Market Janurary

G Model
ARTICLE IN PRESS
QUAECO-852; No. of Pages 18
The Quarterly Review of Economics and Finance xxx (2015) xxxxxx
Contents lists available at ScienceDirect
The Quarterly Review of Economics and Finance

journal homepage: www.elsevier.com/locate/qref
Is the January effect rational? Insights from the accounting valuation

model
Kathryn E. Easterday a, , Pradyot K. Sen b
a
b
Wright State University, United States

University of Washington-Bothell, United States
a r t i c l e
i n f o
Article history:
Received 30 May 2014
Received in revised form 23 April 2015
Accepted 18 May 2015
Available online xxx
Keywords:
January effect
Permanent earnings
Tax-loss selling
a b s t r a c t
Employing a permanent earnings valuation model and a novel sample partition, we nd evidence that the
January effect anomaly is consistent with rational economic market behavior. Investors in rms which
experience January effect return premiums appear to discount rst quarter earnings performance but
reward permanent earnings and expectations of future improvements. Our evidence also supports a taxloss selling explanation for the January effect. We nd that the January effect is experienced by relatively
few rms in the sample overall, but a substantial percentage of January effect rms are identied as
potential tax-loss sellers. Our results complement prior research suggesting that the January effect is
neither a result of irrational noise traders nor consistent with systemic risk factor explanations. Our
study reconciles the assumption of arbitrage inherent in trading studies with a fundamental accounting
valuation approach and offers some further insights into the nature of this market phenomenon.
2015 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.
1. Introduction
This paper nds that the January effect anomaly is associated
with accounting earnings and expectations about future earnings, in a manner both economically rational and consistent with
accounting theory. This work extends that of Henker and Debapriya
(2012), who argue against an irrational noise trader explanation for the January effect. It complements that of Klein and
Rosenfeld (1991), who present evidence that the January effect can
be explained at least in part by new information in January about
upcoming earnings announcements. Finally, our accounting valuation approach complements Mashruwala and Mashruwala (2011),
who argue that return seasonality is incompatible with systemic
risk explanations.
Fama (1998) and Gerlach (2007, 2010) both argue that many
so-called market anomalies are tenuous in the sense that they are
sensitive to the methodologies used to detect or measure them.
Far from being tenuous, the January effect a capital markets
Corresponding author at: Wright State University, Raj Soin College of Business,
Department of Accountancy, 298 Rike Hall, 3640 Col. Glenn Highway, Dayton, OH
45435-0001, United States. Tel.: +1 937 775 3304.
E-mail addresses: kathryn.easterday@wright.edu (K.E. Easterday),
pksen@uwb.edu (P.K. Sen).
phenomenon in which return premiums are on average higher

in January than in other months of the year1 persists in deance of economic theory which says it should be arbitraged away.
Although some studies suggest that the January effect is disappearing (Gu, 2003; He & He, 2011; Hensel & Ziemba, 2000; Mehdian &
Perry, 2002), numerous others provide evidence that the January
effect continues to appear in modern US capital markets (Anderson,
Gerlach, & DiTraglia, 2007; Brown & Luo, 2006; Ciccone, 2011;
Dzhabarov & Ziemba, 2010; Easterday, Sen, & Stephan, 2009; Haug
& Hirschey, 2006; Mashruwala & Mashruwala, 2011; Ziemba, 2011)
although it does not occur every year (Easterday et al., 2009;
Hulbert, 2008).
Tax management is the most common rationalization for the
January effect: investors take advantage of capital losses at year
end for tax purposes, resulting in temporary downward mispricings that create large January returns when prices rebound after
the turn of the year (Branch, 1977; Brown, Ferguson, & Sherry,
2010; Chen & Singal, 2004; Dalton, 1993; Givoly & Ovadia, 1983;
Grifths & White, 1993; Grinblatt & Keloharju, 2004; Jones, Lee, &
Apenbrink, 1991; Koogler & Maberly, 1994; Ma, Rao, & Weinraub,
1988; Phua, Chan, Faff, & Hudson, 2010; Sikes, 2014; Starks, Yong,
1
Some researchers call it the turn of the year effect. Both terms are widely used
throughout the literature, often interchangeably.
http://dx.doi.org/10.1016/j.qref.2015.05.001
1062-9769/ 2015 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.
Please cite this article in press as: Easterday, K. E., & Sen, P.K. Is the January effect rational? Insights from the accounting valuation
model. The Quarterly Review of Economics and Finance (2015), http://dx.doi.org/10.1016/j.qref.2015.05.001
G Model
2
ARTICLE IN PRESS
K.E. Easterday, P.K. Sen / The Quarterly Review of Economics and Finance xxx (2015) xxxxxx
& Zheng, 2006).2 However, there is evidence that tax minimizing

behavior by itself is not enough to drive the January effect (Brown,
Keim, Kleidon, & Marsh, 1983; Fountas & Segredakis, 2002; Jones &
Wilson, 1989; Pettingill, 1989; Reinganum, 1983; Ritter, 1988; Sias
& Starks, 1997; van den Bergh & Wessels, 1985).
Rather than attempting to explain the January effect,
Mashruwala and Mashruwala (2011) exploit this seasonal
increase in stock prices to examine whether accruals quality
measures proxy for information risk. Their ndings suggest that
such measures proxy more for rm attributes associated with
tax-loss selling than for information risk. Studies by Brauer and
Chang (1990), Peterson (1990), and Reinganum and Gangopadhyay
(1991) also provide evidence that information risk is not related to
the January effect.
Extant research into the January effect itself appears primarily
in the nance literature where it is often explained as a temporary
mispricing anomaly resulting from various market inefciencies
and risks resulting in arbitrage opportunities. However, some studies (Loughran, 1997; Mashruwala & Mashruwala, 2011; Roll, 1983;
Seyhun, 1993; Tinic & West, 1984) argue that systemic risk factor
explanations are not compatible with seasonal market behaviors.
A recent study by Henker and Debapriya (2012) provides evidence
that the January effect is not driven by irrational noise traders.
CAPM models neither predict nor explain risk, especially (or only) in
January (Best, Hodges, & Yoder, 2006; Corhay, Hawawini, & Michel,
1987; Gultekin & Gultekin, 1987; Kryzanowski & Zhang, 1992;
Ritter & Chopra, 1989; Thaler, 1987).
Insights from the fundamental valuation theory of accounting
suggest that under a no arbitrage condition, returns in January
or any time period should be positively associated with
contemporaneous accounting earnings and information affecting
expectations about future accounting performance (Feltham &
Ohlson, 1995; Ohlson, 1995, 2001). With the exception of Klein
and Rosenfeld (1991) there is little research considering how or
whether the January effect anomaly is associated with accounting earnings information in a market-efcient, rational economic
manner.3 Their evidence shows that low-PE stocks with low annual
earnings forecasts in December outperform in January relative
to other low-PE stocks and they argue that the prices of these
stocks rise in January because it becomes apparent to investors
then that actual earnings for the just-completed year will be better than was forecasted in December. Their analyses employ a
trailing-earnings-to-price ratio and focus on earnings forecasts and
investors expectations for the earnings announcement for the year
immediately past.
We extend Klein and Rosenfeld (1991) by employing a form
of the permanent earnings model developed in Easterday, Sen,
and Stephan (2011)4 to examine the association between January
returns and earnings in the rst quarter. The ESS model expresses
returns as a function of contemporaneous earnings level, earnings growth, and a term representing the sustainability of earnings
growth, and the model derives directly from Ohlsons (1995, 2001)
valuation framework. Employing an accounting valuation model
rather than an ad hoc trading model enables us to forgo an assumption of arbitrage and demonstrate that, consistent with economic
and accounting theory, earnings information plays an important
role in the economic intuition of the January effect phenomenon.
2
Additional studies focus on the January effect and its relation to tax rules for
individual and institutional investors (Lynch, Puckett, & Yan, 2014; Poterba and
Weisbenner, 2001; Slemrod, 1982).
3
Lakonishok, Shleifer, and Vishny (1994) include current P/E ratio as one indicator of possible mispricing but their study does not examine fundamental valuation
implications of accounting earnings.
4
Hereafter, ESS.
Our model is consistent with the notions that (1) earnings change
not earnings level captures the permanent component of earnings
(Ali & Zarowin, 1992; Ohlson & Shroff, 1992); and (2) information
about future earnings is essential to valuation because it adjusts for
transitory components of current earnings. Valuation depends critically on permanent earnings (Pan, 2007), as well as the expectation
that permanent earnings will be sustained into the future. In addition, eschewing a CAPM approach avoids the uncertainties inherent
in estimating required rates of return, a feature of valuation based
on asset pricing models.5
In order to examine the association of these anomalous returns
with accounting earnings information we introduce an innovative
sample partition, forming an ex post categorization of rms that
experience the January effect (JE rms) versus those that do not
(NJE rms).6 Thus, we specically identify rms that exhibit January
effect return premiums rather than relying on some rm characteristic(s) presumed to be associated with the January effect. NJE
rms act as a kind of comparison group; but under our model and
approach, evidence of rational economic behavior in one group
does not negate or preclude rational economic behavior in the
other.
We nd that JE rms represent approximately nine percent of
all rms in our sample and range across all market caps, suggesting
that the January effect is driven by relatively few rms overall and is
frequently but not exclusively a small rm phenomenon. Our JE/NJE
partition delivers some intriguing results when implemented in our
valuation model.
For JE rms the coefcient on rst quarter contemporaneous
earnings level is signicantly negative, while the coefcients for
contemporaneous earnings growth and expectations for future
earnings growth remain signicantly positive. Although a negative earnings level coefcient may seem at rst irrational, it may
indeed be consistent with rational behavior. First and most importantly, our valuation model is more comprehensive in that it does
not rely only on current or past earnings information, but includes
all other information as captured by the construction of the analysts forecast variable. The inclusion of the term for information
about expected future earnings captures the reality that market
decisions are based in large part on expectations for the future.
Second, it is well established that price leads earnings (Ball &
Brown, 1968; Beaver, Lambert, & Morse, 1980; Beaver, Lambert,
& Ryan, 1987; Collins, Kothari, Shanken, & Sloan, 1994; DeBondt
& Thaler, 1985, 1987; Kothari, 2001). We contend that poor year
end returns followed by superior January returns foreshadow poor
rst quarter earnings followed by an earnings improvement. Our
examination of earnings for the quarters immediately preceding
and following the rst quarter, as well as a correlation analysis of
sequential quarterly earnings, both support this contention. These
results are also compatible with those of Beaver et al. (1980) and
DeBondt and Thaler (1987).7
Third, we argue that permanent earnings and their sustainability
should be quite relevant to higher return premiums in January,
a proposal in keeping with the tax-loss trading explanation for
the January effect advanced in so many prior studies. Firms are
5
Penman (2004, p. 96) reminds us that a capital asset pricing model (CAPM)
generates a required rate of return, not asset value. Further, valuation models relying
on estimated rates of return can be highly sensitive to the underlying assumptions
used in the CAPM.
6
If a rms January return premium is the highest of all 12 months of the year
then it is classied as a JE rm for that year. Otherwise, the rm is categorized as
NJE.
7
Beaver et al. (1980) demonstrate that returns are positively associated with
earnings of the following period. DeBondt and Thaler (1987) present evidence that
earnings improve in subsequent periods for loser rms. They also observe that
January and December return premiums are negatively associated.
G Model
ARTICLE IN PRESS
tax-loss sellers when their prices have fallen, i.e., they are losers,
and investors sell them off at year end to capture capital losses for
income tax purposes. If a low December stock price drives the high
January return and also foreshadows poor rst quarter earnings,
then it is natural that (increased) January return and rst quarter
earnings, however transient, should be negatively correlated. The
evidence in our added correlation analysis also indicates that the
unannounced earnings in January related to the year t 1 are positively correlated with the rst quarter earnings for JE rms, but
not for others. Thus the bad news embedded in these two earnings
numbers seems to be captured in the lowered December return of
the previous year. The idea that JE rms possibly tax-loss selling losers in the prior year offer investors optimistic expectations
for future sustainability of permanent earnings is not unreasonable if we admit the possibility that bargain hunting investors in
loser rms focus more on critical value generation capabilities in
the long run and less on current earnings numbers that may reect
transitory components. Given their weak (strong) near term (long
term) prospects these rms become candidates for tax-loss transactions in December. Then their repurchase in January increases
their price hence, the high January returns. Evidence for such a
scenario can be found in Beyer, Garcia-Feijoo, and Jensen (2013),
who show that a trading strategy targeting small, out of favor rms
achieves superior return performance in January.
It is true that without the connections between tax-loss selling and repurchasing, and the notion that poor December returns
foreshadow short term earnings troubles, the negative correlation
between January returns and rst quarter earnings may appear
at rst glance to be irrational. However, our scenario proposed
above provides a reasonable explanation for why we see evidence
of tax-loss selling intertwined with the (apparently) counterintuitive result of negative correlation between high January return and
poor rst quarter earnings performance.
Our valuation model, which anchors on both contemporaneous permanent earnings and expectations for their sustainability,
coupled with our sample partition that isolates JE rms from NJE
rms, offers the opportunity to examine whether an accounting
earnings valuation approach that does not admit arbitrage can
provide some insight into both the observed return premiums characteristic of the January effect and the tax-loss selling hypothesis
for them.
An effective JE/NJE partition should result in a considerable
proportion of JE rms also being identiable as probable tax-loss
sellers. And if January effect returns are rationally associated with
information about accounting earnings, then JE rms that are taxloss sellers could be expected to have more emphasis placed on
permanent earnings and expectations for the sustainability of permanent earnings in the future, and less emphasis on current (i.e.,
rst quarter) earnings that are likely to be poor.
Following Dalton (1993) we identify rms in our sample whose
previous end of years price performance makes them likely tax loss
sellers, and then implement our model using partitions for both taxloss selling candidacy and occurrence of the January effect. About
45% of our samples JE rms were tax-loss selling candidates at the
end of the previous year, suggesting that although tax-loss selling is
an important market dynamic in understanding the January effect,
other factors likely play in as well.
Our partitions once more deliver interesting results. The signicantly negative coefcient on earnings level appears only for
JE tax-loss sellers; the permanent earnings coefcients are signicantly positive and approximately four times larger in magnitude
than the coefcient values for either NJE tax-loss sellers or any
non-tax-loss sellers. The explanatory value of our earnings model
increases by a factor of approximately 10 when we implement the
JE/NJE partition on our tax-loss selling rms. We interpret these
results as additional evidence that the January effect anomaly is
linked to economically rational market behavior: rms that are ex

