Page 1 of 43 Limited Attention, Analyst Forecasts, and Price Discovery Abdullah Shahid 1 Cornell University [email protected]Rajib Hasan 2 University of Houston-Clear Lake [email protected]***Preliminary version; please do not quote. *** ABSTRACT Post-earnings announcement drift (PEAD), i.e. stock price’s delayed response to earnings news, is a well-documented evidence of market inefficiency. One explanation for such phenomenon is that sell-side analysts, an important information intermediary in capital markets, fail to process earnings news. In this study, we investigate the limited attention of sell-side analysts to understand why analysts might fail to process earnings news and whether such limited attention has any implications for PEAD. Specifically, we examine the attention-limiting role of competing tasks (number of earnings announcements for firms in an analyst’s portfolio and number of earnings forecasts by an analyst within a short period) and distracting events (number of earnings announcements for firms outside an analyst’s portfolio within a short period) in influencing analysts’ forecast accuracy. Then, we test whether limited attention of analysts owing to competing tasks and distracting events affects firms’ price discovery after earnings news. Using a sample of 5,136 firms and 10,798 analysts over the period 2000-2012, we find the following. First, competing tasks worsen analysts’ forecast accuracy, whereas distracting events do not seem to affect analysts’ forecast accuracy. Second, competing tasks of analysts delay firms’ price adjustment process after earnings announcements. Overall, our findings suggest that context-bound (limited) attention of analysts has implications for their task performance as well as market efficiency. Key Words: analyst forecast, context-bound rationality, limited attention, market efficiency, post earnings announcement drift, price discovery 1 Abdullah Shahid acknowledges funding from the Institute for the Social Sciences’ “Creativity, Innovation, and Entrepreneurship” theme project, Cornell University. 2 Rajib Hasan acknowledges funding from the University of Texas-Dallas and comments from the workshop at UHCL business school.
43
Embed
Limited Attention, Analyst Forecasts, and Price Discovery Abdullah ...socialsciences.cornell.edu/wp-content/uploads/2015/11/Limited... · Abdullah Shahid1 Cornell University [email protected]
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1 of 43
Limited Attention, Analyst Forecasts, and Price Discovery
1 Abdullah Shahid acknowledges funding from the Institute for the Social Sciences’ “Creativity, Innovation, and Entrepreneurship” theme project, Cornell University. 2 Rajib Hasan acknowledges funding from the University of Texas-Dallas and comments from the workshop at UHCL business school.
Page 2 of 43
1.0 Introduction
Prior studies suggest that stock prices underreact to recent earnings news. The most prominent
evidence of such underreaction is post earnings announcement drift3 (PEAD) (e.g., Ball and Brown
1968; Bernard and Thomas 1989; Bernard and Thomas 1990; Chan, Jegadeesh, and Lakonishok
1996). One explanation for market-level underreaction is that analysts, an important information
intermediary for dissemination of earnings news, underreact to such news. Indeed, Abarnabell and
Bernard (1992) show that analysts’ underreaction could potentially explain about half of the
magnitude of the delayed stock price response to earnings news. However, it is not clear why
analysts fail to fully process earnings news and provide more precise earnings forecasts.
In this study, we examine whether and the extent to which analysts’ failure to process earnings
news and to provide accurate forecasts stems from their limited attention. Studies in social sciences
suggest that limited attention is a major reason why individuals might fail to process information.
Limited attention refers to a condition in which individuals fail to pay due attention to the necessary
stimuli in the environment. It could arise, among many possible others, in the following ways.
First, when individuals are faced with competing information processing tasks within a limited
amount of time, they cannot pay appropriate attention to all the tasks at hand (i.e. competing task
hypothesis). Hence, performance on all tasks is likely to suffer. Second, when there are a lot of
distracting stimuli during performance of a task, individuals might get distracted. Hence,
performance could suffer (i.e. distracting event hypothesis). Although analysts are sophisticated
information professionals, they are “decidedly human” and could suffer from processing
limitations that, in turn, affect other less sophisticated market participants (DeBondt and Thaler
1990). So, limited attention of analysts, considered vital actors in helping investors process and
interpret corporate information, could be an important driver of the market-wide underreaction in
stock prices to earnings news.
We test the competing task hypothesis and distracting event hypothesis in the following two
contexts of analysts’ forecast revisions around an earnings announcement respectively: (1) there
are competing earnings announcements for analysts’ portfolio firms4 and/or analysts perform many
3 Note that throughout the paper we use “stock price’s underreaction to earnings news” and PEAD interchangeably. 4 Portfolio firms of an analyst are the firms for which the analyst has provided at least one earnings forecast in the previous quarter. We use the words “cover” and “follow” interchangeably in this paper to mean that a firm is
Page 3 of 43
earnings forecasts in a day (or within a short period of time); (2) there are earnings announcements
from outside-portfolio firms, i.e. the firms not followed by an analyst. Using these two contexts
we address the following questions: (1) Does the quality (proxied by accuracy) of analysts’
earnings forecasts worsen with the presence of competing and distracting earnings
announcements? (2) Is underreaction to earnings news associated with the presence of competing
and distracting events for analysts?
