Content uploaded by John Morgan
Author content
All content in this area was uploaded by John Morgan
Content may be subject to copyright.
Analyst Information Acquisition and
Communication∗
Paul E. Fischer
The Pennsylvania State University
Phillip C. Stocken
Dartmouth College
May 28, 2008
Abstract
We examine a communication game between an analyst and a decision-
maker and investigate how the presence of public information affects the preci-
sion of the information the analyst gathers and credibly communicates to the
decision-maker. We characterize conditions under which public information
causes the analyst to under-invest or over-invest in the information gathered
relative to the case where analyst credibility is not an issue. The model also
provides circumstances where the presence of public information causes the an-
alyst to drop coverage of the firm, suggesting that the introduction of public
information can make the decision-maker strictly worse off.
∗We thank Tim Baldenius, Ron Dye, Steve Huddart, Nahum Melumad, John Morgan, Jonathan
Rogers, Richard Sansing, Jerry Salamon, Ulf Schiller, Reed Smith, and Ro Verrecchia for helpful
discussion. We also thank workshop participants at the University of Bern, Chicago-Minnesota
Accounting Theory Conference, Columbia University, Dartmouth College, Indiana University, and
University of Michigan.
Please address correspondence to: Phillip Stocken, Tuck School of Business
at Dartmouth, 100 Tuck Hall, Hanover, NH 03755; e-mail: phillip.c.stocken@dartmouth.edu.
Analyst Information Acquisition and Communication
1 Introduction
Decision-makers obtain information from various sources. Some sources have incen-
tives to communicate strategically in a self-serving fashion and face limited regulation
regarding their communication. Other sources have little incentive to communicate
strategically or face fairly stringent regulation regarding their communication. In-
vestors, for instance, can obtain information from sell-side analysts, who are thought
to have incentives to communicate strategically and who face relatively little regula-
tion regarding their communication, and also from a firm’s audited financial state-
ments, where regulation plays a significant role in limiting strategic communication
of certain information. This study analyzes how changes in the information disclosed
by a non-strategic source influences the information gathering and communication
decisions of an information intermediary who is strategic.
The model we employ in this analysis has four stages and two players: an infor-
mation intermediary, which we call an analyst, and a decision-maker. In the first
stage, both players observe an imperfect public signal about the state of nature. In
the second stage, the analyst engages in costly information gathering that determines
the precision of a signal about the state of nature that he will privately observe. In
the third stage, the analyst privately observes a signal and costlessly sends a mes-
sage to the decision-maker. In the fourth stage, the decision-maker, given his beliefs
about the state of nature, takes an action that affects both players’ payoffs. The
players have divergent preferences regarding the decision-maker’s action. Within the
context of this model, we explore how the precision of the public information that
the players observe in the first stage affects the analyst’s information gathering and
communication in the second and third stages.
While the model can be applied to various settings where decision-makers obtain
information from sources that strategically release information and also from other
sources that release information in an impartial manner, we find this model useful for
framing the recent evolution in the analyst market. The amount of information that
firms make available directly to investors has exploded. Following the introduction of
Reg FD in 2000, for instance, firms can no longer privately reveal information to an-
alysts and hence are more likely to publicly release their forward-looking information
2
(see Heflin, Subramanyam, and Zhang 2003). Further, firms now routinely hold con-
ference calls following their earnings releases that are open to investors (see Bushee,
Matsumoto, and Miller 2003). In addition, the $1.4 billion settlement between the
large brokerage firms and the State of New York in 2002 required the investment
firms to provide information to investors from unbiased sources together with their
own analysis. Following this settlement, there has been a flight of equity analysts
from the research departments at the large investment firms, and entire industry sec-
tors have lost analyst coverage. As a consequence, several commentators have argued
that these changes are likely to hurt individual investors and reduce stock price effi-
ciency (see Schack 2003). A contribution of this study, therefore, is to illustrate how
alternative sources of information might affect analyst information gathering activi-
ties, the integrity of their communication with investors, and whether investors are
likely to be better offas a consequence of changes in the amount of publicly available
information.
This paper highlights that changes in publicly observed information affects an
analyst’s behavior in two ways. First, when the analyst can credibly communicate
privately observed information, the public information serves as a substitute for the
analyst’s information. As a consequence, improvements in the quality of the public in-
formation reduce the marginal value of the analyst information causing him to gather
less precise information. Second, when the analyst’s credibility is in doubt–because
the analyst’s incentives are misaligned with those of the decision-maker–improving
the quality of the public information makes it more difficult for the analyst to cred-
ibly communicate. Specifically, an increase in the quality of the other information
reduces the decision-maker’s responsiveness to the analyst’s information, which causes
the analyst to exaggerate his report in an attempt to influence the decision-maker.
This exaggeration undermines the analyst’s credibility. As a consequence, when the
quality of the public information increases, either the analyst gathers more precise
information to increase the decision-maker’s responsiveness to his information and
thereby facilitate credible communication or he discontinues gathering information
(i.e., leaves the market).
This analysis suggests that competition arising from improvements in the quality
of public information has subtle consequences for analyst behavior. When credibil-
ity issues are minimal, which occurs when information gathering costs are low, an
increase in the precision of public information leads to a marginal reduction in an-
3
alyst information gathering. Although more precise public information crowds out
some analyst information, this crowding out is not sufficient to offset the benefitof
more precise public information. Consequently, the decision-maker is better off.Al-
ternatively, when credibility issues are more pronounced and information gathering
costs are moderate, more precise public information leads the analyst to improve his
information gathering. The heightened precision of the analyst information coupled
with more precise public information combine to make the decision maker better off.
Finally, when credibility issues are more pronounced and information gathering costs
are high, an increase in the precision of public information completely crowds out the
analyst information and the analyst leaves the market. Although the decision-maker
obtains more precise public information, the crowding out of the analyst information
can be so severe that it actually reduces the total information the decision-maker
obtains thereby making the decision-maker worse off.
The primary theoretical antecedent of our paper is Crawford and Sobel (1982).
They consider a communication game between an informed expert and an uninformed
decision-maker. The expert perfectly observes the realization of a payoffrelevant
state variable and sends a costless message to the decision-maker who then takes an
action that affects both players. Their cheap-talk model has been extended in various
directions. In work most closely related to our paper, Austen-Smith (1994) considers
a communication game in which the expert chooses whether to perfectly observe the
stateofnatureatsomefixed cost and decides whether to reveal being informed, but
is unable to prove being uninformed. He establishes that there exist circumstances
under which the players cannot communicate when the receiver knows the expert
is informed although they can do so when the receiver is uncertain about whether
the expert is informed. In contrast to Austen-Smith, we consider a model in which
the expert, after observing a public signal, can choose the precision of his privately
observed signal; further, we allow for the possibility that the precision of the expert’s
signal is either publicly observed or privately chosen.
The extant literature also has extended the model in Crawford and Sobel (1982)
by examining the effect of multiple experts on the communication game (e.g., Austen-
Smith 1993; Krishna and Morgan 2001; Wolinsky 2002; Battaglini 2004; Morgan and
Stocken 2008), by introducing multiple decision-makers (e.g., Farrell and Gibbons
1989; Newman and Sansing 1993; Gigler 1994), or by allowing the players to exchange
messages (e.g., Aumann and Hart 2003; Krishna and Morgan 2004). In these studies,
4
the expert is exogenously endowed with information. We, in contrast, examine a
setting where it is costly for the expert to gather information and consider how the
presence of a non-strategic information source affects the information that the expert
gathers and communicates.
Our paper is also related to the voluntary disclosure literature. Verrecchia (1990)
examines a discretionary disclosure setting in which the quality of information with
which a manager is endowed is exogenously specified. Analogous to our result that
more public information can thwart credible communication, Verrecchia establishes
that an increase in the precision of the market’s prior beliefs, which corresponds with
an increase in the other information in our model, reduces the probability that the
manager discloses. Penno (1996) considers a mandatory disclosure setting in which a
manager, after observing a public signal, privately chooses the precision of the report
that the firm will release. He establishes that the precision of the manager’s report is
inversely related to the favorableness of the public information and that the strength
of this relationship diminishes as the public signal becomes less precise. Hence, like
our paper, he links precision of a public signal with the amount of information com-
municated in a subsequent report. Both of these papers, unlike our paper, assume
that the precision of the manager’s information is exogenous. In more recent work,
Pae (1999) explores a manager’s acquisition of costly private information about the
consequence of the manager’s productive effort in the context of a voluntary disclosure
setting. Hughes and Pae (2004) consider a setting in which an entrepreneur gathers
and voluntarily discloses information truthfully about the precision of a signal of an
asset’s value that the entrepreneur is offering for sale. Lastly, Arya and Mittendorf
(2007) explore how public disclosure might affect analyst behavior. They consider a
setting where competing firms can voluntarily release information to encourage an-
alyst following. Heightened analyst following enriches the information environment
and thereby allows firms to better coordinate their production choices. A common
feature of these studies, like much of the extant literature examining voluntary disclo-
sure behavior (e.g., Verrecchia 1983; Dye 1985; Einhorn and Ziv 2008), is that private
information, if it is disclosed, must be disclosed truthfully. In the analyst reporting
environment, however, analysts often report in a self-serving fashion and face little
regulation regarding how they report (e.g., Michaely and Womack 1999; Lin and Mc-
Nichols 1998). Indeed, it was precisely this concern that motivated key provisions of
the Sarbanes-Oxley Act and the legal settlement between the large brokerage firms
5
and the State of New York in 2002.1We examine a reporting environment in which
the analyst’s disclosure is not constrained by particular disclosure rules and find the
desire to communicate credibly fundamentally influences the analyst’s information
gathering behavior. Hence, a contribution of our paper is to provide insights into
information gathering and communication behavior when reporting credibility is a
key feature of the environment.2
Our paper is also reminiscent of the disclosure literature that considers the credi-
bility of a firm manager’s communication when investors can gather other information
that allows them to assess the veracity of the manager’s disclosure. For instance, Sans-
ing (1992) and Stocken (2000) examine managers’ earnings forecasting behavior when
investors can use an audited earnings report to assess a forecast’s credibility. In these
papers, the investors are exogenously endowed with the earnings report. Further, the
modeling frameworks and the questions they address differ substantially from ours.
