Same-Weekday Momentum*

Zhi Da† Xiao Zhang‡

First Draft: Mar 14, 2024; This Draft: Apr 24, 2024

Abstract

A disproportionately large fraction (70%) of stock momentum reects return

continuation on the same weekday (e.g., Mondays to Mondays), or the same-weekday
momentum. Even accounting for partial reversals in other weekdays, the same-weekday
momentum still contributes to a signicant fraction (20% to 60%) of the momentum
effect. This pattern is robust to different size lters, weighing schemes, time periods,
and sample cuts. The same-weekday momentum is hard to square with traditional
momentum theories based on investor mis-reaction. Instead, we provide direct and
novel evidence that links it to within-week seasonality and persistence in institutional
trading. Overall, our ndings highlight institutional trading as an important driver
of the stock momentum.

Key Words: Momentum, Same-Weekday, Return Seasonality.

JEL Codes: G12

* The authors are grateful for advice from Steven Heston, and helpful comments from Robert Battalio, Qinhua Chen, Ian
Dew-Becker, Thomas Ernst, Stefano Giglio, Brandon Han, Serhiy Kozak, Pete Kyle, Jonathan Lewellen, Juhani Linnainmaa, Jianan
Liu, Mark Loewenstein, Dong Lou, Andreas Neuhierl, Markus Pelger, Shrihari Santosh, David Thesmar, Ruslan Tuneshev, Chenyang
Wang, Russ Wermers, Jing Sophia Xue, Rafa Zambrana, Wei Zhou and seminar participants at the Maryland brownbag. All errors are
our own.
†
Finance Department, University of Notre Dame. E-mail: [email protected]
‡
Finance Department, University of Maryland. E-mail: [email protected].

Electronic copy available at: https://ssrn.com/abstract=4806275

1 Introduction

Stocks that outperform (underperform) in the past one year tend to produce higher
future returns over the medium term (Jegadeesh and Titman (1993)). Such a stock
momentum is probably the most well-studied asset pricing anomaly.1 Stock momentum
can be illustrated using a standard Fama and MacBeth (1973) cross-sectional regression
of (log) return ri,t in month t on the (log) past return ri,t−2,t−12 over prior 11 months from
t − 12 to t − 2, skipping the most recent one month t − 1:

ri,t = αt + βt ri,t−2,t−12 + ui,t , (1)

where β̂t = Cov(ri,t , ri,t−2,t−12 )V ar(ri,t−2,t−12 ). A positive and signicant average β̂t
conrms the stock momentum.
The term Cov(ri,t , ri,t−2,t−12 ) can be expressed as the sum of covariance terms between
a daily return in the holding period ( month t) and a daily return in the formation period
(months from t − 12 to t − 2). In addition, we can separate covariance terms involving
daily returns in the same weekday (Monday to Monday, Tuesday to Tuesday, etc.) from
those involving daily returns across different weekdays (Monday to Tuesday, Monday to
Wednesday, ..., Tuesday to Monday, Tuesday to Wednesday, etc.):

5
 5
 5

Cov(ri,t , ri,t−2,t−12 ) = Cov(ri,t (k1 ), ri,t−2,t−12 (k1 )) + Cov(ri,t (k1 ), ri,t−2,t−12 (k2 )),
k =1 k1 =1 k2 =1
1   k2 ̸=k1
Same-Weekday (5×1=5 items)   
Other-Weekday (5×4=20 items)

(2)

where k1 and k2 = 1 to 5, denoting the ve weekdays.

If stock momentum is distributed evenly, we would expect the same-weekday
1
As of 2024, Jegadeesh and Titman (1993) has received more than 15,000 Google citations. For a
comprehensive literature survey, please refer to Jegadeesh and Titman (2011) and Subrahmanyam (2018).

Electronic copy available at: https://ssrn.com/abstract=4806275

covariances to account for about 20% of β̂t . In reality, almost 70% of β̂t comes from the
same-weekday covariances.
To be clear, we are not the rst to discover the same-weekday momentum. Keloharju,
Linnainmaa, and Nyberg (2016, 2021) have already shown that the average daily return
on a particular weekday in the past strongly and positively predicts future returns on the
same weekday. Keloharju, Linnainmaa, and Nyberg (2016, 2021) also document a reversal
effect associated with the same-weekday momentum. For example, past Monday returns
positively predict future Monday returns, but negatively predict future non-Monday
returns. If the reversal is complete during the holding period, then the same-weekday
momentum does not contribute to the momentum effect at all, but simply “redistributes”
it from other weekdays to the same weekday.
To estimate the degree of reversal (x), we decompose the total momentum effect (β̂t )
into three parts: a standard momentum effect (m), the same weekday price pressure (p),
and its reversal (−xp). We assume that the standard momentum effect (m) does not
vary across weekdays but the same weekday price pressure does and we denote them
as p1 , p2 , , p5 for the ve weekdays. We also assume the reversal to spread evenly across
the ve weekdays during the holding period.
Under these assumptions, we can estimate the 7 parameters (m, p1 , p2 , p3 , p4 , p5 , x)
using the Generalized Method of Moments (GMM). We nd the reversal (x) during the
holding period to be only partial, about 28% with its 5th and 95th percentiles to be -31.2%
and 75.9%, respectively. As a result, the net contribution of same-weekday momentum
to the momentum effect, even after accounting for the reversals, is about 47.4%. Such a
momentum decomposition pattern is robust to different size lters, weighing schemes,
time periods, and sample cuts. The net contribution from the same weekday momentum
is always positive and ranges from 18% to 62% of the overall momentum effect.
The decomposition result adds novel insights to our understanding of stock mo-
mentum. A large body of momentum theories is based on some forms of investor

Electronic copy available at: https://ssrn.com/abstract=4806275

mis-reaction to past information or trading signals. This includes both underreaction
(Chan, Jegadeesh, and Lakonishok (1996), Barberis, Shleifer, and Vishny (1998), Hong and
Stein (1999), Hong, Lim, and Stein (2000), Grinblatt and Han (2005), Antoniou, Doukas,
and Subrahmanyam (2013), Da, Gurun, and Warachka (2014), Luo, Subrahmanyam,
and Titman (2021) among others) or continuing overreaction (Daniel, Hirshleifer, and
Subrahmanyam (1998), Lou and Polk (2022) among others). Ex-ante, however, there is no
strong reason why investor mis-reaction should display such a within-week seasonality
pattern.
Instead, we provide direct evidence that links within-week seasonality and persis-
tence in institutional trading to the same-weekday momentum. Using Morningstar daily
fund ow data from 2008 to 2023 for the sample of active US equity mutual funds, we
identify a fund with seasonal ow if it experienced signicantly larger (absolute) fund
ow on a particular weekday than other days of the week in the past one year. Each
month, we nd 16.4% of mutual funds to experience seasonal ows and their (absolute)
ows account for 14.9% of the total fund ows in the past one year. Among these
“seasonal” funds, more than 36.6% of their past (absolute) fund ows occur on the same
weekday.
Similarly, using institutional trading data from ANcerno from 1999 to 2008, we
identify an institution with seasonal trading if it traded signicantly larger volume on
a particular weekday than other days of the week in the past one year. Each month,
We nd 25.6% of ANcerno institutions to display within-week seasonality in trading and
their trading account for 30.1% of the total ANcerno institutional trading volume in the
past one year. Among “seasonal institutions,” more than 35.5% of their past trading occur
on the same weekday.
Both seasonal ows and seasonal trading are persistent. In other words, if a mutual
fund experienced signicantly more (absolute) ow on Mondays in the past one year,
it’s more likely to experience more (absolute) ow on Mondays in the next month as well.

Electronic copy available at: https://ssrn.com/abstract=4806275

Similarly, if an institution traded more on Mondays in the past, it trades more on Mondays
in the future. Moreover, the directions of both seasonal ow and seasonal trading are
persistent as well. If a fund experienced more inow (outow) on Mondays in the past, it
is more likely to experience inow (outow) on Mondays in the future. Consistent with
the ow persistence, if an institution bought (sold) stocks on Mondays in the past, it is
also more likely to buy (sell) stocks on Mondays in the future.
Finally, we directly link seasonal trading to the same-weekday momentum. For each
month and each stock, we rst identify seasonal ANcerno institutions who have traded
the stock in the momentum formation period (months from t − 12 to t − 2). While
past winners and losers are associated with a similar number of seasonal institutions
as other stocks, their past returns are strongly consistent with the trading by seasonal
institutions. For example, seasonal institutions bought (sold) winners (losers) during the
formation period. In other words, seasonal institutional trading could contribute to the
past return. Second, we show that a seasonal institution who bought (sold) a winner
(loser) on Mondays in the formation period is also more likely to buy (sell) the same stock
on Mondays in the holding period. Third, aggregate seasonal trading on a stock in the
formation period positively predicts future returns of that stock in the holding period.
Our paper contributes to the momentum literature. Our simple decomposition
exercise attributes a signicant fraction of stock momentum to the same-weekday
momentum, even after accounting for its reversal. This nding suggests that a large
group of explanations based on investor mis-reaction, while relevant, do not offer a
complete explanation of the momentum prot. Instead, our evidence is more consistent
with momentum theories based on institutional trading (see Grinblatt, Titman, and
Wermers (1995), Goetzmann and Massa (2002), Lou (2012), Vayanos and Woolley (2013),
Cremers and Pareek (2015), Dong, Kang, and Peress (2023) among others). In particular,
persistent institutional trading, combined with its within-week seasonality, seems an
important ingredient of momentum effect. Note that the same weekday momentum

Electronic copy available at: https://ssrn.com/abstract=4806275

underestimates the overall contribution of institutional trading to the momentum effect,
since it only reects a special type of persistent institutional trading.
Our paper also adds to the literature on seasonality in fund ow and institutional
trading. For example, Kamstra et al. (2017) study within-year seasonality in fund ows.
To our best knowledge, we are the rst to examine within-week seasonality.
Finally, our paper also contributes to an emerging literature on seasonality in stock
returns. Some examples include Heston and Sadka (2008, 2010), Heston, Korajczyk, and
Sadka (2010), Keloharju, Linnainmaa, and Nyberg (2016, 2021), Bogousslavsky (2016,
2021), and Lou, Polk, and Skouras (2019). We provide novel evidence on within-week
seasonality and persistence in both mutual fund ow and institutional trading and we
directly link within-week seasonality in trading to within-week seasonality in returns.
The rest of the paper contains two main sections. In Section 2, we present
our momentum decomposition results. Section 3 examines within-week seasonality
in mutual ows and institutional trading, and links such trading seasonality to the
same-weekday momentum. Section 4 concludes.