post identied as poor price performers at the end of the year (i.e.,
tax-loss selling candidates), but whose future earnings outlook is
expected to improve, are rewarded by investors.
Robustness testing for other quarters of the year provides similar results, but they are much more pronounced for January than
for other rst months of quarter (April, July, or October). Overall, we interpret this as support for the concepts that the January
effect anomaly appears to be a rational economic response to value
relevant accounting earnings information; that earnings levels do
not capture permanent earnings; and that the market focuses on
and rewards valuation implications of permanent earnings that are
expected to be sustainable into the future.
This study contributes to the literature in four ways. First, we
extend both Klein and Rosenfeld (1991) and Henker and Debapriya
(2012) by using a valuation model rather than a trading approach. In
doing so we nd evidence that the January effect is linked to fundamental valuation information represented in permanent earnings
and expectations for earnings growth. Second, we complement the
work of both Mashruwala and Mashruwala (2011) and DeBondt and
Thaler (1987) by investigating tax-loss selling rms and demonstrating that the January effect is related to information captured in
accrual accounting. Our results offer additional support for the tax
management story in a December year end tax environment such
as the US, yet are also consistent with the argument that there is
more to the January effect than just tax-loss selling (Bley & Saad,
2010; Brown et al., 1983; Corhay et al., 1987; Fountas & Segredakis,
2002; Heston & Sadka, 2010; Su, Dutta, Xu, & Ma, 2011). Third, we
extend Penman (1987) by testing the association between returns
and quarterly earnings performance without relying on an ex-post
categorization of earnings as good news or bad news.8 Finally,
partitioning our sample between rms that experience a January
effect and those that do not is an innovation that offers a more
precise inquiry into the nature of this market phenomenon.
The study proceeds as follows. Section 2 explains the development of our empirical model. Analysis and results are in Section 3.
Section 4 discusses robustness testing. Section 5 concludes.
2. Theoretical model and empirical application
A rigorous theoretical discussion of the link between price and
accounting earnings is provided in Feltham and Ohlson (1995) and
Ohlson (1995), beginning with the following assumptions:
I. The value of the rm is equal to the present value of future
expected dividends (PVED).
Pt =
R Et d t+
(1)
=1
Pt = price at date t; dt = net dividends paid at date t; R = 1 + r = the discount rate plus one; Et [.] = the expected value operator, conditioned
on information at date t.
II Clean surplus accounting. That is, change in book value is equal
to earnings less dividends:
bt = bt1 + xt dt .
(2)
Using (2) to substitute recursively for the dividend term in

(1) yields the abnormal earnings model which expresses PVED as
current book value plus capitalized abnormal earnings, dening
8
Penman (1987) appears to acknowledge a potentially distorting effect of January
returns in rst quarter data and focuses his analysis on the other three quarters of
the year.
G Model
ARTICLE IN PRESS
abnormal earnings as accounting earnings less a charge for use of

capital.
Pt = bt +
a
R Et x t+
, provided that Et b t+ /R 0
=1
as
(3)
bt = book value at date t; xt = accounting earnings during the period

t; xta xt (R 1)bt1 = abnormal earnings.
III. Appeal to a rst order autoregressive process for abnormal
earnings and information other than current abnormal earnings.
(Ohlson, 1995 refers to this as assumption [A3].)
a
= xta + t + 1,t+1
x t+1
t+1 = t + 2,t+1

t = other value relevant information; , = parameters known by
the market but unknown to researchers.
Combining (3) and [A3] yields an expression dening rm value
as a function of book value, abnormal earnings, and other information not yet captured in earnings but relevant for forecasting future
earnings:
Pt = bt + 1 xta + 2 t
(4a)
which can be expressed equivalently as

Pt = (1 k)bt + k (xt dt ) + 2 t
(4b)
1 = /(R ) 0
2 = R/(R )(R ) > 0
1 + 2 + 3 = 1
The problematic term t is thus transformed into terms that are
all readily observable and measurable. Both the second and third
terms on the right hand side are divided by the cost of capital, consistent with evidence in Ali and Zarowin (1992) and Ohlson and
Shroff (1992) that earnings changes capture permanent earnings.
f (t+1)
f (t)
Note that the term xtt+1 = xt
xt1 does not represent a forecast revision, but rather the difference between future earnings
forecasts for two consecutive periods.9 Because analysts forecast
total earnings including any transitory components of earnings
the third term provides a correction for the portion of current
earnings change that may not be permanent. Returns increase with
larger earnings changes from the prior period [second expression
in Eq. (5)]. When these changes are accompanied by an expectation that future earnings growth will be of even greater magnitude
[third expression in Eq. (5)], an additional premium
is placed
on

the realized earnings change. If the quantity xtt+1 xt is negative, i.e., earnings growth is not judged sustainable or a decline
is expected to accelerate, then current period realized earnings
change is discounted.
ESS show for both annual and quarterly time frames that this
expression of the other information variable substantially improves
the explanatory power of the returns model relative to the assumption that t is equal to zero. They also provide evidence that the
dividend terms can be ignored without sacricing explanatory
power. In addition, they show that contemporaneous measurement of returns and earnings is both theoretically and empirically
appropriate and that proper specication of the other information
variable in the returns model removes the need for ad hoc control
variables. Based on their derivations and results the basic form of
our empirical model is as follows:
k = r/(R ) 0
xt
xt xt1
Rt =
0 +
1
+
2
+
3
Pt1
Pt1
= R/r
Eq. (4b) presents a challenge to empirical researchers because
other value relevant information, t , is not observable. A common
empirical approach is to assume that t is equal to zero (Easton &
Harris, 1991; Easton, Harris, & Ohlson, 1992; Penman & Sougiannis,
1998 are three well known examples). Ohlson (2001) warns that
although this assumption is analytically convenient it may be
overly simplistic. ESS exploit Ohlsons assumption that earnings
expectations are observable in analysts forecasts to show that in
returns form, t can be captured as the difference between the
change in forecasted future earnings and contemporaneous change
in earnings:
RETt =
RETt =
1
Pt1
1 xt +
3
2 +
r
xt = xt xt1
f (t+1)
f (t)
xtt+1 = xt
xt1 = the forecast in period t for earnings per
share of period t + 1, minus the forecast in period t 1 for earnings per share of period t; Pt1 = price per share at beginning of
period t.
1 = R(1 )(1 )/(R )(R )
f (t)
xt1
(xt xt1 )
Pt1
(6)
t = time period of interest; Rt = return in period t, computed

using CRSP monthly holding returns; Pt1 = price per share
at beginning of period t; xt = earnings per share for quarf (t+1)
ter t;
0 = intercept;
1 = 1 ;
2 = 2 + r3 ;
3 = r3 ; xt
=
the latest earnings forecast for period t + 1 made during period
f (t)
t; xt1 = the latest earnings forecast for period t made during
period t 1.

3
t+1
xt +
xt xt + 3 dt + 2 dt1
r
Pt Pt1 + dt
Pt1
f (t+1)
xt

(5)
Because our focus is on the January effect, our tests ideally would
be carried out by mapping monthly returns into earnings of the
same period. Of course this is not possible because rms report
earnings information on quarterly and annual bases, not monthly.
The next best candidate is to map monthly returns to corresponding
quarterly earnings. This modication ts well with the idea that
prices lead earnings, and to the extent that earnings of February
and March (May and June, August and September, November and
December) are also included in earnings of the rst (second, third,
2 = r/(R )(R )
3 = Rr/(R )(R )
9
ESS discuss this point in detail and it is presented pictorially in their Fig. 1 (page
1131).
G Model
ARTICLE IN PRESS
Fig. 1. Timeline of events and measurement points for quarterly returns, earnings and forecasts as modeled in Eq. (7). Rqn is holding return during the quarter. m1 (m2, m3)
f (n+1)
is the rst (second, third) month of quarter n and xqn is earnings per share in quarter n. xqn
is the latest analyst forecast for quarter n + 1 earnings per share that comes out
f (n)
in quarter n (denoted by the short solid vertical marker), and is after the announcement of quarter n 1 earnings (denoted by the short dashed vertical marker). xq(n1) is the
latest analyst forecast for quarter n earnings per share that comes out in quarter n 1 after the announcement of quarter n 2 earnings.
fourth) quarter there is a bias against nding an effect for returns

of the rst month only.
There is strong evidence that rm size is negatively correlated
with the magnitude of the January effect (Blume & Stambaugh,
1983; Easterday et al., 2009; Haug & Hirschey, 2006; Hensel &
Ziemba, 2000; Keim, 1983; Lamoureux & Sanger, 1989; Reinganum,
1983), and we add rm size as a control variable.10 The general
empirical form of our quarterly data focused, expanded model thus
becomes
Rqn =
0 +
1

+
3
xqn
Pq(n1)
f (n+1)
+
2
f (t)
Pq(n1)
xq(n1)
xqn
xqn xq(n1)
Pq(n1)

+
4 Size
(7)
qn = quarter, n = 1, 2, 3, 4; Rqn = return in quarter n, computed using CRSP monthly holding returns; Pq(n1) = price
per share at the end of quarter (n 1); xqn = earnings per
share
3 =
in
3
r ;
quarter
f (n+1)
xqn
n;
0 = intercept;
Rqn,m1 =
0 +
1

+
3
xqn xq(n1)
xqn
+
2
Pq(n1)
Pq(n1)
f (n+1)
f (n)
xqn,m1 xq(n1)
xqn xq(n1)
Pq(n1)

