Finance Research Letters 17 (2016) 88–92

Contents lists available at ScienceDirect

Finance Research Letters

journal homepage: www.elsevier.com/locate/frl

Identifying portfolio-based systematic risk factors in equity

marketsR
Klaus Grobys a, Jesper Haga b,∗
a
University of Vaasa, Wolffintie 34, 65200 Vaasa, Finland
b
Hanken School of Economics, Biblioteksgatan 16, 65100 Vaasa, Finland

a r t i c l e i n f o a b s t r a c t

Article history: Four prominent new asset pricing factors have recently been proposed. We test whether
Received 24 November 2015 these factors fulfill the necessary conditions to qualify as risk factors. We show that the in-
Revised 29 January 2016
vestment and betting-against-beta factors fulfill these conditions. However, the profitability
Accepted 31 January 2016
and quality factors do not fulfill these conditions pointing towards non-risk-based expla-
Available online 5 February 2016
nations.
JEL classification: © 2016 Elsevier Inc. All rights reserved.
G10
G12
G14
C58

Keywords:
Asset pricing model
Betting-against-beta factor
Quality factor
Investment factor
Profitability factor

1. Introduction

The academic literature devotes considerable attention to exploring risk factors that may link stock returns to systematic
risk. In their seminal paper, Fama and French (1993) proposed a three-factor asset pricing model accounting for the size and
value factor in addition to the market factor. The authors showed that the three-factor model provides a better description
of stock returns than the traditional Capital Asset Pricing Model (CAPM). In the wake of Fama and French’s (1993) research,
and motivated by multifactor asset pricing model theory, several other portfolio-based risk factors have been proposed in
the academic literature. From a theoretical point of view, a risk factor should capture systematic risk and should be ideally
linked to consumption risk, as pointed out in Cochrane (2001, p. 41).
In line with Cochrane’s (2001) stochastic discount factor framework, Charoenrook and Conrad (2008) propose an interest-
ing addition to the debate on the risk–return relationship in asset markets. The authors set out the necessary conditions for a
portfolio-based zero-cost portfolio to be a candidate risk factor. First, a priced factor’s conditional variance should be related
to the factor’s conditional mean, whereas a positive relation between the conditional mean and variance signals a positive

R
We are grateful to an anonymous reviewer for helpful comments.
∗
Corresponding author. Tel.: +358 443797918.
E-mail address: jesper.haga@hanken.fi (J. Haga).

http://dx.doi.org/10.1016/j.frl.2016.01.010
1544-6123/© 2016 Elsevier Inc. All rights reserved.
 K. Grobys, J. Haga / Finance Research Letters 17 (2016) 88–92 89

risk premium for the factor. Second, a conditional factor’s mean should be explained by its conditional variance. Charoenrook
and Conrad (2008) employ their proposed approach to test whether the size, book-to-market, momentum and liquidity fac-
tor qualify as risk factors, in the sense that they satisfy the necessary conditions. They find that the necessary conditions
are fulfilled for the size, liquidity and book-to-market factors over the period 1963–2003. As an alternative to Charoenrook
and Conrad’s (2008) procedure, Pukthuanthong and Roll (2014) extend Moskowitz’s (2003) research by proposing a protocol
for determining which factor candidates are related to risks and which candidates are related to mean returns. Their proce-
dure is very different from Charoenrook and Conrad (2008) and requires accounting for considerable data restrictions. Their
results indicate that only the risk premium associated with momentum appears to be statistically significant.
More recently, other prominent factors have been discussed. Among those, four interesting factors are Franzzini and
Pedersen’s (2014) betting-against-beta zero-cost strategy (BAB), Asness et al. (2014) quality factor (QMJ), and Fama and
French’s (2015) investment and profitability factors (CMA and RMW). While some papers provide alternative explanations
for those factors, it may be surprising that no study has yet explored whether these factors actually satisfy the condition
necessary for characterizing those as risk factors in line with asset pricing theory. This paper addresses this gap in the
literature.1
The purpose of this study is to test whether a positive conditional risk–return relationship exists for the BAB, QMJ, CMA,
or RMW factors. We use a sample period from July 1963 to August 2015 and employ GARCH-in-mean models in the spirit
of Charoenrook and Conrad (2008) to estimate and test whether the necessary conditions are fulfilled for those potential
candidates. We also perform a further robustness check in the form of a sample-split test and examine orthogonalized
factors.
The current research contributes to the existing literature as follows. This is the first study to formally test whether
prominent zero-cost strategies recently proposed in the literature, which are featured as risk factors because they may
be associated with some systematic risk, fulfill the necessary conditions derived from asset pricing theory in the spirit of
Charoenrook and Conrad (2008) and Cochrane (2001). The study extends that of Charoenrook and Conrad (2008) and tests
whether the conditional variances of the BAB, QMJ, CMA, and RMW factors are related to the corresponding factor’s condi-
tional mean. Finally, we argue that a rejection of the necessary conditions would point toward a non-risk-based explanation.
Our results indicate that the BAB and the CMA factors incorporate a positive relation between conditional mean and
conditional variance, whereas the intercept is statistically not different from zero, implying that the conditional risk fully
explains its conditional mean. These results are in line with Charoenrook and Conrad’s (2008) findings for the size, liquidity,
and the book-to-market factor, and suggest that the necessary conditions for those factors are fulfilled. Surprisingly, we
do not find such evidence for the QMJ or the RMW factors suggesting that future research may be needed to investigate
whether non-risk-based arguments could explain those phenomena.
Our paper is organized as follows. The next section presents the data. The third section provides a brief overview of the
theoretical background. The fourth section presents the results, whereas the last section concludes.