2 Weekday Momentum Decomposition

In this section, we show that a disproportionately large fraction of stock momentum

reects return continuation on the same weekday. Such a same-weekday momentum is
both statistically and economically signicant.

2.1 Data and sample construction

Our baseline sample covers individual U.S. stocks listed in NYSE, Nasdaq, and Amex
from 1963 through 2021. To alleviate the impact of market micro-structure noise, we
exclude small stocks and penny stocks from our baseline sample. Specically, at the end

Electronic copy available at: https://ssrn.com/abstract=4806275

of each month, we exclude stocks with a price less than $5. We also exclude stocks whose
market capitalization is less than 10th size percentile based on the NYSE breakpoints. We
conrm that our results are robust to different denitions of small and penny stocks. We
obtain price, return, trading volume and market value data from CRSP, book equity data
from Compustat.

2.2 Momentum decomposition: Baseline results

Eq.1 measures the standard momentum effect as the slope coefcient from regressing
(log) return ri,t in month t on the (log) past return ri,t−2,t−12 over prior 11 months from
t − 12 to t − 2 in a Fama and MacBeth (1973) cross-sectional regression. Following Eq.2,
we can decompose such a coefcient in each cross-section into a term reecting return
continuation across the same weekdays and a term reecting return continuation across
different weekdays:

Cov(ri,t , ri,t−2,t−12 )
β̂t =
V ar(ri,t−2,t−12 )
5  5  5
Cov(ri,t (k1 ), ri,t−2,t−12 (k1 )) Cov(ri,t (k1 ), ri,t−2,t−12 (k2 )) (3)
= + ,
k1 =1
V ar(ri,t−2,t−12 ) k =1 k =1
V ar(ri,t−2,t−12 )
1 2
   k2 ̸=k1
Same-Weekday (5×1=5 items)   
Other-Weekday (5×4=20 items)

where k1 and k2 = 1 to 5, denoting the ve weekdays.

Table 1 Panel A reports the decomposition results under different size lters. The rst
row corresponds to the baseline sample which excludes stocks with market capitalization
smaller than the 10th percentile of the NYSE size break point. The average β̂t is 1.17
(t-value = 6.8), conrming a signicant momentum effect during the period from 1963 to
2021.
If the momentum effect spreads evenly across days, we would expect the average
“Same Weekday” component to be 20% × 117 = 023. In reality, it is 0.85, accounting

Electronic copy available at: https://ssrn.com/abstract=4806275

for more than 72% of β̂t on average. This fraction is signicantly higher than 20% with
a t-value of 13.33. In sharp contrast, the “Other Weekday” component, while accounting
for 80% of the covariance terms, is only 28% of β̂t .
Table 1 Panel B further reports the decomposition results by each weekday of the
formation period. Specically, we report the average “Same Weekday” and “Other
Weekday” components for each k2 = 1 to 5. In other words, we decompose the total
momentum coefcient into 10 components. The momentum effect is the strongest on
Mondays of the formation period (accounting for 33% the total momentum coefcient).
The same weekday momentum is also strongest on Mondays (accounting for 30% of
the total momentum coefcient). Put differently, 30% of the momentum effect reects
return continuation from Mondays during the formation period to Mondays during the
holding period. The same weekday momentum is also present in other four weekdays
of the holding period, since the “Same Weekday” components on these weekdays are
also signicantly higher than 4%, or the expected fraction of the momentum coefcient.
The heterogeneity of momentum effect across weekdays allows us to estimate the net
contribution of the same weekday momentum in the next subsection.

2.3 Net contribution of the same-weekday momentum

There could be a seasonal reversal effect associated with the same-weekday mo-
mentum, which offsets the same-weekday momentum and lower their net contribution
to total momentum. As we will demonstrate in the next section, the same-weekday
momentum could come from persistent seasonal trading. An investor who has bought a
stock on Mondays during the momentum formation period is likely to buy the same stock
again on Mondays during the momenutm holding period. As a result, a past Monday
winner is likely to have higher Monday returns during the holding period, reecting
a price pressure. Such a Monday price pressure reverts on other weekdays during the

Electronic copy available at: https://ssrn.com/abstract=4806275

holding period. In this case, past Monday returns positively predict future Monday
returns, but negatively predict future non-Monday returns, as documented by Keloharju,
Linnainmaa, and Nyberg (2016, 2021).
If the price pressure reverts completely during the holding period, then the same-weekday
momentum does not contribute to the momentum effect at all, but simply “redistributes”
it from other weekdays to the same weekday. On the other hand, if the price pressure
only partially reverts during the holding period, then the same-weekday momentum has
a net positive contribution to the momentum effect.
To estimate the degree of reversal (x), we decompose the total momentum effect (β̂t )
into three parts: a standard momentum effect (m), same weekday price pressure (p), and
its reversal (−xp). If reversal is incomplete during the holding period (x < 1), then the
same weekday momentum is a net contributor to the momentum effect.
We assume the standard momentum effect does not vary across weekdays. The
cross-weekday decomposition results in Table 1 Panel B then suggests the magnitude
of price pressure to vary across weekdays. For example, investors may concentrate their
trading on certain days of a week. We denote the same-weekday momentum effect from
past Monday, Tuesday, ..., and Friday as p1 , p2 , , p5 accordingly. Finally, we assume the
reversal to spread evenly across the ve weekdays.
Under these assumptions, the observed covariance between past Monday and future
Monday (scaled by total past return variance) would be m + p1 − 15 xp1 , and the scaled
covariance between past Monday and future Tuesday (or Wednesday, Thursday, and
Friday) is m − 51 xp1 . Similarly, the scaled covariance between past Tuesday and future
Tuesday is m + p2 − 15 xp2 , and the scaled covariance between past Tuesday and future
non-Tuesday (Monday, Wednesday, ..., and Friday) is m − 15 xp2 . The net contribution of
the same weekday momentum is (p1 + p2 + p4 + p4 + p5 )(1 − x).
We therefore have 7 parameters in total and 25 observed scaled covariances between
ve past weekdays and ve future weekdays. We will estimate these parameters

Electronic copy available at: https://ssrn.com/abstract=4806275

θ = {m, p1 , p2 , p3 , p4 , p5 , x} with 25 moment conditions using the Generalized Method of
Moments (GMM). The 25 moment conditions are:
1
E[Covt,k1 ,k1 − (m + pk1 − xpk1 )] = 0 for k1 = 1, 2, 5
5 (4)
1
E[Covt,k1 ,k2 − (m − xpk2 )] = 0 for k1 = 1, 2, 5; k2 = 1, 2, 5 and k1 ̸= k2
5
Cov(ri,t (k1 ),ri,t−2,t−12 (k2 ))
where Covt,k1 ,k2 = V ar(ri,t−2,t−12 )
is the covariance between past weekday k2 ’s
return over months t − 12 to t − 2 and weekday k1 ’s return at current month t, scaled
by past overall return variance. Then the objective function in GMM is:
N N
1  1 
Q(θ) = ( ′
(gt (θ))) W ( (gt (θ))) (5)
N t=1 N t=1

where gt (θ) is the vector of 25 moment conditions for month t, and W is the identify
weighting matrix.
In Table 1 Panel C we report the estimates of 7 parameters based on GMM and
their 5th and 95th percentiles based on a bootstrap of 1000 samples by resampling with
replacement from the full sample of 708 months. The key parameter, x, the reversal effect
as a percentage of the same-weekday momentum, is about 28%, and its 5th and 95th
percentiles of x based on resampling are -31.2% and 75.9% respectively. As a result, the
net contribution of same-weekday momentum after the adjustment of the reversal effect
is about 47.4%, with a lower bound (5th percentile) of 15.5%, which is still positive. The
evidence conrms that the same weekday momentum is a positive net contributor to the
momentum effect, even after accounting for the reversal during the holding period.
In Figure 1 we plot the net contribution of same-weekday momentum in a 10-year
rolling sample, that is, (p1 + p2 + p3 + p4 + p5 ) × (1 − x), given a xed estimate of x = 28%
based on full sample (in red line), or the 5th and 95th percentiles of x (in the shaded area)
from our bootstrap resamples, which represent the lower bound and upper bound of the
net effect respectively. It shows the net contribution from same-weekday momentum still
accounts almost half of total momentum effect in most time of our sample period.

Electronic copy available at: https://ssrn.com/abstract=4806275

2.4 Momentum decomposition: Robustness

Different size lters The rst three rows of Table 2 report the decomposition results
under different size lters. In the rst row, we exclude the smallest 10% of all stocks
in each month. We nd the net contribution of the same weekday momentum to the
total momentum coefcient is 46%. In the third row, we exclude stocks whose market
capitalization are smaller than 20th percentile of the NYSE break point in each month. In
this case, the net contribution of the same weekday momentum is 48%. These numbers
are very similar to 47%, the net contribution of the same weekday momentum in the
baseline case (the second row) that uses the 10% NYSE breakpoints as the size lters.
Equal- vs. value-weighting The baseline results reported in Table 1 weigh each stock
equally in the Fama-MacBeth regressions. The fth row of Table 2 reports the momentum
decomposition results when we weigh each stock by its market capitalization. We nd
the net contribution of the same weekday momentum to be 36%.
Small vs. large stocks We sort stocks in the baseline sample on their market
capitalization into “small,” “medium,” and ”large” groups each month, and then repeat
the decomposition exercise in each group. The results are reported in rows 6-8 of Table
2. The pattern is similar across the three groups, with the same-weekday momentum’s
net contribution ranges from 33% to 58%. In other words, neither small stocks nor a few
large-cap stocks are driving our result.
Liquid vs. illiquid stocks In rows 9-11, we sort stocks in the baseline sample on their
Amihud liquidity measures into “liquid,” “medium,” and ”illiquid” groups each month,
and then repeat the decomposition exercise in each group. The net contribution of the
same-weekday momentum ranges from 15% to 41%.
Different sub-periods In the last three rows of Table 2, we break our baseline sample
period into three sub-periods: 1927-1962, 1963-1992, and 1993-2021. The net contribution
of the same-weekday momentum is always positive. It was 39% pre-1963, increased to

10

Electronic copy available at: https://ssrn.com/abstract=4806275

62% during 1963-1992, and then declined to 18% during the more recent 1993-2021 period.
Past intraday vs. overnight returns A recent paper by Barardehi, Bogousslavsky,
and Muravyev (2023) shows that momentum effect is primarily driven by past intraday
returns. Unreported results conrm the importance of past intraday returns in our setting
as well. Indeed, we nd the same weekday momentum to mostly come from past
intraday returns.
Implications Keloharju, Linnainmaa, and Nyberg (2016, 2021) have shown that
average daily returns in a particular weekday in the past strongly and positively predict
future returns in the same weekday. While such a same-weekday momentum itself is
not new, our contribution is to quantify its net contribution to the standard momentum
effect, after accounting for the reversals, via a simple decomposition exercise. We nd
robust evidence that the same-weekday momentum, in net, drives a signicant fraction
(20% - 60%) of the standard stock momentum.
The decomposition result sheds new lights on the driver of stock momentum. A
large body of momentum theories is based on some forms of investor mis-reaction to
past information or trading signals. Ex-ante, there is no reason why investor mis-reaction
should display a strong within-week seasonality pattern. Put differently, why should
the stock price on Monday mis-reacts only to information or trading signals in prior
Mondays?

