Time Will Tell

Time Will Tell: Information in
the Timing of Scheduled Earnings News

Eric C. So
Massachusetts Institute of Technology

Sloan School of Management
July 2014
Abstract
This study examines information in the schedule of future earnings announcements.
Using a novel dataset of daily earnings calendar data, I show rm-initiated revisions in
expected earnings announcement dates predict rms future earnings news and returns.
Firms that signicantly advance their announcement subsequently report better earn-
ings news relative to rms that delay, consistent with rms strategically timing news.
Advancers also outperform delayers by 260 basis points in the month after calendar
revisions, where returns are concentrated during earnings announcements and mirror
the nature of earnings news. Together, the results indicate investors fail to unravel
information embedded in the timing of scheduled news releases.
JEL Classications: G10, G11, G12, G14, M40, M41
I thank Ed deHaan, Chris Noe, and seminar participants at the 2014 Citigroup Quant Research Confer-
ence and Nasdaq Economic Research for helpful comments and suggestions. I also thank Wall Street Horizon
for generously providing data on expected earnings announcement dates. Corresponding author: Eric So,
ESo@mit.edu, E62-677 100 Main Street, Cambridge MA 02142.
Time Will Tell: Information in the Timing of Scheduled Earnings News 1
1. Introduction
News releases are often scheduled in advance. Discretion over this schedule allows for
the possibility that the timing and content of news releases are systematically related. This
systematic relation can reect strategic reasons, such as attempts to hide negative news,
and/or non-strategic reasons, such as processing costs that vary with the underlying news.
1
To the extent the scheduled time of a new release is a reliable ex ante indicator of its content,
rational agents should update their expectations of the news to reect this information (e.g.,
Milgrom (1981), Dye and Sridhar (1995), and Acharya, DeMarzo, and Kremer (2011)).
This study focuses on the schedule of rms earnings announcements, which often convey
value-relevant information to investors and whose timing is subject to managerial discretion.
Using information extracted from the schedule of future earnings announcements, this study
address two central questions. First, does the schedule of earnings announcements provide
an ex ante signal regarding subsequently reported performance? Second, do agents unravel
information embedded in the timing of scheduled news releases in a timely fashion?
To address these questions, this study uses a novel dataset containing daily snapshots
of earnings calendar data, which provides a rolling list of expected earnings announcement
dates for a broad cross-section of rms. A key feature of this data is that it identies rm-
initiated revisions in expected announcement dates and thus helps capture managers eorts
to alter the schedule of news releases. Using this forward-looking calendar data, I develop
and implement a new methodology for characterizing the relative timeliness of earnings
announcements before the actual earnings news is explicitly announced (or not announced),
which permits testing whether the schedule of rms earnings announcements is informative
about their subsequently reported performance. Additionally, the daily snapshots provide
a proxy for investors information set in real time, which facilitates testing whether market
prices adjust to signals in the schedule of earnings announcements in a timely fashion.
1
For example, auditors may apply higher verication thresholds for good versus bad earnings news, leading
to systematic variation in when a rm announces earnings (Bagnoli, Kross, and Watts (2002)).
Throughout the paper, I refer to calendar revisions as being rm-initiated if they are
triggered by a rms disclosure that explicitly states when they intend to announce earnings
(see Section 2 for more details and Appendix A for an example of a rm-initiated revision).
The main sample for my analysis consists of 18,959 rm-initiated calendar revisions between
2006 and 2013 observable at least two weeks prior to a rms expected announcement date.
I categorize each observation based on the extent to which the calendar revision advances
(i.e., moves forward) or delays (i.e., moves back) the rms expected announcement date. To
illustrate the ease of extracting signals from calendar revisions, I create a simple summary
metric, referred to as R-Score, that is highest for rms advancing their announcement by
more than one week and lowest for rms delaying by more than one week.
My rst tests show high R-Score rms (i.e., advancers) subsequently report better
earnings news than low R-Score rms (i.e., delayers) during their earnings announcements.
Specically, advancers report greater return-on-assets (ROA), same-quarter growth in ROA,
and analyst-based earnings surprises, compared to delayers. These dierences are both
statistically and economically signicant. For example, average earnings innovations and
analyst-based earnings surprises increase monotonically across R-Score portfolios, where
the average ROA is positive for advancers and negative for delayers. These results are
consistent with prior evidence that managers delay announcing bad news (Kothari, Shu, and
Wysocki (2009)) and indicate that earnings calendar revisions have strong predictive power
for subsequently reported earnings news.
Given the predictive power of calendar revisions for earnings news, I next examine
whether investors impound signals from earnings calendars into prices in a timely fashion.
To conduct these tests, I examine dierences in returns across high and low R-Score rms.
These tests show that although there is no signicant dierence in returns across advancers
and delayers at the time of their calendar revisions, there is a striking dierence in returns
following the revisions. Specically, advancers subsequently outperform delayers by more
than 260 basis points (i.e., 2.6%) in the month after calendar revisions.
The spread in returns across advancers and delayers is also symmetric. Advancers, on
average, outperform the market by approximately 138 basis points and delayers underperform
by 130 basis points over the month following calendar revisions, mitigating concerns that
returns are only predictable on the short-side of the portfolio. This symmetry highlights a
benet of using calendar revisions to identify both advancers and delayers ex ante, rather
than realized announcement dates which can only identify advancers ex post. These ndings
appear robust to a variety of standard risk-adjustments as well as value-weighted portfolios,
indicating that calendar revisions have strong predictive power for future returns.
My next analyses examine the return spread across advancers and delayers in event-
time within the month (21 trading days) surrounding rms earnings announcement dates.
These tests show that over 60% of the monthly event-time return is earned in the three-day
announcement window, suggesting that prices adjust to the information content of calendar
revisions at the time earnings are announced, rather than at the time of the revisions.
Perhaps most importantly, the predictable spread in returns across advancers versus delayers
aligns with the sign and magnitude of rms reported earnings news. The combination
of these results poses a signicant challenge to the assumption in theoretical models that
investors fully unravel information embedded in the timing of news releases.
Subsequent tests show that the predictive power of calendar revisions for future returns is
intuitively pronounced among subsets of rms, for example, those with greater sensitivities
to earnings news. The results are also pronounced in small rms with low analyst coverage,
indicating prices are less likely to reect the content of calendar revisions among rms with
weaker information environments. Finally, I show the results are stronger in rms where
managers face greater career concerns, suggesting the timing of news is more informative
when the news is likely to aect the managers wealth and/or human capital.
Additional analyses show the papers ndings are robust to alternative implementations.
First, I show rm-initiated changes in expected announcement dates continue to predict
earnings news and returns when dening revisions relative to the date a rm reported same-
quarter earnings in the prior year. These results mitigate concerns that the results hinge upon
the expected announcement date in the earnings calendar prior to the rm-initiated revision.
Second, I nd similar results when ranking rms into portfolios using the distribution of
calendar revisions from the prior calendar quarter, which mitigates concerns the results are
sensitive to the use of R-Score or intertemporal shifts in the distribution of revisions. Finally,
I show the papers results hold when using a larger sample that includes both rm-initiated
revisions and non-initiated revisions (e.g., those triggered by a peer rms behavior), but
that non-initiated revisions have signicantly less predictive power for future returns. These
ndings indicate calendar revisions are more likely to convey information about subsequently
reported earnings when the revision originates from within the rm.
This papers ndings are consistent with Chambers and Penman (1984) which shows that
announcement delays accompany bad news and Penman (1984) and Bagnoli, Kross, and
Watts (2002) which show that prices drift downward when rms do not announce on time.
My ndings compliment and extend these studies in at least four ways. First, this paper
introduces a new forward-looking methodology for quantifying the timeliness of earnings
announcements, which expands the scope of ex ante tests to include early announcements.
Second, using this new methodology, this study is the rst to show that advancements
predict positive earnings news and returns ahead of the actual announcements. Third, by
identifying the timeliness of announcements multiple weeks in advance, the return results
in this study are not only notably larger than in prior studies but are also less sensitive to
the concern applicable to strategies based on announcement delays that transaction costs
increase dramatically prior to announcements (e.g., So and Wang (2014)). Finally, my tests
focus on how investors respond specically to rms disclosures that state the timing of a
future announcement without explicit mention of its content, and thus provides a more direct
test of whether investors unravel signals embedded in the timing of scheduled news releases.
As a result, this paper provides new evidence of a sizable anomaly that appears to be driven
by investors underweighting information related to the timing of future announcements.
This papers ndings also align with evidence in Graham, Harvey, and Rajgopal (2005)
that some managers delay releasing bad news in hopes of receiving osetting good news
(e.g., pairing poor current earnings with promising signs of future earnings), as well as
media accusations that managers operate within a corporate culture reluctant to pass along
bad news (Fletcher and Mufson (2014)). These ndings compliment prior evidence that
managers strategically schedule news releases (e.g., Damodaran (1989) and Gennotte and
Trueman (1996)) and extends these studies by showing that managers may postpone releasing
negative news without prices fully reecting this information in a timely fashion.
Additionally, this paper relates to growing evidence that prices underreact to low saliency
signals (e.g., Hirshleifer, Lim, and Teoh (2009), Da, Engelberg, and Gao (2011), Cohen,
Diether, and Malloy (2013), and Giglio and Shue (2014)) and suggest that investors may
underweight rms disclosures regarding expected announcement dates due to low saliency
but that calendar revisions should be treated as signicant sources of information ahead of
the actual announcements.
More broadly, the ndings of this paper relate to a substantial literature spanning eco-
nomics, nance, and accounting studying informed agents who possess discretion over their
communication with outsiders. A common assumption in these studies is that outsiders
rationally infer insiders motives and/or private information using signals embedded in the
insiders actions. For example, the evidence in this paper relates to the unraveling result,
which predicts agents favor disclosing information because outsiders rationally interpret the
absence of disclosure as a negative signal (Grossman and Hart (1980), Grossman (1981)).
This study provides evidence that investors do not always rationally uncover signals con-
veyed by the timing of news releases and thus highlights a need for further research into
investors ability to infer information and motives embedded in managerial actions.
The rest of the paper is organized as follows. Section 2 details the data and main results.
Section 3 contains additional analyses and robustness checks. Section 4 concludes.
2. Empirical Tests
This section provides details on the earnings calendar data used throughout the paper,
discusses the sample selection process, and details the main tests and empirical results.
2.1. Earnings Calendar Data and Sample Selection
The main analyses of this paper examine information in rms earnings calendar revisions.
To calculate calendar revisions, I use daily snapshots of earnings calendar data provided by
Wall Street Horizon from 2006 through 2013. Wall Street Horizon began disseminating
earnings calendar data in 2006, where each snapshot lists expected announcement dates for
a broad cross-section of rms. The calendar data reects information available to investors
by 4am ET of each trading day. I use the data to proxy for investors daily information
set regarding expected announcement dates, which is likely conservative because Wall Street
Horizon provides this data to clients at much higher frequencies through streaming feeds.
2
Some clients license the calendar data and post it online as a service to their customers.
The earnings calendar provides a rolling view of expected announcement dates by con-
tinually updating the calendar in response to new information. A key feature of the data is
that it indicates whether an expected announcement date stems from a rm explicitly stating
when they intend to announce earnings (See Appendix A for an example). In these cases, I
refer to the resulting expected date as being conrmed (or, equivalently, rm-initiated).
All conrmed expected dates are primary sourced based on public information including,
but not limited to, rms investor relations webpage, press releases, and direct correspon-
dence. Prior to an announcement becoming conrmed, the calendar contains unconrmed
expected dates that are forecasts based on information from several sources including a rms
past reporting behavior. Throughout the text, I use the term rm-initiated revisions to
refer to changes in the earnings calendar that accompany an expected announcement date
changing status from unconrmed to conrmed.
2
The calendar revisions used in this study are summarized via Wall Street Horizons DateBreaks Moni-
toring Service. See WallStreetHorizon.com for more details regarding the companys data products.
My primary tests focus on rm-initiated revisions in the earnings calendar.
3
I calculate
earnings calendar revisions as the change in a rms expected announcement date from one
trading day to the next. To capture economically meaningful calendar revisions, I focus
the analysis on revisions that alter the expected announcement date by at least two days.
Because calendar revisions can occur at any time prior to earnings announcements, I use the
timeline below to detail the sample requirements and structure of my main tests:
t-31
Calendar revision (REV ):
Conrmed date, EXDT
r
, on day r (t-31, t-11)
compared to prior expected date, EXDT
r1
,
REV =(EXDT
r
-EXDT
r1
)

