Description
This study aims to test the effects of forecast specificity on the asymmetric short-window
share market response to management earnings forecasts (MEF).
Accounting Research Journal
The effects of forecast specificity on the asymmetric short-window share market
response to management earnings forecasts
Howard Chan Robert Faff Yee Kee Ho Alan Ramsay
Article information:
To cite this document:
Howard Chan Robert Faff Yee Kee Ho Alan Ramsay, (2009),"The effects of forecast specificity on the
asymmetric short-window share market response to management earnings forecasts", Accounting
Research J ournal, Vol. 22 Iss 3 pp. 237 - 261
Permanent link to this document:http://dx.doi.org/10.1108/10309610911005572
Downloaded on: 24 January 2016, At: 21:09 (PT)
References: this document contains references to 37 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 1977 times since 2009*
Users who downloaded this article also downloaded:
Haiyan J iang, Ahsan Habib, (2009),"The impact of different types of ownership concentration on annual
report voluntary disclosures in New Zealand", Accounting Research J ournal, Vol. 22 Iss 3 pp. 275-304http://dx.doi.org/10.1108/10309610911005590
Parulian Silaen, Robert Williams, (2009),"Management control systems: a model for R&D units",
Accounting Research J ournal, Vol. 22 Iss 3 pp. 262-274http://dx.doi.org/10.1108/10309610911005581
Ning Du, J ohn E. McEnroe, Kevin Stevens, (2014),"The joint effects of management incentive and
information precision on perceived reliability in fair value estimates", Accounting Research J ournal, Vol. 27
Iss 2 pp. 188-206http://dx.doi.org/10.1108/ARJ -10-2012-0081
Access to this document was granted through an Emerald subscription provided by emerald-srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The effects of forecast speci?city
on the asymmetric short-window
share market response to
management earnings forecasts
Howard Chan
Department of Finance, The University of Melbourne, Melbourne, Australia
Robert Faff
Department of Accounting and Finance, Monash University, Melbourne, Australia
Yee Kee Ho
Department of Finance and Accounting, National University of Singapore,
Singapore, and
Alan Ramsay
Department of Accounting and Finance, Monash University, Melbourne, Australia
Abstract
Purpose – This study aims to test the effects of forecast speci?city on the asymmetric short-window
share market response to management earnings forecasts (MEF).
Design/methodology/approach – The paper examines a large sample of hand-checked Australian
data over the period 1994 to 2001. Using an analyst news benchmark, it estimates a series of regressions
to investigate whether the short-term impact from bad news announcements is greater in magnitude
than from good news announcements and whether this differs between routine and non-routine MEFs.
Additionally, it examines whether (after controlling for news content of MEF) there is a differential
market impact conditional on speci?city: minimum versus maximum versus range versus point.
Findings – The results indicate that an asymmetric response is evident for the overall sample and a
sub-set of non-routine forecasts. Contrary to predictions, the results show that forecast speci?city,
minimum, maximum, range and point MEFs make no additional contribution to the differences in the
market reaction to bad or good news.
Originality/value – The study extends the research investigating the short-run market impact of
MEFs. The main element of innovation derives from the interaction between speci?city and news
content, as well as distinguishing between routine versus non-routine cases. Notably, it found little
support for the view that more speci?c forecasts elicit greater market responses. What the results do
suggest is that managers appear to choose the form of the forecast to suit the news being delivered. In
particular, bad news delivered in a minimum forecast seems to be ignored by the market.
Keywords Australia, Financial forecasting, Investors
Paper type Research paper
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1030-9616.htm
TheauthorswouldliketothankPeter Clarkson, Peter Easton, AbeHerzberg, GregPoundandPhilip
Sinnadurai andworkshopparticipants at the AAA2005Conference at SanFrancisco, the AFAANZ
2004 Conference at Alice Springs, Monash University, University of Auckland, University of
Adelaide, and University of Melbourne for comments on earlier drafts. They also thank Alex
Featherstone for researchassistance. Financial assistance providedbyMonashUniversityFaculty
of Business and Economics research grant is gratefully acknowledged. The ?rst author gratefully
acknowledges ?nancial assistance provided by a Monash University Research Fellowship.
Effects of
forecast
speci?city
237
Accounting Research Journal
Vol. 22 No. 3, 2009
pp. 237-261
qEmerald Group Publishing Limited
1030-9616
DOI 10.1108/10309610911005572
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
1. Introduction
We investigate the share price impact of management earnings forecasts (hereafter,
MEFs). Speci?cally, we document and analyse whether there is a differential (and
greater in absolute magnitude) response to bad news MEFs as compared to good news
MEFs given the different possible types of MEFs. Moreover, our main focus is to
investigate the effect of alternative degrees of speci?city in MEFs (point, range,
minimum, maximum) on the asymmetric share market response to bad news compared
to good news MEFs[1]. We designate this the “conditional speci?city” hypothesis. It
extends the “news asymmetry” hypothesis that has been previously documented for
US markets.
For situations in which markets are incomplete and investors have behavioural
biases (Daniel et al., 1998), it is not unreasonable to assume that the market may react
asymmetrically to different types of MEFs or forecast speci?city[2]. The main prior
research in this area (Baginski et al., 1993 and Pownall et al., 1993) relies on the
expectations adjustment hypothesis (Ajinkya and Gift, 1984) which argues for a
symmetric relation between good and bad news MEFs and security prices. However,
Zhang (2006) stresses the importance of conditioning tests of market reaction on the
news content of the information. The existence of and reasons for asymmetric response
to MEFs have recently been investigated by Kothari et al. (2007)[3]. Skinner (1994)
provides limited evidence on the effect of forecast speci?city on the asymmetric share
market response to bad news MEFs. However, his results are inconclusive. Thus the
effect of forecast speci?city on the market response to MEFs remains an open question.
The effect of forecast speci?city on the share market response to the unexpected
component of MEFs is important from the perspective of information theory. Kim and
Verrecchia (1991) develop a theoretical model, which predicts that the extent of price
change at the time of a public announcement (for example, a MEF) is proportional to
both the unexpected portion of the announcement and the precision of the information
contained in the announcement. Baginski et al. (1993) test this proposition using the
share market reaction to MEFs that vary in speci?city. They ?nd evidence consistent
with the Kim and Verrecchia prediction, that the share price reaction is a positive
function of forecast speci?city. However, the Baginski et al. (1993) analysis does not
condition the market response on the types of the news contained in the MEFs. Further,
it has been shown that the method used by both Baginski et al. (1993) and Pownall et al.
(1993) to determine the news component of minimum and maximum forecasts is
mis-speci?ed. We contribute to the literature in the area of assessing the news content
of MEFs by using the deviation from analysts’ forecasts as a measure of the
unexpected portion of a public announcement. We also contribute by using a
regression model that allows all four forms of forecast speci?city to interact with both
good and bad news. Using this framework, we test the Kim and Verrecchia (1991)
proposition that forecast speci?city is positively related to market reaction.
From the perspective of disclosure policy, the effect of forecast speci?city on the
share market response is also important. Managers must choose not only whether to
make a MEF but the form (speci?city) of the MEF. Evidence indicates that managers’
forecast speci?city choices are related to the news content of the forecast. Baginski et al.
(1994), Skinner (1994) and Kasznik and Lev (1995) document that forecast news content
and forecast speci?city are related. For example, maximum (upper bound) forecasts are
usually associated with bad news, while minimum (lower bound) forecasts are usually
ARJ
22,3
238
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
associated with good news. Libby et al. (2006, p. 208) argue that “. . . psychology
research suggests a number of possible interactive effects of guidance errors and
guidance form” and their consequential impact on actual earnings announcement and
we will extend this to share price responses. Therefore, managers face a correlated
challenge in releasing a MEF.
Australia provides a different institutional and reporting environment to examine
the asymmetric share prices effect of MEFs and the interactions of forecast speci?city
and news content. First, there is a longer period between routine earnings
announcements as compared to the USA since it has a quarterly reporting
requirement while Australia has a half-yearly reporting requirement for listed
companies. As a result, management earnings forecasts should have a greater role in
informing the market than in the USA since the Australian rule requires a less frequent
routine reporting. Second, the continuous disclosure requirements[4] and reporting to
the Australian Stock Exchange (ASX) results in a central depository of announcements
recorded in the ASX Signal G database that is released simultaneously to all investors
by the ASX. Thus, a further feature of our paper is that, we source our MEF data
directly from the of?cial company announcement lodged with the ASX. We examine
the content of announcements to ensure that all earnings-related cases are identi?ed.
Accordingly, our investigation covers an extensive sample of over 1,600 MEFs for the
period September 1994 to December 2001.
The mainthrust of our ?ndings canbe summarisedas follows. We ?ndsupport for the
“news asymmetry” hypothesis in that the absolute magnitude of the market response to
badnewsMEFsissigni?cantlylarger thanthemagnitudeof thepositiveresponsetogood
news MEFs. Thus we con?rm Skinner’s (1994) ?nding and the recent results of Kothari
et al. (2007). We extend the prior literature by showing that news asymmetry is largely
con?ned to forecasts issued in response to the ?rm’s continuous disclosure obligations
(non-routine forecasts) and is not observed as strongly for forecasts accompanying
routine information events (routine forecasts). In addition, we have also found that the
asymmetry continues to persist in different forecast speci?city.
In relation to our “conditional speci?city” hypothesis, our initial univariate analysis
?nds bad news asymmetry for maximum, range and point forecasts but for forecasts in
the form of a minimum, the asymmetry is reversed – there being a signi?cantly larger
response to good news rather than bad news minimum MEFs. After conditioning our
regression analysis on the types of news, the results of our multivariate analysis differ
from those of Baginski et al. (1993) and are contrary to the predictions of the Kim and
Verrecchia (1991) model. Speci?cally, using the analysts forecast news benchmark, for
good news MEFs there is no evidence that forecast speci?city has any differential
effect on the market response. Our results show that while the market responds equally
strongly to bad news MEFs in the form of point, range or maximum style MEFs,
minimum style MEFs containing bad news are essentially ignored by the market. This
effect is observed overall and for both routine and non-routine MEFs. Our ?ndings are
consistent with Pownall et al. (1993) and Hirst et al. (1999).
The results are consistent with the view that market participants focus more on the
news content of non-routine forecasts (news asymmetry is mainly found for
non-routine forecasts) and less on the form of the forecast (no signi?cant conditional
speci?city effects are observed for this group of forecasts). Routine forecasts are
planned information releases and, as such, the market may well have already
Effects of
forecast
speci?city
239
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
anticipated the information contained in these events. For routine forecasts, market
participants seem to focus more on the forecasts form (signi?cant conditional
speci?city effects are found) and less on the forecast news content (little news
asymmetry is found for routine forecasts).
The remainder of the paper is structured as follows: section 2 outlines prior research
leading to the formulation of testable hypotheses. Section 3 outlines our research
method, data and sample, section 4 outlines and discusses our results and section 5
presents a summary and conclusions.
2. Theory development
2.1 Asymmetric market reaction
Our paper is based on two main premises:
(1) The market responds asymmetrically to good and bad news MEFs.
(2) The forecast speci?city of MEFs affects the asymmetric market response to
good and bad news forecasts.
Analytical research on discretionary information disclosure indicates that, in the
absence of costs, full disclosure will occur (Grossman, 1981; Milgrom, 1981). However,
in the presence of costs and given uncertainty about management’s private
information, management may resort to voluntary disclosure to reduce the level of
information asymmetry (Lennox and Park, 2006). On the other hand, the failure by
management to release a MEF does not necessarily imply no news. For example,
Waymire (1985) proposes that managers are unwilling to release good news especially
when earnings volatility is high because of the risk of litigation if actual earnings fall
short of the expected forecast. Whatever the incentives for disclosure, prima facie,
management’s release of a MEF is a major information event and a signi?cant share
market response might be expected.
Early research (for example, Patell, 1976) found that MEFs have information
content in the sense that such forecasts elicit a signi?cant share market response. This
is con?rmed by Pownall et al. (1993) who conclude that MEFs are highly informative
(although less so than earnings announcements). Ajinkya and Gift (1984) hypothesize
that managers issue voluntary MEFs when they believe that the market’s earnings
expectations are signi?cantly out of line with their own earnings expectations. In order
to avoid large movements in share prices at earnings announcement, management is
motivated to voluntarily release a MEF to adjust the market’s expectations more in line
with their proprietary information. Ajinkya and Gift (1984) argue that this motivation
for forecast disclosure applies equally to both good and bad news MEFs.
In relation to voluntary MEFs, Penman (1980) concludes that in an unregulated
market, on average, ?rms with good news are more likely to disclose MEFs than those
with bad news. However, in the light of possible signi?cant market over-reactions and
lawsuits by investors, ?rms are wary of not informing the market about bad news in a
timely manner. Skinner (1994) investigates the voluntary disclosure of bad news MEFs
by smaller US listed companies. He ?nds that voluntary MEFs are more likely to occur
when there are large negative earnings surprises. He argues that, contrary to Ajinkya
and Gift (1984) and Penman (1980), managers behave as if they bear large costs when
investors are surprised by large negative earnings surprises[5]. Skinner (1994)
attributes the increased propensity to preannounce bad news to a reduction in
ARJ
22,3
240
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
litigation costs resulting from the voluntary release of bad earnings news. Investors
who lose money may sue managers for failing to promptly disclose bad earnings news.
Lawsuits may result in personal liability on behalf of managers. A further reason for
voluntary disclosure is that managers incur reputational costs if they fail to disclose
bad news in a timely fashion. Firms that fail to keep the market informed may be
shunned by fund managers and ?nancial analysts. In a related line of argument,
Matsumoto (2002) and Brown and Higgins (2005) report the use of management
forecast guidance to avoid negative earnings surprises.
More recent research has focused on differentiating the share price effects of different
categories or types of forecasts. The issue of whether the share price effects differ between
good news and bad news MEFs is addressed by Skinner (1994). He investigates the
two-day (0, þ1) share price effects to MEFs in smaller US listed ?rms. Skinner de?nes
good and bad news relative to previous year earnings. He predicts and ?nds that the
absolute value of the share price effect is larger for bad news MEFs than for their good
newscounterparts[6]. KasznikandLev(1995) haveasomewhat different focus bylooking
at the issue of whether investors respond differently to large earnings surprises
dependingonwhether the ?rmhas releasedthe informationgradually(issueda warning)
or not. They ?nd a signi?cantly more negative share price effect for ?rms that issue
warnings. Kothari et al. (2007) investigate the asymmetric share price effect of MEFs
using more recent US data (1995-2002). They ?nd strong evidence of a greater reaction to
bad news MEFs. Australian evidence on the share price effect to voluntary MEFs comes
fromGallery et al. (2006). They test the share price effect in the ?ve days (25, þ5) before
and after the release of the MEFs. They ?nd a signi?cant share market response for their
overall sample and for the bad news MEFs, but not for the good news MEFs.
In addition, the market reaction to a MEF may depend on the circumstances
surroundingthe release of the forecast. ASXlistingrules andtheAustralianCorporations
Act require Australian listed companies to continuously disclose information likely to
haveamaterial effect onshare price. MEFscanbereleasedaspart of aroutine information
event (for example the company’s annual general meeting or the mandatory half-yearly
earnings announcement). Onthe other hand, MEFs couldbe standalone andindependent
of the routine reporting of Australian companies. It can be argued that such non-routine
MEFs may be less anticipated by the market and thus may be more likely to exhibit
greater asymmetric reactions to bad news because of greater information content.
Given the previous, we propose the following hypotheses conditional on the context
in which the MEF was released:
H1A. The short-window negative share price effect to bad news MEFs is
signi?cantly larger in absolute magnitude than the positive share price
effect to good news MEFs when the MEF is part of a non-routine
announcement.
H1B. The short-window share price effect to bad or good news non-routine MEFs
is signi?cantly larger in absolute magnitude than the bad or good news
routine MEFs, respectively.
2.2 The effect of forecast speci?city on the asymmetric market reaction
We draw on prior studies which ?nd that management will choose the type of MEFs to
match the precision of their own information about the ?rm’s future (Libby et al., 2006;
Effects of
forecast
speci?city
241
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
King et al., 1998). The effect of forecast speci?city on the share market reaction to
MEFs is investigated by Baginski et al. (1993). In deriving their hypotheses, Baginski
et al. rely on the expectations adjustment hypothesis (Ajinkya and Gift, 1984), which
posits a symmetric relation between management forecast news and security prices.
In terms of the effect of forecast precision, Baginski et al. (1993) also draw on the
Kim and Verrecchia (1991) two-period rational expectation model. Using this model,
Kim and Verrecchia (1991) ?nd that the price change at the time of an announcement
such as a MEF is a function of the unexpected portion of the announcement (“news”)
and the relative importance of the announcement to the prior beliefs of traders. The
relative importance is positively related to the precision of the announcement and
inversely related to the precision of pre-announcement information. Based on this
analysis, Baginski et al. (1993) hypothesise and ?nd evidence that the stock price
response to a MEF is a decreasing function of forecast imprecision.
The arguments and evidence of Baginski et al. (1993) can be criticised on the
grounds that they assume a symmetric market response to good and bad news MEF,
yet subsequent evidence shows a signi?cantly larger reaction to bad news MEFs
(Skinner, 1994; Kothari et al., 2007). Further, the choice of forecast form (minimum,
maximum, range, point) is not independent of news content (Pownall et al., 1993)[7].
