# Electronic copy available at: http://ssrn.

**com/abstract=1787732
**

Feedback E¤ects and the Limits to Arbitrage

+

Alex Edmans

Wharton, NBER, and ECGI

Itay Goldstein

Wharton

Wei Jiang

Columbia

October 27, 2011

Abstract

This paper identi…es a limit to arbitrage that arises from the fact that a …rm’s fundamental

value is endogenous to the act of exploiting the arbitrage. Trading on private information

reveals this information to managers and helps them improve their real decisions, in turn

enhancing fundamental value. While this increases the pro…tability of a long position, it

reduces the pro…tability of a short position – selling on negative information reveals that

…rm prospects are poor, causing the manager to cancel investment. Optimal abandonment

increases …rm value and may cause the speculator to realize a loss on her initial sale.

Thus, investors may strategically refrain from trading on negative information, and so bad

news is incorporated more slowly into prices than good news. The e¤ect has potentially

important real consequences – if negative information is not incorporated into stock prices,

negative-NPV projects may not be abandoned, leading to overinvestment.

Keywords: Limits to arbitrage, feedback e¤ect, overinvestment

JEL Classification: G14, G34

**aedmans@wharton.upenn.edu, itayg@wharton.upenn.edu, wj2006@columbia.edu. For helpful
**

comments, we thank Philip Bond, Mike Fishman, Kathleen Hanley, Dirk Jenter, Pete Kyle, Sam

Taylor, James Thompson, Dimitri Vayanos, Kostas Zachariadis, and seminar participants at the Fed-

eral Reserve Board, Wharton, the LSE Paul Woolley Centre Conference, and the Theory Conference

on Corporate Finance and Financial Markets. We thank Ali Aram, Guojun Chen, Chong Huang and

Edmund Lee for excellent research assistance. AE gratefully acknowledges …nancial support from

the Dorinda and Mark Winkelman Distinguished Scholar award and the Goldman Sachs Research

Fellowship from the Rodney L. White Center for Financial Research.

1

Electronic copy available at: http://ssrn.com/abstract=1787732

1 Introduction

Whether …nancial markets are informationally e¢cient is one of the most hotly-contested de-

bates in …nance. Proponents of market e¢ciency argue that pro…t opportunities in the …nancial

market will lead speculators to trade in a way that eliminates any mispricing. For example, if

speculators have negative information about a stock, and this information is not re‡ected in

the price, they will …nd it pro…table to sell the stock. This will push down the price, caus-

ing it to re‡ect speculators’ information. However, a sizable literature identi…es various limits

to arbitrage, which may deter speculators from trading on their information. (This notion of

“arbitrage” is broader than the traditional textbook notion of risk-free arbitrage from trading

two identical securities. Here, we use “arbitrage” to refer to investors trading on their private

information.) For example, De Long, Shleifer, Summers, and Waldmann (1990) and Shleifer

and Vishny (1997) show that the slow convergence of price to fundamental value may render

arbitrage activities too risky. This in turn dissuades trading if the speculator has a short hori-

zon, which may in turn arise from informational asymmetries with her own investors. Other

explanations for limited arbitrage rely on market frictions such as short-sales constraints. All

of these mechanisms treat the …rm’s fundamental value as exogenous to the arbitrage process

and rely on market imperfections to explain why speculators will not drive the price towards

fundamental value. Thus, as …nancial markets develop, these limits to arbitrage may weaken.

In this paper, we identify a quite di¤erent limit to arbitrage, which does not rely on exoge-

nous forces but is instead generated endogenously as part of the arbitrage process. It stems from

the fact that the value of the asset being arbitraged is endogenous to the act of exploiting the ar-

bitrage. By trading, speculators cause prices to move, which in turn reveals information to real

decision makers, such as managers, board members, corporate raiders, and regulators. These

decision makers then take actions based on the information revealed in the price, and these

actions change the underlying asset value. This may make the initial trading less pro…table,

deterring it from occurring in the …rst place.

To …x ideas, consider the following example. Suppose that a …rm (acquirer) announces the

acquisition of a target. Also assume that some speculators conducted some analysis suggesting

that this acquisition will be value-destructive. Traditional theory suggests that these speculators

should sell the acquirer’s stock. However, large-scale selling will convey to the acquirer that

speculators believe the acquisition is a bad idea. As a result, the acquirer may end up cancelling

the acquisition. In turn, cancellation of a bad acquisition will boost …rm value, reducing the

speculator’s pro…t from her short position and in some cases causing her to su¤er a loss. Put

di¤erently, the acquirer’s decision to cancel the acquisition means that the negative information

possessed by speculators is now less relevant, and hence they should not trade on it. Thus, the

information ends up not being re‡ected in the price.

Our mechanism is based on the presence of a feedback e¤ect from the …nancial market

to real economic decisions – that real decision makers learn from the market when deciding

their actions. A common perception is that managers know more about their own …rms than

2

outsiders (e.g. Myers and Majluf (1984)). While this is likely plausible for internal information

about the …rm in isolation, optimal managerial decisions also depend on external information

(such as market demand for a …rm’s products, or potential synergies with a target) about which

outsiders may be more informed. A classic example of how information from the stock market

can shape real decisions is Coca-Cola’s attempted acquisition of Quaker Oats. On November

20, 2000, the Wall Street Journal reported that Coca-Cola was in talks to acquire Quaker

Oats. Shortly thereafter, Coca-Cola con…rmed such discussions. The market reacted negatively,

sending Coca-Cola’s shares down 8% on November 20th and 2% on November 21st. Coca-Cola’s

board rejected the acquisition later on November 21st, potentially due to the negative market

reaction. The following day, Coca-Cola’s shares rebounded 8%. Thus, speculators who had

short-sold on the initial merger announcement, based on the belief that the acquisition would

destroy value, lost money – precisely the e¤ect modeled by this paper. In the same context,

Luo (2005) provides large-sample evidence that acquisitions are more likely to be cancelled if

the market reacts negatively to them, and that the e¤ect is more pronounced when the acquirer

is more likely to have something to learn from the market, e.g., for non-high-tech deals and

where the bidder is small. Relatedly, Edmans, Goldstein, and Jiang (2011) demonstrate that

a …rm’s market price a¤ects the likelihood that it becomes a takeover target, which may arise

because potential acquirers learn from the market price. More broadly, Chen, Goldstein, and

Jiang (2007) show that the sensitivity of investment to price is higher when the price contains

more private information not known to managers.

Moreover, our model can apply to corrective actions (i.e., actions that improve …rm value

upon learning negative information about …rm prospects) undertaken by stakeholders other

than the manager. Such stakeholders likely have less information than the manager and may

be more reliant on information held by outsiders. Examples include managerial replacement

(undertaken by the board, or by shareholders who lobby the board), a disciplinary takeover

(undertaken by an acquirer), or the granting of a subsidy or a bail-out (undertaken by the

government). We demonstrate a barrier to the feedback e¤ect, that hinders decision makers

from learning from the market.

An important aspect of our theory is that it generates asymmetry between trading on

positive and negative information. The feedback e¤ect delivers an equilibrium where speculators

trade on good news but do not trade on bad news. Yet, it does not give rise to the opposite

equilibrium, where speculators trade on bad news only. The intuition is as follows. When

speculators trade on information, they improve the e¢ciency of the …rm’s decisions – regardless

of the direction of their trade. If the speculator has positive information on a …rm’s prospects,

trading on it will reveal to the manager that investment is pro…table. This will in turn cause

the …rm to invest more, thus increasing its value. If the speculator has negative information,

trading on it will reveal to the manager that investment is unpro…table. This will in turn cause

the …rm to invest less, also increasing its value as contraction is the correct decision. When a

speculator buys and takes a long position in a …rm, she bene…ts further from increasing its value

3

via the feedback e¤ect. By contrast, when she sells and takes a short position, she loses from

increasing the …rm’s value via the feedback e¤ect. Note that, for the speculator to lose from

the feedback e¤ect, she must end up with a short position. If she ends with a long position, the

value of the shares she still holds onto are enhanced by the feedback e¤ect. Thus, the model

implies that investors are less likely to engage in short-sales than sales – even though the model

contains no short-sale constraints.

Even though the speculator’s trading behavior is asymmetric, in general it is not automatic

that the impact on prices is asymmetric. The market maker is fully rational and takes into

account the fact that the speculator buys on positive information and does not trade on negative

information. Thus, he adjusts his pricing function accordingly. Therefore, it may seem that

negative information will be impounded in prices to the same degree as positive information

– even though it may lead to a neutral rather than negative order ‡ow, the market maker

knows that a neutral order ‡ow can stem from the speculator having negative information but

choosing not to trade, and may decrease the price accordingly. By contrast, we show that

the asymmetry in trading behavior does translate into asymmetry in price impact. The crux is

that the market maker cannot distinguish the case of a speculator who has negative information

but chooses to withhold it, from the case in which the speculator is absent (i.e. there is no

information). Thus, a neutral order ‡ow does not lead to a large stock price decrease, and so

negative information has a smaller e¤ect on prices. Indeed, Hong, Lim, and Stein (2000) show

empirically that bad news is incorporated in prices more slowly than good news. They speculate

that this arises because it is …rm management that possesses value-relevant information, and

they will publicize it more enthusiastically for favorable than unfavorable information. Our

paper presents a formal model that o¤ers an alternative explanation. Here, key information is

held by a …rm’s investors rather than its managers, who “publicize” it not through public news

releases, but by trading on it. They also choose to disseminate good news more readily than

bad news, but for a very di¤erent reason from …rm management, i.e., because of the feedback

e¤ect.

In standard models of underreaction, if bad news has a smaller e¤ect on short-run returns

(i.e. between t = 0 and t = 1) than good news, this must be counterbalanced by bad news

generating a larger long-run drift (between t = 1 and t = 2) than good news. We show that

this need not be the case in a model with feedback. It is indeed true that, if the state is bad

and little bad news comes out in the short-run (due to the speculator not trading on it), there

is more bad news still to come out in the long-run. However, in a feedback model where …rm

value is endogenous, the manager can take a corrective action to mitigate the negative impact

of the state on …rm value. If the feedback e¤ect is su¢ciently strong, bad news has a smaller

e¤ect than good news in both the short-run and the long-run.

While the above considers the returns to good and bad news, the model also generates

predictions regarding the returns to good and bad investment decisions. Naturally, the returns

to investment are positive (both in the short-run and long-run) if the state is good and negative

4

if the state is bad. More interestingly, we show that the returns to good investment are more

front-loaded than the returns to a bad investment – i.e., a higher proportion of the returns

manifests at t = 1 than at t = 2. This result again stems from the asymmetry of the speculator’s

trading strategy. Even if the speculator is aware that the investment is bad at t = 1, she may

not trade on this information due to the feedback e¤ect. Thus, the value-destructiveness of

the investment seeps out ex post at t = 2. Thus, our model provides an explanation for the

negative long-run returns to M&A, documented by Agrawal, Ja¤e, and Mandelker (1992) and

Rau and Vermaelen (1998).

In addition to its interesting e¤ects on stock returns, the asymmetry of the speculator’s

trading strategy can also generate important real consequences. Since negative information is

not incorporated into prices, it does not in‡uence management decisions. Thus, while positive-

NPV projects will be encouraged, some negative-NPV projects will not be canceled – even

though there is an agent in the economy who knows with certainty that the project is negative-

NPV – leading to overinvestment overall. In contrast to standard overinvestment theories based

on the manager’s private bene…ts (e.g., Jensen (1986), Stulz (1990), Zwiebel (1996)), here the

manager is fully aligned with …rm value and there are no agency problems. The manager wishes

to maximize …rm value by learning from prices, but is unable to do so since speculators refrain

from revealing their information. Applied to M&A as well as organic investment, the theory

may explain why M&A appears to be “excessive” and a large fraction of acquisitions destroy

value (see, e.g., Andrade, Mitchell, and Sta¤ord (2001).)

As mentioned above, the primary motivation for our paper is to identify a limit to arbitrage.

Di¤erent authors have emphasized di¤erent factors that lead to limits on arbitrage activities.

Campbell and Kyle (1993) focus on fundamental risk, i.e., the risk that …rm fundamentals will

change while the arbitrage strategy is being pursued. In their model, such changes are unrelated

to speculators’ arbitrage activities. De Long, Shleifer, Summers, and Waldmann (1990) argue

that noise-trading risk, i.e., the risk that noise trading will increase the degree of mispricing,

may render arbitrage activities unpro…table. Noise trading only a¤ects the asset’s market price

and not its fundamental value, which is again exogenous to the act of arbitrage. Shleifer and

Vishny (1997) show that, even if an arbitrage strategy is sure to converge in the long-run, the

possibility that mispricing may widen in the short-term may deter speculators from trading on

it, if they are concerned with redemptions by their own investors. Similarly, Kondor (2009)

demonstrates that …nancially-constrained arbitrageurs may stay out of a trade if they believe

that it may become more pro…table in the future. Many authors (e.g., Ponti¤ (1996), Mitchell

and Pulvino (2001), and Mitchell, Pulvino, and Sta¤ord (2002)) focus on the transaction costs

and holding costs that arbitrageurs have to incur while pursuing an arbitrage strategy. Others

(Geczy, Musto, and Reed (2002), and Lamont and Thaler (2003)) discuss the importance of

short-sales constraints. While these papers emphasize market frictions as the source of limits to

arbitrage, our paper shows that limits to arbitrage arise when the market performs its utmost

e¢cient role: guiding the allocation of real resources. Thus, while limits to arbitrage based

5

on market frictions tend to attenuate with the development of …nancial markets, the e¤ect

identi…ed by this paper may strengthen – as investors become more sophisticated, managers

will learn from them to a greater degree. Our model deliberately shuts down the sources of the

limits to arbitrage identi…ed by prior theories: the speculator is risk-neutral, there are no agency

problems due to portfolio delegation, and there is no exogenous friction on trading (other than

a standard transaction cost), such as short-sale constraints. With all these forces switched o¤, a

limit to arbitrage nevertheless arises because the speculator endogenously chooses not to trade

due to the feedback e¤ect.

Our paper is related to the literature exploring the theoretical implications of the feedback

e¤ects from market prices to real decision making. Several papers in this literature have shown

that the feedback e¤ect can be harmful for real e¢ciency. Most closely related is Goldstein and

Guembel (2008), who show that it provides an incentive for uninformed speculators to short sell

a stock, reducing its value by inducing a real decision (investment) based on false information.

Their paper also highlights an asymmetry between buy-side and sell-side speculation, but only

with respect to uninformed trading; here, we show that informed speculators are less likely

to trade on bad news rather than good news, in turn generating implications for the speed of

incorporation of news into prices.

1

Bond, Goldstein, and Prescott (2010), Dow, Goldstein, and

Guembel (2010), and Goldstein, Ozdenoren, and Yuan (2011) also model complexities arising

from the feedback e¤ect. Overall, the point in our paper – that negatively informed speculators

will strategically withhold information from the market, because they know that the release of

negative information will lead managers to …x the underlying problem – is new in this literature.

This paper proceeds as follows. Section 2 presents the model. Section 3 contains the core

analysis, demonstrating the asymmetric limit to arbitrage. Section 4 investigates the extent to

which information a¤ects beliefs and prices, Section 5 discusses potential applications of the

model, and Section 6 concludes. Appendix A contains all proofs not in the main text.

2 The Model

The model has three dates, t ¸ ¦0. 1. 2¦. There is a …rm whose stock is traded in the …nancial

market. The …rm’s manager needs to take a decision as to whether to continue or abandon an

investment project. The manager’s goal is to maximize expected …rm value; since there are no

agency problems between the manager and the …rm, we will use these two terms interchangeably.

At t = 0, a risk-neutral speculator may be present in the …nancial market. If present, she

is informed about the state of nature o that determines the pro…tability of continuing vs.

abandoning the project. Trading in the …nancial market occurs at t = 1. In addition to the

speculator, two other types of agents participate in the …nancial market: noise traders whose

trades are unrelated to the realization of o, and a risk-neutral market maker. The latter collects

1

In addition, while they require two trading periods to generate the asymmetry, our model is simpler and

requires only a single period.

6

the orders from the speculator and noise traders, and sets a price at which he executes the orders

out of his inventory. At t = 2, the manager takes the decision, which may be a¤ected by the

trading in the …nancial market at t = 1. Finally, all uncertainty is resolved and payo¤s are

realized. We now describe the …rm’s investment problem and the trading process in more detail.

2.1 The Firm’s Decision

Suppose that the …rm has an investment project that can be either continued or abandoned

at t = 2. We denote the …rm’s decision as d ¸ ¦i. :¦, where d = i represents continuing

the investment and d = : represents no investment (also referred to as “abandonment” or

“correction”). The …rm faces uncertainty over the realization of value under each possible

action. In particular, there are two possible states o ¸ = ¦H. 1¦ (“high” and “low”). We

denote the value of the …rm realized in t = 2 as · = 1

d

**, which depends on both the state of
**

nature o and the manager’s action d.

We assume that whether continuation or abandonment is desirable depends on the state of

nature (i.e., there is no dominant action). Without loss of generality, we set:

1

i

H

1

n

H

(1)

1

n

L

1

i

L

. (2)

that is, continuation is optimal in state H, while abandonment is optimal in state 1. We also

set:

1

i

H

1

n

L

, (3)

that is, under the optimal action, the highest …rm value is achieved in state H, consistent with

this being labeled as the “high” state. This assumption is also without loss of generality as,

if it is not satis…ed, the highest …rm value is achieved in state 1 and we can simply reverse

notations.

Note that equations (1) and (2) imply:

1

i

H

÷1

i

L

1

n

H

÷1

n

L

. (4)

Equation (4) is the driving force behind our results. It means that taking the corrective action

reduces the negative e¤ect of state 1 on …rm value. Put di¤erently, if the state is 1 rather than

H, the reduction in …rm value is lower if the manager has taken action :. In turn, equation (4)

incorporates two cases, depending on whether …rm value is monotonic in the underlying state:

Case 1: 1

n

H

1

n

L

. In this case, state H is better for …rm value, no matter what action

has been taken by the …rm. Hence, the corrective action attenuates, but does not eliminate,

the e¤ect of the state on …rm value. Abandonment reduces the volatility of …rm value, i.e., the

dependence of …rm value on the state. For example, state H can represent high demand for the

…rm’s products, while state 1 represents low demand. Whether the …rm continues to invest in

7

its production process or not, its value will be lower in state 1, but the negative e¤ect of state

1 is attenuated if the …rm does not invest.

Case 2: 1

n

L

1

n

H

. In this case, if the corrective action is taken, …rm value is higher in state

1. Put di¤erently, the corrective action is su¢ciently powerful to overturn the e¤ect of the

state on …rm value. Importantly, this second case does not require that abandonment reduces

the volatility of …rm value: it could be that c/: (1

n

H

÷1

n

L

) c/: (1

i

H

÷1

i

L

) so volatility is

higher under correction. Instead, the case 1

n

L

1

n

H

implies non-monotonicity of …rm value

in the state: one state does not dominate the other. For example, consider the case where

continuation implies proceeding with a takeover decision, and abandonment implies keeping

the cash for future opportunities. State H corresponds to a state in which current acquisition

opportunities dominate future ones, and state 1 refers to the reverse. Under continuation, …rm

value is higher in state H, whereas if the …rm chooses to postpone acquisitions, its value is higher

in state 1 where future acquisition opportunities are superior. Another example is related to

Aghion and Stein (2008): d = i corresponds to a growth strategy, and d = : corresponds to

a strategy focused on current pro…t margins. Growth prospects are good if o = H and bad if

o = 1. If the …rm eschews the growth strategy (d = :), its value is higher in the low state

where there are no growth opportunities, since in the high state, its rivals could pursue the

growth opportunities, in turn worsening its competitive position.

The prior probability that the state is o = H is ¸ =

1

2

, which is common knowledge. We use

¡ to denote the posterior probability the manager assigns to the case o = H. The manager’s

decision is conditioned on ¡, which in turn is calculated using information arising from trades

in the …nancial market. Let ¸ denote the posterior belief that the state is H such that the

manager is indi¤erent between continuation and abandonment, i.e.:

¸1

i

H

+ (1 ÷¸)1

i

L

= ¸1

n

H

+ (1 ÷¸)1

n

L

. (5)

The value of ¸ represents a “cuto¤” that determines the manager’s action. If and only if ¡ ¸,

he will continue the project. We will distinguish between two cases. The …rst case is where

¸ <

1

2

. Since the prior ¸ is

1

2

, the manager would continue the investment without further

information, i.e., ex ante, the investment has a positive net present value. The second case is

where ¸

1

2

, and so the ex-ante net present value of the investment is negative.

2.2 Trade in the Financial Market

In t = 0, with probability 0 < ` < 1, a speculator arrives in the …nancial market. Whether the

speculator is present or not is unknown to anyone else.

2

If the speculator is present, she observes

the state of nature o with certainty. We will use the term “positively-informed speculator” to

2

Since private information is not public knowledge, its existence is also unlikely to be public knowledge.

Chakraborty and Yilmaz (2004) also feature uncertainty on whether the speculator is present, in an equilibrium

in which informed insiders manupulate the market by trading in the wrong direction.

8

describe a speculator who observes o = H, and “negatively-informed speculator” to describe a

speculator who observes o = 1. The variable ` is a measure of market sophistication or the

informedness of outside investors and will generate a number of comparative statics.

Trading in the …nancial market happens in t = 1. Always present is a noise trader, who

trades . = ÷1, 0, or 1 with equal probabilities. If the speculator is present, she makes an

endogenous trading choice : ¸ ¦÷1. 0. 1¦. Trading either ÷1 or 1 is costly for the speculator

and entails paying a cost of i. Unless otherwise speci…ed, we refer to trading pro…ts and losses

gross of the cost i. If the speculator is indi¤erent between trading and not trading (because

her expected pro…ts from trading exactly equal i), we assume that she will not trade.

Following Kyle (1985), orders are submitted simultaneously to a market maker who sets the

price and absorbs order ‡ows out of his inventory. The orders are market orders and are not

contingent on the price. The competitive market maker sets the price equal to expected asset

value, given the information contained in the order ‡ow. The market maker can only observe

total order ‡ow A = : +., but not its individual components : and .. Possible order ‡ows are

A ¸ ¦÷2. ÷1. 0. 1. 2¦ and the pricing function is j (A) = 1(·[A). A critical departure from

Kyle (1985) is that …rm value here is endogenous, because the manager’s action is based on

information revealed during the trading process.

Speci…cally, the manager observes total order ‡ow A, and uses the information in A to form

his posterior ¡, which is then used in the investment decision. Allowing the manager to observe

order ‡ow A, rather than just the price j, simpli…es the analysis without a¤ecting its economic

content. In the equilibria that we analyze, there is a one-to-one correspondence between the

price and the order ‡ow so it does not matter which variable the manager observes. Under the

alternative assumption that the manager observes j, other, non-interesting, equilibria can arise,

where the price is essentially uninformative. Since this paper’s focus is to analyze the feedback

e¤ect, which requires the price to be informative, we do not analyze such equilibria here. It

is also realistic to assume that managers have access to information about trading quantities

in the …nancial market: …rst, market making is competitive and so there is little secrecy in

the order ‡ow; second, microstructure databases (such as TAQ) provide such information at a

short lag – rapidly enough to guide investment decisions.

