Description
This paper aims to use Australian analysts’ forecast data to compare the relative accuracy
of consensus and the most recent forecast in the month before the earnings announcement.
Accounting Research Journal
Australian evidence on the accuracy of analysts' expectations: The value of consensus
and timeliness prior to the earnings announcement
Xiaomeng Chen
Article information:
To cite this document:
Xiaomeng Chen, (2010),"Australian evidence on the accuracy of analysts' expectations", Accounting
Research J ournal, Vol. 23 Iss 1 pp. 94 - 116
Permanent link to this document:
http://dx.doi.org/10.1108/10309611011060542
Downloaded on: 24 January 2016, At: 21:10 (PT)
References: this document contains references to 39 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 985 times since 2010*
Users who downloaded this article also downloaded:
Katherine Uylangco, Steve Easton, Robert Faff, (2010),"The equity and efficiency of the Australian share
market with respect to director trading", Accounting Research J ournal, Vol. 23 Iss 1 pp. 5-19 http://
dx.doi.org/10.1108/10309611011060506
Raghavan J . Iyengar, J udy Land, Ernest M. Zampelli, (2010),"Does board governance improve
the quality of accounting earnings?", Accounting Research J ournal, Vol. 23 Iss 1 pp. 49-68 http://
dx.doi.org/10.1108/10309611011060524
Wendy Green, Richard D. Morris, Haiping Tang, (2010),"The split equity reform and corporate
financial transparency in China", Accounting Research J ournal, Vol. 23 Iss 1 pp. 20-48 http://
dx.doi.org/10.1108/10309611011060515
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Australian evidence
on the accuracy of analysts’
expectations
The value of consensus and timeliness prior
to the earnings announcement
Xiaomeng Chen
Department of Accounting and Finance, Macquarie University,
North Ryde, Australia
Abstract
Purpose – This paper aims to use Australian analysts’ forecast data to compare the relative accuracy
of consensus and the most recent forecast in the month before the earnings announcement.
Design/methodology/approach – Cross-sectional regression is used on a sample of 4,358
company-year observations of annual analyst forecasts to examine whether the number of analysts
following and the timeliness of an individual analyst’s forecast is more strongly associated with the
superior forecast measure.
Findings – The results suggest that whilst in the late 1980s the most recent forecast was more
accurate than the consensus, since the early 1990s the accuracy of the consensus forecast has
outperformed the most recent forecast in 15 out of 17 years, and the differences are signi?cant for nine
out of 15 years. The forecasting superiority of the consensus can be attributed to the aggregating value
of the consensus outweighing the small timing advantage of the most recent forecast over the short
forecast horizon examined in this paper.
Research limitations/implications – Given the consistent use of analysts’ forecasts as proxies for
expected earnings in Australian research, this paper provides insights to what extent the expected
level of forecast accuracy is realised and the reasons for the greater accuracy in the superior forecast
measure.
Practical implications – The ?ndings con?rm market practitioners’ views that the consensus
forecast is a better measure of the market’s earnings expectations.
Originality/value – This paper provides direct evidence of the accuracy of alternative forecast
measures and the importance of diversifying idiosyncratic individual error across analyst forecasts.
Keywords Earnings, Financial analysis, Financial forecasting, Australia
Paper type Research paper
1. Introduction
It is well established in the literature that analysts’ earnings forecasts are used as
proxies for market expectations of future earnings because they are more accurate and
have a stronger association with excess returns on the date of the earnings
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1030-9616.htm
The author gratefully acknowledges helpful comments from Sue Wright, Neil Fargher, Andrew
Ferguson, Paul Healy, Egon Kalotay, Terry Walter, Hai Wu, two anonymous referees and
participants at the Sydney Summer School, University of Technology, 31st EAA Annual
Congress 2008 and AFAANZ Conference 2008. The author also thanks I/B/E/S for providing
analyst forecast data.
ARJ
23,1
94
Accounting Research Journal
Vol. 23 No. 1, 2010
pp. 94-116
qEmerald Group Publishing Limited
1030-9616
DOI 10.1108/10309611011060542
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
announcement than time-series models of earnings (Brown and Rozeff, 1978; Fried and
Givoly, 1982; Brown et al., 1987a, b). Analysts’ earnings expectations can however be
measured in a number of ways. Two widely used approaches are a consensus forecast
that aggregates individual analyst forecasts at any point in time and a single, most
recent forecast provided by an individual analyst[1].
A consensus forecast diversi?es away idiosyncratic individual error to gain value
from the aggregation. The most recent forecast made over a shorter forecast horizon
than the consensus is more timely. Tradeoffs between the bene?ts of forecast
aggregation and timeliness of forecasts motivate this study to compare the relative
accuracy of the consensus andthe most recent forecast. This studyspeci?cally examines
whether the number of analysts following a company or the timeliness of an individual
analyst’s forecast is more strongly associated with the superior forecast measure.
Prior research in the late 1980s and early 1990s using data from the US market
examines the relative accuracy of alternative earnings forecast measures provided by
standard sources of analysts forecast data such as I/B/E/S and shows that the most
recent forecast is relatively more accurate than the consensus forecast (O’Brien, 1988;
Brown, 1991). Brown and Kim (1991) ?nd that the most recent forecast is more closely
related to share prices than the consensus forecast. Based on these results, many
studies use the most recent forecast as a measure of the market’s earnings expectations
(Brown, 2001a; Bartov et al., 2002; Brown and Caylor, 2005). However, recent studies
document that the consensus has become a more timely measure in the past decade due
to improvements in analyst forecast data (Barron and Stuerke, 1998; Brown, 2001a;
Ramnath et al., 2005). These studies suggest that attempts have been made to include
only relatively recent forecasts in the consensus to improve the timeliness of consensus
forecasts. Taken together, these ?ndings suggest that prior conclusions of the
superiority of the most recent forecast may no longer apply due to the changing nature
of consensus forecasts in more recent years.
Research in the Australian market has increasingly used analysts’ forecasts as
proxies for expected earnings (Brown et al., 1999; Beekes and Brown, 2006; Habib and
Hossain, 2008)[2]. Australian press reports overwhelmingly cite analysts’ earnings
expectations using the consensus rather than the most recent forecast[3]. While
extensive research on analysts’ forecasts is available for the US market, there is
relatively limited research related to analysts’ forecasts using Australian data
( Jackson, 2005; Brown et al., 2007; Aitken et al., 2008)[4].
This study contributes to prior research on Australian analysts’ forecasts in several
ways. First, this paper provides direct evidence of the accuracy of alternative forecast
measures, the consensus and the most recent forecast, as measures of the market’s
earnings expectations prior to earnings announcements. Recent studies document
improvements in timeliness of consensus forecasts, but these studies do not directly
compare the accuracy of the consensus and the most recent forecast. Given the consistent
use of analysts’ forecasts as proxies for expected earnings in Australian research and
press reports, it is important to understand to what extent the expected level of forecast
accuracy is realised and the reasons for the greater accuracy in the superior forecast
measure. Second, this study provides further evidence on the accuracy of the consensus
in reducing idiosyncratic error by diversifying across analyst forecasts in a market with
relatively few analyst coverage and a different disclosure regime. Last, it also con?rms
Accuracy
of analysts’
expectations
95
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the market practitioners’ views as evidenced by press reports, that the consensus
forecast is a better measure of the market’s earnings expectations.
The results suggest that whilst in the late 1980s, the most recent forecast is more
accurate than the consensus during the period immediately prior to the earnings
announcement, since the early 1990s the consensus forecast outperforms the most recent
forecast. That is, the most recent forecast is, on average, less accurate than the most recent
consensus available inthe month prior to the earnings announcement. The accuracy of the
consensus forecast consistently outperforms the most recent forecast in 15 out of these 17
more recent years and the differences are signi?cant for nine out of these 15 years.
The number of analysts following explains the greater accuracy of the consensus. The
aggregating value of the consensus outweighs the small timing advantage of the most
recent forecast over the short forecast horizon examined in this study.
The results from the late 1980s are consistent with those of earlier studies (O’Brien,
1988; Brown, 1991) using US data that ?nd that the most recent forecast is more accurate
than the consensus. However, the results in more recent years indicate the opposite,
consistent with improvements in timeliness of forecasts included in consensus forecasts
identi?ed in prior studies (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005).
The recent results suggest that the greater accuracy of the consensus forecast comes from
diversifying away idiosyncratic error in individual forecasts, conditional on only
relatively recent forecasts being included in the consensus.
The remainder of the paper is organised as follows. Section 2 reviews the related
literature. Section 3 develops the hypotheses. Section 4 describes the sample selection
and data. Section 5 discusses the variable de?nitions and research methods. Section 6
reports the results of tests. Section 7 presents the results of additional analysis. Section 8
concludes.
2. Prior research
2.1 The aggregation value of consensus forecasts
Analysts make andrevise their earnings forecasts throughout the year as theyincorporate
new information into their forecasts. O’Brien (1988, p. 53) suggests that, “Since a diverse
set of forecasts is available at any time for a given ?rm’s earnings, composites are used to
distil the diverse set into a single expectation”. A consensus forecast is a forecast that
aggregates all information available to analysts. It is often de?ned as the mean or median
of outstanding individual analyst forecasts at any point in time.
An aggregate forecast is expected to average out potential inef?ciencies in how
individual analysts process information and therefore provide more accurate future
earnings expectations. For example, the I/B/E/S consensus forecast in the USA is more
accurate and offers a better proxy for the market’s earnings expectations than a single
forecaster (value line) immediately before a quarterly earnings announcement (Ramnath
et al., 2005). Their study shows that most of the consensus forecasting superiority
can be attributed to the aggregation value. Since the consensus forecast aggregates
expectations from various analysts and stockbroking ?rms who are covering a
company, the idiosyncratic analyst error is diminished through the aggregation process.
Thus, the accuracy of the consensus is improved.
As suggested by Barron et al. (2008), the larger the number of analysts following
and contributing to the consensus, the more the idiosyncratic analyst error is averaged
out in determining the consensus forecast, and the higher is the accuracy of the
ARJ
23,1
96
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
consensus forecast. This suggestion motivates an examination of the association
between the number of analysts following and the accuracy of consensus forecasts in
this study.
The timeliness of individual forecasts included in the consensus forecast is
important when evaluating its accuracy. Since not all analysts update their forecasts in
a timely manner, the consensus forecast at any point in time includes both recent and
potentially stale forecasts (Kothari, 2001). The inclusion of stale forecasts is likely to
reduce the accuracy of consensus forecasts.
Many studies document that the consensus has become a more timely measure over
time due to recent improvements in the timeliness and quality of analyst forecast data
included (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005)[5]. These
studies suggest that attempts have been made to include only relatively recent
forecasts in the consensus. O’Brien (1988) ?nds that the consensus is signi?cantly
better than the most recent forecast when the consensus is relatively timely. That is,
conditional on only reasonably timely forecasts being included in the consensus, an
aggregate or consensus forecast is expected to diversify away idiosyncratic individual
error and therefore provide more accurate future earnings expectations.
2.2 Timeliness of analysts’ forecasts
O’Brien (1988) uses I/B/E/S individual analyst forecast data to compute and compare
three alternative forecast measures: the mean, the median, and the most recent
individual analyst forecast. She ?nds that the most recent forecast is more accurate than
both the mean and median forecasts in the 1975-1981 period. More recent studies
con?rmthat the accuracy of earnings forecasts improves as the earnings announcement
date approaches (Lim, 2001; Ivkovic and Jegadeesh, 2004). These studies indicate that
analysts are able to incorporate newinformation into their forecasts. This highlights the
importance of timeliness of forecasts for improving forecast accuracy.
In particular, the forecasts made over the short forecast horizon (i.e. in the period
immediately prior to the earnings announcements) will be the most informative and
accurate. As suggested by Ivkovic and Jegadeesh (2004), the analysts who update their
forecasts most recently have access to all prior forecasts made by other analysts and
will use them rationally in making their own forecasts. In addition, they may have
early access to earnings information such as management’s guidance on future
earnings. Although company management can choose to provide earnings guidance at
any point in time, any guidance they provide will be more accurate the closer it is to the
earnings announcement. If some analysts obtain early access to such information, then
their earnings forecasts will be superior to others[6]. In this study, the most recent
forecast is used to examine its accuracy in comparison with the consensus forecast.
2.3 Tradeoffs between forecast aggregation and timeliness of forecasts
The consensus forecast is characterised by aggregation value from diversifying across
idiosyncratic individual error. The most recent forecast made over a shorter forecast
horizon is more timely. Brown (1991) investigates tradeoffs between the bene?ts of
forecast aggregation and timeliness of forecasts. He adopts an approach of dropping
stale forecasts from the consensus by using three timely forecast measures (i.e. the
most recent forecast, an average of the three most recently issued forecasts and an
average of all forecasts issued within the past 30 days). He ?nds that the comparative
Accuracy
of analysts’
expectations
97
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
advantage of each forecast measure depends on company size. For large companies, the
30-day average is signi?cantly more accurate than the most recent forecast; for small
companies, the most recent forecast is more accurate than the other two alternative
forecast measures. Brown’s results suggest that the forecast aggregation outweighs the
timeliness of forecasts for large companies with more analysts following. In contrast,
the most recent forecast shows its timing advantage for small companies for which the
bene?t of aggregation of individual analyst forecasts is ineffective. However, Brown
does not directly examine the association between the number of analysts following
and the accuracy of these forecast measures.
Tradeoffs between the bene?ts of forecast aggregation and timeliness of forecasts
motivate this study to compare the accuracy of the aggregate consensus forecast and
the most recent forecast. Speci?cally, this study examines whether the number of
analyst following is more strongly associated with the superior forecast measure.
3. Hypothesis development
Since analysts do not issue forecasts at prescribed times, there is variation in the age of
forecasts included in the consensus. Forecast accuracy generally improves as the
earnings announcement date approaches because analysts incorporate newinformation
into their forecasts (O’Brien, 1988, 1990). If forecast age is the single most important
factor associated with forecast accuracy (Clement, 1999; Jacob et al., 1999), then the more
recent forecast is expected to be more accurate than older ones. Brown (1991) argues that
the consensus forecast is less accurate than more timely forecast measures, including
the most recent forecast, because the consensus includes stale forecasts. Stale forecasts
reduce forecast accuracy because recent earnings information is omitted.
On the other hand, the consensus forecast is expected to average out the individual
analyst’s idiosyncratic error through the aggregation process, which improves forecast
accuracy (Brown, 1993; Ramnath et al., 2005; Barron et al., 2008). If diversifying across
individual idiosyncrasies is more important than discarding stale forecasts, then the
consensus forecast that aggregates multiple analysts’ forecasts may be more accurate
than a single recent forecast.
Relative timeliness of the consensus forecast is also important when evaluating its
accuracy. O’Brien (1988) shows that the consensus forecast is more accurate than the
most recent forecast only when relatively recent forecasts are included in the consensus.
Recent improvements in analyst forecast data including I/B/E/S database are re?ected in
the consensus being a more timely forecast measure (Barron and Stuerke, 1998; Brown,
2001a; Ramnath et al., 2005). Aggregating to reduce idiosyncratic error in the consensus
is more effective when more timely individual forecasts are included in the consensus.
This study investigates how the number of analysts following a company and the
timeliness of an individual analyst’ forecast impacts on the differential accuracy of
the consensus andthe most recent forecast inAustralia. Like manysmaller economies, the
Australian market has relatively few brokerage ?rms and few analysts covering
companies[7]. The limited number of brokerage ?rms and analysts tend to cover
companies with high market capitalisation, leaving small companies with a thin coverage
at best. Since the market relies on the limited number of analysts providing coverage, it is
important to purge idiosyncratic error from analysts’ individual forecasts.
The disclosure environment, enhanced by continuous disclosure regulation since
1994, prohibits companies from brie?ng individual analysts with price-sensitive
ARJ
23,1
98
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
information, which mitigates the ability of individual analysts to gain incremental
information. For the short forecast horizon examined in this study, that is, during the
period immediately prior to the earnings announcement, the aggregation value of the
consensus forecast is expected to outweigh the timing advantage of the most recent
forecast in the Australian context. Therefore, the consensus forecast is expected to be
more accurate than the most recent forecast, and the ?rst hypothesis is:
H1. The consensus forecast is more accurate than the most recent individual
analyst earnings forecast.
A possible explanation for the greater accuracy of the consensus forecast is the
aggregation value from including expectations of multiple analysts. If the relatively
greater accuracy of the consensus forecast is largely explained by the aggregation
value of the consensus forecast, then the difference in forecast accuracy should be
related to the number of analysts contributing to the consensus. Speci?cally, the larger
the number of analysts following and contributing to the consensus, the greater is the
accuracy of the consensus forecast. Based on this, the second hypothesis is formed as:
H2. The greater forecast accuracy of the consensus forecast is due to the number
of analysts contributing to the consensus.
4. Sample and descriptive statistics
4.1 Sources of data
Data on one year ahead analysts’ forecasts of annual earnings per share (EPS) are
obtained from the I/B/E/S International Summary and Detail History ?les. The
Summary ?les contain the summary statistics on analyst forecasts, such as means,
medians and standard deviations. The Detail ?les provide individual analyst forecasts
and the date of each forecast issued. The summary data are calculated and reported by
I/B/E/S on the basis of all outstanding forecasts as of the third Thursday of each month
using the individual forecasts in the Detail ?les[8].
The mean and median consensus forecasts are calculated using individual analyst
forecasts to match the I/B/E/S summary consensus. The number of individual forecasts
available to calculate forecast statistics, as at the publication date of the last I/B/E/S
summary report prior to the earnings announcement, is matched against the number of
individual forecasts included in the I/B/E/S consensus[9]. This approach enables the
identi?cation of individual forecasts included in the consensus and their forecast ages
when considering the timeliness of the consensus. This consensus forecast measure is
checked against the I/B/E/S summary measure and the two measures are found to be
closely matched[10]. Since the empirical results using the reconstructed consensus
measure or the I/B/E/S summary measure are very similar, onlythe results obtained using
the reconstructed consensus measure are reported in this study.
The corresponding actual earnings are obtained from I/B/E/S for comparability
with the forecast. Earnings announcement dates are sourced from the Securities
Industry Research Centre of Asia-Paci?c (SIRCA) database[11]. The constituent list for
the Australian Stock Exchange (ASX) 100 Index is obtained from the SIRCA database.
Share prices and market capitalisation information are obtained from the CRIF Share
Price and Price Relative database. Accounting information was sourced from the
ASPECT database.
Accuracy
of analysts’
expectations
99
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
4.2 Sample selection
The initial sample includes companies[12] traded on the ASX, and with at least one
I/B/E/S consensus forecast available and two analysts following for the period from
?scal 1987 to ?scal 2007. Consistent with prior studies (O’Brien, 1988; Mikhail et al.,
1999; Ramnath et al., 2005), the most recent I/B/E/S consensus forecast prior to the
earnings announcement is retained. The initial sample comprises 5,694 company-year
observations. Notably, many companies listed on ASX are not covered by I/B/E/S.
Company-year observations are eliminated if the actual earnings in I/B/E/S or the
earnings announcement dates from SIRCA are missing. Since the consensus forecast is
compared with the most recent forecast for a particular company and year, the most
recent company-year individual analyst forecasts prior to the earnings announcement
are extracted and matched from the Detail ?les. After observations with a mismatch of
?nancial-year end between reported actual earnings and forecasted earnings are
excluded, outliers are eliminated by omitting observations with price-de?ated forecast
error greater than 10 per cent (Richardson et al., 2004; Clement and Tse, 2005). These
observations are likely to be the result of a data entry error. Table I lists the sample
selection criteria and their effects on the sample size. As shown in Table I, the ?nal
annual earnings forecast sample yields 4,358 company-year observations, representing
862 unique companies[13].
4.3 Sample descriptive statistics
Panel A of Table II reports the year-by-year sample descriptive statistics for all
company-year observations. The number of companies followed by at least two
analysts varies across years and ranges from a low of 21 companies in 1987 to a high of
357 companies in 2007. The number of companies covered generally increases towards
the later years, re?ecting the increased coverage of I/B/E/S for the Australian market.
Companies in the sample have an average (median) market capitalisation of $2 ($0.5)
billion, re?ecting the skewed distribution of companies covered by analysts. That is,
analysts follow a limited number of very large companies. They also selectively cover
small or medium size companies.
Number of company-year
observations remaining in sample
Percentage of total
consensus forecasts
Consensus forecasts from 1987 to 2007
for Australian companies with at least
two analysts following 5,694 100
Actual earnings (from I/B/E/S) and
earnings announcement date (from
SIRCA) available 4,650 82
Excluding companies with change in
?nancial year-end 4,636 81
Excluding outliers ¼ ?nal sample (862
unique companies) 4,358 77
Notes: Consensus forecasts are the means or medians of all the individual analyst forecasts available
as at the publication date of the last I/B/E/S consensus before earnings announcement; individual
analyst forecasts are extracted from the I/B/E/S Detail History ?les; outliers are de?ned as the
observations where AFEs are greater than 10 per cent
Table I.
