A Mean Variance Pricing Model

James W. Kolari∗, Jianhua Z. Huang†, Wei Liu‡, and Huiling Liao§

September 15, 2025

Abstract

We propose a Mean Variance Pricing (MVP) model. Our model is comprised of two macrofactors
related to the Markowitz (1952, 1959) investment parabola: (1) the cross-sectional mean of stock returns
in the market, and (2) their cross-sectional standard deviation. Based on the theoretical ZCAPM of
Kolari, Liu, and Huang (2021), a simplified modeling approach is specified. Rather than estimate
the empirical ZCAPM using the more complex iterative expectation-maximization (EM) algorithm, a
standard linear regression relation with no intercept is employed. Our parsimonious MVP model produces
Fama and MacBeth (1973) out-of-sample cross-sectional regression results that are similar to those
for the more complex EM ZCAPM. Additionally, the MVP model consistently outperforms prominent
multifactor models. Further analyses of average mispricing errors confirm these findings. Implications to
asset pricing research are discussed.

Keywords: anomalies, asset pricing models, mispricing error, ZCAPM

JEL: G12, C20

∗ The corresponding author is JP Morgan Chase Professor of Finance, Texas A&M University, Adam C. Sinn ’00 Department of

Finance, Mays Business School, College Station, TX 77843-4218. Office phone: 979-845-4803. Email address: j-kolari@[Link]
† Presidential Chair Professor, Director, Technology and Innovation Center for Digital Economy (TIDE), School of Data

Science, The Chinese University of Hong Kong, Shenzhen, China, 51872. Office phone: (86) 0755-2351700. Email address:
jhuang@[Link]
‡ Clinical Associate Professor of Finance, Texas A&M University, Adam C. Sinn ’00 Department of Finance, Mays Business

School, College Station, TX 77843-4218. Office phone: 979-845-1224. Email address: wliu@[Link]
§ Assistant Professor of Applied Mathematics, Applied Mathematics, Illinois Institute of Technology, Chicago, IL 60616 Office

phone: 312-567-8990. Email address: hliao13@[Link]

We have benefited from helpful comments by Ihsan Badshah, Hilal Anwar Butt, Klaus Grobys, Yao Han, Seppo Pyn-
nönen, David Veal, and Ivo Welch. We thank participants at academic conferences with respect to related papers, including the
Financial Management Association (FMA) 2012 and 2017, Midwest Finance Association (MFA) 2012 and 2018, Academy of
Financial Services (AFS) 2012, Southern Finance Association 2020, University of Otego (Dunedin, New Zealand) 2021, Academy
of Finance 2022, Southwestern Finance Association (SWFA) 2023 and 2025, and Western Economic Association International
(WEAI) 2023. Best Paper in Investments awards were received at the FMA 2012 and SWFA 2025 conferences.

1
 A Mean Variance Pricing Model

1 Introduction
Black (1972) developed his now famous zero-beta CAPM by relaxing riskless rate and short selling restrictions
in the general equilibrium CAPM of Sharpe (1964), Lintner (1965), Mossin (1966), and Treynor (1961,
1962). Extending his model, Kolari, Liu, and Huang (2021) derived a special case dubbed the ZCAPM that
contains two macromarket risk factors related to the Markowitz (1952, 1959) mean variance investment
parabola: (1) the cross-sectional mean return (related to beta risk), and (2) cross-sectional standard deviation
of returns (associated with zeta risk). The theoretical ZCAPM identifies two orthogonal portfolios on the
investment parabola that have equal time-series return variance: (1) I ∗ on the efficient frontier, and (2)
zero-beta counterpart ZI ∗ on the inefficient lower boundary. Upon substituting expected returns for these
portfolios into the zero-beta CAPM, an alternative form is specified that does not require proxies for market
and zero-beta portfolios. Instead, the cross-sectional first and second moments of returns for all stocks in the
U.S. market (for example) on a given day can be used as macromarket factors.
While the theoretical ZCAPM has the advantage of readily estimable factors, thereby avoiding the pitfall
of finding a good proxy for the market portfolio in the CAPM1 , it is relatively difficult to empirically estimate.
Because zeta risk in the ZCAPM has unknown positive or negative effects on asset returns, a hidden +1, −1
dummy variable is introduced to capture latent risk effects. To estimate the probability pi of the dummy
variable’s sign for the ith asset, the iterative expectation-maximization (EM) algorithm of Dempster, Laird,
and Rubin (1977) is utilized. Subsequently, zeta risk (denoted Z ∗ ) is computed in the empirical ZCAPM
by multiplying pi by the zeta coefficient Zi , or Z ∗ = pi Zi . Numerous publications by the authors have
documented that zeta risk loadings are highly significant in out-of-sample cross-sectional tests.2
Despite promising results using EM regression that consistently outperform prominent multifactor models
(at times by large margins), as well as free software made available on internet3 , the ZCAPM has not been
widely adopted by researchers. A major drawback that likely limits its acceptance is the complexity of the
EM algorithm. While well-known is the hard sciences, EM regression methods are unfamiliar to most finance
academics and professionals. For example, one often-cited researcher commented to the authors that the
cross-sectional test results using the ZCAPM were so good that something must have gone wrong in their
computation. Another disadvantage of the EM regression model (or EM ZCAPM) is that it is difficult to add
other factors to the model due to extensive changes in the iterative EM algorithm.
The present paper seeks to overcome barriers to ZCAPM adoption by proposing an easily estimated
and understood regression relation for empirical tests of the ZCAPM. This new modeling approach produces
Fama and MacBeth (1973) cross-sectional regression results that are almost the same as those using more
complicated EM regression methods. Like the EM regression, our proposed model contains two macrofactors
based on the cross-sectional mean and standard deviation of returns of all U.S. stocks on any given day.
We argue that, in both theoretical and empirical ZCAPM specifications, if these two macrofactors equal
zero, all assets must have zero return. As such, we re-specify the empirical ZCAPM as a standard linear
regression with no intercept. To distinguish this simple econometric approach from the more sophisticated
EM regression approach, we refer to it as the mean variance pricing (MVP) model. For comparative purposes,
1 Thewidely-cited Roll (1977) critique contends that the CAPM cannot be tested without an efficient market portfolio.
2 See
Liu, Kolari, and Huang (2012), Liu (2013), Kolari, Huang, Butt, and Liao (2022), Kolari, Huang, Liu, and Liao (2022),
Kolari, Huang, Liu, and Liao (2025), Kolari, Liu, Huang, and Liao (2026), and Kolari, Liu, and Pynnonen (2024).
3 Matlab, R, and Python programs are available at GitHub ([Link]

1
in forthcoming empirical tests, we report results using alternative test asset portfolios for firm characteristic
and industry portfolios as well as large numbers of anomaly portfolios. Using out-of-sample tests, the MVP
model produces cross-sectional regression results based on factor loadings that are similar to those for the
more complex EM regression form of the empirical ZCAPM (or EM ZCAPM). Further graphical comparisons
of average predicted excess returns to average realized returns show that the MVP model has comparable
average mispricing errors to those for the EM ZCAPM. Also, these ZCAPM-based models have noticeably
smaller mispricing errors compared to prominent multifactor models.
The next section reviews the theoretical and empirical ZCAPM. Section 3 specifies our MVP model.
Section 4 describes the data and empirical tests. Section 5 gives the comparative empirical ZCAPM results
for MVP model and EM ZCAPM in addition to prominent multifactor models. The last section concludes.

