic benefits. In this article, we focus on
the benefit decisions.
From the perspective of capital, the
United States has the largest pension endow-
ment in the world—$25.4 trillion at the end
of 2017,1 or 131.2% of gross domestic product
(GDP) (Willis Towers Watson 2017). In com-
parison, France has an endowment of 6.5%
of GDP, and Germany has 12.9% of GDP.2
It would appear, therefore, that the US pen-
sion system is much better capitalized than
that in most countries. Unfortunately, many
defined benefit (DB) pensions in the United
States are severely underfunded, even with
high discount rates by public sector plans:
7% to 8%.
1
 The United States has 61% of the total pen-
sion assets among the 22 countries with the largest
pension plans.
2
 The second and third largest global pensions
are the United Kingdom and Japan, with roughly
$3.1 trillion assets under management each (Willis
Towers Watson 2017).
In 1998, most DB pensions were either
close to fully funded or overfunded. Over
the past 20 years, the funding ratios have
trended downward. Part of the plunge can
be attributed to the decrease in interest
rates over this period and to two equity
crashes—2001–2002 and 2008–2009. Nev-
ertheless, equities have rebounded from these
events. On the asset side, DB pensions in the
United States have had a strong tilt to high
returning assets, including equities (around
50%) and alternative investments (28%), with
solid asset performance.3 In comparison, the
Netherlands, for example, has 33% equi-
ties and 17% alternatives and possesses the
largest endowment relative to GDP at 193.8%
(Willis Towers Watson 2017). Hence, it is
difficult to blame asset performance. Instead,
we attribute the underfunding to two pri-
mary sources. First, historical contributions
have been too low to generate adequate cap-
ital relative to the plan promises. Exhibit 1
shows the contributions over time for one of
the largest plans within the California Public
Employees’ Retirement System (CalPERS):
15% of salary with much variability. Second,
plan benefits (liabilities) have been too gen-
erous, relative to the average contributions
and demographic conditions. Exhibit 2 shows
the long-term growth of the present value of
3
 For example, the average annual performance
by CalPERS has been 8.5% over the past 30+ years.
The Journal of Portfolio Management   165
Exhibit 1
Contribution Rate per Salary for California State Employees’ Miscellaneous Plan; Note Severe
Drop in Contribution Rate over 1998–2003 Period
(PSOR\HU&RQWULEXWLRQ5DWH















       

Source: CalPERS [2017].

Exhibit 2
CalPERS Liabilities over Time
+LVWRULFDO)XQGLQJ6WDWXVRI&DO3(56±
 





 







 




 
        

$VVHWVEQ $FWXDULDO/LDELOLWLHVEQ )XQGLQJ5DWLRV

Source: CalPERS [2019].