r
t-11

Calendar-time returns:
R Score portfolio returns
starting 1-day after revision, r+1

Event-time returns:
R Score portfolio returns
relative to announcement date t
t-10 t t+10
observed on announcement date t

Protability and earnings surprise
Outcomes observed:
The above timeline helps emphasize that the empirical tests are constructed to avoid look
ahead bias. I calculate a sample of revisions where the revision date, r, occurs in the month
(21 trading days) ending two weeks prior to rms conrmed expected announcement date, t.
Notationally, I require that r falls between t-31 and t-11. The requirement that the revision
occurs no earlier than t-31 helps identify revisions occurring after a rms scal period and
thus those which are potentially informed by managers knowledge of the rms performance.
Similarly, the requirement that the revision occurs no later than t-11 helps mitigate the risk
that investors learn about earnings through other sources such as pre-announcement media
coverage and, more importantly, as depicted in the timeline, facilitates examining event-time
returns surrounding rms announcements without subjecting the results to look ahead bias.
The results reported below regarding the predictive power of calendar revisions for earnings
news and returns do not appear sensitive to this sample requirement.
3
My sample consists of unique rm-quarter observations. If there are more than one conrmed earnings
announcement date that meets the sample requirements, I only include calendar revisions based on the rst
conrmation to avoid look ahead bias. In untabulated results, I nd that rms rarely conrm alternative
announcement dates for the same quarter (less than 2% of all rm-quarters).
To characterize the extent of a given revision, I create a variable, REV , which equals
the number of trading days the rms expected announcement date shifts when the date
changes status from unconrmed to conrmed. Notationally, REV =(EXDT
r
-EXDT
r1
),
where EXDT is the expected date and the subscripts indicate that REV is measured on the
date of rm-initiated revisions (r) relative to the immediately preceding trading day (r-1).
Higher values of REV indicate the rm moved forward their expected date and vice versa
for lower values of REV . I merge the calendar revision sample with return data from CRSP,
nancial information from Compustat, and analyst-based earnings surprise data from IBES.
The nal sample consists of 18,959 observations spanning 2006 through 2013.
I categorize each observation based on the extent to which the calendar revision advances
(i.e., moves forward) or delays (i.e., moves back) a rms expected announcement date.
To illustrate the ease of extracting signals from calendar revisions, I implement a simple
summary metric, referred to as R-Score, that is highest for rms advancing their expected
announcement date by more than one week and lowest for rms delaying by more than one
week. Specically, for each value of REV , I dene R-Score as follows:
R-Score =
0 (Delay) for REV < 5