+
4 Size
(8)
f (n+1)
xqn xq(n1)
the rst month of the current quarter, no earlier than the day of the
previous quarters earnings announcement, as follows:
1 = 1 ;
2 = 2 +
3
r ;
the latest earnings forecast for quarter n +

f (n)
1 made during quarter n; xq(n1) = the latest earnings forecast

for quarter n made during quarter n 1; Size = logarithm of
rm assets at the close of quarter n.
Although we focus on monthly rather than quarterly returns
and our addition of the rm size control variable also deviates
slightly from the ESS model, we strictly follow their efforts to
avoid uncertainty related to availability of value relevant information. We adopt the identical data measurement timeline plan as
ESS, as shown in Fig. 1 for our analyses of quarterly returns and
earnings.
However, for our analyses of monthly returns we require that
our current period latest forecast of future earnings occurs during
10
Seasonal earnings change also has been shown to capture cyclical behavior in
accounting earnings (Bathke, Lorek, & Willinger, 1989; Bernard & Thomas, 1990;
Brown & Rozeff, 1979; Foster, 1977; Wareld & Wild, 1992). ESS provide evidence
that adding seasonal change as a control variable to the model adds little explanatory
value to the model, and we obtained the same results when we ran our analyses
including seasonal change in our modied version of the ESS model. As they were
not signicant, those results are not presented for the sake of brevity.
Rqn,m1 = holding return for the rst month in quarter n; xqn,m1 =

the latest analyst forecast for quarter n + 1 earnings per share that
comes out during the rst month of quarter n and is no earlier than
the same day as the earnings announcement for quarter n 1.
All other variable are as previously dened. Fig. 2 illustrates the
timing of earnings, earnings announcements, and earnings forecasts, and measurement points in this study.
3. Analysis and results
3.1. Data sample and primary analysis
Our data sample consists of domestic rms trading ordinary
common shares on NYSE, AMEX or NASDAQ from 1991 through
2011. We impose a December year end requirement in order to simplify the alignment of calendar and scal quarter dates. Selection
parameters are:
1. Monthly holding return, price and outstanding share data available in CRSP.
2. Share price $1, to avoid very small price deators creating
extreme values in regression variables.
3. In order for a rm to be included in any year t, CRSP data must
be available for all 12 months in year t, and also for December of
year t 1.
4. Earnings per share excluding extraordinary items (EPSXQ),
report date of quarterly earnings, cash dividends per share,
quarterly revenues, and end of quarter assets available in the
Compustat quarterly database.
5. In order for a rm to be included in any rm-quarter, earnings per
share and report date of quarterly earnings for the immediately
previous quarter must also be available.
G Model
ARTICLE IN PRESS
Fig. 2. Timeline of events and measurement points for monthly returns, quarterly earnings and forecasts as modeled in Eq. (8). Rqn,m1 is holding return during the rst month
f (n+1)
of quarter n. m1 (m2, m3) is the rst (second, third) month of quarter n and xq n is earnings per share in quarter n. xqn
is the latest analyst forecast for quarter n + 1 earnings
per share that comes out in month 1 (denoted by the short solid vertical marker), and is after the announcement of quarter n 1 earnings (denoted by the short dashed
f (n)
vertical marker). xq(n1) is the latest analyst forecast for quarter n earnings per share that comes out in quarter n 1 after the announcement of quarter n 2 earnings.
6. At least two consecutive one-quarter-ahead earnings forecasts

available in I/B/E/S: the forecast for quarter n earnings per
share that was announced during quarter n 1, and the forecast announced in quarter n for earnings per share of quarter
n + 1. Individual rm-quarter observations are eliminated if the
quarter n 1 earnings report date in Compustat is later than the
announcement date from I/B/E/S of the n + 1 earnings forecast.
Stock prices and EPS are adjusted for stock splits and dividends
using the cumulative adjustment factors in CRSP and Compustat. In
order to alleviate the distorting effects of outliers we remove the top
1% of returns and prices and the top and bottom 1% of earnings and
earnings forecasts. Firms having all the necessary pricing, earnings
and forecast data for at least one quarter in any year t are included
in the sample.
Matching of data obtained from all three datasets results in
74,871 rm-quarter observations (23,716 rm-years) representing
3950 unique rms. The number of rms in each year ranges from
447 rms (1991) to 1590 (2010). Summary statistics for the sample
are presented in Table 1, Panels A and B.
Our data constraints, especially the requirement for I/B/E/S
earnings forecast data, tend to instill in our sample a tendency
toward larger, more established rms. Evidence from numerous
prior studies suggests that the magnitude of the January effect
return premium is negatively associated with rm size (Blume &
Stambaugh, 1983; Easterday et al., 2009; Haug & Hirschey, 2006;
Hensel & Ziemba, 2000; Keim, 1983; Lamoureux & Sanger, 1989;
Reinganum, 1983). In order to establish that the small rm January
effect is present in our sample, we divide all rm-years in the
sample into deciles based on beginning of year market value of
equity, then compute mean value weighted return premiums for
each month and rm size decile. Fig. 3 shows that the January
effect is present as described in prior research. Mean return premiums for January decrease monotonically and range from 7.8%
for the smallest rms to negative 0.5% for the largest rms in the
sample.11,12
11
Later in this study we examine returns for the rst month, rather than for the
entire quarter, and our regression sample size shrinks due to more restrictive data
requirements for earnings forecast data. The January effect is present in the reduced
sample also, ranging from a high of 7.6% for the smallest rms to 0.5% for the largest
rms.
12
It is important to note that we use return premium as a categorization tool only,
in order to partition our sample between JE and NJE rms. Using return premium as
Table 1a
Sample descriptive information. Panel A: Number of rms and rm-quarter observations by year and in total.
Year
Firms
Firm-quarter observations
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Total
447
483
570
735
798
881
1,025
1,144
1,118
1,014
1,135
1,179
1,242
1,312
1,418
1,504
1,500
1,516
1,530
1,590
1,575
23,716
1,230
1,316
1,494
1,913
2,208
2,485
2,839
3,302
3,243
2,491
3,397
3,805
4,050
4,412
4,709
5,052
5,132
5,270
5,401
5,567
5,555
74,871
The dataset consists of rm-quarter observations having 12 months of CRSP data

for each year t as well as for December of year t 1, and share price $1. EPS and
report date of quarterly earnings are available in Compustat for the current and prior
quarter. One-quarter-ahead earnings forecast and forecast announcement dates are
available in I/B/E/S for the current and prior quarter.
Some studies assert that most market anomalies do not survive after being made known to investors and that the January
effect has disappeared altogether (Gu, 2003, 2004; Gu & Simon,
2007). In a second examination, we followed a methodological
example included in Gu and Simons (2007) investigation into the
September phenomenon.13 Correspondingly, we hypothesized that
a dependent variable in our regression model would be inappropriate because the

model relies upon raw returns, as demonstrated in our Eq. (5) and in Appendix B of
ESS.
13
Gu and Simon (2007, page 292) argue the following: If all months had an equal
likelihood of being the worst performing month of the year over our sample period,
September would be the worst about one-twelfth of the time, or 8.3% of the time.
They proceed to show that in their sample, September was the worst month between
10.7% and 17.1% of the time.
G Model
ARTICLE IN PRESS
Table 1b
Sample descriptive information. Panel B: Descriptive statistics for quarterly earnings levels and changes, and rm size in the entire sample.
N
Mean
Quarter 1 EPS
Quarter 2 EPS
Quarter 3 EPS
Quarter 4 EPS
Quarter 1 change in EPS
16,046
19,672
20,622
18,531
16,046
19,672
20,622
18,531
Total assets
Total revenues
Market value of equity
23,716
23,716
23,716
Std. dev.
0.226
0.247
0.241
0.181
0.036
0.037
0.000
0.073
5,927.7
3,202.0
3,943.3
0.406
0.463
0.491
0.643
0.496
0.384
0.405
0.587
32,927.8
12,028.1
16,348.2
Median
0.180
0.200
0.200
0.190
0.000
0.027
0.010
0.000
645.9
591.8
688.3
mean value-weighted return premium
Quarter n EPS is earnings per share excluding extraordinary items (EPSXQ in the Compustat quarterly dataset), adjusted for effects of stock splits and dividends. Market value
of equity is measured at the beginning of the year. Total assets, Total revenues and Market value of equity are in $millions. Descriptive statistics are cross-sectional averages
over the years 19912011.
0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
0.010
0.000
-0.010
-0.020
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
sm-1
4
5
6
7
decile of firm size
lg-10
Oct
Nov
Fig. 3. Mean monthly return premiums for the sample, 19912011. Regression sample rm-years are divided into deciles based on beginning-of-year market value of
equity. Return premium = CRSP monthly holding return value-weighted market
return. There are 23,716 rm-years representing 3,950 unique rms.
if all months have an equal likelihood of being a rms best performing month, then JE rms should make up approximately one
twelfth (8.3%) of our sample.
We conducted chi-square tests to evaluate whether the number
of JE rms in our sample is within the expected range. Inability to
reject the null hypothesis would indicate that JE rms are no more
frequent than other month effect rms, and suggest that there is
nothing special about January.14 We examined our entire sample.
Then we separated the rms by size using market value of equity,
categorizing them as small, medium, or large, and assessed each
size category. The results of our chi-square analysis are as follows:
1. For the entire sample, we reject the null hypothesis that JE rms
are no more frequent than other month effect rms, with a
probability of < 0.0001. For our time period of study, 9.1% of
all rms in our sample are classied as JE rms, more often than
chance would permit.
2. For small rms, we reject the null hypothesis with a probability
of < 0.0001. Of the small rms in our sample, 10.7% are classied
as JE rms, more often than chance would permit.
3. For medium (large) rms, 8.0% (8.4%) are classied as JE rms,
but we cannot reject the null hypothesis for either subset.
14
We are grateful to an anonymous reviewer for suggesting this opportunity to
present additional evidence of the January effects relevance as a topic worthy of
study.
Following ESS we address both potential cross-correlation

between rms in each year and serial correlation across years
by using the regression methodology recommended in Gow,
Ormazabal, and Taylor (2010). We report our regression results
using two-way cluster robust standard errors, clustering by time
and by rm.
Table 2 shows the results of regressing full quarterly returns
both pooled and for individual quarters on quarterly earnings
level, earnings change, and other value relevant information as
shown in Eq. (7). The ESS model, as well as a more restricted form
of it that uses only contemporaneous earnings level and earnings
change, are also analyzed and included in Table 2 for comparison
purposes.
The ESS model improves explanatory power substantially over
earnings levels and changes only: increases in adjusted R2 range
from 137.5% for the fourth quarter to 275.0% for the rst quarter
and 360% for the pooled sample. Adding the control variable size
provides marginal increases in adjusted R2 over the ESS model for
quarterly returns and earnings data. Adjusted R2 values for these
regressions remain generally low even after such improvement, in
the range of 2.13.8%. Although sample size and time period of the
current study differ somewhat from ESS, our coefcient estimates
and adjusted R2 values for contemporaneous quarterly returns and
earnings variables are similar to their results.
In testing the association of rst month returns with quarterly
earnings variables we ensure that current period analysts forecasts
of next quarter earnings on which our expression of the expected
sustainability of permanent earning relies are available during
the return month. We use the latest earnings forecast for quarter
n + 1 that is announced during the rst month of quarter n as the
f (n+1)
value of xqn
shown in Eq. (8). We eliminate observations for
which no such rst month analyst forecast is available, or for which
the only available forecast occurs prior to the announcement of the
previous quarters earnings.15 Our reduced sample contains 39,601
observations distributed across all four quarters as shown in the
second column of Table 3, and our results using only rst month
returns are still consistent with those of ESS.
Coefcients on
1 are positive but not signicantly so.
Coefcients on earnings change,
2 , and the value relevant information variable,
3 , are reliably positive and signicant across all
individual quarters as well as for the pooled sample. The rm size
coefcient
4 is statistically signicant only in the rst quarter.
Again consistent with ESS, our results here suggest that information about permanent earnings growth is important in explaining
15
We invite the reader to refer again to Fig. 2.
G Model
ARTICLE IN PRESS
Table 2
Results of regressions of quarterly returns on price-deated earnings level, earnings change, and other value relevant information. The ESS model, as well as a more restricted
form of it that uses only contemporaneous earnings level and earnings change, are also included for comparison purposes.
Rqn =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
xqn
f (t)
xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(7)
N
Pooled
74,871
Quarter 1
16,046
Quarter 2
19,672
Quarter 3
20,622
Quarter 4
*
**
***
Rqn
18,531
0.034*
0.037*
0.057*
0.008
0.020
0.055
0.064
0.060
0.082
0.007
0.006
0.002
0.073*
0.076*
0.105
0.255
0.215
0.224
0.563**
0.537***
0.585***
0.305
0.241
0.226
0.518
0.429
0.433
0.061
0.198
0.209
0.166
2.532***
2.257***
0.050
2.642***
2.652***
0.419***
2.498***
2.492***
0.062
2.316***
2.311***
0.047
3.067*
3.076*
2.470***
2.468***
0.003
***
2.676
2.685***
0.005
2.155***
2.153***
0.003
***
2.402
2.399***
0.001
2.907*
2.923*
0.004
Adj. R2
0.005
0.023
0.023
0.008
0.030
0.032
0.007
0.021
0.021
0.008
0.024
0.024
0.016
0.038
0.038
Signicant at < 0.10.