2.5 Trading strategies

Before we examine the economic driver of the same-weekday momentum in the next
Section, we rst evaluate its economic signicance using a trading strategy approach.
Again, our objective is not to rediscover the within-week return seasonality as a protable
trading strategy, but rather to quantify its economic magnitude relative to that of the
standard stock momentum. For this reason, our trading strategies will differ slightly

11

Electronic copy available at: https://ssrn.com/abstract=4806275

from those considered in Keloharju, Linnainmaa, and Nyberg (2016, 2021).
Specically, we consider a daily rebalanced trading strategy. Each day during the
momentum holding period (month t), we long (short) stocks in our baseline sample
whose average daily returns on the same weekday during the momentum formation
period (months t − 12 to t − 2) are in the top (bottom) decile. For example, on Mondays
during month t, we buy (sell) stocks whose average Monday returns during months t−12
to t − 2 are high (low); on Tuesdays during month t, we buy (sell) stocks whose average
Tuesday returns during months t − 12 to t − 2 are high (low), etc. We label this strategy
the “same-weekday momentum strategy.”
For comparison, we also consider a “other-weekday momentum strategy.” Each day
during the momentum holding period (month t), we long (short) stocks in our baseline
sample whose average daily returns on other weekdays during the momentum formation
period (months t − 12 to t − 2) are in the top (bottom) decile.
Finally, our benchmark is the standard monthly rebalanced momentum strategy.
Each month t, we long (short) stocks in our baseline sample whose average returns during
the formation period (months t − 12 to t − 2) are in the top (bottom) decile. The trading
strategy results are reported in Table 3.
As reported in column (1) of Panel A, in our baseline sample from 1963 to 2021,
the standard momentum strategy generates a signicant prot of 1.28% per month
(t-value = 5.19). Its risk-adjusted returns are also highly signicant both statistically and
economically. For example, the Fama-French three- and ve-factor alphas are 1.65% and
1.53% per month with respective t-values of 7.32 and 6.79.
In column (2), the daily rebalanced same-weekday momentum strategy generates
much higher prot. The monthly return, three- and ve-factor alphas are 2.05% (t-value =
10.94), 2.18% (t-value = 11.93), and 2.18% (t-value = 11.86), accordingly. In sharp contrast,
the other-weekday momentum strategy is much less protable. In column (3), its monthly
return of 0.28% is not even signicant. The three- and ve-factors are higher, but only

12

Electronic copy available at: https://ssrn.com/abstract=4806275

about one-fourth of those of the same-weekday momentum strategy.
Figure 2 provides a visual illustration of the performance to the three momentum
trading strategies. It plots their cumulative returns (in log scale) since 1963. It is clear that
the daily-rebalanced same-weekday momentum performs the best. A dollar invested in
this strategy in 1963 will grow to almost 106 =1 million dollars in 2021. In sharp contrast,
a dollar invested in the other-weekday momentum strategy in 1963 is less than 2 dollars
in 2021.
Another way to evaluate the contribution of the same-weekday momentum to the
standard momentum return is to exclude the same weekday winners (losers) from the
standard momentum winner (loser) portfolio. For example, on Mondays during the
holding periods, we exclude stocks in the momentum winner (loser) decile that also
belong to the top (bottom) decile of past average Monday returns; on Tuesdays, we
exclude stocks in the momentum winner (loser) decile that also belong to the top (bottom)
decile of past average Tuesday returns, etc. Column (5) of Table 3 Panel A shows that
about 31% of momentum winners and 33% of momentum losers are excluded. Excluding
these stocks signicantly reduces the return to the momentum strategy. For example, its
ve-factor alpha decreases from 1.53% (column 1) to 1.04% (column 5).
In sharp contrast, excluding other-weekday winners or losers actually improves the
protability of the momentum strategy. Specically, on Mondays during the holding
periods, we exclude stocks in the momentum winner (loser) decile that also belong to the
top (bottom) decile of past average Non-Monday returns; on Tuesdays, we exclude stocks
in the momentum winner (loser) decile that also belong to the top (bottom) decile of past
average Non-Tuesday returns, etc. Column (7) of Table 3 Panel A shows that about 66% of
momentum winners and 68% of momentum losers are excluded. Excluding these stocks
signicantly increases the return to the momentum strategy. For example, its ve-factor
alpha increases from 1.53% (column 1) to 2.06% (column 6).
Table 3 Panel B reports the strategy returns by weekdays. The same-weekday

13

Electronic copy available at: https://ssrn.com/abstract=4806275

momentum performs particularly well on Mondays, followed by Fridays. Again,
excluding same-weekday winners & losers reduces the momentum prot while removing
other-weekday winners & losers increases it.

2.6 Horserace in cross-sectional regressions

In Table 4, we conduct a horerace among the three momentum effects using

Fama-MacBeth cross-sectional regressions. Speccally, we regress a daily return in the
holding period (month t) on a standard momentum variable (past return during month
t − 12 to t − 2), a same-weekday momentum variable (past return on the same weekdays
during month t − 12 to t − 2) and a other-weekday momentum variable (past return on
other weekdays during month t − 12 to t − 2). The regressions also control for other
stock characteristics with return predictive power. The results from the value-weighted
regressions (Panel A) and equal-weighted regressions (Panel B) are very similar.
Three patterns emerge from these regressions. First, consistent with the decompo-
sition and trading strategy results, it is the same-weekday rather than other-weekday
momentum that reliably predicts future daily returns. Second, when we include the
standard momentum variable and the same-weekday momentum variable in the same
regression, the predictive power of the standard momentum is reduced signicantly.
Third, controlling for other stock characteristics does not change the result. Besides,
unreported results conrm that controlling for 25 dummy variables for 5 × 5 size and
book-to-market portfolio does change our results, which is suggested by Kamstra (2017)
to alleviate the concerns that momentum (or same-weekday momentum) is driven by
cross-sectional variation of mean return.
To summarize, in this section, we document a novel empirical pattern: a dispropor-
tionately large fraction (70%) of stock momentum reects return continuation on the same
weekday. Even accounting for partial reversals in other weekdays, the same-weekday

14

Electronic copy available at: https://ssrn.com/abstract=4806275

momentum still contributes to a signicant fraction (20% to 60%) of the momentum effect.
The same-weekday momentum is robust to different size lters, weighing schemes, time
periods, and sample cuts. This within-week seasonality pattern is hard to explain using
traditional momentum theories based on investor mis-reaction. Next, we investigate its
potential economic driver.

3 Seasonality and Persistence in Institutional Trading

Many theories of momentum have been proposed in the past. They are not
necessarily mutually exclusive and could all contribute to the momentum. But if
70% of stock momentum reects return continuation on the same weekday, then the
primary economic force underlying momentum should also generate such a within-week
seasonality pattern. In this section, we provide novel empirical evidence suggesting that
seasonality and persistence in institutional trading could be one such force.

3.1 Data

Mutual fund daily fund ow data during 2008-2023 is downloaded from Morningstar
Direct. We focus on active US equity mutual funds (both dead and currently alive). For
each fund, we focus on the oldest share class. The key daily dollar ow variable is named
“estimated fund level net ow (comprehensive) (daily).” We divide the daily dollar fund
ow by the fund size at the end of the previous day to compute the daily percentage
fund ow. We require an fund to have at least 6-month ows history and at least 20 ow
days in the past 11 months. To compare the daily ow at a particular weekday with that
at other weekdays for a fund, we rst rell the missing daily ow value with 0 to get a
balanced number of observations for each of ve weekdays.
Institutional trading data during 1999-2008 is from ANcerno database. We dene

15

Electronic copy available at: https://ssrn.com/abstract=4806275

an institution at the client-manager level by at looking at the items “clientcode” and
“clientmgrcode” in the dataset. We require an institution to have at least 6-month trading
history and at least 10 trading days in the past 11 months. We merge each trade from an
institution in ANcerno with the stock identity in CRSP by matching the items “cusip” and
“symbol” from ANcerno with items “NCUSIP” and “TICKER” from CRSP at the same
time. Again, to compare the daily trading at one day of the week with other days, we
rell 0 for the day without any trading for the sake of a balanced daily trading subsample
of ve weekdays.

3.2 Seasonal ow and trading

Return continuation across the same weekday can be consistent with concentrated
trading on the same weekday. Hence we rst examine the prevalence of such “seasonal”
trading by institutions. Such a concentrated trading could in turn arise from concentrated
investor fund ow on that weekday. As a result, we also look into “seasonal” ows to
mutual funds.
Specically, for each Morningstar mutual fund and each month, we identify a
“seasonal” fund if its average (absolute) fund ow on a particular weekday in the past
one year is signicantly higher than the average (absolute) fund ow on other weekdays.
For example, if 40% of past one-year (absolute) fund ow occurs on Monday, which is
signicantly higher than the percentages on the other four days of the week, then the
fund is identied as a “seasonal” fund, or more specically a “Monday seasonal” fund in
that month. The corresponding ow concentration ratio is 40%.
Table 5 Panel A reports that on average 304 funds (or 16.4% of the cross-section) are
classied as the “seasonal” funds each month during the 2009-2023 Morningstar sample
period. Their (absolute) fund ows account for 14.9% of the total (absolute) fund ows in
our sample, so seasonable funds are representative in terms of their fund ow size.