0.25 for REV [3, 5]
0.5 for REV [2, +2]
0.75 for REV [+3, +5]
1 (Advance) for REV > +5
(1)
where the cuto points are selected to provide a simple classication rule that creates sym-
metry in the average magnitude of REV across R-Score portfolios (see Table 1 for details).
Using static cuto points also ensures that subsequent tests can be implemented without
referencing the full sample of calendar revisions within a given period, however a potential
concern is that the classication rule is ad hoc and can lead to unequal sample partitions.
To mitigate this concern, additional analyses show the papers inferences are not specic to
this classication rule and hold when using the cross-sectional distribution of REV .
As depicted in the above timeline, my main analyses examine the predictive power of
rms R-Score for earnings news and returns that become observable after calendar revisions.
My main tests focus on dierences between high R-Score rms (hereafter referred to as
advancers) and low R-Score rms (hereafter referred to as delayers). Additionally, I refer
to tests implemented relative to the calendar revision date (r) as calendar-time analyses and
tests implemented relative to the expected announcement date (t) as event-time analyses.
2.2. Descriptive Statistics
Table 1 contains descriptive statistics of the main sample used throughout the paper.
Panel A presents annual descriptive statistics, where the rst two columns indicate the
number of unique rm-quarters and rms, respectively. The sample consists of approximately
2,300 rms-quarters per year and an average of 1,524 unique rms. HORIZON equals the
number of trading days between the revision date r and expected announcement date t. Panel
A shows that the average revision date in my sample occurs approximately 16 trading days
prior to their expected announcement date, which is closer to the announcement than the
midpoint of the sample requirement that r falls between t-31 and t-11. The REV column
of Panel A shows the average revision shifts the expected date back by one to two days,
suggesting rms are more likely to delay than advance their expected announcement date.
The |REV | column contains absolute values of REV and indicates the average revision shifts
the expected announcement date by approximately 4.5 days on an absolute basis.
The nal two columns of Panel A contain descriptive statistics on rms deviations from
their expected announcement dates. Specically, DEV equals the number of days between
the conrmed expected announcement date and the actual announcement date, and |DEV |,
equals the absolute value of DEV . The averages of DEV and |DEV | demonstrate that
rm-initiated expected announcement dates are highly accurate. Specically, the average
value of DEV is insignicantly dierent than zero and the average value of |DEV | indicates
that less than one in four rms announce earnings on a day that diers from their expected
announcement date.
Panel B of Table 1 contains descriptive statistics across R-Score portfolios. The REV
column shows revisions are nearly symmetric across R-Score portfolios, where the average
delayer (advancer) moves back (forward) expected announcement dates by 8.8 days. The
OBS column indicates the average number of rm-quarters within each portfolio, and shows
there are approximately twice as many low R-Score rms than high R-Score rms, which is
consistent with the evidence in Panel A that rms are more likely to delay than advance their
announcements. The HORIZON column in Panel B shows that delayers revise the earnings
calendar approximately one day closer to their expected date compared to advancers.
The next two columns of Panel B contain descriptive statistics for each R-Score portfolio.
MCAP equals rms market capitalization reported in millions and MOMEN equals a
rms cumulative market-adjusted return over the twelve months ending on r-11. The market
capitalization statistics indicate the average rm in my sample is a mid-cap rm, though
rm size does not vary signicantly with R-Score. By contrast, the MOMEN results show
delayers signicantly underperform advancers over the year prior to the calendar revision.
The average value of DEV indicates that delayers are signicantly more likely to deviate
from their expected announcement date by further delaying the announcement, though the
average dierence in DEV across advancers and delayers is less than one-third of a day.
The nal column of Panel B contains descriptive statistics of rms market-adjusted
return in the three-day window surrounding their calendar revision, RET(r-1,r+1). The
results show that there is no signicant dierence in returns across advancers and delayers
at the time of the calendar revision.
4
In the analysis below, I examine dierences in reported
earnings and future returns that occur after the calendar revision date.
Panel A of Table 2 contains average earnings metrics across R-Score portfolios based on
(1) the reported level of protability, ROA; (2) changes in ROA; and (3) reported earnings
relative to consensus analyst forecasts. Specically, ROA equals the rms return on assets
4
These results contrast with the ndings in Duarte-Silva et al. (2013) that prices signicantly decline at
the time of press releases explicitly mention delaying, postposing, or deferring an earnings announcement,
suggesting that the type and nature of announcement may make delays more salient to investors.
dened as net income scaled by beginning-of-quarter total assets, ROA equals the rms
annual change in ROA for the same scal quarter, and SURP equals the rms EPS reported
in IBES minus the consensus EPS forecast available immediately prior to the announcement,
and scaled by beginning-of-quarter assets. Additionally, ROA<0, ROA<0, and SURP<0
are indicator variables that equal one when the correspond variable is negative.
Panel A captures the rst main result of the paper. Specically, advancers subsequently
announce greater ROA, changes in ROA, and earnings surprises, compared to delayers.
These dierences are both statistically and economically signicant. For example, earnings
declines and negative analyst-based surprises are concentrated among delayers, where average
ROA is positive for advancers and negative for delayers. Additionally, average earnings
innovations and earnings surprises increase monotonically across R-Score portfolios. This
evidence relates to Bagnoli, Kross, and Watts (2002), which nds earnings news arrives late
in proportion to the severity of bad news but that there is no signicant relation between
reporting early and the extent of good news. In contrast, by using calendar revisions instead
of the timing of news arrival as in Bagnoli, Kross, and Watts (2002), I nd the link between
announcement timeliness and earnings news is symmetric; the extent to which rms advance
announcement dates is also proportional to the magnitude of good news.
Panel B of Table 2 presents results from regressing each earnings metric on R-Score
and controls for rms log market capitalization (SIZE), log book-to-market ratio (LBM),
return momentum (MOMEN), and historical return volatility (V LTY ). Panel B shows that
R-Score has signicant predictive power for all three earnings metrics (t-statistics ranging
from 3.46 to 5.90) that is not subsumed by standard controls.
Taken together, these results show calendar revisions predict subsequently reported earn-
ings, which is inconsistent with Skinner (1994) which argues that managers advance bad news
to avoid litigation but consistent with survey evidence in Graham, Harvey, and Rajgopal
(2005), empirical evidence in Kothari, Shu, and Wysocki (2009), and experimental evidence
in Legg and Sweeny (2014) that, on average, individuals delay announcing bad news.
2.3. Return Prediction
Given the evidence that calendar revisions predict subsequently reported earnings news,
my next set of tests examines whether investors impound signals from earnings calendars
into prices in a timely fashion. To conduct these tests, I examine dierences in returns across
advancers and delayers following earnings calendar revisions.
Table 3 contains average equal- and value-weighted returns to each R-Score portfolio
following the revision date r, using four return metrics measured over the month following
the calendar revision from r+1 to r+21. Specically, the rst two columns contain raw
and market-adjusted returns denoted as RR(r+1,r+21) and RET(r+1,r+21), respectively.
SAR(r+1,r+21) refers to size-adjusted returns which equals the rms raw return minus the
contemporaneous size-matched portfolio return. Finally, FAR(r+1,r+21) refers to factor-
adjusted returns dened as the rms raw return minus the return calculated by estimating a
rms daily sensitivity to the market (MKTRF), small-minus-big (SMB), high-minus-low
(HML), and up-minus-down momentum (UMD) factors over the year prior to the revision
and applying those sensitivities to the contemporaneous factors.
Panel A of Table 3 captures the second main result of the paper. Specically, the results
demonstrate a robust positive relation between calendar revisions and future returns. For
each return metric, the average return spread across advancers and delayers exceeds 260
basis points in the month after calendar revisions, with corresponding t-statistics ranging
from 3.95 to 5.17. The spread in market-adjusted returns is also fairly symmetric across
advancers and delayers, mitigating concerns that returns are limited to the short side of the
portfolio. Advancers, on average, outperform the market by approximately 138 basis points