166   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
liabilities for CalPERS. The present value of liabilities
has exceeded inf lation and average salaries.4
To address the mismatch of contributions and
liabilities, we propose a factor-consistent and goal-driven
framework for managing a DB pension system. The deci-
sions regarding the size of the benefit promises should
be evaluated in conjunction with realistic assumptions
about the investment returns and the ability of the spon-
sors to make adequate contributions. Otherwise, there
will be an eventual crisis. Although the situation will
be most difficult to reverse in the United States,5 there
are opportunities to establish realistic pension systems
in fast-emerging countries such as China (1.5% of GDP)
and India. Furthermore, individuals will face similar
issues as they manage their own defined contribution
(DC) plans.
The usual order of business for a DB pension is to
compute its funding ratio6 annually with a prespecified
discount rate, amortize any deficits over multiple years,
and render required contributions. This process has been
unable to preserve the long-term health of DB pension
systems in the United States: It leads to undercontribu-
tions and a slow realization that the survivability of the
pension system is threatened. Furthermore, the volatility
of contributions can be high (Exhibit 1) and thereby
difficult to manage from the standpoint of the sponsor.
Today, there are DB plans that must contribute over
50% of employee salary to catch up with previous low
contribution levels.
We propose that a micro–macro simulation
be conducted to assess the factor/asset risks that the
planned contributions and investment returns achieve
the target promised benefits. An innovation in this
article is to evaluate the factor and goal-based analysis
from the standpoint of individuals (micro level). Here,
we evaluate a set of representative individuals to see if
there is an equilibrium and a reasonable chance that the
4
 The population of California has roughly doubled since
1970 (from 20 million to 40 million), along with inf lation that has
gone up 6.5 times. In concert, the number of state employees has
almost doubled as well—from 183,000 to 351,000 over the same
period. Liability growth has much exceeded these values.
5
 Modifying pension benefits ex post is most difficult in a
number of high-profile plans, such as the Illinois State pension
systems, because the benefits are virtually guaranteed by the Illinois
State constitution, even though the plan has an approximately 30%
funding ratio.
6
 The funding ratio is equal to the market value of assets
reduced by the present value of expected liabilities.
Quantitative Special Issue 2019
representatives, when aggregated as a whole group, are
able to achieve the desired goals at retirement as stand-
alone entities. This achieves two purposes. First, it helps
with the goal-setting process. If the promises are too
high relative to a majority of individuals, the pension
plan will be unable to support itself. Of course, if the
demographics are positive in the sense that workers sig-
nificantly outnumber retirees, the system can maintain
a generous retiree payment system; conversely, if the
demographics are negative (as they are in most of the
world), there will need to be extra contribution or lower
promises. Second, the approach provides a consistent
linkage between the factor risks and the critical invest-
ment, contribution, and benefit decisions. These issues
can be evaluated by means of the factor and goal-based
micro–macro system.
MULTIREGIME SCENARIO GENERATOR
In this section, we take up the critical task of
modeling the performance of the asset categories. First,
we employ the traditional approach for setting capital
market assumptions for macro risk factors, taking into
account historical performance and current conditions
(Ilmanen 2011; Bogle and Nolan 2015). We employ five
macro risk factors: (1) global equities, (2) US government
bonds, (3) high-yield bonds, (4) inf lation protection,
and (5) currency protection. Drawing from the factor
investing framework, we generate scenarios and param-
eter values for all asset classes that are consistent with
the scenarios generated for the five risk factors. In an
effort to obtain more robust estimates, the factor load-
ings are determined by means of a regularized regression
algorithm with a LASSO penalty. Blyth, Szigety, and
Xia (2016) applied a similar approach to managing the
Harvard University endowment.
The output is a set of stochastic scenarios: 10,000
in this study. Each scenario depicts a plausible path across
the planning period. Scenario generators similar to ours
have been in widespread use over the past 30+ years for
pension plans, insurance companies, and wealth man-
agement firms.7
To generate realistic scenarios relative to history,
we employ a two-regime Markov switching approach.
The idea is to address contraction/crash periods as
7
 Mulvey et al. (2004, 2009) and Mulvey, Bae, and Kim
(2013) have provided details about the underlying equations and
the methodology.
The Journal of Portfolio Management   167
Exhibit 3
Historical Performance of Equity Markets
Historical Equity Market Drawdowns
0.00%
–10.00%
–20.00%
–30.00%
Drawdown
–40.00%
–50.00%
–60.00%
–70.00%
–80.00%
–90.00%
1929 1939 1949
distinct from normal economic growth periods. Abun-
dant research shows that both volatility and correlation
increase considerably during crash regimes. Exhibit 3
shows the large drawdowns in equity that have occurred
since the 1920s in the United States. A single multi-
normal distribution cannot generate scenarios that depict
this type of performance. Furthermore, DB pension
plans with a tilt to equities and equity-like assets display
similar characteristics (see Exhibit 4).
There are advantages to identifying and employing
economic regimes within a consistent framework. The
multiple regime approach improves the realism of the
forward projections.8 In addition, machine learning
approaches can be applied to identifying regimes
(Tibshirani 1996, 2014; Zou and Hastie 2005; Hastie,
Tibshirani, and Friedman 2009).
The summary statistics of our scenario generator
are shown in Exhibit 5. First, we list the expected returns
and volatilities for the asset categories over the long
horizon needed for pension planning. Second, we sepa-
rate the expected returns for the two regimes—normal
8 the impact of asset allocation and other major deci-
 For example, see Kyle and Xiong (2001); Ang and Bekaert
(2002); Bauer, Haerden, and Molenaar (2004); Guidolin and
Timmermann (2007); Bilgili (2009); Tu (2010); Song (2014);
Goldstein and Razin (2015); Nystrup et al. (2015); and Mulvey
and Liu (2016).
168   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
S&P Composite Declines form All-Time High
1959 1969 1979 1989 1999 2009
Date
and crash. This breakdown is consistent with historical
performance (Mulvey and Liu 2016).
The underlying Markov model generates scenarios
based on a random transition matrix. At each period, we
fix the regime for the next month based on the transi-
tion probabilities: If we are in a normal regime, the
probability of remaining so for the next period (month)
is 98%. If we are in a crash regime, the probability of
remaining so is 87.5%. Thus, within a given scenario,
there should be a mixture of normal and crash periods.
On average, the steady state probability of a normal
regime is 86.2%, which is also consistent with history.
AN OVERVIEW OF MICRO–MACRO
ANALYSIS
This section considers the basic framework for
conducting factor-driven asset and liability management
(ALM) studies, with attention to the internal factor con-
sistency of the contribution decisions to the projected
benefit obligations for a DB pension plan. The output
of an effective ALM system should aim at measuring
sions, such as benefit selection, on portfolio priorities,
including ensuring the survivability of the plan. As men-
tioned, we are primarily interested in setting realistic
pension benefits.
Exhibit 4
Historical Performance for a Large DP Pension Plan
&DO3(56+LVWRU\RI,QYHVWPHQW5HWXUQV±





5HWXUQ






































































±

±

±
<HDU

'LVWULEXWLRQRI+LVWRULFDO5HWXUQRI&D,3(56±


)UHTXHQF\


± ± ±    
5HWXUQ

Source: CalPERS (2019).

To start, we make several assumptions within a
steady-state environment. The first assumption is that
across time, a pension plan should balance the sum of
expected contributions plus investment returns with the
expected outf lows to the retirees. Thus, on average, the
contributions by the employees and employer plus invest-
ment returns should equal expected benefit payments.
This macro-level consistency is required for a pension
plan to stay solvent. We also assume that the annual
contributions will be a fixed proportion of salary under

Quantitative Special Issue 2019 The Journal of Portfolio Management   169

Exhibit 5
Expected Returns for the Asset Categories Projected Nominal Returns for Asset