The Kim and Verrecchia (1991) model relied on by Baginski et al. (1993) indicates that
the share price effect is initially determined by the unexpected portion of the
announcement, thus providing further support for the need to control for the types of
news. Finally, the method used by Baginski et al. (1993), to determine the unexpected
portion of minimum and maximum forecasts, have been shown to be misspeci?ed
(Pownall et al., 1993, Table 7). These arguments cast doubt on the reliability of the
Baginski et al. (1993) conclusions.
A further source of support for the possible interaction between the form of MEFs
and the precision of the signal and their impact on share prices can be drawn from the
psychology literature. Speci?cally, this literature ?nds a general preference for more
precision over less precision (Kuhn and Budescu, 1996; Kuhn et al., 1999) and the
consequence of a less precise signal is that it results in less extreme reactions (Curley
et al., 1986; Einhorn and Hogarth, 1985).
A review of the empirical research indicates a lack of consistent ?ndings concerning
the relationship between forecast speci?city, types of news from MEFs and
asymmetric share price responses. Pownall et al. (1993) investigate the share price
effects of forecasts that differ by horizon and form (minimum, maximum, range, point;
qualitative forecasts were not included). They conclude that there is no signi?cant
difference in market reaction between minimum, maximum, range and point
quantitative MEFs. This is further supported by the ?ndings in Hirst et al. (1999).
Baginski et al. (1993) investigate whether a more imprecise MEF leads to a more
muted share market response. Consistent with their hypothesis, they ?nd that MEF
imprecision is negatively related to the strength of the market reaction. That is,
minimum and maximum forecasts (less speci?c) produce lower share price responses.
Both Pownall et al. (1993) and Baginski et al. (1993) do not condition their analyses on
the news content of the MEF. Further, both rely on a method of assessing the
unexpected portion of the MEF that amounts to assuming that the lower bound for
minimums and the upper bound for maximums correspond to the mean of the
manager’s expectations. As such, in their method, the fact that management have
ARJ
22,3
242
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
identi?ed an upper or lower bound to their MEF is effectively ignored in determining
the news content of that MEF.
Skinner (1994) investigates the effect of forecast speci?city on the differential share
price response for bad versus good news MEFs. He ?nds a signi?cantly greater share
price response to bad news MEFs for the overall sample and for point forecasts, but no
signi?cant difference for bad versus good news MEFs for range, lower/upper bound or
qualitative forecasts. Small sample sizes (between 34 and 67 observations within each
of his ?ve forecast categories) may have resulted in low power for his statistical tests.
There is a compelling argument that point forecasts are the most precise and thus they
should evoke the strongest reactions – stronger than range or minimum or maximum
forecast (Libby et al., 2006; see Wallsten et al., 1986; and Highhouse, 1994; from the
psychology literature). Skinner (1994) does not test for differences in response across
different forecast speci?city categories. Thus the share price effects of forecast
speci?city on the asymmetric response to MEFs remain an open question.
Given the lack of strong theory and the lack of consistent empirical results in the
prior literature, we hypothesize H2, a “conditional speci?city” hypothesis, in null form:
H2. After controlling for the news content of the MEF, there is no difference in
the short-window share price response to different forms (minimum,
maximum, range, point) of MEFs.
On the presumption that point forecasts are more precise in comparison to range,
minimum and maximum forecast, the following is an alternative hypothesis to H2:
H2A. After controlling for the news content of the MEF, the short-window share
price response to point MEFs is greater than the share price response for
minimum, maximum or range MEFs.
3. Research method
3.1 Data and sample
Our sample comprises of Australian listed companies that are followed by analysts,
over the period September 1994 to December 2001. Our primary motivation for
imposing the analyst coverage restriction is that the central core of our analysis
focuses on using analysts’ earnings forecasts as a benchmark for calibrating the types
of news for the MEFs[8]. Analyst earnings forecast data are sourced from the
Institutional Broker Estimates System (I/B/E/S) summary earnings ?le. MEFs data are
sourced from the of?cial Signal G electronic records of all company announcements to
the ASX[9].
For documentation and coding of MEFs, the following announcements were
manually checked and read:
.
announcements within one week prior to and following the company’s AGM for
the prior ?nancial year (especially the Chairman’s address);
.
announcements within one week prior to and following the release of the current
half yearly earnings announcement; and
.
all announcements within three months prior to the current annual earnings
announcement.
Effects of
forecast
speci?city
243
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In addition, the headlines of all announcements made by analyst-followed companies
between the AGM for the prior ?nancial year and three months prior to the current
annual earnings announcement were checked and, where necessary, the full text of the
announcement was read. Finally, share price data were obtained from SIRCA.
Initially, we de?ned any Signal G announcement in which management referred to
current period or future period pro?ts, revenues, distributions/dividends or production
as an earnings-related prior announcement. This process identi?ed 2,424 such
announcements. Subsequently, announcements that did not speci?cally refer to pro?t
(totalling 781 in number) were removed and not analysed. This left an initial sample of
1,643 MEFs made by 464 different companies. To run the regression models reported
later in the paper, we restricted the sample to those cases where:
.
valid share price data were available;
.
the forecast horizon exceeds three days; and
.
a valid news classi?cation (either bad, neutral or good) could be assigned from
either or both the analysts and prior year earnings benchmarks or the
observation related to a qualitative forecast.
This resulted in a ?nal sample size of 1,472. Where multiple pro?t measures were
referred to in a single announcement, we analyse the measure closest to net pro?t after
tax[10].
3.2 Variable de?nitions
Forecast speci?city. We classify MEFs into point, range, maximum, minimum and
qualitative categories (Skinner, 1994). In a point forecast, a precise single numerical
?gure is given ($X) or can be readily quanti?ed (for example, pro?t is expected to equal
that of the previous year). A range forecast contains a precise numeric range of pro?t
(for example, “pro?t will be between $X and $Y”) or the range can be readily quanti?ed
(for example, “pro?t will be within 10 per cent of last year”).
A maximum forecast sets a maximum or upper bound to pro?t, whereas in a
minimum forecast, a minimum or lower bound to pro?t is speci?ed. Qualitative
forecasts provide a general statement that is not capable of any precise numeric
interpretation.
In testing H2, we allow the different speci?cities that can be assigned a news
outcome to be recognised in their own right as a series of individually de?ned dummy
variables: D
Point
, D
Range
, D
Min
and D
Max
.
Classi?cation into good/neutral/bad news MEFs. This study determines good/
neutral/bad news MEFs with respect to the nearest analysts’ forecasts. Using the
analysts’ forecasts, our preferred method was to compare the MEF (typically an
aggregate pro?t number) to the mean analyst earnings forecast (an earnings per share
value) at the date of the announcement (Gallery et al., 2006). The following procedure as
summarised in Figure 1 was employed.
First, using I/B/E/S data, the number of shares on issue was obtained. This was
multiplied by the mean analyst EPS forecast obtained immediately prior to the release
of the MEF. The number of shares and the mean analyst EPS forecast from the I/B/E/S
detail ?le result in an implied analyst net pro?t after tax – this is our analysts
benchmark (AB). Next, the implied analyst net pro?t after tax was compared to the
MEF. If the point earnings forecast (MF
p
) or the mid-point of a range earnings
ARJ
22,3
244
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast[11] (MF
r
) was more than 10 per cent above (below) the implied analyst net
pro?t after tax (AB), this was classi?ed as good (bad) news, while all remaining valid
cases of point/range MEF are classed as neutral news.
For MEFs in the form of a maximum or minimum[12], we ?rst determine the most
speci?c dollar amount we could attach to the forecast (S
min
and S
max
, respectively).
Figure 1.
Characterisation of news
categories associated with
management earnings
forecasts
Effects of
forecast
speci?city
245
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
S
min
and S
max
were then compared to the implied analyst net pro?t after tax as follows:
If S
min
is greater than 10 per cent above (40 per cent below) the implied analyst pro?t, it
is classi?ed as good (bad) news, and otherwise it is deemed neutral news. If S
max
is
greater than 40 per cent above (10 per cent below) the implied analyst pro?t, it is
classi?ed as good (bad) news, and otherwise it is deemed neutral news[13,14].
For the hypothesis testing, the news variables based on analyst forecasts are a
series of dummy variables labelled D
Good
; D
Neut
; and D
Bad
and are scored one when the
decision rules mentioned previously are applied and zero otherwise.
Routine/non-routine announcements. The market reactions to MEFs may depend on
whether the MEF is released as part of a routine information event. Routine
announcements containing MEFs are those made at or accompanying the annual
general meeting (AGM) or the half-yearly pro?t announcement. ASX listing rules and
the Australian Corporations Act require Australian companies to continuously
disclose information likely to have a material effect on share price. Announcements
containing MEFs other than those mentioned previously are considered non-routine
and deemed to arise from the company’s continuous disclosure obligations. For
hypothesis testing, we use a dummy variable labelled D
NR
which is scored one when
the announcement is non-routine and zero otherwise.
Event dates. The timing of MEFs is taken as the date of the Signal G announcement
that contained the forecast. The annual earnings announcement is the date of the
preliminary annual earnings announcement from IRESS. Consensus analysts’
forecasts dates immediately prior to the release of the MEF are selected from the
I/B/E/S database.
Abnormal returns. A measure of the share price response to MEFs is required to test
investors’ reaction. A ?ve-day buy and hold return (BH5) measured from two days
before to two days (22, þ2) after the issue of the MEF. The buy-hold stock returns are
calculated as the stock return less the market return (proxied by the All Ordinaries
Accumulation Index) over the corresponding period. Prior literature such as Brown
and Warner (1985) and Shevlin (1981) suggest that such a parsimonious model does a
good job at isolating the market impact of events and typically produce results that are
at least as reliable as the more complex counterparts. Finally, to help guard against the
undue in?uence of outliers the top and bottom 1 per cent of BH5 observations were
trimmed.
Other variables. We also include a number of control variables, namely, ?rm size
(SIZE), forecast horizon measure (FHOR), year dummies (D_YR_MEF_96 –
D_YR_MEF_01) and prior year earnings loss dummy (D_PRIORYR_LOSS)[15].
SIZE is measured as the natural logarithm of the market value of the ?rm’s equity[16],
FHOR is the number of calendar days between the release of the earnings forecast and
the date of the release of the preliminary annual earnings announcement, D_YR are
calendar year dummies for the year in which MEFs occur and D_PRIORYR_LOSS is a
dummy variable equal to 1 if the ?rm recorded an earnings loss in the prior year.
4. Results
4.1 Sample descriptives
Tables I and II provide some descriptive information about our sample of MEFs.
Table I shows the composition of our sample forecasts in terms of speci?city.
Minimum and point forecasts together represent around 76.4 per cent of our sample,
ARJ
22,3
246
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
with maximum, range and qualitative forecasts making up the remainder. In Table II
we show the breakdown of news and its relationship to forecast speci?city using the
analysts forecast benchmark. In Table II, we see that minimum forecasts represent
nearly 50 per cent of good news cases, but only 22.7 per cent of bad news forecasts. On
the other hand, while maximum forecasts represent only 3.9 per cent of good news
cases, they constitute 19.9 per cent of bad news cases. Therefore, we can conclude that
management’s choice of forecast speci?city seems to be clearly related to the news that
the MEF is communicating. Thus, it is critical to analyse the interaction effects of
forecast speci?city and the types of news communicated[17].
4.2 News asymmetry hypothesis
First, we test for the Australian market that the absolute negative stock price response
to bad news MEFs is signi?cantly larger in magnitude than the positive stock price
response to good news MEFs. Using the analyst forecast basis of assessing news
content, the relevant speci?cation is found in equation (1)[18]. Testing hypotheses H1A
and H1B is conducted using equation (2)[19]:
BH5 ¼ a þ b
1
D
Bad
þb
2
D
Good
þ1 ð1Þ
BH5 ¼ a þ g
1
D
Bad
*
D
NR
þ g
2
D
Bad
*
D
R
þg
3
D
Good
*
D
NR
þg
4
D
Good
*
D
R
þ 1 ð2Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
(D
Good
) is a dummy variable set to
unity when MEFs are deemed bad (good) news relative to the mean analysts earnings
benchmark; D
NR
(D
R
) is a dummy variable taking a value of unity when the
announcement is non-routine (routine) and zero otherwise; Newey-West HAC standard
errors and covariance was used (lag truncation ¼ 7).
Good news
“Indeterminate”
news Bad news Totals
Number (%) Number (%) Number (%) Number (%)
Minimum 88 14.4 441 72.3 81 13.3 610 49.6
48.6 63.7 22.7
Maximum 7 5.1 60 43.5 71 51.4 138 11.2
3.9 8.7 19.9
Range 8 11.4 25 35.7 37 52.9 70 5.7
4.4 3.6 10.4
Point 78 18.9 166 40.3 168 40.8 412 33.5
43.1 24.0 47.1
Totals 181 692 357 1,230 100
100 100 100
Table II.
Distributional properties
of management earnings
forecasts – analyst
benchmarked news
versus speci?city
Qualitative Minimum Maximum Range Point Total
Number 115 661 154 79 463 1472
(%) 7.8 44.9 10.5 5.3 31.5 100.0
Table I.
Distributional properties
of management earnings
forecasts – forms of
speci?city
Effects of
forecast
speci?city
247
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
After con?rming that b
1
, 0 and b
2
. 0, we test whether the absolute value of b
1
is
signi?cantly greater thanthe absolute value of b
2
. The results are shown inTable III[20].
Table III shows that b
1
ðb
2
Þ, the mean effect of bad (good) news MEFs on BH5, is
negative (positive) at 24.0 per cent (þ1.9 per cent) and signi?cant at the 1 per cent
level. While these results are as expected, prior Australian research (Gallery et al., 2006)
has failed to document a signi?cant positive impact for good news MEFs[21]. We also
conduct a Wald test of the equality of the absolute values of (a þb
1
) and (a þ b
2
).
This test of equality produces a p-value of 0.013 and, hence, we document that there is a
differential effect to good and bad news similar to the USA. Therefore, we can conclude
that the magnitude of the share price effect of bad news MEFs is signi?cantly greater
than that of good news MEFs (where news is benchmarked against analyst forecasts).
Table IV shows the results for testing H1A and H1B. It shows that for non-routine
news, the previously mentioned asymmetry results are likewise documented. However,
the asymmetry is not found for routine news. This provides support for H1A but only
in the context of non-routine news, which is consistent with the notion that non-routine
voluntary disclosure may have much more information content and thus invoke
greater share price responses. Table IV also shows that only for bad news, is the share
price response signi?cantly larger for non-routine news as compared to routine news.
Therefore, H1B is supported for the bad news case.
4.3 Conditional speci?city hypothesis tests
The conditional speci?city hypothesis (H2) tests whether, after controlling for news,
forecast speci?city affects the share price response to MEFs. We begin with an
Variable Coef?cient t-statistic p-value
C a 0.020 0.784 0.433
D
Bad
b
1
20.040
* *
26.772 0.000
D
Good
b
2
0.019
* *
3.755 0.000
SIZE b
3
20.002 21.588 0.112
FHOR b
4
0.000
* *
3.398 0.001
D_PRIORYR_LOSS b
5
20.008 20.631 0.528
D_YR_MEF_96 b
6
0.015
* *
2.836 0.005
D_YR_MEF_97 b
7
0.010 1.512 0.131
D_YR_MEF_98 b
8
0.006 1.054 0.292
D_YR_MEF_99 b
9
0.008 1.173 0.241
D_YR_MEF_00 b
10
0.000 20.016 0.987
D_YR_MEF_01 b
11
0.006 0.848 0.396
Overall (good versus bad news) ja þ b
1
j ¼ ja þ b
2
j
( p-value)
6.226
*
(0.013)
Adjusted R
2
0.076
Notes: This table reports the result of running the following regression model: BH5 ¼ aþ
b
1
D
Bad;j
þ b
2
D
Good;j
þ 1. where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy
variable set to unity when MEFs are deemed bad (good) news relative to the mean analysts earnings
benchmark; Newey-West HAC standard errors and covariance was used (lag truncation ¼ 7).
The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast horizon);
D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year dummies).
*
and
* *
represents 5 and 1 per cent level of signi?cance, respectively
Table III.
Basic test of news
asymmetry content
ARJ
22,3
248
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
analysis similar to Skinner (1994). Using the analysts forecast news benchmark we
conduct univariate tests of the absolute value of BH5 market response to MEFs
differing in forecast speci?city. The null form of this hypothesis suggests that after
controlling for the news content of the MEF, there is no difference in the short-window
share price response to different forms (minimum, maximum, range, point) of MEFs.
The results of this analysis are shown in Table V.
There are two notable results from Table V. The ?rst is that for minimum forecasts
the asymmetry is reversed 2 the response to good news MEFs is larger in absolute
magnitude than the response to bad news MEFs. For the sample overall, and for the
sub-samples of routine forecasts and non-routine forecasts, the absolute value of the
response to good news minimum style MEFs exceeds that of bad news minimum style
MEFs at the 1 per cent level. There is asymmetry for minimum forecasts, but in the
opposite direction to that argued by Skinner and found overall in the earlier analysis.