f (n)
is holding period return for quarter n, Pq(n1) is beginning of period price, xqn is earnings per share in quarter n, xq(n1) is the latest analyst forecast that comes out in
quarter n1 for quarter n earnings per share. Earnings and prices are adjusted for stock splits and dividends. Regressions utilize two-way cluster robust standard errors.
Table 3
Results of regressions of returns for the rst month of each quarter on price-deated earnings level, earnings change, other value relevant information, and rm size.
Rqn,m1 =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
f (n)
xqn,m1 xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(8)
N
Pooled
Quarter 1
Quarter 2
Quarter 3
Quarter 4
39,601
4,744
11,598
12,657
10,602
2
*
0.017
0.067*
0.032
0.032
0.036
0.141
0.194
0.088
0.081
0.222
3
***
1.802
1.915*
1.425***
1.608**
2.297*
***
1.719
1.998**
1.272**
1.635**
2.217*
Adj. R2
0.000
0.008*
0.001
0.004
0.002
0.021
0.027
0.011
0.020
0.037

f (n)
Rqn,m1 is holding period return for the rst month in quarter n, Pq(n1) is beginning of period price, xqn is earnings per share in quarter n, xq(n1) is the latest analyst forecast
**
***
f (n+1)
that comes out in quarter n 1 for quarter n earnings per share, xqn,m1 is the latest analyst forecast that comes out in the rst month of quarter n for quarter n + 1 earnings per
share. Size is measured as the logarithm of total assets. Earnings and prices are adjusted for stock splits and dividends. Regressions utilize two-way cluster robust standard
errors.
contemporaneous returns, even when limited to the rst month of

the quarter.
3.2. JE rms versus NJE rms

In order to more closely examine whether the January effect is
associated with accounting earnings information, it is important
to identify rms whose returns are representative of this market
anomaly. Rather than apply an ad hoc return premium variable to
our regression that is unsupported by the economic intuition of
the ESS model, we partition the sample into rms that experience
the January effect (JE rms) and those that do not (NJE rms); the
NJE rms act as a comparison sample. Our partition method proceeds as follows: for each rm observation in year t, we compute its
monthly return premiums by subtracting the value weighted market return from the CRSP monthly holding return. If the January
return premium is the highest of all 12 months in year t then
the rm is categorized as a JE rm for that year; otherwise it is
categorized as NJE.16 Table 4, Panel A shows the distribution of JE

and NJE rms in the sample by year and overall. Over the entire
time period of study 9.1% of our sample rms are classied as JE
rms; the proportion of JE rms varies from 4.4% in 2000 to 22.2%
in 1992.17 This is consistent with previously cited evidence that the
January effect is more pronounced in some years than in others.
Table 4 Panels B and C show summary statistics for the partitioned sample. Unsurprisingly, rms that experience January effect
returns are on average smaller in terms of assets, annual revenues,
and market value of equity but there is little discernible difference in price-deated earnings or change in earnings between JE
and NJE. The mean January return premium for JE rms is 22.6%,
versus negative 1.6% for the much larger population of NJE rms.
16
If January is tied with any other month(s) for highest premium, the rm is
classied as JE for that year.
17
We conducted a similar analysis, not presented here for the sake of brevity, of
the 93,112 rms that met our CRSP selection criteria prior to being matched with
Compustat or I/B/E/S data. JE rms made up 10.8% of that CRSP dataset.
G Model
ARTICLE IN PRESS
Table 4a
Descriptive information for the partitioned sample. Panel A: Distribution of JE and NJE rms by year and overall.
Year
Firms
JE rms
NJE rms
% JE rms
% NJE rms
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Total
447
483
570
735
798
881
1,025
1,144
1,118
1,014
1,135
1,179
1,242
1,312
1,418
1,504
1,500
1,516
1,530
1,590
1,575
23,716
81
107
67
73
60
64
78
64
75
45
191
83
89
155
90
265
123
71
164
119
106
2,170
366
376
503
662
738
817
947
1,080
1,043
969
944
1,096
1,153
1,157
1,328
1,239
1,377
1,445
1,366
1,471
1,469
21,546
18.1%
22.2%
11.8%
9.9%
7.5%
7.3%
7.6%
5.6%
6.7%
4.4%
16.8%
7.0%
7.2%
11.8%
6.3%
17.6%
8.2%
4.7%
10.7%
7.5%
6.7%
9.1%
81.9%
77.8%
88.2%
90.1%
92.5%
92.7%
92.4%
94.4%
93.3%
95.6%
83.2%
93.0%
92.8%
88.2%
93.7%
82.4%
91.8%
95.3%
89.3%
92.5%
93.3%
90.9%
JE rms are those for which the January return premium is higher than the return premium of any other month in the calendar year. NJE rms are those for which the return
premium of at least one other month is higher than the return premium for January in the calendar year. Return premium = CRSP monthly holding return value weighted
market return.
Table 4b
Descriptive information for the partitioned sample. Panel B: Descriptive statistics for earnings levels, earnings changes, and rm size (JE rms).
January return premium

Quarter 1 EPS
Quarter 2 EPS
Quarter 3 EPS
Quarter 4 EPS
Total assets
Total revenues
Mean
Std. dev.
Median
1,423
1,423
1,777
1,886
1,676
1,423
1,777
1,886
1,676
2,170
2,170
2,170
0.226
0.225
0.232
0.209
0.186
0.015
0.023
0.013
0.042
4,834.6
3,156.4
3,317.2
0.179
0.449
0.541
0.535
0.630
0.463
0.371
0.375
0.502
24,537.6
13,254.1
13,715.7
0.180
0.173
0.178
0.160
0.160
0.000
0.025
0.000
0.000
505.2
506.1
549.3
market return. Quarter n EPS is earnings per share excluding extraordinary items (EPSXQ in the Compustat quarterly dataset) deated by beginning of quarter share price
obtained from CRSP, and adjusted for effects of stock splits and dividends. Market value of equity is measured at the beginning of the year. Total assets, Total revenues and
Market value of equity are in $millions. Descriptive statistics are cross-sectional averages over the years 19912011.
Table 4c
Descriptive information for the partitioned sample. Panel C: Descriptive statistics for earnings levels, earnings changes, and rm size (NJE rms).
January return premium

Quarter 1 EPS
Quarter 2 EPS
Quarter 3 EPS
Quarter 4 EPS
Total assets
Total revenues
Mean
Std. dev.
Median
14,623
14,623
17,895
18,736
16,855
14,623
17,895
18,736
16,855
21,546
21,546
21,546
0.016
0.226
0.249
0.244
0.180
0.038
0.038
0.001
0.077
6,038.0
3,206.6
4,014.2
0.108
0.402
0.454
0.486
0.644
0.499
0.386
0.408
0.594
33,656.5
11,897.7
16,250.4
0.016
0.180
0.204
0.205
0.195
0.000
0.027
0.010
0.000
658.484
602.341
709.853
market return. Quarter n EPS is earnings per share excluding extraordinary items (EPSXQ in the Compustat quarterly dataset) deated by beginning of quarter share price
obtained from CRSP, and adjusted for effects of stock splits and dividends. Market value of equity is measured at the beginning of the year. Total assets, Total revenues and
Market value of equity are in $millions. Descriptive statistics are cross-sectional averages over the years 19912011.
G Model
10
ARTICLE IN PRESS
We also evaluate whether a rms experience of January effect

returns results in a halo effect on the returns-earnings association
for the entire quarter. In Table 5, Panel B we examine the association between quarterly holding returns and the earnings variables
on the partitioned sample. The extraordinary explanatory value of
the model for quarterly returns of JE rms is much lower than
for January returns alone, dropping by 81.9%. Firm size remains
negatively signicant for JE rms only. Based on this evidence we
surmise that differentiating between JE and NJE rms is important
when considering January effect returns but its usefulness does not
extend beyond our month of interest.
3.3. Tax-loss selling

Fig. 4. Distribution of JE and NJE by rm size across all years in the sample,
19912011. JE rms are those for which the return in January return premium is
higher than the return premium of any other month in the calendar year. NJE rms
are those for which the return premium of at least one other month is higher than the
return premium for January in the calendar year. Premium = CRSP monthly holding
return value weighted market return. For each year in the sample (19912011),
every rm was ranked according to assets ($millions) into one of three categories:
small rms, medium rms or large rms. The total number of rms in each rm size
category was then divided between JE and NJE rms to produce the graph shown.
Intriguingly, this data also suggests that the January effect phenomenon is driven primarily by the returns of relatively few rms.
The negative correlation between rm size and the magnitude
of January returns could lead to concern that our JE rms all come
from the bottom of our sample, i.e., it is only the smallest rms in
our sample that are categorized as JE rms. To allay this concern,
for each year in our time period of study we divided the rms into
three categories (small rms, medium rms or large rms) based
on their asset size. We then computed the percentage of JE vs. NJE
rms that appeared in each size category across all years. Results
are shown graphically in Fig. 4.
At 10.8%, small rms do make up the highest proportion of JE
rms. However, the proportions of JE rms in the medium and large
rm categories are only slightly lower, at 9.4% and 9.5%, respectively. We interpret this as evidence that although the magnitude
of the January effect may be negatively correlated with rm size,
the phenomenon is not experienced exclusively by small rms.
In addition, our earnings forecast data requirement already tilts
our sample toward larger rms in the market. Even small rms in
our sample are larger than small rms when considering a less
restricted sample that may be more representative of the entire
market, and this biases against our success in nding results.
Regression results for Eq. (8) on the partitioned sample are
shown in Table 5, Panel A. JE rms comprise 11.3% of the usable rst
quarter observations. The differences are striking between adjusted
R2 values for JE rms and NJE rms. Explanatory power is more than
eleven times greater for JE rms than for NJE rms. Results from
Table 3 for Quarter 1, which is the same regression on the unpartitioned sample, are shown for comparison. Adjusted R2 for JE rms
increases more than ten times, from 0.027 to 0.293.
Table 5 Panel A also shows that January returns are negatively
(positively) and signicantly associated with earnings level for JE
(NJE) rms, compared to an insignicant association in the unpartitioned Quarter 1 observations. All coefcients on
2 are positive
and signicant; on
3 they are positive but signicance is slightly
weaker for JE rms.18 Controlling for rm size has strong significance for JE rms, but not for NJE rms, an unsurprising result
given copious evidence that the magnitude of the January effect is
inversely related to rm size.
18
The P-value on
3 for JE rms in Table 5, Panel A is 0.053.
The appearance of a signicantly negative coefcient on earnings level for JE rms in Table 5, Panel A is somewhat unexpected
because it suggests that returns for these rms are decreasing in
earnings levels. In view of evidence in our Tables 2 and 319 that
coefcients on quarterly earnings levels are positive or insignicantly different from zero, the contrast for JE rms is striking. As
discussed earlier in this paper, there are potential explanations for
this, one of which is that the January effect is at least in part related
to tax-loss selling. Tax-loss selling rms are generally poorly performing losers. If low stock prices of tax-loss selling rms lead to
high January returns, then it is reasonable that January return and
rst quarter earnings of such losers should be negatively correlated.
Bargain hunting investors at the turn of the year may be willing to
overlook poor performance in the short term and focus on permanent earnings and expectations for the future. Similar to Dalton
(1993), we categorize rms that exhibit negative holding returns
for December of year t 1 as potential tax-loss sellers (PTLS).20
All other rms are classied as non-tax loss sellers (NTLS). The
mechanics of our classication scheme are such that the same rm
may be classied as a tax-loss seller in one year and as a non-taxloss seller in a different year.
Table 6 Panel A shows the distribution of NTLS and PTLS by year
and overall. Cross-sectionally, the distribution of JE/NJE rms by
tax-loss selling status is slightly different from the distribution in
the general population of rms: 11.2% (9.1%) of PTLS (all) rms are
also JE rms and only 8.2% of NTLS are identied as JE rms.
Table 6 Panel B shows the distribution of NTLS and PTLS rms,
after partitioning according to JE/NJE status. Overall, 44.7% of JE
rms are also PTLS, and there are ve years in which the proportion
is in excess of 60%. In our sample there is only one year (2000) in
which the proportion of NJE PTLS rms exceeds the proportion of
JE PTLS rms. In short, our sample indicates that potential tax-loss
sellers in December of year t 1 may be either JE or NJE rms in
year t, but JE rms are a higher proportion of PTLS rms than are
NJE rms (Table 6 Panel A). Although a substantial proportion of JE
rms are also identied as PTLS in this sample, more JE rms are
NTLS than are PTLS.
Table 6 Panel C shows descriptive statistics for PTLS and NTLS
rms. First quarter earnings of PTLS rms are 21% lower than those
of NTLS rms, consistent with the logic that losers from year t 1
are not likely to be strong performers in the rst quarter of year
19
And also in ESS.