2. Data

Data were downloaded from Kenneth French’s and AQR’s data libraries.2 We examined four factors. For each one of
these prominent factors the prior literature presents some theoretical explanations. First, Franzzini and Pedersen (2014)
show that funding constrains flattens the original capital-allocation-line. The underlying reason is that funding constrained
investors tend to overweight high beta assets. Adrian et al. (2014) connect the leverage of financial intermediary with the
return of assets. Moreover, the leverage of financial intermediaries is related to the funding constrains driving the betting-
against-beta factor. Franzzin and Pedersen (2014) suggest that funding constrains can be accounted for in the asset pricing
model by adding their BAB factor. Second, Asness et al. (2014) propose a QMJ factor. Thereby, quality is, motivated by
the Gordon growth formula, measured by evaluating four characteristics: profitability, payout, safety and growth. Third,
valuation theory states that firm value is a function of investments and the profitability of those investments. According to
Fama and French (2006), firms invest more when the return on investments is high in comparison to the cost of capital.
As a consequence, after controlling for firms’ investments, profitability should be positively related to expected returns.
Further, after controlling for firms’ profitability, investments should be negatively related to expected returns. Fama and
French (2015) use these arguments for establishing a new asset-pricing model incorporating CMA and RMW, which they
derived via a valuation model.
Table 1 reports the descriptive statistics and correlation matrix for the four proposed factors. The BAB factor has the
highest unconditional mean with an economic magnitude of 0.90% per month. Moreover, we observe that both the BAB
factor and the RMW factor have excess kurtosis and negative skewness suggesting that those factors are exposed to crash
risks. The correlation matrix shows that the RMW factor and the QMJ factor have the highest correlation. Moreover, the
CMA is least correlated with the other factors.

1
In a more recent paper, Buchner and Wagner (2016) suggest an alternative explanation for the BAB factor. They derive the pricing errors that are
induced by the standard CAPM’s linearity assumption and argue that the BAB factor could be an artefact of the spurious OLS regression effect.
2
AQR data library: https://www.aqr.com/library/data-sets. Kenneth French data library: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_
library.html.
90 K. Grobys, J. Haga / Finance Research Letters 17 (2016) 88–92

Table 1
Descriptive statistics.

BAB QMJ CMA RMW

Mean 0.90 0.31 0.39 0.25

Std. Dev. 3.15 2.02 2.44 2.13
Skewness −0.54 0.28 −0.01 −0.40
Kurtosis 7.17 4.63 6.34 14.60
Maximum 15.60 9.51 12.30 12.19
Minimum −15.67 −6.81 −12.53 −17.57
Correlation matrix
BAB QMJ CMA RMW
BAB 1
QMJ 0.25 1
CMA 0.33 0.13 1
RMW 0.28 0.72 −0.09 1

3. Theoretical model

If defining the realized excess return of asset i at time t + 1 as Rexcess i,t+1

, the pricing kernel as ξt+1 = 1 − b1 f1,t+1 − . . . −
bK fK,t+1 , and where f1,t+1 , . . . , fK,t+1 denotes the risk factor in excess returs, then the equilibrium condition Et [ξt+1 Rexcess
i,t+1
]=
0 implies that an asset’s risk premium is linearly related to the covariance of its excess returns with the pricing kernel,
   
