Dissecting Anomalies with a Five-Factor Model

Author(s): Eugene F. Fama and Kenneth R. French

Source: The Review of Financial Studies , January 2016, Vol. 29, No. 1 (January 2016),
pp. 69-103
Published by: Oxford University Press. Sponsor: The Society for Financial Studies.

Stable URL: https://www.jstor.org/stable/43866012

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

and Oxford University Press are collaborating with JSTOR to digitize, preserve and extend
access to The Review of Financial Studies

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor
Model

Eugene F. Fama
Booth School of Business, University of Chicago

Kenneth R. French
Amos Tuck School of Business, Dartmouth College

A five-factor model that adds profitability ( RMW ) and investment ( CMA ) factors to the
three-factor model of Fama and French (1993) suggests a shared story for several average-
return anomalies. Specifically, positive exposures to RMW and CMA (stock returns that
behave like those of profitable firms that invest conservatively) capture the high average
returns associated with low market ß , share repurchases, and low stock return volatility.
Conversely, negative RMW and CMA slopes (like those of relatively unprofitable firms that
invest aggressively) help explain the low average stock returns associated with high ß , large
share issues, and highly volatile returns. {JEL G 1 , G 1 1 , G 1 2)

Received November 11, 2014; accepted April 27, 2015 by Editor Andrew Karolyi.

Motivated by the dividend discount valuation model, Fama and French

(FF; 2015) add profitability and investment factors to the market, Size , and
value/growth factors of the three-factor model of Fama and French (FF; 1993).
In FF (2015) the left-hand-side (LHS) assets used to test the resulting five-
factor model are portfolios formed using sorts on Size (market capitalization or
market cap) and combinations of the book-to-market equity ratio, profitability,
and investment. The LHS portfolios are thus just finer sorts on the variables
used to construct the factors.
Here we follow the advice of Lewellen, Nagel, and Shanken (2010) and
consider anomalies not targeted by the five-factor model and known to cause
problems for the FF three-factor model. Accruals (Sloan 1996), net share issues
(Ikenberry, Lakonishok, and Vermaelen 1995; Loughran and Ritter 1995),
momentum (Jegadeesh and Titman 1993), and volatility (Ang et al. 2006) are
prominent examples. There is also long-standing evidence (Black, Jensen, and
Scholes 1972; Fama and MacBeth 1973) that the relation between average
return and market ß is flatter than predicted by the Sharpe (1964)-Lintner

The authors thank Savina Rizova for constructing the data files and Andrew Karolyi (the editor) and two referees
for comments that substantially improved the paper. E. F. F. and K. R. F. are consultants to, board members
of, and shareholders in Dimensional Fund Advisors. Send correspondence to Kenneth R. French, Dartmouth
College, Hanover, NH 03755; telephone: 603-643-5750. E-mail: kfrench@dartmouth.edu.

©The Author 2015.Published by Oxford University Press on behalf of The Society for Financial Studies.
All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.
doi : 1 0. 1 093/rfs/hhv043 Advance Access publication August 1 0, 20 1 5

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

(1965) CAPM. The goal here is to examine whether the five-factor model and
models that use subsets of its factors capture average returns from sorts on these
variables and whether portfolios that signal model problems have exposures
to the size, profitability, and investment factors typical of stocks that cause
problems for the five-factor model in many sorts in FF (2015).
Given the large number of anomalies researchers have discovered in stock
returns, one might ask why the additions to the FF (1993) three-factor model
are profitability and investment factors. FF (2015) argue that these are natural
choices, suggested by the dividend discount model. Miller and Modigliani
(1961) show that in the dividend discount model, Mt, the time t market cap of
a firm's stock, is,
00

M, = J2W,+T-dBt+T)/(l+ry. (1)
r=l

In this equation, Yt+r is equity earnings for period t + r , d Bt+r = Bt+Z - Bt+r _ i
is the change in book equity, and r, the internal rate of return on expected cash
flows to shareholders, is approximately the long-term expected stock return.
Dividing by time t book equity gives
oo

£E (Yt+r-dB
)/d +ry
t+r

- = -
B, B,

Equation (2) implies that if

value of the stock, M,, and
Mt , or equivalently a highe
higher expected return. Simi
investments (d Bt+I ) fixed, h
cash flows to shareholders an
Mt, Bt, and the stream of fu
lower expected cash flows an
Bt/Mt is a noisy proxy for
responds to forecasts of earn
Most of our tests use variant

Rit - Rpt -ū[ +bi(RMt - Rfí)+

(3)
In this equation Ru is the month t return on one of the portfolios from sorts
of stocks on Size and an anomaly variable, RFt is the risk-free rate (the one-
month U.S. Treasury-bill rate observed at the beginning of month t ), RMt is
the return on the value- weight (VW) portfolio of NYSE-AMEX-NASDAQ
stocks, SMBt (small minus big) and HMLt (high minus low B/M ) are the Size
and value factors of the FF three-factor model, RMW t (robust minus weak) is a
profitability factor, and CMAt (conservative minus aggressive) is an investment
factor.

70

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor Model

The bottom line from our tests is that the list of anomalies shrinks when
we use the five-factor model, in part because anomaly returns become less
anomalous and in part because the returns for different anomalies have similar
five-factor exposures (regression slopes in (3)) that suggest they are related
phenomena. With two exceptions, accruals and momentum, the five-factor
model shrinks anomaly average returns left unexplained by the FF three-factor
model. Moreover, the successes and failures of the model are linked to patterns
in the slopes for RMWt and CMAt that are common to the sorts on ß , net share
issues, and volatility. The high average returns associated with low ß , share
repurchases, and low volatility that are left unexplained by the three-factor
model are absorbed by positive five-factor exposures to RMWt and CMAt ,
typical of profitable firms that invest conservatively. At the other extreme, the
low average returns associated with high ß , large share issues, and high return
volatility that are left unexplained by the three-factor model are substantially
captured by negative five-factor exposures to RMWt and CMAt , typical of less
profitable firms that invest aggressively.
In the sorts on net share issues and volatility, the portfolios that cause
the most serious problems for the five-factor model are in the smaller Size
quintiles and the highest quintiles of share issues and volatility. These portfolios
have negative exposures to RMWt and CMAt that lower estimates of their
expected returns, but not enough to explain their low average returns. Most
interesting, the common patterns in the five-factor slopes for these portfolios
suggest they share the lethal traits - small stocks whose returns behave like
those of relatively unprofitable firms that invest aggressively - that plague the
five-factor model in FF (2015).
Accruals pose special problems. For other anomalies, the five-factor model
improves the description of average returns of the FF three-factor model. For
accruals the five-factor model does worse. The problem is that in the sorts
on accruals, portfolios in the smallest Size quintile (microcaps) have negative
RMWt slopes but they do not have the predicted low average returns. Hou,
Xue, and Zhang (2015) also find that sorts on accruals produce average returns
that escape explanation by a model similar to ours.
For the anomalies discussed above, adding a momentum factor to the
five-factor model has little effect on performance, simply because the sorts
do not produce portfolios with large momentum tilts. For portfolios formed
on momentum, however, the five-factor model does poorly, with regression
intercepts about as disperse as average returns on the portfolios. Adding
a momentum factor improves model performance, but leaves nontrivial
unexplained momentum returns among small stocks.

1. The Factors

Dropping the time subscript on the variables, the tests of the five-factor model
use the Rm - Rf,SMB , and HML factors of the three-factor model of FF (1993),

71

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

Table 1
Averages, standard deviations, and /-statistics for monthly factor returns, July 1963-December 2014 (618
months)

rm-Rf SMB HML RMW CMA MOM

Mean ÕJI 027 036 025 032 Õ69~
SD 4.46 3.07 2.86 2.14 1.99 4.22
/-statistic 2.83 2.20 3.15 2.88 4.04 4.05

rM - rF 1S the value- weight return on the m

bill rate. At the end of each June, NYSE, AM
big) using the NYSE median market-cap bre
(low to high), using NYSE 30th and 70th per
value-weight Size-B/M portfolios. In the sor
ending in year / - 1 and M is market cap at
outstanding between the measurement of B
equity, defined as ( 1 ) stockholders equity (or
liabilities, in that order) minus (2) the redem
balance sheet deferred taxes, if available, min
equity data for NYSE stocks as in Davis, Fam
two high B/M portfolios from the 2 x 3 sor
The profitability and investment factors, RM
are formed in the same way as HML, except
Operating profitability, OP, in the sort for
ending in year t - 1 and is revenue minus t
expenses, minus interest expense, all divided
the fiscal year ending in year t - 2 to the fis
Size-B/M, Size-OP, and Size-Inv sorts, there
average of the three, or equivalently, it is t
the three sorts minus the average of the nine
same way as HML, except the factor is upda
portfolios at the end of month t - 1, Size is
cumulative return for the 1 1 months from
standard deviations of monthly returns (SD),

augmented with similar profitab

factors of the original model use
and three book-to-market equ
Size breakpoint is the NYSE m
are the 30th and 70th percenti
of the sorts produce six VW po
of the three small stock portf
stock portfolio returns. The va
B/M portfolio returns minus th
The profitability and investme
the same way as HML , except
(OP) or investment (Inv). Defini
construction are in Table 1 .
The 2x3 sorts used to construct RMW and CMA produce two additional
Size factors: SMBop and SMB¡nv. The Size factor SMB used in the tests is the
average of the returns on the nine small stock portfolios of the three 2x3 sorts
minus the average of the returns on the nine big stock portfolios.
No combination of the factors in (3) explains average returns on portfolios
formed on momentum. Thus, in the tests in which the LHS asset returns to
be explained are for momentum portfolios, we include a momentum factor,

72

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor Model

MOM , among the right-hand-side (RHS) explanatory returns. MOM is defined

like HML , except that it is updated monthly rather than annually, and the sort for
portfolios formed at the end of month / - 1 is based on the cumulative average
returns from /- 12 to /- 2, called Prior 2-12. Note that MOM is reconstituted
monthly using the most recently available data, whereas SMB , HML , RMW,
and CMA are updated annually using data that, except for Size , are at least six
months old. (See Table 1.)
Our sample is the 61 8 months from July 1963 to December 2014 (henceforth
1963-2014). The average monthly returns on the factors for this period are
all more than two standard errors above zero (Table 1). The average equity
premium ( RM - Rf) for 1963-2014 is large, 0.5 1 % per month, but the monthly
standard deviation is also substantial, 4.46%, and the /-statistic for the average
premium is 2.83. The average HML return has a larger /-statistic, 3.15, the
result of combining a smaller average premium, 0.36% per month, with a
smaller standard deviation, 2.86%. The profitability factor, RMW , has the lowest
average premium, 0.25% per month, but because of its low standard deviation,
2.14%, its /-statistic is 2.88. The investment and momentum factors, CMA and
MOM , have the largest /-statistics, 4.04 and 4.05. The large /-statistic for CMA
combines a moderate factor premium of 0.32% per month with the lowest factor
standard deviation, 1.99%. In contrast, the large /-statistic for MOM combines
the second highest standard deviation, 4.22%, with the highest average return,
0.69% per month. FF (2015) provide more detail on factor construction and the
behavior of factor returns.

The momentum factor plays a critical role when the LHS returns in asset
pricing regressions are for momentum portfolios. But including MOM produces
small changes in model performance when the LHS portfolios (here and in FF
2015) are not formed on momentum. Thus, we put the momentum factor aside
except when we address the momentum anomaly. We have also tried models that
add the liquidity factor of Pastor and Stambaugh (2003) to different versions
of regression (3). Skipping the details, the portfolios examined here (and in
FF 2015) have trivial loadings on the traded and nontraded versions of the
liquidity factor, and including the traded version produces only tiny changes in
regression intercepts.

2. Summary Asset Pricing Tests

We turn now to our central task: examining how well variants of the five-factor
model capture average returns on portfolios formed on Size and each of the
anomaly variables. Two-way sorts on Size and an anomaly variable allow us
to see how anomaly returns, and explanations of them provided by different
models, vary across Size groups. This section presents summary tests. Later
sections examine regression intercepts and pertinent slopes for each anomaly
variable. We begin by introducing the left-hand-side portfolios in the tests.

73

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

2.1 The LHS anomaly portfolios

2.1.1 Market ß. Many studies, from Black, Jensen, and Scholes (1972) and
Fama and MacBeth (1973) to Frazzini and Pedersen (2014) find that the relation
between univariate market ß and average stock return is flatter than predicted by
the C APM of Sharpe ( 1 964) and Lintner ( 1 965). We construct 25 VW portfolios
at the end of June each year from independent sorts of stocks into quintiles of
Size at the end of June and ß estimated using the preceding five years (two years
minimum) of past monthly returns. As in all our sorts, the quintile breakpoints
use NYSE stocks, but the sample is NYSE, AMEX, and NASDAQ stocks on
both CRSP and Compustat with data for the variables in the sort and share
codes 10 or 11. For a bit of color, stocks in the bottom and top Size quintiles
are often called microcaps and megacaps.

2.1.2 Net share issues. Share repurchases tend to be followed by large

average returns (Ikenberry, Lakonishok, and Vermaelen 1995), and average
returns after share issues tend to be low (Loughran and Ritter 1995). We form
35 portfolios from independent sorts of stocks into Size quintiles and seven
net share issues (NI) groups. Portfolio formation follows the same rules as
the Size-ß sorts, except the second sort is on NI and there are seven groups -
negative NI (net repurchases), zero NI, and quintiles of positive NI (net issues).
The choice of one repurchase group is a bit arbitrary but in line with the fact
that net repurchases are less frequent than net issues, and for big stocks a
finer breakdown of repurchases would produce undi versified portfolios. For
portfolios formed in June of year t, NI is the change in the natural log of
split-adjusted shares outstanding from the fiscal year-end in t - 2 to the fiscal
year-end in ř - 1 .

2.1.3 Volatility. Ang et al. (2006) find that stocks with highly volatile
returns tend to have low average returns whether volatility is measured as
the variance of daily returns or as the variance of the residuals from the FF
three-factor model. We construct VW portfolios using monthly sorts on Size
and Var or Size and RVar , where Var is the variance of daily returns, and RVar is
the variance of daily residuals from the FF three-factor model, both estimated
using 60 (with a minimum 20) days of lagged returns. We examine quintiles of
Size and Var or RVar but in contrast to other sorts, the NYSE breakpoints for
Var and RVar are set separately for each Size quintile. This reflects the fact that
with unconditional NYSE breakpoints, the highest Var and RVar quintiles are
mostly microcaps, and the megacap portfolios in the highest volatility quintiles
are thin and sometimes empty.