2.3 Equilibrium

The equilibrium concept we use is the Perfect Bayesian Nash Equilibrium. Here, it is de…ned

as follows: (i) A trading strategy by the speculator: o : ÷ ¦÷1. 0. 1¦ that maximizes his

expected …nal payo¤ :(· ÷j) ÷[:[i, given the price setting rule, the strategy of the manager,

and his information about the realization of o. (ii) An investment strategy by the …rm 1 :

Q ÷ ¦i. :¦ (where Q = ¦÷2. ÷1. 0. 1. 2¦), that maximizes expected …rm value · = 1

d

given

the information in the order ‡ow and all other strategies. (iii) A price setting strategy by the

market maker j : Q ÷ R that allows him to break even in expectation, given the information

in the price and all other strategies. Moreover, (iv) the …rm and the market maker use Bayes’

9

rule in order to update their beliefs from the order they observe in the …nancial market, and

(v) beliefs on outcomes not observed on the equilibrium path satisfy the Cho and Kreps (1987)

intuitive criterion. Finally, (vi) all agents have rational expectations in that each player’s belief

about the other players’ strategies is correct in equilibrium.

3 Feedback E¤ect and Asymmetric Limits to Arbitrage

In this section, we characterize the pure-strategy equilibria in our model. We demonstrate the

emergence of asymmetric limits to arbitrage as a result of the feedback from market trading

outcomes to the …rm’s investment decision. We consider Case 1 (1

n

H

1

n

L

) …rst and then

proceed to Case 2 (1

n

H

< 1

n

L

).

3.1 Case 1: Firm Value is Monotone in the State: 1

n

H

1

n

L

We start with the case where ¸ <

1

2

, i.e., without further information, the …rm will choose to

invest. Later, we will show that our main insight carries through to the case where ¸

1

2

. In

our characterization, we make use of three di¤erent threshold levels of the cost of trading i:

i

1

=

1

3

_

1

2

_

1

i

H

÷1

i

L

_

+

1 ÷`

2 ÷`

(1

n

H

÷1

n

L

)

_

. (6)

i

2

=

1

3

_

1

2

+

1 ÷`

2 ÷`

_

_

1

i

H

÷1

i

L

_

. (7)

i

3

=

1

3

_

1

i

H

÷1

i

L

_

, where (8)

i

1

< i

2

< i

3

. and (9)

i

2

÷i

1

=

1

3

1 ÷`

2 ÷`

__

1

i

H

÷1

i

L

_

÷(1

n

H

÷1

n

L

)

¸

0. (10)

The results also depend on whether order ‡ow is su¢ciently informative to overturn the decision

to invest, which is the ex-ante optimal decision. Hence, we distinguish between two cases

depending on whether the cuto¤ ¸ is higher or lower than

1

2

. As we will show, the quantity

1

2

is relevant as, in some equilibria, it represents the posterior probability of state H under

an order ‡ow of A = ÷1. The …rst case is

1

2

< ¸. Here, the probability ` that the speculator

is present is su¢ciently high that a negative order of A = ÷1 is su¢ciently informative to

deter the manager from investing. Thus, there is feedback from the market to real decisions for

the case of A = ÷1.

3

Second,

1

2

¸. Here, a negative order of A = ÷1 is not su¢ciently

informative to lead the manager to abandon the default plan of investing. Thus, there is no

feedback e¤ect for A = ÷1.

3

While X = ÷2 is also a negative order ‡ow, the …rm’s decision in this case is not relevant for equilibrium

trading strategies as the speculator’s information is fully revealed and so she never makes a pro…t. Thus, this

node is not relevant for determining the equilibrium trading strategies.

10

As we show, depending on the values of i, four equilibrium outcomes can arise:

1. No Trade Equilibrium `1: the speculator does not trade,

2. Trade Equilibrium 1: the speculator buys when she knows that o = H and sells when

she knows that o = 1,

3. Partial Trade Equilibrium 1`o (Buy - Not Sell): the speculator buys when she knows

that o = H and does not trade when she knows that o = 1,

4. Partial Trade Equilibrium o`1 (Sell - Not Buy): the speculator does not trade when

she knows that o = H and sells when she knows that o = 1.

Proposition 1 provides the characterization of equilibrium outcomes.

Proposition 1 (Equilibrium, …rm value is monotone in the state, investment is ex-ante desir-

able). Suppose that 1

n

H

1

n

L

and ¸ <

1

2

. Then the trading game has the following pure-strategy

equilibria:

When i < i

1

, the only pure-strategy equilibrium is 1.

When i

1

_ i < i

2

: in the case of feedback (

1

2

< ¸), the only pure-strategy equilibrium is

1`o; in the case of no feedback (

1

2

¸), the only pure-strategy equilibrium is 1.

When i

2

_ i < i

3

, there are two pure-strategy equilibria: 1`o and o`1.

When i _ i

3

, the only pure-strategy equilibrium is `1.

That is, if and only if there is feedback (

1

2

< ¸), there is a strictly positive range of

parameter values (i

1

_ i < i

2

) for which the 1`o equilibrium exists but the o`1 equilibrium

does not exist. There is no range of parameter values for which the o`1 equilibrium exists but

the 1`o equilibrium does not exist.

Proof. Given that …rm value is always higher when o = H than when o = 1, it is straight-

forward to show that the speculator will never buy when she knows that o = 1 and will never

sell when she knows that o = H. Then, the only possible pure-strategy equilibria are `1, 1,

1`o, and o`1. Below, we identify the conditions under which each one of these equilibria

holds. If an order ‡ow of A = ÷2 (A = 2) is observed o¤ the equilibrium path, the beliefs

of the market maker and the manager are that the speculator knows that the state is 1 (H).

Given that speculators always lose if they trade against their information, this is the only belief

that is consistent with the intuitive criterion.

No Trade Equilibrium `1:

For a given order ‡ow A, the posterior ¡, the manager’s decision d and the price j are given

by the following table (see Appendix A for the full calculations):

A ÷2 ÷1 0 1 2

¡ 0

1

2

1

2

1

2

1

d : i i i i

j 1

n

L

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

i

H

11

As shown in Appendix A, the pro…t for the negatively-informed speculator from deviating

to selling is

1

3

(1

i

H

÷1

i

L

), and this is also the pro…t for the positively-informed speculator from

deviating to buying. Thus, this equilibrium holds if and only if i _ i

3

.

Partial Trade Equilibrium o`1:

For a given order ‡ow A, the posterior ¡, the manager’s decision d and the price j are given

by the following table:

A ÷2 ÷1 0 1 2

¡ 0

1

2

1

2

1

2

1

d : i i i i

j 1

n

L

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

i

H

Calculating the pro…t for the negatively-informed speculator from deviating to not trad-

ing and for the positively-informed speculator from deviating to buying, we can see that this

equilibrium holds if and only if i

2

_ i < i

3

.

Partial Trade Equilibrium 1`o:

For a given order ‡ow A, the posterior ¡, the manager’s decision d and the price j are given

by the following table:

A ÷2 ÷1 0 1 2

¡ 0

1

2

1

2

1

2

1

d :

_

: if

1

2

< ¸

i if

1

2

¸

i i i

j 1

n

L

_

1

2

1

n

H

+

1

2

1

n

L

if

1

2

< ¸

1

2

1

i

H

+

1

2

1

i

L

if

1

2

¸

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

i

H

Calculating the pro…t for the negatively-informed speculator from deviating to selling and

for the positively-informed speculator from deviating to not trading, we can see that this equi-

librium holds if and only if i

2

_ i < i

3

for the case of no feedback (

1

2

¸) and if and only

if i

1

_ i < i

3

for the case of feedback (

1

2

< ¸).

Trade Equilibrium 1:

For a given order ‡ow A, the posterior ¡, the manager’s decision d and the price j are given

by the following table:

A ÷2 ÷1 0 1 2

¡ 0

1

2

1

2

1

2

1

d :

_

: if

1

2

< ¸

i if

1

2

¸

i i i

j 1

n

L

_

1

2

1

n

H

+

1

2

1

n

L

if

1

2

< ¸

1

2

1

i

H

+

1

2

1

i

L

if

1

2

¸

1

2

1

i

H

+

1

2

1

i

L

1

2

1

i

H

+

1

2

1

i

L

1

i

H

12

Calculating the pro…t for the negatively-informed speculator from deviating to not trading

and for the positively-informed speculator from deviating to not trading, we can see that this

equilibrium holds if and only if i < i

2

for the case of no feedback (

1

2

¸) and if and only if

i < i

1

for the case of feedback (

1

2

< ¸).

Thus, there is a range of i for which the only equilibrium is 1`o if i

1

< i

2

and

1

2

< ¸.

From (4) and (10), i

1

< i

2

requires ` < 1. In turn,

1

2

< ¸ requires `

12

1

. Thus, there

exist values of ` that satisfy both of the above conditions if

12

1

< 1, which always holds.

Proposition 1 demonstrates the sources of limits to arbitrage in our model, one of which is

the feedback e¤ect that is the focus of our paper. To understand the various forces, we start

by describing the equilibrium outcomes in the case of no feedback, i.e., when

1

2

¸. Here,

an order ‡ow of A = ÷1 may convey (depending on the equilibrium) negative information, but

not su¢ciently negative to deter the manager from abandoning the default plan of investing.

In this case, there are three regions of the parameter i. When i < i

2

, the only pure-strategy

equilibrium is one where the speculator always trades on her information. When i

2

_ i < i

3

,

there are two pure strategy equilibria, exhibiting limited trade, one in which the speculator

buys on good news but does not trade on bad news, and one in which she sells on bad news

but does not trade on good news. When i _ i

3

, the only pure-strategy equilibrium entails no

trade at all by the speculator.

Two sources of limits to arbitrage are present in the no-feedback case, both of which are

common in the literature. The …rst source is the trading cost i. As i increases, we move

to equilibria where speculators trade less on their information. Clearly, when speculators are

subject to greater transaction costs, they have lower incentives to trade. The second source is the

price impact that speculators exert when they trade on their information. In the intermediate

region i

2

_ i < i

3

, there are equilibria where the speculator trades on one type of information

but not the other. There is symmetry in that both types of asymmetric equilibria are possible

in exactly the same range of parameters. To understand the intuition behind these asymmetric

equilibria, consider the 1`o equilibrium without feedback (the case of the o`1 equilibrium

is analogous). Given that the market maker believes that the speculator buys on good news,

a negative order ‡ow is very revealing that the speculator is negatively informed and the price

moves sharply to re‡ect this. Speci…cally, A = ÷1 is inconsistent with the speculator having

positive information, and so the speculator only receives

1

2

1

i

H

+

1

2

1

i

L

from selling. Thus, the

speculator makes little pro…t from selling on bad news; knowing this, she chooses not to trade

on bad news. Conversely, given that the market maker believes that the speculator does not sell

on bad news, a positive order ‡ow is consistent with the speculator being negatively informed:

A = 1 is consistent with the noise trader buying, and the speculator being negatively informed

and choosing not to trade. Thus, the market maker sets a relatively low price of

1

2

1

i

H

+

1

2

1

i

L

,

which allows the speculator to make high pro…ts by buying. Thus, the equilibriumis sustainable.

In sum, in both partial trade equilibria, the order ‡ow in the direction in which the speculator

does not trade becomes particularly informative, leading to larger price impact which reduces

13

the potential trading pro…ts. Thus, not trading in this direction is sustained in equilibrium.

This force is symmetric in the absence of feedback.

We now move to the case of feedback, i.e., when

1

2

< ¸. Here, an order ‡ow of A = ÷1

provides enough negative information for the manager to abandon the investment. Abandon-

ment is the optimal decision in state 1; thus, improving the manager’s decision reduces the

speculator’s pro…t in the node of A = ÷1 from

1

2

(1

i

H

÷1

i

L

) (in the case of no-feedback) to

only

1

2

(1

n

H

÷1

n

L

). This reduced pro…t a¤ects the speculator’s equilibrium trading strategy

and causes her not to sell on bad news if i

1

_ i. Our main result is that the feedback e¤ect

introduces an additional limit to arbitrage that is distinct from those identi…ed in prior litera-

ture – arbitrage is limited because the value of the asset being arbitraged is endogenous to the

act of arbitrage. Unlike trading costs and price impact, the limit to arbitrage arising from the

feedback e¤ect is asymmetric: it reduces the extent of selling on bad news but not the extent of

buying on good news. Indeed, the di¤erence between equilibrium outcomes in the two cases of

no-feedback and feedback is that in the range i

1

_ i < i

2

, the Trade Equilibrium 1 is replaced

with the Partial Trade Equilibrium 1`o. However, there is no range of parameters where the

o`1 equilibrium exists but the 1`o equilibrium does not exist.

The intuition behind the asymmetry of our e¤ect is as follows. In the case of feedback, when

the speculator sells on bad news, she may lead the manager to abandon a bad investment. By

that, she improves …rm value, since 1

n

L

1

i

L

. Since she is holding a short position, this

increase in …rm value reduces her pro…t. Hence, it deters the speculator from selling on bad

news. On the other hand, the feedback e¤ect does not deter the positively-informed speculator

from buying on good news. Buying on good news may reveal to the manager that the state

is good, which (weakly) causes him to increase investment; since investment is desirable in the

high state, this augments …rm value. The speculator will then pro…t from the increase in the

value of her long position, which will further increase her incentive to trade.

4

Overall, trading on her information in either direction – whether it is buying on positive

information or selling on negative information – conveys information to the manager. This

improves his decision making and thus fundamental …rm value. Increased …rm value augments

the pro…tability of a long position but reduces the pro…tability of a short position. Hence, the

feedback e¤ect leads to an asymmetric limit to arbitrage that deters selling on bad news but

not buying on good news. By contrast, the two limits to arbitrage studied in prior research

are symmetric. A high trading cost i leads to the `1 equilibrium where there is no trading

in either direction. Price impact leads to the two partial trade equilibria, 1`o and o`1, but

there is symmetry in that both equilibria are possible in exactly the same range of parameters.

In particular, without feedback (i.e., if ¸

1

2

), there is no value of i in which there is one

4

In the case discussed so far ( <

1

2

) the default option for the manager is to invest, and so positive news

from the market does not change his decision and does not a¤ect …rm value. Hence we state that buying on

good information causes the manager to weakly increase investment. As we will show later, if >

1

2

, buying on

good news causes the manager to strictly increase investment, in turn strictly improving …rm value. This e¤ect

is the driving force behind our results in the case of >

1

2

.

14

partial trade equilibrium but not the other.

The reason for why the feedback e¤ect reduces trading pro…ts is nuanced. Intuition may

suggest that the market maker’s pricing function can “undo” the feedback e¤ect: the market

maker is fully rational and takes into account the fact that the order ‡ow will a¤ect the man-

ager’s decision; since he is competitive, he sets a price that re‡ects this decision. Because the

price that the speculator receives from selling will always re‡ect the action taken by the manager

(be it continuation or investment), it may seem that the action does not matter. Such intuition

turns out to be incorrect. The key to our result is that the source of the speculator’s pro…ts is

not superior knowledge of the manager’s action (since the action is always perfectly predicted

by the market maker), but superior knowledge of the state. In turn, superior knowledge of the

state results from fact that ` < 1, i.e., the speculator is not always present. To see this, consider

the market maker’s inference from seeing A = ÷1 in the 1`o equilibrium. This order ‡ow is

consistent with either the speculator being absent (in which case the state may be H or 1), or

the speculator being present and negatively informed. If ` = 1, the …rst case is ruled out, and

so the market maker knows for certain that o = 1. Thus, the order ‡ow of A = ÷1 is fully

revealing: the market maker knows both that correction will occur, and that the state is 1, and

so sets price exactly equal to the fundamental value of 1

n

L

. Thus, the speculator makes zero

pro…t. Indeed, if ` = 1, then i

1

= i

2

and there is no range of parameter values in which there

is a 1`o equilibrium only. By contrast, if ` < 1, the market maker predicts the manager’s

action but does not know the state. Since A = ÷1 can be consistent with the speculator being

absent and the state being H, the market maker allows for the possibility that the state may

be H and sets a price of

1

2

1

d

H

+

1

2

1

d

L

. Because the speculator knows both the manager’s

action and the state is 1, she makes a pro…t of

1

2

_

1

d

H

÷1

d

L

_

, which in turn depends on the

decision d. The source of her pro…t is her superior information on the state; since the state has

a lower e¤ect on …rm value under correction, her pro…ts are lower in this case. Put di¤erently,

the manager’s action d (and thus the feedback e¤ect on the manager’s action) matters for the

speculator’s trading pro…ts, not because the speculator’s pro…ts stem from superior knowledge

of the manager’s action, but because the action a¤ects the value of the speculator’s superior

knowledge on the state.

We now wish to verify that the asymmetry between buy-side speculation and sell-side specu-

lation, driven by the feedback e¤ect, is not an artifact of the fact that investment is the default

decision, i.e. the case ¸ <

1

2

. The next proposition shows that when ¸

1

2

, i.e., when the

default decision is abandonment, our results are qualitatively similar: without feedback, 1`o

and o`1 equilibria occur over the same range of parameters, whereas with feedback, the 1`o

equilibrium occurs over a wider range than the o`1 equilibrium. In the case of ¸ <

1

2

, the

source of the limit to arbitrage was that the feedback e¤ect reduces the pro…tability of a short

position but does not a¤ect the pro…tability of a long position, since positive order ‡ow leads

to investment but the investment would be undertaken in the absence of further information

anyway. Here, the source is that the feedback e¤ect increases the pro…tability of a long position

15

but does not a¤ect the pro…tability of a short position, since abandonment would be undertaken

in the absence of further information anyway. In both cases (for both ¸ <

1

2

and ¸

1

2

), the

intuition is the same: the feedback e¤ect (weakly) increases the pro…tability of a long position

and (weakly) decreases the pro…tability of short position, as discussed above.

For this proposition, de…ne new threshold levels of the cost of trading i:

i

0

1

=

1

3

_

1

2

(1

n

H

÷1

n

L

) +

1 ÷`

2 ÷`

_

1

i

H

÷1

i

L

_

_

. (11)

i

0

2

=

1

3

_

1

2

+

1 ÷`

2 ÷`

_

(1

n

H

÷1

n

L

) . (12)

i

0

3

=

1

3

(1

n

H

÷1

n

L

) . c:d (13)

i

0

2

< i

0

3

. i

0

2

< i

0

1

(14)

The cuto¤ for feedback e¤ect to exist is also adjusted here. In some equilibria,

1

2

represents

the posterior probability of state H if A = 1. If

1

2

¸, the probability ` that the speculator is

present is su¢ciently high that an order ‡ow of A = 1 contains enough information to lead the

manager to invest (as opposed to the default option of abandoning). Hence, there is feedback.

If

1

2

< ¸, an order ‡ow of A = 1 is not informative enough to lead the manager to invest.

This is the case where there is no feedback.

The following proposition provides the characterization of equilibrium outcomes.

Proposition 2 (Equilibrium, …rm value is monotone in the state, investment is ex-ante unde-

sirable). Suppose that 1

n

H

1

n

L

and ¸

1

2

, then the trading game has the following pure-strategy

equilibria:

When i < i

0

2

, the only pure-strategy equilibrium is 1.

When i

0

2

_ i < i

0

3

: in the case of no feedback (

1

2

< ¸), there are two pure-strategy

equilibria, 1`o and o`1; in the case of feedback (

1

2

¸), the 1`o equilibrium always

exists, whereas the o`1 equilibrium exists only in the sub-range i

0

1

_ i < i

0

3

or does not exist

(if i

0

1

i

0

3

).

When i _ i

0

3

, the only pure-strategy equilibrium is `1.

That is, if and only if there is feedback (

1

2

¸), there is a range of parameter values for

which the 1`o equilibrium exists but the o`1 equilibrium does not exist. If i

0

1

i

0

3

, this

range is i

0

2

_ i < i

0

3

; if i

0

1

< i

0

3

, this range is i

0

2

_ i < i

0

1

. There is no range of parameter

values for which the o`1 equilibrium exists but the 1`o equilibrium does not exist.

Proof. The proof repeats similar steps to those in the proof of Proposition 1, and is thus

omitted for brevity.

In the case of ¸ <

1

2

, the role of the feedback e¤ect can be seen in the 1`o equilibrium:

it reduces the pro…ts that the negatively-informed speculator would earn by deviating and

selling, and so the 1`o equilibrium is sustainable over a wider range of parameters than the

16

o`1 equilibrium. Here, where ¸

1

2

, the feedback e¤ect impacts the o`1 equilibrium.

Since buying improves the …rm’s fundamental value, the feedback e¤ect increases the pro…t

that the positively-informed speculator would earn by deviating and buying, and so the o`1

equilibrium is sustainable over a narrower range of parameters than the 1`o equilibrium

(indeed, if i

0

1

i

0

3

, it is not sustainable at all). In both cases (for ¸ <

1

2

and ¸

1

2

), the end

result is the same: the feedback e¤ect increases the pro…ts from informed buying and reduces

the pro…ts from informed selling, leading to the 1`o equilibrium being sustainable over a

wider range of transactions costs than the o`1 equilibrium.

3.2 Case 2: Firm Value is Non-Monotone in the State: 1

n

H

< 1

n

L

In this subsection, we consider the case where, if the …rm does not invest, its value is higher

in state o = 1 (1

n

H

< 1

n

L

). Hence, the corrective action is su¢ciently powerful to outweigh

the e¤ect of the state on …rm value and lead to a higher value in the low state. We start by

characterizing equilibrium outcomes for the case where ¸ <

1

2

, i.e., without further information,

the …rm will choose to invest.

The analysis of equilibriumoutcomes becomes more complicated in the case of non-monotonicity.

In the previous subsection, where …rmvalue is monotone in the state, a positively-informed spec-

ulator always loses money by selling and a negatively-informed speculator always loses money

by buying, since …rm value is always higher in state H than in state 1. However, now that …rm

value may be higher in state 1, a positively-informed speculator may …nd it optimal to sell and

a negatively-informed speculator may …nd it optimal to buy. Hence, there are nine possible

pure-strategy equilibria (each type of speculator – positively-informed and negatively-informed

– may either buy, sell, or not trade). The following lemma simpli…es the equilibrium analysis,

moving us closer to the analysis conducted in the previous subsection.

Lemma 1 Suppose that 1

n

H

< 1

n

L

and ¸ <

1

2

, then:

(i) The trading game has no pure-strategy equilibrium where the speculator sells when she

knows that o = H.

(ii) The trading game has no pure-strategy equilibrium where the speculator buys when she

knows that o = 1.

Proof. (i) Suppose that the speculator sells when she knows that o = H, then A ¸ ¦÷2. ÷1. 0¦.

In each one of these nodes, posterior probability ¡ of state H is at least

1

2

(given that these

nodes are consistent with the action of the positively-informed speculator and may or may not be

consistent with the action of the negatively-informed speculator, depending on her equilibrium

action). Then, since ¸ <

1

2

, investment will occur, and so …rm value is 1

i

H

. The price, however,

will be between 1

i

L

and 1

i

H

, and so the speculator makes a loss from selling.

(ii) Suppose that the speculator buys when she knows that o = 1, then A ¸ ¦0. 1. 2¦.

Given that the positively-informed speculator does not sell, the posterior probability ¡ is

1

2

at

A ¸ ¦0. 1¦. Hence, since ¸ <

1

2

, investment will occur, and so …rm value is 1

i

L

. Since the price

17

is

1

2

1

i

H

+

1

2

1

i

L

, the speculator will lose money on these nodes. When A = 2, there are two

possibilities. If the positively-informed speculator buys in equilibrium, then the outcome is the

same as on the other nodes. If she does not trade in equilibrium, then the negatively-informed

speculator is revealed, buying a security worth 1

n

L

for a price of 1

n

L

. Thus, in expectation she

makes a loss, given she loses at A ¸ ¦0. 1¦.