2 ZCAPM Review

2.1 Theoretical ZCAPM

Black (1972) derived the zero-beta CAPM from Markowitz’s (1952, 1959) mean variance parabola by
relaxing some assumptions in Sharpe’s CAPM. The single period CAPM assumes: perfect capital markets,
homogeneous investor expectations, two-parameter probability distributions of returns, investor risk aversion,
no short selling, and borrowing at the riskless rate (Rf ). Black allowed investors to borrow at the risky rate
E(RZ ) > Rf and short sell assets. Under these conditions, the CAPM can be re-written as follows:

E(Ri ) = E(RZ ) + βi [E(RM ) − E(RZ )], (1)

where E(RM ) = the expected market portfolio return; E(RZ ) = the expected zero-beta portfolio return that
is uncorrelated (orthogonal) to the market portfolio; and βi = beta risk. Figure 1 illustrates the zero-beta
CAPM in the the context of the parabola. According to Roll (1980), by extending a ray from the market
portfolio M through the global minimum variance portfolio G, we can locate E(RZ ) on the Y-axis. Portfolio
Z is shown on the inefficient lower boundary of the parabola.
Since short positions are allowed, Roll (1980) argued that portfolios M and Z can be constructed from
different combinations of an efficient portfolio I and its zero-beta portfolio counterpart ZI. Consequently,
Black’s model can be more generally specified as:4

E(Ri ) = E(RZI ) + βiI [E(RI ) − E(RZI )], (2)

where I and ZI are any efficient index and its orthogonal zero-beta counterpart, respectively, and βiI is beta
risk. Lastly, Black allowed the existence of a riskless asset. Given that investors can buy but not short sell
(borrow) the riskless asset, after rearrangement of terms, we obtain:5

E(Ri ) − Rf = βiI [E(RI ) − Rf ] + βiZI [E(RZI ) − Rf ], (3)

where βiI and βiZI are beta risks related to portfolios I and ZI, respectively.
Kolari, Liu, and Huang showed that efficient and inefficient portfolios on the boundary of parabola with
equal time-series variance of returns are orthogonal to one another. Figure 2 illustrates these portfolios
4 See Copeland and Weston (1980, p. 184) for this specification.
5 See Black (1972, p. 454, equation (40)) for this three-asset form of the zero-beta CAPM.

2
labeled I ∗ and ZI ∗ , respectively. In their book, they provide two mathematical proofs that the vertical width
or span of the parabola is approximately equal to the cross-sectional standard deviation of returns of all
assets in the market (denoted σm ). This new insight implies that the mean market return (denoted Rm ) lies
approximately on the axis of symmetry that divides the parabola into upper and lower halves. Using these
variables, they define the expected returns of orthogonal portfolios I ∗ and ZI ∗ as follows::6

E(RI ∗ ) ≈ E(Rm ) + σm (4)

E(RZI ∗ ) ≈ E(Rm ) − σm , (5)

Substituting these expected returns into equation (3) for the zero-beta CAPM, the theoretical ZCAPM is
obtained:

E(Ri ) = E(RZI ∗ ) + βiI ∗ [E(RI ∗ ) − E(RZI ∗ )]

= E(Rm ) − σm + βiI ∗ {[E(Rm ) + σm ] − [E(Rm ) − σm ]}
= E(Rm ) + (2βiI ∗ − 1)σm
∗
E(Ri ) = E(Rm ) + Zim σm , (6)

∗
where Zim = 2βiI ∗ − 1. Lastly, similar to Black, assuming that investors can buy (but not short) a riskless
asset, the final form of the ZCAPM becomes:7

∗
E(Ri ) − Rf = βim [E(Rm ) − Rf ] + Zim σm , (7)

∗
where βim = beta risk related to average (or mean) market excess returns; and Zim = zeta risk associated
with cross-sectional market return dispersion of all assets in the market (i.e., market return dispersion).
Again, it is important to note that zeta risk is positive or negative in sign in the cases of portfolios I ∗ and
ZI ∗ , respectively.
Two macromarket forces emerge from the investment parabola. First, as mean market returns change,
the overall level of the parabola changes affecting the expected returns of all assets therein, ceteris paribus.
Second, as market return dispersion changes, the width of the parabola changes affecting the expected returns
of assets in the top half of the parabola in an opposite manner than assets in the bottom half. For example,
assume that market return dispersion increases. Assets located in the top half of the parable would experience
increasing expected returns, and vice versa for those in the bottom half with decreasing expected returns,
ceteris paribus. The magnitudes of these level and width parabola effects (i.e., the first and second moments
of expected returns, respectively) on any given asset are determined by its beta and zeta systematic risk
measures.

2.2 Empirical ZCAPM based on the EM algorithm

Kolari, Liu, and Huang conjectured that the directional positive or negative effects of market return dispersion
on asset returns is unknown. For this reason, they posited the following empirical regression model to estimate
6 More specifically, they employed the term f (θ)σm instead of σm , wherein f (θ) > 0 is a complex function of other terms.
7 For this mathematical derivation, see Kolari, Liu, and Huang (2021, p. 71) and Kolari, Huang, Liu, and Liao (2025, p. 25,
footnote 13).

3
the ZCAPM in the real world:

Rit − Rf t = αi + βim (Rmt − Rf t ) + Zim Dit σmt + uit , t = 1, . . . , T, (8)

where a dummy signal variable Dit = +1 or −1 is introduced to capture potential positive and negative zeta
risk, and other notation is as before. Since it is not known a priori whether an asset is located in the upper
or lower half of the parabola, Dit is a latent or hidden variable. To estimate the probability that Dit = +1,
they estimate equation (8) via the well-known expectation-maximization (EM) algorithm of Dempster, Laird,
and Rubin (1977).8
∗
Defining the zeta risk coefficient Zim = Zim (2pi − 1), which incorporates the probability estimate for
signal variable Dit , they re-write the empirical ZCAPM as:.

∗
Rit − Rf t = βim (Rmt − Rf t ) + Zim σmt + uit , t = 1, . . . , T, (9)