Sample selection
ARJ
23,1
100
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
M
e
a
n
o
r
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
N
u
m
b
e
r
o
f
?
r
m
s
M
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
(
$
0
0
0
)
N
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
F
i
s
c
a
l
y
e
a
r
f
o
l
l
o
w
e
d
M
e
a
n
M
e
d
i
a
n
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
P
a
n
e
l
A
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
a
l
l
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
t
h
e
s
a
m
p
l
e
1
9
8
7
2
1
3
1
4
,
0
9
9
9
5
,
0
3
4
4
2
3
5
1
2
1
8
9
1
2
4
1
4
3
5
7
2
5
4
1
7
9
1
9
8
8
6
7
5
2
5
,
3
6
5
2
0
3
,
4
6
7
6
2
4
1
0
1
4
1
9
7
1
4
5
1
7
8
5
6
2
2
3
7
7
1
1
9
8
9
1
3
5
9
0
2
,
8
7
6
3
3
0
,
7
1
4
8
4
7
1
1
1
0
0
7
5
9
7
1
2
3
3
0
1
7
2
6
3
0
1
9
9
0
1
0
9
1
,
2
5
8
,
6
0
4
5
1
7
,
2
9
4
8
5
8
1
1
1
2
8
9
6
1
3
0
1
5
5
3
7
2
0
3
3
5
0
1
9
9
1
1
1
9
1
,
0
7
1
,
7
9
8
3
3
2
,
9
5
2
7
3
7
1
0
8
9
5
7
8
8
1
1
7
2
4
1
4
2
0
2
8
1
9
9
2
1
3
6
1
,
2
1
1
,
6
4
6
3
7
1
,
5
7
8
6
3
5
8
9
5
7
5
9
4
1
1
7
3
2
1
8
2
8
3
6
1
9
9
3
1
5
0
1
,
1
7
3
,
0
4
0
4
1
9
,
0
6
4
8
4
8
1
2
1
0
3
7
3
1
0
2
1
3
0
2
5
6
1
4
2
8
1
9
9
4
1
6
7
1
,
4
7
8
,
2
1
0
4
8
3
,
7
2
4
7
3
6
1
0
1
0
2
7
8
9
9
1
2
9
2
3
6
1
2
3
4
1
9
9
5
2
2
7
1
,
0
9
4
,
6
7
4
3
2
5
,
2
2
9
7
3
6
1
1
8
2
5
7
7
6
1
0
0
3
1
7
2
1
4
1
1
9
9
6
2
2
3
1
,
2
3
9
,
1
2
0
3
6
8
,
6
4
6
7
3
6
1
0
8
6
5
5
8
2
1
1
0
2
1
5
1
1
2
3
1
9
9
7
2
4
8
1
,
2
7
7
,
2
0
0
3
7
1
,
6
3
2
7
4
7
1
1
9
1
6
4
9
2
1
1
6
2
7
6
1
4
3
6
1
9
9
8
2
7
2
1
,
5
9
7
,
2
7
6
3
8
5
,
8
7
0
8
4
7
1
2
8
6
6
0
8
5
1
0
3
2
6
5
1
6
3
7
1
9
9
9
2
7
9
1
,
6
3
0
,
7
8
0
3
5
5
,
0
0
0
8
4
7
1
1
9
8
6
5
9
6
1
2
3
3
3
7
2
1
3
9
2
0
0
0
2
8
6
1
,
8
2
0
,
8
6
6
4
6
0
,
0
9
1
7
4
6
1
0
1
0
5
7
6
1
0
0
1
2
8
3
5
8
2
1
4
9
2
0
0
1
2
7
3
2
,
2
2
6
,
8
5
8
4
1
0
,
7
9
3
8
4
8
1
1
1
1
3
7
8
1
0
5
1
4
1
3
9
7
2
0
5
7
2
0
0
2
2
3
4
2
,
5
1
6
,
3
5
9
5
1
8
,
1
2
5
7
4
7
1
0
1
1
2
7
8
1
1
0
1
3
8
3
6
8
2
1
4
8
2
0
0
3
2
3
7
2
,
3
8
1
,
1
7
3
5
9
4
,
2
4
3
6
3
6
9
8
4
5
8
7
6
1
0
4
2
8
6
1
6
3
8
2
0
0
4
2
5
5
2
,
5
0
0
,
3
8
6
6
1
1
,
1
0
2
6
3
6
9
1
0
4
7
0
9
9
1
3
1
3
9
8
2
2
4
6
2
0
0
5
2
7
4
2
,
9
3
0
,
0
0
5
7
4
0
,
1
6
1
5
3
5
7
1
1
1
5
8
9
8
1
4
8
4
0
8
2
3
5
4
2
0
0
6
2
8
9
3
,
5
3
8
,
0
8
6
9
2
4
,
2
5
2
6
3
5
8
1
1
1
6
0
1
0
5
1
4
6
4
6
1
2
2
8
5
2
2
0
0
7
3
5
7
3
,
3
3
8
,
7
9
0
6
9
8
,
5
8
9
6
3
5
9
1
1
4
7
4
1
1
0
1
5
1
4
7
1
4
2
9
6
5
A
l
l
y
e
a
r
s
4
,
3
5
8
1
,
9
9
2
,
6
0
3
4
6
5
,
2
5
2
7
3
6
1
0
1
0
2
6
7
9
6
1
2
9
3
4
8
2
1
4
3
P
a
n
e
l
B
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
A
S
X
1
0
0
I
n
d
e
x
1
9
9
7
9
4
2
,
9
5
3
,
6
3
6
1
,
4
2
3
,
6
3
4
1
1
1
0
1
2
1
3
8
5
5
5
8
3
1
1
2
1
3
5
8
1
5
1
9
9
8
9
8
3
,
9
5
6
,
0
0
9
1
,
9
8
9
,
3
7
7
1
2
1
0
1
3
1
6
7
6
5
5
8
1
9
6
1
2
2
8
1
7
1
9
9
9
9
9
4
,
1
5
3
,
4
0
6
1
,
6
6
9
,
0
4
0
1
1
9
1
2
1
4
8
7
6
1
8
5
1
1
0
1
8
3
1
0
2
6
2
0
0
0
1
0
0
4
,
5
9
0
,
1
5
7
2
,
0
5
1
,
8
6
7
1
1
8
1
1
1
3
9
1
6
8
8
9
1
1
1
1
4
4
9
2
0
2
0
0
1
8
9
6
,
1
6
1
,
5
3
7
2
,
0
2
9
,
3
4
7
1
2
1
0
1
2
1
3
9
2
6
6
9
3
1
1
3
1
3
4
8
1
8
2
0
0
2
9
2
5
,
8
3
1
,
7
7
1
2
,
4
3
9
,
5
5
1
1
1
9
1
1
1
3
9
7
6
5
9
6
1
1
9
1
5
2
1
0
2
0
(
c
o
n
t
i
n
u
e
d
)
Table II.
Descriptive statistics
Accuracy
of analysts’
expectations
101
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
M
e
a
n
o
r
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
N
u
m
b
e
r
o
f
?
r
m
s
M
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
(
$
0
0
0
)
N
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
F
i
s
c
a
l
y
e
a
r
f
o
l
l
o
w
e
d
M
e
a
n
M
e
d
i
a
n
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
2
0
0
3
9
2
5
,
5
1
4
,
5
8
5
2
,
4
7
3
,
0
9
7
9
8
9
1
0
7
6
5
5
7
1
8
9
1
4
3
9
2
3
2
0
0
4
9
4
5
,
9
3
7
,
7
3
2
2
,
7
7
0
,
7
9
4
9
8
9
1
1
9
5
7
0
9
4
1
1
4
1
5
3
8
2
1
2
0
0
5
9
2
7
,
5
4
7
,
4
8
4
4
,
1
3
5
,
1
8
5
7
6
7
9
9
8
4
8
8
0
1
3
1
2
1
5
1
3
3
2
2
0
0
6
9
8
8
,
8
7
7
,
1
6
7
4
,
5
5
9
,
9
0
8
8
7
9
1
0
1
0
2
5
8
8
9
1
3
0
2
6
6
1
5
3
0
2
0
0
7
9
8
1
0
,
4
4
4
,
4
1
4
4
,
9
6
4
,
3
2
3
1
0
7
1
0
1
2
1
1
1
6
7
1
0
7
1
4
7
2
6
7
1
7
3
6
A
l
l
y
e
a
r
s
1
,
0
4
6
5
,
9
9
7
,
9
0
2
2
,
6
7
3
,
7
3
0
1
0
8
1
0
1
3
9
2
6
0
8
6
1
1
6
1
7
4
1
0
2
3
P
a
n
e
l
C
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
o
u
t
s
i
d
e
t
h
e
A
S
X
1
0
0
I
n
d
e
x
1
9
9
7
1
5
4
2
5
3
,
9
2
1
1
9
9
,
3
3
4
5
3
4
7
9
5
6
7
9
5
1
1
9
3
5
8
2
2
5
4
1
9
9
8
1
7
4
2
6
8
,
7
9
3
2
2
6
,
7
7
7
6
3
5
8
9
1
6
2
8
9
1
0
9
3
4
1
0
2
4
4
9
1
9
9
9
1
8
0
2
4
3
,
3
3
6
1
7
0
,
6
9
7
6
3
5
8
1
0
5
7
3
1
0
3
1
3
3
4
1
1
3
2
8
4
7
2
0
0
0
1
8
6
3
3
2
,
0
0
0
2
3
5
,
9
7
9
5
3
4
7
1
1
3
8
3
1
0
8
1
4
0
4
7
1
3
3
3
6
4
2
0
0
1
1
8
4
3
2
3
,
6
7
1
2
1
1
,
1
7
9
6
3
5
8
1
2
2
8
6
1
1
8
1
5
4
5
2
1
4
3
3
7
6
2
0
0
2
1
4
2
3
6
8
,
3
4
6
2
2
3
,
8
9
3
5
3
4
7
1
2
2
8
9
1
1
9
1
4
2
4
9
1
6
3
3
6
5
2
0
0
3
1
4
5
3
9
3
,
0
7
6
2
3
5
,
3
5
2
4
2
4
6
8
9
6
0
8
2
1
1
3
3
7
9
2
2
5
1
2
0
0
4
1
6
1
4
9
3
,
4
8
8
3
1
9
,
9
8
3
4
2
4
6
1
0
9
7
1
1
0
2
1
4
9
5
3
1
6
3
4
6
6
2
0
0
5
1
8
2
5
9
5
,
8
9
5
3
6
1
,
8
9
1
4
2
3
6
1
1
8
6
8
1
0
6
1
5
2
4
9
1
0
3
2
6
6
2
0
0
6
1
9
1
7
9
8
,
6
6
3
4
3
7
,
3
8
8
5
2
4
6
1
1
6
6
5
1
1
3
1
5
5
5
7
1
9
3
6
7
2
2
0
0
7
2
5
9
6
5
0
,
1
7
6
3
9
3
,
4
0
3
5
2
4
7
1
1
6
7
6
1
1
2
1
5
3
5
4
1
7
3
6
7
7
A
l
l
y
e
a
r
s
1
,
9
5
8
4
4
3
,
8
8
5
2
7
0
,
0
9
9
5
3
4
7
1
0
9
7
2
1
0
3
1
3
9
4
7
1
3
3
0
6
2
N
o
t
e
s
:
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
d
i
s
t
r
i
b
u
t
i
o
n
o
f
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
s
a
m
p
l
e
;
P
a
n
e
l
A
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
a
l
l
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
;
P
a
n
e
l
B
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
A
S
X
1
0
0
I
n
d
e
x
;
P
a
n
e
l
C
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
o
u
t
s
i
d
e
t
h
e
A
S
X
1
0
0
I
n
d
e
x
;
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
d
a
t
a
i
s
o
b
t
a
i
n
e
d
f
r
o
m
I
/
B
/
E
/
S
;
a
v
e
r
a
g
e
m
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
i
s
t
h
e
e
q
u
i
t
y
v
a
l
u
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
c
o
m
p
a
n
y
’
s
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
;
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
e
q
u
a
l
s
e
i
t
h
e
r
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
c
o
n
t
r
i
b
u
t
i
n
g
t
o
t
h
e
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
o
r
o
n
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
Table II.
ARJ
23,1
102
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Panel A reports analyst coverage statistics. The average number of analyst forecasts
included in the consensus is approximately seven. It is about one third of that reported
by Ke and Yu (2006) for US analyst coverage statistics, re?ecting relatively fewer
brokerage ?rms and analysts provide company coverage in the Australian market.
Panel A also presents descriptive statistics for the distribution of timeliness of the
consensus and the most recent forecast. Despite the last I/B/E/S consensus being
published in the month prior to the earnings announcement, the median age of the
consensus forecast is 96 calendar days before the earnings announcement. The median
age of the most recent forecast is 21 calendar days. The most recent forecast is
approximately 75 calendar days more recent than the median age of the consensus
forecast, suggesting that the most recent forecast should be more accurate if new
information has been incorporated into the forecast.
Consensus forecasts in the early years of the sample period, notably from 1987 to
1990, are on average 20 days staler than in more recent years. The inclusion of stale
forecasts in the consensus is likely to reduce its accuracy. In more recent years, the
relative timeliness of consensus forecasts is improving, consistent with previous
?ndings (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005).
Panels B and C present descriptive statistics for the company-year observations
included in the ASX 100 Index and outside the ASX 100 Index, respectively. Partitioning
the sample into the ASX 100 companies and companies outside the Index reduces the
sample size to 3,004 observations because the constituent list for the ASX 100 Index is
unavailable in SIRCA prior to ?scal 1997. The ASX 100 companies have higher market
capitalisation, are covered by more analysts, and their forecasts are updated more timely,
as compared with companies outside the Index. The ASX100 companies have an average
(median) market capitalisation of $6 ($2.7) billion and are followed by ten analysts on
average. The median age of the consensus (most recent) forecast for these companies is 86
(ten) days. By comparison, companies outside the ASX 100 Index have an average
(median) market capitalisation of $0.4 ($0.3) billion and are covered by an average of ?ve
analysts. The median age of the consensus (most recent) forecast for these companies is
103 (30) days. The ASX 100 companies and companies outside the Index show different
company and forecast characteristics. These differences may have effects on forecast
accuracy. Further analysis is conducted in Section 6.2.
5. Evaluating forecast accuracy
5.1 Variable de?nitions
The absolute forecast error (AFE) is used to measure forecast accuracy:
AFE
jts
¼
A
jt
2F
jts
P
j;t21
ð1Þ
Following Richardson et al. (2004), the AFE
jts
is de?ned as the absolute value of the
difference between A
jt
, actual annual EPS of company j in year t, and F
jts
, the forecast
EPS using each of the alternative forecast measures, s, and is de?ated by company j’s
share price[14] 11 months before the earnings announcement month[15], P
j,t21
.
Each of the alternative forecast measures, denoted by s, is one of the following: the
mean consensus forecast (s ¼ mean), the median consensus forecast (s ¼ median) or
the most recent forecast (s ¼ mr). F
jtmean
, the mean consensus forecast, is the mean of
Accuracy
of analysts’
expectations
103
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
all the individual analyst forecasts available as at the publication date of the last
I/B/E/S consensus before the earnings announcement for company j in year t. F
jtmedian
,
the median consensus forecast, is the median of all the individual analyst forecasts
available as at the publication date of the last I/B/E/S consensus before the earnings
announcement for company j in year t. F
jtmr
, the most recent forecast, is the latest
individual analyst forecast EPS reported to I/B/E/S before the earnings announcement
for company j in year t[16].
To compare the accuracy of the consensus and the most recent forecast for each
year of the sample period, AFE
jts
is computed to measure forecast accuracy at the
company level for each year and then aggregate these results across companies:
MAFE
ts
¼
1
N
X
N
j¼1
AFE
jts
ð2Þ
For each forecast measure, s, the mean of the AFE
jts
in year t, MAFE
ts
, is averaged
across all available company observations ( j ¼ 1, . . . , N) in year t:
MAFE
s
¼
1
N
X
N
n¼1
1
T
X
T
t¼1
AFE
jts
( )
ð3Þ
For each forecast measure, s, the pooled mean of the AFE
jts
, MAFE
s
, is averaged across
years (t ¼ 1, . . . , T) for each company j, and then averaged across companies to
evaluate forecast accuracy at an aggregate level across years and companies.
MAFE
ts
and MAFE
s
, are calculated to compare the accuracy of each forecast
measure for each year and overall for the sample period. Signi?cance tests of the
differences in accuracy are used to test whether the consensus is more accurate than
the most recent forecast (H1).
The timeliness of the consensus forecast is measured by taking the average value of
timeliness of individual analyst forecasts included in the last consensus prior to the
earnings announcement, where the timeliness of an individual analyst forecast is
measured with references to the number of calendar days between the date of the
individual analyst forecast issued prior to the earnings announcement and the earnings
announcement date. The timeliness of the most recent forecast is measured with
references to the number of calendar days between the date of the most recent forecast
issued prior to the earnings announcement and the earnings announcement date.
The number of analysts following is either the number of analysts contributing to the
consensus forecast or one for the most recent forecast.
5.2 Forecast accuracy and pairwise differences in forecast accuracy
Table III reports the accuracy of the mean consensus, the median consensus and the
most recent forecast. The forecast accuracy is measured by the mean of the AFEs
across all available company-year observations for the year. For the 1987-1990 period,
the most recent forecast is more accurate than the consensus forecast. By comparison,
O’Brien (1988)’s 1975-1981 sample and Brown (1991)’s 1984-1988 sample show similar
effects. In the more recent period from 1991 to 2007, the results suggest that both the
mean and median consensus forecasts are more accurate than the most recent forecast.
The accuracy of the median (mean) consensus forecast outperforms that of the most
ARJ
23,1
104
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
A
v
e
r
a
g
e
A
F
E
(
%
)
t
-
s
t
a
t
i
s
t
i
c
f
o
r
p
a
i
r
w
i
s
e
t
e
s
t
o
f
d
i
f
f
e
r
e
n
c
e
s
F
i
s
c
a
l
y
e
a
r
N
u
m
b
e
r
o
f
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
M
e
a
n
c
o
n
s
e
n
s
u
s
M
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
M
e
a
n
–
m
e
d
i
a
n
M
e
a
n
–
m
o
s
t
r
e
c
e
n
t
M
e
d
i
a
n
–
M
o
s
t
r
e
c
e
n
t
1
9
8
7
2
1
1
.
7
1
2
1
.
6
8
4
1
.
3
9
8
0
.
2
5
0
.
5
7
0
.
5
3
1
9
8
8
6
7
1
.
4
2
5
1
.
2
4
6
0
.
8
1
6
2
.
2
2
*
*
2
.
5
8
*
*
1
.
7
9
*
1
9
8
9
1
3
5
1
.
0
9
2
0
.
9
5
0
0
.
8
9
1
2
.
8
7
*
*
*
1
.
8
2
*
0
.
5
3
1
9
9
0
1
0
9
1
.
1
5
6
0
.
9
5
8
0
.
6
9
5
2
.
3
9
*
*
2
.
7
4
*
*
*
1
.
6
6
*
1
9
9
1
1
1
9
0
.
9
3
7
0
.
5
9
4
0
.
8
2
2
3
.
8
4
*
*
*
0
.
8
8
2
2
.
0
2
*
*
1
9
9
2
1
3
6
0
.
9
8
4
0
.
9
7
6
1
.
1
7
9
0
.
1
9
2
1
.
8
8
*
2
1
.
9
8
*
*
1
9
9
3
1
5
0
0
.
8
6
6
0
.
8
2
4
0
.
9
5
3
1
.
6
2
2
0
.
9
7
2
1
.
4
7
1
9
9
4
1
6
7
0
.
7
1
3
0
.
6
8
7
0
.
7
6
3
1
.
6
8
*
2
0
.
7
4
2
1
.
2
1
1
9
9
5
2
2
7
0
.
8
2
3
0
.
8
0
3
0
.
8
8
2
1
.
4
2
2
1
.
0
5
2
1
.
3
7
1
9
9
6
2
2
3
0
.
7
6
8
0
.
7
3
0
0
.
7
0
9
2
.
0
2
*
*
1
.
0
7
0
.
3
8
1
9
9
7
2
4
8
0
.
6
5
5
0
.
6
7
4
0
.
7
7
6
2
0
.
8
0
2
3
.
0
6
*
*
*
2
3
.
0
1
*
*
*
1
9
9
8
2
7
2
0
.
7
9
8
0
.
7
4
1
0
.
7
9
3
2
.
2
6
*
*
0
.
0
9
2
0
.
8
4
1
9
9
9
2
7
9
0
.
9
8
8
0
.
9
8
0
0
.
9
4
9
0
.
3
4
0
.
7
0
0
.
5
8
2
0
0
0
2
8
6
1
.
1
7
7
1
.
1
8
1
1
.
2
0
7
2
0
.
2
4
2
0
.
5
5
2
0
.
4
7
2
0
0
1
2
7
3
1
.