Minimum forecasts are generally associated with good news and markets react more
strongly to good news minimum MEFs than to bad news minimum forecasts. The
Variable Coef?cient t-statistic p-value
C a 0.040 1.501 0.134
D
Bad
* D
NR
g
1
20.062
* *
26.491 0.000
D
Bad
* D
R
g
2
20.018
* *
22.842 0.005
D
Good
* D
NR
g
3
0.025
* *
2.851 0.004
D
Good
* D
R
g
4
0.015
*
2.498 0.013
SIZE g
5
20.002
*
21.995 0.046
FHOR g
6
0.000 1.271 0.204
D_PRIORYR_LOSS g
7
20.010 20.829 0.407
D_YR_MEF_96 g
8
0.014
* *
2.653 0.008
D_YR_MEF_97 g
9
0.009 1.502 0.133
D_YR_MEF_98 g
10
0.005 0.869 0.385
D_YR_MEF_99 g
11
0.006 1.023 0.306
D_YR_MEF_00 g
12
0.000 20.059 0.953
D_YR_MEF_01 g
13
0.005 0.736 0.462
Non-routine H1A: ja þg
1
j ¼ ja þ g
3
j 17.825
* *
(Good versus bad news) ( p-value) (0.000)
Routine H1A : ja þ g
2
j ¼ ja þ g
4
0.377
(Good versus bad news) ( p-value) (0.539)
Bad news H1B : ja þ g
1
j ¼ ja þ g
2
j 44.207
* *
(Non-routine versus routine) ( p-value) (0.000)
Good news H1B: ja þ g
3
j ¼ ja þ g
4
j 0.912
(Non-routine versus routine) ( p-value) (0.340)
Adjusted R
2
0.091
Notes: This table reports the result of running the following regression model: BH5 ¼
a þ g
1
D
Bad
*D
NR
þ g
2
D
Bad
*D
R
þ g
3
D
Good
*D
NR
þ g
4
D
Good
*D
R
þ 1 where BH5 is a ?ve-day buy
and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs are deemed bad (good)
news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy variable taking a value
of unity when the announcement is non-routine (routine) and zero otherwise; Newey-West HAC
standard errors and covariance was used (lag truncation ¼ 7). The regression is estimated with
control variables: SIZE (?rm size); FHOR (forecast horizon); D_PRIORYR_LOSS (prior year loss
dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year dummies).
*
and
* *
represents 5 and 1 per cent
level of signi?cance, respectively
Table IV.
Test of news asymmetry:
Non-routine versus
routine forecasts
Effects of
forecast
speci?city
249
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
second notable result of the univariate analysis is that weakly for the overall sample
and more strongly for the non-routine forecasts, maximum, range and point forecasts
all show an asymmetrically large response to bad news MEFs. This is not observed for
routine forecasts.
The univariate approach considers each form of forecast speci?city separately and
does not combine all forms of forecast speci?city with all news outcomes in a single
regression model. We now formulate such a model to assess the effect of forecast
speci?city on returns after controlling for news (H2). Using analyst forecasts as the
basis for assessing news, the form of the model is given by equation (3):
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ1 ð3Þ
where D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum,
maximum, range and point forecasts, respectively. As mentioned previously, the
regression is estimated with control variables: SIZE; FHOR; D_PRIORYR_LOSS and
D_YR_MEF_96 – D_YR_MEF_01.
We chose this path of separating the types of speci?city for the following reasons.
First, if we use an aggregate concept of speci?city (for example the IMPRECISE
variable[22] as employed by Baginski et al., 1993) in the current context, it may well
mask important variations in market impact across different types of forecast. As such,
equation (3) provides a robust means by which we can simply but effectively model a
non-linear relation. Second, an aggregate form of speci?city imposes an
ordering/weighting and linearity, which in the context of assessing the market
impact may be overly restrictive. This is particularly an issue given the results in
Table V, which show strong good news asymmetry for minimum forecasts and strong
Pairwise comparison results
Overall sample Non-routine MEFs Routine MEFs
Minimum forecasts 22.91
* * *
(0.000) 6.95
* * *
(0.001) 7.19
* * *
(0.008)
(G . B) (G . B) (G . B)
Maximum forecasts 3.42
*
(0.065) 6.72
* *
(0.010) 1.04 (0.308)
(B . G) (B . G) (B . G)
Range forecasts 4.20
* *
(0.041) 6.82
* * *
(0.009) 2.49 (0.115)
Point forecasts 1.01 (0.315) 5.43
* *
(0.020) 0.13 (0.714)
(B . G)
Notes: H0: BH5
Good
s
2 BH5
Bad
s
¼ 0 versus H
a
: BH5
Good
s
2 BH5
Bad
s
– 0. Wald tests of the
difference in value between the absolute value of BH5
N
s
for MEFs of differing forecast speci?city
where superscript “N” is the type of news (good or bad) and the subscript “s” is the speci?city of the
MEF, namely, minimum, maximum, range and point forecast. Cases, in which the value of BH5
Good
s
is
smaller than the counterpart value of BH5
Bad
s
are indicated by (B . G), whereas the converse
situation is shown as (G . B). The values of the Wald statistic are shown in the Table, while the
p-value is in parentheses.
*
,
* *
and
* * *
represents 10 per cent, 5 per cent and 1 level of signi?cance,
respectively
Table V.
Test of asymmetric
speci?city hypothesis:
pairwise tests of the
difference in absolute
values between good and
bad news MEFs of
differing forecast
speci?city
ARJ
22,3
250
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bad news asymmetry for non-routine maximum forecasts. It should also be noted that
prior evidence documents that US markets react to minimum style MEFs as conveying,
on average, good news and maximum style MEFs as conveying, on average, bad news
(Baginski et al., 1994).
We determine the news content of MEFs independent of forecast speci?city and
then, after controlling for news content, assess whether the share market response is
affected by forecast speci?city. As such, the model outlined in equation (3), by allowing
each quantitative type of forecast to interact with good and bad news in its own right,
permits us to incorporate the asymmetry effect of bad news into our exploration of the
market reaction effects of forecast speci?city.
The interaction terms capture the effects of changes in forecast speci?city for good
and bad news MEFs and a variety of tests of joint equality of coef?cients are explored
to assess the validity of H2. The results of estimating this regression equation (3) using
the analyst forecast method to assess news content are shown in Tables VI-VIII.
Table VI shows the basic regression results for the overall sample, while Table VII
shows the results of testing the asymmetry in news content, controlling for speci?city.
Variable Coef?cient t-statistic p-value
C a 0.044
*
1.676 0.094
D
Min
* D
Bad
b
Bad,Min
20.001 20.129 0.897
D
Max
* D
Bad
b
Bad,Max
20.068
* * *
25.282 0.000
D
Range
* D
Bad
b
Bad,Range
20.073
* * *
23.272 0.001
D
Point
* D
Bad
b
Bad,Point
20.043
* * *
25.388 0.000
D
Min
* D
Good
b
Good,Min
0.023
* * *
3.765 0.000
D
Max
* D
Good
b
Good,Max
0.013 0.702 0.483
D
Range
* D
Good
b
Good,Range
0.001 0.072 0.942
D
Point
* D
Good
b
Good,Point
0.015
*
1.781 0.075
SIZE g
9
20.003
* *
22.367 0.018
FHOR g
10
0.001
* *
2.423 0.016
D_PRIORYR_LOSS g
11
20.014 21.139 0.255
D_YR_MEF_96 g
12
0.017
* * *
3.242 0.001
D_YR_MEF_97 g
13
0.011
*
1.739 0.082
D_YR_MEF_98 g
14
0.008 1.345 0.179
D_YR_MEF_99 g
15
0.007 1.104 0.270
D_YR_MEF_00 g
16
0.001 0.085 0.932
D_YR_MEF_01 g
17
0.008 1.184 0.237
Adjusted R
2
0.097
Notes: This table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance, respectively
Table VI.
Test of news asymmetry
and conditional
speci?city hypotheses –
basic estimation results
Effects of
forecast
speci?city
251
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table VIII, Panels A and B report the results of testing the “conditional speci?city”
hypothesis conditioned for bad and good news, respectively.
An interesting observation from Table VI is that three out of the four forecast
speci?cities for bad news, namely, maximum, range and point MEFs are signi?cantly
negative. Focusing on the asymmetry in news content for different forms of forecast
speci?city in Table VII, we ?nd that the absolute magnitude of the share price response
for bad news is greater than the counterpart reaction for good news, except in the case
of minimum form MEFs[23]. The differences are economically important. This result
further strengthens the asymmetry results documented in the earlier section for each of
the forecast speci?city types.
We test H2 for the difference in share price responses between forecast speci?city,
respectively controlling for bad and good news, in Table VIII, Panels A and
B. Speci?cally, Panel A reveals that the market impact of maximum, range and point
forecasts are all signi?cantly different from the minimum category, when assessed on a
pairwise basis (at the 1 per cent level). The group of three (maximum, range and point)
however, when assessed on a pairwise basis, are not signi?cantly different from each
other. Thus, in the case of bad news, it is the minimum forecast category, which drives
the speci?city result. To con?rm the overall rejection of H2 in the context of bad news,
Panel A reports the rejection of the full joint test of equality across all four-interaction
terms (with a p-value of 0.000). We test the Kim and Verrecchia (1991) model
predictions in H2A. We hypothesize that the share price reactions of point MEFs are
larger in magnitude in comparison to the other three types of MEFs because point
MEFs are the most precise. For bad news MEFs, this is only observed when point MEF
is compared with minimum MEF but not for the other two. Therefore we do not ?nd
general support for H2A.
News speci?city ja þ b
Good;I
j ¼ ja þ b
Bad;i
j
Minimum 12.762
*
(0.000)
Maximum 9.690
*
(0.002)
Range 39.503
*
(0.000)
Point 11,184
*
(0.001)
Notes: This table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5> is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
*
represents the 10 per cent level of signi?cance
Table VII.
Test of news asymmetry
and conditional
speci?city hypotheses –
Test of “news
asymmetry” hypothesis
conditional on news
speci?city (F-statistic and
p-value)
ARJ
22,3
252
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Now turning to Panel B for the tests of H2 conditioned on good news MEFs, the results
show that none of the pairwise combinations of speci?city are signi?cantly different
from each other at the 5 per cent level. Moreover, in a full joint test of equality across all
four-interaction terms for good news, this restriction cannot be rejected. Therefore,
H2A is not supported for the good news MEF case.
These results are inconsistent with the predictions of the Kim and Verrecchia (1991)
model. Minimum, maximum and range forecasts are less speci?c than point forecasts.
Again, other things being equal, the Kim and Verrecchia (1991) analysis would argue
for a declining impact on share prices as forecast precision falls. For bad news, this is
found for the minimum and point MEFs pair and the minimum and range MEFs pair.
For good news forecasts, this is not observed. What we do ?nd is that all forms of
forecast which contain good news, essentially have equal market impact.
Tables IX and X report results for tests of H2 and H2A for non-routine and routine
MEFs using equation (4):
Pairwise tests Min Max Range
Panel A: Test of speci?city hypothesis conditional on bad news (chi-square statistic and p-value)
Max 19.771
* *
(0.000) – –
Range 8.869
* *
0.037
(0.003) (0.848) –
Point 11.914
* *
2.880 1.691
(0.001) (0.090) (0.194)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
25.309
* *
(0.000)
Panel B: Test of speci?city hypothesis conditional on good news (chi-square statistic and p-value)
Max 0.343
(0.558) – –
3.274
*
0.313
Range (0.071) (0.576) –
0.732 0.013 1.042
Point (0.393) (0.908) (0.307)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
3.465
(0.325)
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* *
and
*
represents 10 and 1 per cent level of signi?cance, respectively
Table VIII.
Test of news asymmetry
and conditional
speci?city hypotheses –
Test of “news
asymmetry” hypothesis
conditional on news
speci?city (F-statistic and
p-value)
Effects of
forecast
speci?city
253
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Bad;i;j
D
Bad
*
D
i
*
D
j
ð4Þ
þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Good; i
D
Good
*
D
i
*
D
j
þ 1
As before, the regression is estimated with control variables: SIZE; FHOR;
D_PRIORYR_LOSS and D_YR_MEF_96 - D_YR_MEF_01.
Variable Coef?cient t-statistic p-value
C a 0.051
*
1.872 0.061
D
Bad
* D
Min
* D
NR
b
Bad,Min,NR
20.022 20.746 0.456
D
Bad
* D
Min
* D
R
b
Bad,Min,R
0.004 0.429 0.668
D
Bad
* D
Max
* D
NR Bad,Max,NR
20.079
* * *
24.308 0.000
D
Bad
* D
Max
* D
R
b
Bad,Max,R
20.05
* * *
23.553 0.000
D
Bad
* D
Range
* D
NR
b
Bad,Range,NR
20.072
* * *
22.736 0.006
D
Bad
* D
Range
* D
R
b
Bad,Range,R
20.083
*
21.906 0.057
D
Bad
* D
Point
* D
NR
b
Bad,Point,NR
20.058
* * *
25.360 0.000
D
Bad
* D
Point
* D
R
b
Bad,Point,R
20.022
*
21.911 0.056
D
Good
* D
Min
* D
NR
b
Good,Min,NR
0.041
* * *
3.532 0.000
D
Good
* D
Min
* D
R
b
Good,Min,R
0.019
* * *
2.650 0.008
D
Good
* D
Max
* D
NR
b
Good,Max,NR
0.012 1.197 0.232
D
Good
* D
Max
* D
R
b
Good,Max,R
0.013 0.540 0.589
D
Good
* D
Range
* D
NR
b
Good,Range,NR
20.005 20.331 0.740
D
Good
* D
Range
* D
R
b
Good,Range,R
0.015 1.016 0.310
D
Good
* D
Point
* D
NR
b
Good,Point,NR
0.022
*
1.755 0.080
D
Good
* D
Point
* D
R
b
Good,Point,R
0.008 0.716 0.474
SIZE g
9
20.003
* *
22.429 0.015
FHOR g
10
0.000 1.372 0.170
D_PRIORYR_LOSS g
11
20.014 21.165 0.244
D_YR_MEF_96 g
12
0.016
* * *
3.045 0.002
D_YR_MEF_97 g
13
0.011
*
1.788 0.074
D_YR_MEF_98 g
14
0.007 1.150 0.250
D_YR_MEF_99 g
15
0.007 1.098 0.273
D_YR_MEF_00 g
16
0.000 0.020 0.984
D_YR_MEF_01 g
17
0.008 1.098 0.272
Adjusted R
2
0.101
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs
are deemed bad (good) news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy
variable taking a value of unity when the announcement is non-routine (routine) and zero otherwise;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range and
point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance, respectively
Table IX.
Test of speci?city
hypothesis conditioned
on news content:
non-routine versus
routine forecasts – basic
estimation results
ARJ
22,3
254
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Pairwise tests Min Max Range
Panel A: Test of speci?city hypothesis conditional on bad/non-routine news (chi-square statistic and p-value)
Max 3.198 – –
(0.074) – –
Range 1.622 0.063
(0.203) (0.802) –
Point 1.372 0.986 0.246
(0.242) (0.321) (0.620)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
3.383
(0.336)
Panel B: Test of speci?city hypothesis conditional on good/non-routine news (chi-square statistic and p-value)
Max 3.957
* *
– –
(0.047)
Range 6.512
* *
1.017 –
(0.011) (0.314) –
Point 1.298 0.428 2.030
(0.255) (0.513) (0.155)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
7.420
*
(0.060)
Panel C: Test of speci?city hypothesis conditional on bad/routine news (chi-square statistic and p-value)
Max 10.350
* * *
– –
(0.001) – –
Range 3.728
*
0.483 –
(0.054) (0.487)
Point 3.309
*
2.573 1.768
(0.069) (0.109) (0.184)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
14.041
(0.003)
Panel D: Test of speci?city hypothesis conditional on good/routine News (chi-square statistic and p-value)
Max 0.054 – –
(0.817) – –
Range 0.070 0.003 –
(0.791) (0.953) –
Point 0.758 0.043 0.172
(0.384) (0.836) (0.678)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
0.766
(0.858)
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Bad;i;j
D
Bad
*
D
i
*
D
j
ð4Þ
þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Good; i
D
Good
*
D
i
*
D
j
þ 1
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs are
deemed bad (good) news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy variable taking
a value of unity when the announcement is non-routine (routine) and zero otherwise; D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range and point forecasts, respectively. Newey-West
HAC standard errors and covariance was used (lag truncation ¼ 7). The regression is estimated with control
variables: SIZE (?rm size); FHOR (forecast horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_
MEF_96 – D_YR_MEF_01 (year dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance,
respectively
Table X.
Test of speci?city
hypothesis conditioned
on news content:
non-routine versus
routine forecasts
Effects of
forecast
speci?city
255
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Panel A shows that six out of the eight coef?cients for bad news are signi?cantly
negative while only three out of the eight coef?cients for good news are signi?cantly
positive (at the 10 per cent level). An interesting observation is that both bad news
coef?cients for minimum MEFs are insigni?cant while the two counterpart coef?cients
for good news are signi?cantly positive (at the 1 per cent level). This observation
reinforces the generally held belief that minimum MEFs are essentially good news.
Tables X, Panel A shows the chi-square test statistics for the ‘conditional speci?city’
hypothesis for bad and non-routine news. None of the pairwise tests are signi?cant and
thus for bad news of a non-routine type, the type of forecast speci?city has no bearing
on the magnitude of the share price response. As shown in Table X, Panel B, the same
general conclusion is true for good news and non-routine type MEFs (except in the
cases between minimum and range MEFs and minimum and maximum MEFs). Panels
C and D show the results of the counterpart tests of Panels A and B, only now the tests
are conducted on routine MEFs. The results are qualitatively similar, except in Panel C
where the price reaction to minimum MEFs is now found to be signi?cantly different
from maximum MEFs (1 per cent) and from range and point MEFs (10 per cent).
Our results are consistent with the ?ndings of Pownall et al. (1993) and Hirst et al.
(1999) where they conclude that there is no signi?cant difference in market reaction
between minimum, maximum, range and point MEFs.