2.1.4 Accruals. Sloan (1996) is the seminal paper in the literature on the
low returns associated with high accruals. Accruals arise because accounting
decisions cause book earnings to differ from cash earnings. Our tests of
the accruals anomaly use 25 VW portfolios formed from the intersection of

74

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

independent sorts of stocks into Size and accrual (AC) quintiles. The portfolios
are formed at the end of June of each year t. Size is market cap at the end of
June and accruals are the change in operating working capital per split-adjusted
share from the fiscal year-end in t - 2 to t - 1 divided by book equity per share
in ř - 1 .

2.1.5 Momentum. Jegadeesh and Titman (1993) document momentum in

stock returns. For example, the relative performance of stocks in months t - 12
tot - 2 tends to persist in month t. We form 25 VW momentum portfolios every
month. The portfolios for month t , formed at the end of month t - 1, are the
intersection of quintiles from independent sorts on Size (market cap at the end
of t - 1), and Prior 2-12 (the sum of a stock's monthly returns from t - 12 to
t-2).
The troublesome portfolios in our asset pricing tests are typically in the
smaller Size quintiles. For perspective, with NYSE Size breaks, the microcap
quintile on average contains 57% of NYSE-AMEX-NASDAQ stocks but only
2.9% of aggregate market cap. The next Size quintile on average includes 14.7%
of stocks but only 3.7% of aggregate market cap. In contrast, the megacap Size
quintile on average accounts for 8.6% of NYSE-AMEX-NASDAQ stocks, but
they are 74.8% of aggregate market cap, and the two largest Size quintiles
together average 87.3% of aggregate market cap.

2.2 Summary tests

If an asset pricing model captures expected returns, the intercept is
indistinguishable from zero in the time-series regression of any asset's excess
return (its return in excess of the risk-free rate) on the model's factor returns.
Table 2 shows Gibbons, Ross, and Shanken's (1989) GRS statistic, which tests
this hypothesis for variants of regression (3). The variants of (3) we examine
include the three-factor model of FF (1993), in which the explanatory returns
are Rm~Rf , SMB, and HML. Also shown are results for three four-factor
models that combine RM - RF and SMB with pairs of HML , RMW , and CM A,
and the five-factor model, which is the full version of (3).
We estimate all regression slopes as constants, so time variation in the slopes
is a potential problem. Like most of the asset-pricing literature, our models and
tests also assume there are no market frictions, for example, transactions costs
and taxes.

The GRS results are easily summarized. The test rejects the models we
consider and, except for one anomaly, the GRS p-v alues for the rejections round
to zero to at least three decimal places. Thus, all the models are incomplete
descriptions of expected returns. Asset pricing models, however, are simplified
propositions about expected returns that are rejected in tests with power. We
are less interested in whether competing models are rejected than in their
relative performance, which we judge using GRS and other statistics. We want
to identify the model that is the best (but imperfect) story for average returns.

75

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

Table 2
Summary statistics for tests of three-, four-, and five-factor models, July 1963-December 2014 (618
months)

Model factors GRS p(GRS ) A|a¡| òityl A(Ä2)

A|ri' Arf Aaj
25 Size-ß portfolios
Mkt 2.26 0.000 0.232 0.98 0.99 0.18 0.75
MktSMBHML 1.61 0.032 0.106 0.45 0.19 0.38 0.89
MktSMBHMLRMW 1.73 0.016 0.083 0.35 0.13 0.56 0.89
MktSMBHML CM A 1.51 0.055 0.095 0.40 0.17 0.44 0.89
MktSMBRMWCMA 1.68 0.021 0.069 0.29 0.10 0.81 0.89
Mkt SMB HML RMW CMA 1.68 0.021 0.072 0.31 0.10 0.76 0.89

35 Size-NI portfolios
MktSMBHML 4.30 0.000 0.136 0.51 0.39 0.18 0.87
MktSMBHMLRMW 3.74 0.000 0.111 0.41 0.24 0.29 0.88
Mkt SMB HML CMA 4.03 0.000 0.130 0.49 0.35 0.21 0.87
MktSMBRMWCMA 3.15 0.000 0.100 0.37 0.19 0.39 0.88
Mkt SMB HML RMW CMA 3.16 0.000 0.098 0.37 0.18 0.38 0.88

25 Size-Var portfolios
MktSMBHML 6.02 0.000 0.217 0.72 0.84 0.06 0.87
MktSMBHMLRMW 5.28 0.000 0.147 0.49 0.49 0.10 0.89
Mkt SMB HML CMA 6.08 0.000 0.213 0.71 0.81 0.07 0.87
MktSMBRMWCMA 5.05 0.000 0.130 0.43 0.36 0.14 0.88
Mkt SMB HML RMW CMA 5.03 0.000 0.131 0.44 0.36 0.14 0.89

25 Size-RVar portfolios
MktSMBHML 7.15 0.000 0.222 0.71 0.80 0.06 0.88
MktSMBHMLRMW 6.33 0.000 0.144 0.46 0.44 0.09 0.90
Mkt SMB HML CMA 7.22 0.000 0.222 0.70 0.77 0.06 0.88
MktSMBRMWCMA 5.95 0.000 0.118 0.37 0.32 0.14 0.89
Mkt SMB HML RMW CMA 5.94 0.000 0.120 0.38 0.32 0.13 0.90

25 Size-AC portfolios
MktSMBHML 3.68 0.000 0.113 0.48 0.26 0.27 0.91
MktSMBHMLRMW 4.57 0.000 0.143 0.61 0.40 0.17 0.91
Mkt SMB HML CMA 3.29 0.000 0.096 0.41 0.21 0.34 0.91
MktSMBRMWCMA 3.70 0.000 0.127 0.54 0.32 0.22 0.91
Mkt SMB HML RMW CMA 3.77 0.000 0.126 0.54 0.31 0.23 0.91

25 Size-Prior 2-12 portfolios

MktSMBHML 5.06 0.000 0.319 0.97 1.11 0.06 0.85
MktSMBHMLRMW 4.69 0.000 0.305 0.93 0.96 0.07 0.85
Mkt SMB HML CMA 4.74 0.000 0.297 0.90 0.97 0.07 0.85
MktSMBRMWCMA 4.25 0.000 0.280 0.85 0.78 0.09 0.84
Mkt SMB HML RMW CMA 4.24 0.000 0.272 0.83 0.74 0.09 0.85

Mkt SMB HML MOM 3.87 0.000 0.133 0.40 0.17 0.20 0.91
Mkt SMB HML RMW MOM 3.72 0.000 0.118 0.36 0.14 0.23 0.92
Mkt SMB HML CMA MOM 3.73 0.000 0.135 0.41 0.17 0.20 0.91
Mkt SMB RMW CMA MOM 3.56 0.000 0.119 0.36 0.15 0.24 0.92
Mkt SMB HML RMW CMA MOM 3.55 0.000 0.117 0.36 0.14 0.23 0.92

This table tests how well three-, four-, and five-factor models explain monthly
(beta) portfolios, the 35 Size-NI (net share issues) portfolios, the 25 Size-Var
25 Size-RVar (residual variance) portfolios, the 25 Size-AC (accruals) portfol
(momentum) portfolios. The table shows (1) the factors in each regression mod
whether the expected values of all 25 or 35 intercept estimates are zero, (3) p{G
statistic, (4) the average absolute value of the intercepts, A'a¡ |, (5) A'a¡ '/A'?i |,
the intercepts over the average absolute value of r, , which is the average exce
the average VW market portfolio excess return, (6) Aaj/Arf, the average squar
squared value of r¡, (7) As2(a¡)/Aaf, the average of the estimates of the varianc
the estimated intercepts over Ar J, and (8) AŘ2, the average value of the regres
freedom. Each sort uses all stocks with data for the two sort variables at the por
excess return on the VW market portfolio, R^ - Rp- The other factors are defin

76

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Using A to indicate an average value, the other statistics we use to evaluate

competing models include the average absolute intercept, A'a¡', and two ratios
that measure the dispersion of the intercepts (unexplained LHS average excess
returns) produced by a model relative to the dispersion of LHS average excess
returns. We require baselines or reference points to measure dispersion. Since
the asset-pricing hypothesis is that the true intercepts are zero, the appropriate
reference point for the intercepts is zero. What is the best reference point for the
dispersion of the LHS average excess returns? Our current answer is different
from that in FF (2015).
In FF (2015), the dispersion of the LHS average excess returns is measured
relative to the simple average of all LHS average excess returns. From an
asset pricing perspective, however, the average VW market excess return is a
better reference point for three reasons: (1) We take Merton's (1973) ICAPM
to be a central motivation for multifactor models. The VW market portfolio
is the centerpiece of the ICAPM: in the language of Fama (1996), the VW
market portfolio is multifactor efficient in all versions of the ICAPM. (2) More
simply, the VW market portfolio is an attractive reference point because it is
the aggregate of the portfolios chosen by investors. (3) Any model that includes
Rm - Rf as a factor perfectly (and trivially) explains the VW market portfolio
excess return. In contrast, the EW average of the LHS portfolio excess returns
has no special role in asset pricing. For example, as a LHS portfolio, its excess
return almost surely produces a nonzero intercept in an asset pricing regression,
and the intercept is different for different asset pricing models and different sets
of LHS portfolios.
Define as the difference between the time-series average excess return on
LHS portfolio i and the average excess return on the VW market. The first
measure of the relative dispersion of the intercepts is A'a¡ '/A'P¡ |, the average
absolute intercept divided by the average absolute value of . The second is
Aaf/Arf , the average squared intercept over Arf, the average squared value
of Pi .

In the end, the denominators of A'ai'/A'Pļ' and Aaf/Arf are just scaling
variables: for a given set of LHS portfolios, the denominators are the same for
all asset pricing models. They just give perspective on the dispersion of the
intercepts in the numerators of the ratios. Switching the reference point from
the average VW market return used here to the EW average of the LHS average
returns used in FF (2015) produces the same ordering of intercept dispersion
for different models.

Finally, in Fama and French (20 1 5), we show results for a variant of Aaf /Arf
that adjusts numerator and denominator for measurement error. In those tests,
adjusted ratios tend to be a bit smaller than unadjusted ratios. In ongoing
tests on international data, the double adjustment sometimes produces extreme
ratios, positive and negative. The sample period in the international tests is
much shorter, and we suspect the problem is measurement error in estimates

77

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies I v 29 n I 2016

of measurement error, which is especially troublesome in the denominator of

the adjusted ratio since it can lead to explosive ratios of either sign.
We do not show double-adjusted ratios here. Instead we show estimates of the
proportion of the dispersion of the intercept estimates attributable to sampling
error. Thus, the intercept estimate a, is the true intercept, plus an estimation
error, £/,
tf ;=(¥/+£/ (4)

Since ct' is a constant, the expecte

E(af)=af+E(sf). (5)
Averaging over the LHS assets, we have

AE(af) = Aaf + AE(ef). (6)

The expected value of et- is zero, so E (ef) is the variance of a¡ due to estim
error, which we estimate with the squared sample standard error of a,-,
The sample estimate of AE(af) is Aaf. The ratio As2 (ai) /Aaf is the
estimate of the proportion of the dispersion (second moment) of the int
estimates due to estimation error.

Note that low values of A'a¡'/A'Fi' and Aaf /Ar f are good news for an
asset pricing model: they say that intercept dispersion (the dispersion of LHS
average returns left unexplained by the model) is low relative to the dispersion
of the LHS average returns. In contrast, high values of As2(aļ)/Aaf are good
news: they say that much of the dispersion of the intercept estimates is due to
sampling error rather than to dispersion of the true intercepts.

2.2.1 Market ß. The GRS rejections of our asset pricing models are weakest
for the Size-ß portfolios. Since the ß anomaly is a purported violation of the
CAPM, we include the CAPM among the models tested. The CAPM is rejected
with a GRS /7-value that is zero to three decimal places. The ratios A 'a¿ '/A |řř- 1
and Aaf/Arf are 0.98 and 0.99, so the dispersion of CAPM intercepts almost
matches the dispersion of average LHS portfolio returns. And As2(a¿)/ Aaf
estimates that only about 18% of the dispersion of the intercepts is due to
sampling error. Similar negative results are observed in tests of the CAPM on
portfolios from the other anomaly sorts, and to save space we show no CAPM
results for other anomalies. We see later that the CAPM is rejected in the ß
sorts because the model predicts that the slope in the relation between average
excess return and ß is the average excess market return, but the actual relation
is essentially flat.
In earlier drafts of this paper, the FF (1993) three-factor model easily passes
the GRS test on the 25 Size-ß portfolios (GRS =1.07; /7-value = 0.371) and,
like Novy-Marx (2014), we conclude that returns on ß-sorted portfolios do not
identify problems for the three-factor model. Adding 2014 to the 1963-2013
sample of earlier drafts changes that inference; the three-factor GRS statistic

78

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

increases to 1.61 (Table 2) and the GRS p- value shrinks to 0.032. The GRS test
on the Size-ß portfolios also rejects our other models at conventional levels,
but the rejections are weaker than for other anomalies.
Adding 2014 to the 1963-2013 sample has little effect on other measures of
performance. Judged on anything but the GRS test, the best performers (in a
dead heat) in the tests on the Size-ß portfolios are the five-factor model and
the four-factor model that drops HML. The average absolute intercepts from
the two models are 0.072% and 0.069%, versus 0. 106% for the FF three-factor
model. The A'a¡'/A'Fi ' ratios are 0.31 and 0.29 for the five- and four-factor
models, so measured in units of return, the dispersion of unexplained average
returns is about 30% as large as the dispersion of average returns. In units
of return squared ( Aaf/Arf ) the dispersion of unexplained average returns is
about 10% as large as the dispersion of average returns. The ratios As2(ai)/Aaf
are 0.76 and 0.81, which suggests that more than three-quarters of the second
moments of the intercept estimates for the two models is due to sampling error
and only about one-fourth is due to dispersion in the true intercepts. All this is
consistent with relatively weak rejections on the GRS test.