Following the lemma, there are four possible pure-strategy equilibria, just as in the previous

subsection: `1, 1, o`1, and 1`o. However, the conditions for these equilibria to hold are

now tighter. The reason that the positively-informed speculator never sells in equilibrium is

that if the market maker and the manager believe that she sells, she cannot make a pro…t from

selling. However, she still might be tempted to deviate to selling in any of the four equilibria

mentioned above. When she sells, she potentially misleads the market maker and the manager

to think that the negatively-informed speculator is present, and so to abandon the investment.

Since abandonment is suboptimal if o = H, this decision reduces …rm value and causes the

speculator to make a pro…t on her short position. Hence, for any of the above four equilibria to

hold, an additional condition must be satis…ed to ensure that the positively-informed speculator

does not have an incentive to deviate to selling. Interestingly, the same issue does not arise with

the negatively-informed speculator, as she never has an incentive to deviate to buying. If she

does so, she misleads the market maker and the manager to think that the positively-informed

speculator is present, and so to (incorrectly) take the investment. Again, this reduces …rm

value, but because the speculator has a long position, this causes her to make a loss.

5

In analyzing deviations from the equilibrium, another issue that arises in this subsection is

the speci…cation of o¤-equilibrium beliefs. In Case 1, due to monotonicity, the only assumption

that satis…ed the intuitive criterion was that an o¤-equilibrium order ‡ow of A = 2 is due

to the positively-informed speculator (and so the posterior is ¡ = 1), while an o¤-equilibrium

order ‡ow of A = ÷2 is due to the negatively-informed speculator (and so the posterior is

¡ = 0). In this subsection, however, the intuitive criterion is not su¢cient to rule out other

o¤-equilibrium beliefs. We nevertheless retain this assumption regarding o¤-equilibrium beliefs,

which is reasonable given the possible equilibria in our model. Our results remain the same for

any other o¤-equilibrium beliefs that are monotone in the order ‡ow.

The following proposition provides the characterization of equilibrium outcomes.

Proposition 3 (Equilibrium, …rm value is non-monotone in the state, investment is ex-ante

desirable). Suppose that 1

n

H

< 1

n

L

and ¸ <

1

2

, and suppose that the belief of the market

maker and the manager is that an o¤-equilibrium order ‡ow of A = ÷2 (A = 2) is associated

with the presence of negatively-informed (positively-informed) speculator. Then, if

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

is

su¢ciently high, the characterization of equilibrium outcomes is identical to that in Proposition

1.

5

Goldstein and Guembel (2008) also derive conditions to ensure that the speculator does not deviate from

the equilibrium to trade against her information.

18

More speci…cally, the following additional conditions are required for the various equilibria

to hold:

Equilibrium `1 and o`1: i

1

3

(÷(1

i

H

÷1

i

L

) + (1

n

L

÷1

n

H

));

Equilibrium 1`o: in the case of feedback (

1

2

< ¸),

63

62

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

1; in the case of no

feedback (

1

2

¸),

83

42

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

1.

Equilibrium 1: in the case of feedback (

1

2

< ¸),

32

3

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

1; in the case of no

feedback (

1

2

¸), 2

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

1.

Proof. The calculations of the posterior ¡, the manager’s decision d and the price j for di¤erent

order ‡ows A in the various possible equilibria are identical to those provided in the proof of

Proposition 1. Hence, the conditions for the positively-informed speculator to choose between

buying and not trading and for the negatively-informed speculator to choose between selling

and not trading are identical to those derived in the proof of Proposition 1. Analyzing the

possible trading pro…ts for the negatively-informed speculator from deviating to buying in each

of the four possible equilibria, it is straightforward to see that she always loses from buying and

hence will never deviate. Appendix A calculates the possible trading pro…ts for the positively-

informed speculator from deviating to selling in each of the four possible equilibria, which yields

obtain the additional conditions stated in the body of the proposition. These conditions are

binding only when

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

is not su¢ciently high.

As the proposition demonstrates, the main force identi…ed in the previous subsection for the

case where 1

n

H

1

n

L

, exists also in the case where 1

n

H

< 1

n

L

. That is, the feedback e¤ect deters

the negatively-informed speculator, but not the positively-informed speculator, from trading on

her information. In this subsection, this force is even stronger because the range of transaction

costs i between i

1

and i

2

, in which the 1`o equilibrium exists due to feedback but the

o`1 equilibrium does not exist, is higher when (1

n

H

÷1

n

L

) is negative: see equation (10). A

strong feedback e¤ect, in which correction not only mitigates the e¤ect of the low state but

also overturns it, implies that the negatively-informed speculator can make a loss – even before

transaction costs – when selling on bad news. This result is in contrast to standard informed

trading models where a speculator can never make a loss (before transactions costs) if she trades

in the direction of her information. This loss occurs at the A = ÷1 node; again, the key to this

result is ` < 1. Even though both the speculator and market maker know that abandonment

will occur if A = ÷1, they have di¤ering views on …rm value conditional on abandonment.

The speculator knows that the corrective action will be taken, and that correction is desirable

for …rm value (since she knows that o = 1), and so …rm value is 1

n

L

. In contrast, the market

maker knows the corrective action will be taken but is not certain that correction is desirable

for …rm value, because she is unsure of the underlying state o. Order ‡ow A = ÷1 is consistent

with a negatively-informed speculator, but also with an absent speculator and selling by noise

traders. Hence, it is possible that o = H, in which case the manager’s corrective action is

19

undesirable, leading to …rm value of 1

n

H

. Therefore, the price set by the market maker is only

1

2

1

n

H

+

1

2

1

n

L

, since he puts weight on the fact that correction may be undesirable, and so

the speculator loses

1

2

(1

n

H

÷1

n

L

) before transaction costs.

However, the proposition also shows that another force that arises from the feedback e¤ect

exists in this subsection, and that this force has implications on the characterization of equi-

librium outcomes. This is the desire of the positively-informed speculator to deviate from her

equilibrium behavior and manipulate the price by selling, even though she has good news. She

can potentially pro…t from leading the manager to take the wrong decision, which enables her

to pro…t from her short position. The manipulation incentive is not strong enough to interfere

with equilibrium conditions as long as

(

R

i

H

R

i

L

)

(

R

n

L

R

n

H

)

is su¢ciently high. In this case, the loss from

trading against good news (which is proportional to (1

i

H

÷1

i

L

)) is high relative to the bene-

…t from manipulation (which is proportional to (1

n

L

÷1

n

H

)). Otherwise, there are additional

conditions for the various possible equilibria, making it relatively more di¢cult to obtain the

1`o equilibrium due to feedback.

Finally, we analyze the case where ¸

1

2

. It turns out that this case is the exact mirror

image of the case where ¸ <

1

2

. Now, e¤ectively, o = H represents bad news and o = 1

represents good news. This is because the default decision is to abandon the investment; under

this decision, …rm value is lower in state H than in state 1. Thus, the speculator now sells if

o = H and buys if o = 1. The next lemma is the mirror image of Lemma 1:

Lemma 2 Suppose that 1

n

H

< 1

n

L

and ¸

1

2

, then:

(i) The trading game has no pure-strategy equilibrium where the speculator sells when she

knows that o = 1.

(ii) The trading game has no pure-strategy equilibrium where the speculator buys when she

knows that o = H.

Proof. The proof is symmetric to the proof of Lemma 1 and hence is not repeated here.

Hence, the possible pure-strategy equilibria here are:

1. No Trade Equilibrium `1: the speculator does not trade,

2. Trade Equilibrium 1

0

: the speculator buys when she knows that o = 1 and sells when

she knows that o = H,

3. Partial Trade Equilibrium 1`o

0

(Buy - Not Sell): the speculator buys when she knows

that o = 1 and does not trade when she knows that o = H.

4. Partial Trade Equilibrium o`1

0

(Sell - Not Buy): the speculator does not trade when

she knows that o = 1 and sells when she knows that o = H.

The characterization of equilibrium outcomes in the following proposition is symmetric to

that in Proposition 3:

20

Proposition 4 (Equilibrium, …rm value is non-monotone in the state, investment is ex-ante

undesirable). Suppose that 1

n

H

< 1

n

L

and ¸

1

2

, and suppose that the belief of the market maker

and the manager is that an o¤-equilibrium order ‡ow of A = ÷2 (A = 2) is associated with the

positively-informed (negatively-informed) speculator. Then, the characterization of equilibrium

outcomes is symmetric to that in Proposition 3: parameters 1

i

H

, 1

i

L

, 1

n

H

, 1

n

L

are replaced with

parameters 1

n

L

, 1

n

H

, 1

i

L

, 1

i

H

, respectively, and equilibria 1, 1`o, o`1 are replaced with

equilibria 1

0

, 1`o

0

, o`1

0

, respectively.

Proof. The proof is symmetric to the proof of Proposition 3 and hence is not repeated here.

Overall, the result is identical to that of the case of ¸ <

1

2

. Due to feedback, the speculator

is deterred from selling when she has bad news, but not from buying when she has good news.

The only di¤erence is that now, bad news entails o = H and good news entails o = 1.

In Case 1 (1

n

H

1

n

L

), for the sub-case of ¸

1

2

, the role of the feedback e¤ect is seen

in the o`1 equilibrium: the feedback e¤ect increases the pro…ts that the positively-informed

speculator would earn by deviating to buying, and so the o`1 is sustainable over a narrower

range of parameters. In the current scenario of ¸

1

2

within Case 2 (1

n

H

< 1

n

L

), just as in the

scenario of ¸ <

1

2

(for both Case 1 and Case 2), the role of the feedback e¤ect is seen in the

1`o/1`o

0

equilibrium: it deters the speculator from deviating to sell on bad news (o = H

in this case).

3.3 Summary and Discussion of Assumptions

The above analysis has shown that the presence of feedback from market trading to …rms’

decisions creates a wedge between buy-side speculation and sell-side speculation, discouraging

speculators from selling on bad news, but encouraging them to buy on good news. Several

assumptions play a key role in generating this result. These assumptions in turn lead to

empirical predictions, since they demonstrate the conditions under which the asymmetric limit

to arbitrage will exist.

First, the trading in the market has to contain su¢cient information to in‡uence the man-

ager’s decision. For example, consider the result in Proposition 1: for the wedge to arise, we

require

1

2

< ¸. Hence, it is important that the probability ` that the speculator is present is

su¢ciently high so that the order ‡ow is su¢ciently informative to change managerial decisions.

In turn,

1

2

< ¸ is more likely to be satis…ed the closer ¸ is to

1

2

, i.e. the closer the NPV of the

project is to 0. When ¸ is close to

1

2

, the desirability of the investment is su¢ciently uncertain

that the manager’s decision will be in‡uenced by the trading in the …nancial market. If ¸ is very

low, the ex-ante NPV of the project is so high that the manager will almost always undertake

the investment, regardless of order ‡ow.

Second, another important assumption is that ` < 1, so there is uncertainty on whether there

is an informed speculator in the market. To see this, consider again the result in Proposition

21

1. The range between i

1

and i

2

, in which a 1`o equilibrium arises and a o`1 equilibrium

does not arise, shrinks to zero if ` = 1. This is because the di¤erence in expected pro…t from

buying on good information and selling on bad information stems entirely from the di¤erence

in pro…t in the node where the speculator is partially revealed (A = ÷1 or 1). If ` = 1, the

speculator is fully revealed in these nodes and her pro…ts are zero, leading to no asymmetry.

Where ` < 1, the speculator is not fully revealed and makes a positive pro…t from her private

information about the state; the value of this information depends on the manager’s decision

(as this a¤ects the dependence of …rm value on the state) and thus the feedback e¤ect. We

would achieve the same result by instead assuming that the speculator is always present and

informed, but can only trade with probability ` – for example, if with probability 1 ÷ ` she

receives a liquidity shock that prevents her from trading.

6

Third, the reason that the speculator loses from increasing the …rm’s value is that she ends

up with a short position. Hence, it is important that the speculator short sells rather than

just sells stocks she previously owned, which in turn requires the speculator’s initial position

to be zero (or, at least, less than the amount sold) and short-sales to be possible. Thus, the

model delivers the result that investors are more likely to engage in sales rather than short-

sales, even in the absence of a short-sales constraint. However, if the speculator maximizes

returns relative to other speculators or market indices rather than absolute returns (e.g. she is

a mutual fund benchmarked against the performance of other mutual funds), then our limit to

arbitrage may exist even if her initial position is strictly positive. For example, if she sells half

of her portfolio, she increases the value of the remaining half, but increases the value of the

entire portfolio held by her competitors, and so loses in relative terms. In this case, the limit

to arbitrage identi…ed by this paper may exist even in the presence of short-sales constraints.

While short-sales constraints do not deter selling to a non-negative …nal position, the feedback

e¤ect can deter such selling if the speculator maximizes relative performance. Therefore, our

model also predicts that investors who are evaluated according to absolute returns are more

likely to sell on negative information than those who are benchmarked to their peers. Indeed,

hedge funds appear to sell (not just short-sell) more readily than mutual funds.

7

Fourth, the real decision is a corrective action in that it improves …rm value in the low

state. This is a natural assumption if the decision maker is the …rm’s manager who attempts

to maximize …rm value via an investment decision; another potential application is to a board

of directors which chooses whether to …re an underperforming manager in the bad state. The

model does not apply to amplifying actions that worsen …rm value in the low state, i.e. violate

6

An alternative assumption would be that the speculator is always present, but sometimes she is unin-

formed. This, however, may introduce other complications, as the uninformed speculator may choose to trade

to manipulate the price and the …rm’s decision, as in Goldstein and Guembel (2008).

7

Note that the existence of benchmarking alone is typically insu¢cient to explain the reluctance of mutual

funds to deviate from their benchmark, since the gains from beating one’s benchmark (by deviating) equal the

losses from underperforming one’s benchmark. Thus, existing explanations typically rely on the asymmetry in

in the ‡ow-performance relation: the in‡ows from beating one’s benchmark are lower than the out‡ows from

underperforming one’s benchmark. Our explanation for benchmarked investors’ unwillingness to sell does not

require such asymmetry.

22

assumption (2). For example, capital providers may withdraw their investment in the low state,

reducing …rm value further (as in Goldstein, Ozdenoren, and Yuan (2011)), or customers or

employees could terminate their relationship with a troubled …rm (Subrahmanyam and Titman

(2001)). Then, our model will have di¤erent implications: the speculator will no longer be

reluctant to sell on bad news, since the information will reduce …rm value further, enabling her

to pro…t more on her short position.

4 E¤ect of Information on Beliefs and Prices

The previous section demonstrated that the feedback e¤ect gives rise to an equilibrium where

a speculator buys on good news and does not trade on bad news. In this section, we study

the implications coming out of this equilibrium. The analysis that follows focuses on the 1`o

equilibrium where investment is ex-ante desirable (¸ <

1

2

) and there is feedback (

1

2

< ¸), and

considers both Case 1 and Case 2 together. Section 4.1 calculates the e¤ect of good and bad

news on the state on the posterior beliefs ¡, in order to study the extent to which information

reaches the manager and a¤ects real decisions. Section 4.2 analyzes the impact of news on

prices to generate stock return predictions.

4.1 Beliefs

Since the manager uses the posterior belief ¡ to guide his investment decision, ¡ measures the

extent to which information reaches the manager and a¤ects his actions. In a world in which

no agent observed the state, or in which the manager did not learn from prices or order ‡ow,

the posterior ¡ would equal the prior ¸ =

1

2

. Conversely, in a world of perfect information

transmission, ¡ = 1 if o = H and ¡ = 0 if o = 1. Our model, where information is partially

revealed through prices, lies in between these two polar cases. The absolute distance between

¡ and

1

2

measures the extent to which information reaches the manager.

Thus far, we have shown that good news received by the speculator has a di¤erent impact

on her trades (and thus total order ‡ow) than bad news. However, it is not obvious that

this will translate into a di¤erential impact on the manager’s beliefs. The manager is rational

and takes into account the fact that the speculator does not sell on negative information: he

updates his beliefs using the asymmetric equilibrium trading strategy. In the 1`o equilibrium

in the proof of Proposition 1, the manager recognizes that A = 1 could be consistent with a

negatively-informed speculator who chooses not to trade, and so ¡ (1) is no higher than ¡ (0)

(where ¡ (A) denotes the posterior at t = 1 upon observing order ‡ow A). Thus, even though

bad news can lead to a positive order ‡ow of A = 1, the manager knows that such an order ‡ow

can stem from a negatively-informed and non-trading speculator, and will decrease his posterior

accordingly. Put di¤erently, although negative information does not cause a negative order ‡ow

(on average), it can still have a negative e¤ect on beliefs and be fully conveyed to the manager.

23

Thus, it may seem still possible for good and bad news to be conveyed symmetrically to the

manager – by taking into account the speculator’s asymmetric trading strategy, he can “undo”

the asymmetry. Indeed, we start by showing that, if we do not condition on the presence of the

speculator, the e¤ects on beliefs of the high and low states being realized are symmetric. This

is a direct consequence of the law of iterated expectations: the expected posterior belief must

be equal to the prior.

Lemma 3 Consider the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback).

(i) If o = H, the manager’s expected posterior probability of the high state is ¡

H

=

(1)

2

63

+

1

3

+

3

and is increasing in `. (ii) If o = 1, the manager’s expected posterior probability of the high

state is ¡

L

=

1

63

+

1

3

and is decreasing in `. (iii) We have

q

H

+q

L

2

=

1

2

: thus, the realization of

state H has the same absolute impact on beliefs as the realization of state 1.

Proof. See Appendix A.

Of greater interest is to study the e¤ect of the state realization conditional upon the spec-

ulator being present. We use the term “good news” to refer to o = H being realized and the

speculator being present, since in this case there is an agent in the economy who directly re-

ceives news on the state; “bad news” is de…ned analogously. While the above analysis studied

the e¤ect of the state being realized (regardless of whether the state is learned by any agent

in the economy), this analysis studies the impact of the speculator receiving information about

the state. The goal is to investigate the extent to which the speculator’s good and bad news is

conveyed to the manager at t = 1. The results are given in Proposition 5 below:

Proposition 5 (Asymmetric e¤ect of positive and bad news on beliefs at t = 1.) Consider the

1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback). (i) If o = H and the

speculator is present, the manager’s expected posterior probability of the high state is ¡

H;spec

=

2

3

and is independent of `. (ii) If o = 1 and the speculator is present, the manager’s expected

posterior probability of the high state is ¡

L;spec

=

1

63

+

1

3

and is decreasing in `. (iii) We have

¡

H;spec

+ ¡

L;spec

2

=

1 +

1

63

2

. (15)

which is decreasing in `. Since

1+

1

63

2

1

2

, (15) implies that c/:

_

¡

H;spec

÷¸

_

÷c/:

_

¡

L;spec

÷¸

_

0, i.e. the absolute increase in the manager’s posterior if the speculator receives good news ex-

ceeds the absolute decrease in his posterior if the speculator receives bad news. The di¤erence

is decreasing in `.

Proof. See Appendix A.

Proposition 5 shows that, conditional upon the speculator being present, the impact on

beliefs of good news is greater in absolute terms than the impact of bad news. The asymmetry

is monotonically decreasing in the frequency of the speculator’s presence `. This result holds

24

even though the manager is rational and takes into account the fact that the speculator trades

asymmetrically when using the order ‡ow to update his prior. The source of the result is that,

even though the manager is rational, he is unable to distinguish the case of a negatively-informed

(and non-trading) speculator from that of an absent speculator (i.e. no information) – both

of these cases lead to the order ‡ow being ¦÷1. 0. 1¦ with uniform probability. Thus, negative

information has a smaller e¤ect on his belief. By contrast, if the speculator is always present,

the manager has no such inference problem and there is no asymmetry. This can be seen by

plugging ` = 1 into equation (15), in which case the average posterior equals the prior of

1

2

and

so we have c/:

_

¡

H;spec

÷¸

_

= c/:

_

¡

L;spec

÷¸

_

. Just as ` < 1 was a necessary condition for the

asymmetric feedback equilibrium to be the only equilibrium in the …rst place, it is a necessary

and su¢cient condition for bad news to have a smaller e¤ect on the manager’s belief than good

news.

The above analysis considered the change in the manager’s posterior at t = 1. At t = 2, the

state is realized and the posterior becomes either 1 (if o = H) or 0 (if o = 1). Since bad news

is conveyed to the manager to a lesser extent at t = 1, it seeps out to a greater extent ex post,

between t = 1 and t = 2. Thus, bad news causes a greater change in the posterior between

t = 1 and t = 2 than good news. This result is stated in Corollary 1 below:

Corollary 1 (Asymmetric e¤ect of high and low state realization on beliefs at t = 2). Consider

the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback). The absolute impact

on beliefs between t = 1 and t = 2 of the realization of the state is greater for the low state

o = 1 than for the high state o = H, i.e.

c/:

_

0 ÷¡

L;spec

_

÷c/:

_

1 ÷¡

H;spec

_

0.

The asymmetry is monotonically decreasing in the frequency of the speculator’s presence `.

Proof. Follows from simple calculations

The smaller e¤ect of bad news on the posterior at t = 1 is counterbalanced by its larger

e¤ect at t = 2. As we will show in Section 4.2, surprisingly this result need not hold when we

examine the e¤ect of news on prices rather than posteriors.

4.2 Stock Returns

We now calculate the impact of the state realization and news on prices, in order to generate

stock return implications. We study short-run stock returns between t = 0 and t = 1, and long-

run drift between t = 1 and t = 2. While this analysis is similar to Section 4.1 but studying

prices rather than beliefs, we will show that not all the results remain the same.

25

4.2.1 Short-Run Stock Returns

Even though the speculator trades asymmetrically, this need not imply that realizations of

the high and low states will have a di¤erential price impact, since the market maker takes

into account the speculator’s trading strategy when devising his pricing function. Lemma 4 is

analogous to Lemma 3 and shows that, unconditionally, the good and bad states have the same

absolute impact on prices. Let j

0

denote the “ex ante” stock price at t = 0, before the state

has been realized.

Lemma 4 Consider the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback):

(i) The stock price impact of the high state being realized is j

H

1

÷j

0

=

6

[j (2) ÷j (÷1)] 0.

(ii) The stock price impact of the low state being realized is j

L

1

÷ j

0

=

6

[j (÷1) ÷j (2)] <

0 = ÷

_

j

H

1

÷j

0

_

.

Proof. See Appendix A.

We have j

H

1

÷ j

0

= ÷

_

j

L

1

÷j

0

_

: the negative e¤ect of the low state equals the positive

e¤ect of the high state. Thus, the unconditional expected return is zero. This is an inevitable

consequence of market e¢ciency. The price at t = 0 is an unbiased expectation of the t = 1

expected price in the high state and the t = 1 expected price in the low state. Since both states

are equally likely, the absolute e¤ect of the high state must equal the absolute e¤ect of the low

state. An uninformed investor cannot trade the stock at t = 0 and expect a non-zero average

return at t = 1.

Proposition 6 is analogous to Proposition 5 and shows that, conditional on the speculator

being present, good news has a greater e¤ect than bad news:

Proposition 6 (Asymmetric e¤ect of positive and bad news on returns between t = 0 and

t = 1.) Consider the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback):

(i) If o = H and the speculator is present, the average return between t = 0 and t = 1 is

j

H;spec

1

÷j

0

=

1

3

_

1 ÷

2

_

(j (2) ÷j (÷1)) 0.

(ii) If o = 1 and the speculator is present, the average return between t = 0 and t = 1 is

j

L;spec

1

÷j

0

=

6

(j (÷1) ÷j (2)) < 0.