In forthcoming analyses, we use the value-weighted CRSP index to proxy Rat on day t, the U.S. Treasury bill
rate for Rf t , and the value-weighted cross-sectional standard deviation of stock returns in the CRSP index
computed as: v
u n
u n X
σmt =t wit−1 (Rit − Rmt )2 , (10)
n − 1 i=1

where n = the total number of stocks; wit−1 = the previous day’s market value weight for the ith stock; and
Rit = the return of the ith stock on day t. Note that, unlike equation (9), the empirical ZCAPM does not
contain an intercept or alpha parameter, as the authors found that residual error variance increased due to
its inclusion.
Precedent for a market return dispersion factor can be found in its economic content. Widespread
consensus in the literature finds that market return dispersion is related to various macroeconomic shocks,
including economic uncertainty, business cycles, market volatility, and unemployment rates.9 Not surprisingly,
previous studies have augmented the market factor with a market return dispersion factor and conducted
empirical tests. In general, studies by Jiang (2010), Demirer and Jategaonkar (2013), and Garcia, Mantilla-
Garcia, and Martellini (2014) obtained mixed findings in terms of statistical significance. By contrast, in
previously cited publications, ZCAPM studies have shown that, on a consistent basis, zeta risk loadings are
highly significant in out-of-sample Fama and MacBeth (1973) cross-sectional regression analyses. Strikingly,
compared to prominent multifactor models, the goodness-of-fit as measured by R2 values are near perfect at
times in out-of-sample tests, and average mispricing errors were noticeably smaller than those for prominent
multifactor models.
While the use of the EM algorithm in regression estimation of the ZCAPM (hereafter the EM ZCAPM)
produces impressive empirical results, as discussed earlier, researchers have been reluctant to implement the
model due to its statistical complexity. To simplify matters, we next propose a standard regression model
with no intercept that is easy to estimate using available regression software (including Excel) and readily
8 See also Jones and McLachlan (1990), McLachlan and Peel (2000), McLachlan and Krishnan (2008), and others. For

applications in finance, see Harvey and Liu (2016), Chen, Cliff, and Zhao (2017), and others. The EM algorithm iterates between
an E-step that computes the conditional expectation of the log-likelihood function and an M-step to maximize the MLE until
convergence is reached. Kolari, Liu, and Huang provide step-by-step equations to estimate the EM regression.
9 See Loungani, Rush, and Tave (1990), Christie and Huang (1994), Bekaert and Harvey (1997), Connolly and Stivers (2003),

Gomes, Kogan, and Zhang (2003), Stivers (2003), Bansal and Yaron (2004), Zhang (2005), Pastor and Veronesi (2009), Bansal,
Kiku, Shaliastovich, and Yaron (2014), Angelidis, Sakkas, and Tessaromatis (2015), among others.

4
understood by users with basic knowledge of regression methods.

3 The MVP Model

Given empirical ZCAPM relation (8) with macrofactors average market excess return and market return
dispersion, consider the special case when market return dispersion σmt = 0. In this case, the width or
span of the Markowitz investment parabola in Figure 2 collapses to the axis of symmetry. Since there is no
investment parabola (and no dummy signal variable D), the average market return E(Rm ) on the axis of
symmetry of the parabola equals the riskless rate Rf . Consequently, the average market excess return equals
zero, i.e., E(Rm ) − Rf = 0. Under these conditions, all assets’ excess returns in the ZCAPM must be zero.
For these reasons, we propose the following mean variance pricing (MVP) model with no intercept:

Rit − Rf t = βim (Rmt − Rf t ) + Zim σmt + uit , t = 1, . . . , T. (11)

The absence of an intercept in the MVP model is consistent with general equilibrium theory in the
CAPM, zero-beta CAPM, and ZCAPM. In empirical tests of the CAPM and zero-beta CAPM, it is common
practice to include an intercept (or alpha) parameter to enable evaluation of model validity.10 Empirical tests
of multifactor models generally include an intercept parameter also. However, the empirical EM ZCAPM and
MVP models have no intercept due to the fact that asset excess returns must be zero if the macrofactors in
these models equal zero. Note that, as discussed in Section 2, previous studies augmented the market factor
with a market return dispersion factor but incorrectly included an intercept term in their models. The reason
for this mispecification is that they did not perceive that the width or span of the investment parabola is
determined by market return dispersion, which Kolari, Liu, and Huang (2021) later posited in their book.
The regression coefficients in the MVP model can be estimated by means of the following well-known
equations:
PT
t=1 (Rit − Rf t − βim (Rmt − Rf t ) − Zim σmt ) (Rmt − Rf t ) = 0, (12)
PT
t=1 (Rit − Rf t − βim (Rmt − Rf t ) − Zim σmt ) σmt = 0, (13)

where the general formula is

 
Rm1 − Rf 1 σm1
!
 Rm2 − Rf 2 σm2
 
βim T −1 T  ∈ RT ×2 .

= (X X) X y with X =  .. .. (14)
Zim 
 . .


RmT − Rf T σmT .

By comparison, with the inclusion of an intercept term, this formula is

 
αi ! !−1 ! !−1 !
1T 1T T 1T X 1T
 βim  = (1 X) y= y, (15)
 
XT XT XT 1 XT X XT
Zim
10 SeeJensen (1968) for discussion of alpha mispricing error in models. Later work by Gibbons, Ross, and Shanken (GRS)
(1989) developed a formal test of model validity based on the joint equality of alphas across test assets.

5
which results in different coefficients than equation (14) calculated as
!  −1  
βim 1 1
= XT X − XT 11T X XT I − 11T y. (16)
Zim T T

Ordinary statistical software (including Excel) can be used to compute standard errors and t-statistics for
regression coefficients.11
Using standard statistical notation, the coefficient of determination in OLS time-series regression with
dependent variable yt , mean value ȳ, and fitted value ŷt is computed as:
PT PT
2 t=1 (ŷt − ȳ)2 (yt − ŷt )2
R = PT = 1 − Pt=1
T
. (17)
t=1 (yt − ȳ)2 t=1 (yt − ȳ)2

The no intercept model counterpart of this statistic is

PT 2
t=1 (yt − ŷt )
R02 =1− PT , (18)
2
t=1 yi

where
T
X T
X T
X T
X T
X
yt2 = (yt − ȳ + ȳ)2 = (yt − ȳ)2 + T ȳ 2 + 2 (yt − ȳ)ȳ > (yt − ȳ)2 . (19)
t=1 t=1 t=1 t=1 t=1

4 Methodology
Because there is no intercept in the MVP model, we do not conduct in-sample GRS tests for the joint equality
of alphas across test assets. Instead, to comparatively evaluate alternative asset pricing models, we employ
out-of-sample Fama and MacBeth (1973) cross-sectional regression tests. According to Simin (2008), Ferson,
Nallareddy, and Xie (2013), Barillas and Shanken (2018), Fama and French (2018), Martin and Nagel (2021),
and others, out-of-sample analyses provide more reliable tests of model validity than in-sample tests.12 Also,
they are consonant with investable strategies in which risk is evaluated, investment decisions are made, and
returns per unit risk are evaluated in the next out-of-sample period.

4.1 Data
Four sets of test asset portfolios are used in our analyses. Following Fama and French (1992, 1993) and
many others, we download from Kenneth French’s website13 daily returns for 25 size-BM (book-to-market)
portfolios, 25 operating profit-capital investment portfolios, and 48 industry portfolios. Additionally, we
download daily returns for 153 stock anomaly portfolios from an online data website provided by Jensen,
Kelly, and Pedersen (2023). These portfolios are considered to be anomalous in the sense that they are not
explained by asset pricing factors. Because the latter daily returns are complete for all anomalies in 1972
11 Given a two-factor model, the t-statistic for a regression coefficient is defined as β̂ /SE(β̂ ). Coefficient cT β̂ follows a normal
j j
distribution N (cT β, σ 2 cT (XT X)−1 c), where MSE is used to estimate σ 2 , and c = (1, 0) Porn c = (0, 1) 2can be used to obtain the
distribution for βim (or Zim ). In the regression model with intercept term, MSE = (y − ŷ) /(n − 3), whereas in the
i=1 i
Pn
MVP regression model without intercept, MSE = (y − ŷ)2 /(n − 2) with corresponding ŷi values.
i=1 i
12 Data snooping, potential manipulation, and other econometric issues are substantially mitigated in out-of-sample tests.
13 See [Link]

6
and thereafter, our sample period for all data series starts in January 1972 and ends in December 2024.
Additional tests are run using 133 anomaly portfolios from Chen and Zimmerman (2022) in the available
sample period July 1972 to December 2021.
To construct daily excess returns for the market factor, we download from the Center for Research
in Security Prices (CRSP) database the value-weighted CRSP stock market index return (denoted Rmt in
the empirical ZCAPM and MVP model) and three-month Treasury bill rate (or Rf t ). Daily returns for
the market return dispersion factor (σmt ) are computed as in equation (10). Finally, we download daily
returns from French’s website for the following popular multifactors: size (SM B), value (HM L), momentum
(M OM ), profitability (RM W ), and capital investment (CM A). Table 1 contains descriptive statistics for
these factors in our sample period.

