Conditional Risk∗

Niels Joachim Gormsen and Christian Skov Jensen†

February 2024

Abstract

We study the extent to which time-variation in market betas influence estimates of

CAPM alphas. Given the observed variation in conditional market betas, market risk
premia, and market variance, the required compensation for conditional market risk
can, in theory, be as large as the unconditional equity premium. We implement the
conditional CAPM using state-of-the-art methods in a broad global sample. We find
that accounting for conditional risk helps explain the return on all the major anomalies
we consider and that conditional risk explains two percentage points of alpha for value,
investment, and momentum strategies in recent years.

Keywords: asset pricing, conditional CAPM, factor models, time-varying discount

rates.
JEL Codes: G10, G12

1
We are grateful for helpful comments from Malcolm Baker, Thummim Cho, Peter Christoffersen, Robin
Greenwood, Sam Hanson, Bryan Kelly, Eben Lazerus, Martin Lettau, Dong Lou, Stefan Nagel, Tobias
Moskowitz, Lasse Heje Pedersen, Andrei Shleifer, Emil Siriwardane, Adi Sunderam, Christian Wagner, and
Paul Whelan, as well as comments from seminar participants at Copenhagen Business School, the American
Finance Association, and Harvard Business School. Vanessa Hu provided excellent research assistance. Both
authors gratefully acknowledge support from the FRIC Center for Financial Frictions (grant no. DNRF102)
and the European Research Council (ERC grant no. 312417) and Niels Gormsen further acknowledge support
from the Asness Junior Faculty Fellowship, the Fama Faculty Fellowship, and the Fama-Miller Center at the
University of Chicago Booth School of Business.
2
Corresponding author is Niels Gormsen, University of Chicago Booth School of Business and Na-
tional Bureau of Economic Research, 5807 South Woodlawn Avenue, Office 401, 60637, Chicago, Illinois.
Phone: 773-834-8689. Email: [Link]@[Link]. Jensen is at Bocconi University, chris-
[Link]@[Link].

Electronic copy available at: [Link]

 According to the conditional CAPM, assets should have higher average returns if they
have high market betas during times when the market risk premium is high or the variance
is low. Following the seminal work by Lewellen and Nagel (2006), it is often argued that
such variation in market betas (conditional risk) cannot plausibly explain the return to the
major cross-sectional risk factors. The argument is often that the time-series variation in
conditional market betas, market risk premium, and market variance cannot be large enough
to materially influence the required return in the conditional CAPM.
Over the last decade, however, research in asset pricing suggests that conditional mo-
ments may be much more volatile than previously understood. Kelly and Pruitt (2013)
and Martin (2017), for instance, provide evidence that the equity premium is “extremely
volatile”, exhibits substantial high-frequency variation, and generally varies more than pre-
viously understood. Kelly, Moskowitz, and Pruitt (2021), moreover, argue that conditional
betas of dynamic trading strategies can be highly volatile, even at very short horizons. Fi-
nally, Moreira and Muir (2017) find that expected returns and volatility on the market are
not as highly correlated as previously expected, which – for technical reasons explained below
– increases the scope for conditional risk to explain asset returns.
Motivated by this apparent change in our understanding of the dynamics of conditional
moments, it is worthwhile revisiting the question of how large a role conditional risk plays
in the cross-section of stock returns. To do so, we first quantify the plausible effect of
conditional risk in the CAPM. When considering the conditional moments uncovered by the
above literature, conditional risk can in theory explain as much as five percentage points of
returns per year for a representative strategy, depending on the correlation between betas,
the risk premium, and variance. These five percentage points are similar in magnitude to the
full equity risk premium and to the unconditional CAPM alphas of equity factors studied
in the literature, meaning there is indeed scope for the conditional CAPM to explain equity
factors. Given the theoretical possibility of a large role for conditional risk, it becomes an
empirical question whether or not conditional risk can explain equity factors. We, therefore,
set out to quantify the exact impact of conditional risk on equity anomalies based on state-
of-the-art methods.
We find that conditional risk is a pervasive feature of the data. Conditional risk helps
explain a meaningful part of the alpha for the major risk factors we consider. For the value
factor, one of the most intensely studied factors in this context, conditional risk explains a
fourth of the unconditional alpha. In the most recent period, where risk premia are arguably
more volatile, conditional risk plays a relatively larger role, explaining almost all of the alpha

Electronic copy available at: [Link]

for the value factor. For most of our regressions, controlling for conditional risk has a larger
impact on alphas than controlling for unconditional risk. Taken together, the results suggest
that conditional risk can be large enough to have a material impact on the alpha of equity
risk factors.
Despite this impact of conditional risk, the conditional CAPM is easily rejected. The
tangency portfolio spanned by the major equity factors remains highly significant after we
control for conditional risk. In this sense, our results strongly support the main point of
Lewellen and Nagel (2006), namely that the conditional CAPM does not explain the cross-
section of stock returns. Moreover, the tests explained above are tests of the unconditional
predictions of the conditional CAPM. Testing the richer conditional implications leads to
even stronger evidence against the CAPM (see also Nagel and Singleton 2011 and Roussanov
2014).
However, the rejection of the conditional CAPM does not necessarily mean that one
should ignore conditional risk: in many factor analyses, we want to control for market risk,
and to do so properly, even though the CAPM itself does not perfectly price assets. Through-
out our tests, conditional market risk has a bigger impact on alphas than unconditional
market risk, emphasizing the relevance of controlling for conditional risk when implementing
the CAPM.

Methodology and high-level summary

Our analysis begins by revisiting the theoretical analysis in Lewellen and Nagel (2006) (LN).
This analysis studies how much conditional risk can explain as a function of the time-series
volatility in conditional betas, conditional market risk premia, and conditional variance.
When using the variation in betas assumed by LN, along with estimates of market risk
premium volatility from Kelly and Pruitt (2013), we find that conditional market timing
(i.e. covariance between market betas and market risk premia) in principle can explain up
to 2.9 percentage points of annualized return. Similarly, volatility timing can explain up to
7 percentage points of annualized return under the most aggressive assumptions.
Market and volatility timing can thus be qualitatively relevant in factor analysis when
considered in isolation. Previous research has argued that one cannot easily combine the
effects of market and volatility timing. In fact, past work has argued that market risk
premia and market variance are likely highly positively correlated, causing market timing
and volatility timing to mechanically counteract each other. We should thus expect a modest
impact of conditional risk even if market timing and volatility can have a large impact when

Electronic copy available at: [Link]

considered in isolation. Recent work by Moreira and Muir (2017), however, provides evidence
that the two moments are not perfectly correlated. Market and variance timing may thus,
in principle, work together to create substantial conditional risk. In particular, if one is
willing to assume that conditional market risk premia and variance are uncorrelated, as we
find empirically, it is possible for market timing and volatility timing to jointly explain up
to 5 percentage points of return.1 Empirically, we find that the two sources indeed tend to
strengthen each other, as further elaborated on below.
Motivated by these simple calculations, we set out to quantify the impact of conditional
risk on anomalies using state-of-the-art methods. Our main methodology implements the
CAPM using scaled factors, where the conditioning variables (instruments) are the condi-
tional market return, the conditional market variance, and the rolling betas of the test assets.
In theory, the conditional market risk premium and variance should be sufficient to capture
conditional risk if we measure the two perfectly. However, we include rolling betas in case
our measures are imperfect.
A key input to our analysis is the conditional market risk premium. We rely on state-of-
the-art methods that capture all of the variation in expected returns, including the very short
horizon variation that would not be captured by rolling betas. Recent research offers a series
of candidate estimators of the conditional market risk premium,2 some of which are limited
to the U.S. or the recent sample. In our main analysis, we rely on the three-stage estimator
of Kelly and Pruitt (2013), because this estimator can be implemented in all the countries
in our sample and over the full sample length, and because it is proven to forecast returns
well both in- and out-of-sample. As to variance, we estimate this based on the assumption
that it follows an AR(1) process.
Based on these moments, we quantify the impact of conditional risk in the cross-section
of U.S. and global equities. We start with the value factor, HML, which has been studied
extensively in the past. As is well known, the HML factor of Fama and French (1993) has
substantial CAPM alpha in the 1964-2022 sample. Controlling for conditional risk explains
around 1.25 percentage points of annual return in the US sample, which amounts to around
27% of the total alpha on the factor. In comparison, controlling for unconditional market
exposure explains only .86 basis points of annual alpha. As such, conditional risk cannot
1
As detailed below, combining volatility and market timing cannot plausibly lead to a much bigger effect
of conditional risk than volatility timing on its own. The reason is that correlation risk premia and variance
cannot both be perfectly (negatively) correlated with betas, if the two themselves are uncorrelated.
2
See e.g. Lettau and Ludvigson (2001a); Campbell and Thompson (2008); Binsbergen and Koijen (2010);
Kelly and Pruitt (2013); Martin (2017).

Electronic copy available at: [Link]

explain the full alpha to the factor, but it explains a meaningful amount.
We find qualitatively similar results in our broad global sample. Conditional risk helps
explain the value factor in most of the large countries. In France, Germany, and Sweden,
conditional risk explains essentially the entire value premium, which is around 2 percentage
points in these countries. In our pooled global sample, conditional risk explains half of the
alpha for the value strategy.
Conditional risk also matters for other factors than the value factor. In our long U.S. sam-
ple, we find that the risk factors profitability, investment, momentum, and betting against
beta all load positively on conditional risk, in the sense that controlling for conditional risk
lowers the estimated alpha. We find similar results in our broad global sample of 23 countries.
In the global sample, all the major risk factors, except profit, load on our conditional-risk
factor, and the conditional-risk factor explains on average 18% of their unconditional CAPM
alpha.
As argued in the introduction, recent research suggests that conditional moments have
been particularly volatile over the last few decades. It is thus possible that conditional
risk has been pronounced over this period. Based on this observation, we study conditional
risk in the post-1996 sample. This is the period in which the equity premium has been
documented to be particularly volatile – and it also happens to be almost completely out of
sample relative to the first analyses of conditional risk in the value factor.
When considering this out-of-sample period, we find an even more pronounced role for
conditional risk. Conditional risk generally explains 2 percentage points of annual alpha
to the major risk factors. This amount is large relative to the unconditional market risk
premium and the alpha on the factors. For instance, it amounts to almost all of the alpha
on the value factor and half the alpha on the investment factor. For momentum and betting
against beta strategies, conditional risk explains around 30% and 10%.
As explained above, previous research has argued that market and volatility timing gener-
ally should counteract each other when estimating conditional risk. To test this assumption,
we introduce two new risk factors that can explicitly detect market and variance timing.
Using these factors, we find that most equity risk factors exhibit both market and volatility
timing. That is, contrary to the previous arguments, market and volatility timing appear
to work together to generate conditional risk. This finding may help explain why condi-
tional risk can account for a larger fraction of expected returns than previously argued. At
a technical level, this complimentary role of market and volatility timing is possible only
because equity premia and variances are not perfectly correlated. This view is supported, as

Electronic copy available at: [Link]

mentioned earlier, by recent work of Moreira and Muir (2017).
We also implement alternative methods by Boguth, Carlson, Fisher, and Simutin (2011)
and Lewellen and Nagel (2006). Using these methods, we again find a large role for condi-
tional risk in our more recent sample. The effect is not always as large as our main methods
suggest, but the effect of conditional risk is similar in magnitude. This finding emphasizes
the robustness of our results across methodologies, but it also highlights the importance of
including the best possible estimates of equity premia and variance to properly account for
conditional risk.
The paper proceeds as follows. Section 1 covers the theory behind conditional risk in
factor models, quantifies the plausible impact of conditional risk, and develops new factors to
account for market and variance timing. Section 2 covers data and methodology. Section 3
quantifies the effect of conditional risk on estimates of alpha for major risk factors. Section 4
provides additional tests using managed portfolios as tests assets. The section also considers
conditional risk with respect to other risk factors. Section 5 discusses the results in relation
to previous implementations of conditional factor models. Section 6 concludes.

1 Theory
In this section, we first introduce the conditional CAPM and define terms in Section 1.1. We
next discuss the plausible effect of conditional risk on equity factors in Section 1.2. In Section
1.3, we introduce conditional-risk factors that researchers can use to quantify conditional risk
when observing conditional market risk premia and variance. These risk factors also allow
for precise estimation of the contribution of market and volatility timing.