16

Electronic copy available at: https://ssrn.com/abstract=4806275

 The average concentration ratio is 36.6% for these “seasonal” funds, meaning that
36.6% of their (absolute) fund ows occur on one particular day of the week. When we
break down the results by weekday, we nd that “Tuesday seasonal” funds are most
common (32.4%) but their average concentration ratio is the lowest (34%). In contrast,
while only 10.1% of the “seasonal” funds are ”Monday seasonal” funds, their average
concentration ratio of 42.1% is the highest.
Figure 3 Panel A plots the percentage of “seasonal” funds, their (absolute) fund ows
as a percentage of total (absolute) fund ows, and their average concentration ratio over
time. While the prevalence of “seasonal” funds has declined from above 20% of the
sample in 2009-2012 to around 10% more recently, their average concentration ratio of
just below 40% is fairly stable.
Seasonal ow could lead to seasonal trading. For each ANcerno institution and each
month, we identify a “seasonal” institution if its average dollar trading volume on a
particular weekday in the past one year is signicantly higher than the average dollar
trading volume on other weekdays. If that weekday is Monday, then the institution is
identied as a “Monday seasonal” institution in that month.
Table 5 Panel B reports that on average 291 institutions (or 25.6% of the cross-section)
are classied as the “seasonal” institutions each month during the 1999-2008 ANcerno
sample period. Their trading volumes account for 30.1% of the total volume in our
sample, so “seasonal” institutions are more active traders.
Their average concentration ratio is 35.5%, meaning that 35.5% of their trading occurs
on one particular day of the week. When we break down the results by weekday,
we nd that “Thursday seasonal” funds are most common (28.5%) but their average
concentration ratio is the lowest (32.3%). In contrast, while only 12.9% and 13.7% of the
“seasonal” institutions are ”Monday” and “Friday seasonal” institutions, their average
concentration ratios of 38.4% and 37.6% are higher.
Figure 3 Panel B plots the percentage of “seasonal” institutions, their dollar trading

17

Electronic copy available at: https://ssrn.com/abstract=4806275

volume as a percentage of total volume in our sample, and their average concentration
ratio over time. The prevalence of “seasonal” institutions is quite stable and even
increased a bit towards 2008. Their average concentration ratio of just below 40% is also
fairly stable.
To conclude this subsection, we nd a large number of mutual funds to experience
concentrated ow on a particular weekday and even more institutions to concentrate their
trading on a particular day of the week.

3.3 Persistence in seasonal ow and trading

We then examine whether the within-week seasonality in fund ow and institutional
trading is persistent over time.
In Table 6, we examine “seasonal” funds. Each month t, we sort “seasonal” funds
into deciles based on their concentration ratio in the prior momentum formation period
(months t − 12 to t − 2). In Panel A, we report the past 11-month average concentration
ratio in column (1) and the average concentration ratio on the same weekday during
month t in column (2). Importantly, all the concentration ratios in column (2) are above
20%. Take decile 10 for example, the funds in this decile experience 64.07% of (absolute)
ow on one particular weekday in the past 11 months. In month t, they continue to
experience 24.83% of their (absolute) ow on the same weekday. The number 24.83% is
also signicantly higher than that for decile 1 (21.53%), meaning that funds with more
concentrated ow in the past continue to have concentrated ow on the same weekday
in the future. Columns (3) to (7) break down the “seasonal” funds by weekdays in each
decile. The pattern is similar to those reported in Table 5 Panel A that Monday and Friday
see more funds with highly concentrated ows than other three weekdays.
In Panel B, we sort “seasonal” funds based on their past 11-month net daily fund
ow on the concentrated weekday into deciles. Column (1) shows that “seasonal” funds

18

Electronic copy available at: https://ssrn.com/abstract=4806275

in decile 10 (1) experience an average net daily inow (outow) of 1.50% (-0.58%) on the
concentrated weekday in the past 11 months. Column (2) reports the average daily net
ow on the same weekday in month t and it shows that the direction of the “seasonal”
ow is also highly persistent. “Seasonal” funds experiencing inow (outow) on the
concentrated weekday in the past continue to experience inow (outow) on the same
weekday in the future. Columns (3) to (7) again break down the “seasonal” funds by
weekdays in each decile. It shows that Monday and Friday see more funds with extreme
(inow and outow) ows than other three weekdays as well.
Persistent “seasonal” fund ows can result in persistent “seasonal” institutional
trading, which we conrm in Panels C and D. In Panel C, we sort “seasonal” institutions
in each month t into deciles based on their concentration ratios in the momentum
formation period (months t − 12 to t − 2). We report the past 11-month average
concentration ratio in column (1) and the average concentration ratio on the same
weekday during month t in column (2). Again, all the concentration ratios in column
(2) are above 20%. Take decile 10 for example, the institutions in this decile conduct
72.89% of their trading on one particular weekday in the past 11 months. In month t, they
continue to conduct 34.77% of their trading on the same weekday. The number 34.77%
is also signicantly higher than that for decile 1 (20.70%), meaning that institutions with
more concentrated trading in the past continue to have concentrated trading on the same
weekday in the future. Compared to the mutual fund results in Panel A, “seasonal”
institutional trading seems even more persistent. Columns (3) to (7) break down the
“seasonal” institutions by weekdays in each decile. The pattern is also similar to those
reported in Table 5 Panel B in that Monday and Friday see more institutions with highly
concentrated trading than Tuesday, Wednesday, and Friday.
In Panel D, we sort “seasonal” institutions based on their past 11-month net trade
imbalance on the concentrated weekday into deciles. Column (1) shows that “seasonal”
institutions in decile 10 bought more stocks than sold on the concentrated weekday in the

19

Electronic copy available at: https://ssrn.com/abstract=4806275

past 11 months. The opposite is true for “seasonal” institutions in decile 1. Column (2)
reports the average net trade imbalance on the same weekday in month t and it shows that
the direction of the “seasonal” institutional trading is also highly persistent. “Seasonal”
institutions who bought more stocks than sold on the concentrated weekday in the past
continue to buy stocks in net on the same weekday in the future. Columns (3) to (7) again
break down the “seasonal” institutions by weekdays in each decile. Similar to the fund
ow results in Panel B, Monday and Friday also see more institutions with extreme (buy
and sell) trading than other three weekdays. These breakdown results on ve weekdays
for fund ow and institutional trading are all consistent with our decomposition results
by weekday in Tables 1 and 3 that Monday and Friday contribute more to same-weekday
momentum than other weekdays.

3.4 Momentum and seasonal trading

In this subsection, we link momentum to “seasonal” institutional trading more

directly. We focus on the 1999-2008 sample where we can use ANcerno data to measure
“seasonal” institutional trading.
Each month t, we construct momentum portfolios by sorting stocks in our baseline
sample on their formation-period (months t−12 to t−2) returns into terciles. We consider
tercile- rather than decile-sorts in order to make sure we have a sufcient number of
“seasonal” institutions in each portfolio so “seasonal” trading can be measured more
precisely. For each stock, we also identify ANcerno institutions who have traded it during
the formation period, and among them those “seasonal” institutions.
Table 7 Panel A reports summary statistics related to institutional trading on these
momentum terciles. We do not observe a large difference between past winners and
losers. On average, each stock has been traded by 5 to 6 “seasonal” institutions in
the formation period. These “seasonal” institutions represent about 5.35% to 5.47% of

20

Electronic copy available at: https://ssrn.com/abstract=4806275

all ANcerno institutions trading the stock. The average concentration ratio is about
35% to 36% for these “seasonal” institutions. Put differently, the amount of “seasonal”
institutional trading does not differ signicantly across the momentum terciles.
When we examine the direction of “seasonal” institutional trading in Panel B, how-
ever, we see a signicant difference across the momentum terciles. Column (1) reports the
average net trade imbalance by the seasonal institutions during the momentum formation
period (months t − 12 to t − 2). The imbalance is computed for each institution-stock
pair rst, before being averaged across all institutions who traded that stock. During
the formation period, “seasonal” institutions bought more winners than losers. Their
trade imbalances are consistent with past returns. More importantly, when we examine
the average net trade imbalance by these seasonal institutions during month t, or the
momentum holding period in column (2), we nd the imbalance pattern to persist. The
evidence suggests that “seasonal” institutions bought winners (sold losers) in the past
and continue to buy winners (sell losers) on the same weekday in the future.
Columns (3) and (4) repeat Columns (1) and (2) but compute trade imbalances
by aggregating across different “seasonal” institutions to each stock rst. In other
words, institutions are weighted by their trading volumes. We nd the same pattern:
“seasonal” institutions bought winners (sold losers) in the past and continue to buy
winners (sell losers) on the same weekday in the future. Put differently, persistent
“seasonal” institutional trading is consistent with the same-weekday momentum.
In Table 8, we conrm the persistence of “seasonal” institutional trading in a panel
regression at stock-institution-month level. We use the past net trading imbalance from
a seasonal institution on a stock in the past 11 months (months t − 12 to t − 2) to predict
its net trading imbalance of the same stock on the same concentrated weekday in month
t. The net trading imbalance of an institution on a stock is calculated as the difference
between the dollar buy and dollar sell at that stock divided by the sum of the buy and
sell. Column (1) presents the baseline results that seasonal institutions are more likely to

21

Electronic copy available at: https://ssrn.com/abstract=4806275

trade a stock in same direction as before, with a positively signicant coefcient of 0.0217,
which means, on average, a Monday seasonal institution will buy (or sell) 2.17% more of
the same stock on future Mondays if it traded the stock with purely 100% buy (or sell) on
the past Mondays. This result is robust to the size and book-to-market ratio of the stock
as controls in column (2).
One thing noteworthy is that the persistent trading coefcient remains positively
signicant even when we control for the past 11-month return of the stock in column (3)
of Table 8. In other words, seasonal institutions keep trading a stock persistently as before,
regardless of the past performance of the stock, which implies the trading persistence of
seasonal institutions is not merely another manifestation of buying winners and selling
losers, but instead a new potential contributor to the persistent performance of stocks in
momentum.
Finally, we directly link past “seasonal” institutional trading to future stock return
using a panel regression at stock-month level. In Table 9, we aggregate the net (buy
minus sell) trading dollars on a stock from all seasonal institutions who traded that stock
at their own concentrated weekday in past 11 months from t − 12 to t − 2 and then scale
this aggregated net dollar trading by the market value of that stock at the end of previous
month t − 1, which we label as “past 11-month seasonal trading imbalance”. In column
(1) of Table 9, we use this stock-level aggregate seasonal trading imbalance in the past to
predict monthly stock return at t in a univariate regression, which shows a signicantly
positive coefcient of 0.00119. This means that if all seasonal institutions together traded
a stock with net positive 100% volume of the total market value over the past 11 months,
it will lead to 0.119% return next month. Consistent with the results in previous table,
the stock-month regression results are also robust to controlling for size, book-to-market
ratio, and the past 11-month return of the stock as shown in columns (2) and (3).
Notably, in column (3), the coefcient for past seasonal trading imbalance represents
roughly a quarter of the coefcient for the past 11-month return, signifying its substantial

22

Electronic copy available at: https://ssrn.com/abstract=4806275

explanatory power on future returns. This signicance is noteworthy given the fact that
ANcerno institutions only account for about 10% of all institutions in the market.
To conclude this section, we rst nd a large fraction of equity funds with seasonal
ows and institutions with seasonal trading. Second, we nd both the seasonal ow and
trading are highly persistent. Last, we directly link the persistent seasonal trading to stock
momentum.

