Risk Momentum: A New Class of Price Patterns∗

Sophia Zhengzi Li†, Peixuan Yuan‡, and Guofu Zhou§

First Draft: October 21, 2021

This Draft: March 22, 2023
Abstract

We uncover a new class of price pattern in terms of risk: the risk component, the

component of stock returns explained by common factors, exhibits strong momentum

intraday. This is important in asset pricing: prices cannot be a random walk if there

is compensation for taking risk. Moreover, this risk momentum implies a return

momentum: the long-short portfolio of stocks sorted by past risk exhibits a momentum

in return, demonstrating that high systematic risk implies high return even intraday. In

comparison with the extremely popular and extensively studied Jegadeesh and Titman

(1993) momentum, which is valid only monthly and only for stocks, our risk-based

return momentum holds intraday, daily, weekly, and monthly, and holds for not only

stocks, but also for corporate bonds and other asset classes.

JEL classification: G11, G12, G40

Keywords: Momentum, factor risk, intraday, arbitrageur participation, limits to

arbitrage

∗
We are grateful to Vincent Bogousslavsky, Youngmin Choi, Zhi Da, Kent Daniel, Ed Fang, Daniel
Giamouridis, Bing Han, Steve Heston, Jiantao Huang (discussant), Bryan Kelly, Jiahan Li, Kunpeng Li,
Victor Liu, Siyuan Ma, Alessandro Melone (discussant), David McLean, Lasse Pedersen, Jeffrey Pontiff,
Larry Schmidt, Karlye Dilts Stedman, Sheridan Titman, Xintong Zhan, Yingquang Zhang, Geng Zhe, and
seminar participants at Boston College, Capital University of Economics and Business, Fudan University,
Georgia State University, Merrill Lynch International, Peking University, Rutgers Business School, University
of Nottingham, Washington University in St. Louis, and conference participants at China Fintech Research
Conference, the 3rd International Fintech Research Forum, FMA Annual Meeting, New Zealand Finance
Meeting, Australasian Finance and Banking Conference, and MFA Annual Meeting for their helpful comments.
†
Rutgers Business School; E-mail: [email protected].
‡
School of Finance, Renmin University of China; E-mail: [email protected].
§
Olin Business School, Washington University in St. Louis; E-mail: [email protected].

Electronic copy available at: https://ssrn.com/abstract=4062260

1. Introduction

The goal of various asset pricing models is to explain asset returns by their risk exposures.
Popular models include the capital asset pricing model (CAPM) of Sharpe (1964) and Lintner
(1965), the Fama and French (1993) three-factor model, Fama and French (2015) five-factor
model, Hou, Xue, and Zhang (2015) q-factor model, and Stambaugh and Yuan (2016)
mispricing-factor model. While countless studies have examined or tested how well these
models work, there is barely any study on the times series property of the risk component,
the component of stock returns explained by common factors.
In this paper, we uncover the first momentum pattern in the risk component of stock
returns. While our results holds daily , weekly, and monthly, we focus on the less studied
intraday data, and estimate the risk from a cross-sectional regression of stock returns on
standardized anomaly variables in each of the 13 intraday periods, including 12 half-hour
periods between 10:00 and 16:00 and one overnight period between 16:00 and 10:00.1
Specifically, we decompose intraday stock returns into two components, a risk component
associated with factors (i.e., the sum of each estimated coefficient times the corresponding
anomaly variable) and a residual component which includes alpha. To capture various
risk exposures, we use three different sets of representative risk factors, respectively. Our
novel finding is that the risk component, regardless of each set, exhibits a strong intraday
momentum pattern: a high risk in the previous period leads to a high risk in the subsequent
period. The risk momentum holds every half hour from 10:00 till 16:00 market close, and
then overnight till 10:00 the next day.
The risk momentum is important in asset pricing: prices cannot be a random walk if
1
Note that our intraday periods extend over day trading hours, including the period from market close to
market open.

Electronic copy available at: https://ssrn.com/abstract=4062260

there is compensation for taking risk. Market efficiency and the random walk hypothesis
have long been a subject of research and debate in finance and its practice. While there is
increasing evidence that market is predictable (see, e.g., Rapach and Zhou, 2022), there are no
explanations that link the predictability to the evolution path of the risk. Our paper provides
perhaps the first momentum pattern for risk. To the extent that there is compensation for
risk taking, as the risk is not a random walk, the return cannot be either, and thus it must
be predictable.
We also find that the risk momentum, though not tradable itself, implies a tradable return
momentum. Sorting stocks by their past risk, the resulted long-short portfolio has an intraday
return momentum pattern. At the monthly frequency, it in particular produces a monthly
return momentum, similar to the monthly momentum discovered by Jegadeesh and Titman
(1993) (JT), which is extremely widely applied and is one of the most important anomalies in
asset pricing.2 However, the JT momentum fails intraday because reversal effect dominates
(Heston, Korajczyk, and Sadka, 2010). In contrast, our risk-based return momentum, robust
to various time frequencies, also implies a monthly risk-based return momentum stronger than
the JT, yielding a much higher annual return and Sharpe ratio. Moreover, our momentum
does not suffer from crash risk, while the JT does, an important fact discovered first by Daniel
and Moskowitz (2016). Our risk-based return momentum differs from the traditional JT in
another key aspect. While it takes years to develop theories and explanations to explain the
JT momentum, including rational and behavioral ones, our momentum has a risk explanation
at the start.3 It reinforces the notion that high total systematic risk is associated with high
returns.
2
The Google Scholar Citation of Jegadeesh and Titman (1993) is over 14,900, and the authors have
won many awards for their discovery of momentum, including the recent Wharton-Jacobs Levy Prize for
Quantitative Financial Innovation.
3
An extensive but incomplete list of rational and behavioral explanations for momentum includes Chan,
Jegadeesh, and Lakonishok (1996), Barberis, Shleifer, and Vishny (1998), Conrad and Kaul (1998), Daniel,
Hirshleifer, and Subrahmanyam (1998), Berk, Green, and Naik (1999), Hong and Stein (1999), Chordia and
Shivakumar (2002), Johnson (2002), Lewellen (2002), Gomes, Kogan, and Zhang (2003), Sagi and Seasholes
(2007), Liu and Zhang (2008), Li, Livdan, and Zhang (2009), Vayanos and Woolley (2013), Da, Gurun, and
Warachka (2014), Liu and Zhang (2014), Andrei and Cujean (2017), Li (2018), Mortal and Schill (2018), and
Kelly, Moskowitz, and Pruitt (2021).

Electronic copy available at: https://ssrn.com/abstract=4062260

 Our risk-based return momentum, in addition, extends to corporate bonds at the monthly
frequency, while it is known that the JT momentum does not generalize to bonds (see, e.g.,
Jostova, Nikolova, Philipov, and Stahel, 2013).4 Interestingly, following the idea of this
paper, Filippou, Li, Yuan, and Zhou (2023), Li, Ye, Yuan, and Zhou (2023), and Beckmeyer,
Filippou, and Zhou (2023) find risk-based momentum also holds for currencies, commodities,
and options, respectively.
The persistence of the risk component is intriguing. Given that our risk component
is aggregated from various anomaly components, persistent anomaly returns can be a
driving source of the continuation of the entire risk component. Indeed, we find that the
implied anomaly factor returns in our return decomposition exhibit positive and significant
autocorrelation. If all factors exhibit momentum, the risk must follow momentum as it is a
linear combination of factors. However, risk premium momentum does not require all factors
to exhibit momentum. As long as a sufficient number of factors exhibit momentum, the risk
will show momentum too.5
Why are anomaly returns persistent? We consider limits to arbitrage as the explanation.
We argue that anomaly returns are realized only when arbitrageurs start to trade on the
perceived mispricing. Due to limits to arbitrage, arbitrageurs may trade on the mispricing
gradually instead of eliminating mispricing all at once, and such gradual trading can result in
persistent anomaly returns. Empirically, we examine how arbitrageur participation and limits
to arbitrage affect the risk-based return momentum. Based on the coordinated arbitrage
model of Abreu and Brunnermeier (2002), rational arbitrageurs delay their arbitrage trading
on known mispricing due to holding costs and synchronicity risk. We find that our risk-based
return momentum becomes notably stronger after more frequent news arrivals, which is
consistent with the view that arbitrageurs are more willing to participate in trading as
more peers become aware of the mispricing in light of the news (i.e., reduced synchronicity
4
One question is whether there is intraday risk momentum for corporate bonds. This is difficult to answer
because corporate bonds are much less frequently traded and intraday data are limited.
5
Our risk momentum results are about the total risk exposures and are different from individual risks, but
they are consistent with the existence of factor momentum as the latter can only help. Our return momentum
is more general than the JT momentum which vanishes at daily and weekly frequencies.

Electronic copy available at: https://ssrn.com/abstract=4062260

risk). We further document stronger risk-based return momentum during a high aggregate
IVOL period, which corresponds to high limits-to-arbitrage time. When limits-to-arbitrage
is high, arbitrageurs are more likely to trade gradually due to market frictions. Both
empirical results are consistent with our hypotheses that increased arbitrageur participation
in mispricing correction and gradual trading due to limits to arbitrage are associated with
stronger risk-based return momentum.
We also propose a simple measure to directly capture arbitrageur participation in any
stock in real time. Our measure focuses on risk concentration, defined as the fraction of a
stock’s return variation explained by the risk component in a given intraday interval. The
idea is that the more a stock’s total return variation can be explained by the risk component
aggregated from various anomaly factors, the more likely that arbitrageurs have already
started trading this stock to correct the associated mispricing. Empirically, we find much
stronger risk-based return momentum among stocks with higher risk concentration.
The risk component, capturing the total systematic risk of stock returns, is a function of
anomaly factors. We consider three sets of anomalies. The first set consists of 15 representative
anomalies, including the 11 major mispricing anomalies from Stambaugh, Yu, and Yuan
(2012) and Beta, Size, Book-to-market ratio, and Reversal. The second set consists of 15
anomalies from Ehsani and Linnainmaa (2022). The third set includes 60 anomalies drawn
from Green, Hand, and Zhang (2017), Freyberger, Neuhierl, and Weber (2020), Gu, Kelly,
and Xiu (2020), and Kozak, Nagel, and Santosh (2020), covering numerous categories, such
as value versus growth, profitability, investment, issuance activity, momentum, and trading
frictions. We find that the risk-based return momentum is quite robust to using different sets
of anomalies to approximate the total systematic risk.6 While the results for the first two
sets are similar, with annualized return varying from 3.82% to 12.06% across overnight and
intraday half-hour intervals, the third set performs the best, with a greater annualized return
between 5.92% and 14.68%. For all three sets of anomalies, the best performance is obtained
6
We also measure the risk component using the 205 anomalies from Chen and Zimmermann (2022) and
the results remain robust.

Electronic copy available at: https://ssrn.com/abstract=4062260

in the mornings rather than in the afternoons.
An interesting question is how the risk-based return momentum performs when the holding
horizon is more than one intraday period. Notably, we find that the momentum based on risk
in the past intraday period is extremely persistent and lasts about 65 intraday periods or 5
days. Consistent with the previous intraday patterns, the risk-based return momentum from
the morning risk signal is more persistent than those from the afternoon signal, consistent
with existing studies (see, e.g, Cushing and Madhavan, 2000; Foucault, Kadan, and Kandel,
2005; Bogousslavsky, 2021) that investors behave differently towards the end of the day.
Our paper is related to Kelly, Moskowitz, and Pruitt (2021). They focus on explaining the
traditional momentum as compensation for time-varying covariance risk with factors based on
the IPCA model of Kelly, Pruitt, and Su (2019), and discovered a risk-based momentum using
the IPCA risk. In contrast, we are the first to study the momentum pattern of the risk. While
their study of risk-based momentum provides an improved momentum over the JT momentum,
they do not generate to alternative time frequency. In general, there are four major differences
between their study and ours. First, they analyze momentum at the monthly frequency,
and we study intraday price patterns on which no momentum has ever been documented
before. Second, they focus on explaining the traditional momentum, whereas we aim at
uncovering risk momentum and its wide implications to different time frequencies and asset
classes. Third, they estimate the IPCA risk component from a time-series perspective with
five latent factors. Instead, we measure the time-varying total risk exposure of stocks based on
a period-by-period cross-sectional regression on a large set of observable firm characteristics
(from 15 to 60). As a result, they assume time-varying factor loadings and constant risk
premia, whereas our factor loadings are constant but the risk premia are allowed to vary
over time. Our approach is what practitioners often use to attribute returns to various risks,
and is what is used by most recent machine learning studies (see, e.g., Gu, Kelly, and Xiu,
2020) due to the large sample size in the cross section. In short, our paper is related to Kelly,
Moskowitz, and Pruitt (2021), but is fundamentally different and contributes to the large

Electronic copy available at: https://ssrn.com/abstract=4062260

momentum literature by discovering a novel risk momentum that has wide implications.
Our paper is also related to Ehsani and Linnainmaa (2022), but is fundamentally different.
First, we rely on total risk of stocks that can have nonlinear and non-tradable risk factors,
and we have both momentum in the total risk and in the stocks. In contrast, their factor
momentum is on a long and short portfolios of the factors, a purely factor momentum without
implications on stock momentum per se. To illustrate, for example, if we add either highly
correlated or useless factors, this should have no effect on our momentum as our risk measure
is fairly robust to the choice of anomaly sets, but that will affect factor momentum greatly.
Second, our risk-based momentum is the long and short portfolios of stocks, which is a large
panel and can be sorted in various manners by the total risk of stocks. In contrast, the total
number of factors is small, and its sorting is limited. From an investment perspective, more
ways of sorting can only make the investment more appealing.
Our paper adds to the growing literature on high-frequency studies as data become
increasingly available. Heston, Korajczyk, and Sadka (2010) document an intraday anomaly
that there is a striking pattern of return continuation at half-hour intervals that are exact
multiples of an earlier trading day. Gao, Han, Li, and Zhou (2018) uncover a time-series
momentum pattern of the market: the first half-hour market return predicts the last half-hour
return, on which Bogousslavsky (2016) explains theoretically that this market intraday
momentum can be driven by investors’ infrequent rebalancing to their portfolios. Baltussen,
Da, Lammers, and Martens (2021) further analyze market intraday momentum in general via
the lens of hedging. Bogousslavsky (2021) explores anomaly returns over the trading hours
and overnight. To the best of our knowledge, our paper is the first to study the intraday
patterns of the risk component of stock returns and discover the associated cross-sectional
risk-based return momentum.
Our paper also adds new directions of research to the large momentum literature. Since the
seminal work of Jegadeesh and Titman (1993), the momentum factor that buys winners and
shorts losers plays an important role in factor models and in explaining mutual fund returns,

Electronic copy available at: https://ssrn.com/abstract=4062260

among others. For example, Griffin, Ji, and Martin (2003) study global stock momentum
and Asness, Moskowitz, and Pedersen (2013) examine momentum across asset classes. Since
our risk-based return momentum is unique and stronger than the traditional momentum, the
great number of questions studied related to the traditional momentum can also be asked for
the risk-based return momentum, generating more studies and deeper understanding about
momentum and risk in general.
The paper is organized as follows. Section 2 discusses the data and methodology of
decomposing stock returns into risk and residual components. Section 3 presents the empirical
results pertaining to the risk momentum and the risk-based return momentum. Section 4
uncovers various conditions under which the risk-based return momentum becomes stronger.
Section 5 performs robustness tests. Section 6 compares the performance of the monthly
risk-based return momentum to that of the traditional momentum across stocks and bonds.
Section 8 concludes.

2. Data and Methodology

2.1. Data

Our stock sample consists of the Russell 1000 index constituents.7 This top 1000 stock sample
comprises more than 90% of the total market cap of all stocks in the US equity market and
also has the advantage of allowing for relatively reliable high-frequency return estimation.
Our intraday price and quote data come from the NYSE trade and quote (TAQ) database,
covering the period from data inception in January 1993 to December 2020. We obtain
data on daily stock returns between January 1970 and December 2020 from the Center for
Research in Security Prices (CRSP) database.8
7
Our results are robust to the Russell 3000 index constituents as shown in Table A.3 in the Internet
Appendix.
8
The Russell 1000 index, launched in January 1984, comprises the top 1,000 stocks by market capitalization
and is rebalanced on the last Friday of June each year based on end-of-May stock capitalization. For the
period before January 1984, we use the end-of-May stock capitalization to select the top 1,000 stocks.

Electronic copy available at: https://ssrn.com/abstract=4062260

 We compute every 30-minute return between 10:00 and 16:00 and the overnight return
between 16:00 on the previous trading day and 10:00 on the current trading day.9 Our
high-frequency returns are based on mid-quote prices to mitigate three undesirable properties
caused by the use of transaction prices: the spurious correlation induced by the bid-ask
bounce (Roll, 1984), the selection bias associated with the occurrence of a trade, and possible
unachievability of transaction prices in the market place.10 Similar to us, the main analysis in
Bogousslavsky (2021) also uses returns computed from quote midpoints. Our results remain
robust to using volume-weighted average prices for computing intraday returns. Since prices
from TAQ are raw prices without adjusting for corporate actions such as dividend payout
and stock splits, we apply the daily “cumulative factor to adjust price” and “dividend cash
amount” variables in the CRSP database to adjust for split and dividend.
We consider three sets of anomalies. The first set consists of 15 representative anomalies,
including 11 mispricing anomalies of Stambaugh, Yu, and Yuan (2012) and Beta, Size,
Book-to-market ratio, and Reversal. The mispricing anomalies are updated monthly following
Stambaugh, Yu, and Yuan (2012), with the exception of Momentum, which is updated daily
and estimated by the cumulative return over the past 22 to 252 days. Beta is the estimated
coefficient by regressing monthly stock excess returns on monthly market excess returns
using a 60-month rolling window. Size is the natural logarithm of the market value of equity,
estimated by the product of the closing price and the number of shares outstanding, and
is updated daily. Book-to-market ratio is the ratio of the book value of common equity to
the market value of equity. Reversal is defined as the cumulative return over the past 21
days. We refer to the first set of anomalies as “15 RP anomalies”, where RP stands for
“representative”.
The second set consists of 15 anomalies investigated by Ehsani and Linnainmaa (2022),
including Accruals, Betting against beta, Book-to-market, Cash-flow to price, Earnings to
9
We use the price at 10:00 for the overnight return calculation to ensure that most securities have traded
at least once after the market open.
10
To see the last point, note that on average transaction prices have a 50% chance of being executed at
the bid and another 50% at the ask price. Suppose for a given day, the last transaction price is at the ask
price. The transaction price is the right price for the long position, but unachievable for the short position.

Electronic copy available at: https://ssrn.com/abstract=4062260

price, Profitability, Residual variance, Liquidity, Investment, Long-term reversals, Momentum,
Short-term reversals, Size, Quality minus junk, and Net share issues. We refer to the second
set of anomalies as “15 EL anomalies”, where EL stands for “Ehsani and Linnainmaa”.11
The third is a comprehensive set of 60 anomalies drawn from Green, Hand, and Zhang
(2017), Freyberger, Neuhierl, and Weber (2020), Gu, Kelly, and Xiu (2020), and Kozak,
Nagel, and Santosh (2020), covering numerous categories, such as value versus growth,
profitability, investment, issuance activity, momentum, and trading frictions. The list of these
60 well-known anomalies is provided in Table A.1 in the Internet Appendix. We refer to the
third set as “60 anomalies”.12
Panel A of Table 1 reports the summary statistics of the 15 RP anomalies, and Panel B
reports the same statistics for the 15 EL anomalies. Descriptive statistics on the last anomaly
set are also calculated but untabulated to conserve space. For all three sets, the anomalies
vary substantially on their means and standard deviations. Thus, we standardize all anomaly
variables before using them for predictive analyses.
Following Kozak, Nagel, and Santosh (2020) and Ehsani and Linnainmaa (2022), we
first transform each raw anomaly variable Vs,d−1,j on each day into a cross-sectional rank,
rank(Vs,d−1,j )
rcs,d−1,j = nd−1 +1
, where s denotes the stock, j denotes the anomaly variable, and nd−1
denotes the number of stocks on day d − 1. Next, we standardize these ranks by first centering
them around zero and then normalizing them by the sum of absolute deviations from the
mean:

rcs,d−1,j − rcs,d−1,j
Cs,d−1,j = Pnd−1 . (1)
s=1 |rcs,d−1,j − rcs,d−1,j |

Note that no future information is used in the transformation, so that any predictive regressions
based on Cs,d−1,j would be truly out-of-sample.
11
The set of 15 characteristics from Lewellen (2015) produces similar results.
12
In Section 5.3, we also examine a much larger set including 205 anomalies from Chen and Zimmermann
(2022), obtaining similar results to the case of 60 anomalies.