2.2.2 Net share issues. All the asset pricing metrics in Table 2 agree that the
five-factor model and the four-factor model that drops HML provide the best
descriptions of average Size-NI portfolio returns. Thus, adding profitability and
investment factors enhances estimates of expected returns for portfolios formed
on Size and net issues. The average absolute intercepts produced by the two
models are 0.098% and 0.100% per month. The ratio A'ai'/A'?¡ ' is 0.37 for
both models, so in units of return, the dispersion of the estimated intercepts is
37% as large as the dispersion of average Size-NI portfolio returns. In units of
return squared, Aaf/Ařf is 0.18 and 0.19 for the two models, and the ratios
As2(ūi)/Aaf estimate that about 40% of Aaf is due to sampling error in the
intercept estimates. These results suggest that, despite rejection on the GRS
test, the two models perform well in the tests on the Size-NI portfolios.
When we later examine the intercepts produced by the five-factor model
(Section 4), we see that its problems are largely in portfolios of small stocks in
the highest NI quintile, which have negative exposures to RMW and CMA like
those of small firms that invest a lot despite low profitability. This is the lethal
combination that plagues the five-factor model in FF (2015).

2.2.3 Volatility. We also see later (Section V) that the same lethal
combination plays a big role in the rejection of the five-factor model in tests
on the Size-Var and Size-RVar (total and residual variance) portfolios. Again,
the GRS test and other summary statistics in Table 2 imply that the five-factor
model and the four-factor model that drops HML provide the best descriptions
of average Size-Var and Size-RVar portfolio returns. On all metrics, however,
the volatility portfolios pose stronger challenges to the two models than the
Size-NI portfolios. The average absolute intercepts and seven of the eight ratios

79

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

comparing the dispersion of the intercepts to the dispersion of the LHS average
returns are larger for the volatility portfolios than for the net issues portfolios,
and less of the dispersion of the intercepts for the volatility portfolios can be
attributed to sampling error.

2.2.4 Accruals. In the tests on the six different sets of LHS portfolios in FF
(2015), and in the tests on other LHS portfolios examined here, the five-factor
model typically performs better than the FF three-factor model. This is not true
for the 25 Size-AC portfolios. The culprit is the profitability factor RMW. The
three models that include RMW produce larger GRS statistics and are weaker
on other metrics than the two models that do not include RMW. In contrast,
models that include the investment factor CMA perform relatively well, except
when they include RMW. For the Size-AC portfolios, the four-factor model that
drops RMW delivers the best performance on all metrics. The performance of
this model in the tests for accruals is similar to that of the best models in the
tests for net issues and volatility. We see later that the poor performance of the
five-factor model in the accruals tests owes a lot to microcaps.

2.2.5 Momentum. Models that do not include MOM fail badly as descrip-
tions of average returns on the 25 Size-Prior 2-12 portfolios. For example, the
estimates of intercept dispersion relative to the dispersion of average excess
returns, A|a/ |/A|ř/ 1, range from 0.97 for the FF three-factor model to 0.83 for
the five-factor model. All are far above the values of this ratio in the sorts for
other anomaly variables.
When we include the momentum factor, MOM , all models are still rejected
on the GRS test, but explanatory power improves. The best models on the GRS
test are the six-factor model that adds MOM to the five-factor model and the
five-factor model that drops HML. The performance of these two models is
essentially identical on all metrics. The average absolute intercept is 0.117%
per month with HML and 0.119% without. In units of return, the dispersion
of the intercepts relative to the dispersion of Size-Prior 2-12 average returns,
A'ai'/A'?i', is 0.36 for both models, and in units of return squared, Aaf/Ařf ,
it is 0. 14 for one and 0. 15 for the other. These are strong numbers, but they are
achieved by adding a momentum factor constructed with a coarser version of the
sorts for the 25 Size-Prior 2-12 portfolios, a luxury not allowed in the tests for
other anomalies. Moreover, when MOM is among the factors, other models,
including Carharťs (1997) model, which adds MOM to the FF three-factor
model, perform almost as well as the six-factor model.

2.3 The five-factor model versus the FF three-factor model: A simple test
For almost all sorts examined here and in FF (2015), the five-factor model
performs better than the FF three-factor model. Are the differences statistically
reliable? If we assume expected returns are governed by a linear factor model
and some stocks have nonzero exposures to RMW and CMA , we can use the

80

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

GRS test to show formally that the profitability and investment factors add
information about expected returns to the three-factor model.
The GRS test on the intercepts from FF three-factor regressions to explain
RMW and CMA tells us whether adding RMW and CMA improves the mean-
variance efficient set produced by combining the risk-free rate, RM - RF, SMB ,
and HML. The regression estimates (t -statistics in parentheses) are

RMW t = 0.33 - 0.05(Rmí - Rfí) - 0.23SMBt +0.01 HMLt + et ,

(4.08H- 2.64) (-8.43) (0.33) fl2=0.14 (7)

CMA, = 0.20 - 0.09(RMt-RFi)+0mSMBl+0A5HML,+el.

(3.62X- 6.70) (0.52) (22.12) fl2=0.53 (8)
The intercepts in these regressions - 0.33 (ř=4.08) for RMW and 0.2
3.62) for CMA - are large, even by the standards suggested by Harve
and Zhu (2015). The GRS statistic (21.09; p-' alue zero to at least five d
places) confirms that RMW and CMA jointly add to the information
expected returns in Rm~Rf> SMB , and HML.
Factor redundancy tests like regressions (7) and (8) are definitive. If a fa
average return is captured by its exposures to the other factors in a mode
factor adds nothing to the model's explanation of average returns, and no
LHS portfolios can overturn this conclusion (Fama 1998; Barillas and Sh
2015). Conversely, if a factor's average return is not captured by its ex
to the other factors in a model, that factor has a role in explaining a
returns in the model. This does not mean this factor is important for exp
average returns for all sets of LHS portfolios. For example, exposure
important factor may be negligible in a particular LHS sort, in which c
factor may not help explain average returns in that sort.

2.4 An equivalent five-factor model

In the summary tests of Table 2, the five-factor model and the four-facto
that drops HML perform almost identically on all metrics. This result is
with the evidence in FF (2015) that HML is redundant for describing
average returns, at least for 1963-2014. Specifically, the large averag
return (0.36% per month; ř = 3. 15 in Table 1) is absorbed by the exp
of HML to other factors, especially the profitability and investment f
RMW and CMA. For the time period and version of the factors used h
regression to explain HML (t -statistics in parentheses) is

HMLt= -0.04 + 0m(RMt-RFt)+QMSMBt+0.22RMWt + 'MCMAt+et.

(-0.47) (0.31) (0.88) (5.37) (23.24) tf2=0.51
(9)

In contrast, Rm - Rf , SMB , RMW , and CMA have substantial marginal

information about average returns. Skipping the details, the intercepts in the

81

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 The Review of Financial Studies / v 29 n J 2016

regressions to explain each of these factor returns with the other four are 0.81
(ř = 5.00) for Rm-Rf , 0.36 (f = 3.09) for SMB , 0.42 (i = 5.33) for RMW, and
0.27 (ř=4.98) for CM A.
The trivial intercept in (9) implies that nothing is lost in the explanation
of average returns if we drop HML from the five-factor model. Exposures to
HML are, however, important for understanding the portfolio types that cause
asset-pricing problems. We want to keep HML , but we also want other factors
to have slopes that reflect the fact that, at least in U.S. data for 1963-2014,
the four-factor model that drops HML captures average stock returns as well
as the five-factor model. A twist on the five-factor model meets these goals.
Define HMLO (orthogonal HML) as the sum of the intercept and residual from
(9). Substituting HMLO for HML in (3) produces an alternative five-factor
regression:

Rit - Rfí ~&i ~ł~bi(RMt - RFt)+SiSMBt +h¡HMLOt +CļCMAļ +£//.

(10)
The intercept and residual in (10) are the same as in the five-factor regression
(3), so the two regressions are equivalent forjudging model performance. For
example, the GRS test and other results for the five-factor model in Table 2 do
not change if we use (10) rather than (3). The HMLO slope in (10) is also the
same as the HML slope in (3), so (10) produces the same estimate of the value
tilt of the LHS portfolio. But the estimated mean of HMLO (the intercept in the
HML regression (9)) is near zero (-0.04; t = - 0.47), so its slope adds little to
the estimate of the expected LHS return from (10). The slopes on other factors
in ( 10) are the same as in the four-factor model that drops HML , so other factors
have slopes that reflect the fact that they capture the information in HML about
average returns.
For more insight into model performance, we next examine the asset pricing
regressions for the anomaly portfolios in more detail. For each anomaly, we first
document the patterns in average returns we seek to explain. We then examine
intercepts and pertinent slopes from asset pricing regressions.

3. Market 0
Panel A of Table 3 shows average monthly excess returns (returns in excess
of the one-month U.S. Treasury-bill rate) on the 25 VW Size-ß portfolios. The
results confirm previous evidence that there is no clear relation between ß and
average return. For example, the portfolio in the highest ß quintile of a Size
quintile tends to have a slightly higher average return than the portfolio in the
lowest ß quintile. But the portfolio in the highest ß quintile also has a lower
average return than portfolios in the middle three ß quintiles of a Size quintile,
which have similar average returns. There is, however, a size effect in every
ß quintile: given ß, average return is highest for microcaps and lowest for
megacaps.

82

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table 3
Average excess returns and characteristics of stocks in the 25 Size-ß portfolios, July 1963-December 2014
(618 months)

Low /5 2 3 4 High ß Low ß 2 3 4 High ß

Panel A: Average excess returns and SD
Mean SD
Small 0.73 0.90 0.92 0.98 0.79 4.41 5.10 5.95 6.51 8.23
2 0.72 0.86 0.95 0.89 0.72 4.30 4.77 5.48 6.21 7.92
3 0.69 0.86 0.84 0.80 0.76 3.85 4.65 5.22 5.96 7.69
4 0.67 0.76 0.74 0.58 0.75 3.89 4.63 5.17 5.84 7.57
Big 0.49 0.52 0.49 0.50 0.41 3.63 4.25 4.88 5.69 7.15
Panel B: Average B/M, OP, Inv , and prior ß characteristics
B/M OP
Small 0.99 1.04 1.04 1.04 0.94 0.15 0.14 0.16 0.37 0.02
2 0.88 0.87 0.84 0.82 0.76 0.26 0.25 0.47 0.27 0.23
3 0.85 0.79 0.75 0.72 0.68 0.26 0.30 0.31 0.31 0.25
4 0.80 0.72 0.69 0.69 0.66 0.29 0.31 0.30 0.29 0.31
Big 0.65 0.51 0.57 0.56 0.63 0.35 0.40 0.34 0.37 0.36
Inv Prior ß
Small 0.10 0.09 0.10 0.11 0.14 0.31 0.84 1.15 1.51 2.49
2 0.12 0.12 0.13 0.15 0.20 0.38 0.84 1.16 1.51 2.40
3 0.12 0.12 0.13 0.15 0.23 0.38 0.84 1.15 1.50 2.32
4 0.10 0.12 0.13 0.14 0.21 0.40 0.84 1.15 1.50 2.24
Big 0.10 0.12 0.14 0.15 0.21 0.42 0.83 1.15 1.48 2.10

At the end of June from 1963 to 2014, we form value- weight portfolios usi
and (beginning in 1973) NASDAQ stocks into Size (market capitalization)
beta), with NYSE breakpoints for both variables. The intersections of the
For portfolios formed in June of year t , Size is market capitalization at
estimated by regressing a stock's monthly return on the current market
minimum 24) months of returns preceding June of t. Panel A shows mea
excess returns on the 25 portfolios. Panel B shows time-series means of
ratio (B/M), operating profitability (OP), and investment (Inv) for the f
preceding portfolio formation. For portfolios formed at the end of June
are value-weight averages (market-cap weights) of the variables for the f
is the value-weight average ratio of book equity at the fiscal year-end in
the end of December of / - 1 ; OP is the value-weight average ratio of ope
fiscal year ending in / - 1 ; and Inv is the value-weight average rate of gr
ending in r - 1 . Panel B also shows the ß estimates used to form portfolio
across the stocks in a portfolio and then averaged across years.