and delayers underperform by 130 basis points over the month following calendar revisions,
which align with the nature of rms subsequently reported earnings news. The magnitude
of these returns compares favorably to other well-known anomalies, however, the amount of
capital deployable to exploit this evidence is limited by the fact that only a subset of rms
revise earnings dates within a given trading period.
To address the possibility that return prediction is limited to small rms where trans-
action costs are highest, Panel B of Table 3 presents value-weighted future returns across
R-Score portfolios. Value-weighting lowers the portfolios performance but the resulting
returns remain economically signicant. The average dierence in value-weighted returns
across advancers and delayers ranges from 153 to 216 basis points in the month after calen-
dar revisions, with corresponding t-statistics ranging from 1.95 to 2.13. This evidence of a
robust post-revision return spread contrasts sharply with the insignicant price reaction at
the time of calendar revisions shown in Table 1, suggesting that investors do not unravel the
implications of calendar revisions for earnings news in a timely fashion.
My next analyses examine the spread in event-time returns within the month (21 trading
days) surrounding rms expected earnings announcement dates.
5
As shown in Table 3, I
nd no evidence that the results signicantly vary across return metrics and, in the remaining
tests, focus on predicting market-adjusted returns. Table 4 contains market-adjusted returns
around earnings announcements, where RET(t+X,t+Y) denotes the cumulative market-
adjusted return from day X to Y relative to the expected earnings announcement date t.
The RET(t-10,t+10) column of Table 4 shows advancers outperform delayers by 253 basis
points in the month centered on rms expected announcement date (i.e., from t-10 to t+10),
consistent with the magnitude of the calendar-time return spread documented in Table 3.
The RET(t-10,t-2) column of Table 4 shows advancers only weakly outperform delayers
by 54 basis points (t-statistic = 1.77) prior to announcements from t-10 to t-2. In contrast,
the RET(t-1,t+1) column of Table 4 shows that event-time returns are heavily concentrated
in the three-day window surrounding rms announcements dates. Specically, over 60%
(=158/253 basis points) of monthly event-time returns are earned during the announcement.
This evidence suggests that prices adjust to the information content of calendar revisions at
the time earnings are announced, rather than at the time of the revisions.
5
The analysis focuses on expected, rather than actual, announcement dates, to avoid look ahead bias.
Use of actual announcement dates yields qualitatively identical results, which is not surprising given the
evidence in Table 1 that rms generally announce earnings on their conrmed expected date.
The RET(t+2,t+10) column of Table 4 also shows that advancers outperform delayers
by approximately 50 basis points following earnings announcements, which is consistent with
prior evidence of post-earnings announcement drift (PEAD). However, Table 4 also shows
that 80% (=211/253 basis points) of monthly event-time returns are earned up through the
announcement window, indicating the results are likely distinct from PEAD.
Related evidence in Figure 1 depicts the spread in returns across advancers and delayers
in event-time. The top graph shows that the cumulative spread in returns reaches 50 basis
points two days prior to the announcement (i.e., t-2), but nearly doubles on the announce-
ment date t, and jumps over four-fold by t+1 to over 200 basis points. The bottom panel of
Figure 1 decomposes the event-time return spread into the average cumulative returns for
advancers and delayers and shows the price reaction at announcements is fairly symmetric
across advancers and delayers.
6
Figure 2 shows the average spread in returns across advancers versus delayers for each
calendar quarter, where RET(r+1,r+21) is shown in black bars and RET(t-1,t+1) is shown
in grey bars. The results show that the average return spread is positively skewed and gen-
erally positive over time, yielding positive average monthly (announcement-window) returns
in 23 (29) of the 32 calendar quarters in my sample window. This evidence helps mitigate
concerns that the return-based results are isolated within a specic period.
Finally, Table 5 contains regression results of monthly and announcement-window returns
on R-Score and standard risk controls. The dependent variable in the rst three columns is
RET(t-1,t+1) and RET(r+1,r+21) in the latter three columns. In both sets of regressions, I
nd that R-Score predicts returns incremental to the control variables, where the coecient
magnitudes align with the return spreads shown in earlier tables. These ndings help mitigate
concerns that the relation between R-Score and future returns is driven by a rms exposure
to standard risk proxies.
6
The graph also depicts rising pre-announcement prices for both advancers and delayers, which is consis-
tent with the evidence in Johnson and So (2014) that prices predictably rise ahead of earnings announcements
due to asymmetrically high costs of acting on negative news.
Taken together, the results of this section indicate that calendar revisions have strong
predictive power for future returns, which relate to evidence in Penman (1984) and Bagnoli,
Kross, and Watts (2002) that prices react negatively when rms do not announce on time but
continue to drift downward through the announcement. The ndings in those studies suggest
that investors recognize the failure to announce on time as a negative signal but are subject
to the concern that the initial price reaction is incomplete due to transaction costs which
increase dramatically prior to announcements (So and Wang (2014)) and particularly so for
demands to sell (Johnson and So (2014)). These same concerns regarding pre-announcement
transaction costs are unlikely to apply to this paper because R-Score portfolios are formed
multiple weeks in advance of announcements based on forward-looking calendar revisions.
The novel data used this study allows me to examine how investors interpret disclosures
that reference only the timing of an announcement, with no explicit indication of its content.
As a result, this study provides a more direct test of, and challenge to, the assumption that
investors fully unravel signals embedded in the timing of scheduled news releases.
7
Moreover,
the methodological innovation in this paper yields returns that are balanced across long and
short positions and signicantly larger than in prior research. At 260 basis points per month,
the return results suggest a massive underreaction to earnings calendar information.
The most likely explanation for the ndings in this study are that investors underreact to
the schedule of future earnings announcements because it is primarily conveyed through low
salience disclosures, in contrast to rms failing to announce on time, which are likely more
attention-grabbing events. As the example in Appendix A shows, it is not immediately ob-
vious that these seemingly boilerplate disclosures contain information regarding the nature
of earnings news; only when the news is to be announced. Thus, identifying value-relevant
signals embedded in these disclosures requires viewing them in the context of past reporting
behavior and with a skeptical lens that managers strategically time news announcements.
7
The idea that investors underweight the information content of calendar revisions is directly related to
prior studies showing that prices underreact to rms nancial ratios (e.g., Lakonishok, Shleifer, and Vishny
(1994), Ou and Penman (1989), and So (2013)) and fundamental signals (e.g., Sloan (1996), Piotroski (2000),
and Piotroski and So (2012)) that have strong predictive power for subsequently reported earnings news.
3. Additional Analyses
This section provides details on extensions of the main results as well as tests designed
to gauge the robustness of the papers ndings to alternative implementations.
3.1. Contextual Analysis
The preceding analysis establishes a robust link between rm-initiated calendar revisions
and future returns. A natural extension of these tests is to examine whether the return
results are predictably concentrated among subsets of rms in which calendar revisions are
more likely to be relevant for prices. To address this possibility, Table 6 contains results
from regressing RET(r+1,r+21) on R-Score interacted with four conditioning variables.
The rst two conditioning variables in Table 6 capture the extent of a rms information
environment and trade frictions. Specically, 1(Small Firm) is an indicator variable that
equals one if the rm is in the lowest tercile of market capitalization and 1(Low Coverage) is
dened analogously for analyst coverage, where terciles are measured each calendar quarter.
I predict that the predictive power of R-Score for returns is concentrated among smaller
rms where investors are less likely to learn about the earnings information embedded in
calendar revisions through other sources such as media coverage.
Abarbanell and Lehavy (2003) and So (2013) provide evidence that analysts investment
recommendations signal a rms incentives to meet or beat analysts forecasts, such that
higher recommendations signal a greater sensitivity of a rms share price to earnings news.
Thus, I dene 1(Buy Recommendation) as an indicator variable that equals one if the rm