Electronic copy available at: https://ssrn.com/abstract=4062260

2.2. Methodology

In this paper, we consider stock returns and their risk decomposition in every intraday period.
Such intervals are common in intraday studies such as Heston, Korajczyk, and Sadka (2010)
and many others. Our first period (i = 1) is the overnight interval from the market close of
the previous day to 10:00, where the ending time is chosen to ensure that almost all securities
are traded at least once by then. The next period (i = 2) is from 10:00 to 10:30, and so on
until the last period (i = 13) which is between 15:30 and 16:00. We obtain risk decomposition
from running the following cross-sectional regression:

p
X
RET s,d,i = αd,i + Cs,d−1,j θd,i,j + ϵs,d,i , (2)
j=1

where RETs,d,i is the return of stock s on day d in intraday period i, αd,i is the intercept
on day d in period i, Cs,d−1,j is the standardized anomaly j of stock s observable at the
market close of day d-1, and ϵs,d,i is the residual. The unknown slope estimates θd,i,j can
be interpreted as factor returns (see, e.g., Fama and French, 2020), and so the second term
captures the total systematic risk of the stocks at intraday frequency.
Using the estimated coefficients θ̂d,i,j , we can then decompose the raw return of stock s
on d in intraday period i into two parts:

RETs,d,i = RISK s,d,i + RES s,d,i , (3)

where
p
X
RISK s,d,i = Cs,d−1,j θ̂d,i,j , (4)
j=1

is the estimated systematic risk explained by common risk factors, which is simply referred
as RISK.
It is important to point out that for simplicity we refer to the total systematic component,
the right hand side of Equation (4), as RISK. This is consistent with Fama and French (1993)

10

Electronic copy available at: https://ssrn.com/abstract=4062260

and Stambaugh, Yu, and Yuan (2012), although they focus on the time series version of the
factors. Our definition of RISK is quite general and can include any cross-sectional tradable
or non-tradable factors that contribute systematically to expected returns. In particular,
RISK can include both risk-based and behavioral factors. As argued by Kozak, Nagel, and
Santosh (2018), factor covariances should explain cross-sectional variation in expected returns
even in a model of sentiment-driven asset prices, because time-varying investor sentiment
can give rise to an ICAPM-like SDF. As a result, the RISK component can reflect both
compensation for risk and exposure to (systematic) mispricing correction.
The residual part RESs,d,i = RETs,d,i − RISKs,d,i captures the return component
unexplained by the factors, including the alpha component. While most existing studies focus
on alpha, there is rarely any analysis on RISK. In contrast, we focus on RISK, a measure of
the total risk contribution of all the systematic factors. Because of this, the new momentum
pattern of RISK discovered below appears to stimulate new theories to explain the risk path,
not just returns, as suggested by existing studies.
Our objective is to study the properties of RISK and its predictive power on future
returns. We exploit its predictive power as follows. At the beginning of each intraday period
i (i = 1, ..., 13) on each day d, we sort stocks into ten portfolios based on their realized RISK
values available at the time (i.e., the risk component of returns in the previous intraday
period). We buy stocks in the top decile with high RISK values and short those in the
bottom decile with low RISK values. We hold this long-short value-weighted portfolio during
period i on day d, resulting in the risk-based portfolio return:

d,i
RMd,i = R10 − R1d,i , i = 1, 2, . . . , 13, (5)

where R1d,i and R10

d,i
are the returns of decile portfolios 1 and 10 during period i on day
d, respectively. As a result, we have 13 long-short portfolios per day corresponding to
the 13 intraday intervals. Each of these portfolios enters position at the beginning of the
corresponding intraday period and exits position at the end of the corresponding intraday

11

Electronic copy available at: https://ssrn.com/abstract=4062260

period (i.e., rebalancing once per day). Then for each intraday period i, we have a time series
of such risk-based long-short portfolio over trading days. Besides the 13 long-short portfolios
that are rebalanced once per day at a fixed time of the day and are held over one intraday
period, we also consider a risk-based long-short portfolio investing in all 13 RISK signals
and obtain its time-series returns. For each long-short portfolio, we further decompose the
holding-period return into risk and residual components. These time series would allow us to
examine whether the risk component itself exhibits momentum and if such risk momentum
can imply a risk-based return momentum.
We use each of the three anomaly sets as Cs,d−1,j and estimate the slope coefficients in
Equation (2) to obtain the corresponding RISK. Panels A and B of Table A.2 in the Internet
Appendix respectively report the correlations among the 15 RP anomalies and the 15 EL
anomalies. The correlations are generally low, suggesting that multicollinearity is unlikely
an issue when we use 15 anomalies jointly to explain the cross-sectional returns of Russell
1000 stocks. Thus, for the first two anomaly sets, we run simple OLS regressions to obtain
the risk decomposition. To improve the efficiency of the slope estimators, we purposely use
inverse variances as regression weights.13
The third and larger set of 60 anomalies potentially contains more predictive information
about future returns while raising challenges of efficiently estimating the increased number
of unknown parameters. To alleviate concerns about overfitting, we apply two solutions.
First, following Kozak, Nagel, and Santosh (2020) and Ehsani and Linnainmaa (2022), we
use 15 principal components (PCs) to reduce the dimensionality. Specifically, we first fit the
PCA model with anomaly data from January 1970 to December 1992 and then construct 15
out-of-sample PCs between January 1993 and December 2020 matching the sample period
in our main analysis. Second, we use a penalized regression with the LASSO method that
encourages sparse estimates of regression coefficients by introducing the L1 penalty. Following
Dong, Li, Rapach, and Zhou (2022), we fit the LASSO model period by period using the
13
The variance for stock s on day d is estimated by the sum of squared returns over the [d − 21, d − 1] day
window.

12

Electronic copy available at: https://ssrn.com/abstract=4062260

Akaike Information Criterion (AIC). Such criteria are useful for selecting the value of the
regularization parameter by making a trade-off between the goodness of fit and the complexity
of the model.

3. Main Results

3.1. Risk Momentum

We first investigate the predictability of RISK for future RISK and RES components, which
are constructed according to Equations (4) and (5). As outlined in Section 2.2, we form 13
long-short portfolios based on the RISK signal in the previous intraday period. In addition,
we also form an “All-together” portfolio that trades all 13 RISK signals and is rebalanced
every intraday period. Then we calculate the value-weighted RISK and RES components
for each portfolio. If risk momentum exists, we expect the long-short portfolios to yield
significantly positive RISK in the future.
Table 2 reports the results of 13 spread portfolios sorted by the RISK signal available at
time “Start” and held over the subsequent intraday period between “Start” and “End” each
day. The last column reports the performance of the “All-together” portfolio that trades all
13 RISK signals and is rebalanced every intraday period. The rows labeled “RISK” (“RES”)
report the annualized RISK (RES ) component of each spread portfolio in percentage (i.e.,
annualized by a multiplier of 252 in percentage points), with Newey-West robust t-statistics
in parentheses. For example, the first column presents the RISK and RES of the spread
portfolio sorted by the previous RISK signal available at 16:00 and held from 16:00 to 10:00
each day.
Panel A reports the results based on RISK estimated from 15 RP anomalies. In each of
the intraday periods, the RISK are decisively positive, suggesting a strong risk momentum.
What is striking is that the pattern holds for every intraday period. The spread portfolios
are very stable with an average RISK ranging from 3.18% for the portfolio held from 14:00

13

Electronic copy available at: https://ssrn.com/abstract=4062260

to 14:30 to 11.50% for the portfolio held from 10:30 to 11:00. Overall, the risk momentum
is relatively stronger during the morning sessions. To have a sense of the risk momentum
through all day, we turn to the “All-together” portfolio in the last column that is rebalanced
13 times a day on investing in the spread portfolio every period. The average annualized
RISK of 79.10% is astronomical. In stark contrast, the average annualized RES is only
15.96%. Therefore, RISK has a much weaker predictability for future stock RES component.
Panel B reports the results for signals constructed based on 15 EL anomalies. The results
indicate that the construction of the spread portfolios is quite robust to a different choice of
the anomaly set. For instance, the overnight spread portfolio formed based on the previous
RISK available at 16:00 has an average RISK of 5.53%. The average RISK of the intraday
spread portfolio ranges from 2.52% when holding the portfolio from 14:00 to 14:30, to 11.22%
when holding the portfolio from 10:30 to 11:00. In addition, the “All-together” spread
portfolio generates an average RISK of 77.12% with a t-statistic of 23.95 and RES of 16.55%
with a t-statistic of 9.78.
Panel C reports the performance of the spread portfolios based on 15 PCs from the
60 well-known anomalies covering numerous categories. The results are almost uniformly
stronger than before. For instance, the “All-together” portfolio has a RISK of 108.92%,
compared with 79.10% and 77.12% from previous panels. With more factors, the resulting
RISK component likely better captures the total systematic risk, and in turn generates an
extra momentum that adds to the already strong momentum based on a smaller anomaly set.
On the contrary, PCA-based RISK shows a relatively weaker predictive power for the future
RES, yielding a smaller RES spread of 10.22% for the “All-together” portfolio.
Panel D reports the performance of the spread portfolios based on RISK estimated
from 60 well-known anomalies with LASSO sparse estimators. There are several noteworthy
findings. First, the results continue to show that the positive predictive power of RISK
on future stock RISK component is robust to different constructions of RISK. Second, the
LASSO-based RISK has an even stronger predictive power. As shown in the last column,

14

Electronic copy available at: https://ssrn.com/abstract=4062260

the “All-together” spread portfolio has an average RISK of 113.77% with a t-statistic of
27.11. Both values are greater than those reported in the other panels. Furthermore, the
increased predictability of RISK can also be observed across 13 individual spread portfolios.
For example, the average RISK of intraday spread portfolios ranges from 5.92% when holding
the portfolio from 14:00 to 14:30, to 13.60% when holding the portfolio overnight. Therefore,
exploiting the 60 well-known anomalies by LASSO can better capture the RISK component
which has significantly positive predictability for future RISK.

3.2. Persistent Risk Momentum

The holding period of our previous risk momentum strategies is either overnight or every
30-minute interval during the regular trading hours. Does the risk momentum only last one
intraday period or much longer? To address this question, we increase the holding periods to
up to five days and compute holding-period returns following an event study approach.
Specifically, at the end of each intraday interval i (i = 1, 2, . . . , 13), we form a long-short
portfolio based on the risk component of the return in interval i, and then hold the portfolio
over the subsequent k intraday periods with k ranging from 1 to 65, corresponding to one
intraday period to 5 days. For each portfolio formation time and each holding horizon k, we
form decile portfolios and compute the cumulative RISK (RES ) for each decile portfolio.
The spread between the cumulative RISK (RES ) in deciles 1 and 10 then forms a time series
of cumulative RISK (RES ) on the spread portfolio constructed by the end of the interval i.
In Table 3, we measure the risk component using 15 RP anomalies, and report for each
portfolio formation time (column labeled “Start”) the cumulative RISK and RES in basis
points (bps) and Newey-West robust t-statistics of the spread portfolios averaged across all
trading days.14 The columns labeled by numbers indicate the number of intraday periods the
portfolios are held. For instance, the cell with the row and column labeled “10:30” and “12”
14
For holding horizons less than or equal to one day (k = 13), there is no overlap between the two
consecutive observations of the spread series. For holding periods equal to 39 (65), there are 26 (42) intervals
of overlap, so we use Newey-West robust t-statistics with lag 26 (42).

15

Electronic copy available at: https://ssrn.com/abstract=4062260

corresponds to the portfolio formed at 10:30 and held for 12 periods until 10:00 the next day.
To understand the overall pattern of the cumulative returns controlling for the time-of-day
effect, we further average the returns and t-statistics across all 13 formation times and report
the results at the bottom of the same table (row labeled “Mean”).
There are several notable patterns. First, the average cumulative RISK of the risk
momentum portfolios generally increases as we extend the holding horizon to 65 intraday
periods (5 days). For example, the row labeled “Mean” reports the average RISK of the
risk momentum portfolios constructed at different times of the day. The cumulative RISK
monotonically increases from 2.41 bps for a one-intraday holding period to 14.22 bps for a
five-day holding period.
Second, the risk continuation is stronger during the morning sessions than the afternoon
sessions, and the contrast is more prominent for longer holding periods. For instance, for
the risk momentum with a holding horizon of 13 intraday periods, the cumulative RISK
decreases from 18.12 bps with a t-statistic of 11.22 for a portfolio formed at 10:00, to 6.08
bps with t-statistic of 4.47 for a portfolio formed at 16:00. Such pattern is consistent with
the earlier results in Table 2 that the one-period RISK s are greater in the mornings than in
the afternoons.
Third, RISK has no predictive power for future RES component of the risk momentum
portfolios even for longer periods. As shown in the last two rows, the cumulative RES
decreases from 0.49 bps for a portfolio held for one intraday period to 0.15 bps for a portfolio
held for 1 day, and it even becomes negative when the portfolio is held for 3 or 5 days.
Table 3 only reports cumulative returns over the subsequent k periods, where k takes every
consecutive integer value between 1 and 13, and then 39 and 65, due to space constraints.
To have a better sense of how cumulative RISK gradually evolves between day 1 and day 5,
we plot in Figure 1 the average cumulative RISK of the risk momentum strategies (using
15 RP anomalies) and their 95% confidence intervals against event time. In Panel A, we
present cumulative RISK averaged across spread portfolios formed at different times, the

16

Electronic copy available at: https://ssrn.com/abstract=4062260

same as the values reported in row “Mean” of Table 3. We find that immediately after the
portfolio formation, the cumulative RISK gradually increases to 12.33 bps after 13 periods,
then slowly goes up to 14.22 bps after 5 days.
To further investigate the intraday pattern of risk momentum, we separately plot the
average RISK of portfolios that are formed in the morning (from 10:00 to 12:30) and afternoon
(from 13:00 to 16:00) in Panels B and C of Figure 1, respectively. Strikingly, the persistence
of RISK momentum highly depends on the portfolio formation time. From Panel B, the
cumulative RISK of the momentum formed in the morning constantly increases to 16 bps
after 13 periods and continues to rise to 20.2 bps until 65 periods. However, Panel B exhibits
a relatively weaker and less persistent RISK momentum formed in the afternoon. The
cumulative RISK only increases to 9.5 bps after 13 periods; it quickly reverts to 7.8 bps
after 18 periods and stays around 8 bps until 5 days. The sharply different momentum
patterns are consistent with the finding that investors behave differently towards the end
of the day (see, .e.g, Cushing and Madhavan, 2000; Foucault, Kadan, and Kandel, 2005;
Bogousslavsky, 2021).15 Later in Section 4.2, we present further evidence showing that
the stronger risk momentum in the morning sessions can be explained by increased overall
arbitrageur participation early in the day.

3.3. Source of Risk Momentum

To understand the source of the risk momentum, we follow Lo and MacKinlay (1990) by
decomposing the holding-period RISK into three components: the average autocovariance of
individual stock RISK, the average cross-autocovariance across stocks, and the cross-sectional
variance in expected stock RISK. The first component captures the serial correlation in
individual stock RISK : a positive value would imply a positive average RISK to a strategy
15
Cushing and Madhavan (2000) and Foucault, Kadan, and Kandel (2005) point out that institutional
traders place an enormous emphasis on closing prices, which are used to calculate portfolio returns, tally
the net asset values of mutual funds, and mark-to-market various financial constraints; in the meanwhile,
market makers prefer to unload inventory to avoid exposures to overnight risk. Bogousslavsky (2021) finds
that mispricing is gradually corrected over the day but worsens at the end of the day when arbitrageurs are
constrained to adjust their portfolios to reduce overnight risk.

17

Electronic copy available at: https://ssrn.com/abstract=4062260

that buys winners and sells losers conditional on past RISK. The second component reflects
the lead-lag effects across stocks: a positive value would imply a negative average RISK
to our risk momentum strategy. The third component measures the dispersion in expected
RISK across stocks; if firms with positive RISK on average have higher expected RISK, the
risk momentum could be profitable due to the difference in expected individual stock RISK.
The requirement of the decomposition is that firms must exist over the entire sample
period, which leaves us with 256 stocks in the Russell 1000 index that have complete return
observations between January 1993 and December 2020. The decomposition results reported
in Table 4 indicate that the majority of the risk momentum profit comes almost entirely from
the positive autocovariance of individual stocks. For example, Panel A shows that, for the
“All-together” strategy based on 15 RP anomalies, the total RISK component is 107.98% and
the autocovariance component is 103.28%, indicating that the autocovariance component
explains 95.64% of the holding-period RISK. Similar conclusions hold for the remaining
panels based on different anomaly sets. Thus, the risk momentum is driven primarily by the
positive autocovariance in individual stock RISK.
According to Equation (4), RISK is the total systematic risk captured by the common
factors, and is equal to the sum of the standardized anomaly (Cs,d−1,j ) times the estimated
intraday factor return (θ̂d,i,j ). Since the anomaly variable is relatively stable over time and,
in fact, constant intraday, the intraday continuation of RISK must partially come from the
persistence of the intraday factor returns. To test if the factor returns are persistent, we turn
to examining their autocorrelations.
Table 5 reports the autocorrelation of intraday factor returns as a function of lag k (in
number of intraday periods), where k varies from 1 to 65. Panel A (B) reports the results for
15 RP (EL) anomaly factors, with *, **, and *** denoting significance at 10%, 5%, and 1%
level, respectively. The column labeled “1” reports the first-order autocorrelation for each time
series of the factor returns. Among all factors, Beta, Distress, Gross profitability, Earnings to
price, and Profitability have a first-order autocorrelation of 0.1 or above. Strikingly, 26 out of

18

Electronic copy available at: https://ssrn.com/abstract=4062260

the total 30 factors have a positive first-order autocorrelation at the 1% significance level,
while none of them have significant negative first-order autocorrelations.16
At lag 2, although the autocorrelations decrease substantially in magnitude, most of
them remain significantly positive. In particular, the autocorrelations at lag 2 for Distress,
Gross profitability, Earnings to price, and Profitability remain at or above 0.05 and are highly
significant. After lag 5, the magnitude of the autocorrelations decreases slowly to near zero.
For instance, at lag 8, the autocorrelation for only one-third of the factors remains statistically
significant. Interestingly, at lag 13, the positive autocorrelations of 13 out of 15 factors in
Panel A and 14 out of 15 factors in Panel B become significant again. Moreover, the majority
of the positive autocorrelations stay highly statistically significant at horizons that are exact
multiples of 13 half hours as shown in the last two columns. The periodicity of the factors
is consistent with the intraday periodicity pattern documented by Heston, Korajczyk, and
Sadka (2010).17 In the Section 4, we provide insights on what drives the continuation of
anomaly factor returns as a whole.

3.4. Risk-based Return Momentum

So far, we have shown a strong risk momentum intraday. However, we cannot trade on RISK
component. To investigate the economic gain of risk momentum, we construct a similar
but tradable risk-based return momentum resulted from sorting stocks by their past risk
components, as described in Section 2.2.
Table 6 reports the performance of 13 spread portfolios sorted by the RISK signal available
at time “Start” and held over the subsequent intraday period between “Start” and “End”
each day. The last column reports the performance of the “All-together” portfolio that trades
all 13 RISK signals and is rebalanced every intraday period. The rows labeled “Return”
16
Because of the cross-sectional correlations among factors, the factor returns of some factors (e.g., Size)
may be different depending on the control variables and thus have different autocorrelations under different
regression specifications in Panels A and B.
17
Based on Lo and MacKinlay (1990) decomposition with either the 15 RP or 15 EL factors, we find
strong intraday autocovariance, consistent with the autocorrelation results, yet there is almost no lead-lag
relation among factors.

19

Electronic copy available at: https://ssrn.com/abstract=4062260

report the annualized return of each spread portfolio in percentage (i.e., annualized by a
multiplier of 252 in percentage points), with Newey-West robust t-statistics in parentheses.
For example, the first column presents the return of the spread portfolio sorted by the RISK
signal available at 16:00 and held from 16:00 to 10:00 each day. The rows labeled “Alpha”
report the annualized CAPM alphas, where the intraday market returns are approximated
by using the returns on SPY.
Panels A–D in Table 6 show the results of the risk momentum constructed from four
different sets of anomalies. The results indicate that, regardless of the choice of anomaly sets,
the lagged RISK can positively predict future stock return. For example, the last column
labeled “All” shows that under the four different RISK specifications, the lagged RISK
component generates a stellar annualized future return of 95.06%, 93.67%, 119.14%, and
125.21%, respectively. Thus, we discover a strong risk-based return momentum intraday. Yet
in the absence of trading costs, this level of profitability will clearly invite arbitrage and will
go down quickly. On the other hand, the cost of trading individual stocks every intraday
period is likely high, and the return may evaporate after trading costs. What we show here is
simply the return patterns on the risk-based return momentum, and we do not suggest in any
way that the patterns can be profitable to investors who trade the strategy intraday. However,
as shown in Table 14 in Section 6 later, our risk-based return momentum also extends to
the monthly frequency with an annual alpha around 20%, and such high value would easily
survive typical transaction costs. The other columns corresponding to 13 different intraday
holding periods further confirm the pervasive risk-based return momentum effect.18
Comparing the results across the four panels in Table 6, we find that using a large set
of 60 anomalies to construct RISK component can result in a stronger momentum effect.
18
In sharp contrast, spread portfolios sorted by the total return in the previous intraday period all
generate reversal, as shown in Table A.5 in the Internet Appendix, consistent with the short-term reversal
documented in the existing literature. Heston, Korajczyk, and Sadka (2010) discover a striking pattern of
return continuation at half-hour intervals that are multiplies of a day. Figure A.1 in the Internet Appendix
confirms their results that there is a large return reversal at lag 1 and there are positive return responses
peaking at horizons that are exact multiples of 13 half-hours. In contrast, RISK exhibits strong positive
predictability for the next-period return. Strikingly, the predictability can last for several periods until one
day. As such, different from the return reversal or periodic continuation, the intraday risk-based return
momentum is a persistent phenomenon and lasts for a trading day.