Table 4 shows intercepts and slopes for the 25 Si

by the CAPM and the five-factor model (10). The
A) show that sorts on prior ß estimates produce l
using post-sort returns. Since average returns do n
with ß , it is not surprising that the CAPM intercep
low ß portfolios. Megacaps aside, most of the CAP
lower quintiles of ß are reliably positive and those
near zero. Megacaps produce the smallest intercept
only reliably negative CAPM intercept, -0.31% (t
in the highest ß quintile.
The five-factor model cures the systematic prob
tests on the 25 Size-ß portfolios. The strong positi
four lower Size and ß quintiles disappear in the fiv
Table 4). The negative CAPM intercept for the mega

83

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

Table 4
Regressions for the 25 Size-ß portfolios, July 1963-December 2014 (618 months)

Low ß 2 3 4 High ß Low ß 2 3 4 High ß

Panel A: CAPM: R¡t -Rpt = aļ + -Rpt) + e
a t(a )
Small 0.34 0.44 0.39 0.36 0.04 2.96 3.45 2.60 2.45 0.22
2 0.33 0.40 0.41 0.28 -0.07 3.11 3.82 3.68 2.23 -0.42
3 0.32 0.39 0.31 0.19 -0.02 3.70 4.66 3.33 1.80 -0.13
4 0.30 0.28 0.19 -0.04 -0.03 3.39 3.74 2.50 -0.41 -0.19
Big 0.15 0.07 -0.03 -0.11 -0.31 1.77 1.03 -0.50 -1.42 -2.29
b m
Small 0.75 0.90 1.04 1.21 1.48 29.13 31.51 31.18 36.27 33.20
2 0.77 0.90 1.07 1.21 1.55 32.61 38.58 43.25 43.53 44.32
3 0.71 0.93 1.05 1.20 1.54 36.51 49.90 50.34 50.08 49.26
4 0.72 0.95 1.08 1.22 1.53 36.46 57.55 63.68 61.60 51.71
Big 0.67 0.89 1.04 1.20 1.42 35.21 61.82 72.92 68.40 46.97
Panel B: Five-factor: R¿t -Rpt = ai + b¡(R'jt -Rpt) + SļSMBt + hļHMLOt +
a t(a )
Small 0.07 0.06 0.04 0.08 -0.03 1.07 0.98 0.63 1.27 -0.31
2 0.04 0.02 0.05 -0.06 -0.15 0.57 0.31 0.90 -0.99 -2.03
3 0.09 0.10 0.00 -0.11 -0.02 1.16 1.70 0.07 -1.55 -0.25
4 0.07 -0.02 -0.09 -0.24 0.08 0.84 -0.24 -1.37 -2.98 0.72
Big -0.07 -0.07 -0.09 -0.09 -0.06 -1.08 -1.38 -1.52 -1.14 -0.45
b m
Small 0.66 0.81 0.89 1.02 1.14 40.86 51.46 56.20 69.41 56.64
2 0.73 0.86 0.99 1.10 1.29 40.23 63.70 69.69 75.42 70.14
3 0.75 0.92 1.01 1.13 1.31 40.85 63.62 66.61 65.78 59.72
4 0.81 1.01 1.10 1.19 1.35 40.68 65.09 69.59 62.12 50.15
Big 0.83 0.98 1.08 1.18 1.30 50.97 75.20 74.85 59.75 39.24
s t(s)
Small 0.77 0.93 1.12 1.14 1.40 34.05 42.18 50.18 54.95 49.30
2 0.60 0.74 0.81 0.95 1.11 23.81 39.13 40.50 46.68 42.85
3 0.27 0.50 0.61 0.72 0.87 10.56 24.94 28.51 29.87 28.29
4 0.04 0.24 0.32 0.38 0.56 1.31 10.88 14.37 14.13 14.88
Big -0.28 -0.17 -0.11 0.06 0.06 -12.27 -9.32 -5.35 2.06 1.25
h m
Small 0.30 0.26 0.22 0.22 -0.09 9.35 8.32 7.18 7.72 -2.38
2 0.33 0.30 0.24 0.17 -0.06 9.37 11.42 8.76 5.99 -1.67
3 0.38 0.19 0.21 0.21 -0.12 10.71 6.80 6.90 6.30 -2.71
4 0.33 0.21 0.13 0.20 -0.09 8.38 6.76 4.15 5.38 -1.63
Big 0.10 0.04 0.06 0.09 -0.10 3.03 1.71 2.07 2.41 -1.57
r tir)
Small 0.05 0.18 0.03 0.01 -0.47 1.49 5.73 0.83 0.44 -11.67
2 0.13 0.32 0.30 0.21 -0.22 3.59 11.71 10.33 7.12 -5.80
3 0.13 0.25 0.29 0.28 -0.17 3.60 8.51 9.57 8.21 -3.95
4 0.17 0.33 0.35 0.23 -0.33 4.26 10.59 10.89 6.05 -6.21
Big 0.31 0.31 0.26 0.11 -0.26 9.41 11.81 8.98 2.77 -3.96
c t(c)
Small 0.33 0.42 0.38 0.23 -0.08 9.32 12.39 11.05 7.28 -1.90
2 0.38 0.42 0.33 0.29 -0.10 9.60 14.25 10.65 9.12 -2.52
3 0.40 0.33 0.30 0.25 -0.24 10.20 10.66 9.08 6.66 -4.98
4 0.45 0.39 0.30 0.17 -0.26 10.52 11.70 8.84 4.16 -4.40
Big 0.44 0.19 0.01 -0.14 -0.44 12.60 6.77 0.17 -3.33 -6.16

The LHS variables in each set of 25 regressions are the monthly excess returns
RHS variables are the excess market return, Rf^ - Rp, the Size factor, SMB, th
the profitability factor, RMW, and the investment factor, CM A. The table show
(panel B) intercepts and regressions slopes.

84

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

ß quintile also becomes inconsequential (-0.06; t = -0.45). The only blemish

on the five-factor model is the intercept for the intersection of the fourth Size
and fourth ß quintiles, -0.24 (/ = -2.98).
The improvements in the description of average returns on the Size-ß
portfolios provided by the five-factor model trace to patterns in the five-factor
regression slopes that absorb the patterns in average returns. Panel B of Table 4
shows that portfolios in the four lower ß quintiles tilt toward value (positive
HMLO slopes) and portfolios in the highest ß quintile have a growth tilt
(negative HMLO slopes), but since the average HMLO return is close to zero
these tilts have little impact on five-factor intercepts. The heavy lifting is done
by the RMW and CMA slopes. Microcaps aside, the five-factor RMW slopes
are strongly positive in the four lower ß quintiles. The RMW slopes become
strongly negative in the highest ß quintile, especially for microcaps. In all Size
quintiles the CMA slopes are positive in the lower ß quintiles but turn strongly
negative in the highest ß quintile. In short, the returns on low ß stocks behave
like those of profitable firms that invest conservatively, whereas the returns on
high ß stocks track those of less profitable firms that invest a lot.
The RMW and CMA slopes of the five-factor model increase the predicted
returns on the low ß portfolios of the Size-ß sorts and reduce the predicted
returns on the high ß portfolios. But the low and high ß portfolios have similar
average returns, so five-factor intercepts close to zero imply that the slopes for
the market and/or SMB lean against the RMW and CMA slopes. Panel B of
Table 4 shows that the spreads in the five-factor market slopes for the lowest
and highest ß portfolios of given Size quintiles are large, from 0.47 to 0.56. The
average market premium for 1963-2014 is 0.51% per month, so the spreads in
average returns predicted by the spreads in Rm-Rf slopes are also large, 0.24%
to 0.29% per month. Surprisingly, in every Size quintile SMB slopes increase
monotonically from low ß to high ß quintiles, and the spreads are again large,
from 0.34 for megacaps to 0.63 for microcaps. The average SMB return for
1963-2014 is 0.27% per month, so the spreads in average returns predicted
by the spreads in the SMB slopes range from 0.09% (megacaps) to 0.17% per
month (microcaps). In short, the five-factor market and SMB slopes, and the
associated premiums, offset RMW and CMA slopes and premiums to capture
average returns that show little tendency to increase with CAPM ß.
The results suggest that average returns vary with multivariate ß (market
slope b) in the way predicted by the five-factor model, even though there is little
relation between CAPM ß and average returns. For perspective, we estimate a
four-factor model that drops the market factor from the five-factor model (3)
and uses returns on the 25 Size-ß portfolios measured in excess of the market
return as LHS variables. In other words, we set all multivariate ßs equal to
1.0. Skipping the details, the intercepts from this model are negative for the
five portfolios in the lowest ß quintile, and three of five are more than two
standard errors below zero. The intercepts for the five portfolios in the highest
ß quintile are positive, and three of five are more than two standard errors

85

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 The Review of Financial Studies / v 29 n 1 2016

above zero. Thus, setting multivariate ßs equal to 1.0 produces forecasts of

average returns that are too high for low ß portfolios and too low for high ß
portfolios. We conclude that there is a positive relation between multivariate
ß and average returns, and the average premium for multivariate ß conforms
well to the five-factor model.

Since lots of what is common in the story for average returns for different
sets of LHS anomaly portfolios centers on the slopes for RMW and CM A, an
interesting question is whether the slopes line up with profitability (OP) and
investment ( Inv ) characteristics. Like the CMA slopes of the Size-ß portfolios,
average investment increases from lower to higher ß quintiles (Table 3). But
contradicting the RMW slopes, profitability (OP) is not systematically lower
for high ß portfolios, except perhaps for microcaps. This is not surprising.
Multivariate regression slopes estimate marginal effects, holding constant
other explanatory variables, so the slopes need not line up with univariate
characteristics.

Since characteristics do not always line up with regression slopes, we are

careful when describing the slopes. For example, we say that strong negative
RMW and CMA slopes for the portfolios in the highest ß quintile imply that
returns on these stocks "behave like" those of unprofitable firms that invest
aggressively. Table 3 shows these firms have grown rapidly, but except for
microcaps, they have not been a lot less profitable than lower ß portfolios of
the same Size quintile.

4. Net Share Issues

Panel A of Table 5 shows average excess returns for the 35 VW portfolios from
independent sorts of stocks into Size quintiles and seven net share issues (NI)
groups. Repurchases (negative NI) are associated with higher average returns.
In all Size groups average returns are similar for the lowest three quintiles of
positive M, but average returns are lower in the fourth quintile. The striking
result, and the result that will be difficult to explain fully, is the extreme low
average returns of the five portfolios in the highest NI quintile (largest net
issues). Though not our main interest, there is a Size effect in every NI group:
microcaps have higher average returns than megacaps.
The summary tests in Table 2 say that the five-factor model improves the
description of average returns on the Size-NI portfolios provided by the FF
three-factor model. Table 6 shows the three-factor and five-factor intercepts
and the five-factor HMLO , RMW , and CMA slopes. We do not show Rm - Rf
and SMB slopes since they are similar for different models and so cannot explain
the intercept improvements produced by the five-factor model.
In previous research, repurchases are associated with positive unexplained
average returns. The three-factor intercepts for the repurchase portfolios are
positive, 0.11% to 0.24% per month, and 1.96 to 3.62 standard errors from
zero. The intercepts are smaller in the five-factor model, and the largest, 0. 10%

86

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table 5
Average excess returns and characteristics of stocks in the 35 Size-NI portfolios, July 1963-December
2014 (618 months)

NI -► Neg Zero Low 2 3 4 High Neg Zero Low 2 3 4 High

Panel A: Average excess returns and SD
Mean SD
Small 1.05 0.78 0.89 0.94 1.02 0.71 0.24 5.67 5.72 6.36 6.42 6.85 7.12 7.73
2 0.92 0.82 0.92 0.90 0.87 0.80 0.32 5.31 5.66 5.84 6.13 6.31 6.52 7.09
3 0.97 0.76 0.84 0.93 0.82 0.72 0.29 5.09 5.46 5.26 5.64 5.84 6.00 6.53
4 0.94 0.57 0.65 0.70 0.81 0.61 0.35 4.91 4.90 5.07 5.32 5.45 5.79 6.17
Big 0.62 0.61 0.49 0.47 0.59 0.43 0.18 4.18 4.78 4.25 4.54 4.85 5.41 5.12
Panel B: Average B/M, OP , Inv, and NI characteristics
B/M OP
Small 1.05 1.17 1.03 0.93 0.86 0.75 0.68 0.31 0.21 0.18 0.22 0.23 0.06 -0.10
2 0.86 1.00 0.86 0.79 0.73 0.67 0.63 0.27 0.31 0.28 0.30 0.54 0.23 0.18
3 0.78 0.92 0.84 0.73 0.68 0.65 0.64 0.31 0.30 0.27 0.26 0.30 0.23 0.26
4 0.69 0.90 0.79 0.68 0.67 0.65 0.68 0.34 0.27 0.28 0.28 0.32 0.31 0.24
Big 0.59 0.81 0.59 0.52 0.53 0.61 0.71 0.37 0.32 0.36 0.35 0.46 0.32 0.30
Inv NI
Small 0.06 0.06 0.07 0.08 0.10 0.14 0.59 -4.99 0.00 0.14 0.52 1.28 3.85 36.60
2 0.08 0.11 0.10 0.11 0.13 0.18 0.58 -4.58 0.00 0.14 0.52 1.28 3.89 34.34
3 0.08 0.11 0.09 0.11 0.14 0.19 0.53 -3.85 0.00 0.14 0.52 1.29 3.94 28.96
4 0.07 0.07 0.09 0.11 0.13 0.18 0.45 -3.34 0.00 0.14 0.52 1.28 3.96 24.82
Big 0.08 0.10 0.09 0.12 0.13 0.18 0.41 -2.00 0.00 0.13 0.51 1.28 3.84 22.65

At the end of June each year from 1963 to 2014, we form value- weight (VW) portfolios using independent so
of NYSE, AMEX, and (beginning in 1973) NASDAQ stocks into Size (market capitalization) quintiles and i
seven NI (net share issues) groups, including stocks with negative NI (repurchases), zero NI, and quintiles of
positive NI (net issues), using NYSE breakpoints for both variables. The intersections of the two sorts prod
35 Size-NI portfolios. For portfolios formed in June of year t , Size is market capitalization at the end of June an
NI is the change in the natural log of split-adjusted shares outstanding from the fiscal year-end in / - 2 to the fis
year-end in / - 1 . Panel A shows means and standard deviations of monthly excess returns on the 35 portfol
Panel B shows time-series means of the portfolio book-to-market equity ratio (B/M), operating profitabilit
(OP), and investment (Inv) for the fiscal year ending in the calendar year preceding portfolio formation, a
defined in Table 3. Panel B also shows the time-series average values of NI used to form portfolios each yea

per month, is only 1.73 standard errors from zero. The intercept improvement
produced by the five-factor model center on the RMW and CM A slopes. Th
repurchase portfolios have strong positive exposures to CM A, RMW , and
megacaps aside, HMLO. In other words, their returns covary positively with
returns of value stocks and stocks of profitable, low investment firms. Posit
exposures to RMW and CM A increase five-factor estimates of expected retur
and lead to negligible intercepts. In short, the repurchase anomaly disappea
in the five-factor model.