World US US High Real Corporate

Asset Equities Equities Treasuries Yield Estate Commodities Bonds
Panel A: Categories—Single Regime
Annualized Return 6.42% 7.31% 3.29% 5.79% 5.54% 5.03% 4.28%
Annualized Volatility 17.31% 15.56% 4.69% 8.27% 19.00% 20.33% 5.42%
Panel B: Projected Nominal Returns for Asset Categories—Normal Regime
Annualized Return 12.68% 14.30% 1.81% 10.03% 10.56% 7.57% 4.03%
Annualized Volatility 14.85% 12.37% 4.46% 6.19% 13.99% 18.30% 5.01%
Panel C: Projected Nominal Returns for Asset Categories—Crash Regime
Annualized Return –26.20% –28.00% 12.68% –17.60% –19.49% –9.19% 5.79%
Annualized Volatility 23.11% 22.36% 5.78% 14.38% 35.33% 29.35% 7.44%

stochastic inf lation. This assumption is implementable
by the plan sponsors and much less disruptive than large
changes in contributions (Exhibit 1).
At the micro-level, any individual, say Ms. Kim,
should have the same consistency—on average. For
example, suppose that an employee begins to work
for a sponsoring organization (X-Big) at the begin-
ning of her career age 25. She expects to work 40
years—until age 65, when she desires to retire. Let’s
assume that Ms. Kim remains alive and working for
X-Big until age 65, whereupon she retires and lives
for the next 20 years. Then, in a steady state whereby
the number of employees remains approximately con-
stant, the actuarial fair value should enforce the con-
straint: Future values of contribution and investment
performance equal the future value of future benefits.
Importantly, individuals in a DC plan cannot violate
this consistency rule.
The basic equation for the micro–macro simula-
tion over the long-term horizon involves the capital Ct,s
at the beginning of time t = [1, 2, …, T ] under each
scenario s e {S}. At each period, we model uncertainties
via a set of stochastic scenarios {S}. In our example, the
horizon T will depict a long time in the future. Thus,
at each time step t, we update the capital by adding
contributions from the population that is alive and
working and subtract the payments to the retirees who
remain alive. Furthermore, capital will be adjusted by
the asset performance during the previous period: rt–1,s.
The equation is defined as follows:
Ct ,s = Ct −1,s ∗ (1 + rt −1,s ) + xtcont
,s − x t ,s
pay
for all t,s
The cash-f low variables indicate the payments to
the retirees, xtpay cont
,s , and the contributions, x t ,s , at each
time period under each scenario. This framework
allows for a wide variety of investment, contribution,
and payment policies. In this article, our goal is to
evaluate the ability of the pension system to support
the promised payments.
Let’s look at a trivial example. Suppose that
Ms. Kim puts aside $20,000 each year, with adjustments
for annual inf lation. Thus, she contributes $20,000 ×
40 years = $800,000 real in total. For this highly simpli-
fied example, we assume that the investment returns are
equal to inf lation. During retirement, again, we assume
that the investment returns are equal to the inf lation
rate (through inf lation-linked government bonds, for
example). Thus, she is able to spend $40,000 adjusted
for inf lation per year, or $800,000, given that the real
discount rate is equal to 0%. A higher level of spending
or lower contributions would suggest that Ms. Kim is
getting more than she deserves, at least from an actu-
arial fair-value basis and with respect to the significant
inf lation risk factor.
We build on this idea by computing the fair value
of a population of individuals at the macro level by simu-
lating the fair values for a set of representative individ-
uals—micro level—with explicit longevity, investment,
and inf lation risks included (Exhibit 6). This type of
analysis fits within the realm of agent-based modeling
(Schelling 1971; Bonabeau 2002; Macal and North 2011;
Lychkina and Morozova 2014). The scenarios depict
uncertainties for inf lation, investment returns, and lon-
gevity risks across the long time period.

170   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
Exhibit 6
Consistent Linkage between Fair-Value of Individuals and Survivability of Pension System
  
:RUNIRUFH :RUNIRUFH :RUNIRUFH
VHOI VHOI VHOI
VXVWDLQLQJ VXVWDLQLQJ VXVWDLQLQJ
0LFUR±0DFUR&RQVLVWHQF\
Clearly, for actual individuals, fair value will be
violated under random circumstances. Ms. Kim may live
longer than age 85 and thus reap greater benefits than
her contributions. Or there might be a larger number of
workers than retirees, and again the benefits can exceed
the costs, as has been the case for DB pensions since their
inception in the late 1880s with Siemens and other firms
in Germany.9 Unfortunately, we are moving toward the
opposite environment, wherein the number of retirees
is beginning to exceed the number of workers. In this
case, a fair-value calculation can assist with setting rea-
sonably attainable benefits because it provides an upper
bound on benefits.
Stylized illustrations of a micro–macro analysis
based on our experiences with actual DB pension
plans are discussed in this section. These examples are
most appropriate for public sector pensions, in which
employees are likely to work within the same spon-
soring organization over their career, and for individual
DC plans.
The first case study shows that the US Social
Security system is reasonably fair valued, at least for
individuals with higher than average incomes, average
longevity risks, stochastic inf lation risks, and a steady-
state population. A primary reason for micro–macro
consistency is the relatively low level of benefits for
the target population: about 25% of final years’ salary
as defined by the Social Security maximum amount
and under stochastic inf lation. The level of benefits is
commensurate with the low returns (and safety) of the
investment mix: 100% government inf lation-linked
bonds. The micro–macro analysis confirms this consis-
9
 The earliest pension plans were created by Otto von
Bismarck in the late 1880s.
Quantitative Special Issue 2019
 
