Firm-specific versus systematic momentum*

Frank Graef †, #, Daniel Hoechle ‡, Markus Schmid §

11 December 2024

Abstract
We decompose stock returns into a systematic and a firm-specific component and
show that the dynamics of the firm-specific return component drives the well-
known stock momentum anomaly. Our results are robust to the use of a variety
of prominent factor models for return decomposition. Furthermore, we find that
momentum profits are largely unaffected when the investment universe is
restricted to stocks with inconspicuous factor loadings. Our empirical findings
call into question the transmission mechanism from factor momentum to stock
momentum proposed in recent research.

JEL classification: G14

Keywords: Factor momentum, Firm-specific momentum, Factor timing

*
We are grateful to Martin Brown, Tim Kroencke, Luca Liebi, Bharat Parajuli, Paul Whelan, Gulnara
Zaynutdinova and conference participants at the 2022 Financial Management Association (FMA)
European Conference, the 2022 Swiss Finance Institute (SFI) Research Days, the 2022 Paris Financial
Management Conference (PFMC), the 2022 European Financial Management Association (EFMA)
Annual Meeting and seminar participants at the University of St. Gallen and the University of Konstanz
for helpful comments.
#
Corresponding author.
†
University of Applied Sciences Northwestern Switzerland, Institute for Finance, CH-4002 Basel,
Switzerland, [email protected]
‡
University of Applied Sciences Northwestern Switzerland, Institute for Finance, CH-4002 Basel,
Switzerland, [email protected]
§
Swiss Institute of Banking and Finance, University of St. Gallen, CH-9000 St. Gallen, Switzerland,
Swiss Finance Institute (SFI), and European Corporate Governance Institute (ECGI),
[email protected]
1. Introduction

Stock momentum (e.g., Jegadeesh and Titman, 1993) appears to be at odds with the weakest
form of market efficiency, according to which future returns should not be predictable based on
historical price movements. Since prominent factor models fail to explain momentum profits,
behavioral explanations have been proposed. However, momentum has been shown to exist in
the cross-section of characteristics-sorted portfolios that contain a large number of stocks and,
as noted by Lewellen (2002), biased reactions to firm-specific news arguably cannot explain
momentum as a portfolio-level phenomenon.
Against this background, Ehsani and Linnainmaa (2022; henceforth "EL") propose an
alternative explanation for the existence of stock momentum: They find that momentum
originates at the level of equity risk factors and that this factor momentum transmits into the
cross-section of stock returns. In this paper, we empirically test the transmission mechanism
proposed by EL, whereby factor momentum translates into stock momentum through
autocovariance in factors, amplified by the cross-sectional variation in betas.
To this end, we first investigate the performance of stock-level strategies, using a holding
period of one month after a formation period over month 𝑡𝑡 −1 (“short-term”) or months 𝑡𝑡 −12
to 𝑡𝑡 −2 (“medium-term”). Our sample consists of monthly U.S. stock returns from CRSP over
the period of July 1963 to December 2019. We then sort portfolios by systematic and
idiosyncratic stock returns. To do so, we estimate Fama and French (2015) five-factor model
betas on a rolling basis and compute expected (i.e., systematic) returns. Firm-specific (i.e.,
idiosyncratic) returns then correspond to the difference between total returns and expected
returns. If EL’s conjecture about the transmission from factor momentum to stock momentum
is correct, systematic returns should be a better predictor for future performance than
idiosyncratic returns. We, however, find the opposite: Over a medium-term formation period,
systematic returns are not informative about future performance, while idiosyncratic returns
are. Alternative rolling window periods for beta estimation and variations in the factor model
used for return decomposition yield qualitatively similar results. Moreover, we replicate the
results of EL, who double-sort by firm size and idiosyncratic returns to construct “UMD-style”
residual momentum strategies. In extensions to this replication, we then find that systematic
returns also lack outperformance when using this alternative construction method.
Second, we split our sample into two subsamples. The first subsample includes all stock-
months with very large (> 80th percentile) or very small (< 20th percentile) loadings on at least
one of the five Fama and French (2015) factors. The second subsample consists of all remaining