The paper proceeds as follows: Section 2 describes the model and includes a time
line that summarizes the model’s notation. Section 3 characterizes the equilibrium.
Section 4 examines how changes in the precision of public information influence an an-
alyst’s information gathering and communication. Section 5 considers how changes in
the precision of public information affect the quality of the decision-maker’s informa-
tion. Section 6 relaxes the assumption that the analyst’s precision choice is common
knowledge and discusses how the analyst’s information gathering and communication
changes. Section 7 concludes. All proofs are in the Appendix.
2Model
We consider a communication game in which an analyst can choose to obtain private
information about an unknown state variable and send a report to a decision-maker
who then takes an action. The unknown state variable is represented by the random
variable ˜α, which has support {s, f }where s>f. The common prior beliefs are that
1Some work has studied analyst information gathering. Hayes (1998), for instance, examines an
analyst’s incentives to gather and report information in an environment in which the analyst’s report
influences an investor’s trading behavior that, in turn, determines the trading commission that the
analyst receives. The analyst always reports truthfully so that issues of strategic disclosure do not
arise.
2In addition, several studies examine how private information acquisition will change in response
to a public information signal within a rational expectations framework (e.g., Verrecchia 1982;
Diamond 1985; Alles and Lundholm 1993).
6
the successful, s, and failure, f, state realizations are equally likely to occur.3
The game has four stages. In the first stage, information represented by the
realization of a random variable ˜xis publicly observed. The support for ˜xis {h, l}.
The probability that ˜x=hconditional on ˜α=sand the probability that ˜x=l
conditional on ˜α=fare both q∈(1/2,1).Thevariableqcaptures the precision
of ˜x, where a higher value for qimplies that ˜xis more informative. Setting q=
1/2is equivalent to supposing that the public information is absent. The public
information may be viewed as the filing of a firm’s audited financial statements or the
release of governmental statistics that are informative about a firm’s future outlook.
The analyst observes the realization of the public information xbefore choosing the
precision of his private information.
In the second or information gathering stage, the analyst chooses whether to invest
in a costly information generating technology to acquire private non-verifiable infor-
mation. The analyst’s private information, if obtained, is represented by the realiza-
tion of a random variable ˜y, which has support {g, b}. Conditional on the realization of
˜α,˜xand ˜yare independent. The probability that ˜y=gconditional on ˜α=sand the
probability that ˜y=bconditional on ˜α=fare both p∈[1/2,1]. The analyst chooses
the precision pand incurs a cost c(p),wherec(p)is a twice differentiable function that
satisfies: (i) c0(p)>0and c00(p)>0for p>1/2; (ii) limp→1/2c0(p)/(p−1/2) →0;
(iii) limp→1c0(p)→∞;and(iv)c00(p)/c0(p)>(1 −3p(1 −p)) /(p(1 −p)(p−1/2))
for all p∈(1/2,1). Condition (i) requires that the analyst’s cost of gathering in-
formation increases in the precision of the information and at an increasing rate.
Conditions (ii) and (iii) are sufficient for the analyst to choose an interior level of
precision when he can credibly communicate. Condition (iv) is sufficient to insure
the analyst’s optimal precision choice is unique. As an example, the simple power
function c(p)=((p−1/2) /(1 −p))n,wheren>2,satisfies these conditions (see
Appendix for details). The analyst choosing not to invest in the information technol-
ogy is equivalent to the case in which the analyst’s private information has precision
p=1/2.
The decision-maker observes the precision of the analyst’s private information, but
not the realization of his private information. This assumption is consistent with the
sell-side equity analyst environment, which is the primary institutional application of
3As in this paper, many cheap-talk models assume the players’ prior beliefs are diffuse (e.g.,
Austen-Smith 1993; Krishna and Morgan 2001; and Battaglini 2004).
7
our model. Analyst stock reports vary in content and often contain detailed analysis
of a company, its key competitors, and its industry. The detail in the analyst’s
stock report, which the investor can readily observe, reflects the precision of the
analyst’s information. Investors, however, do not observe the analyst’s procedures for
evaluating and interpreting the information that is gathered. It is this interpretation
and evaluation that provides the analyst with private information y.Inthefinal
section of the paper, we suppose the analyst’s choice of precision is unobservable.
In the third or reporting stage, the analyst costlessly sends a report rto the
decision-maker. The analyst need not report truthfully. The analyst’s ability to
communicate the privately observed signal yin his report is potentially thwarted,
however, because the commonly known preferences of the decision-maker regarding
her action choice differ from the commonly known preferences of the analyst. In
particular, the decision-maker chooses an action a∈<given her information Ωdto
maximize
Ud=−E£(˜α−a)2|Ωd¤,(1)
where E[·|·]is the expectation operator. Given this objective function, it is opti-
mal for the decision-maker to choose an action a=E[˜α|Ωd], which is analogous
to investors valuing a firm at its expected value.4The analyst’s objective function
also depends on the decision-maker’s action; the analyst chooses a report rgiven his
information Ωito maximize
Ui=E£φa −(˜α−a)2−c(p)|Ωi¤,(2)
where φ>0.5When φ=0, the players’ interests are perfectly aligned because,
given the same information, both players would prefer the same action. When φ>
0, however, their interests are misaligned because, for a given information set, the
analyst’s preferred action exceeds the decision-maker’s preferred action. These utility
representations are broadly descriptive of institutional settings where an analyst wants
to induce a higher action than a decision-maker would prefer, but is constrained
from inducing an action that is too high because of, say, (unmodeled) reputation or
litigation concerns associated with misleading the decision-maker (e.g., Dugar and
4The presumption that investors value a firm at its expected value is standard in the voluntary
disclosure literature (e.g., Verrecchia 1990; Penno 1996; and Einhorn 2007).
5We can also assume φ<0without altering the flavor of the results.
8
Nathan 1995; Grossman and Helpman 2001; Morgan and Stocken 2003).6
In the fourth stage, the state of nature αis realized and the analyst’s and decision-
maker’s payoffs are determined. All aspects of game are common knowledge except
the analyst’s private signal y. The time line of events and the model’s notation is
summarized in Figure 1.
[Figure 1]
We st ud y Perfect Bayesian Equilibria, which require that: the players’ beliefs
satisfy Bayes’ rule whenever possible; and, given beliefs, the decision-maker’s action
amaximizes her expected payoffUd=−E[(˜α−a)2|Ωd]and the analyst’s choice
of information precision pmaximizes Ui=E[φa −(˜α−a)2−c(p)|Ωi]and, given
the cost of gathering information is sunk, the analyst’s report rmaximizes Ui=
E[φa −(˜α−a)2|Ωi].
Like most cheap-talk games, there are multiple equilibria. In our model, there
are, at most, two classes of equilibria: one class of equilibria in which the analyst
chooses the same precision p>1/2(i.e., the analyst gathers some information) and
communicates that information to the decision-maker, and one class of equilibria in
which the analyst chooses p=1/2(i.e., the analyst does not gather any information),
which is analogous to a babbling equilibrium in a cheap-talk game without endogenous
information acquisition. There always exists equilibria in which the analyst does
not gather any information. There may or may not exist equilibria in which the
analyst gathers and communicates information. When there exist equilibria in which
information is gathered, we focus on this class because these equilibria ex ante Pareto
dominate any equilibrium in which information is not gathered.7Within the class of
equilibria in which information is gathered, we focus on the truthful communication
equilibrium–the equilibrium in which r=yfor all y. This focus is without loss of
6While the Sarbanes-Oxley Act enacted on July 25, 2002 was drafted to strengthen the inde-
pendence of security analysts (Razaee, 2007) and the legal settlement between the large brokerage
firms and the State of New York in 2002 has altered the way in which analysts are compensated,
separated research and investment banking activities within investment firms, and mandated disclo-
sure in stock reports of conflicts where they might exist, there still is evidence suggesting analysts
have incentives to curry favor with firm management. For instance, Mayew (2008) finds that firm
management favors those analysts in conference calls who have bullish stock recommendations on
the firm.
7It is standard in the cheap-talk literature to focus on an ex ante Pareto dominant equilibrium;
see, for instance, Crawford and Sobel (1982).
9
generality because when there exists an equilibrium with full revelation that involves
the analyst reporting something other than y(e.g., r∈{b, g}and r6=yfor all y), it
is economically equivalent to the truthful communication equilibrium.8
3 Equilibrium
The analyst chooses the precision of his private information after observing the public
information and then decides whether to truthfully reveal his private information. We
use backward induction to characterize the equilibrium and begin with the reporting
stage before considering the information gathering stage.
In the reporting stage, the analyst decides whether to truthfully reveal his private
information after having observed the public signal and his private information. To
assess whether the analyst can communicate yto the decision-maker, note that the
analyst prefers a higher action than the decision-maker given y, and that the decision-
maker would take a higher action if she believes the analyst has observed ˜y=g.It
follows that the analyst is always willing to reveal when ˜y=g, but may want to
mislead the decision-maker when ˜y=bby claiming that ˜y=g. Recalling that the
value of the analyst’s objective is increasing and then decreasing in the decision-
maker’s action implies that the analyst will not try to mislead when ˜y=bif the
action induced from claiming that ˜y=gis much higher than the action preferred by
the analyst when ˜y=b. Hence, the analyst can communicate yif the action he would
induce by misrepresenting his information is sufficiently high relative to the one he
induces if he truthfully communicates his information.
For either realization of ˜x, the difference between the actions the decision-maker
might choose when she believes she knows yis
∆(p)≡E[˜α|x, g]−E[˜α|x, b]= q(1 −q)(2p−1) (s−f)
(p(1 −q)+(1−p)q) ((1 −p)(1−q)+pq).(3)
The difference ∆(p)is an increasing function of p, which implies that the decision-
maker’s action choice is more responsive to the analyst’s information when the an-
alyst’s information is more precise. As discussed above, the analyst can credibly
8To the extent that it is institutionally inappropriate to focus on the properties of the most
informative equilibrium when there exist a multiplicity of equilibria, then the predictive power of
the claims in this paper will be reduced.