20

Electronic copy available at: https://ssrn.com/abstract=4062260

For example, for the “All-together” portfolios in the last column, the holding-period returns
based on 15 anomalies in the first two panels are around 90% per year, versus the returns
based on 60 anomalies at around 120% per year in the last two panels. The comparison
reaffirms that the additional information in the 60 anomalies can better capture the total
systematic risk of stocks and thus generate stronger momentum effects.
The long-short spread portfolio in Table 6 only shows the return difference between the
top and bottom decile portfolios. To examine whether the results are entirely driven by
these two extreme deciles, Table 7 reports the performance across all deciles for strategies
based on 15 RP anomalies or 15 EL anomalies. Interestingly, the monotonic pattern that
high risk goes hand in hand with high return holds very well across all deciles for any given
formation period. As a result, the long-short spread portfolio captures well the cross-sectional
differences in RISK. While Table 7 presents the results for the first two sets of anomalies
only, similar patterns also hold for the two cases of 60 anomalies via PCA and LASSO, as
shown in Table A.4 in the Internet Appendix.
To shed light on the risk-based return momentum over longer horizons, we increase the
holding periods to up to five days and compute holding-period returns following an event
study approach. In Table 8, we measure the risk component using 15 RP anomalies, and
report for each portfolio formation time (column labeled “Start”) the cumulative returns in
basis points (bps) and Newey-West robust t-statistics of the spread portfolios averaged across
all trading days. Table 8 reveals several notable patterns. First, the average cumulative
return of the risk-based return momentum portfolios generally increases as we extend the
holding horizon to 13 intraday periods (one day); it declines afterward yet remains positive
and significant for up to 65 intraday periods (5 days). For example, the row labeled “Mean”
reports the average performance of the risk-based return momentum portfolios constructed
at different times of the day. The cumulative return monotonically increases from 2.9 bps
for a one-intraday holding period to 12.48 bps for a one-day holding period. After the first
day, the risk-based return momentum drops to 9.39 bps by day 3 and stays relatively flat at

21

Electronic copy available at: https://ssrn.com/abstract=4062260

9.43 bps by day 5. Therefore, the risk-based return momentum is persistent and robust to
different portfolio formation times as shown in other rows.
Figure 2 plots the average cumulative returns of the risk-based return momentum strategies
(using 15 RP anomalies) and their 95% confidence intervals against event time. In Panel A,
we present cumulative returns averaged across spread portfolios formed at different times, the
same as the values reported in row “Mean” of Table 8. We find that immediately after the
portfolio formation, the cumulative return gradually increases to 12.48 bps after 13 periods,
then slowly goes down to 9.39 bps after 39 periods, and stays relatively flat until 65 periods
after.
Furthermore, the persistence of momentum highly depends on the portfolio formation time.
From Panel B, the cumulative return of the momentum formed in the morning constantly
increases to 16.2 bps after 13 periods and continues to rise to 17.8 bps until 20 periods. After
that, the momentum slightly drops down to 15.8 bps by 36 periods and stays at the same
level until the end of day 5. However, Panel B exhibits a relatively weaker and less persistent
momentum formed in the afternoon. The cumulative return only increases to 9.6 bps after 11
periods; it quickly and surprisingly reverts to 5.8 bps after 18 periods and continues to go
down to 2.9 bps after 46 periods. Eventually, the momentum stays at around 4 bps until the
end of day 5.
In summary, we find that the risk-based spread portfolios exhibit strong return momentum
beyond one intraday period, and such risk-based return momentum is driven by the continuation
of the risk component itself.19 In the next section, we offer further explanations on the risk
momentum.

4. Explanation

From Equation (4), our risk component is aggregated from individual anomaly components. As
further shown in Table 5, anomaly factor returns exhibit positive and significant autocorrelation,
19
Our main results have been replicated by some doctoral students and researchers.

22

Electronic copy available at: https://ssrn.com/abstract=4062260

leading to continuation of risk and risk-based return momentum. Since anomaly returns are
realized only when arbitrageurs trade on the perceived mispricing, to generate persistent
anomaly returns, arbitrageurs need to participate in trading to correct the mispricing, and they
also need to trade gradually instead of aggressively. Due to limits to arbitrage, arbitrageurs
tend to trade gradually instead of eliminating all mispricing at once. In this section, we
formally examine how the degree of arbitrageur participation and limits to arbitrage affects the
risk-based return momentum , and propose a simple measure to directly capture arbitrageur
participation level in real time.

4.1. News Effect

Abreu and Brunnermeier (2002) propose a coordinated arbitrage model in which rational
arbitrageurs may be reluctant to trade on known mispricing because of holding costs and
synchronicity risk – the risk that other arbitrageurs do not trade and so it will take time
for anomaly returns to realize. More recently, Engelberg, McLean, and Pontiff (2018) find
that anomaly returns are much higher on earnings announcement days, consistent with
the notion that mispricing is driven by biased expectations which are at least partially
corrected upon news arrival. Motivated by both studies, we conjecture that after news arrival,
arbitrageurs as a group become more aware of the mispricing in light of the news, and thus
the overall synchronicity risk is reduced. As a result, they are more willing to participate in
trading to correct the perceived mispricing. Arbitrageurs still trade gradually due to limits
to arbitrage, leading to persistent anomaly returns. Thus, we expect a stronger risk-based
return momentum upon news arrival.
To empirically test this idea, we investigate the performance of our risk-based return
momentum strategies conditional on firm news arrivals. We begin with collecting the
high-frequency firm news data from the RavenPack news database between January 2000
(data inception) and December 2020. The RavenPack news data provide a comprehensive

23

Electronic copy available at: https://ssrn.com/abstract=4062260

sample of firm-specific news stories from the Dow Jones News Wire.20 Following the standard
practice of the literature, we only keep fundamental news that is most fresh and relevant
to a particular company. To consider only news about company fundamentals, we select 12
news groups of acquisitions-mergers, analyst ratings, assets, bankruptcy, loans, credit ratings,
dividends, earnings, corporate actions, labor issues, product services, and revenue from a
total of 29 news groups. To keep only fresh news about a company, we exclude repeated news
by requiring news in our sample to have an “event novelty score” of 100. To ensure that the
news is company-specific, we select news with a “relevance score” of 100, meaning that the
entity is mentioned predominantly in a news article.21 After processing the firm news, we
construct a time series of news arrival measures using the total number of news counts for
Russell 1000 stocks in each intraday interval over the period from January 2000 to December
2020. For simplicity, we construct risk signals based on the 15 RP anomalies throughout this
section.
We examine the effect of firm news on our risk-based return momentum by conducting
a similar event study as described in Section 3.2. Specifically, we compute the time-series
median of the news arrivals for each of the 13 intraday periods, and the time-series median
of the arrivals using all intraday periods and all days. In other words, we have 13 different
median values for 13 strategies formed by signals in a specific intraday period, and a median
value for the “All-together” strategy that trades on signals from all intraday periods. Then
we compute the average returns of the risk-based return momentum strategies conditional
on whether the news arrival in the signal formation period is above or below the median.
Figure 3 presents the firm news results during the period of January 2000 and December 2020
when the RavenPack news data are available. Strikingly, we find a distinct difference between
the risk-based return momentum performance conditional on high and low news arrival. For
example, the risk-based return momentum conditional on high news arrival rapidly increases
20
Recent studies using this data set include Kelley and Tetlock (2017), Jiang, Li, and Wang (2021), and
Jiang, Li, and Yuan (2022).
21
Applying these filters will not result in look-ahead bias, as all news articles are processed by RavenPack
within milliseconds of receipt, and the resulting data are sent to subscribers immediately.

24

Electronic copy available at: https://ssrn.com/abstract=4062260

to 12.8 bps after 13 periods, then gradually decreases to 8.7 bps after 43 periods, and rises to
around 11 bps until day 5. On the contrary, the momentum conditional on low news arrival
only slowly increases to a much lower level at 7.5 bps after 13 periods, continues to decline to
3.8 bps after 33 periods, and stays at the same level until day 5. Thus, we identify a strong
and instantaneous effect of firm news arrivals on the risk-based return momentum , consistent
with the idea of improved arbitrageur participation after news arrivals. Their gradual trading
due to limits to arbitrage leads to higher autocorrelations in anomaly returns and stronger
risk-based return momentum.

4.2. Aggregate Idiosyncratic Volatility

The recent literature has emphasized the important role of idiosyncratic volatility (IVOL)
in affecting asset prices.22 Garcia, Mantilla-Garcı́a, and Martellini (2014) argue that
aggregate IVOL is related to consumption volatility, a measure of economic uncertainty
in the inter-temporal asset pricing model of Bansal and Yaron (2004). Motivated by these
studies, we conjecture that our risk-based return momentum is also stronger during high
aggregate IVOL periods due to limits to arbitrage. Specifically, high IVOL leads to a shortage
of funds available for arbitrageurs, resulting in slow responsiveness of arbitrage capital to
mispricing. The greater limits to arbitrage they face will render more gradual mispricing
correction and thus enhance the performance of the risk-based return momentum strategy.
To verify our conjecture, we follow Garcia, Mantilla-Garcı́a, and Martellini (2014) and
use return dispersion (RD), defined as the cross-sectional standard deviation of stock returns,
to measure aggregate IVOL. There are two advantages of the RD measure: being model-free
and measurable at any return frequency. We first calculate RD for each intraday period on
all trading days from January 1993 to December 2020 to obtain a time series of RD measures.
Then we compute the time-series median of RD for each of the 13 intraday periods, and the
time-series median of RD using all intraday periods on all days. That is, we have 13 different
22
See, e.g., Campbell, Lettau, Malkiel, and Xu (2001), Bali, Cakici, Yan, and Zhang (2005), Ang, Hodrick,
Xing, and Zhang (2006), and Arena, Haggard, and Yan (2008), among others.

25

Electronic copy available at: https://ssrn.com/abstract=4062260

median values corresponding to 13 strategies that trade during a specific intraday period,
and a median value for the “All-together” strategy that trades every intraday period. Finally,
we calculate the average annualized returns during high (i.e., above median) and low (i.e.,
below median) RD periods for each strategy.
Table 9 reports the results for the momentum strategies during high and low RD periods
formed by different RISK signals. Interestingly, although the risk-based return momentum
exists in both high and low RD periods, the effect is much stronger during the high RD
period. Across Panels A to D, the “All-together” spread portfolio during high RD periods
delivers an annualized return of 157.57%, 156.81%, 203.46%, and 210.28%, about five times
as much as the annualized return of 32.16%, 30.34%, 38.50%, and 42.79% during low RD
periods. The returns on the spread portfolios during high RD periods are highly statistically
significant with t-statistics all above 21. The higher momentum during high RD periods is
also robust to all intraday holding periods. In sum, the results lend strong support to our
conjecture that a stronger risk-based return momentum exists during periods with elevated
aggregate IVOL.
In earlier sections, we have documented a stronger risk-based return momentum in the
morning. It is of interest to explore whether the result can be explained by higher return
dispersion early in the day. In Figure 4, we plot the time-series average of intraday RD in
percentage along with the 95% confidence intervals over the course of the day. Notably, we
observe a decreasing smirk pattern: the average RD starts from the highest value of 1.3% at
10:00, monotonically decreases to 0.38% at 13:30, stays around 0.4%, and eventually slightly
rises to 0.48% at the market close. Overall, the average RD is much higher in the morning
than in the afternoon. Moreover, more than half of the firm news are released overnight,
which could spur arbitrageurs to correct mispricing right after market open. Therefore,
both the time-of-day variation of RD and the great amount of news released overnight can
potentially explain the stronger risk-based return momentum in the morning in addition to
the clientele effect.

26

Electronic copy available at: https://ssrn.com/abstract=4062260

4.3. Risk Concentration

From Section 4.1, news arrivals correspond to more arbitrageur participation in trading and
stronger risk-based return momentum. In this section, we directly measure the degree of
arbitrageur participation in any stock in real time using a simple risk concentration measure,
i.e., the fraction of a stock’s return variation explained by risk in a given intraday interval. The
idea is that the more a stock’s total return variation can be explained by the risk component
aggregated from various anomaly factors, the more likely arbitrageurs have already started
trading this stock to correct the associated mispricings. As a result, greater arbitrageur
participation would lead to stronger risk-based return momentum among stocks with higher
risk concentration.
To test the above hypothesis, we measure a stock’s risk concentration as follows,

RiskCons,d,i = RISK 2s,d,i /RETs,d,i

2
, (6)

where RISK s,d,i and RET s,d,i are the risk component and the total return of stock s in
interval i on day d, respectively.23 A higher value of RiskCons,d,i indicates stock s has more
risk concentration by the end of intraday period i on day d. To put the risk concentration
into perspective, its average value upon high news arrivals is 29.26%, versus the average value
of 28.68% upon low news arrivals, generating a difference of 0.58% with a t-statistic of 5.46.
The contrast indicates that our risk concentration measure appears to capture the degree of
arbitrageur participation.
We next test the impact of risk concentration on the risk-based return momentum as
follows. By the end of each intraday period, we sort stocks into quintile portfolios on the basis
of their previous risk concentration; we classify stocks in quintile 1 as low-risk-concentration
stocks and those in quintile 5 as high-risk-concentration stocks. Within each group of stocks,
we form a spread portfolio that longs stocks in the top RISK decile and shorts stocks in the
23
We assume the population mean of intraday returns is zero as the cross-sectional intraday mean is almost
negligible in comparison with the cross-sectional variance.

27

Electronic copy available at: https://ssrn.com/abstract=4062260

bottom RISK decile, and we hold the spread portfolio over the subsequent intraday period.
Table 10 reports the risk-based portfolio returns formed by low or high risk-concentration
stocks based on different anomaly sets in Panels A to D. Strikingly, the high risk-concentration
stocks produce an even stronger risk-based return momentum universally, whereas the
low risk-concentration stocks generate a risk-based reversal instead of momentum. For
example, across the four panels, the “All-together” spread portfolio constructed based on
high risk-concentration stocks (rows labeled by “High”) yields an annualized return between
151.64% and 185.65%, about 50% to 60% higher than the return between 95.06% and 125.21%
based on the full sample in Table 6. In contrast, the spread portfolio constructed using low
risk-concentration stocks (rows labeled by “Low”) produces a negative return between -41.83%
and -30.28% with t-statistics all below -5. The stronger risk-based return momentum among
high risk-concentration stocks and the reversal among low risk-concentration stocks largely
hold for each of the 13 intraday holding periods shown in the other columns. Consistent
with Table 6, the risk-based return momentum effect conditional on high risk-concentration
stocks is the strongest based on 60 anomalies via the LASSO method. Overall, the results of
Table 10 confirm our conjecture that a stronger risk-based return momentum exists among
high-risk concentration stocks, supporting our initial hypothesis.24
Given the superior performance of the risk-based return momentum among high risk-concentration
stocks, it is of interest to see how its cumulative return evolves over time. From a practical
perspective, the “All-together” portfolio that is rebalanced every 30 minutes can also result
in high trading costs. Thus, for a given RISK signal, we form a daily-rebalanced portfolio
sorted by the corresponding close-to-close RISK signal from the previous day. Specifically, at
the market close of each day t, we aggregate the 13 intraday RISK signals between market
24
To mitigate the concern that stocks in the high (low) risk-concentration quintile simply have more (less)
extreme RISK in absolute magnitude (|RISK |) and thus tend to generate greater (smaller) RISK -based
return spread, we sort stocks into five portfolios by |RISK |, and then, within each quintile, we sort them
further into quintiles based on their risk concentration (e.g., 5 × 5 grouping). Next, we average across the
five |RISK | portfolios to produce five risk-concentration portfolios with large cross-portfolio variation in risk
concentration but little variation in |RISK |, and construct risk-based long-short portfolios within the highest
and lowest risk-concentration portfolios, respectively. The results are reported in Table A.6 in the Internet
Appendix, confirming the robustness of the finding that the risk-based return momentum is stronger among
high risk-concentration stocks.

28

Electronic copy available at: https://ssrn.com/abstract=4062260

close on day t − 1 and market close on day t into one cumulative RISK signal, then form
a long-short portfolio based on the close-to-close RISK, and hold the portfolio over the
subsequent close-to-close period. Finally, we compute the cumulative profits for the resulting
value-weighted daily-rebalanced strategy based on an initial investment of W1 = $1. That is,
we compound the returns of the spread portfolio on day d as follows,

Wd = Wd−1 × (1 + Rhigh,d − Rlow,d ), d = 1, 2, . . . , (7)

where Rhigh,d and Rlow,d are the returns of decile 10 and decile 1 on day d, respectively.
Figure 5 plots the trajectory of Wd starting from a given W1 = $1 initial investment
at the start of January 1993. We further take logarithm of the portfolio value with base
ten, so the trajectory in the plot starts from 0. The portfolio value based on all four
RISK strategies continues to rise throughout the entire sample without experiencing major
drawdowns. Furthermore, consistent with Table 10, the LASSO-based momentum strategy
exhibits the best performance throughout the sample period.
Table 11 further reports the performance of all decile portfolios formed on the basis of
various daily RISK signals using either all stocks in our sample in Panel A or stocks in the top
risk-concentration quintile in Panel B. Across different signals and stock samples, we observe
a strong monotonically increasing relation between past-day RISK and next-day return. For
the full sample in Panel A, the annualized returns of the RISK -sorted spread portfolios are
between 44.37% and 48.01% with t-statistics all above 9. For the high-risk-concentration
stocks in Panel B, the annualized returns uniformly increase to between 57.24% and 64.87%
with higher t-statistics all above 11. The return spreads based on Fama and French (2015)
five-factor alphas are numerically similar.25
In this section, we have explored two possible drivers for the risk-based return momentum.
25
Figure A.2 and Table A.7 in the Internet Appendix show that the risk-based return momentum continues
to hold at daily, weekly, and monthly frequencies over the expanded sample period from January 1970 to
December 2020. Figure A.3 in the Internet Appendix further shows that the risk-based return momentum
phenomenon exists in the corporate bond market at the monthly frequency.

29

Electronic copy available at: https://ssrn.com/abstract=4062260

Understanding further the economic forces of the heterogeneity in the factor return continuation
across anomalies is important but difficult. This will be an interesting topic for future research.

5. Robustness

5.1. Different Portfolio Formation and Holding Periods

The risk-based return momentum strategy forms portfolios based on the one-period RISK
component computed from returns over the previous overnight or 30-minute interval. We
would like to ensure that the results are not sensitive to this particular choice of portfolio
formation period. In this section, we investigate the performance of different portfolio
formation periods. Specifically, we first estimate one-period RISK using 15 RP anomalies and
compute M -period RISK by aggregating the one-period RISK in the prior M periods. We
consider M to be 1, 3, 5, 8, 13, 26, 39, and 65 periods, corresponding to a formation period
of 30-, 90-, 150-, 240-minutes, 1-, 2-, 3-, and 5-days. We also consider different portfolio
holding periods by following the portfolio rebalancing schedule of Jegadeesh and Titman
(1993). That is, to construct a portfolio with a holding period of N , we revise the weights on
1/N of the stocks in our strategy in any intraday interval and carry over the rest from the
previous interval.
Panels A and B in Figure 6 show the average annualized returns and their annualized
Sharpe ratios of the long-short risk-based return momentum portfolios based on different
combinations of portfolio formation period M (rows) and holding period N (columns). For
example, the entry with row labeled “3” (M = 3) and the column labeled “5” (N = 5) reports
the performance of the long-short portfolio that is formed based on sorting the past 3-period
RISK and is held for 5 intraday periods. The dark (light) color in the figure indicates a
high (low) value of the entry. As can be seen, almost all long-short portfolios have positive
returns, except for the 65/39 and 65/65 strategies. Many of the individual strategies have
both high returns and high Sharpe ratios. Out of the total 64 strategies, 56 strategies have an

30

Electronic copy available at: https://ssrn.com/abstract=4062260

annualized return of 10% or above, and 54 strategies have an annualized Sharpe ratio greater
than 1. The most successful strategies seem to be selecting stocks based on their RISK
signal over the past one-intraday period to one-day period and then holding the long-short
portfolios for one-intraday period to one day. Such strategies, as shown by the upper-left
regions of Panels A and B in Figure 6, can generate an annualized return between 37.51%
and 100.88% with an annualized Sharpe ratio between 2.38 and 4.51. In particular, at the
one-day holding horizon, the signal from the past one-intraday period generates the highest
Sharpe ratio of 4.51.
Another noticeable pattern is that there is a significant performance drop in terms of both
return and Sharpe ratio when the signal formation period or the holding horizon is longer
than one day. For example, the strategies based on the RISK signal over the past 2-day
period (row labeled “26”) yield a return of 8.48% to 67.21 % with a Sharpe ratio between 0.55
and 2.81, both of which are much lower than those based on the RISK signal over the past
one-intraday period to one-day period. These results suggest that our use of high-frequency
intraday data is crucial in effectively uncovering the strong risk-based return momentum.

5.2. Risk Momentum Attribution

How do different anomalies contribute to the risk-based return momentum strategy? To

answer this question, we decompose the RISK component of the holding-period return for
the risk-based spread portfolio into a sum of individual anomaly components according to
Equation (4), and estimate the average value of each anomaly component Cs,d−1,j θ̂d,i,j during
each intraday interval over all trading days, which captures the contribution of the associated
anomaly to the spread portfolio return.
Panel A of Figure 7 shows the results for each of the 15 RP anomalies. The color gradients
indicate the average spread portfolio returns of a given anomaly component over trading days,
with dark blue and white indicating the highest and lowest values. For example, in the sixth
column for the spread portfolio held between 12:00 and 12:30, Beta, Distress, Momentum,

31

Electronic copy available at: https://ssrn.com/abstract=4062260

and Reversal are associated with the darkest colors (greatest values). This means that the
RISK calculated in the previous period between 11:30 to 12:00 significantly and positively
predicts the Beta, Distress, Momentum, and Reversal components of the following period
RISK. Noticeably, the order of the component importance remains stable as we move from
the left to the right, indicating that the results are robust to the intraday holding periods.
Panel B reports the results for the 15 EL anomalies. The most important components
across intraday periods include Betting against beta, Residual variance, Momentum, and
Short-term reversals. Interestingly, both sets of anomalies point out the significant role of Beta,
Momentum, and Reversal in explaining the high-frequency risk-based return momentum.
Overall, most anomalies contribute to the risk-based return momentum strategies over
time, which further corroborates the positive autocorrelation of the vast majority of factors
as shown in Table 5.

