analysts’ responsiveness and market …...analysts’ responsiveness and market underreaction to...
TRANSCRIPT
Analysts’ Responsiveness and Market Underreaction
to Earnings Announcements
Yuan Zhang
611 Uris Hall, 3022 Broadway Columbia Business School
Columbia University New York, NY 10027
Email: [email protected]
Phone: 212-854-0159 Fax: 212-316-9219
October 2004
Preliminary. Comments Welcome.
I thank Mei Cheng, Stephen Penman, K.R. Subramanyam, Richard Willis, and Paul Zarowin for helpful comments and suggestions. I also thank I/B/E/S for providing analysts’ forecast information. All errors are my own.
Analysts’ Responsiveness and Market Underreaction
to Earnings Announcements
Abstract This study shows that analysts vary significantly in their responsiveness to earnings announcements, where responsiveness is defined as promptness of analysts’ first forecast revisions for the next quarter since the prior quarterly earnings announcements. Further evidence indicates that analysts’ responsiveness improves the efficiency of their expectations of future earnings immediately after the earnings announcements, which in turn mitigates the magnitude of the post-earnings-announcement drift. The results provide direct support for the “delayed response” hypothesis that prior research proposes to explain market underreactions.
1
Analysts’ Responsiveness and Market Underreaction to Earnings Announcements
1. Introduction
This study seeks to examine (i) how responsive sell-side security analysts (hereafter,
“analysts”) are to quarterly earnings announcements in revising their forecasts for future
earnings and (ii) whether their responsiveness is associated with the extent to which they, as
well as the market, underreact to earnings announcements.
Stylized valuation models frequently posit that stock price is a function of expected
(permanent) earnings based on available information. The efficient market hypothesis
suggests that upon receiving new information, investors instantaneously adjust their
expectations of earnings, which is in turn reflected instantaneously in stock prices.
However, researchers have been able to document evidence inconsistent with the
efficient market hypothesis. One of the most persistent anomalies is the post-earnings-
announcement drift, where stock prices continue to drift for a long period after the earnings
announcements. Since the phenomenon was first reported by Ball and Brown (1968), it has
survived robustness checks, including extension to more recent data (e.g., Bernard and
Thomas 1989; Chan, Jegadeesh, and Lakonishok 1996; Doyle, Lundholm, and Soliman 2003).
As Fama (1998) puts it, the post-earnings-announcement drift is an anomaly that is “above
suspicion.”
A number of studies have attempted to explain the post-earnings-announcement drift.
Bernard and Thomas (1989) suggest that the “delayed response” hypothesis is a more likely
explanation for the drift than the “risk premium” hypothesis. Hong and Stein (1999) propose
2
that market underreacts because private information diffuses gradually across investors.1 In
explaining why the market underreacts to public information such as earnings announcements,
Hong and Stein suggest that although the news itself is public, it might require some other,
private, information to convert this news into a judgment about value, thus the “gradual
information diffusion” explanation continues to apply.
Barberis, Shleifer, and Vishny (1998) propose a model of investor sentiment to
explain market under- and overreaction. Their model is based on literature on the psychology
of decision making. In particular, they suggest that market underreaction is consistent with a
phenomenon documented in psychology, namely conservatism, defined as the slow updating
of models in the face of new information.
While these papers have different perspectives in explaining market underreaction,
one common and important implication of their explanations is that investors are slow in
adjusting their expectations for future earnings upon receiving new information, which I
generally refer to as the “delayed response” hypothesis. Few studies, however, have directly
tested this hypothesis by focusing on the speed at which investors adjust their expectations
after news releases.
This study seeks to directly test the “delayed response” hypothesis. Specifically, I
examine the responsiveness of analyst forecast revisions after quarterly earnings
announcements. I focus on analysts because general investors’ earnings expectations are not
directly observable. On the other hand, one of analysts’ major responsibilities is to issue
1 In Hong and Stein’s (1999) model, there are two types of investors: news watchers and momentum traders. It is the gradual information diffusion among the news watchers that they suggest leads to market underreaction.
3
earnings forecasts to guide the general investors, and their forecasts have significant influence
on investors (Schipper 1991).2
I define analysts’ responsiveness as the promptness of their first forecast revisions for
the next quarter after the prior quarterly earnings announcements. Consistent with the
“delayed response” hypothesis, I find that, based on analyst-firm-quarter specific observations,
about 44% of analysts (hereafter, “responsive analysts”) issue forecast revisions within two
trading days after the earnings announcement, whereas the average number of calendar days
between the earnings announcements and the first forecast revisions is thirty-four days for the
other 56% of analysts (hereafter, “non-responsive analysts”). Additional analysis shows that
relative to those for the responsive analysts, the absolute forecast errors are in fact
significantly larger for the non-responsive analysts, suggesting their lack of prompt responses
is not because the earnings announcements convey relatively less new information with
respect to their prior information set. This also suggests that the responsive analysts’ forecasts
are more accurate. Further, the responsive analysts not only react more promptly, but also
more completely to the earnings announcements—their first forecast revisions have higher
correlations with the earning surprises than do those of the non-responsive analysts.
Abarbanell and Bernard (1992) examine whether the post-earnings-announcement
drift can be explained by analysts’ underreaction to earnings announcements and find
supporting evidence that analyst forecast errors exhibit positive serial correlations.
Accordingly, to investigate the effect of analysts’ responsiveness on the extent of market
underreaction, I first examine its effect on the serial correlation of analyst forecast errors two 2 It is not uncommon that researchers use analyst forecasts to proxy for market expectations (e.g., Conrad, Cornell, and Landsman 2002; Liang 2003). Fried and Givoly (1982) suggest that analysts’ forecasts provide a better surrogate for market expectations than forecasts generated by time-series models.
4
trading days after the earnings announcements, the starting point of my measure of the post-
earnings-announcement drift. While both responsive and non-responsive analysts appear to
underreact to earnings announcements, responsive analysts underreact to a significantly lesser
extent. In fact, the serial correlation in their forecast errors is less than half of that in non-
responsive analysts’ forecast errors. This result also holds at the firm level with mean forecast
errors, where the firm-level analysts’ responsiveness is measured by the percentage of
responsive analysts following the firm.
Finally, I investigate the effects of analysts’ responsiveness on the magnitude of the
post-earnings-announcement drift. I find that the drift is significantly lower when the
percentage of responsive analysts following the firm is higher. Specifically, over the sixty-
trading-day period starting from the third trading day after the earnings announcements, the
drift is about one third lower for firms followed by responsive analysts only than firms
followed by non-responsive analysts only.
In sum, the results of this study show that a majority of analysts are not responsive to
earnings announcements in revising their forecasts for future earnings. Further, analysts vary
significantly in their responsiveness (and completeness) in incorporating information in
earnings announcements into their forecast revisions. Most importantly, the difference in
responsiveness affects the efficiency of their expectations, and hence the efficiency of market
expectations, of future earnings immediately after the earnings announcement, which in turn
affects the magnitude of the post-earnings-announcement drift.
Thus, this study provides direct support for the “delayed response” hypothesis for the
post-earnings-announcement drift discussed above. It suggests that the speed at which market
participants incorporate new information into their forecasts for future earnings and into stock
5
prices is indeed associated with the extent of market underreaction. As suggested by Hong
and Stein (1999) and Barberis et al. (1998) respectively, the delayed response may be caused
by analysts requiring additional, private information to convert the news in the earnings
announcements into a judgment about future earnings, or simply by their cognitive
conservatism.
This study also extends the literature by focusing on an alternative aspect of market
efficiency. Unlike many prior studies that focus on the instantaneity of stock prices in
incorporating new information (e.g., Ball and Brown 1968), this study examines the
instantaneity of analysts in incorporating new information into their forecast revisions. While
focusing on stock prices speaks to market efficiency aggregately, focusing on analysts’
forecast revisions has the advantage of providing information about individual investors’
behavior. Since stock prices are ultimately driven by individual market participants’ behavior,
understanding their responses to information releases helps us understand the reasons for
market efficiency or inefficiency.