The first context of analysts’ limited attention, which is used here to test the competing task
hypothesis, is when analysts face competing earnings announcements for their portfolio firms and
/or when analysts perform many earnings forecasts within a short period of time. The context
captures simultaneous processing of multiple earnings news. We expect competing tasks to affect
the quality of analyst forecasts negatively. The reason is that while attending to many tasks at a
time, analysts fail to pay attention to the relevant aspects of all the tasks. Hence, the situation would
lead to an analyst’s failing to properly react to earnings news or provide correct earnings forecasts.
In this context of competing tasks, we have the following specific predictions. First, there is a
negative relationship between competing tasks and analysts’ forecast accuracy. Second, there is a
positive association between the average competing tasks faced (and/or performed) by the analysts
covering a firm and stock price’s underreaction to earnings news of that firm.
The second context of limited attention, which is used here to test the distracting event
hypothesis, is when there are many earnings announcements by outside-portfolio firms on the day
an analyst’s portfolio firm announces earnings. The context may limit attention since it presents
analysts with many irrelevant stimuli. To facilitate flow of discussion, let us term “days with many
earnings announcements by other firms” as “high distraction days”. On high distraction days,
analyst attention could be limited in the following ways (Forster and Lavie 2008). First, distracting
(in other words, less-than-relevant) noise in the workplace could limit attention just like roadside
billboards might distract drivers from the task of driving (Wallace 2003; McEvoy, Stevenson and
Woodward 2007). Likewise, many earnings news on the same day could increase overall noise the
economic environment from which analysts collect information and thus, could distract their
attention away from the relevant task. Second, actions of colleagues could distract an individual.
incorporated in an analyst’s portfolio. Also, an analyst following a firm is called “following analyst” or “covering analyst”.
Page 4 of 43
High distractions days could increase overall activities and discussions in the analyst colleague
network inside and outside their offices. This could occupy some attention of analysts, disrupting
the rapt attention needed for analysis of relevant news. Our expectations are as follows. First,
analyst forecast accuracy is negatively related with the number of distracting earnings
announcements. Second, market-wide underreaction to earnings news of a firm is positively
associated with the average number of distracting news faced by analysts following the firm.
Using a sample of 5,136 firms and 10,798 analysts over the period from 2000 to 2012 we find
the following for the competing task hypothesis. First, the greater is the number of competing
earnings announcements for portfolio firms on a forecast day, the lower is the forecast accuracy of
an analyst. Moreover, the negative relationship between competing earnings news and forecast
accuracy is larger in magnitude for complicated (proxied by presence of multiple segments of
business) firms. This relationship between competing tasks and forecasting accuracy remains the
same in direction, even after controlling for various firm-specific, analyst-specific and earnings-
news specific factors, and year-quarter fixed effects. The result is robust to varying window of
competing tasks (such as the number of earnings news faced in the week preceding the forecast
day (inclusive)). To find out whether analysts can self-select and focus attention by limiting the
number of tasks to be performed on a particular forecast day, we use the following measures:
number of competing tasks performed on a forecast day and number of competing tasks performed
in the week prior to the forecast day (inclusive). Yet, the negative relationship between forecast
accuracy and competing tasks remain. Second, there is a statistically significant positive
relationship (p-value<0.0001) between the average competing tasks faced by following analysts
on a firm’s earnings announcement day and the delay in price discovery process of the firm
(proxied by absolute cumulative abnormal return in various windows following earnings
announcements of the firm). The result is robust to inclusion of various control variables from the
literature, firm-fixed effects, and quarter-fixed effects. Overall, results provide strong support for
the competing task hypothesis.
Using the same dataset, we find the following for the distracting event hypothesis. First, there
is no significant relationship (p-value>0.60) between the number of outside-portfolio earnings
announcements faced by an analyst on a forecast day and forecast error. The relationship remains
even after controlling for an interaction effect of firm complexity, various firm-specific, analyst-
Page 5 of 43
specific and earnings-news-specific factors, and year-quarter fixed effects. The results remain
qualitatively similar even after we use the number of earnings-announcements over a week (prior
to the forecast day, inclusive) as an alternative measure for distracting events. Second, there is no
significant relationship between average earnings-announcements by outside-portfolio firms of
analysts on an earnings announcement day and the delay in price discovery process of a firm). The
result is robust to inclusion of various control variables from literature, firm-fixed effects, and
quarter-fixed effects.
Our study contributes to asset pricing, empirical capital markets in accounting, and new
institutionalism5 in social sciences literature by illustrating that limited attention of analysts from
competing tasks lead to their forecasting inefficiency as well as information processing
inefficiency by the entire market (i.e. the investors in general). At this point, it is to be duly noted
that we do not have sufficient evidence for the distracting event hypothesis for analysts, while
Hirshleifer, Lim and Teoh (2009) show that distracting events (proxied by similar measures) lead
to limited attention of the market in general. However, the finding that the context of analysts
support the competing task hypothesis while providing no sufficient evidence for the distracting
event hypothesis, point, to some extent, to the merit of taking a closer look at the work context in
understanding limited attention of expert actors in the market. The approach encourages us to look
beyond attributing investors’ failure to process some information to analysts’ failure to process the
same information. Rather, we ask “why such failure by analysts” and to find answers we delve
deeper into the constraints from the work-context of analysts. Indeed, in a recent compendium of
scholarly debates, asset pricing scholars have argued that understanding institutional constraints is
where both schools (traditional and behavioral) can have amicable and productive conversations
(Bloomfield 2010).