1.1 Conditional Risk in the CAPM

The conditional CAPM is the following statement:
i m
i covt (rt+1 ; rt+1 ) m
Et [rt+1 ]= m
Et [rt+1 ] (1)
vart (rt+1 )

i
where rt+1 is the excess return to asset i between period t and t + 1, with m indexing the
market, and Et is the conditional expectation at time t.
To quantify the conditional risk in the CAPM, note first that taking unconditional ex-

Electronic copy available at: [Link]

pectations of (1) gives

i m m


E[rt+1 ] = E[βt ]E[rt+1 ] + cov βt ; Et [rt+1 ] (2)

We show in the Appendix that the average beta can be written as

m
 
vart (r̃t+1 )
E[βt ] = β̃ − cov βt ; m
(3)
var(r̃t+1 )

m m m
where r̃t+1 = rt+1 − Et [rt+1 ] is the shock to the market portfolio and

i m
cov(rt+1 ; r̃t+1 )
β̃ = m
(4)
var(r̃t+1 )

is the asset’s unconditional shock-beta. Inserting (3) into (2) gives

i m m m


E[rt+1 ] = β̃E[rt+1 ] + cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) (5)
| {z }
Conditional Risk

where the covariance term summarizes the conditional risk and

m
E[rt+1 ]
b= m
(6)
var(r̃t+1 )

is the unconditional price of risk. The expression intuitively conveys what conditional risk is:
conditional risk is the tendency for an asset to have a higher conditional beta when either the
conditional market risk premium is high or the conditional market variance is low. Lewellen
and Nagel (2006), and the literature in general refers to these terms as market and volatility
timing.
The expression for conditional risk in (5) features conditional betas, but we do not need
to observe these conditional betas to calculate conditional risk: we only need the part of
conditional betas that is spanned by the conditional market risk premium and variance. In
fact, there is an intuitive factor representation that captures the effect of time-varying betas.
Defining the conditional risk factor as

m
ct+1 = r̃t+1 (bt − b) (7)

Electronic copy available at: [Link]

where
m
Et rt+1
bt = m
vart (r̃t+1 )

is the conditional price of risk, we show in the appendix that can write the covariance term
for conditional risk in (5) as

m m i


cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) = cov(rt+1 ; ct+1 ) (8)

and therefore write equation (5) as

i m i
E[rt+1 ] = β̃E[rt+1 ] + cov(rt+1 ; ct+1 ) (9)
| {z }
Conditional Risk

Equation (9) shows that the impact of conditional risk on unconditional expected returns
can be completely summarized through the covariance with a conditional risk factor. This
factor is akin to a market timing strategy that is more exposed to the shock to the market
when the conditional risk price of risk is high relative to the unconditional average. Section
1.3 further expands on how one can use conditional-risk factors to capture conditional risk in
unconditional implementations of factor models. Before doing so, we quantify the plausible
effect of conditional risk on unconditional risk premia.

1.2 How Much Can Conditional Risk Explain? Revisiting Lewellen

and Nagel (2006)
The expression in equation (5) shows that conditional risk is the sum of two components:
(1) the unconditional covariance between conditional betas and conditional market risk pre-
mia and (2) the unconditional covariance between conditional market betas and conditional
variance,

i m m m


E[rt+1 ] − β̃E[rt+1 ] = cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) (10)
m m



= cov βt ; Et [rt+1 ] + cov βt ; −b vart (r̃t+1 ) (11)
| {z } | {z }
Market timing Volatility timing

The expression states how much of the unconditional return premium on a given asset can
be accounted for by conditional risk. The amount explained by market timing is determined

Electronic copy available at: [Link]

by the volatility of conditional market betas, the volatility of the conditional market risk
premium, and the correlation between the two. Similarly, the amount explained by volatility
timing is determined by the volatility of conditional market betas, the volatility of the
conditional market variance, the correlation between the two, and the unconditional price of
risk b.
To quantify the effect of conditional risk, we first examine how much these conditional
moments may vary over time. Recent work by Kelly and Pruitt (2013) and Martin (2017)
suggests that the market risk premium is “extremely volatile,” even at very short horizons.
Martin finds that the volatility of the monthly horizon market risk premium is around 0.4%.
These results are obtained in the post-1996 sample, which is the period where the option
prices studied by Martin are available. In our paper, we instead focus on the measure by
Kelly and Pruitt (2013), which we describe in detail in the upcoming Section 2.1. The market
risk premium coming from this measure similarly has a standard deviation of 0.4% in the
post-1996 sample. In the full sample, the standard deviation is 0.35%, reflecting a slightly
less volatile equity premium in the earlier parts of the sample.
To understand how these estimates relate to previous estimates, Figure 1 plots the ex-
pected return from the Kelly and Pruitt-measure along with a traditional estimate of ex-
pected returns. The traditional estimate is the inverse of the CAPE ratio plus the expected
inflation from the Michigan survey. The figure shows that the measure from Kelly and
Pruitt (2013) is notably more volatile than the estimate from the CAPE ratio. The Kelly
and Pruitt-measure has a standard deviation of 0.5% in this sample, whereas the measure
from the CAPE has a standard deviation of only 0.3%. The two estimates comove substan-
tially over time, with a correlation of 0.5. Importantly, there is much more short-horizon
variation in the measure by Kelly and Pruitt. This short horizon makes it important to use
methods that can capture the effects of such short-horizon variation in market risk premia.
LN consider the possibility that the volatility of the equity premium is as high as docu-
mented above. In their calculations, they consider a volatility of the equity premium as high
as 0.5%. In this sense, the new sample does not lead to substantially higher estimates of
the volatility of the market risk premia than those entertained by past research. However,
as Figure 1 shows, this volatility was historically considered to come from long-run fluctu-
ations in expected returns (see also discussion in Campbell and Cochrane 1999), meaning
conditional risk could only matter over regressions with long samples. The new estimates,
however, suggest substantial variation over short horizons, which means that conditional risk
can matter even in short-period regressions and that one must, when estimating conditional

Electronic copy available at: [Link]

risk, use methods that can capture the effects of such short-horizon variation in market risk
premia.
We next consider how much conditional betas may vary over time. We use two estimates
of conditional market betas: conditional rolling betas (calculated using the Fama and French
(1992) method) along with conditional betas using the state-of-the-art machinery from Kelly,
Moskowitz, and Pruitt (2021). Figure 2 Panel A plots a histogram of the time-series volatility
in these betas across firms. The figure shows that the volatility of conditional betas is around
0.2 on average when using the Kelly, Moskowitz, and Pruitt (2021) betas and 0.3 when using
rolling betas. However, we cannot rule out a substantially higher variation for subsets of
stocks. Considering the volatility of strategy-level betas yields roughly similar results.
While Kelly, Moskowitz, and Pruitt (2021) betas are not more volatile overall than rolling
betas, they vary more on the very short horizon for the median firm. To visualize this, Figure
2 Panel B plots the standard deviation of monthly changes in betas. Here, the volatility of
Kelly, Moskowitz, and Pruitt (2021) betas for the median firm is above that obtained using
rolling betas. The average volatility of the changes is, however, fairly similar across the
two measures. In addition, Kelly, Moskowitz, and Pruitt (2021) show that the variation in
betas is not idiosyncratic across firms and can translate into substantial variation in factor
portfolios. Taken together, the new research does not change our estimates the of volatility
of the conditional market betas, but they highlight that the betas can be volatility on the
very short horizon, meaning the methods we use to test for conditional risk must be able to
account for such variation.
In Table 1, we study how much market and volatility timing can plausibly influence esti-
mates of required returns in the CAPM. We calculate conditional risk for a range of different
values of the volatility of conditional moments, motivated by the discussion above. We first
consider the effect of market timing. In our first calculations, we consider a correlation
between conditional betas and market risk premia of 1 to obtain an upper bound on the
effect of market timing. Assuming that the volatility of market risk premium is 0.6 and
the volatility of market betas is 0.4, which is at the very high end based on the discussions
above, conditional market timing can account for 2.9 percentage points of annual return
on the strategies. This is a substantial amount, given that the average factor in the Fama
and French model has an unconditional CAPM alpha of 3.7 percentage points per year (and
given the unconditional equity premium of 5 to 6 percentage points).
We next consider volatility timing. The impact of volatility timing is given by the negative
covariance between conditional market betas and conditional market variance, multiplied by

Electronic copy available at: [Link]

the unconditional price of risk b. Empirically, this price of risk is around 2.5. Because
variance and risk premia vary roughly by the same amount over time, the multiplication on
the volatility timing term leads to a higher potential effect of volatility timing than expected
return timing. Table 1 shows that, assuming a correlation of -1 between conditional betas
and variance, volatility timing may explain as much as 7 percentage points of annual return
under the most aggressive assumptions. This is large in magnitude, larger than the alpha
to most risk factors, and similar to the equity risk premium itself. These estimates are
consistent with those in Boguth, Carlson, Fisher, and Simutin (2011).
As a novel part of the quantification, we consider the joint effect of market and volatility
timing. When considering the joint effect of market and volatility timing, a key determinant
is the correlation between the conditional market risk premium and variance. If the two are
perfectly correlated, then market and volatility timing must counteract each other: any asset
that loads positively on market timing must load negatively on volatility timing. However, if
the two are imperfectly correlated, or even uncorrelated, the two effects need not counteract
each other. As we shall see in the upcoming empirical analysis, the conditional market risk
premium and variance are close to uncorrelated under our main measure, meaning the two
need not counteract each other.
To estimate an upper bound on the joint impact of market timing and volatility timing,
we relax the assumption of a perfectly positive correlation between betas and market premia
and a perfectly negative correlation between betas and variance. We instead assume that the
correlation between conditional betas and conditional market risk premia is 0.5 and similarly
that the correlation between conditional betas and conditional variance is -0.5. We note that
under such a correlation structure, the conditional market premia and conditional variance
can remain uncorrelated.
Table 1 Panel B shows the joint effect market and volatility timing, i.e. conditional
risk, under the above assumptions. Based on the exact assumptions about the variance of
conditional moments, we see conditional risk explains up to 5 percentage points. This is
almost the entire equity premium, emphasizing that conditional risk can, in theory, be a
quite powerful force.
Another way to illustrate the potential importance of conditional risk is by comparing
the Sharpe ratios of the market and a managed market portfolio for the market. If there is
substantial variation in expected returns and variance, and if the variations in these objects
do not offset each other, we should expect substantial Sharpe ratio gains from timing the
market (Moreira and Muir 2017). We indeed find that such market timing leads to substantial

10

Electronic copy available at: [Link]

Sharpe ratio gains. When scaling the position in the market by the ratio of the conditional
market risk premium to the conditional variance, we obtain a portfolio that has a Sharpe ratio
of 0.68 in our sample. In comparison, the market has a Sharpe ratio of 0.34 in our sample.
Incorporating the information in the conditional moments thus expands the mean-variance
frontier substantially, highlighting the scope of conditional risk to help explain anomalies.
In conclusion, the above estimates suggest that conditional risk can, in principle, ac-
count for a large fraction of risk premia. Under a sufficiently strong correlation structure,
the variation in conditional moments appears large enough that it can generate substantial
conditional risk. Whether the correlation structure is such that conditional risk matters
is thus ultimately an empirical question. We will embrace this question in the upcoming
sections. We note, based on the discussions above, that properly accounting for conditional
risk will require methods that allow us to account for the potential impact of high-frequency
variation in conditional risk.
Finally, we emphasize that we use largely similar assumptions as LN. The main reason
we differ in our conclusions is that we consider it plausible that market timing and volatility
timing both contribute positively to conditional risk. We will present direct evidence of
this assumption in the upcoming empirical section. Moreover, there is a slight difference in
semantics between our studies: We consider a potential impact of 2.9 percentage points from
market timing a large impact on expected returns, particularly considering that the equity
premium is likely around 5 percentage points and the average Fama and French factor has
historically had an unconditional alpha of 3.7 percentage points. The focus of LN is different:
LN focus on whether conditional risk can explain all of the alpha on the value premium, in
which case the effect of market timing is indeed too small.

1.3 Conditional Risk Factors

This section introduces a set of conditional risk factors that can help researchers account for
conditional risk. We also introduce risk factors that allow us to explicitly estimate the role
of conditional risk premia and conditional variance in generating conditional risk (i.e. split
conditional risk into expected return and volatility timing).

1.3.1 Conditional Risk in the CAPM

We can arrive at the above expression for conditional risk more easily if we use the stochas-
tic discount factor language instead of the beta language. The stochastic discount factor

11

Electronic copy available at: [Link]

approach is also useful when generalizing the results to a multi-factor model.
The stochastic discount factor of the conditional CAPM3 is

1 1
mt+1 = − b r̃m
f t t+1
(12)
Rtf Rt

which can be written as

1 1 m 1 m
mt+1 = − br̃t+1 − (bt − b)r̃t+1 (13)
Rtf Rtf Rtf

The law of one price implies that

i
0 = Et [mt+1 rt+1 ] = Et [Rtf mt+1 rt+1
i
] (14)

By the law of iterated expectations, we have

0 = E[Rtf mt+1 rt+1

i
] (15)
i
= E[rt+1 i
] + cov(rt+1 ; Rtf mt+1 ) (16)

meaning that

i
E[rt+1 i
] = −cov(rt+1 ; Rtf mt+1 ) (17)
m i
= β̃E[rt+1 ] + cov(rt+1 ; ct+1 ) (18)
| {z }
Conditional Risk

which is the same expression as in (9). In the following section, we use the stochastic discount
factor language to more formally derive a multi-factor model with conditional risk.
3
The notation for the stochastic discount factor for the CAPM in expression (12) differs slightly from the
one usually used. Cochrane (2001) uses
M
mt+1 = At + Bt Rt+1

where At = 1/Rtf − Bt Et [Rt+1

M
] and Bt = −bt /Rtf . But this expression is of course the same as ours:

M 1 M M 1 1
mt+1 = At + Bt Rt+1 = + Bt (Rt+1 − Et [Rt+1 ]) = − b r̃m
f t t+1
Rtf Rtf Rt

12

Electronic copy available at: [Link]

1.3.2 Conditional Risk in Factor Models

We now derive a general statement for conditional risk in factor models. Consider the class
of factor models captured by the following stochastic discount factor for k = 1, . . . , K traded
risk factors:
K
1 1 X
mt+1 = − bkt r̃t+1
k
(19)
Rtf f
Rt k=1

where

k k k
r̃t+1 = rt+1 − Et [rt+1 ] (20)

and
k
Et [rt+1 ]
bkt = k
vart (r̃t+1 )

is the time t shock and price of risk for factor k. The expression in (19) can be rewritten as

K K
1 1 X 1 X
mt+1 = − bk r̃t+1
k
− (bkt − bk )r̃t+1
k
(21)
Rtf Rtf k=1 Rtf k=1

where bk is the unconditional price of risk for factor k

k
k E[rt+1 ]
b = k
var(r̃t+1 )

By applying the law of one price and taking unconditional expectations, we can state
an unconditional model that incorporates conditional risk. Before doing so, we define the
conditional risk factors ckt+1 = r̃t+1
k
(bkt − bk ).