10
communicate yiftheactionheinduceswhenheclaimstohaveobservedrealizationg
is sufficiently large relative to the action he induces when he claims to have observed
realization b,orformally
∆(p)−φ≥0.
Hence, if pis sufficiently high, the analyst can credibly communicate y. This obser-
vation is formalized in the next lemma.
Lemma 1 The analyst can communicate yin the presence of the other information
xif and only if the precision of the analyst’s information is sufficiently high; that
is, there exists a minimum threshold ¯psuch that, for any x, the analyst can credibly
communicate yif and only if p≥¯p.
Having examined the reporting stage, we now step back and consider the infor-
mation gathering stage. In this stage, the analyst chooses the precision of his private
information after having observed the public signal realization x∈{l, h}. To deter-
mine the analyst’s choice, it is useful to first examine how the analyst’s objective
function behaves in pwhen we assume the analyst’s information yis publicly ob-
served. Recalling that the decision-maker’s action choice equals the expectation of ˜α,
the analyst’s objective function is
Ey£φE[˜α|x, y]−(˜α−E[˜α|x, y])2−c(p)|x, y¤
=φEy[E[˜α|x, y]] −p(1 −p)(s−f)∆(p)
2p−1−c(p).
The next lemma establishes how the analyst’s objective function behaves in the ana-
lyst’s precision choice p.
Lemma 2 When the analyst’s information yis public, the analyst’s objective function
is strictly quasiconcave and attains a maximum at p∗∈(1/2,1) where p∗is such that
∆2(p∗)/(2p∗−1) −c0(p∗)=0.(4)
Lemma 2 implies the analyst has a single-peaked objective function that is max-
imized at p∗when the analyst’s signal yis public information. The analyst’s signal
y, however, is not publicly observed. Lemma 1 implies that pmust be at least ¯pfor
credible communication to occur when the analyst’s signal is not publicly observed.
11
It follows from these two lemmas that, when yis not public information, the analyst
chooses p=p∗if p∗≥¯p.Alternatively,ifp∗<¯p, the analyst chooses p=1/2or
p=¯p. The next proposition extends this discussion and characterizes the analyst’s
optimal choice.
Proposition 3 Suppose the analyst privately observes yand both players observe the
other information x.
(i) If φ∆(p∗)/(2p∗−1) ≤c0(p∗), then the analyst chooses a level of precision
p∗∈[¯p, 1) and communicates truthfully in the reporting stage.
(ii) If φ∆(p∗)/(2p∗−1) >c
0(p∗)and c(¯p)≤∆(¯p)q(1 −q)(2¯p−1) (s−f),
then the analyst chooses a level of precision ¯pand communicates truthfully in the
reporting stage.
(iii) If φ∆(p∗)/(2p∗−1) >c
0(p∗)and c(¯p)>∆(¯p)q(1 −q)(2¯p−1) (s−f),
then the analyst chooses to not collect any private information.
Proposition 3 assumes the analyst chooses the precision of his private information
after observing the realization of the public information. Analysts, however, often
choose to gather information in anticipation of a public information event. Because
∆(p)does not depend on whether the public signal realization is either ˜x=hor
l, it follows that the characterization of the equilibrium in Proposition 3 would not
change if we instead assumed that the analyst chose the precision of his information
before observing the public information. Further, there also exists an equilibrium in
which the analyst does not gather any information in cases (i) and (ii) in Proposition
3. As we stated earlier, if there exists an equilibrium in which information gathering
occurs, we assume it is the equilibrium that is played.
4AnalystInformation
While Proposition 3 characterizes the analyst’s information gathering decision, it
offers little insight into the determinants of the information the analyst chooses to
gather and disseminate. To address our primary research question, we examine how
the quality of the other information available to the decision-maker, q,affects the
analyst’s information gathering decision, p. From Proposition 3, we know that the
analyst chooses information quality p∗if credible communication of the resulting
information is possible, specifically p=p∗≥¯p. Otherwise, the analyst chooses
12
a higher level of information quality to allow communication, p=¯p>p
∗,orhe
chooses to exit the market if information of higher quality is too costly to obtain,
that is p=1/2. Assessing how the quality of the public information, q,influences
the analyst’s behavior entails determining how qaffects the critical values p∗and ¯p.
The following corollary, which follows directly from Lemmas 1 and 2, characterizes
the relation between qand the critical precision values p∗and ¯p.
Corollary 4 The minimum precision of the analyst’s private information necessary
for the analyst to credibly communicate, ¯p,isincreasing in the precision of public
information q;thatis,∂¯p/∂q > 0. The precision of the analyst’s information chosen
when the analyst’s information is public, p∗,isdecreasing in the precision of the public
information q;thatis,∂p∗/∂q < 0when q6=1/2and ∂p∗/∂q =0when q=1/2.
To develop intuition for the relation between qand ¯p, note that when the precision
of the public information is low, the decision-maker’s action choice is more responsive
to the analyst’s information because the analyst information is relatively precise. As
a consequence of this greater responsiveness, the analyst is less inclined to exaggerate
and report gwhen bis actually observed because, if the report is believed, the decision-
maker takes an action that is undesirably high from the analyst’s perspective. On the
other hand, the analyst is less capable of credibly communicating when the public
information is precise because the decision-maker’s action choice is less responsive to
the analyst’s information. In this case, the analyst is more inclined to exaggerate and
report gwhen bis actually observed because, if the report is believed, the decision-
maker takes a higher action but one that is not so high that the analyst finds it
undesirable. If there are incentives for such exaggeration, however, the analyst is not
believed and communication does not occur in equilibrium. The relation between
qand p∗is straight forward. As the precision of public information increases, the
marginal benefit of the analyst’s information decreases. Consequently, the level of p
that maximizes the analyst’s expected utility falls.
With Corollary 4 in hand, consider the case when the precision of the public
information is relatively low so that p∗≥¯p. As the precision of the public information
increases, ¯prises and p∗falls. Because the analyst chooses p∗when p∗≥¯p,the
analyst’s choice of pfalls in qbecause p∗falls in q. Eventually the precision of the other
information rises to a point where p∗=¯p.Asqcontinues to rise, communication is no
longer credible if the analyst chooses p∗. Hence, the analyst must choose between the
13
level of precision necessary for credible communication ¯por drop out of the market by
choosing p=1/2. Initially, the analyst responds by choosing ¯pand over-investing in
the quality of information he gathers to allow credible communication. Given that the
¯pis increasing in q, it follows that the precision of the analyst information increases
in q. Atsomepoint,however,implementing¯pbecomes so costly that the analyst
chooses not to collect any information. In these circumstances, the improvement in
the public information causes the analyst to under-invest in the quality of information
he gathers relative to that which he would gather if he could commit to credibly reveal
his private information. In summary, we have the following observation.
Remark 1 The precision of the information the analyst collects is decreasing in the
precision of the public information, then increasing in the precision of the public
information, and, at some sufficiently high level of precision of the public information,
the analyst stops collecting private information; that is, there exists a q∗and ¯q>q
∗
such that the equilibrium precision of the analyst’s private information is decreasing in
qover the range (1/2,q
∗],increasinginqover the range (q∗,¯q], and is uninformative
for q>¯q.
Remark 1 provides some insight into the recent evolution of the sell-side analyst
industry. Recent regulatory changes have greatly increased the amount of information
firms make available directly to investors. For instance, following the introduction of
Reg FD, firms can no longer privately reveal information to analysts so firms are more
likely to publicly release their forward-looking information (see Heflin, et al., 2003). In
addition, the $1.4 billion settlement between ten large brokerage firms and the State of
New York requires these firms to distribute research reports that independent analysts
have prepared along with their own analysis (see Schack 2003). Finally, firms now
routinely host conference calls following earnings releases that are open to investors
(see Bushee, et al. 2003). These changes have eroded stock analysts’ information
advantage. Remark 1 suggests that, in response to these changes in the information
environment, some intermediaries have reduced the information they collect, others
have increased the quality of their analysis in response to the heightened competition
from other sources, and yet others have discontinued coverage of firms.
Remark 1 is empirically unsatisfying since it suggests “anything can happen”.
To offer more definitive guidance as to how intermediaries might alter the quality of
their analysis in response to changes in public information, we consider how some
14
environmental characteristics might influence analyst behavior. In particular, we
assess how the extent of the incentive conflict, captured by φ,andthedegreeofprior
uncertainty about the firm’s payoffs, as captured by (s−f), would alter the analyst
behavior. Doing so, however, requires that we first examine how changes in these
characteristics affect the critical values of precision ¯pand p∗.
Corollary 5 The minimum precision of the analyst’s private information necessary
for the analyst to credibly communicate, ¯p, is increasing in the extent of incentive
misalignment φand decreasing in the difference between the state payoffs(s−f);
that is ∂¯p/∂φ > 0and ∂¯p/∂ (s−f)<0. The precision of the analyst’s information
chosen when the analyst’s information is public, p∗,isunaffected by the extent of
incentive misalignment φand is increasing in the difference between the state payoffs
(s−f);thatis∂p∗/∂φ =0and ∂p∗/∂ (s−f)>0.
To develop intuition for how ¯pbehaves, observe that the analyst is more inclined
to truthfully reveal his privately observed signal when the decision-maker’s actions are
highly responsive to his disclosure. The decision-maker is more responsive when the
analyst’s information is relatively more precise or the state payoffs are more divergent.
When the extent of incentive misalignment is large, the minimal precision of analyst
information necessary for credible communication must be high. Alternatively, when
the state payoffs are far apart and prior uncertainty is large, the decision-maker will
be more responsive to the analyst’s information. Therefore, the minimal precision of
the analyst’s information necessary for credible communication need not be high.
