The Expectations Management Game: Do Analysts Act Independently of
Explicit Management Earnings Guidance?
Julie Cotter
University of Southern Queensland
Toowoomba Queensland 4350, Australia
E-mail: cotter@usq.edu.au
A. Irem Tuna
University of Michigan Business School
701 Tappan Street, Ann Arbor, Michigan, 48109-1234, USA
E-mail: atuna@umich.edu
Peter D. Wysocki
University of Michigan Business School
701 Tappan Street, Ann Arbor, Michigan, 48109-1234, USA
E-mail: wysockip@umich.edu
June, 2000 - Preliminary
Abstract
The financial press and recent academic research have made claims that firms are guiding
security analysts toward "beatable" earnings forecasts. This paper re-examines these claims by
investigating security analysts' reactions to explicit management earnings guidance. We use a
large sample of quarterly management earnings forecasts to examine the content of management
guidance and the timing, extent and independence of analysts' reactions to this guidance. Our
empirical analysis shows that (a) management guidance tends to be overly pessimistic relative to
actual earnings outcomes, and (b) analysts react quickly to management guidance that contains
new information. However, we find that analysts do not simply "parrot" the information
provided by management. Analysts act independently and adjust their forecasts to compensate
for bias in management's earnings guidance. Further, while analysts tend to shift from optimism
to pessimism in response to ‘bad news’ management guidance, the opposite is true for ‘good
Key Words: Analysts, Expectations management, Management earnings forecasts.
_________
We would like to thank Stan Levine at First Call for providing the analyst and management forecast data and for
helpful discussions. The comments of Andreas Gintschel and Scott Richardson are gratefully acknowledged.
1. Introduction
Recent research has argued that, during the 1990's, firms have successfully engaged in
earnings manipulation and expectations management games that allow them to consistently beat
analysts' earnings predictions. For example, Brown (1999), Brown and Higgins (1999), and
Burgstahler and Eames (1999) provide indirect evidence of earnings manipulation that allows
firms to beat analysts' targets. On the other hand, Matsumoto (1999) and Richardson, Teoh and
Wysocki (1999) provide indirect evidence that firms appear to be guiding analysts toward
"beatable" earnings forecasts. In particular, these studies document an apparent shift from
systematic analyst optimism at the beginning of a fiscal period to systematic analyst pessimism
just prior to an earnings announcement. These findings have been interpreted as prima facie
evidence that firms opportunistically guide analysts toward beatable targets. However, these
studies have not been able to provide direct evidence of actual management guidance. In
addition, it is implicitly assumed that analysts are naïve information disseminators who can be
easily directed toward any target.
This study re-examines the evidence on expectations management games. Our approach
is new because we directly investigate security analysts' reactions to explicit management
earnings guidance. Our key finding is that analysts act independently of management when
issuing revised earnings forecasts following explicit management guidance. Analysts identify
and adjust for both pessimistic and optimistic bias in management earnings forecasts. Our direct
tests provide new insights into the role of security analysts as independent information
intermediaries in financial markets. Our approach also extends prior research on earnings
expectations management by documenting direct communications between firm management
and analysts instead of relying on indirect proxies for firm-provided guidance.
2
There is a widespread belief that "meeting or beating" analysts' consensus earnings target
is crucial for firm management. For example, Skinner and Sloan (1999) provide evidence that the
stock market severely punishes firms, especially growth firms, that miss an earnings target. This
emphasis on earnings targets motivates the empirical tests in this paper.
We examine both the content of management guidance and the timing, extent and
independence of analysts' reactions to this guidance. This direct analysis breaks down the
expectations management game into its fundamental components. We first determine if
management is, in fact, providing pessimistic earnings guidance. We then examine whether
analysts react to management's guidance by issuing a revised earnings forecast, and the speed of
this reaction. For those analysts that do revise their forecasts, we investigate their independence
from management in terms of whether they merely “parrot” the forecasts of management.
Finally, we examine whether analysts become pessimistic relative to actual earnings outcomes. If
analysts are acting independently of management, we expect that they will discount any
excessive pessimism by managers.
Our data sample consists of a large set of firms that issued quarterly management
earnings forecasts in the three months leading up to quarterly earnings announcements for the
period 1993-1999. We gather information on the actual date and content of all quarterly
management earnings forecasts from the comprehensive First Call database of management and
security analyst earnings forecasts. We then track the daily forecasting activities of individual
security analysts following explicit management guidance. The overall sample consists of 2,422
management earnings forecast events for 1,424 firms.
As a first step, we examine explicit management guidance and determine whether
management is attempting to guide analysts toward a beatable target by being overly pessimistic
3
in their earnings forecasts (the first necessary part of the expectations management game).
Consistent with prior research, we find that most management earnings forecasts tend to contain
"bad news" relative to analysts' prior consensus estimates of quarterly earnings. Moreover, we
find that management guidance tends to be overly pessimistic relative to ex post actual earnings
outcomes (see also Soffer, Thiagaragan and Walther, 1999). This finding is consistent with
claims that management is trying to guide analysts toward beatable targets.
The second important question is whether analysts "take the bait" and naïvely revise their
earnings forecasts to mirror managers' pessimistic forecasts. Our examination of this issue
involves the timing, extent and independence of analyst forecast revisions. In relation to the
timing of forecast revisions, we find that the vast majority of analyst activity in a fiscal quarter
takes place in direct response to explicit management communications with the market. This is
inconsistent with prior claims that analysts tend to revise their forecasts based on unobservable
guidance from management. In our sample, analysts appear to primarily react to explicit public
guidance from management. In regard to the extent of analyst revisions, we find that and
average of 41.6% of analysts following our sample firms revise their forecasts within five days
of a management guidance event. Analysts are more likely to react to management earnings
forecasts that contain new information, particularly if the guidance indicates that earnings will be
lower than expected (bad news).
We then examine the independence of analysts' reactions to management earnings
guidance. First, we determine whether analysts merely "parrot" the information provided by
management. Our results indicate that analysts are more than mere information disseminators.