Proposition 1 (conditional risk in factor models)

The unconditional expected excess return on an asset i is given by

K
X K
X
i k k
E[rt+1 ] = β̃ λ + βck λkc (22)
k=1 k=1

13

Electronic copy available at: [Link]

where
i k
cov(rt+1 ; r̃t+1 )
β̃ k = k
, λk = E[rt+1
k
], (23)
var(r̃t+1 )
i
k cov(rt+1 ; ckt+1 )
βc = , λkc = var(ckt+1 ) (24)
var(ckt+1 )

In the factor model above, each factor k is represented by two betas: one for its uncondi-
tional risk and the other for its conditional risk. These factors capture all of the unconditional
return implications of the stochastic discount factor in (19). The following proposition sum-
marizes the properties of the conditional-risk factors and their betas.

Proposition 2 (properties of conditional risk factors and betas)

2.a (zero mean factors): The means of all conditional-risk factors are zero:

k
E[r̃t+1 ] = E[ckt+1 ] = 0 (25)

2.b (uncorrelated factors): For each factor k, the return and shock to the risk factor is
uncorrelated with the conditional risk factor:

k
cov(rt+1 ; ckt+1 ) = cov(r̃t+1
k
; ckt+1 ) = 0 (26)

2.c (shock betas for the factors): The factor k has a loading of one on its own shock:

k k
cov(rt+1 ; r̃t+1 )
k
=1 (27)
var(r̃t+1 )

2.d (constant-beta equivalence): If an asset j has a constant conditional beta, the expected
return is given by the usual unconditional beta. That is, if
j k
covt (rt+1 ; rt+1 )
βtk = k
=c (28)
vart (rt+1 )

then

β̃ k = β k and βck = 0 (29)

14

Electronic copy available at: [Link]

 While Proposition 1 allows for the estimation of a k factor model, we will focus on the
one-factor CAPM model in most of the empirical analysis. We do so because the conditional
risk with respect to the market portfolio has the most tangible interpretation and because
the market factor is the most widely used factor. However, we extend the analysis in Section
4.2 to also cover multi-factor models.

1.3.3 Expected return and variance timing

Recall that conditional risk comes from covariance between conditional betas and two terms,
namely the conditional expected returns and conditional variance (multiplied by a constant):

i m m m


E[rt+1 ] = β̃E[rt+1 ] + cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) (30)
| {z }
Conditional Risk

The conditional risk factors introduced above allow one to estimate the net impact of these
two terms. We next introduce two factors that allow one to effectively separate the two terms.
In particular, we decompose the k th conditional-risk factor into two parts, ckt+1 = ck,e k,v
t+1 + ct+1 ,
where we define the two factors as
k
 
k,e k E[rt+1 ]
ct+1 = r̃t+1 bt − k
(31)
vart (rt+1 )
k
 
k,v k E[rt+1 ]
ct+1 = −r̃t+1 b − k
. (32)
vart (rt+1 )

These two conditional-risk factors can be used to summarize how conditional betas of test
assets covary with expected returns and variance for factor k, as summarized in the next
proposition.

Proposition 3 (market and volatility timing)

The effect of market timing and volatility timing can be captured by the unconditional co-
variance between expected excess return on asset i and the two factors ck,e k,v
t+1 and ct+1 :

i
; ck,e k k


cov(rt+1 t+1 ) = cov βt ; Et [rt+1 ] (33)
i
; ck,v k k


cov(rt+1 t+1 ) = cov βt ; −b vart (r̃t+1 ) , (34)

with total conditional risk given by the sum of the two covariances.

15

Electronic copy available at: [Link]

2 Methodology
In our empirical analysis, we control for conditional risk by including a number of condition-
ing variables in our regressions. Motivated by the theory section above, we include estimates
of the conditional market risk premium, the conditional variance, and conditional betas. If
we observe these variables perfectly, we can use the conditional-risk factors from Proposition
1 and need not include conditional betas or the conditional variance and market risk premia
in isolation. However, given that our measures of these inputs are likely to be imperfect,
we include both the conditional betas and conditional market moments to ensure that we
capture as much conditional risk as possible. More precisely, we implement the following
time-series regression for each asset i:

i Mkt Mkt Mkt Mkt i

rt+1 = α + a1 rt+1 + a2 rt+1 Et + a3 rt+1 vart +a4 rt+1 βt + ϵt+1 . (35)

We outline our estimation strategy for conditional market betas, risk premia, and variance
below.

2.1 Identifying Conditional Moments

In order to estimate our factor model, we must estimate the conditional mean and variance
of the factors. In this section, we outline the identifying assumptions we rely on in doing so.
To estimate the conditional market risk premium, we use the three-pass estimator sug-
gested by Kelly and Pruitt (2013). The estimator uses the cross-section of valuation ratios
to estimate the expected return. By using the cross-section of valuation ratios rather than
just the valuation ratio for the market, it is possible to separate the effect of expected growth
rates and expected discount rates. Accordingly, the methodology consistently recovers the
conditional market risk premium based on two simple identifying assumptions: (1) the ex-
pected log return and log growth rates are linear in a set of latent factors, and (2) these
factors evolve according to a first-order vector autoregression.
We rely on the Kelly and Pruitt estimator for multiple reasons. Most importantly, the
method is proven to predict the one-month expected market return well both in- and out-
of-sample, and it is proven to work in both the U.S. and internationally. Indeed, Kelly
and Pruitt (2013) show that the estimator predicts the one-month expected return on the

16

Electronic copy available at: [Link]

U.S. market portfolio with an R2 of 2.38 in-sample and 0.93 out-of-sample, and it predicts
the global market portfolio with an R2 of 1.5 out-of-sample. In addition, the estimator
consistently recovers the market risk premium under assumptions that are consistent with
the null hypothesis we test against when we are testing for conditional risk.
The estimator by Kelly and Pruitt (2013) is a three-stage estimator that extracts the
conditional market risk premium using information from the cross-section of book-to-market
values. By first estimating the sensitivity of valuation ratios of different portfolios to changes
in expected returns, the methodology afterwards aggregates the information into a single
estimate of expected one-period stock returns.
With respect to the variance, we similarly assume that the market variance evolves ac-
cording to a first-order autoregression. We rely on this assumption because it is transparent
and in line with recently published papers revolving around time-varying variance, such as
Campbell, Giglio, Polk, and Turley (2017).

2.2 Data
Our sample consists of 75,274 stocks covering 23 countries between January 1964 and De-
cember 2022. The 23 markets in our sample correspond to the countries belonging to the
MSCI World Developed Index as of December 31, 2018. We report summary statistics in
Table 2. Stock returns are from the union of the CRSP tape and the XpressFeed Global
Database. All returns are in USD and do not include any currency hedging. All excess
returns are measured as excess returns above the U.S. Treasury bill rate.
We study conditional risk in each country in our sample, a broad global sample, and
an international sample. Our broad sample of global equities contains all available common
stocks on the union of the CRSP tape and the XpressFeed Global database. Our interna-
tional sample excludes U.S. firms from the global sample. For companies traded in multiple
markets, we use the primary trading vehicle identified by XpressFeed. Our global sample
runs from June 1990 to December 2022, based on factor availability on Ken French’s website.
In some individual countries, we start our sample earlier if data is available (see Table 2 for
an overview of country-level start dates).
For the U.S. and global samples, we download all risk factors except betting against beta
from Ken French’s webpage. For other samples, we construct our own version of these risk

17

Electronic copy available at: [Link]

factors.4 We use the version of betting against beta from the authors’ webpage.5
The Kelly and Pruitt (2013) estimator takes as input portfolios sorted on size and book-
to-market. In the U.S., we use 100 portfolios sorted unconditionally on size and book-to-
market from Ken French’s website. In the global sample, we similarly create 100 portfolios
sorted unconditionally on size and book-to-market. In the individual international countries,
we create 25 portfolios sorted first on size and then conditionally on book-to-market.6 We
use only 25 portfolios and conditional sorts because some of the countries have few firms
at the beginning of the sample and the conditional sorts into 25 portfolios helps ensure an
adequate number of firms in each portfolio.
We calculate monthly variance as the sum of squared daily residuals over the month with
a degree of freedom adjustment for the estimation of the mean.
n
m n X m
var
c t (r̃t+1 ) = (r − r̄m )2 (36)
n − 1 i=1 i

where n is the number of trading days in the month. The estimation assumes that the
expected return is constant during each month.
The expected time t variance is then calculated as:

m
vart (r̃t+1 c t−1 (r̃tm )
) = θ̂0 + θ̂1 var (37)

where θ̂0 and θ̂1 are parameter estimates from the following regression:

m
var
c t (r̃t+1 c t−1 (r̃tm )
) = θ0 + θ1 var (38)

We rely on in-sample estimations for the expected variance, but the results are generally
robust to using out-of-sample estimates of the variance as in Bollerslev, Tauchen, and Zhou
(2009).
Finally, in order to estimate the betas of a given portfolio, we use what Boguth, Carlson,
Fisher, and Simutin (2011) refer to as lagged-component betas. These are estimated as the
portfolio-weighted average of the ex-ante beta of the stocks in the portfolio. We follow Fama
and French (1996) and use 60 months of monthly returns to calculate betas on individual
4
We use the methodology of Asness and Frazzini (2011) for constructing Fama and French portfolios
outside the U.S., although we use the traditional Fama and French measure of value.
5
[Link]
6
Note that, unless stated otherwise, we use global risk factors on the right-hand side except in the U.S.

18

Electronic copy available at: [Link]

stocks.7 See Boguth, Carlson, Fisher, and Simutin (2011) for details.

3 Conditional Risk in Major Risk Factors

In this section, we study the effect of conditional market risk on the major risk factors
studied in the literature. We first study the behavior of our estimates of conditional market
moments before quantifying the impact of conditional risk.

3.1 Variation in Conditional Moments of Market Returns

Table 2 offers summary statistics of the 23 exchanges in our sample along with the interna-
tional and global samples. The first three columns show the starting year of the sample, the
time-series median number of firms, and the time-series average weight of the given country
in the global portfolio. The U.S. has a high average weight in the global portfolio, but this is
in part driven by the early years where the U.S. constitutes most of the sample. The weight
of the U.S. market is downward trending throughout the sample and towards the end of the
sample, the weight of the U.S. is closer to .2. The fifth and sixth columns in Table 2 show
the average standard deviation and market risk premium in annualized terms.
The last three columns of Table 2 show the R2 of the expected variance and return to the
market portfolio. Regarding the variance, the R2 is generally around 20% to 50%, with the
U.S. and the global portfolio being on the low end. This high R2 corresponds to previous
studies on predicting variance (Bollerslev, Tauchen, and Zhou, 2009; Bollerslev, Hood, Huss,
and Pedersen, 2016), suggesting that the simple AR(1) method for predicting variance works
well.
The two last columns of Table 2 summarize the R2 of the expected return on the market
portfolio. The first column shows the R2 of the expected log return to the market portfolio,
which is what the Kelly Pruitt estimator extracts. The last column shows the expected excess
returns, which are calculated under the assumption of log-normally distributed returns by
adding one-half the conditional log-variance to the log-return, taking the exponential, and
subtracting the risk-free rate.
The table shows that the R2 for the excess returns in the U.S. and the global sample is
1.3% and 1.5%, which is around the same as reported by Kelly and Pruitt (2013). Inter-
7
An exception is for BAB, where we use the Frazzini and Pedersen (2014) betas, which means the
conditional beta is always zero (the strategy is hedged ex-ante to have a zero beta). Using Fama and French
betas means betas are not exactly zero, although conceptually they should be.