5. Conclusions
This paper documents that the response to bad news MEFs is greater than the
response to good news MEFs. This ?nding is similar to that found in the US.
Furthermore, we investigate hypotheses regarding the short-window market reaction
to MEFs. First, we consider whether the negative stock price response to bad news
MEFs is signi?cantly larger in magnitude than the positive stock price response to
good news MEFs for non-routine announcements. Next, we test whether the magnitude
of the stock price response is larger for non-routine as compared to routine
announcements. In further analysis we consider a “conditional speci?city” hypothesis,
which proposes that after controlling for the news content of the MEF, there is no
difference in the stock price response to different forms (point, range, minimum,
maximum) of MEF.
Our tests, based on a large sample of Australian MEFs that are extensively checked
and sourced from the of?cial ASX database of company announcements, document
that there is a greater response to bad news MEFs in our overall sample and for the
sub-set of non-routine forecasts, but generally not for routine forecasts. This suggests
that the “unexpected” nature of non-routine MEFs is what surprises the market. Our
univariate tests of the “conditional speci?city” hypothesis ?nd bad news asymmetry
for maximum, range and point forecasts but for forecasts in the form of a minimum, the
asymmetry is reversed – there being a signi?cantly larger response to good news
rather than bad news minimum MEFs. This is a new ?nding in the literature.
Our paper also contributes to an important theoretical issue. Speci?cally, we use a
model that allows four forms of forecast speci?city to interact with both good and bad
news to test the Kim and Verrecchia (1991) proposition that the precision of MEFs is
positively related to the share market reaction to that forecast. Our results are
inconsistent with those of Baginski et al. (1993) and they do not support Kim and
Verrecchia’s prediction. We ?nd that, using the analyst forecast news benchmark, for
ARJ
22,3
256
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bad news MEFs, the market reacts equally negatively to point, range and maximum
forecasts, but is mute to bad news minimum forecasts. For good news forecasts, we are
unable to identify any difference in the market reaction based on the speci?city of the
forecast. Our further analysis shows that for non-routine forecasts, the form of the
forecast has no signi?cant effect on the market reaction. Our ?ndings are consistent
with Pownall et al. (1993) and Hirst et al. (1999), which ?nd that forecast speci?city has
little to contribute to the difference in market reaction of MEFs.
Notably, none of our results support the view that more speci?c forecasts elicit
greater market responses – except when we compare point and minimum MEFs. What
our results do suggest is that managers appear to choose the form of the forecast to suit
the news being delivered. In particular, bad news delivered in a minimum forecast
seems to be ignored by the market.
Notes
1. Skinner and Sloan (2002) document an asymmetric share price response of ?rms that fail to
meet analyst expectations as compared to those ?rms that meet or better analyst forecasts.
Libby et al. (2006, p. 207) ?nd, in an experimental market setting, that “the form of
management’s earnings guidance (point, narrow range, wide range) affects analysts’
earnings forecasts”.
2. Speci?city re?ects the precision of management’s earnings forecast. We analyse the effects
of forecast speci?city in four categories: point forecasts, range forecasts (both upper and
lower bound are speci?ed), minimum forecasts (lower bound only), and maximum forecasts
(upper bound only).
3. In studies pertaining to analysts earnings forecasts and market responses, many have found
support for the asymmetric responses to good and bad news (Ho and Sequeira, 2007; Skinner
and Sloan, 2002; Lopez and Rees, 2001; Conrad et al., 2002).
4. The continuous reporting requirement in Australia does not necessarily results in more
frequent and greater quantity of reporting. It basically requires companies to be more-timely
in reporting price sensitive information. This paper leverages on this unique disclosure
regime to test the share price impact of management earnings forecasts.
5. This argument is supported by the subsequent work of Skinner and Sloan (2002) which
argues that large negative surprises can “torpedo” the share price of a ?rm.
6. Hutton et al. (2003), document that good news MEFs generate signi?cant market reaction
only when accompanied by other veri?able forward looking statements, whereas there is a
signi?cant reaction to bad news MEFs irrespective of any accompanying disclosures.
7. Baginski et al. (1994) use the share price response to identify the “news” content of forecasts
differing in speci?city. They ?nd that the market reaction to minimum forecasts indicates
that they are, on average, perceived as good news, whereas the market reaction to maximum
forecasts indicates that they are, on average, bad news. As we wish to test the effect of
forecast speci?city on share returns, after controlling for the news content of the forecast,
their approach is not appropriate for our purposes.
8. Like any situation involving research design tradeoffs, restricting our sample to
analyst-covered ?rms comes at a cost. Speci?cally, our analysis is potentially impacted
by a selection bias due to the non-randomness in analysts’ stock coverage selection (Rajan
and Servaes, 1997). We gratefully acknowledge an anonymous referee for drawing this issue
to our attention.
Effects of
forecast
speci?city
257
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
9. We use Signal G data throughout but access this data in two different ways: First, for the
period September 1994 to 31 December 1995 all Signal G announcements are obtained from
the Securities Industry Research Centre for Asia Paci?c (SIRCA). Companies with analysts’
earnings forecasts available were manually read for any reference to earnings outlook,
forecasts or upgrades and such cases were carefully recorded in the database. Second, for the
period 1 January 1996 to 31 December 2001, the Signal G company announcements by all
analyst-followed ASX companies on the I/B/E/S database were accessed on Integrated Real
Time Equity System (IRESS). After 1 July 2001 the PDF signal on IRESS is used.
10. Measures of pro?t referred to include net pro?t before/after tax, before and after interest,
before and after abnormal items, before and after depreciation and amortisation.
11. We are mindful of the caution in Libby et al. (2006, p. 207) where they state “treating the
mean of the range endpoints as equivalent to a point estimate and failing to consider effects
after the release of actual earnings may paint an incomplete picture of how management
guidance affects analysts and investors”. There is currently no acceptable methodology to
capture the quantitative estimate for an MEF expressed as a range.
12. Some prior studies exclude maximum and minimum forecasts on the grounds that it is
dif?cult to unequivocally identify their news content (Ajinkya and Gift, 1984). Baginski et al.
(2002) rely on Baginski et al. (1994) and assume all minimum forecasts are good news and all
maximum forecasts are bad news. Such an approach would not allow us to assess the impact
of forecast speci?city after controlling for news (H2). Our approach to assessing the news
content of maximum and minimum forecasts differs from Baginski et al. (1993) in that they
do not use a neutral category and simply compare the upper bound (for maximum forecasts)
or the lower bound (for minimum forecasts) to the median analyst forecast.
13. The news content of qualitative forecasts cannot be determined and they are not considered
further.
14. One major criticism of this approach in classifying bad and good news is that the cut-off
score seems arbitrary. While our conservative approach is appealing (we allow a larger
margin for bad news minimum/good news maximum forecasts) it is not without concern. For
example, a company announcing that it expects to have an EPS of at least 5.9 cents against a
current analyst forecast benchmark of 10 cents per share, will be classi?ed as bad news.
However, the managers of this hypothetical company could be habitually conservative, and
the open-ended forecast still leaves open the potential for an actual pro?t that exceeds
consensus forecasts (we thank an anonymous referee for drawing our attention to this issue).
Our reaction to this general concern has two elements. First, there is currently no generally
“acceptable” methodology in applying a systematic cut-off score for classi?cation purposes.
Second, by varying the cut-off score, our results are generally qualitatively similar for
minimum, maximum and range MEFs. Details are suppressed in the interests of brevity.
15. Current year earnings loss is not used because it is highly correlated with other variables.
16. Given that our sample is con?ned to analyst-followed cases, it is biased toward larger
Australian companies. Further details, while suppressed to conserve space, are available
from the authors on request.
17. The small sample size for range, and maximum forecasts for good news events, means that
the more meaningful comparisons are between minimum and point forecasts for good news
cases. Also, the comparison between maximum and point forecasts for bad news cases is
more meaningful due to the concern about classifying minimum forecasts as bad news. We
thank an anonymous referee for drawing these issues to our attention.
ARJ
22,3
258
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
18. To enhance readability of all regression models, the control variables are not included in the
formal written equations (SIZE; FHOR; D_PRIORYR_LOSS and D_YR_MEF_96 –
D_YR_MEF_01).
19. To assess the effect of the three classes of news: good, neutral and bad; we specify dummy
variables for good and bad news in the regression model. The neutral news content is
embedded within the intercept term.
20. A correlation table was also constructed to check for possible problems with
multi-collinearity between the explanatory variables. The correlation results, although not
reported here, do not suggest that the ?ndings will be affected by any multi-collinearity
problems.
21. A possible reason why Gallery et al. (2006) do not ?nd a statistically signi?cant impact is due
to their small sample size. They only have in total 45 observations that are point, range,
minimum or maximum out of a total sample of 180 observations. Within this total sample
only 63.3 per cent are good news and within the good news only 15.8 per cent are correctional
forecasts (quantitative forecasts with deviations greater than 5 per cent from analysts’
forecasts).
22. Baginski et al. (1993) combine the four forms of forecast speci?city into a single variable
(IMPRECISE), which is given a value of zero for point forecasts, one for closed interval
forecasts (range) and two for open interval forecasts (minimum/maximum).
23. While it is true that the minimum case also reveals a signi?cant test result, readers should
note that it is re?ective of the converse scenario – namely, that the absolute magnitude of the
share price response for bad news is less than the counterpart reaction for good news.
References
Ajinkya, B.P. and Gift, M.J. (1984), “Corporate managers’ earnings forecasts and symmetrical
adjustments of market expectations”, Journal of Accounting Research, Vol. 22, pp. 425-44.
Baginski, S.P., Conrad, E.J. and Hassell, J.M. (1993), “The effects of management forecast
precision on equity pricing and on the assessment of earnings uncertainty”, Accounting
Review, Vol. 68, pp. 913-27.
Baginski, S.P., Hassell, J.M. and Kimborough, M.D. (2002), “The effect of legal environment on
voluntary disclosure: evidence from management earnings forecasts issued in US and
Canadian markets”, Accounting Review, Vol. 77, pp. 25-50.
Baginski, S., Hassell, J. and Waymire, G. (1994), “Some evidence on the news content of
preliminary earnings estimates”, Accounting Review, Vol. 69, pp. 265-73.
Brown, L.D. and Higgins, H.N. (2005), “Managers’ forecast guidance of analysts: international
evidence”, Journal of Accounting and Public Policy, Vol. 24, pp. 280-99.
Brown, S. and Warner, J.B. (1985), “Using daily stock returns: the case of event studies”, Journal
of Financial Economics, Vol. 14, pp. 3-31.
Conrad, J., Cornell, B. and Landsman, W.R. (2002), “When is bad news really bad news?”, Journal
of Finance, Vol. 57, pp. 2507-32.
Curley, S.P., Yates, J.F. and Abrams, R.A. (1986), “Psychological sources of ambiguity
avoidance”, Organizational Behavior and Human Decision Processes, Vol. 38, pp. 230-56.
Daniel, K., Hirshleifer, D. and Subrahmanyam, A. (1998), “Investor psychology and security
market under- and overreactions”, Journal of Finance, Vol. 53, pp. 1839-85.
Einhorn, H.J. and Hogarth, R.M. (1985), “Ambiguity and uncertainty in probabilistic inference”,
Psychological Review, Vol. 92, pp. 433-61.
Effects of
forecast
speci?city
259
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Gallery, G., Gallery, N. and Hsu, C-M.G. (2006), “The association between management and
analysts’ earnings forecasts in the Australian continuous disclosure environment”,
working paper, Queensland University of Technology and University of Queensland,
Brisbane.
Grossman, S. (1981), “The informational role of warranties and private disclosure about product
quality”, Journal of Law and Economics, Vol. 24, pp. 461-83.
Highhouse, A.A. (1994), “Verbal protocol analysis of choice under ambiguity”, Journal of
Economic Psychology, Vol. 15, pp. 621-35.
Hirst, D.E., Koonee, L. and Miller, J. (1999), “The joint effect of management’s prior forecast
accuracy and the form of its ?nancial forecasts on investor judgment”, Journal of
Accounting Research, Vol. 37, Supplement, pp. 101-24.
Ho, Y.K. and Sequeira, J. (2007), “Earnings surprises, asymmetry of returns and market level
changes – an industry study”, Journal of Accounting, Auditing and Finance, Vol. 22 No. 1.
Hutton, A.P., Miller, G.S. and Skinner, D.J. (2003), “The role of supplementary statements with
management earnings forecasts”, Journal of Accounting Research, Vol. 41, pp. 867-90.
Kasznik, R. and Lev, B. (1995), “To warn or not to warn? Management disclosures in the face of
an earnings surprise”, Accounting Review, Vol. 70, pp. 113-34.
Kim, O. and Verrecchia, R.E. (1991), “Trading volume and price reactions to public
announcements”, Journal of Accounting Research, Vol. 29, pp. 302-21.
King, R., Pownall, G. and Waymire, G. (1998), “How frequently do managers disclose prospective
earnings information?”, Journal of Financial Statement Analysis, Vol. 3, pp. 5-16.
Kothari, S.P., Shu, S. and Wysocki, P.D. (2007), “Do managers withhold bad news?”, research
paper, MIT Sloan, Cambridge, MA.
Kuhn, K.M. and Budescu, D.V. (1996), “The relative importance of probabilities, outcomes and
vagueness in hazard risk decisions”, Organizational Behavior and Human Decision
Processes, Vol. 68, pp. 301-17.
Kuhn, K.M., Budescu, D.V., Hershey, J.R., Kramer, K.M. and Rantilla, A.K. (1999), “Attribute
tradeoffs in low probability/high consequence risks: the joint effects of dimension
preference and vagueness”, Risk, Decision and Policy, Vol. 4, pp. 31-46.
Lennox, C.S. and Park, C.W. (2006), “The informativeness of earnings and management’s
issuance of earnings forecasts”, Journal of Accounting and Economics, Vol. 42, pp. 439-58.
Libby, R., Tan, H.T. and Hunton, J.E. (2006), “Does the form of management’s earnings guidance
affect analysts’ earnings forecasts?”, Accounting Review, Vol. 81, pp. 207-25.
Lopez, T.J. and Rees, L.L. (2001), The Effect of Meeting Analyst Forecasts and Systematic Positive
Forecast Errors on the Information Content of Unexpected Earnings, available at: http://
ssrn.com/abstract¼181929.
Matsumoto, D.A. (2002), “Management’s incentives to avoid negative earnings surprises”,
Accounting Review, Vol. 77, pp. 483-514.
Milgrom, P.R. (1981), “Good news and bad news: representation theorems and applications”, Bell
Journal of Economics, Vol. 12, pp. 380-91.
Patell, J. (1976), “Corporate forecasts of earnings per share and stock price behaviour: empirical
tests”, Journal of Accounting Research, Vol. 14, pp. 246-76.
Penman, S.H. (1980), “An empirical investigation of the voluntary disclosure of corporate
earnings forecasts”, Journal of Accounting Research, Vol. 18, pp. 132-60.
Pownall, G., Wasley, C. and Waymire, G. (1993), “The stock price effects of alternative types of
management earnings forecasts”, Accounting Review, Vol. 68, pp. 896-912.
ARJ
22,3
260
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Rajan, R. and Servaes, H. (1997), “Analyst following of initial public offerings”, Journal of
Finance, Vol. 52, pp. 507-29.
Shevlin, T. (1981), “Measuring abnormal performance on the Australian equity market”,
Australian Journal of Management, Vol. 6, No. 2, pp. 67-107.
Skinner, D.J. (1994), “Why ?rms voluntarily disclose bad news”, Journal of Accounting Research,
Vol. 32, pp. 38-60.
Skinner, D.J. and Sloan, R.G. (2002), “Earnings surprises, growth expectations and stock returns:
don’t let an earnings torpedo sink your portfolio”, Review of Accounting Studies, Vol. 7,
No. 2/3, pp. 289-312.
Wallsten, T., Budescu, D., Rapoport, A., Zwick, R. and Forsyth, B. (1986), “Measuring the vague
meanings of probability phrases”, Journal of Experimental Psychology: General, Vol. 115,
No. 4, pp. 348-65.
Waymire, G. (1985), “Earnings volatility and voluntary management forecast disclosure”,
Journal of Accounting Research, Vol. 23, pp. 268-95.
Zhang, X.F. (2006), “Information uncertainty and stock returns”, Journal of Finance, Vol. 61,
pp. 105-36.
About the authors
Howard Chan is an Associate Professor in the Department of Finance at the University of
Melbourne, Australia. Howard Chan is the corresponding author and can be contacted at:
[email protected]
Robert Faff is a Professor of Finance, in the Department of Accounting and Finance, at
Monash University, Victoria, Australia.
Yee Kee Ho is an Associate Professor in the Department of Accounting, National University
of Singapore, Singapore.
Alan Ramsay is an Associate Professor in the Department of Accounting and Finance,
Monash University, Victoria, Australia.
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
Effects of
forecast
speci?city
261
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_927040461.pdf
This study aims to test the effects of forecast specificity on the asymmetric short-window
share market response to management earnings forecasts (MEF).