Indeed, they tend to adjust their forecasts to compensate for the excessive pessimism of
4
management. This result provides new evidence on the independence of analysts from
management. Analysts determine and then adjust for the direction of management bias.
While our evidence that analysts do not simply “parrot” management forecasts indicates
that they act independently of management earnings guidance, management may still have been
successful in guiding analysts to “beatable” forecasts. That is, while analyst forecasts following
explicit management earnings guidance are less pessimistic than management’s forecasts they
may still be pessimistic relative to actual earnings outcomes. Therefore, our final set of tests
examines the extent of analyst pessimism following management earnings guidance. We find
that while analysts tend to be pessimistic following ‘bad news’ management guidance, they are
generally slightly optimistic following ‘good news’ management guidance. However, given that
the majority of management forecasts convey bad news, our results tend to indicate that
management is generally successful in guiding analysts towards forecasts that they can ‘meet’ or
The remainder of the paper is organized as follows. Section 2 reviews the related
literature and develops the hypotheses. Section 3 describes the sample and our empirical
analysis, while section 4 concludes.
2. Hypothesis development and literature review
Recent research has argued that firms have been engaging in a game of earnings
manipulation and expectations management that allows them to consistently beat analysts'
earnings predictions. There is evidence that investors tend to focus on earnings targets and
penalize firms that miss these targets. For example, Skinner and Sloan (1999) provide evidence
that the stock market severely punishes firms, especially growth firms, that miss an earnings
5
target. Given these penalties, it has been argued that firm management has strong incentives to
avoid these negative surprises by manipulating earnings upward to meet or exceed analysts’
expectations. Indirect evidence of earnings manipulation to beat analysts' targets includes Brown
(1999), Brown and Higgins (1999), and Burgstahler and Eames (1999).
On the other hand, firms can attempt to beat an earnings target by actually lowering the
target set by analysts. In other words, firm management can manage expectations downward.
Matsumoto (1999) and Richardson, Teoh and Wysocki (1999) provide indirect evidence that, in
recent years, firms have been guiding analysts toward "beatable" earnings forecasts. These
studies document an apparent shift from systematic analyst optimism at the beginning of a fiscal
period to systematic analyst pessimism just prior to an earnings announcement. This pattern has
become more pronounced during the 1990's. These findings have been interpreted as prima facie
evidence that management has been opportunistically guiding analysts toward beatable targets.
However, these studies have not been able to provide direct evidence of actual management
guidance. For example, Matsumoto (1999) asserts that "… much of the guidance provided to
analysts occurs in private conversations between managers and analysts, the content of which is
not directly observable."
Implicit in prior work by Brown (1999), Burgstahler and Eames (1998), Masumoto
(1999), and Richardson et al. (1999), is the idea of a smooth and continual downward trend in the
level of optimism over a fiscal period that finally leads to pessimism at the end of the period.
This downward trend appears to be consistent with management gradually lowering expectations
of analysts through informal communications throughout the fiscal period.
The purpose of this paper is to examine explicit guidance by management and determine
if prior theories on ex ante expectations management are correct. Past research on analyst
6
forecast activity has shown that analyst activity tends to come in bursts, followed by long lulls in
activity. For example, Stickel (1989) finds that analysts tend to update around interim earnings
announcements. However, there has been little evidence on how analysts cluster around
unscheduled management announcements. 1 If analyst revisions tend to cluster around explicit
management earnings forecasts, then this is inconsistent with management gradually lowering
analysts' expectations through informal communications throughout the fiscal period.
2.1 Managers' attempts to lead analysts
Our analysis begins with an examination of the nature of explicit management guidance.
In particular, we want to determine if management earnings forecasts contain overly-pessimistic
information intended to guide analysts toward a beatable target. If management is attempting to
deflate analysts' expectations and lead analysts toward a beatable target, then:
Hypothesis H1: Management forecasts are pessimistic relative to realized earnings.
2.2 Analysts’ reactions to explicit management earnings guidance
Finding empirical evidence that managers are overly pessimistic in their earnings
forecasts would lend support to the claim that managers are explicitly trying to guide analysts
toward beatable targets. However, how analysts react to this guidance is the more critical
component to the game. If analysts act independently of management and fully adjust for
management bias in their revised forecasts, then the game falls apart. Our examination of
1
Stan Levine of First Call Corporation has performed an analysis of the First Call dataset and also finds evidence
that analysts’ forecasts tend to cluster in time.
7
analysts’ reactions to management earnings guidance includes the timing and extent of analyst
forecast revisions, as well as the independence of analysts' reactions to this guidance.
Stickel (1989) finds that analysts tend to revise their forecasts following interim earnings
announcements. Based on this finding, we expect that analysts will update their earnings
forecasts following other forms of explicit management guidance, including earnings forecasts.
If analysts participate in the expectations management game, then:
Hypothesis H2: Analysts issue revised earnings forecasts following explicit management
earnings guidance.
We examine the independence of analysts by determining whether they react naively to
explicit management earnings guidance. The expectations management game implicitly assumes
that analysts act like "parrots" that simply disseminate information from management forecasts.
If this is the case, then:
Hypothesis H3a "Naive Analyst Hypothesis": Analysts simply repeat earnings numbers that
are forecasted by management.
However, Chung and Jo (1996) propose that analysts act as independent monitors of
managers’ disclosures. Financial analysts play an important role as information intermediaries
between company management and capital markets (Schipper, 1991). Their stock of knowledge
about the firms that they follow allows them to assess the reasonableness of management
earnings forecasts. However, whether analysts adjust for any bias in management earnings
8
forecasts will be a function of their independence from management. Analysts have incentives to
nurture relationships with managers to ensure continued access to company information. If
analysts are independent of management, then:
Hypothesis H3b "Independent Analyst Hypothesis": Analysts’ earnings forecasts following
explicit management guidance are less biased than managers’ earnings forecasts.