19

Electronic copy available at: [Link]

nationally, the R2 varies between 0.02% to 3.5%, with the median being 1.5%. The results
reported by Kelly and Pruitt for the U.S. and global sample thus appear to extend to most
individual exchanges.
The expected variance and market return are used to calculate the relative price of risk
bt − b, which is an important input for the conditional-risk factors used in Section 3.4. Figure
3 visually inspects this relative price of risk in the U.S. (Panel A) and the global sample
(Panel B). The price of risk varies substantially on both the short and long horizon. The
substantial short-horizon variation in the price of risk underlines the importance of using a
forward-looking measure of the price of risk. Indeed, an alternative to our approach is to
implement the conditional CAPM over short horizons for which the price of risk is assumed
to be constant. If daily data are available, the horizon is often around three to six months,
and if daily data are not available, the horizon is substantially longer. The price of risk in
Figure 3 exhibits substantial variation over these horizons, which, if statistically significant
and not driven by forecast errors, challenges this nonparametric approach.
The price of risk in Figure 3 also shows the substantial long-run variation that appears
linked to economic conditions. In the U.S. in particular, the price of risk tends to be the
highest in the years after economic recessions. The price of risk peaks a few years after the
recessions in 1973-1975, 1981–1982, 1990–1991, 2001, and 2007–2009. On the other hand,
the price of risk is lowest during the tech bubble. The price of risk is also low during the
onset of the financial crisis. The low price of risk at the onset of the financial crisis appears
to run counter to the notion of counter-cyclical risk aversion, but it is consistent with the
findings in Moreira and Muir (2017). Moreira and Muir argue that in the beginning of the
financial crisis, and crises more generally, the variance increases by more than the market
risk premium which causes the price of risk to go down.

3.2 Conditional Risk in Major Equity Risk Factors

In this section, we quantify how much of the unconditional alpha to major equity factors
can be explained by conditional risk. We consider six cross-sectional portfolios throughout
the section: value (HML), profitability (RMW), investment (CMA), momentum (UMD),
betting against beta (BAB), and the tangency portfolio (TAN) spanned by the market and
these five factors.
Table 3 Panel A shows the results of the U.S. sample. The first row shows the un-
conditional CAPM alpha, revealing the well-known empirical fact that these factors have
substantial CAPM alpha. The next row shows the intercept obtained from estimating the

20

Electronic copy available at: [Link]

model in equation (35). If our conditioning variables capture all the relevant variation in
conditional market risk premia and variance, this intercept is equal to the average condi-
tional alpha over the given sample. Controlling for conditional risk lowers the alpha for all
strategies.
In the rows below, we specify the conditional risk premia for the different factors (in annu-
alized terms). Conditional risk explains between 38 basis points and 1.25 percentage points
across the factors. The t-statistics for these estimates are calculated using the methodology
in Boguth, Carlson, Fisher, and Simutin (2011). The largest effect of conditional risk is for
the value factor, where conditional risk explains 27% of the unconditional alpha.
In Panel B, we zoom in on our the post-1996 sample. We focus on this period for
two reasons. First, recent research suggests that the equity premium may be particularly
volatile during this period (as can be seen in Figure 1). Second, it represents essentially an
out-of-sample analysis relative to the original study by Lewellen and Nagel (which ends in
June 2001). When considering this sample period, we find a substantially larger impact of
conditional risk. For the value, investment, and momentum factors, we find that conditional
risk can explain around 2% points of alpha.
The impact of conditional risk in Panel B is large relative to the impact of market
risk in general. The compensation is close to half the market risk premium. Moreover,
the compensation for conditional risk is substantially higher than the compensation for
unconditional risk. The conditional risk and unconditional risk appear to work in opposite
directions. In fact, the raw average returns are substantially closer to the true alpha of the
factors than the unconditional CAPM alphas are.
The magnitude is also large relative to the unconditional alpha on the factors. For
instance, the 2 percentage points represent 83% of the unconditional alpha on the HML
factor. We note that the unconditional alpha to the HML factor is insignificant in this
sample even before controlling for conditional risk. This is partly because value is slightly
weaker in the modern sample but of course also because the sample is shorter.
For the other factors, conditional risk has a substantial impact as well. For the investment
factor, the alpha is cut almost in half and becomes insignificant. For momentum, the impact
on the alpha, relative to the unconditional alpha, is more modest, although the momentum
factor also becomes insignificant.
Taken together, the above results from the U.S. sample emphasize the importance of
controlling for conditional risk when doing performance evaluation. Conditional risk can
have a material impact on alphas.

21

Electronic copy available at: [Link]

 To further illustrate the importance of conditional risk, Figure 4 plots country-level re-
sults for the value factor. The figure focuses on the G10 countries, excluding Italy because
the value factor has negative unconditional alpha in this country. The figure plots the com-
pensation for conditional risk in the value factor in each country along with the percentage
of the unconditional alpha that is explained by conditional risk. The compensation for con-
ditional risk is as high as 2% in France, Germany, and Sweden. This amounts to almost the
entire value premium in France and Germany and more than the entire premium in Sweden.
This finding reaffirms one of the key points of the paper: there is enough conditional risk
in the data to potentially have a material influence on estimates of alphas on major risk
factors. One must therefore be careful to control for such conditional risk whenever doing
performance evaluation.
We next turn our attention to our broad global sample. In Table 3 Panel C, we study
the effect of conditional in our broad global sample spanning 23 countries. We find very
similar results to those in the U.S. sample. All factors except profitability continue to load
on conditional risk, in the sense that controlling for conditional risk reduces the estimated
alpha.
The effect is again most pronounced for the value factor and the investment factor.
Controlling for conditional risk lowers the alpha of HML by 50 %. Similarly, controlling for
conditional risk lowers the alpha of CMA by 30 % and makes the alpha insignificant. These
findings further highlight the ways in which controlling for conditional risk has an impact
on our estimates of the performance of the major risk factors.
We also study conditional risk in individual countries. In Table 4, we report the impact
of conditional risk on the tangency portfolio in each country in our sample. For all but 3
countries, we find that controlling for conditional risk reduces the alpha of the tangency
portfolio. The median effect of conditional risk is 58 basis points. However, in none of the
countries considered is alpha for the tangency portfolio insignificant, emphasizing the clear
rejection of the conditional CAPM.

3.3 Timing of Expected Returns and Volaility

The previous section documents that conditional risk can have a substantial impact on
estimates of alphas on major factors. In this section, we study to what extent these results
are driven by market timing or volatility timing – and whether the two reinforce or counteract
each other in the generation of conditional risk.
As discussed in Section 1.2, one may be worried that market timing and volatility timing

22

Electronic copy available at: [Link]

counteract each other in the generation of conditional risk. In particular, if the conditional
market risk premium and variance are perfectly positively correlated, market timing and
volatility timing will counteract each other and reduce the impact of conditional risk. In
such a setting, conditional risk is unlikely to have a large impact on alphas. However, if
the two moments are uncorrelated, market timing and volatility timing need not counteract
each other. In fact, the two may work together to generate variation in conditional risk.
To better understand how market timing and volatility timing work together in the
creation of conditional risk – and thus help assess the overall scope for conditional risk in
explaining equity returns – we directly estimate the two terms using the conditional risk
factors from Proposition 3. Proposition 3 provides two precisely defined factors, for which
the covariance between these factors and the test asset tells us the exact impact of market
timing and volatility timing. These factors have the advantage that we can estimate the
impact of market and volatility timing without first estimating conditional betas.
Table 5 shows the loadings of the major risk factors on the market timing and the
volatility timing factors. The first asset we consider is the value factor. The value factor
loads positively on the market timing factor but negatively on the volatility timing factor.
This result suggests that volatility timing cannot help explain the return to the value factor,
consistent with Lewellen and Nagel (2006).
For the other factors, however, we find positive loadings on both the market timing factor
and the volatility timing factor. This finding suggests that market and volatility timing
contribute to conditional risk, i.e., the two factors work together in generating conditional
risk.
Table 6 reports similar results using rolling conditional betas instead of the conditional
risk factors. In this table, we regress the estimates of conditional betas at a given time
onto the estimate of conditional market premia and conditional market variance. The slope
coefficients on the market risk premium are generally positive in the US sample, and the
coefficients on the variance are negative. This finding is consistent with the idea that market
and volatility timing work together in generating conditional risk (a negative relation between
variance and betas increases the amount of conditional risk. The results are more mixed in
the global sample, but this could reflect that the methodology based on rolling betas does not
perfectly capture the behavior of realized betas (as the risk factors used in Table 5 should,
according to Proposition 3).

23

Electronic copy available at: [Link]

3.4 Estimates Using Alternative Methods
The above analysis suggests that conditional risk has a material impact on conditional alphas,
particularly in recent years. To ensure that these conclusions are not driven by the specific
implementation of the conditional CAPM that we have chosen, we redo the analysis using
the methods in Lewellen and Nagel (2006) and Boguth, Carlson, Fisher, and Simutin (2011).

3.4.1 Conditional Risk Using Methods in Lewellen and Nagel (2006)

Lewellen and Nagel (2006) suggests estimating alphas as the average alpha across a long
series of short-horizon regressions. The intuition behind this method is simple: if betas are
constant over the short horizon on which the regression is estimated, the average alpha will
be an unbiased estimate of the true CAPM alpha of the test asset.
Lewellen and Nagel (2006) implement the short-horizon regressions over various horizons,
finding largely similar results across the horizons. We consider here the annual horizon, which
– unlike the shorter horizons considered by the authors – has the advantage that it can be
implemented using monthly return data.
Table 7 Panel A reports the results in the post-1996 regressions. The results are not
too dissimilar to those obtained in our main specifications: conditional risk explains 1 to 3
percentage points of return (annualized) across the different factors. For value, the estimates
are quite similar, with conditional risk explaining 83% of the unconditional alpha. The main
difference relative to our main specification is that conditional risk has a large effect on
profitability effect but a more modest effect on momentum and investment. We note that
these differences can potentially arise from an overconditioning bias identified by Boguth,
Carlson, Fisher, and Simutin (2011).
We also consider the alphas in the full sample. As in our main specification, the impact
of conditional risk is smaller in the full sample. But the conditional risk remains substantial.
For the value factor, conditional risk explains 50% of the unconditional alpha in the long
sample, again emphasizing a large role for conditional risk in the return to the value strategy.

3.4.2 Conditional Risk Using Methods in Boguth, Carlson, Fisher, and Simutin
(2011)

We next consider the methods of Boguth, Carlson, Fisher, and Simutin (2011). Boguth,
Carlson, Fisher, and Simutin argue that using short-horizon windows as Lewellen and Nagel
(2006) potentially induces what they refer to as an over-conditioning bias, which arises

24

Electronic copy available at: [Link]

because small-sample estimates of beta are different from the ex-ante expected beta. Instead,
they use ex ante betas as instruments to capture conditional risk. They propose both a proxy
approach and an IV approach. The proxy method constructs an ex ante beta-estimate and
estimates average conditional alphas as the average difference between the realized excess
returns and the product of the beta and the realized market excess returns. The IV method
augments the usual CAPM regressions with the market returns times a vector of ex-ante
instruments, one of which is the lagged betas. See Boguth, Carlson, Fisher, and Simutin
(2011) page 372 for details on the procedures.
The IV procedure used by Boguth, Carlson, Fisher, and Simutin is very similar to our
main implementation. The only difference is in the choice of conditioning variables. The
authors use conditional betas, like we do, along with three measures of equity premia, namely
the dividend-price ratio, the risk-free rate, and the term premium. We instead use the
measure from Kelly and Pruitt (2013) to capture the conditional market risk premium and
we directly include the conditional market variance. These choices are guided by the theory
discussed in Section 1.
Our measure of conditional betas is what Boguth, Carlson, Fisher, and Simutin (2011)
refer to as lagged-component betas. These are estimated as the portfolio-weighted average
of the ex ante beta of the stocks in the portfolio. As explained earlier, we follow Fama and
French (1996) and use 60 months of monthly returns to calculate betas on individual stocks.8
Table 8 Panel A shows the results in the post-1996 U.S. sample. We again find a substan-
tial impact of conditional risk on estimates of alpha in this sample. For value, conditional
risk explains the entire risk premium. The effect is also large for the investment factor,
with conditional risk explaining 50% of the unconditional alpha in the best specifications.
The only factor for which the estimated alpha is substantially worse using the methods in
Boguth, Carlson, Fisher, and Simutin (2011) is the betting against beta factor. These results
emphasize the robustness of our empirical finding that conditional risk materially influences
estimates of alpha in the recent period. For the full sample in Panel B, the results are similar
as when using our methods, although they are slightly weaker.
8
An exception is for BAB, where we use the Frazzini and Pedersen (2014) betas, which means the
conditional beta is always zero (the strategy is hedged ex ante to have a zero beta). Using Fama and French
betas means betas are not exactly zero, although conceptually they should be.

25

Electronic copy available at: [Link]

4 Conditional Risk in Managed Portfolios and Multi-
factor Models
The previous section documents a large role for conditional market risk in explaining the
returns to major cross-sectional risk factors. In this section, we go beyond the tests of the
major risk factors in the conditional CAPM. In Section 4.1, we test whether the CAPM can
price managed portfolios. In Section 4.2, we explore the empirical importance of conditional
risk with respect to other factors than the market.