4 Conclusion

In this paper, we document a new empirical fact about the stock momentum. A
signicant fraction (47.4%) of stock momentum reects return continuation on the same
weekday, even after accounting for reversals in other weekdays. This pattern is extremely
robust to different size lters, weighing schemes, time periods, and sample cuts. The net
contribution of the same weekday momentum to the overall stock momentum ranges
from about 20% to 60%.
The same-weekday momentum is hard to explain using traditional momentum
theories based on investor mis-reaction. Instead, we nd within-week seasonality and
persistence in institutional trading to be its driver. A large number of institutions
experience disproportionately large ows on a particular day of the week and concentrate
their trading on that weekday. Such a seasonal trading tends to be highly persistent,
which drives the same-weekday momentum. Overall, our evidence suggests that
institutional trading is an important ingredient of the stock momentum effect.

23

Electronic copy available at: https://ssrn.com/abstract=4806275

References
Antoniou, Constantinos, John A. Doukas, and Avanidhar Subrahmanyam. 2013. Cognitive Dissonance,
Sentiment, and Momentum. Journal of Financial and Quantitative Analysis 48 (1):245–275.

Barardehi, Yashar H., Vincent Bogousslavsky, and Dmitriy Muravyev. 2023. What Drives Momentum and
Reversal? Evidence from Day and Night Signals.

Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. 1998. A model of investor sentiment. Journal of
Financial Economics 49 (3):307–343.

Bogousslavsky, Vincent. 2016. Infrequent Rebalancing, Return Autocorrelation, and Seasonality. The Journal
of Finance 71 (6):2967–3006.

———. 2021. The cross-section of intraday and overnight returns. Journal of Financial Economics
141 (1):172–194.

Chan, Louis K. C., Narasimhan Jegadeesh, and Josef Lakonishok. 1996. Momentum Strategies. The Journal
of Finance 51 (5):1681–1713.

Cremers, Martijn and Ankur Pareek. 2015. Short-Term Trading and Stock Return Anomalies: Momentum,
Reversal, and Share Issuance. Review of Finance 19 (4):1649–1701.

Da, Zhi, Umit G. Gurun, and Mitch Warachka. 2014. Frog in the Pan: Continuous Information and
Momentum. Rev. Financ. Stud. 27 (7):2171–2218.

Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam. 1998. Investor Psychology and Security
Market Under- and Overreactions. The Journal of Finance 53 (6):1839–1885.

Dong, Xi, Namho Kang, and Joel Peress. 2023. Fast and Slow Arbitrage: The Predictive Power of Capital
Flows for Factor Returns.

Fama, Eugene F. and James D. MacBeth. 1973. Risk, Return, and Equilibrium: Empirical Tests. Journal of
Political Economy 81 (3):607–636.

Goetzmann, William N. and Massimo Massa. 2002. Daily Momentum and Contrarian Behavior of Index
Fund Investors. Journal of Financial and Quantitative Analysis 37 (3):375–389.

Grinblatt, Mark and Bing Han. 2005. Prospect theory, mental accounting, and momentum. Journal of
Financial Economics 78 (2):311–339.

Grinblatt, Mark, Sheridan Titman, and Russ Wermers. 1995. Momentum Investment Strategies, Portfolio
Performance, and Herding: A Study of Mutual Fund Behavior. The American Economic Review
85 (5):1088–1105.

Heston, Steven L., Robert A. Korajczyk, and Ronnie Sadka. 2010. Intraday Patterns in the Cross-section of
Stock Returns. The Journal of Finance 65 (4):1369–1407.

Heston, Steven L. and Ronnie Sadka. 2008. Seasonality in the cross-section of stock returns. Journal of
Financial Economics 87 (2):418–445.

24

Electronic copy available at: https://ssrn.com/abstract=4806275

———. 2010. Seasonality in the Cross Section of Stock Returns: The International Evidence. Journal of
Financial and Quantitative Analysis 45 (5):1133–1160.

Hong, Harrison, Terence Lim, and Jeremy C. Stein. 2000. Bad News Travels Slowly: Size, Analyst Coverage,
and the Protability of Momentum Strategies. The Journal of Finance 55 (1):265–295.

Hong, Harrison and Jeremy C. Stein. 1999. A Unied Theory of Underreaction, Momentum Trading, and
Overreaction in Asset Markets. The Journal of Finance 54 (6):2143–2184.

Jegadeesh, Narasimhan and Sheridan Titman. 1993. Returns to Buying Winners and Selling Losers:
Implications for Stock Market Efciency. The Journal of Finance 48 (1):65–91.

———. 2011. Momentum. Annual Review of Financial Economics 3 (1):493–509.

Kamstra, Mark J. 2017. Momentum, Reversals, and other Puzzles in Fama-MacBeth Cross-Sectional
Regressions.

Kamstra, Mark J., Lisa A. Kramer, Maurice D. Levi, and Russ Wermers. 2017. Seasonal Asset Allocation:
Evidence from Mutual Fund Flows. Journal of Financial and Quantitative Analysis 52 (1):71–109.

Keloharju, Matti, Juhani T. Linnainmaa, and Peter Nyberg. 2016. Return Seasonalities. The Journal of Finance
71 (4):1557–1590.

———. 2021. Are return seasonalities due to risk or mispricing? Journal of Financial Economics
139 (1):138–161.

Lou, Dong. 2012. A Flow-Based Explanation for Return Predictability. Rev. Financ. Stud. 25 (12):3457–3489.

Lou, Dong and Christopher Polk. 2022. Comomentum: Inferring Arbitrage Activity from Return
Correlations. The Review of Financial Studies 35 (7):3272–3302.

Lou, Dong, Christopher Polk, and Spyros Skouras. 2019. A tug of war: Overnight versus intraday expected
returns. Journal of Financial Economics 134 (1):192–213.

Luo, Jiang, Avanidhar Subrahmanyam, and Sheridan Titman. 2021. Momentum and Reversals When
Overcondent Investors Underestimate Their Competition. The Review of Financial Studies 34 (1):351–393.

Subrahmanyam, Avanidhar. 2018. Equity market momentum: A synthesis of the literature and suggestions
for future work. Pacic-Basin Finance Journal 51:291–296.

Vayanos, Dimitri and Paul Woolley. 2013. An Institutional Theory of Momentum and Reversal. Rev. Financ.
Stud. 26 (5):1087–1145.

25

Electronic copy available at: https://ssrn.com/abstract=4806275

 Figure 1: Historical decomposition of momentum covariance

This gure shows the historical monthly average values of the total momentum effect and net contribution
of same-weekday momentum with consideration of potential reversal effect in a 10-year rolling window.
The dashed green line denotes the average monthly coefcient of total momentum effect, and the red
line depicts the net contribution of the same-weekday momentum given a xed percentage of reversal
effect estimated from full sample. We also plot lower bound and upper bound of the net contribution in
shaded area given the 5% and 95% percentile of the percentage of reversal effect based on bootstrap. We
estimate the raw momentum effect (m), same-weekday momentum (p1 , p2 , p3 , p4 , p5 ), and the percentage of
reversal effect (x) of the same-weekday momentum by minimizing the 25 weekday-to-weekday covariance
moment conditions based on GMM in Eq.5. We rst estimate the percentage of reversal effect x given
the full sample of 708 months, and then estimate the 5% and 95% percentiles of x by resampling with
replacement from the original sample for 1000 times. Then we estimate the net contribution using 6
parameters: m, p1 , p2 , p3 , p4 , p5 (with x given from full-sample estimate or 5% and 95% percentile from
resampling) in each of 10-year rolling subsample. The net contribution of same-weekday momentum
would be (p1 +p2 +p3 +p4 +p5 )×(1−x) and the total momentum would be (p1 +p2 +p3 +p4 +p5 )×(1−x)+25m.
The sample includes all individual stocks listed on NYSE, Amex, and NASDAQ. The penny stocks with
price below $5 and small-cap stocks below NYSE 10% breakpoints are excluded each month. The original
sample spans period from 1963 through 2021 and the rst 10-year rolling average value starts at 1972.

26

Electronic copy available at: https://ssrn.com/abstract=4806275

 Figure 2: Historical cumulative return of three trading strategies

This gure presents the cumulative return (in log scale) of investing $1 since 1963 in long-short portfolios
of three strategies—momentum, same-weekday, and other-weekday—by sorting individual stocks equally
into 10 decile portfolios based on past 11-month (skipping recent one month) overall return, same-weekday
return and other-weekday return respectively. The decile portfolios are value-weighted. The sample
includes all individual stocks listed on NYSE, Amex, and NASDAQ. The penny stocks with price below
$5 and small-cap stocks below NYSE 10% breakpoints are excluded each month. The sample covers years
1963 through 2021.