(iii) The di¤erence in the absolute average returns between the speculator learning o = H

and o = 1 is given by:

c/:

_

j

H;spec

1

÷j

0

_

÷c/:

_

j

L;spec

1

÷j

0

_

=

1

3

(1 ÷`) (j (2) ÷j (÷1)) 0. (16)

i.e. the stock price increase upon good news exceeds the stock price decrease upon bad news.

This di¤erence is decreasing in `.

(iv) The average return, conditional on the speculator being present, is positive:

j

spec

1

÷j

0

=

1

3

1 ÷`

2

(j (2) ÷j (÷1)) 0. (17)

26

This di¤erence is decreasing in `.

Proof. See Appendix A.

Proposition 6 states that the average return, conditional on the speculator being present, is

positive – i.e. the stock price increase upon positive information exceeds the stock price decrease

upon negative information (part (iii)). Put di¤erently, positive information is impounded into

prices to a greater degree than negative information. Since good and bad news are equally likely,

this means that the average return, conditional on the speculator being present, is positive (part

(iv)). As with Proposition 5, the key to this result is that, even though the market maker is

rational, he is unable to distinguish the case of a negatively-informed speculator from that of

an absent speculator (i.e. no information). If ` = 1, equations (16) and (17) become zero

and there is no asymmetry; the asymmetry is monotonically decreasing in `. Note that the

positive average return given in part (iv) is not inconsistent with market e¢ciency, because it

is conditional upon the speculator being present, which is private information. An uninformed

investor cannot buy the stock at t = 0 and expect to earn a positive return at t = 1 because

she will not know whether the speculator is present.

4.2.2 Long-Run Drift

We now move from short-run returns to calculating the long-run drift of the stock price, to

analyze the stock return analog of Corollary 1, i.e., the impact of the state realization on prices

between t = 1 and t = 2. Corollary 1 showed that the smaller e¤ect of bad news on beliefs at

t = 1 is counterbalanced by a larger e¤ect on beliefs at t = 2. Corollary 2 below shows that

this need not be the case for returns: it is possible for bad news to have a smaller e¤ect than

good news at both t = 1 and t = 2.

Corollary 2 (Asymmetric e¤ect of positive and bad news on returns between t = 1 and t = 2).

Consider the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback):

(i) If o = H and the speculator is present, the average return between t = 1 and t = 2 is

j

H;spec

2

÷j

H;spec

1

=

1

3

(1

i

H

÷1

i

L

) 0.

(ii) If o = 1 and the speculator is present, the average return between t = 1 and t = 2 is

j

L;spec

2

÷j

L;spec

1

= ÷

1

3

_

1

i

H

÷1

i

L

_

÷

1

3

_

1 ÷`

2 ÷`

(1

n

H

÷1

n

L

)

_

. (18)

which is negative in Case 1, but can be positive or negative in Case 2.

(iii) If (18) < 0, the di¤erence in the absolute average returns between the speculator learning

o = H and o = 1 is given by:

c/:

_

j

H;spec

2

÷j

H;spec

1

_

÷c/:

_

j

L;spec

2

÷j

L;spec

1

_

=

1

3

_

1 ÷`

2 ÷`

(1

n

L

÷1

n

H

)

_

.

27

which is positive in Case 2 and negative in Case 1. The magnitude of the di¤erence is decreasing

in `.

(iv) Expected …rm value at t = 2, conditional upon the speculator being present, is :

j

spec

2

=

1

2

1

i

H

+

1

3

1

i

L

+

1

6

1

n

L

.

and the average return between t = 1 and t = 2 if the speculator is present is:

j

spec

2

÷j

spec

1

=

1

6

1 ÷`

2 ÷`

(1

n

L

÷1

n

H

) ,

which is positive in Case 2 and negative in Case 1. The magnitude of the di¤erence is decreasing

in `.

Proof. See Appendix A.

Corollary 1 showed that, if the speculator is present, good news has a larger e¤ect on beliefs

at t = 1 than bad news, because she trades on the former but not the latter; thus, the expected

change in beliefs between t = 0 and t = 1 is positive. Since bad news has a smaller e¤ect at

t = 1, it must have a larger e¤ect at t = 2 (since the truth about the state comes out at t = 2),

and so the average increase in beliefs between t = 0 and t = 1 is reversed by an average decrease

in beliefs between t = 1 and t = 2. Corollary 2 shows that this need not be the case when we

study prices rather than beliefs: the speculator’s presence can lead to positive average returns

in both the short-run (between t = 0 and t = 1) and also in the long-run (between t = 1 and

t = 2). This is because the stock price depends not only on the beliefs about the state, but

also the manager’s action. Thus, there is an additional e¤ect of the speculator on prices that

does not exist in the analysis of beliefs: not only does she convey information about the state,

but also this information improves the manager’s decision-making and enhances …rm value –

the essence of the feedback e¤ect. In Case 2 (1

n

H

< 1

n

L

), this feedback e¤ect is su¢ciently

strong to turn the average return between t = 1 and t = 2 positive. In state 1, little bad news

emerges about the state between t = 0 and t = 1, which means that there is a large amount

of bad news to come out between t = 1 and t = 2; this in turn leads to the large downward

revision in beliefs in Corollary 1. However, the e¤ect on prices in Corollary 2 is muted because

the damage to …rm value caused by state 1 can be mitigated by taking the corrective action.

Thus, the negative e¤ect of bad news is smaller than the positive e¤ect of good news between

t = 1 and t = 2 as well as between t = 0 and t = 1 due to the feedback e¤ect. Indeed, if the

feedback e¤ect is su¢ciently strong, i.e. 1

n

L

is much higher than 1

n

H

, the return to bad news

between t = 1 and t = 2 can be positive ((18) 0). By contrast, in Case 1, (1

n

H

1

n

L

), the

long-run drift to the low state is larger in magnitude, analogous to Corollary 1. Since state 1

is bad for …rm value regardless of whether the manager takes the corrective action or not, the

realization of state 1 at t = 2 leads to a large decrease in prices.

28

The analysis thus far has considered the impact of news on prices at t = 1 and t = 2. We

now consider the impact of investment (a real variable) on prices; speci…cally, the extent to

which it is impounded into prices at t = 1 or at t = 2. While Section 4.2.1 showed that good

news received by the speculator has a greater short-run price impact than bad news, Proposition

7 now demonstrates a related result: the proportion of the total returns to an investment that

is realized in the short-run (at t = 1) rather than the long-run (at t = 2) is greater for a good

investment (o = H) than a bad investment (o = 1). In other words, the price impact of a good

investment is more front-loaded than for a bad investment.

Proposition 7 (Faster incorporation into prices of good investment than bad investment.)

Consider the 1`o equilibrium where ¸ <

1

2

and

1

2

< ¸ (i.e., there is feedback):

(i) If investment is undertaken in state H:

(ia) The average return between t = 0 and t = 1 is

j

i;H

1

÷j

0

=

1

6(2 + `)

[(2 + 2` ÷`

2

)1

i

H

+ (2 ÷2`)1

i

L

÷(2 ÷` ÷`

2

)1

n

H

÷(2 + `)1

n

L

] 0. (19)

(ib) The average return between t = 1 and t = 2 is

1

i

H

÷j

i;H

1

=

1

i

H

÷1

i

L

2 + `

0. (20)

(ii) If investment is undertaken in state 1:

(iia) The average return between t = 0 and t = 1 is

1

6

[(1 ÷`)

_

1

i

H

÷1

n

H

_

+ 1

i

L

÷1

n

L

] < 0. (21)

(iib) The average return between t = 1 and t = 2 is

1

i

L

÷j

i;H

1

= ÷

1

2

(1

i

H

÷1

i

L

) < 0. (22)

(iii) The returns to a good investment manifest more rapidly (i.e., to a greater degree at

t = 1) than the returns to a bad investment, i.e., c/: ((22) ÷(21)) c/: ((20) ÷(19)).

Proof. See Appendix A.

Parts (i) and (ii) of Proposition 7 show that investing in the high state leads to both positive

short-run returns between t = 0 and t = 1 and also positive long-run drift between t = 1 and

t = 2. Investing in the low state leads to negative short-run returns and negative long-run drift.

Part (iii) demonstrates that the returns to a good investment are realized to a greater extent at

t = 1 rather than t = 2, compared to a bad investment. Thus, the returns to a good investment

manifest more rapidly than the returns to a bad investment, i.e., are more front-loaded. To our

knowledge, this prediction has not yet been tested.

29

The intuition behind the asymmetry is di¤erent from Proposition 6. In both Propositions

6 and 7, the asymmetry occurs because the low state has a lesser impact on prices than the

high state. In Proposition 6, this arises from the fact that ` < 1, which means that the

market maker cannot distinguish the case of a negatively-informed speculator from that of an

uninformed speculator. Here, the intuition is as follows. If the investment is bad, the negative

returns cannot manifest too strongly at t = 1, otherwise the decline in the stock price will have

led to the investment being canceled. Thus, the negative returns must manifest predominantly

at t = 2. Put di¤erently, there are bad investments that do not lead to a sharply negative

reaction at t = 1 because the speculator did not trade on the bad news. Instead, the value-

destructiveness of the investment seeps out ex post. Note that the long-term drift in returns

does not violate market e¢ciency. The key to reconciling this result with market e¢ciency is

that …rm value is endogenous to trading. If speculators sold aggressively in response to a bad

investment, the decline in the stock price will lead to the investment being cancelled. Thus,

the negative returns must manifest predominantly at t = 2.

5 Summary of Implications

This section discusses several implications of our model. The …rst is that this paper identi…es

a limit to arbitrage which, in contrast to alternative explanations, is likely to persist over

time even as markets evolve and investors become more sophisticated. One existing source of

limited arbitrage is market frictions such as short-sales constraints, which will likely diminish

with the development of …nancial markets. A second is that investors in professional money

managers make their allocation decisions based on short-run measures of performance, which

leads to mutual funds avoiding arbitrage trading that will only converge in the long run (Shleifer

and Vishny (1997)). Such behavior can either be irrational over-extrapolation, or rational if

investors have limited information on the fund manager’s quality but instead must infer it

imperfectly from short-run performance. Either way, if investor sophistication and information

improve over time, this force will also diminish.

By contrast, the limit to arbitrage analyzed by this paper stems from …rm value being

endogenous to the act of arbitrage. This is a fundamental force that does not rely on short-sale

constraints, investor irrationality or investors’ limited information on the quality of a portfolio

manager, and so may continue to persist over time. (The only market imperfection that our

model requires is trading costs, which exist even in developed …nancial markets). There is no

exogenous friction preventing the arbitrageur from trading; instead, she endogenously chooses

not to trade because of the feedback e¤ect. All agents in the model act with full rationality:

the market maker takes into account the manager’s learning when setting the price, and this in

turn a¤ects the speculator’s decision to trade; the market maker knows that the speculator is

pursuing an asymmetric trading strategy. If anything, the limit to arbitrage may increase with

investor sophistication, as this augments the extent to which speculators have value-relevant

30

information which the manager attempts to learn by observing the price.

The second main category of applications stems from the fact that the limit to arbitrage

is asymmetric. While the speculator buys on good information, she does not sell on bad in-

formation. This prediction has implications for trading volume, suggesting that volume should

be higher upon good investments (such as M&A or capital expenditure) than bad investments.

Such a relation is consistent with the well-documented positive correlation between trading

volume and stock returns (see, e.g., Karpo¤ (1987)). Moreover, even though the market maker

takes the asymmetric trading volume into account, Proposition 6 shows that negative informa-

tion will enter into prices more slowly, as found empirically by Hong, Lim, and Stein (2000).

While Hong, Lim, and Stein’s results are consistent with the Hong and Stein (1999) model

that news travels more slowly in small …rms with low analyst coverage, Hong and Stein do not

predict an asymmetry between good and bad news.

8

Hong, Lim, and Stein speculate that the

asymmetry arises because key information is held by the …rm’s managers, and they dissem-

inate favorable information more enthusiastically than unfavorable information because they

are evaluated according to the stock price. Our theoretical model o¤ers a potential alternative

explanation. Key information is held by a …rm’s investors, who disseminate information not

through public news releases, but by trading on it. Their reluctance to disseminate bad news

is not because they are evaluated according to the stock price, but due to the limit to arbitrage

created by the feedback e¤ect. Another di¤erence is that, in an underreaction model, if bad

news has a smaller e¤ect on short-run returns than good news, it must be counterbalanced by

a larger long-run drift. By contrast, Corollary 2 shows that, in some cases, bad news generates

a smaller e¤ect on returns in both the short-run and long-run. Even though less bad news is

transmitted to the market at t = 1, meaning that there is more to come out at t = 2, stock

returns depend not only on the state, but also the manager’s decision. In our feedback model,

the manager can take a corrective action to mitigate the negative impact of the state on …rm

value, so the e¤ect of bad news on returns is lower at t = 2 as well as at t = 1. While the

above results are unconditional on investment occurring, the model also generates implications

for the short- and long-run returns to investment. Proposition 7 shows that the returns to

good investment are more front-loaded than the returns to a bad investment, because the spec-

ulator trades more readily on good news than bad news. Thus, the value-destructiveness of a

bad investment seeps out to a greater extent ex post, leading to negative long-run returns as

documented by Agrawal, Ja¤e, and Mandelker (1992) and Rau and Vermaelen (1998).

Moreover, the feedback e¤ect means that the lack of negative information in prices will

have further consequences on real decisions. In particular, if speculators choose not to trade on

negative information, then such negative information does not become incorporated into stock

prices and fails to in‡uence the manager’s behavior. Thus, some negative-NPV projects will not

be optimally abandoned, leading to overinvestment – even though there is an agent who knows

8

Note that our paper focuses on the e¤ect of news on stock prices. It does not address the predictability of

future returns from past returns, which is another component of the Hong, Lim, and Stein (2000) …ndings.

31

with certainty that the investment is undesirable, it still takes place. In the model, even if

o = 1, we have d = i if the noise trader does not sell. Critically, overinvestment does not occur

because the manager is pursuing private bene…ts, as in the standard theories of Jensen (1986),

Stulz (1990) and Zwiebel (1996). In contrast, the manager is fully aligned with …rm value and

there are no agency problems. The manager wishes to maximize …rm value by learning from

prices, but is unable to do so since speculators refrain from impounding their information into

prices. Overinvestment occurs even though the manager is fully aware that the speculator does

not trade on negative information and takes this into account.

The above overinvestment result can apply to M&A as well as organic expansion. Luo

(2005) shows that managers sometimes use the market reaction to announced M&A deals to

guide whether they should cancel the acquisition. While he …nds that some transactions are

canceled in equilibrium, our model suggests that there are other negative-NPV deals that should

optimally be canceled but are not because speculators do not impound their negative views into

prices. This may explain why a large proportion of M&A deals destroy value (see, e.g., Andrade,

Mitchell and Sta¤ord (2001).)

6 Conclusion

This paper has modeled a limit to arbitrage that stems from the fact that …rm value is endoge-

nous to the act of exploiting the arbitrage. Even if a speculator has negative information on the

state, she may strategically refrain from trading on it, because doing so conveys her information

to the manager. The manager may then take a corrective action that improves …rm value but

reduces the pro…ts from her short position below the cost of trading, and sometimes causes her

to realize a loss. There are several important di¤erences between the feedback-driven limit to

arbitrage that we study, and the limits to arbitrage identi…ed by prior literature. First, unlike

limits to arbitrage based on fundamental risk, noise trader risk, or holding costs, our e¤ect is

asymmetric. Trading in either direction impounds information into prices, which improves the

manager’s decision-making and increases fundamental value. This increases the pro…tability of

a long position but reduces the pro…tability of a short position, thus encouraging buying on

good news but discouraging selling on bad news. Second, unlike limits to arbitrage based on

short-sale constraints, holding costs or portfolio delegation, our model does not rely on exoge-

nous forces or agency problems; instead, the e¤ect is generated endogenously as part of the

arbitrage process.

The asymmetry of our e¤ect has implications for both stock returns and real investment.

In terms of stock returns, bad news has a smaller e¤ect on short-run prices than good news,

even though the market maker is rational and takes the speculator’s trading strategy into

account when devising his pricing function. Interestingly, in contrast to underreaction models,

the smaller short-run reaction to bad news may also coincide with smaller long-run drift, since

the manager can take a corrective action to attenuate the negative e¤ect of the state on …rm

32

value. In addition, the returns to a good investment are more front-loaded than the returns

to a bad investment – since the speculator does not trade on negative information, the value-

destructiveness of a bad investment seeps out ex post. In terms of real investment, the manager

may overinvest in negative-NPV projects, even though there are no agency problems and he

is attempting to learn from the market to take the e¢cient decision. Even though there is

an agent in the economy who knows with certainty that the investment is undesirable, and

the manager is aware of the speculator’s asymmetric trading strategy, this information is not

conveyed to the manager and so the project is not abandoned.

33

References

[1] Aghion, Philippe and Jeremy C. Stein (2008): “Growth versus Margins: Destabilizing

Consequences of Giving the Stock Market What It Wants.” Journal of Finance 63, 1025–

1058.

[2] Agrawal, Anup, Je¤rey F. Ja¤e, and Gershon N. Mandelker (1992): “The Post-Merger

Performance of Acquiring Firms: A Re-Examination of an Anomaly.” Journal of Finance

47, 1605–1621.

[3] Andrade, Gregor, Mark Mitchell, and Erik Sta¤ord (2001): “New Evidence and Perspec-

tives on Mergers.” Journal of Economic Perspectives 15, 103–120.

[4] Barberis, Nicholas and Richard Thaler (2003): “A Survey of Behavioral Finance,” in

George Constantinides, Milton Harris and René Stulz (eds), Handbook of Economics of

Finance (Elsevier/North-Holland, New York).

[5] Bond, Philip, Itay Goldstein, and Edward S. Prescott (2010): “Market-Based Corrective

Actions.” Review of Financial Studies 23, 781–820.

[6] Campbell, John and Albert S. Kyle (1993): “Smart Money, Noise Trading and Stock Price

Behavior.” Review of Economic Studies 60, 1–34.

[7] Chakraborty, Archishman and Bilge Yilmaz (2004): “Manipulation in Market Order Mod-

els.” Journal of Financial Markets 7, 187–206.

[8] Chen, Qi, Itay Goldstein, and Wei Jiang (2007): “Price Informativeness and Investment

Sensitivity to Stock Price.” Review of Financial Studies 20, 619–650.

[9] Cho, In-Koo and David M. Kreps (1987): “Signaling Games and Stable Equilibria.” Quar-

terly Journal of Economics 102, 179–221

[10] DeLong, Bradford, Andrei Shleifer, Lawrence Summers, and Robert Waldmann (1990):

“Noise Trader Risk in Financial Markets.” Journal of Political Economy 98, 703–738.

[11] Dow, James, Itay Goldstein, and Alexander Guembel (2010): “Incentives for Information

Production in Markets where Prices A¤ect Real Investment.” Working paper, University

of Pennsylvania.

[12] Edmans, Alex, Itay Goldstein, and Wei Jiang (2011): “The Real E¤ects of Financial

Markets: The Impact of Prices on Takeovers.” Journal of Finance, forthcoming.

[13] Geczy, Christopher, David K. Musto, and Adam V. Reed (2002), “Stocks Are Special Too:

An Analysis of the Equity Lending Market.” Journal of Financial Economics 66, 241–269.

34

[14] Goldstein, Itay, and Alexander Guembel (2008): “Manipulation and the Allocational Role

of Prices.” Review of Economic Studies 75, 133–64.

[15] Goldstein, Itay, Emre Ozdenoren, and Kathy Yuan (2011): “Learning and Complementar-

ities in Speculative Attacks.” Review of Economic Studies, forthcoming.

[16] Hong, Harrison, Terence Lim and Jeremy C. Stein (2000): “Bad News Travels Slowly: Size,

Analyst Coverage and the Pro…tability of Momentum Strategies.” Journal of Finance 55,

265–296.

[17] Hong, Harrison and Jeremy C. Stein (1999): “A Uni…ed Theory of Underreaction, Mo-

mentum Trading and Overreaction in Asset Markets.” Journal of Finance 54, 2143–2184.

[18] Jensen, Michael C. (1986): “The Agency Costs of Free Cash Flow: Corporate Finance and

Takeovers.” American Economic Review 76, 323–329.

[19] Karpo¤, Jonathan. M. (1987), “A Relation between Price Changes and Trading Volume:

A Survey.” Journal of Financial and Quantitative Analysis 22, 109–126.

[20] Kondor, Peter (2009): “Risk in Dynamic Arbitrage: The Price E¤ects of Convergence

Trading.” Journal of Finance 64, 631–655.

[21] Kyle, Albert S. (1985): “Continuous Auctions and Insider Trading.” Econometrica 53,

1315–1336.

[22] Lamont, Owen, and Richard Thaler (2003): “Can the Market Add and Subtract?: Mis-

pricing in Tech-Stock Carve-Outs.” Journal of Political Economy 111, 227–268.

[23] Luo, Yuanzhi (2005): “Do Insiders Learn from Outsiders? Evidence from Mergers and

Acquisitions.” Journal of Finance 60, 1951–1982

[24] Mitchell, Mark, and Todd Pulvino (2001): “Characteristics of Risk and Return in Risk

Arbitrage.” Journal of Finance 56, 2135–2176.

[25] Mitchell, Mark, Todd Pulvino, and Erik Sta¤ord (2002): “Limited Arbitrage in Equity

Markets.” Journal of Finance 57, 551–584.

[26] Myers, Stewart C. and Nicolas Majluf (1984): “Corporate Financing and Investment De-

cisions When Firms Have Information That Investors Do Not Have.” Journal of Financial

Economics 13, 187–221

[27] Ponti¤, Je¤rey (1996): “Costly Arbitrage: Evidence from Close-End Funds.” Quarterly

Journal of Economics 111, 1135–1151.

35

[28] Rau, P. Raghavendra and Theo Vermaelen (1998): “Glamour, Value and the Post-

Acquisition Performance of Acquiring Firms.” Journal of Financial Economics 49, 223–

253.

[29] Shleifer, Andrei and Robert W. Vishny (1997): “The Limits of Arbitrage.” Journal of

Finance 52, 35–55.

[30] Stulz, René M. (1990): “Managerial Discretion and Optimal Financing Policies.” Journal

of Financial Economics 26, 3–26.

[31] Subrahmanyam, Avanidhar and Sheridan Titman (2001): “Feedback From Stock Prices to

Cash Flows.” Journal of Finance 56, 2389–2413.

[32] Zwiebel, Je¤rey (1996): “Dynamic Capital Structure Under Managerial Entrenchment.”

American Economic Review 86, 1197–1215.

36

A Proofs

Proof of Proposition 1

This proof only provides supplementary material to what is in the text.

No Trade Equilibrium NT. The order ‡ows of A = ÷2 and A = 2 are o¤ the equilibrium

path and the posteriors are given by 0 and 1, respectively, as these are the only posteriors that

satisfy the Intuitive Criterion (as stated in the main proof). The order ‡ows of A ¸ ¦÷1. 0. 1¦

are observed on the equilibrium path and so the posteriors can be calculated by Bayes’ rule:

¡(A) = Pr(H[A)

=

Pr(A[H)

Pr(A[H) + Pr(A[1)

.

We thus have:

¡(÷1) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

.

and ¡ (0) and ¡ (1) are calculated in exactly the same way. Sequential rationality leads to the

decisions d and prices j as given by the Table.

We now turn to calculating trading pro…ts. If the positively-informed speculator chooses to

deviate from not trading to buying:

« With probability (w.p.)

1

3

, A = 2 and she is fully revealed. Thus, trading pro…ts are zero.

« W.p.

1

3

, A = 1 and she pays

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each share

is 1

i

H

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

« W.p.