4.1 Testing the Conditional CAPM using Managed Factors

To go beyond the unconditional tests in Section 3, we next consider managed versions of the
major risk factors as test asset on the left-hand side. For each test asset i, we consider the
managed portfolio,

i,managed bit i
rt+1 = × rt+1 . (39)
bi

These portfolios increase the position in the test asset when the asset has a high conditional
price of risk. We expect such timing to cause the factors to have unconditional alpha with
respect to the original test asset. The upcoming analysis tests whether such unconditional
alpha can be explained by conditional market risk. We divide the managed portfolios by the
unconditional price of risk, such that the average weight in the input portfolio is close to one
(see discussion in Moreira and Muir 2017).
To construct these managed portfolios, we need estimates of the conditional expected
return and variance on the test assets. We estimate these moments in the same way we
estimate the moments for the market. That is, we estimate conditional variance assuming
an AR(1) process and we estimate the conditional expected return using the method by
Kelly and Pruitt (2013) (see Section 2). These methods predict moments fairly well in-
sample as shown in Appendix A1, although the predictability of expected returns on the
factors is slightly weaker than that on the market reported in Table 2, particularly in the
global sample.
Table 9 reports results from two sets of regressions. In the first, unconditional regressions,
we estimate

i,managed
rt+1 = αu + β1 rt+1
i m
+ β2 rt+1 + ϵit+1 . (40)

26

Electronic copy available at: [Link]

In the second set of regressions, we estimate what we refer to as conditional alphas. We
do so by implementing our main specification (equation (35)) augmented with the baseline
i
version of the managed factor i (that is, rt+1 ). In other words, for the managed version of
the value factor, we estimate its alpha in our baseline specification augmented with the value
factor.
The table shows that most managed portfolios have significant alpha with respect to the
market and the baseline portfolio. The unconditional alpha is significant for the momentum
and betting against beta factors. In general, the effect of conditional risk is fairly modest for
the conditional versions of the test assets. Controlling for conditional risk lowers the alpha
for value, profit, and momentum, but it increases the alpha on betting against beta and
profitability. The very strong performance to the scaled momentum portfolio is consistent
with the findings in Barroso and Santa-Clara (2015). Overall, the results suggest that a large
part of the time-series variation in the conditional moments of these factors is independent
of the movement in the conditional moments with respect to the market portfolio.
The results also emphasize an additional dimension along which the conditional CAPM
is rejected. We refer to Nagel and Singleton (2011) and Roussanov (2014) for in-depth tests
of the conditional implications of the CAPM.

4.2 Conditional Risk with Respect to Other Factors

We next consider the conditional risk with respect to other risk factors than the market. To
this end, we test how much of the return on the tangency portfolio that can be explained by
conditional risk with respect to the HML, RMW, CMA, UMD, and BAB factors.
For each factor, we first calculate the unconditional alpha by regressing the tangency
portfolio on the market portfolio and the factor in question. We next estimate alphas in the
conditional model by augmenting the unconditional model with variables capturing condi-
tional dynamics of the given risk factor. For each factor, we include two additional right-hand
side variables, namely the factor in question multiplied by the conditional expected return on
the factor and the conditional variance of the factor. We continue to estimate the conditional
moments of the given risk factor as described in the previous section.
Table 10 reports the results. Conditional risk with respect to the major risk factors
generally has a trivial impact on the alpha of the tangency portfolio. The largest effect
of conditional risk is from conditional risk with respect to the value factor, for which the
reduction in alphas is 5%.
The results contrast the results on the market factor. Conditional risk with respect to the

27

Electronic copy available at: [Link]

market factor appears unique relative to conditional risk with respect to other well-known
factors, as the impact of conditional market risk is an order of magnitude larger than the
impact of the other conditional risk factors.

5 Further Relation to the Literature

Our results relate to and extend a long strand of literature on conditional CAPM. A large
literature including Ferson and Harvey (1991), Ferson and Schadt (1996), Ferson and Harvey
(1999), and Jagannathan and Wang (1996) document that a series of conditioning variables
predict time-variation in returns and betas in the cross-section of equities and use these con-
ditioning variables as instruments in factor models. Lettau and Ludvigson (2001b) show that
using the cay variable as an instrument in the CAPM explains the returns to size and value
sorted portfolio, but Lewellen and Nagel (2006) argue that the effect is overestimated and
that the conditional CAPM cannot explain the cross-section of stocks. Lewellen and Nagel
further advocate the use of short-horizon regressions as an instrument-free way of testing the
conditional CAPM. However, Boguth, Carlson, Fisher, and Simutin (2011) argue that the
short-horizon regressions have certain small-sample issues and instead advocate the use of
an instrumental approach that uses past betas and state variables as instruments. Using this
approach, they show that momentum portfolios load on conditional risk. In addition, Ceder-
burg and O’Doherty (2016) argue that the conditional CAPM explains the low-risk anomaly
documented by Black, Jensen, and Scholes (1972) and Frazzini and Pedersen (2014). Going
beyond unconditional expected returns, Nagel and Singleton (2011) test the additional im-
plication that conditional expected returns must be consistent with the conditional factor
models. More recently, Kelly, Pruitt, and Su (2018) and Fama and French (2018) advocate
the use of characteristics as measures of conditional betas.
Our results on the low-risk effect differ substantially from those by Cederburg and
O’Doherty, as they find that the low-risk effect is statistically insignificant once control-
ling for conditional risk. One potential reason for this discrepancy is that we study the
returns to the monthly betting against beta factor and not quarterly beta sorted portfolios
as Cederburg and O’Doherty do. The advantage of studying the betting against beta factor
is that the factor is hedged ex ante to have a conditional beta of zero, mitigating the risk of
missing variation in conditional betas. In addition, the fact that the factor is hedged condi-
tionally to have a beta of zero, and an alpha of 10 percentage points per year, means that
it is unlikely that the conditional CAPM can explain its average return in the first place. It

28

Electronic copy available at: [Link]

would require the estimated conditional betas to be far from the true betas.
Finally, we note that Liu, Stambaugh, and Yuan (2018) redo the analysis in Cederburg
and O’Doherty (2016) using a slightly different methodology, and, consistent with our results,
they find that the conditional CAPM does not explain the low-risk anomaly.9

6 Conclusion
This paper documents a substantial impact of conditional risk on the alpha for major risk
factors. Across 23 developed countries, risk factors generally load on conditional risk. The
impact of conditional risk is substantial in many instances. In the long U.S. sample, condi-
tional risk explains around 30% of the unconditional alpha for the value factor. For the more
recent U.S. sample, as well as the French, German, and Swedish samples, conditional risk
explains essentially all the alpha to the value factor. In the global and recent U.S. sample,
conditional risk explains half the alpha to the investment factor and in general commands
premia of around 2 percentage points annualized across the major factors. These results
challenge the view that conditional risk cannot plausibly influence estimates of CAPM al-
pha.
The conditional CAPM is strongly rejected, in that conditional risk does not explain all
the alpha to the factors we consider. However, this rejection does not necessarily imply that
we should not control for conditional risk in factor analyses. It is, for instance, very common
to control for the market factor in factor analysis, even though the CAPM is rejected.
Overall, the impact of conditional market risk on anomaly alphas is larger than the impact
of unconditional market risk, suggesting that researchers and practitioners should seriously
consider controlling for conditional risk whenever implementing the CAPM in the future.
Controlling for conditional risk in the usual CAPM implementations can have broad
economic implications. For instance, a CFO of a value firm who discounts cash flows using
the unconditional CAPM would use the company’s beta of, say, 1 times the global market
risk premium, which gives an annual discount rate of around 5% in excess of the appropriate
risk-free rate. However, given the conditional risk in global value firms, the CFO should in
fact use an annual discount rate of around 7% in excess of the risk-free rate to also reflect the
conditional-risk premium. Such an increase in the perceived cost of equity of 2 percentage
points may have a material influence on a firm’s discount rate and ultimately its investment
9
See also Asness, Frazzini, Gormsen, and Pedersen (2020) for discussion of the role of conditional betas
in the low-risk anomaly.

29

Electronic copy available at: [Link]

decisions (Gormsen and Huber 2023, 2024). In addition to influencing corporate investment,
controlling for conditional risk is also important for judging the economic importance of
different anomalies, understanding market efficiency, evaluating the performance of asset
managers, and in financial analysis more generally.

30

Electronic copy available at: [Link]

References
Asness, Cliff, and Andrea Frazzini, 2011, The devil in hml’s details, AQR Capital Manage-
ment Paper.

Asness, Cliff, Andrea Frazzini, Niels Joachim Gormsen, and Lasse Heje Pedersen, 2020,
Betting against correlation: Testing theories of the low-risk effect, Journal of Financial
Economics 135, 629–652.

Barroso, Pedro, and Pedro Santa-Clara, 2015, Momentum has its moments, Journal of Fi-
nancial Economics 116, 111–120.

Binsbergen, Jules H. van, and Ralph S. J. Koijen, 2010, Predictive regressions: A present-
value approach, The Journal of Finance 65, 1439–1471.

Black, Fischer, Michael Jensen, and Myron Scholes, 1972, The capital asset pricing model:
Some empirical tests, Studies in the Theory of Capital Markets.

Boguth, Oliver, Murray Carlson, Adlai Fisher, and Mikhail Simutin, 2011, Conditional risk
and performance evaluation: Volatility timing, overconditioning, and new estimates of
momentum alphas, Journal of Financial Economics 102, 363–389.

Bollerslev, Tim, Benjamin Hood, John Huss, and Lasse Heje Pedersen, 2016, Risk every-
where: Modeling and managing volatility, Duke, working paper.

Bollerslev, Tim, George Tauchen, and Hao Zhou, 2009, Expected stock returns and variance
risk premia, Review of Financial Studies 22, 4463–4492.

Campbell, John Y., and John H Cochrane, 1999, By force of habit: A consumption based
explanation of aggregate stock market behavior, Journal of Political Economy 107, 205–
251.

Campbell, John Y., Stefano Giglio, Christopher Polk, and Robert Turley, 2017, An intertem-
poral capm with stochastic volatility, Journal of Financial Economic forthcoming.

Campbell, John Y., and Robert J. Shiller, 1988, The dividend-price ratio and expectations
of future dividends and discount factors, The Review of Financial Studies 1, 195–228.

Campbell, John Y., and S. Thompson, 2008, Predicting excess stock returns out of sample:
Can anything beat the historical average?, Review of Financial Studies 21, 1509––1531.

31

Electronic copy available at: [Link]

Campbell, John Y., and Tuomo Vuolteenaho, 2004, Bad beta, good beta, American Eco-
nomic Review 94, 1249–1275.

Cederburg, Scott, and Michael S O’Doherty, 2016, Does it pay to bet against beta? on the
conditional performance of the beta anomaly, The Journal of Finance 71, 737–774.

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

Cochrane, John H, 2001, Asset pricing, .

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

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

Fama, Eugene F, and Kenneth R French, 1996, Multifactor explanations of asset pricing
anomalies, The journal of finance 51, 55–84.

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

Fama, Eugene F, and Kenneth R French, 2018, Comparing cross-section and time-series
factor models, working paper.

Ferson, Wayne E, and Campbell R Harvey, 1991, The variation of economic risk premiums,
Journal of political economy 99, 385–415.

Ferson, Wayne E, and Campbell R Harvey, 1999, Conditioning variables and the cross section
of stock returns, The Journal of Finance 54, 1325–1360.

Ferson, Wayne E, and Rudi W Schadt, 1996, Measuring fund strategy and performance in
changing economic conditions, The Journal of finance 51, 425–461.

Frazzini, Andrea, and Lasse Heje Pedersen, 2014, Betting against beta, Journal of Financial
Economics 111, 1–25.

Gormsen, Niels Joachim, and Kilian Huber, 2023, Corporate discount rates, NBER Working
Paper.

32

Electronic copy available at: [Link]

Gormsen, Niels Joachim, and Kilian Huber, 2024, Firms’ perceived cost of capital, University
of Chicago Working Paper.

Jagannathan, Ravi, and Zhenyu Wang, 1996, The conditional capm and the cross-section of
expected returns, The Journal of Finance 51, 3–53.

Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling
losers: Implications for stock market efficiency, The Journal of Finance 48, 65–91.

Kelly, Bryan, and Seth Pruitt, 2013, Market expectations in the cross-section of expected
values, Journal of Finance 68, 1721–1756.

Kelly, Bryan, Seth Pruitt, and Yinan Su, 2018, Characteristics are covariances: A unified
model of risk and return, Journal of Financial Economics, forthoming.

Kelly, Bryan T, Tobias J Moskowitz, and Seth Pruitt, 2021, Understanding momentum and
reversal, Journal of Financial Economics 140, 726–743.

Lettau, Martin, and Sydney Ludvigson, 2001a, Consumption, aggregate wealth, and ex-
pected stock returns, The Journal of Finance 56, 815–849.

Lettau, Martin, and Sydney Ludvigson, 2001b, Resurrecting the (c) capm: A cross-sectional
test when risk premia are time-varying, Journal of Political Economy 109, 1238–1287.