Our study also contributes to the growing body of literature that examines the impact of limited
attention on financial markets. Hirshleifer and Teoh (2003) argue that limited attention makes it
difficult to process news that requires analysis in conscious thought. They analytically show that
due to presence of limited attention among investors, alternative ways of presenting corporate
5 In various branches of social sciences such as political science and sociology, new institutionalism calls for understanding actions of actors subject to various context-bound constraints (stemming from any formal and information institutional conditions). See Brinton and Nee (1998) for further discussion.
Page 6 of 43
information could affect price discovery differently. Empirical work finds that stock prices behave
as if investors underreact to earnings news released on Fridays as well as on the days with many
extraneous events (DellaVigna and Pollet 2009; Hirshleifer et al. 2009). Stock prices also do not
incorporate important information such as demographics (DellaVigna and Pollet 2007) and signals
in oil prices (Pollet 2005). In this paper, we provide an explanation for why prices underreact to
earnings news. We find that analysts significantly suffer from limited attention particularly in the
face of competing tasks and such limited attention of analysts has significant association with
market-wide underreaction to earnings news.
Our study is also related to the literature on efficiency of analyst forecasts. Abarnabell and
Bernard (1992) find analysts’ underreaction to past earnings information. Mikhail, Walther, and
Willis (1997) find that accuracy of analysts’ earnings forecasts improves with their firm-specific
forecasting experience. Jacob, Lys, and Neale (1999) show that analysts forecast accuracy is a
function of analyst aptitude, learning-by-doing, and internal environment (i.e. size of the brokerage
house). The contribution of our paper is that we show how limited attention from competing tasks,
a natural aspect of analysts’ task environment, could affect analysts’ efficiency in providing correct
earnings forecast. Moreover, we examine how limited attention of these sophisticated
professionals affects overall information processing efficiency of the market. Hence, in a sense,
we take the debate from behavioral economics and psychology that attention is a precious and
limited cognitive resource (Nisbett and Ross 1980; Kahneman 2013) from mere individuals to
professionals who are known as experts in information processing.
In the remainder of the paper we review the related literature and develop hypotheses (Section
2), discuss data and variables (Section 3), present methods and results (Section 4) and conclude
with summarizing contributions and limitations (Section 5).
2.0 Review of Literature and Development of Hypotheses
This section is organized in the following subsections. First, we briefly discuss the history of
finance literature (particularly, the asset pricing) from market efficiency to the rise of behavioral
finance. Second, we devote a subsection to discuss how new institutionalism perspectives
combined with traditional and behavioral finance, can be useful in further understanding the debate
of market efficiency. Third, we discuss limited attention of analysts and their earnings forecast, a
Page 7 of 43
case that we empirically examine in the paper. Fourth, we review the prior literature that discuss
the relationship between analyst forecast and price discovery. Fifth, we develop empirical
predictions for the competing task hypothesis and the distracting event hypothesis.
2.1 Market Efficiency, Anomalous Empirical Patterns and the Rise of Behavioral Finance
The neoclassical financial theory, based on assumptions of decision makers’ rationality, common
risk aversion, perfect markets with no frictions, and costless access to information for all market
participants has long dominated finance (Szyska 2010). The efficient market hypothesis (EMH)
and the capital asset pricing model (CAPM) are the foundational theories of the neoclassical
financial theory. EMH argues that market is informationally efficient, i.e. one cannot consistently
attain return in excess of market, since prices already contain all relevant information. The capital
asset pricing model suggest a theoretically appropriate required rate of return on a stock is
determined by its beta, i.e. sensitivity of the stock’s return to a non-diversifiable risk (i.e. the
market risk).
However, finance researchers started finding patterns in stock prices that are not consistent
with the EMH. These patterns are often called “anomalies” in the asset pricing literature. Some
examples of such findings follow. The January effect (Reinganum 1983) shows that much of the
abnormal return to small firms occur during the first two weeks in January. Jaegadeesh and Titman
(1993) find that returns on stock are significantly correlated over three to twelve month time
horizon. These findings suggest that excess returns can be earned using even widely available
information like price. Researchers further show that based on public information such as firm
characteristics (size, book to market ratios, growth in sales, earnings to price ratio, accruals, asset
growth) it is possible to form portfolios that can earn returns in excess of the market (or in excess
of return suggested by the CAPM) (Fama and French 1992; Lakonishaok, Shleifer and Vishny
1994; Sloan 1996). The central anomaly discussed in this paper, PEAD is another such anomalous
empirical pattern. These findings raised serious doubts about the EMH.
At the same time, findings by behavioral economists suggest that the assumptions of decision
maker’s rationality in the neoclassical financial theory is false. Limited in cognitive capacity,
humans are rather found to be relying on various shortcuts and heuristics. Drawing on these
insights from behavioral economics and psychology (such as Daniel Kahneman, Amos Tversky,
Page 8 of 43
Richard Thaler, and Paul Slovic), some finance researchers started finding biases and irrational
behavior in stock prices, forming the field of behavioral finance. In the early papers of behavioral
finance, for example, DeBondt and Thaler (1985) show investors’ overreaction to news and
Shefrin and Statman (1985) provide evidence that prices show a disposition effect.
