Professional Documents
Culture Documents
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)).
Time Will Tell: Information in the Timing of Scheduled Earnings News 2
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.
Time Will Tell: Information in the Timing of Scheduled Earnings News 3
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-
Time Will Tell: Information in the Timing of Scheduled Earnings News 4
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.
Time Will Tell: Information in the Timing of Scheduled Earnings News 5
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.
Time Will Tell: Information in the Timing of Scheduled Earnings News 6
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.
Time Will Tell: Information in the Timing of Scheduled Earnings News 7
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