In untabulated results we varied the start of the tax-loss selling categorization
period, identifying PTLS rms over 3-month, 6-month and 11-month return periods
for robustness. Similar to Fant and Peterson (1995), in whose analysis January was
partitioned from the rest of the year due to the seasonal pattern of January returns,
we measured returns in year t 1 over 11 months (February through December)
rather than 12 months (January through December) in order to avoid possible distorting effects resulting from the inclusion of year t 1s JE rms returns. The results
of these analyses were qualitatively similar.
20
G Model
ARTICLE IN PRESS
11
Table 5a
Regressions on the partitioned sample. Panel A: Results of regressing January returns on price deated rst quarter earnings level, earnings change, other value relevant
information, and rm size.
Rqn,m1 =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
f (n)
xqn,m1 xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(8)
All rms
JE rms
NJE rms
Adj. R2
4,744
538
4,206
0.067*
0.530***
0.000
0.194
2.434**
0.421*
1.915*
3.683*
1.339***
1.998**
2.836
1.409***
0.008*
0.039***
0.003
0.027
0.293
0.026

**
***
premium of at least one other month is higher than the return premium for January in the calendar year. Return premium = CRSP monthly holding returnvalue weighted
f (n)
market return. Rqn,m1 is holding period return for the rst month in quarter n, Pq(n-1) is beginning of period price, xqn is earnings per share in quarter n, xq(n1) is the latest
*
f (n+1)
analyst forecast that comes out in quarter n-1 for quarter n earnings per share, xqn,m1 is the latest analyst forecast that comes out in the rst month of quarter n for quarter
n + 1 earnings per share. Size is measured as the logarithm of total assets. Earnings and prices are adjusted for stock splits and dividends. Regressions utilize two-way cluster
robust standard errors.
Table 5b
Regressions on the partitioned sample. Panel B: Results of regressing rst quarter returns on price-deated earnings level, earnings change, other value relevant information,
and rm size after partitioning the sample into JE and NJE rms. Results from the unpartitioned sample of rst quarter observations shown in Table 2 are presented in the
third row for comparison.
Rqn =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
xqn
f (t)
xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(7)
N
All rms
JE rms
NJE rms
15,819
1,377
14,442
Adj. R2
0.055
0.347***
0.012
0.585***
0.299
0.672***
2.652***
2.529*
2.669***
2.685***
2.288*
2.707***
0.005
0.024**
0.002
0.032
0.053
0.038

***
premium of at least one other month is higher than the return premium for January in the calendar year. Return premium = CRSP monthly holding returnvalue weighted
f (n)
market return. Rqn,m1 is holding period return for the rst month in quarter n, Pq(n1) is beginning of period price, xqn is earnings per share in quarter n, xq(n1) is the latest
**
f (n+1)
analyst forecast that comes out in quarter n-1 for quarter n earnings per share, xqn.m1 is the latest analyst forecast that comes out in the rst month of quarter n for quarter
n + 1 earnings per share. Size is measured as the logarithm of total assets. Earnings and prices are adjusted for stock splits and dividends. Regressions utilize two-way cluster
robust standard errors.
Table 6a
Potential tax-loss selling and non-tax-loss selling rms. Panel A: Distribution of PTLS and NTLS in the full sample and for JE rms.
Year
PTLS
NTLS
% PTLS
% NTLS
%PTLS and JE
%NTLS and JE
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Total
52
43
120
122
177
245
305
377
303
213
242
226
620
319
249
437
378
489
305
136
111
5,469
208
292
187
274
296
317
332
387
479
374
523
676
319
642
641
452
448
420
588
710
682
9,247
20.0%
12.8%
39.1%
30.8%
37.4%
43.6%
47.9%
49.3%
38.7%
36.3%
31.6%
25.1%
66.0%
33.2%
28.0%
49.2%
45.8%
53.8%
34.2%
16.1%
14.0%
37.2%
80.0%
87.2%
60.9%
69.2%
62.6%
56.4%
52.1%
50.7%
61.3%
63.7%
68.4%
74.9%
34.0%
66.8%
72.0%
50.8%
54.2%
46.2%
65.8%
83.9%
86.0%
62.8%
19.2%
30.2%
12.5%
13.1%
9.0%
9.4%
9.5%
6.9%
8.6%
3.3%
29.8%
8.4%
7.4%
15.7%
6.4%
21.1%
11.6%
7.0%
12.1%
8.8%
7.2%
11.2%
15.9%
21.9%
9.6%
8.4%
6.1%
5.7%
6.3%
3.4%
5.4%
4.5%
9.0%
7.0%
6.0%
9.8%
6.2%
17.3%
6.0%
4.5%
11.1%
8.0%
6.5%
8.2%
Firms represented are those for which 12 months of CRSP holding returns are available for year t 1. Firms with negative holding returns for December of year t 1 are
categorized as potential tax-loss-sellers (PTLS). All other rms are classied as non-tax-loss sellers (NTLS). JE rms are those for which the January return premium is
higher than the return premium of any other month in the calendar year. Return premium = CRSP monthly holding return value weighted market return.
G Model
ARTICLE IN PRESS
12
Table 6b
Potential tax-loss selling and non-tax-loss selling rms. Panel B: Distribution of PTLS and NTLS for both JE and NJE rms, by year and overall.
Year
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
Total
JE rms
43
77
33
39
34
41
50
39
52
24
119
66
65
113
56
170
71
53
102
69
52
1,368
% PTLS
% NTLS
23.3%
16.9%
45.5%
41.0%
47.1%
56.1%
58.0%
66.7%
50.0%
29.2%
60.5%
28.8%
70.8%
44.2%
28.6%
54.1%
62.0%
64.2%
36.3%
17.4%
15.4%
44.7%
76.7%
83.1%
54.5%
59.0%
52.9%
43.9%
42.0%
33.3%
50.0%
70.8%
39.5%
71.2%
29.2%
55.8%
71.4%
45.9%
38.0%
35.8%
63.7%
82.6%
84.6%
55.3%
NJE rms
% PTLS
% NTLS
217
258
274
357
439
521
587
725
730
563
646
836
874
848
834
719
755
856
791
777
741
13,348
19.4%
11.6%
38.3%
29.7%
36.7%
42.6%
47.0%
48.4%
37.9%
36.6%
26.3%
24.8%
65.7%
31.7%
27.9%
48.0%
44.2%
53.2%
33.9%
16.0%
13.9%
36.4%
80.6%
88.4%
61.7%
70.3%
63.3%
57.4%
53.0%
51.6%
62.1%
63.4%
73.7%
75.2%
34.3%
68.3%
72.1%
52.0%
55.8%
46.8%
66.1%
84.0%
86.1%
63.6%
market return. Firms represented are those for which 12 months of CRSP holding returns are available for year t 1. Firms are categorized as potential tax-loss sellers
(PTLS) if they have negative holding returns for December of year t 1. All other rms are classied as not tax-loss sellers (NTLS).
t. PTLS rms are also on average smaller and have lower revenues
than NTLS rms.
We perform the regression analysis in Eq. (8) on rms having sufcient available data, examining the association between
January returns and rst quarter earnings information for all categories of tax-loss seller status and JE/NJE experience. Results are
shown in Table 7.
For both tax-loss sellers and non-tax-loss sellers overall the
model performs as expected: the coefcient on earnings level is
insignicantly different from zero, coefcients related to permanent earnings performance are signicantly positive, explanatory
value does not exceed 4.3%. However, the JE/NJE partition once
more delivers interesting results. The large and signicantly negative coefcient on earnings level again appears, but only for JE
tax-loss sellers. For all other categories it is insignicantly different
from zero or slightly positive. Permanent earnings coefcients for JE
PTLS are signicantly positive and approximately four times larger
in magnitude than the coefcient values for either NJE PTLS or any
Table 6c
Potential tax-loss selling and non-tax-loss selling rms. Panel C: Descriptive statistics for PTLS and NTLS.
N
PTLS
Quarter 1 EPS
Quarter 2 EPS
Quarter 3 EPS
Quarter 4 EPS
Mean
Std. dev.
Median
1,781
4,043
4,356
3,616
0.253
0.257
0.233
0.182
0.440
0.460
0.533
0.692
0.180
0.220
0.207
0.200

1,781
4,043
4,356
3,616
0.003
0.035
0.020
0.069
0.460
0.364
0.436
0.623
0.010
0.030
0.010
0.005
Total assets
Total revenues
5,469
5,469
5,469
6765.2
3890.1
5001.0
35,367.4
14,434.3
20,252.7
NTLS
Quarter 1 EPS
Quarter 2 EPS
Quarter 3 EPS
Quarter 4 EPS
Total assets
Total revenues
2,963
6,796
7,400
6,195
2,963
6,796
7,400
6,195
9,247
9,247
9,247
0.321
0.326
0.321
0.269
0.023
0.043
0.006
0.064
7,414.0
4,233.9
5,278.0
0.433
0.455
0.466
0.591
0.417
0.345
0.346
0.535
37,506.6
14,055.1
18,610.2
765.2
749.7
827.1
0.250
0.270
0.270
0.250
0.000
0.030
0.010
0.000
1,020.1
983.6
1,117.0
Firms are categorized as potential tax-loss sellers (PTLS) if they have negative holding returns for the month of December in year t-1. All other rms are classied as not
tax-loss sellers (NTLS). Quarter n EPS is earnings per share excluding extraordinary items (EPSXQ in the Compustat quarterly dataset) deated by beginning of quarter share
price obtained from CRSP, and adjusted for effects of stock splits and dividends. Market value of equity is measured at the beginning of the year. Total assets, Total revenues
and Market value of equity are in $millions. Descriptive statistics are cross-sectional averages over the years 19912011.
G Model
ARTICLE IN PRESS
13
Table 7
Results of regressing January returns on price-deated rst-quarter earnings level, earnings change, other value relevant information, and rm size, after partitioning the
sample as to tax loss-selling status and JE/NJE rms, 19912011.
Rqn,m1 =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
f (n)
xqn,m1 xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(8)
All PTLS
JE PTLS
NJE PTLS
All NTLS
JE NTLS
NJE NTLS
Adj. R2
1,781
233
1,548
2,963
305
2,658
0.113*
0.547***
0.020
0.035
0.456***
0.014
0.157
4.131***
0.432*
0.422
0.913
0.492*
2.609*
5.753***
1.379*
1.442*
1.005
1.344**
2.786*
3.915*
1.538*
1.433*
0.799
1.326*
0.013*
0.038***
0.005
0.005
0.033***
0.002
0.043
0.395
0.028
0.020
0.203
0.024