4.2 Cross-sectional regression tests

We estimate cross-sectional regressions for the following four models: CAPM, MVP with intercept, MVP
(with no intercept), and EM ZCAPM. While the latter two models are the main focus of this paper, for
comparative purposes, we also estimate the following prominent (often-cited) multifactor models: Fama and
French (1992, 1993) three-factor model (FF3) with market, size and value factors, Carhart (1997) four-factor
model (C4) that adds the momentum factor to the FF3 model, Fama and French (2015) five-factor model
(FF5) that adds profitability and capital investment factors to the FF3 model, and the Fama and French
(2018) six-factor model (FF6) that adds the momentum factor to the FF5 model.
The standard Fama and MacBeth (1973) two-step procedure is used to cross-sectionally test each model.
In the first step, a time-series regression is run using one year of daily excess returns from January 1972 to
December 1972 for the test asset portfolios:
K
X
Rit − Rf t = αi + bik Fkt + eit , t = 1, . . . , T, (20)
k=1

where Rit − Rf t is the realized excess return on the ith test asset portfolio for day t; Rf t is the riskless
rate proxied by the Treasury bill rate; αi is the intercept term or alpha; Fkt are asset pricing factors for
k = 1, . . . , K factors; bik are corresponding K beta risk loadings for the ith portfolio; i = 1, . . . , N correspond
to the number of portfolios; t = 1, . . . , T is the estimation period; and eit ∼ iid (0, σi2 ).
In the second step, the following cross-sectional regression in the nest out-of-sample month of January
1973:

RiT +1 − Rf T +1 = α̂ + λ1 b̂i1 + λ2 b̂i2 + . . . + λK b̂iK + uiT +1 , i = 1, . . . , N, (21)

where RiT +1 − Rf T +1 is the realized excess return on the ith portfolio in out-of-sample (one-month-ahead)
month T + 1; α̂ is the intercept; b̂ik , k = 1, . . . , K are K beta risk loadings estimated for N anomaly portfolios
in the estimation period t = 1, . . . , T ; λk are estimated market prices of beta risk loadings14 ; and uiT +1 are
zero mean and independent of the independent variables.15
We roll the two-step procedure forward one month at a time to the end of the sample period in December
2024. Monthly time-series estimates of the market prices of risk, or λk s, from January 1973 to December
2024 are averaged and associated t-statistics computed. To study average mispricing errors for each model,
14 See Cochrane (1996), Back, Kapadia, and Ostdiek (2013, 2015), Ferson (2019), and others.
15 For the MVP model with no intercept and EM ZCAPM, the zeta risk coefficient or factor loading is multiplied by 21 to
rescale the loading from a daily to monthly basis (i.e., 21 trading days per month). See Kolari, Liu, and Huang (2021, p. 127).

7
we generate monthly realized and predicted (one-month-ahead) returns for test asset portfolios from January
1973 to December 2024. In line with Jagannathan and Wang (1996) and Lettau and Ludvigson (2001), a
simple cross-sectional regression of average realized returns on average predicted returns is estimated to
obtain an R2 goodness-of-fit value. Also, average realized returns are graphically plotted against predicted
returns to visually show average mispricing errors for test asset portfolios (i.e., their dispersion above or
below a 45 degree line in the graph).

4.3 Cross-sectional regression results

Table 2 reports the cross-sectional regression results for the CAPM, MVP with intercept, MVP (no intercept),
and EM ZCAPM. In Panel A, we see that the MVP with intercept model boosts the estimated R2 of the
CAPM from 40 percent to 51 percent. Also, the estimated market price of zeta risk loadings related to the
market return dispersion factor in the MVP with intercept model, or λ̂zeta , is not significant. By contrast, for
the MVP model (with no intercept), λ̂zeta is highly significant (i.e., t = 3.24) with an economically meaningful
risk premium of 0.42 percent per month. Also notice that the R2 value markedly increases to 79 percent.
While the EM ZCAPM outperforms the other models with R2 of 90 percent and t = 4.02, the MVP model
results are similar to the EM ZCAPM
Panel B of Table 2 shows the results for 25 profit-investment portfolios. Here the CAPM is not much
improved by the MVP with intercept model. Adding the market return dispersion factor to the market
factor only increases the estimated R2 from 29 percent to 30 percent, and associated zeta risk loading are
insignificant. Notably, the MVP model has a very high R2 value equal to 93 percent that is close to the EM
ZCAPM at 95 percent. This near perfect goodness-of-fit is exceptional compared to other multifactor models
in published cross-sectional regression tests. Estimated magnitudes of market prices of zeta risk loadings, or
λ̂zeta , ar 0.28 percent and 0.33 percent per month for the MVP model and EM ZCAPM, respectively.
In Panel C for 48 industry portfolios, which are exogenous test assets unrelated to any market or firm
characteristic factors, we see that the MVP model dominates the other models with R2 at 64 percent that
well exceeds the EM ZCAPM at 50 percent. The λ̂zeta estimate is 0.34 percent per month for both the MVP
model and EM ZCAPM, both highly significant with t = 4.06 and 4.19, respectively. Moreover, the CAPM
and MVP with intercept model perform poorly with respect to these exogenous assets with R2 values of only
5 percent at most.
Lastly, Panel D gives the results for the 153 anomaly portfolios. These test assets should be difficult to
price as researchers have intensively searched for long/short portfolios of stocks that are not well explained
by asset pricing models. The MVP model produces an R2 value equal to 59 percent compared to 68 percent
for the EM ZCAPM. The CAPM and MVP with intercept model offer no explanatory power with R2 values
of only 2 percent at most.
In unreported results, we repeat these tests for 153 anomalies for 133 anomalies from Chen and
Zimmerman (2022). Their sample period is July 1972 to December 2021. The estimated R2 values for the
MVP model is fairly high at 86 percent compared to 83 percent for the EM ZCAPM. By contrast, the CAPM
and MVP with intercept model yielded at most a 2 percent R2 value.
Referring to the MVP model and EM ZCAPM in Panel D of Table 2, the t-statistics associated with the
estimated market prices of zeta risk loadings, or λ̂zeta , are 5.94 and 5.96, respectively. Moreover, estimated
zeta risk premiums are relatively high at 0.59 percent per month for both models. These zeta risk loading
results well exceed those of various risk loadings tested by researchers in extant multifactor model literature.