7LPHOLQH
tency. It fits a conservative liability-driven investment
(LDI) framework.
STYLIZED EXAMPLE: THE US SOCIAL
SECURITY SYSTEM
This analysis extends the previous example for
Ms. Kim. We approximate the contribution and the
benefits at the maximum Social Security level, then
apply this rule across a homogeneous population using
current longevity tables and the inf lation risk factor.
The purpose is to illustrate the process and provide an
example with approximate micro–macro consistency.
Next, we highlight the primary assumptions and issues.
The stylized micro–macro model of a small subset
of the Social Security Trust Fund10 focuses on the
matching of net contributions and benefits. Exhibit 7
provides high-level information on the current status
of the Trust Fund. We simulate individuals with some
wage distribution and life expectancy estimates by fol-
lowing the assumptions mentioned in the old age, sur-
vivors, and disability insurance (OASDI) annual report
10
 The Social Security program started in 1935 and has been
paying on a timely schedule since then. The program is funded by
contribution from both employers and employees in the form of pay-
roll taxes or self-employment taxes. It pays for retirement and other
benefits, such as disability. Beginning in 1941, the Social Security
Board of Trustees is required to present to the Congress a financial
report and detailed actuarial estimates of the fund. According to
Goss (2010), the board has six members, including the Secretary of
the Treasury as the managing trustee, the Secretary of Labor, the
Secretary of Health and Human Services, and the Commissioner of
Social Security, plus two public trustees appointed by the president
and confirmed by the Senate. Exhibit 7 shows the current financial
statement of the fund.
The Journal of Portfolio Management   171
Exhibit 7
Summary of 2017 Social Security Trust Fund Operations
>,QELOOLRQV@
2$6, ', 2$6',
$VVHWUHVHUYHVDWWKHHQGRI   
7RWDOLQFRPHLQ   
1HWSD\UROOWD[FRQWULEXWLRQV   
5HLPEXUVHPHQWIURP*HQHUDO)XQGRIWKH7UHDVXU\ D D D
7D[DWLRQRIEHQHILWV   
,QWHUHVW   
7RWDOH[SHQGLWXUHVLQ   
%HQHILWSD\PHQWV   
5DLOURDG5HWLUHPHQWILQDQFLDOLQWHUFKDQJH    
$GPLQLVWUDWLYHH[SHQVHV    
1HWLQFUHDVHLQDVVHWUHVHUYHVLQ    
$VVHWUHVHUYHVDWWKHHQGRI   

Notes: aLess than $50 million. Totals do not necessary equal the sums of rounded components.
Source: OASDI Board of Trustees (2018).

(OASDI Board of Trustees 2018) and applying some Economic Assumptions

simplifications. Starting with a population of 100,000
people who just joined the workforce at age 25, we We set wages at the maximum level to start, and
are interested in calculating the present value of their wages increase at the stochastic inf lation rate, which is
total net real contributions. Herein, we do not con- modeled by a normally distributed random variable with
sider demographic changes except the mortality rate. a mean of 2.1% and standard deviation of 1%. We assume
For example, for simplicity, the expansion/contraction that capital is invested in Treasury inf lation-protected
of the workforce is not modeled. securities (TIPS), which gives an assumed yield around
The current mortality rate, ages 25–110, is taken 0.25% on top of the inf lation rate. Thus, the discount
from the Social Security website. However, the mor- rate is simply the inf lation rate plus 0.25%. A sensitivity
tality rate declines as time goes by. In the three pos- analysis will be conducted to evaluate this assumption.
sible scenarios given by the Board of Trustees, the rate Social Security benef its follow cost-of-living
will decline each year by 0.41%, 0.77%, and 1.15% on adjustments (COLAs). From historical data, we found
average. To be consistent with the board’s assumptions, that the average COLA is about 0.05% higher than inf la-
we apply a rate of decline at 0.75% per year. tion, with a higher standard deviation. The correlation
between the inf lation rate and COLA is 0.7—a high
level, as expected.
Demographic Assumptions

Males and females have different expected mortality Program-Specific Assumptions

rates, so we simulate male and female employees separately.
The proportion is 53% male and 47% female. Currently,
the life expectancy at age 25 for men and women is about
53 years and 57 years, respectively. Individuals are evalu-
ated each year with respect to the number of remaining
workers and retirees throughout their simulated lives,
along with the accompanying contribution and benefit
cash flows adjusted for the stochastic inflation risk-factor.
Social Security benefits are roughly proportional
to the salary of the individual, so we just consider the
cap of the Social Security tax and correspondingly the
maximum full benefit employees will receive upon
retirement. As for 2018, the cap of income that is taxable
for Social Security is $132,900. The effective tax rate for
contribution (for brevity, we call it the contribution rate)

172   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
Exhibit 8
Sensitivity Analysis with 0.25% and 0.5% Real Returns on TIPS
3DQHO$3UREDELOLW\RI5HDFKLQJ*RDO5HDO5HWXUQV 3DQHO&3UREDELOLW\RI5HDFKLQJ*RDO5HDO5HWXUQV
           
             
             
             
             

3DQHO%0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO 3DQHO'0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO
5HDO5HWXUQV 5HDO5HWXUQV

 

3U5HDFK*RDO

3U5HDFK*RDO
 

 
 

 
 
 
 