A number of studies examine the efficiency of analysts’ forecasts. However, as
discussed in more details later, while these studies focus on whether and to what extent
analysts underreact or overreact to prior information by examining the serial correlation of
their forecast errors (e.g., Abarbanell and Bernard 1992; Easterwood and Nutt 1999), they do
not specifically address the promptness of analysts’ incorporating public information into
their forecasts. Both the timing and the magnitude of analysts’ reaction to public information
are important because the market efficiency hinges on both the instantaneity and the
completeness of the stock prices in reflecting available information.
6
This study also underscores the importance of examining analysts’ responsiveness to
information releases in order to fully understand the characteristics of analysts and their
forecasts. In addition to pointing out that analysts’ responsiveness per se is an important
aspect of analyst forecast efficiency, this study also finds that responsive analysts not only
respond more promptly, their responses are also more complete and their forecasts more
accurate. Thus, this study suggests that analysts’ responsiveness to information releases is
potentially a good indicator of analysts’ forecasting ability, which can help investors
differentiate among analysts. This implication is consistent with Cooper, Day, and Lewis
(2001) who find that lead analysts can be identified by the timeliness of their earnings
forecasts.
The paper proceeds as follows. Section 2 describes the data. Section 3 examines the
timing and magnitude of analysts’ responses to earnings announcements. Section 4 examines
the effects of analysts’ responsiveness on their underreaction to the earnings announcements,
and on the post-earnings-announcement drift. Section 5 concludes.
2. Data and Descriptive Statistics
I focus on analysts’ forecast revision for the next quarter after the current quarterly
earnings announcement, since prior studies (e.g., Bernard and Thomas 1990) suggest that the
post-earnings-announcement drift is caused by investors’ failure to recognize the
autocorrelation structure of quarterly earnings. I obtain analyst forecast revision and earnings
announcement data from I/B/E/S detail file3 and stock return and price data from CRSP. As
3 The I/B/E/S data is obtained from WRDS, where the detailed file provides not only the analyst forecast information but also corresponding earnings announcement information.
7
discussed below, I also use information from I/B/E/S identification file and adjustment file to
make certain adjustments.
The sample starts with all I/B/E/S individual analyst forecasts for quarterly earnings
per share with fiscal period ending between 1988 and 2002. I delete observations with zero
analyst-specific identification code4 or missing CUSIP. I also delete observations with (i)
forecast date on or after the corresponding earnings announcement date or (ii) earnings
announcement date before or more than ninety days after the corresponding fiscal period end,
as these observations are potentially subject to data error or other irregularities. For firm-
quarters that are followed by I/B/E/S on a diluted basis, I use the dilution/primary adjustment
factor in the identification file to convert the forecast and actual earnings per share to a
primary basis.
For each analyst-firm-quarter, I retain only the most recent forecast before the
corresponding earnings announcement. I then require that this analyst has at least one forecast
for the next quarter of the same firm issued before the current earnings announcement and at
least one issued after. At this stage, 396,001 analyst-firm-quarters remain in the sample. All
EPS variables (forecasts or actuals) are then unadjusted for stock splits based on information
from the I/B/E/S adjustment file. I next require fiscal period end stock price and return
information necessary to calculate size-adjusted returns (SARj,t, defined in Section 4)
available from CRSP, yielding a sample of 333,758 observations. To prevent undue
influences by outliers, I delete observations with extreme 1% forecast errors (FEi,j,t, defined
4 I/B/E/S assigns a zero identification code if the broker did not provide an analyst name to be associated with the estimate.
8
below) at both tails, which results in a final sample of 327,084 analyst-firm-quarters,
representing 103,681 firm-quarters.
Figure 1 plots the timeline of various information events. For each analyst i who
follows firm j, I denote her latest forecast for quarter t as Fi,j,t and the corresponding actual
earnings per share as Ej,t. The forecast error for quarter t, FEi,j,t, is calculated as (Ej,t - Fi,j,t)/P,
where P is the firm j’s stock price at the end of fiscal quarter t. Additionally, I denote her
latest (first) forecast for quarter t+1 before (after) quarter t earnings announcement as Fi,j,t+1a
(Fi,j,t+1b), and the corresponding actual earnings per share as Ej,t+1. The forecast revision for
quarter t+1 upon quarter t earnings announcement (REVi,j,t+1) is thus calculated as (Fi,j,t+1b -
Fi,j,t+1a)/P.
Table 1 provides descriptive statistics for the sample. The number of analyst-firm-
quarters increases steadily over the sample period, consistent with prior research (e.g., Ivković
and Jegadeesh 2004). The average number of calendar days from quarter t earnings
announcement to the analyst’s first forecast revision for quarter t+1 since (inclusive), on the
other hand, decreases steadily from 33 days in 1988 to 13 days in 2002. The decrease in
medians is even more salient, from 27 days in 1988 to merely 2 days in 2002. This suggests
that analysts have gotten more responsive over the sample period, although a considerable
number of analysts still do not react immediately upon the earnings announcements.
Table 1 also presents the means and medians of analysts’ forecast errors (FEi,j,t) and
their forecast revisions (REVi,j,t+1). The mean and median forecast errors at the time of the
earnings announcements are more frequently negative in the early years and more frequently
positive in later years, suggesting analysts have gone from being relatively optimistic to
9
relatively pessimistic over the sample period.5 Finally, both the mean and median of analysts’
first forecast revisions for the next quarter after the current quarter’s earnings announcements
are negative for every year in the sample period. This is consistent with evidence presented in
Ivković and Jegadeesh (2004, Table 3 Panel B) that more of the first forecast revisions after
the earnings announcements are downward revisions than upward revisions.
A couple of econometric issues warrant discussion before I move on to the empirical
evidence. To correct for the autocorrelation and generalized conditional heteroskedasticity
embedded in the sample, the t-statistics in the regressions are adjusted using generalized
method of moments (GMM) as described in Newey and West (1987) with six lags. To
minimize the effects of outliers, I delete observations with extreme forecast errors as
described in the sample selection process. In addition, all regressions reported in the tables are
estimated after deleting observations with absolute studentised residuals greater than 2 (e.g.,
see SAS 1989).
3. How Responsive are Analysts to Earnings Announcements?
3.1. Timing of Analyst Forecast Revisions
In light of the stylized model discussed at the beginning of the paper, the efficient
market hypothesis predicts that upon receiving new information, rational investors
instantaneously update their expectations for future earnings. However, while new
5 Richardson, Teoh, Wysocki (forthcoming) also find analyst forecast errors at the time of earnings announcements are on average optimistic in earlier years and pessimistic in recent years, although their focus is the change in bias (optimism to pessimism) during the year. They suggest that the pessimistic bias in analyst forecasts outstanding at time of earnings announcements in recent years is driven by the “earnings-guidance game” in which analysts walkdown their estimates to a level the firm can beat at the earnings announcements.
10
information may become available to the market everyday, analysts do not necessarily revise
and publish their forecasts everyday. Their revision activities potentially depend on the extent
to which the new information alters their expectations for future earnings.
Bagnoli, Levine, and Watts (2004) find that among various corporate information
events, analyst forecast revisions tend to cluster to a greater extent after earnings
announcements than after earnings guidance or other events. They argue that this is because
the earnings announcements deliver news in a clear and relatively consistent format at
predictable times and consist largely of financial information prepared in accordance with
generally accepted accounting principles. They also argue that earnings announcements may
induce more analyst forecast revisions because of the directness of the link between earnings
and firm valuation. In addition, as Francis et al. (2002) document, there is a growing tendency
for managers to include additional, significant information in an earnings release such as
income statement line items, balance sheets, cash flow information, and forecasts. This
additional information also helps analysts in revising their forecasts for future earnings.