1

3

, A = 0 and she pays

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each share

is 1

i

H

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

Thus, her expected gross pro…t is given by:

1

3

1

2

_

1

i

H

÷1

i

L

_

+

1

3

1

2

_

1

i

H

÷1

i

L

_

=

1

3

_

1

i

H

÷1

i

L

_

= i

3

. (23)

A similar calculation shows that, if a negatively-informed speculator sells, her gross pro…t

is also given by (23). Thus, if and only if i _ i

3

, the no-trade equilibrium is sustainable.

Partial Trade Equilibrium SNB. The order ‡ow of A = 2 is o¤ the equilibrium path and

37

the posterior is given by 1. The posteriors of the other order ‡ows are given as follows:

¡(÷2) =

0

`(1,3)

= 0

¡ (÷1) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

¡ (0) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

¡(1) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + (1 ÷`)(1,3)

=

1

2 ÷`

.

Under this equilibrium, the negatively-informed speculator sells.

« W.p.

1

3

, A = ÷2 and she is fully revealed. Thus, trading pro…ts are zero.

« W.p.

1

3

, A = ÷1 and she receives

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each

share is 1

i

L

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

« W.p.

1

3

, A = 0 and she receives

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each

share is 1

i

L

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

Thus, her expected gross pro…t is given by:

1

3

_

1

i

H

÷1

i

L

_

= i

3

.

If the positively-informed speculator deviates to buying:

« W.p.

1

3

, A = 2 and she is fully revealed. Thus, trading pro…ts are zero.

« W.p.

1

3

, A = 1 and she pays

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each

share is 1

i

H

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

« W.p.

1

3

, A = 0 and she pays

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each share

is 1

i

H

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

Thus, her expected gross pro…t is given by:

1

3

_

1

2

+

1 ÷`

2 ÷`

_

_

1

i

H

÷1

i

L

_

= i

2

.

Thus, the SNB equilibrium is sustainable if and only if i

2

_ i < i

3

.

Partial Trade Equilibrium BNS. The order ‡ow of A = ÷2 is o¤ the equilibrium path and

38

the posterior is given by 0. The posteriors of the other order ‡ows are given as follows:

¡ (÷1) =

(1 ÷`)(1,3)

(1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1 ÷`

2 ÷`

¡ (0) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

¡(1) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

¡(2) =

`(1,3)

`(1,3)

= 1.

There are two sub-cases to consider. In the sub-case of no feedback (

1

2

¸), decision

d = i is taken for all of the order ‡ows on the equilibrium path. Thus, analogous to the SNB

equilibrium, informed trading (in this case, buying on good information) yields pro…ts of i

3

;

if the negatively-informed speculator deviates to selling, she earns pro…ts of i

2

. Hence, this

equilibrium is sustainable if and only if i

2

_ i < i

3

. For the sub-case of feedback (

1

2

< ¸),

the manager now takes decision d = : upon observing order ‡ow A = ÷1. The pro…ts from

trading on positive information are unchanged. The pro…ts from deviating to selling on negative

information are now given as follows:

« W.p.

1

3

, A = ÷2 and she is fully revealed. Thus, trading pro…ts are zero.

« W.p.

1

3

, A = ÷1 and she receives

1

2

1

n

H

+

1

2

1

n

L

per share. The fundamental value

of each share is 1

n

L

because the manager is now taking the corrective action, and so her

pro…t is

1

2

(1

n

H

÷1

n

L

) 0.

« W.p.

1

3

, A = 0 and she receives

1

2

1

i

H

+

1

2

1

i

L

per share. The fundamental value of each

share is 1

i

L

, and so her pro…t is

1

2

(1

i

H

÷1

i

L

) 0.

Thus, her expected gross pro…t is given by:

1

3

_

1

2

_

1

i

H

÷1

i

L

_

+

1 ÷`

2 ÷`

(1

n

H

÷1

n

L

)

_

= i

1

.

Thus, this equilibrium is sustainable if and only if i

1

_ i < i

3

.

Trade Equilibrium T. All order ‡ows are on the equilibrium path and so the posteriors are

39

given as follows:

¡(÷2) =

0

`(1,3)

= 0

¡ (÷1) =

(1 ÷`)(1,3)

(1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1 ÷`

2 ÷`

¡ (0) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + `(1,3) + (1 ÷`)(1,3)

=

1

2

¡(1) =

`(1,3) + (1 ÷`)(1,3)

`(1,3) + (1 ÷`)(1,3) + (1 ÷`)(1,3)

=

1

2 ÷`

¡(2) =

`(1,3)

`(1,3)

= 1.

The pro…ts from buying on positive information are given by i

2

, as in the o`1 equilibrium.

For the pro…ts from selling on negative information, there are two sub-cases to consider, which

correspond to the two sub-cases in the 1`o equilibrium. Without feedback (

1

2

¸), the

pro…ts are given by i

2

and so the equilibrium is sustainable if and only if i < i

2

. With feedback

(

1

2

¸), the pro…ts are given by i

1

and so the equilibrium is sustainable if and only if i < i

1

.

Proof of Proposition 3

This proof only provides supplementary material to what is in the text. As discussed in the

main text, it is straightforward to show that the negatively-informed speculator will not deviate

to buying. Here we calculate the pro…ts made if the positively-informed speculator deviates to

selling, to derive the necessary conditions to prevent such a deviation.

No Trade Equilibrium NT.

If the positively-informed speculator deviates to selling:

« W.p.

1

3

, A = ÷2 and she receives 1

n

L

for a share that is worth 1

n

H

, which yields a pro…t

of (1

n

L

÷1

n

H

) 0.

« W.p.

1

3

, A = ÷1 and she receives

1

2

1

i

H

+

1

2

1

i

L

for a share that is worth 1

i

H

, which yields

a pro…t of

1

2

(1

i

L

÷1

i

H

) < 0.

« W.p.

1

3

, A = 0 and she receives

1

2

1

i

H

+

1

2

1

i

L

for a share that is worth 1

i

H

, which yields a

pro…t of

1

2

(1

i

L

÷1

i

H

) < 0.

Thus, her overall pro…ts are given by

1

3

(1

n

L

÷1

n

H

) +

1

3

_

1

i

L

÷1

i

H

_

.

For the positively-informed speculator not to deviate to selling, we require

1

3

_

÷

_

1

i

H

÷1

i

L

_

+ (1

n

L

÷1

n

H

)

_

< i.

40

The calculations for the Partial Trade Equilibrium o`1 are identical.

Partial Trade Equilibrium BNS.

If the positively-informed speculator deviates to selling:

« W.p.

1

3

, A = ÷2 and she receives 1

n

L

for a share that is worth 1

n

H

, which yields a pro…t

of (1

n

L

÷1

n

H

) 0.

« W.p.

1

3

, A = ÷1. In the case of feedback (

1

2

< ¸ ), she receives

1

2

1

n

H

+

1

2

1

n

L

for

a share that is worth 1

n

H

, which yields a pro…t of

1

2

(1

n

L

÷1

n

H

) 0. In the case of no

feedback (

1

2

¸ ), she receives

1

2

1

i

H

+

1

2

1

i

L

for a share that is worth 1

i

H

, which

yields a pro…t of

1

2

(1

i

L

÷1

i

H

) < 0.

« W.p.

1

3

, A = 0 and she receives

1

2

1

i

H

+

1

2

1

i

L

for a share that is worth 1

i

H

, which yields a

pro…t of

1

2

(1

i

L

÷1

i

H

) < 0.

In the case of feedback, her overall pro…ts are given by

1

3

(1

n

L

÷1

n

H

) +

1

3

1

2 ÷`

(1

n

L

÷1

n

H

) +

1

3

_

1

2

_

1

i

L

÷1

i

H

_

_

=

1

3

_

3 ÷`

2 ÷`

(1

n

L

÷1

n

H

)

_

÷

1

3

_

1

2

_

1

i

H

÷1

i

L

_

_

.

For the positively-informed speculator to choose buying over selling, her pro…ts must be

greater under the former. This requires:

1

3

_

1

i

H

÷1

i

L

_

1

3

_

3 ÷`

2 ÷`

(1

n

L

÷1

n

H

)

_

÷

1

3

_

1

2

_

1

i

H

÷1

i

L

_

_

3

2

_

1

i

H

÷1

i

L

_

3 ÷`

2 ÷`

(1

n

L

÷1

n

H

) .

The …rst term is the “fundamental” e¤ect, which represents the pro…ts from trading in the

direction of one’s private information. The second term is the “feedback” e¤ect, which arises

because selling manipulates the order ‡ow and causes the manager to take the wrong decision.

This yields the condition

63

62

(1

i

H

÷1

i

L

) 1 in the Proposition.

In the case of no feedback, her overall pro…ts are given by

1

3

(1

n

L

÷1

n

H

) +

1

3

1

2 ÷`

_

1

i

L

÷1

i

H

_

+

1

3

_

1

2

_

1

i

L

÷1

i

H

_

_

=

1

3

(1

n

L

÷1

n

H

) ÷

1

3

_

4 ÷`

4 ÷2`

_

1

i

H

÷1

i

L

_

_

.

41

For the positively-informed speculator to choose buying over selling, we require:

1

3

_

1

i

H

÷1

i

L

_

1

3

(1

n

L

÷1

n

H

) ÷

1

3

_

4 ÷`

4 ÷2`

_

1

i

H

÷1

i

L

_

_

8 ÷3`

4 ÷2`

1

i

H

÷1

i

L

1

n

L

÷1

n

H

1.

Trade Equilibrium T.

If the positively-informed speculator deviates to selling, the calculations are exactly the

same as in the Partial Trade Equilibrium 1`o. However, the pro…ts from buying (that we

need to compare against the pro…ts from selling) are di¤erent.

In the case of feedback, for the positively-informed speculator to choose buying over selling,

we require:

1

3

_

1

2

+

1 ÷`

2 ÷`

_

_

1

i

H

÷1

i

L

_

1

3

_

3 ÷`

2 ÷`

(1

n

L

÷1

n

H

)

_

÷

1

3

_

1

2

_

1

i

H

÷1

i

L

_

_

3 ÷2`

3 ÷`

1

i

H

÷1

i

L

1

n

L

÷1

n

H

1.

In the case of no feedback, for the positively-informed speculator to choose buying over

selling, we require:

1

3

_

1

2

+

1 ÷`

2 ÷`

_

_

1

i

H

÷1

i

L

_

1

3

(1

n

L

÷1

n

H

) ÷

1

3

_

4 ÷`

4 ÷2`

_

1

i

H

÷1

i

L

_

_

2

1

i

H

÷1

i

L

1

n

L

÷1

n

H

1.

Proof of Lemma 3

For part (i), if o = H, the expected posterior is given by:

¡

H

= (1 ÷`)

_

1

3

¡ (÷1) +

1

3

¡ (0) +

1

3

¡ (1)

_

+ `

_

1

3

¡ (0) +

1

3

¡ (1) +

1

3

¡ (2)

_

=

1 ÷`

3

¡ (÷1) +

1

3

¡ (0) +

1

3

¡ (1) +

`

3

¡ (2)

=

(1 ÷`)

2

6 ÷3`

+

1

3

+

`

3

. (24)

42

We have:

J¡

H

J`

=

1

3

+

1

3

_

÷2(1 ÷`)(2 ÷`) + (1 ÷`)

2

(2 ÷`)

2

_

=

1

3

_

1 +

_

1 ÷`

2 ÷`

_

2

÷2

1 ÷`

2 ÷`

_

=

1

3

_

1 ÷

_

1 ÷`

2 ÷`

__

2

0.

The expected posterior is increasing in `: if the speculator is more likely to be present, she

is more likely to impound her information into prices by trading.

Moving to part (ii), if o = 1, we have:

¡

L

=

1

3

(¡ (÷1) + ¡ (0) + ¡ (1))

=

1 ÷`

6 ÷3`

+

1

3

. (25)

This is decreasing in `. Even though the speculator does not trade upon o = 1 if she is present,

her information is still partially incorporated into prices. With o = 1, there is a

1

3

probability

that the order ‡ow is A = ÷1. This is consistent with the speculator being absent (in which

case the state may be either H or 1) or her being present and observing o = 1; it is not

consistent with the speculator observing o = H. The greater the likelihood that the speculator

is present, the greater the likelihood that A = ÷1 stems from o = 1, and thus the greater the

decrease in the market maker’s posterior. Part (iii) follows from simple calculations.

Proof of Proposition 5

For parts (i) and (ii), we have:

¡

H;spec

=

1

3

(¡ (0) + ¡ (1) + ¡ (2))

=

2

3

. (26)

¡

L;spec

=

1

3

(¡ (÷1) + ¡ (0) + ¡ (1))

=

1 ÷`

6 ÷3`

+

1

3

. (27)

Note that ¡

H;spec

is independent of `, but ¡

L;spec

is decreasing in `. The variable ` can a¤ect

the expected posterior in two ways: …rst, it can change the relative likelihood of the di¤erent

order ‡ows, and second, it can change the actual posterior given a certain order ‡ow. Since we

are conditioning on the speculator being present, the …rst channel is ruled out: conditional on

the speculator being present and o = H, A ¸ ¦0. 1. 2¦ with uniform probability regardless of `;

43

conditional on the speculator being present and o = 1, A ¸ ¦÷1. 0. 1¦ with uniform probability

regardless of `. Turning to the second channel, the only posterior that depends on ` is ¡ (÷1):

since A = ÷1 is inconsistent with the speculator being present and seeing o = H, it has a

particularly negative impact on the likelihood of o = H if the speculator is more likely to be

present. By contrast, A ¸ ¦÷2. 2¦ is fully revealing and so the posterior is independent of `;

A ¸ ¦0. 1¦ is completely uninformative and so the posterior is again independent of `. Since

A = ÷1 can only occur in the presence of a speculator if she has received bad news, only ¡

L;spec

depends on ` but ¡

L;spec

does not. Part (iii) follows from simple calculations.

Proof of Lemma 4

We start by calculating j

0

. With probability

1

2

, the state will be o = 1 and there is no trade,

regardless of whether the speculator is present. Thus, A ¸ ¦÷1. 0. 1¦ with equal probability.

With probability

1

2

, the state will be o = H. If the speculator is absent (w.p. (1 ÷`)), there is

no trade and we again have A ¸ ¦÷1. 0. 1¦. If the speculator is present, A ¸ ¦0. 1. 2¦. Letting

j (A) denote the stock price set by the market maker after observing order ‡ow A at t = 1,

the price at t = 0 will be the expectation over all possible future prices at t = 1, and is given

as follows:

j

0

=

`

2

_

1

3

j (0) +

1

3

j (1) +

1

3

j (2)

_

+

_

1 ÷

`

2

__

1

3

j (÷1) +

1

3

j (0) +

1

3

j (1)

_

=

1

3

__

1 ÷

`

2

_

j (÷1) + j (0) + j (1) +

`

2

j (2)

_

=

1

6

_

(1 ÷`)1

n

H

+ 1

n

L

+ (2 + `)1

i

H

+ 21

i

L

¸

. (28)

Even though the initial belief ¸ is independent of `, the initial stock price j

0

is increasing in

`, because the speculator provides information to improve the manager’s decision. Moreover,

@p

0

@

is increasing in (1

i

H

÷1

n

H

), the increase in …rm value from taking the e¢cient continuation

decision in the high state. This is intuitive. If the speculator is present, she always buys in

state H, which guarantees that A _ 0 and the investment is undertaken. If she is absent, there

is a possibility that A = ÷1. This leads the manager to take the suboptimal corrective action,

reducing …rm value by (1

i

H

÷1

n

H

). By contrast,

@p

0

@

is independent of 1

i

L

and 1

n

L

, the …rm

values in the low state. This is because, if the low state is realized, the speculator’s presence

does nothing to help the manager’s decision, since she does not trade.

For part (i), if o = H is realized, the expected price at t = 1 is given by:

j

H

1

= (1 ÷`)

_

1

3

j (÷1) +

1

3

j (0) +

1

3

j (1)

_

+ `

_

1

3

j (0) +

1

3

j (1) +

1

3

j (2)

_

=

1 ÷`

3

j (÷1) +

1

3

j (0) +

1

3

j (1) +

`

3

j (2) . (29)

44

Note that:

Jj

H

1

J`

=

1

3

j(2) ÷

1

3

j(÷1) +

1 ÷`

3

Jj(÷1)

J`

=

1

3

_

1

i

H

÷

1

(2 ÷`)

2

1

n

L

÷

_

1 ÷

1

(2 ÷`)

2

_

1

n

H

_

1

3

[1

i

H

÷1

n

L

]

0.

i.e. j

H

1

is increasing in `, since the speculator impounds information about the high state into

prices.

Turning to part (ii), if o = 1is realized, the expected price at t = 1 is given by:

j

L

1

=

1

3

(j (÷1) + j (0) + j (1)) . (30)

We have

@p

L

1

@

=

R

n

L

R

n

H

3(2)

2

. If the speculator is more likely to be present, then A = ÷1 is more

likely to result from o = 1. Thus, the price is higher if and only if …rm value is higher in this

state, i.e., 1

n

L

1

n

H

(Case 2).

The calculations of j

H

1

÷j

0

and j

L

1

÷j

0

follow automatically.

Proof of Proposition 6

For part (i), if the speculator receives positive information, she will buy one share and so

the expected price becomes:

j

H;spec

1

=

1

3

(j (0) + j (1) + j (2)) . (31)

Unlike j

H

1

(equation (29)), this is independent of `, for the same reasons that ¡

H;spec

(equa-

tion (26)) is independent of `. The stock return realized when the speculator receives good

information is thus given by:

j

H;spec

1

÷j

0

=

1

3

(j (0) + j (1) + j (2)) ÷

1

3

__

1 ÷

`

2

_

j (÷1) + j (0) + j (1) +

`

2

j (2)

_

=

1

3

_

1 ÷

`

2

_

(j (2) ÷j (÷1))

=

1

3

_

1 ÷

`

2

__

1

i

H

÷

1 ÷`

2 ÷`

1

n

H

÷

1

2 ÷`

1

n

L

_

0, (32)

and we have

J

_

j

H;spec

1

÷j

0

_

J`

=

1

6

[1

n

H

÷1

i

H

] < 0.

Equation (32) is decreasing in `, whereas the stock return not conditioning on the specula-

45

tor’s presence, (32), was increasing in `. This is because j

0

is increasing in `, but j

H;spec

1

is

independent of `.

For part (ii), if the speculator is present and receives negative information, we have:

j

L;spec

1

=

1

3

(j (÷1) + j (0) + j (1)) = j

L

1

. (33)

and

j

L;spec

1

÷j

0

=

1

3

(j (÷1) + j (0) + j (1)) ÷

1

3

__

1 ÷

`

2

_

j (÷1) + j (0) + j (1) +

`

2

j (2)

_

=

`

6

(j (÷1) ÷j (2)) = j

L

1

÷j

0

< 0.

Parts (iii) and (iv) follow from simple calculations.

Dropping constants, both equation (16) (the asymmetry between the price impact of good

and bad news) and equation (17) (the average return, conditional on the speculator being

present) become:

(1 ÷`)

_

1

i

H

÷

1 ÷`

2 ÷`

1

n

H

÷

1

2 ÷`

1

n

L

_

.

Di¤erentiating with respect to ` gives:

÷1

i

H

+

1 ÷`

2 ÷`

1

n

H

+

1

2 ÷`

1

n

L

+ (1 ÷`)

_

1

(2 ÷`)

2

1

n

H

÷

1

(2 ÷`)

2

1

n

L

_

= ÷1

i

H

+

3 ÷4` + `

2

(2 ÷`)

2

1

n

H

+

1

(2 ÷`)

2

1

n

L

The coe¢cients of 1

n

H

and 1

n

L

are positive and add up to one. That is, we have a convex

combination of 1

n

H

and 1

n

L

, which is smaller that 1

i

H

, since 1

i

H

1

n

H

and 1

i

H

1

n

L

. Thus,

both equations (16) and (17) are decreasing in `.

Proof of Corollary 2

We start with part (i). If the speculator receives good news, she will buy and so the project

will always be undertaken. We thus have j

H;spec

2

= 1

i

H

. This yields:

j

H;spec

2

÷j

H;spec

1

= 1

i

H

÷

1

3

(j (0) + j (1) + j (2))

=

1

3

_

1

i

H

÷1

i

L

_

.

Moving to part (ii), if the speculator receives bad news, she will not trade. The project will

be cancelled if the noise trader sells, else it will be continued. We thus have j

L;spec

2

=

2

3

1

i

L

+

1

3

1

n

L

.

46

This yields:

j

L;spec

2

÷j

L;spec

1

=

2

3

1

i

L

+

1

3

1

n

L

÷

1

3

(j (÷1) + j (0) + j (1))

= ÷

1

3

_

1

i

H

÷1

i

L

_

÷

1

3

_

1 ÷`

2 ÷`

(1

n

H

÷1

n

L

)

_

.

which can be positive or negative. Part (iii) follows from simple calculations. For part (iv), we

…rst calculate the expected …rm value at t = 2 if the speculator is present, not conditioning on

the state. If o = H, the project is always undertaken, regardless of the order ‡ow at t = 1,

and so …rm value · = 1

i

H

. If o = 1, whether the project is undertaken depends on the order

‡ow: if A = ÷1, we have d = : and so · = 1

n

L

; if A ¸ ¦0. 1¦, we have d = i and so · = 1

i

L

.

Expected …rm value at t = 2 is thus given by:

j

spec

2

=

1

2

1

i

H

+

1

3

1

i

L

+

1

6

1

n

L

.

and so we have

j

spec

2

÷j

spec

1

=

1

6

1 ÷`

2 ÷`

(1

n

L

÷1

n

H

) ,

which is positive if we are in Case 2 and negative if we are in Case 1.

Proof of Proposition 7

We start with some preliminary results that will be useful in the main proof. From

1

2

< ¸

we have

` 1 ÷

¸

1 ÷¸

,

which implies that ` ¸ (1 ÷

1

. 1). Let `

min

= 1 ÷

1

and `

max

= 1.

De…ne [

~

1

i

H

~

1

i

L

~

1

n

L

] = [1

i

H

÷1

n

H

1

i

L

÷1

n

H

1

n

L

÷1

n

H

]. Note that

~

1

i

H

0 from equation (1).

Equation (5) thus yields:

~

1

n

L

= (1 ÷`

min

)

~

1

i

H

+

~

1

i

L

. (34)

For part (i), we …rst calculate j

i;H

1

, the expected stock price at t = 1 if investment has been

undertaken and the state is good. We have o = H w.p.

1

2

. W.p. `, the speculator is present

and buys, so A ¸ ¦0. 1. 2¦ with uniform probability and the investment is always undertaken.

W.p. 1 ÷ ` the speculator is absent, so there is no trade, which yields A ¸ ¦÷1. 0. 1¦. If

A = ÷1, we have d = : so we exclude this case. We therefore have:

j

i;H

1

=

2

1

3

(j(0) + j(1) + j(2)) +

1

2

1

3

(j(0) + j(1))

1

2

_

` + (1 ÷`)

2

3

_

=

j(0) + j(1) + `j(2)

2 + `

.

Simple calculations show that j

i;H

1

is increasing in `: if the speculator is present, she will trade

on her positive signal and impound it into prices. Note that this argument did not apply to

47

¡

H;spec

(equation (26)) and j

H;spec

1

(equation (31)) because those quantities are conditional on

the speculator being present, so A = ¦0. 1. 2¦ were all equally likely. Here, A = 2 is particularly

likely if the speculator is present.