 
 JH 
    $JH
  H$   H
  H WLU   WLU
&RQWULE
XWLRQ5

  5 &RQWULE
XWLRQ5   5H
DWH,Q3   DWH,Q3   
HUFHQWD HUFHQWD
JH JH
now is 12.4%. This means, in real terms, each person
in the new workforce (together with the employer) is
paying a maximum fixed amount of $16,479 per year.
In addition, the maximum benefit a person can receive
at full retirement age is $2,861 per month in today’s
dollars. We will assume this amount to increase at our
simulated stochastic COLA values.
Based on all of these assumptions, we generate
10,000 scenarios for each of the retirement age ranges
from 65 to 75 and calculate the probability of reaching the
goal of sustainability by calculating the probability that the
simulated present values of net contribution are larger than
zero across the 100,000-member population. The simula-
tion algorithm appears in the supplemental materials.
Baseline
In this section, we first show the baseline results
with the previous model specifications. Assuming the
contribution rate is 12.4% (current level), under different
retirement ages from 65 to 71 with an increment of
2, we calculate the probability of reaching the goal of
Quantitative Special Issue 2019

sustainability. Exhibit 8 shows that the probability of
reaching macro-consistency is 44.2% at age 65, 98%
at age 67, and virtually 100% at ages 69 and 71. In the
United States, the full retirement age for Social Secu-
rity benefits is moving toward 67, which corresponds
to a 98% probability of micro–macro consistency for
the target population. Thus, we see that the current
contribution and benefits have micro–macro consis-
tency from the standpoint of these 100,000 individuals
under steady-state assumptions. In addition, Exhibit 8
provides sensitivity analyses along two lines: changes in
the contribution amounts (from 12% to 13%) and two
return values (0.25% and 0.5%) above inf lation for the
TIPS. Clearly, the chance of achieving the modest goal
(25% of maximum social security salary) improves with
incremental increases in contribution and higher returns.
Moreover, as we observe, given a fixed contribution
rate and TIPS real return (0.25% per annum), the later
people retire, the higher the probability of reaching the
consistency goal. The probability increases from 44.2%
to 98% when the age of retirement increases from 65
The Journal of Portfolio Management   173
Exhibit 9
DB Pension Plan (10,000 scenarios), Traditional 60% Bond, 40% Equity
3DQHO$3UREDELOLW\RI5HDFKLQJ*RDO5HWLUHDW 3DQHO&3UREDELOLW\RI5HDFKLQJ*RDO5HWLUHDW
5HWLUHDW      5HWLUHDW     
           
           
           
           
           

3DQHO%0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO5HWLUHDW 3DQHO'0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO5HWLUHDW

 

3U5HDFK*RDO

3U5HDFK*RDO
 

 

 
 
 
 
 H  H
  DW   DW
   Q 5 H  
 Q 5 H
LR J LR J
   XW WD    XW WD
3D\RXW
5DWH,   W U LE UFHQ 3D\RX
W5DWH   W U LE UFHQ
Q3HUFH   Q H ,Q3H   Q H
QWDJH &R ,Q3 UFHQWD
to 67. The approach shows the benefits of LDI with a
stochastic inf lation risk factor.
EMPIRICAL TEST #2: EXAMPLES WITH
A PORTFOLIO OF RISKY ASSETS
In this section, we expand the previous analysis
to address the consistency issue with two portfolios of
risky assets.
First, we employ a 60/40 mix, in which 60% is
invested in US equities and 40% in US long-duration
government securities. Given the higher expected values
and the long-term horizon, the probability of reaching a
consistent plan is higher than that by investing in TIPS
alone—compare Exhibits 8 and 9. Even in this case,
however, the benefits must be restricted to approxi-
mately the 40% threshold under 15% contribution if
we want to achieve the goals with 90% confidence.
In comparison, financial planners suggest that retirees
should aim for 60%–80% of final salary during retire-
ment. Clearly, this goal would require at least 20%–25%
174   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
JH &R ,Q3
contribution per year, with a max cap to retire at age 65.
Extending retirement to age 70 is much more attainable
for contributions in the 15%–20% range.
In the next analysis, we introduce a portfolio11 that
is closer to those that larger DB pension plans employ
with a mix of asset categories, including world equity,
government and high-yield bonds, real estate, com-
modities, and corporate bonds. As mentioned, the factor
approach12 estimates the projected performance based on
the five macro factors. For simplicity, we assume that the
portfolio is rebalanced annually. As expected as shown
in Exhibit 10, this portfolio assumes higher expected
returns and thus achieves its goals with larger probabili-
ties, such as a 70% chance of retiring with 15% annual
contribution at age 65 versus a 63% chance. However, at
the higher levels of contribution, the traditional 60/40
mix provides more consistency and thereby generates
11
 Our sample portfolio is obtained via robust optimization
method. See Fabozzi et al. (2007).
12
 See Simonian et al. (2018).
Exhibit 10
DB Pension Plan (10,000 scenarios), Stylized Pension Asset Mix (world equity 25%, US equity 35%,
US Treasury 10%, high yield 5%, real estate 10%, commodity 5%, corporate bonds 10%)
3DQHO$3UREDELOLW\RI5HDFKLQJ*RDO5HWLUHDW 3DQHO&3UREDELOLW\RI5HDFKLQJ*RDO5HWLUHDW
5HWLUHDW      5HWLUHDW     
           
           
           
           
           

3DQHO%0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO5HWLUHDW 3DQHO'0HVK3ORWIRU3UREDELOLWLHVRI5HDFKLQJ*RDO5HWLUHDW

 

3U5HDFK*RDO

3U5HDFK*RDO
 

 

 
 
 
 