1
observations. If the transmission from factor momentum to stock momentum works as EL
hypothesize, we would expect momentum profits to be higher, when there is large variation in
stocks’ betas. However, we find that momentum profits in the two subsamples are quite similar.
We contribute to the literature on factor momentum (Gupta and Kelly, 2019; Ehsani and
Linnainmaa, 2022; Arnott et al., 2023), by showing that stock momentum is driven by
momentum in firm-specific returns and that the dispersion in betas does not affect momentum
performance. Our finding that a larger fluctuation in the factor betas of a momentum strategy
does not necessarily imply better performance also contributes to a literature that examines the
time-varying risk exposures of momentum strategies (Grundy and Martin, 2001; Wang and Wu,
2011; Daniel and Moskowitz, 2016) and to a literature that analyzes idiosyncratic momentum
and short-term reversal (Gutierrez and Prinsky, 2007; Blitz et al., 2011; Da et al., 2011, 2014).
The remainder of the paper is organized as follows: Section 2 describes the transmission
mechanism proposed by EL, Section 3 describes our data, Section 4 presents our analysis and
results, and Section 5 concludes.

2. The transmission from factor momentum and idiosyncratic momentum to

stock momentum

Consider a stock momentum strategy that chooses investment weights in proportion to stocks’
past returns relative to the cross-sectional mean (Lo and MacKinlay, 1990). If stocks’ expected
returns are generated by some 𝐹𝐹-factor model, EL show that the following relationship holds
for the expected performance of this strategy, implemented in a cross-section of 𝑁𝑁 stocks:

𝐹𝐹 𝑁𝑁
𝑓𝑓 𝑓𝑓 1
𝐸𝐸[𝜋𝜋𝑡𝑡𝑚𝑚𝑚𝑚𝑚𝑚 ] = �[𝑐𝑐𝑐𝑐𝑐𝑐(𝑟𝑟−𝑡𝑡 , 𝑟𝑟𝑡𝑡 ) 𝜎𝜎𝛽𝛽2𝑓𝑓 ] + �[ 𝑐𝑐𝑐𝑐𝑐𝑐(𝜀𝜀𝑖𝑖,−𝑡𝑡 , 𝜀𝜀𝑖𝑖,𝑡𝑡 )]
𝑁𝑁
𝑓𝑓=1 𝑖𝑖=1
(1)
𝐹𝐹 𝐹𝐹
𝑓𝑓 𝑔𝑔
+ � �[𝑐𝑐𝑐𝑐𝑐𝑐(𝑟𝑟−𝑡𝑡 , 𝑟𝑟𝑡𝑡 ) 𝑐𝑐𝑐𝑐𝑐𝑐(𝛽𝛽 𝑓𝑓 , 𝛽𝛽 𝑔𝑔 )] + 𝜎𝜎𝜂𝜂2 ,
𝑓𝑓=1 𝑔𝑔≠𝑓𝑓

𝑓𝑓
where 𝑟𝑟𝑡𝑡 is the return of factor 𝑓𝑓 in month 𝑡𝑡, 𝛽𝛽 𝑓𝑓 are stocks’ loadings towards factor 𝑓𝑓, 𝜀𝜀𝑖𝑖,𝑡𝑡 are
stocks’ idiosyncratic returns, 𝜂𝜂 represents stocks’ unconditional expected returns and −𝑡𝑡 refers
to the formation period (e.g., months 𝑡𝑡 −12 to 𝑡𝑡 −2).
Factor momentum may translate into stock momentum via the first term in Equation (1),
𝑓𝑓 𝑓𝑓
i.e., via the multiplication of factor autocorrelations 𝑐𝑐𝑐𝑐𝑐𝑐(𝑟𝑟−𝑡𝑡 , 𝑟𝑟𝑡𝑡 ) with the cross-sectional
variance in betas 𝜎𝜎𝛽𝛽2𝑓𝑓 . This is the channel proposed by EL. To illustrate this transmission

mechanism, assume that expected returns are determined by stocks’ loadings against a serially

2
uncorrelated market factor and a positively autocorrelated size factor (SMB). Further assume
no autocorrelation in firm-specific returns and no persistent differences in unconditional mean
returns. If small-cap stocks then outperform large-cap stocks in the past, they will continue to
do so in the following months. Stock momentum may arise in this scenario, because stocks with
a large positive (negative) SMB loading tend to be sorted into the top (bottom) momentum
portfolio. The strength of this channel depends on the cross-sectional dispersion in betas. For
example, one would expect a stock momentum strategy to become less profitable if it is
restricted to the subsample of large-cap stocks with similar SMB betas. According to EL, it is
therefore the expected (i.e., systematic) component of stock returns that picks up autocorrelation
in factors and drives stock momentum. We empirically test this conjecture by running a horse
race between this first component and the second component in equation (1), the latter
representing idiosyncratic (or firm-specific) momentum. We thus directly test the relative
importance of systematic (or factor) versus idiosyncratic momentum.