8
In this regard, to avoid false discoveries in asset pricing, Harvey, Liu, and Zhu (2016) and Chordia, Goyal, and
Saretto (2020) have recommended a t-statistic greater than 3. Because the t-statistics associated with zeta
risk loadings are well beyond this threshold at close to 6, we can reliably infer that market return dispersion
is a salient asset pricing factor. Confirming these results, in unreported results for 133 anomalies in Chen and
Zimmerman (2022), the t-statistics related to zeta risk loadings in the MVP model and EM ZCAPM equal
6.80 and 6.67, respectively, and estimated zeta risk premiums are 0.74 percent and 0.72 percent per month.
In all four sets of test asset portfolios in Table 2, with the exception of the 153 anomalies in Panel D for
the MVP model and EM ZCAPM, the market price of beta risk, or λ̂beta , is insignificant. At least from a
cross-sectional standpoint, these results suggest that changes in the market factor (corresponding to changes
in the mean level of the investment parabola) are normally not useful in pricing stock portfolios under study.
However, the market price of zeta risk, or λ̂zeta , is always significant in the MVP model and EM ZCAPM
(but not the MVP with intercept model). Thus, upon dropping the intercept in the regression model, changes
in market return dispersion have significant impacts on the cross section of average stock returns.
For comparative purposes, Table 3 provides cross-sectional regression results for the prominent Fama and
French three-, five-, and six-factor models and the Carhart four-factor model (denoted FF3, FF5, FF6, and
C4, respectively). These models perform fairly well for the 25 size-BM and 25 profit-investment portfolios
in Panels A and B, respectively. For example, the R2 values for the FF6 model for these two sets of firm
characteristic portfolios are relatively high at 70 percent and 88 percent, respectively. However, a problem
with interpreting these results is that the test assets are endogenous with respect to the factors in the FF6
model to a considerable extent. Daniel and Titman (2012) have pointed out that cross-sectional tests have low
power when test assets are highly correlated with proposed factors by construction using firm characteristics
in both instances.
Turning to the results in Panels C and D of Table 3 that are based on exogenous test assets unrelated
to the factors in the models, we see that the FF6 model achieves much lower R2 values of 34 percent and
32 percent for 48 industry and 153 anomaly portfolios, respectively. These results confirm the Daniel and
Titman’s caution about test assets that are correlated by construction with multifactors.

4.4 Graphical mispricing error results

Fama and MacBeth (1973) have stated that a normative model uses past information to explain future returns.
Consistent with this approach, Cochrane (1996) and Lettau and Ludvigson (2001) have recommended that
researchers examine mispricing errors of models by graphically comparing their out-of-sample predicted to
realized (actual) returns. As discussed earlier in Section 4.2, we compute average mispring errors for different
asset pricing models and test asset portfolios.
For selected models, we graphically display average mispricing error results for exogenous industry and
anomaly portfolios. Figures 3 to 5 provide the results for 48 industry portfolios. The CAPM is compared to
the FF3 model in Figure 3; as shown there, average mispricing errors are large for the CAPM but somewhat
improved by FF3. In Figure 4, FF5 results look similar to those for FF3 but FF6 shows noticeably lower
average mispricing errors. Lastly, by casual inspection of Figure 5, it is obvious that average mispricing errors
are reduced further by the EM ZCAPM and markedly so by the MVP model.
Using the 153 anomaly portfolios, Figures 6 to 8 illustrate the average pricing errors for different models.
In Figure 6, both FF3 and C4 models exhibit fairly large average mispricing errors. As shown in Figure 7,
the FF5 and FF6 models do not improve mispricing errors for the most part. However, the EM ZCAPM and
MVP model graphs in Figure 8 clearly reduce average mispricing errors compared to prominent multifactor

9
models. The higher density of portfolios near the 45 degree line is evident for these ZCAPM-based models.
Lastly, Figure 9 shows the average mispricing errors for the EM ZCAPM and MVP model using the 133
anomaly portfolios in Chen and Zimmerman (2022). Unlike other test assets under study, the sample period
is July 1972 to December 2021. There we see that average mispricing errors for these ZCAPM-based models
are even smaller than those in Figure 8 for 153 anomalies in Jensen, Kelly, and Pedersen (2023). Indeed, these
mispricing results are exceptional in the sense that they suggest that anomaly portfolios are not anomalous
to our ZCAPM-based models. In effect, on an out-of-sample basis that eliminates potential manipulation
of results, risk factors in the ZCAPM are able to almost completely explain long/short anomalies in line
with the efficient market hypothesis of Fama (1970, 2013). By implication, behavioral finance explanations
(grounded in psychological biases of investors) are not needed to explain these anomalies.

5 Conclusion
This paper proposed a mean variance pricing (MVP) model to estimate the empirical ZCAPM. Previous
studies employed the relatively complex iterative expectation-maximization (EM) algorithm to estimate the
regression model for the ZCAPM. By contrast, our MVP model is easy to estimate using standard regression
methods and, therefore, can be readily understood by researchers unfamiliar with EM regression methods.
Unlike ordinary least squares (OLS) regression, a zero intercept is imposed in the MVP model. The rationale
for this restriction is that, when the macrofactors in the ZCAPM comprised of cross-sectional mean market
returns and their standard deviation are zero, the expected returns for all assets must be zero.
In out-of-sample cross-sectional regression tests using different test asset portfolios and models, we
showed that the MVP model performed almost the same as the more complicated EM regression form of the
empirical ZCAPM (or EM ZCAPM). In comparative tests, zeta risk loadings were highly significant with
similar magnitudes in both regression approaches, and estimated R2 measures of goodness-of-fit were both
fairly high. Notably, adding an intercept to the MVP model yielded poor cross-sectional regression results.
Also, comparisons to a number of prominent multifactor models indicated that the MVP model (as well as
EM ZCAPM) consistently produced better cross-sectional regression results. Further graphical analyses of
different test asset portfolios provided evidence that average mispricing errors were similar for the MVP
model and EM ZCAPM but smaller compared to other multifactor models. We conclude that the MVP
model represents a parsimonious econometric approach to estimating the ZCAPM that is easy to implement
and understand.
A number of implications of the MVP model to future research are suggested. First, a parsimonious
two-factor model performed well compared to other multifactor models with more factors. As the number of
anomalies identified by researchers has increased over time, additional factors have been added to multifactor
models in an attempt to explain them. Commenting on these trends, Cochrane (2011) opined that a factor
zoo has emerged that complicates the field of asset pricing. The MVP model substantially reduces this
problem. Second, unlike the empirical EM ZCAPM, researchers can readily augment the MVP model with
other multifactors in an effort to possibly improve its ability to price different test assets. While no industry
factor exists in the literature, most professional investment managers consider industry risk to be important
to firm valuation. In our opinion, the inclusion of an industry factor would be advantageous to price industry
portfolios in particular and other assets in general. Third, applications of our empirical MVP model analyses
to other asset classes traded on a daily basis, including bonds, real estate, commodities, etc., as well as other
countries are warranted. Fourth, further tests of anomaly portfolios can be conducted in an effort to find

10
long/short portfolios that are anomalous (or not explained) by the MVP model. These portfolios would be
economically interesting to behavioral finance researchers, who can investigate abnormally high returns due
to potential psychological investor biases. Fifth, it is possible that conditional MVP model specifications that
allow time-variant beta and zeta risk parameters would be beneficial. These and other possible extensions of
our MVP model model are recommended.