Et Rexcess
i,t+1 = −R f COVt ξt+1 , Rexcess
i,t+1 . (1)

Charoenrook and Conrad (2008) point out that if a well-diversified zero-cost portfolio Rzero−cost
k,t+1
has a positive loading on
factor fk,t+1 , and a beta of zero with respect to other factors and no residual risk, we obtain3
   
Et Rzero−cost
k,t+1
= R f bk βkVARt fk,t+1 . (2)
VARt [Rzero−cost ]
k,t+1
Since VARt [ fk,t+1 ]= , Eq. (2) can be easily re-written as
βk2
   
zero−cost
Et Rk,t+1 = R f ψkVARt Rzero−cost
k,t+1
. (3)
Charoenrook and Conrad (2008) argue that in a time-series setting, if ψ k is constant, one should observe that a higher
forecast conditional variance of return to the factor-mimicking portfolio flows through a higher forecast mean return of the
portfolio.

4. Results

We followed Charoenrook and Conrad’s (2008) approach to test whether a link between a factor’s conditional variance
and conditional mean exists, as derived by employing Cochrane’s (2001) stochastic discount factor model framework. We
used GARCH-in-mean models and tested if the conditional variance is significantly positively related to the conditional mean
(i), and whether the conditional mean can be fully explained by its conditional mean (ii), as implied by Eq. (3). In doing so,
we estimate the GARCH-in-mean model as:
rk,t+1 = αk + γk hk,t+1 + εk,t+1 , (1a)
where rk,t+1 is the return for factor k at time t+1 and hk,t+1 is the expected conditional variance of factor k’s returns, and
γk = R f ψk (see Eq. (3)). The expected conditional variance (hk,t+1 ) is then estimated using the following equation:
hk,t+1 = ωk + δk εk,t
2
+ ηk hk,t + μk,t+1 , (2a)

where εk,t
2 is the squared error term for time t from the mean equation, and hk, t is the conditional variance from time t.
Table 2 reports the parameter estimates for the calibrated models for both the full sample and two sub samples. The early
subsample covers July 1963 to December 1989 and the later subsample covers January 1990 to August 2015.
We start by examining the time variation of the factors’ expected conditional variance, as time variation is necessary for
the relationship between the conditional variance and conditional mean to exist. The variance equation in Table 2 reports
large and significant δ k and ηk coefficients for all four factors and for all samples suggesting that the variance for all four
factors is both time varying and persistent. The sum of δ k and ηk is less than one in all regressions, with only one exception
that is the later period for the BAB factor.
Next, we explore the relationship between the factors conditional variance and conditional mean. Over the full sample,
the BAB factor has a positive and significant relationship with conditional variance and the conditional mean implied by the

3
Note zero-cost factor-mimicking portfolios are given in terms of excess return by construction.
 K. Grobys, J. Haga / Finance Research Letters 17 (2016) 88–92 91

Table 2
Risk–return relation for the factors.
Mean equation Full sample Early sample Later sample
BAB QMJ CMA RMW BAB QMJ CMA RMW BAB QMJ CMA RMW

α 0.10 0.71 ∗∗∗

−0.17 0.19 −1.54 ∗
0.39 −0.15 0.22 0.87∗∗∗ ∗∗∗
1.14 −0.20 0.20
(0.39) (3.29) (−1.16) (1.57) (−1.77) (1.17) (−0.68) (0.96) (3.74) (2.98) (−1.11) (1.16)
γ 0.42∗∗ −0.28∗ 0.39∗∗∗ 0.05 1.35∗∗ −0.05 0.42∗∗ −0.07 0.06 −0.49∗∗ 0.36∗ 0.10
(2.41) (−1.69) (2.91) (0.35) (2.57) (−0.17) (2.22) (−0.22) (0.35) (−2.01) (1.91) (0.57)
Variance equation
ω 0.53∗∗∗ 0.27∗∗∗ 0.16∗∗∗ 0.19∗∗∗ 0.53 0.13 0.22 0.22 0.48∗∗ 0.46∗∗ 0.12 0.19∗∗
(3.43) (2.98) (2.27) (2.90) (1.64) (1.39) (1.62) (1.40) (2.51) (2.38) (1.44) (2.07)
δ 0.25∗∗∗ 0.15∗∗∗ 0.17∗∗∗ 0.17∗∗∗ 0.10∗∗∗ 0.05∗ 0.21∗∗∗ 0.11∗∗ 0.45∗∗∗ 0.21∗∗∗ 0.14∗∗∗ 0.21∗∗∗
(5.69) (6.41) (5.06) (7.24) (2.66) (1.94) (3.86) (2.12) (4.80) (5.21) (3.01) (5.91)
η 0.72∗∗∗ 0.81∗∗∗ 0.79∗∗∗ 0.78∗∗∗ 0.82∗∗∗ 0.91∗∗∗ 0.74∗∗∗ 0.80∗∗∗ 0.59∗∗∗ 0.74∗∗∗ 0.83∗∗∗ 0.76∗∗∗
(17.06) (25.26) (20.25) (22.01) (11.39) (20.03) (11.99) (8.52) (10.77) (13.99) (14.39) (16.52)