has a consensus BUY recommendation in IBES and predict that it has a positive interaction
eect with R-Score in predicting future returns because prices should react more strongly
to subsequently announced earnings news for this subset of rms.
Finally, Gilson (1989) shows that managers career concerns increase when a rm ap-
proaches distress and Kothari, Shu, and Wysocki (2009) argues that managers of distressed
rms have a heightened incentive to suppress bad news. I dene 1(High Distress) as an
indicator variable that equals one if the rm is in the lowest tercile of the Zmijewski (1984)
Z-Score nancial distress measure and predict that it has a positive interaction eect with
R-Score in predicting returns. This prediction is based on the idea that managers are more
likely to use discretion over the timing of earnings news when it has a greater impact on
their human capital and/or personal wealth.
Consistent with these predictions, the interaction terms in Table 6 show that return
prediction increases among rms with greater sensitivities to earnings news, small rms
with low analyst coverage, and in rms whose management faces greater career concerns.
These results show the predictive power of calendar revisions for future returns is intuitively
correlated with rm characteristics through contextual analysis.
3.2. Robustness Tests
This subsection presents three sets of tests designed to extend and gauge the robustness
of the papers main ndings using alternative empirical implementations. First, although the
conrmed announcement dates in this study are primary sourced to public information, one
potential concern is that unconrmed expected announcement dates are based on proprietary
forecasting techniques and/or are specic to the earnings calendar data used in this study.
To mitigate this concern, Table 7 re-examines the main analyses when dening calendar
revisions as the dierence between a conrmed date and the date a rm reported same-
quarter earnings in the prior year.
Panel A of Table 7 contains earnings metrics and future returns across ve RW-Score
portfolios (the RW is short for random-walk), where the portfolio assignments are based
on revisions in expected announcement dates relative to rms prior year announcement
date (i.e., assuming announcement dates follow a random-walk). Using these random-walk
expected announcement dates as the benchmark, I assign rms to RW-Score portfolios using
the same cuto points as in Equation (1) when calculating R-Score. Table 7 shows that
rm-initiated revisions in expected announcement dates continue to signicantly predict both
earnings news and returns when benchmarking to random-walk dates, though the predictive
power of revisions for RET(r+1,r+21) declines to 220 basis points (t-statistic = 3.79). These
results mitigate concerns that the results hinge upon the expected announcement date in
the earnings calendar prior to the rm-initiated revision.
To supplement the tests based on random-walk dates, Panel B contains descriptive statis-
tics of RET(r+1,r+21) when partitioning the sample across rms with above and below
median announcement date consistency. I dene announcement date consistency using the
historical standard deviation of the dierence between random-walk forecasts and rms
actual earnings announcement dates over the prior ten quarters, where lower standard de-
viations indicate greater consistency (i.e., the rms announcement pattern conforms more
closely to a random-walk). Panel B captures the intuitive result that the predictive power of
RW-Score for returns increases among rms that consistently reported earnings on the same
day in prior years. Specically, the return spread more than doubles among high consistency
rms (360 bps, t-statistic = 2.95) compared to low consistency rms (159 bps, t-statistic =
1.88). These results indicate that historical reporting patterns are also a useful conditioning
variable for strategies based on calendar revisions.
An additional potential concern with the papers main results is that rms are categorized
as advancers and delayers using static cuto points for calendar revisions rather than allowing
the distribution to shift over time. To mitigate this concern, Table 8 contains average
earnings metrics and future returns across portfolios where rms are sorted into quintiles
(Panel A) and terciles (Panel B) of calendar revisions each calendar quarter, where higher
(lower) values are assigned to quintile 1 (0).
Both Panels A and B of Table 8 show the papers main results hold when ranking rms
into portfolios using the distribution of calendar revisions from the prior calendar quarter.
Specically, using both quintile and tercile portfolios, I nd rms with relatively large positive
calendar revisions tend to subsequently announce better earnings performance and have
higher returns, compared to rms with relatively large negative calendar revisions. For
example, use of terciles in Panel B produces a spread in RET(r+1,r+21) of 144 basis points
(t-statistic = 3.20). These ndings mitigate concerns the results are sensitive to intertemporal
shifts in the distribution of calendar revisions.
In my nal analyses, I relax the sample requirement that calendar-revisions be rm-
initiated. In the introduction, I argued that a key feature of the calendar data is that it
identies rm-initiated revisions in expected announcement dates and thus helps capture
managers eorts to alter the schedule of news releases. In Table 9, I examine whether this
feature of the data is important by re-examining my main tests using a sample that aggregates
both rm-initiated and non-initiated calendar revisions, where non-initiated revisions reect
changes in the earnings calendar that result for reasons other than a rm explicitly stating
when they intend to announce earnings (e.g., a peer rm delays their announcement).
Table 9 reports the spread in RET(r+1,r+21) and RET(t-1,t+1) across R-Score port-
folios for 30,265 observations that includes both rm-initiated and non-initiated revisions.
The rst two columns contain returns for this combined sample. Although the return spread
across advancers and delayers remains economically and statistically signicant, the magni-
tude and signicance decline. Specically, compared to the results in Table 3, returns in the
month following revisions fall from 269 basis points (t-statistic = 4.15) in the rm-initiated
sample to 226 basis points (t-statistic = 3.47) in the aggregate sample, which represents a
15% decline. Similarly, announcement-window returns fall from 158 basis points (t-statistic
= 4.85) to 115 basis points (t-statistic = 3.46), which represents a 27% decline.
The latter two columns of Table 9 show that the average monthly return falls to 160 basis
points (t-statistic = 1.54) and the announcement-window return falls to 82 basis points (t-
statistic = 1.95) among the subsample of revisions that are not rm-initiated. Together, the
results in Table 9 show that calendar revisions continue to predict returns when aggregating
rm-initiated and non-initiated revisions, though non-initiated revisions have signicantly
lower predictive power. These results highlight the benet of identifying the subset of cal-
endar revisions that are rm-initiated and suggests calendar revisions are more likely to be
informative when they originate from within the rm.
4. Conclusion
This study examines information in the schedule of rms earnings announcements using
a novel dataset containing daily snapshots of expected announcement dates. A key feature of
this data is that it identies rm-initiated revisions in expected earnings announcement dates
and thus helps capture managers eorts to alter the schedule of news releases. Firms that
signicantly advance their expected announcement date subsequently report better earnings
news than those that delay, consistent with rms strategically timing their announcements.
Despite calendar revisions predicting earnings news, I nd no evidence prices react at the time
of the revisions but nd a striking dierence in post-revision returns. Specically, advancers
outperform delayers by more than 260 basis points in the month after revisions, which aligns
with the sign and magnitude of subsequently reported earnings news. Additionally, the
return spread is concentrated during rms earnings announcements, suggesting that prices
adjust to the information content of calendar revisions at the time earnings are announced,
rather than at the time of the revisions. The central implication of these ndings is that
investors fail to fully unravel information embedded in the schedule of news releases.
Appendix A. Example of Firm-Initiated Calendar Revisions
This appendix uses the change in the expected earnings announcement date of Oracle
Corporation (ORCL) in March of 2010 as an example of a rm-initiated calendar revision.
On March 3, 2010, Oracle issued a press release with the following information:
In the prior year, Oracle announced its 2009 third quarter earnings on March 18th, 2009
(the third Thursday of the month) and, prior to the above press release, the earnings calendar
forecasted Oracles 2010 third quarter expected earnings announcement date as March 17,
2010 (also the third Thursday of the month).
In response to Oracles press release, Wall Street Horizon revised the expected announce-
ment date to March 25, 2010. Because the calendar revision resulted from the rm explicitly
indicating when they intend to announce earnings, I categorize the revision as being rm-
initiated (or, equivalently, conrmed). The calendar data appears as follows:
Date Ticker Fyear FQtr NextDt etype
3/1/2010 ORCL 2010 3 3/17/2010 T
3/2/2010 ORCL 2010 3 3/17/2010 T
3/3/2010 ORCL 2010 3 3/17/2010 T
3/4/2010 ORCL 2010 3 3/25/2010 V
3/5/2010 ORCL 2010 3 3/25/2010 V
3/8/2010 ORCL 2010 3 3/25/2010 V
Note that the press release is issued on March 3, 2010 but is recorded one day later
in the March 4, 2010 earnings calendar data, which Wall Street Horizon dissiminated by
4am ET on March 4. Accompanying this date change, the etype column of the calendar
data changes from T, indicating it was unconrmed, to V, indicating it was based on
information directly conveyed by the rm regarding when they intend to announce earnings.
In this example, my measure of the calendar revision, REV , equals negative six because
Oracles press release caused the expected announcement date in the earnings calendar to
shift back by six trading days. Because the value of REV is greater than one week, Oracles
revision is assigned an R-Score of zero and Oracle would be treated as a delayer.
References
Abarbanell, J., Lehavy, R., 2003. Biased forecasts or biased earnings? the role of reported
earnings in explaining apparent bias and over/underreaction in analysts earnings forecasts.
Journal of Accounting and Economics 36, 105146.
Acharya, V.V., DeMarzo, P., Kremer, I., 2011. Endogenous information ows and the
clustering of announcements. American Economic Review 101, 29552979.
Bagnoli, M., Kross, W., Watts, S.G., 2002. The information in managements expected
earnings report date: A day late, a penny short. Journal of Accounting Research 40,
12751296.
Chambers, A.E., Penman, S.H., 1984. Timeliness of reporting and the stock price reaction
to earnings announcements. Journal of Accounting Research 2147.
Cohen, L., Diether, K., Malloy, C., 2013. Misvaluing innovation. Review of Financial Studies
Forthcoming.
Da, Z., Engelberg, J., Gao, P., 2011. In search of attention. Journal of Finance 66, 14611499.
Damodaran, A., 1989. The weekend eect in information releases: A study of earnings and
dividend announcements. Review of Financial Studies 2, 607623.
Duarte-Silva, T., Fu, H., Noe, C.F., Ramesh, K., 2013. How do investors interpret announce-
ments of earnings delays? Journal of Applied Corporate Finance 25, 6471.
Dye, R.A., Sridhar, S.S., 1995. Industry-wide disclosure dynamics. Journal of Accounting
Research 157174.
Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds.
Journal of Financial Economics 33, 356.
Fletcher, M., Mufson, S., 2014. Why did GM take so long to respond to deadly defect?
corporate culture may hold answer. The Washington Post March 30, 2014.
Gennotte, G., Trueman, B., 1996. The strategic timing of corporate disclosures. Review of
Financial Studies 9, 665690.
Giglio, S., Shue, K., 2014. No news is news: Do markets underreact to nothing? Review of
Financial Studies Forthcoming.
Gilson, S.C., 1989. Management turnover and nancial distress. Journal of nancial Eco-
nomics 25, 241262.
Graham, J.R., Harvey, C.R., Rajgopal, S., 2005. The economic implications of corporate
nancial reporting. Journal of Accounting and Economics 40, 373.
Grossman, S., Hart, O., 1980. Disclosure laws and takeover bids. Journal of Finance 35,
323334.
Grossman, S.J., 1981. The informational role of warranties and private disclosure about
product quality. Journal of Law and Economics 461483.
Hirshleifer, D., Lim, S., Teoh, S., 2009. Driven to distraction: Extraneous events and
underreaction to earnings news. Journal of Finance 64, 22892325.
Johnson, T., So, E.C., 2014. Earnings announcement premia: The role of asymmetric liq-
uidity provision. Working paper, University of Texas at Austin.
Kothari, S.P., Shu, S., Wysocki, P.D., 2009. Do managers withhold bad news? Journal of
Accounting Research 47, 241276.
Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment, extrapolation, and
risk. Journal of Finance 49, 15411578.
Legg, A., Sweeny, K., 2014. Do you want the good news or the bad news rst? the nature
and consequences of news order preferences. Personality and Social Psychology Bulletin
40, 279288.
Milgrom, P.R., 1981. Good news and bad news: Representation theorems and applications.
Bell Journal of Economics 380391.
Ou, J., Penman, S., 1989. Financial statement analysis and the prediction of stock returns.
Journal of Accounting and Economics 11, 295329.
Penman, S.H., 1984. Abnormal returns to investment strategies based on the timing of
earnings reports. Journal of Accounting and Economics 6, 165183.
Piotroski, J., 2000. Value investing: The use of historical nancial statement information to
separate winners from losers. Journal of Accounting Research 38, 141.
Piotroski, J.D., So, E.C., 2012. Identifying expectation errors in value/glamour strategies:
A fundamental analysis approach. Review of Financial Studies 25, 28412875.
Skinner, D.J., 1994. Why rms voluntarily disclose bad news. Journal of accounting research
3860.
Sloan, R., 1996. Do stock prices fully reect information in accruals and cash ows about
future earnings? Accounting Review 289315.
So, E.C., 2013. A new approach to predicting analyst forecast errors: Do investors overweight
analyst forecasts? Journal of Financial Economics 108, 615640.
So, E.C., Wang, S., 2014. News-driven return reversals: Liquidity provision ahead of earnings
announcements. Journal of Financial Economics Forthcoming.
Zmijewski, M.E., 1984. Methodological issues related to the estimation of nancial distress
prediction models. Journal of Accounting Research 5982.
Figure 1. Cumulative Returns Around Earnings Announcements
Panel A contains the average spread in cumulative returns across high and low R-Score portfolios in event-time in the month
surrounding expected earnings announcement dates. Firms are assigned to R-Score portfolios on date r using conrmed
revisions in their expected expected announcement dates, REV , dened as the number of days between a conrmed expected
announcement date and the immediately preceding expected announcement date. R-Score equals 0 for rms with REV <-5
and 1 for rms with REV >5. Firms in the highest R-Score portfolio are deemed advancers and rms in the lowest R-Score
portfolio are deemed delayers. In Panel A, the value on day d equals the average cumulative return spread from day t-10 to
day d. Panel B contains the average cumulative, market-adjusted returns to high and low R-Score rms. The sample consists
of 18,959 rm-quarters spanning 2006 through 2013 in which rms issued a conrmed earnings announcement date at least two
weeks prior to their expected announcement date.
Figure 2. Quarterly Strategy Returns
The gure contains the average spread in returns across high and low R-Score rms (i.e., advancers vs. delayers) each calendar
quarter. Firms are assigned to R-Score portfolios on date r using conrmed revisions in their expected expected announcement
dates, REV , dened as the number of days between a conrmed expected announcement date and the immediately preceding
expected announcement date. R-Score equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with
-2REV 2; 0.75 for rms with 3REV 5; and 1 for rms with REV >5. Firms in the highest R-Score portfolio are deemed
advancers and rms in the lowest R-Score portfolio are deemed delayers. RET(r+1,r+21), shown in black bars, equals the
market-adjusted return over the month following the calendar revision. RET(t-1,t+1), shown in grey bars, equals the three-
day return surrounding the expected earnings announcement date t. The sample consists of 18,959 rm-quarters spanning
2006 through 2013 in which rms issued a conrmed earnings announcement date at least two weeks prior to their expected
announcement date.
Table 1. Descriptive statistics
Panel A presents annual descriptive statistics of the main variables used throughout the paper. N equals the number of rm-
quarters and Firms indicates the number of unique rms. HORIZON equals the number of days between the date on which
a rm issues a conrmed date and the expected announcement date. REV equals the number of days between a conrmed
expected announcement date and the immediately preceding expected announcement date. |REV | equals the absolute value of
REV . DEV equals the number of days between the conrmed expected announcement date and the actual announcement date.
|DEV | equals the absolute value of DEV . Panel B contains descriptive statistics across R-Score portfolios. Firms are assigned
to R-Score portfolios on date r using conrmed revisions in their expected expected announcement dates, REV , dened as
the number of days between a conrmed expected announcement date and the immediately preceding expected announcement
date. R-Score equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with -2REV 2; 0.75 for
rms with 3REV 5; and 1 for rms with REV >5. OBS denotes the average number of rm-quarters within each portfolio.
MCAP equals rms market capitalization reported in millions. MOMEN is the cumulative market-adjusted return over the
prior 12-months ending on r-11. RET(r-1,r+1) is the three-day return surrounding the calendar revision. Reported t-statistics
are based on the dierence in high and low R-Score portfolios over the time-series of calendar quarters. The sample consists of
18,959 rm-quarters spanning 2006 through 2013 in which rms issued a conrmed earnings announcement date at least two
weeks prior to their expected announcement date.
Panel A: Summary Statistics by Year
N Firms HORIZON REV |REV | DEV |DEV |
2006 1,926 1,357 15.069 -2.965 5.372 0.080 0.287
2007 2,536 1,652 16.123 -1.647 4.744 0.062 0.265
2008 3,031 1,843 16.747 -1.638 4.551 0.021 0.274