3. Data and factor construction

We merge monthly U.S. stock returns from CRSP with annual accounting information from
Compustat. We collect data for all NYSE, AMEX, and Nasdaq securities listed as ordinary
common stock at any time during the sample period from July 1963 to December 2019.
Following Hou et al. (2020), we exclude financial firms and firms with negative book equity.
To avoid a look-ahead bias, we lag Compustat accounting information by six months, such that
year-end figures become available at the end of June of next year (Fama and French, 1992). We
use CSRP delisting returns. If delisting returns are missing and the delisting is performance-
related, we set them to -30% (Shumway, 1997; Beaver et al., 2007). We retrieve factor returns
of the Fama and French (1993, 2015) three- and five-factor models from Ken French’s
webpage. 1 Stambaugh and Yuan (2017) mispricing factors are obtained from Robert
0F

Stambaugh’s webpage. 2 Hou et al. (2021) augmented q-model factors are taken from the global-
1F

q data library. 3
2F

1
https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html
2
http://finance.wharton.upenn.edu/~stambaug/
3
http://global-q.org/index.html

3
4. Empirical results
4.1 Return-based strategies using individual stocks and factor portfolios

To implement the stock momentum strategy, we rank stocks by their returns over the last year,
skipping the month before the ranking (𝑟𝑟𝑡𝑡−12,𝑡𝑡−2 ). To test for short-term reversal, we rank stocks
by their returns in the last month (𝑟𝑟𝑡𝑡−1 ). We then sort stocks into quintile portfolios. In the value-
weighted specification, we use NYSE breakpoints and compute value-weighted average returns
for each quintile, using stocks’ lagged market equity. In the equal-weighted specification, we
use unconditional breakpoints. To compute strategy returns, we go long (short) the top (bottom)
quintile portfolio for the current month, with monthly rebalancing.
Table 1 reports the results. The value-weighted momentum (short-term reversal) strategy
achieves a statistically significant monthly return of 0.66% (-0.34%). After adjusting for
exposure to the Fama and French (2015) factors, momentum returns increase, whereas short-
term reversal returns become smaller and statistically insignificant. In the equal-weighted setup,
momentum (short-term reversal) performance amounts to 0.85% (-1.55%) per month, with a
statistically significant alpha. 4 3F

4.2 Portfolio sorts by systematic and idiosyncratic returns

We decompose stock returns into a systematic and an idiosyncratic component, where the
former represents the expected return, given stocks’ current factor loadings, and the latter
represents the firm-specific return that is unexplained by factor loadings. We estimate firms’
𝑓𝑓
loadings 𝛽𝛽̂𝑖𝑖,𝑡𝑡 against the 𝐹𝐹 = 5 factors of Fama and French (2015), using a five-year rolling
period from month 𝑡𝑡 −60 to 𝑡𝑡 −1. Each beta estimate is based on at least 36 past-return
observations. Systematic and idiosyncratic returns are computed as follows:

𝐹𝐹
𝑓𝑓 𝑓𝑓
Short-term systematic returns: 𝑟𝑟̂𝑖𝑖,𝑡𝑡−1 = � 𝛽𝛽̂𝑖𝑖,𝑡𝑡 𝑟𝑟𝑡𝑡−1 (2)
𝑓𝑓=1
𝑒𝑒
Short-term idiosyncratic returns: 𝜀𝜀̂𝑖𝑖,𝑡𝑡−1 = 𝑟𝑟𝑖𝑖,𝑡𝑡−1 − 𝑟𝑟̂𝑖𝑖,𝑡𝑡−1 (3)
−2

Medium-term systematic returns: 𝑟𝑟̂𝑖𝑖,𝑡𝑡−12,𝑡𝑡−2 = � 𝑟𝑟̂𝑖𝑖,𝑡𝑡−𝑗𝑗 (4)

𝑗𝑗=−12
−2
𝑒𝑒
Medium-term idiosyncratic returns: 𝜀𝜀̂𝑖𝑖,𝑡𝑡−12,𝑡𝑡−2 = � 𝑟𝑟𝑖𝑖,𝑡𝑡−𝑗𝑗 − 𝑟𝑟̂𝑖𝑖,𝑡𝑡−𝑗𝑗 , (5)
𝑗𝑗=−12

4
We report short-term reversal returns with a negative sign for analytical reasons. Of course, by switching the long
and short positions, we get an equivalent positive return.