Thus, given the significant implications of earnings announcements for future earnings
as well as for firm values, although analysts do not necessarily revise their forecasts each time
they receive new information, they are more likely to do so after the earnings announcements
than after other corporate information events. Consistent with Bagnoli et al. (2004), Stickel
(1989) finds that analysts avoid revising for two weeks before an earnings announcement and
more frequently revise immediately after the announcement. Similarly, Ivković and Jegadeesh
(2004) find that analysts’ forecast revisions concentrate on the days immediately after the
earnings announcements. However, unlike the current study, none of the above studies is
specifically interested in the effects of analysts’ responsiveness on market underreaction.
11
To examine analysts’ responsiveness to earnings announcements, I first present
distributions of the number of calendar days from the earnings announcement of quarter t to
the analyst’s first forecast revision for quarter t+1 since. Panel A of Table 2 shows that
inconsistent with the prediction that analysts instantaneously adjust their expectations upon
the earnings announcements, on average it takes 20 days before an analyst revises her forecast.
The median is 7 days. The first and third quartiles are 2 and 31 days respectively, whereas the
standard deviation is 25 days. The distribution is very similar after I delete observations with
zero analyst-specific forecast error where the earnings announcement potentially conveys
minimal information to the analyst.
Figure 2 plots the timing of analysts’ issuance of forecast revision with respect to the
earnings announcement date. The pattern mirrors that of the post-earnings-announcement drift
documented in prior studies (e.g., Ball and Brown 1968). Specifically, a relatively large
portion of revisions takes place immediately after the earnings announcement, yet the
activities of revisions continue into months after the earnings announcements. In fact, about
50% of analysts do not revise their forecasts within five calendar days after the earnings
announcements. About 14% of analysts revise their forecasts during the second month and
11% do so during the third month and beyond after the earnings announcements.
I formerly measure an analyst’s responsiveness by RESPi,j,t, which equals 1 if she
issues a forecast revision within two trading days after the earnings announcement (i.e., the
event window is from trading day 0 to trading day 2), and 0 otherwise. To the extent that one
expects stock prices to have reflected the information immediately after the event, one would
also expect investors to have processed the information and revised their expectations for
future earnings by the same time. Event studies typically use one or two trading days after the
12
event as the end of the event window. I choose two trading days to allow for the possibility
that analysts may need more time to convert the information in earnings announcement into a
formal forecast for future earnings or they may need more time to acquire additional
information from the firm.
Panel B of Table 2 shows that only 43.67% of analysts revise their quarter t+1
forecasts during the event window (i.e., RESPi,j,t=1). This percentage is virtually unchanged at
43.82% if I delete observations with FEi,j,t=0.6 I next analyze analysts’ responsiveness by
industry, where firms are classified into 48 industries following Fama and French (1997).
Panel B lists industries with at least 1% of analyst-firm-quarters in the sample that have the
highest or lowest percentage of responsive analysts. Electronic Equipment has the highest
percentage of responsive analysts (56%), followed by Measuring and Control Equipment
(53%), Business Services (51%), and Electrical Equipment (51%). On the other hand,
Chemicals has the lowest percentage of responsive analysts (35%), followed by Trading
(38%), Automobiles and Trucks (38%), and Food Products (38%). While there is some degree
of variation in analysts’ responsiveness across different industries, the evidence suggests that
even in industries with the highest percentage of responsive analysts, only slightly more than
half of the analysts revise their forecasts for future quarters promptly.
Panel C examines analysts’ responsiveness at the firm level, where the responsiveness
is measured by FRESPj,t, the percentage of responsive analysts among all those who follow
the firm for the quarter. Only about 18% of firm-quarters have 100% of responsive analysts
6 Ivković and Jegadeesh (2004) find frequency of responsive revisions similar to that reported in the current study when they examine only the first forecast revisions by the analysts since the earnings announcements.
13
following the firm. In contrast, 46% of firm-quarters have no responsive analysts following at
all.
Overall, the descriptive statistics in Table 2 suggest considerable variations in
analysts’ responsiveness and more importantly, they suggest a majority of analysts, despite
their expertise, do not adjust their expectations instantaneously after earnings announcements
which potentially provide significant, new information. This evidence is consistent with the
“delayed response” explanation for the post-earnings-announcement drift, raising questions
regarding the efficiency of the analyst forecasts as well as the efficiency of the market.
I next examine certain analyst-specific or firm-specific variables conditional on
analysts’ responsiveness. Table 3 shows that while mean and median number of calendar days
from firms’ announcing Ej,t to analysts’ issuing Fi,j,t+1b are only 2.50 and 2 respectively for
responsive analysts, they are as high as 34 and 27 days respectively for non-responsive
analysts.7 The next variable of interest is AFEi,j,t, the absolute value of FEi,j,t, which reflects
the amount of new information the earnings announcement conveys to the analyst with
respect to her prior information set reflected in Fi,j,t. One would expect higher probability for
the analyst to react immediately to the earnings announcement with higher AFEi,j,t.
In contrast to this expectation, responsive analysts’ forecasts are in fact more accurate
than non-responsive analysts, for both means and medians. Similarly, the absolute value of
forecast error for quarter t+1 before the quarter t earnings announcement, calculated as
AFEi,j,t+1a=Abs(Ej,t+1 - Fi,j,t+1
a)/P, is also significantly lower for responsive analysts than for
7 The average number of calendar days between fiscal period end and the first forecast revision by the non-responsive analysts is 59 days and the median is 53 days. Thus, considering that the SEC requires firms to file 10-Q forms within 45 calendar days after the fiscal period end, it is unlikely the lack of prompt reaction by non-responsive analysts is completely driven by the need for additional, comprehensive information from the 10-Q forms.
14
non-responsive analysts. These results suggest that the lack of reactions by the non-responsive
analysts are not because the earnings announcements convey less information with respect to
their prior information set or they have relative information advantage over the responsive
analysts before the earnings announcements.8 Further, AFEi,j,t+1b, the counterpart of AFEi,j,t+1
a
after the earnings announcements, is significantly lower for responsive analysts than for non-
responsive analysts. This suggests that the first forecasts of the responsive analysts since the
earnings announcements are more accurate than those of the non-responsive analysts, even
though on average the non-responsive analysts issue forecast revisions almost one month after
the earnings announcements, and potentially have access to additional information during this
period.
The results regarding the absolute forecast errors so far suggest that the lack of prompt
reaction by non-responsive analysts is not driven by less new information contained in the
earnings announcements relative to their prior information set. Instead, the results seem to
suggest that non-responsive analysts generally have lower ability to forecast earnings
accurately. This raises the possibility that non-responsive analysts’ forecasts are of lower
quality (e.g., Cooper et al. 2001) and that their lack of prompt reaction is due to their lower
ability to understand the implications of current earnings for future earnings.
I also examine the signed forecast errors, FEi,j,t, conditional on analysts’
responsiveness. The evidence reported in Table 3 suggests that analysts are more likely to
revise their forecasts promptly when the earnings announcements convey “good news.” The
average forecast error is 0.0002 for responsive analysts but -0.0004 for non-responsive
8 For example, prior to the enactment of Regulation Fair Disclosure in October 2000, certain analysts may gain information advantage over other analysts because of selective disclosures by managers.
15
analysts. This is consistent with Hong, Lim, and Stein (2000) that the market seems to
underreact more to “bad news”—“bad news travels slowly.” Untabulated results show that
34% of the non-responsive analyst-firm-quarters have negative earnings surprises, versus 27%
of the responsive analyst-firm-quarters.
Another analyst-specific variable reported in Table 3 is analysts’ firm-specific
experience. Mikhail, Walther, and Willis (1997) suggest that analysts’ forecast accuracy
improves as they gain firm-specific experience. In addition, Mikhail, Walther, and Willis
(2003a) find that analysts with longer firm-specific experience underreact to prior information
to a lesser extent. Accordingly, I next examine EXPi,j,t, measured as the number of quarters
that the analyst has followed the firm by quarter t. Consistent with these studies, on average,
responsive analysts have 9-quarter firm-specific experience, longer than the 8-quarter firm-
specific experience of the non-responsive analysts. The difference is statistically significant.