We now turn to how p∗behaves. This value is derived assuming the analyst’s
information is publicly observed. Accordingly, it is clear that the extent of the in-
centive misalignment between the players φshould not influence the choice of p∗.In
contrast, as the prior variance of the firm’s payoffs, which is proportional to (s−f),
increases, both players prefer more precise information.
Given the intuition underlying Corollary 5, we consider how the extent of incentive
misalignment φinfluences the analyst’s response to increases in the precision of other
information. If the extent of incentive misalignment is low, then p∗will exceed the
minimum level of precision necessary for credible communication ¯p. The analyst
therefore will choose p∗. Corollary 4 establishes that an increase in the quality of
public information is associated with a marginal decrease in the analyst’s precision
choice p∗. If the extent of misalignment is moderate, so that the analyst chooses
15
the minimum level of precision to facilitate credible communication ¯p, Corollary 4
shows that an increase in the quality of the public information is associated with a
marginal increase in the precision of the analyst’s information. Finally, if the extent of
misalignment is high, the analyst responds to an increase in the precision of the public
information by not gathering any addition information or performing any analysis.
Within the sell-side analyst context, these observations suggest that intermediaries
with large incentive conflicts, such as those whose employers do a great deal of banking
business (see Lin and McNichols 1998) or those who have substantial equity positions
in the firm’s stock, may be more incline to drop coverage of that firm in response to the
increase in public information. In contrast, analysts that face few conflicts of interest
will respond more modestly and simply reduce their analysis. Finally, analysts with
more moderate conflicts of interest will be more incline to increase their analysis in
response to increases in the availability of public information.
The extent of prior uncertainty also influences the response of intermediaries to
the increase in public information. Decision-makers might be more uncertain about
afirm’s payoffsifthefirm is in an industry that uses an innovative and unproven
technology. Corollaries 4 and 5 suggest that, in response to an increase in public
information, firms with highly risky payoffs will face relatively modest declines in
analyst attention, those with low risk payoffs will have analysts drop coverage, and
those with moderately risky payoffs will receive greater analyst attention.
In summary, the paper highlights that public information influences analysts in
two ways. First, improvements in public information make it more difficult for the
analyst to communicate (i.e., ¯pis increasing in q). Second, improvements in public
information reduce the benefit of more precise analyst information because the public
information substitutes for the analyst’s information (i.e., p∗is decreasing in q). When
communication credibility is not an issue (i.e., ¯pis small relative to p∗), the analyst
reduces the quality of his analysis in response to more precise public information. In
contrast, when analyst communication credibility is an issue, the analyst responds
to more precise public information by either gathering more precise information or
alternately dropping coverage of firms.
Our model, which emphasizes the role of reporting credibility for understanding
the information intermediary’s behavior, contrasts with the discretionary disclosure
models commonly examined in the accounting literature that assume any disclosure
must be truthful, although the manager can choose to withhold information. These
16
papers offer reasons for why managers might not voluntarily release private informa-
tion. In this paper, we suggest analysts might not disclose non-verifiable information
because, in anticipation of being unable to credibly reveal it, they choose to not
gather any information. Further, we find the presence of a reporting stage in a model
in which the analyst’s credibility is an important ingredient affects the precision of
information that the analyst gathers in subtle ways. It can cause an analyst to under
or over-invest in information collection relative to the model where any disclosure is
always truthful.
5 Quality of Decision-Maker Information
We considered how the precision of the analyst’s information changes in response
to increases in the precision of the other information. In some cases, we demon-
strate that the analyst reduces the quality of analysis in response to an increase in
precision of public information. An unanswered question in these cases is: What
happens to the precision of the decision-maker’s information? In this section, we
examine how the quality of the decision-maker’s information varies with the precision
of the public information. We define the quality of information as a function of the
decision-maker’s expected posterior variance after the public signal and analyst re-
port, −Ey[var [˜α|Ωd]]. It can be shown that decision-maker’s objective function and
action choice is such that the quality of the decision-maker’s information is equivalent
to the decision-maker’s expected utility.
Consider the case in which the analyst chooses a level of precision that exceeds
the minimum level of precision necessary to credibly communicate in the reporting
stage prior to an increase in the precision of the other information, that is p∗>¯p.
The decision-maker’s quality of information when the analyst chooses p=p∗is given
by
−Ey[var [˜α|x, y;p∗]] = −p∗(1 −p∗)q(1 −q)(s−f)2
(p∗(1 −q)+(1−p∗)q) ((1 −p∗)(1−q)+p∗q).
As the precision of the public information qincreases, it follows from our previous
analysis that the precision of the analyst information decreases because the other in-
formation serves as a substitute for the analyst information. Nevertheless, we observe
that an increase in precision of the public information qleads to an increase in the
17
decision-maker’s information; that is, ∂(−Ey[var [˜α|x, y;p∗]]) /∂q > 0.
Consider the case where the precision of the other information qhas increased to a
point where the analyst optimally chooses the minimum level of precision that allows
credible communication, that is p=¯p. In this case, the decision-maker’s quality of
information is given by
−Ey[var [˜α|x, y;¯p]] = −¯p(1 −¯p)q(1 −q)(s−f)2
(¯p(1 −q)+(1−¯p)q) ((1 −¯p)(1−q)+ ¯pq).
When the truth-telling condition in the reporting stage is binding, the analyst in-
creases the precision of information he collects as the precision of the public in-
formation increases. Here the presence of the public information has a comple-
mentary effect on the analyst’s information gathering activities. The increase in
the precision of the analyst’s information coupled with the increase in the precision
of the public information continues to cause the decision-maker’s information qual-
ity to increase with an increase in the precision of the public information; that is
∂(−Ey[var [˜α|x, y;¯p]]) /∂q > 0.
Finally, consider the case where the precision of the other information becomes
so large that the analyst’s expected utility when ¯pis chosen is less that the analyst’s
expected utility when no information is collected, that is p=1/2. In this case, the
analyst simply decides not to gather any information about the company–the analyst
drops firm coverage. Since the decision-maker only observes the public information
x, the decision-maker’s quality of information when p=1/2is given by
−Var[˜α|x]=−q(1 −q)(s−f)2.
This latter observation implies that, an increase in qabove some value ¯qcauses the
decision-maker’s information quality to fall discontinuously at ¯q.Thus,theincrease
in public information drives out the analyst’s willingness to collect and communicate
his private information, which makes the decision-maker worse off. Of course, as the
precision of the public information continues to increase, the quality of the decision-
maker’s information increases and the decision-maker’s information quality attains a
maximum value at q=1. This observation is formalized in the next Corollary.
Corollary 6 As the precision of the public information increases from q=1/2,the
total information the decision-maker obtains initially increases, falls discontinuously
18
at ¯qwhen the analyst no longer gathers private information, and then increases as q
approaches one.
Corollary 6 suggests that improving the quality of public information can crowd
out the analyst’s ability to credibly communicate his private information. The ex-
pected quality of the decision-maker’s information does not monotonically increase in
the precision of the public information. In the financial reporting environment, how-
ever, policy-makers do not choose the precision of the public information. Instead,
they decide whether or not information of a given precision should be gathered and
disclosed. In essence, a policy-maker’s choice is between disclosure and no disclosure.
To assess whether the crowding out phenomenon can make no disclosure the policy-
maker’s preferred choice, we analyze the difference between the decision-maker’s in-
formation quality in the absence of disclosure, which is equivalent to q=1/2,and
the information quality when other information of precision qis publicly disclosed.
The goal now is to determine whether there are values of the precision of the public
information qfor which the decision-maker is better offin the absence of the public
information.
We show that the introduction of public information can make the decision-
maker worse offwithin the context of an example with a cost function c(p)=
((p−1/2) /(1 −p))nwhere n>2.Sets=2,f=3/20,φ=1/2,andn=3.
On one hand, consider the setting in which public information is absent. The analyst
chooses ¯p=1/2+φ(2 (s−f)) = 47/74 >p
∗; he gathers more precise information
than he would gather if he could commit to truthfully reveal his privately observed
information. Given the analyst chooses a level of precision ¯pand communicates truth-
fully in the reporting game, the decision-maker’s quality of information is given by
−Ey[var [˜α|y;¯p]] = −1
4¡(s−f)2−φ2¢.
On the other hand, consider an environment where the public disclosure of information
is mandatory. It is sufficiently costly to gather information in this setting that the
analyst prefers to not gather any private information at all; the analyst chooses
p=1/2for all q∈(1/2,1). Although the presence of any public information crowds
out the analyst’s information, the decision-maker benefits from observing the public
19
information. The decision-maker’s quality of information when p=1/2is given by
−Var[˜α|x]=−q(1 −q)(s−f)2.
When the precision of the public information is such that q∈(1/2,¯p), the benefit
to the decision-maker of observing the public information is not sufficient to offset
the analyst’s information that it displaces. The introduction of public information
reduces the total quality of the decision-maker’s information and makes her worse off.
Alternatively, when the precision of the public information is such that q∈(¯p, 1),
then the public information is sufficiently precise that its introduction makes the
decision-maker strictly better off, even though it squashes the analyst’s willingness to
gather private information.
This observation that additional public information may reduce the players welfare
is reminiscent of a result in Christensen’s (1982) study of performance standards in an
agency setting. In that study, Christensen shows that the principal may be worse-off
when the players observe additional contractual information before the agent takes an
action. Our setting differs from an agency setting because, in our setting, a receiver
cannot commit (i.e., contract) to use a report in a particular way. Consequently, the
character of our equilibrium and the economic forces delivering our results differ from
those in Christensen (1982).
To digress briefly, we have raised the possibility that introducing mandatory disclo-
sures of other information can reduce the total information available to the decision-
maker when the state space is binary. One might be concerned that this result can
arise only when the state space is discrete thereby allowing the analyst either to com-
municate truthfully or not to communicate at all. This possibility result, however,
can also be established when the state space and message space is continuous.