27

Electronic copy available at: https://ssrn.com/abstract=4806275

 Figure 3: Seasonal fund ow and seasonal institutional trading

This gure plots the percentage of seasonal fund ow and institutional trading out of all fund ow and
institutional trading every month. The blue line and orange line denote the number of and the ow
(or trading volume) of seasonal funds (or institutions) as a percentage of all funds (or institutions). The
green line plots the average concentration ratio of the seasonal funds (or institutions) at their concentrated
weekday to all ve weekdays. Every month, we dene a seasonal fund (or institution) at a specic weekday
by comparing its daily absolute ows (or dollar trading volume) on that particular weekday and on other
four weekdays in the past year based on T-test at 10% signicance level. Panel A covers all equity funds
from 2009 to 2023 from Morningstar and Panel B all institutions from 1999 to 2008 from ANcerno.
Panel A: Seasonal fund ow

Panel B: Seasonal institutional trading

28

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 1: Weekday Momentum Decomposition

This table reports the decomposition of Fama-Macbeth regression coefcient into 5 same-weekday items and 20 other-weekday items:
5
 5
 5

Cov(ri,t ,ri,t−2,t−12 ) Cov(ri,t (k1 ), ri,t−2,t−12 (k1 )) Cov(ri,t (k1 ), ri,t−2,t−12 (k2 ))
β̂t,l = V ar(ri,t−2,t−12 )
= +
k1 =1
V ar(ri,t−2,t−12 ) k1 =1 k2 =1
V ar(ri,t−2,t−12 )
   k2 ̸=k1
  
Same-Weekday (5×1=5 items)
Other-Weekday (5×4=20 items)
where ri,t and ri,t (k) are the log monthly return and the log weekday k’s return at month t. And ri,t−2,t−12 and ri,t−1,t−12 (k)
are the log past 11-month return and its log weekday k’s return. Panel A columns (1), (2), and (3) present the average value of
total momentum, the same-weekday, and the other-weekday respectively. The t-stat in below parentheses of columns (2) and (3) are
based on the test of whether the same-weekday (and other-weekday) component contributes more than 20% (and 80%) to total effect.
Columns (4) and (5) present the contribution of same-weekday and other-weekday in percentage out of the total momentum. Panel
B reports the monthly average value of total momentum, the same-weekday and the other-weekday by ve weekdays in formation
period, and their contribution to total momentum in percentage. The t-stat in below parentheses of columns (2) and (3) refers to
the test of whether the same-weekday (and other-weekday) component contributes more than 4% (and 16%) to total momentum
effect. In panel C, columns (1) through (7) report the estimates of 7 parameters: total raw momentum effect (25 × m), same-weekday
momentum (p1 , p2 , p3 , p4 , p5 ), and the percentage of reversal effect (x) of the same-weekday momentum by minimizing the 25
weekday-to-weekday covariance moment conditions based on GMM in Eq.5. Columns (8) and (9) report the net contribution of
the same-weekday momentum ((p1 + p2 + p3 + p4 + p5 ) × (1 − x)) and its contribution to total moenntum in percentage. In below
brackets we report the 5% and 95% percentiles of these parameters respectively based on 1000 samples resampled with replacement
from original sample of 708 months. Our data covers sample period from 1963 to 2021, and includes all individual stock listed in
NYSE, Nasdaq, and Amex, except for penny stocks with price below $5 and small stocks below NYSE 10% breakpoints. All scaled
covariance values are multiplied by 100.

Panel A: Decomposition of monthly covariance

(1) (2) (3) (4) (5)
Total momentum Same-Weekday Other-Weekday % of same-weekday % of other-weekday
mean 1.17 0.85 0.32 72% 28%
(6.80) (13.33) -(13.33)
# of items per month 25 5 20 20% 80%
# of month 708 708 708

Panel B: Decomposition of monthly covariance by ve weekdays in formation period

(1) (2) (3) (4) (5) (6)
Weekday Total momentum Same-Weekday Other-Weekday % of total momentum % of same-weekday % of other-weekday
Monday 0.39 0.35 0.03 33% 30% 3%
(10.20) -(10.20)
Tuesday 0.32 0.14 0.17 27% 12% 15%
(5.01) -(5.01)
Wednesday 0.18 0.12 0.06 15% 10% 5%
(3.94) -(3.94)
Thursday 0.16 0.07 0.09 14% 6% 8%
(2.06) -(2.06)
Friday 0.12 0.16 -0.04 10% 14% -4%
(7.62) -(7.62)
# of items per month 5 1 4 20% 4% 16%
# of month 708 708 708

Panel C: Estimation of parameters based on GMM

(1) (2) (3) (4) (5) (6) (7) (8) (9)
m (total) p1 p2 p3 p4 p5 x Net % of net
contribution contribution
full sample estimate 0.61 0.35 0.12 0.10 0.05 0.15 28.0% 0.55 47.4%
[5% percentile, 95% percentile] [0.03, 1.10] [0.29, 0.40] [0.08, 0.16] [0.05, 0.14] [0.01, 0.09] [0.11, 0.19] [-31.2%, 75.9%] [0.19, 1.01] [15.5%, 97.2%]

29

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 2: Robustness check

This table reports the decomposition results of momentum Fama-Macbeth regression coefcient into same-weekday and
other-weekday components in cross-section with different lter of small stocks, based on equal- or value- weight, within three
market-cap (size) or Amihud liquidity subgroups, and by three subperiods, respectively. Columns (1), (2), and (3) report the average
monthly value of total momentum, same-weekday and other-weekday respectively. The t-stats are provided in parentheses below
(for columns (2) and (3)) based on a test of whether the same-weekday (and other-weekday) component contributes more than 20%
(and 80%) to total effect. Columns (4) through (6) report the total raw momentum effect (25 × m), total same-weekday momentum
(p1 +p2 +p3 +p4 +p5 ) and reversal effect (x) as a percentage of same-weekday momentum based on an estimation of these 7 parameters
by minimizing the 25 weekday-to-weekday covariance moment conditions in Eq.5. Columns (7) and (8) are the net contribution of the
same-weekday momentum ((p1 + p2 + p3 + p4 + p5 ) × (1 − x)) and its contribution of percentage to total momentum effect. The
sample includes all individual stocks listed in NYSE, Nasdaq, and Amex from 1963 to 2021 (except for sub-period 1: 1927 to 1962),
exluding penny stocks with price below $5 and small stocks below NYSE 10% breakpoints. All covariance values are multiplied by
100.

Different weight, size, liquidity, and subperiods

(1) (2) (3) (4) (5) (6) (7) (8)
Category Total momentum Same-Weekday Other-Weekday m (total) p (total) x Net % of net
contribution contribution
Size lter 10% all-sample 1.33 0.88 0.45 0.72 0.77 20.8% 0.61 46%
- (15.26) -(15.26)
10% NYSE 1.17 0.85 0.32 0.61 0.77 28.0% 0.55 47%
- (13.33) -(13.33)
20% NYSE 1.13 0.81 0.32 0.59 0.73 26.2% 0.54 48%
- (11.75) -(11.75)
Weight equal 1.17 0.85 0.32 0.61 0.77 28.0% 0.55 47%
(13.33) - (13.33)
value 1.09 0.96 0.13 0.70 0.93 57.5% 0.39 36%
(10.41) - (10.41)
Size small 1.18 0.86 0.32 0.50 0.78 12.5% 0.68 58%
(12.48) - (12.48)
medium 1.14 0.79 0.35 0.77 0.71 46.2% 0.38 33%
(10.67) - (10.67)
large 1.05 0.74 0.31 0.54 0.66 23.7% 0.51 48%
(8.35) - (8.35)
Amihud liquid 1.38 0.81 0.57 0.81 0.66 14.2% 0.57 41%
liquidity (8.11) - (8.11)
medium 1.36 0.81 0.55 1.00 0.67 47.3% 0.36 26%
(10.58) - (10.58)
illiquid 0.91 0.77 0.14 0.77 0.74 80.9% 0.14 15%
(12.15) - (12.15)
Subperiods period 1: 1927-1962 1.71 1.69 0.02 1.22 1.34 41.3% 0.79 39%
(10.08) - (10.08)
period 2: 1963-1992 1.67 1.27 0.40 0.64 1.17 11.2% 1.04 62%
(13.16) - (13.16)
period 3: 1993-2021 0.70 0.43 0.27 0.57 0.36 65.2% 0.13 18%
(5.26) - (5.26)

30

Electronic copy available at: https://ssrn.com/abstract=4806275

Table 3: Decile portfolios performance of momentum, same-weekday and
other-weekday trading strategies

This table reports the average value-weighted monthly return (computed from daily return series) of decile portfolios constructed
based on past 12-month overall, same-weekday, and other-weekday returns, skipping the most recent one-month, as shown in
columns (1), (2), and (3), respectively. Portfolios are daily rebalanced since stocks are equally sorted based on different past
return information every day for same-weekday and other-weekday strategies. In columns (4) or (6), we exclude the top 10% of
same-weekday or other-weekday winners for momentum decile 10, and the bottom 10% of same-weekday or other-weekday losers
from momentum decile 1. The columns (5) and (7) indicate the percentage of stocks excluded from the original momentum decile
portfolios in columns (4) and (6) respectively. Panel A presents the average monthly returns for ten deciles and long-short portfolios
for ve strategies respectively. The last three rows present the long-short average return, along with the Fama-French three- and ve-
factor adjusted alphas. Panel B reports the long-short average monthly returns on ve weekdays for these ve strategies respectively.
The sample includes individual stocks listed in NYSE, Nasdaq, and Amex from 1963 to 2021, exluding penny stocks with price below
$5 and small stocks below NYSE 10% breakpoints. T-statistics are provided in parentheses below.

Panel A: Average monthly return of three strategies

(1) (2) (3) (4) (5) (6) (7)
Deciles Momentum Same- Other- Mom. excl. % of stocks Mom. excl. % of stocks
weekday weekday same-weekday excluded other-weekday excluded
1 0.04 -0.33 0.62 0.32 31% -0.35 66%
2 0.46 0.10 0.77
3 0.60 0.25 0.79
4 0.57 0.45 0.62
5 0.58 0.59 0.57
6 0.60 0.64 0.63
7 0.70 0.73 0.65
8 0.79 1.03 0.61
9 0.92 1.26 0.59
10 1.32 1.72 0.90 1.13 33% 1.46 68%
Long-short 1.28 2.05 0.28 0.81 1.81
(5.19) (10.94) (1.22) (3.24) (6.99)
FF3 1.65 2.18 0.55 1.14 2.13
(7.32) (11.93) (2.58) (4.99) (8.71)
FF5 1.53 2.18 0.49 1.04 2.06
(6.79) (11.86) (2.28) (4.52) (8.41)

Panel B: The average holding return on ve weekdays respectively

(1) (2) (3) (4) (5)
Weekdays Momentum Same- Other- Mom. excl. Mom. excl.
weekday weekday same-weekday other-weekday
Monday 1.03 4.04 -1.39 -0.11 2.67
(1.64) (8.12) -(2.54) -(0.18) (3.83)
Tuesday 1.69 1.36 1.17 1.46 1.78
(3.07) (3.58) (2.22) (2.59) (3.23)
Wednesday 1.59 1.52 1.38 1.56 1.27
(2.78) (3.48) (2.66) (2.73) (2.12)
Thursday 0.56 1.00 0.12 0.21 0.88
(1.03) (2.55) (0.23) (0.38) (1.62)
Friday 1.53 2.45 0.00 0.83 2.51
(3.25) (6.40) -(0.01) (1.71) (5.06)

31

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 4: Fama-Macbeth regression

This table reports the Fama-Macbeth regression of different past return components in predicting the daily return in next month.
”Ret all”, ”Ret same-weekday”, and ”Ret other-weekday” are the overall return, same-weekday return, and other-weekday return in
the past 11 months (skipping recent one month) respectively. Control variables include Size (market value by the end of last month,
B/M (book value divided by market value by the end of last month), Amihud (absolute daily return divided by daily dollar volume
and then averaged over previous 11 months), Turnover (daily turnover ratio and then averaged over previous 11 months), Amihud
same-weekday (absolute daily return divided by daily dollar volume and then averaged over same weekdays in previous 11 months),
Turnover same-weekday (daily turnover ratio and then averaged over same weekdays in previous 11 months). Panel A and B report
the regression results based on value- or equal-weight of each stock respectively. The sample includes individual stocks listed in NYSE,
Nasdaq, and Amex from 1963 to 2021, exluding penny stocks with price below $5 and small stocks below NYSE 10% breakpoints. All
coefcients are multiplied by 1000. T-statistics are provided in parentheses below.