Finally, I examine three firm-specific variables including age, market capitalization,
and analyst coverage. Prior studies suggest that firms with older age have lower information
uncertainty and lower price momentum (Jiang, Lee, and Zhang 2004), and that the market
incorporates information more efficiently for larger firms and firms with greater analyst
coverage (e.g., Elgers, Lo, and Pfeiffer 2001; Foster, Olsen, and Shevlin 1984; Hong et al.
2000). To the extent that analysts’ responsiveness is associated with the information
efficiency of the firm, one would expect that firms followed by responsive analysts are older,
larger, and followed by more analysts. The descriptive statistics in Table 3 show that
consistent with expectations, firms followed by responsive analysts have significantly higher
analyst coverage (NUMANAj,t) and larger market capitalization (LOGMVj,t) than those
16
followed by non-responsive analysts.9 However, analysts seem to be more responsive to
younger firms as opposed to older firms (AGEj,t).
3.2. Magnitude of Analyst Forecast Revisions
The previous subsection examines how promptly analysts revise their earnings
forecasts for the next quarter. In this subsection, I focus on the extent to which analysts
incorporate the news in the earnings announcements to their forecast revisions. Prior studies
suggest that market reactions to earnings announcements are correlated with the “surprises” or
new information contained in the announcements (e.g., Fried and Givoly 1982). To the extent
that market reactions reflect investors’ revisions of expectations of future earnings, one would
expect that analysts’ forecast revisions are also correlated with the earnings surprises.
Consistent with this, Easterwood and Nutt (1999) find that analysts forecast revisions are
correlated with their previous forecast errors. However, the measure of forecast revisions in
their study is relatively “stale,” as it is based on the changes of analysts’ consensus forecasts
from eight months before to four months after the earnings announcements.
In the context of the current study, I am specifically interested in the effects of quarter
t earnings surprises on analysts’ forecast revisions for quarter t+1, and in particular, whether
the magnitude of this effect varies with analysts’ responsiveness. If an analyst responds more
promptly after an earnings announcement in revising her forecast for future earnings, it is
probably because she understands better the implications of current earnings for future
9 Note that the descriptive statistics of analysts following are based on per analyst-firm-quarter, as opposed to per firm-quarter.
17
earnings, as discussed in the previous subsection. Thus, her responsiveness is expected to be
indicative of the extent to which she incorporates the information into her forecasts revision.
The forecast revision REVi,j,t+1 is as defined in Table 1. I use FEi,j,t, the analyst-
specific forecast error for quarter t, to measure the news in the earnings announcements to the
analyst, since different analysts may have different information set and probably an analyst’s
individual information set is best reflected in her own forecast. I investigate the relation
between forecast revisions and the news in the earnings announcements by estimating the
following model:
REVi,j,t+1 = β0 + β1FEi,j,t + β2FEi,j,t × RESPi,j,t + β3RESPi,j,t + β4FEi,j,t × EXPi,j,t
+ β5EXPi,j,t + β6FEi,j,t × NUMANAj,t + β7NUMANAj,t (1)
I start the analysis by estimating a base model which simply regresses REVi,j,t+1 on
FEi,j,t with the pooled sample. The results are reported in Column (1) of Table 4. The
coefficient on FEi,j,t is significantly positive with a magnitude of 0.486, suggesting analysts
revise their forecasts in accordance with the new information conveyed in the earnings
announcements. The next two columns in Table 4 report this regression estimated for
responsive analysts and non-responsive analysts separately. For responsive analysts, the
coefficient on FEi.j,t is significant at 0.528, and the R-squared is 28.93%. In contrast, the
coefficient is only 0.457 and R-squared 22.44% for non-responsive analysts.
To test the statistical significance of the difference in the coefficient on FEi,j,t, I next
include an interaction term between FEi,j,t and RESPi,j,t, allowing FEi,j,t to have different
coefficients conditional on analysts’ responsiveness. The results are reported in Column (4).
The interaction term has a significantly positive coefficient of 0.071, suggesting the difference
18
between the extent to which responsive and non-responsive analysts incorporate the news into
their revisions is statistically significant.
Finally, I estimate the complete version of model (1) which controls for the effects of
the analyst’s firm-specific experience and the firm’s analyst coverage on the correlation
between REVi,j,t+1 and FEi,j,t. It is necessary to include these control variables because Table 3
shows that analysts’ responsiveness are correlated with these two variables and prior studies
suggest these variables are associated with the information efficiency of the security market
(e.g., Mikhail, Walther, and Willis 2003b; Elgers et al. 2001).10 The results, presented in the
last column of Table 4, show that RESPi,j,t continues to have a significantly positive, albeit
somewhat smaller, effect on the relation between analysts’ forecast revisions and earnings
surprises, suggesting that this effect is not subsumed by firm-specific experience or analyst
coverage. On the other hand, EXPi,j,t has an insignificant effect, while NUMANAj,t has a
significantly positive effect as expected.
4. Does Analysts’ Responsiveness Affect Market Underreaction?
4.1. Analysts’ Responsiveness and Analysts’ Underreaction
In this Section, I explicitly test whether delayed analysts’ response (i.e., lack of
analysts’ responsiveness) contributes to market underreaction. I start by examining whether
analysts’ responsiveness mitigates analysts’ underreaction to the earnings announcements,
where analysts’ underreaction is measured by the serial correlation in their forecast errors.
10 I include analysts’ coverage, but not firm size, as a control variable because prior research has shown that analysts’ coverage is highly correlated with firm size. However, including firm size as an additional control variable in all my tests does not qualitatively alter the implications of my empirical results.
19
The results in Section 3 suggest that in comparison to those of non-responsive analysts,
responsive analysts’ forecast revisions are associated with new information in earnings
announcements to a greater extent. However, this result per se does not necessarily imply that
the responsive analysts underreact to a lesser extent than do non-responsive analysts. As
Easterwood and Nutt (1999) point out, to test whether analysts systematically underreact (or
overreact) to news in earnings announcements about future earnings, one needs to benchmark
analysts’ reactions against the true earnings innovation series. To achieve this, one needs to
examine the autocorrelation in analysts’ forecasts errors. If the analyst fully understands the
implications of current earnings surprise for future earnings and instantaneously adjusts her
forecast for future earnings accordingly, her forecast errors should not be autocorrelated.
A number of studies find that analysts’ forecast errors are correlated with prior
information, suggesting they underreact to that information. For example, Abarbanell (1991)
finds that analysts’ forecast errors are positively correlated with prior returns and surprises in
recent earnings announcements (see also, e.g., Mendenhall 1991). Abarbanell and Bernard
(1992) specifically examine if the post-earnings-announcement drift is caused by analysts’
underreaction to the earnings announcements. They present evidence that analysts’ forecast
errors exhibit positive serial correlation, but conclude that analysts’ underreaction is at best
only a partial explanation for stock price underreaction to earnings. Finally, Mikhail et al.
(2003a) find that analysts underreact less to prior information as they gain firm-specific
experience and Mikhail et al. (2003b) show that this effect mitigates the post-earnings-
announcement drift.
As Easterwood and Nutt (1999) suggest, the literature on security analysts defines
“forecast efficiency” as analysts accurately incorporating new information on a timely and
20
unbiased fashion. Research on analysts’ underreaction such as that discussed about, however,
has largely ignored the timeliness perspective of the efficiency of analysts’ forecasts. For
example, Abarbanell and Bernard (1992) measure the serial correlation in analysts forecast
errors only as of the last forecast revisions before next quarter’s earnings announcements.
While this research design is capable of documenting the extent of analysts’ underreaction as
of the measurement date, if any, it fails to capture analysts’ immediate reaction to earnings
announcement and the possibility that the extent of analysts’ underreaction may be correlated
with their responsiveness.