3
3
5
1
.
3
3
0
1
.
4
2
2
0
.
2
9
2
1
.
4
0
2
1
.
5
7
2
0
0
2
2
3
4
1
.
0
4
0
1
.
0
3
2
1
.
2
1
0
0
.
3
2
2
2
.
6
7
*
*
*
2
2
.
6
6
*
*
*
2
0
0
3
2
3
7
0
.
9
8
7
1
.
0
2
0
1
.
1
5
5
2
1
.
7
6
*
2
2
.
4
8
*
*
2
2
.
0
0
*
*
2
0
0
4
2
5
5
0
.
8
9
3
0
.
8
2
4
0
.
9
8
0
2
.
0
0
*
*
2
1
.
4
1
2
2
.
7
2
*
*
*
2
0
0
5
2
7
4
0
.
8
5
1
0
.
8
2
0
0
.
9
9
1
1
.
5
1
2
2
.
2
5
*
*
2
2
.
6
6
*
*
*
2
0
0
6
2
8
9
0
.
7
4
2
0
.
7
0
6
0
.
8
1
9
1
.
4
8
2
1
.
5
0
2
2
.
2
1
*
*
2
0
0
7
3
5
7
0
.
7
6
6
0
.
7
1
1
1
.
0
0
9
2
.
6
5
*
*
*
2
4
.
0
8
*
*
*
2
4
.
7
2
*
*
*
A
l
l
y
e
a
r
s
4
,
3
5
8
0
.
9
2
7
0
.
8
8
5
0
.
9
7
5
6
.
3
9
*
*
*
2
2
.
9
9
*
*
*
2
5
.
6
8
*
*
*
N
o
t
e
s
:
S
t
a
t
i
s
t
i
c
a
l
s
i
g
n
i
?
c
a
n
c
e
a
t
t
h
e
*
0
.
1
0
,
*
*
0
.
0
5
a
n
d
*
*
*
0
.
0
1
l
e
v
e
l
s
(
t
w
o
-
t
a
i
l
t
e
s
t
)
,
r
e
s
p
e
c
t
i
v
e
l
y
;
f
o
r
e
c
a
s
t
a
c
c
u
r
a
c
y
i
s
m
e
a
s
u
r
e
d
b
y
t
h
e
A
F
E
;
A
F
E
i
s
d
e
?
n
e
d
a
s
t
h
e
a
b
s
o
l
u
t
e
v
a
l
u
e
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
a
c
t
u
a
l
E
P
S
a
n
d
t
h
e
f
o
r
e
c
a
s
t
E
P
S
d
e
?
a
t
e
d
b
y
s
h
a
r
e
p
r
i
c
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
;
a
v
e
r
a
g
e
A
F
E
i
s
t
h
e
m
e
a
n
o
f
t
h
e
A
F
E
s
a
c
r
o
s
s
a
l
l
a
v
a
i
l
a
b
l
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
f
o
r
t
h
e
y
e
a
r
;
t
h
e
m
e
a
n
c
o
n
s
e
n
s
u
s
i
s
t
h
e
m
e
a
n
o
f
a
l
l
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
a
v
a
i
l
a
b
l
e
a
s
a
t
t
h
e
p
u
b
l
i
c
a
t
i
o
n
d
a
t
e
o
f
t
h
e
l
a
s
t
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
;
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
i
s
t
h
e
m
e
d
i
a
n
o
f
a
l
l
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
a
v
a
i
l
a
b
l
e
a
s
a
t
t
h
e
p
u
b
l
i
c
a
t
i
o
n
d
a
t
e
o
f
t
h
e
l
a
s
t
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
;
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
i
s
t
h
e
l
a
t
e
s
t
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
E
P
S
r
e
p
o
r
t
e
d
t
o
I
/
B
/
E
/
S
p
r
i
o
r
t
o
t
h
e
a
n
n
o
u
n
c
e
m
e
n
t
o
f
a
c
t
u
a
l
e
a
r
n
i
n
g
s
;
t
h
e
t
-
s
t
a
t
i
s
t
i
c
i
s
a
t
e
s
t
o
f
t
h
e
p
a
i
r
w
i
s
e
d
i
f
f
e
r
e
n
c
e
i
n
a
c
c
u
r
a
c
y
b
e
t
w
e
e
n
t
h
e
c
o
n
s
e
n
s
u
s
a
n
d
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
s
;
a
n
e
g
a
t
i
v
e
s
i
g
n
o
n
a
t
-
s
t
a
t
i
s
t
i
c
i
n
t
h
i
s
t
a
b
l
e
i
n
d
i
c
a
t
e
s
t
h
a
t
t
h
e
?
r
s
t
o
f
t
h
e
p
a
i
r
o
f
f
o
r
e
c
a
s
t
m
e
a
s
u
r
e
s
c
o
m
p
a
r
e
d
i
s
m
o
r
e
a
c
c
u
r
a
t
e
;
t
h
e
c
o
n
v
e
r
s
e
i
s
t
r
u
e
f
o
r
a
p
o
s
i
t
i
v
e
s
i
g
n
Table III.
Accuracy of the
consensus and the most
recent earnings forecasts
by year
Accuracy
of analysts’
expectations
105
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
recent forecast in 15 (13) out of these 17 most recent years. The accuracy of the median
consensus is greater than the mean consensus in 18 out of total 21 years.
Table III also reports results of statistical tests for differences in accuracy among
the mean consensus, the median consensus and the most recent forecast. A negative
sign on a t-statistic indicates that the ?rst of the pair of forecast measures compared is
more accurate. For example, in the ?scal year 2007, the t-statistic for the pairwise test
of differences in accuracy between the median consensus and the most recent forecast
is 24.72, which favours the median consensus, and is statistically signi?cant at the
0.01 level. The results con?rm that whilst in the late 1980s the most recent forecast is
signi?cantly more accurate than the consensus, the consensus forecast outperforms the
most recent forecast in more recent years. The results show that the median (mean)
consensus forecast dominates the most recent forecast where signi?cant differences
exist for nine (six) out of these 15 years[17].
In terms of economic signi?cance, Table III shows that the most recent forecast is on
average less accurate than the consensus by 0.09 per cent of share price[18]. For a
company with a market capitalisation of $2 billion, the average market capitalisation in
the sample, this translates into a most recent forecast that misses actual earnings by
$1.8 million relative to the consensus forecast[19].
6. Explanations for the greater accuracy of consensus forecasts
Following Ramnath et al. (2005), to examine whether the number of analysts following
and the timeliness of analysts’ forecasts explain the relative greater forecast accuracy
of consensus forecasts using a cross-sectional regression (H2), the following
cross-sectional regression is estimated:
AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ 1
jts
ð4Þ
where:
AFE
jts
¼ Is the absolute value of the median consensus[20] (s ¼ median)
or the most recent (s ¼ mr) forecast error de?ated by share
price 11 months before the earnings announcement month for
company j’s annual EPS in year t.
MEASURE
jts
¼ Is an indicator variable, coded one if AFE is sourced from the
median consensus forecast (s ¼ median); coded zero if AFE is
sourced from the most recent forecast (s ¼ mr) for company j
in year t.
ANALYST
jts
¼ Equals the number of analysts contributing to the consensus
forecast (s ¼ median) or one for the most recent forecast
(s ¼ mr) for company j in year t.
TIMELINESS
jts
¼ Equals the average value of timeliness of individual analyst
forecasts included in the last consensus prior to the earnings
announcement, where the timeliness of an individual analyst
forecast is measured by the number of calendar days between
the date of the individual analyst forecast issued prior to the
ARJ
23,1
106
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
earnings announcement and the earnings announcement date
for the median consensus forecast (s ¼ median) or the number
of calendar days between the date of the most recent forecast
issued prior to the earnings announcement and the earnings
announcement date for the most recent forecast (s ¼ mr) for
company j in year t.
Cross-sectional regression is estimated for each year t, respectively, and the mean of
the annual coef?cient estimates across the sample period is calculated.
Prior research demonstrates that the larger the number of analysts following, the
greater is the accuracy of consensus forecasts. The AFE is expected to decrease as
the number of analyst following increases, b
2
, 0. Increasing forecast accuracy is also
associated with the timeliness of analysts’ forecasts. The accuracy of analysts’ forecasts
improves when the earnings announcement date approaches. Hence, the AFE is
expected to decrease as the timeliness of analysts’ forecasts is shorter, b
3
. 0, for the
regression model (4).
Consistent with Ramnath et al. (2005) and Ke and Yu (2006), company-speci?c and
macroeconomic control variables are not included in the model. This may seem
unusual, however, the model examines determinants of the relative accuracy between
the consensus and the most recent forecast given the underlying economic conditions.
Both forecast measures are exposed to the same company-speci?c factors and
macroeconomic effects.
Since the aggregation value of consensus forecasts is expected to outweigh the timing
advantage of the most recent forecast, the forecasting superiority of the consensus
forecast over the most recent forecast is expected to be reduced after controlling for the
forecast aggregation and timing. In other words, the most recent forecast would
outperform the consensus forecast after controlling for these factors. b
1
, the coef?cient
on the MEASURE indicator variable, measuring the difference in accuracy of the
consensus forecast versus the most recent forecast, is expected to be positive (i.e. the
consensus forecast generates larger AFEs than the most recent forecast), b
1
. 0. b
0
,
the intercept, is expected to be positive, b
0
. 0, since the AFEis greater than or equal to
zero. The cross-sectional regression is estimated for each year t, respectively. Coef?cient
estimates are presented as the mean across the sample period following the Fama and
MacBeth (1973) procedure. The t-statistics and signi?cance levels are obtained under the
null that the mean of the coef?cient distributions across the sample period equals zero.
To control for size effects, and therefore, information environment effects on
forecast accuracy, the sample is partitioned into observations included in the ASX 100
Index and outside the Index to examine whether the same factors are associated with
the greater accuracy of consensus forecasts for these two groups.
Table IV reports the results of the regression model that examines whether the
number of analysts following and the timeliness of analysts’ forecasts explain the
greater accuracy of consensus forecasts for the overall sample period, the pre-1991 and
post-1991 periods, and observations included in the ASX 100 Index and outside the
Index. For the overall sample period, consistent with H2, the coef?cient on number of
analysts following is signi?cantly negative (20.0788, t ¼ 26.28), indicating forecast
accuracy increases with the number of analysts following. The coef?cient on timeliness
of analysts’ forecasts is close to zero (0.0004) and is not statistically signi?cant,
indicating that the timeliness of analysts’ forecasts does not contribute to increasing
Accuracy
of analysts’
expectations
107
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
:
A
F
E
V
a
r
i
a
b
l
e
P
r
e
d
i
c
t
e
d
s
i
g
n
O
v
e
r
a
l
l
p
e
r
i
o
d
P
r
e
-
1
9
9
1
p
e
r
i
o
d
P
o
s
t
-
1
9
9
1
p
e
r
i
o
d
A
S
X
1
0
0
O
u
t
s
i
d
e
A
S
X
1
0
0
I
n
t
e
r
c
e
p
t
þ
1
.
0
2
9
*
*
*
(
2
0
.
6
5
)
0
.
8
8
0
7
*
*
*
(
5
.
2
9
)
1
.
0
5
8
8
*
*
*
(
2
2
.
5
5
)
0
.
7
8
6
1
*
*
*
(
1
2
.
2
0
)
1
.
3
6
3
6
*
*
*
(
2
1
.
4
2
)
M
E
A
S
U
R
E
þ
0
.
3
2
8
4
*
*
*
(
6
.
3
4
)
0
.
3
6
1
9
(
2
.
2
4
)
0
.
3
2
0
5
*
*
*
(
5
.
8
9
)
0
.
6
6
6
9
*
*
*
(
4
.
2
9
)
0
.
3
5
2
1
*
*
*
(
6
.
8
5
)
A
N
A
L
Y
S
T
2
2
0
.
0
7
8
8
*
*
*
(
2
6
.
2
8
)
2
0
.
0
8
8
9
(
2
1
.
5
0
)
2
0
.
0
7
6
5
*
*
*
(
2
8
.
1
9
)
2
0
.
0
5
8
6
*
*
*
(
2
3
.
8
0
)
2
0
.
1
1
5
4
*
*
*
(
2
8
.
9
4
)
T
I
M
E
L
I
N
E
S
S
þ
0
.
0
0
0
4
(
0
.
7
6
)
0
.
0
0
2
8
(
1
.
2
9
)
2
0
.
0
0
0
1
(
2
0
.
2
5
)
2
0
.
0
0
3
*
*
(
2
2
.
8
4
)
2
0
.
0
0
0
8
(
2
1
.
0
8
)
N
u
m
b
e
r
o
f
o
b
s
e
r
v
a
t
i
o
n
s
4
,
3
5
8
3
3
2
4
,
0
2
6
1
,
0
4
6
1
,
9
5
8
A
d
j
u
s
t
e
d
R
2
(
%
)
1
.
9
8
1
.
0
1
2
.
2
2
.
1
8
1
.
9
0
N
o
t
e
s
:
S
t
a
t
i
s
t
i
c
a
l
s
i
g
n
i
?
c
a
n
c
e
a
t
t
h
e
*
0
.
1
0
,
*
*
0
.
0
5
a
n
d
*
*
*
0
.
0
1
l
e
v
e
l
s
(
t
w
o
-
t
a
i
l
t
e
s
t
)
,
r
e
s
p
e
c
t
i
v
e
l
y
;
A
F
E
j
t
s
¼
b
0
þ
b
1
ð
M
E
A
S
U
R
E
j
t
s
Þ
þ
b
2
ð
A
N
A
L
Y
S
T
j
t
s
Þ
þ
b
3
ð
T
I
M
E
L
I
N
E
S
S
j
t
s
Þ
þ
1
j
t
s
;
A
F
E
j
t
s
i
s
t
h
e
a
b
s
o
l
u
t
e
v
a
l
u
e
o
f
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
(
s
¼
m
e
d
i
a
n
)
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
(
s
¼
m
r
)
f
o
r
e
c
a
s
t
e
r
r
o
r
d
e
?
a
t
e
d
b
y
s
h
a
r
e
p
r
i
c
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
f
o
r
c
o
m
p
a
n
y
j
’
s
a
n
n
u
a
l
E
P
S
i
n
y
e
a
r
t
;
t
h
i
s
v
a
r
i
a
b
l
e
i
s
m
u
l
t
i
p
l
i
e
d
b
y
1
0
0
,
s
o
t
h
e
c
o
e
f
?
c
i
e
n
t
e
s
t
i
m
a
t
e
s
a
r
e
a
s
a
p
e
r
c
e
n
t
a
g
e
o
f
s
h
a
r
e
p
r
i
c
e
;
M
E
A
S
U
R
E
j
t
s
i
s
a
n
i
n
d
i
c
a
t
o
r
v
a
r
i
a
b
l
e
,
c
o
d
e
d
o
n
e
i
f
A
F
E
i
s
s
o
u
r
c
e
d
f
r
o
m
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
;
c
o
d
e
d
z
e
r
o
i
f
A
F
E
i
s
s
o
u
r
c
e
d
f
r
o
m
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
A
N
A
L
Y
S
T
j
t
s
e
q
u
a
l
s
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
c
o
n
t
r
i
b
u
t
i
n
g
t
o
t
h
e
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
o
r
o
n
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
s
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
T
I
M
E
L
I
N
E
S
S
j
t
s
e
q
u
a
l
s
t
h
e
a
v
e
r
a
g
e
v
a
l
u
e
o
f
t
i
m
e
l
i
n
e
s
s
o
f
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
i
n
c
l
u
d
e
d
i
n
t
h
e
l
a
s
t
c
o
n
s
e
n
s
u
s
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
,
w
h
e
r
e
t
h
e
t
i
m
e
l
i
n
e
s
s
o
f
a
n
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
i
s
m
e
a
s
u
r
e
d
b
y
t
h
e
n
u
m
b
e
r
o
f
c
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
t
h
e
d
a
t
e
o
f
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
i
s
s
u
e
d
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
a
n
d
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
f
o
r
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
o
r
t
h
e
n
u
m
b
e
r
o
f
c
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
t
h
e
d
a
t
e
o
f
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
i
s
s
u
e
d
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
a
n
d
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
t
h
e
r
e
g
r
e
s
s
i
o
n
i
s
e
s
t
i
m
a
t
e
d
f
o
r
e
a
c
h
y
e
a
r
t
,
r
e
s
p
e
c
t
i
v
e
l
y
;
c
o
e
f
?
c
i
e
n
t
e
s
t
i
m
a
t
e
s
a
r
e
p
r
e
s
e
n
t
e
d
a
s
t
h
e
m
e
a
n
a
c
r
o
s
s
t
h
e
s
a
m
p
l
e
p
e
r
i
o
d
f
o
l
l
o
w
i
n
g
t
h
e
F
a
m
a
a
n
d
M
a
c
B
e
t
h
(
1
9
7
3
)
p
r
o
c
e
d
u
r
e
Table IV.
Association between the
number of analyst
following and the greater
forecast accuracy of the
consensus forecast
ARJ
23,1
108
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast accuracy. The coef?cient on the MEASURE indicator variable that measures
the difference in accuracy of the consensus forecast (coded one) versus the most recent
forecast (coded zero) is signi?cant and positive (0.3284, t ¼ 6.34), after controlling for
forecast aggregation and timing. This result indicates that the consensus forecast
generates larger AFEs and is less accurate than the most recent forecast after
controlling for the effects of the number of analysts following and the timeliness of
analysts’ forecasts. It also suggests that the aggregation value of the consensus
forecast outweighs the timing advantage of the most recent forecast, and the number of
analysts following largely explains the forecasting superiority of the consensus
forecast. The model’s explanatory power is low (R
2
¼ 1.98 per cent), but consistent
with previous ?ndings (Ramnath et al., 2005). The results from the post-1991 period are
consistent with the results for the overall sample period.
Both sets of regression results for the ASX 100 companies and companies outside
the Index show that the aggregating value of the consensus outweighs the timing
advantage of the most recent forecast. The coef?cients on number of analysts
following for both subsamples are signi?cantly negative (20.0586, t ¼ 23.80 for the
ASX 100 companies; 20.1154, t ¼ 28.94 for companies outside the index). The
association between forecast accuracy and the number of analyst following is stronger
for the companies outside the ASX 100 Index because these companies are usually
covered by fewer analysts as compared to the ASX 100 companies. The incremental
analyst coverage more effectively purges the individual analyst’s idiosyncratic error.
Overall, the results suggest that the consensus forecast is more accurate than the
most recent forecast. The forecast superiority of the consensus forecast can be
attributed to the aggregating value of the consensus forecast outweighing the timing
advantage of the most recent forecast. These results show that the greater accuracy of
the consensus forecast is due to the number of analysts following and contributing to
the consensus. The results are robust to using the mean consensus rather than the
median consensus, constructed by using individual analyst forecasts. They are also
robust to using the I/B/E/S summary consensus measure.
7. Robustness checks
7.1 Excluding the most recent forecast from the consensus
If the most recent forecast is issued before the formation date of the last consensus prior
to the earnings announcement, it will be included in the calculation of the consensus
forecast. In the ?nal sample, 2,888 observations (66 per cent of total observations) have
the most recent forecast included in the consensus for the corresponding company-year.
Given this, the question arises whether, the relative timeliness of the consensus is
attributed to its increasing accuracy together with its aggregation value. In this section,
the relative accuracy between the consensus and the most recent forecast is re-examined
using a subsample which excludes the most recent forecast from the calculation of the
consensus.
Column (1) of Table V repeats the analysis in Table IV. The control variables for
number of analysts following and timeliness are the same as those described in the
regression model (4). The descriptive statistics (not tabulated) show that the median
age of the consensus excluding the most recent forecast is 104 calendar days before the
earnings announcement, suggesting that timeliness of consensus forecasts is improved
by including recently-updated forecasts.
Accuracy
of analysts’
expectations
109
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The results in column (1) of Table V are similar to those in Table IV, showing that
the coef?cient on number of analysts following is negative and statistically signi?cant;
whereas, the coef?cient on timeliness of analysts’ forecasts is close to zero and
statistically insigni?cant, indicating that the number of analysts following explains the
consensus forecasting advantage. The results demonstrate that, conditional on relative
timeliness of consensus forecasts, there are gains in accuracy from aggregating to
reduce idiosyncratic error.