Accounting Research Journal
The effects of forecast specificity on the asymmetric short-window share market
response to management earnings forecasts
Howard Chan Robert Faff Yee Kee Ho Alan Ramsay
Article information:
To cite this document:
Howard Chan Robert Faff Yee Kee Ho Alan Ramsay, (2009),"The effects of forecast specificity on the
asymmetric short-window share market response to management earnings forecasts", Accounting
Research J ournal, Vol. 22 Iss 3 pp. 237 - 261
Permanent link to this document:http://dx.doi.org/10.1108/10309610911005572
Downloaded on: 24 January 2016, At: 21:09 (PT)
References: this document contains references to 37 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 1977 times since 2009*
Users who downloaded this article also downloaded:
Haiyan J iang, Ahsan Habib, (2009),"The impact of different types of ownership concentration on annual
report voluntary disclosures in New Zealand", Accounting Research J ournal, Vol. 22 Iss 3 pp. 275-304http://dx.doi.org/10.1108/10309610911005590
Parulian Silaen, Robert Williams, (2009),"Management control systems: a model for R&D units",
Accounting Research J ournal, Vol. 22 Iss 3 pp. 262-274http://dx.doi.org/10.1108/10309610911005581
Ning Du, J ohn E. McEnroe, Kevin Stevens, (2014),"The joint effects of management incentive and
information precision on perceived reliability in fair value estimates", Accounting Research J ournal, Vol. 27
Iss 2 pp. 188-206http://dx.doi.org/10.1108/ARJ -10-2012-0081
Access to this document was granted through an Emerald subscription provided by emerald-srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The effects of forecast speci?city
on the asymmetric short-window
share market response to
management earnings forecasts
Howard Chan
Department of Finance, The University of Melbourne, Melbourne, Australia
Robert Faff
Department of Accounting and Finance, Monash University, Melbourne, Australia
Yee Kee Ho
Department of Finance and Accounting, National University of Singapore,
Singapore, and
Alan Ramsay
Department of Accounting and Finance, Monash University, Melbourne, Australia
Abstract
Purpose – This study aims to test the effects of forecast speci?city on the asymmetric short-window
share market response to management earnings forecasts (MEF).
Design/methodology/approach – The paper examines a large sample of hand-checked Australian
data over the period 1994 to 2001. Using an analyst news benchmark, it estimates a series of regressions
to investigate whether the short-term impact from bad news announcements is greater in magnitude
than from good news announcements and whether this differs between routine and non-routine MEFs.
Additionally, it examines whether (after controlling for news content of MEF) there is a differential
market impact conditional on speci?city: minimum versus maximum versus range versus point.
Findings – The results indicate that an asymmetric response is evident for the overall sample and a
sub-set of non-routine forecasts. Contrary to predictions, the results show that forecast speci?city,
minimum, maximum, range and point MEFs make no additional contribution to the differences in the
market reaction to bad or good news.
Originality/value – The study extends the research investigating the short-run market impact of
MEFs. The main element of innovation derives from the interaction between speci?city and news
content, as well as distinguishing between routine versus non-routine cases. Notably, it found little
support for the view that more speci?c forecasts elicit greater market responses. What the results do
suggest is that managers appear to choose the form of the forecast to suit the news being delivered. In
particular, bad news delivered in a minimum forecast seems to be ignored by the market.
Keywords Australia, Financial forecasting, Investors
Paper type Research paper
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1030-9616.htm
TheauthorswouldliketothankPeter Clarkson, Peter Easton, AbeHerzberg, GregPoundandPhilip
Sinnadurai andworkshopparticipants at the AAA2005Conference at SanFrancisco, the AFAANZ
2004 Conference at Alice Springs, Monash University, University of Auckland, University of
Adelaide, and University of Melbourne for comments on earlier drafts. They also thank Alex
Featherstone for researchassistance. Financial assistance providedbyMonashUniversityFaculty
of Business and Economics research grant is gratefully acknowledged. The ?rst author gratefully
acknowledges ?nancial assistance provided by a Monash University Research Fellowship.
Effects of
forecast
speci?city
237
Accounting Research Journal
Vol. 22 No. 3, 2009
pp. 237-261
qEmerald Group Publishing Limited
1030-9616
DOI 10.1108/10309610911005572
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
1. Introduction
We investigate the share price impact of management earnings forecasts (hereafter,
MEFs). Speci?cally, we document and analyse whether there is a differential (and
greater in absolute magnitude) response to bad news MEFs as compared to good news
MEFs given the different possible types of MEFs. Moreover, our main focus is to
investigate the effect of alternative degrees of speci?city in MEFs (point, range,
minimum, maximum) on the asymmetric share market response to bad news compared
to good news MEFs[1]. We designate this the “conditional speci?city” hypothesis. It
extends the “news asymmetry” hypothesis that has been previously documented for
US markets.
For situations in which markets are incomplete and investors have behavioural
biases (Daniel et al., 1998), it is not unreasonable to assume that the market may react
asymmetrically to different types of MEFs or forecast speci?city[2]. The main prior
research in this area (Baginski et al., 1993 and Pownall et al., 1993) relies on the
expectations adjustment hypothesis (Ajinkya and Gift, 1984) which argues for a
symmetric relation between good and bad news MEFs and security prices. However,
Zhang (2006) stresses the importance of conditioning tests of market reaction on the
news content of the information. The existence of and reasons for asymmetric response
to MEFs have recently been investigated by Kothari et al. (2007)[3]. Skinner (1994)
provides limited evidence on the effect of forecast speci?city on the asymmetric share
market response to bad news MEFs. However, his results are inconclusive. Thus the
effect of forecast speci?city on the market response to MEFs remains an open question.
The effect of forecast speci?city on the share market response to the unexpected
component of MEFs is important from the perspective of information theory. Kim and
Verrecchia (1991) develop a theoretical model, which predicts that the extent of price
change at the time of a public announcement (for example, a MEF) is proportional to
both the unexpected portion of the announcement and the precision of the information
contained in the announcement. Baginski et al. (1993) test this proposition using the
share market reaction to MEFs that vary in speci?city. They ?nd evidence consistent
with the Kim and Verrecchia prediction, that the share price reaction is a positive
function of forecast speci?city. However, the Baginski et al. (1993) analysis does not
condition the market response on the types of the news contained in the MEFs. Further,
it has been shown that the method used by both Baginski et al. (1993) and Pownall et al.
(1993) to determine the news component of minimum and maximum forecasts is
mis-speci?ed. We contribute to the literature in the area of assessing the news content
of MEFs by using the deviation from analysts’ forecasts as a measure of the
unexpected portion of a public announcement. We also contribute by using a
regression model that allows all four forms of forecast speci?city to interact with both
good and bad news. Using this framework, we test the Kim and Verrecchia (1991)
proposition that forecast speci?city is positively related to market reaction.
From the perspective of disclosure policy, the effect of forecast speci?city on the
share market response is also important. Managers must choose not only whether to
make a MEF but the form (speci?city) of the MEF. Evidence indicates that managers’
forecast speci?city choices are related to the news content of the forecast. Baginski et al.
(1994), Skinner (1994) and Kasznik and Lev (1995) document that forecast news content
and forecast speci?city are related. For example, maximum (upper bound) forecasts are
usually associated with bad news, while minimum (lower bound) forecasts are usually
ARJ
22,3
238
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
associated with good news. Libby et al. (2006, p. 208) argue that “. . . psychology
research suggests a number of possible interactive effects of guidance errors and
guidance form” and their consequential impact on actual earnings announcement and
we will extend this to share price responses. Therefore, managers face a correlated
challenge in releasing a MEF.
Australia provides a different institutional and reporting environment to examine
the asymmetric share prices effect of MEFs and the interactions of forecast speci?city
and news content. First, there is a longer period between routine earnings
announcements as compared to the USA since it has a quarterly reporting
requirement while Australia has a half-yearly reporting requirement for listed
companies. As a result, management earnings forecasts should have a greater role in
informing the market than in the USA since the Australian rule requires a less frequent
routine reporting. Second, the continuous disclosure requirements[4] and reporting to
the Australian Stock Exchange (ASX) results in a central depository of announcements
recorded in the ASX Signal G database that is released simultaneously to all investors
by the ASX. Thus, a further feature of our paper is that, we source our MEF data
directly from the of?cial company announcement lodged with the ASX. We examine
the content of announcements to ensure that all earnings-related cases are identi?ed.
Accordingly, our investigation covers an extensive sample of over 1,600 MEFs for the
period September 1994 to December 2001.
The mainthrust of our ?ndings canbe summarisedas follows. We ?ndsupport for the
“news asymmetry” hypothesis in that the absolute magnitude of the market response to
badnewsMEFsissigni?cantlylarger thanthemagnitudeof thepositiveresponsetogood
news MEFs. Thus we con?rm Skinner’s (1994) ?nding and the recent results of Kothari
et al. (2007). We extend the prior literature by showing that news asymmetry is largely
con?ned to forecasts issued in response to the ?rm’s continuous disclosure obligations
(non-routine forecasts) and is not observed as strongly for forecasts accompanying
routine information events (routine forecasts). In addition, we have also found that the
asymmetry continues to persist in different forecast speci?city.
In relation to our “conditional speci?city” hypothesis, our initial univariate analysis
?nds bad news asymmetry for maximum, range and point forecasts but for forecasts in
the form of a minimum, the asymmetry is reversed – there being a signi?cantly larger
response to good news rather than bad news minimum MEFs. After conditioning our
regression analysis on the types of news, the results of our multivariate analysis differ
from those of Baginski et al. (1993) and are contrary to the predictions of the Kim and
Verrecchia (1991) model. Speci?cally, using the analysts forecast news benchmark, for
good news MEFs there is no evidence that forecast speci?city has any differential
effect on the market response. Our results show that while the market responds equally
strongly to bad news MEFs in the form of point, range or maximum style MEFs,
minimum style MEFs containing bad news are essentially ignored by the market. This
effect is observed overall and for both routine and non-routine MEFs. Our ?ndings are
consistent with Pownall et al. (1993) and Hirst et al. (1999).
The results are consistent with the view that market participants focus more on the
news content of non-routine forecasts (news asymmetry is mainly found for
non-routine forecasts) and less on the form of the forecast (no signi?cant conditional
speci?city effects are observed for this group of forecasts). Routine forecasts are
planned information releases and, as such, the market may well have already
Effects of
forecast
speci?city
239
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
anticipated the information contained in these events. For routine forecasts, market
participants seem to focus more on the forecasts form (signi?cant conditional
speci?city effects are found) and less on the forecast news content (little news
asymmetry is found for routine forecasts).
The remainder of the paper is structured as follows: section 2 outlines prior research
leading to the formulation of testable hypotheses. Section 3 outlines our research
method, data and sample, section 4 outlines and discusses our results and section 5
presents a summary and conclusions.
2. Theory development
2.1 Asymmetric market reaction
Our paper is based on two main premises:
(1) The market responds asymmetrically to good and bad news MEFs.
(2) The forecast speci?city of MEFs affects the asymmetric market response to
good and bad news forecasts.
Analytical research on discretionary information disclosure indicates that, in the
absence of costs, full disclosure will occur (Grossman, 1981; Milgrom, 1981). However,
in the presence of costs and given uncertainty about management’s private
information, management may resort to voluntary disclosure to reduce the level of
information asymmetry (Lennox and Park, 2006). On the other hand, the failure by
management to release a MEF does not necessarily imply no news. For example,
Waymire (1985) proposes that managers are unwilling to release good news especially
when earnings volatility is high because of the risk of litigation if actual earnings fall
short of the expected forecast. Whatever the incentives for disclosure, prima facie,
management’s release of a MEF is a major information event and a signi?cant share
market response might be expected.
Early research (for example, Patell, 1976) found that MEFs have information
content in the sense that such forecasts elicit a signi?cant share market response. This
is con?rmed by Pownall et al. (1993) who conclude that MEFs are highly informative
(although less so than earnings announcements). Ajinkya and Gift (1984) hypothesize
that managers issue voluntary MEFs when they believe that the market’s earnings
expectations are signi?cantly out of line with their own earnings expectations. In order
to avoid large movements in share prices at earnings announcement, management is
motivated to voluntarily release a MEF to adjust the market’s expectations more in line
with their proprietary information. Ajinkya and Gift (1984) argue that this motivation
for forecast disclosure applies equally to both good and bad news MEFs.
In relation to voluntary MEFs, Penman (1980) concludes that in an unregulated
market, on average, ?rms with good news are more likely to disclose MEFs than those
with bad news. However, in the light of possible signi?cant market over-reactions and
lawsuits by investors, ?rms are wary of not informing the market about bad news in a
timely manner. Skinner (1994) investigates the voluntary disclosure of bad news MEFs
by smaller US listed companies. He ?nds that voluntary MEFs are more likely to occur
when there are large negative earnings surprises. He argues that, contrary to Ajinkya
and Gift (1984) and Penman (1980), managers behave as if they bear large costs when
investors are surprised by large negative earnings surprises[5]. Skinner (1994)
attributes the increased propensity to preannounce bad news to a reduction in
ARJ
22,3
240
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
litigation costs resulting from the voluntary release of bad earnings news. Investors
who lose money may sue managers for failing to promptly disclose bad earnings news.
Lawsuits may result in personal liability on behalf of managers. A further reason for
voluntary disclosure is that managers incur reputational costs if they fail to disclose
bad news in a timely fashion. Firms that fail to keep the market informed may be
shunned by fund managers and ?nancial analysts. In a related line of argument,
Matsumoto (2002) and Brown and Higgins (2005) report the use of management
forecast guidance to avoid negative earnings surprises.
More recent research has focused on differentiating the share price effects of different
categories or types of forecasts. The issue of whether the share price effects differ between
good news and bad news MEFs is addressed by Skinner (1994). He investigates the
two-day (0, þ1) share price effects to MEFs in smaller US listed ?rms. Skinner de?nes
good and bad news relative to previous year earnings. He predicts and ?nds that the
absolute value of the share price effect is larger for bad news MEFs than for their good
newscounterparts[6]. KasznikandLev(1995) haveasomewhat different focus bylooking
at the issue of whether investors respond differently to large earnings surprises
dependingonwhether the ?rmhas releasedthe informationgradually(issueda warning)
or not. They ?nd a signi?cantly more negative share price effect for ?rms that issue
warnings. Kothari et al. (2007) investigate the asymmetric share price effect of MEFs
using more recent US data (1995-2002). They ?nd strong evidence of a greater reaction to
bad news MEFs. Australian evidence on the share price effect to voluntary MEFs comes
fromGallery et al. (2006). They test the share price effect in the ?ve days (25, þ5) before
and after the release of the MEFs. They ?nd a signi?cant share market response for their
overall sample and for the bad news MEFs, but not for the good news MEFs.
In addition, the market reaction to a MEF may depend on the circumstances
surroundingthe release of the forecast. ASXlistingrules andtheAustralianCorporations
Act require Australian listed companies to continuously disclose information likely to
haveamaterial effect onshare price. MEFscanbereleasedaspart of aroutine information
event (for example the company’s annual general meeting or the mandatory half-yearly
earnings announcement). Onthe other hand, MEFs couldbe standalone andindependent
of the routine reporting of Australian companies. It can be argued that such non-routine
MEFs may be less anticipated by the market and thus may be more likely to exhibit
greater asymmetric reactions to bad news because of greater information content.
Given the previous, we propose the following hypotheses conditional on the context
in which the MEF was released:
H1A. The short-window negative share price effect to bad news MEFs is
signi?cantly larger in absolute magnitude than the positive share price
effect to good news MEFs when the MEF is part of a non-routine
announcement.
H1B. The short-window share price effect to bad or good news non-routine MEFs
is signi?cantly larger in absolute magnitude than the bad or good news
routine MEFs, respectively.
2.2 The effect of forecast speci?city on the asymmetric market reaction
We draw on prior studies which ?nd that management will choose the type of MEFs to
match the precision of their own information about the ?rm’s future (Libby et al., 2006;
Effects of
forecast
speci?city
241
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
King et al., 1998). The effect of forecast speci?city on the share market reaction to
MEFs is investigated by Baginski et al. (1993). In deriving their hypotheses, Baginski
et al. rely on the expectations adjustment hypothesis (Ajinkya and Gift, 1984), which
posits a symmetric relation between management forecast news and security prices.
In terms of the effect of forecast precision, Baginski et al. (1993) also draw on the
Kim and Verrecchia (1991) two-period rational expectation model. Using this model,
Kim and Verrecchia (1991) ?nd that the price change at the time of an announcement
such as a MEF is a function of the unexpected portion of the announcement (“news”)
and the relative importance of the announcement to the prior beliefs of traders. The
relative importance is positively related to the precision of the announcement and
inversely related to the precision of pre-announcement information. Based on this
analysis, Baginski et al. (1993) hypothesise and ?nd evidence that the stock price
response to a MEF is a decreasing function of forecast imprecision.
The arguments and evidence of Baginski et al. (1993) can be criticised on the
grounds that they assume a symmetric market response to good and bad news MEF,
yet subsequent evidence shows a signi?cantly larger reaction to bad news MEFs
(Skinner, 1994; Kothari et al., 2007). Further, the choice of forecast form (minimum,
maximum, range, point) is not independent of news content (Pownall et al., 1993)[7].
The Kim and Verrecchia (1991) model relied on by Baginski et al. (1993) indicates that
the share price effect is initially determined by the unexpected portion of the
announcement, thus providing further support for the need to control for the types of
news. Finally, the method used by Baginski et al. (1993), to determine the unexpected
portion of minimum and maximum forecasts, have been shown to be misspeci?ed
(Pownall et al., 1993, Table 7). These arguments cast doubt on the reliability of the
Baginski et al. (1993) conclusions.
A further source of support for the possible interaction between the form of MEFs
and the precision of the signal and their impact on share prices can be drawn from the
psychology literature. Speci?cally, this literature ?nds a general preference for more
precision over less precision (Kuhn and Budescu, 1996; Kuhn et al., 1999) and the
consequence of a less precise signal is that it results in less extreme reactions (Curley
et al., 1986; Einhorn and Hogarth, 1985).