For part (ia), the short-run return to an investment (d = i) in the high state is given by:

j

i;H

1

÷j

0

=

`j(2) + j(1) + j(0)

2 + `

÷

1

3

__

1 ÷

`

2

_

j (÷1) + j (0) + j (1) +

`

2

j (2)

_

=

4` ÷`

2

6(2 + `)

j(2) +

1 ÷`

3(2 + `)

j(1) +

1 ÷`

3(2 + `)

j(0) ÷

_

1

3

÷

`

6

_

j(÷1)

=

1

6(2 + `)

[(2 + 2` ÷`

2

)1

i

H

+ (2 ÷2`)1

i

L

÷(2 ÷` ÷`

2

)1

n

H

÷(2 + `)1

n

L

]. (35)

The sign of (35) is the same as the sign of

(2 + 2` ÷`

2

)

_

1

i

H

÷1

n

H

_

+ (2 ÷2`)

_

1

i

L

÷1

n

H

_

÷(2 ÷` ÷`

2

) (1

n

H

÷1

n

H

) ÷(2 + `) (1

n

L

÷1

n

H

)

= (2 + 2` ÷`

2

)

~

1

i

H

+ (2 ÷2`)

~

1

i

L

÷(2 + `)

~

1

n

L

. (36)

Equation (36) is quadratic and concave in `, and so its minimum occurs at either `

min

or

`

max

. Thus, to prove that (36) 0, it is su¢cient to prove that it is positive at both `

min

and `

max

. At ` = `

max

, equation (36) reduces to 3

~

1

i

H

÷ 3

~

1

n

L

0. At ` = `

min

, it reduces to

3`

min

(

~

1

i

H

÷

~

1

i

L

) 0. Thus, equation (35) is positive.

For part (ib), the long-run drift is given by:

1

i

H

÷j

i;H

1

= 1

i

H

÷

`j(2) + j(1) + j(0)

2 + `

=

1

i

H

÷1

i

L

2 + `

0. (37)

For part (ii), we …rst calculate j

i;L

1

, the expected stock price at t = 1 if investment has been

undertaken and the state is bad. Regardless of whether the speculator is present, the order

‡ow will be A ¸ ¦÷1. 0. 1¦ with uniform probability. If A = ÷1, we have d = : so we exclude

this case. We therefore have:

j

i;L

1

=

1

2

_

1

3

j (0) +

1

3

j (1)

_

1

2

2

3

=

1

2

(j(0) + j(1)) .

This is independent of ` since the presence of the speculator does not change the order ‡ow.

Unlike in the earlier cases of j

L

1

and j

L;spec

1

which did depend on `, here we are conditioning

upon the investment being undertaken. This rules out the case of j (÷1) which is the only price

that depends on `.

48

For part (iia), the short-run return to an investment in the low state is:

j

i;L

1

÷j

0

=

1

2

(j(0) + j(1)) ÷

_

`

6

j(2) +

1

3

j(1) +

1

3

j(0) +

_

1

3

÷

`

6

_

j(÷1)

_

= ÷

`

6

j(2) +

1

6

j(1) +

1

6

j(0) ÷(

1

3

÷

`

6

)j(÷1)

=

1

6

[(1 ÷`)

_

1

i

H

÷1

n

H

_

+ 1

i

L

÷1

n

L

]. (38)

The sign of (38) is the same as the sign of:

(1 ÷`)

_

1

i

H

÷1

n

H

÷(1

n

H

÷1

n

H

)

_

+ 1

i

L

÷1

n

H

÷(1

n

L

÷1

n

H

)

= (1 ÷`)

~

1

i

H

+

~

1

i

L

÷

~

1

n

L

. (39)

Equation (39) is decreasing in `, since

~

1

i

H

0. This implies that

(1 ÷`)

~

1

i

H

+

~

1

i

L

÷

~

1

n

L

< (1 ÷`

min

)

~

1

i

H

+

~

1

i

L

÷

~

1

n

L

= 0.

from equation (34). Thus, equation (38) is negative.

Since j

i;L

1

is independent of `, and j

0

is increasing in ` (because the speculator improves

the manager’s decisions), the absolute return is decreasing in `.

For part (iib), the long-run drift is given by:

1

i

L

÷j

i;L

1

= 1

i

L

÷

1

2

(j(0) + j(1))

= ÷

1

2

(1

i

H

÷1

i

L

) < 0 (40)

and independent of `, since j

i;L

1

is independent of `.

For part (iii), we …rst calculate the di¤erence between the long-run drift and the short-run

return to a good investment, i.e. (37)-(35). This yields:

1

2 + `

(

~

1

i

H

÷

~

1

i

L

) ÷

1

6(2 + `)

[(2 + 2` ÷`

2

)

~

1

i

H

+ (2 ÷2`)

~

1

i

L

÷(2 + `)

~

1

n

L

]

=

1

6(2 + `)

[(4 ÷2` + `

2

)

~

1

i

H

÷(8 ÷2`)

~

1

i

L

+ (2 + `)

~

1

n

L

].

In order to calculate c/: ((37) ÷(35)), we must …rst sign (37)-(35). To prove this is positive,

we must prove that

(4 ÷2` + `

2

)

~

1

i

H

÷(8 ÷2`)

~

1

i

L

+ (2 + `)

~

1

n

L

(41)

is positive. If

~

1

i

L

_ 0, it is automatic that (41) 0. Suppose

~

1

i

L

0. Evaluating (41) at `

min

49

and `

max

, respectively, yields

` = `

min

:(4 ÷2`

min

+ `

2

min

)

~

1

i

H

÷(8 ÷2`

min

)

~

1

i

L

+ (2 + `

min

)[(1 ÷`

min

)

~

1

i

H

+

~

1

i

L

]

= (6 ÷3`

min

)

~

1

i

H

÷(6 ÷3`

min

)

~

1

i

L

0.

` = `

max

:3

~

1

i

H

÷6

~

1

i

L

+ 3

~

1

n

L

0.

If (41) is monotonic in ` in its feasible range, it follows that (41) 0. Suppose that (41) is not

monotonic in `. Then its derivative with respect to ` must have a root `

0

¸ (`

min

. `

max

). The

derivative of (41) is equal to

÷(2 ÷2`)

~

1

i

H

+ 2

~

1

i

L

+

~

1

n

L

.

Then `

0

= 1 ÷

2

~

R

i

L

+

~

R

n

L

2

~

R

i

H

. The condition `

0

`

min

, together with (34), yields

1 ÷

2

~

1

i

L

+

~

1

n

L

2

~

1

i

H

1 ÷

~

1

n

L

÷

~

1

i

L

~

1

i

H

=

2

~

1

i

L

+

~

1

n

L

2

~

1

i

H

<

~

1

n

L

÷

~

1

i

L

~

1

i

H

=4

~

1

i

L

<

~

1

n

L

.

which implies

(41) (4 ÷2` + `

2

)

~

1

i

H

÷(8 ÷2`)

~

1

i

L

+ 4(2 + `)

~

1

i

L

= (4 ÷2` + `

2

)

~

1

i

H

~

1

i

H

+ 6`

~

1

i

L

0 for

~

1

i

L

0.

We now calculate the di¤erence between the long-run drift and the short-run return to a

good investment, i.e. (40) ÷(38). This yields:

÷

1

2

(1

i

H

÷1

i

L

) ÷

1

6

[(1 ÷`)1

i

H

+ 1

i

L

÷1

n

L

]

= ÷

1

6

[(4 ÷`)1

i

H

÷21

i

L

÷1

n

L

].

In order to calculate c/: ((40) ÷(38)), we must …rst sign (40) ÷ (38). To prove this is

negative, we must prove that

(4 ÷`)

~

1

i

H

÷2

~

1

i

L

÷

~

1

n

L

(42)

is positive. At ` = `

max

, equation (42) becomes 3

~

1

i

H

÷2

~

1

i

L

÷

~

1

n

L

0. Since (42) is decreasing

in lambda, we have (42) 0.

Finally, we wish to showthat c/: ((40) ÷(38)) c/: ((37) ÷(35)). The sign of c/: ((40) ÷(38))÷

50

c/: ((37) ÷(35)) is equal to the sign of:

(2 + `)[(4 ÷`)

~

1

i

H

÷2

~

1

i

L

÷

~

1

n

L

] ÷[(4 ÷2` + `

2

)

~

1

i

H

÷(8 ÷2`)

~

1

i

L

+ (2 + `)

~

1

n

L

]

=[(8 + 2` ÷`

2

)

~

1

i

H

÷(4 + 2`)

~

1

i

L

÷(2 + `)

~

1

n

L

] ÷[(4 ÷2` + `

2

)

~

1

i

H

÷(8 ÷2`)

~

1

i

L

+ (2 + `)

~

1

n

L

]

=2(2 + 2` ÷`

2

)

~

1

i

H

+ 4(1 ÷`)

~

1

i

L

÷2(2 + `)

~

1

n

L

. (43)

(43) is quadratic and concave in `. Then its minimum occurs at either `

min

or `

max

. At

` = `

max

, (43) reduces to 6(

~

1

i

H

÷

~

1

n

L

) 0. For ` = `

min

,

(43) = 2(2 + 2`

min

÷`

2

min

)

~

1

i

H

+ 4(1 ÷`

min

)

~

1

i

L

÷2(2 + `

min

)[(1 ÷`

min

)

~

1

i

H

+

~

1

i

L

]

= 6`

min

(

~

1

i

H

÷

~

1

i

L

) 0.

Thus, (43) is always positive, and so c/: ((40) ÷(38)) ÷c/: ((37) (35)).

51

1

Introduction

Whether …nancial markets are informationally e¢ cient is one of the most hotly-contested debates in …nance. Proponents of market e¢ ciency argue that pro…t opportunities in the …nancial market will lead speculators to trade in a way that eliminates any mispricing. For example, if speculators have negative information about a stock, and this information is not re‡ ected in the price, they will …nd it pro…table to sell the stock. This will push down the price, causing it to re‡ speculators’information. However, a sizable literature identi…es various limits ect to arbitrage, which may deter speculators from trading on their information. (This notion of “arbitrage” is broader than the traditional textbook notion of risk-free arbitrage from trading two identical securities. Here, we use “arbitrage” to refer to investors trading on their private information.) For example, De Long, Shleifer, Summers, and Waldmann (1990) and Shleifer and Vishny (1997) show that the slow convergence of price to fundamental value may render arbitrage activities too risky. This in turn dissuades trading if the speculator has a short horizon, which may in turn arise from informational asymmetries with her own investors. Other explanations for limited arbitrage rely on market frictions such as short-sales constraints. All of these mechanisms treat the …rm’ fundamental value as exogenous to the arbitrage process s and rely on market imperfections to explain why speculators will not drive the price towards fundamental value. Thus, as …nancial markets develop, these limits to arbitrage may weaken. In this paper, we identify a quite di¤erent limit to arbitrage, which does not rely on exogenous forces but is instead generated endogenously as part of the arbitrage process. It stems from the fact that the value of the asset being arbitraged is endogenous to the act of exploiting the arbitrage. By trading, speculators cause prices to move, which in turn reveals information to real decision makers, such as managers, board members, corporate raiders, and regulators. These decision makers then take actions based on the information revealed in the price, and these actions change the underlying asset value. This may make the initial trading less pro…table, deterring it from occurring in the …rst place. To …x ideas, consider the following example. Suppose that a …rm (acquirer) announces the acquisition of a target. Also assume that some speculators conducted some analysis suggesting that this acquisition will be value-destructive. Traditional theory suggests that these speculators should sell the acquirer’ stock. However, large-scale selling will convey to the acquirer that s speculators believe the acquisition is a bad idea. As a result, the acquirer may end up cancelling the acquisition. In turn, cancellation of a bad acquisition will boost …rm value, reducing the speculator’ pro…t from her short position and in some cases causing her to su¤er a loss. Put s di¤erently, the acquirer’ decision to cancel the acquisition means that the negative information s possessed by speculators is now less relevant, and hence they should not trade on it. Thus, the information ends up not being re‡ ected in the price. Our mechanism is based on the presence of a feedback e¤ect from the …nancial market to real economic decisions – that real decision makers learn from the market when deciding their actions. A common perception is that managers know more about their own …rms than 2

Electronic copy available at: http://ssrn.com/abstract=1787732

outsiders (e.g. Myers and Majluf (1984)). While this is likely plausible for internal information about the …rm in isolation, optimal managerial decisions also depend on external information (such as market demand for a …rm’ products, or potential synergies with a target) about which s outsiders may be more informed. A classic example of how information from the stock market can shape real decisions is Coca-Cola’ attempted acquisition of Quaker Oats. On November s 20, 2000, the Wall Street Journal reported that Coca-Cola was in talks to acquire Quaker Oats. Shortly thereafter, Coca-Cola con…rmed such discussions. The market reacted negatively, sending Coca-Cola’ shares down 8% on November 20th and 2% on November 21st. Coca-Cola’ s s board rejected the acquisition later on November 21st, potentially due to the negative market reaction. The following day, Coca-Cola’ shares rebounded 8%. Thus, speculators who had s short-sold on the initial merger announcement, based on the belief that the acquisition would destroy value, lost money – precisely the e¤ect modeled by this paper. In the same context, Luo (2005) provides large-sample evidence that acquisitions are more likely to be cancelled if the market reacts negatively to them, and that the e¤ect is more pronounced when the acquirer is more likely to have something to learn from the market, e.g., for non-high-tech deals and where the bidder is small. Relatedly, Edmans, Goldstein, and Jiang (2011) demonstrate that a …rm’ market price a¤ects the likelihood that it becomes a takeover target, which may arise s because potential acquirers learn from the market price. More broadly, Chen, Goldstein, and Jiang (2007) show that the sensitivity of investment to price is higher when the price contains more private information not known to managers. Moreover, our model can apply to corrective actions (i.e., actions that improve …rm value upon learning negative information about …rm prospects) undertaken by stakeholders other than the manager. Such stakeholders likely have less information than the manager and may be more reliant on information held by outsiders. Examples include managerial replacement (undertaken by the board, or by shareholders who lobby the board), a disciplinary takeover (undertaken by an acquirer), or the granting of a subsidy or a bail-out (undertaken by the government). We demonstrate a barrier to the feedback e¤ect, that hinders decision makers from learning from the market. An important aspect of our theory is that it generates asymmetry between trading on positive and negative information. The feedback e¤ect delivers an equilibrium where speculators trade on good news but do not trade on bad news. Yet, it does not give rise to the opposite equilibrium, where speculators trade on bad news only. The intuition is as follows. When speculators trade on information, they improve the e¢ ciency of the …rm’ decisions –regardless s of the direction of their trade. If the speculator has positive information on a …rm’ prospects, s trading on it will reveal to the manager that investment is pro…table. This will in turn cause the …rm to invest more, thus increasing its value. If the speculator has negative information, trading on it will reveal to the manager that investment is unpro…table. This will in turn cause the …rm to invest less, also increasing its value as contraction is the correct decision. When a speculator buys and takes a long position in a …rm, she bene…ts further from increasing its value

3

via the feedback e¤ect. By contrast, when she sells and takes a short position, she loses from increasing the …rm’ value via the feedback e¤ect. Note that, for the speculator to lose from s the feedback e¤ect, she must end up with a short position. If she ends with a long position, the value of the shares she still holds onto are enhanced by the feedback e¤ect. Thus, the model implies that investors are less likely to engage in short-sales than sales –even though the model contains no short-sale constraints. Even though the speculator’ trading behavior is asymmetric, in general it is not automatic s that the impact on prices is asymmetric. The market maker is fully rational and takes into account the fact that the speculator buys on positive information and does not trade on negative information. Thus, he adjusts his pricing function accordingly. Therefore, it may seem that negative information will be impounded in prices to the same degree as positive information – even though it may lead to a neutral rather than negative order ‡ ow, the market maker knows that a neutral order ‡ can stem from the speculator having negative information but ow choosing not to trade, and may decrease the price accordingly. By contrast, we show that the asymmetry in trading behavior does translate into asymmetry in price impact. The crux is that the market maker cannot distinguish the case of a speculator who has negative information but chooses to withhold it, from the case in which the speculator is absent (i.e. there is no information). Thus, a neutral order ‡ does not lead to a large stock price decrease, and so ow negative information has a smaller e¤ect on prices. Indeed, Hong, Lim, and Stein (2000) show empirically that bad news is incorporated in prices more slowly than good news. They speculate that this arises because it is …rm management that possesses value-relevant information, and they will publicize it more enthusiastically for favorable than unfavorable information. Our paper presents a formal model that o¤ers an alternative explanation. Here, key information is held by a …rm’ investors rather than its managers, who “publicize”it not through public news s releases, but by trading on it. They also choose to disseminate good news more readily than bad news, but for a very di¤erent reason from …rm management, i.e., because of the feedback e¤ect. In standard models of underreaction, if bad news has a smaller e¤ect on short-run returns (i.e. between t = 0 and t = 1) than good news, this must be counterbalanced by bad news generating a larger long-run drift (between t = 1 and t = 2) than good news. We show that this need not be the case in a model with feedback. It is indeed true that, if the state is bad and little bad news comes out in the short-run (due to the speculator not trading on it), there is more bad news still to come out in the long-run. However, in a feedback model where …rm value is endogenous, the manager can take a corrective action to mitigate the negative impact of the state on …rm value. If the feedback e¤ect is su¢ ciently strong, bad news has a smaller e¤ect than good news in both the short-run and the long-run. While the above considers the returns to good and bad news, the model also generates predictions regarding the returns to good and bad investment decisions. Naturally, the returns to investment are positive (both in the short-run and long-run) if the state is good and negative

4

if the state is bad. More interestingly, we show that the returns to good investment are more front-loaded than the returns to a bad investment – i.e., a higher proportion of the returns manifests at t = 1 than at t = 2. This result again stems from the asymmetry of the speculator’ s trading strategy. Even if the speculator is aware that the investment is bad at t = 1, she may not trade on this information due to the feedback e¤ect. Thus, the value-destructiveness of the investment seeps out ex post at t = 2. Thus, our model provides an explanation for the negative long-run returns to M&A, documented by Agrawal, Ja¤e, and Mandelker (1992) and Rau and Vermaelen (1998). In addition to its interesting e¤ects on stock returns, the asymmetry of the speculator’ s trading strategy can also generate important real consequences. Since negative information is not incorporated into prices, it does not in‡ uence management decisions. Thus, while positiveNPV projects will be encouraged, some negative-NPV projects will not be canceled – even though there is an agent in the economy who knows with certainty that the project is negativeNPV –leading to overinvestment overall. In contrast to standard overinvestment theories based on the manager’ private bene…ts (e.g., Jensen (1986), Stulz (1990), Zwiebel (1996)), here the s manager is fully aligned with …rm value and there are no agency problems. The manager wishes to maximize …rm value by learning from prices, but is unable to do so since speculators refrain from revealing their information. Applied to M&A as well as organic investment, the theory may explain why M&A appears to be “excessive” and a large fraction of acquisitions destroy value (see, e.g., Andrade, Mitchell, and Sta¤ord (2001).) As mentioned above, the primary motivation for our paper is to identify a limit to arbitrage. Di¤erent authors have emphasized di¤erent factors that lead to limits on arbitrage activities. Campbell and Kyle (1993) focus on fundamental risk, i.e., the risk that …rm fundamentals will change while the arbitrage strategy is being pursued. In their model, such changes are unrelated to speculators’arbitrage activities. De Long, Shleifer, Summers, and Waldmann (1990) argue that noise-trading risk, i.e., the risk that noise trading will increase the degree of mispricing, may render arbitrage activities unpro…table. Noise trading only a¤ects the asset’ market price s and not its fundamental value, which is again exogenous to the act of arbitrage. Shleifer and Vishny (1997) show that, even if an arbitrage strategy is sure to converge in the long-run, the possibility that mispricing may widen in the short-term may deter speculators from trading on it, if they are concerned with redemptions by their own investors. Similarly, Kondor (2009) demonstrates that …nancially-constrained arbitrageurs may stay out of a trade if they believe that it may become more pro…table in the future. Many authors (e.g., Ponti¤ (1996), Mitchell and Pulvino (2001), and Mitchell, Pulvino, and Sta¤ord (2002)) focus on the transaction costs and holding costs that arbitrageurs have to incur while pursuing an arbitrage strategy. Others (Geczy, Musto, and Reed (2002), and Lamont and Thaler (2003)) discuss the importance of short-sales constraints. While these papers emphasize market frictions as the source of limits to arbitrage, our paper shows that limits to arbitrage arise when the market performs its utmost e¢ cient role: guiding the allocation of real resources. Thus, while limits to arbitrage based

5

and Section 6 concludes. There is a …rm whose stock is traded in the …nancial market. Goldstein. in turn generating implications for the speed of incorporation of news into prices. If present. but only with respect to uninformed trading. we will use these two terms interchangeably. Trading in the …nancial market occurs at t = 1. Our paper is related to the literature exploring the theoretical implications of the feedback e¤ects from market prices to real decision making. demonstrating the asymmetric limit to arbitrage. Overall. Appendix A contains all proofs not in the main text. Section 2 presents the model. 2g. The manager’ goal is to maximize expected …rm value. who show that it provides an incentive for uninformed speculators to short sell a stock. the point in our paper –that negatively informed speculators will strategically withhold information from the market. Several papers in this literature have shown that the feedback e¤ect can be harmful for real e¢ ciency. and a risk-neutral market maker. such as short-sale constraints.
1
6
.on market frictions tend to attenuate with the development of …nancial markets. Most closely related is Goldstein and Guembel (2008). there are no agency problems due to portfolio delegation. Goldstein. and there is no exogenous friction on trading (other than a standard transaction cost). t 2 f0. our model is simpler and requires only a single period. Their paper also highlights an asymmetry between buy-side and sell-side speculation. Ozdenoren. In addition to the speculator. and Goldstein.
2
The Model
The model has three dates. 1. because they know that the release of negative information will lead managers to …x the underlying problem –is new in this literature. and Guembel (2010). Section 3 contains the core analysis.1 Bond. she is informed about the state of nature that determines the pro…tability of continuing vs. here. two other types of agents participate in the …nancial market: noise traders whose trades are unrelated to the realization of . and Prescott (2010). Our model deliberately shuts down the sources of the limits to arbitrage identi…ed by prior theories: the speculator is risk-neutral. Section 5 discusses potential applications of the model. and Yuan (2011) also model complexities arising from the feedback e¤ect. abandoning the project. reducing its value by inducing a real decision (investment) based on false information. a limit to arbitrage nevertheless arises because the speculator endogenously chooses not to trade due to the feedback e¤ect. a risk-neutral speculator may be present in the …nancial market. With all these forces switched o¤. At t = 0. Dow. since there are no s agency problems between the manager and the …rm. managers will learn from them to a greater degree. while they require two trading periods to generate the asymmetry. This paper proceeds as follows. The …rm’ manager needs to take a decision as to whether to continue or abandon an s investment project. The latter collects
In addition. Section 4 investigates the extent to which information a¤ects beliefs and prices. the e¤ect identi…ed by this paper may strengthen – as investors become more sophisticated. we show that informed speculators are less likely to trade on bad news rather than good news.