***
market return. Firms are categorized as potential tax-loss sellers (PTLS) if they have negative holding returns for the month of December in year t 1. All other rms are
classied as not tax-loss sellers (NTLS). Rqn,m1 is holding period return for the rst month in quarter n, in this case January returns. Pq(n1) is beginning of period price, xqn
**
f (n)
f (n+1)
is earnings per share in quarter n, xq(n1) is the latest analyst forecast that comes out in quarter n1 for quarter n earnings per share, xqn,m1 is the latest analyst forecast that
comes out in the rst month of quarter n for quarter n + 1 earnings per share. Size is measured as the logarithm of total assets. Earnings and prices are adjusted for stock splits
and dividends. Regressions utilize two-way cluster robust standard errors.
NTLS. The explanatory value of our earnings model increases by a

factor of about ten for the JE rms in both the PTLS and NTLS groups;
the JE PTLS explanatory value is nearly twice that of JE NTLS. Further, none of the earnings variables are signicantly different from
zero for the JE NTLS. Together with its adjusted R2 value, this suggests that in the case of JE NTLS rms the model does a good job of
explaining the variability in January effect returns but we cannot
sort out which, if any, variable(s) matter(s), apart from rm size.
In summary, consistent with previous studies suggesting that
tax-loss selling is at least a partial explanation for January return
premiums, we nd a sizable, though not complete, correspondence
between JE rms and those we categorize as potential tax-loss sellers. This suggests that our categorization scheme is reasonable.
Further, complementing the evidence of Henker and Debapriya
(2012), our partitioning methodology teases out evidence that
January effect investors behavior is economically rational: exceptionally high January returns on PTLS loser rms may occur as
investors discount comparatively poor current earnings performance and give extra weight to permanent earnings and their
sustainability in the future. Finally, the existence of JE rms that
are not tax-loss sellers and our inability to determine signicant
explanatory variables for them suggest that there are still some
aspects of this market puzzle that are not well understood, and
that invite future investigation.
4. Robustness testing
4.1. Expectations and EPS performance of JE rms
As a reasonableness check on our ndings, we investigate
how expectations and actual performance change over the time
period from the fourth quarter of year (t 1) through the rst
and second quarters of year (t). Table 8 presents mean quarterly
for permanent
earnings sustainability
earnings and expectations

f (n+1)
xqn
f (n)
xq(n1)
xqn xq(n1)
21
for the fourth through
second quarters for JE rms. The outlook for permanent earnings

is very negative in the fourth quarter for JE rms and then highly
21
This is the other information term described previously. It is the difference
between the change in forecasted future earnings and contemporaneous change in
earnings, and it captures expectations for the sustainability of earnings growth.
Table 8
Mean quarterly earnings per share and expectations for permanent earnings sustainability (JE rms).
N = 1,781
Quarter 4 EPS (year t 1)
Quarter 1 EPS
Quarter 2 EPS
Quarter 4 expectation for permanent earnings sustainability
Quarter 1 expectation for permanent earnings sustainability
0.391***
0.340***
0.398***
0.062**
0.089***
**

JE rms are those for which the January return premium is higher
than the return premium of any other month
in the calendar
year.
***
f (n+1)
The expectation for permanent earnings sustainability = xqn
xqn xq(n1) , where xqn is earnings per share in quarter n,
f (n)
xq(n1)
f (n)
xq(n1)
is the latest
analyst forecast that comes out in quarter n-1 for quarter n earnings per share, and
f (n+1)
xqn
is the latest analyst forecast for quarter n + 1 earnings per share that comes
out in quarter n.
positive, and this is borne out in the behavior of mean earnings over
the three quarters it drops, and then recovers.
We also examine correlations between the earnings of individual quarters. If our story is feasible and investors in JE rms
are willing to overlook poor earnings in the current (rst) quarter because they expect good performance in the second, then we
should expect a negative correlation between rst and second quarter earnings of JE rms. On the other hand, rst quarter earnings
should be positively correlated with fourth quarter earnings of the
previous year as poor earnings performance drags into the rst
quarter before things improve. We employ Spearman correlations,
as they are not dependent upon assumptions of the datas normal
distribution or heteroskedasticity.
Results are shown in Table 9. Panel A shows that for JE rms in
general, rst quarter earnings in the January effect year are positively correlated with earnings of the previous quarter (the fourth
quarter of year t 1), as expected. The signicant correlation coefcient of 0.0389 between rst and second quarter earnings for JE
rms is consistent with the anticipated future earnings turnaround.
In contrast, for NJE rms the correlation between the rst and
fourth quarters is insignicantly different from zero while the
rst quarter-second quarter correlation is a signicantly positive
0.0344.
G Model
ARTICLE IN PRESS

14
Table 9a
Correlations of quarterly earnings. Panel A: JE versus NJE rms, 19912011.
NJE rms
Quarter 4 (t 1)
Quarter 1 (t)
Quarter 2 (t)
Quarter 3 (t)
JE rms
Quarter 4 (t1)
Quarter 1 (t)
1.000
0.0057
0.0457*
1.000
0.0344***
Quarter 2 (t)
Quarter 3 (t)
0.0389*
1.000
0.0508***
0.0366***
1.000
JE rms are those for which the January return premium is higher than the return premium of any other month in the calendar year. NJE rms are those for which the
return premium of at least one other month is higher than the return premium for January in the calendar year. Return premium = CRSP monthly holding return value
weighted market return. The year in which the comparison quarter falls is designated as t. N = 3,909 for JE rms. N = 35,645 for NJE rms. Correlations presented are Spearman
correlations, which do not depend upon assumptions of the datas normal distribution or heteroskedasticity.
*
***
Table 9b
Correlations of quarterly earnings. Panel B: JE versus NJE rms, 19912011 (PTLS).
NJE rms
Quarter 4 (t 1)
Quarter 1 (t)
Quarter 2 (t)
Quarter 3 (t)
JE rms
Quarter 4 (t 1)
Quarter 1 (t)
1.000
0.0229*
0.0650*
1.000
0.0326**
Quarter 2 (t)
Quarter 3 (t)
0.1127***
1.000
0.0548***
0.0474
1.000

market return. Firms are categorized as potential tax-loss sellers (PTLS) if they have negative holding returns for the month of December in year t 1. All other rms
are classied as not tax-loss sellers (NTLS). The year in which the comparison quarter falls is designated as t. N = 1,607 for JE rms. N = 12,171 for NJE rms. Correlations
presented are Spearman correlations, which do not depend upon assumptions of the datas normal distribution or heteroskedasticity.
**
***
In Panels B and C, we partition the sample along the dual dimensions of tax-loss selling status and JE versus NJE rms. For JE PTLS
rms the earnings correlation coefcients are of the predicted signs,
and stronger. On the other hand, Panel C shows that the rst quarter (t)-fourth quarter (t 1), as well as the second quarter (t)-rst
quarter (t) correlation coefcients are both insignicantly different
from zero for JE NTLS rms. NJE rms in Panels B and Panel C are
positively signicant for the rst-second quarter earnings correlations, but the correlation between the rst and fourth quarters for
PTLS (NTLS) rms is signicantly negative (insignicantly different
from zero). Earnings correlations between quarters two and three
are positively signicant in all cases except NJE PTLS, where the
correlation is insignicantly different from zero.
Taken together, we interpret these results as evidence supporting our argument that in the case of JE rms, investors seem to
rationally discount relatively poor earnings performance in the rst
quarter and focus on increased expectations for future improved
earnings in awarding large return premiums.
4.2. The role of dividends

Hand and Landsman (2005) argue that dividends could proxy
for other value relevant information because they are included
in the linear information dynamics supporting the Ohlson valuation framework. Their argument is consistent with evidence that
investors may look for information about expected future performance in rms dividend policies (Brucato & Smith, 1997; Jin, 2000).
ESS test the role of dividends in their returns specication and nd
no evidence that current or past period dividends add explanatory
power, nor that dividends proxy for other value relevant information in the absence of their term representing
about
f (t+1) information

f (t)
xt
x
(xt xt1 )
t1
expected growth in permanent earnings,
. We
P
t1
re-examine dividends as a potential proxy for

value relevant
inforf (t+1)
f (t)
xt
x
(xt xt1 )
t1
mation here for two reasons. First, measuring
P
t1
requires analysts forecasts, the availability of which introduces a
Table 9c
Correlations of quarterly earnings. Panel C: JE versus NJE rms, 19912011 (NTLS).
NJE rms
Quarter 4 (t 1)
Quarter 1 (t)
Quarter 2 (t)
Quarter 3 (t)
JE rms
Quarter 4 (t 1)
Quarter 1 (t)
1.000
0.0011
0.0234
1.000
0.0354***
Quarter 2 (t)
Quarter 3 (t)
0.0114
1.000
0.0408***
0.0813***
1.000
***
market return. Firms are categorized as potential tax-loss sellers, PTLS if they have negative holding returns for the month of December in year t 1. All other rms are
classied as not tax-loss sellers, NTLS. The year in which the comparison quarter falls is designated as t. N = 1,991 for JE rms. N = 21,336 for NJE rms. Correlations presented
are Spearman correlations, which do not depend upon assumptions of the datas normal distribution or heteroskedasticity.
G Model
ARTICLE IN PRESS
15
Table 10
Results of regressing January returns on price-deated rst quarter earnings level, earnings change, dividends, and rm size on the unpartitioned and partitioned sample.
Rqn,m1 =
0 +
1
xqn xq(n1)
dq(n1)
xqn
dqn
+
2
+
4 Size +
5
+
6
Pq(n1)
Pq(n1)
Pq(n1)
Pq(n1)
(9)
All rms
JE rms
NJE rms
Adj. R2
3,707
418
3,289
0.059
0.509***
0.008
0.304
3.139*
0.431*
0.079
1.155*
0.038
0.007*
0.033***
0.002
1.062
0.759
0.742
0.145
1.923
0.027
0.010
0.275
0.008

JE rms are those for which the return in January return premium is higher than the return premium of any other month in the calendar year. NJE rms are those for which
the return premium of at least one other month is higher than the return premium for January in the calendar year. Return premium = CRSP monthly holding return value
weighted market return. Rqn,m1 is holding period return for the rst month in quarter n, in this case January returns. Pq(n1) is beginning of period price, xqn is earnings per
share in quarter n, dqn is cash dividend per share in quarter n. Size is measured as the logarithm of total assets. Earnings and prices are adjusted for stock splits and dividends.
Regressions utilize two-way cluster robust standard errors.
***
selection bias toward larger, more established rms. Second, the

main difference in specication between our empirical model in
this study and that proposed by ESS is our focus on the rst month of
the quarter, when investors may be struggling to determine expectations for future earnings performance. In such a setting there
may be an informational function for dividends. We investigate the
role of dividends in explaining January returns using the following
empirical form of the model, which follows from Eq. (5) and in
which current (rst quarter) and
the previous
years fourth quarter
f (t+1)
f (t)
xt
(xt xt1 )
x
t1
price deated dividends replace
, as follows:
P
t1
Rqn,m1 =
0 +
1
+
6
xqn
+
2
Pq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size +
5
dqn
Pq(n1)
dq(n1)
observations in quarter 1 with analysts forecasts to 418 (3,707).