 H  H
  DW   DW
   Q 5 H    Q 5 H
LR J LR J
   XW WD    XW WD
3D\RXW
5DWH,   W U LE UFHQ 3D\RX
W5DWH   W U LE UFHQ
Q3HUFH   Q H ,Q3H   Q H
QWDJH &R ,Q3 UFHQWD
slightly higher probabilities. If the age of retirement is
delayed to age 70, the 60/40 strategy largely dominates
the second broader asset mix at age 65. It seems wise,
therefore, to expect increased retirement ages in the
future, even in the case of a steady-state population.
CONCLUSIONS
A critical objective for any liability- or goal-driven
asset owner is to evaluate the size of the benefits (prom-
ises or goals) relative to the chance of meeting these
promises and with respect to the underlying major risk
factors—longevity and inf lation, in our study. This
objective is attainable, in our opinion, for a DB pension
system within the developed factor- and goal-driven
framework. The first example suggests that even a
highly conservative asset allocation for the Social Secu-
rity system, 100% TIPs, can achieve a baseline goal for
a portion of the steady-state population. In the second
set of examples, we show that the selected benefit can be
Quantitative Special Issue 2019
JH &R ,Q3
increased above the Social Security 25% level if higher-
returning assets are part of the asset portfolio. How-
ever, even by increasing contribution to 15% per year
from 12.4%, we are confident of reaching this goal with
90% only if the benefits are set at 40% of the final max
salary at age 65. Of course, higher contributions and
later retirement will generate greater levels of attainable
benefits. Additional real-world considerations,13 such as
goal-risk measures and dynamic investment strategies
(Martellini, Malhau, and Mulvey 2018, 2019), can be
included in the micro–macro modeling approach.
In summary, the determination of appropriate
pension benefits should be evaluated on several fronts.
First, the benefits need to be fairly attainable for the
sponsoring organization and the employees. Under most
circumstances, a long-term contribution rate above a
13
 The average years of service for all retirees in some of the
plans in CalPERS is 19.8 years (Fox 2017). The micro–macro simu-
lation can include real-world considerations of this type.
The Journal of Portfolio Management   175
threshold, such as 30%-40% of salary, is unreasonable
in our opinion. In addition, there may be a need for an
upper bound on the salary levels for fairness and eco-
nomic consistency.14 Finally, it is becoming obvious
that the traditional retirement age 65 will need to be
increased going forward. These policy trade-offs should
be evaluated by means of systematic analysis. Our factor-
and goal-based framework provides a benchmark (or
upper bound under demographic headwinds) for these
important societal discussions.
REFERENCES
Ang, A., and G. Bekaert. 2002. “International Asset Allo-
cation with Regime Shifts.” The Review of Financial Studies
15 (4): 1137–1187.
Bauer, R., R. Haerden, and R. Molenaar. 2004. “Asset Allo-
cation in Stable and Unstable Times.” The Journal of Investing
13 (3): 72–80.
Bilgili, M. “Clustering Techniques and Multi Regime Sto-
chastic Optimization with Applications in Finance.” Ph.D.
dissertation, Princeton University, 2009.
Blyth, S., M. C. Szigety, and J. Xia. 2016. “Flexible Indeter-
minate Factor-Based Asset Allocation.” The Journal of Portfolio
Management 42 (5): 79–93.
Bogle, J., and M. Nolan Jr. 2015. “Occam’s Razor Redux:
Establishing Reasonable Expectations for Financial Market
Returns.” The Journal of Portfolio Management 42 (1): 119–134.
Bonabeau, E. 2002. “Agent-Based Modeling: Methods and
Techniques for Simulating Human Systems.” Proceedings of
the National Academy of Sciences of the United States of America
99 (3): 7280–7287.
CalPERS. “California Public Employees’ Retirement System
Investments.” 2019, https://www.calpers.ca.gov/page/
investments.
CalPERS. “CalPERS State Actuarial Valuation Report.” 2017,
https://www.calpers.ca.gov/docs/forms-publications/2017-
state-valuation.pdf.
Fabozzi, F., P. Kolm, D. Pachamanova, and S. Focardi. Robust
Optimization and Management. Hoboken: John Wiley, 2007.
14
 Similar to the Social Security system, CalPERS recently
added a retirement cap to new employees.
176   A Factor- and Goal-Driven Model for Defined Benefit Pensions: Setting R ealistic Benefits Quantitative Special Issue 2019
Fox, K. “CalPERS Update and Path Forward.” Presentation
to the League of California Cities, December 13, 2017.
Goldstein, I., and A. Razin. 2015. “Three Branches of Theo-
ries of Financial Crises.” Foundations and Trends in Finance
10 (2): 113–180.
Goss, S. C. 2010. “The Future Financial Status of Social
Security System.” Social Security Bulletin 70 (3): 111–125.
Guidolin, M., and A. Timmermann. 2007. “Asset Allocation
under Multivariate Regime Switching.” Journal of Economic
Dynamics and Control 31 (11): 3503–3544.
Hastie, T., R. Tibshirani, and J. Friedman. The Elements of
Statistical Learning: Data Mining, Inference, and Prediction. 2nd
ed. New York: Springer, 2009.
Ilmanen, A. Expected Returns: An Investor’s Guide to Harvesting
Market Rewards. Chichester: John Wiley & Sons, 2011.
Kyle, A., and Xiong, W. 2001. “Contagion as a Wealth
Effect.” The Journal of Finance 56 (4): 1401–1440.
Lychkina, N., and Y. Morozova “Dynamic Simulation of
Pension System Development Processes.” 32nd Interna-
tional Conference of the System Dynamics Society, Delft,
the Netherlands, July 24–28, 2014.
Macal, C., and M. North. “Introductory Tutorial: Agent-
Based Modeling Simulation.” Proceedings of the 2011 Winter
Simulation Conference, 2011.
Martellini, L., V. Malhau, and J. Mulvey. “Securing Replace-
ment Income with Goal-Based Retirement Income Strate-
gies.” Working paper, 2018.
——. 2019. “Flexicure Retirement Solutions: A Part of
the Answer to the Pension Crisis?” The Journal of Portfolio
Management (forthcoming).
Mulvey, J., G. Bae, and W. Kim. 2013. “Dynamic Asset Allo-
cation for Varied Financial Markets under Regime Switching
Framework.” European Journal of Operational Research 234 (2):
450–458.
Mulvey, J., F. Fabozzi, W. Pauling, and K. Simsek. 2004.
“Modernizing the Defined Benefit Pension System.” The
Journal of Portfolio Management 31 (2): 73–82.
Mulvey, J. M., and H. Liu. 2016. “Identifying Economic
Regimes: Reducing Downside Risks for University Endow-
ments and Foundations.” The Journal of Portfolio Management
43 (1): 100–108.
Mulvey, J., K. Simsek, Z. Zhang, and F. Fabozzi. 2009.
“Assisting Defined-Benefit Pension Plans.” Operations Research
56 (5): 1066–1078.
Nystrup, P., B. W. Hansen, H. Madsen, and E. Lindstrom.
2015. “Regime-Based versus Static Asset Allocation: Letting
the Data Speak.” The Journal of Portfolio Management 42 (1):
103–109.
OASDI Board of Trustees. “The 2018 OASDI Trustees
Report.” 2018, https://www.ssa.gov/OACT/TR/2018/
tr2018.pdf.
Schelling, T. C. 1971. “Dynamic Models of Segregation.”
Journal of Mathematical Sociology 1 (2): 143–186.
Simonian, J., O. Sosa, S. Chandrashekar, and D. Nielson.
2018. “LDI-Sensitive Equity Factor Portfolios: The ALM
Perspective to Smart Beta Investing.” The Journal of Portfolio
Management 44 (4): 106–112.
Tibshirani, R. 1996. “Regression Shrinkage and Selection
via the LASSO.” Journal of the Royal Statistical Society. Series B
(Methodological) 58 (1): 267–288.
——. 2014. “Adaptive Piecewise Polynomial Estimation via
Trend Filtering.” The Annals of Statistics 42 (1): 285–323.
Tu, J. 2010. “Is Regime Switching in Stock Returns Impor-
tant in Portfolio Decisions?” Management Science 56 (7):
1198–1215.
Willis Towers Watson. “Global Pension Asset Study.”
2017, https://w w w.wil listowerswatson.com/en-US/
insights/2017/01/global-pensions-asset-study-2017.
Zou, H., and T. Hastie. 2005. “Regularization and Variable
Selection via the Elastic Net.” Journal of the Royal Statistical
Society. Series B (Statistical Methodology) 67 (2): 301–320.
To order reprints of this article, please contact David Rowe at
d.rowe@pageantmedia.com or 646-891-2157.