5.3. A 205 Anomaly Set

In this section, we extend our anomaly set to the comprehensive one studied by Chen and
Zimmermann (2022), which includes 205 anomalies in total. The results are reported in
Table 12.
Panel A reports the results for signals constructed based on 15 PCs from the 205 anomalies.
The results are very similar to our main findings in Table 6. For instance, the overnight spread
portfolio formed on the basis of the last half-hour RISK from the prior day has an average
annualized return of 10.00% and a t-statistic of 3.69. The average returns of the intraday
spread portfolios range from 4.31% with a t-statistic of 5.94, when holding the portfolio from
13:30 to 14:00, to 10.07% with a t-statistic of 9.54, when holding the portfolio from 10:30 to
11:00. In addition, the “All-together” spread portfolio generates an average return of 88.33%
with a t-statistic of 22.10. Panel B reports the performance of the spread portfolios based
on the 205 anomalies with the LASSO dimension reduction technique. The results clearly
show that the resulting RISK continues to have strong positive predictability for future stock

32

Electronic copy available at: https://ssrn.com/abstract=4062260

returns, and the corresponding spread portfolio returns are almost uniformly higher than
those in Panel A. For instance, the “All-together” spread portfolio has an average return
of 105.26% with a t-statistic of 21.75, and both values are greater than those reported in
Panel A. The increased predictability of RISK can be observed in other 13 individual spread
portfolios as well. Overall, the results are largely similar to those in Table 6. Taking together,
this additional analysis with the 205 anomaly set reaffirms our earlier results, suggesting that
the risk-based return momentum is quite robust to the choice of the anomaly set.

5.4. Day-of-Week Effect

In this section, we test if the risk-based return momentum is affected by a weekend effect or
any other day-of-week effects.
Table 13 summarizes the performance of the long-short portfolios on different days of the
week formed by stocks in the top risk-concentration quintile. The “All” column indicates
that, on average, the risk-based return momentum is statistically significant and economically
large across all days of the week: the average annualized returns of the spread portfolios
are all above 130% and with t-statistics all above 13. The other columns report the results
of the portfolios held during overnight or one of the 30-minute intervals. As can be seen,
positive and statistically significant portfolio returns are present in almost all scenarios,
indicating the risk-based return momentum is robust across not only days of the week but
also times of the day. While the momentum is strong on each day of the week, the annualized
return almost monotonically decreases from Monday (162.8%) to Friday (131.49%). One
possibility is that arbitragers are reluctant to carry positions over the weekend, resulting in
less persistent mispricing correction and weaker anomaly return continuation as the weekend
approaches. Indeed, the annualized returns of the intraday risk-based return momentum
portfolios held on Friday afternoons are much smaller than the afternoon returns on other
weekdays. Furthermore, the strategy formed based on the last half-hour RISK signal on
Friday and held over the weekend produces an annualized return of 11.56%, only one-third

33

Electronic copy available at: https://ssrn.com/abstract=4062260

to one-half of the overnight returns on other days as shown in the first column. In contrast,
the “All-together” strategy on Monday still yields the highest return spread, likely because
arbitrager participation greatly increases after the weekend is over. For example, the strategy
that is held between 10:00 and 10:30 on Monday generates an annualized return spread of
24.62%, about two to three times those during the same time interval on other weekdays.

6. Monthly Momentum

So far, we have focused on the intraday or daily analysis of the risk momentum effect. This
section shows that this effect exists even at the lower monthly frequency and is stronger than
traditional momentum documented in Jegadeesh and Titman (1993). We first exploit the
predictive power of RISK on future returns as follows. At the end of each month t, we sort
assets (either bonds or stocks) into decile portfolios based on the RISK component of returns
in month t. We buy assets in the top decile with high RISK values and short those in the
bottom decile with low RISK values. We hold this long-short value-weighted portfolio over
month t+1 and rebalance it every month. The RISK components are constructed by using
60 stock anomalies via LASSO. Following Jegadeesh and Titman (1993), we also form two
return momentum strategies, namely JT 6m and JT 12m. JT 6m (12m) is the traditional
momentum strategy that long (short) stocks with the highest (lowest) cumulative returns
over the past 2 to 6 (12) months.
Panel A of Table 14 reports the average returns, alphas, and annualized Sharpe ratio for
the three momentum strategies. The returns and alphas are in percentage per month, and
Newey and West (1987) t-statistics are reported in parenthesis. Panel A.1 reports the results
from the full-sample period. Consistent with Jegadeesh and Titman (1993), we find that the
average raw returns of the long-short portfolios based on the JT momentum strategies are
significantly positive, indicating a strong momentum effect in the stock market. The return
momentum strategies are highly profitable, with average returns of 11.33% and 14.57% per
year for JT 6m and 12m strategies, respectively. Furthermore, the risk-adjusted return based

34

Electronic copy available at: https://ssrn.com/abstract=4062260

on the five-factor model from Fama and French (2015) continues to be significantly positive,
with an alpha of 12.14% for JT 6m and 15.65% for JT 12m. Importantly, the risk-based
return momentum exhibits a much better performance, which is evidenced by a higher annual
return of 19.44% with a t-statistic of 6.44. Furthermore, the annual alpha of 19.68% with a
t-statistic of 5.78 indicate the traditional pricing factors cannot explain the risk momentum
effect. Impressively, the risk-based return momentum yields an annualized Sharpe ratio of
0.83, much higher than the Shape ratios of 0.47 and 0.55 for the JT 6m and 12 strategies.
Panels B.2 and B.3 report the results from the first-half sample between January 1970
and December 1995 and the second-half sample between January 1996 and December 2020,
respectively. Notably, the JT momentum strategies perform quite well during the first-half
sample, which is more evident for JT 12m. Specifically, the JT 12m strategy yields a much
higher alpha of 19.27% per year with a t-statistic of 3.98. Furthermore, the Sharpe ratio
of JT 12m almost doubles during the first-sample period. The performance of risk-based
return momentum generates a relatively higher Sharpe ratio of 1.01. Not surprisingly, the JT
momentum becomes much weaker during the second-half sample, with a marginally significant
annual return of 11.01% for JT 6m and an insignificant positive return of 10.00% for JT
12m. The weaker return momentum effect is also indicated by the lower Sharpe ratios of
0.38 and 0.31 for JT 6m and 12m strategies, respectively. Importantly, the risk-based return
momentum continues to be strong even in the most recent two decades. It yields an average
return of 20.67% with a t-statistic of 4.39 and an alpha of 20.97% with a t-statistic of 4.88.
The resulting Sharpe ratio remains at a high level of 0.74.
Panel A in Figure 8 further plots the cumulative portfolio value from investing in different
momentum strategies. Notably, the portfolio value based on the past one-month RISK
continues to rise throughout the entire sample without experiencing significant drawdowns.
However, the JT momentum strategies fail to generate profits during the recent two decades
and exhibit momentum crash around 2009 as documented by Daniel and Moskowitz (2016).
Our risk-based return momentum strategy yields surprisingly better performance than the

35

Electronic copy available at: https://ssrn.com/abstract=4062260

JT momentum, and its performance even strengthens during the momentum-crash period.
The stronger risk-based return momentum is possibly due to the more constrained arbitrage
capital available to correct mispricing, enhancing the correction persistency.

7. Other Assets: Bonds

In this section, we extend our analysis to the corporate bond market. We first obtain
corporate bond data from the enhanced version of the Trade Reporting and Compliance
Engine (TRACE) over the sample period from July 2002 to December 2020.26 We supplement
the TRACE data with the Lehman Brothers Fixed Income (LBFI), Datastream, and the
National Association of Insurance Commissioners (NAIC) database such that the extended
sample starts from January 1973.27
We follow the same procedure outlined in Section 2.2 to decompose monthly bond returns
into their risk and residual components based on 25 bond characteristics, including “Bond age”,
“Amihud illiquidity”, “Bond market beta”, “Credit risk beta”, “Coupon”, “Default beta”,
“Downside risk beta”, “Duration”, “Illiquidity beta”, “Illiquidity”, “Kurtosis”, “Long-term
reversal”, “Momentum”, “Six-month momentum”, “Credit rating”, “Reversal”, “Issuance
size”, “Skewness”, “Bid-ask spread”, “Downside risk”, “Term beta”, “Time-to-maturity”,
“Macroeconomic uncertainty beta”, “Volatility beta”, and “Volatility”.
Panel B Table 14 reports the results for the three momentum strategies. JT 6m (12m)
is the traditional momentum strategy from Jegadeesh and Titman (1993) that long (short)
stocks with the highest (lowest) cumulative returns over the past 2 to 6 (12) months. The
risk-based return momentum is formed based on the RISK component in the last month. All
26
The TRACE Enhanced dataset contains corporate bond transactions with intraday price, uncapped
trading volume records, and information on whether the trade is a buy, a sell, or an inter-dealer transaction,
in addition to the information contained in standard TRACE. We then merge the TRACE Enhanced dataset
with the Mergent Fixed Income Securities Database (FISD) to obtain bond characteristics.
27
The LBFI database covers monthly data for corporate bond issues from January 1973 to March 1998;
Datastream has the daily corporate bond price averaged across all dealers for that bond on a given day,
and the NAIC database covers all transactions of corporate bonds by life, property, and casualty insurance
companies and health maintenance organizations beginning from January 1994.

36

Electronic copy available at: https://ssrn.com/abstract=4062260

portfolios are rebalanced monthly. Panel B.1 reports the results for the full-sample period.
The average raw return of the long-short portfolios based the JT momentum strategies are
all negative, indicating no momentum effect in the corporate bond market. In sharp contrast,
the portfolio formed on the basis of RISK yields a significantly positive average return of
6.01% per year with a t-statistic of 3.82, suggesting a strong risk-based return momentum in
corporate bonds. In addition, the risk-adjusted return based on the five-factor model from
Bai, Bali, and Wen (2019) continues to be significantly positive, with an alpha of 6.30% and
a t-statistic of 4.15. The risk-based return momentum also yields a high annualized Sharpe
ratio of 0.64.
Panels B.2 and B.3 report the results from the first-half sample between January 1974
and December 1999 and the second-half sample between January 2000 and December 2020,
respectively. Notably, the sub-sample findings are similar to those from the full sample.
Specifically, the traditional JT momentum strategies fail in both sub-sample periods. The
risk-based return momentum, on the other hand, is highly profitable during the first-half
sample, with a significantly positive average return of 5.35% per year and an alpha of 5.72%.
Importantly, it also exists during the second-half sample, with a significantly positive average
return of 6.80% per year and an alpha of 6.85%.
Consistent with Table 14, Panel B in Figure 8 shows that risk-based return momentum
makes a considerable profit throughout the entire sample. However, the JT momentum
strategies fail to generate positive returns and perform much worse during the momentum
crash period. Like stock risk-based return momentum, bond risk-based return momentum
exhibits better performance during severe limits-to-arbitrage periods, supporting the persistent
mispricing correction as a source for risk momentum. Overall, risk momentum is a pervasive
effect that exists across different trading frequencies and asset classes.

37

Electronic copy available at: https://ssrn.com/abstract=4062260

8. Conclusion

One of the fundamental insights of asset pricing theory is that greater systematic risk implies
greater expected return. In this paper, we uncover the first cross-sectional risk momentum in
the intraday literature. We find that there is not only a risk momentum (i.e., the continuation
of the risk component itself), but also a strong return momentum when stocks are sorted
by their past risk components. Both risk momentum and the risk-based return momentum
hold across different times of the day. In addition, we find that the risk component of daily
(weekly, monthly) stock returns also exhibits momentum, and generates a risk-based return
momentum at the corresponding frequency. Furthermore, the same phenomenon well exists in
the corporate bond market at the monthly frequency. Lastly, the monthly risk-based return
momentum is crash-free and much stronger than the widely cited Jegadeesh and Titman
(1993) momentum.
While the positive return of our risk-based return momentum is a manifestation of asset
pricing theory on compensation for bearing risk even intraday, the continuation of the risk
component itself is intriguing. Since our risk component is aggregated from various anomaly
components, the continuation of the entire risk component can be driven by persistent
anomaly returns, which can be further explained by arbitrageurs’ participation in mispricing
correction and gradual trading due to limits to arbitrage. Particularly, we argue that anomaly
returns are realized only when arbitrageurs trade on the perceived mispricing, who participate
in trading to correct mispricing, and trade only gradually instead of eliminating mispricing at
once due to market frictions. Indeed, we document stronger risk-based return momentum in
the morning sessions, during periods with more frequent firm news arrivals, when aggregate
idiosyncratic volatility is high, and among stocks with greater risk concentration. All the
evidence appears consistent with the explanation from limits to arbitrage.
Numerous explanations have been proposed over the past decades after the discovery of
the Jegadeesh and Titman (1993) (JT) momentum, and we anticipate more explanations to
be developed on the economic channels for our risk momentum pattern. Regardless whether

38

Electronic copy available at: https://ssrn.com/abstract=4062260

one agrees or not on our current explanations, the bottom line is that our paper provides
what seems the strongest momentum pattern in asset pricing, which holds intraday, daily,
weekly, and monthly, versus the JT momentum which holds only at monthly frequency. Also,
our risk momentum outperforms JT momentum in terms of higher Sharpe ratio and being
crash-free. Our risk-based momentum also holds for corporate bonds while the JT momentum
does not. Exploring our risk-based momentum in other asset markets is an important avenue
for future research. Given that momentum is one of the major anomalies in finance and is
extensively studied and applied, our discovery of a new class of momentum patterns may
provide a new stimulant of research in the area, and may be of wide interest.

39

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) All Day

(B) Morning Session (C) Afternoon Session

Fig. 1 Risk Momentum

This figure plots the cumulative risk (RISK ) component of the momentum strategy based on the risk signal
constructed from the 15 RP anomalies and their 95% confidence intervals for each event time. The cumulative
risk is calculated by compounding all hold-period risk components derived according to Equation (4). Each
trading day is divided into 13 intraday periods, including one overnight period from 16:00 on day d − 1 to
10:00 on day d and 12 half-hour periods. The portfolios are formed at the beginning of each 13 intraday
period and held up to 65 periods. Panel A shows the cumulative return averaged across all 13 formation
periods. Panels B and C display the cumulative return averaged from portfolios that are formed in the
morning (from 10:00 to 12:30) and afternoon (from 13:00 to 16:00) sessions, respectively. The sample includes
all stocks in the Russell 1000 index over the period from January 1993 to December 2020.

40

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) All Day

(B) Morning Session (C) Afternoon Session

Fig. 2 Risk-based Return Momentum

This figure plots the cumulative returns of the momentum strategy based on the risk component constructed
from the 15 RP anomalies and their 95% confidence intervals for each event time. Each trading day is divided
into 13 intraday periods, including one overnight period from 16:00 on day d − 1 to 10:00 on day d and 12
half-hour periods. The portfolios are formed at the beginning of each 13 intraday period and held up to 65
periods. Panel A shows the cumulative return averaged across all 13 formation periods. Panels B and C
display the cumulative return averaged from portfolios that are formed in the morning (from 10:00 to 12:30)
and afternoon (from 13:00 to 16:00) sessions, respectively. The sample includes all stocks in the Russell 1000
index over the period from January 1993 to December 2020.

41

Electronic copy available at: https://ssrn.com/abstract=4062260

Fig. 3 Risk-based Return Momentum: News Arrivals
This figure plots the cumulative returns of the momentum strategy based on RISK constructed from the 15
RP anomalies for each event time conditional on news arrivals. Each trading day is divided into 13 intraday
periods, including one overnight period from 16:00 on day d − 1 to 10:00 on day d and 12 half-hour periods.
The portfolios are formed at the beginning of each 13 intraday period and held up to 65 periods. The solid
(dash-dotted) line corresponds to the period with above (below) median number of news arrivals. The sample
includes all stocks in the Russell 1000 index over the period from January 2000 to December 2020 when the
RavenPack news data are available.

42

Electronic copy available at: https://ssrn.com/abstract=4062260

Fig. 4 Intraday Return Dispersion Pattern
This figure plots the time-series average of return dispersion (RD) over the course of the day. Each trading
day is divided into 13 intraday periods, including one overnight period from 16:00 to 10:00 and 12 half-hour
periods. We first measure RD in each intraday period on each day as the cross-sectional standard deviation
of stock returns and then calculate the average RD for each intraday period over all trading days. The x-axis
labels the end of each intraday period. The sample includes all stocks in the Russell 1000 index over the
period from January 1993 to December 2020.

43

Electronic copy available at: https://ssrn.com/abstract=4062260

Fig. 5 Performance of Daily Conditional Risk-based Return Momentum
This figure plots the cumulative portfolio value from investing in different daily conditional risk-based return
momentum strategies from January 1993 to December 2020. Conditional risk-based return momentum is
based on Russell 1000 stocks in the top risk-concentration quintile, where risk concentration is defined in
Equation (6). Daily strategies are formed on the basis of the past-day close-to-close RISK estimated from 15
RP anomalies, 15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO, respectively, and are
held over the subsequent day. The initial investment is $1. We further take logarithm of the portfolio value
with base ten so the trajectory starts from 0.

44

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) Annualized Return in Percentage

(B) Annualized Sharpe Ratio

Fig. 6 Risk-based Return Momentum: Different Signal Formation and Return Holding Periods
This figure shows the performance of the long-short portfolio based on the risk component constructed from
the 15 RP anomalies for various combinations of signal formation and portfolio holding periods. The x-axis
denotes the number of intraday periods (N ) in holding the long-short portfolios. The y-axis denotes the
number of intraday periods (M ) in calculating the signal. Following Jegadeesh and Titman (1993), our
trading strategy includes portfolios with overlapping holding periods. That is, for return holding period
N , we revise the weights on 1/N of the securities in our strategy on any given interval and carry over the
rest from the previous interval. Panel A shows the annualized return in percentage and Panel B shows the
annualized Sharpe ratio. The sample includes all stocks in the Russell 1000 index over the period from
January 1993 to December 2020.
45

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) 15 RP Anomalies

(B) 15 EL Anomalies

Fig. 7 Risk Momentum Attribution

We decompose the RISK component of the holding-period return for the risk-based spread portfolio into a
sum of individual anomaly components according to Equation (4). The x-axis on top of each figure denotes
the start and end of the portfolio holding period, and the y-axis denotes the anomaly component. The color
gradients within each square indicate the average value of the anomaly component across all trading days,
with dark blue and white indicating the highest and lowest values. Panels A and B display the results based
on 15 RP anomalies and 15 EL anomalies, respectively. The sample includes all stocks in the Russell 1000
index over the period from January 1993 to December 2020.

46

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) Momentum in Stocks

(B) Momentum in Bonds

Fig. 8 Monthly Momentum Across Asset Classes

This figure plots the cumulative portfolio value from investing in different momentum strategies, which include
the risk-based return momentum (RISK ) and the traditional return momentum from Jegadeesh and Titman
(1993). JT 6m (12m) is the traditional momentum strategy that long (short) stocks with the highest (lowest)
cumulative returns over the past 2 to 6 (12) months. Panel A reports the results for the stock market. The
stock RISK signal is constructed based on the 60 stock anomalies via LASSO. The sample period is from
January 1970 to December 2020. Panel B reports the results for the corporate bond market with the sample
from January 1974 to December 2020. The bond RISK signal is constructed based on the 25 bond anomalies.

47

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 1 Summary Statistics of Anomalies
Panel A reports the time-series average of the cross-sectional mean, standard deviation, and quantiles of 15 RP anomalies. Panel
B reports the same for 15 EL anomalies. RP anomalies are 15 representative anomalies including 11 mispricing anomalies of
Stambaugh, Yu, and Yuan (2012) and Beta, Size, Book-to-market ratio, and Reversal. EL anomalies are the 15 anomalies
investigated by Ehsani and Linnainmaa (2022). The sample includes all stocks in the Russell 1000 index over the period from
January 1993 to December 2020.

Panel A: 15 RP Anomalies
Acronym Anomaly Mean Std P1 P25 Median P75 P99

acc Accruals 0.00 0.05 -0.15 -0.02 0.00 0.02 0.16

agr Asset growth 0.17 0.36 -0.26 0.01 0.08 0.20 2.04
beta Beta 0.99 0.52 0.13 0.62 0.90 1.27 2.57
bm Book-to-market 0.64 1.58 -0.12 0.23 0.40 0.66 5.81
cei Composite equity issues -0.08 0.26 -0.99 -0.11 -0.05 -0.02 0.76
dis Distress -6.05 1.58 -8.04 -6.99 -6.38 -5.53 -0.52
gpf Gross profitability 0.29 0.23 -0.03 0.11 0.25 0.42 1.00
inta Investment-to-assets 0.10 0.29 -0.23 0.01 0.04 0.10 1.90
mom Momentum 0.20 0.38 -0.44 -0.02 0.13 0.33 1.59
rev Reversal 0.02 0.09 -0.19 -0.03 0.01 0.06 0.28
size Size 15.71 0.86 14.59 14.97 15.51 16.32 17.44
noa Net operating assets 0.58 0.36 -0.18 0.38 0.58 0.74 1.88
oscore O-score -3.64 1.70 -7.59 -4.65 -3.71 -2.79 1.19
roa Return on assets 0.01 0.03 -0.07 0.00 0.01 0.02 0.09
nsi Net stock issues 0.30 0.77 -0.13 -0.01 0.01 0.15 3.38

Panel B: 15 EL Anomalies
Acronym Anomaly Mean Std P1 P25 Median P75 P99

acc Accruals 0.00 0.05 -0.15 -0.02 0.00 0.02 0.16

bab Betting against beta 1.03 0.44 0.27 0.72 0.97 1.27 2.24
bm Book-to-market 0.64 1.58 -0.12 0.23 0.40 0.66 5.81
cfp Cash-flow to price 0.10 0.11 -0.16 0.05 0.08 0.13 0.49
ep Earnings to price 0.04 0.10 -0.29 0.03 0.05 0.07 0.18
prof Profitability 0.33 0.22 -0.03 0.16 0.29 0.45 1.02
rvar Residual variance 0.04 0.02 0.02 0.03 0.04 0.05 0.10
liq Liquidity 0.00 0.00 0.00 0.00 0.00 0.00 0.00
invest Investment 1.01 0.36 0.22 0.81 0.98 1.15 2.29
lrev Long-term reversals 0.48 0.75 -0.51 0.06 0.32 0.67 3.68
mom Momentum 0.20 0.38 -0.44 -0.02 0.13 0.33 1.59
srev Short-term reversals 0.02 0.09 -0.19 -0.03 0.01 0.06 0.28
size Size 15.71 0.86 14.59 14.97 15.51 16.32 17.44
qmj Quality minus junk 0.65 0.85 -1.48 0.10 0.87 1.36 1.70
nsi Net share issues 0.30 0.77 -0.13 -0.01 0.01 0.15 3.38

48

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 2 Risk Momentum
This table shows the predictability of RISK for future RISK and RES components of returns. The risk (RISK ) and residual
(RES ) components are constructed according to Equations (3) and (4). We form 13 long-short portfolios over a given intraday
period based on the RISK signal in the previous intraday period. Each of these portfolios enters position at time “Start” and
exits position at time “End” once per day. The last column “All” reports the performance of the long-short portfolio that is held
during all 13 intraday periods but is rebalanced every intraday period based on the RISK signal in the previous intraday period
(i.e., investing in all 13 signals). The row labeled “RISK ” (“RES ”) report the value-weighted RISK (RES ) component for each
long-short portfolio. We report the annualized return of the two components in percentage with Newey and West (1987) robust
t-statistics in parentheses. Panels A, B, C, and D report the results based on RISK estimated from 15 RP anomalies, 15 EL
anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO, respectively. The sample includes all stocks in the Russell 1000
index over the period from January 1993 to December 2020.