Panel A: Value weight

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Ret all 3.38 1.06 11.99 3.37 1.56 1.87 1.83
(3.86) (1.23) (10.54) (3.73) (1.73) (2.17) (2.14)
Ret same-weekday 17.30 15.48 16.90 13.88 12.00 11.09 11.10
(11.98) (12.26) (11.91) (9.68) (9.48) (9.49) (9.48)
Ret other-weekday 0.91 -11.30 1.36 1.42
(1.00) -(10.62) (1.50) (1.52)
Size -0.19 -0.21 -0.22 -0.21 -0.28 -0.26
-(1.50) -(1.59) -(1.69) -(1.69) -(2.17) -(2.09)
B/M -0.14 -0.13 -0.17 -0.13 -0.18 -0.19
-(1.37) -(1.29) -(1.71) -(1.38) -(2.02) -(2.12)
Amihud -4.43
-(1.44)
Turnover -2.36
-(1.48)
Amihud same-weekday -3.23
-(0.95)
Turnover same-weekday -2.17
-(1.35)
R2 2.14% 1.27% 1.86% 2.94% 2.84% 3.01% 4.57% 3.70% 4.30% 5.32% 6.86% 6.82%

Panel B: Equal weight

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Ret all 3.09 1.27 9.49 3.22 1.66 1.98 1.98
(4.95) (2.09) (12.14) (4.91) (2.56) (3.24) (3.22)
Ret same-weekday 13.89 12.19 13.73 12.08 10.31 9.16 9.28
(14.52) (15.35) (14.16) (12.90) (13.30) (13.58) (13.67)
Ret other-weekday 1.05 -8.18 1.57 1.24
(1.68) -(12.75) (2.47) (1.92)
Size -0.34 -0.33 -0.34 -0.34 -0.43 -0.42
-(2.38) -(2.28) -(2.38) -(2.41) -(2.96) -(2.89)
B/M 0.12 0.11 0.11 0.11 0.04 0.03
(2.00) (1.89) (2.00) (1.93) (0.96) (0.67)
Amihud -3.88
-(2.43)
Turnover -4.25
-(2.82)
Amihud same-weekday -4.08
-(2.37)
Turnover same-weekday -3.73
-(2.54)
R2 0.94% 0.48% 0.80% 1.23% 1.20% 1.26% 1.84% 1.33% 1.69% 2.10% 3.32% 3.28%

32

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 5: Seasonal fund ow and seasonal institutional trading by day of the week

This table reports the number of seasonal funds and seasonal institutions (in total and by the day of the week), respectively. To dene
a seasonal fund-weekday, at the end of each month, we look at the daily ows (absolute ow value scaled by the total net asset by
the end of previous day, relled with 0 if missing) of a fund in the past year to test whether the daily ows on a specic weekday is
signicantly larger than that on other four weekdays based on a T-test of the mean of these two samples. For example, in January, to test
whether a fund is a seasonal fund on Monday, we compare the mean of its Monday daily ows and the mean of Tuesday, Wednesday,
Wednesday, Thursday, and Friday daily ows in last year using the T-statistics (T =  2 µ1 −µ22 , where µ1 , σ12 and µ2 , σ22 are the
σ1 /n1 +σ2 /n2
mean and variance of daily ows on Monday and on other four weekdays, respectively) at 10% signicance level. We only take one
weekday with largest daily ow for a fund at a given month if there are multiple weekdays with signicantly larger ows in past year.
The seasonal institution-weekday is dened in a similar way but using the daily absolute trading volume in dollars (relled with 0 if
missing). Panel A covers all equity funds from Morningstar from 2009 to 2023 and Panel B all institutions from ANcerno from 1999 to
2008. Columns (1) and (2) present the average number of all funds (or institutions) and seasonal funds (or institutions) across months;
Columns (3) and (4) present the average percentage of the number and of the ow volume (or trading volume) in past year of seasonal
funds (or institutions) out of all funds (or institutions). And column (5) reports the average concentration ratio of the weekday with
signicantly larger daily ows out of all ve weekdays among all dened seasonal funds (or institutions) across months. We also
breakdown the seasonal funds (or institutions) by their concentrated weekday in ve rows Mon, Tue, Wed, Thu, and Fri, respectively.

Panel A: Seasonal fund ow

(1) (2) (3) (4) (5)
# of # of % of % of ow volume avg. concentration ratio
funds seasonal funds seasonal funds of seasonal funds of seasonal funds
% of all 1824.1 303.7 16.4% 14.9% 36.6%
by weekday Mon 29.8 10.1% 15.1% 42.1%
(% of all seasonal) Tue 103.1 32.4% 27.6% 34.0%
Wed 72.4 24.0% 23.0% 35.7%
Thu 51.5 18.0% 17.5% 38.1%
Fri 46.2 15.6% 16.8% 38.6%

Panel B: Seasonal institutional trading

(1) (2) (3) (4) (5)
# of # of % of % of trading volume avg. concentration ratio
institutions seasonal institutions seasonal institutions of seasonal institutions of seasonal institutions
% of all 1144.3 291.2 25.6% 30.1% 35.5%
by weekday Mon 37.8 12.9% 9.1% 38.4%
(% of all seasonal) Tue 59.6 19.6% 18.7% 35.6%
Wed 78.9 25.3% 28.4% 34.9%
Thu 91.2 28.5% 32.4% 32.3%
Fri 42.7 13.7% 11.4% 37.6%

33

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 6: Persistence of seasonal fund ow and seasonal institutional trading

This table presents the monthly sorting results to test the persistence of the concentration and of the direction at the same weekday in
the past and in the future for seasonal fund ow (Panels A and B) and seasonal institutional trading (Panels C and D), respectively.
In Panel A, at each month, we sort all dened seasonal funds equally into ten deciles based on their concentration ratio of the daily
absolute ow at the weekday with signicantly larger ow out of all ve weekdays in past 11 months (skipping most recent month),
as shown in column (1), and report the next-month concentration ratio of the absolute daily ow at the same weekday out of all ve
weekdays in column (2). The rows in “decile” from 1 to 10 report the average values over institutions in same decile and across
months. And the row “10-1” calculates the difference between decile 10 and 1 and the below row presents the t-value accordingly.
Columns (3) to (7) report the percentage of seasonal funds at each of ve weekdays out of all seasonal funds within a given decile
group, respectively. In Panel B, we sort seasonal funds equally into ten deciles based on their past average net daily ow at the
weekday with signicantly larger absolute daily ow in past 11 months as shown in column (1) and report the next-month average
daily ow at the same weekday in column (2). Again, deciles 1 to 10 report the average values over institutions and across months and
row “10-1” the difference between decile 10 and 1 with t-stat in below parenthesis. Columns (3) through (7) present the percentage of
seasonal funds at ve weekdays respectively within each decile group. Panels A and B cover all dened seasonal equity funds from
Morningstar from 2009 to 2023.

Panel A: Persistence of seasonal fund ow concentration

(1) (2) (3) (4) (5) (6) (7)
sorting past 11-month next-month
decile concentration ratio concentration ratio Mon Tue Wed Thu Fri
1 23.95% 21.53% 3.2% 43.2% 27.1% 14.5% 12.0%
2 26.33% 21.71% 4.3% 42.1% 25.4% 14.7% 13.5%
3 28.26% 21.81% 6.1% 39.3% 24.8% 16.0% 13.8%
4 30.23% 22.39% 8.0% 36.3% 25.2% 15.6% 14.9%
5 32.43% 22.08% 8.7% 34.2% 22.8% 18.3% 15.9%
6 35.01% 22.39% 11.5% 31.3% 23.7% 17.7% 15.8%
7 38.08% 22.69% 13.5% 26.8% 23.0% 20.5% 16.2%
8 42.12% 22.60% 14.9% 26.6% 21.5% 19.4% 17.6%
9 48.59% 23.00% 14.5% 23.4% 20.7% 23.6% 17.8%
10 64.07% 24.83% 18.5% 20.8% 22.5% 19.9% 18.2%
10-1 40.12% 3.30%
t-stat (6.67)

Panel B: Persistence of seasonal fund ow direction

(1) (2) (3) (4) (5) (6) (7)
sorting past 11-month next-month
decile daily ow daily ow Mon Tue Wed Thu Fri
1 -0.58% -0.27% 13.9% 22.6% 24.3% 20.9% 18.3%
2 -0.19% -0.11% 10.4% 32.7% 22.7% 20.7% 13.5%
3 -0.11% -0.08% 7.1% 36.5% 24.0% 19.4% 12.9%
4 -0.07% -0.05% 5.7% 41.1% 24.3% 14.1% 14.9%
5 -0.04% -0.03% 7.0% 38.4% 26.0% 14.7% 13.9%
6 0.00% 0.00% 10.6% 34.1% 25.7% 16.5% 13.2%
7 0.04% 0.01% 11.3% 31.9% 22.5% 18.1% 16.2%
8 0.12% 0.03% 13.0% 32.3% 22.3% 16.6% 15.9%
9 0.26% 0.09% 12.0% 29.5% 22.8% 18.4% 17.3%
10 1.50% 0.23% 11.6% 25.6% 21.5% 21.4% 19.9%
10-1 2.08% 0.51%
t-stat (10.84)

34

Electronic copy available at: https://ssrn.com/abstract=4806275

Table 6: Persistence of seasonal fund ow and seasonal institutional trading
(continued)

In Panel C, we sort all dened seasonal institutions every month equally into ten deciles based on their concentration ratio of the
daily trading volume in dollars at the weekday with signicantly larger trading volume out of all ve weekdays in past 11 months
(skipping most recent month), as shown in column (1), and report the next-month concentration ratio of the daily trading volume at
the same weekday out of all ve weekdays in column (2). The rows in “decile” from 1 to 10 report the average values across months.
And the row “10-1” calculates the difference between decile 10 and 1 and the below row presents the t-value accordingly. Columns
(3) to (7) report the percentage of seasonal institutions at each of ve weekdays out of all seasonal institutions within a given decile
group, respectively. In Panel D, we sort seasonal institutions equally into ten deciles based on their trading imbalance (the difference
between buy and sell in dollars scaled by the sum of buy and sell) at the weekday with signicantly larger trading volume in past 11
months as shown in column (1) and report the next-month trading imbalance at the same weekday in column (2). Again, deciles 1 to
10 report the average values across months and row “10-1” the difference between decile 10 and 1 with t-stat in below parenthesis.
Columns (3) through (7) presents the percentage of seasonal institutions at ve weekdays respectively within each decile group. Panel
C and D cover all dened seasonal institutions from ANcerno from 2000 to 2008.