The analysis indicates that substituting rst and fourth quarter dividends for the other information term results in a model with little
explanatory power for either NJE rms or rms overall, as their
respective adjusted R2 values are only 0.008 and 0.010. Consistent
with the idea that dividends represent a decrease in future valuegenerating power, the coefcients on
5 and
6 , are negative in
all cases but nowhere signicant. In the case of JE rms adjusted
R2 using dividends is higher than for NJE rms or for the pooled
sample. Adjusted R2 values for JE rms are slightly lower for the
dividend model than for the ESS model (0.275 versus 0.293). Taken
together, we nd little support for dividends as an explanatory variable that is preferable to the other information term specied by
ESS.
(9)
Pq(n1)
4.3. April (July, October) effects
dqn = cash dividends paid in quarter n, in this case quarter 1 and all
other variables are as previously dened.
Results are shown in Table 10. As we are limited to using rms
for which we have two-consecutive-quarter dividend data available our sample size decreases slightly from 538 (4,206) JE (NJE)
As a nal check, we perform similar partitions for rms that

experience an April (July, October) effect; that is, April (July,
October) return premiums are higher than return premiums of
all other months in the year. Consistent with the methodology
Table 11
Results of regressing April (July, October) returns on price-deated second (third, fourth) quarter earnings level, earnings change, other value relevant information, and rm
size, after partitioning the sample into AE (JULE, OE) and NAE (NJULE, NOE) rms. Results for JE and NJE rms from Table 5, Panel A are shown in the top line of each section
for comparison purposes.
Rqn,m1 =
0 +
1
xqn
Pq(n1)
+
2
xqn xq(n1)
Pq(n1)
+
3
f (n+1)
f (n)
xqn,m1 xq(n1)
xqn xq(n1)
Pq(n1)
+
4 Size
(8)
N
JE rms
AE rms
JULE rms
OE rms
NJE rms
NAE rms
NJULE rms
NOE rms
0
538
1,427
1,154
1,202
4,206
10,171
11,503
9,400
1
***
0.530
0.488***
0.414***
0.560***
0.000
0.019
0.066*
0.026
Adj. R2
2.434
0.677
0.541***
0.800*
3.683
1.356***
3.248*
2.451*
2.836
0.672
2.918*
2.095*
0.039
0.029*
0.029***
0.040**
0.293
0.074
0.208
0.192
0.421*
0.092
0.445*
0.296
1.339***
1.125**
1.385***
1.572*
1.409***
0.853*
1.513***
1.525*
0.003
0.004*
0.006*
0.002
0.026
0.014
0.037
0.034
**
***

**
***
JE (AE, JULE, OE) rms are those for which the return premium in January (April, July, October) is higher than the return premium of any other month in the calendar year.
NJE (NAE, NJULE, NOE) rms are those for which the return premium of at least one other month is higher than the return premium for January (April, July, October) in
f (n)
the calendar year. Rqn,m1 is holding period return for rst month in quarter n, Pq(n1) is beginning of period price, xqn is earnings per share in quarter n, xq(n1) is the latest
*
f (n+1)
analyst forecast that comes out in quarter n 1 for quarter n earnings per share, xqn,m1 is the latest analyst forecast that comes out in the rst month of quarter n for quarter
n + 1 earnings per share. Size is measured as the logarithm of total market value of equity. Earnings and prices are adjusted for stock splits and dividends. Regressions utilize
two-way cluster robust standard errors.
G Model
16
ARTICLE IN PRESS
used in Table 5, Panel A, April (July, October) returns are regressed

against the earnings and size control variables for the second (third,
fourth) quarter. Results are presented in Table 11, along with the
JE/NJE regression results from Table 5, Panel A for comparison.
Although the results can also generally be interpreted as supportive
of an association between April (July, October) effect returns and
accounting earnings variables, there are some differences between
these results and those for JE rms. The most obvious contrast is
that, compared to JE rms, the explanatory powers in Table 11
are considerably lower across the second, third and fourth quarters. The highest adjusted R2 value in Table 11 is for July effect
rms; even so, it is 29.0% lower than the adjusted R2 value for
JE rms.
Table 11 also shows that
1 is signicantly negative for all of
the effect rms. Still, the magnitude of the JE earnings level coefcient dwarfs the others. It appears that rst-month-of-quarter
investors may in general discount earnings level and reward permanent earnings, but the effect is more noticeable around the
turn of the calendar year. The literature relating these other rstmonth-of-quarter effects to tax-loss explanations is scant; the
possible existence of other economic explanations for such systematic behavior offers motivation for future studies.
2 and
3 ,
the coefcients on earnings changes and expected future earnings growth, are reliably positive and signicant for all quarters.
We interpret this as additional support for our original argument
that valuation depends primarily upon permanent earnings and
importantly upon the perceived sustainability of permanent
earnings.
5. Conclusions
Theoretical and empirical research holds that returns reect
information in contemporaneous earnings, as well as expectations
about future earnings performance. Consistent with empirical and
anecdotal evidence, we document that the January effect is not
exclusively a small rm effect and seems to be driven by a relatively
small number of rms overall. Overall, our inquiry contributes to
the literature by reconciling the assumption of market inefciency
inherent in the tax-loss selling explanation with a fundamental
accounting valuation approach, and offers some further insights
into the nature of the January effect.
Our study has limitations. In using the highest return premium
of the calendar year we have chosen a strict denition to categorize January effect rms. There are perhaps other denitions that
could increase the sample size of JE rms. Our data requirements
bias our sample toward larger, more established rms. Further, the
existence in our sample of January effect rms that do not appear to
be related to tax-loss selling nor are explained well in our valuation
model suggests that the January effect anomaly itself bears further
investigation.
We utilize a permanent earnings valuation model rather than a
trading approach and nd evidence that the January effect is associated with rational economic behavior by market participants. That
is, investors in JE rms appear to discount rst quarter earnings
performance but do reward permanent earnings and expectations
for future improvements with high return premiums. Examining future earnings suggests that these rms earn those return
premiums via higher second quarter earnings. It appears that
in general, JE rms reward investors who identify future swans
amongst the ugly ducklings.
The explanatory power of our model is very high for JE rms,
relative to those that do not experience the effect. It is also higher
than for April (July, October) effect rms. Our ndings suggest
that January effect premiums reect rational pricing but of course
do not establish market efciency, because the ESS model rests on