Return-Holding Period

Start Ret 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End Component 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

RISK 4.52 10.22 11.50 9.54 8.56 5.28 4.37 3.80 3.79 3.18 4.57 5.70 3.95 79.10
(2.57) (8.53) (12.17) (12.17) (12.79) (8.45) (8.40) (6.96) (6.57) (4.97) (7.25) (8.74) (5.40) (24.60)
RES 4.93 -0.39 0.55 1.05 0.89 0.79 0.53 0.67 0.98 1.31 1.23 1.45 2.17 15.96
(4.74) (-0.72) (1.19) (2.50) (2.62) (2.47) (1.72) (2.05) (2.96) (3.74) (3.33) (3.62) (4.98) (9.51)

Panel B: 15 EL Anomalies

RISK 5.53 9.68 11.22 9.73 8.54 4.75 4.07 3.71 3.76 2.52 4.30 5.58 3.73 77.12
(2.79) (8.01) (11.92) (12.50) (12.85) (7.62) (8.08) (6.93) (6.53) (4.01) (7.01) (8.76) (5.21) (23.95)
RES 3.36 -0.74 0.74 0.85 1.22 1.07 1.01 1.08 1.02 1.30 1.55 1.58 2.60 16.55
(3.15) (-1.38) (1.54) (2.10) (3.61) (3.13) (3.38) (3.22) (2.97) (3.74) (4.09) (4.00) (5.92) (9.78)

Panel C: 60 Anomalies via PCA

RISK 12.14 8.76 12.07 11.12 10.02 6.89 6.61 6.07 5.90 5.89 7.68 8.31 7.17 108.92
(4.62) (6.59) (10.44) (11.29) (11.95) (8.91) (10.02) (8.26) (7.75) (6.85) (8.87) (9.47) (7.39) (26.47)
RES 0.61 1.55 1.54 1.12 0.98 1.04 0.40 0.30 0.21 0.03 0.36 0.78 1.53 10.22
(0.78) (4.26) (4.55) (3.49) (3.62) (3.85) (1.54) (1.22) (0.79) (0.12) (1.27) (2.70) (4.49) (8.20)

Panel D: 60 Anomalies via LASSO

RISK 13.60 10.16 12.95 11.55 9.95 7.21 6.50 6.08 6.19 5.92 7.80 8.27 7.15 113.77
(5.03) (7.46) (11.13) (11.51) (11.92) (9.33) (9.81) (8.26) (8.17) (6.93) (9.01) (9.31) (7.05) (27.11)
RES 1.08 1.52 1.24 1.20 0.88 0.97 0.70 0.12 0.71 0.24 0.68 1.00 1.33 11.44
(1.31) (3.93) (3.69) (4.07) (3.29) (3.63) (2.76) (0.51) (2.90) (0.90) (2.46) (3.49) (4.14) (9.28)

49

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 3 Risk Momentum over Longer Horizon
This table decomposes the event-time return (with different portfolio formation time and holding period) into risk (RISK ) and
residual (RES ) components according to Equations (4) and (5), and reports the cumulative return of the two components in
basis points (bps) with Newey and West (1987) robust t-statistics in parentheses. The column labeled “Start” indicates when
the portfolios are formed. The portfolios are held with a different number of intraday periods ranging from 1 to 65, as shown in
the columns. The row labeled “Mean” reports the average return and t-statistic across different signal times. The results are
based on RISK estimated from 15 RP anomalies. The sample includes all stocks in the Russell 1000 index over the period from
January 1993 to December 2020.

Return-Holding Horizon (in Number of Intraday Periods)

Start Ret Component 1 2 3 5 8 12 13 39 65

10:00 RISK 4.05 5.43 7.31 9.93 11.84 13.74 18.12 25.13 24.87
(8.53) (8.46) (10.09) (11.91) (12.59) (11.79) (11.22) (9.42) (7.35)
RES -0.15 -0.34 -0.03 0.50 0.28 -0.06 0.33 -3.51 -5.41
(-0.72) (-0.95) (-0.07) (1.09) (0.54) (0.09) (0.39) (-2.64) (-3.33)
10:30 RISK 4.56 5.79 6.57 8.96 10.70 16.13 17.44 20.57 21.09
(12.17) (11.34) (10.72) (12.43) (12.57) (10.75) (10.80) (7.53) (6.39)
RES 0.22 0.02 -0.10 0.06 0.10 1.31 0.95 -3.24 -4.56
(1.19) (0.08) (-0.26) (0.14) (0.21) (1.64) (1.14) (-2.51) (-2.85)
11:00 RISK 3.78 5.64 6.79 7.98 9.35 16.26 17.52 21.70 21.48
(12.17) (12.58) (12.86) (12.49) (12.55) (11.39) (11.72) (8.00) (6.10)
RES 0.42 0.19 0.27 -0.11 -0.37 0.08 -0.35 -4.34 -6.47
(2.50) (0.67) (0.85) (-0.29) (-0.78) (0.10) (-0.43) (-3.18) (-3.95)
11:30 RISK 3.40 4.94 5.73 6.56 8.49 14.91 15.88 20.69 25.57
(12.79) (13.68) (12.99) (12.03) (11.80) (10.81) (10.98) (7.85) (7.78)
RES 0.36 0.09 -0.02 -0.45 -0.46 1.07 1.02 -2.41 -4.23
(2.62) (0.40) (-0.08) (-1.27) (-1.05) (1.42) (1.31) (-1.87) (-2.71)
12:00 RISK 2.10 3.00 3.48 3.70 4.75 12.06 12.55 12.98 15.23
(8.46) (8.85) (8.68) (6.49) (6.48) (8.24) (8.44) (5.33) (5.00)
RES 0.31 0.14 -0.09 -0.35 -0.66 0.15 -0.06 -3.67 -3.92
(2.47) (0.63) (-0.33) (-0.91) (-1.26) (0.19) (-0.08) (-2.75) (-2.47)
12:30 RISK 1.73 2.72 3.22 4.49 9.54 12.43 13.41 13.83 13.45
(8.40) (9.42) (8.90) (8.87) (8.83) (9.08) (9.53) (5.69) (4.34)
RES 0.21 0.18 0.10 -0.03 1.51 1.21 0.98 -4.26 -5.37
(1.72) (0.81) (0.40) (-0.08) (2.33) (1.58) (1.24) (-3.27) (-3.35)
13:00 RISK 1.51 2.12 2.81 3.53 6.41 8.71 9.25 10.00 10.17
(6.97) (6.70) (6.97) (6.61) (5.14) (6.13) (6.37) (4.21) (3.19)
RES 0.26 0.30 0.13 -0.03 0.46 0.55 0.41 -3.08 -4.45
(2.05) (1.31) (0.47) (-0.09) (0.65) (0.68) (0.51) (-2.37) (-2.75)
13:30 RISK 1.50 2.37 3.02 4.09 8.15 10.50 10.80 10.47 10.86
(6.57) (7.11) (7.81) (7.38) (6.61) (7.50) (7.59) (4.38) (3.63)
RES 0.39 0.36 0.02 -0.23 -0.41 -0.55 -0.57 -2.65 -4.13
(2.96) (1.54) (-0.06) (-0.59) (-0.56) (-0.71) (-0.73) (-2.05) (-2.58)
14:00 RISK 1.26 2.00 2.42 6.18 7.61 8.36 8.52 9.69 12.02
(4.97) (5.82) (5.24) (5.86) (5.85) (5.98) (6.03) (3.88) (4.05)
RES 0.52 0.44 0.34 1.22 1.14 1.16 0.84 -2.64 -3.30
(3.74) (1.78) (1.06) (1.81) (1.44) (1.36) (0.99) (-1.96) (-2.05)
14:30 RISK 1.82 3.27 3.78 9.25 11.04 11.91 11.75 7.79 9.10
(7.25) (8.67) (7.26) (7.74) (8.05) (8.05) (7.84) (3.12) (3.01)
RES 0.49 0.96 1.11 2.05 1.42 1.03 0.70 -4.77 -5.92
(3.33) (3.57) (3.12) (2.67) (1.73) (1.20) (0.81) (-3.44) (-3.51)
15:00 RISK 2.26 3.49 4.64 6.87 8.45 8.99 8.60 4.61 7.06
(8.74) (8.06) (4.38) (5.37) (5.80) (5.82) (5.50) (1.87) (2.28)
RES 0.58 0.84 1.55 1.15 1.04 0.37 -0.20 -4.42 -4.05
(3.62) (2.50) (2.31) (1.51) (1.29) (0.42) (-0.22) (-3.01) (-2.40)
15:30 RISK 1.57 5.50 7.53 9.34 10.72 10.53 10.35 9.48 11.89
(5.40) (5.77) (6.87) (7.36) (7.70) (7.05) (6.91) (3.81) (3.87)
RES 0.86 1.82 1.50 1.33 0.72 -0.48 -0.93 -4.35 -6.46
(4.98) (2.89) (2.21) (1.80) (0.90) (-0.55) (-1.04) (-3.22) (-4.00)
16:00 RISK 1.79 3.73 4.56 5.40 5.86 6.29 6.08 2.37 2.03
(2.57) (3.86) (4.40) (4.70) (4.63) (4.72) (4.47) (1.03) (0.71)
RES 1.96 1.87 1.92 1.20 0.92 0.09 -1.20 -3.93 -3.90
(4.74) (3.04) (2.91) (1.75) (1.24) (0.11) (-1.43) (-2.90) (-2.42)
Mean RISK 2.41 3.85 4.76 6.64 8.69 11.60 12.33 13.02 14.22
(8.08) (8.49) (8.24) (8.40) (8.35) (8.26) (8.26) (5.09) (4.44)
RES 0.49 0.53 0.52 50
0.49 0.44 0.46 0.15 -3.64 -4.78
(2.71) (1.40) (0.96) (0.58) (0.51) (0.58) (0.19) (-2.71) (-2.95)
Electronic copy available at: https://ssrn.com/abstract=4062260
Table 4 RISK Component Decomposition
This table reports the three components of the Lo and MacKinlay (1990) decomposition of the holding-period RISK from
the risk momentum strategy. The row labeled “Total” denotes the total holding-period RISK ; “Auto” is the first component
attributable to the autocovariance of RISK ; “Cross” is the second component attributable to the cross-autocovariance; and
“Dispersion” is the third component representing the dispersion in expected RISK captured by past average RISK. Newey and
West (1987) robust t-statistics are reported in parentheses. Panels A, B, C, and D report the results based on RISK estimated
from 15 RP anomalies, 15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO, respectively. The sample period
is from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

Total 12.49 13.03 13.46 12.94 9.28 7.79 5.83 4.26 3.61 4.17 6.39 8.11 6.36 107.98
(4.33) (9.93) (12.63) (13.44) (11.86) (10.87) (8.65) (6.31) (5.09) (4.99) (7.42) (8.81) (6.51) (24.83)
Auto 13.83 11.81 12.63 11.80 8.17 7.07 5.45 3.40 3.91 2.79 6.00 8.61 6.96 103.28
(4.11) (7.22) (10.22) (11.85) (8.47) (8.42) (6.82) (4.67) (4.48) (2.83) (5.36) (7.32) (5.65) (20.29)
Cross -1.88 1.84 0.78 1.05 1.06 0.68 0.36 0.85 -0.29 1.38 0.37 -0.55 -0.58 4.70
(-0.86) (1.59) (0.93) (1.50) (1.39) (1.30) (0.74) (1.47) (-0.50) (2.12) (0.46) (-0.56) (-0.65) (1.37)
Dispersion 0.54 -0.62 0.05 0.09 0.05 0.04 0.01 0.01 0.00 0.01 0.02 0.04 -0.02 0.00

Panel B: 15 EL Anomalies

Total 11.59 12.69 12.76 12.88 9.27 7.39 5.83 4.17 3.33 3.63 5.89 8.09 6.10 103.70
(4.02) (9.38) (11.75) (13.03) (11.60) (10.18) (8.57) (6.20) (4.63) (4.36) (7.03) (8.88) (6.21) (23.75)
Auto 12.31 11.70 11.86 11.57 8.14 6.79 5.39 3.06 3.82 2.18 5.56 8.37 6.32 97.30
(3.64) (6.77) (9.42) (11.31) (8.28) (8.27) (6.87) (4.21) (4.26) (2.18) (5.03) (6.91) (4.99) (18.87)
Cross -1.07 1.81 0.84 1.23 1.08 0.57 0.43 1.10 -0.49 1.43 0.30 -0.35 -0.22 6.41
(-0.47) (1.49) (0.99) (1.65) (1.39) (1.10) (0.90) (1.87) (-0.82) (2.16) (0.39) (-0.35) (-0.25) (1.82)
Dispersion 0.35 -0.82 0.05 0.08 0.05 0.03 0.01 0.01 0.00 0.02 0.03 0.06 0.00 -0.02

Panel C: 60 Anomalies via PCA

Total 15.32 13.53 14.82 14.39 10.45 9.75 7.75 6.08 4.91 5.95 8.74 10.51 8.88 132.64
(4.62) (9.88) (12.33) (13.27) (11.37) (11.31) (9.66) (7.11) (5.77) (5.79) (8.78) (9.42) (7.77) (26.41)
Auto 14.21 12.84 13.66 12.89 9.15 8.53 7.06 4.61 5.30 4.09 7.06 8.69 6.98 117.13
(4.04) (7.45) (11.23) (12.50) (9.48) (9.68) (8.65) (5.98) (5.94) (3.86) (6.44) (7.68) (5.62) (22.32)
Cross 0.76 1.56 1.09 1.44 1.26 1.22 0.69 1.46 -0.38 1.81 1.67 1.75 1.87 15.47
(0.33) (1.18) (1.26) (1.88) (1.50) (2.13) (1.33) (2.11) (-0.58) (2.43) (2.07) (1.76) (2.02) (4.20)
Dispersion 0.35 -0.87 0.07 0.06 0.04 0.01 0.01 0.02 0.00 0.06 0.01 0.07 0.04 0.04

Panel D: 60 Anomalies via LASSO

Total 15.95 14.67 15.58 15.56 10.57 9.73 7.65 6.01 5.36 6.21 9.13 11.11 8.66 137.84
(4.67) (10.87) (13.07) (13.68) (11.94) (11.04) (9.68) (7.03) (6.26) (6.00) (8.75) (10.09) (7.45) (26.94)
Auto 18.17 14.00 14.53 14.37 9.24 8.79 7.23 4.47 5.79 4.40 8.10 10.94 8.18 130.75
(4.82) (8.41) (11.39) (13.14) (8.84) (8.95) (8.32) (5.22) (5.86) (3.97) (6.29) (8.65) (5.99) (23.14)
Cross -2.59 1.34 0.97 1.13 1.27 0.90 0.42 1.53 -0.43 1.75 1.00 0.10 0.49 7.02
(-1.05) (1.05) (1.07) (1.43) (1.48) (1.46) (0.76) (2.21) (-0.61) (2.29) (1.08) (0.10) (0.49) (1.81)
Dispersion 0.37 -0.68 0.08 0.06 0.06 0.03 0.00 0.01 0.00 0.06 0.03 0.07 0.00 0.07

51

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 5 Intraday Factor Return Autocorrelation
This table reports the autocorrelation of intraday factor returns. Each trading day is divided into 13 intraday periods, including
one overnight period from 16:00 on day d − 1 to 10:00 on day d and 12 half-hour periods. For each intraday period, factor
returns are estimated by the slope coefficients in Equation (2). The order of the autocorrelations ranges from 1 to 65 intraday
periods. *, **, and *** denote significance at 10%, 5%, and 1%, respectively. Panel A reports the results for 15 RP anomalies
and Panel B reports the same for 15 EL anomalies. The sample includes all stocks in the Russell 1000 index over the period
from January 1993 to December 2020.

Order of Autocorrelation (in Number of Intraday Periods)

Factor 1 2 3 5 8 13 39 65

Panel A: 15 RP Anomalies

Accruals 0.01* -0.01 0.00 0.00 0.00 0.02*** 0.01* 0.00

Asset growth 0.02*** 0.01* 0.00 0.01*** 0.00 0.02*** 0.01* 0.00
Beta 0.10*** 0.03*** 0.02*** 0.03*** 0.01*** 0.00 0.01* 0.00
Book-to-market 0.04*** 0.01** 0.01*** 0.01*** 0.01 0.02*** 0.02*** 0.02***
Composite equity issues 0.02*** 0.01 0.01*** 0.01*** 0.00 0.02*** 0.02*** 0.02***
Distress 0.11*** 0.05*** 0.03*** 0.03*** 0.02*** 0.05*** 0.02*** 0.02***
Gross profitability 0.12*** 0.05*** 0.04*** 0.03*** 0.02*** 0.07*** 0.06*** 0.04***
Investment-to-assets 0.06*** 0.01*** 0.02*** 0.01*** 0.00 0.01*** 0.01** 0.01***
Momentum 0.05*** 0.01*** 0.00 0.00 0.00 0.01*** 0.00 0.00
Reversal 0.03*** -0.01 0.00 0.00 0.00 0.03*** 0.01** 0.01**
Size -0.01 -0.01 -0.01 0.00 0.00 0.01** 0.00 0.01 ***
Net operating assets 0.06*** 0.02*** 0.01*** 0.01*** 0.01* 0.02*** 0.01*** 0.01***
O-score 0.02*** 0.01 0.00 0.01** 0.00 0.00 0.02*** 0.01***
Return on assets 0.03*** 0.00 0.00 0.00 0.00 0.01*** 0.00 0.00
Net stock issues 0.02*** 0.00 0.00 0.01*** 0.00 0.03*** 0.02*** 0.02***

Panel B: 15 EL Anomalies

Accruals 0.00 0.00 0.00 0.00 0.00 0.02*** 0.01** -0.01

Betting against beta 0.09*** 0.02*** 0.02*** 0.02*** 0.01*** -0.01 0.00 -0.01
Book-to-market 0.02*** 0.00 0.00 0.00 0.00 0.01** 0.01*** 0.00
Cash-flow to price 0.03*** 0.01** 0.01*** 0.00 0.00 0.02*** 0.01*** 0.01***
Earnings to price 0.12*** 0.06*** 0.03*** 0.03*** 0.02*** 0.05*** 0.02*** 0.01***
Profitability 0.12*** 0.05*** 0.04*** 0.03*** 0.02*** 0.08*** 0.06*** 0.04***
Residual variance 0.04*** 0.01*** 0.01* 0.01** 0.01** 0.03*** 0.01*** 0.00
Liquidity 0.06*** 0.01** 0.01*** 0.01*** 0.01* 0.01*** 0.01*** 0.01***
Investment 0.07*** 0.02*** 0.01*** 0.01*** 0.01*** 0.03*** 0.01*** 0.03***
Long-term reversals 0.04*** 0.00 0.01*** 0.01*** 0.00 0.01*** 0.01*** 0.01***
Momentum 0.01*** 0.00 0.00 0.00 0.01* 0.01*** 0.01* 0.00
Short-term reversals 0.06*** 0.02*** 0.01*** 0.01*** 0.00 0.01*** 0.02*** 0.02***
Size 0.05*** 0.01** 0.01** 0.01*** 0.00 0.01*** 0.02*** 0.01***
Quality minus junk 0.04*** 0.01*** 0.01** 0.00 0.00 0.01** 0.00 0.00
Net share issues 0.00 -0.01 0.00 0.00 0.00 0.04*** 0.02*** 0.02***

52

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 6 Risk-based Return Momentum
This table reports the performance of 13 long-short portfolios over a given intraday period based on the RISK signal in the
previous intraday period. Each of these portfolios enters position at time “Start” and exits position at time “End” once per day.
The last column “All” reports the performance of the long-short portfolio that is held during all 13 intraday periods but is
rebalanced every intraday period based on the RISK signal in the previous intraday period (i.e., investing in all 13 signals). The
rows labeled “Return” report the annualized return of the long-short portfolio in percentage with Newey and West (1987) robust
t-statistics in parentheses. The rows labeled “Alpha” report the annualized CAPM alpha. Panels A, B, C, and D report the
results based on RISK estimated from 15 RP anomalies, 15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO,
respectively. The sample includes all stocks in the Russell 1000 index over the period from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