Lewellen, Jonathan, and Stefan Nagel, 2006, The conditional capm does not explain asset-
pricing anomalies, Journal of Financial Economics 82, 289–314.

Liu, Jianan, Robert F. Stambaugh, and Yu Yuan, 2018, Absolving beta of volatility’s effects,
Journal of Financial Economics 128, 1 – 15.

Martin, Ian, 2017, What is the expected return on the market?, Quarterly Journal og Eco-
nomics 132, 367–433.

Moreira, Alan, and Tyler Muir, 2017, Volatility-managed portfolios, The Journal of Finance
72, 1611–1644.

Nagel, Stefan, and Kenneth J. Singleton, 2011, Estimation and evaluation of conditional
asset pricing models, Journal of Finance.

Roussanov, Nikolai, 2014, Composition of wealth, conditioning information, and the cross-
section of stock returns, Journal of Financial Economics 111, 352–380.

33

Electronic copy available at: [Link]

Santos, Tano, and Pietro Veronesi, 2004, Conditional betas, Working paper, National Bureau
of Economic Research.

34

Electronic copy available at: [Link]

 Appendix
A Conditional Cash Flow and Discount Rate Risk
Conditional market risk arises because conditional market betas are higher when the price of
risk is higher. As shown by Campbell and Vuolteenaho (2004), conditional market betas are
the sum of the given asset’s conditional cash flow and discount rate betas. Accordingly, the
conditional risk must come from either conditional cash flow or discount rate betas being
high when the price of risk is high. In this section, we show how to estimate these two
sources of conditional risk by decomposing the conditional risk factor into two.
m
First note that shocks to the market portfolio, r̃t+1 , are given by cash flow news and
discount rate news (Campbell and Shiller, 1988):

m
r̃t+1 = NCF,t+1 + NDR,t+1 (41)

The beta of an individual stock can then be expressed as:

βt = βtCF + βtDR (42)

covt (ri ;N ) covt (ri ;N )

where βtCF = t+1 CF,t+1
m )
vart (r̃t+1
and βtDR = t+1 DR,t+1
m )
vart (r̃t+1
.
Similarly, the market’s conditional-risk factor can be decomposed into two parts:

m
ct+1 = r̃t+1 (bm m
t −b ) (43)
= cCF DR
t+1 + ct+1 (44)

where cCF m m DR m
t+1 = NCF,t+1 (bt −b ) is the conditional cash-flow-risk factor and ct+1 = NDR,t+1 (bt −
bm ) is the conditional discount-rate-risk factor. Loading on conditional cash flow risk and
conditional discount rate risk has a tangible economic interpretation. Indeed, the uncon-
ditional covariance with the two risk factors summarizes the covariance of cash flow- and
discount rate betas with the expected return and variance:

i
, cCF CF m m


cov(rt+1 t+1 ) = cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) (45)

35

Electronic copy available at: [Link]

and

i
, cDR DR m m


cov(rt+1 t+1 ) = cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) (46)

Table A2 reports empirical results of the decomposition. Statistical significance is gen-

erally limited, but we find evidence of conditional discount rate risk for the investment and
value factors. Both in the U.S. and globally, factors load positively on the factor for con-
ditional discount rate risk. The loadings on the factor for cash flow risk are, on the other
hand, generally negative. These results imply that conditional risk comes mostly from time
variation in conditional discount rate betas, not cash-flow betas.

B Proofs
Proof of (3). Note that we can write the conditional beta as βt = E[β] + ηt . We can then
write the unconditional covariance between the excess return to asset i and the shock to the
market portfolio as

i m m m m
cov(rt+1 ; r̃t+1 ) = cov(E[β]r̃t+1 + ηt r̃t+1 ; r̃t+1 )
m m
= E[β] var(r̃t+1 ) + cov(ηt ; var(r̃t+1 ))

m m m 2 m 2 m 2 m 2
given that cov(ηt r̃t+1 ; r̃t+1 ) = E[ηt (r̃t+1 ) ] = cov(ηt ; (r̃t+1 ) ) and using that (r̃t+1 ) = Et [(r̃t+1 ) ]+
m
ϵt+1 = vart (r̃t+1 ) + ϵt+1 where cov(ηt ; ϵt+1 ) = 0. By dividing both sides by the unconditional
m
variance of r̃t+1 we obtain the expression in (3).

Proof of (9). Note that the covariance term in (5) can be written as

m m m m

 
cov βt ; Et [rt+1 ] − b vart (r̃t+1 ) =E βt (Et [rt+1 ] − b vart (r̃t+1 ))
m m
 

− E [βt ] E (Et [rt+1 ] − b vart (r̃t+1 ))

where the first term is equal to

i m
 
Et [rt+1 r̃t+1 ] m m
 i   m  i m
E m
(Et [rt+1 ] − b vart (r̃t+1 ) =E rt+1 E r̃t+1 (bt − b) + cov(rt+1 ; r̃t+1 (bt − b))
vart (r̃t+1 )
i
= cov(rt+1 ; ct+1 ),

36

Electronic copy available at: [Link]

given that E[ct+1 ] = 0 (shown later), and the second term is equal to zero:

m m m m
 

E [βt ] E (Et [rt+1 ] − b vart r̃t+1 ) = E [βt ] E[rt+1 ] − b var(r̃t+1 ) =0

Finally, to see that E[ct+1 ] = 0, note that

 m  m
E[ct+1 ] = E r̃t+1 E[bt − b] + cov(r̃t+1 ; bt − b)

which is equal to zero because the shock to the market portfolio has a zero mean and is
uncorrelated with (unpredictable by) the ex ante price of risk, bt .

Proof of Proposition 2.b. The covariance between the conditional-risk factor and the
shock to the market is zero:

m m
 m 2 
cov(r̃t+1 ; ct+1 ) = E[r̃t+1 ct+1 ] = E (r̃t+1 ) (bt − b) = 0

Proof of Proposition 3. Because the conditional-risk factors have zero mean, we have
that
k
  
i k,e i k E[rt+1 ]
cov(rt+1 ; ct+1 ) = E rt+1 r̃t+1 bt − k
vart (rt+1 )
k
  
i k E[rt+1 ]
= E Et [rt+1 r̃t+1 ] bt − k
vart (rt+1 )
k k


= E βt Et [rt+1 ] − E[rt+1 ]
= cov(βtk ; Et [rt+1
k
]).

Similarly, for the variance factor, we have that

k
  
E[rt+1 ]
i
cov(rt+1 ; ck,v
t+1 ) = −E i
rt+1 k
b−
r̃t+1 k
vart (rt+1 )
k
  
i k E[rt+1 ]
= −E Et [rt+1 r̃t+1 ] b − k
vart (rt+1 )
k k


= −E βt b vart (rt+1 ) − E[rt+1 ]
= cov βtk ; −b vart (r̃t+1
k


) .

37

Electronic copy available at: [Link]

 Table 1
Unconditional Alphas Implied by the Conditional CAPM
This table reports the unconditional alphas in annual percentage points implied by the conditional CAPM for varying
values of: (1) 𝜎"# , the unconditional standard deviation of the conditional expected excess return on the market, (2) 𝜎%&'# ,
the unconditional standard deviation of the conditional variance of the market, and (3) 𝜎(# , the unconditional standard
deviation of the conditional CAPM beta. The correlations between the conditional CAPM beta and the conditional market
risk premia and variance are denoted by 𝜌(#,"# and 𝜌(#,%&'# . The unconditional alpha in the conditional CAPM is given
by:
𝛼 , = cov1 (𝛽4 , 𝐸4 ) − 𝑏 × cov1 (𝛽4 , 𝑣𝑎𝑟4 ),

where 𝑏 is the ratio of the unconditional market risk premium to the unconditional market variance. Panel A reports the
unconditional alphas when considering either the market timing strategy, cov4 (𝛽4 , 𝐸4 ), or the volatility timing strategy,
− 𝑏cov1 (𝛽4 , 𝑣𝑎𝑟4 ), in isolation. Panel B reports the total unconditional alpha when combining the market and volatility
timing effects. In Panel B, we set 𝜌(#,"# = −𝜌(#,%&'# = 0.5. In the left part of Panel B, we set 𝜎%&'# = 0.5 and compute
the unconditional alpha for a range of unconditional standard deviations of the conditional market risk premium and betas.
In the right part of Panel B, we set 𝜎"# = 0.5 and compute the unconditional alpha for a range of unconditional standard
deviations of the conditional market variance and the betas. All returns are measured in annualized percentage points.

Panel A: Upper bounds for market timing and volatility timing

Market timing Volatility timing

𝜎"# 𝜎%&'#
𝜌(#,"# = 1 0.2 0.4 0.6 𝜌(#,%&'# = −1 0.2 0.4 0.6
0.2 0.24 0.48 0.72 0.2 0.60 1.20 1.80

𝜎(# 0.3 0.72 1.44 2.16 𝜎(# 0.3 1.80 3.60 5.40

0.4 0.96 1.92 2.88 0.4 2.40 4.80 7.20

Panel B: Joint effect of market and volatility timing

𝜎"# 𝜎%&'#
𝜎%&'# = 0.5 0.2 0.4 0.6 𝜎"# = 0.5 0.2 0.4 0.6
0.2 1.74 1.98 2.22 0.2 1.20 1.80 2.40

𝜎(# 0.3 2.61 2.97 3.33 𝜎(# 0.3 1.80 2.70 3.60

0.4 3.48 3.96 4.44 0.4 2.40 3.60 4.80

38

Electronic copy available at: [Link]

 Table 2
Summary Statistics
This table reports summary statistics for the 24 exchanges in our sample. Our sample consist of the union of all U.S.
common stocks on CRSP tape (“shrcd” equal to 10 or 11) and all global stocks in the Xpressfeed global database (“tcpi”
equal to 0). The expected market risk premium is calculated using the Kelly and Pruitt (2013) estimator. Expected variance
is calculated using an AR(1) regression. All returns are in USD. The standard deviation of the market risk premium is
annualized. The market risk premium is in annual percent. The R2 is based on monthly regressions. Outside the U.S., we
use the global measures of expected return and variance in all samples. Returns are total log returns. Excess returns are
simple returns in excess of the risk-free rate.

Mean Market risk premium R2 in predictive regressions

Median weight in
Starting number global Excess
Exchange year of firms portfolio St. dev Average Variance Returns returns
AUS 1994 1733 0.018 0.192 8.0% 0.462 0.002 0.004
AUT 1992 90 0.007 0.185 3.5% 0.441 0.020 0.020
BEL 1991 157 0.009 0.175 8.0% 0.419 0.017 0.017
CAN 1986 484 0.023 0.161 6.7% 0.516 0.010 0.011
CHE 1991 268 0.027 0.149 7.9% 0.328 0.025 0.022
DEU 1991 1208 0.084 0.174 5.0% 0.379 0.010 0.009
DNK 1992 165 0.005 0.182 9.0% 0.375 0.013 0.014
ESP 1991 154 0.014 0.204 6.7% 0.384 0.005 0.002
FIN 1991 140 0.004 0.242 12.1% 0.530 0.032 0.027
FRA 1991 766 0.044 0.191 6.6% 0.433 0.012 0.011
GBR 1988 2105 0.162 0.162 4.4% 0.381 0.013 0.015
HKG 1995 1209 0.052 0.212 10.9% 0.380 0.005 0.005
IRL 1995 51 0.002 0.231 6.0% 0.339 0.023 0.015
ISR 1996 369 0.002 0.206 8.4% 0.353 0.008 0.007
ITA 1992 279 0.018 0.203 3.9% 0.389 0.000 0.000
JPN 1989 3612 0.146 0.197 0.7% 0.227 0.003 0.011
NLD 1991 160 0.017 0.181 7.7% 0.469 0.039 0.032
NOR 1991 244 0.004 0.231 8.8% 0.454 0.022 0.022
NZL 1999 131 0.003 0.198 7.9% 0.437 0.017 0.020
PRT 1997 56 0.002 0.214 4.3% 0.370 0.011 0.009
SGP 2000 699 0.009 0.170 8.0% 0.398 0.057 0.061
SWE 1991 380 0.012 0.220 8.4% 0.433 0.014 0.015
USA 1964 4656 0.420 0.174 6.3% 0.196 0.007 0.013
WOR 1990 19698 1.000 0.146 5.3% 0.188 0.013 0.015

39

Electronic copy available at: [Link]

 Table 3
Conditional Risk in Equity Factors
This table reports results from the evaluation of different equity trading strategies when controlling for conditional risk.
We control for conditional risk by including a number of conditioning variables in our regression: 𝐸4 , the conditional
market premium, 𝑣𝑎𝑟4 , the conditional market variance, and 𝛽4A , the conditional beta of the factor. For each asset i, we
implement the following time-series regression:
A
𝑟4BC = 𝛼 A + 𝑎C 𝑟4BC
EFG EFG
+ 𝑎H 𝑟4BC EFG
𝐸4 + 𝑎I 𝑟4BC EFG A
𝑣𝑎𝑟4 + 𝑎J 𝑟4BC 𝛽4 + 𝜀4BC
A
where 𝑟4BC is the excess return to the risk factor i. TAN refers to the tangency portfolio spanned by the other five portfolios
and the market. “Compensation for conditional risk” is the difference between the unconditional CAPM alpha of the
factor and the alpha stemming from the regression above. “Compensation for unconditional risk” is the factor’s
unconditional CAPM beta times the market risk premium. Alphas and compensation for conditional risk are in annual
percentage points. Below parameter estimates we report t-statistics based on Newey-West standard errors. Statistical
significance at the 5% level is indicated in bold. The Full U.S. sample is 1964-2022 and the global sample is 1986-2022.