A review of the empirical research indicates a lack of consistent ?ndings concerning
the relationship between forecast speci?city, types of news from MEFs and
asymmetric share price responses. Pownall et al. (1993) investigate the share price
effects of forecasts that differ by horizon and form (minimum, maximum, range, point;
qualitative forecasts were not included). They conclude that there is no signi?cant
difference in market reaction between minimum, maximum, range and point
quantitative MEFs. This is further supported by the ?ndings in Hirst et al. (1999).
Baginski et al. (1993) investigate whether a more imprecise MEF leads to a more
muted share market response. Consistent with their hypothesis, they ?nd that MEF
imprecision is negatively related to the strength of the market reaction. That is,
minimum and maximum forecasts (less speci?c) produce lower share price responses.
Both Pownall et al. (1993) and Baginski et al. (1993) do not condition their analyses on
the news content of the MEF. Further, both rely on a method of assessing the
unexpected portion of the MEF that amounts to assuming that the lower bound for
minimums and the upper bound for maximums correspond to the mean of the
manager’s expectations. As such, in their method, the fact that management have
ARJ
22,3
242
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
identi?ed an upper or lower bound to their MEF is effectively ignored in determining
the news content of that MEF.
Skinner (1994) investigates the effect of forecast speci?city on the differential share
price response for bad versus good news MEFs. He ?nds a signi?cantly greater share
price response to bad news MEFs for the overall sample and for point forecasts, but no
signi?cant difference for bad versus good news MEFs for range, lower/upper bound or
qualitative forecasts. Small sample sizes (between 34 and 67 observations within each
of his ?ve forecast categories) may have resulted in low power for his statistical tests.
There is a compelling argument that point forecasts are the most precise and thus they
should evoke the strongest reactions – stronger than range or minimum or maximum
forecast (Libby et al., 2006; see Wallsten et al., 1986; and Highhouse, 1994; from the
psychology literature). Skinner (1994) does not test for differences in response across
different forecast speci?city categories. Thus the share price effects of forecast
speci?city on the asymmetric response to MEFs remain an open question.
Given the lack of strong theory and the lack of consistent empirical results in the
prior literature, we hypothesize H2, a “conditional speci?city” hypothesis, in null form:
H2. After controlling for the news content of the MEF, there is no difference in
the short-window share price response to different forms (minimum,
maximum, range, point) of MEFs.
On the presumption that point forecasts are more precise in comparison to range,
minimum and maximum forecast, the following is an alternative hypothesis to H2:
H2A. After controlling for the news content of the MEF, the short-window share
price response to point MEFs is greater than the share price response for
minimum, maximum or range MEFs.
3. Research method
3.1 Data and sample
Our sample comprises of Australian listed companies that are followed by analysts,
over the period September 1994 to December 2001. Our primary motivation for
imposing the analyst coverage restriction is that the central core of our analysis
focuses on using analysts’ earnings forecasts as a benchmark for calibrating the types
of news for the MEFs[8]. Analyst earnings forecast data are sourced from the
Institutional Broker Estimates System (I/B/E/S) summary earnings ?le. MEFs data are
sourced from the of?cial Signal G electronic records of all company announcements to
the ASX[9].
For documentation and coding of MEFs, the following announcements were
manually checked and read:
.
announcements within one week prior to and following the company’s AGM for
the prior ?nancial year (especially the Chairman’s address);
.
announcements within one week prior to and following the release of the current
half yearly earnings announcement; and
.
all announcements within three months prior to the current annual earnings
announcement.
Effects of
forecast
speci?city
243
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In addition, the headlines of all announcements made by analyst-followed companies
between the AGM for the prior ?nancial year and three months prior to the current
annual earnings announcement were checked and, where necessary, the full text of the
announcement was read. Finally, share price data were obtained from SIRCA.
Initially, we de?ned any Signal G announcement in which management referred to
current period or future period pro?ts, revenues, distributions/dividends or production
as an earnings-related prior announcement. This process identi?ed 2,424 such
announcements. Subsequently, announcements that did not speci?cally refer to pro?t
(totalling 781 in number) were removed and not analysed. This left an initial sample of
1,643 MEFs made by 464 different companies. To run the regression models reported
later in the paper, we restricted the sample to those cases where:
.
valid share price data were available;
.
the forecast horizon exceeds three days; and
.
a valid news classi?cation (either bad, neutral or good) could be assigned from
either or both the analysts and prior year earnings benchmarks or the
observation related to a qualitative forecast.
This resulted in a ?nal sample size of 1,472. Where multiple pro?t measures were
referred to in a single announcement, we analyse the measure closest to net pro?t after
tax[10].
3.2 Variable de?nitions
Forecast speci?city. We classify MEFs into point, range, maximum, minimum and
qualitative categories (Skinner, 1994). In a point forecast, a precise single numerical
?gure is given ($X) or can be readily quanti?ed (for example, pro?t is expected to equal
that of the previous year). A range forecast contains a precise numeric range of pro?t
(for example, “pro?t will be between $X and $Y”) or the range can be readily quanti?ed
(for example, “pro?t will be within 10 per cent of last year”).
A maximum forecast sets a maximum or upper bound to pro?t, whereas in a
minimum forecast, a minimum or lower bound to pro?t is speci?ed. Qualitative
forecasts provide a general statement that is not capable of any precise numeric
interpretation.
In testing H2, we allow the different speci?cities that can be assigned a news
outcome to be recognised in their own right as a series of individually de?ned dummy
variables: D
Point
, D
Range
, D
Min
and D
Max
.
Classi?cation into good/neutral/bad news MEFs. This study determines good/
neutral/bad news MEFs with respect to the nearest analysts’ forecasts. Using the
analysts’ forecasts, our preferred method was to compare the MEF (typically an
aggregate pro?t number) to the mean analyst earnings forecast (an earnings per share
value) at the date of the announcement (Gallery et al., 2006). The following procedure as
summarised in Figure 1 was employed.
First, using I/B/E/S data, the number of shares on issue was obtained. This was
multiplied by the mean analyst EPS forecast obtained immediately prior to the release
of the MEF. The number of shares and the mean analyst EPS forecast from the I/B/E/S
detail ?le result in an implied analyst net pro?t after tax – this is our analysts
benchmark (AB). Next, the implied analyst net pro?t after tax was compared to the
MEF. If the point earnings forecast (MF
p
) or the mid-point of a range earnings
ARJ
22,3
244
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast[11] (MF
r
) was more than 10 per cent above (below) the implied analyst net
pro?t after tax (AB), this was classi?ed as good (bad) news, while all remaining valid
cases of point/range MEF are classed as neutral news.
For MEFs in the form of a maximum or minimum[12], we ?rst determine the most
speci?c dollar amount we could attach to the forecast (S
min
and S
max
, respectively).
Figure 1.
Characterisation of news
categories associated with
management earnings
forecasts
Effects of
forecast
speci?city
245
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
S
min
and S
max
were then compared to the implied analyst net pro?t after tax as follows:
If S
min
is greater than 10 per cent above (40 per cent below) the implied analyst pro?t, it
is classi?ed as good (bad) news, and otherwise it is deemed neutral news. If S
max
is
greater than 40 per cent above (10 per cent below) the implied analyst pro?t, it is
classi?ed as good (bad) news, and otherwise it is deemed neutral news[13,14].
For the hypothesis testing, the news variables based on analyst forecasts are a
series of dummy variables labelled D
Good
; D
Neut
; and D
Bad
and are scored one when the
decision rules mentioned previously are applied and zero otherwise.
Routine/non-routine announcements. The market reactions to MEFs may depend on
whether the MEF is released as part of a routine information event. Routine
announcements containing MEFs are those made at or accompanying the annual
general meeting (AGM) or the half-yearly pro?t announcement. ASX listing rules and
the Australian Corporations Act require Australian companies to continuously
disclose information likely to have a material effect on share price. Announcements
containing MEFs other than those mentioned previously are considered non-routine
and deemed to arise from the company’s continuous disclosure obligations. For
hypothesis testing, we use a dummy variable labelled D
NR
which is scored one when
the announcement is non-routine and zero otherwise.
Event dates. The timing of MEFs is taken as the date of the Signal G announcement
that contained the forecast. The annual earnings announcement is the date of the
preliminary annual earnings announcement from IRESS. Consensus analysts’
forecasts dates immediately prior to the release of the MEF are selected from the
I/B/E/S database.
Abnormal returns. A measure of the share price response to MEFs is required to test
investors’ reaction. A ?ve-day buy and hold return (BH5) measured from two days
before to two days (22, þ2) after the issue of the MEF. The buy-hold stock returns are
calculated as the stock return less the market return (proxied by the All Ordinaries
Accumulation Index) over the corresponding period. Prior literature such as Brown
and Warner (1985) and Shevlin (1981) suggest that such a parsimonious model does a
good job at isolating the market impact of events and typically produce results that are
at least as reliable as the more complex counterparts. Finally, to help guard against the
undue in?uence of outliers the top and bottom 1 per cent of BH5 observations were
trimmed.
Other variables. We also include a number of control variables, namely, ?rm size
(SIZE), forecast horizon measure (FHOR), year dummies (D_YR_MEF_96 –
D_YR_MEF_01) and prior year earnings loss dummy (D_PRIORYR_LOSS)[15].
SIZE is measured as the natural logarithm of the market value of the ?rm’s equity[16],
FHOR is the number of calendar days between the release of the earnings forecast and
the date of the release of the preliminary annual earnings announcement, D_YR are
calendar year dummies for the year in which MEFs occur and D_PRIORYR_LOSS is a
dummy variable equal to 1 if the ?rm recorded an earnings loss in the prior year.
4. Results
4.1 Sample descriptives
Tables I and II provide some descriptive information about our sample of MEFs.
Table I shows the composition of our sample forecasts in terms of speci?city.
Minimum and point forecasts together represent around 76.4 per cent of our sample,
ARJ
22,3
246
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
with maximum, range and qualitative forecasts making up the remainder. In Table II
we show the breakdown of news and its relationship to forecast speci?city using the
analysts forecast benchmark. In Table II, we see that minimum forecasts represent
nearly 50 per cent of good news cases, but only 22.7 per cent of bad news forecasts. On
the other hand, while maximum forecasts represent only 3.9 per cent of good news
cases, they constitute 19.9 per cent of bad news cases. Therefore, we can conclude that
management’s choice of forecast speci?city seems to be clearly related to the news that
the MEF is communicating. Thus, it is critical to analyse the interaction effects of
forecast speci?city and the types of news communicated[17].
4.2 News asymmetry hypothesis
First, we test for the Australian market that the absolute negative stock price response
to bad news MEFs is signi?cantly larger in magnitude than the positive stock price
response to good news MEFs. Using the analyst forecast basis of assessing news
content, the relevant speci?cation is found in equation (1)[18]. Testing hypotheses H1A
and H1B is conducted using equation (2)[19]:
BH5 ¼ a þ b
1
D
Bad
þb
2
D
Good
þ1 ð1Þ
BH5 ¼ a þ g
1
D
Bad
*
D
NR
þ g
2
D
Bad
*
D
R
þg
3
D
Good
*
D
NR
þg
4
D
Good
*
D
R
þ 1 ð2Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
(D
Good
) is a dummy variable set to
unity when MEFs are deemed bad (good) news relative to the mean analysts earnings
benchmark; D
NR
(D
R
) is a dummy variable taking a value of unity when the
announcement is non-routine (routine) and zero otherwise; Newey-West HAC standard
errors and covariance was used (lag truncation ¼ 7).
Good news
“Indeterminate”
news Bad news Totals
Number (%) Number (%) Number (%) Number (%)
Minimum 88 14.4 441 72.3 81 13.3 610 49.6
48.6 63.7 22.7
Maximum 7 5.1 60 43.5 71 51.4 138 11.2
3.9 8.7 19.9
Range 8 11.4 25 35.7 37 52.9 70 5.7
4.4 3.6 10.4
Point 78 18.9 166 40.3 168 40.8 412 33.5
43.1 24.0 47.1
Totals 181 692 357 1,230 100
100 100 100
Table II.
Distributional properties
of management earnings
forecasts – analyst
benchmarked news
versus speci?city
Qualitative Minimum Maximum Range Point Total
Number 115 661 154 79 463 1472
(%) 7.8 44.9 10.5 5.3 31.5 100.0
Table I.
Distributional properties
of management earnings
forecasts – forms of
speci?city
Effects of
forecast
speci?city
247
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
After con?rming that b
1
, 0 and b
2
. 0, we test whether the absolute value of b
1
is
signi?cantly greater thanthe absolute value of b
2
. The results are shown inTable III[20].
Table III shows that b
1
ðb
2
Þ, the mean effect of bad (good) news MEFs on BH5, is
negative (positive) at 24.0 per cent (þ1.9 per cent) and signi?cant at the 1 per cent
level. While these results are as expected, prior Australian research (Gallery et al., 2006)
has failed to document a signi?cant positive impact for good news MEFs[21]. We also
conduct a Wald test of the equality of the absolute values of (a þb
1
) and (a þ b
2
).
This test of equality produces a p-value of 0.013 and, hence, we document that there is a
differential effect to good and bad news similar to the USA. Therefore, we can conclude
that the magnitude of the share price effect of bad news MEFs is signi?cantly greater
than that of good news MEFs (where news is benchmarked against analyst forecasts).
Table IV shows the results for testing H1A and H1B. It shows that for non-routine
news, the previously mentioned asymmetry results are likewise documented. However,
the asymmetry is not found for routine news. This provides support for H1A but only
in the context of non-routine news, which is consistent with the notion that non-routine
voluntary disclosure may have much more information content and thus invoke
greater share price responses. Table IV also shows that only for bad news, is the share
price response signi?cantly larger for non-routine news as compared to routine news.
Therefore, H1B is supported for the bad news case.
4.3 Conditional speci?city hypothesis tests
The conditional speci?city hypothesis (H2) tests whether, after controlling for news,
forecast speci?city affects the share price response to MEFs. We begin with an
Variable Coef?cient t-statistic p-value
C a 0.020 0.784 0.433
D
Bad
b
1
20.040
* *
26.772 0.000
D
Good
b
2
0.019
* *
3.755 0.000
SIZE b
3
20.002 21.588 0.112
FHOR b
4
0.000
* *
3.398 0.001
D_PRIORYR_LOSS b
5
20.008 20.631 0.528
D_YR_MEF_96 b
6
0.015
* *
2.836 0.005
D_YR_MEF_97 b
7
0.010 1.512 0.131
D_YR_MEF_98 b
8
0.006 1.054 0.292
D_YR_MEF_99 b
9
0.008 1.173 0.241
D_YR_MEF_00 b
10
0.000 20.016 0.987
D_YR_MEF_01 b
11
0.006 0.848 0.396
Overall (good versus bad news) ja þ b
1
j ¼ ja þ b
2
j
( p-value)
6.226
*
(0.013)
Adjusted R
2
0.076
Notes: This table reports the result of running the following regression model: BH5 ¼ aþ
b
1
D
Bad;j
þ b
2
D
Good;j
þ 1. where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy
variable set to unity when MEFs are deemed bad (good) news relative to the mean analysts earnings
benchmark; Newey-West HAC standard errors and covariance was used (lag truncation ¼ 7).
The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast horizon);
D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year dummies).
*
and
* *
represents 5 and 1 per cent level of signi?cance, respectively
Table III.
Basic test of news
asymmetry content
ARJ
22,3
248
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
analysis similar to Skinner (1994). Using the analysts forecast news benchmark we
conduct univariate tests of the absolute value of BH5 market response to MEFs
differing in forecast speci?city. The null form of this hypothesis suggests that after
controlling for the news content of the MEF, there is no difference in the short-window
share price response to different forms (minimum, maximum, range, point) of MEFs.
The results of this analysis are shown in Table V.
There are two notable results from Table V. The ?rst is that for minimum forecasts
the asymmetry is reversed 2 the response to good news MEFs is larger in absolute
magnitude than the response to bad news MEFs. For the sample overall, and for the
sub-samples of routine forecasts and non-routine forecasts, the absolute value of the
response to good news minimum style MEFs exceeds that of bad news minimum style
MEFs at the 1 per cent level. There is asymmetry for minimum forecasts, but in the
opposite direction to that argued by Skinner and found overall in the earlier analysis.
Minimum forecasts are generally associated with good news and markets react more
strongly to good news minimum MEFs than to bad news minimum forecasts. The
Variable Coef?cient t-statistic p-value
C a 0.040 1.501 0.134
D
Bad
* D
NR
g
1
20.062
* *
26.491 0.000
D
Bad
* D
R
g
2
20.018
* *
22.842 0.005
D
Good
* D
NR
g
3
0.025
* *
2.851 0.004
D
Good
* D
R
g
4
0.015
*
2.498 0.013
SIZE g
5
20.002
*
21.995 0.046
FHOR g
6
0.000 1.271 0.204
D_PRIORYR_LOSS g
7
20.010 20.829 0.407
D_YR_MEF_96 g
8
0.014
* *
2.653 0.008
D_YR_MEF_97 g
9
0.009 1.502 0.133
D_YR_MEF_98 g
10
0.005 0.869 0.385
D_YR_MEF_99 g
11
0.006 1.023 0.306
D_YR_MEF_00 g
12
0.000 20.059 0.953
D_YR_MEF_01 g
13
0.005 0.736 0.462
Non-routine H1A: ja þg
1
j ¼ ja þ g
3
j 17.825
* *
(Good versus bad news) ( p-value) (0.000)
Routine H1A : ja þ g
2
j ¼ ja þ g
4
0.377
(Good versus bad news) ( p-value) (0.539)
Bad news H1B : ja þ g
1
j ¼ ja þ g
2
j 44.207
* *
(Non-routine versus routine) ( p-value) (0.000)
Good news H1B: ja þ g
3
j ¼ ja þ g
4
j 0.912
(Non-routine versus routine) ( p-value) (0.340)
Adjusted R
2
0.091
Notes: This table reports the result of running the following regression model: BH5 ¼
a þ g
1
D
Bad
*D
NR
þ g
2
D
Bad
*D
R
þ g
3
D
Good
*D
NR
þ g
4
D
Good
*D
R
þ 1 where BH5 is a ?ve-day buy
and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs are deemed bad (good)
news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy variable taking a value
of unity when the announcement is non-routine (routine) and zero otherwise; Newey-West HAC
standard errors and covariance was used (lag truncation ¼ 7). The regression is estimated with
control variables: SIZE (?rm size); FHOR (forecast horizon); D_PRIORYR_LOSS (prior year loss
dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year dummies).