2.3 Do analysts become pessimistic following explicit management guidance?
Even if analysts act independently of management and adjust for management bias in
their forecast revisions, management may still have been successful in guiding analysts to
overly-pessimistic or “beatable” forecasts. That is, while analyst forecasts following explicit
management earnings guidance may be less pessimistic than management’s’ forecasts, they may
still be pessimistic relative to actual earnings outcomes. Prior research has shown that analysts
tend to be optimistic in their earnings forecasts, especially early in the year (O'Brien, 1988).
However, if managers can successfully guide analysts toward beatable targets just before an
earnings announcement, then:
Hypothesis H4: Analysts become pessimistic (relative to realized earnings) following explicit
management earnings guidance.
3. Analysis
This section outlines the empirical tests of our hypotheses. In particular, we test the
relation between management earnings forecasts and analysts’ forecast revisions following these
9
forecasts. The empirical tests examine each stage of the earnings expectations management
game and are divided into three main categories. We first test whether there is a pessimistic bias
in management's earnings guidance. In other words, is management attempting to lead analysts
toward beatable targets? This is a necessary condition for the existence of an expectations
management game. We then examine whether analysts naively follow management guidance.
This stage of the research commences with an examination of the timing and extent of analyst's
reactions to guidance events, followed by an examination of analyst independence from
management. Finally, we examine whether analyst forecasts become pessimistic following
explicit management guidance. That is, have managers been successful in guiding earnings
expectations to a level that they can meet or beat?
3.1 Sample and descriptive statistics
In this study, we examine explicit earnings guidance provided by management to security
analysts. The data is obtained from the First Call database of “company issued guidelines"
(basically management forecasts). The sample is limited to forecasts of quarterly earnings per
share for the current quarter. Each quarter is defined by earnings announcements. Rather than
using the fiscal dates, we define the previous earnings announcement date as the start of the
quarter and the current earnings announcement date as the end of the quarter.
We record the date and the exact value of each management earnings forecast. The date
of the management earnings forecast must take place after the announcement of last quarter's
earnings and be related to earnings for the current quarter. We merge this data with the First Call
database of analyst estimates of earnings per share for the current fiscal quarter. The date of each
analyst forecast revision is recorded. Analysts are allowed to make multiple revisions during the
10
quarter. The data set is designed so that all events happen within a given quarter. The sample of
management and analysts forecasts are then merged with the First Call “official actuals” file that
contains the actual ex post earnings per share for the current fiscal quarter. Finally, the data is
merged with financial information from the Compustat database to obtain information on firm
characteristics.
Panel A of Table 1 summarizes the sample for the final set of observations that have
complete information on management earnings forecasts, analyst revisions, and firm
characteristics. The final sample consists of 2,422 management guidance events for 1,424 firms.
There are a total of 6,760 analysts following our sample firms.
Panel B of Table 1 shows descriptive statistics for our sample of firms issuing
management earnings forecasts. Nanalyst is a measure of the number analysts actively following
each sample firm in the quarter (tabulated as the number of individual analyst forecasts made
within 11 days of the previous quarter's earnings announcement). The mean (median) number of
analysts following each of our sample firms is 2.58 (2), with a range of 1 to 22. MVE is the
market value of common shareholders' equity at the beginning of the year. Growth captures
growth opportunities and is measured as the firm's market to book ratio. Our sample of firms
exhibits considerable variation in both size and growth.
Panel C of Table 1 categorizes management earnings forecasts according to their “news
content”. The "news content" of earnings guidance is measured as the management forecast
value relative to analysts' prior consensus estimates of earnings. Good news (bad news, no news)
indicates that the management forecast value is higher than (lower than, equal to) the prior
consensus. Consistent with prior research (see Skinner 1994, 1997) there are substantially more
bad news forecasts than good news forecasts in our sample. 1119 of our sample of 2422
11
management forecasts contain bad news, while 190 contain good news and 1113 contain no
news. That is, they are equal to the prior consensus.
3.2 Do managers attempt to lead analysts?
To commence our analysis of the expectations management game, we first want to
determine if managers attempt to guide analysts toward a beatable earnings target. In particular,
we examine management earnings forecasts where there is explicit, public information
communicated to security analysts and investors. We examine the bias in management earnings
forecasts relative to actual ex post earnings outcomes for each management earnings forecast in
the sample. Pessimism is indicated where the management forecast is lower than ex post
earnings, optimism is indicated where the management forecast is higher than ex post earnings,
while unbiased management forecasts are equal to ex post earnings.
A table cross tabulating the directional bias in management earnings forecasts and “news
content” is presented in Panel A of Table 2. The majority of management forecasts are
pessimistic relative to actual earnings outcomes. Indeed, 1351 of our 2422 (55.8%) of the
management forecasts in our sample are pessimistic, while 124 (5.1%) are unbiased and 947
(39.1%) are optimistic. The large number of pessimistic management forecasts is consistent with
anecdotal evidence and the commonly held perception that management may be leading analysts
toward beatable earnings targets. That is, this result provides evidence in support of hypothesis
one.
Given the differing incentives of managers in relation to announcements of good and bad
news (see Skinner, 1994) we examine the impact of “news content” on managers’ propensity to
provide pessimistic forecasts. The results shown in Panel A of Table 2 indicate that the
12
directional bias in management earnings forecasts is related to the news content of the forecast
(Chi-square = 27.346, p = 0.001). 128 (67.4%) of the 190 good news forecasts are pessimistic,
while only 568 (51%) of the 1113 bad news forecasts are pessimistic.
Panel B of Table 2 shows the extent of bias in management earnings forecasts. MgmtBias
captures the magnitude of bias in management earnings forecasts, and is measured as the amount
of the management earnings forecast less actual earnings, all scaled by actual earnings. 2 The
mean value of our MgmtBias variable for our full sample of 2,422 management forecasts is
11.5% relative to actual earnings. An examination of the pessimistic and optimistic sub-samples
reveals that the average amount of optimism in management forecasts (76.3%) is much greater
than the average amount of pessimism (32.9%), although both of these are high.