4
 𝑒𝑒 𝑓𝑓
where 𝑟𝑟𝑖𝑖,𝑡𝑡 are monthly stock returns in excess of the risk-free rate and 𝑟𝑟𝑡𝑡 denotes monthly factor
premia. We rank stocks by one of the return measures defined in Equations (2) to (5) each
month, sort them into value-weighted quintile portfolios and invest long (short) in the top
(bottom) quintile portfolio for the subsequent month.
Table 2 reports the returns of the quintile portfolios and of the high-low quintile strategies.
Sorting by short-term idiosyncratic returns (𝜀𝜀̂𝑖𝑖,𝑡𝑡−1 ) delivers a statistically significant short-term
reversal return of -0.78% per month, which is more than twice as large compared to the
conventional value-weighted short-term reversal strategy analyzed in Table 1. In contrast,
sorting by short-term systematic returns (𝑟𝑟̂𝑖𝑖,𝑡𝑡−1) results in a statistically significant positive
return of 0.38%. Sorting by medium-term idiosyncratic returns (𝜀𝜀̂𝑖𝑖,𝑡𝑡−12,𝑡𝑡−2 ) yields a statistically
significant return of 0.57%, which is only slightly smaller than the 0.66% return of the
conventional value-weighted momentum strategy in Table 1. When we instead sort by medium-
term systematic returns (𝑟𝑟̂𝑖𝑖,𝑡𝑡−12,𝑡𝑡−2), the return of 0.20% is statistically insignificant. Results are
qualitatively identical for the five-factor model alphas. 5 Figure 1 tracks the cumulative
4F

performance of a $1 investment into the (value-weighted) individual-stock strategies analyzed

in Tables 1 and 2.
The finding of statistically insignificant returns and alphas when we sort by medium-term
systematic returns is noteworthy: If stock momentum is caused by factor momentum, we would
expect positive predictability for exactly this measure. However, this is not what we find. To
the contrary, sorting by medium-term idiosyncratic returns largely captures the performance of
a conventional momentum strategy.
Taken together, the results in Table 2 suggest that stock momentum is unlikely to be driven
by the transmission mechanism from factor momentum to stock momentum as proposed by EL.
Stock momentum rather appears to stem from firm-specific return patterns. If there is a spillover
from factor momentum to stock momentum, it is limited to a one-month time frame, as
evidenced by significant short-term systematic momentum. Furthermore, Table 2 shows that
this effect is dominated by strong short-term reversal in idiosyncratic returns, leading to an
overall negative alpha of the short-term reversal strategy in Table 1.

4.3 Robustness

In this section, we conduct a series of robustness tests. First, we test whether our results are
sensitive to the estimation window over which betas are estimated. To this end, we repeat the

5
For the strategies with a significant positive (negative) performance, portfolio returns increase (decrease) fairly
monotonously from Q1 to Q5, making it less likely that the results are spurious (Patton and Timmermann, 2010).

5
analysis in Table 2. However, we now estimate betas using rolling window regressions over
months 𝑡𝑡 −73 to 𝑡𝑡 −13 or months 𝑡𝑡 −36 to 𝑡𝑡 −1, rather than over months 𝑡𝑡 −60 to 𝑡𝑡 −1. The
results, which are displayed in Panel A of Table 3, are qualitatively similar to those in Table 2.
Next, we test the robustness of our results against the choice of the factor model used for
the return decomposition. In particular, we use the CAPM, the Fama and French (1993) three-
factor model, the augmented q-factor model of Hou et al. (2021), and the Stambaugh and Yuan
(2017) mispricing factor model to derive systematic and idiosyncratic returns. We then re-
estimate the analysis in Table 2. Results in Panel B of Table 3 show that medium-term
systematic returns (𝑟𝑟̂𝑖𝑖,𝑡𝑡−12,𝑡𝑡−2) consistently lack any return predictability, whereas medium-term
idiosyncratic momentum (𝜀𝜀̂𝑡𝑡−12,t−2) returns and alphas are positive and statistically significant
for all factor models.
In Table 7 of their paper, EL construct “UMD-style” residual momentum strategies. They
interpret the results as evidence for idiosyncratic momentum arising due to “omitted-factor”
momentum, since the performance of these strategies decreases when additional factors are
added to the factor model used to compute idiosyncratic returns. Comparing our results for the
Fama and French (1993, 2015) three- and five-factor models in Tables 2 and 3, this is not what
we find.
To address concerns that the differences between our results and those of EL are due to
differences in methodology or sample construction, we replicate their results from Table 7. To
this end, we sort stocks into six portfolios by size (< median; ≥ median) and past idiosyncratic
returns (< P30; ≥ P30 & < P70; ≥ P70), computed against three different factor models over
months 𝑡𝑡 −73 to 𝑡𝑡 −13. We then invest long (short) in the small-cap and large-cap portfolios
with the best (worst) performance. Comparing the first two columns in Panel C of Table 3, we
find that our estimates for “UMD-style” idiosyncratic momentum are very similar to those of
EL. However, the third column shows that “UMD-style” strategies based on systematic returns,
which are not tested in EL, fail to outperform, which is at odds with the transmission mechanism
propagated by EL and suggests that factor momentum may not the root cause of stock
momentum.