I incorporate the timeliness perspective by focusing on the extent of analysts’
underreaction immediately after the earnings announcements. I measure analysts’
expectations of future earnings immediately after the earnings announcements using Fi,j,t+1,
their forecasts outstanding at the end of the second trading day after the announcements.
Specifically, Fi,j,t+1 equals to Fi,j,t+1b for responsive analysts and Fi,j,t+1
a for non-responsive
analysts. Since non-responsive analysts have not updated their forecasts, their latest forecasts
prior to the earnings announcements are still considered valid. I examine the autocorrelation
in analyst forecast errors using the following model:
FEi,j,t+1 = β0 + β1FEi,j,t + β2FEi,j,t × RESPi,j,t + β3RESPi,j,t + β4FEi,j,t × EXPi,j,t
+ β5EXPi,j,t + β6FEi,j,t × NUMANAj,t + β7NUMANAj,t (2)
where FEi,j,t+1=(Ej,t+1 - Fi,j,t+1)/P. Other variables are as defined previously.
As in Section 3.2, I first estimate a base model which regresses FEi,j,t+1 on FEi,j,t. The
coefficient on FEi,j,t is 0.82, indicating significant underreaction, on average, by analysts at the
end of second trading day after the earnings announcements. Prior studies that use the latest
analyst forecasts before the next quarter’s earnings announcements typically report a
21
coefficient around 0.15.11 The decrease in the serial correlation from immediately after
quarter t’s earnings announcements to immediately before quarter t+1’s earnings
announcements suggests that analysts only slowly incorporate implications of current
earnings for future earnings into their forecasts over time, again consistent with the “delayed
response” explanation for market underreactions.
A more striking result emerges when I estimate the model for the two types of analysts
separately. Columns (2) and (3) in Table 5 show that for non-responsive analysts, the
autocorrelation is as high as 0.844, while for responsive analysts, it is only 0.369, less than
half of that for non-responsive analysts. The R-squared is also in sharp contrast: it is almost
21% for non-responsive analysts, more than double that for responsive analysts (8.61%). This
result suggests that the responsive analysts’ prompt forecast revisions significantly mitigate
the correlation of their forecast errors immediately after the earnings announcements.
I next estimate the model after adding the interaction terms of FEi,j,t with RESPi,j,t, and
subsequently, the interaction terms of FEi,j,t with EXPi,j,t and NUMANAj,t respectively. The
results appear in the last two columns in Table 5. Consistent with expectations, the coefficient
on the interaction term of FEi,j,t and RESPi,j,t is significantly negative, with or without the
control variables. When the model controls for the effects of analysts’ firm-specific
experience and analysts’ coverage, non-responsive analysts’ forecast errors have a serial
correlation of 0.809, in comparison to 0.327 (=0.809-0.482) for responsive analysts. In other
words, immediately (i.e., two trading days) after the earnings announcements, the serial
correlation in forecast errors for responsive analysts’ is only 40% of that for non-responsive
11 For example, Abarbanell and Bernard (1992) report a coefficient of 0.18 while Mikhail et al. (2003) report a coefficient of 0.14.
22
analysts, suggesting responsive analysts’ forecasts better incorporate the implications of
current earnings for future earnings. Regarding the control variables, inconsistent with
Mikhail et al. (2003a), the coefficient on FEi,j,t × EXPi,j,t is significantly positive.12 In addition,
the effect of NUMANAj,t on the degree of underreaction is negative, albeit insignificant.
The above analyses are performed at the analyst level; that is, the unit of observation
is analyst-firm-quarter specific. Since the post-earnings-announcement drift is necessarily
examined at the firm level, I also repeat the analyses above using firm level data. Specifically,
I estimate the following regression:
MFEj,t+1 = β0 + β1MFEj,t + β2MFEj,t × FRESPj,t + β3FRESPj,t + β4MFEj,t × FEXPj,t
+ β5FEXPj,t + β6MFEj,t × NUMANAj,t + β7NUMANAj,t (3)
where MFEj,t is the firm-quarter mean of FEi,j,t (MFEj,t+1 is calculated likewise). Following
Mikhail et al. (2003b), FEXPj,t is the firm-quarter median of EXPi,j,t, where EXPi,j,t is as
defined previously. FRESPj,t and NUMANAj,t are as defined previously.
The results, presented in Panel B of Table 5, are generally consistent with those
reported in Panel A. In Column (1), before the inclusion of the interaction terms, MFEj,t has a
positive coefficient of 0.722 for the pooled sample. In Columns (2) and (3), without or with
control variables respectively, the coefficient on the interaction term of MFEj,t and FRESPj,t is
significantly negative. The results suggest that at the firm level, the autocorrelation in mean
analyst forecast errors after the earnings announcements decreases as the percentage of
responsive analysts following the firm increases. In fact, based on Column (3), for firms
12 There are at least two differences in sample selection and research design between the Mikhail et al. paper and the current paper. First, Mikhail et al. use analyst forecast information from Zacks while I use analyst forecast information from I/B/E/S. Second, they measure the serial correlation in analyst forecast errors immediately before the next quarter’s earnings announcements while I measure the serial correlation immediately after current quarter’s earnings announcements.
23
followed by responsive analysts only, the autocorrelation in analyst forecast errors is only
42% of that for firms followed by non-responsive analysts only. Finally, the effect of
experience on the autocorrelation at the firm level is insignificantly positive, and the effect of
analyst coverage is significantly positive.
4.2. Analysts’ Responsiveness and Post-Earnings-Announcement Drift
The results presented in the previous subsection indicate that in comparison to non-
responsive analysts, responsive analysts underreact to earnings announcements to a lesser
extent. In other words, their forecasts at the end of two trading days after the earnings
announcements better reflect the implications of current earnings for future earnings. To the
extent that analyst forecasts mirror market expectations for future earnings and that the post-
earnings-announcement drift is caused by investors’ failure to promptly incorporate the
implications of current earnings for future earnings into stock prices (e.g., Bernard and
Thomas 1990), this result implies that the post-earnings-announcement drift would be lower
for firms followed by responsive analysts. I now explicitly test this prediction.
To be consistent with my definition of the event window, the post-earnings-
announcement period starts from the third trading after the earnings announcements.
Following prior studies (e.g., Liang 2003; Mikhail et al. 2003b), I focus on stock returns over
the sixty-trading-day period after the earnings announcements and measure stock returns over
this period using size-adjusted buy-and-hold returns (SARj,t), where the adjustment is based
on equally-weighted returns of NYSE / AMEX / NASDAQ firm size decile to which the firm
belongs at the beginning of the calendar year.
24
I estimate the following regression to examine the effect of analysts’ responsiveness
on the magnitude of the post-earnings-announcement drift.
SARj,t = β0 + β1RMFEj,t + β2RMFEj,t × FRESPj,t + β3FRESPj,t + β4MFEj,t × FEXPj,t
+ β5FEXPj,t + β6MFEj,t × NUMANAj,t + β7NUMANAj,t (4)
To minimize problems associated with outliers as well as non-linearity, I follow prior
literature to use deciles of MFEj,t, namely, RMFEj,t, as opposed the raw MFEj,t, in the
regression (e.g., Bernard and Thomas 1990; Bartov, Radhakrishnan, and Krinsky 2000; Doyle
et al. 2003). Specifically, I rank MFEj,t by fiscal quarters into ten deciles indexed from 0 to 9
and then divide the index by 9 so that RMFEj,t, the ranked surprise, ranges between 0 and 1.
Thus, the coefficient on RMFEj,t can be readily interpreted as the size-adjusted return one can
earn by taking a long position in the highest decile and a short positive in the lowest decile.
Other variables are as defined in the previous sections.