Song, Y. 2014. “Modelling Regime Switching and Structural

Breaks with an Infinite Hidden Markov Model.” Journal of
Applied Econometrics 29 (5): 825–842.

Quantitative Special Issue 2019 The Journal of Portfolio Management   177

 PATRO N SPO N SO R

178    The Journal of Portfolio M anagement Quantitative Special Issue 2019

 Jacobs Levy
Equity Management
100 Campus Drive
P.O. Box 650
Florham Park, NJ 07932-0650
www.jacobslevy.com
jacobslevy@jlem.com
(973) 410-9222
Contacts
BRUCE I. JACOBS
Principal, Co-Founder
KENNETH N. LEVY
Principal, Co-Founder
About Jacobs Levy
Jacobs Levy Equity Management focuses exclusively
on managing U.S. equity portfolios for institutional
clients. Strategies offered by the firm include long eq-
uity, defensive equity, 130-30 long-short, and absolute
return portfolios. Building on the pioneering research
of founders Bruce Jacobs and Ken Levy, the firm has
developed a unique, multidimensional, dynamic ap-
proach to investing that combines human insight and
intuition, finance and behavioral theory, leading-edge
quantitative and statistical methods, and over 30 years
of proprietary research. Jacobs Levy manages assets
for a prestigious global roster of corporate defined ben-
efit and defined contribution plans, public retirement
systems, sub-advised funds, and endowments/founda-
tions, many of which are Pensions & Investments’ “Top
200 Pension Funds/Sponsors.”
We are proud to have established the Bernstein
Fabozzi/Jacobs Levy Awards and the Wharton-Jacobs
Levy Prize for Quantitative Financial Innovation, and to
have introduced many investment concepts in the fol-
lowing articles:
Disentangling Market Inefficiencies
• “Investing in a Multidimensional Market”
• “Ten Investment Insights That Matter”
Quantitative Special Issue 2019 The Journal of Portfolio M anagement    179
• “The Complexity of the Stock Market”
• “Disentangling Equity Return Regularities:
New Insights and Investment Opportunities”
Portfolio Engineering
• “Smart Beta: Too Good to be True?”
• “Smart Beta versus Smart Alpha”
• “Residual Risk: How Much is Too Much?”
• “Alpha Transport with Derivatives”
• “The Law of One Alpha”
Market Neutral Long-Short Equity Strategies
• “20 Myths about Long-Short”
• “Long-Short Portfolio Management: An Integrated
Approach”
• “On the Optimality of Long-Short Strategies”
Enhanced Active Equity (130-30) Strategies
• “20 Myths about Enhanced Active 120-20 Strategies”
• “Enhanced Active Equity Strategies: Relaxing the
Long-Only Constraint in the Pursuit of Active Return”
Leverage Aversion and Portfolio Optimality
• “The Unique Risks of Portfolio Leverage: Why Modern
Portfolio Theory Fails and How to Fix It”
• “Traditional Optimization Is Not Optimal for Leverage-
Averse Investors”
• “A Comparison of the Mean-Variance-Leverage
Optimization Model and the Markowitz General
Mean-Variance Portfolio Selection Model”
• “Leverage Aversion, Efficient Frontiers, and the
Efficient Region”
Market Simulation
• “Simulating Security Markets in Dynamic and
Equilibrium Modes”
• “Financial Market Simulation”
Submission Guidelines