2009 2,402 1,524 16.622 -1.993 4.635 -0.001 0.248
2010 2,146 1,418 16.224 -1.137 4.754 -0.017 0.150
2011 2,076 1,378 16.205 -1.665 4.564 0.013 0.211
2012 2,395 1,495 16.250 -1.857 4.466 0.030 0.194
2013 2,447 1,528 16.145 -1.884 4.620 -0.094 0.182
Panel B: Descriptive Statistics by R-Score Portfolios
R-Score REV OBS HORIZON MCAP MOMEN DEV RET(r-1,r+1)
0 (Delay) -8.849 84.6 16.495 3,579 -8.525 -0.040 -0.230
0.25 -4.158 248.8 16.434 5,272 -2.871 -0.058 -0.180
0.5 -0.019 134.1 15.894 5,275 0.579 -0.052 -0.052
0.75 4.118 83.8 15.442 4,106 1.404 0.095 -0.081
1 (Advance) 8.836 41.2 15.487 3,025 6.311 0.247 -0.047
Advance-Delay 17.685 -43.5 -1.009 -554 14.835 0.289 0.183
t-statistic -(9.16) -(4.17) -(1.40) (8.62) (4.45) (1.52)
Table 2. Protability and Earnings Surprises
Panel A contains average earnings metrics (shown as percentages) across R-Score portfolios. Firms are assigned to R-Score
portfolios on date r using conrmed revisions in their expected expected announcement dates, REV , dened as the number of
days between a conrmed expected announcement date and the immediately preceding expected announcement date. R-Score
equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with -2REV 2; 0.75 for rms with 3REV 5;
and 1 for rms with REV >5. ROA is the rms return on assets dened as net income scaled by beginning-of-quarter total
assets and ROA<0 equals one for rms with negative ROA. ROA equals same-quarter annual change in ROA and ROA<0
equals one for rms with annual decreases in ROA. SURP equals the actual EPS number reported in IBES minus the last
consensus forecast available immediately prior to the announcement, and scaled by beginning-of-quarter assets, and SURP<0
equals one for rms with negative SURP. Reported t-statistics are based on the dierence in high and low R-Score portfolios
over the time-series of calendar quarters. Panel B contains regression results of earnings metrics on R-Score and additional
rm controls. LBM and SIZE are the log of one plus the book-to-market ratio and log of market capitalization, respectively.
MOMEN is the cumulative market-adjusted return and V LTY is the standard deviation of returns over the prior 12-months
ending on r-11. The reported t-statistics are based on two-way cluster robust standard errors, clustered by rm and quarter.
***, **, and * indicate signicance at the 1, 5, and 10% level, respectively. Industry xed eects are based on two-digit SIC
codes. The sample consists of 18,959 rm-quarters spanning 2006 through 2013 in which rms issued a conrmed earnings
announcement date at least two weeks prior to their expected announcement date.
Panel A: Earnings Metrics by R-Score Portfolios
R-Score ROA ROA<0 ROA ROA<0 SURP SURP<0
0 (Delay) -0.316 0.418 -0.824 0.600 -0.028 0.374
0.25 0.507 0.338 -0.254 0.549 0.055 0.319
0.5 0.636 0.312 -0.059 0.501 0.058 0.310
0.75 0.394 0.353 0.285 0.479 0.116 0.287
1 (Advance) 0.525 0.377 0.854 0.423 0.124 0.283
Advance-Delay 0.841 -0.041 1.678 -0.177 0.152 -0.091
t-statistic (4.80) -(2.79) (6.08) -(9.70) (4.95) -(6.58)
Panel B: Regression Results of Earnings Metrics
ROA ROA SURP
(1) (2) (3) (4) (5) (6)
R-Score 0.481*** 0.439*** 1.326*** 1.162*** 0.128*** 0.118***
(3.79) (3.62) (5.58) (5.07) (5.59) (5.10)
SIZE 0.628*** 0.420*** 0.065 0.112*** 0.016 0.024***
(12.46) (10.00) (1.04) (2.97) (1.61) (2.82)
LBM -0.588** -0.226 -0.997*** -0.649*** -0.304*** -0.205***
(-2.72) (-1.04) (-6.31) (-3.20) (-8.35) (-5.73)
MOMEN 0.013*** 0.016*** 0.001***
(6.51) (10.58) (4.34)
V LTY -0.581*** 0.197* -0.001
(-8.85) (1.78) (-0.05)
R
2
0.074 0.110 0.015 0.040 0.015 0.016
Year FE? Y Y Y Y Y Y
Industry FE? N Y N Y N Y
Table 3. Equal- and Value-Weighted Future Returns
This table contains average equal- and value-weighted returns to each R-Score portfolio. All returns are shown as percentages.
Firms are assigned to R-Score portfolios on date r using conrmed revisions in their expected expected announcement dates,
REV , dened as the number of days between a conrmed expected announcement date and the immediately preceding expected
announcement date. R-Score equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with -2REV 2;
0.75 for rms with 3REV 5; and 1 for rms with REV >5. The reported returns are calculated over the month following the
calendar revision from r+1 to r+21. The rst two columns contain raw and market-adjusted returns denoted as RR(r+1,r+21)
and RET(r+1,r+21), respectively. SAR(r+1,r+21) refers to size-adjusted returns dened as the rms raw return minus the
rms decile-matched return as the benchmark return. FAR(r+1,r+21) refers to factor-adjusted returns dened as the rms raw
return minus the return calculated by estimating a rms daily sensitivity to the market (MKTRF), small-minus-big (SMB),
high-minus-low (HML), and up-minus-down momentum (UMD) factors over the year prior to the earnings announcement and
applying those sensitivities to the contemporaneous factors, following Fama and French (1993). Reported t-statistics are based
on the dierence in high and low R-Score portfolios over the time-series of calendar quarters. The sample consists of 18,959
rm-quarters spanning 2006 through 2013 in which rms issued a conrmed earnings announcement date at least two weeks
prior to their expected announcement date.
Panel A: Equal-Weighted Returns in Month Following Revision
R-Score RR(r+1,r+21) RET(r+1,r+21) SAR(r+1,r+21) FAR(r+1,r+21)
0 (Delay) -0.773 -1.304 -0.743 -1.165
0.25 0.173 -0.677 -0.267 -0.300
0.5 0.750 -0.094 0.335 0.319
0.75 1.365 0.189 0.656 0.607
1 (Advance) 2.263 1.384 1.896 1.627
Advance-Delay 3.036 2.688 2.639 2.792
t-statistic (4.59) (4.15) (3.95) (5.17)
Panel B: Value-Weighted Returns in Month Following Revision
R-Score RR(r+1,r+21) RET(r+1,r+21) SAR(r+1,r+21) FAR(r+1,r+21)
0 (Delay) 0.584 -0.496 -0.247 0.098
0.25 0.618 -0.639 -0.185 0.086
0.5 0.695 -0.657 -0.260 0.246
0.75 1.148 -0.490 0.065 0.219
1 (Advance) 2.745 1.261 1.716 1.624
Advance-Delay 2.162 1.757 1.962 1.526
t-statistic (2.13) (1.95) (2.07) (2.01)
Table 4. Returns Around Earnings Announcements
This table contains market-adjusted returns around earnings announcements across R-Score portfolios. All returns are shown
as percentages. Firms are assigned to R-Score portfolios on date r using conrmed revisions in their expected expected an-
nouncement dates, REV , dened as the number of days between a conrmed expected announcement date and the immediately
preceding expected announcement date. R-Score equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for
rms with -2REV 2; 0.75 for rms with 3REV 5; and 1 for rms with REV >5. RET(t+X,t+Y) equals the cumulative
market-adjusted return from day X to Y relative to the expected earnings announcement date t. Reported t-statistics are based
on the dierence in high and low R-Score portfolios over the time-series of calendar quarters. The sample consists of 18,959
rm-quarters spanning 2006 through 2013 in which rms issued a conrmed earnings announcement date at least two weeks
prior to their expected announcement date.
R-Score RET(t-10,t+10) RET(t-10,t-2) RET(t-1,t+1) RET(t+2,t+10) RET(t-10,t+1)
0 (Delay) -0.911 0.078 -0.707 -0.294 -0.675
0.25 -0.323 -0.175 -0.083 -0.081 -0.288
0.5 -0.016 -0.105 0.149 -0.046 0.021
0.75 0.412 0.044 0.627 -0.253 0.656
1 (Advance) 1.623 0.616 0.870 0.198 1.403
Advance-Delay 2.534 0.538 1.578 0.493 2.077
t-statistic (4.36) (1.77) (4.85) (2.13) (4.17)
Table 5. Cross-Sectional Return Regressions
This table contains results from regressing future returns on R-Score and additional controls. R-Score is measured on date
r using conrmed revisions in rms expected expected announcement dates, REV , dened as the number of days between
a conrmed expected announcement date and the immediately preceding expected announcement date. R-Score equals 0 for
rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with -2REV 2; 0.75 for rms with 3REV 5; and
1 for rms with REV >5. RET(r+X,r+Y) equals the cumulative market-adjusted return from day X to Y relative to the
calendar revision date. Similarly, RET(t+X,t+Y) equals the cumulative market-adjusted return from day X to Y relative to
the expected earnings announcement date t. LBM and SIZE are the log of one plus the book-to-market ratio and log of
market capitalization, respectively. MOMEN is the cumulative market-adjusted return and V LTY is the standard deviation
of returns over the prior 12-months ending on r-11. The reported t-statistics are based on two-way cluster robust standard
errors, clustered by rm and quarter. ***, **, and * indicate signicance at the 1, 5, and 10% level, respectively. Industry xed
eects are based on two-digit SIC codes. The sample consists of 18,959 rm-quarters spanning 2006 through 2013 in which
rms issued a conrmed earnings announcement date at least two weeks prior to their expected announcement date.
RET(t-1,t+1) RET(r+1,r+21)
(1) (2) (3) (4) (5) (6)
R-Score 1.520*** 1.499*** 1.522*** 2.404*** 2.336*** 2.351***
(6.66) (6.95) (6.82) (4.77) (4.80) (4.89)
SIZE -0.067 0.007 -0.038 0.017 -0.009 0.019
(-1.13) (0.12) (-0.55) (0.11) (-0.06) (0.12)
LBM -0.353* -0.182 -0.207 -0.280 -0.228 -0.120
(-1.79) (-0.92) (-1.04) (-0.34) (-0.30) (-0.16)
MOMEN 0.004* 0.003 0.004* 0.001 0.000 0.001
(1.95) (1.55) (1.87) (0.26) (0.09) (0.09)
V LTY -0.281** -0.140 -0.272** -0.140 -0.136 -0.069
(-2.40) (-1.33) (-2.18) (-0.30) (-0.35) (-0.14)
R
2
0.004 0.003 0.004 0.003 0.002 0.002
Year FE? Y N Y Y N Y
Industry FE? N Y Y N Y Y
Table 6. Interaction Eects
This table contains results from regressing RET(r+1,r+21) on R-Score and additional controls. Firms are assigned to R-Score