Panel C: Persistence of seasonal institutional trading concentration

(1) (2) (3) (4) (5) (6) (7)
sorting past 11-month next-month
decile concentration ratio concentration ratio Mon Tue Wed Thu Fri
1 22.98% 20.70% 20.4% 12.8% 19.6% 37.9% 9.3%
2 24.61% 21.25% 10.5% 16.7% 24.0% 38.3% 10.4%
3 25.87% 21.67% 8.6% 17.1% 27.7% 34.4% 12.2%
4 27.21% 21.86% 9.0% 18.8% 28.1% 30.6% 13.4%
5 28.86% 22.43% 10.6% 20.3% 26.7% 29.4% 13.0%
6 31.07% 22.12% 10.7% 22.7% 26.7% 27.0% 12.9%
7 34.17% 22.50% 11.9% 23.8% 24.9% 25.4% 14.1%
8 38.99% 24.94% 13.8% 23.4% 23.7% 22.8% 16.2%
9 48.00% 24.44% 16.1% 20.8% 21.4% 22.5% 19.1%
10 72.89% 34.77% 18.2% 20.8% 25.0% 17.7% 18.3%
10-1 49.91% 14.08%
t-stat (12.72)

Panel D: Persistence of seasonal institutional trading direction

(1) (2) (3) (4) (5) (6) (7)
sorting past 11-month next-month
decile trading imbalance trading imbalance Mon Tue Wed Thu Fri
1 -0.57 -0.10 15.8% 23.7% 21.8% 22.8% 15.8%
2 -0.22 -0.03 11.4% 19.6% 24.9% 31.4% 12.6%
3 -0.12 -0.04 11.1% 16.4% 28.3% 31.9% 12.3%
4 -0.06 0.00 11.9% 17.0% 26.5% 33.7% 10.9%
5 -0.02 -0.01 12.2% 20.9% 25.5% 30.9% 10.5%
6 0.01 -0.01 13.8% 22.2% 26.4% 25.3% 12.3%
7 0.05 0.01 13.0% 20.9% 25.1% 28.2% 12.8%
8 0.11 0.02 12.2% 18.4% 23.5% 31.1% 14.8%
9 0.22 0.05 13.2% 17.4% 22.9% 29.6% 16.9%
10 0.57 0.20 15.3% 20.7% 23.0% 21.3% 19.8%
10-1 1.13 0.30
t-stat (10.76)

35

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 7: Seasonal trading of momentum

This table presents the persistence of seasonal institutional trading in momentum tercile portfolios. We rst equally sort individuals
stocks into three terciles every month based on their past 11-month return (skipping recent one month). Then we look at how many
(seasonal) institutions trading on these stocks at each of ve weekdays in the past 11 months. Panel A column (1) reports the total
number of month-weekday in the sample. Panel A columns (2) and (3) report the average number of institutions who traded the
stocks on a weekday of past 11 month and the average number of seasonal institutions who traded the stocks on their concentrated
weekday of past 11 month. Columns (4) and (5) present the percentage of seasonal institutions out of all institutions and the average
trading concentration ratio of seasonal institutions at their concentrated weekday. Panel B reports the average institution-level trading
imbalance (the difference between buy and sell divided by the sum of buy and sell) on a stock across seasonal institutions who traded
the stock at its own concentrated weekday in past 11 months in column (1) and the next-month average institution-level trading
imbalance among these seasonal institutions on the stock at future same weekday in column (2). We then average the averaged
institution-level trading imbalance of each stock across stocks within same tercile, and nally average across ve weekdays of months.
Panel B columns (3) and (4) reports the trading imbalance of seasonal institutions aggregated at stock level (the difference between
total buy and total sell in dollars on a stock from all seasonal institutions at a concentrated weekday divided by the sum of the total buy
and total sell) in the past 11 months and in the next one month respectively. We then average the stock-level trading imbalance across
stocks within same tercile and nally average across ve weekdays of months. The row “3-1” indicates the difference between tercile
3 and 1, and with the t-stat in below parentheses for next-month trading imbalance. The sample includes individual stocks listed in
NYSE, Nasdaq, and Amex, excluding penny stocks with price below $5 and small stocks with size below NYSE 10% breakpoints,
and covers all dened seasonal institutions with specic concentrated weekday at each month from ANcerno, from 2000 to 2008. We
require that each stock has at least 50 institutions who traded it in one weekday of past 11 months.

Panel A: % seasonal traders for momentum portfolios

(1) (2) (3) (4) (5)
Tercile # of avg. # of avg. # of % of avg. concentration ratio
by momentum month-weekday institutions seasonal institutions seasonal institutions of seasonal institutions
1 535 117.5 6.1 5.35% 35.1%
2 535 118.6 6.2 5.39% 36.0%
3 535 103.4 5.7 5.47% 35.0%

Panel B: Imbalance persistence for winners and losers

(1) (2) (3) (4)
Tercile past 11-month next-month past 11-month next-month
by momentum trading imbalance trading imbalance trading imbalance trading imbalance
(institution-level) (institution-level) (stock-level) (stock-level)
1 0.99% -2.65% 2.14% -3.00%
2 3.95% 0.04% 5.03% -0.37%
3 10.74% 0.98% 10.49% 0.24%
3-1 9.74% 3.64% 8.35% 3.24%
t-stat (4.38) (3.78)

36

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 8: Regression: Persistence in seasonal trading at individual stock

This table presents the regression results of the persistence of trading for seasonal institutions at stock-institution-month level. The
regression model is: T radeImbi,j,t = T radeImbi,j,t−12,t−2 + Controls + ϵi,j,t , where i refers to any stocks in our baseline sample
at month t, j is the dened seasonal institution at month t who also traded stock i in past 11 months and the independent variable
T radeImbi,j,t−12,t−2 is therefore the trading imbalance (the difference between buy and sell in dollars divided by the sum of buy
and sell) of seasonal institution j on stock i in months t − 12 to t − 2. The dependent variable T radeImbi,j,t is the next-month trading
imbalance of seasonal institution j on stock i at month t. The control variables include the log market value and book to market ratio
at the end of last month, and the past 11-month return of the stock respectively in columns (2), and (3). We impose institution-, month-,
and stock-level xed effects in each regression and cluster the standard errors at institution level. The clustered standard errors are
attached in below parentheses. The institution sample covers all dened seasonal institutions with specic concentrated weekday at
each month from ANcerno from 2000 to 2008. The stock sample includes all individual stocks listed in NYSE, Nasdaq, and Amex over
the same period, excluding penny stocks with price below $5 and small stocks with size below NYSE 10% breakpoints. We denote the
10%, 5%, and 1% signicance levels with *, **, and *** respectively.

(1) (2) (3)

Indep. Var. next-month next-month next-month
trading imbalance trading imbalance trading imbalance
Dep. Var.
past 11-month trading imbalance 0.0217*** 0.0208*** 0.0206***
(0.00713) (0.00711) (0.00714)
log(ME) 0.0114** 0.00582
(0.00545) (0.00526)
B/M 0.000275 -0.000536
(0.00148) (0.00133)
past 11-month return 0.0126***
(0.00457)
Constant -0.0245*** -0.204** -0.118
(0.00203) (0.0856) (0.0826)

# of Obs 1,203,735 1,181,005 1,181,005

R-squared 0.177 0.177 0.177
Institution FE. YES YES YES
Month FE. YES YES YES
Stock FE. YES YES YES
Clu. Std. Institution Institution Institution

37

Electronic copy available at: https://ssrn.com/abstract=4806275

 Table 9: Regression: Past seasonal trading imbalance and future stock return

This table presents the regression results of predicting next-month individual stock returns using past aggregate seasonal institutional
trading in a regression: Ri,t = T radeImbi,t−12,t−2 + Controls + ϵi,t , where the independent variable, the past 11-month seasonal
trading imbalance T radeImbi,t−12,t−2 , is the sum of trading imbalance of all seasonal institutions who traded the stock i at their
concentrated weekday in the past 11 months from t − 12 to t − 2 (the difference between total buy and total sell in dollars from
all seasonal institutions at their concentrated weekday in past 11 months) scaled by the market value of the stock by the end of the
previous month t − 1, and the dependent variable Ri,t is the next-month return of stock i at month t. The control variables include the
log market value and book to market ratio at the end of last month, and the past 11-month return of the stock respectively in columns
(2), and (3). We impose month-, and stock-level xed effects in each regression and cluster the standard errors at stock level. The
clustered standard errors are attached in below parentheses. The stock sample includes all individual stocks listed in NYSE, Nasdaq,
and Amex, excluding penny stocks with price below $5 and small stocks with size below NYSE 10% breakpoints. The institution
sample covers all dened seasonal institutions with specic concentrated weekday at each month from ANcerno, from 2000 to 2008.
We denote the 10%, 5%, and 1% signicance levels with *, **, and *** respectively.

(1) (2) (3)

Indep. Var. next-month return next-month return next-month return
Dep. Var.
past 11-month seasonal trading imbalance 0.00119*** 0.00107** 0.00105**
(0.000410) (0.000421) (0.000419)
log(ME) -0.0410*** -0.0434***
(0.00121) (0.00128)
B/M 1.07e-05 -0.000150
(0.000163) (0.000144)
past 11-month return 0.00451***
(0.000666)
Constant 0.00373*** 0.585*** 0.618***
(2.32e-05) (0.0171) (0.0182)

# of Obs 238,502 231,940 231,940

R-squared 0.189 0.203 0.203
Month FE. YES YES YES
Stock FE. YES YES YES
Clu. Std. Stock Stock Stock

38

Electronic copy available at: https://ssrn.com/abstract=4806275