4.4 Dispersion in betas and momentum performance

In this section, we examine the relationship between momentum profits and the cross-sectional
dispersion in betas. The transmission mechanism proposed by EL implies lower (higher)
momentum profits when stocks in the investment universe are characterized by similar (widely
varying) factor exposures.

6
 To empirically test this hypothesis, we perform a monthly ranking of stocks by their
loadings against each of the five Fama and French (2015) factors (i.e., each stock receives five
different ranks). We then construct an “extreme-beta” subsample by selecting all stock-months,
which rank above the 80th or below the 20th percentiles in terms of their exposures against at
least one factor. All remaining observations with non-missing factor loadings are included in
the “modest-beta” subsample. We then re-estimate the value-weighted momentum and short-
term reversal strategies analyzed in Table 1 within these two subsamples.
Results in Table 4 show that momentum returns are only marginally higher in the “extreme-
beta” subsample (0.58%), compared to the “modest-beta” subsample (0.54%). The same holds
true for the risk-adjusted returns. In contrast, the performance of the short-term reversal
strategies differs substantially across the two subsamples.
In other words, we find that the cross-sectional variation in betas does not significantly
affect the profitability of stock momentum, even though Figure 2 illustrates that the momentum
strategy implemented in the “extreme-beta” subsample displays a much larger variation in its
net factor loadings.

5. Conclusion

EL find that the profitability of stock momentum strategies stems from factor momentum. They
explain their finding by aid of a transmission mechanism from factor momentum to stock
momentum. In this paper, we empirically test EL’s transmission mechanism and present results
at odds with it. First, we show that a momentum strategy based on firm-specific (i.e.,
idiosyncratic) returns yields a statistically significant outperformance, while a strategy based
on expected (i.e., systematic) returns does not. This outcome is the opposite of what we would
expect, if factor momentum drives stock momentum as in EL’s transmission mechanism.
Second, we find that the cross-sectional dispersion in factor loadings, which should be
positively correlated with momentum returns, leaves stock momentum performance largely
unaffected. While we do not dispute EL’s findings that point to factor momentum being related
to stock momentum, our results call into question whether EL’s propagated transmission
mechanism is responsible for the existence of stock momentum.