7.2 Controlling for company and analyst characteristics
Consistent with Ramnath et al. (2005) and Ke and Yu (2006), company-speci?c
variables are not included in the model when comparing the accuracy of the two
Dependent variable: AFE
Variable Predicted sign (1) (2)
Intercept þ 0.9797
* * *
(21.81) 4.3494
* * *
(6.42)
MEASURE þ 0.4127
* * *
(4.27) 0.1610
* *
(2.28)
ANALYST 2 0.0775
* * *
(26.71) 20.0247
* * *
(23.44)
TIMELINESS þ 20.0003
*
(20.29) 20.0013 (21.91)
PASTAFE þ 23.3971
* * *
(7.06)
LNSIZE 2 20.1802
* * *
(25.35)
ASX100 2 0.1264 (1.67)
TACC 2 20.8971
* * *
(23.52)
Number of observations 3,766 2,346
Adjusted R
2
(%) 2.35 11.44
Notes: Statistical signi?cance at the
*
0.10,
* *
0.05 and
* * *
0.01 levels (two-tail test),
respectively; AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ 1
jts
ð1Þ
for the subsample excluding the most recent forecast from the calculation of the consensus
AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ b
4
ðPASTAFE
jts
Þ þ
b
5
ðLNSIZE
jts
Þ þ b
6
ðASX100
jts
Þ þ b
7
ðTACC
jts
Þ þ 1
jts
ð2Þ for the subsample including additional
variables: past forecast accuracy, ?rm size, whether the company observation is included in the ASX 100
Index, and the level of accruals AFE
jts
is the absolute value of the median consensus (s ¼ median) or the
most recent (s ¼ mr) forecast error de?ated by share price 11 months before the earnings announcement
month for company j’s annual EPS in year t; this variable is multiplied by 100, so, the coef?cient estimates
are as a percentage of share price; MEASURE
jts
is an indicator variable, coded one if AFEis sourced from
the median consensus forecast (s ¼ median); coded zero if AFE is sourced from the most recent forecast
(s ¼ mr) for company j in year t; ANALYST
jts
equals the number of analysts contributing to the
consensus forecast (s ¼ median) or one for the most recent forecast (s ¼ mr) for company j in year t;
TIMELINESS
jts
equals the average value of timeliness of individual analyst forecasts included in the last
consensus prior to the earnings announcement, where the timeliness of an individual analyst forecast is
measured by the number of calendar days between the date of the individual analyst forecast issued prior
to the earnings announcement and the earnings announcement date for the median consensus forecast
(s ¼ median) or the number of calendar days between the date of the most recent forecast issued prior to
the earnings announcement and the earnings announcement date for the most recent forecast (s ¼ mr) for
company j in year t; PASTAFE
jts
is the AFE
jts
for the prior year; LNSIZE
jts
is the log market capitalisation
11 months before the earnings announcement month for company j in year t; ASX100
jts
is an indicator
variable, coded one if company j is included in the ASX100 Index; coded zero if company is outside the
ASX100 Index in year t; TACC
jts
equals the difference between net income and cash ?owfromoperations
for company j in year t; the regression is estimated for each year t, respectively; coef?cient estimates are
presented as the mean across the sample period following the Fama and MacBeth (1973) procedure
Table V.
Association between the
number of analyst
following and the greater
forecast accuracy of the
consensus forecast
ARJ
23,1
110
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast measures. That is, the basic model compares the accuracy of the two measures
irrespective of the source of the differences in accuracy. To ensure the results are
robust to the inclusion of other factors associated with the accuracy of analysts’
forecasts, an additional set of variables are included in the analysis: past forecast
accuracy, ?rm size, whether or not the company observation is included in the ASX 100
Index, and the level of accruals.
Lim (2001) and Brown (2001b) show that analysts’ forecast accuracy is signi?cantly
positively correlated with their past forecast accuracy. Brown (2001b) suggests that
analysts’ past forecasting performance has a greater predictive power for forecast
accuracy than all analyst characteristics combined. Accordingly, a control for past
forecast accuracy to proxy for analyst characteristics is included. Since high-reputation
analysts provide more accurate and timely forecasts and have a greater impact on share
prices (Stickel, 1992; Brown et al., 2007), past forecast accuracy is also used to proxy for
the reputation of analysts.
Brown et al. (1999) and Lim(2001) ?nd that ?rmsize is negatively related to analysts’
forecast errors, indicating that the general information environment is likely to be
richer for large companies resulting in analysts issuing more accurate forecasts for
them. An indicator variable to control for the possible different information
environment for companies included in the ASX 100 Index as compared to companies
outside the index is also incorporated (Chan et al., 2006; Brown et al., 2007).
A company’s earnings management behaviour may affect quality of the reported
earnings and hence analysts’ forecast accuracy. Since managers perceive analysts as
one of the most important groups in?uencing the share price of their companies
(Graham et al., 2005), they could use accruals, or issue management earnings guidance
to manage earnings in an attempt to meet or beat analysts’ forecasts (Habib and
Hossain, 2008). For robustness, a control for total accruals as a proxy for the quality of
the reported earnings is incorporated into the regression.
Imposing the requirement that past forecast data, the constituent list for the ASX
100 Index, and accruals data are available reduces the sample size to 2,346
observations. Consistent with prior research, column (2) of Table V illustrates that
analysts’ forecast accuracy is signi?cantly positively correlated with their past forecast
accuracy and increases with ?rm size. Although it is in the opposite direction of the
predicted sign and is statistically insigni?cant, the coef?cient on index is signi?cantly
negative after the ?rm size is dropped from the regression (not tabulated); this indicates
that ?rm size and the index are highly correlated. Analysts’ forecast accuracy is
signi?cantly negatively correlated with total accruals, suggesting that earnings
management has an effect on analysts’ forecast accuracy. Adjusted R
2
increases to 11
per cent. The results remain robust to the inclusion of additional sets of variables. That
is, the number of analysts following continues to explain the greater accuracy of the
consensus forecast.
8. Conclusions
This study provides direct evidence of the accuracy of the consensus forecast versus
the most recent forecast, as measures of the market’s earnings expectations prior to
earnings announcements in the Australian market by examining 4,358 company-year
annual analyst forecasts between 1987 and 2007. Consistent with US evidence (O’Brien,
1988; Brown, 1991) the results suggest that in the late 1980s there is some evidence that
Accuracy
of analysts’
expectations
111
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the most recent forecast is more accurate than the consensus. This is, however, the
results in more recent years show that the consensus forecast is more accurate than
the most recent forecast. For the sample period from 1991 to 2007, the accuracy of the
consensus forecast consistently outperforms that of the most recent forecast in 15 out
of the 17 years; statistically signi?cant differences are shown for nine out of those 15
years. These results supports the ?ndings reported by recent US studies (Barron and
Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005) and is consistent with the improving
timeliness of forecasts included in consensus forecasts. The forecasting superiority of
the consensus forecast can be attributed to the number of analysts following and
contributing to the consensus. The aggregation value of the consensus outweighs the
small timing advantage of the most recent forecast over the short forecast horizon
examined in this study.
This study contributes to prior research on Australian analysts’ forecasts by
providing evidence on the importance of diversifying idiosyncratic individual error
across analyst forecasts in the consensus forecast, in a non-US setting with relatively
thin analyst coverage and a different disclosure environment. It also con?rms the
market practitioners’ views as evidenced by press reports, that the consensus forecast
is a better measure of the market’s earnings expectations.
Future research could examine how analysts change their forecasting behaviour to
maintain their forecast accuracy in an environment of increased regulation over the
dissemination of company information. Future research might also pro?tably consider
whether consensus forecasts can be improved by forming a consensus based on
forecasts of a subset of highly skilled analysts or analysts possessing certain attributes.
Notes
1. There is an alternative approach to derive a consensus forecast which is to vary
component-forecast weights as a function of the expected accuracy of each forecast such as
forecast age, broker size and analyst experience (Kim et al., 2001; Brown and Mohd, 2003;
Butler et al., 2007). Since this approach is not widely used in the investment community
or academic research, this study focuses on the consensus forecast de?ned as the mean or
median of outstanding individual analyst forecasts at any point in time.
2. For example, Brown et al. (1999) examine properties of analysts’ forecasts in different
corporate disclosure environments. Beekes and Brown (2006) use analyst forecast accuracy
as one of the indicators for the informativeness of a ?rm’s disclosure to investigate the
association between a listed Australian ?rm’s corporate governance quality and its
disclosure and the market response. Habib and Hossain (2008) use analysts’ forecasts to
examine whether Australian managers manage earnings in an attempt to meet or beat
analysts’ forecasts.
3. For example, a search using text terms of “consensus forecast(s) or consensus estimate(s)”
?nds 800 articles in Australian key newspapers from 1 January 2007 to 31 December 2008
while a search using text terms of “most recent analyst forecast or most recent analyst
expectations” ?nds two articles in Factiva database.
4. An example of prior research relating to analysts’ forecasts in Australia is Jackson (2005).
Jackson presents evidence that analysts build their reputations by providing accurate
forecasts, and that high-reputation analysts generate more trading volume. Recent studies
examine share price responses to Australian analysts’ research quality and their stock
recommendations. For example, Brown et al. (2007) study the market response to initiating
ARJ
23,1
112
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
recommendations made by analysts in terms of returns and share price responses. Aitken
et al. (2008) examine the relation between analysts’ research quality and price discovery.
5. Barron and Stuerke (1998) note improvements in the integrity of the I/B/E/S database. Brown
(2001a) documents improvement over time in the accuracy of I/B/E/S forecasts. Ramnath
et al. (2005) ?nd increased reliability of I/B/E/S forecasts. The increasing competition
between forecast data providers may lead to these improvements. For example, First Call
Corporation began to provide analyst forecast data in the early 1900s (Brown, 2001a;
Ramnath et al., 2005).
6. This argument appears to assume, however, that the few analysts who update their forecasts
have access to information that is not available to the majority of analysts who choose not to
update their forecasts, all else being equal. It is unlikely that selective brie?ngs are
widespread given the disclosure environment brie?y discussed in the following section. Any
ability of individual analysts to outperform the consensus close to the earnings
announcement does however raise interesting questions regarding the source of any
superior performance, and hence the need to examine previous results using US data within
other regulatory environments.
7. Using I/B/E/S data for US companies, Ke and Yu (2006) report that 21.07 analysts on average
cover a company during 1983-2000. Barron and Stuerke (1998) show that the average
number of analyst following from 1990 to 1994 is about 16 in their sample. In this study, the
average number of analysts per company is seven.
8. While the policy is stated in terms of all forecasts, I/B/E/S does drop off stale forecasts when
calculating its monthly summary statistics. I/B/E/S reports the number of analysts’ forecasts
included in the consensus and this number of forecasts is used to identify the forecasts
assumed to be included in the consensus in this study.
9. For example, if I/B/E/S reports that 12 individual forecasts are included in the calculation of
its summary consensus, then the most recent forecasts issued by 12 individual analysts as
of the publication date of the I/B/E/S summary report will be used in the computation of
forecast statistics.
10. This study is interested in mimicking the I/B/E/S consensus forecast that is broadly
available to users on a monthly basis, rather than in creating a superior consensus.
The reconstruction of the I/B/E/S consensus is necessary as both the aggregated value of the
consensus and the attributes of individual forecasts that comprise the consensus are
required by the study.
11. Earnings announcement dates are sourced from I/B/E/S between 1987 and 1992 because they
are not available in the SIRCA database. Of the sample, observations with earnings
announcement dates differing by more than one day between SIRCA and I/B/E/S are less
than 10 per cent of observations between 2003 and 2007, more than 10 per cent but less
than 20 per cent of observations in 1999, 2000 and 2002, more than 20 per cent but less than
30 per cent of observations in 2001, and the discrepancies increase substantially before 1999
(between 1993 and 1998). Earnings announcement dates reported by I/B/E/S tend to be later
than those reported by SIRCA. Earnings announcement dates reported in SIRCA tend to
correspond with the ASX announcement dates wherever available on the ASX web site.
12. If a company is traded on multinational stock exchanges and followed by multinational
analysts, only Australian analyst forecasts are included in the sample.
13. As expected, the sample re?ects companies with analyst coverage and not all companies on
the ASX are followed by analysts. The results must be interpreted with respect to this
limitation.
Accuracy
of analysts’
expectations
113
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
14. Brown (1996) argues the share price is the appropriate de?ator to use when valuing
companies because the primary use of analysts’ earnings forecasts for security analysis is to
make investment decisions.
15. For companies listed less than 11 months, their forecast errors are de?ated by the ?rst listed
month closing share prices.
16. Consistent with Bartov et al. (2002) and Brown and Caylor (2005), the average value of the
analysts’ forecasts is used if more than one analyst forecast is issued on the most recent day.
17. The qualitative results remain unchanged when signed forecast errors are used to measure
forecast accuracy.
18. The difference between the average AFE for the most recent forecast (0.885 per cent) and the
average AFE for the median consensus (0.975 per cent) for the full sample is 0.09 per cent of
share price.
19. The average market capitalisation ($2 billion) multiplied by 0.09 per cent is $1.8 million.
20. The same conclusion is reached if the mean consensus forecast is used.
References
Aitken, M., Almeida, N., Harris, F.H.D. and McInish, T.H. (2008), “Financial analysts and price
discovery”, Accounting & Finance, Vol. 48, pp. 1-24.
Barron, O.E. and Stuerke, P. (1998), “Dispersion in analysts’ earnings forecasts as a measure of
uncertainty”, Journal of Accounting, Auditing & Finance, Vol. 13, pp. 243-68.
Barron, O.E., Byard, D. and Yu, Y. (2008), “Earnings surprises that motivate analysts to reduce
average forecast error”, The Accounting Review, Vol. 83, pp. 303-25.
Bartov, E., Givoly, D. and Hayn, C. (2002), “The rewards to meeting or beating earnings
expectations”, Journal of Accounting and Economics, Vol. 33, pp. 173-204.
Beekes, W. and Brown, P. (2006), “Do better governed Australian ?rms make more informative
disclosures?”, Journal of Business Finance & Accounting, Vol. 33, pp. 422-50.
Brown, L.D. (1991), “Forecast selection when all forecasts are not equally recent”, International
Journal of Forecasting, Vol. 7, pp. 349-56.
Brown, L.D. (1993), “Earnings forecasting research: its implications for capital markets
research”, International Journal of Forecasting, Vol. 9, pp. 295-320.
Brown, L.D. (1996), “Analyst forecasting errors and their implications for security analysis:
an alternative perspective”, Financial Analysts Journal, Vol. 52, pp. 40-7.
Brown, L.D. (2001a), “A temporal analysis of earnings surprises: pro?ts versus losses”, Journal of
Accounting Research, Vol. 39, pp. 221-41.
Brown, L.D. (2001b), “How important is past analyst forecast accuracy?”, Financial Analysts
Journal, Vol. 57, p. 44.
Brown, L.D. and Caylor, M.L. (2005), “A temporal analysis of quarterly earnings thresholds:
propensities and valuation consequences”, The Accounting Review, Vol. 80, pp. 423-40.
Brown, L.D. and Kim, K.J. (1991), “Timely aggregate analyst forecasts as better proxies for
market earnings expectations”, Journal of Accounting Research, Vol. 29, pp. 382-5.
Brown, L.D. and Mohd, E. (2003), “The predictive value of analyst characteristics”, Journal of
Accounting, Auditing & Finance, Vol. 18, pp. 625-47.
Brown, L.D. and Rozeff, M.S. (1978), “The superiority of analyst forecasts as measures of
expectations: evidence from earnings”, Journal of Finance, Vol. 33, pp. 1-16.
ARJ
23,1
114
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Brown, L.D., Hagerman, R.L., Grif?n, P.A. and Zmijewski, M.E. (1987a), “An evaluation of
alternative proxies for the market’s assessment of unexpected earnings”, Journal of
Accounting and Economics, Vol. 9, pp. 159-93.
Brown, L.D., Hagerman, R.L., Grif?n, P.A. and Zmijewski, M.E. (1987b), “Security analyst
superiority relative to univariate time-series models in forecasting quarterly earnings”,
Journal of Accounting and Economics, Vol. 9, pp. 61-87.
Brown, P., Taylor, S.L. and Walter, T.S. (1999), “The impact of statutory sanctions on the level
and information content of voluntary corporate disclosure”, Abacus, Vol. 35, pp. 138-62.
Brown, R., Chan, H.W.H. and Ho, Y.K. (2007), “Initiating coverage, broker reputation and
management earnings forecasts in Australia”, Accounting & Finance, Vol. 47, pp. 401-21.
Butler, M.B., Kraft, A.G. and Markov, S. (2007), “On the weighting of individual analyst forecasts
in the consensus”, working paper, Emory University, Atlanta, GA.
Chan, H.W.H., Brown, R. and Ho, Y.K. (2006), “Initiation of brokers’ recommendations, market
predictors and stock returns”, Journal of Multinational Financial Management, Vol. 16,
pp. 213-31.
Clement, M.B. (1999), “Analyst forecast accuracy: do ability, resources, and portfolio complexity
matter?”, Journal of Accounting and Economics, Vol. 27, pp. 285-303.
Clement, M.B. and Tse, S.Y. (2005), “Financial analyst characteristics and herding behavior in
forecasting”, Journal of Finance, Vol. 60, pp. 307-41.
Fama, E.F. and MacBeth, J.D. (1973), “Risk, return, and equilibrium: empirical tests”, Journal of
Political Economy, Vol. 81, pp. 607-36.
Fried, D. and Givoly, D. (1982), “Financial analysts’ forecasts of earnings: a better surrogate for
market expectations”, Journal of Accounting and Economics, Vol. 4, pp. 85-107.
Graham, J.R., Harvey, C.R. and Rajgopal, S. (2005), “The economic implications of corporate
?nancial reporting”, Journal of Accounting and Economics, Vol. 40, pp. 3-73.
Habib, A. and Hossain, M. (2008), “Do managers manage earnings to ‘just meet or beat’ analyst
forecasts? Evidence from Australia”, Journal of International Accounting, Auditing and
Taxation, Vol. 17, pp. 79-91.
Ivkovic, Z. and Jegadeesh, N. (2004), “The timing and value of forecast and recommendation
revisions”, Journal of Financial Economics, Vol. 73, pp. 433-63.
Jackson, A.R. (2005), “Trade generation, reputation, and sell-side analysts”, Journal of Finance,
Vol. 60, pp. 673-717.
Jacob, J., Lys, T.Z. and Neale, M.A. (1999), “Expertise in forecasting performance of security
analysts”, Journal of Accounting and Economics, Vol. 28, pp. 51-82.
Ke, B. and Yu, Y. (2006), “The effect of issuing biased earnings forecasts on analysts’ access to
management and survival”, Journal of Accounting Research, Vol. 44, pp. 965-99.
Kim, O., Lim, S.C. and Shaw, K.W. (2001), “The inef?ciency of the mean analyst forecast as a
summary forecast of earnings”, Journal of Accounting Research, Vol. 39, pp. 329-35.
Kothari, S.P. (2001), “Capital markets research in accounting”, Journal of Accounting and
Economics, Vol. 31, pp. 105-231.
Lim, T. (2001), “Rationality and analysts’ forecast bias”, Journal of Finance, Vol. 56, pp. 369-85.
Mikhail, M.B., WaIther, B.R. and Willis, R.H. (1999), “Does forecast accuracy matter to security
analysts?”, The Accounting Review, Vol. 74, pp. 185-200.
O’Brien, P.C. (1988), “Analysts’ forecasts as earnings expectations”, Journal of Accounting and
Economics, Vol. 10, pp. 53-83.
Accuracy
of analysts’
expectations
115
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
O’Brien, P.C. (1990), “Forecast accuracy of individual analysts in nine industries”, Journal of
Accounting Research, Vol. 28, pp. 286-304.
Ramnath, S., Rock, S. and Shane, P. (2005), “Value line and I/B/E/S earnings forecasts”,
International Journal of Forecasting, Vol. 21, pp. 185-98.
Richardson, S., Teoh, S.H. and Wysocki, P.D. (2004), “The walk-down to beatable analyst
forecasts: the role of equity issuance and insider trading incentives”, Contemporary
Accounting Research, Vol. 21, pp. 885-924.
Stickel, S.E. (1992), “Reputation and performance among security analysts”, The Journal of
Finance, Vol. 47, pp. 1811-36.
Corresponding author
Xiaomeng Chen can be contacted at: [email protected]
ARJ
23,1
116
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
This article has been cited by:
1. Rafael Moreira Antônio, Fabiano Guasti Lima, Tabajara Pimenta Junior. 2015. Stock recommendations
and investment portfolio formation: A study in the Brazilian market. Contaduría y Administración 60,
874-892. [CrossRef]
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_174750723.pdf
This paper aims to use Australian analysts’ forecast data to compare the relative accuracy
of consensus and the most recent forecast in the month before the earnings announcement.