depending on whether …rm value is monotonic in the underlying state: n n Case 1: RH > RL . the corrective action attenuates. Without loss of generality. We now describe the …rm’ investment problem and the trading process in more detail. Whether the …rm continues to invest in s 7
.. while abandonment is optimal in state L. This assumption is also without loss of generality as.e. under the optimal action. Lg (“high” and “low” We ). d denote the value of the …rm realized in t = 2 as v = R . Note that equations (1) and (2) imply:
i RH i n RL > R H n RL :
(4)
Equation (4) is the driving force behind our results. if the state is L rather than H. In particular.the orders from the speculator and noise traders. ng. the reduction in …rm value is lower if the manager has taken action n.. Hence. Finally. all uncertainty is resolved and payo¤s are realized. the highest …rm value is achieved in state L and we can simply reverse notations. where d = i represents continuing s the investment and d = n represents no investment (also referred to as “abandonment” or “correction” The …rm faces uncertainty over the realization of value under each possible ). but does not eliminate.e.1
The Firm’ Decision s
Suppose that the …rm has an investment project that can be either continued or abandoned at t = 2.
(1) (2)
that is. continuation is optimal in state H. action. we set:
i n RH > R H n i RL > R L . In this case. which may be a¤ected by the trading in the …nancial market at t = 1. At t = 2. We denote the …rm’ decision as d 2 fi. the highest …rm value is achieved in state H. the manager takes the decision. which depends on both the state of nature and the manager’ action d. s
2. We also set: i n RH > R L . state H can represent high demand for the …rm’ products. there are two possible states 2 fH. Put di¤erently. while state L represents low demand. Abandonment reduces the volatility of …rm value. equation (4) incorporates two cases. if it is not satis…ed. and sets a price at which he executes the orders out of his inventory. the e¤ect of the state on …rm value. consistent with this being labeled as the “high” state. there is no dominant action). s We assume that whether continuation or abandonment is desirable depends on the state of nature (i. It means that taking the corrective action reduces the negative e¤ect of state L on …rm value. i. In turn. the dependence of …rm value on the state. no matter what action has been taken by the …rm. For example. (3) that is. state H is better for …rm value.

n n Case 2: RL > RH . Under continuation. whereas if the …rm chooses to postpone acquisitions. The …rst case is where < 1 . the investment has a positive net present value. i. this second case does not require that abandonment reduces n n i i the volatility of …rm value: it could be that abs (RH RL ) > abs (RH RL ) so volatility is n n higher under correction.. For example. she observes the state of nature with certainty. its value is higher in state L where future acquisition opportunities are superior. Since the prior y is 1 . Another example is related to Aghion and Stein (2008): d = i corresponds to a growth strategy. and so the ex-ante net present value of the investment is negative. Let denote the posterior belief that the state is H such that the manager is indi¤erent between continuation and abandonment. s he will continue the project.
2
8
. its value is higher in the low state where there are no growth opportunities. Whether the speculator is present or not is unknown to anyone else.e. since in the high state. We will use the term “positively-informed speculator” to
Since private information is not public knowledge.e. The manager’ s decision is conditioned on q. Instead. If the …rm eschews the growth strategy (d = n). 1 The prior probability that the state is = H is y = 2 .2
Trade in the Financial Market
In t = 0. and d = n corresponds to a strategy focused on current pro…t margins. In this case. Put di¤erently. the manager would continue the investment without further 2 2 information.2 If the speculator is present. Importantly. The second case is where > 1 .:
i RH + (1 i n )RL = RH + (1 n )RL :
(5)
The value of represents a “cuto¤”that determines the manager’ action. which in turn is calculated using information arising from trades in the …nancial market. but the negative e¤ect of state L is attenuated if the …rm does not invest. in an equilibrium in which informed insiders manupulate the market by trading in the wrong direction. its rivals could pursue the growth opportunities. Growth prospects are good if = H and bad if = L. If and only if q > . ex ante. …rm value is higher in state H. and abandonment implies keeping the cash for future opportunities. We use q to denote the posterior probability the manager assigns to the case = H. 2
2. its value will be lower in state L. Chakraborty and Yilmaz (2004) also feature uncertainty on whether the speculator is present. in turn worsening its competitive position.its production process or not. consider the case where continuation implies proceeding with a takeover decision. which is common knowledge. and state L refers to the reverse. State H corresponds to a state in which current acquisition opportunities dominate future ones. its existence is also unlikely to be public knowledge. the case RL > RH implies non-monotonicity of …rm value in the state: one state does not dominate the other. We will distinguish between two cases. a speculator arrives in the …nancial market. the corrective action is su¢ ciently powerful to overturn the e¤ect of the state on …rm value. …rm value is higher in state L. with probability 0 < < 1. i. if the corrective action is taken.

she makes an endogenous trading choice s 2 f 1. The variable is a measure of market sophistication or the informedness of outside investors and will generate a number of comparative statics. (iv) the …rm and the market maker use Bayes’ 9
. ng (where Q = f 2. The market maker can only observe total order ‡ X = s + z. Unless otherwise speci…ed. but not its individual components s and z. the strategy of the manager. and “negatively-informed speculator”to describe a speculator who observes = L. It is also realistic to assume that managers have access to information about trading quantities in the …nancial market: …rst. In the equilibria that we analyze. microstructure databases (such as TAQ) provide such information at a short lag –rapidly enough to guide investment decisions. simpli…es the analysis without a¤ecting its economic ow content. non-interesting. Allowing the manager to observe order ‡ X. second. 1g that maximizes his expected …nal payo¤ s(v p) jsj . who trades z = 1. 1g. it is de…ned as follows: (i) A trading strategy by the speculator: S : ! f 1.describe a speculator who observes = H.
2. orders are submitted simultaneously to a market maker who sets the price and absorbs order ‡ ows out of his inventory. Trading either 1 or 1 is costly for the speculator and entails paying a cost of . 0. given the information contained in the order ‡ ow. Speci…cally. Following Kyle (1985). rather than just the price p. (ii) An investment strategy by the …rm D : Q ! fi. Since this paper’ focus is to analyze the feedback s e¤ect. other. If the speculator is indi¤erent between trading and not trading (because her expected pro…ts from trading exactly equal ). A critical departure from Kyle (1985) is that …rm value here is endogenous. 2g). equilibria can arise. Trading in the …nancial market happens in t = 1. given the information in the price and all other strategies. 0. (iii) A price setting strategy by the ow market maker p : Q ! R that allows him to break even in expectation. the manager observes total order ‡ X. 1. we do not analyze such equilibria here.3
Equilibrium
The equilibrium concept we use is the Perfect Bayesian Nash Equilibrium. Here. Under the ow alternative assumption that the manager observes p. 1. 1. 0. we assume that she will not trade. which requires the price to be informative. where the price is essentially uninformative. given the price setting rule. If the speculator is present. Always present is a noise trader. 2g and the pricing function is p (X) = E(vjX). and uses the information in X to form ow his posterior q. market making is competitive and so there is little secrecy in the order ‡ ow. we refer to trading pro…ts and losses gross of the cost . 0. or 1 with equal probabilities. there is a one-to-one correspondence between the price and the order ‡ so it does not matter which variable the manager observes. The orders are market orders and are not contingent on the price. The competitive market maker sets the price equal to expected asset value. Moreover. 0. that maximizes expected …rm value v = Rd given the information in the order ‡ and all other strategies. and his information about the realization of . 1. because the manager’ action is based on s information revealed during the trading process. Possible order ‡ ow ows are X 2 f 2. which is then used in the investment decision.

3 Second. We consider Case 1 (RH > RL ) …rst and then s n n proceed to Case 2 (RH < RL ). without further information. and (v) beliefs on outcomes not observed on the equilibrium path satisfy the Cho and Kreps (1987) intuitive criterion. we make use of three di¤erent threshold levels of the cost of trading :
1
2
3
1 1 i 1 i RH RL + 3 2 2 1 1 1 i + RH 3 2 2 1 i i RL .
3
Feedback E¤ect and Asymmetric Limits to Arbitrage
In this section.e. Thus. Finally. The …rst case is 2 < . Here. As we will show. this s node is not relevant for determining the equilibrium trading strategies. i. we will show that our main insight carries through to the case where > 1 . i RL
and
n (RH n RL ) > 0:
(9) (10)
2
1
=
11 32
i RH
The results also depend on whether order ‡ is su¢ ciently informative to overturn the decision ow to invest. the …rm will choose to invest. the …rm’ decision in this case is not relevant for equilibrium s trading strategies as the speculator’ information is fully revealed and so she never makes a pro…t. it represents the posterior probability of state H under 2 1 an order ‡ of X = 1. the probability that the speculator ow is present is su¢ ciently high that a negative order of X = 1 is su¢ ciently informative to deter the manager from investing. Hence. we distinguish between two cases 1 depending on whether the cuto¤ is higher or lower than 2 . where R 3 H
1
n (RH i RL .
While X = 2 is also a negative order ‡ ow..
3
10
. Here. in some equilibria.
(6) (7) (8)
<
2
<
3. the quantity 1 is relevant as. we characterize the pure-strategy equilibria in our model.
n RL ) . In 2 our characterization.rule in order to update their beliefs from the order they observe in the …nancial market. We demonstrate the emergence of asymmetric limits to arbitrage as a result of the feedback from market trading n n outcomes to the …rm’ investment decision. there is no feedback e¤ect for X = 1. Thus. 2 > . a negative order of X = 1 is not su¢ ciently informative to lead the manager to abandon the default plan of investing.1
n n Case 1: Firm Value is Monotone in the State: RH > RL
1 We start with the case where < 2 . which is the ex-ante optimal decision. there is feedback from the market to real decisions for 1 the case of X = 1. (vi) all agents have rational expectations in that each player’ belief s about the other players’strategies is correct in equilibrium.
3. Later. Thus.

if and only if there is feedback ( 2 < ).As we show. Below. the only possible pure-strategy equilibria are N T . it is straightforward to show that the speculator will never buy when she knows that = L and will never sell when she knows that = H. in the case of no feedback ( 2 > ). 4. this is the only belief that is consistent with the intuitive criterion. four equilibrium outcomes can arise: 1. = H and sells when
3. T . the posterior q. the beliefs ow of the market maker and the manager are that the speculator knows that the state is L (H). the only pure-strategy equilibrium is T . there are two pure-strategy equilibria: BN S and SN B. the only pure-strategy equilibrium is N T .Not Buy): the speculator does not trade when she knows that = H and sells when she knows that = L. Then. 2. When 2 < 3 . Then the trading game has the following pure-strategy 2 equilibria: When < 1 . Given that speculators always lose if they trade against their information. No Trade Equilibrium N T : For a given order ‡ X. There is no range of parameter values for which the SN B equilibrium exists but the BN S equilibrium does not exist. Suppose that RH > RL and < 1 . Proof. If an order ‡ of X = 2 (X = 2) is observed o¤ the equilibrium path. investment is ex-ante desirn n able). BN S. we identify the conditions under which each one of these equilibria holds. the only pure-strategy equilibrium is 2 1 BN S. No Trade Equilibrium N T : the speculator does not trade. depending on the values of .Not Sell): the speculator buys when she knows that = H and does not trade when she knows that = L. Partial Trade Equilibrium BN S (Buy . the only pure-strategy equilibrium is T . When 1 < 2 : in the case of feedback ( 1 < ). …rm value is monotone in the state. Proposition 1 provides the characterization of equilibrium outcomes. and SN B. 1 That is. Proposition 1 (Equilibrium. Given that …rm value is always higher when = H than when = L. there is a strictly positive range of parameter values ( 1 < 2 ) for which the BN S equilibrium exists but the SN B equilibrium does not exist. the manager’ decision d and the price p are given ow s by the following table (see Appendix A for the full calculations): X q d p 2 0 n n RL
1 2
1 i
1 i R 2 H i + 1 RL 2
0
1 2
1
1 2
i
1 i R 2 H i + 1 RL 2
i
1 i R 2 H i + 1 RL 2
2 1 i i RH
11
. When 3 . Partial Trade Equilibrium SN B (Sell . Trade Equilibrium T : the speculator buys when she knows that she knows that = L.

this equilibrium holds if and only if 3. we can see that this equilibrium holds if and only if 2 < 3. the pro…t for the negatively-informed speculator from deviating i i to selling is 1 (RH RL ).As shown in Appendix A. 2 Trade Equilibrium T : For a given order ‡ X. Partial Trade Equilibrium BN S: For a given order ‡ X. and this is also the pro…t for the positively-informed speculator from 3 deviating to buying. we can see that this equilibrium holds if and only if 2 < 3 for the case of no feedback ( 1 > ) and if and only 2 if 1 < 3 for the case of feedback ( 1 < ). Partial Trade Equilibrium SN B: For a given order ‡ X. the posterior q. the manager’ decision d and the price p are given ow s by the following table: X q d p 2 0 n
n RL
1
1 2 (
0
1 2 1 2 1 2 2 1 2
1
1 2
2 1 i
i + 1 RL 2 i RH
(
n if i if
1 2 1 2
< >
1 n RL i RL
i if if
1 2 1 2
i
i + 1 RL 2 1 i R 2 H
n RH + i RH +
< >
1 i R 2 H
Calculating the pro…t for the negatively-informed speculator from deviating to selling and for the positively-informed speculator from deviating to not trading. the manager’ decision d and the price p are given ow s by the following table: X q d p 2 0 n
n RL 1
1
2 (
0
1 2 1 2 1 2 2 1 2
1
1 2
2 1 i
(
n if i if
1 2 1 2
< >
1 n RL i RL
i if if
1 2 1 2
i
1 i + 2 RL 1 2 i RH + 1 2 i RL
n RH + i RH +
< > 12
1 i R 2 H
i RH
. Thus. the posterior q. the posterior q. the manager’ decision d and the price p are given ow s by the following table: X q d p 2 0 n n RL
1 2
1 i
1 i R 2 H 1 i + 2 RL
0
1 2
1
1 2
i
1 i R 2 H 1 i + 2 RL
i
1 2 i RH + 1 2 i RL
2 1 i i RH
Calculating the pro…t for the negatively-informed speculator from deviating to not trading and for the positively-informed speculator from deviating to buying.

there are two pure strategy equilibria. when speculators are subject to greater transaction costs. Thus. and the speculator being negatively informed i i and choosing not to trade. The …rst source is the trading cost . when 2 > . 1 < 2 requires < 1. As increases. consider the BN S equilibrium without feedback (the case of the SN B equilibrium is analogous). leading to larger price impact which reduces
13
. we can see that this equilibrium holds if and only if < 2 for the case of no feedback ( 1 > ) and if and only if 2 1 < 1 for the case of feedback ( 2 < ). When < 2 . given that the market maker believes that the speculator does not sell on bad news. knowing this. Here. In turn. exhibiting limited trade.. Conversely. which always holds. When 3 . there is a range of for which the only equilibrium is BN S if 1 < 2 and 2 < . To understand the intuition behind these asymmetric equilibria. a negative order ‡ is very revealing that the speculator is negatively informed and the price ow moves sharply to re‡ this. There is symmetry in that both types of asymmetric equilibria are possible in exactly the same range of parameters. Thus. but ow not su¢ ciently negative to deter the manager from abandoning the default plan of investing. In this case. Thus.Calculating the pro…t for the negatively-informed speculator from deviating to not trading and for the positively-informed speculator from deviating to not trading. the order ‡ in the direction in which the speculator ow does not trade becomes particularly informative. When 2 < 3. and one in which she sells on bad news but does not trade on good news. the speculator makes little pro…t from selling on bad news. the only pure-strategy equilibrium entails no trade at all by the speculator. In the intermediate region 2 < 3 . 1 Thus. both of which are common in the literature.e. an order ‡ of X = 1 may convey (depending on the equilibrium) negative information. Clearly. Proposition 1 demonstrates the sources of limits to arbitrage in our model. In sum. i. they have lower incentives to trade. 2 2 which allows the speculator to make high pro…ts by buying. The second source is the price impact that speculators exert when they trade on their information. the only pure-strategy equilibrium is one where the speculator always trades on her information. one of which is the feedback e¤ect that is the focus of our paper. and so the speculator only receives 2 RH + 2 1 RL from selling. Given that the market maker believes that the speculator buys on good news. Thus. X = 1 is inconsistent with the speculator having ect 1 i i positive information. 1 < requires > 11 2 . the market maker sets a relatively low price of 1 RH + 1 RL . Two sources of limits to arbitrage are present in the no-feedback case. the equilibrium is sustainable. a positive order ‡ is consistent with the speculator being negatively informed: ow X = 1 is consistent with the noise trader buying. From (4) and (10). Speci…cally. there are three regions of the parameter . one in which the speculator buys on good news but does not trade on bad news. there 2 exist values of that satisfy both of the above conditions if 11 2 < 1. there are equilibria where the speculator trades on one type of information but not the other. we start 1 by describing the equilibrium outcomes in the case of no feedback. in both partial trade equilibria. she chooses not to trade on bad news. we move to equilibria where speculators trade less on their information. To understand the various forces.

The intuition behind the asymmetry of our e¤ect is as follows. Hence. the di¤erence between equilibrium outcomes in the two cases of no-feedback and feedback is that in the range 1 < 2 . 1 We now move to the case of feedback. there is no value of in which there is one
In the case discussed so far ( < 1 ) the default option for the manager is to invest. Unlike trading costs and price impact. which (weakly) causes him to increase investment. it deters the speculator from selling on bad news. By contrast. Thus. the feedback e¤ect leads to an asymmetric limit to arbitrage that deters selling on bad news but not buying on good news. if > 2 ). In the case of feedback. Abandonment is the optimal decision in state L.. This reduced pro…t a¤ects the speculator’ equilibrium trading strategy s and causes her not to sell on bad news if 1 . This improves his decision making and thus fundamental …rm value. This e¤ect is the driving force behind our results in the case of > 1 .the potential trading pro…ts. BN S and SN B. On the other hand. and so positive news 2 from the market does not change his decision and does not a¤ect …rm value. if > 1 . 2
4
14
. the limit to arbitrage arising from the feedback e¤ect is asymmetric: it reduces the extent of selling on bad news but not the extent of buying on good news. Price impact leads to the two partial trade equilibria. not trading in this direction is sustained in equilibrium. this increase in …rm value reduces her pro…t. Hence. this augments …rm value. This force is symmetric in the absence of feedback. in turn strictly improving …rm value. Here. the feedback e¤ect does not deter the positively-informed speculator from buying on good news. However. Our main result is that the feedback e¤ect introduces an additional limit to arbitrage that is distinct from those identi…ed in prior literature – arbitrage is limited because the value of the asset being arbitraged is endogenous to the act of arbitrage. when the speculator sells on bad news. since RL > RL . 1 In particular.e. without feedback (i. there is no range of parameters where the SN B equilibrium exists but the BN S equilibrium does not exist.e. Indeed.. the two limits to arbitrage studied in prior research are symmetric. improving the manager’ decision reduces the s 1 i i speculator’ pro…t in the node of X = 1 from 2 (RH RL ) (in the case of no-feedback) to s 1 n n only 2 (RH RL ). Hence we state that buying on good information causes the manager to weakly increase investment. since investment is desirable in the high state. Buying on good news may reveal to the manager that the state is good. As we will show later. but there is symmetry in that both equilibria are possible in exactly the same range of parameters. Increased …rm value augments the pro…tability of a long position but reduces the pro…tability of a short position. A high trading cost leads to the N T equilibrium where there is no trading in either direction. she improves …rm value. trading on her information in either direction – whether it is buying on positive information or selling on negative information – conveys information to the manager. Since she is holding a short position. buying on 2 good news causes the manager to strictly increase investment. an order ‡ of X = 1 ow provides enough negative information for the manager to abandon the investment. which will further increase her incentive to trade. By n i that. The speculator will then pro…t from the increase in the value of her long position. when 2 < .4 Overall. she may lead the manager to abandon a bad investment. the Trade Equilibrium T is replaced with the Partial Trade Equilibrium BN S. i. thus.

the case < 1 . the BN S equilibrium occurs over a wider range than the SN B equilibrium.partial trade equilibrium but not the other. Because the speculator knows both the manager’ 1 d d action and the state is L. consider the market maker’ inference from seeing X = 1 in the BN S equilibrium. Thus. she makes a pro…t of 2 RH RL .. our results are qualitatively similar: without feedback. but superior knowledge of the state. if = 1. her pro…ts are lower in this case. We now wish to verify that the asymmetry between buy-side speculation and sell-side speculation. when the 2 2 default decision is abandonment.. The key to our result is that the source of the speculator’ pro…ts is s not superior knowledge of the manager’ action (since the action is always perfectly predicted s by the market maker). since positive order ‡ leads ow to investment but the investment would be undertaken in the absence of further information anyway.e. Thus. Indeed. which in turn depends on the decision d. the manager’ action d (and thus the feedback e¤ect on the manager’ action) matters for the s s speculator’ trading pro…ts. This order ‡ is s ow consistent with either the speculator being absent (in which case the state may be H or L). The source of her pro…t is her superior information on the state. but because the action a¤ects the value of the speculator’ superior s s knowledge on the state. Put di¤erently. Intuition may suggest that the market maker’ pricing function can “undo” the feedback e¤ect: the market s maker is fully rational and takes into account the fact that the order ‡ will a¤ect the manow ager’ decision. if < 1. whereas with feedback. or the speculator being present and negatively informed. The reason for why the feedback e¤ect reduces trading pro…ts is nuanced. i. i. the order ‡ of X = 1 is fully ow revealing: the market maker knows both that correction will occur. superior knowledge of the state results from fact that < 1. the market maker allows for the possibility that the state may 1 d d s be H and sets a price of 2 RH + 2 1 RL . The next proposition shows that when > 1 . In the case of < 1 . Because the price that the speculator receives from selling will always re‡ the action taken by the manager ect (be it continuation or investment). the speculator is not always present. is not an artifact of the fact that investment is the default decision. driven by the feedback e¤ect. and that the state is L. the market maker predicts the manager’ s action but does not know the state. In turn. since he is competitive. i. Since X = 1 can be consistent with the speculator being absent and the state being H. By contrast. If = 1. Such intuition turns out to be incorrect. BN S and SN B equilibria occur over the same range of parameters. the speculator makes zero pro…t. To see this. Here. since the state has a lower e¤ect on …rm value under correction. the 2 source of the limit to arbitrage was that the feedback e¤ect reduces the pro…tability of a short position but does not a¤ect the pro…tability of a long position.e. it may seem that the action does not matter. and so the market maker knows for certain that = L. the …rst case is ruled out. the source is that the feedback e¤ect increases the pro…tability of a long position
15
. not because the speculator’ pro…ts stem from superior knowledge s s of the manager’ action. then 1 = 2 and there is no range of parameter values in which there is a BN S equilibrium only. he sets a price that re‡ s ects this decision. and n so sets price exactly equal to the fundamental value of RL .e.

as discussed above. the 2 intuition is the same: the feedback e¤ect (weakly) increases the pro…tability of a long position and (weakly) decreases the pro…tability of short position. the only pure-strategy equilibrium is N T . The proof repeats similar steps to those in the proof of Proposition 1.but does not a¤ect the pro…tability of a short position. The following proposition provides the characterization of equilibrium outcomes. There is no range of parameter values for which the SN B equilibrium exists but the BN S equilibrium does not exist. whereas the SN B equilibrium exists only in the sub-range 01 < 03 or does not exist (if 01 > 03 ). …rm value is monotone in the state. Hence. 02 < 01
i RH n RL ) . If 01 > 03 . For this proposition. there is feedback. there is a range of parameter values for which the BN S equilibrium exists but the SN B equilibrium does not exist. BN S and SN B. When 02 < 03 : in the case of no feedback ( 2 1 < ). Suppose that RH > RL and > 1 . and so the BN S equilibrium is sustainable over a wider range of parameters than the 16
.
i RL
. if and only if there is feedback ( 2 1 > ). investment is ex-ante unden n sirable). ow This is the case where there is no feedback. if 01 < 03 . and 3 < 03 . this range is 02 < 01 . In some equilibria. there are two pure-strategy equilibria. If 2 1 < . the BN S equilibrium always exists. and is thus omitted for brevity. then the trading game has the following pure-strategy 2 equilibria: When < 02 . If 2 1 > . the probability that the speculator is present is su¢ ciently high that an order ‡ of X = 1 contains enough information to lead the ow manager to invest (as opposed to the default option of abandoning).
(11) (12) (13) (14)
The cuto¤ for feedback e¤ect to exist is also adjusted here. in the case of feedback ( 2 1 > ). 2 1 represents the posterior probability of state H if X = 1. an order ‡ of X = 1 is not informative enough to lead the manager to invest. de…ne new threshold levels of the cost of trading :
0 1 0 2 0 3 0 2
1 1 1 n n (RH RL ) + 3 2 2 1 1 1 n + (RH 3 2 2 1 n n (RH RL ) . That is. Proposition 2 (Equilibrium. the only pure-strategy equilibrium is T . this range is 02 < 03 . In the case of < 1 . In both cases (for both < 2 and > 1 ). since abandonment would be undertaken 1 in the absence of further information anyway. Proof. 0 When 3 . the role of the feedback e¤ect can be seen in the BN S equilibrium: 2 it reduces the pro…ts that the negatively-informed speculator would earn by deviating and selling.