There are no serious problems in the three-factor and five-factor intercepts

for portfolios with zero NI and in the two lowest quintiles of positive NI. Three
portfolios in the third quintile of NI have three-factor and five-factor intercepts
near or more than 2.0 standard errors above zero. These unexplained average
returns are positive, but the net issues anomaly is about the low average stock
returns of firms that issue stock. Chance is a possible explanation.
The net issues anomaly is strong in the three-factor regressions for portfolios
in the highest NI quintile, with negative intercepts from -0.28% per month
(t = - 3.03) for the portfolio in the fourth Size quintile to -0.57% (t = - 6.20)

87

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 The Review of Financial Studies / v 29 n 1 2016

a w
.c o o m Tt so-^-vomm io no io rf ^ vo o *o on oo r- on n td
ņjņ (N r-; 'n o fN fļ rļ in - < (N Tt-^- oņ in - • o Is; >o oņ n ťNvqinr^vq «o
S no iri rfr ri rť Tt" en - Ö ri vi ri o - oo iri - ri ri ri 0' ^ E

I ■?
't h in Tt h m omo « o m - ■ r-- r> m - mvor^vo Tt rj - 00 °o îj
00 »0 - < « - >n r~; (N r~; Tt on o »n o o a ri 'o o h oo vo cļ vq uT«-
- O O O ¿ Ö ; ri o " " rn rn no iri ri OŇ OŃ ri tj-' rt o' O ■£
I I lili I I I I I I I I I I t> [Ł
<2 E
u
• üí -C
Co ~
(N m r- 00 - on m ri n m h « 00 *0 00 n 00 m 00 m r- m -<t r- e
«npvor^OŅ - - *0 - ri 00 r-^ - ; r-; r-; -• 1^ 00 os (S *0 r| - ; - ¿20
ri - - ri - no- ri ri - 0000 - o - O - - ö ri -*t ~ ¿2
I I I I I I I I I
CK H,
iIo
iI

«I

'Jņoop^rnr-
^ O - ri Ö Ö ^00-00 ^ ri ri Tf ri o ^ ^ rň iri ri ^ K K iri vo ri _- c
II I I g c o
3 '53
2Ž g
oj 00
n^tnvooN t ON o' n ON 10 10 M» h ON ^ r- - mr-^oo .*22
g O (N h 00 ON ^ O ri 10 on 00 vO OO tí O 00 rn (N 00 Tt r- Nû On Tt Jg
S O - O ¿ II
O öööriö iri h O NÒ
lili -
O - oo oo K oo oo O - iri Cj- ®
« >
<u >
O o
X JZ
<D CÄ

rtnrļOO
2on rtnrļOO no
nú m on no no io - o> -Tto
m io > Tto^- Tt
n oo - h
^ Tt ro
m o r- oo
r- -no
io h-ro; -- oo
rir^r^r^-
Oņ Oņ no -oo ri- ji:
; - ri no <n ^ - oo oo n ri ji: <n
Oņ Oņ
Nv 0 0 0 0« oòd-ò iri NO iri - - ri no - ri K iri no K K t"! *ā>
11 I I I I - - - no t"! g *ā> § c/5 Ò-*

- -t
- 'S -t §
ftr,
ftr, ®*
®* r-
tj-no
oņ(Nno
- >Tf
- n ^ on
r-; t - ono
io n
ooooo onoo
- - no- on m -iom
o (N onio- no
o mo no r> m
io no m <n
- E.2Qj
Z ri - ri ri ri _ - : ö - - Ö oŇ OŇ K d o! ^ -t' ri oo rf' no - Tf ri > lì
S I

+ ssr
^ až <*- «
S
2
_£
r~
e
Ł)
o
c
G 2 r~ H Ł) s c Ē
¡$ « o
^ .2 u tí
j3 r- n r- oo no + no -«t ~ - oo <n n m no ri « - io h m n Tf no m o õ c 5
OJD lOTtmnm *- n <n - o - ri - o O co notJ-^io- n 4 io ^ n •+- - •
>5 ® dodoo^ o o o o o d o d o o o o o o o o o o o o fe y >
5 ® I I I I ! ^ III I III I I I I I I I I I I EL'S <c >
s
^ "O <D
e
e ř+ e ^ s sT ^ f
4" ^ sT
Tt - - « - ^ 5 (N on rt o m n no - (N «o r- o io o - no r- o r> .52 ^ o
C
00
jr - o o o o ^ o - o - n o o o - - ■ ri - - m rt - - - <n ^ u ^ ¿
s S o o o o o >5 o o o o o o o o o o o o o o o o o o o o u iaScA
S o i o o o o g o i o o o o o i o i o o i o i o i o i o i o i o i o t o ! o
o
^
>St
S ^1.
~ W
2 O
3ai
ba
+ -S S<2 n;
V *» oo r- n o + n io - m no vonornr» rN no m r- - ■ no io - oo no
Xi oq T- o - n - (no- 'n- o o o o o noooo o o o o - <.t:n
ododo fiq dddod ddddo ddddo dddod

0
¡ Ř
««s.
+
s fiq
S £
^ s
è
© ^ + u ex |5<
E no r~- r- (N io ^ ri - nt^n n r- -*t n r> Ti- - on on oo n oo o £ u ?
n ^ E a q no P r~- - r- O p io ^ g. PP-PP nt^n -P--P r- -*t r> ^ p Ti-
1 ooooo ¿7 ooooo ooooo ^ ooooo ooooo S rC
I ooooo , ooooo 1 1 ooooo 1 ooooo ooooo i 2 o rC 2
s S
BÍ
X
5 í &5
5 í
Lì
3 Š
^
ÏC
-a
i. ^ o on io no > w m - n on r- - io oo n o r- on On oo on m on - ■ r- on ^ - c
i i. > ^ + j' ooooo .¿r o o o - o <n n m n O o rj n n n m m Tt Tt - ^ Ç d
J ^ + ooooo , ooooo ooooo ooooo ooooo «LiS
6
z
J B ooooo I I t , ooooo I I I I ooooo ooooo ooooo « §■§.
II Q ^ G 2
Ł
s
iv 'L
g; 2§S
±j o ±j
"" c
s I o O o o
2, I - "5(N
o o O - o o
(Nmroo-
O o m ^
- <n nmmm-^ioio
o o - ed
o c - io
,Ä
N N .t: Iooooo
ft¡ i ooooo ooooo ooooo ooooo cg.2
I - lili
il0303 ..a -Sou
$
I
I y
M ,2 - - ^ n io 'S o rņ oo no - NO«onno r- r- on r4 ý no h on n «Í2S
sis
M ®®T n - n ri - es - o o o o n n n n o n n ^ n n 4 ^ n rt m i: - >
«S Z ģ ooooo ^ ooooo ooooo ooooo ooooo ■§ g 2
h gř i 'S f M
JO
g H E 'S N® f
V© ™
a> 2 í= « _ _ _ Ï*1
S g» ■111 M § 1 .Sf _ 1 .Sf _ 1 ,2f _ ! .S S f-o
0u ç/JíSf^^-eQ fiu c/3<s«*>TtPQ cßisro^tPQ CÄr^f^-^fiQ c«<Sf0^eQ h<*Hcj

88

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor Model

for microcaps. The five-factor intercepts for these portfolios are less extreme
due to negative RMW and CMA slopes that lower five-factor estimates of
expected returns. But the net issues anomaly survives in the five-factor model:
the intercepts for four of the five portfolios in the highest NI quintile are negative
and three are more than 2.2 standard errors below zero.

The unexplained average returns associated with large net issues have lots in
common with the five-factor asset pricing problems in the sorts on Size , B/M ,
OP , and Inv in FF (2015). The portfolios in the highest M quintile have negative
RMW and CMA slopes, so their returns behave like those of the stocks of firms
with low profitability and high investment. Small stocks with this combination
of RMW and CMA exposures are the major problem for the five-factor model
in many LHS sorts in FF (2015). But the highest NI megacap portfolio also
has a negative five-factor intercept, -0.18% (t = - 2.23), and high investment
despite low profitability is not a problem among large stocks in FF (2015).
The RMW and CMA slopes for the Size-NI portfolios in Table 6 line up with
their average profitability and investment characteristics, OP and Inv , in Table 5.
Firms that repurchase are on average more profitable than firms that make large
share issues (Table 5), but the decline in RMW slopes from repurchasers to
extreme issuers is sharper (Table 6). There is stronger correspondence between
CMA slopes and Inv. Firms that repurchase on average have the lowest rates of
investment, which is in line with strong positive CMA slopes, and large share
issues signal high rates of investment matched by strong negative CMA slopes.
The jumps in NI and Inv from the fourth to the fifth quintile of NI are
impressive. Net issues average less than 4% of stock outstanding in the fourth
NI quintile, rising to 22.65% (megacaps) to 36.60% (microcaps) in the fifth
quintile. Investment is 14% to 19% of assets in the fourth NI quintile, rising
to 41% (megacaps) to 59% (microcaps) in the fifth. The extreme rates of
investment and net issues in the fifth NI quintile suggest that lots of these
firms do mergers financed with stock, a combination known to be associated
with low stock returns (Loughran and Vijh 1997). The overlap among new
issues, mergers financed with stock, and the combination of low profitability
and high investment that is a five-factor asset-pricing problem here and in FF
(2015) is an interesting topic for future research.

5. Volatility

Table 7 shows summary statistics for the 25 VW Size-RVar (residual variance)

portfolios. Table 8 shows intercepts and slopes for the portfolios from the
five-factor regression (10), along with the intercepts from the FF three-factor
model. The corresponding results for the 25 Size -Var (total variance) portfolios
are similar and are in Tables Al and A2 of the Appendix.
For megacaps there is no relation between average return and RVar (Table 7).
For microcaps, the portfolios in the two highest RVar quintiles have lower
average returns and the average excess return of the portfolio in the highest

89

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

Table 7
Average excess returns and characteristics of stocks in the 25 Size-RVar portfolios, July 1963-December
2014 (618 months)

RVar Low 2 3 4 High Low 2 3 4 High

Panel A: Means and SD of portfolio excess returns
Mean SD
Small 1.01 1.17 1.08 0.81 -0.20 4.18 5.64 6.48 7.52 8.99
2 0.92 1.03 1.07 0.97 0.18 4.12 5.30 5.96 6.82 8.45
3 0.75 0.91 0.93 0.93 0.36 3.79 4.86 5.42 6.20 7.79
4 0.72 0.75 0.75 0.78 0.44 3.84 4.53 5.17 5.72 7.41
Big 0.47 0.52 0.51 0.48 0.45 3.65 4.15 4.55 5.06 6.40
Panel B: Average B/M, OP , Inv, and RVar characteristics
B/M OP
Small 1.01 0.96 0.93 0.90 0.94 0.24 0.28 0.33 0.28 -0.09
2 0.89 0.83 0.80 0.77 0.73 0.34 0.31 0.30 0.28 0.18
3 0.85 0.77 0.73 0.71 0.68 0.28 0.30 0.32 0.31 0.27
4 0.82 0.73 0.69 0.68 0.64 0.29 0.32 0.35 0.31 0.30
Big 0.60 0.58 0.58 0.59 0.55 0.35 0.36 0.36 0.35 0.40
Inv RVar
Small 0.11 0.14 0.17 0.21 0.22 2.05 4.64 7.54 12.94 41.16
2 0.10 0.13 0.16 0.21 0.30 1.41 2.90 4.40 6.66 16.81
3 0.10 0.12 0.14 0.18 0.29 1.14 2.23 3.35 5.03 12.50
4 0.09 0.11 0.13 0.15 0.25 1.03 1.84 2.68 3.95 9.70
Big 0.10 0.11 0.12 0.15 0.21 0.83 1.40 1.94 2.76 5.85

Panel A shows means and standard deviations of monthly excess returns

monthly using a first pass sort of NYSE, AMEX, and (beginning in 1
capitalization) quintiles and second-pass sorts into quintiles of RVar (res
for both variables. The RVar sorts are conditional on Size quintile. The in
Size-RVar portfolios. For portfolios formed at the beginning of month t,
beginning of t and RVar is the variance of its daily residuals from the FF
(with a minimum 20) days of lagged returns. Panel B shows time-series m
equity ratio (B/M), operating profitability (OP), and investment (Inv) for
year preceding portfolio formation, as defined in Table 3. Panel B also s
RVar used to form portfolios each month.

RVar quintile is -0.20% per month. For the middle

is no clear relation between average return and v
volatility quintiles, but the portfolios in the high
much lower average returns.
The summary tests in Table 2 say that the five-
better description of average returns on the Size-R
three-factor model. Table 8 shows a clear pattern in t
In every Size quintile, the portfolios in the lowes
positive three-factor intercepts and the portfolios i
have negative intercepts. The pattern is weak for m
stronger for smaller Size quintiles. For microcaps,
for the portfolio in the lowest RVar quintile is 0
and the intercept for the portfolio in the highes
(ř = - 7.66).
Problems remain, but the five-factor model shrinks the troublesome
intercepts of the three-factor model. Four of the five three-factor intercepts
for portfolios in the lowest RVar quintile are more than two standard errors

90

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table 8
Regressions for the 25 Size-RVar portfolios, July 1963 to December 2014 (618 months)

RVar -* Low 2 3 4 High Low 2 3 4 High

Panel A: Three-factor: Rn-Rpt = a,- + -Rfí) + SļSMBt + hļHMLi + e¡t
a t(a )
Small 0.34 0.31 0.13 -0.19 -1.23 5.20 4.33 1.75 -1.86 -7.66
2 0.26 0.20 0.20 0.04 -0.72 4.16 3.04 2.77 0.56 -7.16
3 0.15 0.17 0.13 0.08 -0.43 2.32 2.53 1.82 1.06 -4.58
4 0.15 0.10 0.03 0.05 -0.29 2.02 1.45 0.45 0.65 -2.87
Big 0.08 0.11 0.01 -0.06 -0.08 1.33 2.03 0.18 -1.02 -0.86
Panel B: Five-factor: R¿t -Rpt-a¡ + bļiR^t - R Ft) + SļSMBt + h¡HMLOt
a tia)
Small 0.22 0.17 0.09 -0.08 -0.85 3.41 2.52 1.10 -0.80 -5.63
2 0.12 0.02 0.05 -0.05 -0.46 1.97 0.34 0.69 -0.66 -4.85
3 0.01 0.03 -0.03 -0.03 -0.20 0.21 0.43 -0.47 -0.42 -2.27
4 0.02 -0.06 -0.12 -0.04 -0.04 0.25 -0.86 -1.82 -0.50 -0.42
Big -0.00 -0.02 -0.10 -0.05 0.15 -0.08 -0.38 -1.91 -0.85 1.68
b m
Small 0.72 0.99 1.10 1.15 1.12 46.57 59.76 58.42 46.41 30.55
2 0.78 1.01 1.11 1.24 1.26 53.64 69.67 69.28 68.73 54.47
3 0.78 1.00 1.10 1.20 1.25 50.88 66.65 70.02 66.09 57.16
4 0.82 1.00 1.13 1.20 1.26 46.39 62.05 70.24 64.42 54.72
Big 0.83 0.96 1.05 1.11 1.18 59.85 77.16 80.24 75.87 55.31
s /(s)
Small 0.69 0.94 1.04 1.19 1.36 31.92 40.56 39.37 34.50 26.57
2 0.56 0.77 0.85 0.94 1.10 27.27 37.81 37.55 37.12 34.05
3 0.31 0.48 0.58 0.69 0.80 14.52 23.10 26.27 27.13 26.18
4 0.10 0.19 0.25 0.32 0.50 4.10 8.41 11.13 12.21 15.49
Big -0.27 -0.25 -0.16 -0.13 0.02 -14.06 -14.08 -8.82 -6.20 0.64
h m
Small 0.37 0.37 0.37 0.30 0.21 12.41 11.63 10.08 6.19 2.90
2 0.32 0.34 0.30 0.24 -0.08 11.27 11.99 9.56 6.71 -1.78
3 0.34 0.38 0.33 0.23 -0.16 11.19 13.13 10.59 6.52 -3.74
4 0.36 0.31 0.25 0.17 -0.16 10.50 9.76 7.99 4.56 -3.47
Big 0.14 -0.04 0.04 0.04 -0.18 5.09 -1.48 1.68 1.49 -4.30
r t(r )
Small 0.39 0.45 0.25 -0.09 -0.81 12.48 13.52 6.69 -1.76 -11.06
2 0.41 0.52 0.48 0.37 -0.56 13.96 17.92 14.90 10.34 -12.02
3 0.38 0.50 0.51 0.41 -0.50 12.29 16.48 16.25 11.31 -11.30
4 0.37 0.47 0.44 0.29 -0.58 10.28 14.56 13.65 7.73 -12.54
Big 0.22 0.25 0.27 0.01 -0.48 7.99 9.87 10.24 0.24 -11.35
c t(c)
Small 0.50 0.47 0.34 0.05 -0.24 15.09 13.26 8.36 0.99 -3.05
2 0.50 0.53 0.39 0.17 -0.51 15.79 16.86 11.07 4.38 -10.09
3 0.53 0.44 0.41 0.21 -0.53 15.99 13.63 11.91 5.40 -11.24
4 0.56 0.45 0.38 0.19 -0.50 14.44 12.84 10.82 4.71 -10.01
Big 0.23 0.17 0.17 0.01 -0.57 7.47 6.43 6.08 0.25 -12.26