Accounting Research Journal
Australian evidence on the accuracy of analysts' expectations: The value of consensus
and timeliness prior to the earnings announcement
Xiaomeng Chen
Article information:
To cite this document:
Xiaomeng Chen, (2010),"Australian evidence on the accuracy of analysts' expectations", Accounting
Research J ournal, Vol. 23 Iss 1 pp. 94 - 116
Permanent link to this document:
http://dx.doi.org/10.1108/10309611011060542
Downloaded on: 24 January 2016, At: 21:10 (PT)
References: this document contains references to 39 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 985 times since 2010*
Users who downloaded this article also downloaded:
Katherine Uylangco, Steve Easton, Robert Faff, (2010),"The equity and efficiency of the Australian share
market with respect to director trading", Accounting Research J ournal, Vol. 23 Iss 1 pp. 5-19 http://
dx.doi.org/10.1108/10309611011060506
Raghavan J . Iyengar, J udy Land, Ernest M. Zampelli, (2010),"Does board governance improve
the quality of accounting earnings?", Accounting Research J ournal, Vol. 23 Iss 1 pp. 49-68 http://
dx.doi.org/10.1108/10309611011060524
Wendy Green, Richard D. Morris, Haiping Tang, (2010),"The split equity reform and corporate
financial transparency in China", Accounting Research J ournal, Vol. 23 Iss 1 pp. 20-48 http://
dx.doi.org/10.1108/10309611011060515
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Australian evidence
on the accuracy of analysts’
expectations
The value of consensus and timeliness prior
to the earnings announcement
Xiaomeng Chen
Department of Accounting and Finance, Macquarie University,
North Ryde, Australia
Abstract
Purpose – This paper aims to use Australian analysts’ forecast data to compare the relative accuracy
of consensus and the most recent forecast in the month before the earnings announcement.
Design/methodology/approach – Cross-sectional regression is used on a sample of 4,358
company-year observations of annual analyst forecasts to examine whether the number of analysts
following and the timeliness of an individual analyst’s forecast is more strongly associated with the
superior forecast measure.
Findings – The results suggest that whilst in the late 1980s the most recent forecast was more
accurate than the consensus, since the early 1990s the accuracy of the consensus forecast has
outperformed the most recent forecast in 15 out of 17 years, and the differences are signi?cant for nine
out of 15 years. The forecasting superiority of the consensus can be attributed to the aggregating value
of the consensus outweighing the small timing advantage of the most recent forecast over the short
forecast horizon examined in this paper.
Research limitations/implications – Given the consistent use of analysts’ forecasts as proxies for
expected earnings in Australian research, this paper provides insights to what extent the expected
level of forecast accuracy is realised and the reasons for the greater accuracy in the superior forecast
measure.
Practical implications – The ?ndings con?rm market practitioners’ views that the consensus
forecast is a better measure of the market’s earnings expectations.
Originality/value – This paper provides direct evidence of the accuracy of alternative forecast
measures and the importance of diversifying idiosyncratic individual error across analyst forecasts.
Keywords Earnings, Financial analysis, Financial forecasting, Australia
Paper type Research paper
1. Introduction
It is well established in the literature that analysts’ earnings forecasts are used as
proxies for market expectations of future earnings because they are more accurate and
have a stronger association with excess returns on the date of the earnings
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1030-9616.htm
The author gratefully acknowledges helpful comments from Sue Wright, Neil Fargher, Andrew
Ferguson, Paul Healy, Egon Kalotay, Terry Walter, Hai Wu, two anonymous referees and
participants at the Sydney Summer School, University of Technology, 31st EAA Annual
Congress 2008 and AFAANZ Conference 2008. The author also thanks I/B/E/S for providing
analyst forecast data.
ARJ
23,1
94
Accounting Research Journal
Vol. 23 No. 1, 2010
pp. 94-116
qEmerald Group Publishing Limited
1030-9616
DOI 10.1108/10309611011060542
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
announcement than time-series models of earnings (Brown and Rozeff, 1978; Fried and
Givoly, 1982; Brown et al., 1987a, b). Analysts’ earnings expectations can however be
measured in a number of ways. Two widely used approaches are a consensus forecast
that aggregates individual analyst forecasts at any point in time and a single, most
recent forecast provided by an individual analyst[1].
A consensus forecast diversi?es away idiosyncratic individual error to gain value
from the aggregation. The most recent forecast made over a shorter forecast horizon
than the consensus is more timely. Tradeoffs between the bene?ts of forecast
aggregation and timeliness of forecasts motivate this study to compare the relative
accuracy of the consensus andthe most recent forecast. This studyspeci?cally examines
whether the number of analysts following a company or the timeliness of an individual
analyst’s forecast is more strongly associated with the superior forecast measure.
Prior research in the late 1980s and early 1990s using data from the US market
examines the relative accuracy of alternative earnings forecast measures provided by
standard sources of analysts forecast data such as I/B/E/S and shows that the most
recent forecast is relatively more accurate than the consensus forecast (O’Brien, 1988;
Brown, 1991). Brown and Kim (1991) ?nd that the most recent forecast is more closely
related to share prices than the consensus forecast. Based on these results, many
studies use the most recent forecast as a measure of the market’s earnings expectations
(Brown, 2001a; Bartov et al., 2002; Brown and Caylor, 2005). However, recent studies
document that the consensus has become a more timely measure in the past decade due
to improvements in analyst forecast data (Barron and Stuerke, 1998; Brown, 2001a;
Ramnath et al., 2005). These studies suggest that attempts have been made to include
only relatively recent forecasts in the consensus to improve the timeliness of consensus
forecasts. Taken together, these ?ndings suggest that prior conclusions of the
superiority of the most recent forecast may no longer apply due to the changing nature
of consensus forecasts in more recent years.
Research in the Australian market has increasingly used analysts’ forecasts as
proxies for expected earnings (Brown et al., 1999; Beekes and Brown, 2006; Habib and
Hossain, 2008)[2]. Australian press reports overwhelmingly cite analysts’ earnings
expectations using the consensus rather than the most recent forecast[3]. While
extensive research on analysts’ forecasts is available for the US market, there is
relatively limited research related to analysts’ forecasts using Australian data
( Jackson, 2005; Brown et al., 2007; Aitken et al., 2008)[4].
This study contributes to prior research on Australian analysts’ forecasts in several
ways. First, this paper provides direct evidence of the accuracy of alternative forecast
measures, the consensus and the most recent forecast, as measures of the market’s
earnings expectations prior to earnings announcements. Recent studies document
improvements in timeliness of consensus forecasts, but these studies do not directly
compare the accuracy of the consensus and the most recent forecast. Given the consistent
use of analysts’ forecasts as proxies for expected earnings in Australian research and
press reports, it is important to understand to what extent the expected level of forecast
accuracy is realised and the reasons for the greater accuracy in the superior forecast
measure. Second, this study provides further evidence on the accuracy of the consensus
in reducing idiosyncratic error by diversifying across analyst forecasts in a market with
relatively few analyst coverage and a different disclosure regime. Last, it also con?rms
Accuracy
of analysts’
expectations
95
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the market practitioners’ views as evidenced by press reports, that the consensus
forecast is a better measure of the market’s earnings expectations.
The results suggest that whilst in the late 1980s, the most recent forecast is more
accurate than the consensus during the period immediately prior to the earnings
announcement, since the early 1990s the consensus forecast outperforms the most recent
forecast. That is, the most recent forecast is, on average, less accurate than the most recent
consensus available inthe month prior to the earnings announcement. The accuracy of the
consensus forecast consistently outperforms the most recent forecast in 15 out of these 17
more recent years and the differences are signi?cant for nine out of these 15 years.
The number of analysts following explains the greater accuracy of the consensus. The
aggregating value of the consensus outweighs the small timing advantage of the most
recent forecast over the short forecast horizon examined in this study.
The results from the late 1980s are consistent with those of earlier studies (O’Brien,
1988; Brown, 1991) using US data that ?nd that the most recent forecast is more accurate
than the consensus. However, the results in more recent years indicate the opposite,
consistent with improvements in timeliness of forecasts included in consensus forecasts
identi?ed in prior studies (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005).
The recent results suggest that the greater accuracy of the consensus forecast comes from
diversifying away idiosyncratic error in individual forecasts, conditional on only
relatively recent forecasts being included in the consensus.
The remainder of the paper is organised as follows. Section 2 reviews the related
literature. Section 3 develops the hypotheses. Section 4 describes the sample selection
and data. Section 5 discusses the variable de?nitions and research methods. Section 6
reports the results of tests. Section 7 presents the results of additional analysis. Section 8
concludes.
2. Prior research
2.1 The aggregation value of consensus forecasts
Analysts make andrevise their earnings forecasts throughout the year as theyincorporate
new information into their forecasts. O’Brien (1988, p. 53) suggests that, “Since a diverse
set of forecasts is available at any time for a given ?rm’s earnings, composites are used to
distil the diverse set into a single expectation”. A consensus forecast is a forecast that
aggregates all information available to analysts. It is often de?ned as the mean or median
of outstanding individual analyst forecasts at any point in time.
An aggregate forecast is expected to average out potential inef?ciencies in how
individual analysts process information and therefore provide more accurate future
earnings expectations. For example, the I/B/E/S consensus forecast in the USA is more
accurate and offers a better proxy for the market’s earnings expectations than a single
forecaster (value line) immediately before a quarterly earnings announcement (Ramnath
et al., 2005). Their study shows that most of the consensus forecasting superiority
can be attributed to the aggregation value. Since the consensus forecast aggregates
expectations from various analysts and stockbroking ?rms who are covering a
company, the idiosyncratic analyst error is diminished through the aggregation process.
Thus, the accuracy of the consensus is improved.
As suggested by Barron et al. (2008), the larger the number of analysts following
and contributing to the consensus, the more the idiosyncratic analyst error is averaged
out in determining the consensus forecast, and the higher is the accuracy of the
ARJ
23,1
96
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
consensus forecast. This suggestion motivates an examination of the association
between the number of analysts following and the accuracy of consensus forecasts in
this study.
The timeliness of individual forecasts included in the consensus forecast is
important when evaluating its accuracy. Since not all analysts update their forecasts in
a timely manner, the consensus forecast at any point in time includes both recent and
potentially stale forecasts (Kothari, 2001). The inclusion of stale forecasts is likely to
reduce the accuracy of consensus forecasts.
Many studies document that the consensus has become a more timely measure over
time due to recent improvements in the timeliness and quality of analyst forecast data
included (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005)[5]. These
studies suggest that attempts have been made to include only relatively recent
forecasts in the consensus. O’Brien (1988) ?nds that the consensus is signi?cantly
better than the most recent forecast when the consensus is relatively timely. That is,
conditional on only reasonably timely forecasts being included in the consensus, an
aggregate or consensus forecast is expected to diversify away idiosyncratic individual
error and therefore provide more accurate future earnings expectations.
2.2 Timeliness of analysts’ forecasts
O’Brien (1988) uses I/B/E/S individual analyst forecast data to compute and compare
three alternative forecast measures: the mean, the median, and the most recent
individual analyst forecast. She ?nds that the most recent forecast is more accurate than
both the mean and median forecasts in the 1975-1981 period. More recent studies
con?rmthat the accuracy of earnings forecasts improves as the earnings announcement
date approaches (Lim, 2001; Ivkovic and Jegadeesh, 2004). These studies indicate that
analysts are able to incorporate newinformation into their forecasts. This highlights the
importance of timeliness of forecasts for improving forecast accuracy.
In particular, the forecasts made over the short forecast horizon (i.e. in the period
immediately prior to the earnings announcements) will be the most informative and
accurate. As suggested by Ivkovic and Jegadeesh (2004), the analysts who update their
forecasts most recently have access to all prior forecasts made by other analysts and
will use them rationally in making their own forecasts. In addition, they may have
early access to earnings information such as management’s guidance on future
earnings. Although company management can choose to provide earnings guidance at
any point in time, any guidance they provide will be more accurate the closer it is to the
earnings announcement. If some analysts obtain early access to such information, then
their earnings forecasts will be superior to others[6]. In this study, the most recent
forecast is used to examine its accuracy in comparison with the consensus forecast.
2.3 Tradeoffs between forecast aggregation and timeliness of forecasts
The consensus forecast is characterised by aggregation value from diversifying across
idiosyncratic individual error. The most recent forecast made over a shorter forecast
horizon is more timely. Brown (1991) investigates tradeoffs between the bene?ts of
forecast aggregation and timeliness of forecasts. He adopts an approach of dropping
stale forecasts from the consensus by using three timely forecast measures (i.e. the
most recent forecast, an average of the three most recently issued forecasts and an
average of all forecasts issued within the past 30 days). He ?nds that the comparative
Accuracy
of analysts’
expectations
97
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
advantage of each forecast measure depends on company size. For large companies, the
30-day average is signi?cantly more accurate than the most recent forecast; for small
companies, the most recent forecast is more accurate than the other two alternative
forecast measures. Brown’s results suggest that the forecast aggregation outweighs the
timeliness of forecasts for large companies with more analysts following. In contrast,
the most recent forecast shows its timing advantage for small companies for which the
bene?t of aggregation of individual analyst forecasts is ineffective. However, Brown
does not directly examine the association between the number of analysts following
and the accuracy of these forecast measures.
Tradeoffs between the bene?ts of forecast aggregation and timeliness of forecasts
motivate this study to compare the accuracy of the aggregate consensus forecast and
the most recent forecast. Speci?cally, this study examines whether the number of
analyst following is more strongly associated with the superior forecast measure.
3. Hypothesis development
Since analysts do not issue forecasts at prescribed times, there is variation in the age of
forecasts included in the consensus. Forecast accuracy generally improves as the
earnings announcement date approaches because analysts incorporate newinformation
into their forecasts (O’Brien, 1988, 1990). If forecast age is the single most important
factor associated with forecast accuracy (Clement, 1999; Jacob et al., 1999), then the more
recent forecast is expected to be more accurate than older ones. Brown (1991) argues that
the consensus forecast is less accurate than more timely forecast measures, including
the most recent forecast, because the consensus includes stale forecasts. Stale forecasts
reduce forecast accuracy because recent earnings information is omitted.
On the other hand, the consensus forecast is expected to average out the individual
analyst’s idiosyncratic error through the aggregation process, which improves forecast
accuracy (Brown, 1993; Ramnath et al., 2005; Barron et al., 2008). If diversifying across
individual idiosyncrasies is more important than discarding stale forecasts, then the
consensus forecast that aggregates multiple analysts’ forecasts may be more accurate
than a single recent forecast.
Relative timeliness of the consensus forecast is also important when evaluating its
accuracy. O’Brien (1988) shows that the consensus forecast is more accurate than the
most recent forecast only when relatively recent forecasts are included in the consensus.
Recent improvements in analyst forecast data including I/B/E/S database are re?ected in
the consensus being a more timely forecast measure (Barron and Stuerke, 1998; Brown,
2001a; Ramnath et al., 2005). Aggregating to reduce idiosyncratic error in the consensus
is more effective when more timely individual forecasts are included in the consensus.
This study investigates how the number of analysts following a company and the
timeliness of an individual analyst’ forecast impacts on the differential accuracy of
the consensus andthe most recent forecast inAustralia. Like manysmaller economies, the
Australian market has relatively few brokerage ?rms and few analysts covering
companies[7]. The limited number of brokerage ?rms and analysts tend to cover
companies with high market capitalisation, leaving small companies with a thin coverage
at best. Since the market relies on the limited number of analysts providing coverage, it is
important to purge idiosyncratic error from analysts’ individual forecasts.
The disclosure environment, enhanced by continuous disclosure regulation since
1994, prohibits companies from brie?ng individual analysts with price-sensitive
ARJ
23,1
98
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
information, which mitigates the ability of individual analysts to gain incremental
information. For the short forecast horizon examined in this study, that is, during the
period immediately prior to the earnings announcement, the aggregation value of the
consensus forecast is expected to outweigh the timing advantage of the most recent
forecast in the Australian context. Therefore, the consensus forecast is expected to be
more accurate than the most recent forecast, and the ?rst hypothesis is:
H1. The consensus forecast is more accurate than the most recent individual
analyst earnings forecast.
A possible explanation for the greater accuracy of the consensus forecast is the
aggregation value from including expectations of multiple analysts. If the relatively
greater accuracy of the consensus forecast is largely explained by the aggregation
value of the consensus forecast, then the difference in forecast accuracy should be
related to the number of analysts contributing to the consensus. Speci?cally, the larger
the number of analysts following and contributing to the consensus, the greater is the
accuracy of the consensus forecast. Based on this, the second hypothesis is formed as:
H2. The greater forecast accuracy of the consensus forecast is due to the number
of analysts contributing to the consensus.
4. Sample and descriptive statistics
4.1 Sources of data
Data on one year ahead analysts’ forecasts of annual earnings per share (EPS) are
obtained from the I/B/E/S International Summary and Detail History ?les. The
Summary ?les contain the summary statistics on analyst forecasts, such as means,
medians and standard deviations. The Detail ?les provide individual analyst forecasts
and the date of each forecast issued. The summary data are calculated and reported by
I/B/E/S on the basis of all outstanding forecasts as of the third Thursday of each month
using the individual forecasts in the Detail ?les[8].
The mean and median consensus forecasts are calculated using individual analyst
forecasts to match the I/B/E/S summary consensus. The number of individual forecasts
available to calculate forecast statistics, as at the publication date of the last I/B/E/S
summary report prior to the earnings announcement, is matched against the number of
individual forecasts included in the I/B/E/S consensus[9]. This approach enables the
identi?cation of individual forecasts included in the consensus and their forecast ages
when considering the timeliness of the consensus. This consensus forecast measure is
checked against the I/B/E/S summary measure and the two measures are found to be
closely matched[10]. Since the empirical results using the reconstructed consensus
measure or the I/B/E/S summary measure are very similar, onlythe results obtained using
the reconstructed consensus measure are reported in this study.
The corresponding actual earnings are obtained from I/B/E/S for comparability
with the forecast. Earnings announcement dates are sourced from the Securities
Industry Research Centre of Asia-Paci?c (SIRCA) database[11]. The constituent list for
the Australian Stock Exchange (ASX) 100 Index is obtained from the SIRCA database.
Share prices and market capitalisation information are obtained from the CRIF Share
Price and Price Relative database. Accounting information was sourced from the
ASPECT database.
Accuracy
of analysts’
expectations
99
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
4.2 Sample selection
The initial sample includes companies[12] traded on the ASX, and with at least one
I/B/E/S consensus forecast available and two analysts following for the period from
?scal 1987 to ?scal 2007. Consistent with prior studies (O’Brien, 1988; Mikhail et al.,
1999; Ramnath et al., 2005), the most recent I/B/E/S consensus forecast prior to the
earnings announcement is retained. The initial sample comprises 5,694 company-year
observations. Notably, many companies listed on ASX are not covered by I/B/E/S.
Company-year observations are eliminated if the actual earnings in I/B/E/S or the
earnings announcement dates from SIRCA are missing. Since the consensus forecast is
compared with the most recent forecast for a particular company and year, the most
recent company-year individual analyst forecasts prior to the earnings announcement
are extracted and matched from the Detail ?les. After observations with a mismatch of
?nancial-year end between reported actual earnings and forecasted earnings are
excluded, outliers are eliminated by omitting observations with price-de?ated forecast
error greater than 10 per cent (Richardson et al., 2004; Clement and Tse, 2005). These
observations are likely to be the result of a data entry error. Table I lists the sample
selection criteria and their effects on the sample size. As shown in Table I, the ?nal
annual earnings forecast sample yields 4,358 company-year observations, representing
862 unique companies[13].
4.3 Sample descriptive statistics
Panel A of Table II reports the year-by-year sample descriptive statistics for all
company-year observations. The number of companies followed by at least two
analysts varies across years and ranges from a low of 21 companies in 1987 to a high of
357 companies in 2007. The number of companies covered generally increases towards
the later years, re?ecting the increased coverage of I/B/E/S for the Australian market.
Companies in the sample have an average (median) market capitalisation of $2 ($0.5)
billion, re?ecting the skewed distribution of companies covered by analysts. That is,
analysts follow a limited number of very large companies. They also selectively cover
small or medium size companies.
Number of company-year
observations remaining in sample
Percentage of total
consensus forecasts
Consensus forecasts from 1987 to 2007
for Australian companies with at least
two analysts following 5,694 100
Actual earnings (from I/B/E/S) and
earnings announcement date (from
SIRCA) available 4,650 82
Excluding companies with change in
?nancial year-end 4,636 81
Excluding outliers ¼ ?nal sample (862
unique companies) 4,358 77
Notes: Consensus forecasts are the means or medians of all the individual analyst forecasts available
as at the publication date of the last I/B/E/S consensus before earnings announcement; individual
analyst forecasts are extracted from the I/B/E/S Detail History ?les; outliers are de?ned as the
observations where AFEs are greater than 10 per cent
Table I.