From its birth in early 1980s, the field of behavioral finance grew by combining behavioral
and cognitive psychology with conventional economics and finance. Baker and Nofsinger (2010)
summarize the findings of this field into four areas: huristics, framing, emotions, and market
impact. The behavioral finance research in huristics shows that actors use cognitive short-cuts or
rules of thumbs in financing decision making, with some famous huristics being
representativeness, availability, anchoring and adjustments, status quo, loss and regret aversion,
conservatism, ambiguity aversion, and mental accounting. The research in framing shows that if
the same problem is stated differently financial decision makers react differently even though
objective facts of the problem are held constant. The behavioral finance research in emotions show
that emotions such as confidence, illusions about the nature of money, and sense of unfairness play
an important role in people’s financial decisions. The behavioral finance on market impact
examine if cognitive errors and biases of individuals and groups affect market prices in aggregate.
But, one might wonder why individuals and groups suffer from these biases and errors. To
understand this “why” question in greater details, a new institutional perspective can be helpful,
since it would seek to understand the context of the actors in the market. That’s exactly where new
institutionalism perspective can complement this market efficiency debate between traditional and
behavioral finance. We discuss this issue next.
2.2 Towards New Institutionalism of Market Efficiency
New institutionalism is an interdisciplinary research agenda seeking to understand and explain the
interactions between institutions and economic actions of various actors in the society. An
important assumption of the new institutionalists is that actors behave subject to various context-
bound constraints. Such constraints, referred to as “institutions”, can cover all formal and informal
social and economic constraints that shape the choice-set of actors (Nee 1998). New
institutionalists reject neoclassical assumptions, while remaining committed to choice-theoretic
tradition explanation in the social sciences where choices and actions are subject to constraints.
Page 9 of 43
We argue that a new institutionalism approach towards the debate of market efficiency between
the traditional finance and the behavioral finance is rather complementary, not conflicting. Before
explicating on this point, let us first discuss a very much related research agenda in experimental
capital markets that Bloomfield and his colleagues (Bloomfield et al 2009; Bloomfield and
Rennekamp 2009) have pioneered7. The basic argument behind their research agenda is that
market efficiency should be viewed as an interplay between disciplinary institutions (strong vs
week) and behavioral forces (strong vs. weak). In this framework, strong disciplinary institutions,
for example, would imply competitive and liquid securities market that have very high ability to
eliminate behavioral biases. Here, behavioral forces would have greater impact on market and firm
behavior in absence of strong disciplinary institutions. The following figure by Bloomfield (2010)
presents a 2X2 research framework that can be used to test the key statement of such research
agenda.
Weak Behavioral Forces
Strong Behavioral Forces
Inst
itut
ions
Strong Disciplinary Institutions
(Cell 1) (Cell 2)
Weak Disciplinary Institutions
(Cell 3) (Cell 4)
Figure 1: A Research Design for Behavioral Finance Studies by Bloomfield (2010). Note: Bloomfield (2010) clarifies the figure by noting “This research design clarifies the interaction between the strength of behavioral forces on individual decision making and the ability of the finance institution in which individuals make decision to eliminate the behavioral forces in aggregate phenomena”.
7 See Bloomfield (2010) and Bloomfield and Anderson (2010) for a more detailed description of this research agenda in experimental capital markets.
Page 10 of 43
A novel proposition in itself, this framework, however, misses an important issue that new
institutionalists would like to know in greater details: the sources of these behavioral forces.
Particularly, institutionalists would inquire about the institutional and extra-institutional sources
of these behavioral forces and in doing so, would examine the context-bound economic actions of
pertinent actors and the relationships among their actions. This perspective of looking emphatically
at the sources of behavioral forces is, indeed, not entirely new to finance. We observe that the
emerging fields of neuroeconomics and neurofinance have been investigating the fundamental
biological and psychological mechanisms underlying the individual biases and irrational behavior
(Peterson 2010). Thus, a new institutionalist framework, reduced to simple representation, would
look like Figure 2. In this framework, three factors, i.e. “actors’ context and connections”,
“behavioral forces”, and “disciplinary market institutions” interact with each other to give rise to
aggregate phenomena such as market efficiency or market inefficiency. We further add that
aggregate market phenomena is not at a mere receiver in this interactive framework; rather,
aggregate market phenomena give back to the rest in a dynamic manner.
One would argue whether we could fit “actors’ context and connections” into the disciplinary
market institutions, defined by long-standing laws, practices and organizations (Bloomfield and
Rennekemp 2009). Such an argument may sound like the hypothetical story of a person who loses
a watch (for not-yet-known reasons) in a dark room and cannot find it, and the search-party blames
the darkness of the room for such loss and to be successful, starts looking for the watch in a brighter
room. So, we argue that such an attempt (of fitting sources into disciplinary institutions) would
Disciplinary Market
Institutions
Behavioral Forces
Actors’ Context
and Connectedness
Aggregate Market Phenomena
Figure 2: A New Institutionalist’s Framework for Aggregate Market Phenomena
Page 11 of 43
hinder a careful understanding of the sources of behavioral biases and apparent irrationality of
actors. While it may be true that disciplinary market institutions may mediate/foster the effect the
effects of behavioral forces on aggregate market phenomenon, such disciplinary institutions are
not necessarily the sources of behavioral biases. In other words, the context of various actors in
the market and their connectedness cannot be entirely subsumed by the disciplinary market
institutions.