*
and
* *
represents 5 and 1 per cent
level of signi?cance, respectively
Table IV.
Test of news asymmetry:
Non-routine versus
routine forecasts
Effects of
forecast
speci?city
249
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
second notable result of the univariate analysis is that weakly for the overall sample
and more strongly for the non-routine forecasts, maximum, range and point forecasts
all show an asymmetrically large response to bad news MEFs. This is not observed for
routine forecasts.
The univariate approach considers each form of forecast speci?city separately and
does not combine all forms of forecast speci?city with all news outcomes in a single
regression model. We now formulate such a model to assess the effect of forecast
speci?city on returns after controlling for news (H2). Using analyst forecasts as the
basis for assessing news, the form of the model is given by equation (3):
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ1 ð3Þ
where D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum,
maximum, range and point forecasts, respectively. As mentioned previously, the
regression is estimated with control variables: SIZE; FHOR; D_PRIORYR_LOSS and
D_YR_MEF_96 – D_YR_MEF_01.
We chose this path of separating the types of speci?city for the following reasons.
First, if we use an aggregate concept of speci?city (for example the IMPRECISE
variable[22] as employed by Baginski et al., 1993) in the current context, it may well
mask important variations in market impact across different types of forecast. As such,
equation (3) provides a robust means by which we can simply but effectively model a
non-linear relation. Second, an aggregate form of speci?city imposes an
ordering/weighting and linearity, which in the context of assessing the market
impact may be overly restrictive. This is particularly an issue given the results in
Table V, which show strong good news asymmetry for minimum forecasts and strong
Pairwise comparison results
Overall sample Non-routine MEFs Routine MEFs
Minimum forecasts 22.91
* * *
(0.000) 6.95
* * *
(0.001) 7.19
* * *
(0.008)
(G . B) (G . B) (G . B)
Maximum forecasts 3.42
*
(0.065) 6.72
* *
(0.010) 1.04 (0.308)
(B . G) (B . G) (B . G)
Range forecasts 4.20
* *
(0.041) 6.82
* * *
(0.009) 2.49 (0.115)
Point forecasts 1.01 (0.315) 5.43
* *
(0.020) 0.13 (0.714)
(B . G)
Notes: H0: BH5
Good
s
2 BH5
Bad
s
¼ 0 versus H
a
: BH5
Good
s
2 BH5
Bad
s
– 0. Wald tests of the
difference in value between the absolute value of BH5
N
s
for MEFs of differing forecast speci?city
where superscript “N” is the type of news (good or bad) and the subscript “s” is the speci?city of the
MEF, namely, minimum, maximum, range and point forecast. Cases, in which the value of BH5
Good
s
is
smaller than the counterpart value of BH5
Bad
s
are indicated by (B . G), whereas the converse
situation is shown as (G . B). The values of the Wald statistic are shown in the Table, while the
p-value is in parentheses.
*
,
* *
and
* * *
represents 10 per cent, 5 per cent and 1 level of signi?cance,
respectively
Table V.
Test of asymmetric
speci?city hypothesis:
pairwise tests of the
difference in absolute
values between good and
bad news MEFs of
differing forecast
speci?city
ARJ
22,3
250
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bad news asymmetry for non-routine maximum forecasts. It should also be noted that
prior evidence documents that US markets react to minimum style MEFs as conveying,
on average, good news and maximum style MEFs as conveying, on average, bad news
(Baginski et al., 1994).
We determine the news content of MEFs independent of forecast speci?city and
then, after controlling for news content, assess whether the share market response is
affected by forecast speci?city. As such, the model outlined in equation (3), by allowing
each quantitative type of forecast to interact with good and bad news in its own right,
permits us to incorporate the asymmetry effect of bad news into our exploration of the
market reaction effects of forecast speci?city.
The interaction terms capture the effects of changes in forecast speci?city for good
and bad news MEFs and a variety of tests of joint equality of coef?cients are explored
to assess the validity of H2. The results of estimating this regression equation (3) using
the analyst forecast method to assess news content are shown in Tables VI-VIII.
Table VI shows the basic regression results for the overall sample, while Table VII
shows the results of testing the asymmetry in news content, controlling for speci?city.
Variable Coef?cient t-statistic p-value
C a 0.044
*
1.676 0.094
D
Min
* D
Bad
b
Bad,Min
20.001 20.129 0.897
D
Max
* D
Bad
b
Bad,Max
20.068
* * *
25.282 0.000
D
Range
* D
Bad
b
Bad,Range
20.073
* * *
23.272 0.001
D
Point
* D
Bad
b
Bad,Point
20.043
* * *
25.388 0.000
D
Min
* D
Good
b
Good,Min
0.023
* * *
3.765 0.000
D
Max
* D
Good
b
Good,Max
0.013 0.702 0.483
D
Range
* D
Good
b
Good,Range
0.001 0.072 0.942
D
Point
* D
Good
b
Good,Point
0.015
*
1.781 0.075
SIZE g
9
20.003
* *
22.367 0.018
FHOR g
10
0.001
* *
2.423 0.016
D_PRIORYR_LOSS g
11
20.014 21.139 0.255
D_YR_MEF_96 g
12
0.017
* * *
3.242 0.001
D_YR_MEF_97 g
13
0.011
*
1.739 0.082
D_YR_MEF_98 g
14
0.008 1.345 0.179
D_YR_MEF_99 g
15
0.007 1.104 0.270
D_YR_MEF_00 g
16
0.001 0.085 0.932
D_YR_MEF_01 g
17
0.008 1.184 0.237
Adjusted R
2
0.097
Notes: This table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance, respectively
Table VI.
Test of news asymmetry
and conditional
speci?city hypotheses –
basic estimation results
Effects of
forecast
speci?city
251
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table VIII, Panels A and B report the results of testing the “conditional speci?city”
hypothesis conditioned for bad and good news, respectively.
An interesting observation from Table VI is that three out of the four forecast
speci?cities for bad news, namely, maximum, range and point MEFs are signi?cantly
negative. Focusing on the asymmetry in news content for different forms of forecast
speci?city in Table VII, we ?nd that the absolute magnitude of the share price response
for bad news is greater than the counterpart reaction for good news, except in the case
of minimum form MEFs[23]. The differences are economically important. This result
further strengthens the asymmetry results documented in the earlier section for each of
the forecast speci?city types.
We test H2 for the difference in share price responses between forecast speci?city,
respectively controlling for bad and good news, in Table VIII, Panels A and
B. Speci?cally, Panel A reveals that the market impact of maximum, range and point
forecasts are all signi?cantly different from the minimum category, when assessed on a
pairwise basis (at the 1 per cent level). The group of three (maximum, range and point)
however, when assessed on a pairwise basis, are not signi?cantly different from each
other. Thus, in the case of bad news, it is the minimum forecast category, which drives
the speci?city result. To con?rm the overall rejection of H2 in the context of bad news,
Panel A reports the rejection of the full joint test of equality across all four-interaction
terms (with a p-value of 0.000). We test the Kim and Verrecchia (1991) model
predictions in H2A. We hypothesize that the share price reactions of point MEFs are
larger in magnitude in comparison to the other three types of MEFs because point
MEFs are the most precise. For bad news MEFs, this is only observed when point MEF
is compared with minimum MEF but not for the other two. Therefore we do not ?nd
general support for H2A.
News speci?city ja þ b
Good;I
j ¼ ja þ b
Bad;i
j
Minimum 12.762
*
(0.000)
Maximum 9.690
*
(0.002)
Range 39.503
*
(0.000)
Point 11,184
*
(0.001)
Notes: This table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5> is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
*
represents the 10 per cent level of signi?cance
Table VII.
Test of news asymmetry
and conditional
speci?city hypotheses –
Test of “news
asymmetry” hypothesis
conditional on news
speci?city (F-statistic and
p-value)
ARJ
22,3
252
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Now turning to Panel B for the tests of H2 conditioned on good news MEFs, the results
show that none of the pairwise combinations of speci?city are signi?cantly different
from each other at the 5 per cent level. Moreover, in a full joint test of equality across all
four-interaction terms for good news, this restriction cannot be rejected. Therefore,
H2A is not supported for the good news MEF case.
These results are inconsistent with the predictions of the Kim and Verrecchia (1991)
model. Minimum, maximum and range forecasts are less speci?c than point forecasts.
Again, other things being equal, the Kim and Verrecchia (1991) analysis would argue
for a declining impact on share prices as forecast precision falls. For bad news, this is
found for the minimum and point MEFs pair and the minimum and range MEFs pair.
For good news forecasts, this is not observed. What we do ?nd is that all forms of
forecast which contain good news, essentially have equal market impact.
Tables IX and X report results for tests of H2 and H2A for non-routine and routine
MEFs using equation (4):
Pairwise tests Min Max Range
Panel A: Test of speci?city hypothesis conditional on bad news (chi-square statistic and p-value)
Max 19.771
* *
(0.000) – –
Range 8.869
* *
0.037
(0.003) (0.848) –
Point 11.914
* *
2.880 1.691
(0.001) (0.090) (0.194)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
25.309
* *
(0.000)
Panel B: Test of speci?city hypothesis conditional on good news (chi-square statistic and p-value)
Max 0.343
(0.558) – –
3.274
*
0.313
Range (0.071) (0.576) –
0.732 0.013 1.042
Point (0.393) (0.908) (0.307)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
3.465
(0.325)
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when
MEFs are deemed bad (good) news relative to the mean analysts earnings benchmark;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range
and point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* *
and
*
represents 10 and 1 per cent level of signi?cance, respectively
Table VIII.
Test of news asymmetry
and conditional
speci?city hypotheses –
Test of “news
asymmetry” hypothesis
conditional on news
speci?city (F-statistic and
p-value)
Effects of
forecast
speci?city
253
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Bad;i;j
D
Bad
*
D
i
*
D
j
ð4Þ
þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Good; i
D
Good
*
D
i
*
D
j
þ 1
As before, the regression is estimated with control variables: SIZE; FHOR;
D_PRIORYR_LOSS and D_YR_MEF_96 - D_YR_MEF_01.
Variable Coef?cient t-statistic p-value
C a 0.051
*
1.872 0.061
D
Bad
* D
Min
* D
NR
b
Bad,Min,NR
20.022 20.746 0.456
D
Bad
* D
Min
* D
R
b
Bad,Min,R
0.004 0.429 0.668
D
Bad
* D
Max
* D
NR Bad,Max,NR
20.079
* * *
24.308 0.000
D
Bad
* D
Max
* D
R
b
Bad,Max,R
20.05
* * *
23.553 0.000
D
Bad
* D
Range
* D
NR
b
Bad,Range,NR
20.072
* * *
22.736 0.006
D
Bad
* D
Range
* D
R
b
Bad,Range,R
20.083
*
21.906 0.057
D
Bad
* D
Point
* D
NR
b
Bad,Point,NR
20.058
* * *
25.360 0.000
D
Bad
* D
Point
* D
R
b
Bad,Point,R
20.022
*
21.911 0.056
D
Good
* D
Min
* D
NR
b
Good,Min,NR
0.041
* * *
3.532 0.000
D
Good
* D
Min
* D
R
b
Good,Min,R
0.019
* * *
2.650 0.008
D
Good
* D
Max
* D
NR
b
Good,Max,NR
0.012 1.197 0.232
D
Good
* D
Max
* D
R
b
Good,Max,R
0.013 0.540 0.589
D
Good
* D
Range
* D
NR
b
Good,Range,NR
20.005 20.331 0.740
D
Good
* D
Range
* D
R
b
Good,Range,R
0.015 1.016 0.310
D
Good
* D
Point
* D
NR
b
Good,Point,NR
0.022
*
1.755 0.080
D
Good
* D
Point
* D
R
b
Good,Point,R
0.008 0.716 0.474
SIZE g
9
20.003
* *
22.429 0.015
FHOR g
10
0.000 1.372 0.170
D_PRIORYR_LOSS g
11
20.014 21.165 0.244
D_YR_MEF_96 g
12
0.016
* * *
3.045 0.002
D_YR_MEF_97 g
13
0.011
*
1.788 0.074
D_YR_MEF_98 g
14
0.007 1.150 0.250
D_YR_MEF_99 g
15
0.007 1.098 0.273
D_YR_MEF_00 g
16
0.000 0.020 0.984
D_YR_MEF_01 g
17
0.008 1.098 0.272
Adjusted R
2
0.101
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
b
Bad;i
D
Bad
*
D
i
þ
i¼{Min;Max;
Range;Point}
X
b
Good;i
D
Good
*
D
i
þ 1 ð3Þ
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs
are deemed bad (good) news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy
variable taking a value of unity when the announcement is non-routine (routine) and zero otherwise;
D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range and
point forecasts, respectively. Newey-West HAC standard errors and covariance was used (lag
truncation ¼ 7). The regression is estimated with control variables: SIZE (?rm size); FHOR (forecast
horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_MEF_96 – D_YR_MEF_01 (year
dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance, respectively
Table IX.
Test of speci?city
hypothesis conditioned
on news content:
non-routine versus
routine forecasts – basic
estimation results
ARJ
22,3
254
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Pairwise tests Min Max Range
Panel A: Test of speci?city hypothesis conditional on bad/non-routine news (chi-square statistic and p-value)
Max 3.198 – –
(0.074) – –
Range 1.622 0.063
(0.203) (0.802) –
Point 1.372 0.986 0.246
(0.242) (0.321) (0.620)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
3.383
(0.336)
Panel B: Test of speci?city hypothesis conditional on good/non-routine news (chi-square statistic and p-value)
Max 3.957
* *
– –
(0.047)
Range 6.512
* *
1.017 –
(0.011) (0.314) –
Point 1.298 0.428 2.030
(0.255) (0.513) (0.155)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
7.420
*
(0.060)
Panel C: Test of speci?city hypothesis conditional on bad/routine news (chi-square statistic and p-value)
Max 10.350
* * *
– –
(0.001) – –
Range 3.728
*
0.483 –
(0.054) (0.487)
Point 3.309
*
2.573 1.768
(0.069) (0.109) (0.184)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Bad news MEFs
14.041
(0.003)
Panel D: Test of speci?city hypothesis conditional on good/routine News (chi-square statistic and p-value)
Max 0.054 – –
(0.817) – –
Range 0.070 0.003 –
(0.791) (0.953) –
Point 0.758 0.043 0.172
(0.384) (0.836) (0.678)
Full joint test H0: b
Min
¼ b
Max
¼ b
Range
¼ b
Point
Good news MEFs
0.766
(0.858)
Notes: This Table reports the result of running the following regression model:
BH5 ¼ a þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Bad;i;j
D
Bad
*
D
i
*
D
j
ð4Þ
þ
i¼{Min;Max;
Range;Point}
X
j¼{NR;R}
X
b
Good; i
D
Good
*
D
i
*
D
j
þ 1
where BH5 is a ?ve-day buy and hold return, D
Bad
ðD
Good
Þ is a dummy variable set to unity when MEFs are
deemed bad (good) news relative to the mean analysts earnings benchmark; D
NR
ðD
R
Þ is a dummy variable taking
a value of unity when the announcement is non-routine (routine) and zero otherwise; D
Min
; D
Max
; D
Range
and D
Point
are dummy variables representing the minimum, maximum, range and point forecasts, respectively. Newey-West
HAC standard errors and covariance was used (lag truncation ¼ 7). The regression is estimated with control
variables: SIZE (?rm size); FHOR (forecast horizon); D_PRIORYR_LOSS (prior year loss dummy) and D_YR_
MEF_96 – D_YR_MEF_01 (year dummies).
* * *
,
* *
and
*
represents 1, 5 and 10 per cent level of signi?cance,
respectively
Table X.
Test of speci?city
hypothesis conditioned
on news content:
non-routine versus
routine forecasts
Effects of
forecast
speci?city
255
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Panel A shows that six out of the eight coef?cients for bad news are signi?cantly
negative while only three out of the eight coef?cients for good news are signi?cantly
positive (at the 10 per cent level). An interesting observation is that both bad news
coef?cients for minimum MEFs are insigni?cant while the two counterpart coef?cients
for good news are signi?cantly positive (at the 1 per cent level). This observation
reinforces the generally held belief that minimum MEFs are essentially good news.