3.3 How do analysts react to explicit management guidance?
Based on our finding that managers tend to issue pessimistic guidance about future
earnings, we wish to determine whether analysts react to management's public information
releases, and if so, whether their reactions are naïve or independent. We commence our analysis
with an examination of the timing and extent of analyst reactions to guidance events. The content
of revised analyst forecasts is then examined to determine whether reacting analysts act
independently of management guidance.
3.3.1 The timing and extent of analysts reactions to management forecasts
2
Very low realized earnings for some sample firms cause large absolute values for this measure and the analyst bias
measures discussed in the following sections. To overcome this problem, extreme outliers are winsorized. We also
calculate alternative measures for these variables that involve scaling by market value of equity rather than realized
earnings. However, we report the results using measures scaled by realized earnings, since the values are more
intuitive.
13
Hypothesis two predicts that analysts issue revised earnings forecasts following explicit
management earnings guidance. In testing this hypothesis, we examine whether explicit
management guidance leads to forecast revisions by analysts, how quickly these revised
forecasts are issued following the management forecast, and the proportion of analysts revising
their forecasts.
Based on the findings of Stickel (1989), we expect that most analysts will update their
earnings forecasts following explicit management communications including interim earnings
announcements and earnings forecasts. On a daily basis, we track the number of analyst forecast
revisions immediately following the last interim earnings announcement and in the days
surrounding a manangement guidance event. The results of this plot are presented in Figure 1.
We find a striking pattern in analyst activity. The vast majority of analyst activity occurs around
firms' public information releases. In the overall sample, we find that of all forecasts made
during a quarter, 34% of these individual analyst forecasts occur within 3 days of the last
quarterly earnings announcement. In addition, we find that for this sample of firms that had a
management forecast, another 41% occur on the day of and during the two days following the
management earnings forecasts. Analysts do not appear to foresee this unscheduled
announcement, but most analysts quickly react to the announcement event. We find a 17-fold
increase in analyst activity the day after a management forecast announcement compared to the
day before the announcement. In summary, the vast majority of analyst activity occurs in direct
response to explicit management announcements.
We measure the extent of analyst reactions to management guidance by counting the
number of analysts issuing revised earnings forecasts following a guidance event relative to the
number of analysts following that firm. The total number of analysts' forecasts within 11 days
14
after the announcement of the previous quarter's earnings captures the number of analysts
following each firm. While Figure 1 indicates that the majority of forecast revisions fall on the
day following interim earnings announcements, we consider eleven days following the
announcements to ensure that we capture all analysts following our sample firms.
The number of analysts issuing revised earnings forecasts following a guidance event is
measured between the day of the management forecast and five days subsequent. This is the
period over which the majority of forecast revisions occur. There are a total of 6,284 analyst
forecast revisions during this post-management guidance period. However, not all of these are
issued by analysts that issue forecasts at the beginning of the quarter, and several analysts revise
more than once during this period. Of the 6,760 analysts following our sample, 3,302 (48.8%)
issue at least one revised forecast within this five day period. The majority of these are issued on
either the day of the management earnings forecast (819) or the day following (1,620). 96 of our
following analysts issue two forecasts over the post-management guidance period. However, we
include only the most recent forecast for these analysts.
Frac is calculated for each management guidance event as the ratio of the number of
following analysts who revise their forecasts within 5 days after a management forecast to the
total number of analysts' forecasts within 11 days of the announcement of the previous quarter's
earnings. As for our examination of management guidance bias, our analysis of the extent of
analyst reactions to management guidance includes an examination of the impact of “news
content” on analysts’ propensity to issue revised earnings forecasts. The type of information
released in a management forecast is expected to affect the likelihood that analysts will react to
management guidance. In particular, analysts are expected to have incentives similar to those of
mangers in relation to the avoidance of potential legal liability and reputation costs associated
15
with overly optimistic forecasts. We therefore expect a greater extent of analyst revisions
following ‘bad news’ guidance events.
Panel A of Table 3 shows that an average of 41.6% of analysts following each firm revise
their forecasts within five days of a management guidance event. However, the extent of
forecast revisions differs dramatically with news content. While an average of 65.3% of
following analysts revise in response to bad news management forecasts, only 37.1% respond to
good news forecasts. Further, a mere 18.8% of analysts respond to management forecasts that
simply confirm consensus analyst forecasts. Untabulated results confirm that these average
percentages are similar for the sub-sample of forecast revisions associated with pessimistic
management forecasts. We test the robustness of this relation by controlling for the extent of
management pessimism and several firm characteristics. Our regression model is:
Frac = α + β1GoodNews + β2BadNews + β3MgmtBias + β4Nanalyst
+ β5Size + β6Growth + ε (1)
where GoodNews and BadNews are categorical variables that equal one if the
management forecast is greater than or less than the prior consensus respectively, and zero where
it is the same (no news). MgmtBias captures the magnitude of bias in management earnings
forecasts, and is measured as the amount of the management earnings forecast less actual
earnings, all scaled by actual earnings. Nanalyst is a measure of the number analysts actively
following each sample firm in the quarter (tabulated as the total number of individual analyst
forecasts made within 11 days of the previous quarter's earnings announcement). Size is the
natural logarithm of market value of common shareholders' equity at the beginning of the year
16
C
(computed as [Compustat item # 199] * [ ompustat item # 25]). We measure firm growth
C
opportunities, Growth, using the firm's market to book ratio (computed as [ ompustat item #
199] * [Compustat item # 25] / [Compustat item # 60]).
The regression results are summarized in Panel B of Table 3. Again, analysts are more
likely to react to management earnings forecasts that contain new information, particularly if the
guidance indicates that earnings will be lower than expected (bad news). The extent of analyst
reaction is not significantly associated with the extent of management bias, the number of
analysts following each firm, or the firm characteristics examined. Overall, our evidence is
consistent with analysts quickly issuing revised earnings forecasts following explicit
management earnings guidance that contains new information, particularly bad news.