Let us further explicate this point by taking the case of limited attention. One argument in this
area is that scarcity captures attention (Mullainathan and Shafir 2013). Scarcity of food, for
example, severely limits an individual’s thoughts to foods. Scarcity of time, for example, could
make an individual further hurried for time. Now let us consider the role of such concept of limited
attention of actors for the aggregate market phenomenon. In doing that consider the finding
(Hirshleifer et al. 2009) that on the days of many earnings announcements investors feel scarcity
of time and thus underreact to earnings news of a firm. And, in absence of strong disciplinary
institutions such as arbitrageurs, such underreaction shows up as a systematic irrational behavior
in the aggregate market phenomena. Should we say that arbitrageurs are the reasons for investors’
inattention? In another example, let’s say during Fridays, sports, lunch hours, festivals, and
elections, investors further fail to pay attention to the news relevant for stock pricing and in absence
of active arbitrage activities (i.e. presence of weak disciplinary institutions) individual investors’
inattention shows up as a systematic aggregate market phenomena. Should we attribute the sources
of inattention to the disciplinary institutions only? While features of the disciplinary institutions
are definitely of high import, another area of investigation would be to look at the context of the
actors and the connectedness among the actors to understand the causes of inattention. It is to note
that we add “connectedness among actors” along with the context of actors to emphasize that
actions of capital market actors (and so is case for any phenomena that is “social”) have
theoretically as well as empirically shown mutual independence8; so context-bound actions of one
actor could end up affecting another actor due to their relevant connectedness.
In a similar vein, we turn our attention to a particular context-bound rationality of analysts, a
very important information intermediary whose actions have been found to have implications for
8 The bulk of literature proving this point in various branches of social sciences is quite huge. We do not provide more discussion about it for the sake of brevity.
Page 12 of 43
investors, i.e. the market in general. In particular, we discuss limited attention from competing
tasks and distracting events and the role of such limited attention in analyst earnings forecast.
2.3 Limited Attention and Analyst Earnings Forecast
Limited attention refers to the neglect of relevant information in performing a task. Studies in
psychology show how individuals’ task performance suffers due to limited attention. These studies
relate to the school of “behaviorism” in psychology. This school, introduced by John D. Watson
in 1913, underscores that environmental stimuli, not the internal proclivities, determine human
behavior. The idea is that we could predict a person’s behavior based on the external stimuli
presented to her (e.g., Daniel, Hirshleifer, and Teoh 2002). Limited attention could arise due to
competing tasks and distracting events.
Limited attention could stem from a situation in which an individual is performing multiple
tasks simultaneously (i.e., competing tasks). The situation occurs particularly when the tasks
require “controlled processing”. A task is said to require “controlled processing” if individuals
need to engage in conscious thinking before giving a response (i.e., individual response is not
automatic, Shiffrin and Schneider 1977). Studies in psychology and ergonomics provide various
reasons why competing tasks could limit attention and affect performance. First, multiple difficult
tasks could use up resources quickly. The multiple resources models argue that conduct of
competing difficult tasks create a competition for resources among the tasks. So, individuals have
to allocate and exchange resources among the tasks. This could create interferences. Such
inferences could lead to individuals not paying due attention to all the relevant aspects of the tasks.
Hence, efficiency could suffer. When analysts have competing earnings news to process, they face
a similar situation, because processing earnings news is a complex professional task and multiple
earnings news creates a stretch on resources. Second, the single channel theories suggest that in
performance of multiple tasks in a time-sharing situation, interferences could arise at various
levels, leading to limited attention. Two prominent levels at which interferences occur are as
follows: (1) when individuals perceive the stimuli; (2) when individuals generate the response
(Broadbent 1958; Briggs, Peter, and Fisher 1972). The idea is that interferences arise because
multiple tasks quickly saturate the channel capacity (i.e., the individual capacity needed to perform
a task). In situations of saturated channel capacity, individuals could switch to serial (i.e. one after
another) processing from parallel (i.e. simultaneous) processing. However, failure to do that at any
Page 13 of 43
levels could result in interferences and thus, limit attention. Likewise, competing news could
saturate analysts’ channel capacity and thus, could lead to limited attention. Third, there are studies
which show how task interruptions, in general, could lead to poor performance. Research shows
that occurrences of a second task affect operators in the nuclear power industry negatively and
lead to incidents that shut-down the operation. Studies find in context of a telephone company
sales office that occurrence of a second task creates interference in the performance of the first
task, leading to increased duration for execution of the first task (Cellier and Eyrolle 1992). Hence,
if analysts face multiple tasks, i.e. multiple earnings news for his/her portfolio firms, the quality
of his/her earnings forecast revisions could suffer.
Limited attention could also arise from a situation in which individuals are faced with
distracting events. The reason is that such events could distract attention away from the relevant
tasks. There is ample evidence in social sciences that show irrelevant stimuli distract attention.
Some examples follow. “Stroop Test” is a famous one (Stroop 1935). The test shows that people
take longer time to read the name of a color if the print color of the word does not match with the
color name. For example, if “blue” is printed in red color it would take readers longer to read.