7
References

Arnott, R., Clements, M., Kalesnik, V., & Linnainmaa, J. (2023). Factor momentum. Review of
Financial Studies, 36(8), 3034-3070.
Beaver, W., McNichols, M., & Price, R. (2007). Delisting returns and their effect on accounting-
based market anomalies. Journal of Accounting and Economics, 43(2-3), 341-368.
Blitz, D., Huij, J., & Martens, M. (2011). Residual momentum. Journal of Empirical Finance,
18(3), 506-521.
Da, Z., Liu, Q., & Schaumburg, E. (2011). Decomposing short-term return reversal. In F. R. B.
o. N. York (Ed.), Federal Reserve Bank of New York Staff Reports (Vol. 513).
Da, Z., Liu, Q., & Schaumburg, E. (2014). A closer look at the short-term return reversal.
Management Science, 60(3), 658-674.
Daniel, K., & Moskowitz, T. (2016). Momentum crashes. Journal of Financial Economics,
122(2), 221-247.
Ehsani, S., & Linnainmaa, J. (2022). Factor Momentum and the Momentum Factor. Journal of
Finance, 77(3), 1877-1919.
Fama, E. F., & French, K. R. (1992). The Cross-Section of Expected Stock Returns. Journal of
Finance, 47(2), 427-465.
Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds.
Journal of Financial Economics, 33(1), 3-56.
Fama, E. F., & French, K. R. (2015). A five-factor asset pricing model. Journal of Financial
Economics, 116(1), 1-22.
Grundy, B. D., & Martin, J. S. (2001). Understanding the Nature of the Risks and the Source of
the Rewards to Momentum Investing. Review of Financial Studies, 14(1), 29-78.
Gupta, T., & Kelly, B. (2019). Factor Momentum Everywhere. Journal of Portfolio
Management, 45(3), 13-36.
Gutierrez, R. C., & Prinsky, C. A. (2007). Momentum, reversal, and the trading behaviors of
institutions. Journal of Financial Markets, 10(1), 48-75.
Hou, K., Mo, H., Xue, C., & Zhang, L. (2021). An Augmented q-Factor Model with Expected
Growth. Review of Finance, 25(1), 1-41.
Hou, K., Xue, C., & Zhang, L. (2020). Replicating Anomalies. Review of Financial Studies,
33(5), 2019-2133.
Jegadeesh, N. (1990). Evidence of Predictable Behavior of Security Returns. Journal of
Finance, 45(3), 881-898.
Jegadeesh, N., & Titman, S. (1993). Returns to Buying Winners and Selling Losers:
Implications for Stock Market Efficiency. Journal of Finance, 48(1), 65-91.
Lehmann, B. (1990). Fads, Martingales, and Market Efficiency. Quarterly Journal of
Economics, 105(1), 1-28.
Lewellen, J. (2002). Momentum and Autocorrelation in Stock Returns. Review of Financial
Studies, 15(2), 533-563.
Lo, A. W., & MacKinlay, A. C. (1990). When Are Contrarian Profits Due to Stock Market
Overreaction? Review of Financial Studies, 3(2), 175-205.
Patton, A. J., & Timmermann, A. (2010). Monotonicity in asset returns: New tests with
applications to the term structure, the CAPM, and portfolio sorts. Journal of Financial
Economics, 98(3), 605-625.
Shumway, T. (1997). The Delisting Bias in CRSP Data. Journal of Finance, 52(1), 327-340.
Stambaugh, R. F., & Yuan, Y. (2017). Mispricing Factors. Review of Financial Studies, 30(4),
1270-1315.
Wang, J., & Wu, Y. (2011). Risk adjustment and momentum sources. Journal of Banking &
Finance, 35(6), 1427-1435.

8
Table 1
Return-based strategies using individual stocks.

Investment Investment Short-term returns Medium-term returns

universe rule (𝑟𝑟𝑡𝑡−1 ) (𝑟𝑟𝑡𝑡−12,𝑡𝑡−2 )
Individual -0.337 0.660
High quintile - 𝑟𝑟̅
stocks (-2.19) (3.05)
Low quintile
(value- -0.190 0.755
(Q5-Q1) 𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
weighted) (-1.06) (2.83)
Individual -1.548 0.846
High quintile - 𝑟𝑟̅
stocks (-7.87) (3.52)
Low quintile
(equal- -1.518 0.765
(Q5-Q1) 𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
weighted) (-5.21) (2.45)

This table shows monthly returns and Fama and French (2015) model alphas of short-term reversal and medium-
term momentum strategies using individual stocks. The sample period is July 1966 to December 2019. 𝑡𝑡-values
(in parentheses) are based on Newey-West standard errors with three lags. Bold numbers indicate statistical
significance at the 5% level or higher.

9
Table 2
Portfolio sorts by systematic and idiosyncratic returns: Baseline results.

Idiosyncratic returns Systematic returns

Quintile returns Short-term Medium-term Short-term Medium-term
(𝜀𝜀̂𝑡𝑡−1 ) (𝜀𝜀̂𝑡𝑡−12,t−2 ) (𝑟𝑟̂𝑡𝑡−1 ) (𝑟𝑟̂𝑡𝑡−12,t−2 )
Q1 1.357 0.635 0.828 0.831
Q2 1.227 0.985 0.930 1.001
Q3 0.926 0.864 1.063 1.064
Q4 0.789 0.926 1.083 1.084
Q5 0.575 1.206 1.211 1.031
-0.783 0.571 0.383 0.200
𝑟𝑟̅
(-6.13) (3.24) (2.19) (1.00)
Q5-Q1
-0.579 0.937 0.414 0.040
𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-4.41) (5.65) (2.35) (0.17)

This table shows monthly returns of quintile portfolios sorted on idiosyncratic or systematic stock returns, as well
as the performance of the corresponding high-low quintile strategies. We compute systematic and idiosyncratic
returns according to Equations (2)-(5). The sample period is July 1966 to December 2019. 𝑡𝑡-values (in parentheses)
are based on Newey-West standard errors with three lags. Bold numbers indicate statistical significance at the 5%
level or higher.