portfolios on date r using conrmed revisions in their expected expected announcement dates, REV , dened as the number of
days between a conrmed expected announcement date and the immediately preceding expected announcement date. R-Score
equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with -2REV 2; 0.75 for rms with 3REV 5;
and 1 for rms with REV >5. RET(r+1,r+21) equals the cumulative market-adjusted return in the month (21 trading days)
following the calendar revision. 1(Small Firm) is an indicator variable that equals one if the rm is in the lowest tercile of
market capitalization within a given calendar quarter, 1(Low Coverage) is an indicator variable that equals one if the rm is in
the lowest tercile of analyst coverage within a given calendar quarter, and 1(High Distress) is an indicator variable that equals
one if the rm is in the lowest tercile of the Zmijewski (1984) Z-Score nancial distress measure within a given calendar quarter.
1(Buy Recommendation) is an indicator variable that equals one if the rm has an outstanding BUY recommendation in the
IBES consensus database. The reported t-statistics are based on two-way cluster robust standard errors, clustered by rm and
quarter. ***, **, and * indicate signicance at the 1, 5, and 10% level, respectively. Year and industry xed eects are included
throughout. Industry xed eects are based on two-digit SIC codes. The sample consists of 18,959 rm-quarters spanning
announcement date.
(1) (2) (3) (4) (5) (6)
R-Score 2.412*** 1.308*** 1.754*** 2.298*** 1.926*** 0.575
(4.81) (2.77) (2.92) (4.79) (4.38) (1.24)
R-Score *1(Small Firm) 2.988*** 2.307***
(3.58) (3.08)
1(Small Firm) -1.387** -1.109***
(-2.27) (-3.47)
R-Score *1(Low Coverage) 2.021** 1.497*
(2.35) (1.80)
1(Low Coverage) -0.785* -0.526*
(-1.75) (-1.96)
R-Score *1(Buy Recommendation) 3.689* 2.370**
(1.74) (2.09)
1(Buy Recommendation) -0.719 -0.108
(-0.63) (-0.16)
R-Score *1(High Distress) 3.273*** 2.834**
(2.78) (2.28)
1(High Distress) -1.890*** -1.670***
(-2.76) (-4.41)
R-square 0.002 0.003 0.003 0.003 0.003 0.004
Table 7. Benchmarking to Random-Walk Expected Announcement Dates
Panel A contains reported earnings metrics and returns across ve RW-Score portfolios based on calendar revisions, where
revisions in expected announcement dates are dened relative to rms historical reporting patterns. For each rm-quarter, I
calculate the expected announcement date by using a random-walk expected announcement date that equals the same date
from the prior year. Using these random-walk expected announcement dates as the benchmark, rms are assigned to RW-Score
portfolios on date r using conrmed revisions in their expected expected announcement dates, REV
RW
, dened as the number
of days between a conrmed expected announcement date and the historically estimated random-walk expected announcement
dates. RW-Score equals 0 for rms with REV
RW
<-5; 0.25 for rms with -5REV
RW
-3; 0.5 for rms with -2REV
RW
2;
0.75 for rms with 3REV
RW
5; and 1 for rms with REV
RW
>5. ROA is the rms return on assets dened as net income
scaled by beginning-of-quarter total assets. ROA equals same-quarter annual change in ROA. SURP equals the actual
EPS number reported in IBES minus the last consensus forecast available immediately prior to the announcement, and scaled
by beginning-of-quarter assets. OBS denotes the average number of rm-quarters within each portfolio. RET(r+1,r+21)
equals the market-adjusted return over the month following the calendar revision. RET(t-1,t+1) equals the three-day return
surrounding the expected earnings announcement date t. All returns are shown as percentages. Reported t-statistics are
based on the dierence in high and low R-Score portfolios over the time-series of calendar quarters. The sample consists of
15,439 rm-quarters spanning 2006 through 2013 with a random-walk forecasts from the prior year and that issue conrmed
earnings announcement dates at least two weeks prior to their expected announcement date. Panel B contains descriptive
statistics of RET(r+1,r+21) across RW-Score portfolios, when the sample is partitioned across rms with above and below
median announcement date consistency. I dene announcement date consistency using the historical standard deviation of the
dierence between random-walk forecasts and rms actual earnings announcement dates over the prior ten quarters, where
lower standard deviations indicate greater consistency. Reported t-statistics are based on the dierence in high and low RW-
Score portfolios over the time-series of calendar quarters. The sample for the analysis in Panel B consists of 15,024 rm-quarters
spanning 2006 through 2013 with at least ve prior quarters to calculate announcement date consistency. Reported t-statistics
are based on the dierence in high and low RW-Score portfolios over the time-series of calendar quarters.
Panel A: Revisions Benchmarked to Random-Walk Dates
RW-Score ROA ROA SURP RET(r+1,r+21) RET(t-1,t+1)
0 (Delay) -0.410 -0.887 -0.089 -1.169 -0.638
0.25 0.525 -0.244 0.048 -0.635 -0.131
0.5 0.674 -0.132 0.044 -0.489 -0.043
0.75 0.398 0.189 0.102 0.305 0.476
1 (Advance) 0.177 0.418 0.094 1.039 0.791
High-Low 0.587 1.305 0.183 2.208 1.429
t-statistic (2.63) (6.61) (4.42) (3.79) (3.27)
Panel B: Conditioning on Historical Consistency
High Consistency Low Consistency
RW-Score N=7,590 N=7,434
0 (Delay) -0.209 -2.007
0.25 -0.486 -0.919
0.5 -0.354 -0.471
0.75 0.852 0.009
1 (Advance) 3.397 -0.418
Advance-Delay 3.606 1.589
t-statistic (2.95) (1.88)
Table 8. Alternative Portfolio Assignments of Earnings Calendar Revisions
This contains average earnings metrics and future returns across earnings calendar revision (REV ) portfolios. REV is dened as
the number of days between a conrmed expected announcement date and the immediately preceding expected announcement
date. In Panel A, rms are assigned to REV quintiles using the distributional breakpoints from the prior quarter, where higher
(lower) values are assigned to quintile 1 (0). In Panel B, rms are analogously assigned to terciles ranging from 1 to 0. ROA
is the rms return on assets dened as net income scaled by beginning-of-quarter total assets. ROA equals same-quarter
annual change in ROA. SURP equals the actual EPS number reported in IBES minus the last consensus forecast available
immediately prior to the announcement, and scaled by beginning-of-quarter assets. RET(r+1,r+21) equals the market-adjusted
return over the month following the calendar revision. RET(t-1,t+1) equals the three-day return surrounding the expected
earnings announcement date t. All returns are shown as percentages. Reported t-statistics are based on the dierence in high
and low R-Score portfolios over the time-series of calendar quarters. The sample consists of 18,959 rm-quarters spanning
announcement date.
Panel A: Earnings Metrics and Returns by REV Quintiles
ROA ROA SURP RET(r+1,r+21) RET(t-1,t+1)
Q1 (Low REV) -0.316 -0.851 -0.033 -1.376 -0.704
Q2 0.522 -0.261 0.073 -0.784 -0.168
Q3 0.477 -0.266 0.042 -0.720 0.047
Q4 0.590 -0.046 0.050 -0.124 0.100
Q5 (High REV) 0.441 0.415 0.092 0.430 0.569
High-Low 0.757 1.266 0.125 1.806 1.273
t-statistic (4.30) (5.26) (3.00) (4.07) (4.44)
Panel B: Earnings Metrics and Returns by REV Terciles
ROA ROA SURP RET(r+1,r+21) RET(t-1,t+1)
T1 (Low REV) -0.079 -0.670 0.016 -1.117 -0.441
T2 0.470 -0.250 0.039 -0.575 -0.091
T3 (High REV) 0.494 0.294 0.088 0.326 0.496
High-Low 0.573 0.964 0.072 1.443 0.937
t-statistic (3.51) (4.53) (2.00) (3.20) (4.45)
Table 9. Initiated Versus Non-Initiated Revisions
This table contains average equal-weighted returns to each R-Score portfolio for a sample that includes both rm-initiated
revisions in announcement dates as well as non-initiated revisions. Specically, the sample consists of 30,265 rm-quarters
spanning 2006 through 2013 where the earnings announcement dates was revised at least two weeks prior to their expected
announcement date. The non-initiated sample consists of those calendar revisions where the rm did not conrm the earnings
calendar change. Firms are assigned to R-Score portfolios on date r using revisions in their expected expected announcement
dates, REV , dened as the number of days between a revised expected announcement date and the immediately preceding
expected announcement date. R-Score equals 0 for rms with REV <-5; 0.25 for rms with -5REV -3; 0.5 for rms with
-2REV 2; 0.75 for rms with 3REV 5; and 1 for rms with REV >5. The reported returns are calculated over the month
following the calendar revision from r+1 to r+21. The rst two columns contain raw and market-adjusted returns for the full
sample. The last two columns contain raw and market-adjusted returns for non-initiated revisions in the earnings calendar. All
returns are shown as percentages. Reported t-statistics are based on the dierence in high and low R-Score portfolios over the
time-series of calendar quarters.
All: N=30,265 Non-Conrmed: N=12,728
R-Score RET(r+1,r+21) RET(t-1,t+1) RET(r+1,r+21) RET(t-1,t+1)
0 (Delay) -1.225 -0.549 -0.919 -0.297
0.25 -0.600 0.022 -0.499 0.142
0.5 -0.389 0.043 -0.700 -0.151
0.75 0.195 0.421 0.011 -0.093
1 (Advance) 1.036 0.605 0.682 0.527
Advance-Delay 2.261 1.154 1.601 0.824
t-statistic (3.47) (3.46) (1.54) (1.95)

Time Will Tell

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Time Will Tell

Uploaded by

Copyright:

Available Formats

Time Will Tell: Information in

the Timing of Scheduled Earnings News

Massachusetts Institute of Technology

observed on announcement date t

0 (Delay) for REV < 5

You might also like