The LHS variables in each set of 25 regressions are the monthly excess ret
variance) portfolios. The RHS variables are the excess market return, R^-R
factor, HML, or its orthogonal counterpart, HMLO, the profitability factor
CMA. Panel A shows intercepts from the FF three-factor model, and pane
slopes from (10).

above zero. In the five-factor model only the interc

per month; ř = 3.41) is more than two standard erro
five three-factor intercepts for portfolios in the high
than 2.8 standard errors below zero. The five-factor m

91

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

intercepts toward zero, but two are still extreme, -0.85% per month, t = -5.63,
for microcaps and -0.46%, t = - 4.85, for the second Size quintile.
Panel B of Table 8 shows the five-factor regression slopes for the 25 Size-
RVar portfolios. Market slopes increase strongly from the low RVar to the high
RVar portfolios. Within Size quintiles, there is a strong positive relation between
SMB slope and RVar ; stocks with higher residual return volatility behave like
smaller stocks. The positive correlations of RVar with market and SMB slopes
help explain why Size-Var and Size-RVar portfolios produce much the same
results in our tests. Megacaps aside, HMLO slopes are strongly positive in the
bottom four quintiles of RVar, but microcaps aside, they turn negative in the
highest RVar quintile. In words, low residual volatility tends to be associated
with value and high residual volatility tends to be associated with growth. Keep
in mind, however, that the average HMLO return is close to zero, so HMLO
slopes add almost nothing to the description of average returns.
Higher five-factor market and SMB slopes for stocks with more volatile
residual returns go in the wrong direction to explain the pattern in the Size-
RVar average portfolio returns. The improvements in the description of average
return provided by the five-factor model come from its RMW (profitability) and
CMA (investment) slopes. Major lifting is done by the RMW slopes, which are
strongly positive in the lower three quintiles of RVar and strongly negative in the
highest RVar quintile. The CMA slopes have a similar though less pronounced
pattern. Thus, the improvements in the explanation of average returns provided
by the five-factor model trace to the fact that the returns of low volatility
stocks behave like those of firms that are profitable but conservative in terms of
investment, whereas the returns of high volatility stocks behave like those of
firms that are relatively unprofitable but nevertheless invest aggressively. Novy-
Marx (2014) also finds that profitability exposures are important in capturing
the low average returns of high volatility stocks. (His model does not include
an investment factor.)
Average values of Inv that increase with RVar (Table 7) confirm the
suggestion from the CMA slopes that higher residual volatility is associated with
more investment. For stocks in the bottom two Size quintiles, lower profitability
(OP) in the highest RVar quintile is roughly consistent with lower RMW slopes.
In the highest three Size quintiles, however, average OP shows no relation to
RVar - another example of multivariate regression slopes that do not line up
with a univariate characteristic.
The five-factor model does not completely capture average returns on the 25
Size-RVar portfolios, but its major problems are familiar. Specifically, strong
negative exposures to RMW and CMA capture the low average returns of
big stocks that have high RVar. But strong negative exposures to RMW and
CMA miss a large part of the lower average returns of high RVar small stocks.
Small stocks with strong negative exposures to RMW and CMA are the lethal
combination that escapes explanation in many of the sorts here and in FF
(2015).

92

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor Model

Table 9
Average excess returns and characteristics of stocks in the 25 Size-AC portfolios, July 1963-December
2014 (618 months)

AC -► Low 2 3 4 High Low 2 3 4 High

Panel A: Means and SD of portfolio excess returns
Mean SD
Small 0.89 0.91 0.83 0.96 0.60 7.14 6.46 6.15 6.51 7.22
2 0.84 0.86 0.80 0.78 0.64 6.58 5.77 5.76 6.20 6.85
3 0.85 0.81 0.85 0.79 0.59 6.25 5.42 5.21 5.64 6.62
4 0.75 0.69 0.65 0.74 0.71 5.73 5.09 4.92 5.15 6.26
Big 0.67 0.48 0.52 0.50 0.26 5.20 4.14 4.05 4.57 5.20
Panel B: Average B/M, OP , 7/iv, and AC characteristics
B/M OP
Small 0.91 0.99 0.99 0.92 0.76 -0.06 0.15 0.16 0.18 0.24
2 0.75 0.85 0.79 0.74 0.62 0.19 0.22 0.24 0.25 0.44
3 0.71 0.77 0.78 0.68 0.56 0.31 0.25 0.25 0.27 0.30
4 0.65 0.75 0.74 0.63 0.51 0.33 0.27 0.27 0.30 0.41
Big 0.52 0.59 0.57 0.47 0.41 0.43 0.33 0.33 0.36 0.67
Inv AC
Small 0.09 0.15 0.16 0.18 0.31 -57.78 -2.24 1.66 6.03 50.26
2 0.14 0.16 0.18 0.18 0.32 -39.77 -2.18 1.66 5.97 47.45
3 0.15 0.16 0.15 0.18 0.31 -66.10 -2.15 1.67 5.90 38.96
4 0.13 0.12 0.14 0.16 0.25 -22.48 -2.12 1.60 5.84 22.28
Big 0.15 0.12 0.12 0.13 0.20 -16.71 -2.06 1.58 5.64 27.88

Panel A shows means and standard deviations of monthly excess returns on 2

formed yearly at the end of June using independent sorts of NYSE, AMEX,
stocks into Size (market capitalization) quintiles and quintiles of AC (accruals),
variables. For portfolios formed in June of year t, Size is market capitalization
change in operating working capital per split-adjusted share from the fiscal ye
book equity per split-adjusted share in t - 1 . Panel B shows time-series means
equity ratio (B/M), operating profitability (OP), and investment (Inv) for the
year preceding portfolio formation, as defined in Table 3. Panel B also shows
AC used to form portfolios each year.

6. Accruals

Panel A of Table 9 shows average excess returns on the 25 VW Size-AC

portfolios. Average returns are similar for the lower four AC quintiles of each
of the four smallest Size quintiles. For megacaps the lowest AC quintile has a
higher average return than the portfolios in the middle three quintiles. In four
of the five Size quintiles, average returns are much lower for the highest AC
quintile. The megacap portfolio in the highest AC quintile has by far the lowest
average excess return in the matrix, 0.26% per month. There is also a Size effect
in every AC quintile: The microcap portfolio in each AC quintile has a higher
average return than the megacap portfolio.
Panel A of Table 10 shows regression intercepts for the 25 Size-AC portfolios
from the FF three-factor model, the four-factor model that adds CMA (the best-
performing model in the summary tests on the Size-AC portfolios in Table 2),
and the five-factor model that also adds RMW. The FF three-factor model
overestimates average returns on four of five portfolios in the highest AC
quintile, producing intercepts from 1.90 to 4.03 standard errors below zero.
The exception is the portfolio in the second largest Size quintile, which, unlike

93

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

the other portfolios in the highest AC quintile, does not have a low average
return (Table 9). The FF three-factor model underestimates average returns on
1 9 of the 20 portfolios in the lower four AC quintiles. The most extreme positive
intercept, 0.28% per month ( t = 3.32), is for the megacap portfolio in the lowest
AC quintile.
The four-factor model that adds CMA to the FF three-factor model moves all
the troublesome negative intercepts in the highest AC quintile toward zero, and
only those for the two smallest Size quintiles are more than two standard errors
from zero. The four-factor intercept for the megacap portfolio in the lowest AC
quintile is a bit larger than the three-factor intercept (0.30, t = 3.47, versus 0.28,

Table 10
Regressions for the 25 Size-AC portfolios, July 1963 to December 2014 (618 months)
a t(a )

AC Low 2 3 4 High Low 2 3 4 High

Panel A: Regression intercepts
Three-factor: Mkt, SMB, and HML
Small 0.02 0.12 0.06 0.16 -0.28 0.26 1.61 0.95 2.48 -4.03
2 0.02 0.12 0.06 -0.00 -0.18 0.28 1.87 0.95 -0.04 -2.86
3 0.13 0.14 0.18 0.08 -0.18 1.68 2.10 2.74 1.25 -2.21
4 0.10 0.06 0.01 0.15 0.03 1.23 0.92 0.09 2.23 0.35
Big 0.28 0.10 0.12 0.07 -0.17 3.32 1.93 2.21 1.15 -1.90
Four-factor: Mkt, SMB, HML, and CMA
Small -0.04 0.05 0.02 0.13 -0.27 -0.47 0.74 0.26 2.01 -3.96
2 0.00 0.10 0.03 -0.00 -0.15 0.07 1.53 0.42 -0.03 -2.38
3 0.13 0.12 0.16 0.10 -0.11 1.70 1.73 2.37 1.57 -1.31
4 0.08 0.02 0.02 0.13 0.07 0.95 0.32 0.24 1.97 0.87
Big 0.30 0.08 0.09 0.10 -0.11 3.47 1.52 1.70 1.67 -1.21
Five-factor: Mkt, SMB, HMLO, RMW, and CMA
Small 0.12 0.20 0.13 0.22 -0.19 1.63 3.09 2.03 3.44 -2.70
2 0.06 0.10 0.04 -0.01 -0.17 0.93 1.52 0.60 -0.10 -2.62
3 0.20 0.16 0.17 0.08 -0.19 2.45 2.31 2.53 1.18 -2.30
4 0.09 0.09 -0.02 0.12 0.07 1.06 1.29 -0.27 1.79 0.82
Big 0.35 0.10 0.04 0.05 -0.20 4.00 1.94 0.64 0.83 -2.21
Panel B: Regression slopes

Three-factor: R¿t -Rft = a¡ + b¡(R¡^ft -Rpt) + s¿SMB( + hļHMLt + elt

h m
Small -0.02 0.00 0.02 0.00 -0.04 -0.87 0.04 0.87 0.04 -1.47
2 -0.00 0.07 0.05 0.03 -0.09 -0.14 3.12 2.32 1.18 -4.20
3 -0.07 0.02 0.08 0.06 -0.05 -2.64 0.97 3.45 2.62 -1.76
4 -0.05 0.10 0.18 -0.02 -0.13 -1.74 3.94 7.88 -0.81 -4.33
Big -0.27 -0.03 0.02 -0.10 -0.19 -8.84 -1.88 1.02 -4.34 -5.91
Four-factor: R¡t -Rpt = a¡ + bļ(R'ft -Rpt) + SļSMBt + hļHMLt + c¿CMA( + en
h m
Small -0.15 -0.14 -0.08 -0.06 -0.03 -4.04 -4.13 -2.54 -2.09 -1.01
2 -0.03 0.03 -0.02 0.03 -0.03 -1.04 0.83 -0.70 0.96 -1.01
3 -0.07 -0.03 0.03 0.11 0.12 -1.77 -0.95 0.98 3.49 3.01
4 -0.10 0.01 0.21 -0.05 -0.03 -2.55 0.16 6.55 -1.69 -0.86
Big -0.23 -0.08 -0.04 -0.03 -0.05 -5.76 -3.22 -1.56 -0.87 -1.23
c tic)
Small 0.28 0.31 0.22 0.14 -0.01 5.07 6.25 4.81 3.18 -0.13
2 0.06 0.10 0.16 -0.01 -0.14 1.40 2.26 3.67 -0.12 -3.23
3 -0.02 0.12 0.12 -0.10 -0.38 -0.30 2.52 2.41 -2.29 -6.56
4 0.11 0.21 -0.05 0.08 -0.21 1.88 4.24 -1.01 1.63 -3.60
Big -0.08 0.10 0.14 -0.16 -0.31 -1.26 2.72 3.50 -3.61 -4.89
( continued )

94

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table 10
Continued

AC Low 2 3 4 High Low 2 3 4 High

Five-factor: -Rpt = ūi + bļ(R^jt -RFt) + SiSMB
h m
Small -0.07 -0.07 -0.03 -0.02 0.01 -2.12 -2.09 -0.85 -0.69 0.29
2 -0.00 0.03 -0.02 0.03 -0.04 -0.16 0.84 -0.50 0.86 -1.27
3 -0.04 -0.01 0.04 0.10 0.08 -0.97 -0.32 1.16 3.05 2.00
4 -0.09 0.04 0.19 -0.06 -0.04 -2.35 1.14 5.92 -1.79 -0.87
Big -0.21 -0.07 -0.07 -0.05 -0.10 -5.07 -2.70 -2.62 -1.68 -2.24
r t(r )
Small -0.38 -0.38 -0.27 -0.22 -0.21 -10.77 -11.64 -8.74 -7.01 -6.23
2 -0.13 0.00 -0.03 0.02 0.03 -4.22 0.04 -1.01 0.54 1.13
3 -0.16 -0.10 -0.03 0.08 0.21 -3.97 -3.01 -0.75 2.39 5.27
4 -0.05 -0.15 0.12 0.01 -0.00 -1.15 -4.49 3.74 0.28 -0.05
Big -0.17 -0.07 0.12 0.11 0.19 -3.94 -2.77 4.57 3.61 4.40
c t(c )
Small 0.04 0.09 0.08 0.03 -0.09 1.08 2.54 2.46 0.95 -2.40
2 0.00 0.13 0.14 0.03 -0.16 0.05 3.69 3.99 0.79 -4.88
3 -0.12 0.07 0.14 0.02 -0.21 -2.80 1.85 3.85 0.55 -4.96
4 -0.00 0.18 0.18 0.03 -0.25 -0.01 4.83 5.20 0.71 -5.48
Big -0.35 0.00 0.12 -0.16 -0.32 -7.55 0.16 4.25 -4.79 -6.68

The LHS variables in each set of 25 regressions are the monthly excess retur
portfolios of Table 2. The RHS variables are the excess market return, Mkt = Ry
value factor, HML, or its orthogonal counterpart, HMLO, the profitability factor, RM
CAM. Panel A shows three-factor, four-factor, and five-factor regression interc
slopes for RMW, CMA, and HML or HMLO (as relevant).

t = 3.32), but most of the intercepts in the first four quin

toward zero.