Wallace and Vodanovich (2003) find that electrical workers who are more distracted by the events
in daily life face greater number of accidents at work. Roadside billboards distract attention of the
drivers (Wallance 2003). According to McEvoy, Stevenson, and Woodward (2007) more than 10%
of the drivers (out of 1,367 sample after accidents) indicate that at the time of accidents they were
absentminded due to various distracting events (such as sight of a person, event, or object outside
the car or animal or insect in inside the car). Forster and Lavie (2008) argue that activities of
colleagues or work-environment in general could distract individuals as well. While the exact
forms of distractions caused by irrelevant stimuli are not known, the mounting evidence shows
that irrelevant stimuli do limit attention. Moreover, studies in “inattentional blindness” and
“dichotic listening” show that individuals cannot attend to and retain multiple stimuli at the same
time (Cherry 1953; Moray 1959; Simons and Chabris 1999). So, once irrelevant stimuli somehow
catch a person’s attention, it is likely that s/he will be distracted from the relevant. Analysts could
face similar distracting stimuli when the environment in which analysts operate has lots of
activities going on. One such example is when there are many earnings announcements by other
firms on the day an analyst’s portfolio firm announces earnings or an analyst provide a forecast
Page 14 of 43
revision. Days with such high distracting events could negatively affect the quality of analysts’
earnings forecast revisions.
2.4 Prior Evidence of Analyst Forecast Quality and Price Discovery
There is a large literature on analyst forecast accuracy. Briefly put, the literature has investigated
analyst-specific characteristics (e.g. experience), brokerage house characteristics (particularly,
size), following-firm characteristics (e.g. firm complexity, analyst following), and various decision
making errors (e.g. overweight of information, underreaction, and faulty input models). However,
Katherine Schipper’s (1991) commentary suggests that there is a dearth of research on
understanding the decision-context of analysts, a gap our paper attempts to fulfill.
Since analysts are a major information intermediary and play a vital role in interpretation and
dissemination of information in capital markets, researchers have also devoted attention to
understanding the role of analysts’ forecasts in price discovery. Chan et al. (1996) show that a
moving average of the forecast revisions over the last six months can significantly predict firms’
returns over the next 6-12 months. Park and Stice (2000) find that market response to analyst
forecast revisions depends on the prior usefulness of analyst forecasts. Our paper further extends
this literature by examining how the limited attention of analysts delays price discovery.
2.5 Empirical Predictions
Given two prominent sources of limited attention pointed out in the relevant literature we divide
our empirical predictions into two categories: (1) competing task hypothesis; (2) distracting event
hypothesis.
2.5.1 Competing Task Hypothesis
The first context of limited attention the number of competing tasks an analyst faces his/her
portfolio. Here, by the number of competing tasks we mean the number of earnings
announcements that an analyst faces for his portfolio firms on a day. As the literature review
suggests this context demands analyst attention to multiple tasks and hence, in such a context,
analyst earnings forecast quality is likely to be worse. At this point, one could argue that such
prediction does not consider the idea that what an analysts face may not necessarily matter if the
analyst can prioritize tasks rationally, i.e. analysts resort to ‘serial processing’ in a way so that
Page 15 of 43
forecast accuracy does not suffer due to competing tasks faced by the analysts. Our argument is
that even if an analyst can switch to such ‘serial processing’ in the face of competing tasks, the
mere thought that there are tasks pending could create a disturbance in the performance of the task
at hand. However, to be thorough in laying the hypothesis we also consider the number of tasks an
analyst actually performs as an additional definition of limited attention arising from competing
tasks
By analyst earnings forecast quality we consider the ‘accuracy’ aspect of earnings forecast.
Hence, we hypothesize the following.
H1A: There is a negative relationship between the level of competing tasks faced and/or performed
by analysts and the accuracy of analysts’ earnings forecast, ceteris paribus.
If analysts fail to process relevant information because of competing tasks, their limited
attention is also likely to be a factor contributing to market-wide underreaction to earnings news,
i.e. PEAD. We argue that the higher the overall analyst (i.e. all the analysts who follow/cover a
firm) limited attention due to competing tasks, the greater will be the market underreaction to
earnings news of a firm. Hence, we hypothesize the following.
H1B: There is a positive association between the market underreaction to earnings news and the
limited attention due to competing tasks faced and/or performed by all the analysts that follow a
firm, ceteris paribus.
2.5.2 Distracting Event Hypothesis
The second prominent source of limited attention is distracting events. Here, we consider the
following distracting events that are closely related to analysts’ work environment: number of
earnings announcements by non-portfolio firms. Consistent with the literature review, we argue
that the higher the level of such distraction on a day (i.e. forecast day), the lower will be the quality
(i.e. accuracy) of analyst forecasts. Hence we hypothesize the following.
H2A: There is a negative relationship between distracting events and analyst forecast accuracy,
ceteris paribus.
Page 16 of 43
If analysts fail to process relevant information because of distracting events, their limited
attention is also likely to be a factor contributing to market-wide underreaction to earnings news.
We argue that the higher the overall analyst (i.e. all the analysts that follow a firm) limited attention
due to distracting events, the greater will be the market underreaction to earnings news of a firm.
Hence, we hypothesize the following.
H2B: There is a positive association between the market underreaction to earnings news and the
limited attention due to distracting events faced by all the analysts that follow a firm, ceteris
paribus.
3.0 Data
We use data from WRDS database for this study. Our sample covers the period from 2000 to 2012.