10
Table 3
Portfolio sorts by systematic and idiosyncratic returns: Robustness.

Panel A: Alternative rolling periods

Idiosyncratic returns Systematic returns
Q5-Q1 Short-term Medium-term Short-term Medium-term
(𝜀𝜀̂𝑡𝑡−1 ) (𝜀𝜀̂𝑡𝑡−12,t−2 ) (𝑟𝑟̂𝑡𝑡−1 ) (𝑟𝑟̂𝑡𝑡−12,t−2 )
Fama & French -0.730 0.476 0.570 0.126
𝑟𝑟̅
(2015) beta (-5.62) (2.74) (3.56) (0.74)
estimation from -0.567 0.744 0.610 -0.056
𝑡𝑡 −72 to 𝑡𝑡 −13 𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-4.14) (3.94) (3.72) (-0.28)
Fama & French -0.694 0.513 0.138 0.321
𝑟𝑟̅
(2015) beta (-5.36) (3.07) (0.79) (1.65)
estimation from -0.483 0.852 0.141 0.230
𝑡𝑡 −36 to 𝑡𝑡 −1 𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-3.74) (5.30) (0.79) (1.00)

Panel B: Alternative factor models

Idiosyncratic returns Systematic returns
Q5-Q1 Short-term Medium-term Short-term Medium-term
(𝜀𝜀̂𝑡𝑡−1 ) (𝜀𝜀̂𝑡𝑡−12,t−2 ) (𝑟𝑟̂𝑡𝑡−1 ) (𝑟𝑟̂𝑡𝑡−12,t−2 )
-0.564 0.701 0.570 -0.126
𝑟𝑟̅
Market model (-3.87) (3.71) (2.55) (-0.50)
(CAPM) -0.410 0.889 0.727 -0.205
𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-2.36) (3.87) (2.93) (-0.73)
-0.758 0.537 0.539 0.191
𝑟𝑟̅
Fama & French (-5.87) (3.11) (2.69) (0.93)
(1993) 3-factor -0.607 0.738 0.681 0.145
𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-4.55) (3.82) (3.12) (0.66)
-0.657 0.458 0.383 0.227
𝑟𝑟̅
Hou et al. (-5.11) (2.76) (2.16) (1.23)
(2021) 5-factor -0.471 0.701 0.550 0.198
𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-3.40) (4.01) (3.03) (0.95)
-0.690 0.496 0.476 0.047
Stambaugh & 𝑟𝑟̅
(-5.48) (2.65) (2.50) (0.23)
Yuan (2016)
-0.506 0.849 0.503 -0.156
4-factor 𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-3.49) (4.10) (2.50) (-0.73)

Panel C: UMD-style strategies

Estimates from E & L Our
High (SC+LC) - Low (SC+LC) (2022) Table 7 estimates
Idiosyncratic returns Idiosyncratic returns Systematic returns
Market model 0.58 0.607 -0.141
𝑟𝑟̅
(CAPM) (4.29) (4.23) (-0.77)
Fama & French 0.44 0.478 0.016
𝑟𝑟̅
(1993) 3-factor (3.83) (3.58) (0.11)
Fama & French 0.37 0.383 0.056
𝑟𝑟̅
(2015) 5-factor (3.39) (3.01) (0.42)

The table presents the results of robustness tests for the strategies analyzed in Table 2. In Panel A, betas against
the Fama and French (2015) model factors are determined using rolling window regressions over months 𝑡𝑡 − 72
to 𝑡𝑡 −13 or months 𝑡𝑡 − 36 to 𝑡𝑡 −1. In Panel B, we replace the Fama and French (2015) model with several widely-
used factor models when computing systematic and idiosyncratic returns. In the second and third column of Panel
C, we implement UMD-style strategies. We form six portfolios by size (< median; ≥ median) and past idiosyncratic
returns (< P30; ≥ P30 & < P70; ≥ P70), computed against different factor models over months 𝑡𝑡 − 72 to 𝑡𝑡 −13.
We invest long (short) in the best (worst) performing small-cap and large-cap portfolios. The first column shows
estimates from Table 7 in Ehsani & Linnainmaa (2022). 𝑡𝑡-values (in parentheses) are based on Newey-West
standard errors with three lags. Bold numbers indicate statistical significance at the 5% level or higher.