The results are presented in Table 6. Column (1) reports the pooled regression with the
base model which regresses SARj,t on RMFEj,t. The coefficient on RMFEj,t is 0.048,
indicating that on average, one can earn about 4.8% size-adjusted return during the sixty
trading days after the earnings announcements by taking the “post-earnings-announcement
trading strategy” with the sample used in this paper. The next column allows the coefficient
on RMFEj,t to vary with the percentage of responsive analysts that follow the firm. The
coefficient on RMFEj,t is significantly positive at 0.055, whereas that on RMFEj,t × FRESPj,t
is significantly negative at -0.022. This suggests that for firms that are followed by responsive
analysts only, the post-earnings-announcement drift is about 40% lower than for firms that are
followed by non-responsive analysts only.
25
The last column confirms that this result is not subsumed by the effects of analysts’
firm-specific experience or analyst coverage. The results suggest that after controlling for
these effects, while the drift for firms that are followed by non-responsive analysts only can
be as high as 6.2%, it is more than one third lower if all analysts who follow the firm are
responsive to the earnings announcements in revising their forecasts. In terms of the control
variables, unlike the results presented in previous sections regarding EXPi,j,t or FEXPj,t, here
FEXPj,t has a significantly negative effect on the magnitude of the post-earnings-
announcement drift, consistent with the results in Mikhail et al. (2003b). Similarly,
NUMANAj,t also has a significantly negative effect, suggesting that the post-earnings-
announcement drift is smaller when the firm has a greater number of analysts following.
In sum, the results in Table 6 show that the post-earnings-announcement drift
decreases as the percentage of responsive analysts following the firm increases. This finding
supports the hypothesis that delayed responses by investors and/or analysts contribute to the
post-earnings-announcement drift.
5. Conclusion
Recently, a number of studies provide behavioral explanations for market
underreactions or overreactions. A common view of the behavioral explanations for
underreaction anomalies such as post-earnings-announcement drift proposes that investors are
slow in updating their expectations upon receiving information, i.e., the “delayed response”
hypothesis. Few studies, however, have provided direct evidence to support this view.
This study addresses this issue by focusing specifically on the speed at which analysts
respond to quarterly earnings announcements in revising their earnings forecasts for future
26
quarters. To directly test the validity of the “delayed response” hypothesis as an explanation
for market underreaction, I also examine the effects of analysts’ responsiveness on market
underreaction to earnings announcements. The results show that more than half of analysts do
not react to earnings surprises instantaneously as the efficient market hypothesis predicts.
Further, analysts’ responsiveness mitigates both the extent of the positive serial correlation in
analyst forecast errors after the earnings announcements and the extent of the post-earnings-
announcement drift. Together, these results provide direct support for the “delayed response”
hypothesis and suggest it is indeed possible that investors’ cognitive incompetence leads to
underreaction anomalies such as the post-earnings-announcement drift.
Given the sophistication levels of security analysts, it is somewhat surprising that a
considerable number of analysts do not promptly adjust their earnings expectations upon the
earnings announcements. Additional analyses in this study suggest that the lack of prompt
reaction by the non-responsive analysts is not because they have information advantage over
the responsive analysts. On the contrary, they seem to in general have lower ability to forecast
accurately and to understand the implications of current earnings for future earnings. This
result sheds light on the importance to examining the responsiveness aspect of analysts
information processing, suggesting that analysts’ responsiveness can potentially be a good
indicator of their overall forecasting abilities. Future work could investigate economic factors
as well as behavioral factors that drive analysts’ responsiveness.
27
REFERENCES
Abarbanell, Jeffery S., 1991, Do analysts’ forecasts incorporate information in prior stock price changes? Journal of Accounting and Economics 14, 147-165.
Abarbanell, Jeffery S., and Victor L. Bernard, 1992, Tests of analysts’ overreaction/ underreaction to earnings information as an explanation for anomalous stock price behavior, Journal of Finance 47, 1181-1207.
Bagnoli, Mark, Stanley, Levine, and Susan G. Watts, 2004, Analyst estimate revision clusters and corporate events. Working paper, Purdue University.
Ball, Ray, and Phillip Brown, 1968, An empirical evaluation of accounting income numbers, Journal of Accounting Research 6, 159-178.
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor sentiment, Journal of Financial Economics 49, 307-343.
Bartov, Eli, Suresh Radhakrishnan, Itzhak Krinsky, 2000, Investor sophistication and patterns in stock returns after earnings announcements, The Accounting Review 75, 43-63.
Bernard, Victor L., and Jacob K. Thomas, 1989, Post-earnings-announcement drift: Delayed price response or risk premium? Journal of Accounting Research 27, 1-36.
Bernard, Victor L., and Jacob K. Thomas, 1990, Evidence that stock prices do not fully reflect the implications of current earnings for future earnings, Journal of Accounting and Economics 13, 305-340.
Chan, Louis K. C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, Journal of Finance 51, 1681-1713.
Conrad, Jennifer, Bradford Cornell, and Wayne R. Landsman, 2002, When is bad news really bad news? The Journal of Finance 57, 2507-2532.
Cooper, Rick A., Theodore E. Day and Craig M. Lewis, 2001, Following the leader: A study of individual analysts’ earnings forecasts, Journal of Financial Economics 61, 383-416.
Doyle, Jeffrey, Russell Lundholm, and Mark Soliman, 2003, The extreme future stock returns following extreme earnings surprises, Working paper, University of Michigan.
Easterwood, John C., and Stacey R. Nutt, 1999, Inefficiency in analysts’ earnings forecasts: Systematic misreaction or systematic optimism? Journal of Finance 54, 1777-1797.
Elgers, Pieter T., May H. Lo, and Ray J. Pfeiffer Jr., 2001, Delayed security price adjustments to financial analysts’ forecasts of annual earnings, The Accounting Review 76, 613-632.
28
Fama, Eugene F., 1998, Market efficiency, long-term returns, and behavioral finance, Journal of Financial Economics 49, 283-306.
Fama, Eugene F., and Kenneth French, 1997, Industry costs of equity, Journal of Financial Economics 43, 153-193.
Foster, George, Chris Olsen, and Terry Shevlin, 1984, Earnings releases, anomalies, and the behavior of securities returns, The Accounting Review 59, 574-603.
Francis, Jennifer, Katherine Schipper, and Linda Vincent, 2002, Expanded disclosures and the increased usefulness of earnings announcements, The Accounting Review 77, 515-546.
Fried, Dov, and Dan Givoly, 1982, Financial analysts' forecasts of earnings: A better surrogate for market expectations, Journal of Accounting and Economics 4, 85-107.
Hong, Harrison, Terence Lim, and Jeremy C. Stein, 2000, Bad news travels slowly: Size, analyst coverage, and the profitability of momentum strategies, Journal of Finance 55, 265-295.
Hong, Harrison, and Jeremy C. Stein, 1999, A unified theory of underreactions, momentum trading and overreaction in asset markets, Journal of Finance 54, 2143-2184.
Ivković, Zoran, and Narasimhan Jegadeesh, 2004, The timing and value of forecast and recommendation revisions, Journal of Financial Economics 73, 433-463.
Jiang, Guohua, Charles M. C. Lee, and Grace Y. Zhang, 2004, Information uncertainty and expected returns, Working paper, Peking University and Cornell University.
Liang, Lihong, 2003, Post-earnings-announcement drift and market participants’ information processing biases, Review of Accounting Studies 8, 321-345.
Mendenhall, Richard R., 1991, Evidence on the possible underweighting of earnings-related information, Journal of Accounting Research 29, 170-179.
Mikhail Michael B., Beverly R. Walther, and Richard H. Willis, 1997, Do security analysts improve their performance with experience? Journal of Accounting Research 35, 131-157.
Mikhail Michael B., Beverly R. Walther, and Richard H. Willis, 2003a, The effect of experience on security analyst underreaction, Journal of Accounting & Economics 35, 101-116.
Mikhail Michael B., Beverly R. Walther, and Richard H. Willis, 2003b, Security analyst experience and post-earnings-announcement drift, Journal of Accounting, Auditing & Finance 18, 529-550.
29
Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix, Econometrica 55, 703-708.
Richardson, Scott, Siew Hong Teoh, and Peter D. Wysocki, forthcoming, The walkdown to beatable analyst forecasts: The roles of equity issuance and insider trading incentives, Contemporary Accounting Research.