Variables are winsorized on both sides of the distribution (1% and 99%). The final sample has
5,136 distinct firms, 10,798 distinct analysts, and 155,443 firm-quarters covered. Next we define
the variables used in the study.
3.1 Dependent Variables
The dependent variables are analyst earnings forecast accuracy (for ease in measure and
interpretation, we rather calculate forecast error, FE in short) and price drift (absolute value of
abnormal return, ABSRET in short) following earnings announcement of a firm. They are defined
below.
���,�,�: Forecast error of analyst ‘i’ for forecast made about the earnings of firm ‘j’ for quarter ‘q’
is calculated by the following formula where ��� is the actual earnings of firm ‘j’ for quarter ‘q’,
��,�,� is the forecast made my analyst ‘i’ for the quarter ‘q’ earnings of firm ‘j’, and ��,� is the
closing day’s (t) stock price on the day of forecast.
���,�,� = |��,� − ��,�,�
��,�|
�������,�: Absolute value of abnormal return of firm ‘j’ on trading day ‘t’ is calculated after
return is adjusted for Cahart four factors, which are momentum factor and Fama-French three
factors (overall market factors and factors related to firm size and book-to-market equity). We
Page 17 of 43
cumulate this absolute abnormal return over various post-announcement windows such as (+2,
+6), (+7,+11) and (+12 to +16) following earnings announcement to find out ABS CAR, i.e.
absolute cumulative abnormal returns. It should be noted that day ‘0’ refers to the earnings
announcement day; so , day +2 refers to 2nd trading day after the earnings announcement day, day
+12 refers to the 12th trading day after the earnings announcement day.
3.2 Key Independent Variables
The key independent variables measures competing tasks (analyst-level and firm-level) and
distracting events (analyst-level and firm-level).
Competing Task Measures: We use two measures of analyst-level competing tasks (i.e.
COMPTASK1, and CTFACED1) and two measures of firm-level competing tasks (i.e.
AVG_COMPTASK1 and AVG_CTFACED1). These measures are defined below.
COMPTASK1: This is an analyst-level competing measure. It is calculated by the number of
earnings forecasts made by analyst ‘i’ on a particular day. In other words, this is a measure of
competing tasks performed.
CTFACED1: This analyst-level competing task measure is calculated by the number of earnings
announcements by the portfolio firms of analyst ‘i’ on a forecast day. In other words, this is a
measure of competing tasks faced.
AVG_COMPTASK1: This is a firm-level competing task measure. It is calculated as the average
COMPTASK1 (on the day of firm’s j’s quarterly earnings announcement) for all the covering
analysts for firm ‘j’.
AVG_CTFACED1: This is a firm level competing task measure. It is calculated as the average
CTFACED1 (on the day of firm’s j’s quarterly earnings announcement) for all the covering
analysts for firm ‘j’.
Distracting Event Measures: We use one analyst-level distracting event measure and one firm-
level distracting event measure. They are defined below.
DISTRACT1: This is an analyst-level distracting event measure. It is calculated by the number of
earnings announcements by firms outside the portfolio of analyst ‘i’ on a forecast day.
Page 18 of 43
AVG_DISTRACT1: This is a firm-level distracting event measure. It is calculated by the average
DISTRACT1 of all following analysts of firm ‘j’ on the day of quarterly earnings announcement
of firm ‘j’.
3.3 Control Variables
Consistent with prior literature we use a number of control variables in the models used for testing
hypotheses (to be discussed in Section 4.0). The control variables can be categorized into the
following: firm and earnings characteristics, analyst characteristics, and information environment.
We also discuss (or provide footnotes) as to how these controls also proxy for the various
components in Figure 2, while we focus on the influence of limited attention from competing tasks
and distracting events.
Firm and Earnings Characteristics: We use the following firm characteristics as control variables
in the study: firm size (SIZE), number of segments (NUMSEG), incidence of restructuring
(RESTRUCTURE), incidence of merger and acquisition (MERGE), incidence of special items
reporting (SPECIAL), institutional ownership (INST), size of earnings news (UE), type of earnings
news (BDNEWS), and quarter of the earnings (4THQTR). The variables are measured as follows.
�����,�: Firm size is measured as natural logarithm of the product of closing stock price and
number of shares outstanding of firm ‘j’ in quarter ‘q’.
�������,�: Number of segments is the number of business segments reported in the quarterly
financial statements of firm ‘j’ in quarter ‘q’.
������������,�: Incidence of restructuring is a dummy variable that is assigned a value of ‘1’
if firm ‘j’ reported restructured cost of quarter ‘q’ and else, it takes on a value of ‘0’.
������,�: Incidence of mergers and acquisitions for firm ‘j’ is a dummy variable that is assigned
a value of ‘1’ if firm ‘j’ reported any mergers/acquisitions in financial statements of quarter ‘q’;
else, the variable takes on a value of ‘0’.
��������,�: Incidence of special items for firm ‘j’ is a dummy variable that is assigned a value
of ‘1’ if firm ‘j’ reported any special items in financial statements of quarter ‘q’; else, it takes on a
value of ‘0’.
Page 19 of 43
�����,�: Institutional ownership is measured by percentage of institutional ownership of firm ‘j’
at the end of quarter ‘q’.