11
References
Angelidis, T., A. Sakkas, and N. Tessaromatis, 2015, Stock market dispersion, the business cycle and expected
factor returns, Journal of Banking and Finance 59, 256–279.

Back, K., N. Kapadia, and B. Ostdiek, 2013, Slopes as factors: Characteristic pure plays, Working paper,
Rice University.

Back, K., N. Kapadia, and B. Ostdiek, 2015, Testing factor models on characteristic and covariance pure
plays, Working paper, Rice University.

Bansal, R., D. Kiku, I. Shaliastovich, and A. Yaron, 2014, Volatility, the macroeconomy, and asset prices,
Journal of Finance 69, 2471–2511.

Bansal, R., and A. Yaron, 2004, Risks for the long run: A potential resolution of asset pricing puzzles, Journal
of Finance 59, 1481–1509.

Barillas, F., and J. Shanken, 2018, Comparing asset pricing models, Journal of Finance 73, 715–754.

Bekaert, G., and C. Harvey, 1997, Emerging equity market volatility, Journal of Financial Economics 43,
29–77.

Black, F., 1972, Capital market equilibrium with restricted borrowing, Journal of Business 45, 444–454.

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

Chen, Y., M. Cliff, and H. Zhao, 2017, Hedge funds: The good, the bad, and the lucky, Journal of Financial
and Quantitative Analysis 52, 1081–1109.

Chen, A. Y., and T. Zimmermann, 2020, Publication bias and the cross-section of stock returns, Review of
Asset Pricing Studies 10, 249–289.

Chen, A. Y., and T. Zimmerman, 2022, Open source cross-sectional asset pricing Critical Finance Review 11,
207–264.

Chordia, T., A. Goyal, and A. Saretto, 2020, Anomalies and false rejections, Review of Financial Studies 33,
2134–2179.

Christie, W., and R. Huang, 1994, The changing functional relation between stock returns and dividend
yields, Journal of Empirical Finance 1, 161–191.

Cochrane, J. H., 1996, A cross-sectional test of an investment-based asset pricing model, Journal of Political
Economy 104, 572–621.

Cochrane, J. H., 2011, Presidential address: Discount rates, Journal of Finance 56, 1047–1108.

Connolly, R., and C. Stivers, 2003, Momentum and reversals in equity index returns during periods of
abnormal turnover and return dispersion, Journal of Finance 58, 1521–1556.

Copeland, T. E., and J. F. Weston, 1980, Financial Theory and Corporate Policy (Addison-Wesley Publishing
Company, Reading, MA).

Daniel, K., and S. Titman, 2012, Testing factor-model explanations of market anomalies, Critical Finance
Review 1, 103–139.

12
Demirer, R. and S. P. Jategaonkar, 2013, The conditional relation between dispersion and return, Review of
Financial Economics 22, 125–134.

Dempster, A. P., N. M. Laird, and D. B. Rubin 1977, Maximum likelihood from incomplete data via the EM
algorithm, Journal of the Royal Statistical Society 39, 1–38.

Fama, E. F., and K. R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47,
427–465.

Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of
Financial Economics 33, 3–56.

Fama, E. F., and K. R. French, 2015, A five-factor asset pricing model, Journal of Financial Economics 116,
1–22.

Fama, E. F., and K. R. French, 2018, Choosing factors, Journal of Financial Economics 128, 234–252. ‘

Fama, E. F., and J. D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests, Journal of Political
Economy 81, 607–636.

Ferson, W. E., 2019, Empirical Asset Pricing: Models and Methods (The MIT Press, Cambridge, MA).

Ferson, W. E., S. K. Nallareddy, and B. Xie, 2013, The "out-of-sample" performance of long-run risk models,
Journal of Financial Economics 107, 537–556 .

Garcia, R., D. Mantilla-Garcia, and L. Martellini, 2014, A model-free measure of aggregate idiosyncratic
volatility and the prediction of market returns, Journal of Financial and Quantitative Analysis 49, 1133–
1165.

Gibbons, M. R., S. A. Ross, and J. Shanken, 1989, A test of the efficiency of a given portfolio, Econometrica
57, 1121–1152.

Gomes, J., L. Kogan, and L. Zhang, 2003, Equilibrium cross section of returns, Journal of Political Economy
111, 693–732.

Harvey, C. R., and Y. Liu, 2016, Rethinking performance evaluation, Working paper no. 22134, National
Bureau of Economic Research, Cambridge, MA.

Harvey, C. R., Y. Liu, and H. Zhu, 2016, ... and the cross-section of expected returns, Review of Financial
Studies 29, 5–68.

Jagannathan, R., and Z. Wang, 1996, The conditional CAPM and the cross-section of asset returns, Journal
of Finance 51, 3–53.

Jensen, M. C.,1968, The performance of mutual funds in the period 1945-1964, Journal of Finance 23,
389–416.

Jensen, T. I., B. Kelly, and L. H. Pedersen, 2023, Is there a replication crisis in finance?, Journal of Finance
78, 2465–2518.

Jiang, X., 2010, Return dispersion and expected returns, Financial Markets and Portfolio Management 24,
107–135.

13
Jones, P. N., and G. J. McLachlan, 1990, Algorithm AS 254: Maximum likelihood estimation from grouped
and truncated data with finite normal mixture models, Applied Statistics 39, 273–282.

Kolari, J. W., J. Z. Huang, H. A. Butt, and H. Liao, 2022, International tests of the ZCAPM asset pricing
model, Journal of International Financial Markets, Institutions &Money 79, 101607.

Kolari, J. W., J. Z. Huang, W. Liu, and H. Liao, 2022, Further tests of the ZCAPM asset pricing model,
Journal of Risk and Financial Management 15, 1-23. Reprinted in Kolari, J. W., and S. Pynnonen, 2022,
eds., Frontiers of Asset Pricing (MDPI, Basel, Switzerland).

Kolari, J. W., J. Z. Huang, W. Liu, and H. Liao, 2025, A quantum leap in asset pricing: Explaining anomalous
returns, Journal of Risk and Financial Management 18, 362, [Link]

Kolari, J. W., W. Liu, and J. Z. Huang, 2021, A New Model of Capital Asset Prices: Theory and Evidence
(Palgrave Macmillan, Cham, Switzerland).

Kolari, J. W., W. Liu, J. Z. Huang, and H. Liao, 2026, Asset Pricing Models and Market Efficiency:
Using Machine Learning to Explain Stock Market Anomalies (Palgrave Macmillan, Cham, Switzerland),
forthcoming.

Kolari, J. W., W. Liu, and S. Pynnonen, 2024, Professional Investment Portfolio Management: Boosting
Performance with Machine-Made Portfolios and Stock Market Evidence (Palgrave Macmillan, Cham,
Switzerland).

Lettau, M., and S. Ludvigson, 2001, Consumption, aggregate wealth, and expected stock returns, Journal of
Finance 56, 815–849.

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.