Return 9.45 9.83 12.06 10.59 9.46 6.06 4.90 4.47 4.77 4.49 5.81 7.15 6.12 95.06
(3.50) (6.94) (10.33) (10.53) (11.26) (7.82) (7.36) (6.23) (6.28) (5.32) (6.86) (7.95) (6.21) (22.74)
Alpha 9.33 9.66 11.87 10.48 9.45 6.01 5.04 4.46 4.74 4.51 5.85 7.22 6.04 95.10
(3.46) (6.77) (10.15) (10.53) (11.44) (7.75) (7.53) (6.28) (6.09) (5.26) (6.89) (8.09) (6.20) (22.50)

Panel B: 15 EL Anomalies

Return 8.88 8.94 11.95 10.58 9.76 5.82 5.08 4.79 4.77 3.82 5.85 7.16 6.33 93.67
(3.25) (6.22) (10.21) (10.69) (11.79) (7.15) (7.80) (6.68) (6.17) (4.58) (7.04) (8.15) (6.43) (22.29)
Alpha 8.73 8.77 11.78 10.46 9.75 5.78 5.24 4.78 4.74 3.83 5.87 7.23 6.26 93.70
(3.19) (6.11) (10.03) (10.69) (11.93) (7.09) (8.04) (6.75) (5.99) (4.53) (7.03) (8.27) (6.41) (22.20)

Panel C: 60 Anomalies via PCA

Return 12.75 10.31 13.61 12.24 11.00 7.93 7.00 6.37 6.11 5.92 8.05 9.09 8.70 119.14
(4.08) (6.92) (10.28) (10.66) (11.32) (8.59) (8.86) (7.35) (6.74) (5.89) (7.86) (8.73) (7.46) (24.69)
Alpha 12.66 10.04 13.40 12.11 10.96 7.87 7.14 6.37 6.06 5.93 8.10 9.19 8.60 119.17
(4.05) (6.78) (10.15) (10.65) (11.53) (8.54) (9.04) (7.45) (6.52) (5.81) (7.88) (8.89) (7.41) (24.36)

Panel D: 60 Anomalies via LASSO

Return 14.68 11.68 14.19 12.76 10.83 8.18 7.20 6.20 6.90 6.16 8.47 9.27 8.48 125.21
(4.62) (7.55) (10.77) (11.23) (11.46) (8.84) (8.99) (7.28) (7.78) (6.14) (8.36) (8.94) (7.08) (25.67)
Alpha 14.58 11.48 13.98 12.62 10.80 8.12 7.34 6.20 6.85 6.16 8.53 9.40 8.41 125.23
(4.59) (7.42) (10.62) (11.21) (11.62) (8.77) (9.12) (7.30) (7.52) (6.04) (8.41) (9.13) (7.11) (25.62)

53

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 7 Decile Portfolio Performance
This table reports the annualized return in percentage of decile portfolios over a given intraday period based on the RISK signal
in the previous intraday period. Each of these portfolios enters position at time “Start” and exits position at time “End” once
per day. The last column “All” reports the performance of the decile portfolios that are held during all 13 intraday periods but
are rebalanced every intraday period based on the RISK signal in the previous intraday period (i.e., investing in all 13 signals).
The row labeled “High-Low” reports the annualized return of the long-short portfolios in percentage with Newey and West
(1987) robust t-statistics in parentheses. Panels A and B report the results based on RISK estimated from 15 RP anomalies and
15 EL anomalies, respectively. The sample includes all stocks in the Russell 1000 index over the period from January 1993 to
December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

1 2.77 -7.80 -7.51 -5.91 -4.25 -3.58 -0.70 -1.46 -2.94 -2.70 -1.37 -2.77 -1.00 -39.14
2 1.51 -4.99 -4.85 -4.09 -2.71 -2.37 -0.53 -0.98 -1.89 -2.22 -0.38 -1.27 -0.29 -25.05
3 2.37 -3.25 -3.99 -2.70 -2.00 -1.30 -0.24 -0.40 -1.27 -1.35 0.61 -0.21 0.18 -13.57
4 4.21 -2.46 -3.02 -1.82 -1.26 -1.12 0.77 0.03 -0.77 -0.94 1.10 0.30 0.49 -4.49
5 4.62 -1.58 -1.70 -0.57 0.13 -0.39 1.09 0.34 -0.20 -0.51 1.51 0.85 1.21 4.79
6 4.68 -0.96 -0.39 0.07 0.55 0.18 1.69 0.80 -0.24 0.00 1.99 1.53 1.90 11.80
7 6.54 -0.10 0.77 0.81 1.59 0.49 1.79 1.40 0.02 0.47 2.70 2.30 2.35 21.11
8 7.31 0.71 2.20 1.60 2.40 0.94 2.46 1.68 0.58 0.63 3.13 3.24 2.82 29.72
9 8.77 1.59 3.43 2.50 3.35 1.67 3.40 2.44 0.87 0.91 3.84 3.67 3.65 40.09
10 12.22 2.03 4.55 4.68 5.21 2.48 4.19 3.01 1.83 1.79 4.43 4.38 5.12 55.92
High-Low 9.45 9.83 12.06 10.59 9.46 6.06 4.90 4.47 4.77 4.49 5.81 7.15 6.12 95.06
(3.50) (6.94) (10.33) (10.53) (11.26) (7.82) (7.36) (6.23) (6.28) (5.32) (6.86) (7.95) (6.21) (22.74)

Panel B: 15 EL Anomalies

1 3.21 -7.41 -7.16 -5.78 -4.31 -3.45 -0.86 -1.65 -2.74 -2.36 -1.20 -2.57 -1.02 -37.24
2 2.02 -4.65 -5.33 -4.14 -2.72 -2.31 -0.17 -0.86 -1.84 -1.75 -0.32 -1.47 -0.34 -23.87
3 2.54 -3.30 -3.69 -2.75 -1.77 -1.63 0.31 -0.35 -1.37 -1.20 0.61 -0.36 -0.17 -13.12
4 3.29 -2.29 -2.16 -1.88 -1.09 -0.90 0.65 0.15 -0.73 -1.04 1.26 0.33 0.76 -3.63
5 4.31 -1.57 -1.64 -1.01 -0.13 -0.56 0.97 0.18 -0.35 -0.30 1.70 1.23 1.39 4.21
6 4.76 -1.23 -0.51 -0.13 0.50 -0.04 1.34 0.87 -0.22 0.03 2.02 1.43 1.83 10.63
7 5.13 -0.11 0.50 0.90 1.28 0.51 2.00 1.44 0.08 0.07 2.14 2.16 2.45 18.57
8 7.22 0.63 1.80 1.80 2.44 1.23 2.52 1.99 0.43 0.47 2.93 3.06 2.77 29.29
9 9.89 1.30 2.86 2.69 3.39 1.69 3.03 1.96 0.84 0.59 3.66 3.74 3.77 39.40
10 12.10 1.52 4.79 4.80 5.45 2.38 4.22 3.14 2.03 1.46 4.65 4.59 5.31 56.43
High-Low 8.88 8.94 11.95 10.58 9.76 5.82 5.08 4.79 4.77 3.82 5.85 7.16 6.33 93.67
(3.25) (6.22) (10.21) (10.69) (11.79) (7.15) (7.80) (6.68) (6.17) (4.58) (7.04) (8.15) (6.43) (22.29)

54

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 8 Risk-based Return Momentum over Longer Horizon
This table reports the event-time cumulative returns to the long-short portfolios based on the intraday RISK signal available at
time “Start” and held up to 65 intraday periods. Each trading day is divided into 13 intraday periods, including one overnight
period from 16:00 on day d − 1 to 10:00 on day d and 12 half-hour periods. The event-time returns are reported in basis points
(bps) with Newey and West (1987) robust t-statistics in parentheses. The row labeled “Mean” reports the average return and
t-statistic across different signal times. The results are based on RISK estimated from 15 RP anomalies. The sample includes
all stocks in the Russell 1000 index over the period from January 1993 to December 2020.

Return-Holding Horizon (in Number of Intraday Periods)

Start 1 2 3 5 8 12 13 39 65

10:00 3.90 5.08 7.28 10.43 12.13 13.68 18.45 21.63 19.46
(6.94) (6.68) (8.47) (10.51) (11.08) (9.87) (9.93) (7.70) (5.79)
10:30 4.78 5.81 6.47 9.01 10.80 17.44 18.39 17.33 16.54
(10.33) (9.25) (8.54) (10.52) (10.69) (9.78) (9.73) (6.05) (5.05)
11:00 4.20 5.83 7.06 7.87 8.97 16.34 17.17 17.36 15.01
(10.53) (10.32) (10.85) (10.23) (9.82) (9.65) (9.81) (6.12) (4.32)
11:30 3.75 5.03 5.71 6.12 8.03 15.98 16.90 18.28 21.34
(11.26) (11.20) (10.34) (8.97) (9.09) (9.65) (9.80) (6.53) (6.54)
12:00 2.41 3.15 3.38 3.35 4.09 12.22 12.49 9.31 11.32
(7.82) (7.42) (6.67) (4.58) (4.30) (7.04) (7.15) (3.53) (3.65)
12:30 1.94 2.90 3.33 4.46 11.05 13.65 14.39 9.56 8.08
(7.36) (7.75) (7.13) (6.86) (8.22) (8.38) (8.63) (3.68) (2.63)
13:00 1.77 2.42 2.95 3.50 6.87 9.27 9.67 6.91 5.72
(6.23) (5.88) (5.70) (5.09) (4.51) (5.47) (5.64) (2.65) (1.80)
13:30 1.89 2.72 3.04 3.86 7.74 9.95 10.23 7.82 6.73
(6.28) (6.38) (5.91) (5.34) (5.06) (5.98) (6.05) (3.00) (2.18)
14:00 1.78 2.44 2.77 7.40 8.75 9.52 9.36 7.04 8.72
(5.32) (5.29) (4.41) (5.42) (5.34) (5.49) (5.41) (2.55) (2.82)
14:30 2.30 4.23 4.88 11.30 12.45 12.94 12.45 3.03 3.18
(6.86) (8.32) (7.10) (7.39) (7.45) (7.38) (7.01) (1.11) (1.02)
15:00 2.84 4.33 6.19 8.02 9.50 9.36 8.40 0.19 3.00
(7.95) (7.09) (4.49) (5.01) (5.39) (5.10) (4.56) (0.07) (0.95)
15:30 2.43 7.32 9.03 10.67 11.44 10.05 9.42 5.13 5.43
(6.21) (5.83) (6.50) (6.85) (6.79) (5.70) (5.35) (1.93) (1.75)
16:00 3.75 5.60 6.48 6.60 6.78 6.38 4.88 -1.57 -1.87
(3.50) (4.59) (4.95) (4.72) (4.48) (4.06) (3.03) (-0.63) (-0.64)

Mean 2.90 4.37 5.27 7.12 9.12 12.06 12.48 9.39 9.43
(7.43) (7.38) (7.00) (7.04) (7.09) (7.20) (7.09) (3.41) (2.91)

55

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 9 Risk-based Return Momentum over High and Low Return Dispersion Periods
This table reports the performance of the risk-based return momentum over high and low return dispersion periods. Each
trading day is divided into 13 intraday periods, including one overnight period from 16:00 on day d − 1 to 10:00 on day d and 12
half-hour periods. For each intraday period on each day, we compute return dispersion (RD) as the cross-sectional standard
deviation of stock returns. We compute the time-series median of RD for each of the 13 intraday periods, and the time-series
median of RD using all intraday periods. The row labeled “High” (“Low”) corresponds to the sample period with above (below)
median RD level. Newey and West (1987) robust t-statistics are reported in parentheses. Panels A, B, C, and D report the
results based on RISK estimated from 15 RP anomalies, 15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO,
respectively. The sample includes all stocks in the Russell 1000 index over the period from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
RD End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

High 16.06 16.35 19.40 17.18 15.60 9.47 7.97 6.87 7.23 7.74 9.95 13.07 11.71 157.57
(3.38) (6.25) (9.23) (9.61) (10.51) (6.69) (6.66) (5.16) (5.13) (4.94) (6.33) (7.83) (6.22) (21.18)

Low 2.79 3.32 4.71 3.99 3.31 2.66 1.82 2.06 2.31 1.23 1.64 1.20 0.50 32.16
(1.13) (3.20) (4.89) (5.17) (4.72) (4.20) (3.33) (4.02) (4.33) (2.02) (2.97) (2.22) (0.97) (8.74)

Panel B: 15 EL Anomalies

High 16.56 14.64 18.98 17.27 16.40 8.78 8.14 7.23 7.49 7.01 9.79 12.69 11.86 156.81
(3.48) (5.53) (8.98) (9.87) (11.33) (5.89) (6.96) (5.45) (5.20) (4.57) (6.34) (7.75) (6.30) (21.11)

Low 1.14 3.24 4.92 3.89 3.12 2.87 2.02 2.34 2.06 0.60 1.89 1.61 0.77 30.34
(0.45) (3.06) (5.06) (5.05) (4.41) (4.47) (3.80) (4.44) (3.77) (1.02) (3.46) (3.07) (1.49) (8.02)

Panel C: 60 Anomalies via PCA

High 22.25 15.72 21.10 20.35 17.64 12.55 11.63 10.04 9.45 10.12 13.26 16.14 16.33 203.46
(3.93) (5.78) (8.79) (9.90) (10.18) (7.37) (8.11) (6.18) (5.54) (5.40) (6.91) (8.20) (7.34) (23.04)

Low 3.45 4.83 6.14 4.16 4.40 3.39 2.46 2.80 2.82 1.75 2.84 2.16 1.16 38.50
(1.28) (4.16) (5.82) (4.89) (5.50) (4.90) (3.99) (4.63) (4.58) (2.51) (4.65) (3.65) (1.95) (9.62)

Panel D: 60 Anomalies via LASSO

High 22.92 18.92 21.87 20.74 17.20 13.33 12.08 9.95 10.66 10.37 14.00 16.30 15.47 210.28
(4.03) (6.70) (9.31) (10.29) (10.25) (7.85) (8.34) (6.21) (6.38) (5.51) (7.36) (8.38) (6.74) (23.64)

Low 6.57 4.38 6.49 4.77 4.49 3.05 2.38 2.56 3.20 1.97 2.98 2.31 1.55 42.79
(2.32) (3.71) (5.89) (5.50) (5.69) (4.28) (3.76) (4.36) (5.38) (2.96) (4.84) (3.83) (2.65) (10.54)

56

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 10 Risk-based Return Momentum Conditional on Risk Concentration
This table reports the performance of risk-based return momentum for Russell 1000 stocks in the top and bottom RiskCon
quintiles. RiskCon, defined in Equation (6), captures a stock’s risk concentration. The rows labeled “High” (“Low”) denotes
stocks in the top (bottom) risk-concentration quintile. Newey and West (1987) robust t-statistics are reported in parentheses.
Panels A, B, C, and D report the results based on RISK estimated from 15 RP anomalies, 15 EL anomalies, 60 anomalies via
PCA, and 60 anomalies via LASSO, respectively. The sample period is from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
RiskCon End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

High 23.21 14.52 16.41 15.17 11.86 9.50 7.97 7.13 7.80 7.58 9.42 11.23 9.98 151.64
(7.70) (9.37) (13.78) (14.91) (13.38) (11.85) (10.52) (9.51) (10.31) (9.27) (10.96) (12.34) (10.19) (33.97)

Low -16.52 -0.72 -2.03 -2.75 1.85 -2.65 -2.89 -1.10 -3.32 -2.11 0.21 -1.65 -1.47 -35.29
(-6.09) (-0.53) (-1.59) (-2.64) (1.88) (-2.68) (-3.31) (-1.25) (-3.59) (-2.26) (0.21) (-1.57) (-1.25) (-7.82)

Panel B: 15 EL Anomalies

High 24.26 15.01 16.27 15.61 12.54 8.93 8.83 7.86 7.92 6.94 9.67 11.17 9.74 154.64
(8.21) (9.65) (13.80) (15.11) (14.52) (10.53) (11.95) (10.37) (10.55) (8.62) (11.74) (12.64) (10.57) (34.99)

Low -17.57 -0.45 -3.16 -2.46 -1.35 -2.85 -2.91 -2.32 -1.54 -2.56 -2.18 -0.93 -1.40 -41.83
(-6.54) (-0.31) (-2.63) (-2.20) (-1.39) (-2.85) (-3.46) (-2.59) (-1.70) (-2.46) (-2.08) (-0.91) (-1.13) (-9.18)

Panel C: 60 Anomalies via PCA

High 28.98 15.50 17.17 16.65 14.13 11.01 9.89 8.37 8.83 9.35 11.97 12.41 11.53 176.00
(8.49) (10.16) (12.82) (14.62) (14.20) (12.38) (11.69) (9.76) (10.63) (10.32) (12.65) (12.38) (10.23) (35.34)

Low -17.89 -1.13 -2.71 -0.65 -0.81 -1.75 -2.18 -0.11 -1.19 -1.05 0.65 0.42 -2.09 -30.28
(-5.41) (-0.70) (-1.77) (-0.52) (-0.72) (-1.52) (-2.11) (-0.11) (-1.02) (-0.95) (0.53) (0.31) (-1.46) (-5.56)

Panel D: 60 Anomalies via LASSO

High 31.85 16.21 18.41 17.65 13.66 11.41 10.32 8.94 10.27 9.34 12.45 13.07 11.70 185.65
(9.02) (10.13) (14.00) (15.39) (14.09) (12.06) (11.67) (10.28) (11.70) (10.04) (12.56) (12.53) (10.63) (36.09)

Low -18.81 1.95 -2.23 -1.16 1.57 -2.94 -2.45 0.31 -1.51 -1.47 -0.41 -1.53 -3.07 -31.50
(-6.88) (1.33) (-1.69) (-1.03) (1.50) (-2.85) (-2.84) (0.33) (-1.50) (-1.41) (-0.38) (-1.40) (-2.48) (-6.67)

57

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 11 Daily Risk-based Return Momentum
This table reports the performance of decile portfolios formed on the basis of daily close-to-close RISK signal and held over
the subsequent day. Panel A reports the annualized portfolio returns in percentage for the unconditional risk-based return
momentum strategies formed on all stocks in the Russell 1000 index. Panel B reports the results for the conditional risk-based
return momentum strategies formed on Russell 1000 stocks in the top risk-concentration quintile, where risk concentration is
defined in Equation (6). The RISK signal is constructed using 15 RP anomalies, 15 EL anomalies, 60 anomalies via PCA, or 60
anomalies via LASSO. The rows labeled “High-Low” and “FF5 Alpha” respectively report the annualized return and the Fama
and French (2015) five-factor alpha of the long-short portfolio in percentage, with Newey and West (1987) robust t-statistics in
parentheses. The sample period is from January 1993 to December 2020.