Performance deteriorates in the five-factor model that adds RMW, especially

for microcaps. Adding RMW pushes the intercept for the microcap portfolio
in the highest AC quintile toward zero (from -0.27, t = - 3.96, to -0.19, t =
-2.70), but it moves the intercepts for the other four microcap portfolios from
values mostly indistinguishable from zero to large positive values that are 1.63
to 3.44 standard errors from zero. The problems of the five-factor model are not
limited to microcaps. Adding RMW increases the intercepts for all portfolios
in the lowest AC quintile, and three of the five intercepts for the portfolios in
the highest AC quintile become more negative.
The regression slopes in panel B of Table 10 help us interpret the intercepts.
(We do not show RM - RF and SMB slopes since they are similar for different
models.) For the troublesome portfolios in the three smallest Size quintiles and
the highest AC quintile, the HML slopes in the FF three-factor model are close to
zero, so HML does not help explain the low average returns of these portfolios.
The HML slope for the megacap portfolio in the highest AC quintile is rather
strongly negative, which helps explain the extremely low average excess return
on this portfolio, 0.26% per month (Table 9), but nevertheless leaves a three-
factor intercept, -0.17% per month that is -1.90 standard errors from zero
(Table 10). The negative HML slope for the megacap portfolio in the lowest AC
quintile in part explains why the three-factor model does a poor job explaining
the high average return of this portfolio.

95

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 The Review of Financial Studies / v 29 n 1 2016

Microcaps aside, the four-factor model that adds CMA to the FF three-factor
model produces strong negative CMA slopes for the portfolios in the highest
AC quintile in Table 10. This is consistent with the Inv evidence in Table 9
that firms in these portfolios tend to invest aggressively, and it helps explain
why this model improves the regression intercepts for these portfolios. Table 9
shows the microcap portfolio in the highest AC quintile also invests a lot, but
its CMA slope is close to zero and its four-factor intercept, -0.27 (t = - 3.96),
is almost unchanged from the three-factor intercept, -0.28. Most of the four-
factor HML slopes are close to zero and so have little effect on the regression
intercepts.
The RMW slopes for microcaps in the five-factor model are strongly negative.
This is consistent with the evidence in Table 9 that controlling for AC,
profitability is lowest for microcaps. Driven by a large negative RMW slope,
the intercept for the microcap portfolio in the highest AC quintile shrinks
from -0.27 in the four-factor model that does not include RMW to -0.19
(t = -2.70) in the five-factor model. The RMW slopes are, however, also largely
responsible for the general deterioration of the intercepts from the four-factor
to the five-factor model for the other four microcap portfolios. The reductions
predicted by negative RMW slopes do not show up in the average returns of
these portfolios. Negative RMW slopes for portfolios in the lowest AC quintile
(which do not have low average returns in Table 9) and positive slopes for some
of the portfolios in the highest AC quintile (which have low average returns)
are responsible for the five-factor model's adverse effect on the intercepts for
these portfolios.
FF (2015) find that in 5 x 5 sorts on Size and Inv , the portfolios in the smaller
Size quintiles and the highest Inv quintile produce intercept problems, even for
asset pricing models that include the investment factor CMA. This suggests
that small firms that invest a lot are a general problem for the asset pricing
models we consider. Table 9 shows these firms are prominent in the highest AC
portfolios of smaller Size quintiles, and the intercepts for these portfolios are
always among the most extreme in Table 10.
The Size-AC portfolios provide a valuable caution. They are the only sorts,
here and in FF (2015), in which the five-factor model performs noticeably worse
than other models. It is also noteworthy that the problems of the five-factor
model in the Size-AC sorts trace to the profitability factor since in other sorts
RMW typically improves the description of average returns, often substantially.

7. Momentum

Table 1 1 shows average excess returns for monthly independent sorts of stocks
into quintiles of Size and momentum {Prior 2-12). With one exception, there
is a Size effect in the Prior 2-12 quintiles; given Prior 2-12 , average returns
are larger for portfolios of small stocks. The exception is the lowest Prior 2-
12 quintile (extreme losers), in which the portfolios in the two smallest Siz

96

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table 11
Average excess returns and characteristics of stocks in the 25 Size-Prior 2-12 portfolios, July
1963-December 2014 (618 months)

Prior 2-12 - » Low 2 3 4 High Low 2 3 4 High

Panel A: Means and SD of portfolio excess returns
Mean SD
Small 0.03 0.67 0.91 1.05 1.39 8.01 5.88 5.43 5.50 6.78
2 0.14 0.66 0.82 1.01 1.23 7.86 5.88 5.27 5.41 6.74
3 0.27 0.62 0.72 0.78 1.19 7.37 5.53 5.07 5.00 6.31
4 0.20 0.59 0.66 0.79 1.04 7.26 5.52 4.87 4.79 5.89
Big 0.17 0.46 0.39 0.55 0.79 6.79 4.88 4.37 4.31 5.26
Panel B: Average B/M, OP , 7/iv, and Prior 2-12 characteristics
B/M OP
Small 0.84 0.95 0.98 1.00 0.99 0.08 0.21 0.28 0.28 0.17
2 0.72 0.80 0.83 0.84 0.82 0.24 0.27 0.28 0.29 0.24
3 0.68 0.75 0.78 0.78 0.73 0.27 0.29 0.29 0.29 0.29
4 0.66 0.71 0.74 0.73 0.71 0.29 0.35 0.33 0.30 0.28
Big 0.56 0.59 0.60 0.60 0.61 0.36 0.34 0.35 0.37 0.39
Inv Prior 2-12
Small 0.24 0.14 0.13 0.13 0.14 -31.87 -4.56 9.45 25.08 90.33
2 0.27 0.16 0.15 0.15 0.19 -27.79 -4.33 9.55 25.08 83.61
3 0.24 0.15 0.14 0.14 0.20 -25.91 -4.13 9.56 24.97 77.08
4 0.19 0.14 0.12 0.13 0.18 -23.80 -4.03 9.66 24.90 71.41
Big 0.19 0.13 0.12 0.12 0.16 -20.94 -3.71 9.68 24.82 58.96

Panel A shows means and standard deviations of monthly excess returns on v

monthly using independent sorts of NYSE, AMEX, and (beginning in 1973) N
capitalization) quintiles and quintiles of Prior 2-12 (momentum), with NYS
The intersections of the two sorts produce 25 Size-Prior 2-12 portfolios. For p
of month t, Size is the market cap of a stock at the beginning of t and Prior
the 1 1 months from t - 12 to t - 2. Panel B shows time-series means of the portf
(B/M), operating profitability (OP), and investment (Inv) for the fiscal year end
portfolio formation, as defined in Table 3. Panel B also shows the time-series
to form portfolios each month.

quintiles have extreme low average excess returns, 0.03%

There is a strong momentum effect in every Size quin
Size increases. The spread in average returns from extr
losers is 1 .36% per month for microcaps and 0.62% fo
monthly excess return for microcap extreme winners
for megacap winners.
Table 12 shows intercepts from the five-factor mode
model that adds MOM to (10) and in which HMLO is th
(0.04; ř = 0.52) and residual from the regression of H
RMW , CM A, and MOM. The five-factor model is not
the average returns produced by momentum sorts. Th
for extreme losers are strongly negative, the intercepts
strongly positive, and the spreads between the intercep
and losers are similar to the spreads in average returns
In the dividend discount model (2) that Fama and
motivate the five-factor model, the internal rate of retur
to shareholders (r in Equation (2)) is approximately th
stock return. Momentum is short-term; the relative pe

97

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

Table 12
Regressions for the 25 Size-Prior 2-12 portfolios, July 1963-December 2014 (618 months)

Prior 2-12 -► Low 2 3 4 High Low 2 3 4 High

Panel A: R¡t -Rpt = ai + M^A/f ~^Ft) + SļSMBt + hļHMLt + riR
a t{a)
Small -0.71 -0.20 0.02 0.20 0.62 -4.68 -2.50 0.39 3.06 6.40
2 -0.62 -0.21 -0.05 0.15 0.51 -4.35 -2.58 -0.80 2.79 5.63
3 -0.37 -0.18 -0.11 -0.07 0.52 -2.40 -2.10 -1.75 -1.06 5.34
4 -0.40 -0.21 -0.13 0.02 0.44 -2.41 -2.19 -1.94 0.34 4.09
Big -0.37 -0.13 -0.17 -0.04 0.33 -2.27 -1.29 -2.72 -0.62 3.08

Panel B: R¡t -Rpt = aļ + £,(^Af/ -Rft) + s¡SMBt + h¡HMLOt + r¡RMWt + Cļ

a t{a)
Small -0.23 -0.01 0.09 0.15 0.40 -2.28 -0.09 1.44 2.31 4.85
2 -0.11 0.03 0.01 0.10 0.24 -1.47 0.45 0.09 1.88 3.73
3 0.16 0.06 -0.00 -0.13 0.22 1.79 0.98 -0.02 -2.11 3.30
4 0.16 0.07 -0.02 -0.03 0.11 1.60 1.09 -0.25 -0.45 1.49
Big 0.17 0.19 -0.09 -0.15 -0.02 1.59 2.76 -1.51 -2.61 -0.24
h m
Small 0.13 0.29 0.31 0.21 0.05 2.62 9.20 10.27 6.55 1.37
2 0.06 0.20 0.25 0.21 0.01 1.66 7.00 8.92 8.33 0.36
3 0.05 0.21 0.27 0.29 0.00 1.07 6.84 9.51 9.87 0.14
4 0.09 0.14 0.24 0.20 0.04 1.94 4.26 7.90 6.68 1.03
Big 0.02 0.04 0.13 0.08 0.06 0.34 1.20 4.28 2.64 1.73
r tir)
Small -0.31 0.15 0.26 0.15 -0.17 -6.37 4.75 8.62 4.70 -4.22
2 -0.16 0.27 0.30 0.22 -0.15 -4.50 9.34 11.03 8.68 -4.87
3 -0.18 0.26 0.34 0.37 -0.08 -4.16 8.46 11.84 12.48 -2.44
4 -0.20 0.29 0.37 0.36 -0.09 -4.05 8.74 12.31 11.64 -2.63
Big 0.03 0.24 0.26 0.25 -0.01 0.55 7.43 8.80 8.89 -0.27
c tic)
Small -0.10 0.33 0.38 0.33 0.02 -1.86 9.81 11.67 9.57 0.38
2 -0.17 0.23 0.32 0.30 -0.13 -4.49 7.44 10.69 10.78 -3.89
3 -0.18 0.22 0.33 0.35 -0.14 -3.84 6.61 10.68 10.91 -3.93
4 -0.09 0.30 0.34 0.30 -0.10 -1.69 8.47 10.42 8.95 -2.46
Big -0.13 0.19 0.15 0.22 -0.17 -2.34 5.47 4.72 7.25 -4.75
m tim)
Small -0.69 -0.30 -0.12 0.05 0.30 -31.07 -20.38 -8.90 3.09 16.16
2 -0.72 -0.35 -0.10 0.04 0.37 -43.43 -26.44 -7.75 3.76 25.35
3 -0.74 -0.36 -0.18 0.05 0.41 -36.98 -25.21 -13.92 4.00 27.62
4 -0.79 -0.41 -0.19 0.05 0.45 -35.34 -26.70 -13.35 3.28 27.17
Big -0.75 -0.45 -0.13 0.15 0.48 -32.18 -29.29 -9.08 11.34 30.72

The LHS variables are the monthly excess returns on the 25 Size-Prior 2-12 portfolios
the excess market return, R^ - Rp, the Size factor, SMB, the value factor, HML, or its
HMLO, the profitability factor, RMW, the investment factor, CM A, and the momentum fa
using independent 2x3 sorts on Size and each of B/M , OP, Inv, and Prior 2-12. The ta
a five-factor model that does not include MOM, and intercepts and slopes for the six-f
MOM. In this table, HMLO is the sum of the intercept (0.04, ř =0.5 1) and the residual
HML on Rm-Rf, SMB, HML, RMW, CM A, and MOM.

months t - 12 to t - 2 tends to persist for only about nine m

t. Because of the long-term return reversals identified by De
(1985), Prior 2-12 is negatively related to longer-term relativ
perhaps it is not surprising that the five-factor model, which is
run expected returns, fails to capture the positive relation be
and current returns.
Adding MOM to the five-factor model improves the regres
but problems remain. Most noticeable is the unexplained m

98

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
 Dissecting Anomalies with a Five-Factor Model

microcaps: the extreme loser portfolio has a rather strong negative intercept,
-0.23% per month (t = - 2.28), the winner portfolio has a strong positive
intercept, 0.40% (ř=4.85), and the intercepts increase monotonically from
losers to winners. There is a weaker momentum pattern in the intercepts of
the second Size quintile, and there is a weak reverse momentum pattern in the
intercepts for the largest (megacap) quintile.
The regression slopes for the six-factor model show why including MOM is
critical in the tests on the Size-Prior 2-12 portfolios. The MOM slopes increase
from strongly negative for losers to strongly positive for winners, which is the
pattern in average returns. But HMLO , RMW , and CM A provide little help. Both
the HMLO premium and the HMLO slopes for extreme winners and extreme
losers are close to zero. Megacaps aside, RMW slopes are negative for extreme
losers, and the negative slopes are helpful for explaining low average returns,
but RMW slopes are also slightly negative for extreme winners. The CM A
slopes have a similar pattern. Panel B of Table 1 1 shows there are also no clear
patterns in B/M , OP , and Inv for the 25 Size-Prior 2-12 portfolios.