Sample selection
ARJ
23,1
100
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
M
e
a
n
o
r
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
N
u
m
b
e
r
o
f
?
r
m
s
M
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
(
$
0
0
0
)
N
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
F
i
s
c
a
l
y
e
a
r
f
o
l
l
o
w
e
d
M
e
a
n
M
e
d
i
a
n
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
P
a
n
e
l
A
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
a
l
l
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
t
h
e
s
a
m
p
l
e
1
9
8
7
2
1
3
1
4
,
0
9
9
9
5
,
0
3
4
4
2
3
5
1
2
1
8
9
1
2
4
1
4
3
5
7
2
5
4
1
7
9
1
9
8
8
6
7
5
2
5
,
3
6
5
2
0
3
,
4
6
7
6
2
4
1
0
1
4
1
9
7
1
4
5
1
7
8
5
6
2
2
3
7
7
1
1
9
8
9
1
3
5
9
0
2
,
8
7
6
3
3
0
,
7
1
4
8
4
7
1
1
1
0
0
7
5
9
7
1
2
3
3
0
1
7
2
6
3
0
1
9
9
0
1
0
9
1
,
2
5
8
,
6
0
4
5
1
7
,
2
9
4
8
5
8
1
1
1
2
8
9
6
1
3
0
1
5
5
3
7
2
0
3
3
5
0
1
9
9
1
1
1
9
1
,
0
7
1
,
7
9
8
3
3
2
,
9
5
2
7
3
7
1
0
8
9
5
7
8
8
1
1
7
2
4
1
4
2
0
2
8
1
9
9
2
1
3
6
1
,
2
1
1
,
6
4
6
3
7
1
,
5
7
8
6
3
5
8
9
5
7
5
9
4
1
1
7
3
2
1
8
2
8
3
6
1
9
9
3
1
5
0
1
,
1
7
3
,
0
4
0
4
1
9
,
0
6
4
8
4
8
1
2
1
0
3
7
3
1
0
2
1
3
0
2
5
6
1
4
2
8
1
9
9
4
1
6
7
1
,
4
7
8
,
2
1
0
4
8
3
,
7
2
4
7
3
6
1
0
1
0
2
7
8
9
9
1
2
9
2
3
6
1
2
3
4
1
9
9
5
2
2
7
1
,
0
9
4
,
6
7
4
3
2
5
,
2
2
9
7
3
6
1
1
8
2
5
7
7
6
1
0
0
3
1
7
2
1
4
1
1
9
9
6
2
2
3
1
,
2
3
9
,
1
2
0
3
6
8
,
6
4
6
7
3
6
1
0
8
6
5
5
8
2
1
1
0
2
1
5
1
1
2
3
1
9
9
7
2
4
8
1
,
2
7
7
,
2
0
0
3
7
1
,
6
3
2
7
4
7
1
1
9
1
6
4
9
2
1
1
6
2
7
6
1
4
3
6
1
9
9
8
2
7
2
1
,
5
9
7
,
2
7
6
3
8
5
,
8
7
0
8
4
7
1
2
8
6
6
0
8
5
1
0
3
2
6
5
1
6
3
7
1
9
9
9
2
7
9
1
,
6
3
0
,
7
8
0
3
5
5
,
0
0
0
8
4
7
1
1
9
8
6
5
9
6
1
2
3
3
3
7
2
1
3
9
2
0
0
0
2
8
6
1
,
8
2
0
,
8
6
6
4
6
0
,
0
9
1
7
4
6
1
0
1
0
5
7
6
1
0
0
1
2
8
3
5
8
2
1
4
9
2
0
0
1
2
7
3
2
,
2
2
6
,
8
5
8
4
1
0
,
7
9
3
8
4
8
1
1
1
1
3
7
8
1
0
5
1
4
1
3
9
7
2
0
5
7
2
0
0
2
2
3
4
2
,
5
1
6
,
3
5
9
5
1
8
,
1
2
5
7
4
7
1
0
1
1
2
7
8
1
1
0
1
3
8
3
6
8
2
1
4
8
2
0
0
3
2
3
7
2
,
3
8
1
,
1
7
3
5
9
4
,
2
4
3
6
3
6
9
8
4
5
8
7
6
1
0
4
2
8
6
1
6
3
8
2
0
0
4
2
5
5
2
,
5
0
0
,
3
8
6
6
1
1
,
1
0
2
6
3
6
9
1
0
4
7
0
9
9
1
3
1
3
9
8
2
2
4
6
2
0
0
5
2
7
4
2
,
9
3
0
,
0
0
5
7
4
0
,
1
6
1
5
3
5
7
1
1
1
5
8
9
8
1
4
8
4
0
8
2
3
5
4
2
0
0
6
2
8
9
3
,
5
3
8
,
0
8
6
9
2
4
,
2
5
2
6
3
5
8
1
1
1
6
0
1
0
5
1
4
6
4
6
1
2
2
8
5
2
2
0
0
7
3
5
7
3
,
3
3
8
,
7
9
0
6
9
8
,
5
8
9
6
3
5
9
1
1
4
7
4
1
1
0
1
5
1
4
7
1
4
2
9
6
5
A
l
l
y
e
a
r
s
4
,
3
5
8
1
,
9
9
2
,
6
0
3
4
6
5
,
2
5
2
7
3
6
1
0
1
0
2
6
7
9
6
1
2
9
3
4
8
2
1
4
3
P
a
n
e
l
B
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
A
S
X
1
0
0
I
n
d
e
x
1
9
9
7
9
4
2
,
9
5
3
,
6
3
6
1
,
4
2
3
,
6
3
4
1
1
1
0
1
2
1
3
8
5
5
5
8
3
1
1
2
1
3
5
8
1
5
1
9
9
8
9
8
3
,
9
5
6
,
0
0
9
1
,
9
8
9
,
3
7
7
1
2
1
0
1
3
1
6
7
6
5
5
8
1
9
6
1
2
2
8
1
7
1
9
9
9
9
9
4
,
1
5
3
,
4
0
6
1
,
6
6
9
,
0
4
0
1
1
9
1
2
1
4
8
7
6
1
8
5
1
1
0
1
8
3
1
0
2
6
2
0
0
0
1
0
0
4
,
5
9
0
,
1
5
7
2
,
0
5
1
,
8
6
7
1
1
8
1
1
1
3
9
1
6
8
8
9
1
1
1
1
4
4
9
2
0
2
0
0
1
8
9
6
,
1
6
1
,
5
3
7
2
,
0
2
9
,
3
4
7
1
2
1
0
1
2
1
3
9
2
6
6
9
3
1
1
3
1
3
4
8
1
8
2
0
0
2
9
2
5
,
8
3
1
,
7
7
1
2
,
4
3
9
,
5
5
1
1
1
9
1
1
1
3
9
7
6
5
9
6
1
1
9
1
5
2
1
0
2
0
(
c
o
n
t
i
n
u
e
d
)
Table II.
Descriptive statistics
Accuracy
of analysts’
expectations
101
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
M
e
a
n
o
r
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
N
u
m
b
e
r
o
f
?
r
m
s
M
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
(
$
0
0
0
)
N
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
C
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
f
o
r
e
c
a
s
t
i
s
s
u
e
d
a
t
e
a
n
d
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
F
i
s
c
a
l
y
e
a
r
f
o
l
l
o
w
e
d
M
e
a
n
M
e
d
i
a
n
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
M
e
a
n
2
5
%
M
e
d
i
a
n
7
5
%
2
0
0
3
9
2
5
,
5
1
4
,
5
8
5
2
,
4
7
3
,
0
9
7
9
8
9
1
0
7
6
5
5
7
1
8
9
1
4
3
9
2
3
2
0
0
4
9
4
5
,
9
3
7
,
7
3
2
2
,
7
7
0
,
7
9
4
9
8
9
1
1
9
5
7
0
9
4
1
1
4
1
5
3
8
2
1
2
0
0
5
9
2
7
,
5
4
7
,
4
8
4
4
,
1
3
5
,
1
8
5
7
6
7
9
9
8
4
8
8
0
1
3
1
2
1
5
1
3
3
2
2
0
0
6
9
8
8
,
8
7
7
,
1
6
7
4
,
5
5
9
,
9
0
8
8
7
9
1
0
1
0
2
5
8
8
9
1
3
0
2
6
6
1
5
3
0
2
0
0
7
9
8
1
0
,
4
4
4
,
4
1
4
4
,
9
6
4
,
3
2
3
1
0
7
1
0
1
2
1
1
1
6
7
1
0
7
1
4
7
2
6
7
1
7
3
6
A
l
l
y
e
a
r
s
1
,
0
4
6
5
,
9
9
7
,
9
0
2
2
,
6
7
3
,
7
3
0
1
0
8
1
0
1
3
9
2
6
0
8
6
1
1
6
1
7
4
1
0
2
3
P
a
n
e
l
C
:
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
o
u
t
s
i
d
e
t
h
e
A
S
X
1
0
0
I
n
d
e
x
1
9
9
7
1
5
4
2
5
3
,
9
2
1
1
9
9
,
3
3
4
5
3
4
7
9
5
6
7
9
5
1
1
9
3
5
8
2
2
5
4
1
9
9
8
1
7
4
2
6
8
,
7
9
3
2
2
6
,
7
7
7
6
3
5
8
9
1
6
2
8
9
1
0
9
3
4
1
0
2
4
4
9
1
9
9
9
1
8
0
2
4
3
,
3
3
6
1
7
0
,
6
9
7
6
3
5
8
1
0
5
7
3
1
0
3
1
3
3
4
1
1
3
2
8
4
7
2
0
0
0
1
8
6
3
3
2
,
0
0
0
2
3
5
,
9
7
9
5
3
4
7
1
1
3
8
3
1
0
8
1
4
0
4
7
1
3
3
3
6
4
2
0
0
1
1
8
4
3
2
3
,
6
7
1
2
1
1
,
1
7
9
6
3
5
8
1
2
2
8
6
1
1
8
1
5
4
5
2
1
4
3
3
7
6
2
0
0
2
1
4
2
3
6
8
,
3
4
6
2
2
3
,
8
9
3
5
3
4
7
1
2
2
8
9
1
1
9
1
4
2
4
9
1
6
3
3
6
5
2
0
0
3
1
4
5
3
9
3
,
0
7
6
2
3
5
,
3
5
2
4
2
4
6
8
9
6
0
8
2
1
1
3
3
7
9
2
2
5
1
2
0
0
4
1
6
1
4
9
3
,
4
8
8
3
1
9
,
9
8
3
4
2
4
6
1
0
9
7
1
1
0
2
1
4
9
5
3
1
6
3
4
6
6
2
0
0
5
1
8
2
5
9
5
,
8
9
5
3
6
1
,
8
9
1
4
2
3
6
1
1
8
6
8
1
0
6
1
5
2
4
9
1
0
3
2
6
6
2
0
0
6
1
9
1
7
9
8
,
6
6
3
4
3
7
,
3
8
8
5
2
4
6
1
1
6
6
5
1
1
3
1
5
5
5
7
1
9
3
6
7
2
2
0
0
7
2
5
9
6
5
0
,
1
7
6
3
9
3
,
4
0
3
5
2
4
7
1
1
6
7
6
1
1
2
1
5
3
5
4
1
7
3
6
7
7
A
l
l
y
e
a
r
s
1
,
9
5
8
4
4
3
,
8
8
5
2
7
0
,
0
9
9
5
3
4
7
1
0
9
7
2
1
0
3
1
3
9
4
7
1
3
3
0
6
2
N
o
t
e
s
:
T
h
i
s
t
a
b
l
e
r
e
p
o
r
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
d
i
s
t
r
i
b
u
t
i
o
n
o
f
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
s
a
m
p
l
e
;
P
a
n
e
l
A
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
a
l
l
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
;
P
a
n
e
l
B
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
i
n
c
l
u
d
e
d
i
n
t
h
e
A
S
X
1
0
0
I
n
d
e
x
;
P
a
n
e
l
C
p
r
e
s
e
n
t
s
d
e
s
c
r
i
p
t
i
v
e
s
t
a
t
i
s
t
i
c
s
f
o
r
t
h
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
o
u
t
s
i
d
e
t
h
e
A
S
X
1
0
0
I
n
d
e
x
;
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
d
a
t
a
i
s
o
b
t
a
i
n
e
d
f
r
o
m
I
/
B
/
E
/
S
;
a
v
e
r
a
g
e
m
a
r
k
e
t
c
a
p
i
t
a
l
i
s
a
t
i
o
n
i
s
t
h
e
e
q
u
i
t
y
v
a
l
u
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
c
o
m
p
a
n
y
’
s
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
;
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
f
o
l
l
o
w
i
n
g
e
q
u
a
l
s
e
i
t
h
e
r
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
c
o
n
t
r
i
b
u
t
i
n
g
t
o
t
h
e
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
o
r
o
n
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
Table II.
ARJ
23,1
102
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Panel A reports analyst coverage statistics. The average number of analyst forecasts
included in the consensus is approximately seven. It is about one third of that reported
by Ke and Yu (2006) for US analyst coverage statistics, re?ecting relatively fewer
brokerage ?rms and analysts provide company coverage in the Australian market.
Panel A also presents descriptive statistics for the distribution of timeliness of the
consensus and the most recent forecast. Despite the last I/B/E/S consensus being
published in the month prior to the earnings announcement, the median age of the
consensus forecast is 96 calendar days before the earnings announcement. The median
age of the most recent forecast is 21 calendar days. The most recent forecast is
approximately 75 calendar days more recent than the median age of the consensus
forecast, suggesting that the most recent forecast should be more accurate if new
information has been incorporated into the forecast.
Consensus forecasts in the early years of the sample period, notably from 1987 to
1990, are on average 20 days staler than in more recent years. The inclusion of stale
forecasts in the consensus is likely to reduce its accuracy. In more recent years, the
relative timeliness of consensus forecasts is improving, consistent with previous
?ndings (Barron and Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005).
Panels B and C present descriptive statistics for the company-year observations
included in the ASX 100 Index and outside the ASX 100 Index, respectively. Partitioning
the sample into the ASX 100 companies and companies outside the Index reduces the
sample size to 3,004 observations because the constituent list for the ASX 100 Index is
unavailable in SIRCA prior to ?scal 1997. The ASX 100 companies have higher market
capitalisation, are covered by more analysts, and their forecasts are updated more timely,
as compared with companies outside the Index. The ASX100 companies have an average
(median) market capitalisation of $6 ($2.7) billion and are followed by ten analysts on
average. The median age of the consensus (most recent) forecast for these companies is 86
(ten) days. By comparison, companies outside the ASX 100 Index have an average
(median) market capitalisation of $0.4 ($0.3) billion and are covered by an average of ?ve
analysts. The median age of the consensus (most recent) forecast for these companies is
103 (30) days. The ASX 100 companies and companies outside the Index show different
company and forecast characteristics. These differences may have effects on forecast
accuracy. Further analysis is conducted in Section 6.2.
5. Evaluating forecast accuracy
5.1 Variable de?nitions
The absolute forecast error (AFE) is used to measure forecast accuracy:
AFE
jts
¼
A
jt
2F
jts
P
j;t21
ð1Þ
Following Richardson et al. (2004), the AFE
jts
is de?ned as the absolute value of the
difference between A
jt
, actual annual EPS of company j in year t, and F
jts
, the forecast
EPS using each of the alternative forecast measures, s, and is de?ated by company j’s
share price[14] 11 months before the earnings announcement month[15], P
j,t21
.
Each of the alternative forecast measures, denoted by s, is one of the following: the
mean consensus forecast (s ¼ mean), the median consensus forecast (s ¼ median) or
the most recent forecast (s ¼ mr). F
jtmean
, the mean consensus forecast, is the mean of
Accuracy
of analysts’
expectations
103
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
all the individual analyst forecasts available as at the publication date of the last
I/B/E/S consensus before the earnings announcement for company j in year t. F
jtmedian
,
the median consensus forecast, is the median of all the individual analyst forecasts
available as at the publication date of the last I/B/E/S consensus before the earnings
announcement for company j in year t. F
jtmr
, the most recent forecast, is the latest
individual analyst forecast EPS reported to I/B/E/S before the earnings announcement
for company j in year t[16].
To compare the accuracy of the consensus and the most recent forecast for each
year of the sample period, AFE
jts
is computed to measure forecast accuracy at the
company level for each year and then aggregate these results across companies:
MAFE
ts
¼
1
N
X
N
j¼1
AFE
jts
ð2Þ
For each forecast measure, s, the mean of the AFE
jts
in year t, MAFE
ts
, is averaged
across all available company observations ( j ¼ 1, . . . , N) in year t:
MAFE
s
¼
1
N
X
N
n¼1
1
T
X
T
t¼1
AFE
jts
( )
ð3Þ
For each forecast measure, s, the pooled mean of the AFE
jts
, MAFE
s
, is averaged across
years (t ¼ 1, . . . , T) for each company j, and then averaged across companies to
evaluate forecast accuracy at an aggregate level across years and companies.
MAFE
ts
and MAFE
s
, are calculated to compare the accuracy of each forecast
measure for each year and overall for the sample period. Signi?cance tests of the
differences in accuracy are used to test whether the consensus is more accurate than
the most recent forecast (H1).
The timeliness of the consensus forecast is measured by taking the average value of
timeliness of individual analyst forecasts included in the last consensus prior to the
earnings announcement, where the timeliness of an individual analyst forecast is
measured with references to the number of calendar days between the date of the
individual analyst forecast issued prior to the earnings announcement and the earnings
announcement date. The timeliness of the most recent forecast is measured with
references to the number of calendar days between the date of the most recent forecast
issued prior to the earnings announcement and the earnings announcement date.
The number of analysts following is either the number of analysts contributing to the
consensus forecast or one for the most recent forecast.
5.2 Forecast accuracy and pairwise differences in forecast accuracy
Table III reports the accuracy of the mean consensus, the median consensus and the
most recent forecast. The forecast accuracy is measured by the mean of the AFEs
across all available company-year observations for the year. For the 1987-1990 period,
the most recent forecast is more accurate than the consensus forecast. By comparison,
O’Brien (1988)’s 1975-1981 sample and Brown (1991)’s 1984-1988 sample show similar
effects. In the more recent period from 1991 to 2007, the results suggest that both the
mean and median consensus forecasts are more accurate than the most recent forecast.
The accuracy of the median (mean) consensus forecast outperforms that of the most
ARJ
23,1
104
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
A
v
e
r
a
g
e
A
F
E
(
%
)
t
-
s
t
a
t
i
s
t
i
c
f
o
r
p
a
i
r
w
i
s
e
t
e
s
t
o
f
d
i
f
f
e
r
e
n
c
e
s
F
i
s
c
a
l
y
e
a
r
N
u
m
b
e
r
o
f
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
M
e
a
n
c
o
n
s
e
n
s
u
s
M
e
d
i
a
n
c
o
n
s
e
n
s
u
s
M
o
s
t
r
e
c
e
n
t
M
e
a
n
–
m
e
d
i
a
n
M
e
a
n
–
m
o
s
t
r
e
c
e
n
t
M
e
d
i
a
n
–
M
o
s
t
r
e
c
e
n
t
1
9
8
7
2
1
1
.
7
1
2
1
.
6
8
4
1
.
3
9
8
0
.
2
5
0
.
5
7
0
.
5
3
1
9
8
8
6
7
1
.
4
2
5
1
.
2
4
6
0
.
8
1
6
2
.
2
2
*
*
2
.
5
8
*
*
1
.
7
9
*
1
9
8
9
1
3
5
1
.
0
9
2
0
.
9
5
0
0
.
8
9
1
2
.
8
7
*
*
*
1
.
8
2
*
0
.
5
3
1
9
9
0
1
0
9
1
.
1
5
6
0
.
9
5
8
0
.
6
9
5
2
.
3
9
*
*
2
.
7
4
*
*
*
1
.
6
6
*
1
9
9
1
1
1
9
0
.
9
3
7
0
.
5
9
4
0
.
8
2
2
3
.
8
4
*
*
*
0
.
8
8
2
2
.
0
2
*
*
1
9
9
2
1
3
6
0
.
9
8
4
0
.
9
7
6
1
.
1
7
9
0
.
1
9
2
1
.
8
8
*
2
1
.
9
8
*
*
1
9
9
3
1
5
0
0
.
8
6
6
0
.
8
2
4
0
.
9
5
3
1
.
6
2
2
0
.
9
7
2
1
.
4
7
1
9
9
4
1
6
7
0
.
7
1
3
0
.
6
8
7
0
.
7
6
3
1
.
6
8
*
2
0
.
7
4
2
1
.
2
1
1
9
9
5
2
2
7
0
.
8
2
3
0
.
8
0
3
0
.
8
8
2
1
.
4
2
2
1
.
0
5
2
1
.
3
7
1
9
9
6
2
2
3
0
.
7
6
8
0
.
7
3
0
0
.
7
0
9
2
.
0
2
*
*
1
.
0
7
0
.
3
8
1
9
9
7
2
4
8
0
.
6
5
5
0
.
6
7
4
0
.
7
7
6
2
0
.
8
0
2
3
.
0
6
*
*
*
2
3
.
0
1
*
*
*
1
9
9
8
2
7
2
0
.
7
9
8
0
.
7
4
1
0
.
7
9
3
2
.
2
6
*
*
0
.
0
9
2
0
.
8
4
1
9
9
9
2
7
9
0
.
9
8
8
0
.
9
8
0
0
.
9
4
9
0
.
3
4
0
.
7
0
0
.
5
8
2
0
0
0
2
8
6
1
.
1
7
7
1
.
1
8
1
1
.
2
0
7
2
0
.
2
4
2
0
.
5
5
2
0
.
4
7
2
0
0
1
2
7
3
1
.
3
3
5
1
.
3
3
0
1
.
4
2
2
0
.
2
9
2
1
.
4
0
2
1
.
5
7
2
0
0
2
2
3
4
1
.
0
4
0
1
.