HML RMW CMA UMD BAB TAN

Panel A: Full U.S. Sample

Alpha in unconditional CAPM 4.58 3.95 4.70 8.63 9.99 4.97

(2.67) (3.43) (4.54) (4.93) (5.11) (7.91)

Alpha in conditional model 3.33 3.57 3.80 8.05 8.95 4.24

(2.08) (3.55) (4.02) (4.61) (4.91) (7.87)

Compensation for conditional risk 1.25 0.38 0.90 0.58 1.04 0.73
(2.12) (0.85) (2.68) (0.67) (1.47) (1.98)
Fraction of alpha explained by conditional risk 0.27 0.10 0.19 0.07 0.10 0.15

Compensation for unconditional risk -0.86 -0.58 -1.04 -0.99 -0.33 0.58

Observations 708 708 708 708 708 708

Adjusted R2 0.11 0.15 0.22 0.29 0.03 0.30

Panel B: Post 1996 U.S. Sample

Alpha in unconditional CAPM 2.45 6.29 4.43 7.25 10.42 5.23

(0.77) (3.30) (2.50) (2.62) (2.92) (4.44)

Alpha in conditional model 0.43 5.62 2.61 5.19 9.19 3.76

(0.15) (3.33) (1.64) (1.75) (2.88) (3.81)

Compensation for conditional risk 2.02 0.68 1.82 2.05 1.24 1.47
(1.63) (0.77) (2.42) (1.19) (1.38) (2.25)
Fraction of alpha explained by conditional risk 0.83 0.11 0.41 0.28 0.12 0.28

Compensation for unconditional risk -0.48 -1.64 -1.17 -2.63 -1.74 0.17

Observations 324 324 324 324 324 324

Adjusted R2 0.15 0.22 0.26 0.32 0.10 0.25

continued

40

Electronic copy available at: [Link]

 Table 3 – Continued
Conditional Risk in Equity Factors

HML RMW CMA UMD BAB TAN

Panel C: Global Sample

Alpha in unconditional CAPM 3.40 4.80 3.50 8.78 9.92 4.90

(1.48) (5.69) (2.29) (3.89) (4.36) (7.06)

Alpha in conditional model 1.71 4.90 2.48 8.70 8.96 4.47

(0.88) (6.10) (1.88) (3.86) (4.20) (7.35)

Compensation for conditional risk 1.69 -0.09 1.03 0.08 0.95 0.43
(1.89) (-0.25) (1.85) (0.09) (1.60) (1.67)
Fraction of alpha explained by conditional risk 0.50 -0.02 0.29 0.01 0.10 0.09

Compensation for unconditional risk -0.38 -0.55 -0.77 -1.54 -0.54 0.29

Observations 390 390 390 390 390 390

Adjusted R2 0.15 0.17 0.21 0.23 0.06 0.11

41

Electronic copy available at: [Link]

 Table 4
Conditional Risk Around the World
This table reports results from the evaluation of the tangency portfolio (TAN) in the unconditional CAPM and in the
conditional model across different exchanges. For each exchange, we regress the monthly excess returns for the tangency
portfolio on the shock to the market factor and the conditional model in equation (41) for the given exchange. TAN refers
to the tangency portfolio spanned the market and HML, RMW, CMA, UMD, and BAB. “Compensation for conditional
risk” is the difference between the unconditional CAPM alpha and the alpha of the conditional model. Alphas and
compensation for conditional risk are in annual percent. We report t-statistics based on Newey-West standard errors.
Statistical significance at the 5% level is indicated in bold.

Fraction
explained by Compensation
conditional for conditional Unconditional Adjusted
CAPM Conditional model risk risk risk premium Observations R2

Exchange alpha 𝑡-stat alpha 𝑡-stat

AUS 10.83 (7.68) 10.75 7.39 0.01 0.08 0.54 342.00 0.09
AUT 6.16 (3.34) 5.13 3.06 0.17 1.02 0.53 366.00 0.18
BEL 5.71 (2.80) 4.45 2.33 0.22 1.26 1.87 378.00 0.30
CAN 11.70 (6.10) 10.77 6.45 0.08 0.93 0.78 438.00 0.16
CHE 4.59 (3.94) 3.67 3.69 0.20 0.92 1.72 378.00 0.40
DEU 6.58 (4.75) 5.79 4.60 0.12 0.78 0.44 378.00 0.12
DNK 7.86 (5.26) 6.82 4.37 0.13 1.05 2.20 366.00 0.30
ESP 4.19 (3.44) 3.96 3.35 0.05 0.22 0.72 378.00 0.23
FIN 8.86 (4.29) 8.44 4.41 0.05 0.42 1.86 378.00 0.34
FRA 10.95 (5.42) 10.21 5.83 0.07 0.74 0.88 378.00 0.11
GBR 3.55 (2.72) 3.27 2.60 0.08 0.28 0.50 414.00 0.15
HKG 9.86 (3.84) 8.88 3.83 0.10 0.97 0.79 330.00 0.16
IRL 5.13 (2.47) 5.23 2.54 -0.02 -0.10 0.86 330.00 0.25
ISR 11.28 (6.28) 12.15 8.09 -0.08 -0.87 0.88 318.00 0.11
ITA 8.16 (4.82) 8.37 5.10 -0.03 -0.21 0.53 366.00 0.14
JPN 3.75 (3.69) 3.13 3.62 0.17 0.62 0.01 402.00 0.08
NLD 5.94 (4.66) 5.51 4.38 0.07 0.43 1.32 378.00 0.24
NOR 9.81 (4.90) 9.65 4.86 0.02 0.17 0.84 378.00 0.11
NZL 11.09 (8.50) 11.03 8.23 0.01 0.06 0.73 282.00 0.06
PRT 12.53 (4.51) 11.98 4.23 0.04 0.55 0.56 306.00 0.06
SGP 8.47 (5.32) 7.58 5.13 0.10 0.89 0.55 270.00 0.09
SWE 6.88 (3.38) 5.50 3.37 0.20 1.39 1.32 378.00 0.32
USA 4.97 (6.54) 4.24 7.07 0.15 0.73 0.58 708.00 0.30
WOR 4.90 (6.01) 4.47 6.75 0.09 0.43 0.29 390.00 0.11

42

Electronic copy available at: [Link]

 Table 5
Conditional Risk in Equity Factors: Decomposing Conditional Risk
This table reports of estimates of market timing and variance timing for the major equity factors. For each of the test
assets, we regress the monthly excess returns on the factor capturing market timing and the factor capturing variance
timing, as defined in Proposition 3. A positive loading on these factors reflects that conditional betas for the test assets
covary positively with the conditional market risk premium or negatively with the conditional market variance,
respectively. Below parameter estimates of betas we report t-statistics based on Newey-West standard errors. Statistical
significance at the 5% level is indicated in bold. TAN refers to the tangency portfolio spanned by the other five portfolios
and the market. The U.S. sample is 1964-2022 and the global sample is 1986-2022.

HML RMW CMA UMD BAB TAN

Panel A: U.S. Sample

Loading on conditional expected return factor 0.30 0.06 0.26 0.40 0.53 0.21
(1.92) (0.67) (2.77) (1.74) (2.43) (1.50)

Loading on conditional market variance factor -0.18 0.24 0.02 0.94 0.55 0.27
(-0.78) (1.86) (0.23) (3.10) (2.13) (2.29)

Panel B: Global sample

Loading on conditional expected return factor 0.64 0.00 0.46 0.06 0.57 0.15
(2.65) (0.57) (3.21) (0.79) (3.04) (0.86)

Loading on conditional market variance factor -0.06 0.01 0.15 0.87 0.18 0.14
(-0.12) (1.08) (2.42) (2.67) (0.99) (1.04)

43

Electronic copy available at: [Link]

 Table 6
Conditional Risk in Equity Factors
This table reports the results when regressing conditional rolling window betas of equity factors onto the market risk
premium and the conditional market variance:

𝑟𝑜𝑙𝑙𝑖𝑛𝑔 𝑤𝑖𝑛𝑑𝑜𝑤 𝛽4A = 𝑎A + 𝑎C 𝐸4 + 𝑎H 𝑣𝑎𝑟4 + 𝜖4

where 𝛽4A is the five-year rolling window CAPM beta of factor i, 𝐸4 is the conditional market risk premium, and var1 . The
table reports the parameter estimates 𝑎C and 𝑎H . Below parameter estimates we report t-statistics based on Newey-West
standard errors. TAN refers to the tangency portfolio spanned by the other five portfolios and the market. Statistical
significance at the 5% level is indicated in bold. The U.S. sample is 1964-2022 and the sample is 1986-2022.

HML RMW CMA UMD TAN

Panel A: U.S. Sample

Loading on conditional market risk premium 5.36 -8.57 12.04 4.02 3.03
(0.77) (-1.44) (2.28) (0.44) (1.14)

Loading on conditional market variance 1.62 -2.59 -8.30 -30.06 -7.34

(0.26) (-0.42) (-2.04) (-2.65) (-2.65)

Observations 702 702 702 708 702

Adjusted R2 0.01 0.01 0.05 0.03 0.05

Panel B: Global sample

Loading on conditional market risk premium 2.01 -6.97 -26.65 -7.82 -8.88
(0.41) (-2.29) (-0.97) (-0.80) (-1.37)

Loading on conditional market variance 2.93 12.77 25.33 -41.29 6.46

(0.27) (2.65) (0.62) (-3.18) (0.78)

Observations 384 384 384 390 384

Adjusted R2 0.03 0.04 0.07 0.02 0.02

44

Electronic copy available at: [Link]

 Table 7
Conditional Risk Using Other Methods: Short-Window Regressions
This table reports results from short-horizon CAPM regressions. For each calendar year, we regress monthly excess return
on a given test asset on the monthly excess returns of the market portfolio. For each test asset, we average across the
intercepts estimates each year to get the average conditional alpha for the test asset. The unconditional CAPM alpha is
the intercept from a full-sample regression of excess returns to the test asset onto the excess returns on the market. The
sample is 1996-2022 in Panel A and 1964-2022 in Panel B. All numbers are annualized.

HML RMW CMA UMD BAB TAN

Panel A: Post 1996 U.S. Sample
Alpha in unconditional CAPM 2.00 6.50 4.24 7.74 9.22 5.11
Alpha in conditional model 0.34 3.58 3.27 6.89 7.44 3.80

Compensation for conditional risk 1.66 2.91 0.97 0.85 1.78 1.31
Fraction of alpha explained by conditional risk 0.83 0.45 0.23 0.11 0.19 0.26

Panel B: Full U.S. Sample

Alpha in unconditional CAPM 4.47 3.98 4.65 8.96 9.10 4.88
Alpha in conditional model 2.27 3.29 3.43 7.98 8.26 4.05

Compensation for conditional risk 2.20 0.69 1.22 0.99 0.84 0.83
Fraction of alpha explained by conditional risk 0.49 0.17 0.26 0.11 0.09 0.17

45

Electronic copy available at: [Link]

 Table 8
Conditional Risk Using Other Methods: Boguth, Carlson, Fisher, and Simutin (2011)
This table reports estimates of conditional risk in equity factors after hedging time-varying betas using the methods in
Boguth et al. (2011). When using the proxy method to hedge, we subtract from the factor the time-series of conditional
betas times the time-series of realized excess market returns. When using the IV method, we subtract from the factor the
estimated betas in the IV regressions times the time-series of the regressors. The simple IV regressions use only the lagged
betas as instruments and the full IV regressions use lagged betas, the dividend-price ratio, the risk-free rate, and the term
spread. Below the estimates we report t-statistics based on Newey-West standard errors. The U.S. and global samples
run from 1964-2022 and 1986-2022.