The fact that variations in the precision of the public information can harm the
decision-maker arises not because the state space or message space is discrete, but
because of the discontinuity in the precision with which the analyst can communicate
his private information. A key feature of our model is that the analyst does not
bear a direct cost from issuing any specific message. The analyst, however, incurs
an indirect cost from inducing the decision-maker to take an action that affects his
expected payoff. When there are no direct reporting costs and the incentives of
the analyst and decision-maker are not perfectly aligned, Crawford and Sobel (1982)
20
established that the unique equilibrium is characterized by a partition of the state
space and the analyst is only able to credibly reveal the interval containing his signal as
opposed to the signal itself. The lengths of the intervals in a partition vary. Therefore,
the precision with which the analyst can communicate his private information varies
discontinuously with his signal realization.
The following example illustrates the possibility of public information harming
the decision-maker within a game that features a continuous state space and message
space. To simplify the illustration, consider the reporting game when the analyst’s
information gathering activity (that depends on the specificcostfunction)istaken
as given. Thus, the decision-maker’s and analyst’s utility functions are specified
in (1) and (2), respectively, but with c(p)=0; set φ=1/4. Assume the state
variable, ˜α, is uniformly distributed on the unit interval, and the analyst privately
observes the state variable y=α. The analyst and decision-maker publicly observe
the sum of two independent signals x=P2
i=1 vi, where each signal, vi,isdrawn
from a Bernoulli distribution with an unknown parameter α;thatis,Pr (˜vi=1)=α
for i=1,2. Therefore, the decision-maker’s posterior distribution of ˜αis a beta
distribution with parameters 1+xand 3−x(see DeGroot 1970). Given these beliefs,
when ˜x∈{0,1}, the analyst cannot reveal his information. In contrast, when ˜x=2,
the analyst’s message rcan reveal, at most, whether the state lies in one of three
intervals {[0,0.06] ,(0.06,0.43] ,(0.43,1]}.Whenthesignalxis publicly observable,
the decision-maker’s information quality −E[Var[˜α|x, r]] = −0.04.However,when
the public signal xis unavailable, the analyst’s message can reveal whether the state
lies in one of two intervals {[0,0.25] ,(0.25,1]}, and the decision-maker’s information
quality −E[Var[˜α|r]] = −0.03. Hence, the decision-maker is ex ante better offwhen
the other information is unavailable.9
9When the state is continuously distributed and the other information variable ˜xis such that
it does not change the support of the decision-maker’s beliefs, analytic solutions to the questions
examined in this paper generally cannot be obtained. Characterizing the partitional equilibria to
this game requires finding solutions to non-linear, second-order difference equations. These equations
generally do not have analytic solutions. To obtain analytic solutions, we accordingly restrict the
support of the information variables in the model.
21
6 Unobservable Choice of Precision
In the primary model, the analyst’s choice of precision pis common knowledge. This
assumption is based on the observation that it is often possible to infer an analyst’s
expertise from the depth and clarity of the stock report, even though it is not possible
to infer exactly what the analyst believes. At times, however, the decision-maker
may be unable to determine the analyst’s expertise. As a consequence, in this section
we modify the model and assume the analyst privately chooses the precision of his
information.
As in the analysis of the primary model, it is again useful to initially consider a
setting where the analyst’s signal yis public information. After understanding that
less constrained setting, we consider the effect of the credibility problem faced by
the analyst. In contrast to the analysis in the primary model in which pis public
information, analyzing the setting in which the analyst’s signal yis public information
involves a bit more of a game. In this game where pis not public information, the
decision-maker conjectures the analyst’s choice of pand the analyst conjectures the
decision-maker’s response to the analyst’s report. In the equilibrium to this game,
the choices of both players must maximize their objective functions conditional on
each having conjectures that are consistent with the other player’s equilibrium choices.
Denote the decision-maker’s conjecture regarding the analyst’s choice of pas bp.When
the public information xand analyst information yis realized, the decision-maker’s
action choice must satisfy
axg =bpt
bpt +(1−bp)(1 −t)s+(1 −bp)(1 −t)
bpt +(1−bp)(1 −t)f(5)
and
axb =(1 −bp)t
(1 −bp)t+bp(1 −t)s+bp(1 −t)
(1 −bp)t+bp(1 −t)f,(6)
where t=qwhen ˜x=hand t=1−qwhen ˜x=l. It follows that, in any equilibrium
s≥axg ≥axb ≥f.
Denote the analyst’s conjecture regarding the decision-maker’s response to realiza-
tion (x, y)as ˆaxy. Since in any equilibrium s≥axg ≥axb ≥f, consider the analyst’s
decision choice given xand conjectured actions s≥ˆaxg ≥ˆaxb ≥f.Theanalyst’s
22
choice of pmust maximize
tp(φbaxg −(baxg −s)2)+(1−t)(1 −p)(φbaxg −(baxg −f)2)
+t(1 −p)(φbaxb −(baxb −s)2)+(1−t)p¡φbaxb −(baxb −f)2¢−c(p).
Given that c(p)is sufficiently convex, an interior choice of precision pthat satisfies the
first-order condition maximizes the analyst’s objective function. Hence, the analyst’s
choice of pin an equilibrium with an interior choice of pmust satisfy
(baxg −baxb)(t(2s−baxg −baxb )+(1−t)(baxg +baxb −2f)+φ(2t−1)) −c0(p)=0.(7)
An equilibrium precision choice pand action choice function axy must satisfy (5),
(6), and (7) for each x∈{h, l}after substituting the players’ equilibrium choices for
their conjectured choices. Substituting in pfor ˆpin (5) and (6) and then taking those
actions and substituting them in for baxg and baxb in (7) yields the following condition
that any equilibrium choice for p,namelyp∗
no,mustsatisfy
∆2(p∗
no)/(2p∗
no −1) −c0(p∗
no)+∆(p∗
no)φ(2t−1) = 0.(8)
Proposition 7 When the analyst’s choice of precision is unobservable and the ana-
lyst’s information yis public, there always exists an equilibrium precision choice, p∗
no,
in which the analyst gathers no information, p∗
no =1/2. In addition, if the public
signal is favorable, ˜x=h, there always exists an equilibrium precision choice, p∗
no,
such that p∗
no ∈(1/2,1). Finally, if the public signal is unfavorable, ˜x=l,theremay
exist one or more equilibrium precision choices such that p∗
no ∈(1/2,1).
The first point worth noting in Proposition 7 is that there always exists an equilib-
rium in which the analyst does not provide any information. This equilibrium, which
is analogous to the babbling equilibrium that always exists in a standard cheap-talk
game, arises because: (1) the analyst has no incentive to gather information if the
decision-maker does not respond to the resulting signal, and (2) the decision-maker
has no reason to respond to a signal that is believed to be uninformative. Thus, in
contrast to cheap-talk games without endogenous information collection that always
feature a babbling equilibrium in which the analyst misrepresents his private infor-
mation, in this setting in which information collection is endogenous, we find that
23
there is always an equilibrium in which the analyst does not provide the decision-
maker with any information even though the analyst’s signal is public information
and hence cannot be misrepresented. The second point worth noting is that there
always exists one, and perhaps more than one, equilibrium in which the analyst does
gather information for at least one of the two realizations of the public signal. It is
these interior equilibrium choices for pthat are the focus of our remaining analysis.
On comparing the equilibrium condition for p∗
no given in (8) with that for p∗given
in (4), we observe that any interior precision of the analyst’s information depends
on the public information, which determines the value for t.Considerfirst the case
when the public information is favorable so ˜x=hand t=q.Inthiscasethere
exists only a signal interior equilibrium choice p∗
no. Furthermore, the observation that
∆(p∗
no)φ(2t−1) >0implies that this choice always exceeds the choice that would
be made if pwere observable, p∗
no >p
∗. The intuition underlying why the precision
choice exceeds the precision choice in the case where pis observable is as follows.
When the favorable public signal is observed, the analyst believes his signal is more
likely to be a good one if he selects a higher level of precision. Because the analyst
prefers a higher action from the decision-maker, it follows that the analyst has an
incentive to choose a higher precision level than he otherwise would choose in order
to increase the probability of getting a higher level of action from the decision-maker.
The cost of precision, however, still provides a disincentive to choosing the maximum
precision level and an interior choice of precision is sustained as an equilibrium.
Consider the case when the public information is unfavorable so ˜x=land t=1−q.
In this case, any equilibrium choice for the precision pis exceed by the choice made
when the the precision choice is observable because ∆(p∗
no)φ(2t−1) <0.Inthiscase
when the unfavorable public signal is observed, the analyst believes his signal is more
likely to be good when the precision level he selects is lower. Given that he prefers
a higher action from the decision-maker, it follows that the analyst has an incentive
to lower the precision level than he otherwise would choose in order to increase the
probability of inducing the higher action. An interior equilibrium can still exist in
face of the incentive to choose a lower precision level, however. For such an interior
equilibrium to exist, the decision maker’s response to the good signal must result in
an action that is too high from the perspective of the analyst who has chosen p=1/2.
Consequently, the analyst chooses a level of precision p∗
no >1/2.10
10These predictions contrast with Penno (1996) who considers a setting where a firm manager
24
With an understanding of the game when yis public information established,
we next consider how credibility issues affect equilibrium outcomes by assuming the
analyst privately observes y.Considerfirst the case when the favorable public signal is
realized, ˜x=h. In this case there exists one interior value of precision that can be an
equilibrium outcome in the game where credibility is not an issue. When credibility is
also an issue, this value of precision, p∗
no,isstillsustainedasanequilibriumoutcome
ifandonlyifitisweaklygreaterthan¯p. Otherwise, the only equilibrium is one in
which the analyst gathers no information, p=1/2. Next consider the case when the
unfavorable signal is realized, ˜x=l. If there exists an interior equilibrium in this
case, it must again be true that p∗
no exceeds ¯p. Note, however, that the likelihood
that this condition is satisfied when ˜x=lis lower because p∗
no must be lower in the
case when ˜x=lthan in the case when ˜x=h. Hence, the analyst is more likely to
not collect any information when the public information is unfavorable.