11
Table 4
Strategy performance in modest-beta and extreme-beta subsamples.

“Modest-beta” sample “Extreme-beta” sample

Short-term total Medium-term total Short-term total Medium-term total
returns (𝑟𝑟𝑡𝑡−1 ) returns (𝑟𝑟𝑡𝑡−12,𝑡𝑡−2 ) returns (𝑟𝑟𝑡𝑡−1 ) returns (𝑟𝑟𝑡𝑡−12,𝑡𝑡−2 )
-0.821 0.544 -0.360 0.585
𝑟𝑟̅
(-5.45) (2.83) (-2.27) (2.61)
Q5-Q1
-0.637 0.642 -0.181 0.682
𝛼𝛼 𝐹𝐹𝐹𝐹5𝐹𝐹
(-3.95) (2.90) (-1.00) (2.45)

This table shows the performance of the (value-weighted) individual-stock momentum and short-term reversal
strategies analyzed in Table 1, implemented within two different subsamples. The “extreme-beta” subsample is
defined as all stock-months, for which at least one of the Fama and French (2015) model factor loadings ranks
above the 80th or below the 20th percentile. The “modest-beta” subsample consists of all remaining observations
with non-missing factor loadings. The sample period is July 1966 to December 2019. 𝑡𝑡-values (in parentheses) are
based on Newey-West standard errors with three lags. Bold numbers indicate statistical significance at the 5%
level or higher.

12
 Cumulative performance of short-term strategies Cumulative performance of medium-term strategies
10 10

5
1

2.5

.1

1
.01

1970m1 1980m1 1990m1 2000m1 2010m1 2020m1 1970m1 1980m1 1990m1 2000m1 2010m1 2020m1

Total ret. Idiosyncratic ret. Systematic ret. Total ret. Idiosyncratic ret. Systematic ret.

Fig. 1. Cumulative performance of individual-stock investment strategies. This figure shows the performance
of a $1 investment into the (value-weighted) individual-stock strategies analyzed in Tables 1 and 2. We adjust the
leverage of each strategy to a realized annual volatility of 10%. The y-axis is shown in logarithmic scale. The
sample period is July 1966 to December 2019.

13
 Market factor (RMRF) Size factor (SMB)
2 2

1 1

0 0

-1 -1

-2 -2

1970m1 1980m1 1990m1 2000m1 2010m1 2020m1 1970m1 1980m1 1990m1 2000m1 2010m1 2020m1

b(H-L); modest-beta stocks b(H-L); extreme-beta stocks b(H-L); modest-beta stocks b(H-L); extreme-beta stocks

Value factor (HML) Profitability factor (RMW)

2 2

1
1

0
0

-1

-1

-2

-2
-3

1970m1 1980m1 1990m1 2000m1 2010m1 2020m1 1970m1 1980m1 1990m1 2000m1 2010m1 2020m1

b(H-L); modest-beta stocks b(H-L); extreme-beta stocks b(H-L); modest-beta stocks b(H-L); extreme-beta stocks

Investment factor (CMA)

3

-1

-2

1970m1 1980m1 1990m1 2000m1 2010m1 2020m1

b(H-L); modest-beta stocks b(H-L); extreme-beta stocks

Fig. 2. Factor loadings of momentum in modest-beta and extreme-beta subsamples. This figure shows Fama
and French (2015) factor loadings of (value-weighted) momentum strategies, implemented within the “modest-
beta” and “extreme-beta” subsamples (see Section 5 and Table 4). We compute the value-weighted average
loadings of composite stocks against each factor for the high and low momentum quintiles (𝛽𝛽𝐻𝐻 and 𝛽𝛽𝐿𝐿 ). Net
loadings then correspond to 𝛽𝛽𝐻𝐻−𝐿𝐿 = 𝛽𝛽𝐻𝐻 − 𝛽𝛽𝐿𝐿 . The sample period is July 1966 to December 2019.

14