3.3.2 Are analysts' forecast revisions independent of management guidance?
We have provided preliminary evidence that is consistent with the expectations
management game. First, management tends to understate expected performance in their public
earnings forecasts. Second, analysts quickly respond to newsworthy management guidance and
update their forecasts following these management announcements. The next part of the puzzle is
to determine whether analysts react naïvely or independently of management guidance.
We measure how analysts react to management guidance by comparing each analyst’s
earnings forecast after a guidance event with the value of management's earnings forecast. The
results presented in Panel A of Table 4 provide strong evidence that analysts do not merely
"parrot" management's forecasts. Indeed, only 406 of the 3,302 analysts forecast revisions in the
five days following the management guidance are equal to the management forecasts. The
remainder are either more optimistic or more pessimistic than the management forecasts.
17
Overall, revised analyst forecasts tend to be more optimistic (less pessimistic) than management
forecasts. 2419 (73.3%) of the revised analyst forecasts relate to bad news guidance from
management. For these bad news announcements, analysts are unbiased or more optimistic than
management 67.4% of the time.
Panel B of Table 4 shows the extent of differences between analyst and management
earnings forecasts. GuidanceBias captures the magnitude of bias in revised analyst forecasts
relative to management earnings forecasts, and is measured as the difference between these two
forecasts scaled by actual earnings. For pessimistic management earnings forecasts, analysts tend
to be less pessimistic (more optimistic), while for optimistic management forecasts, analysts tend
to be less optimistic (more pessimistic). These results indicate that not only do analysts act
independently of management, but that they are able to determine and adjust for the direction of
management bias. 3
We test for the robustness of this finding by using a multivariate regression model to
control for the news content of the announcement as well as several firm characteristics. Our
regression model is:
GuidanceBias = α + β1 GoodNews + β2 BadNews + β3 MgmtBias + β4 Nanalyst
+ β5 Size + β6Growth + ε (2)
where all variables are as previously defined. The regression results are summarized in
Panel C of Table 4. The result that analysts act independently of management by identifying and
adjusting for the extent of management bias is supported in a multivariate context. Further,
18
analysts are more likely to be optimistic relative to managers after bad news guidance. In
addition, analysts are more likely to be optimistic when there is higher analyst following for a
given firm.
Overall, the results above reject H3a, “naïve analyst hypothesis” and support H3b
“independent analysts hypothesis”. There is a substantial difference between the analysts’
forecast value and management forecast value. In other words, analysts do not simply
disseminate the same information that is contained in management forecasts. Rather, their
forecast revisions exhibit independence from management guidance. Analysts identify and adjust
for both pessimistic and optimistic bias in explicit management earnings forecasts. Moreover, the
difference between the analysts’ forecasts and management forecasts is changing in the type of
news and number of analysts following the firm.
3.4 Do analysts become pessimistic following management guidance?
The preceding results indicate that analysts tend to adjust for the bias in explicit
management guidance. However, analysts' revised forecasts could still be pessimistic enough to
create a "beatable" target for management at the end of the fiscal period. Therefore, we examine
revised analyst forecasts relative to actual ex post earnings to determine whether analysts become
pessimistic following management guidance. Hypothesis four predicts that analysts become
pessimistic following explicit management guidance.
The results shown in Panel A of Table 5 support hypothesis four and indicate that
management was able to meet or beat a total of 2,113 of the 3,302 revised analyst forecasts
(64%). Therefore, while analysts tend to act independently of managers to the extent that they
3
Deleting all but the most recent forecast revision for each analysts biases against us finding that analysts act
independently of management, since the median ‘second revision’ (0.047) is closer to the management forecast than
19
reduce the bias in management earnings forecasts, managers are still able to meet or beat revised
analysts forecasts in the majority of cases. Further, these results indicate that analyst bias is
associated with the news content of the management forecast; with analysts tending to become
overly optimistic following good news and overly pessimistic following bad news. Management
was less successful in beating analyst forecasts that were pre-empted by ‘good news’
announcements.
Panel B of Table 5 shows the extent of bias in revised analyst forecasts. AnalystBias
captures the magnitude of bias in analyst earnings forecasts, and is measured as the amount of
the analyst earnings forecast less actual earnings, all scaled by actual earnings. While the mean
value of our AnalystBias variable is negative, the median of zero indicates that revised analysts
forecasts following explicit management guidance tend to be reasonably accurate indicators of
expected earnings. For pessimistic management earnings forecasts, analysts tend to be optimistic
relative to actual earnings; while for optimistic management forecasts, analysts tend to be
pessimistic relative to actual earnings. These results tend to indicate that analysts overcorrect in
their adjustments for management bias. That is, while they are able to determine the direction of
management forecast bias, they are less accurate in determining the extent of management bias.
We test for the robustness of this finding by using a multivariate regression model to
control for the news content of the announcement as well as several firm characteristics. Our
regression model is:
AnalystBias = α + β1 GoodNews + β2 BadNews + β3 MgmtBias + β4 Nanalyst +
β5 Size + β6Growth + ε (3)
the first revision (0.323)
20
where all variables are as previously defined. The regression results are summarized in
Panel C of Table 5. These results confirm that the bias in analyst forecasts takes the opposite
direction to that in management forecasts. Further, analysts tend to be pessimistic in relation to
bad news management guidance and optimistic in relation to good news management guidance.
Finally, analysts tend to be more pessimistic when there are a greater number of analysts
following the firm. Analyst bias is not related to the size or growth characteristics of the firm.
The final stage of our tests involves a determination of whether explicit management
guidance has caused a ‘switch’ from analyst optimism to pessimism. We examine our ‘good’ and
‘bad’ news groups separately, since by definition, analysts are overly optimistic prior to Bad
News management guidance and overly pessimistic prior to Good News guidance. Panel A of
Table 5 indicates that only 33.3% (805) of analysts remain optimistic following explicit bad
news guidance. The remainder issue revised forecasts that managers are able to meet or beat.