The observation that the analyst’s precision choice varies with firm performance
provides an explanation of empirical evidence regarding analyst coverage that Mc-
Nichols and O’Brien (1997) document. They find that analysts are more likely to
discontinue coverage on a firm that is performing poorly. The story they offer as an
explanation for this finding is that analysts want to avoid jeopardizing their prospects
of winning investment banking business by avoiding issuing a sell recommendation.
Our model provides another explanation for the correlation they observe in the data:
when news is unfavorable, analysts are less likely to be able to credibly communicate
their information and, as a consequence, they do not gather any information–they
discontinue coverage.
7Conclusion
The Sarbanes-Oxley Act of 2002 and the settlement between the large brokerage
firms and the State of New York in 2002 have led to shrinking of research depart-
ments at investment firms. The exodus of analysts has left many firms and, indeed,
observes a public signal about the firm and then privately chooses the precision of a report that the
firm will subsequently issue. He finds that the precision of the report that the manager chooses is
decreasing in the favorableness of the public signal because it lowers the weight investors will place
on the subsequent report, which is likely to be less favorable than the previously observed public
signal. In our model, the public signal and the analyst’s private signal are positively correlated.
Therefore, the analyst chooses to gather more precise information following a favorable public signal
because he is more likely to observe a favorable signal.
25
entire industry sectors without analyst coverage, raising the ironic possibility that the
new regulations might have made investors that rely on these information intermedi-
aries worse off(see Schack 2003). Furthermore, recent regulatory changes, including
Reg FD released in 2000, have greatly increased the amount of information available
directly to investors. Against this background, this paper examines analyst informa-
tion gathering and reporting activities in a setting in which the analyst chooses the
precision of information to collect in response to changes in the precision of public
information about a firm. A key feature of the model that distinguishes it from the
extant reporting literature is that the analyst is not restricted to truthfully report.
Rather, recognizing that analyst credibility is an important feature of the analyst
environment, we assume that the analyst can report in a self-serving fashion.
Using endogenous information collection coupled with reporting credibility as pri-
mary ingredients, this model offers the following insights. First, we consider a set-
ting where the analyst’s precision choice is publicly observed and examine how the
analyst’s information gathering activities vary with the precision of the public in-
formation. When the analyst’s reporting credibility is not an issue, increases in the
precision of the public information cause the analysts to reduce the precision of in-
formation they privately collect and communicate. Nevertheless, because the public
information substitutes for the private information, investors are better off.Onthe
other hand, when an analyst reporting credibility becomes salient, an increase in the
precision of public information causes the analyst to collect more precise information.
In this case, the public information has a complementary effect on the analyst infor-
mation gathering activities and an investor is again made better off. However, if the
precision of the public information continues to increase, then a threshold is reached
where an analyst can no longer economically gather sufficient information necessary
to credibly communicate with investors. As a consequence, the analyst drops coverage
of the firm or leaves the market. The analyst’s decision to drop coverage makes the
investors worse off–which aligns with the concern that the recent regulations might
make investors worse off.Moreover,wefind, in some circumstances, that investors
would be better offif the public information was never introduced than to have the
public information that drives analysts from the market.
While an analyst stock report might often evidence the precision of the analyst’s
information, at times investors might be uncertain about the precision of the analyst
information. We find that the analyst, after observing favorable public information,
26
gathers more precise information when the precision choice is not publicly observable
than when the analyst’s precision choice is publicly observable, and conversely, the
analyst, after observing unfavorable public information, chooses less precise informa-
tion or is more likely to discontinue coverage when the precision choice is not publicly
observable relative to when the analyst’s precision choice is common knowledge.
Because of the import of the recent changes in the structure of the analyst envi-
ronment, the paper has focused on the information gathering and reporting activities
of analysts. The model, however, is quire general. The insights might be applied to
the voluntary information gathering activities of a firm manager in the presence of
changes in the mandatory reporting environment, particularly when the manager’s
and investors’ interests are imperfectly aligned and the manager’s voluntary disclosure
is not verifiable–such as in the case of management earnings forecasts. The insights
also might be applied to the information gathering and communicating behavior of a
seller of an experience good when the potential buyer has access to another sources of
unbiased information–such as Consumer Reports–that reveals the properties of the
seller’s product. Alternatively, the insights might be applied to the interaction be-
tween a venture capitalist and a firm that has private information about an innovative
technology for which it is seeking venture financing.
27
8 Appendix
This appendix contains the proofs of the lemmas, propositions, and corollaries.
Proof establishing power cost function is well defined:
Consider the power cost function
c(p)=µp−1/2
1−p¶n
where n>2.The first and second derivatives are
c0(p)= n
2(p−1/2)(1 −p)c(p)>0
and
c00(p)= n(n+4p−3)
(2(p−1/2)(1 −p))2c(p)>0if n>1.
The second derivative divided by the first derivative is
c00(p)
c0(p)=n+4p−3
2(p−1/2)(1 −p).
Observe that
(1 −3p(1 −p))
(p(1 −p)(p−1/2)) <n+4p−3
2(p−1/2)(1 −p)if and only if 2p2−3p+2
p<n
Since the expression (2p2−3p+2)/p is strictly decreasingly in p, it follows that the
inequality is always satisfied if n>2.
ProofofLemma1:
At this stage the analyst’s cost of gathering information is sunk and p,whichis
commonly observed, is fixed. If the analyst communicates his private information
ywhen the other information xis realized, then the decision-maker’s action after
receiving the analyst’s report is a=E[˜α|x, y]. Given the decision-maker’s action,
the analyst reports observing gif and only if
E£φE[˜α|x, g]−(˜α−E[˜α|x, g])2|g, x¤≥E£φE[˜α|x, b]−(˜α−E[˜α|x, b])2|g, x¤.(9)
This truth-telling condition is trivially satisfied because the analyst never wants to
28
dissemble and induce a lower action than the one the decision-maker would take given
the analyst’s information. The analyst reveals observing bif and only if
E£φE[˜α|x, b]−(˜α−E[˜α|x, b])2|b, x¤≥E£φE[˜α|x, g]−(˜α−E[˜α|x, g])2|b, x¤.
(10)
This condition can be shown to be satisfied if and only if expression
E[˜α|x, g]−E[˜α|x, b]−φ≥0(11)
is satisfied.
It follows from expression (11) that a necessary and sufficient condition for the
analyst to reveal ywhen ˜x=lis
E[˜α|l, g]−E[˜α|l, b]−φ
=(2p−1) q(1 −q)(s−f)
(p(1 −q)+(1−p)q) ((1 −p)(1−q)+pq)−φ≥0,(12)
and for the analyst to reveal ywhen ˜x=his
E[˜α|h, g]−E[˜α|h, b]−φ
=(2p−1) q(1 −q)(s−f)
(p(1 −q)+(1−p)q) ((1 −p)(1−q)+pq)−φ≥0.(13)
The incentive compatibility expressions (12) and (13) are identical; hence, without
loss of generality consider (12). When q=1/2, then (12) can be expressed as
(2p−1) (s−f)−φ≥0.Solving for pyields p≥1/2+φ/ (2 (s−f)). On the other
hand, when q∈(1/2,1),Solvingforpin (12) yields two roots:
p1=1
2−2q(1 −q)(s−f)+q4q2(1 −q)2(s−f)2+φ2(2q−1)2
2φ(2q−1)2<1
2
and
p2=1
2+q4q2(1 −q)2(s−f)2+φ2(2q−1)2−2q(1 −q)(s−f)
2φ(2q−1)2>1
2.
29
Hence, the truth-telling constraint requires
p≥¯p≡1
2+q4q2(1 −q)2(s−f)2+φ2(2q−1)2−2q(1 −q)(s−f)
2φ(2q−1)2.(14)
ProofofLemma2:
Given the first-order condition, the optimal choice of p∗is such that
∂Ui(.)/∂p =(2p−1) q2(1 −q)2(s−f)2
((1 −p)q+(1−q)p)2(qp +(1−q)(1−p))2−c0(p)
=∆2(p∗)
(2p−1) −c0(p∗)=0.
Since Uiis a twice differentiable function, evaluating ∂2Ui/∂p2yields
∂(∆2(p)/(2p−1) −c0(p))
∂p =2∆(p)∆0(p)
(2p−1) −2∆2(p)
(2p−1)2−c00 (p).
Evaluating the expression ∂2Ui/∂p2at p∗where p∗is such that c0(p∗)=∆2(p∗)/(2p−1)
yields
2c0(p∗)∆0(p)
∆(p)−2c0(p∗)
(2p−1) −c00 (p).
For a unique maximum it must be the case that
c00 (p)
c0(p∗)>µ2∆0(p)
∆(p)−2
(2p−1)¶.
Substituting in for ∆(p)in the right hand side of the inequality gives
µ2∆0(p)
∆(p)−2
(2p−1)¶=2
(2p−1) µ1
p(1 −p)(2q−1)2+q(1 −q)−3¶.
Since the expression is increasing in q,thefollowingisasufficient condition for the
objective function to be strictly quasiconcave
c00 (p)
c0(p∗)>1−3p(1 −p)
p(1 −p)(p−1/2),
which the cost function c(p)is assumed to satisfy.
30
ProofofProposition3:
For the analyst to credibly communicate his private information yin the reporting
game, it follows from (13) that pmust be such that
q2(1 −q)2(2p−1) (s−f)2
((1 −p)q+(1−q)p)2(qp +(1−q)(1−p))2
−φq (1 −q)(s−f)
((1 −p)q+(1−q)p)(qp +(1−q)(1−p))
≥0.
Further, it follow from (4) that p∗is such that
q2(1 −q)2(2p∗−1) (s−f)2
((1 −p∗)q+(1−q)p∗)2(qp∗+(1−q)(1−p∗))2−c0(p∗)=0.
It follows that p∗≥¯pif and only if
φq (1 −q)(s−f)
((1 −p∗)q+(1−q)p∗)(qp∗+(1−q)(1−p∗)) ≤c0(p∗).