15 RP Anomalies 15 EL Anomalies 60 Anomalies via PCA 60 Anomalies via LASSO

Panel A: Unconditional risk-based return momentum

1 (Low) -17.58 -16.82 -16.93 -19.42

2 -9.89 -8.60 -10.13 -11.45
3 -5.45 -3.48 -4.12 -4.99
4 -0.08 -0.57 -0.07 -0.20
5 4.45 4.15 4.94 3.59
6 7.62 8.27 7.72 9.16
7 12.85 11.80 11.34 12.00
8 15.07 15.15 15.38 16.01
9 20.01 18.88 19.28 21.59
10 (High) 27.24 27.56 28.22 28.59
High-Low 44.82 44.37 45.15 48.01
(9.02) (9.05) (9.16) (9.11)
FF5 Alpha 43.72 44.62 45.34 47.83
(8.70) (9.12) (9.22) (9.17)

Panel B: Conditional risk-based return momentum

1 (Low) -22.77 -23.05 -22.15 -27.52

2 -14.56 -14.07 -17.00 -17.96
3 -5.29 -8.35 -7.46 -8.04
4 1.24 -1.48 -1.68 -0.18
5 4.30 3.77 4.30 4.65
6 9.95 9.45 10.11 10.58
7 14.69 13.69 15.47 14.42
8 21.22 23.01 19.97 21.91
9 26.66 27.81 29.44 27.80
10 (High) 34.47 36.87 36.69 37.35
High-Low 57.24 59.93 58.85 64.87
(11.46) (11.63) (11.96) (12.24)
FF5 Alpha 57.21 60.15 59.15 64.80
(11.64) (11.73) (12.04) (12.35)

58

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 12 Risk-based Return Momentum : 205 Anomalies
This table reports the performance of risk-based return momentum , where the RISK signal is estimated from 205 anomalies in
Chen and Zimmermann (2022). Panels A and B report the results based on RISK estimated via PCA and LASSO, respectively.
Newey and West (1987) robust t-statistics are reported in parentheses. The sample includes all stocks in the Russell 1000 index
over the period from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 205 Anomalies via PCA

10.00 8.32 10.07 9.42 8.31 5.62 4.92 4.89 4.31 4.47 6.42 6.51 5.03 88.33
(3.69) (6.87) (9.54) (10.08) (10.63) (7.46) (7.55) (6.54) (5.94) (5.38) (7.75) (7.63) (5.20) (22.10)

Panel B: 205 Anomalies via LASSO

4.22 11.96 13.36 11.94 10.96 7.36 6.38 5.64 5.44 4.79 7.17 8.29 7.35 105.26
(1.39) (7.46) (9.74) (10.07) (11.01) (7.63) (8.11) (6.31) (5.99) (4.70) (7.01) (7.86) (6.24) (21.75)

59

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 13 Conditional Risk-based Return Momentum: Day-of-week Effect
This table reports the performance of conditional risk-based return momentum on each day of the week. Conditional risk-based
return momentum is formed on Russell 1000 stocks in the top risk-concentration quintile, where risk concentration is defined in
Equation (6). Panels A, B, C, and D report the results based on RISK estimated from 15 RP anomalies, 15 EL anomalies,
60 anomalies via PCA, and 60 anomalies via LASSO, respectively. Newey and West (1987) robust t-statistics are reported in
parentheses. The sample period is from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
Weekday End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Mon 11.56 24.62 20.40 16.70 11.06 11.48 8.03 8.57 7.62 7.43 12.41 12.87 9.57 162.80
(1.81) (6.15) (8.18) (8.28) (5.44) (6.57) (5.44) (5.19) (5.11) (4.02) (7.89) (5.87) (4.41) (16.55)

Tue 21.66 12.09 17.18 14.68 11.36 8.91 6.40 7.00 7.86 6.93 11.19 15.62 17.06 158.13
(3.51) (3.63) (5.97) (6.57) (5.87) (5.44) (4.03) (4.27) (4.37) (4.24) (5.53) (7.70) (6.20) (16.10)

Wed 23.06 14.85 16.10 17.79 14.12 9.51 10.75 7.19 8.81 7.97 8.54 11.95 9.90 160.59
(3.45) (5.28) (4.84) (7.70) (6.52) (5.27) (5.90) (3.70) (5.19) (4.20) (4.59) (6.38) (4.81) (15.92)

Thu 26.54 13.61 14.82 12.92 9.87 6.59 9.24 7.88 8.33 10.19 8.22 10.87 7.91 146.46
(3.58) (4.30) (5.69) (5.54) (4.74) (3.43) (5.72) (4.66) (4.77) (5.63) (4.45) (5.51) (4.00) (14.26)

Fri 32.74 8.07 13.80 13.79 12.80 11.17 5.39 5.08 6.31 5.36 6.85 4.71 5.22 131.49
(4.43) (2.61) (5.33) (6.06) (5.92) (7.42) (3.79) (3.26) (4.23) (3.66) (4.26) (2.51) (2.77) (13.25)

60

Electronic copy available at: https://ssrn.com/abstract=4062260

Table 14 Monthly Momentum Across Asset Classes
This table reports annualized returns, alphas, and Sharpe ratios for different monthly momentum strategies, including the
risk-based return momentum (RISK ) strategy and the traditional return momentum strategies from Jegadeesh and Titman
(1993). JT 6m (12m) is the traditional momentum strategy that longs stocks with the highest cumulative returns and shorts the
stocks with the lowest cumulative returns over the past 2 to 6 (12) months. Panel A reports the results for the stock market
from January 1970 to December 2020. The stock RISK signal is constructed based on the 60 stock anomalies via LASSO. The
stock alphas are calculated using the FF5 model from Fama and French (2015). Panel B reports the results for the corporate
bond market from January 1974 to December 2020. The bond RISK signal is constructed based on the 25 bond anomalies. The
bond alphas are calculated based on the five factors from Bai, Bali, and Wen (2019). All returns and alphas are reported in
percentage, with Newey-West robust t-statistics in parentheses.

Panel A: Stock market Panel B: Corporate Bond market

Return Alpha SR Return Alpha SR

Panel A1: Full sample: Jan. 1970 - Dec. 2020 Panel B1: Full sample: Jan. 1974 - Dec. 2020

JT 6m 11.33 12.14 0.47 JT 6m -1.92 -1.94 -0.25

(3.47) (2.98) (-1.42) (-1.33)
JT 12m 14.57 15.65 0.55 JT 12m -2.84 -2.65 -0.37
(3.96) (3.49) (-1.97) (-1.90)
RISK 19.44 19.68 0.83 RISK 6.01 6.30 0.64
(6.44) (5.78) (3.82) (4.15)

Panel A2: Sub-sample: Jan. 1970 - Dec. 1995 Panel B2: Sub-sample: Jan. 1974 - Dec. 1999

JT 6m 11.63 11.81 0.65 JT 6m -1.60 -0.67 -0.28

(3.41) (2.84) (-1.31) (-0.62)
JT 12m 18.99 19.27 0.94 JT 12m -1.09 -0.04 -0.21
(4.80) (3.98) (-0.94) (-0.04)
RISK 18.25 18.34 1.01 RISK 5.35 5.72 0.67
(5.31) (4.47) (3.34) (3.96)

Panel A3: Sub-sample: Jan. 1996 - Dec. 2020 Panel B3: Sub-sample: Jan. 2000 - Dec. 2020

JT 6m 11.01 13.86 0.38 JT 6m -2.29 -2.64 -0.25

(1.96) (2.48) (-0.89) (-0.95)
JT 12m 10.00 12.04 0.31 JT 12m -4.93 -5.33 -0.50
(1.61) (1.96) (-1.74) (-1.93)
RISK 20.67 20.97 0.74 RISK 6.80 6.85 0.63
(4.39) (4.88) (3.37) (3.21)

61

Electronic copy available at: https://ssrn.com/abstract=4062260

References
Abreu, Dilip and Markus K. Brunnermeier, 2002, Synchronization risk and delayed arbitrage,

Journal of Financial Economics 66, 341–360.

Andrei, Daniel and Julien Cujean, 2017, Information percolation, momentum and reversal, Journal

of Financial Economics 123, 617–645.

Ang, Andrew, Robert J. Hodrick, Yuhang Xing, and Xiaoyan Zhang, 2006, The cross-section of

volatility and expected returns, The Journal of Finance 61, 259–299.

Arena, Matteo P., K. Stephen Haggard, and Xuemin (Sterling) Yan, 2008, Price momentum and

idiosyncratic volatility, Financial Review 43, 159–190.

Asness, Clifford S., Tobias J. Moskowitz, and Lasse Heje Pedersen, 2013, Value and momentum

everywhere, The Journal of Finance 68, 929–985.

Bai, Jennie, Turan G. Bali, and Quan Wen, 2019, Common risk factors in the cross-section of

corporate bond returns, Journal of Financial Economics 131, 619–642.

Bali, Turan G, Nusret Cakici, Xuemin Yan, and Zhe Zhang, 2005, Does idiosyncratic risk really

matter? The Journal of Finance 60, 905–929.

Bali, Turan G, Amit Goyal, Dashan Huang, Fuwei Jiang, and Quan Wen, 2022, Predicting corporate

bond returns: Merton meets machine learning, working paper Georgetown University, University

of Lausanne, Swiss Finance Institute, Singapore Management University, and Central University

of Finance and Economics.

Baltussen, Guido, Zhi Da, Sten Lammers, and Martin Martens, 2021, Hedging demand and market

intraday momentum, Journal of Financial Economics 142, 377–403.

Bansal, Ravi and Amir Yaron, 2004, Risks for the long run: A potential resolution of asset pricing

puzzles, The Journal of Finance 59, 1481–1509.

Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor sentiment, Journal

of Financial Economics 49, 307–343.

62

Electronic copy available at: https://ssrn.com/abstract=4062260

Beckmeyer, Heiner, Ilias Filippou, and Guofu Zhou, 2023, A new option momentum: Compensation

for risk, Working paper University of Munster and Washington University in St. Louis.

Berk, Jonathan B., Richard C. Green, and Vasant Naik, 1999, Optimal investment, growth options,

and security returns, The Journal of Finance 54, 1553–1607.

Bogousslavsky, Vincent, 2016, Infrequent rebalancing, return autocorrelation, and seasonality, The

Journal of Finance 71, 2967–3006.

Bogousslavsky, Vincent, 2021, The cross-section of intraday and overnight returns, Journal of

Financial Economics 141, 172–194.

Campbell, John Y., Martin Lettau, Burton G. Malkiel, and Yexiao Xu, 2001, Have individual stocks

become more volatile? An empirical exploration of idiosyncratic risk, The Journal of Finance 56,

1–43.

Chan, Louis K. C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, The

Journal of Finance 51, 1681–1713.

Chen, Andrew Y. and Tom Zimmermann, 2022, Open source cross sectional asset pricing, Critical

Finance Review 27, 207–264.

Chordia, Tarun and Lakshmanan Shivakumar, 2002, Momentum, business cycle, and time-varying

expected returns, The Journal of Finance 57, 985–1019.

Conrad, Jennifer and Gautam Kaul, 1998, An anatomy of trading strategies, Review of Financial

Studies 11, 489–519.

Cushing, David and Ananth Madhavan, 2000, Stock returns and trading at the close, Journal of

Financial Markets 3, 45–67.

Da, Zhi, Umit G. Gurun, and Mitch Warachka, 2014, Frog in the pan: Continuous information and

momentum, Review of Financial Studies 27, 2171–2218.

Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam, 1998, Investor psychology and

security market under-and overreactions, The Journal of Finance 53, 1839–1885.

63

Electronic copy available at: https://ssrn.com/abstract=4062260

Daniel, Kent D. and Tobias J. Moskowitz, 2016, Momentum crashes, Journal of Financial Economics

122, 221–247.

Dong, Xi, Yan Li, David E. Rapach, and Guofu Zhou, 2022, Anomalies and the expected market

return, The Journal of Finance 77, 639–681.

Ehsani, Sina and Juhani T. Linnainmaa, 2022, Factor momentum and the momentum factor, The

Journal of Finance 77, 1877–1919.

Engelberg, Joseph, R. David McLean, and Jeffrey Pontiff, 2018, Anomalies and news, The Journal

of Finance 73, 1971–2001.

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, 2015, A five-factor asset pricing model, Journal of

Financial Economics 116, 1–22.

Fama, Eugene F. and Kenneth R. French, 2020, Comparing cross-section and time-series factor

models, Review of Financial Studies 33, 1891–1926.

Filippou, Ilias, Sophia Zhengzi Li, Peixuan Yuan, and Guofu Zhou, 2023, Risk momentum in

currencies, Working paper Washington University in St. Louis, Rutgers University, and Renmin

University of China.

Foucault, Thierry, Ohad Kadan, and Eugene Kandel, 2005, Limit order book as a market for

liquidity, Review of Financial Studies 18, 1171–1217.

Freyberger, Joachim, Andreas Neuhierl, and Michael Weber, 2020, Dissecting characteristics

nonparametrically, Review of Financial Studies 33, 2326–2377.

Gao, Lei, Yufeng Han, Sophia Zhengzi Li, and Guofu Zhou, 2018, Market intraday momentum,

Journal of Financial Economics 129, 394–414.

64

Electronic copy available at: https://ssrn.com/abstract=4062260

Garcia, René, Daniel Mantilla-Garcı́a, and Lionel Martellini, 2014, A model-free measure of aggregate

idiosyncratic volatility and the prediction of market returns, Journal of Financial and Quantitative

Analysis 49, 1133–1165.

Gomes, Joao, Leonid Kogan, and Lu Zhang, 2003, Equilibrium cross section of returns, Journal of

Political Economy 3, 693–732.

Green, Jeremiah, John R.M. Hand, and X. Frank Zhang, 2017, The characteristics that provide

independent information about average US monthly stock returns, Review of Financial Studies

30, 4389–4436.

Griffin, John. M., Xiuqing Ji, and Spencer Martin, 2003, Momentum investing and business cycle

risk: Evidence from pole to pole, The Journal of Finance 58, 1515–1547.

Gu, Shihao, Bryan Kelly, and Dacheng Xiu, 2020, Empirical asset pricing via machine learning,

Review of Financial Studies 33, 2223–2273.

Heston, Steven L., Robert A. Korajczyk, and Ronnie Sadka, 2010, Intraday patterns in the

cross-section of stock returns, The Journal of Finance 65, 1369–1407.

Hong, Harrison and Jeremy C Stein, 1999, A unified theory of underreaction, momentum trading,

and overreaction in asset markets, The Journal of Finance 54, 2143–2184.

Hou, Kewei, Chen Xue, and Lu Zhang, 2015, Digesting anomalies: An investment approach, Review

of Financial Studies 28, 650–705.

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.

Jiang, Hao, Sophia Zhengzi Li, and Hao Wang, 2021, Pervasive underreaction: Evidence from

high-frequency data, Journal of Financial Economics 141, 573–599.

Jiang, Hao, Sophia Zhengzi Li, and Peixuan Yuan, 2022, Granular information and sectoral

movements, Working paper Michigan State University, Rutgers University, and Renmin University

of China.

65

Electronic copy available at: https://ssrn.com/abstract=4062260

Johnson, Timothy C., 2002, Rational momentum effects, The Journal of Finance 57, 585–608.

Jostova, Gergana, Stanislava Nikolova, Alexander Philipov, and Christof W. Stahel, 2013, Momentum

in corporate bond returns, Review of Financial Studies 26, 1649–1693.

Kelley, Eric K. and Paul C. Tetlock, 2017, Retail short selling and stock prices, Review of Financial

Studies 30, 801–834.

Kelly, Bryan T., Seth Pruitt, and Yinan Su, 2019, Characteristics are covariances: A unified model

of risk and return, Journal of Financial Economics 134, 501–524.

Kelly, Bryan T., Tobias J. Moskowitz, and Seth Pruitt, 2021, Understanding momentum and reversal,

Journal of Financial Economics 140, 726–743.

Kelly, Bryan T, Diogo Palhares, and Seth Pruitt, 2022, Modeling corporate bond returns, The

Journal of Finance, forthcoming.

Kozak, Serhiy, Stefan Nagel, and Shrihari Santosh, 2018, Interpreting factor models, The Journal

of Finance 73, 1183–1223.

Kozak, Serhiy, Stefan Nagel, and Shrihari Santosh, 2020, Shrinking the cross-section, Journal of

Financial Economics 135, 271–292.

Lewellen, Jonathan, 2002, Momentum and autocorrelation in stock returns, Review of Financial

Studies 15, 533–564.

Lewellen, Jonathan, 2015, The cross section of expected stock returns, Critical Finance Review 4,

1–44.

Li, Erica X. N., Dmitry Livdan, and Lu Zhang, 2009, Anomalies, Review of Financial Studies 22,

4301–4334.

Li, Jun, 2018, Explaining momentum and value simultaneously, Management science 64, 4239–4260.

Li, Sophia Zhengzi, Xiaoxia Ye, Peixuan Yuan, and Guofu Zhou, 2023, Risk momentum in

commodities, Working paper Rutgers University, University of Exeter, Renmin University of

China, and Washington University in St. Louis.

66

Electronic copy available at: https://ssrn.com/abstract=4062260

Lintner, John, 1965, The valuation of risky assets and the selection of risky investments in stock

portfolios and capital budgets, Review of Economics and Statistics 47, 13–37.

Liu, Laura Xiaolei and Lu Zhang, 2008, Momentum profits, factor pricing, and macroeconomic risk,

Review of Financial Studies 21, 2417–2448.

Liu, Laura Xiaolei and Lu Zhang, 2014, A neoclassical interpretation of momentum, Journal of

Monetary Economics 67, 109–128.

Lo, Andrew W. and Archie Craig MacKinlay, 1990, When are contrarian profits due to stock market

overreaction? Review of Financial Studies 3, 175–205.

Mortal, Sandra C. and Michael J. Schill, 2018, The role of firm investment in momentum and

reversal, Journal of Empirical Finance 48, 255–278.

Newey, Whitney and Kenneth West, 1987, A simple, positive semi-definite, heteroskedasticity and

autocorrelation consistent covariance matrix, Econometrica 55, 703–08.

Rapach, David and Guofu Zhou, 2022, Asset pricing: Time-series predictability, Oxford Research

Encyclopedia of Economics and Finance pp. 1–34.

Roll, Richard, 1984, A simple implicit measure of the effective bid-ask spread in an efficient market,

The Journal of Finance 39, 1127–1139.

Sagi, Jacob S. and Mark S. Seasholes, 2007, Firm-specific attributes and the cross-section of

momentum, Journal of Financial Economics 84, 389–434.

Sharpe, William F., 1964, Capital asset prices: A theory of market equilibrium under conditions of

risk, The Journal of Finance 19, 425–42.

Stambaugh, Robert F. and Yu Yuan, 2016, Mispricing factors, Review of Financial Studies 30,

1270–1315.

Stambaugh, Robert F., Jianfeng Yu, and Yu Yuan, 2012, The short of it: Investor sentiment and

anomalies, Journal of Financial Economics 104, 288–302.

67

Electronic copy available at: https://ssrn.com/abstract=4062260

Vayanos, Dimitri and Paul Woolley, 2013, An institutional theory of momentum and reversal,

Review of Financial Studies 26, 1087–1145.

68

Electronic copy available at: https://ssrn.com/abstract=4062260

 Internet Appendix

Fig. A.1 Intraday Patterns based on Lagged Return, Risk, and Residual
This figure shows the performance of the long-short portfolios based on a given intraday (30-minute or
overnight) signal in lag K, where all portfolios are held for one intraday period. The x-axis denotes the lag
of the signals. The RET signal is the total return in lag K; the RISK signal is the risk component of the
return and constructed based on the 15 RP anomalies; the RES signal is the difference between RET and
RISK. The sample includes all stocks in the Russell 1000 index over the period from January 1993 through
December 2020.

Electronic copy available at: https://ssrn.com/abstract=4062260

 (A) Daily Risk-based Momentum

(B) Weekly Risk-based Momentum (C) Monthly Risk-based Momentum

Fig. A.2 Performance of Low-Frequency Risk-based Momentum for Stocks

The figures plot the cumulative portfolio value from investing in different risk-based momentum strategies
at lower frequencies, with portfolios formed by the past-day (-week, -month) RISK estimated from 15 RP
anomalies, 15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO, respectively. The portfolios
are held over the subsequent day (week, month) with an initial investment of $1. We further take logarithm
of the portfolio value with base ten so the trajectory starts from 0. The sample period is from January 1970
to December 2020.

Electronic copy available at: https://ssrn.com/abstract=4062260

Fig. A.3 Performance of Monthly Risk-based Momentum for Corporate Bonds
This figure plots the cumulative portfolio value from investing in different momentum strategies for corporate
bonds, with portfolios formed by the last-month bond RISK estimated from 25 bond anomalies, 60 stock
anomalies, and 85 stock + bond anomalies, respectively. The portfolios are held over the subsequent month
with an initial investment of $1. We further take logarithm of the portfolio value with base ten so the
trajectory starts from 0. The corporate data are from the TRACE Enhanced database over the period from
July 2002 to December 2020. The 25 corporate bond characteristics are drawn from Kelly, Palhares, and
Pruitt (2022) and Bali, Goyal, Huang, Jiang, and Wen (2022), including “Bond age”, “Amihud illiquidity”,
“Bond market beta”, “Credit risk beta”, “Coupon”, “Default beta”, “Downside risk beta”, “Duration”,
“Illiquidity beta”, “Illiquidity”, “Kurtosis”, “Long-term reversal”, “Momentum”, “Six-month momentum”,
“Credit rating”, “Reversal”, “Issuance size”, “Skewness”, “Bid-ask spread”, “Downside risk”, “Term beta”,
“Time-to-maturity”, “Macroeconomic uncertainty beta”, “Volatility beta”, and “Volatility”. The 60 stock
characteristics are the same as in the paper.

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.1 List of 60 Anomalies
This table lists the 60 anomalies drawn from Green, Hand, and Zhang (2017), Freyberger, Neuhierl, and Weber (2020), Gu, Kelly,
and Xiu (2020), and Kozak, Nagel, and Santosh (2020), covering numerous categories, such as value versus growth, profitability,
investment, issuance activity, momentum, and trading frictions.

No. Anomaly Acronym No. Anomaly Acronym

1 Abnormal earnings announcement return abr 31 Employee growth rate hire

2 Accruals acc 32 Industry momentum indmom
3 Accrual volatility acvol 33 Investment-to-assets inta
4 Abnormal earnings announcement volume aeavol 34 Investment invest
5 # years since first Compustat coverage age 35 Leverage lev
6 Asset growth agr 36 Growth in long-term debt lgr
7 Bid-ask spread baspread 37 Liquidity liq
8 Beta beta 38 Maximum daily return maxret
9 Betting against beta bab 39 Long-term reversals lrv
10 Book-to-market bm 40 Momentum mom
11 Cash holdings cash 41 Short-term reversals rev
12 Cash flow to debt cashdebt 42 Residual variance rvr
13 Cash productivity cashpr 43 Size size
14 Cash-flow to price cfp 44 Net operating assets noa
15 Change in shares outstanding chcsho 45 Net stock issues nis
16 Change in inventory chinv 46 Profitability prof
17 Change in 6-month momentum chmom 47 O-score oscore
18 Industry-adjusted change in profit margin chpmia 48 Price delay pricedelay
19 Corporate investment cinvest 49 Quality minus junk qmj
20 Composite equity issues cei 50 Return on assets roa
21 Current ratio currat 51 Return on equity roe
22 Dividend initiation divi 52 Return on invested capital roic
23 Dollar trading volume dolvol 53 Revenue surprise rsup
24 Dividend to price dy 54 Sales growth sgr
25 Expected growth eg 55 Sales to price sp
26 Growth in common shareholder equity egr 56 Volatility of liquidity voliq
27 Earnings to price ep 57 Debt capacity/rm tangibility tang
28 Distress dis 58 Tax income to book income tb
29 Gross profitability gpf 59 Share turnover turn
30 Industry sales concentration herf 60 Zero trading days zerotrade

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.2 Correlations of Standardized Anomalies
Panel A reports the time-series average of the cross-sectional correlations of 15 standardized RP anomalies. Panel B reports
the same for 15 standardized EL anomalies. RP anomalies are 15 representative anomalies including 11 mispricing anomalies
of Stambaugh, Yu, and Yuan (2012) and Beta, Size, Book-to-market ratio, and Reversal. EL anomalies are the 15 anomalies
investigated by Ehsani and Linnainmaa (2022), including Accruals, Betting against beta, Book-to-market, Cash-flow to price,
Earnings to price, Profitability, Residual variance, Liquidity, Investment, Long-term reversals, Momentum, Short-term reversals,
Size, Quality minus junk, and Net share issues. Each anomaly is standardized according to Equation (1). The sample includes
all stocks in the Russell 1000 index over the period from January 1993 to December 2020.