HML RMW CMA UMD BAB TAN

Panel A: Post 1996 U.S. Sample

Unconditional alpha 2.45 6.29 4.43 7.25 10.42 5.23

Conditional alpha (proxy) 0.91 5.94 2.66 5.87 10.42 4.37

Conditional alpha (simple IV) 0.92 5.97 2.75 5.52 10.42 3.98
Conditional alpha (full IV) 0.42 5.25 2.35 6.26 9.01 3.54

Fraction of alpha explained by conditional risk:

Proxy 0.63 0.06 0.40 0.19 0.00 0.16
Simple IV 0.62 0.05 0.38 0.24 0.00 0.24
Full IV 0.83 0.17 0.47 0.14 0.14 0.32
Panel B: Full U.S. Sample

Unconditional alpha 4.58 3.95 4.70 8.63 9.99 4.97

Conditional alpha (proxy) 3.53 3.91 3.95 8.61 9.99 4.70

Conditional alpha (simple IV) 3.86 3.91 4.16 8.61 9.99 4.60
Conditional alpha (full IV) 3.14 3.46 3.79 9.53 9.70 4.58

Fraction of alpha explained by conditional risk:

Proxy 0.23 0.01 0.16 0.00 0.00 0.05
Simple IV 0.16 0.01 0.11 0.00 0.00 0.07
Full IV 0.31 0.12 0.19 -0.10 0.03 0.08

46

Electronic copy available at: [Link]

 Table 9
Managed Portfolios
This table reports results of the performance of managed portfolios in the conditional CAPM. For each of the five major
risk factors, we construct managed portfolios that scales the position in the portfolio based on the conditional price of risk
on the portfolio. We regress the excess returns to these portfolios on the market portfolio and the underlying factor, as
explained in the text, to estimate unconditional alphas. We next estimate alphas in the conditional model by augmenting
the unconditional model by the conditional market risk factors from our main specification. The table reports t-statistics
for the two different alphas along with the fraction of conditional risk explained by the inclusion of the conditional factors.
The sample is U.S. 1964-2022.

HML RMW CMA UMD BAB

Alpha in unconditional model 2.39 0.25 0.61 9.50 11.70

(1.41) (0.24) (0.48) (4.62) (7.09)
Alpha in conditional model 1.77 0.32 -0.07 8.55 11.97
(1.04) (0.32) (-0.05) (4.14) (7.22)

Compensation for conditional risk 0.62 -0.07 0.68 0.95 -0.27

Fraction of unconditional alpha explained by 0.26 -0.28 1.11 0.10 -0.02
conditional risk:

Observations 708 708 708 708 708

Adjusted R2 0.63 0.71 0.59 0.62 0.61

47

Electronic copy available at: [Link]

 Table 10
Conditional Risk in Multifactor Models
This table reports results from evaluation of the tangency portfolio in conditional multifactor models. For each of the
factors HML, RMW, CMA, UMD, and BAB, we study how much of the return to the tangency portfolio that can be
explained by the conditional risk with respect to the factors. For each factor, we first estimate the unconditional alpha of
the tangency portfolio with respect to the market and the given factor. We next estimate alphas in a conditional model by
augmenting the unconditional model with variables capturing conditional dynamics of the given risk factor. For each
factor, we include two additional right-hand side variables, namely the factor in question multiplied by the conditional
expected return on the factor and the conditional variance of the factor. TAN refers to the tangency portfolio spanned by
the market, HML, RMW, CMA, UMD, and BAB. Compensation for conditional risk is the annualized difference in the
alphas estimated with and without controlling for conditional factor risk. Statistical significance at the 5% level is
indicated in bold. The U.S. sample is 1964-2022 and the global and international samples are 1986-2022.

Tangency portfolio

Factor used on the right-hand side: HML RMW CMA UMD BAB

Alpha without controls for conditional factor risk 4.26 4.03 3.26 3.72 2.33
(8.47) (8.23) (7.33) (7.88) (6.11)

Alpha with controls for conditional factor risk 4.06 4.07 3.14 3.69 2.47
(8.44) (8.81) (7.62) (7.73) (6.46)

Compensation for conditional factor 𝑘 risk 0.20 -0.04 0.12 0.02 -0.14
Fraction of alpha explained by conditional risk 0.05 -0.01 0.04 0.01 -0.06
conditional risk:
Observations 708 708 708 708 708
Adjusted R2 0.31 0.36 0.50 0.35 0.61

48

Electronic copy available at: [Link]

 Table A1
Inputs to Conditional-Risk Factors
This table shows the R2 for the expected return and variance for the individual risk factors. These inputs are used for
constructing conditional-risk factors for these risk factors. The procedure for estimating expected return and variance
follows that for the market used in the main conditional-risk factor.

US World
R2 excess returns R2 variance R2 excess returns R2 variance

HML 0.02 0.51 0.00 0.43

RMW 0.02 0.48 0.01 0.24

CMA 0.01 0.37 0.00 0.34

UMD 0.01 0.48 0.01 0.48

BAB 0.03 0.56 0.02 0.53

49

Electronic copy available at: [Link]

 Table A2
Conditional Cash Flow- and Discount Rate Risk
This table reports results from evaluation of different equity strategies in the conditional CAPM. We regress quarterly
excess returns of different factors on the shock to the market portfolio, the conditional discount-rate-risk factor and the
conditional cash-flow-risk factor. TAN refers to the tangency portfolio spanned by the market, HML, RMW, CMA, UMD,
and BAB. “Compensation for conditional risk” is the conditional risk beta multiplied by the risk premium on the
conditional risk factor. Alphas and compensation for conditional risk are in annual percent. Below parameter estimates
we report t-statistics based on Newey-West standard errors. Statistical significance at the 5% level is indicated in bold.
The U.S. and global samples run from 1964-2015 and 1986-2015.

HML RMW CMA UMD BAB TAN

Panel A: U.S. Sample

Alpha 0.93 0.75 0.79 2.02 2.28 1.28

(2.04) (2.43) (2.85) (3.03) (3.67) (7.53)

Market beta -0.29 -0.17 -0.25 -0.12 -0.02 0.07

(-2.35) (-1.77) (-3.03) (-0.84) (-0.14) (1.31)

Conditional cash flow beta -0.01 -0.08 -0.02 0.02 -0.02 -0.03
factorfactorflowalph(simple IV) (-0.07) (-1.20) (-0.37) (0.20) (-0.26) (-1.02)

Conditional discount rate beta 0.09 0.07 0.09 0.06 0.08 0.04
Fraction of alpha explained by (1.72) (1.76) (3.44) (0.84) (1.27) (1.65)
conditional risk:
Observations 192 192 192 192 192 192
Adjusted R2 0.08 0.10 0.17 0.00 0.01 0.04

Panel B: Global Sample

Alpha 1.23 0.25 0.60 1.69 1.85 1.21

(2.53) (0.72) (1.89) (2.08) (2.64) (4.41)

Market beta -0.27 -0.23) -0.17 -0.14 -0.15 0.03

(-1.86) (-3.24) (-2.06) (-0.70) (-0.97) (0.41)

Conditional cash flow beta 0.01 -0.02) -0.03 -0.06 0.05 0.00
factorfactorflowalph(simple IV) (0.11) (-0.59) (-0.73) (-0.98) (1.08) (-0.13)

Conditional discount rate beta 0.07 0.04) 0.04 -0.01 0.09 0.04
Fraction of alpha explained by (1.83) (2.28) (2.37) (-0.29) (3.69) (3.81)
conditional risk:
Observations 102 102 102 102 99 102
Adjusted R2 0.14 0.15 0.14 -0.01 0.12 0.10

50

Electronic copy available at: [Link]

 Figure 1
Time-Variation in the Market Risk Premium
This figure shows two measures of the (annualized) market risk premium. The solid blue line shows the estimate obtained
using the methodology in Kelly and Pruitt (2013). The dotted green line is expected returns estimated as the inverse of
the CAPE ratio plus expected inflation on the 10-year horizon from the Michigan survey. The figure shows the equity
premium as of December each year between 1982 and 2022 in the US.

0.2
0.25

0.2
0.15

0.15
0.1

0.1
0.05

0.05

0
0

-0.05
9/1/1965
6/1/1967
3/1/1969

9/1/1972
6/1/1974
3/1/1976

9/1/1979
6/1/1981
3/1/1983

9/1/1986
6/1/1988
3/1/1990

9/1/1993
6/1/1995
3/1/1997

9/1/2000
6/1/2002
3/1/2004

9/1/2007
6/1/2009
3/1/2011

9/1/2014
6/1/2016
3/1/2018

9/1/2021
12/1/1963

12/1/1970

12/1/1977

12/1/1984

12/1/1991

12/1/1998

12/1/2005

12/1/2012

12/1/2019
-0.05
2/1/84
4/1/85
6/1/86
8/1/87

2/1/91
4/1/92
6/1/93
8/1/94

2/1/98
4/1/99
6/1/00
8/1/01

2/1/05
4/1/06
6/1/07
8/1/08

2/1/12
4/1/13
6/1/14
8/1/15

2/1/19
4/1/20
6/1/21
12/1/82

10/1/88
12/1/89

10/1/95
12/1/96

10/1/02
12/1/03

10/1/09
12/1/10

10/1/16
12/1/17
Expected
Expectedreturn
returnfrom
from Kelly
Kelly Pruitt Expectedreturn
Expected returnfrom
fromCAPE
CAPE

51

Electronic copy available at: [Link]

 Figure 2
Standard Deviation in Levels and Changes of Conditional Betas
For each stock, we compute the monthly horizon conditional market beta in two ways: 1) using a five-year rolling window
to compute the CAPM beta or 2) using the IPCA method of Kelly, Moskowitz, and Pruitt (2021). For each stock, we use
the conditional betas to compute the unconditional standard deviations of the levels and changes in the betas. Subfigure
(a) shows the distribution of the unconditional standard deviation of the levels in the conditional betas for individual
stocks. Subfigure (b) shows the distribution of the unconditional standard deviation of the changes in the conditional
betas. The sample is US 1963-2014.

Panel A: Unconditional standard deviation of conditional betas

0.06
Rolling window betas
Kelly, Moskowitz, and Pruitt (2021) betas
0.05
Percentage of stocks

0.04

0.03

0.02

0.01

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

Panel B: Unconditional standard deviation of changes in conditional betas

0.2
Rolling window betas
0.18 Kelly, Moskowitz, and Pruitt (2021) betas

0.16

0.14
Percentage of stocks

0.12

0.1

0.08

0.06

0.04

0.02

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

52

Electronic copy available at: [Link]

 Figure 3
Conditional Price of Risk
This figure plots the time series of the conditional price of risk minus the unconditional price of risk. The (conditional)
price of risk is the (conditional) market risk premium relative to the (conditional) variance. Panel A plots the price of
risk in the U.S. sample and Panel B plots the price of risk in the global sample.

Panel A: U.S. conditional price of risk (𝑏4 − 𝑏)

6.00

4.00

2.00

0.00

-2.00

-4.00

-6.00
9/1/1969
8/1/1971
7/1/1973
6/1/1975
5/1/1977
4/1/1979
3/1/1981
2/1/1983
1/1/1985

9/1/1992
8/1/1994
7/1/1996
6/1/1998
5/1/2000
4/1/2002
3/1/2004
2/1/2006
1/1/2008

9/1/2015
8/1/2017
7/1/2019
6/1/2021
12/1/1963
11/1/1965
10/1/1967

12/1/1986
11/1/1988
10/1/1990

12/1/2009
11/1/2011
10/1/2013
Panel B: Global conditional price of risk (𝑏4 − 𝑏)

10.00

5.00

0.00

-5.00

-10.00

-15.00
6/1/1990
7/1/1991
8/1/1992
9/1/1993
10/1/1994
11/1/1995
12/1/1996
1/1/1998
2/1/1999
3/1/2000
4/1/2001
5/1/2002
6/1/2003
7/1/2004
8/1/2005
9/1/2006
10/1/2007
11/1/2008
12/1/2009
1/1/2011
2/1/2012
3/1/2013
4/1/2014
5/1/2015
6/1/2016
7/1/2017
8/1/2018
9/1/2019
10/1/2020
11/1/2021

53

Electronic copy available at: [Link]

 Figure 4
Compensation for Conditional Risk in HML for G10 Countries
This figure shows the compensation for conditional risk and the fraction of the unconditional CAPM alpha in HML that
is explained by conditional risk in ten G10 countries. “Compensation for conditional risk” is the difference between the
alpha of the unconditional CAPM and the alpha of the conditional model in equation (41), measured in annual percentage
points. HML is the Fama and French (1993) value factor. We exclude Italy because the value factor has negative alpha in
this country.

2.5 1
0.9

Fraction of alpha explained by conditional

Compensation for conditional risk (%

2 0.8
0.7
1.5 0.6
0.5
points)

1 0.4

risk
0.3
0.5 0.2
0.1
0 0
BEL CAN CHE DEU FRA GBR JPN NLD SWE USA

Compensation for conditional risk Fraction of alpha explained by conditional risk

54

Electronic copy available at: [Link]

 Figure 5
Compensation for Conditional Risk Around the World
This figure shows the annualized compensation for conditional risk in percentage points. We regress the tangency
portfolio (TAN) in each country on: i) the unconditional CAPM and ii) the conditional model in equation (41). The
compensation for conditional risk is the difference between the unconditional alpha and the conditional alpha. TAN refers
to the tangency portfolio spanned by the five equity factors HML, RMW, CMA, UMD, and BAB along with the market.

1.50

1.00
Compensation for conditional risk

0.50
(% points)

0.00
EU

LD

ZL
BR
T

G
L

SA
L
R
N
E
S

A
N

R
P

E
A

T
P
N

R
U

SG
ES
U

SW
BE

CH

PR
IR

O
IS

O
JP
N

K
CA

IT
FI
FR

N
A

U
A

N
N

W
D

-0.50

-1.00

55

Electronic copy available at: [Link]