���,�: Unexpected earnings of firm ‘j’ in quarter ‘q’ (���,�) is calculated as the actual earnings
minus the latest analyst earnings revision prior to earnings announcement.
��������,�: Bad news indicator variable, ��������,� for firm ‘j’ in quarter ‘q’ equals ‘1’ if
���,� < 0 and 0 otherwise.
4������: Quarter of the earnings, 4������ is an indicator variable that equals ‘1’ if the
earnings is for the 4th quarter of the fiscal year of firm ‘j’ and ‘0’ otherwise.
Analyst Characteristics: We use the following analyst characteristics as control variables for the
study: experience (EXPERIENCE) and size of the analyst’s brokerage house (BRSIZE). The
variables are measured as follows.
�����������,�: The experience of analyst ‘i’ in forecasting earnings of firm ‘j’, is calculated as
the number of prior quarters of revisions made for firm ‘j’ by analyst ‘i’.
�������,�: Size of the brokerage house of analyst ‘i’ in quarter ‘q’ is the number of analysts
employed by the analyst’s brokerage house in that quarter.
Information Environment: We use the following variables to proxy for firm information
environment: analyst following (NANALYST) and management guidance (GUIDE). The
variables are measured as follows.
���������,�: Number of analysts following the firm, ���������� is calculated as the number
distinct analysts who provided at least one forecast for firm ‘j’ in quarter ‘q’.
������,�: Management guidance indicator variable, ������,� equals ‘1’ if the firm ‘j’ provided
guidance for quarter ‘q’ and ‘0’ otherwise.
Page 20 of 43
4.0 Models and Results
First we discuss the descriptive statistics and correlations among the study variables. Then, we
discuss models and results into two subsections: competing task hypothesis and distracting event
hypothesis. All tables containing results are provided in the Appendix.
4.1 Descriptive Statistics
Table 1 presents the descriptive statistics of the key study variables. Some notable aspects from
descriptive statistics are discussed next. We find that on an average analysts are less than accurate.
Also the magnitude of the standard deviation of forecast error is almost 7 times the mean forecast
error. There is a large deviation in forecast errors since mean and median (50th percentile) error is
largely distant from each other. COMPTASK1, a measure of limited attention from competing
tasks performed on a forecast day has a mean of 1.44 with a median of 1, suggesting a wide
variation among analysts in the number of competing tasks they (analysts) perform on a forecasting
day. CTFACED1, a measure of limited attention due to competing tasks faced, has a mean 1.9
times its median. This suggests that analysts also vary in the number of competing tasks they face
for their portfolio firms. DISTRACT1, a measure of distraction of analysts from earnings
announcements by outside-portfolio firms, does not have a wide variation (mean is closer to
median). This makes sense, since an individual analyst’s portfolio is very small compared to the
universe of firms in the US capital markets as well as in our sample.
Table 1 further shows that various control variables used in this study (firm and earnings
characteristics, analyst characteristics, and information environment) have variation in them as
well. Some notable aspects are discussed next. In only about 7% of the firm-quarters in our sample
managers provide guidance. Average institutional shareholding for sample firms is about 70%
whereas at 10th percentile such shareholding is about 30%. In about 27% of the firm-quarters,
earnings announcement is bad news. In about 12.9% of firm-quarters there are incidences of
mergers and acquisitions and in about 9% of the firm-quarters there are incidences of restructuring
charges. On an average we observe more than 5 segments (NUMSEG) per firm-quarter, with some
variation in the measure (1 segment in 10th percentile and 11 segments in the 90th percentile). An
average analyst in the sample has about 11 quarters of firm-specific forecasting experience and an
average brokerage house employs about 47 analysts (with a median of 40 analysts).
Page 21 of 43
Table 2 presents results of spearman correlation analysis among the study variables. Some
notable findings from this analysis are discussed next. There is a significantly positive correlation
between forecast error (FE) and number of competing tasks performed by analysts on a forecast
day (COMPTASK1). Similarly, FE has a significantly positive relationship with DISTRACT1, i.e.
the number of earnings announcement for outside-portfolio firms of analysts on a forecast day.
The number of competing tasks faced (CTFACED1) has a negative relationship with forecast
error, which, in bivariate comparison weaken our competing task hypothesis and contradict with
the relationship between COMPTASK1 and FE. However, one must note that analysts’ forecast
errors are significantly correlated with various control variables as well. For example, firm size
(SIZE) and incidences of mergers and acquisitions (MERGE) are significantly negatively
correlated with forecast error. Also, extent of disciplinary institutions (or better information
environment), as proxied by management guidance and institutional ownership, is significantly
negatively correlation with forecast error. We find (in unreported results) that pearson correlation
provide qualitatively similar results (the direction of the relationship is mostly similar but
magnitude different).
Overall, the results of Table 2 correlation analysis suggest that we must consider the entire
gamut of pertinent factors in multivariate models to find careful results for our hypothesized
relations.
4.2 Models and Results for the Competing Task Hypothesis
We use the following models (Model 1 and Model 2) to test the relationship between competing
tasks and analyst earnings forecast error. The variables are defined in Section 3. We further control
for year-quarter fixed effects to proxy for any time-dependent aggregate market phenomena9.
9 In the future version of this paper we will consider more specific incidences of aggregate market phenomena, for example, market liquidity (which may also be considered as disciplinary market institutions, in the definition of Bloomfield (2010)).