Panel A: 15 RP Anomalies
acc agr beta bm cei dis gpf inta mom rev size noa oscore roa nsi
acc 1.00 0.10 0.03 -0.01 0.03 0.02 0.06 0.13 -0.03 0.00 -0.03 0.17 -0.02 0.08 0.04
agr 0.10 1.00 0.14 -0.05 0.11 0.14 -0.03 0.43 0.04 0.00 -0.04 0.50 0.03 0.00 0.21
beta 0.03 0.14 1.00 -0.02 -0.11 0.37 -0.03 0.08 0.07 0.04 -0.15 0.00 0.10 -0.09 0.10
bm -0.01 -0.05 -0.02 1.00 0.05 -0.02 -0.45 -0.01 -0.05 -0.01 -0.08 -0.02 0.00 -0.10 -0.09
cei 0.03 0.11 -0.11 0.05 1.00 -0.12 -0.07 0.09 -0.30 -0.03 0.01 0.10 -0.01 -0.04 0.09
dis 0.02 0.14 0.37 -0.02 -0.12 1.00 -0.05 0.10 0.00 0.02 -0.22 0.04 0.21 -0.15 0.10
gpf 0.06 -0.03 -0.03 -0.45 -0.07 -0.05 1.00 -0.03 0.05 0.01 0.00 0.06 -0.27 0.44 0.03
inta 0.13 0.43 0.08 -0.01 0.09 0.10 -0.03 1.00 0.01 0.00 -0.05 0.40 0.04 -0.01 0.15
mom -0.03 0.04 0.07 -0.05 -0.30 0.00 0.05 0.01 1.00 0.03 -0.01 -0.03 0.08 0.04 0.06
rev 0.00 0.00 0.04 -0.01 -0.03 0.02 0.01 0.00 0.03 1.00 -0.01 -0.01 0.04 -0.01 0.00
size -0.03 -0.04 -0.15 -0.08 0.01 -0.22 0.00 -0.05 -0.01 -0.01 1.00 -0.12 -0.28 0.07 0.02
noa 0.17 0.50 0.00 -0.02 0.10 0.04 0.06 0.40 -0.03 -0.01 -0.12 1.00 0.19 0.02 0.10
oscore -0.02 0.03 0.10 0.00 -0.01 0.21 -0.27 0.04 0.08 0.04 -0.28 0.19 1.00 -0.44 -0.04
roa 0.08 0.00 -0.09 -0.10 -0.04 -0.15 0.44 -0.01 0.04 -0.01 0.07 0.02 -0.44 1.00 0.06
nsi 0.04 0.21 0.10 -0.09 0.09 0.10 0.03 0.15 0.06 0.00 0.02 0.10 -0.04 0.06 1.00

Panel B: 15 EL Anomalies
acc agr beta bm cei dis gpf invest mom rev size noa oscore roa nsi
acc 1.00 0.03 -0.03 -0.20 0.06 0.07 0.01 0.02 0.05 0.07 -0.03 0.00 -0.04 0.19 0.04
bab 0.03 1.00 -0.13 -0.15 -0.17 0.09 0.53 -0.10 -0.01 0.06 -0.03 0.01 0.00 0.22 0.15
bm -0.03 -0.13 1.00 0.45 0.30 -0.45 -0.19 0.16 -0.01 -0.27 -0.09 -0.01 -0.10 -0.25 -0.06
cfp -0.20 -0.15 0.45 1.00 0.44 -0.20 -0.18 0.04 0.00 -0.21 -0.02 0.00 0.01 -0.05 -0.17
ep 0.06 -0.17 0.30 0.44 1.00 0.03 -0.32 -0.01 0.04 -0.01 -0.04 -0.01 0.06 0.15 -0.24
prof 0.07 0.09 -0.45 -0.20 0.03 1.00 0.06 -0.08 -0.01 0.11 0.05 0.02 0.03 0.34 -0.13
rvar 0.01 0.53 -0.19 -0.18 -0.32 0.06 1.00 0.05 -0.03 0.06 0.08 0.04 -0.20 0.01 0.24
liq 0.02 -0.10 0.16 0.04 -0.01 -0.08 0.05 1.00 0.00 -0.03 -0.04 0.02 -0.78 -0.18 0.06
invest 0.05 -0.01 -0.01 0.00 0.04 -0.01 -0.03 0.00 1.00 0.08 -0.04 0.00 0.01 0.03 0.01
lrev 0.07 0.06 -0.27 -0.21 -0.01 0.11 0.06 -0.03 0.08 1.00 0.00 0.01 0.02 0.14 0.22
mom -0.03 -0.03 -0.09 -0.02 -0.04 0.05 0.08 -0.04 -0.04 0.00 1.00 0.01 0.02 0.00 0.05
srev 0.00 0.01 -0.01 0.00 -0.01 0.02 0.04 0.02 0.00 0.01 0.01 1.00 0.00 -0.01 0.00
size -0.04 0.00 -0.10 0.01 0.06 0.03 -0.20 -0.78 0.01 0.02 0.02 0.00 1.00 0.13 -0.08
qmj 0.19 0.22 -0.25 -0.05 0.15 0.34 0.01 -0.18 0.03 0.14 0.00 -0.01 0.13 1.00 -0.07
nsi 0.04 0.15 -0.06 -0.17 -0.24 -0.13 0.24 0.06 0.01 0.22 0.05 0.00 -0.08 -0.07 1.00

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.3 Risk-based Momentum: Russell 3000 Stock Universe
This table reports the performance of 13 long-short portfolios over a given intraday period based on the RISK signal in the
previous intraday period. Each of these portfolios enters position at time “Start” and exits position at time “End” once per day.
The last column “All” reports the performance of the long-short portfolio that is held during all 13 intraday periods but is
rebalanced every intraday period based on the RISK signal in the previous intraday period (i.e., investing in all 13 signals).
We report the annualized return of the long-short portfolio in percentage with Newey and West (1987) robust t-statistics in
parentheses. Panels A, B, C, and D report the results based on RISK estimated from 15 RP anomalies, 15 EL anomalies, 60
anomalies via PCA, and 60 anomalies via LASSO, respectively. The sample includes all stocks in the Russell 3000 index over
the period from January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

9.42 8.63 11.30 9.97 8.79 5.85 4.84 4.36 4.91 4.04 5.41 6.82 6.29 90.48
(3.88) (6.61) (10.80) (11.02) (11.38) (8.06) (8.23) (6.63) (7.28) (5.34) (7.12) (8.22) (6.87) (23.89)

Panel B: 15 EL Anomalies

10.54 8.23 11.82 10.14 9.81 5.89 5.12 4.76 4.96 4.21 5.97 6.95 6.87 94.95
(4.34) (6.24) (11.16) (11.19) (12.36) (8.00) (8.55) (6.96) (7.27) (5.60) (7.74) (8.45) (7.38) (24.89)

Panel C: 60 Anomalies via PCA

14.88 9.62 13.47 12.50 10.83 7.90 6.93 6.41 6.51 6.20 7.91 9.21 8.80 121.35
(5.18) (6.84) (10.83) (11.55) (11.93) (9.01) (9.39) (7.78) (7.74) (6.59) (8.21) (9.39) (8.01) (26.89)

Panel D: 60 Anomalies via LASSO

13.42 11.63 15.02 14.10 12.15 8.61 7.92 7.11 7.00 6.51 8.74 9.83 9.49 131.52
(4.65) (7.79) (11.68) (12.55) (12.68) (9.50) (10.38) (8.29) (8.16) (6.83) (8.80) (9.62) (8.46) (28.51)

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.4 Decile Portfolio Performance: 60 Anomalies
This table reports the annualized return in percentage of decile portfolios over a given intraday period based on the RISK signal
in the previous intraday period. Each of these portfolios enters position at time “Start” and exits position at time “End” once
per day. The last column “All” reports the performance of the decile portfolios that are held during all 13 intraday periods but
are rebalanced every intraday period based on the RISK signal in the previous intraday period (i.e., investing in all 13 signals).
The row labeled “High-Low” reports the annualized return of the long-short portfolios in percentage with Newey and West
(1987) robust t-statistics in parentheses. Panels A and B report the results based on RISK estimated from 60 anomalies via
PCA and 60 anomalies via LASSO, respectively. The sample includes all stocks in the Russell 1000 index over the period from
January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 60 Anomalies via PCA

1 -0.47 -7.35 -8.33 -6.97 -5.24 -4.85 -2.02 -2.11 -3.66 -3.89 -2.64 -3.82 -2.33 -53.75
2 0.54 -4.77 -5.43 -4.98 -3.36 -2.98 -1.38 -1.62 -2.60 -2.60 -1.21 -2.20 -1.09 -33.69
3 2.13 -3.40 -3.76 -3.45 -2.20 -1.76 -0.27 -0.70 -1.68 -1.72 0.16 -0.76 -0.12 -17.53
4 2.87 -2.62 -3.08 -1.86 -1.03 -1.02 0.12 -0.42 -1.26 -1.10 0.86 0.15 0.39 -8.01
5 3.90 -1.60 -1.48 -0.67 -0.42 -0.63 1.00 0.12 -0.63 -0.99 1.71 0.87 1.32 2.51
6 5.10 -0.82 -0.43 0.02 0.38 0.15 1.33 1.00 -0.29 -0.29 2.34 2.11 1.71 12.31
7 5.98 0.04 0.58 1.17 1.36 0.83 1.65 1.37 0.14 0.14 2.94 2.19 2.37 20.76
8 7.10 0.55 1.96 1.95 2.38 1.39 2.88 2.02 0.80 1.20 3.76 3.45 3.07 32.52
9 10.08 1.50 3.47 3.38 3.74 1.97 3.75 3.07 1.18 1.54 4.28 4.19 4.05 46.18
10 12.28 2.97 5.28 5.27 5.76 3.08 4.99 4.26 2.45 2.03 5.41 5.27 6.36 65.40
High-Low 12.75 10.31 13.61 12.24 11.00 7.93 7.00 6.37 6.11 5.92 8.05 9.09 8.70 119.14
(4.08) (6.92) (10.28) (10.66) (11.32) (8.59) (8.86) (7.35) (6.74) (5.89) (7.86) (8.73) (7.46) (24.69)

Panel B: 60 Anomalies via LASSO

1 -1.00 -8.71 -8.76 -7.40 -5.03 -4.67 -2.04 -2.27 -4.25 -3.91 -2.74 -3.68 -2.12 -56.79
2 0.38 -5.62 -5.60 -5.13 -3.45 -2.94 -1.08 -1.83 -2.49 -2.62 -1.06 -1.42 -1.01 -33.89
3 1.33 -3.62 -4.00 -3.25 -1.96 -1.89 -0.26 -1.08 -1.81 -2.06 0.32 -0.33 0.26 -18.36
4 3.02 -2.58 -2.93 -2.04 -1.24 -0.86 0.63 -0.42 -1.24 -1.09 1.01 0.29 0.42 -7.02
5 3.16 -1.60 -1.76 -1.04 -0.31 -0.54 1.25 0.30 -0.89 -0.65 1.41 1.35 0.98 1.67
6 4.81 -0.79 -0.58 0.34 0.54 0.03 1.49 0.77 -0.46 -0.22 2.53 1.96 1.46 11.89
7 6.14 -0.25 0.60 1.10 1.26 0.67 2.24 1.18 -0.02 0.34 2.78 2.76 2.31 21.11
8 7.51 1.02 2.30 1.77 2.54 1.10 2.71 2.03 0.46 0.62 3.40 3.24 3.07 31.78
9 9.41 2.36 3.80 3.39 3.82 2.30 3.65 2.83 1.15 1.39 4.44 4.55 4.35 47.43
10 13.68 2.97 5.43 5.35 5.80 3.51 5.17 3.93 2.65 2.25 5.73 5.59 6.35 68.42
High-Low 14.68 11.68 14.19 12.76 10.83 8.18 7.20 6.20 6.90 6.16 8.47 9.27 8.48 125.21
(4.62) (7.55) (10.77) (11.23) (11.46) (8.84) (8.99) (7.28) (7.78) (6.14) (8.36) (8.94) (7.08) (25.67)

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.5 Predictability of Past Returns
This table shows the predictability of past returns for future total returns (Panel A) and RISK and RES components of
holding-period returns (Panel B). The risk (RISK ) and residual (RES ) components are constructed according to Equations (4)
and (5). We form 13 long-short portfolios over a given intraday period based on the returns in the previous intraday period. Each
of these portfolios enters position at time “Start” and exits position at time “End” once per day. The last column “All” reports
the performance of the long-short portfolio that is held during all 13 intraday periods but is rebalanced every intraday period
based on the returns in the previous intraday period (i.e., investing in all 13 signals). The row labeled “RISK ” (“RES ”) report
the value-weighted RISK (RES ) component for each long-short portfolio. In Panel A, the row labeled “Return” reports the
annualized return of the long-short portfolio in percentage with Newey and West (1987) robust t-statistics in parentheses. The
row labeled “Alpha” reports the annualized CAPM alpha. In Panel B, we report the annualized return of the two components
in percentage with t-statistics in parentheses. The sample includes all stocks in the Russell 1000 index over the period from
January 1993 to December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: Return Prediction

Return -50.77 -8.07 -1.89 -3.22 -1.86 -6.06 -6.93 -7.91 -8.44 -10.31 -7.74 -8.45 -13.38 -135.03
(-20.15) (-6.72) (-1.99) (-3.85) (-2.44) (-8.13) (-10.72) (-11.53) (-11.08) (-13.42) (-9.79) (-10.55) (-14.56) (-38.77)
Alpha -50.78 -8.37 -2.00 -3.30 -1.86 -6.09 -6.83 -7.92 -8.47 -10.29 -7.73 -8.38 -13.40 -134.97
(-20.14) (-6.98) (-2.09) (-3.95) (-2.47) (-8.18) (-10.60) (-11.65) (-10.92) (-13.23) (-9.69) (-10.55) (-14.79) (-38.70)

Panel B: Decomposition

RISK 1.11 4.70 4.49 3.73 3.00 2.04 1.87 1.95 1.82 1.76 2.23 2.58 2.20 32.16
(2.23) (8.56) (10.84) (11.02) (11.39) (8.29) (9.72) (8.40) (8.07) (6.32) (8.62) (9.38) (7.56) (25.99)
RES -51.88 -12.78 -6.37 -6.95 -4.85 -8.10 -8.80 -9.85 -10.26 -12.07 -9.97 -11.03 -15.58 -167.19
(-22.23) (-13.58) (-8.65) (-10.53) (-7.62) (-12.96) (-15.11) (-16.97) (-15.78) (-19.03) (-14.97) (-16.47) (-19.79) (-48.25)

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.6 Risk-based Momentum Conditional on Risk Concentration: Controlling for |RISK|
Level
This table reports the performance of risk-based momentum for Russell 1000 stocks in the top and bottom RiskCon quintiles
controlling for |RISK| level. RiskCon, defined in Equation (6), captures a stock’s risk concentration. We first sort stocks
into five portfolios by |RISK|, and then within each |RISK| quintile we sort the stocks into quintiles based on their risk
concentration (5 × 5 grouping). Next, we average across the five |RISK| portfolios to produce five RiskCon portfolios with
large cross-portfolio variation in risk concentration but little variation in |RISK|. Finally, we construct risk-based long-short
portfolios within the highest and lowest RiskCon portfolios, respectively. The rows labeled “High” (“Low”) denotes stocks in
the top (bottom) RiskCon quintile. Panels A, B, C, and D report the results based on RISK estimated from 15 RP anomalies,
15 EL anomalies, 60 anomalies via PCA, and 60 anomalies via LASSO, respectively. The sample period is from January 1993 to
December 2020.

Return-Holding Period

Start 16:00 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30
All
RiskCon End 10:00 10:30 11:00 11:30 12:00 12:30 13:00 13:30 14:00 14:30 15:00 15:30 16:00

Panel A: 15 RP Anomalies

High 21.30 14.42 13.32 12.88 11.09 8.63 6.35 6.70 7.19 6.80 7.99 9.85 8.62 134.94
(7.93) (10.64) (12.77) (14.13) (14.36) (12.21) (9.92) (10.09) (11.14) (9.81) (11.08) (12.86) (10.03) (34.49)

Low -7.47 6.05 4.32 3.98 5.46 -0.17 -0.07 -0.69 0.91 -0.08 1.73 2.17 0.61 16.97
(-2.14) (3.45) (2.93) (2.86) (4.57) (-0.14) (-0.07) (-0.67) (0.81) (-0.07) (1.47) (1.71) (0.42) (3.05)

Panel B: 15 EL Anomalies

High 21.57 14.33 12.97 12.34 10.78 7.55 6.77 7.24 6.61 6.30 8.06 9.47 7.99 131.89
(8.08) (9.96) (12.50) (13.50) (14.17) (10.54) (10.94) (10.59) (10.48) (8.98) (11.40) (12.84) (9.72) (33.61)

Low -13.14 5.20 3.33 4.93 5.50 0.24 -0.29 0.22 0.55 -2.06 2.35 1.34 2.91 11.29
(-3.74) (2.93) (2.19) (3.61) (4.61) (0.20) (-0.28) (0.21) (0.49) (-1.76) (1.90) (1.08) (2.11) (2.02)

Panel C: 60 Anomalies via PCA

High 22.45 14.17 14.90 14.62 11.53 9.30 9.11 7.44 7.59 7.71 10.06 10.66 9.77 149.61
(7.42) (9.94) (12.45) (14.32) (13.48) (11.74) (12.25) (9.90) (10.51) (9.68) (11.81) (12.15) (10.18) (33.95)

Low -8.85 6.55 4.15 4.82 7.29 3.74 1.40 2.54 1.37 2.30 4.72 4.45 5.24 40.36
(-2.37) (3.41) (2.42) (3.08) (5.33) (2.76) (1.17) (2.14) (1.05) (1.63) (3.35) (3.21) (3.33) (6.44)

Panel D: 60 Anomalies via LASSO

High 25.61 17.05 15.86 15.06 12.06 10.49 8.12 8.01 8.93 8.11 10.32 11.11 9.64 160.67
(8.21) (11.71) (14.11) (14.91) (14.21) (12.91) (11.38) (10.51) (11.96) (10.40) (12.02) (12.67) (9.87) (35.85)

Low -7.31 7.44 6.46 5.26 7.59 2.61 1.89 1.16 3.23 1.43 4.76 3.74 5.49 43.93
(-1.90) (3.88) (3.77) (3.42) (5.59) (2.01) (1.57) (0.95) (2.50) (1.05) (3.51) (2.69) (3.43) (6.95)

Electronic copy available at: https://ssrn.com/abstract=4062260

Table A.7 Monthly Risk-based Momentum
This table reports the performance of decile portfolios formed on the basis of the monthly RISK signal and held over the
subsequent month. Panel A reports the annualized portfolio returns in percentage for the unconditional risk-based momentum
strategies formed on all stocks in the Russell 1000 index. Panel B reports the results for the conditional risk-based momentum
strategies formed on Russell 1000 stocks in the top risk-concentration quintile. Risk concentration is defined in Equation (6). The
RISK signal is constructed using 15 RP anomalies, 15 EL anomalies, 60 anomalies via PCA, or 60 anomalies via LASSO. The
rows labeled “High-Low”, “FF5 Alpha” and “q-factor Alpha” respectively report the annualized return, the Fama and French
(2015) five-factor alpha, and the Hou, Xue, and Zhang (2015) q-factor alpha of the long-short spread portfolio in percentage,
with Newey-West robust t-statistics in parentheses. The sample period is from January 1970 to December 2020.

15 RP Anomalies 15 EL Anomalies 60 Anomalies via PCA 60 Anomalies via LASSO

Panel A: Unconditional Risk-based Momentum

1 (Low) 2.99 1.96 3.46 1.14

2 7.92 6.63 7.98 6.20
3 9.22 8.98 8.68 8.93
4 10.55 11.12 11.38 10.43
5 12.73 12.50 13.03 12.03
6 14.14 13.03 14.35 14.24
7 15.40 16.11 14.86 16.06
8 15.88 16.79 15.67 16.76
9 16.16 17.42 17.33 17.89
10 (High) 19.26 20.56 17.15 20.58
High-Low 16.27 18.60 13.69 19.44
(5.88) (6.30) (4.31) (6.44)
FF5 Alpha 17.48 19.16 14.08 19.68
(5.77) (5.87) (3.56) (5.78)
q-factor Alpha 18.83 20.73 14.79 20.59
(5.56) (5.69) (3.39) (5.47)

Panel B: Conditional Risk-based Momentum

1 (Low) 0.61 -1.50 1.29 -1.47

2 1.95 3.34 6.24 2.54
3 6.70 6.43 7.06 6.64
4 9.35 11.41 9.69 11.75
5 11.91 12.79 12.76 11.97
6 15.39 15.08 14.15 15.67
7 17.44 17.60 17.39 16.23
8 19.01 17.41 18.48 20.38
9 20.86 22.46 21.79 21.14
10 (High) 25.64 24.69 22.48 25.37
High-Low 25.03 26.19 21.19 26.84
(7.01) (7.22) (5.66) (7.44)
FF5 Alpha 25.02 27.00 20.54 26.73
(6.62) (6.86) (4.51) (6.46)
q-factor Alpha 26.34 27.72 21.08 27.61
(6.47) (6.57) (4.42) (6.17)

10

Electronic copy available at: https://ssrn.com/abstract=4062260