∗
Statistically significant on 10% level.
∗∗
Statistically significant on 5% level.
∗∗∗
Statistically significant on 1% level.

Table 3
Risk–return relation for the orthogonal factors.

Mean equation Orthogonal against the market factor

BAB QMJ CMA RMW

α 0.17 0.78 ∗∗∗

0.03 0.33∗∗∗
(0.62) (3.70) (0.17) (2.63)
γ 0.40∗∗ −0.16 0.36∗∗ 0.00
(2.26) (−0.87) (2.14) (0.02)
Variance equation
ω 0.52∗∗∗ 0.33∗∗∗ 0.19∗∗ 0.20∗∗∗
(3.42) (3.20) (2.30) (3.08)
δ 0.24∗∗∗ 0.16∗∗∗ 0.14∗∗∗ 0.17∗∗∗
(5.52) (6.09) (4.17) (7.51)
η 0.72∗∗∗ 0.76∗∗∗ 0.81∗∗∗ 0.78∗∗∗
(17.27) (20.88) (17.40) (24.62)

coefficient γ k . This relationship is in line with the first necessary condition. Moreover, the constant is small and insignificant
which suggests that the conditional variance explains the conditional mean implying that the second necessary condition is
also fulfilled. We consider this result to be evidence that that the BAB factor does indeed satisfy the qualifying conditions
for a risk factor. A subsample test shows that this relationship appears to be mainly driven by the first subsample.
Unlike the BAB factor, the QMJ factor has a significant negative relationship with conditional variance and the conditional
mean. This relationship holds on a 10% significance level for the full sample and a 5% significance level for the later sample.
Moreover, in both the full sample and the later subsample the QMJ factor has a significant positive constant value implying
that the necessary conditions are not fulfilled.
Furthermore, Table 2 shows that the CMA factor’s conditional mean has a significant positive relationship with its condi-
tional variance, as it is stable across both subsamples. Moreover, the conditional mean is not significantly different from zero
and smaller than the unconditional mean in all three models, suggesting that the conditional mean and conditional variance
are able to fully explain the factor’s return. Finally, the RMW factor exhibits both an insignificant constant and insignificant
variance coefficient suggesting that the RMW factor does not fulfill the necessary conditions to qualify as a risk factor.
To check the robustness of our results, we orthogonalized all four factors by regressing them successively on the market
factor, and also repeated the previous analysis for the sample from 1963 to 2015. The results are reported in Table 3. The
conclusions do not change.4

5. Conclusions

In our paper, we test whether a positive conditional risk–return relationship exists for the BAB, QMJ, CMA, or RMW
factors. We focus on Charoenrook and Conrad (2008), who proposed a theoretically motivated approach and derived the
relation between the conditional variance of a true factor and that factor’s associated risk premium. Future research may
explore these factors employing other approaches, such as proposed in Pukthuanthong and Roll (2014), for instance. We
argue that if a potential portfolio-based zero-cost strategy is a risk factor, the factor should not only be related to some
systematic risk, but also satisfy the necessary conditions derived from asset pricing theory. Considering a sample from 1963

4
In an additional robustness check, we employed a threshold GARCH-in-mean (TGARCH) model to account for asymmetric volatility effects. The results
are available on request and confirm our findings.
92 K. Grobys, J. Haga / Finance Research Letters 17 (2016) 88–92

to 2015, we find that the CMA and BAB zero-cost strategies satisfy both necessary conditions to qualify as risk factors. How-
ever, we do not find such evidence for either the QMJ or RMW. We argue that the risk-based asset pricing models proposed
in the literature should embrace our findings. However, future research would be necessary to clarify the underlying sources
of non-risk-based factors.