When p∗∈(1/2,¯p), the analyst will only gather information if he can credibly
communicate it in the reporting game. Thus, to determine the precision of informa-
tion the analyst will choose to gather, we compare the analyst’s expected utility from
gathering the minimum precision of information that will allow him to credibly com-
municate in the reporting game, i.e., p=¯p,denotedUi(¯p), with the expected payoff
from not gathering any private information, i.e., p=1/2,denotedUi(1/2).When
p∗∈(1/2,¯p), the analyst prefers ¯pif and only if
Ui(1/2) −Ui(¯p)<0,
or equivalently,
c(¯p)<q2(2 ¯p−1)2(1 −q)2(s−f)2
(¯p(1 −q)+(1−¯p)q)((1−¯p)(1−q)+ ¯pq).(15)
Hence, if p∗∈(1/2,¯p)and if (15) is satisfied, then the analyst gathers more
information than he would if his incentives were not misaligned with those of the
decision-maker–i.e., the analyst over-invests in the quality of information he gathers.
31
Conversely, if p∗∈(1/2,¯p)and if (15) is not satisfied, then the analyst gathers
less information than he would if his incentives were not misaligned with those of
the decision-maker–i.e., the analyst under-invests in the quality of information he
gathers.
Proof of Corollary 4:
First consider ∂¯p/∂q.Whenq∈(1/2,1) and ¯pgiven in (14), observe that
∂¯p
∂q =(s−f)q4q2(1 −q)2(s−f)2+φ2(2q−1)2−2q(s−f)2(1 −q)−φ2(2q−1)2
q4q2(1 −q)2(s−f)2+φ2(2q−1)2φ(2q−1)3
∝(s−f)q4q2(1 −q)2(s−f)2+φ2(2q−1)2−2q(s−f)2(1 −q)−φ2(2q−1)2
>0.
To establish the last inequality note that
(s−f)q4q2(1 −q)2(s−f)2+φ2(2q−1)2−2q(s−f)2(1 −q)+φ2(2q−1)2>0,
or equivalently,
(s−f)2¡4q2(1 −q)2(s−f)2+φ2(2q−1)2¢−¡2q(s−f)2(1 −q)+φ2(2q−1)2¢2
=φ2(2q−1)4¡(s−f)2−φ2¢
>0.
To establish that φ2(2q−1)4((s−f)2−φ2)>0, observe that
(2p−1) q(1 −q)
(p(1 −q)+(1−p)q) ((1 −p)(1−q)+pq)<1
for all p. Therefore, for the truth-telling constraint to be satisfied,itmustbethecase
that φ/ (s−f)<1; otherwise the truth-telling condition can never be satisfied.
Second ∂p∗/∂q. It follows from Lemma 2 that p∗is such that
∂Ui/∂p|p∗=q2(1 −q)2(2p∗−1) (s−f)2
((1 −p∗)q+(1−q)p∗)2(qp∗+(1−q)(1−p∗))2−c0(p∗)=0.
32
Therefore, the implicit function theorem yields
∂p∗
∂q =2q(1 −q)(2p−1) (s−f)2p(1 −p)(2q−1)
((1 −p)q+(1−q)p)3(qp +(1−q)(1−p))3/∂(∂Ui(.)/∂p|p∗)
∂p∗.
Since Ui(p)is a strictly quasi-concave and twice differentiable function, it follows that
the denominator is strictly negative. Hence,
∂p∗
∂q ∝(−2q(1−q)(2p−1)(s−f)2p(1−p)(2q−1)
((1−p)q+(1−q)p)3(qp+(1−q)(1−p))3<0when q∈(1/2,1)
0when q=1/2.
Proof of Corollary 5:
The arguments showing ∂¯p/∂φ > 0,∂¯p/∂ (s−f)<0,∂¯p/∂φ =0are straightfor-
ward. Consider ∂p∗/∂ (s−f). Applying the implicit function theorem to Lemma 2
yields
∂p∗
∂(s−f)=−2q2(1 −q)2(2p−1) (s−f)
((1 −p)q+(1−q)p)2(qp +(1−q)(1−p))2/∂(∂Ui(.)/∂p|p∗)
∂p∗.
Since Uiis a strictly quasi-concave and twice differentiable function, it follows that
the denominator is strictly negative. Hence,
∂p∗
∂(s−f)∝2q2(1 −q)2(2p−1) (s−f)
((1 −p)q+(1−q)p)2(qp +(1−q)(1−p))2>0.
Proof of Corollary 6:
First, we establish that ∂(−Ey[var [˜α|x, y;p∗]]) /∂q > 0. Observe that
∂(−Ey[var [˜α|x, y;p∗]])
∂q =∂(−Ey[var [˜α|x, y;p∗]])
∂q +∂(−Ey[var [˜α|x, y;p∗]])
∂p∗
∂p∗
∂q
∝(p∗)2(1 −p∗)2(2q−1) + q2(1 −q)2(2p−1) ∂p∗
∂q .(16)
33
Consider ∂p∗/∂q. Applying the implicit function theorem to (4) yields
∂p∗
∂q =−
∂(∂Ui/∂p|p∗)
∂q
∂(∂Ui/∂p|p∗)
∂p∗
=−
−2q(1−q)(2p∗−1)p∗(1−p∗)(2q−1)(s−f)2
((1−p∗)q+(1−q)p)3(qp∗+(1−q)(1−p∗))3
2q2(1−q)2(s−f)2(3p∗(2q−1)2(p−1)+1−3q+3q2)
((1−p∗)q+(1−q)p∗)3(qp∗+(1−q)(1−p∗))3−c00 (p∗)
.(17)
Observe that
∂(∂Ui/∂p|p∗)
∂p∗=2q2(q−1)2(s−f)2¡3p∗(2q−1)2(p∗−1) + 1 −3q+3q2¢
((1 −p∗)q+(1−q)p∗)3(qp∗+(1−q)(1−p∗))3−c00 (p∗)
<0
because Uiis a strictly quasi-concave and twice differentiable function.
Substituting (17) into (16), factoring, and using the fact that ∂(∂Ui/∂p|p∗)/∂p∗<
0yields
∂(−Ey[var [˜α|x, y;p∗]])
∂q
∝p∗(1 −p∗)c00 (p∗)−2q2(1 −q)2(1 −3p∗(1 −p∗)) (s−f)2
((1 −p∗)q+(1−q)p∗)2(qp∗+(1−q)(1−p∗))2.
On substituting c0(p∗)=∆2(p∗)/(2p∗−1), given in (4), into the assumption that
c00(p)/c0(p)>(1 −3p(1 −p)) /(p(1 −p)(p−1/2)), yields
c00(p∗)>2(s−f)2(1 −q)2q2¡1−3p∗+3(p∗)2¢
(1 −p∗)p∗((1 −p∗)(1−q)+p∗q)2(p∗(1 −q)+(1−p∗)q)2.(18)
Using the relation in (18), it follows that
∂(−Ey[var [˜α|x, y;p∗]])
∂q
∝p∗(1 −p∗)c00 (p∗)−2q2(1 −q)2(1 −3p∗(1 −p∗)) (s−f)2
((1 −p∗)q+(1−q)p∗)2(qp∗+(1−q)(1−p∗))2
>p
∗(1 −p∗)Ã2(s−f)2(1 −q)2q2¡1−3p∗+3(p∗)2¢
(1 −p∗)p∗((1 −p∗)(1−q)+p∗q)2(p∗(1 −q)+(1−p∗)q)2!
−2q2(1 −q)2(1 −3p∗(1 −p∗)) (s−f)2
((1 −p∗)q+(1−q)p∗)2(qp∗+(1−q)(1−p∗))2
=0.
34
Second, we show the decision-maker’s quality of information falls discontinuously
at ¯q. Observe that when c(¯p)=∆(¯p)q(1 −q)(2¯p−1) (s−f)and q=¯q,then
Ui(1/2) = Ui(¯p),orequivalently,
φE[˜α|x]−q(1 −q)(s−f)2
=φEy[E[˜α|x, y]] −¯p(1 −¯p)q(1 −q)(s−f)2
(¯p(1 −q)+(1−¯p)q)((1−¯p)(1−q)+¯pq)−c(¯p).(19)
The law of iterated expectations, E[˜α|x]=Ey[E[˜α|x, y]] implies (19) may be ex-
pressed as
−Var[˜α|x]≡−q(1 −q)(s−f)2
=−¯p(1 −¯p)q(1 −q)(s−f)2
(¯p(1 −q)+(1−¯p)q) ((1 −¯p)(1−q)+ ¯pq)−c(¯p)
<−¯p(1 −¯p)q(1 −q)(s−f)2
(¯p(1 −q)+(1−¯p)q) ((1 −¯p)(1−q)+ ¯pq)≡−Ey[var [˜α|x, y;¯p]] .
Hence, we observe there exists a q=q+ε>q
∗for ε>0such that
−Var[˜α|x]<−Ey[var [˜α|x, y;¯p]] .
ProofofProposition7:
Because ∆(p=1/2) = 0,it follows from equation (8) that p=1/2is an equi-
librium level of precision for t=qor t=1−q. Consider the case in which t=q.
Rewrite equation (??)asfollows
(1/2)∆(p)2−c0(p)(p−1/2) + ∆(p)φ(2t−1)(p−1/2) = 0.
The first derivative with respect to pis
2∆0(p)
∆(p)¡(1/2)∆(p)2+∆(p)φ(2t−1)(p−1/2)¢
−1
(p−1/2) (c0(p)(p−1/2) −∆(p)φ(2t−1)(p−1/2))
−c00(p)(p−1/2) −
∆0(p)
∆(p)∆(p)φ(2t−1)(p−1/2).
35
Iftheequationissatisfied at p, this expression can be rewritten as
2∆0(p)
∆(p)c0(p)(p−1/2) −c00(p)(p−1/2) −(1/2)∆(p)2
(p−1/2)
−
∆0(p)
∆(p)∆(p)φ(2t−1)(p−1/2),
where ∆0(p)
∆(p)=1
2(p−1/2)
1−2p(1 −p)(2t−1)2−2t(1 −t)
p(1 −p)(2t−1)2+t(1 −t).