Tables X, Panel A shows the chi-square test statistics for the ‘conditional speci?city’
hypothesis for bad and non-routine news. None of the pairwise tests are signi?cant and
thus for bad news of a non-routine type, the type of forecast speci?city has no bearing
on the magnitude of the share price response. As shown in Table X, Panel B, the same
general conclusion is true for good news and non-routine type MEFs (except in the
cases between minimum and range MEFs and minimum and maximum MEFs). Panels
C and D show the results of the counterpart tests of Panels A and B, only now the tests
are conducted on routine MEFs. The results are qualitatively similar, except in Panel C
where the price reaction to minimum MEFs is now found to be signi?cantly different
from maximum MEFs (1 per cent) and from range and point MEFs (10 per cent).
Our results are consistent with the ?ndings of Pownall et al. (1993) and Hirst et al.
(1999) where they conclude that there is no signi?cant difference in market reaction
between minimum, maximum, range and point MEFs.
5. Conclusions
This paper documents that the response to bad news MEFs is greater than the
response to good news MEFs. This ?nding is similar to that found in the US.
Furthermore, we investigate hypotheses regarding the short-window market reaction
to MEFs. First, we consider whether the negative stock price response to bad news
MEFs is signi?cantly larger in magnitude than the positive stock price response to
good news MEFs for non-routine announcements. Next, we test whether the magnitude
of the stock price response is larger for non-routine as compared to routine
announcements. In further analysis we consider a “conditional speci?city” hypothesis,
which proposes that after controlling for the news content of the MEF, there is no
difference in the stock price response to different forms (point, range, minimum,
maximum) of MEF.
Our tests, based on a large sample of Australian MEFs that are extensively checked
and sourced from the of?cial ASX database of company announcements, document
that there is a greater response to bad news MEFs in our overall sample and for the
sub-set of non-routine forecasts, but generally not for routine forecasts. This suggests
that the “unexpected” nature of non-routine MEFs is what surprises the market. Our
univariate tests of the “conditional speci?city” hypothesis ?nd bad news asymmetry
for maximum, range and point forecasts but for forecasts in the form of a minimum, the
asymmetry is reversed – there being a signi?cantly larger response to good news
rather than bad news minimum MEFs. This is a new ?nding in the literature.
Our paper also contributes to an important theoretical issue. Speci?cally, we use a
model that allows four forms of forecast speci?city to interact with both good and bad
news to test the Kim and Verrecchia (1991) proposition that the precision of MEFs is
positively related to the share market reaction to that forecast. Our results are
inconsistent with those of Baginski et al. (1993) and they do not support Kim and
Verrecchia’s prediction. We ?nd that, using the analyst forecast news benchmark, for
ARJ
22,3
256
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bad news MEFs, the market reacts equally negatively to point, range and maximum
forecasts, but is mute to bad news minimum forecasts. For good news forecasts, we are
unable to identify any difference in the market reaction based on the speci?city of the
forecast. Our further analysis shows that for non-routine forecasts, the form of the
forecast has no signi?cant effect on the market reaction. Our ?ndings are consistent
with Pownall et al. (1993) and Hirst et al. (1999), which ?nd that forecast speci?city has
little to contribute to the difference in market reaction of MEFs.
Notably, none of our results support the view that more speci?c forecasts elicit
greater market responses – except when we compare point and minimum MEFs. What
our results do suggest is that managers appear to choose the form of the forecast to suit
the news being delivered. In particular, bad news delivered in a minimum forecast
seems to be ignored by the market.
Notes
1. Skinner and Sloan (2002) document an asymmetric share price response of ?rms that fail to
meet analyst expectations as compared to those ?rms that meet or better analyst forecasts.
Libby et al. (2006, p. 207) ?nd, in an experimental market setting, that “the form of
management’s earnings guidance (point, narrow range, wide range) affects analysts’
earnings forecasts”.
2. Speci?city re?ects the precision of management’s earnings forecast. We analyse the effects
of forecast speci?city in four categories: point forecasts, range forecasts (both upper and
lower bound are speci?ed), minimum forecasts (lower bound only), and maximum forecasts
(upper bound only).
3. In studies pertaining to analysts earnings forecasts and market responses, many have found
support for the asymmetric responses to good and bad news (Ho and Sequeira, 2007; Skinner
and Sloan, 2002; Lopez and Rees, 2001; Conrad et al., 2002).
4. The continuous reporting requirement in Australia does not necessarily results in more
frequent and greater quantity of reporting. It basically requires companies to be more-timely
in reporting price sensitive information. This paper leverages on this unique disclosure
regime to test the share price impact of management earnings forecasts.
5. This argument is supported by the subsequent work of Skinner and Sloan (2002) which
argues that large negative surprises can “torpedo” the share price of a ?rm.
6. Hutton et al. (2003), document that good news MEFs generate signi?cant market reaction
only when accompanied by other veri?able forward looking statements, whereas there is a
signi?cant reaction to bad news MEFs irrespective of any accompanying disclosures.
7. Baginski et al. (1994) use the share price response to identify the “news” content of forecasts
differing in speci?city. They ?nd that the market reaction to minimum forecasts indicates
that they are, on average, perceived as good news, whereas the market reaction to maximum
forecasts indicates that they are, on average, bad news. As we wish to test the effect of
forecast speci?city on share returns, after controlling for the news content of the forecast,
their approach is not appropriate for our purposes.
8. Like any situation involving research design tradeoffs, restricting our sample to
analyst-covered ?rms comes at a cost. Speci?cally, our analysis is potentially impacted
by a selection bias due to the non-randomness in analysts’ stock coverage selection (Rajan
and Servaes, 1997). We gratefully acknowledge an anonymous referee for drawing this issue
to our attention.
Effects of
forecast
speci?city
257
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
9. We use Signal G data throughout but access this data in two different ways: First, for the
period September 1994 to 31 December 1995 all Signal G announcements are obtained from
the Securities Industry Research Centre for Asia Paci?c (SIRCA). Companies with analysts’
earnings forecasts available were manually read for any reference to earnings outlook,
forecasts or upgrades and such cases were carefully recorded in the database. Second, for the
period 1 January 1996 to 31 December 2001, the Signal G company announcements by all
analyst-followed ASX companies on the I/B/E/S database were accessed on Integrated Real
Time Equity System (IRESS). After 1 July 2001 the PDF signal on IRESS is used.
10. Measures of pro?t referred to include net pro?t before/after tax, before and after interest,
before and after abnormal items, before and after depreciation and amortisation.
11. We are mindful of the caution in Libby et al. (2006, p. 207) where they state “treating the
mean of the range endpoints as equivalent to a point estimate and failing to consider effects
after the release of actual earnings may paint an incomplete picture of how management
guidance affects analysts and investors”. There is currently no acceptable methodology to
capture the quantitative estimate for an MEF expressed as a range.
12. Some prior studies exclude maximum and minimum forecasts on the grounds that it is
dif?cult to unequivocally identify their news content (Ajinkya and Gift, 1984). Baginski et al.
(2002) rely on Baginski et al. (1994) and assume all minimum forecasts are good news and all
maximum forecasts are bad news. Such an approach would not allow us to assess the impact
of forecast speci?city after controlling for news (H2). Our approach to assessing the news
content of maximum and minimum forecasts differs from Baginski et al. (1993) in that they
do not use a neutral category and simply compare the upper bound (for maximum forecasts)
or the lower bound (for minimum forecasts) to the median analyst forecast.
13. The news content of qualitative forecasts cannot be determined and they are not considered
further.
14. One major criticism of this approach in classifying bad and good news is that the cut-off
score seems arbitrary. While our conservative approach is appealing (we allow a larger
margin for bad news minimum/good news maximum forecasts) it is not without concern. For
example, a company announcing that it expects to have an EPS of at least 5.9 cents against a
current analyst forecast benchmark of 10 cents per share, will be classi?ed as bad news.
However, the managers of this hypothetical company could be habitually conservative, and
the open-ended forecast still leaves open the potential for an actual pro?t that exceeds
consensus forecasts (we thank an anonymous referee for drawing our attention to this issue).
Our reaction to this general concern has two elements. First, there is currently no generally
“acceptable” methodology in applying a systematic cut-off score for classi?cation purposes.
Second, by varying the cut-off score, our results are generally qualitatively similar for
minimum, maximum and range MEFs. Details are suppressed in the interests of brevity.
15. Current year earnings loss is not used because it is highly correlated with other variables.
16. Given that our sample is con?ned to analyst-followed cases, it is biased toward larger
Australian companies. Further details, while suppressed to conserve space, are available
from the authors on request.
17. The small sample size for range, and maximum forecasts for good news events, means that
the more meaningful comparisons are between minimum and point forecasts for good news
cases. Also, the comparison between maximum and point forecasts for bad news cases is
more meaningful due to the concern about classifying minimum forecasts as bad news. We
thank an anonymous referee for drawing these issues to our attention.
ARJ
22,3
258
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
18. To enhance readability of all regression models, the control variables are not included in the
formal written equations (SIZE; FHOR; D_PRIORYR_LOSS and D_YR_MEF_96 –
D_YR_MEF_01).
19. To assess the effect of the three classes of news: good, neutral and bad; we specify dummy
variables for good and bad news in the regression model. The neutral news content is
embedded within the intercept term.
20. A correlation table was also constructed to check for possible problems with
multi-collinearity between the explanatory variables. The correlation results, although not
reported here, do not suggest that the ?ndings will be affected by any multi-collinearity
problems.
21. A possible reason why Gallery et al. (2006) do not ?nd a statistically signi?cant impact is due
to their small sample size. They only have in total 45 observations that are point, range,
minimum or maximum out of a total sample of 180 observations. Within this total sample
only 63.3 per cent are good news and within the good news only 15.8 per cent are correctional
forecasts (quantitative forecasts with deviations greater than 5 per cent from analysts’
forecasts).
22. Baginski et al. (1993) combine the four forms of forecast speci?city into a single variable
(IMPRECISE), which is given a value of zero for point forecasts, one for closed interval
forecasts (range) and two for open interval forecasts (minimum/maximum).
23. While it is true that the minimum case also reveals a signi?cant test result, readers should
note that it is re?ective of the converse scenario – namely, that the absolute magnitude of the
share price response for bad news is less than the counterpart reaction for good news.
References
Ajinkya, B.P. and Gift, M.J. (1984), “Corporate managers’ earnings forecasts and symmetrical
adjustments of market expectations”, Journal of Accounting Research, Vol. 22, pp. 425-44.
Baginski, S.P., Conrad, E.J. and Hassell, J.M. (1993), “The effects of management forecast
precision on equity pricing and on the assessment of earnings uncertainty”, Accounting
Review, Vol. 68, pp. 913-27.
Baginski, S.P., Hassell, J.M. and Kimborough, M.D. (2002), “The effect of legal environment on
voluntary disclosure: evidence from management earnings forecasts issued in US and
Canadian markets”, Accounting Review, Vol. 77, pp. 25-50.
Baginski, S., Hassell, J. and Waymire, G. (1994), “Some evidence on the news content of
preliminary earnings estimates”, Accounting Review, Vol. 69, pp. 265-73.
Brown, L.D. and Higgins, H.N. (2005), “Managers’ forecast guidance of analysts: international
evidence”, Journal of Accounting and Public Policy, Vol. 24, pp. 280-99.
Brown, S. and Warner, J.B. (1985), “Using daily stock returns: the case of event studies”, Journal
of Financial Economics, Vol. 14, pp. 3-31.
Conrad, J., Cornell, B. and Landsman, W.R. (2002), “When is bad news really bad news?”, Journal
of Finance, Vol. 57, pp. 2507-32.
Curley, S.P., Yates, J.F. and Abrams, R.A. (1986), “Psychological sources of ambiguity
avoidance”, Organizational Behavior and Human Decision Processes, Vol. 38, pp. 230-56.
Daniel, K., Hirshleifer, D. and Subrahmanyam, A. (1998), “Investor psychology and security
market under- and overreactions”, Journal of Finance, Vol. 53, pp. 1839-85.
Einhorn, H.J. and Hogarth, R.M. (1985), “Ambiguity and uncertainty in probabilistic inference”,
Psychological Review, Vol. 92, pp. 433-61.
Effects of
forecast
speci?city
259
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Gallery, G., Gallery, N. and Hsu, C-M.G. (2006), “The association between management and
analysts’ earnings forecasts in the Australian continuous disclosure environment”,
working paper, Queensland University of Technology and University of Queensland,
Brisbane.
Grossman, S. (1981), “The informational role of warranties and private disclosure about product
quality”, Journal of Law and Economics, Vol. 24, pp. 461-83.
Highhouse, A.A. (1994), “Verbal protocol analysis of choice under ambiguity”, Journal of
Economic Psychology, Vol. 15, pp. 621-35.
Hirst, D.E., Koonee, L. and Miller, J. (1999), “The joint effect of management’s prior forecast
accuracy and the form of its ?nancial forecasts on investor judgment”, Journal of
Accounting Research, Vol. 37, Supplement, pp. 101-24.
Ho, Y.K. and Sequeira, J. (2007), “Earnings surprises, asymmetry of returns and market level
changes – an industry study”, Journal of Accounting, Auditing and Finance, Vol. 22 No. 1.
Hutton, A.P., Miller, G.S. and Skinner, D.J. (2003), “The role of supplementary statements with
management earnings forecasts”, Journal of Accounting Research, Vol. 41, pp. 867-90.
Kasznik, R. and Lev, B. (1995), “To warn or not to warn? Management disclosures in the face of
an earnings surprise”, Accounting Review, Vol. 70, pp. 113-34.
Kim, O. and Verrecchia, R.E. (1991), “Trading volume and price reactions to public
announcements”, Journal of Accounting Research, Vol. 29, pp. 302-21.
King, R., Pownall, G. and Waymire, G. (1998), “How frequently do managers disclose prospective
earnings information?”, Journal of Financial Statement Analysis, Vol. 3, pp. 5-16.
Kothari, S.P., Shu, S. and Wysocki, P.D. (2007), “Do managers withhold bad news?”, research
paper, MIT Sloan, Cambridge, MA.
Kuhn, K.M. and Budescu, D.V. (1996), “The relative importance of probabilities, outcomes and
vagueness in hazard risk decisions”, Organizational Behavior and Human Decision
Processes, Vol. 68, pp. 301-17.
Kuhn, K.M., Budescu, D.V., Hershey, J.R., Kramer, K.M. and Rantilla, A.K. (1999), “Attribute
tradeoffs in low probability/high consequence risks: the joint effects of dimension
preference and vagueness”, Risk, Decision and Policy, Vol. 4, pp. 31-46.
Lennox, C.S. and Park, C.W. (2006), “The informativeness of earnings and management’s
issuance of earnings forecasts”, Journal of Accounting and Economics, Vol. 42, pp. 439-58.
Libby, R., Tan, H.T. and Hunton, J.E. (2006), “Does the form of management’s earnings guidance
affect analysts’ earnings forecasts?”, Accounting Review, Vol. 81, pp. 207-25.
Lopez, T.J. and Rees, L.L. (2001), The Effect of Meeting Analyst Forecasts and Systematic Positive
Forecast Errors on the Information Content of Unexpected Earnings, available at: http://
ssrn.com/abstract¼181929.
Matsumoto, D.A. (2002), “Management’s incentives to avoid negative earnings surprises”,
Accounting Review, Vol. 77, pp. 483-514.
Milgrom, P.R. (1981), “Good news and bad news: representation theorems and applications”, Bell
Journal of Economics, Vol. 12, pp. 380-91.
Patell, J. (1976), “Corporate forecasts of earnings per share and stock price behaviour: empirical
tests”, Journal of Accounting Research, Vol. 14, pp. 246-76.
Penman, S.H. (1980), “An empirical investigation of the voluntary disclosure of corporate
earnings forecasts”, Journal of Accounting Research, Vol. 18, pp. 132-60.
Pownall, G., Wasley, C. and Waymire, G. (1993), “The stock price effects of alternative types of
management earnings forecasts”, Accounting Review, Vol. 68, pp. 896-912.
ARJ
22,3
260
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Rajan, R. and Servaes, H. (1997), “Analyst following of initial public offerings”, Journal of
Finance, Vol. 52, pp. 507-29.
Shevlin, T. (1981), “Measuring abnormal performance on the Australian equity market”,
Australian Journal of Management, Vol. 6, No. 2, pp. 67-107.
Skinner, D.J. (1994), “Why ?rms voluntarily disclose bad news”, Journal of Accounting Research,
Vol. 32, pp. 38-60.
Skinner, D.J. and Sloan, R.G. (2002), “Earnings surprises, growth expectations and stock returns:
don’t let an earnings torpedo sink your portfolio”, Review of Accounting Studies, Vol. 7,
No. 2/3, pp. 289-312.
Wallsten, T., Budescu, D., Rapoport, A., Zwick, R. and Forsyth, B. (1986), “Measuring the vague
meanings of probability phrases”, Journal of Experimental Psychology: General, Vol. 115,
No. 4, pp. 348-65.
Waymire, G. (1985), “Earnings volatility and voluntary management forecast disclosure”,
Journal of Accounting Research, Vol. 23, pp. 268-95.
Zhang, X.F. (2006), “Information uncertainty and stock returns”, Journal of Finance, Vol. 61,
pp. 105-36.
About the authors
Howard Chan is an Associate Professor in the Department of Finance at the University of
Melbourne, Australia. Howard Chan is the corresponding author and can be contacted at:
[email protected]
Robert Faff is a Professor of Finance, in the Department of Accounting and Finance, at
Monash University, Victoria, Australia.
Yee Kee Ho is an Associate Professor in the Department of Accounting, National University
of Singapore, Singapore.
Alan Ramsay is an Associate Professor in the Department of Accounting and Finance,
Monash University, Victoria, Australia.
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
Effects of
forecast
speci?city
261
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
0
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_927040461.pdf