Liu, W., 2013, A New Asset Pricing Model Based on the Zero-Beta CAPM: Theory and Evidence, Doctoral
dissertation, Texas A&M University.

Liu, W., J. W. Kolari, and J. Z. Huang, 2012, A new asset pricing model based on the zero-beta CAPM,
Best paper award in investments at the 2012 annual meetings of the Financial Management Association,
Atlanta, GA (October).

Loungani, P., M. Rush, and W. Tave, 1990, Stock market dispersion and unemployment, Journal of Monetary
Economics 25, 367–388.

Markowitz, H. M., 1952, Portfolio selection, Journal of Finance 7, 77–91.

Markowitz, H. M., 1959, Portfolio Selection: Efficient Diversification of Investments (John Wiley & Sons,
New York, NY).

Martin, I. and S. Nagel, 2021, Market efficiency in the age of big data, Journal of Financial Economics 45,
154–177.

McLachlan, G. J., and T. Krishnan, 2008, The EM Algorithm and Extensions, Second edition (John Wiley &
Sons, New York, NY).

McLachlan, G., and D. Peel, 2000, Finite Mixture Models (Wiley Interscience, New York, NY).

14
McLean, R. D., and J. Pontiff, 2016, Does academic publication destroy predictability?, Journal of Finance
71, 5–32.

Mossin, J., 1966, Equilibrium in a capital asset market, Econometrica 34, 768–783.

Pastor, L., and P. Veronesi, 2009, Technological revolutions and stock prices, American Economic Review 99,
1451–1483.

Roll, R., 1977, A critique of the asset pricing theory’s tests, part I: On past and potential future testability of
the theory, Journal of Financial Economics 4, 129–176.

Roll, R., 1980, Orthogonal portfolios, Journal of Financial and Quantitative Analysis 15, 1005–1012.

Sharpe, W. F., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk, Journal
of Finance 19, 425–442.

Simin, T., 2008, The poor performance of asset pricing models, Journal of Financial and Quantitative Analysis
43, 335–380.

Stivers, C., 2003, Firm-level return dispersion and the future volatility of aggregate stock market returns,
Journal of Financial Markets 6, 389–411.

Treynor, J. L., 1961, Market value, time, and risk, Unpublished manuscript.

Treynor, J. L., 1962, Toward a theory of market value of risky assets, Unpublished manuscript.

Zhang, L., 2005, The value premium, Journal of Finance 60, 67–103.

15
Table 1: Descriptive statistics for asset pricing factors
This table summarizes the descriptive statistics for the asset pricing factors. Daily factor returns (in percent) for
different asset pricing models are as follows: Rm − Rf (value-weighted CRSP index return minus Treasury bill rate),
σm (market return dispersion), SMB (size), HML (value), MOM (momentum), RMW (profit), and CMA (capital
investment). The sample period is from January 1972 to December 2024.

Statistic Rm − Rf σa SMB HML MOM RMW CMA

Mean 0.03 1.86 0.00 0.01 0.03 0.01 0.01
Standard deviation 1.07 0.65 0.58 0.61 0.81 0.42 0.39
Maximum 11.36 10.77 6.08 6.73 7.14 4.57 2.48
Minimum -17.44 0.68 -11.15 -5.03 -14.37 -2.97 -5.31

16
Table 2: Out-of-sample Fama-MacBeth cross-sectional tests for different models and test asset portfolios in
the sample period January 1972 to December 2024
This table reports the cross-sectional regression test results for different models and test asset portfolios. The models
are: CAPM, MVP with intercept, MVP, and EM ZCAPM. The CAPM and MVP model with intercept are estimated
via OLS regression methods. The MVP model (with no intercept) is estimated by means of regression through the
origin. The ZCAPM is estimated by means of expectation-maximization (EM) regression. The test asset portfolios from
Kenneth French’s website are: 25 size-BM (book-to-market), 25 operating profit-capital investment, and 48 industries.
Also, a total of 153 anomaly portfolios are downloaded from the data website of Jensen, Kelly, and Pedersen (2023).
The sample period is January 1972 to December 2024. First, time-series regression models are estimated using one year
of daily returns and rolled forward one month at a time. Second, monthly cross-sectional regressions are estimated
using one-month-ahead (out-of-sample) excess returns and factor loadings. We compute averages of the estimated
market prices of risk denoted λ̂k for the kth factor for all months from January 1973 to December 2024. Asterisks
on t-statistics in parentheses indicate the level of statistical significance: *–10 percent, **–5 percent, and ***–1 percent.

Panel A: 25 size-BM
Model α̂ λ̂beta λ̂zeta R2
CAPM 1.23 -0.53 0.40
(4.78***) (-2.08)
MVP with intercept 0.95 -0.29 -0.03 0.51
(3.75***) (-1.09) (-0.91)
MVP 0.91 -0.22 0.42 0.79
(3.63***) (-0.86) (3.24***)
EM ZCAPM 0.84 -0.15 0.46 0.90
(3.34***) (-0.60) (4.02***)
Panel B: 25 profit-investment
Model α̂ λ̂beta λ̂zeta R2
CAPM 0.71 -0.05 0.29
(2.95***) (-0.18)
MVP with intercept 0.61 0.04 -0.02 0.30
(2.47***) (0.16) (-1.27)
MVP 0.61 0.04 0.28 0.93
(2.51***) (0.13) (3.27***)
EM ZCAPM 0.58 0.08 0.33 0.95
(2.36***) (0.30) (4.17***)
Panel C: 48 industries
Model α̂ λ̂beta λ̂zeta R2
CAPM 0.42 0.26 0.04
(2.07***) (1.08)
MVP with intercept 0.31 0.32 -0.02 0.05
(1.53) (1.33) (-1.46)
MVP 0.18 0.49 0.34 0.64
(0.90) (2.05**) (4.06***)
EM ZCAPM 0.29 0.40 0.34 0.50
(1.44) (1.70*) (4.19***)
Panel D: 153 anomaly portfolios
Model α̂ λ̂beta λ̂zeta R2
CAPM 0.21 0.25 0.00
(11.51***) (0.92)
MVP with intercept 0.20 0.26 -0.00 0.02
(11.63***) (1.00) (-0.19)
MVP 0.16 0.51 0.59 0.59
(8.65***) (2.02**) (5.94***)
EM ZCAPM 0.15 0.46 0.59 0.68
(8.82***) (1.83*) (5.96***)

17
Table 3: Out-of-sample Fama-MacBeth cross-sectional tests for prominent multifactor models and test asset
portfolios in the sample period January 1972 to December 2024
This table reports the cross-sectional regression test results for prominent multifactor models and test asset portfolios.
The models are: Fama and French three-factor model (FF3), Carhart four-factor model (C4), Fama and French
five-factor model (FF5), and Fama and French six-factor model (FF6). The test asset portfolios from Kenneth
French’s website are: 25 size-BM (book-to-market), 25 operating profit-capital investment, and 48 industries. Also, a
total of 153 anomaly portfolios are downloaded from the data website of Jensen, Kelly, and Pedersen (2022). The
sample period is January 1972 to December 2024. First, time-series regression models are estimated using one year of
daily returns and rolled forward one month at a time. Second, monthly cross-sectional regressions are estimated
using one-month-ahead (out-of-sample) excess returns and factor loadings. We compute averages of the estimated
market prices of risk denoted λ̂k for the kth factor for all months from January 1973 to December 2024. Asterisks
on t-statistics in parentheses indicate the level of statistical significance: *–10 percent, **–5 percent, and ***–1 percent.