SAS Institute Inc., 1989. SAS/STAT User’s Guide (SAS Institute Inc., Cary, NC.).
Schipper, Katherine, 1991, Analysts’ forecasts, Accounting Horizons 5, 105-121.
Stickel, Scott E., 1989, The timing of and incentives for annual earnings forecasts near interim earnings announcements, Journal of Accounting and Economics 11, 275-292.
30
Figure 1. Timeline of information events. Ej,t is the actual earnings per share for firm j quarter t. Fi,j,t is analyst i’s most recent forecast for firm j quarter t before the corresponding earnings announcement. Fi,j,t+1
a is analyst i’s most recent forecast for firm j quarter t+1 issued before the announcement of Ej,t. Fi,j,t+1
b is analyst i’s first forecast for firm j quarter t+1 issued after the announcement of Ej,t. Ej,t+1 is the actual earnings per share for firm j quarter t+1.
Latest Forecast for Quarter t: F i,j,t
Latest Forecast before day 0 for Quarter t+1: F i,j,t+1
a
First Forecast after day 0 for Quarter t+1: F i,j,t+1
b
Day 0: Earnings Announcement for Quarter t: E j,t
Earnings Announcement for Quarter t+1: E j,t+1
31
0
20
40
60
80
100
0 10 20 30 40 50 60 70 80 90
Number of Calendar Days After the Earnings Announcements
Cum
ulat
ive
Perc
enta
ge
Figure 2. Timing of first forecast revision after the earnings announcements.
32
Table 1. Descriptive Statistics
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. FEi,j,t=( Ej,t - Fi,j,t)/P and REVi,j,t+1=(Fi,j,t+1
b - Fi,j,t+1a)/P, where P is
firm j’s stock price at the end of fiscal quarter t. See Figure 1 for timeline and definitions of Ej,t, Fi,j,t, Fi,j,t+1
b, and Fi,j,t+1a.
# calendar days
[Ej,t, Fi,j,t+1b]
FEi,j,t × 100 REVi,j,t+1 × 100 Year # analyst- firm-quarter Mean Median Mean Median Mean Median
1988 4,485 33.18 27 -0.12 0.00 -0.11 -0.03 1989 5,531 30.41 23 -0.23 -0.06 -0.28 -0.11 1990 7,210 30.02 23 -0.22 -0.04 -0.40 -0.13 1991 9,807 30.17 24 -0.13 -0.02 -0.30 -0.10 1992 11,494 27.21 20 -0.09 0.00 -0.19 -0.06 1993 12,465 25.33 17 -0.08 0.00 -0.17 -0.06 1994 19,024 26.45 16 0.00 0.04 -0.12 -0.03 1995 21,763 24.01 13 -0.03 0.03 -0.17 -0.05 1996 23,547 23.61 11 -0.02 0.03 -0.19 -0.04 1997 27,196 22.73 8 0.00 0.03 -0.16 -0.04 1998 33,430 20.59 6 -0.03 0.02 -0.26 -0.06 1999 35,489 17.64 4 0.02 0.04 -0.18 -0.02 2000 36,236 17.02 3 0.03 0.04 -0.20 -0.03 2001 42,091 13.88 3 0.01 0.03 -0.26 -0.09 2002 a 37,316 12.69 2 0.05 0.05 -0.19 -0.05 Over All 327,084 20.32 7 -0.01 0.03 -0.21 -0.05
a Year 2002 has observations for three quarters only, due to the requirement for quarter t+1 information.
33
Table 2. Analysts’ Responsiveness to Earnings Announcements
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. FEi,j,t=( Ej,t - Fi,j,t)/P, where P is firm j’s stock price at the end of fiscal quarter t. RESPi,j,t equals 1 if Fi,j,t+1
b is issued within two trading days after the announcement of Ej,t and 0 otherwise. FRESPj,t is the percentage of analysts with RESPi,j,t=1 among all analysts following firm j for the quarter t. See Figure 1 for timeline and definitions of Ej,t, Fi,j,t, Fi,j,t+1
a, and Fi,j,t+1b. Industries are
classified in accordance to Fama and French (1997). Panel A: Descriptive Statistics of Number of Calendar Days between Et and Ft+1
Mean Median STD P25 P75 All observations 20 7 25 2 31 Excluding obs. w/ FEi,j,t=0 20 7 24 2 30 Panel B: Analysts’ Responsiveness to Earnings Announcements—Analyst Level
RESPi,j,t=1 RESPi,j,t=0 All observations 43.67% 56.33% Excluding obs. w/ FEi,j,t=0 43.82% 56.18% Industries with highest percentage of responsive analysts a Electronic Equipment (8.29%) 56.01% 43.99% Measuring and Control Equipment (1.54%) 52.65% 47.35% Business Services (12.54%) 51.33% 48.67% Electrical Equipment (2.34%) 50.73% 49.27% Industries with lowest percentage of responsive analysts Chemicals (2.36%) 34.50% 65.50% Trading (3.49%) 37.51% 62.49% Automobiles and Trucks (1.72%) 37.72% 62.28% Food Products (1.37%) 38.15% 61.85%
Panel C: Analysts’ Responsiveness to Earnings Announcements—Firm Level FRESPj,t=1 0<FRESPj,t<
1FRESPj,t=0
All observations 18.42% 35.17% 46.41% a Only industries with at least one percent of analyst-firm-quarters in the sample that have highest or lowest percentage of responsive analysts are reported. Numbers in parentheses are the percentage of observations represented in the corresponding industries.
34
Table 3. Descriptive Statistics Conditional on Analysts’ Responsiveness
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. FEi,j,t=(Ej,t - Fi,j,t)/P, AFEi,j,t=Abs(FEi,j,t), AFEi,j,t+1
a=Abs(Ej,t+1 - Fi,j,t+1
a)/P, and AFEi,j,t+1b=Abs(Ej,t+1 - Fi,j,t+1
b)/P, where P is firm j’s stock price at the end of fiscal quarter t. RESPi,j,t equals 1 if Fi,j,t+1
b is issued within two trading days after the announcement of Ej,t and 0 otherwise. EXPi,j,t is the number of quarters that analyst i has been following firm j by quarter t. AGEj,t is the number of years firm j has been included in the CRSP database as of the end of quarter t. LOGMVj,t is the log of market value of firm j as of the end of quarter t. NUMANAj,t is the number of analysts following firm j for quarter t. See Figure 1 for timeline and definitions of Ej,t, Fi,j,t, Fi,j,t+1
a, and Fi,j,t+1
b. Numbers in parentheses are two-sided p-values. For means, it is from t-test; for medians, it is from Wilcoxon test.
Mean Median RESPi,j,t=1 RESPi,j,t=0 RESPi,j,t=1 RESPi,j,t=0 # calendar days [Et, Ft+1
b] 2.50 34.13 2.00 27.00 (0.00) (0.00) AFEi,j,t 0.0025 0.0030 0.0010 0.0011 (0.00) (0.00) AFEi,j,t+1
a 0.0065 0.0075 0.0023 0.0027 (0.00) (0.00) AFEi,j,t+1
b 0.0045 0.0050 0.0013 0.0014 (0.00) (0.00) FEi,j,t 0.0002 -0.0004 0.0004 0.0002 (0.00) (0.00) EXPi,j,t 8.51 7.62 5.00 5.00 (0.00) (0.00) AGEj,t 15.64 16.12 11.00 13.00 (0.00) (0.00) LOGMVj,t 14.42 13.98 14.35 13.91 (0.00) (0.00) NUMANAj,t 7.60 5.56 6.00 4.00
(0.00) (0.00)
35
Table 4. Analysts’ Responsiveness and Forecast Revisions
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. FEi,j,t= (Ej,t - Fi,j,t)/P and REVi,j,t+1=(Fi,j,t+1
b - Fi,j,t+1a)/P,
where P is firm j’s stock price at the end of fiscal quarter t. RESPi,j,t equals 1 if Fi,j,t+1b is issued
within two trading days after the announcement of Ej,t and 0 otherwise. EXPi,j,t is the number of quarters that analyst i has been following firm j by quarter t. NUMANAj,t is the number of analysts following firm j for quarter t. See Figure 1 for timeline and definitions of Ej,t, Fi,j,t, Fi,j,t+1
a, and Fi,j,t+1b. All regressions are estimated after deleting observations with absolute
studentised residuals greater than 2. Numbers in parentheses are two-sided p-values. They are based on t-statistics adjusted using the Newey and West (1987) procedure with six lags.