0
3
2
1
.
2
1
0
0
.
3
2
2
2
.
6
7
*
*
*
2
2
.
6
6
*
*
*
2
0
0
3
2
3
7
0
.
9
8
7
1
.
0
2
0
1
.
1
5
5
2
1
.
7
6
*
2
2
.
4
8
*
*
2
2
.
0
0
*
*
2
0
0
4
2
5
5
0
.
8
9
3
0
.
8
2
4
0
.
9
8
0
2
.
0
0
*
*
2
1
.
4
1
2
2
.
7
2
*
*
*
2
0
0
5
2
7
4
0
.
8
5
1
0
.
8
2
0
0
.
9
9
1
1
.
5
1
2
2
.
2
5
*
*
2
2
.
6
6
*
*
*
2
0
0
6
2
8
9
0
.
7
4
2
0
.
7
0
6
0
.
8
1
9
1
.
4
8
2
1
.
5
0
2
2
.
2
1
*
*
2
0
0
7
3
5
7
0
.
7
6
6
0
.
7
1
1
1
.
0
0
9
2
.
6
5
*
*
*
2
4
.
0
8
*
*
*
2
4
.
7
2
*
*
*
A
l
l
y
e
a
r
s
4
,
3
5
8
0
.
9
2
7
0
.
8
8
5
0
.
9
7
5
6
.
3
9
*
*
*
2
2
.
9
9
*
*
*
2
5
.
6
8
*
*
*
N
o
t
e
s
:
S
t
a
t
i
s
t
i
c
a
l
s
i
g
n
i
?
c
a
n
c
e
a
t
t
h
e
*
0
.
1
0
,
*
*
0
.
0
5
a
n
d
*
*
*
0
.
0
1
l
e
v
e
l
s
(
t
w
o
-
t
a
i
l
t
e
s
t
)
,
r
e
s
p
e
c
t
i
v
e
l
y
;
f
o
r
e
c
a
s
t
a
c
c
u
r
a
c
y
i
s
m
e
a
s
u
r
e
d
b
y
t
h
e
A
F
E
;
A
F
E
i
s
d
e
?
n
e
d
a
s
t
h
e
a
b
s
o
l
u
t
e
v
a
l
u
e
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
a
c
t
u
a
l
E
P
S
a
n
d
t
h
e
f
o
r
e
c
a
s
t
E
P
S
d
e
?
a
t
e
d
b
y
s
h
a
r
e
p
r
i
c
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
;
a
v
e
r
a
g
e
A
F
E
i
s
t
h
e
m
e
a
n
o
f
t
h
e
A
F
E
s
a
c
r
o
s
s
a
l
l
a
v
a
i
l
a
b
l
e
c
o
m
p
a
n
y
-
y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
f
o
r
t
h
e
y
e
a
r
;
t
h
e
m
e
a
n
c
o
n
s
e
n
s
u
s
i
s
t
h
e
m
e
a
n
o
f
a
l
l
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
a
v
a
i
l
a
b
l
e
a
s
a
t
t
h
e
p
u
b
l
i
c
a
t
i
o
n
d
a
t
e
o
f
t
h
e
l
a
s
t
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
;
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
i
s
t
h
e
m
e
d
i
a
n
o
f
a
l
l
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
a
v
a
i
l
a
b
l
e
a
s
a
t
t
h
e
p
u
b
l
i
c
a
t
i
o
n
d
a
t
e
o
f
t
h
e
l
a
s
t
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
;
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
i
s
t
h
e
l
a
t
e
s
t
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
E
P
S
r
e
p
o
r
t
e
d
t
o
I
/
B
/
E
/
S
p
r
i
o
r
t
o
t
h
e
a
n
n
o
u
n
c
e
m
e
n
t
o
f
a
c
t
u
a
l
e
a
r
n
i
n
g
s
;
t
h
e
t
-
s
t
a
t
i
s
t
i
c
i
s
a
t
e
s
t
o
f
t
h
e
p
a
i
r
w
i
s
e
d
i
f
f
e
r
e
n
c
e
i
n
a
c
c
u
r
a
c
y
b
e
t
w
e
e
n
t
h
e
c
o
n
s
e
n
s
u
s
a
n
d
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
s
;
a
n
e
g
a
t
i
v
e
s
i
g
n
o
n
a
t
-
s
t
a
t
i
s
t
i
c
i
n
t
h
i
s
t
a
b
l
e
i
n
d
i
c
a
t
e
s
t
h
a
t
t
h
e
?
r
s
t
o
f
t
h
e
p
a
i
r
o
f
f
o
r
e
c
a
s
t
m
e
a
s
u
r
e
s
c
o
m
p
a
r
e
d
i
s
m
o
r
e
a
c
c
u
r
a
t
e
;
t
h
e
c
o
n
v
e
r
s
e
i
s
t
r
u
e
f
o
r
a
p
o
s
i
t
i
v
e
s
i
g
n
Table III.
Accuracy of the
consensus and the most
recent earnings forecasts
by year
Accuracy
of analysts’
expectations
105
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
recent forecast in 15 (13) out of these 17 most recent years. The accuracy of the median
consensus is greater than the mean consensus in 18 out of total 21 years.
Table III also reports results of statistical tests for differences in accuracy among
the mean consensus, the median consensus and the most recent forecast. A negative
sign on a t-statistic indicates that the ?rst of the pair of forecast measures compared is
more accurate. For example, in the ?scal year 2007, the t-statistic for the pairwise test
of differences in accuracy between the median consensus and the most recent forecast
is 24.72, which favours the median consensus, and is statistically signi?cant at the
0.01 level. The results con?rm that whilst in the late 1980s the most recent forecast is
signi?cantly more accurate than the consensus, the consensus forecast outperforms the
most recent forecast in more recent years. The results show that the median (mean)
consensus forecast dominates the most recent forecast where signi?cant differences
exist for nine (six) out of these 15 years[17].
In terms of economic signi?cance, Table III shows that the most recent forecast is on
average less accurate than the consensus by 0.09 per cent of share price[18]. For a
company with a market capitalisation of $2 billion, the average market capitalisation in
the sample, this translates into a most recent forecast that misses actual earnings by
$1.8 million relative to the consensus forecast[19].
6. Explanations for the greater accuracy of consensus forecasts
Following Ramnath et al. (2005), to examine whether the number of analysts following
and the timeliness of analysts’ forecasts explain the relative greater forecast accuracy
of consensus forecasts using a cross-sectional regression (H2), the following
cross-sectional regression is estimated:
AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ 1
jts
ð4Þ
where:
AFE
jts
¼ Is the absolute value of the median consensus[20] (s ¼ median)
or the most recent (s ¼ mr) forecast error de?ated by share
price 11 months before the earnings announcement month for
company j’s annual EPS in year t.
MEASURE
jts
¼ Is an indicator variable, coded one if AFE is sourced from the
median consensus forecast (s ¼ median); coded zero if AFE is
sourced from the most recent forecast (s ¼ mr) for company j
in year t.
ANALYST
jts
¼ Equals the number of analysts contributing to the consensus
forecast (s ¼ median) or one for the most recent forecast
(s ¼ mr) for company j in year t.
TIMELINESS
jts
¼ Equals the average value of timeliness of individual analyst
forecasts included in the last consensus prior to the earnings
announcement, where the timeliness of an individual analyst
forecast is measured by the number of calendar days between
the date of the individual analyst forecast issued prior to the
ARJ
23,1
106
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
earnings announcement and the earnings announcement date
for the median consensus forecast (s ¼ median) or the number
of calendar days between the date of the most recent forecast
issued prior to the earnings announcement and the earnings
announcement date for the most recent forecast (s ¼ mr) for
company j in year t.
Cross-sectional regression is estimated for each year t, respectively, and the mean of
the annual coef?cient estimates across the sample period is calculated.
Prior research demonstrates that the larger the number of analysts following, the
greater is the accuracy of consensus forecasts. The AFE is expected to decrease as
the number of analyst following increases, b
2
, 0. Increasing forecast accuracy is also
associated with the timeliness of analysts’ forecasts. The accuracy of analysts’ forecasts
improves when the earnings announcement date approaches. Hence, the AFE is
expected to decrease as the timeliness of analysts’ forecasts is shorter, b
3
. 0, for the
regression model (4).
Consistent with Ramnath et al. (2005) and Ke and Yu (2006), company-speci?c and
macroeconomic control variables are not included in the model. This may seem
unusual, however, the model examines determinants of the relative accuracy between
the consensus and the most recent forecast given the underlying economic conditions.
Both forecast measures are exposed to the same company-speci?c factors and
macroeconomic effects.
Since the aggregation value of consensus forecasts is expected to outweigh the timing
advantage of the most recent forecast, the forecasting superiority of the consensus
forecast over the most recent forecast is expected to be reduced after controlling for the
forecast aggregation and timing. In other words, the most recent forecast would
outperform the consensus forecast after controlling for these factors. b
1
, the coef?cient
on the MEASURE indicator variable, measuring the difference in accuracy of the
consensus forecast versus the most recent forecast, is expected to be positive (i.e. the
consensus forecast generates larger AFEs than the most recent forecast), b
1
. 0. b
0
,
the intercept, is expected to be positive, b
0
. 0, since the AFEis greater than or equal to
zero. The cross-sectional regression is estimated for each year t, respectively. Coef?cient
estimates are presented as the mean across the sample period following the Fama and
MacBeth (1973) procedure. The t-statistics and signi?cance levels are obtained under the
null that the mean of the coef?cient distributions across the sample period equals zero.
To control for size effects, and therefore, information environment effects on
forecast accuracy, the sample is partitioned into observations included in the ASX 100
Index and outside the Index to examine whether the same factors are associated with
the greater accuracy of consensus forecasts for these two groups.
Table IV reports the results of the regression model that examines whether the
number of analysts following and the timeliness of analysts’ forecasts explain the
greater accuracy of consensus forecasts for the overall sample period, the pre-1991 and
post-1991 periods, and observations included in the ASX 100 Index and outside the
Index. For the overall sample period, consistent with H2, the coef?cient on number of
analysts following is signi?cantly negative (20.0788, t ¼ 26.28), indicating forecast
accuracy increases with the number of analysts following. The coef?cient on timeliness
of analysts’ forecasts is close to zero (0.0004) and is not statistically signi?cant,
indicating that the timeliness of analysts’ forecasts does not contribute to increasing
Accuracy
of analysts’
expectations
107
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
:
A
F
E
V
a
r
i
a
b
l
e
P
r
e
d
i
c
t
e
d
s
i
g
n
O
v
e
r
a
l
l
p
e
r
i
o
d
P
r
e
-
1
9
9
1
p
e
r
i
o
d
P
o
s
t
-
1
9
9
1
p
e
r
i
o
d
A
S
X
1
0
0
O
u
t
s
i
d
e
A
S
X
1
0
0
I
n
t
e
r
c
e
p
t
þ
1
.
0
2
9
*
*
*
(
2
0
.
6
5
)
0
.
8
8
0
7
*
*
*
(
5
.
2
9
)
1
.
0
5
8
8
*
*
*
(
2
2
.
5
5
)
0
.
7
8
6
1
*
*
*
(
1
2
.
2
0
)
1
.
3
6
3
6
*
*
*
(
2
1
.
4
2
)
M
E
A
S
U
R
E
þ
0
.
3
2
8
4
*
*
*
(
6
.
3
4
)
0
.
3
6
1
9
(
2
.
2
4
)
0
.
3
2
0
5
*
*
*
(
5
.
8
9
)
0
.
6
6
6
9
*
*
*
(
4
.
2
9
)
0
.
3
5
2
1
*
*
*
(
6
.
8
5
)
A
N
A
L
Y
S
T
2
2
0
.
0
7
8
8
*
*
*
(
2
6
.
2
8
)
2
0
.
0
8
8
9
(
2
1
.
5
0
)
2
0
.
0
7
6
5
*
*
*
(
2
8
.
1
9
)
2
0
.
0
5
8
6
*
*
*
(
2
3
.
8
0
)
2
0
.
1
1
5
4
*
*
*
(
2
8
.
9
4
)
T
I
M
E
L
I
N
E
S
S
þ
0
.
0
0
0
4
(
0
.
7
6
)
0
.
0
0
2
8
(
1
.
2
9
)
2
0
.
0
0
0
1
(
2
0
.
2
5
)
2
0
.
0
0
3
*
*
(
2
2
.
8
4
)
2
0
.
0
0
0
8
(
2
1
.
0
8
)
N
u
m
b
e
r
o
f
o
b
s
e
r
v
a
t
i
o
n
s
4
,
3
5
8
3
3
2
4
,
0
2
6
1
,
0
4
6
1
,
9
5
8
A
d
j
u
s
t
e
d
R
2
(
%
)
1
.
9
8
1
.
0
1
2
.
2
2
.
1
8
1
.
9
0
N
o
t
e
s
:
S
t
a
t
i
s
t
i
c
a
l
s
i
g
n
i
?
c
a
n
c
e
a
t
t
h
e
*
0
.
1
0
,
*
*
0
.
0
5
a
n
d
*
*
*
0
.
0
1
l
e
v
e
l
s
(
t
w
o
-
t
a
i
l
t
e
s
t
)
,
r
e
s
p
e
c
t
i
v
e
l
y
;
A
F
E
j
t
s
¼
b
0
þ
b
1
ð
M
E
A
S
U
R
E
j
t
s
Þ
þ
b
2
ð
A
N
A
L
Y
S
T
j
t
s
Þ
þ
b
3
ð
T
I
M
E
L
I
N
E
S
S
j
t
s
Þ
þ
1
j
t
s
;
A
F
E
j
t
s
i
s
t
h
e
a
b
s
o
l
u
t
e
v
a
l
u
e
o
f
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
(
s
¼
m
e
d
i
a
n
)
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
(
s
¼
m
r
)
f
o
r
e
c
a
s
t
e
r
r
o
r
d
e
?
a
t
e
d
b
y
s
h
a
r
e
p
r
i
c
e
1
1
m
o
n
t
h
s
b
e
f
o
r
e
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
m
o
n
t
h
f
o
r
c
o
m
p
a
n
y
j
’
s
a
n
n
u
a
l
E
P
S
i
n
y
e
a
r
t
;
t
h
i
s
v
a
r
i
a
b
l
e
i
s
m
u
l
t
i
p
l
i
e
d
b
y
1
0
0
,
s
o
t
h
e
c
o
e
f
?
c
i
e
n
t
e
s
t
i
m
a
t
e
s
a
r
e
a
s
a
p
e
r
c
e
n
t
a
g
e
o
f
s
h
a
r
e
p
r
i
c
e
;
M
E
A
S
U
R
E
j
t
s
i
s
a
n
i
n
d
i
c
a
t
o
r
v
a
r
i
a
b
l
e
,
c
o
d
e
d
o
n
e
i
f
A
F
E
i
s
s
o
u
r
c
e
d
f
r
o
m
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
;
c
o
d
e
d
z
e
r
o
i
f
A
F
E
i
s
s
o
u
r
c
e
d
f
r
o
m
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
A
N
A
L
Y
S
T
j
t
s
e
q
u
a
l
s
t
h
e
n
u
m
b
e
r
o
f
a
n
a
l
y
s
t
s
c
o
n
t
r
i
b
u
t
i
n
g
t
o
t
h
e
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
o
r
o
n
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
s
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
T
I
M
E
L
I
N
E
S
S
j
t
s
e
q
u
a
l
s
t
h
e
a
v
e
r
a
g
e
v
a
l
u
e
o
f
t
i
m
e
l
i
n
e
s
s
o
f
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
s
i
n
c
l
u
d
e
d
i
n
t
h
e
l
a
s
t
c
o
n
s
e
n
s
u
s
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
,
w
h
e
r
e
t
h
e
t
i
m
e
l
i
n
e
s
s
o
f
a
n
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
i
s
m
e
a
s
u
r
e
d
b
y
t
h
e
n
u
m
b
e
r
o
f
c
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
t
h
e
d
a
t
e
o
f
t
h
e
i
n
d
i
v
i
d
u
a
l
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
i
s
s
u
e
d
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
a
n
d
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
f
o
r
t
h
e
m
e
d
i
a
n
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
(
s
¼
m
e
d
i
a
n
)
o
r
t
h
e
n
u
m
b
e
r
o
f
c
a
l
e
n
d
a
r
d
a
y
s
b
e
t
w
e
e
n
t
h
e
d
a
t
e
o
f
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
i
s
s
u
e
d
p
r
i
o
r
t
o
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
a
n
d
t
h
e
e
a
r
n
i
n
g
s
a
n
n
o
u
n
c
e
m
e
n
t
d
a
t
e
f
o
r
t
h
e
m
o
s
t
r
e
c
e
n
t
f
o
r
e
c
a
s
t
(
s
¼
m
r
)
f
o
r
c
o
m
p
a
n
y
j
i
n
y
e
a
r
t
;
t
h
e
r
e
g
r
e
s
s
i
o
n
i
s
e
s
t
i
m
a
t
e
d
f
o
r
e
a
c
h
y
e
a
r
t
,
r
e
s
p
e
c
t
i
v
e
l
y
;
c
o
e
f
?
c
i
e
n
t
e
s
t
i
m
a
t
e
s
a
r
e
p
r
e
s
e
n
t
e
d
a
s
t
h
e
m
e
a
n
a
c
r
o
s
s
t
h
e
s
a
m
p
l
e
p
e
r
i
o
d
f
o
l
l
o
w
i
n
g
t
h
e
F
a
m
a
a
n
d
M
a
c
B
e
t
h
(
1
9
7
3
)
p
r
o
c
e
d
u
r
e
Table IV.
Association between the
number of analyst
following and the greater
forecast accuracy of the
consensus forecast
ARJ
23,1
108
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast accuracy. The coef?cient on the MEASURE indicator variable that measures
the difference in accuracy of the consensus forecast (coded one) versus the most recent
forecast (coded zero) is signi?cant and positive (0.3284, t ¼ 6.34), after controlling for
forecast aggregation and timing. This result indicates that the consensus forecast
generates larger AFEs and is less accurate than the most recent forecast after
controlling for the effects of the number of analysts following and the timeliness of
analysts’ forecasts. It also suggests that the aggregation value of the consensus
forecast outweighs the timing advantage of the most recent forecast, and the number of
analysts following largely explains the forecasting superiority of the consensus
forecast. The model’s explanatory power is low (R
2
¼ 1.98 per cent), but consistent
with previous ?ndings (Ramnath et al., 2005). The results from the post-1991 period are
consistent with the results for the overall sample period.
Both sets of regression results for the ASX 100 companies and companies outside
the Index show that the aggregating value of the consensus outweighs the timing
advantage of the most recent forecast. The coef?cients on number of analysts
following for both subsamples are signi?cantly negative (20.0586, t ¼ 23.80 for the
ASX 100 companies; 20.1154, t ¼ 28.94 for companies outside the index). The
association between forecast accuracy and the number of analyst following is stronger
for the companies outside the ASX 100 Index because these companies are usually
covered by fewer analysts as compared to the ASX 100 companies. The incremental
analyst coverage more effectively purges the individual analyst’s idiosyncratic error.
Overall, the results suggest that the consensus forecast is more accurate than the
most recent forecast. The forecast superiority of the consensus forecast can be
attributed to the aggregating value of the consensus forecast outweighing the timing
advantage of the most recent forecast. These results show that the greater accuracy of
the consensus forecast is due to the number of analysts following and contributing to
the consensus. The results are robust to using the mean consensus rather than the
median consensus, constructed by using individual analyst forecasts. They are also
robust to using the I/B/E/S summary consensus measure.
7. Robustness checks
7.1 Excluding the most recent forecast from the consensus
If the most recent forecast is issued before the formation date of the last consensus prior
to the earnings announcement, it will be included in the calculation of the consensus
forecast. In the ?nal sample, 2,888 observations (66 per cent of total observations) have
the most recent forecast included in the consensus for the corresponding company-year.
Given this, the question arises whether, the relative timeliness of the consensus is
attributed to its increasing accuracy together with its aggregation value. In this section,
the relative accuracy between the consensus and the most recent forecast is re-examined
using a subsample which excludes the most recent forecast from the calculation of the
consensus.
Column (1) of Table V repeats the analysis in Table IV. The control variables for
number of analysts following and timeliness are the same as those described in the
regression model (4). The descriptive statistics (not tabulated) show that the median
age of the consensus excluding the most recent forecast is 104 calendar days before the
earnings announcement, suggesting that timeliness of consensus forecasts is improved
by including recently-updated forecasts.
Accuracy
of analysts’
expectations
109
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The results in column (1) of Table V are similar to those in Table IV, showing that
the coef?cient on number of analysts following is negative and statistically signi?cant;
whereas, the coef?cient on timeliness of analysts’ forecasts is close to zero and
statistically insigni?cant, indicating that the number of analysts following explains the
consensus forecasting advantage. The results demonstrate that, conditional on relative
timeliness of consensus forecasts, there are gains in accuracy from aggregating to
reduce idiosyncratic error.