Panel A: 25 size-BM
Model α̂ λ̂beta λ̂size λ̂value λ̂mom λ̂prof it λ̂investment R2
FF3 1.04 -0.41 0.10 0.24 0.50
(5.21***) (-1.78*) (0.82) (1.98**)
C4 1.00 -0.36 0.11 0.22 0.22 0.50
(4.87***) (-1.59) (0.87) (1.79*) (0.84)
FF5 1.04 -0.39 0.12 0.24 0.27 0.16 0.69
(4.98***) (-1.71*) (0.95) (1.88*) (2.14**) (1.27)
FF6 1.10 -0.46 0.13 0.25 0.27 0.26 0.11 0.70
(5.29***) (-2.02**) 1.05 (1.96**) (1.01) (2.08**) (0.88)
Panel B: 25 profit-investment
Model α̂ λ̂beta λ̂size λ̂value λ̂mom λ̂prof it λ̂investment R2
FF3 0.21 0.42 0.03 0.12 0.34
(0.87) (1.55) (0.23) (0.93)
C4 0.34 0.29 0.03 0.09 0.35 0.47
(1.32) (1.07) (0.16) (0.70) (1.60)
FF5 0.21 0.43 0.27 0.07 0.27 0.19 0.86
(0.80) (1.53) (1.58) (0.58) (2.94***) (2.29**)
FF6 0.35 0.28 0.27 0.06 0.03 0.29 0.19 0.88
(1.27) (0.98) (1.58) (0.47) (0.13) (3.17***) (2.22**)
Panel C: 48 industries
Model α̂ λ̂beta λ̂size λ̂value λ̂mom λ̂prof it λ̂investment R2
FF3 0.32 0.30 0.05 0.14 0.13
(1.53) (1.17) (0.33) (1.05)
FF4 0.33 0.29 0.03 0.07 0.45 0.32
(1.57) (1.19) (0.24) (0.55) (2.10**)
FF5 0.28 0.35 0.05 0.13 0.27 0.17 0.13
(1.31) (1.40) (0.36) (1.04) (2.50**) (1.62)
FF6 0.43 0.18 0.04 0.11 0.44 0.25 0.12 0.34
(2.04**) (0.76) (0.26) (0.88) (2.04**) (2.33**) (1.15)
Panel D: 153 anomaly portfolios
Model α̂ λ̂beta λ̂size λ̂value λ̂mom λ̂prof it λ̂investment R2
FF3 0.20 0.61 -0.19 0.13 0.16
(12.90***) (2.77***) (-1.28) (0.90)
FF4 0.18 0.50 -0.17 -0.17 0.31 0.19
(12.89***) (2.34**) (-1.17) (1.38) (1.70*)
FF5 0.17 0.71 -0.13 0.10 0.14 0.27 0.30
(13.12***) (3.00***) (-0.95) (0.75) (1.08) (2.31**)
FF6 0.16 0.46 -0.10 0.18 0.33 0.16 0.21 0.32
(13.94***) (1.93*) (-0.70) (1.34) (1.85*) (1.19) (1.82*)

18
 𝐸(𝑅 ሻ

M
𝐸(𝑅𝑀 ሻ •

•G

E(𝑅𝑍 ሻ •Z

0
𝜎2 𝑅

Figure 1: Black’s zero-beta CAPM: Locating orthogonal zero-beta portfolio Z

19
 𝐸(𝑅𝑃 )

𝐸(𝑅𝐼∗ )
• I*
Cross-sectional
σa standard deviation
of returns in the
market
𝐸(𝑅𝐺 ) •G
E(Ra) Mean market
𝐸(𝑅𝑍𝐼∗ ) • ZI* return near
the axis of symmetry

0  I2* =  ZI2 * σ2𝑃

Time-series return variance

Figure 2: Orthogonal portfolios I ∗ and ZI ∗ with equal time-series return variance of returns
Source: Adapted from Kolari, Liu, and Huang (2021, Figure 3.2, p. 59).

20
 A. CAPM B. Fama and French three-factor model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 3: Out-of-sample cross-sectional CAPM (Panel A) and Fama and French three-factor model (Panel B)
mispricing errors comparing average one-month-ahead realized excess returns in percent (Y-axis) to average
one-month-ahead predicted (fitted) excess returns in percent (X-axis) for 48 industry portfolios. The sample
period is January 1972 to December 2024.

21
 A. Fama and French five-factor model B. Fama and French six-factor model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 4: Out-of-sample cross-sectional Fama and French five-factor model (Panel A) and six-factor model
(Panel B) mispricing errors comparing average one-month-ahead realized excess returns in percent (Y-axis)
to average one-month-ahead predicted (fitted) excess returns in percent (X-axis) for 48 industry portfolios.
The sample period is January 1972 to December 2024.

22
 A. EM ZCAPM B. MVP model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 5: Out-of-sample cross-sectional EM ZCAPM (Panel A) and MVP model (Panel B) mispricing errors
comparing average one-month-ahead realized excess returns in percent (Y-axis) to average one-month-ahead
predicted (fitted) excess returns in percent (X-axis) for 48 industry portfolios. The sample period is January
1972 to December 2024.

23
 A. Fama and French thee-factor model B. Carhart four-factor model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 6: Out-of-sample cross-sectional Fama and French three-factor model (Panel A) and Carhart four-factor
model (Panel B) mispricing errors comparing average one-month-ahead realized excess returns in percent
(Y-axis) to average one-month-ahead predicted (fitted) excess returns in percent (X-axis) for 153 anomaly
portfolios from Jensen, Kelly, and Pedersen (2023). The sample period is January 1972 to December 2024.

24
 A. Fama and French five-factor model B. Fama and French six-factor model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 7: Out-of-sample cross-sectional Fama and French five-factor model (Panel A) and six-factor model
(Panel B) mispricing errors comparing average one-month-ahead realized excess returns in percent (Y-axis)
to average one-month-ahead predicted (fitted) excess returns in percent (X-axis) for 153 anomaly portfolios
from Jensen, Kelly, and Pedersen (2023). The sample period is January 1972 to December 2024.

25
 A. EM ZCAPM B. MVP model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 8: Out-of-sample cross-sectional EM ZCAPM (Panel A) and MVP model (Panel B) mispricing errors
comparing average one-month-ahead realized excess returns in percent (Y-axis) to average one-month-ahead
predicted (fitted) excess returns in percent (X-axis) for 153 anomaly portfolios from Jensen, Kelly, and
Pedersen (2023). The sample period is January 1972 to December 2024.

26
 A. EM ZCAPM B. MVP model
Realized E(Ri)

Fitted E(Ri) Fitted E(Ri)

Figure 9: Out-of-sample cross-sectional EM ZCAPM (Panel A) and MVP model (Panel B) mispricing errors
comparing average one-month-ahead realized excess returns in percent (Y-axis) to average one-month-ahead
predicted (fitted) excess returns in percent (X-axis) for 133 anomaly portfolios from Chen and Zimmerman
(2022). The sample period is July 1972 to December 2021.

27