A
ll submitted manuscripts must be original work that hasn’t been published in another form such
as another journal, magazine, or book chapter. Upon submitting a manuscript for consideration
by Institutional Portfolio Research Journals (IPR Journals), an author is restricted from submitting
their manuscript elsewhere until an editorial decision on their work has been made. Submissions are
only accepted by uploading to our online article submission system. We do not require any specific style
or format for initial submission. However, upon acceptance of an article for publication, the author will
be expected to bring their manuscript into conformity with the stylistic and mechanical specifications
described in the full submission guidelines. Also at that time, the author will be expected to sign the IPR
Journals copyright agreement and to remove all prior versions of the article from everywhere that they
have been posted, including SSRN and all other websites.

Target Length: The target length for articles is 4,000 words,
not including the abstract, exhibits, and references. How-
ever, the editor of a journal may accept articles that are
substantially longer or shorter. Articles shorter than 2,500
words or longer than 7,500 words are rarely accepted.
Title: You should give your article a short descriptive title. In
general, an article’s title should be no more than 12 words.
You should try to avoid including words that carry little or
no meaning.
Abstract: You should write a strong abstract for an article
appearing in an IPR Journals publication. The target length
for abstracts is 160 words. The abstract should be non-tech-
nical, should not contain any references, and should define
any acronyms on first use. An article’s abstract is the only
part of the article that appears on the unrestricted portion
of the related journal’s website. In the printed version of a
journal, the abstract appears in the table of contents (as
if it is written by the editorial team, not the authors). It does
not appear directly before the article’s main text.
Introductory Material: Please write strong introductory para-
graphs. The introductory material is an important part of an
article. It serves as a substitute for the abstract at the start
of the article. Because an abstract is not presented right
before its associated article, a reader will not necessarily
have read the abstract before reading the main text. An
article’s introductory material should describe the article’s
main points and briefly explain the reasoning or analysis
that supports them. The introductory material should explain
why the findings, results, or conclusions are important and
how they add to prior research in the area. If applicable,
the introductory material should convey how an investment
professional can use the findings, results, or conclusions in a
practical way. In most cases it is helpful to include a “road-
map” paragraph at the end of the introductory material.
The purpose of the roadmap paragraph is to inform readers
about how the article is organized into different sections and
what is covered in each one.
Headings: Articles in IPR Journals publications can have up to
three levels of headings. The first heading in an article should
appear after the introductory material. Articles in IPR Journals
publications do not include any heading before an article’s
introductory material.
Conclusion: Please end the main text of an article with a short
conclusion that briefly summarizes the article’s main point.
The conclusion may resemble the article’s abstract, though
it need not omit references and technical terms used in the
article. The conclusion should not introduce new material.
Footnotes: IPR Journals publications now use footnotes rather
than endnotes. Footnotes generally should not be used for
references but rather for peripheral or tangential commen-
tary that would not be suitable for inclusion in the main text.
References: IPR Journals follows the “author-date” reference
format of The Chicago Manual of Style (CMS), 17th Edition,
Chapter 15 both for references included in the text of an arti-
cle and for the list of references at the end of an article. The
list of references should immediately follow the main text.
Style, Usage, and Grammar: IPR Journals follow CMS with
respect to matters of style, usage, and grammar. This includes
standards for hyphenation, abbreviations, and capitalization.
Exhibits: Please call all tabular or graphical material Exhibits.
Number exhibits with Arabic numbers consecutively in order
of appearance in the text. Do not call them Table 1 and
Figure 1. An article’s text should refer to all of its exhibits. In
general, the text should clearly explain the point supported
or demonstrated by each exhibit. Additionally, except for
the simplest tables and charts, the text should explain how
to read each exhibit. For tables that report statistical results,
the text should direct readers to the key values and clearly
explain what they mean. For an equation, the text should
describe and explain the relationship represented by the
equation and clearly define all terms used in the equation.
In many cases, it is better to place content that is heavy with
equations in an appendix. Within the IPR Journals family of
journals, The Journal of Fixed Income and The Journal of
Derivatives generally allow for the highest proportions of
quantitative material in the main text.
Exhibit Presentation: Authors are encouraged to submit arti-
cles with exhibits in color. Although IPR Journals publications
are generally printed in black and white, the online version
can show color exhibits. Also, upon article acceptance, if
you wish to have your exhibits printed in color, you may do
so at a cost of $1,200. Please make sure that all categories
in an exhibit can be distinguished from each other and
match the categories in any legend connected with an
exhibit. If you use color, please make sure that each exhibit
will be legible if printed in black-and-white. This is important
for all articles because a reader may print an article on a
black-and-white printer.
Copyright Agreement: IPR Journals’ copyright agreement
form must be signed prior to publication. Only one author’s
signature is necessary. Should your article not be accepted
the copyright agreement will be considered null and void.
Upon acceptance of the article, no further changes are
allowed, except with the permission of the editor.
Submit Your Article to JPM

jpm.iprjournals.com

1
Go to jpm.iprjournals.com
and choose Submit an article

2
Click to SUBMIT AN ARTICLE and
follow the prompts to register with
Editorial Manager

3
Scroll down to download
the complete version of the
Submission Guidelines
 We’re proud to support and
contribute to the advancement
of investment management research.

aqr.com