7.2 Controlling for company and analyst characteristics
Consistent with Ramnath et al. (2005) and Ke and Yu (2006), company-speci?c
variables are not included in the model when comparing the accuracy of the two
Dependent variable: AFE
Variable Predicted sign (1) (2)
Intercept þ 0.9797
* * *
(21.81) 4.3494
* * *
(6.42)
MEASURE þ 0.4127
* * *
(4.27) 0.1610
* *
(2.28)
ANALYST 2 0.0775
* * *
(26.71) 20.0247
* * *
(23.44)
TIMELINESS þ 20.0003
*
(20.29) 20.0013 (21.91)
PASTAFE þ 23.3971
* * *
(7.06)
LNSIZE 2 20.1802
* * *
(25.35)
ASX100 2 0.1264 (1.67)
TACC 2 20.8971
* * *
(23.52)
Number of observations 3,766 2,346
Adjusted R
2
(%) 2.35 11.44
Notes: Statistical signi?cance at the
*
0.10,
* *
0.05 and
* * *
0.01 levels (two-tail test),
respectively; AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ 1
jts
ð1Þ
for the subsample excluding the most recent forecast from the calculation of the consensus
AFE
jts
¼ b
0
þ b
1
ðMEASURE
jts
Þ þ b
2
ðANALYST
jts
Þ þ b
3
ðTIMELINESS
jts
Þ þ b
4
ðPASTAFE
jts
Þ þ
b
5
ðLNSIZE
jts
Þ þ b
6
ðASX100
jts
Þ þ b
7
ðTACC
jts
Þ þ 1
jts
ð2Þ for the subsample including additional
variables: past forecast accuracy, ?rm size, whether the company observation is included in the ASX 100
Index, and the level of accruals AFE
jts
is the absolute value of the median consensus (s ¼ median) or the
most recent (s ¼ mr) forecast error de?ated by share price 11 months before the earnings announcement
month for company j’s annual EPS in year t; this variable is multiplied by 100, so, the coef?cient estimates
are as a percentage of share price; MEASURE
jts
is an indicator variable, coded one if AFEis sourced from
the median consensus forecast (s ¼ median); coded zero if AFE is sourced from the most recent forecast
(s ¼ mr) for company j in year t; ANALYST
jts
equals the number of analysts contributing to the
consensus forecast (s ¼ median) or one for the most recent forecast (s ¼ mr) for company j in year t;
TIMELINESS
jts
equals the average value of timeliness of individual analyst forecasts included in the last
consensus prior to the earnings announcement, where the timeliness of an individual analyst forecast is
measured by the number of calendar days between the date of the individual analyst forecast issued prior
to the earnings announcement and the earnings announcement date for the median consensus forecast
(s ¼ median) or the number of calendar days between the date of the most recent forecast issued prior to
the earnings announcement and the earnings announcement date for the most recent forecast (s ¼ mr) for
company j in year t; PASTAFE
jts
is the AFE
jts
for the prior year; LNSIZE
jts
is the log market capitalisation
11 months before the earnings announcement month for company j in year t; ASX100
jts
is an indicator
variable, coded one if company j is included in the ASX100 Index; coded zero if company is outside the
ASX100 Index in year t; TACC
jts
equals the difference between net income and cash ?owfromoperations
for company j in year t; the regression is estimated for each year t, respectively; coef?cient estimates are
presented as the mean across the sample period following the Fama and MacBeth (1973) procedure
Table V.
Association between the
number of analyst
following and the greater
forecast accuracy of the
consensus forecast
ARJ
23,1
110
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
forecast measures. That is, the basic model compares the accuracy of the two measures
irrespective of the source of the differences in accuracy. To ensure the results are
robust to the inclusion of other factors associated with the accuracy of analysts’
forecasts, an additional set of variables are included in the analysis: past forecast
accuracy, ?rm size, whether or not the company observation is included in the ASX 100
Index, and the level of accruals.
Lim (2001) and Brown (2001b) show that analysts’ forecast accuracy is signi?cantly
positively correlated with their past forecast accuracy. Brown (2001b) suggests that
analysts’ past forecasting performance has a greater predictive power for forecast
accuracy than all analyst characteristics combined. Accordingly, a control for past
forecast accuracy to proxy for analyst characteristics is included. Since high-reputation
analysts provide more accurate and timely forecasts and have a greater impact on share
prices (Stickel, 1992; Brown et al., 2007), past forecast accuracy is also used to proxy for
the reputation of analysts.
Brown et al. (1999) and Lim(2001) ?nd that ?rmsize is negatively related to analysts’
forecast errors, indicating that the general information environment is likely to be
richer for large companies resulting in analysts issuing more accurate forecasts for
them. An indicator variable to control for the possible different information
environment for companies included in the ASX 100 Index as compared to companies
outside the index is also incorporated (Chan et al., 2006; Brown et al., 2007).
A company’s earnings management behaviour may affect quality of the reported
earnings and hence analysts’ forecast accuracy. Since managers perceive analysts as
one of the most important groups in?uencing the share price of their companies
(Graham et al., 2005), they could use accruals, or issue management earnings guidance
to manage earnings in an attempt to meet or beat analysts’ forecasts (Habib and
Hossain, 2008). For robustness, a control for total accruals as a proxy for the quality of
the reported earnings is incorporated into the regression.
Imposing the requirement that past forecast data, the constituent list for the ASX
100 Index, and accruals data are available reduces the sample size to 2,346
observations. Consistent with prior research, column (2) of Table V illustrates that
analysts’ forecast accuracy is signi?cantly positively correlated with their past forecast
accuracy and increases with ?rm size. Although it is in the opposite direction of the
predicted sign and is statistically insigni?cant, the coef?cient on index is signi?cantly
negative after the ?rm size is dropped from the regression (not tabulated); this indicates
that ?rm size and the index are highly correlated. Analysts’ forecast accuracy is
signi?cantly negatively correlated with total accruals, suggesting that earnings
management has an effect on analysts’ forecast accuracy. Adjusted R
2
increases to 11
per cent. The results remain robust to the inclusion of additional sets of variables. That
is, the number of analysts following continues to explain the greater accuracy of the
consensus forecast.
8. Conclusions
This study provides direct evidence of the accuracy of the consensus forecast versus
the most recent forecast, as measures of the market’s earnings expectations prior to
earnings announcements in the Australian market by examining 4,358 company-year
annual analyst forecasts between 1987 and 2007. Consistent with US evidence (O’Brien,
1988; Brown, 1991) the results suggest that in the late 1980s there is some evidence that
Accuracy
of analysts’
expectations
111
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the most recent forecast is more accurate than the consensus. This is, however, the
results in more recent years show that the consensus forecast is more accurate than
the most recent forecast. For the sample period from 1991 to 2007, the accuracy of the
consensus forecast consistently outperforms that of the most recent forecast in 15 out
of the 17 years; statistically signi?cant differences are shown for nine out of those 15
years. These results supports the ?ndings reported by recent US studies (Barron and
Stuerke, 1998; Brown, 2001a; Ramnath et al., 2005) and is consistent with the improving
timeliness of forecasts included in consensus forecasts. The forecasting superiority of
the consensus forecast can be attributed to the number of analysts following and
contributing to the consensus. The aggregation value of the consensus outweighs the
small timing advantage of the most recent forecast over the short forecast horizon
examined in this study.
This study contributes to prior research on Australian analysts’ forecasts by
providing evidence on the importance of diversifying idiosyncratic individual error
across analyst forecasts in the consensus forecast, in a non-US setting with relatively
thin analyst coverage and a different disclosure environment. It also con?rms the
market practitioners’ views as evidenced by press reports, that the consensus forecast
is a better measure of the market’s earnings expectations.
Future research could examine how analysts change their forecasting behaviour to
maintain their forecast accuracy in an environment of increased regulation over the
dissemination of company information. Future research might also pro?tably consider
whether consensus forecasts can be improved by forming a consensus based on
forecasts of a subset of highly skilled analysts or analysts possessing certain attributes.
Notes
1. There is an alternative approach to derive a consensus forecast which is to vary
component-forecast weights as a function of the expected accuracy of each forecast such as
forecast age, broker size and analyst experience (Kim et al., 2001; Brown and Mohd, 2003;
Butler et al., 2007). Since this approach is not widely used in the investment community
or academic research, this study focuses on the consensus forecast de?ned as the mean or
median of outstanding individual analyst forecasts at any point in time.
2. For example, Brown et al. (1999) examine properties of analysts’ forecasts in different
corporate disclosure environments. Beekes and Brown (2006) use analyst forecast accuracy
as one of the indicators for the informativeness of a ?rm’s disclosure to investigate the
association between a listed Australian ?rm’s corporate governance quality and its
disclosure and the market response. Habib and Hossain (2008) use analysts’ forecasts to
examine whether Australian managers manage earnings in an attempt to meet or beat
analysts’ forecasts.
3. For example, a search using text terms of “consensus forecast(s) or consensus estimate(s)”
?nds 800 articles in Australian key newspapers from 1 January 2007 to 31 December 2008
while a search using text terms of “most recent analyst forecast or most recent analyst
expectations” ?nds two articles in Factiva database.
4. An example of prior research relating to analysts’ forecasts in Australia is Jackson (2005).
Jackson presents evidence that analysts build their reputations by providing accurate
forecasts, and that high-reputation analysts generate more trading volume. Recent studies
examine share price responses to Australian analysts’ research quality and their stock
recommendations. For example, Brown et al. (2007) study the market response to initiating
ARJ
23,1
112
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
recommendations made by analysts in terms of returns and share price responses. Aitken
et al. (2008) examine the relation between analysts’ research quality and price discovery.
5. Barron and Stuerke (1998) note improvements in the integrity of the I/B/E/S database. Brown
(2001a) documents improvement over time in the accuracy of I/B/E/S forecasts. Ramnath
et al. (2005) ?nd increased reliability of I/B/E/S forecasts. The increasing competition
between forecast data providers may lead to these improvements. For example, First Call
Corporation began to provide analyst forecast data in the early 1900s (Brown, 2001a;
Ramnath et al., 2005).
6. This argument appears to assume, however, that the few analysts who update their forecasts
have access to information that is not available to the majority of analysts who choose not to
update their forecasts, all else being equal. It is unlikely that selective brie?ngs are
widespread given the disclosure environment brie?y discussed in the following section. Any
ability of individual analysts to outperform the consensus close to the earnings
announcement does however raise interesting questions regarding the source of any
superior performance, and hence the need to examine previous results using US data within
other regulatory environments.
7. Using I/B/E/S data for US companies, Ke and Yu (2006) report that 21.07 analysts on average
cover a company during 1983-2000. Barron and Stuerke (1998) show that the average
number of analyst following from 1990 to 1994 is about 16 in their sample. In this study, the
average number of analysts per company is seven.
8. While the policy is stated in terms of all forecasts, I/B/E/S does drop off stale forecasts when
calculating its monthly summary statistics. I/B/E/S reports the number of analysts’ forecasts
included in the consensus and this number of forecasts is used to identify the forecasts
assumed to be included in the consensus in this study.
9. For example, if I/B/E/S reports that 12 individual forecasts are included in the calculation of
its summary consensus, then the most recent forecasts issued by 12 individual analysts as
of the publication date of the I/B/E/S summary report will be used in the computation of
forecast statistics.
10. This study is interested in mimicking the I/B/E/S consensus forecast that is broadly
available to users on a monthly basis, rather than in creating a superior consensus.
The reconstruction of the I/B/E/S consensus is necessary as both the aggregated value of the
consensus and the attributes of individual forecasts that comprise the consensus are
required by the study.
11. Earnings announcement dates are sourced from I/B/E/S between 1987 and 1992 because they
are not available in the SIRCA database. Of the sample, observations with earnings
announcement dates differing by more than one day between SIRCA and I/B/E/S are less
than 10 per cent of observations between 2003 and 2007, more than 10 per cent but less
than 20 per cent of observations in 1999, 2000 and 2002, more than 20 per cent but less than
30 per cent of observations in 2001, and the discrepancies increase substantially before 1999
(between 1993 and 1998). Earnings announcement dates reported by I/B/E/S tend to be later
than those reported by SIRCA. Earnings announcement dates reported in SIRCA tend to
correspond with the ASX announcement dates wherever available on the ASX web site.
12. If a company is traded on multinational stock exchanges and followed by multinational
analysts, only Australian analyst forecasts are included in the sample.
13. As expected, the sample re?ects companies with analyst coverage and not all companies on
the ASX are followed by analysts. The results must be interpreted with respect to this
limitation.
Accuracy
of analysts’
expectations
113
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
14. Brown (1996) argues the share price is the appropriate de?ator to use when valuing
companies because the primary use of analysts’ earnings forecasts for security analysis is to
make investment decisions.
15. For companies listed less than 11 months, their forecast errors are de?ated by the ?rst listed
month closing share prices.
16. Consistent with Bartov et al. (2002) and Brown and Caylor (2005), the average value of the
analysts’ forecasts is used if more than one analyst forecast is issued on the most recent day.
17. The qualitative results remain unchanged when signed forecast errors are used to measure
forecast accuracy.
18. The difference between the average AFE for the most recent forecast (0.885 per cent) and the
average AFE for the median consensus (0.975 per cent) for the full sample is 0.09 per cent of
share price.
19. The average market capitalisation ($2 billion) multiplied by 0.09 per cent is $1.8 million.
20. The same conclusion is reached if the mean consensus forecast is used.
References
Aitken, M., Almeida, N., Harris, F.H.D. and McInish, T.H. (2008), “Financial analysts and price
discovery”, Accounting & Finance, Vol. 48, pp. 1-24.
Barron, O.E. and Stuerke, P. (1998), “Dispersion in analysts’ earnings forecasts as a measure of
uncertainty”, Journal of Accounting, Auditing & Finance, Vol. 13, pp. 243-68.
Barron, O.E., Byard, D. and Yu, Y. (2008), “Earnings surprises that motivate analysts to reduce
average forecast error”, The Accounting Review, Vol. 83, pp. 303-25.
Bartov, E., Givoly, D. and Hayn, C. (2002), “The rewards to meeting or beating earnings
expectations”, Journal of Accounting and Economics, Vol. 33, pp. 173-204.
Beekes, W. and Brown, P. (2006), “Do better governed Australian ?rms make more informative
disclosures?”, Journal of Business Finance & Accounting, Vol. 33, pp. 422-50.
Brown, L.D. (1991), “Forecast selection when all forecasts are not equally recent”, International
Journal of Forecasting, Vol. 7, pp. 349-56.
Brown, L.D. (1993), “Earnings forecasting research: its implications for capital markets
research”, International Journal of Forecasting, Vol. 9, pp. 295-320.
Brown, L.D. (1996), “Analyst forecasting errors and their implications for security analysis:
an alternative perspective”, Financial Analysts Journal, Vol. 52, pp. 40-7.
Brown, L.D. (2001a), “A temporal analysis of earnings surprises: pro?ts versus losses”, Journal of
Accounting Research, Vol. 39, pp. 221-41.
Brown, L.D. (2001b), “How important is past analyst forecast accuracy?”, Financial Analysts
Journal, Vol. 57, p. 44.
Brown, L.D. and Caylor, M.L. (2005), “A temporal analysis of quarterly earnings thresholds:
propensities and valuation consequences”, The Accounting Review, Vol. 80, pp. 423-40.
Brown, L.D. and Kim, K.J. (1991), “Timely aggregate analyst forecasts as better proxies for
market earnings expectations”, Journal of Accounting Research, Vol. 29, pp. 382-5.
Brown, L.D. and Mohd, E. (2003), “The predictive value of analyst characteristics”, Journal of
Accounting, Auditing & Finance, Vol. 18, pp. 625-47.
Brown, L.D. and Rozeff, M.S. (1978), “The superiority of analyst forecasts as measures of
expectations: evidence from earnings”, Journal of Finance, Vol. 33, pp. 1-16.
ARJ
23,1
114
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Brown, L.D., Hagerman, R.L., Grif?n, P.A. and Zmijewski, M.E. (1987a), “An evaluation of
alternative proxies for the market’s assessment of unexpected earnings”, Journal of
Accounting and Economics, Vol. 9, pp. 159-93.
Brown, L.D., Hagerman, R.L., Grif?n, P.A. and Zmijewski, M.E. (1987b), “Security analyst
superiority relative to univariate time-series models in forecasting quarterly earnings”,
Journal of Accounting and Economics, Vol. 9, pp. 61-87.
Brown, P., Taylor, S.L. and Walter, T.S. (1999), “The impact of statutory sanctions on the level
and information content of voluntary corporate disclosure”, Abacus, Vol. 35, pp. 138-62.
Brown, R., Chan, H.W.H. and Ho, Y.K. (2007), “Initiating coverage, broker reputation and
management earnings forecasts in Australia”, Accounting & Finance, Vol. 47, pp. 401-21.
Butler, M.B., Kraft, A.G. and Markov, S. (2007), “On the weighting of individual analyst forecasts
in the consensus”, working paper, Emory University, Atlanta, GA.
Chan, H.W.H., Brown, R. and Ho, Y.K. (2006), “Initiation of brokers’ recommendations, market
predictors and stock returns”, Journal of Multinational Financial Management, Vol. 16,
pp. 213-31.
Clement, M.B. (1999), “Analyst forecast accuracy: do ability, resources, and portfolio complexity
matter?”, Journal of Accounting and Economics, Vol. 27, pp. 285-303.
Clement, M.B. and Tse, S.Y. (2005), “Financial analyst characteristics and herding behavior in
forecasting”, Journal of Finance, Vol. 60, pp. 307-41.
Fama, E.F. and MacBeth, J.D. (1973), “Risk, return, and equilibrium: empirical tests”, Journal of
Political Economy, Vol. 81, pp. 607-36.
Fried, D. and Givoly, D. (1982), “Financial analysts’ forecasts of earnings: a better surrogate for
market expectations”, Journal of Accounting and Economics, Vol. 4, pp. 85-107.
Graham, J.R., Harvey, C.R. and Rajgopal, S. (2005), “The economic implications of corporate
?nancial reporting”, Journal of Accounting and Economics, Vol. 40, pp. 3-73.
Habib, A. and Hossain, M. (2008), “Do managers manage earnings to ‘just meet or beat’ analyst
forecasts? Evidence from Australia”, Journal of International Accounting, Auditing and
Taxation, Vol. 17, pp. 79-91.
Ivkovic, Z. and Jegadeesh, N. (2004), “The timing and value of forecast and recommendation
revisions”, Journal of Financial Economics, Vol. 73, pp. 433-63.
Jackson, A.R. (2005), “Trade generation, reputation, and sell-side analysts”, Journal of Finance,
Vol. 60, pp. 673-717.
Jacob, J., Lys, T.Z. and Neale, M.A. (1999), “Expertise in forecasting performance of security
analysts”, Journal of Accounting and Economics, Vol. 28, pp. 51-82.
Ke, B. and Yu, Y. (2006), “The effect of issuing biased earnings forecasts on analysts’ access to
management and survival”, Journal of Accounting Research, Vol. 44, pp. 965-99.
Kim, O., Lim, S.C. and Shaw, K.W. (2001), “The inef?ciency of the mean analyst forecast as a
summary forecast of earnings”, Journal of Accounting Research, Vol. 39, pp. 329-35.
Kothari, S.P. (2001), “Capital markets research in accounting”, Journal of Accounting and
Economics, Vol. 31, pp. 105-231.
Lim, T. (2001), “Rationality and analysts’ forecast bias”, Journal of Finance, Vol. 56, pp. 369-85.
Mikhail, M.B., WaIther, B.R. and Willis, R.H. (1999), “Does forecast accuracy matter to security
analysts?”, The Accounting Review, Vol. 74, pp. 185-200.
O’Brien, P.C. (1988), “Analysts’ forecasts as earnings expectations”, Journal of Accounting and
Economics, Vol. 10, pp. 53-83.
Accuracy
of analysts’
expectations
115
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
O’Brien, P.C. (1990), “Forecast accuracy of individual analysts in nine industries”, Journal of
Accounting Research, Vol. 28, pp. 286-304.
Ramnath, S., Rock, S. and Shane, P. (2005), “Value line and I/B/E/S earnings forecasts”,
International Journal of Forecasting, Vol. 21, pp. 185-98.
Richardson, S., Teoh, S.H. and Wysocki, P.D. (2004), “The walk-down to beatable analyst
forecasts: the role of equity issuance and insider trading incentives”, Contemporary
Accounting Research, Vol. 21, pp. 885-924.
Stickel, S.E. (1992), “Reputation and performance among security analysts”, The Journal of
Finance, Vol. 47, pp. 1811-36.
Corresponding author
Xiaomeng Chen can be contacted at: [email protected]
ARJ
23,1
116
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
This article has been cited by:
1. Rafael Moreira Antônio, Fabiano Guasti Lima, Tabajara Pimenta Junior. 2015. Stock recommendations
and investment portfolio formation: A study in the Brazilian market. Contaduría y Administración 60,
874-892. [CrossRef]
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
:
1
0
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_174750723.pdf