Indeed, Panel B of Table 5 indicate that the mean (median) extent of analysts forecast bias five
days after the management guidance is –0.110 (0.000). 4 There appears to have been a switch
from optimism to unbiased or slightly pessimistic analyst forecasts in relation to bad news
management guidance. On the other hand, the majority of revised analyst forecasts following
good news guidance are optimistic, with only 10.4% (23) remaining pessimistic. Further, the
mean (median) extent of analyst forecast bias five days after the management guidance is 0.016
(0.019), indicating a switch from pessimism to slight optimism in relation to good news
announcements. Overall, the results shown in Table 5 indicate a switch from optimism to slight
pessimism following bad news management guidance. However, they indicate the opposite trend
in relation to good news guidance.
4
Our alternative measure of analyst bias that is scaled by market value rather than realized earnings shows similar
results.
21
4. Summary and conclusions
This paper re-examines the claim that firms have been successfully guiding security
analysts toward "beatable" earnings forecasts. We use a novel approach to this empirical problem
by examining security analysts' reactions to explicit management earnings guidance. The
empirical tests are implemented using a sample of quarterly management earnings forecasts for a
large number of U.S. firms between 1993 and 1999. We examine the content of both
management's guidance and revised analysts' forecasts following this guidance.
Our empirical tests examine each stage of the earnings expectations management game
and are divided into three main categories. We first test whether there is a pessimistic bias in
management's earnings guidance and find that the majority of management forecasts are
pessimistic. We then examine whether analysts naively follow management guidance. This stage
of the research commences with an examination of the timing and extent of analyst's reactions to
guidance events, followed by an examination of analyst independence from management. We
find that the vast majority of analyst activity occurs quickly and in direct response to explicit
management announcements. Further, we find that analysts are more likely to react to
management guidance that contains bad news.
However, our results indicate that analysts are more than mere information disseminators.
Indeed, it appears that analysts act as monitors of management information releases and make
adjustments to management's overly pessimistic predictions of firm performance. Analysts
appear to limit manager's ability to mislead the market with negative forecasts. This is of
particular importance because management earnings forecasts are not "audited" by independent
22
parties. Analysts are able to identify and adjust for both pessimistic and optimistic bias in explicit
management earnings forecasts.
Finally, we examine whether analyst forecasts become pessimistic following explicit
management guidance. Despite the apparent independence of analysts, our findings are generally
consistent with prior claims that management has been successful in systematically guiding
analysts toward "beatable" earnings forecasts. However, this is not the case for good news
management guidance. Analysts tend to become overly optimistic following good news
guidance, causing a switch from analyst pessimism to slight optimism. Our results in relation to
bad news management guidance are consistent with Matsumoto (1999), Richardson et al. (1999),
and Brown (1999). That is, we find that analysts appear to shift from optimism to slight
pessimism in response to bad news management guidance.
23
References
Baginski, S., E. Conrad, and J. Hassell, 1993, The effects of management forecast precision on
equity pricing and on the assessment of earnings uncertainty, The Accounting Review 68,
913-927.
Baginski, S., J. Hassell, and G. Waymire, 1994, Some evidence on the news content of
preliminary earnings estimates, The Accounting Review 69, 265-271.
Baginski, S., and J. Hassell, 1997, Determinants of management forecast precision, The
Accounting Review 72, 303-312.
Bauman, W.S., S. Datta, and M. Iskandar-Datta, 1995, Investment analyst recommendations: A
test of 'the announcement effect' and 'the valuable information effect', Journal of Business
Finance and Accounting 22, 659-70.
Bhushan, R., 1989, Firm characteristics and analyst following, Journal of Accounting and
Economics 11, 255-74.
Brown, L., 1999, Managerial behavior and the bias in analysts' earnings forecasts, Working
paper, Georgia State University.
Brown, L., and H. Higgins, 1999, Earnings surprise games: International evidence, Working
Paper, Georgia State University.
Burgstahler, D. and M. Eames, 1999, Management of earnings and analyst forecasts, Working
paper, University of Washington.
Chung, K.H. and H. Jo, 1996, The impact of security analysts' monitoring and marketing
functions on the market value of firms, Journal of Financial and Quantitative Analysis
31, 493-512.
Elliott, J., D. Philbrick, and C. Wiedman, 1995, Evidence from archival data on the relation
between security analysts' forecast errors and prior forecast revisions, Contemporary
Accounting Research 11, 919-938.
Frost, C., "Characteristics and information value of corporate disclosures of forward-looking
information in global equity markets", 1996, Unpublished working paper, Washington
University in St. Louis.
Kasnik, R. and B. Lev, 1995, To warn or not to warn: Management disclosures in the face of an
earnings surprise, The Accounting Review 70, 113-34.
Lang, M.H. and R. Lundholm, 1996, Corporate disclosure policy and analyst behavior, The
Accounting Review 71, 467-92.
24
Lev, B., 1992, Information disclosure strategy, California Management Review, Summer, 9-32.
Lys, T. and S. Sohn, 1990, The association between revisions of financial analysts' earnings
forecasts and security-price changes, Journal of Accounting and Economics 13, 341-363.
McNichols, M. and O'Brien, P.C., 1997, Self-selection and analyst coverage, Journal of
Accounting Research 35, Supplement, 167-99.
O'Brien, P.C. and Bhushan, R., 1990, Analyst following and institutional ownership, Journal of
Accounting Research 28, Supplement, 55-76.
Pownall, G. and G. Waymire, 1993, Alternate forms of management earnings forecasts:
Incidence and stock price effects, The Accounting Review 68, 896-912.
Previts, G., R. Bricker, T. Robinson, S. and Young, 1994, A content analysis of sell-side
financial analyst company reports, Accounting Horizons 8, 55-70.