Hence, the above expression can be written as
1−2p(1 −p)(2t−1)2−2t(1 −t)
p(1 −p)(2t−1)2+t(1 −t)c0(p)−c00(p)(p−1/2) −(1/2)∆(p)2
(p−1/2)
−(1/2)1−2p(1 −p)(2t−1)2−2t(1 −t)
p(1 −p)(2t−1)2+t(1 −t)∆(p)φ(2t−1).
Exploiting the fact that (8) holds in equilibrium allows this expression to be rewritten
as
µ1
p(1 −p)(2t−1)2+t(1 −t)−3¶c0(p)−c00(p)(p−1/2)
−µ1
2p(1 −p)(2t−1)2+2t(1 −t)−2¶∆(p)φ(2t−1).
Noting that p(1 −p)(2t−1)2+t(1 −t)is decreasing in tand that 2t−1>0because
t=q>1/2implies that the above expression is greater than
µ1−3p(1 −p)
p(1 −p)¶c0(p)−c00(p)(p−1/2),
which is proportional to
(1 −3p(1 −p))c0(p)−p(1 −p)(p−1/2)c00(p).
Hence, there exists only an single equilibrium in which p∈(1/2,1) if
(1 −3p(1 −p)) /(p(1 −p)(p−1/2)) ≤c00(p)/c0(p)
for all p, which is true by assumption.
36
Finally, consider the case in which t=1−q. In this case, there may or may not
exist an equilibrium in which p∈(1/2,1) and, sometimes there may exist more than
one equilibrium in which p∈(1/2,1). The proof follows from an example in which
there is no equilibrium when p∈(1/2,1),one equilibrium when p∈(1/2,1),and
two equilibria when p∈(1/2,1).Fortheexamples,letc(p)=((p−1/2) /(1 −p))3
and q=.85.Ifφ=2and s−f=2, then there does not exist a p∈(1/2,1) that
satisfies the equilibrium condition. If φ=1and s−f=2,thenp=.5227 is the only
p∈(1/2,1) that satisfies the equilibrium condition. If φ=110,s−f= 150,then
p=.9072 and p=.5584 both satisfy the equilibrium condition.
37
References
[1] Alles, M., and R. Lundholm. “On the Optimality of Public Signals in the Pres-
ence of Private Information.” The Accounting Review, Vol. 68 (1993): 93-112.
[2] Arya, A., and B.G. Mittendorf. “The Interaction among Disclosure, Competition
between Firms, and Analyst Following.” Journal of Accounting and Economics,
Vol. 43, No 2-3 (July 2007): 321-339.
[3] Aumann, R. J., and S. Hart. “Long Cheap Talk.” Econometrica, Vol. 71, No. 6
(November 2003): 1619-60.
[4] Austen-Smith, D. “Interested Experts and Policy Advice: Multiple Referrals
under Open Rule.” Games and Economic Behavior, Vol. 5 (1993): 3-43.
[5] Austen-Smith, D. “Strategic Transmission of Costly Information.” Econometrica,
Vol. 62, No. 4 (July 1994): 955-963.
[6] Battaglini, M. “Policy Advice with Imperfectly Informed experts.” Advances in
Theoretical Economics, Vol. 4, No. 1 (2004): article 1.
[7] Bushee, B.J., D. Matsumoto. and G. Miller. “Open versus Closed Conference
Calls: The Determinants and Effects of Broadening Access to Disclosure.” Jour-
nal of Accounting and Economics, Vol. 34 (2003): 149-180.
[8] Christensen, J. “The Determination of Performance Standards and Participa-
tion.” Journal of Accounting Research, Vol. 20, No. 2 (Autumn, 1982): 589-603.
[9] Crawford, V. P., and J. Sobel. “Strategic Information Transmission.” Economet-
rica, Vol. 50 (1982): 1431-51.
[10] DeGroot, M. H. Optimal Statistical Decisions.McGraw-Hill,Inc.(1970).
[11] Diamond, D.W. “Optimal Release of Information by Firms.” Journal of Finance,
Vol. 40 (1985): 1071-94.
[12] Dugar, A. and Nathan, S. “The Effect of Investment Banking Relationships
on Financial Analyst’s Earnings Forecasts and Investment Recommendations.”
Contemporary Accounting Research, Vol. 12 (1995): 131-160.
38
[13] Dye, R. A. “Disclosure of nonproprietary information”, Journal of Accounting
Research, Vol. 23 (1985): 123-145.
[14] Einhorn, E. “Voluntary disclosure under uncertainty about the reporting ob-
jective.” Journal of Accounting & Economics, Vol. 43, No. 273 (June 2007):
245-274.
[15] Einhorn. E. and A. Ziv. “Intertemporal Dynamics of Corporate Voluntary Dis-
closures.” Journal of Accounting Research, Vol. 46, No. 3. (June 2008): 567-589.
[16] Farrell, J., and R. Gibbons. “Cheap talk with two audiences.” American Eco-
nomic Review, Vol. 79 (1989): 1214-23.
[17] Gigler, F. “Self-enforcing voluntary disclosures.” Journal of Accounting Research,
Vol. 32 (1994): 224-40.
[18] Grossman, G. M., and E. Helpman. Special Interest Politics. The MIT Press
(2001).
[19] Hayes, R.M. “The Impact of Trading Commission Incentives on Analysts’ Stock
Coverage Decisions and Earnings Forecasts.” Journal of Accounting Research,
Vol. 36, No. 2 (Autumn, 1998): 299-320.
[20] Heflin, F., K.R. Subramanyam, and Y. Zhang. “Regulation FD and the Financial
Information Environment: Early Evidence.” TheAccountingReview,Vol.78,No.
1 (2003): 1-37.
[21] Hughes, J.S. and S. Pae. “Voluntary Disclosure of Precision Information.” Jour-
nal of Accounting & Economics, Vol. 37, No. 2 (June 2004): 261-289.
[22] Krishna, V., and J. Morgan. “A model of expertise.” Quarterly Journal of Eco-
nomics, Vol. 116 (2001): 747-75.
[23] Krishna, V., and J. Morgan. “The Art of Conversation: Eliciting information
from Experts through Multi-Stage Communication.” Journal of Economic The-
ory, Vol. 117 ( 2004): 147-79.
[24] Lin, H.-W. and McNichols, M.F. “Underwriting Relationships, Analysts’ Earn-
ings Forecasts and Investment Recommendations.” Journal of Accounting and
Economics, Vol. 25 (1998): 101-127.
39
[25] Mayew, William J. “Evidence of Management Discrimination Among Analysts
During Earnings Conference Calls.” Journal of Accounting Research, Vol. 46, No.
3(2008): 627-659.
[26] McNichols, M. and P.C. O’Brien. “Self-Selection and Analyst Coverage.” Journal
of Accounting Research, Vol. 35, Supplement (1997): 167-199.
[27] Michaely, R. and Womack, K. “ConflictofInterestandtheCredibilityofUn-
derwriter Analyst Recommendations.” The Review of Financial Studies, Vol . 12
(1999): 653-686.
[28] Morgan, J., and P. C. Stocken. “An Analysis of Stock Recommendations.” RAND
Journal of Economics, Vol. 34, No. 1 (2003): 183-203.
[29] Morgan, J., and P.C. Stocken. “Information Aggregation in Polls.” American
Economic Review, Vol 98, No. 3 (June 2008): Forthcoming.
[30] Newman, P., and R. Sansing. “Disclosure policies with multiple users.” Journal
of Accounting Research, Vol. 31 (1993): 92-112.
[31] Pae, S. “Acquisition and Discretionary Disclosure of Private Information and Its
Implications for Firms’ Productive Activities.” Journal of Accounting Research,
Vol. 37, No. 2 (Autumn 1999): 465-474.
[32] Penno, M. “Unobservable Precision Choices in Financial Reporting.” Journal of
Accounting Research, Vol. 34, No. 1 (Spring, 1996): 141-149.
[33] Razaee, Z. Corporate Governance Post—Sarbanes-Oxley: Regulations, Require-
ments, and Integrated Processes, John Wiley & Sons: Hoboken, NJ. (2007).
[34] Sansing, R. C. “Accounting and the credibility of management forecasts.” Con-
temporary Accounting Research, Vol. 9, No. 1 (1992): 33-45.
[35] Schack, J. “Unsettled on Wall Street.” Institutional Investor, (October 2003): 1.
[36] Stocken, P. C. “Credibility of voluntary disclosure.” RAND Journal of Eco-
nomics, Vol. 31, No. 2 (2000): 359-74.
[37] Verrecchia, R. E. “Mathematical models in financial accounting.” Journal of
Accounting Research, Vol. 20 (1982): 1-42.
40
[38] Verrecchia, R. E. “Discretionary disclosure.” Journal of Accounting and Eco-
nomics, Vol. 5 (1983): 179-194.
[39] Verrecchia, R. E. “Information Quality and Discretionary Disclosure.” Journal
of Accounting and Economics, Vol. 12 (1990): 365-380.
[40] Wolinsky, A. “Eliciting information from multiple experts.” Games and Eco-
nomic Behavior, Vol. 41 (2002): 141-60.
41
42
Figure 1: Time line of events
The unknown state α assumes the
values s or f with equal probability.
In stage one, the analyst and
decision-maker observe the public
information x,whichiseither
h
or
l
,withprobability
Prh|s
Prl|f
q
.
Information gathering stage
In stage two, the analyst chooses the
p
recision p of his private information
y
,whichiseither
g
or
b
,with
probability
Prg|sPrb|fp
.
The cost function c(p) reflects the
analyst’s cost of gathering
information with precision p.
Reporting stage
In stage three, the analyst observes
his private information y and sends a
report r, which need not be truthful,
to the decision-maker. The decision-
maker then chooses an action a. The
p
arameter φ denotes extent to which
the players’ interests are misaligned.
In stage four, the state
of nature α is realized,
and the analyst’s payoff
Ui and the decision-
maker’s payoff Ud are
determined.