an assumption of market efciency.
Our results also support a tax-loss selling explanation for the
January effect. First, we nd that a substantial percentage of January
effect rms are potential tax-loss sellers at the end of the previous
year. Second, discounting of the earnings level variable using our
model occurs only for tax-loss selling rms, which are generally
poor earnings performers relative to non-tax loss sellers.
Results for April (July, October) effect rms in the second
(third, fourth) quarter are similar to those of JE rms but generally weaker. The difference we observe in rst quarter results for
the January effect may be linked to ndings that aggregate accruals
explain aggregate market returns (Cready & He, 2014; Hirshleifer,
Hou, & Teoh, 2009; Kothari, Lewellen, & Warner, 2006), but more
investigation is needed. The timing of earnings forecast changes
and their possible association with returns in the month of January
offers yet another avenue for research.
Acknowledgements
We are grateful to participants at the 2012 and 2014 AAA Ohio
Regional Meetings, the 2014 AAA Annual meeting, the Wright State
University research workshop, and several anonymous reviewers
for helpful comments. Dennis J. Easterday graciously provided editing assistance. All errors are ours.
References
Ali, A., & Zarowin, P. (1992). The role of earnings levels in annual earnings-returns
studies. Journal of Accounting Research, 30(2), 286296.
Anderson, L. R., Gerlach, J. R., & DiTraglia, F. J. (2007). Yes, Wall Street, there is
a January effect! Evidence from laboratory auctions. The Journal of Behavioral
Finance, 8(1), 18.
Ball, R., & Brown, P. (1968). An empirical evaluation of accounting income numbers.
Journal of Accounting Research, 6(2), 159177.
Bathke, A. W., Lorek, K. S., & Willinger, G. L. (1989). Firm size and the predictive
ability of quarterly earnings data. The Accounting Review, 64(1), 4968.
Beaver, W. H., Lambert, R. A., & Morse, D. (1980). The information content of security
prices. Journal of Accounting and Economics, 2(1), 328.
Beaver, W. H., Lambert, R. A., & Ryan, S. G. (1987). The information content of security
prices: A second look. Journal of Accounting and Economics, 9(2), 139157.
Bernard, V. L., & Thomas, J. K. (1990). Evidence that stock prices do not fully reect the
implications of current earnings for future. Journal of Accounting and Economics,
13(4), 305340.
Best, R. W., Hodges, C. W., & Yoder, J. A. (2006). Expected earnings growth and portfolio performance. Review of Quantitative Finance and Accounting, 26(4), 431437.
Beyer, S., Garcia-Feijoo, L., & Jensen, G. R. (2013). Can you capitalize on the turn-ofthe-year effect? Applied Financial Economics, 23(18), 14571468.
Bley, J., & Saad, M. (2010). Cross-cultural differences in seasonality. International
Review of Financial Analysis, 19(4), 306312.
Blume, M. E., & Stambaugh, R. F. (1983). Biases in computed returns: An application
to the size effect. Journal of Financial Economics, 12(3), 387404.
Branch, B. (1977). A tax loss trading rule. Journal of Business, 50(2), 198207.
Brauer, G. A., & Chang, E. C. (1990). Return seasonality in stocks and their underlying assets: Tax-loss selling versus information explanations. Review of Financial
Studies, 3(2), 255280.
Brown, L. D., & Luo, L. (2006). The January barometer: Further evidence. The Journal
of Investing, 15(1), 2531.
Brown, L. D., & Rozeff, M. S. (1979). Univariate time-series models of quarterly
accounting earnings per share: A proposed model. Journal of Accounting Research,
17(1), 179189.
Brown, P., Ferguson, A., & Sherry, S. (2010). Investor behaviour in response to
Australias capital gains tax. Accounting and Finance, 50(4), 783808.
Brown, P., Keim, D. B., Kleidon, A. W., & Marsh, T. A. (1983). Stock return seasonalities
and the tax-loss selling hypothesis: Analysis of the arguments and Australian
evidence. Journal of Financial Economics, 12(1), 105127.
Brucato, P. F., & Smith, D. M. (1997). An analysis of the role of rm reputation in
the markets reaction to corporate dividends. Quarterly Review of Economics &
Finance, 37(3), 647665.
Chen, H., & Singal, V. (2004). All things considered, taxes drive the January effect.
Journal of Financial Research, 27(3), 351372.
Ciccone, S. J. (2011). Investor optimism, false hopes and the January effect. Journal
of Behavioral Finance, 12(3), 158168.
Collins, D. W., Kothari, S. P., Shanken, J., & Sloan, R. G. (1994). Lack of timeliness and
noise as explanations for the low contemporaneous return-earnings association.
Journal of Accounting and Economics, 18(3), 289324.
G Model
ARTICLE IN PRESS
Corhay, A., Hawawini, G., & Michel, P. (1987). Seasonality in the risk-return relationship: Some international evidence. The Journal of Finance, 42(1), 4968.
Cready, W. M., & He, T. (2014). Seasonal patterns in the relation between aggregate
quarterly accruals and aggregate market return. Working paper. University of Texas
at Dallas.
Dalton, T. (1993). An empirical investigation of the relation between tax-loss selling and the January effect. Journal of the American Taxation Association, 15(2),
2639.
DeBondt, W. F. M., & Thaler, R. H. (1985). Does the stock market overreact? The
Journal of Finance, 40(3), 793805.
DeBondt, W. F. M., & Thaler, R. H. (1987). Further evidence on investor overreaction
and stock market seasonality. The Journal of Finance, 42(3), 557580.
Dzhabarov, C., & Ziemba, W. T. (2010). Do seasonal anomalies still work? Journal of
Portfolio Management, 36(3), 93104.
Eakins, S., & Sewell, S. (1993). Tax-loss selling, institutional investors, and the January
effect: A note. Journal of Financial Research, 16(4), 377384.
Easterday, K. E., Sen, P. K., & Stephan, J. A. (2009). The persistence of the small
rm/January effect: Is it consistent with investors learning and arbitrage
efforts? Quarterly Review of Economics and Finance, 49(3), 11721193.
Easterday, K. E., Sen, P. K., & Stephan, J. A. (2011). Another specication of Ohlsons
other information term for the earnings/returns association: Theory and some
evidence. Journal of Business Finance and Accounting, 38(910), 11231155.
Easton, P. D., & Harris, T. S. (1991). Earnings as an explanatory variable for returns.
Journal of Accounting Research, 29(1), 1936.
Easton, P. D., Harris, T. S., & Ohlson, J. A. (1992). Aggregate accounting earnings can
explain most of security returns. Journal of Accounting and Economics, 15(23),
119142.
Fama, E. F. (1998). Market efciency, long-term returns, and behavioral nance.
Journal of Financial Economics, 49(3), 283306.
Fant, L. F., & Peterson, D. R. (1995). The effect of size, book-to-market equity, prior
returns, and beta on stock returns: January versus the remainder of the year.
Journal of Financial Research, 18(2), 129142.
Feltham, G. A., & Ohlson, J. A. (1995). Valuation and clean surplus accounting
for operating and nancial activities. Contemporary Accounting Research, 11(2),
689731.
Foster, G. (1977). Quarterly accounting data: Time-series properties and predictiveability results. The Accounting Review, 52(1), 121.
Fountas, S., & Segredakis, K. N. (2002). Emerging stock markets returns seasonalities: The January effect and the tax-loss selling hypothesis. Applied Financial
Economics, 12(4), 291300.
Gerlach, J. R. (2007). Macroeconomic news and stock market calendar and weather
anomalies. Journal of Financial Research, 30(2), 283300.
Gerlach, J. R. (2010). Daylight and investor sentiment: A second look at two
stock market behavioral anomalies. Journal of Financial Research, 33(4),
429462.
Givoly, D., & Ovadia, A. (1983). Year-end tax-induced sales and stock market seasonality. The Journal of Finance, 38(1), 171185.
Gow, I. D., Ormazabal, G., & Taylor, D. J. (2010). Correcting for cross-sectional and
time-series dependence in accounting research. The Accounting Review, 85(2),
483513.
Grifths, M. D., & White, R. W. (1993). Tax-induced trading and the turn-of-the-year
anomaly: An intraday study. The Journal of Finance, 48(2), 575598.
Grinblatt, M., & Keloharju, M. (2004). Tax-loss trading and wash sales. Journal of
Financial Economics, 71(1), 5175.
Gu, A. Y. (2003). The declining January effect: Evidences from the U.S. equity markets.
Quarterly Review of Economics and Finance, 43(2), 395404.
Gu, A. Y. (2004). The reversing weekend effect: Evidence from the US equity markets.
Review of Quantitative Finance and Accounting, 22(1), 514.
Gu, A. Y., & Simon, J. T. (2007). The September phenomenon of US equity
market. Advances in Quantitative Analysis of Finance and Accounting, 5(2),
283297.
Gultekin, M. N., & Gultekin, N. B. (1987). Stock return anomalies and the tests of the
APT. The Journal of Finance, 42(5), 12131224.
Hand, J. R. M., & Landsman, W. R. (2005). The pricing of dividends in equity valuation.
Journal of Business Finance and Accounting, 32(34), 435469.
Haug, M., & Hirschey, M. (2006). The January effect. Financial Analysts Journal, 62(5),
7888.
He, L. T., & He, S. C. (2011). Has the November effect replaced the January effect in
stock markets? Managerial and Decision Economics, 32(7), 481486.
Henker, J., & Debapriya, J. P. (2012). Retail investors exonerated: The case of the
January effect. Accounting and Finance, 52(4), 10831099.
Hensel, C. R., & Ziemba, W. T. (2000). Anticipation of the January small rm effect
in the US futures markets. In D. B. Keim, & W. T. Ziemba (Eds.), Security market
imperfections in world wide equity markets (pp. 179202). New York: Cambridge
University Press.
Heston, S. L., & Sadka, R. (2010). Seasonality in the cross section of stock returns:
The international evidence. Journal of Financial and Quantitative Analysis, 45(5),
11331160.
Hirshleifer, D., Hou, K., & Teoh, S. H. (2009). Accruals, cash ows, and aggregate stock
returns. Journal of Financial Economics, 91(3), 389406.
Hulbert, M. (2008). The January effects disappearing act. In Barrons Online,
January 11. http://www.marketwatch.com/story/january-effect-has-failed-tomaterialize-this-year
Jin, Z. (2000). On the differential market reaction to dividend initiations. Quarterly
Review of Economics & Finance, 40(2), 263277.
17
Jones, C. P., & Wilson, J. W. (1989). An analysis of the January effect in stocks and
interest rates under varying monetary regimes. Journal of Financial Research,
12(4), 341354.
Jones, S. L., Lee, W., & Apenbrink, R. (1991). New evidence on the January
effect before personal income taxes. The Journal of Finance, 46(5),
19091924.
Keim, D. B. (1983). Size-related anomalies and stock return seasonality: Further
empirical evidence. Journal of Financial Economics, 12(1), 1332.
Klein, A., & Rosenfeld, J. (1991). PE ratios, earnings expectations, and abnormal
returns. Journal of Financial Research, 14(1), 5164.
Koogler, P. R., & Maberly, E. D. (1994). Additional evidence of year-end tax-motivated
trading by individual investors, 19621986. Journal of the American Taxation
Association, 16(1), 122137.
Kothari, S. P. (2001). Capital markets research in accounting. Journal of Accounting
and Economics, 31(13), 105231.
Kothari, S. P., Lewellen, J., & Warner, J. B. (2006). Stock returns, aggregate earnings surprises, and behavioral nance. Journal of Financial Economics, 79(3),
537568.
Kryzanowski, L., & Zhang, H. (1992). Economic forces and seasonality in security
returns. Review of Quantitative Finance and Accounting, 2(3), 227244.
Lakonishok, J., Shleifer, A., & Vishny, R. W. (1994). Contrarian investment, extrapolation, and risk. The Journal of Finance, 49(5), 15411578.
Lamoureux, C. G., & Sanger, G. C. (1989). Firm size and turn-of-the-year effects in
the OTC/NASDAQ market. The Journal of Finance, 44(5), 12191245.
Loughran, T. (1997). Book-to-market across rm size, exchange, and seasonality: Is there an effect? Journal of Financial and Quantitative Analysis, 32(3),
249268.
Lynch, A., Puckett, A., & Yan, X. S. (2014). Institutions and the turn-of-the-year
effect: Evidence from actual institutional trades. Journal of Banking and Finance,
49(Complete), 5668.
Ma, C. K., Rao, R. P., & Weinraub, H. J. (1988). The seasonality in convertible bond
markets: A stock effect or bond effect? Journal of Financial Research, 11(4),
335347.
Mashruwala, C. A., & Mashruwala, S. D. (2011). The pricing of accruals
quality: January vs. the rest of the year. The Accounting Review, 86(4),
13491381.
Mehdian, S., & Perry, M. J. (2002). Anomalies in US equity markets: A re-examination
of the January effect. Applied Financial Economics, 12(2), 141145.
Ohlson, J. A. (1995). Earnings, book values, and dividends in equity valuation. Contemporary Accounting Research, 11(2), 661687.
Ohlson, J. A. (2001). Earnings, book values and dividends in equity valuation: An empirical perspective. Contemporary Accounting Research, 18(1),
107120.
Ohlson, J. A., & Shroff, P. K. (1992). Changes versus levels in earnings as explanatory variables for returns: Some theoretical considerations. Journal of Accounting
Research, 30(2), 210226.
Pan, M. S. (2007). Permanent and transitory components of earnings, dividends, and stock prices. Quarterly Review of Economics & Finance, 47(4),
535549.
Penman, S. H. (1987). The distribution of earnings news over time and seasonalities in aggregate stock returns. Journal of Financial Economics, 18(2),
199228.
Penman, S. H. (2004). Financial statement analysis and security valuation (2013) (5th
ed.). New York, NY: McGrawHill Irwin.
Penman, S. H., & Sougiannis, T. (1998). A comparison of dividend, cash ow, and earnings approaches to equity valuation. Contemporary Accounting Research, 15(3),
343383.
Peterson, D. R. (1990). Stock return seasonalities and earnings information. Journal
of Financial and Quantitative Analysis, 25(2), 187201.
Pettingill, G. N. (1989). Holiday closings and security returns. Journal of Financial
Research, 12(1), 5767.
Phua, V., Chan, H., Faff, R., & Hudson, R. (2010). The inuence of time, seasonality and
market state on momentum: Insights from the Australian stock market. Applied
Financial Economics, 20(20), 15471563.
Poterba, J. M., & Weisbenner, S. J. (2001). Capital gains tax rules, tax-loss trading,
and turn-of-the-year returns. The Journal of Finance, 56(1), 353368.
Reinganum, M. R. (1983). The anomalous stock market behavior of small rms in
January. Journal of Financial Economics, 12(1), 89104.
Reinganum, M. R., & Gangopadhyay, P. (1991). On information release and the
January effect: Accounting-information hypothesis. Review of Quantitative
Finance and Accounting, 1(2), 169176.
Ritter, J. R. (1988). The buying and selling behavior of individual investors at the turn
of the year. The Journal of Finance, 43(3), 701717.
Ritter, J. R., & Chopra, N. (1989). Portfolio rebalancing and the turn-of-the-year effect.
The Journal of Finance, 44(1), 149166.
Roll, R. (1983). Vas ist das? Journal of Portfolio Management, 9(2), 1828.
Seyhun, H. N. (1993). Can omitted risk factors explain the January effect? A stochastic dominance approach. Journal of Financial and Quantitative Analysis, 28(2),
195212.
Sias, R. W., & Starks, L. T. (1997). Institutions and individuals at the turn of the year.
The Journal of Finance, 52(4), 15431562.
Sikes, S. A. (2014). The turn-of-the-year effect and tax-loss-selling by institutional
investors. Journal of Accounting and Economics, 57(1), 2242.
Slemrod, J. (1982). The effect of capital gains taxation on year-end stock market
behavior. National Tax Journal, 35(1), 6977.
G Model
18
ARTICLE IN PRESS
Starks, L. T., Yong, L., & Zheng, L. (2006). Tax-loss selling and the January effect:
Evidence from municipal bond closed-end funds. The Journal of Finance, 61(6),
30493067.
Su, R., Dutta, A. S., Xu, M., & Ma, J. (2011). Financial anomalies: Evidence from Chinese
A-share markets. International Journal of Economics & Finance, 3(2), 7688.
Thaler, R. (1987). Anomalies: The January effect. Economic Perspectives, 1(1),
197201.
Tinic, S. M., & West, R. R. (1984). Risk and return. Journal of Financial Economics, 13(4),
561574.
van den Bergh, W. M., & Wessels, R. E. (1985). Stock market seasonality and taxes:
An examination of the tax-loss selling hypothesis. Journal of Business Finance and
Accounting, 12(14), 515530.
Wareld, T. D., & Wild, J. J. (1992). Accounting recognition and the relevance of
earnings as an explanatory variable for returns. The Accounting Review, 67(4),
821842.
Ziemba, W. T. (2011). Investing in the turn-of-the-year effect. Financial Markets and
Portfolio Management, 25(4), 455472.

Market Janurary

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Market Janurary

Transféré par

Droits d'auteur :

Formats disponibles

G Model

QUAECO-852; No. of Pages 18

The Quarterly Review of Economics and Finance xxx (2015) xxxxxx

Contents lists available at ScienceDirect

The Quarterly Review of Economics and Finance

Is the January effect rational? Insights from the accounting valuation

Wright State University, United States

phenomenon in which return premiums are on average higher

& Zheng, 2006).2 However, there is evidence that tax minimizing

linked to economically rational market behavior: rms that are ex

Using (2) to substitute recursively for the dividend term in

QUAECO-852; No. of Pages 18

abnormal earnings as accounting earnings less a charge for use of

bt = book value at date t; xt = accounting earnings during the period

t+1 = t + 2,t+1

which can be expressed equivalently as

t = time period of interest; Rt = return in period t, computed

QUAECO-852; No. of Pages 18

fourth) quarter there is a bias against nding an effect for returns

the latest earnings forecast for quarter n +

1 made during quarter n; xq(n1) = the latest earnings forecast

Rqn,m1 = holding return for the rst month in quarter n; xqn,m1 =

6. At least two consecutive one-quarter-ahead earnings forecasts

The dataset consists of rm-quarter observations having 12 months of CRSP data

a dependent variable in our regression model would be inappropriate because the

QUAECO-852; No. of Pages 18

mean value-weighted return premium

Following ESS we address both potential cross-correlation

We invite the reader to refer again to Fig. 2.

QUAECO-852; No. of Pages 18

Signicant at < 0.10.

Signicant at < 0.05.

contemporaneous returns, even when limited to the rst month of

3.2. JE rms versus NJE rms

categorized as NJE.16 Table 4, Panel A shows the distribution of JE

January return premium

January return premium

We also evaluate whether a rms experience of January effect

3.3. Tax-loss selling

And also in ESS.

QUAECO-852; No. of Pages 18

Signicant at < 0.10.

Signicant at < 0.05.

Quarter 1 change in EPS

QUAECO-852; No. of Pages 18

Signicant at < 0.05.

NTLS. The explanatory value of our earnings model increases by a

for the fourth through

second quarters for JE rms. The outlook for permanent earnings

Signicant at < 0.001.

The expectation for permanent earnings sustainability = xqn

xqn xq(n1) , where xqn is earnings per share in quarter n,

QUAECO-852; No. of Pages 18

Signicant at < 0.05.

4.2. The role of dividends

re-examine dividends as a potential proxy for

requires analysts forecasts, the availability of which introduces a

QUAECO-852; No. of Pages 18

Signicant at < 0.05.

selection bias toward larger, more established rms. Second, the

observations in quarter 1 with analysts forecasts to 418 (3,707).

4.3. April (July, October) effects

t+1 = t + 2,t+1