Appendix

Table A.1.

Table A.1
Risk–return relation estimated with TGARCH.
Mean equation Full sample Early sample Late sample
BAB QMJ CMA RMW BAB QMJ CMA RMW BAB QMJ CMA RMW

α −0.10 0.63∗∗∗ −0.18 0.20 −2.01∗∗∗ 0.38 −0.16 0.25 0.87∗∗∗ 1.02∗∗∗ −0.21 0.22
(−0.39) (2.90) (−1.28) (1.64) (−3.27) (1.15) (−0.72) (1.13) (3.70) (2.62) (−1.15) (0.24)
γ 0.56∗∗∗ −0.21 0.43∗∗∗ 0.00 1.78∗∗∗ −0.04 0.47∗∗ −0.12 0.05 −0.41 0.39∗∗ 0.05
(3.29) (−1.27) (3.13) (0.02) (4.80) (−0.15) (2.38) (−0.42) (0.28) (−1.64) (2.01) (0.78)
Variance equation
ω 0.49∗∗∗ 0.28∗∗∗ 0.16∗∗ 0.19∗∗∗ 0.27∗ 0.13 0.22 0.19 0.49∗∗ 0.48∗∗ 0.12 0.20∗∗
(3.18) (2.93) (2.31) (3.00) (1.77) (1.37) (1.63) (1.35) (2.41) (2.38) (1.47) (2.32)
δ 0.30∗∗∗ 0.17∗∗∗ 0.20∗∗∗ 0.13∗∗∗ 0.26∗∗∗ 0.06∗ 0.26∗∗∗ 0.08 0.44∗∗∗ 0.23∗∗∗ 0.16∗∗∗ 0.16∗∗
(4.78) (4.82) (4.58) (3.26) (3.00) (1.72) (3.42) (1.64) (4.24) (3.91) (2.66) (2.52)
η 0.71∗∗∗ 0.81∗∗∗ 0.79∗∗∗ 0.79∗∗∗ 0.83∗∗∗ 0.91∗∗∗ 0.74∗∗∗ 0.82∗∗∗ 0.59∗∗∗ 0.74∗∗∗ 0.83∗∗∗ 0.77∗∗∗
(17.13) (23.36) (19.17) (20.67) (19.43) (19.13) (11.22) (9.27) (10.18) (12.99) (13.94) (14.95)
β −0.08 −0.05 −0.06 0.07∗ −0.20∗∗ −0.00 −0.09 0.05 0.02 −0.04 −0.03 0.08
(−1.47) (−1.37) (−1.18) (1.85) (−2.47) (−0.11) (−1.13) (0.68) (0.19) (−0.76) (−0.41) (1.38)

∗
Statistically significant on 10% level.
∗∗
Statistically significant on 5% level.
∗∗∗
Statistically significant on 1% level.

References

Adrian, T., Etula, E., Muir, T., 2014. Financial intermediaries and the cross-section of asset returns. J. Financ. 69, 2557–2596.
Asness, C., Frazzini, A., Pedersen, L. H., 2014. Quality minus junk, Working paper. Available at: http://papers.ssrn.com/abstract_id=2312432 (accessed
09.02.2016).
Buchner, A., Wagner, N., 2016, The betting against beta anomaly: Fact or fiction?, Financ. Res. Letts., Forthcoming.
Charoenrook, A., Conrad, J., 2008. Identifying risk-based factors, Working paper. Available at: https://apps.olin.wustl.edu/workingpapers/pdf/2008- 12- 002.pdf
(accessed 09.02.2016).
Cochrane, J., 2001. Asset Pricing. Princeton University Press, Princeton, N.J., p. 41.
Fama, E., French, K., 1993. Common risk factors in the returns of stocks and bonds. J. Financ. Econ. 59, 3–32.
Fama, E., French, K., 2006. Profitability, investment, and average returns. J. Financ. Econ. 73, 491–518.
Fama, E., French, K., 2015. A five-factor asset pricing model. J. Financ. Econ. 116, 1–22.
Franzzini, A., Pedersen, L.H., 2014. Betting against beta. J. Financ. Econ. 111, 1–25.
Moskowitz, T.J., 2003. An analysis of covariance risk and pricing anomalies. Rev. Financ. Stud. 16, 417–457.
Pukthuanthong, K., Roll, R., 2014. A Protocol for Factor Identification. Working Paper. University of Missouri.