Richardson, S., S. Teoh, and P. Wysocki, 1999, Tracking Analysts’ Forecasts over the Annual
Earnings Horizon: Are Analysts’ Forecasts Optimistic or Pessimistic?, Working Paper,
University of Michigan.
Schipper, K., 1991, Commentary on analysts' forecasts, Accounting Horizons, December, 105-
21.
Skinner, D. J. , 1994, Why Firms Voluntarily Disclose Bad News, Journal of Accounting
Research, Vol. 32, 38-60.
Skinner, D. J. and R. G. Sloan, 1999, Earnings Surprises, Growth Expectations and Stock
Returns, Working Paper, University of Michigan.
Soffer, L., S. Thiagarajan, and B. Walther, 1998, Earnings preannouncements, Working Paper,
Kellogg Graduate School of Management, Northwestern University.
Stickel, S., 1989, The timing of and incentives for annual earnings forecasts near earnings
announcements, Journal of Accounting and Economics 11, 275-292.
Trueman, B., 1996, The impact of analyst following on stock prices and the implications for
firms' disclosure policies, Journal of Accounting, Auditing and Finance 11, 333-54.
Williams, P., 1996, The relation between a prior earnings forecast by management and analyst
response to a current management forecast, The Accounting Review 71, pages?
25
Table 1
Descriptive Statistics
Panel A: Sample Information
Sample Period 1993-1999
Number of management 2,422
earnings forecasts
Number of firms issuing 1,424
management forecasts
Total number of analysts 6760
following sample firms
Panel B: Descriptive Statistics for Sample of 1424 Firms
Variable Name Mean Median Std. Dev. Min Max
NAnalyst 2.58 2 2.49 1 22
MVE 2236 446 7037 23 100231
Growth 3.36 2.52 2.69 0.41 20.45
Panel C: News Content of Management Earnings Forecastsa
Good News 190
Bad News 1113
No News 1119
Total Number of Management
Earnings Forecasts 2422
Nanalyst captures the number of analysts actively following each firm, and is calculated as the number of analysts’
forecasts within 11 days after the previous quarter’s earnings announcement.
MVE is the beginning market value of equity [computed as Compustat Item #199 * Compustat Item #25].
Growth captures growth opportunities and is measured as the market to book ratio [computed as MVE / Compustat
Item #60].
a
The news content of management earnings forecasts is determined by a comparison with the prior analyst
consensus. Good news (bad news, no news) indicates that the management forecast value is higher than (lower than,
equal to) the prior consensus.
26
Table 2
Bias in Management Earnings Forecasts Compared to Actual Earnings Outcomes for
Sample of 2422 Management Earnings Forecasts
Panel A: Bias in Management Earnings Forecasts by News Content
News Contenta Pessimisticb Unbiasedb Optimisticb Total
Good news 128 4 58 190
Bad news 568 72 473 1113
No news 655 48 416 1119
1351 124 947 2422
Chi-square = 27.346, p = 0.001
Panel B: Extent of Bias in Management Earnings Forecasts (MgmtBias)
Total (includes
b b
Pessimistic Optimistic unbiased forecasts)
Mean -0.329 0.763 0.115
Median -0.200 0.707 -0.057
Standard Deviation 0.328 0.643 0.705
N 1351 947 2422
a
The news content of management earnings forecasts is determined by a comparison with the prior analyst
consensus. Good news (bad news, no news) indicates that the management forecast value is higher than (lower than,
equal to) the prior consensus.
b
The directional bias in management forecasts is determined by a comparison with actual earnings. Pessimistic
(optimistic, unbiased) indicates that the management forecast value is lower than (higher than, equal to) the ex-post
earnings value for a given quarter.
MgmtBias captures the extent of bias in management earnings forecasts and is measured as the management
earnings forecast less actual earnings, all scaled by actual earnings.
27
Table 3
Fraction of Analysts Reacting to Management Earnings Forecast (FRAC)
Panel A: Descriptive Statistics for FRAC for Full Sample of 2422 Management Forecasts
News content a Mean Median Std. Dev. Min Max
Good news 0.371 0.250 0.406 0 1
Bad news 0.653 0.800 0.389 0 1
No News 0.188 0 0.341 0 1
Total 0.416 0.333 0.431 0 1
Panel B: Determinants of Fraction of Analysts Reacting to Management Earnings Forecast
(Frac = α + β1 GoodNews + β2 BadNews + β3 MgmtBias + β4 Nanalyst + β5Size + β6 Growth + ε)
Coefficient Estimate t-statistic p-value
Intercept 0.124 3.592 0.001
GoodNews 0.187 6.441 0.001
BadNews 0.470 29.311 0.001
MgmtBias -0.017 -1.590 0.112
Nanalyst 0.004 1.269 0.204
Size 0.007 1.342 0.180
Growth 0.001 0.472 0.637
N 2422
Adj. R2 0.269
a
The news content of management earnings forecasts is determined by a comparison with the prior analyst
consensus. Good news (bad news, no news) indicates that the management forecast value is higher than (lower than,
equal to) the prior consensus.
Frac is calculated as the ratio of the total number of individual analysts’ forecasts within 5 days after a management
forecast to the total number of individual analysts’ forecasts within 11 days after previous quarter’s earnings
announcement.
GoodNews is a categorical variable which equals 1 if the management forecast is greater than the prior consensus
and 0 otherwise.
BadNews is a categorical variable which equals 1 if the management forecast is less than the prior consensus and 0
otherwise.
MgmtBias captures the extent of bias in management earnings forecasts and is measured as the management
earnings forecast less actual earnings, all scaled by actual earnings.
Nanalyst captures the number of analysts actively following each firm, and is calculated as the number of analysts’
forecasts within 11 days after the previous quarter’s earnings announcement.
Size is proxied by Log(MVE) which is the natural logarithm of beginning market value of equity and is computed as
ln (Compustat Item #199 * Compustat Item #25).
Growth captures growth opportunities and is measured as the market to book ratio [computed as MVE / Compustat
Item #60].