8. Conclusions

The list of anomalies shrinks in the five-factor model, in part because anomalous
returns become less anomalous and in part because the returns associated with
different anomaly variables share factor exposures that suggest they are in large
part the same phenomenon.
The flat relation between market ß and average return that has long plagued
tests of the CAPM is captured in the five-factor model by RMW and CMA slopes
that offset the average return predictions of market and SMB slopes. Stocks
with higher CAPM market ßs have higher five-factor market and SMB slopes
that raise predictions of their average returns. But low ß stocks have positive
exposures to the profitability and investment factors of the five-factor model
that raise predictions of their average returns, and high ß stocks have negative
exposures to RMW and CMA that lower their predicted returns. Thus, low ß
stock returns behave like those of profitable firms that invest conservatively,
whereas high ß returns behave like those of less profitable firms that invest
aggressively.
The high average returns associated with share repurchases, which are a
problem for the FF three-factor model, cease to be an anomaly in the five-factor
model. The reason again is that the returns of repurchasers behave like those
of profitable firms that invest conservatively. Positive exposures to RMW and
CMA also go a long way toward capturing the average returns of low volatility
stocks, whether volatility is measured in terms of total returns or residuals from
the FF three-factor model.
Like the returns of relatively unprofitable firms that invest aggressively, the
returns of high ß stocks, stocks with highly volatile returns, and stocks of firms
that make large share issues load negatively on RMW and CMA. Unlike the

99

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n 1 2016

average returns of high ß portfolios, however, negative five-factor exposures

to RMW and CMA do not fully capture the low average returns associated with
large share issues and high volatility. Unexplained average returns are largely
concentrated in small stocks, especially microcaps. Small stocks with negative
exposures to RMW and CMA are also a problem for the five-factor model in
many of the tests in FF (2015), leading them to dub it the lethal combination.
The five-factor model typically performs better than the FF three-factor
model when applied to different sets of LHS portfolios here and in FF (2015).
Portfolios formed on Size and accruals are an exception. The pricing problems
associated with accruals do not seem to have much to do with the lethal
combination of slopes that is a common problem in other sorts.
All models that do not include a momentum factor fare poorly in the tests
on the 25 Size-Prior 2-12 portfolios. A six-factor model that includes MOM
performs well, but by playing a home game; the momentum factor, MOM , is
just a coarse (2x3 rather than 5x5) version of the sorts used to construct the
25 Size-Prior 2-12 portfolios. Nevertheless, the six-factor model leaves lots of
momentum in microcap returns unexplained.

Table Al
Summary statistics for the 25 Size-Var portfolios, July 1963-December 2014 (618 months)

Var Low 2 3 4 High Low 2 3 4 High

Panel A: Means and SD of portfolio excess return
Mean SD
Small 1.00 1.18 1.09 0.81 -0.18 4.07 5.67 6.50 7.53 9.22
2 0.90 1.03 1.06 0.91 0.27 4.02 5.26 5.90 6.83 8.78
3 0.75 0.85 0.97 0.88 0.44 3.68 4.78 5.39 6.23 8.08
4 0.67 0.74 0.77 0.77 0.49 3.72 4.48 5.10 5.79 7.69
Big 0.43 0.53 0.54 0.46 0.46 3.43 3.99 4.48 5.10 6.74
Panel B: Average B/M , OP , Inv, and Var characteristics
~B/M OP
Small 1.01 0.96 0.93 0.90 0.94 0.24 0.28 0.33 0.30 -0.10
2 0.89 0.83 0.80 0.78 0.73 0.34 0.31 0.30 0.29 0.18
3 0.84 0.77 0.73 0.72 0.68 0.29 0.30 0.31 0.31 0.27
4 0.81 0.72 0.69 0.68 0.65 0.29 0.33 0.35 0.31 0.30
Big 0.65 0.55 0.56 0.58 0.57 0.34 0.37 0.36 0.36 0.37
Inv Var
Small 0.11 0.14 0.17 0.21 0.22 2.53 5.69 9.01 14.88 43.98
2 0.10 0.13 0.16 0.21 0.30 1.89 3.79 5.70 8.52 20.08
3 0.10 0.12 0.14 0.18 0.30 1.54 2.98 4.47 6.69 15.70
4 0.10 0.11 0.12 0.15 0.26 1.41 2.52 3.70 5.43 12.57
Big 0.10 0.11 0.12 0.14 0.21 1.33 2.16 2.98 4.19 8.51

This table shows means and standard deviations of monthly excess retur
monthly using a first pass sort of NYSE, AMEX, and (beginning in 1
capitalization) quintiles and second-pass sorts into quintiles of Var (to
for both variables. The Var sorts are conditional on Size quintile. The int
Size-Var portfolios. For portfolios formed at the beginning of month t,
beginning of t and Var is the variance of its daily returns estimated using 6
returns. Panel A shows means and standard deviations of monthly exces
B shows time-series means of the portfolio book-to-market equity ratio
and investment (Inv) for the fiscal year ending in the calendar year prec
Table 3. Panel B also shows the time-series average values of Var used to

100

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Table A2
Regressions for the 25 Size-Var portfolios, July 1963 to December 2014 (618 months)

Var Low 2 3 4 High Low 2 3 4 High

Panel A: Three-factor: Rit -RFt = aļ + M^A/í -
a t{a)
Small 0.35 0.30 0.13 -0.22 -1.25 5.34 4.09 1.68 -2.33 -7.64
2 0.27 0.21 0.18 -0.03 -0.68 4.24 3.05 2.51 -0.38 -6.08
3 0.18 0.11 0.18 0.01 -0.38 2.68 1.69 2.36 0.08 -3.76
4 0.13 0.10 0.06 0.00 -0.26 1.68 1.31 0.83 0.06 -2.46
Big 0.05 0.11 0.05 -0.06 -0.12 0.73 1.85 0.89 -1.06 -1.21

Panel B: Five-factor: R¡t -Rpt = a¡ + b¿(R'ft - Rpt) + SiSMBt + h¡HMLOt

a t(a)
Small 0.23 0.17 0.07 -0.14 -0.87 3.65 2.42 0.95 -1.42 -5.62
2 0.12 0.03 0.04 -0.15 -0.42 2.02 0.48 0.55 -1.97 -3.96
3 0.04 -0.02 0.02 -0.14 -0.16 0.61 -0.34 0.23 -1.87 -1.60
4 0.01 -0.06 -0.11 -0.10 -0.01 0.08 -0.90 -1.67 -1.20 -0.07
Big -0.04 -0.05 -0.06 -0.11 0.13 -0.57 -0.91 -1.11 -1.69 1.30
b m
Small 0.69 0.99 1.10 1.17 1.16 44.19 58.48 57.56 50.14 30.95
2 0.76 1.00 1.10 1.24 1.32 51.99 66.59 68.05 69.30 50.63
3 0.75 0.98 1.09 1.22 1.30 47.76 65.14 65.50 67.60 55.13
4 0.78 0.98 1.12 1.22 1.32 41.83 57.53 67.74 62.51 54.81
Big 0.76 0.93 1.03 1.14 1.24 44.97 70.40 75.24 74.54 52.07
s í(s)
Small 0.67 0.94 1.05 1.20 1.39 30.34 39.61 39.38 36.69 26.31
2 0.55 0.76 0.84 0.96 1.13 26.85 35.93 36.90 38.19 30.80
3 0.31 0.48 0.56 0.70 0.83 13.94 22.66 24.23 27.56 25.19
4 0.09 0.20 0.25 0.32 0.51 3.55 8.30 10.71 11.66 15.09
Big -0.25 -0.22 -0.17 -0.18 0.01 -10.54 -12.08 -8.84 -8.35 0.38
h m
Small 0.35 0.42 0.39 0.33 0.24 11.30 12.68 10.54 7.29 3.21
2 0.30 0.36 0.33 0.25 -0.07 10.48 12.43 10.49 7.00 -1.33
3 0.31 0.38 0.34 0.26 -0.17 10.13 13.09 10.57 7.34 -3.62
4 0.36 0.30 0.27 0.21 -0.17 9.78 9.01 8.30 5.62 -3.59
Big 0.16 0.06 0.07 -0.02 -0.09 4.94 2.25 2.51 -0.57 -1.83
r t(r )
Small 0.37 0.44 0.27 -0.04 -0.79 11.74 12.97 7.04 -0.86 -10.52
2 0.40 0.52 0.47 0.43 -0.54 13.78 17.39 14.46 11.85 -10.24
3 0.38 0.45 0.54 0.47 -0.47 12.06 14.98 16.05 12.80 -9.96
4 0.36 0.45 0.50 0.33 -0.57 9.60 13.17 15.05 8.42 -11.80
Big 0.19 0.38 0.30 0.09 -0.46 5.66 14.31 11.11 2.85 -9.69
c t(c )
Small 0.47 0.50 0.39 0.13 -0.23 13.72 13.60 9.30 2.61 -2.84
2 0.49 0.54 0.40 0.22 -0.47 15.44 16.63 11.41 5.56 -8.32
3 0.50 0.45 0.40 0.32 -0.55 14.79 13.94 11.20 8.08 -10.69
4 0.52 0.47 0.42 0.26 -0.52 12.85 12.71 11.66 6.04 -10.06
Big 0.35 0.24 0.16 0.04 -0.59 9.66 8.51 5.34 1.12 -11.51

The LHS variables in each set of 25 regressions are the monthly excess retur
portfolios. The RHS variables are the excess market return, Rm - Rp, the S
HML, or its orthogonal counterpart, HMLO, the profitability factor, RMW, and
A shows intercepts from the FF three-factor model, and panel B shows fiv
(10).

101

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
The Review of Financial Studies / v 29 n J 2016

References

Ang, A., R. J. Hodrick, J. Xing, and X. Zhang. 2006. The cross-section of volatility and expected returns. Journal
of Finance 5 1 :259-99.

Barillas, F., and J. Shanken. 2015. Which alpha? Manuscript, Goizueta Business School, Emory University.

Black, F., M. C. Jensen, and M. Scholes. 1972. The capital asset pricing model: Some empirical tests. In Studies
in the theory of capital markets, 79-121. Ed. M. C. Jensen. New York: Praeger.

Carhart, M. M. 1997. On persistence in mutual fund performance. Journal of Finance 52:57-82.

Davis, J. L., E. F. Fama, and K. R. French. 2000. Characteristics, covariances, and average returns: 1929-97.
Journal of Finance 55:389-406.

DeBondt, W. F. M., and R. Thaler. 1985. Does the stock market overreact? Journal of Finance 40:793-805.

Fama, E. F. 1996. Multifactor portfolio efficiency and multifactor asset pricing. Journal of Financial and
Quantitative Analysis 3 1 :44 1-65.

Quantitative Analysis 33:2 1 7-3 1 .

Fama, E. F., and K. R. French. 1993. Common risk factors in the return
Economics 33:3-56.

105:457-72.

Fama, E. F., and J. D. MacBeth. 1 973. Risk, return, and equilibrium: Empirical
81:607-36.

Frazzini, A., and L. H. Pedersen. 2014. Betting against beta. Journal of Financial Economics 1 1 1:1-25.

Gibbons, M. R., S. A. Ross, and J. Shanken. 1989. A test of the efficiency of a given portfolio. Econometrica
57:1121-52.

Harvey, C. R., Y. Liu, and H. Zhu. 2015. .. . and the cross-section of expected returns. Review of Financial Studies
29:5-68.

Hou, K., C. Xue, and L. Zhang. 2015. Digesting anomalies: An investment approach. Review of Financial Studies
28:650-705.

Ikenberry, D., J. Lakonishok, and T. Vermaelen. 1995. Market underreaction to open market share repurchases.
Journal of Financial Economics 39:181-208.

Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers: Implications for stock market
efficiency. Journal of Finance 48:65-91.

Lewellen, J., S. Nagel, and J. Shanken. 2010. A skeptical appraisal of asset pricing tests. Journal of Financial
Economics 96: 1 75-94.

Lintner, J. 1965. The valuation of risk assets and the selection of risky investments in stock portfolios and capital
budgets. Review of Economics and Statistics 47: 13-37.

Loughran, T., and J. R. Ritter. 1995. The new issues puzzle. Journal of Finance 50:23-51.

Loughran, T., and A. M. Vijh. 1997. Do long-term shareholders benefit from corporate acquisitions? Journal of
Finance 52: 1 765-90.

Merton, R. 1973. An intertemporal asset pricing model. Econometrica 41:867-77.

Miller, M., and F. Modigliani. 1961. Dividend policy, growth, and the valuation of shares. Journal of Business
34:411-33.

102

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms
Dissecting Anomalies with a Five-Factor Model

Novy-Marx, R. 2014. Understanding defensive equity. Manuscript, Simon Graduate School of Business,
University of Rochester, September.

Pastor, L'., and R. F. Stambaugh. 2003. Liquidity risk and expected stock returns. Journal of Political Economy
111:642-85.

Sharpe, W. F. 1964. Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of
Finance 19:425^12.

Sloan, R. G 1996. Do stock prices fully reflect information in accruals and cash flows about future earnings?
The Accounting Review 7 1 : 289-3 15.

103

This content downloaded from

203.78.15.149 on Sat, 10 May 2025 05:34:16 UTC
All use subject to https://about.jstor.org/terms