REVi,j,t+1 = β0 + β1FEi,j,t + β2 FEi,j,t × RESPi,j,t + β3RESPi,j,t + β4 FEi,j,t × EXPi,j,t + β5EXPi,j,t + β6 FEi,j,t × NUMANAj,t + β7NUMANAj,t (1)
Predicted Sign
(1) Pooled
(2) RESP=1
(3) RESP=0
(4) Pooled
(5) Pooled
Intercept -0.001 (0.00)
-0.001 (0.00)
-0.002 (0.00)
-0.002 (0.00)
-0.002 (0.00)
FEi,j,t + 0.486 (0.00)
0.528 (0.00)
0.457 (0.00)
0.458 (0.00)
0.402 (0.00)
FEi,j,t × RESPi,j,t + 0.071 (0.00)
0.047 (0.00)
RESPi,j,t ? 0.000 (0.00)
0.000 (0.00)
FEi,j,t × EXPi,j,t + 0.001 (0.24)
EXPi,j,t ? -0.00 (0.80)
FEi,j,t × NUMANAj,t + 0.013 (0.00)
NUMANAj,t ? 0.00 (0.45)
Adj. R-Squared 25.06% 28.93% 22.44% 25.25% 25.45%
36
Table 5. Analysts’ Responsiveness and Analyst Underreaction
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. FEi,j,t= (Ej,t - Fi,j,t)/P, where P is firm j’s stock price at the end of fiscal quarter t. RESPi,j,t equals 1 if Fi,j,t+1
b is issued within two trading days after the announcement of Ej,t and 0 otherwise. FEi,j,t+1= (Ej,t+1 - Fi,j,t+1)/P, where Fi,j,t+1 equals Fi,j,t+1
b if RESPi,j,t equals 1 and Fi,j,t+1a
otherwise. EXPi,j,t is the number of quarters that analyst i has been following firm j by quarter t. NUMANAj,t is the number of analysts following firm j for quarter t. MFE j,t and MFE j,t+1 are firm-quarter means of FEi,j,t and FE i,j,t+1 respectively. FRESPj,t is the percentage of analysts with RESPi,j,t=1 among all analysts following firm j for the quarter t. FEXPj,t is median of EXPi,j,t for firm j quarter t. See Figure 1 for timeline and definitions of Ej,t, Ej,t+1, Fi,j,t, Fi,j,t+1
a, and Fi,j,t+1b. All regressions are
estimated after deleting observations with absolute studentised residuals greater than 2. Numbers in parentheses are two-sided p-values. They are based on t-statistics adjusted using the Newey and West (1987) procedure with six lags.
Panel A: Analyst Level Analysis
FEi,j,t+1 = β0 + β1FEi,j,t + β2FEi,j,t × RESPi,j,t + β3RESPi,j,t + β4FEi,j,t × EXPi,j,t + β5EXPi,j,t + β6FEi,j,t × NUMANAj,t + β7NUMANAj,t (2)
Predicted Sign
(1) Pooled
(2) RESP=1
(3) RESP=0
(4) Pooled
(5) Pooled
Intercept -0.002 (0.00)
-0.001 (0.00)
-0.003 (0.00)
-0.003 (0.00)
-0.002 (0.00)
FEi,j,t + 0.820 (0.00)
0.369 (0.00)
0.844 (0.00)
0.850 (0.00)
0.809 (0.00)
FEi,j,t × RESPi,j,t - -0.470 (0.00)
-0.482 (0.00)
RESPi,j,t ? 0.002 (0.00)
0.002 (0.00)
FEi,j,t × EXPi,j,t - 0.002 (0.00)
EXPi,j,t ? -0.000 (0.09)
FEi,j,t × NUMANAj,t - 0.006 (0.00)
NUMANAj,t ? -0.00 (0.16)
Adj. R-Squared 7.06% 8.61% 20.84% 19.73% 19.75%
37
Panel B: Firm Level Analysis
MFEj,t+1 = β0 + β1MFEj,t + β2MFEj,t × FRESPj,t + β3FRESPj,t + β4MFEj,t × FEXPj,t + β5FEXPj,t + β6MFEj,t × NUMANAj,t + β7NUMANAj,t (3)
Predicted Sign
(1) Pooled
(2) Pooled
(3) Pooled
Intercept -0.002 (0.00)
-0.003 (0.00)
-0.003 (0.00)
MFEj,t + 0.722 (0.00)
0.849 (0.00)
0.767 (0.00)
MFEj,t × FRESPj,t - -0.423 (0.00)
-0.442 (0.00)
FRESPj,t ? 0.002 (0.00)
0.002 (0.00)
MFEj,t × FEXPj,t - 0.003 (0.23)
FEXPj,t ? -0.000 (0.65)
MFEj,t × NUMANAj,t - 0.036 (0.00)
NUMANAj,t ? -0.000 (0.25)
Adj. R-Squared 16.03% 17.63% 17.77%
38
Table 6. Analysts’ Responsiveness and Post-Earnings-Announcement Drift
The sample includes 327,084 analyst-firm-quarters with fiscal period ending between 1988 and 2002, representing 103,681 firm-quarters. SARj,t is size-adjusted buy-and-hold returns over sixty trading days since the 3rd trading day after the earnings announcement of firm j for quarter t. FEi,j,t= (Ej,t - Fi,j,t)/P, where P is firm j’s stock price at the end of fiscal quarter t. RMEFt is the decile of MFEt ranked by quarter and ranges from 0 to 1, where MFE j,t is firm-quarter means of FEi,j,t. FRESPj,t is the percentage of analysts with RESPi,j,t=1 among all analysts following firm j for the quarter t, where RESPi,j,t equals 1 if Fi,j,t+1
b is issued within two trading days after the announcement of Ej,t and 0 otherwise. FEXPj,t is median of EXPi,j,t for firm j quarter t, where EXPi,j,t is the number of quarters that analyst i has been following firm j by quarter t. NUMANAj,t is the number of analysts following firm j for quarter t. See Figure 1 for timeline and definitions of Ej,t, Fi,j,t, Fi,j,t+1
b, and Fi,j,t+1a. All regressions are
estimated after deleting observations with absolute studentised residuals greater than 2. Numbers in parentheses are two-sided p-values. They are based on t-statistics adjusted using the Newey and West (1987) procedure with six lags.
SARj,t = β0 + β1RMFEj,t + β2RMFEj,t × FRESPj,t + β3FRESPj,t + β4MFEj,t × FEXPj,t + β5FEXPj,t + β6MFEj,t × NUMANAj,t + β7NUMANAj,t (4)
Predicted Sign
(1) Pooled
(2) Pooled
(3) Pooled
Intercept -0.037 (0.00)
-0.041 (0.00)
-0.047 (0.00)
RMFEj,t + 0.048 (0.00)
0.055 (0.00)
0.062 (0.00)
RMFEj,t × FRESPj,t - -0.022 (0.00)
-0.021 (0.00)
FRESPj,t ? 0.014 (0.00)
0.014 (0.00)
RMFEj,t × FEXPj,t - -0.004 (0.01)
FEXPj,t ? 0.001 (0.00)
RMFEj,t × NUMANAj,t - -0.001 (0.34)
NUMANAj,t ? -0.001 (0.01)
Adjusted R-Squared 0.74% 0.77% 0.88%