28
Table 4
Differences between Analyst and Management Forecasts
Panel A: Analyst Earnings Forecasts Compared to Management Earnings Forecasts by News Content
Analysts more Analysts more
a
News Content Pessimistic b Unbiased b
Optimistic b Total
Good news 91 18 113 222
Bad news 788 327 1304 2419
No news 270 61 330 661
1149 406 1747 3302
χ2 = 25.732, p < 0.001
Panel B: Extent of differences between analyst and management earnings forecasts (GuidanceBias)
Total
Pessimistic Optimistic (includes unbiased
Management Management management
Forecasts Forecasts forecasts)
Mean 0.300 -0.524 -0.057
Median 0.167 -0.333 0.029
Standard Deviation 0.381 0.891 0.749
N 1716 1362 3302
Panel C:
(GuidanceBias = α + β1 GoodNews + β2 BadNews + β3MgmtBias + β4Nanalyst + β5Size + β6 Growth + ε)
Coefficient Estimate t-statistic p-value
Intercept 0.041 0.936 0.349
GoodNews -0.065 -1.758 0.079
BadNews 0.085 4.019 0.001
MgmtBias -0.792 -68.280 0.001
Nanalyst 0.006 3.314 0.001
Size -0.012 -1.949 0.051
Growth 0.000 0.041 0.967
N 3302
Adj. R2 0.599
29
a
The news content of management earnings forecasts is determined by a comparison with the prior analyst
consensus. Good news (bad news, no news) indicates that the management forecast value is higher than (lower than,
equal to) the prior consensus.
b
The directional bias in analyst forecasts relative to management forecasts is determined by a comparison of the
two. Analysts more pessimistic (optimistic, unbiased) indicates that the analysts forecast value is lower than (higher
than, equal to) the management forecast.
GuidanceBias captures the extent of bias in analyst forecasts relative to management forecasts and is measured as
the analysts forecasts less the management forecast, all scaled by actual earnings.
GoodNews is a categorical variable which equals 1 if the management forecast is greater than the prior consensus
and 0 otherwise.
BadNews is a categorical variable which equals 1 if the management forecast is less than the prior consensus and 0
otherwise.
MgmtBias captures the extent of bias in management earnings forecasts and is measured as the management
earnings forecast less actual earnings, all scaled by actual earnings.
Nanalyst captures the number of analysts actively following each firm, and is calculated as the number of analysts’
forecasts within 11 days after the previous quarter’s earnings announcement.
Size is proxied by Log(MVE) which is the natural logarithm of beginning market value of equity and is computed as
ln (Compustat Item #199 * Compustat Item #25).
Growth captures growth opportunities and is measured as the market to book ratio [computed as MVE / Compustat
Item #60].
30
Table 5
Analyst Earnings Forecasts Compared to Actual Earnings Outcomes
Panel A: Analyst Earnings Forecasts Compared to Actual Earnings Outcomes by News
Content
News Content a Pessimistic b Unbiasedb Optimistic b Total
Good news 23 74 125 222
Bad news 946 668 805 2419
No news 225 177 259 661
1194 919 1189 3302
χ2 = 82.475, p < 0.001
Panel B: Extent of Bias in Revised Analyst Forecasts (AnalystBias)
Pessimistic Optimistic Bad News Good News Total
Management Management Management Management
Forecasts Forecasts Forecasts Forecasts
Mean 0.024 -0.244 -0.110 0.016 -0.091
Median 0.000 -0.042 0.000 0.019 0.000
Std Deviation 0.209 0.724 0.545 0.198 0.507
N 1716 1362 2419 222 3302
Panel C: Determinants of Extent of Analyst Bias
(AnalystBias = α + β1 GoodNews + β2 BadNews + β3MgmtBias + β4 Nanalyst + β5Size + β6 Growth + ε)
Coefficient Estimate t-statistic p-value
Intercept -0.017 -0.395 0.693
GoodNews 0.078 2.157 0.031
BadNews -0.091 -4.371 0.001
MgmtBias -0.270 -23.691 0.001
Nanalyst -0.006 -3.222 0.001
Size 0.009 1.485 0.138
Growth -0.001 -0.338 0.736
N 3302
Adj. R2 0.154
31
a
The news content of management earnings forecasts is determined by a comparison with the prior analyst
consensus. Good news (bad news, no news) indicates that the management forecast value is higher than (lower than,
equal to) the prior consensus.
b
The directional bias in analyst forecasts relative to management forecasts is determined by a comparison of the
two. Analysts more pessimistic (optimistic, unbiased) indicates that the analysts forecast value is lower than (higher
than, equal to) the management forecast.
AnalystBias is a categorical variable that equals –1 (1, 0) if the individual analyst’s forecast (after management
guidance) is lower than (higher than, equal to) the ex-post earnings value for a given quarter.
GoodNews is a categorical variable which equals 1 if the management forecast is greater than the prior consensus
and 0 otherwise.
BadNews is a categorical variable which equals 1 if the management forecast is less than the prior consensus and 0
otherwise.
MgmtBias captures the extent of bias in management earnings forecasts and is measured as the management
earnings forecast less actual earnings, all scaled by actual earnings.
Nanalyst captures the number of analysts actively following each firm, and is calculated as the number of analysts’
forecasts within 11 days after the previous quarter’s earnings announcement.
Size is proxied by Log(MVE) which is the natural logarithm of beginning market value of equity and is computed as
ln (Compustat Item #199 * Compustat Item #25).
Growth captures growth opportunities and is measured as the market to book ratio [computed as MVE / Compustat
Item #60].
32
Figure 1: Frequency Distribution of Analysts' Forecasts
4500
4000
3500
3000
frequency
2500
2000
1500
1000
500
0
1 2 3 4 5 6 7 8 9 10 11 -5 -4 -3 -2 -1 0 1 2 3 4 5
days
Day after earnings Day of management
announcement for quarter q-1 forecast for quarter q
33