2
n n Case 2: Firm Value is Non-Monotone in the State: RH < RL
In this subsection. moving us closer to the analysis conducted in the previous subsection. we consider the case where. now that …rm value may be higher in state L. its value is higher n n in state = L (RH < RL ). then X 2 f 2. We start by characterizing equilibrium outcomes for the case where < 1 . investment will occur. posterior probability q of state H is at least 1 (given that these 2 nodes are consistent with the action of the positively-informed speculator and may or may not be consistent with the action of the negatively-informed speculator. since …rm value is always higher in state H than in state L. depending on her equilibrium 1 i action). In the previous subsection. if the …rm does not invest. since < 2 . where > 1 . and so the speculator makes a loss from selling. and so …rm value is RL . (ii) Suppose that the speculator buys when she knows that = L. there are nine possible pure-strategy equilibria (each type of speculator –positively-informed and negatively-informed –may either buy. a positively-informed speculator may …nd it optimal to sell and a negatively-informed speculator may …nd it optimal to buy.e. However. The price. i i will be between RL and RH . the feedback e¤ect impacts the SN B equilibrium. then: 2 (i) The trading game has no pure-strategy equilibrium where the speculator sells when she knows that = H. 0g. Then. 2 Since buying improves the …rm’ fundamental value. Given that the positively-informed speculator does not sell. where …rm value is monotone in the state. since < 2 . 2g. however. (i) Suppose that the speculator sells when she knows that = H. Hence. then X 2 f0. or not trade). and so the SN B equilibrium is sustainable over a narrower range of parameters than the BN S equilibrium 1 (indeed.
3. and so …rm value is RH . In both cases (for < 2 and > 1 ).
Proof. investment will occur. (ii) The trading game has no pure-strategy equilibrium where the speculator buys when she knows that = L. i. the end 2 result is the same: the feedback e¤ect increases the pro…ts from informed buying and reduces the pro…ts from informed selling. it is not sustainable at all).
n n Lemma 1 Suppose that RH < RL and < 1 . the feedback e¤ect increases the pro…t s that the positively-informed speculator would earn by deviating and buying..SN B equilibrium. Here. Since the price 17
. In each one of these nodes. The analysis of equilibrium outcomes becomes more complicated in the case of non-monotonicity. the posterior probability q is 1 at 2 1 i X 2 f0. Hence. without further information. The following lemma simpli…es the equilibrium analysis. leading to the BN S equilibrium being sustainable over a wider range of transactions costs than the SN B equilibrium. sell. 1. the corrective action is su¢ ciently powerful to outweigh the e¤ect of the state on …rm value and lead to a higher value in the low state. a positively-informed speculator always loses money by selling and a negatively-informed speculator always loses money by buying. 1g. Hence. 2 the …rm will choose to invest. 1. if 01 > 03 .

and BN S. the conditions for these equilibria to hold are now tighter. When she sells. Again. We nevertheless retain this assumption regarding o¤-equilibrium beliefs. and so to abandon the investment. If she does not trade in equilibrium. this decision reduces …rm value and causes the speculator to make a pro…t on her short position. …rm value is non-monotone in the state. the characterization of equilibrium outcomes is identical to that in Proposition 1. buying a security worth RL for a price of RL . an additional condition must be satis…ed to ensure that the positively-informed speculator does not have an incentive to deviate to selling. however. However. Then. and suppose that the belief of the market 2 maker and the manager is that an o¤-equilibrium order ‡ow of X = 2 (X = 2) is associated i (Ri RL ) with the presence of negatively-informed (positively-informed) speculator. If the positively-informed speculator buys in equilibrium. she cannot make a pro…t from selling. the intuitive criterion is not su¢ cient to rule out other o¤-equilibrium beliefs. the speculator will lose money on these nodes. then the negatively-informed n n speculator is revealed. Our results remain the same for any other o¤-equilibrium beliefs that are monotone in the order ‡ ow. due to monotonicity. Following the lemma. T . In Case 1. but because the speculator has a long position. there are two 2 2 possibilities. In this subsection. given she loses at X 2 f0. Thus. investment is ex-ante n n desirable). Hence.i i is 1 RH + 1 RL . she misleads the market maker and the manager to think that the positively-informed speculator is present. However. she potentially misleads the market maker and the manager to think that the negatively-informed speculator is present. Interestingly.5 In analyzing deviations from the equilibrium. there are four possible pure-strategy equilibria. while an o¤-equilibrium order ‡ of X = 2 is due to the negatively-informed speculator (and so the posterior is ow q = 0).
Proposition 3 (Equilibrium. When X = 2. as she never has an incentive to deviate to buying. Suppose that RH < RL and < 1 . she still might be tempted to deviate to selling in any of the four equilibria mentioned above. if RH Rn is n ( L H) su¢ ciently high. Since abandonment is suboptimal if = H. just as in the previous subsection: N T . this causes her to make a loss. 1g. If she does so. for any of the above four equilibria to hold. which is reasonable given the possible equilibria in our model. The reason that the positively-informed speculator never sells in equilibrium is that if the market maker and the manager believe that she sells. then the outcome is the same as on the other nodes. this reduces …rm value.
5
18
. the only assumption that satis…ed the intuitive criterion was that an o¤-equilibrium order ‡ of X = 2 is due ow to the positively-informed speculator (and so the posterior is q = 1). SN B. the same issue does not arise with the negatively-informed speculator. another issue that arises in this subsection is the speci…cation of o¤-equilibrium beliefs. and so to (incorrectly) take the investment. The following proposition provides the characterization of equilibrium outcomes. in expectation she makes a loss.
Goldstein and Guembel (2008) also derive conditions to ensure that the speculator does not deviate from the equilibrium to trade against her information.

Hence. n ( L H) As the proposition demonstrates. The speculator knows that the corrective action will be taken. 2 Rn Rn > 1. because she is unsure of the underlying state . the main force identi…ed in the previous subsection for the n n n n case where RH > RL . That is. this force is even stronger because the range of transaction costs between 1 and 2 . the market maker knows the corrective action will be taken but is not certain that correction is desirable for …rm value. again. the manager’ decision d and the price p for di¤erent s order ‡ ows X in the various possible equilibria are identical to those provided in the proof of Proposition 1. the key to this result is < 1. In contrast. they have di¤ering views on …rm value conditional on abandonment. implies that the negatively-informed speculator can make a loss –even before transaction costs –when selling on bad news. the feedback e¤ect deters the negatively-informed speculator. the following additional conditions are required for the various equilibria to hold: i i n n Equilibrium N T and SN B: > 1 ( (RH RL ) + (RL RH )). in the case of no n 2 ( L H) i i (RH RL ) 1 feedback ( 2 > ). it is possible that = H. ( L H) Proof.More speci…cally. 4 2 ( L H) i (Ri RL ) Equilibrium T : in the case of feedback ( 1 < ). the conditions for the positively-informed speculator to choose between buying and not trading and for the negatively-informed speculator to choose between selling and not trading are identical to those derived in the proof of Proposition 1. In this subsection. Hence. in which correction not only mitigates the e¤ect of the low state but also overturns it. Order ‡ X = 1 is consistent ow with a negatively-informed speculator. and that correction is desirable n for …rm value (since she knows that = L). Appendix A calculates the possible trading pro…ts for the positivelyinformed speculator from deviating to selling in each of the four possible equilibria. which yields obtain the additional conditions stated in the body of the proposition. but also with an absent speculator and selling by noise traders. in which case the manager’ corrective action is s 19
. from trading on her information. it is straightforward to see that she always loses from buying and hence will never deviate. Even though both the speculator and market maker know that abandonment will occur if X = 1. This loss occurs at the X = 1 node. in which the BN S equilibrium exists due to feedback but the n n SN B equilibrium does not exist. Analyzing the possible trading pro…ts for the negatively-informed speculator from deviating to buying in each of the four possible equilibria. These conditions are i (Ri RL ) binding only when RH Rn is not su¢ ciently high. in the case of no 6 2 ( n L H) i i (RH RL ) 1 feedback ( 2 > ). and so …rm value is RL . exists also in the case where RH < RL . is higher when (RH RL ) is negative: see equation (10). 33 2 RH Rn > 1. 6 3 RH Rn > 1. 8 3 Rn Rn > 1. A strong feedback e¤ect. but not the positively-informed speculator. 3 i (Ri RL ) 1 Equilibrium BN S: in the case of feedback ( 2 < ). The calculations of the posterior q. This result is in contrast to standard informed trading models where a speculator can never make a loss (before transactions costs) if she trades in the direction of her information.

the price set by the market maker is only 1 n n RH + 2 1 RL . The characterization of equilibrium outcomes in the following proposition is symmetric to that in Proposition 3: 20
. then: 2 (i) The trading game has no pure-strategy equilibrium where the speculator sells when she knows that = L. It turns out that this case is the exact mirror image of the case where < 1 . Trade Equilibrium T 0 : the speculator buys when she knows that she knows that = H. The next lemma is the mirror image of Lemma 1: n n Lemma 2 Suppose that RH < RL and > 1 . (ii) The trading game has no pure-strategy equilibrium where the speculator buys when she knows that = H. Partial Trade Equilibrium SN B 0 (Sell . the speculator now sells if = H and buys if = L. which enables her to pro…t from her short position. Therefore.Not Sell): the speculator buys when she knows that = L and does not trade when she knows that = H.Not Buy): the speculator does not trade when she knows that = L and sells when she knows that = H. and that this force has implications on the characterization of equilibrium outcomes. Thus. making it relatively more di¢ cult to obtain the BN S equilibrium due to feedback.n undesirable. the loss from n ( L H) i i trading against good news (which is proportional to (RH RL )) is high relative to the benen n …t from manipulation (which is proportional to (RL RH )). Otherwise. the possible pure-strategy equilibria here are: 1. Hence. Partial Trade Equilibrium BN S 0 (Buy . the proposition also shows that another force that arises from the feedback e¤ect exists in this subsection. leading to …rm value of RH . In this case. No Trade Equilibrium N T : the speculator does not trade. The proof is symmetric to the proof of Lemma 1 and hence is not repeated here. …rm value is lower in state H than in state L. under this decision. = L and sells when
3. there are additional conditions for the various possible equilibria. 4. since he puts weight on the fact that correction may be undesirable. we analyze the case where > 2 . 1 Finally. This is the desire of the positively-informed speculator to deviate from her equilibrium behavior and manipulate the price by selling. e¤ectively. = H represents bad news and = L 2 represents good news. This is because the default decision is to abandon the investment. However. 2. She can potentially pro…t from leading the manager to take the wrong decision. Now.
Proof. The manipulation incentive is not strong enough to interfere i (Ri RL ) with equilibrium conditions as long as RH Rn is su¢ ciently high. even though she has good news. and so 2 1 n n the speculator loses 2 (RH RL ) before transaction costs.

another important assumption is that < 1. consider again the result in Proposition 21
. For example. RH . and so the SN B is sustainable over a narrower n n range of parameters. the result is identical to that of the case of < 1 . Hence.Proposition 4 (Equilibrium. RH . so there is uncertainty on whether there is an informed speculator in the market. 1 n n In Case 1 (RH > RL ). The proof is symmetric to the proof of Proposition 3 and hence is not repeated here. we s require 1 < . bad news entails = H and good news entails = L. respectively. i. First. Overall. the ex-ante NPV of the project is so high that the manager will almost always undertake the investment. the desirability of the investment is su¢ ciently uncertain that the manager’ decision will be in‡ s uenced by the trading in the …nancial market. the role of the feedback e¤ect is seen in the SN B equilibrium: the feedback e¤ect increases the pro…ts that the positively-informed speculator would earn by deviating to buying. When is close to 2 . RL are replaced with n n i i parameters RL . the speculator 2 is deterred from selling when she has bad news.
3. since they demonstrate the conditions under which the asymmetric limit to arbitrage will exist. BN S. Then. Several assumptions play a key role in generating this result. but encouraging them to buy on good news. investment is ex-ante n n undesirable). discouraging speculators from selling on bad news.3
Summary and Discussion of Assumptions
The above analysis has shown that the presence of feedback from market trading to …rms’ decisions creates a wedge between buy-side speculation and sell-side speculation. SN B 0 . the role of the feedback e¤ect is seen in the 2 BN S/BN S 0 equilibrium: it deters the speculator from deviating to sell on bad news ( = H in this case). RL . and suppose that the belief of the market maker 2 and the manager is that an o¤-equilibrium order ‡ow of X = 2 (X = 2) is associated with the positively-informed (negatively-informed) speculator. SN B are replaced with equilibria T 0 . just as in the 2 scenario of < 1 (for both Case 1 and Case 2). and equilibria T . Second. BN S 0 . These assumptions in turn lead to empirical predictions. …rm value is non-monotone in the state. To see this. the characterization of equilibrium i i n n outcomes is symmetric to that in Proposition 3: parameters RH . RL . the closer the NPV of the 2 1 project is to 0. RH . for the sub-case of > 2 . Suppose that RH < RL and > 1 . Proof. it is important that the probability that the speculator is present is 2 su¢ ciently high so that the order ‡ is su¢ ciently informative to change managerial decisions. consider the result in Proposition 1: for the wedge to arise. In the current scenario of > 1 within Case 2 (RH < RL ). but not from buying when she has good news. If is very low. ow 1 In turn.e. regardless of order ‡ ow. the trading in the market has to contain su¢ cient information to in‡ uence the manager’ decision. Due to feedback. respectively. The only di¤erence is that now. 2 < is more likely to be satis…ed the closer is to 1 .

if with probability 1 she 6 receives a liquidity shock that prevents her from trading. However. as the uninformed speculator may choose to trade to manipulate the price and the …rm’ decision. s 7 Note that the existence of benchmarking alone is typically insu¢ cient to explain the reluctance of mutual funds to deviate from their benchmark. if she sells half of her portfolio. she increases the value of the remaining half. she is a mutual fund benchmarked against the performance of other mutual funds). since the gains from beating one’ benchmark (by deviating) equal the s losses from underperforming one’ benchmark. which in turn requires the speculator’ initial position s to be zero (or. the model delivers the result that investors are more likely to engage in sales rather than shortsales. the speculator is fully revealed in these nodes and her pro…ts are zero. This is a natural assumption if the decision maker is the …rm’ manager who attempts s to maximize …rm value via an investment decision. may introduce other complications. the value of this information depends on the manager’ decision s (as this a¤ects the dependence of …rm value on the state) and thus the feedback e¤ect. Indeed.7 Fourth. In this case.1.e.g. in which a BN S equilibrium arises and a SN B equilibrium does not arise. less than the amount sold) and short-sales to be possible. the speculator is not fully revealed and makes a positive pro…t from her private information about the state.
6
22
. While short-sales constraints do not deter selling to a non-negative …nal position. at least. For example. it is important that the speculator short sells rather than just sells stocks she previously owned. but increases the value of the entire portfolio held by her competitors. even in the absence of a short-sales constraint. the real decision is a corrective action in that it improves …rm value in the low state. shrinks to zero if = 1. Our explanation for benchmarked investors’unwillingness to sell does not s require such asymmetry. Therefore. our model also predicts that investors who are evaluated according to absolute returns are more likely to sell on negative information than those who are benchmarked to their peers. This. The model does not apply to amplifying actions that worsen …rm value in the low state. hedge funds appear to sell (not just short-sell) more readily than mutual funds. as in Goldstein and Guembel (2008). violate
An alternative assumption would be that the speculator is always present. another potential application is to a board of directors which chooses whether to …re an underperforming manager in the bad state. The range between 1 and 2 . but can only trade with probability – for example. Where < 1. Hence. but sometimes she is uninformed. existing explanations typically rely on the asymmetry in s in the ‡ ow-performance relation: the in‡ ows from beating one’ benchmark are lower than the out‡ s ows from underperforming one’ benchmark. If = 1. the feedback e¤ect can deter such selling if the speculator maximizes relative performance. however. and so loses in relative terms. Thus. We would achieve the same result by instead assuming that the speculator is always present and informed. Thus. Third. leading to no asymmetry. This is because the di¤erence in expected pro…t from buying on good information and selling on bad information stems entirely from the di¤erence in pro…t in the node where the speculator is partially revealed (X = 1 or 1). then our limit to arbitrage may exist even if her initial position is strictly positive. the limit to arbitrage identi…ed by this paper may exist even in the presence of short-sales constraints. i. the reason that the speculator loses from increasing the …rm’ value is that she ends s up with a short position. if the speculator maximizes returns relative to other speculators or market indices rather than absolute returns (e.

The manager is rational s and takes into account the fact that the speculator does not sell on negative information: he updates his beliefs using the asymmetric equilibrium trading strategy. we have shown that good news received by the speculator has a di¤erent impact on her trades (and thus total order ‡ ow) than bad news. it is not obvious that this will translate into a di¤erential impact on the manager’ beliefs.
23
.
4
E¤ect of Information on Beliefs and Prices
The previous section demonstrated that the feedback e¤ect gives rise to an equilibrium where a speculator buys on good news and does not trade on bad news. the manager knows that such an order ‡ ow ow can stem from a negatively-informed and non-trading speculator. q measures the extent to which information reaches the manager and a¤ects his actions. Put di¤erently. Section 4.assumption (2). Then. our model will have di¤erent implications: the speculator will no longer be reluctant to sell on bad news. q = 1 if = H and q = 0 if = L. reducing …rm value further (as in Goldstein. 1 the posterior q would equal the prior y = 2 . Ozdenoren. in a world of perfect information transmission. or in which the manager did not learn from prices or order ‡ ow. In the BN S equilibrium in the proof of Proposition 1. in order to study the extent to which information reaches the manager and a¤ects real decisions. or customers or employees could terminate their relationship with a troubled …rm (Subrahmanyam and Titman (2001)). lies in between these two polar cases.1
Beliefs
Since the manager uses the posterior belief q to guide his investment decision.
4. Thus. and Yuan (2011)). In this section. and 2 2 considers both Case 1 and Case 2 together. where information is partially revealed through prices. In a world in which no agent observed the state. Conversely. capital providers may withdraw their investment in the low state. and so q (1) is no higher than q (0) (where q (X) denotes the posterior at t = 1 upon observing order ‡ X). Section 4. 2 Thus far. The absolute distance between q and 1 measures the extent to which information reaches the manager. and will decrease his posterior accordingly. even though ow bad news can lead to a positive order ‡ of X = 1. the manager recognizes that X = 1 could be consistent with a negatively-informed speculator who chooses not to trade. enabling her to pro…t more on her short position.1 calculates the e¤ect of good and bad news on the state on the posterior beliefs q. since the information will reduce …rm value further.2 analyzes the impact of news on prices to generate stock return predictions. For example. However. although negative information does not cause a negative order ‡ ow (on average). The analysis that follows focuses on the BN S equilibrium where investment is ex-ante desirable ( < 1 ) and there is feedback ( 1 < ). we study the implications coming out of this equilibrium. Our model. it can still have a negative e¤ect on beliefs and be fully conveyed to the manager.

2
(i) If = H. Proof. since in this case there is an agent in the economy who directly receives news on the state.e. (iii) We have q +q = 1 : thus. the manager’ expected s 1 1 posterior probability of the high state is q L. the manager’ expected posterior probability of the high state is q H.. The results are given in Proposition 5 below: Proposition 5 (Asymmetric e¤ect of positive and bad news on beliefs at t = 1. See Appendix A. Indeed.e. This result holds s 24
.spec = 6 3 + 3 and is decreasing in . the impact on beliefs of good news is greater in absolute terms than the impact of bad news. the absolute increase in the manager’ posterior if the speculator receives good news exs ceeds the absolute decrease in his posterior if the speculator receives bad news. “bad news” is de…ned analogously. conditional upon the speculator being present. The goal is to investigate the extent to which the speculator’ good and bad news is s conveyed to the manager at t = 1. This is a direct consequence of the law of iterated expectations: the expected posterior belief must be equal to the prior.spec y > 2 2 0. this analysis studies the impact of the speculator receiving information about the state. Of greater interest is to study the e¤ect of the state realization conditional upon the speculator being present.Thus. Lemma 3 Consider the BN S equilibrium where <
1 2
and
1 2
<
(i. if we do not condition on the presence of the speculator.spec + q L. (ii) If = L. the realization of 2 2 state H has the same absolute impact on beliefs as the realization of state L. the manager’ expected posterior probability of the high state is q H = (1 3 ) + 1 + 3 s 6 3 and is increasing in . We use the term “good news” to refer to = H being realized and the speculator being present. i.. there is feedback). Since 6 3 > 1 .spec = 3 s and is independent of . (15) implies that abs q H. the manager’ expected posterior probability of the high s H L 1 state is q L = 61 3 + 3 and is decreasing in . Proposition 5 shows that. (ii) If = L and the speculator is present. he can “undo” s the asymmetry. The di¤erence is decreasing in . we start by showing that. the e¤ects on beliefs of the high and low states being realized are symmetric. it may seem still possible for good and bad news to be conveyed symmetrically to the manager –by taking into account the speculator’ asymmetric trading strategy. The asymmetry is monotonically decreasing in the frequency of the speculator’ presence . Proof.
(15)
which is decreasing in .) Consider the 1 BN S equilibrium where < 2 and 1 < (i. (iii) We have 1 + 61 q H.spec = 2 2
1+
1
3
.e. (i) If = H and the 2 2 speculator is present.spec y abs q L. See Appendix A. there is feedback). While the above analysis studied the e¤ect of the state being realized (regardless of whether the state is learned by any agent in the economy).

As we will show in Section 4. This can be seen by 1 plugging = 1 into equation (15). and longrun drift between t = 1 and t = 2.e. there is feedback).2. Consider the BN S equilibrium where < 1 and 1 < (i.
4. it seeps out to a greater extent ex post.
25
.2
Stock Returns
We now calculate the impact of the state realization and news on prices. he is unable to distinguish the case of a negatively-informed (and non-trading) speculator from that of an absent speculator (i. it is a necessary and su¢ cient condition for bad news to have a smaller e¤ect on the manager’ belief than good s news.spec abs 1 q H. bad news causes a greater change in the posterior between t = 1 and t = 2 than good news. Follows from simple calculations The smaller e¤ect of bad news on the posterior at t = 1 is counterbalanced by its larger e¤ect at t = 2. ow even though the manager is rational. Since bad news is conveyed to the manager to a lesser extent at t = 1. Thus.even though the manager is rational and takes into account the fact that the speculator trades asymmetrically when using the order ‡ to update his prior. 1g with uniform probability. We study short-run stock returns between t = 0 and t = 1. Thus. s Proof. between t = 1 and t = 2. no information) – both of these cases lead to the order ‡ being f 1. i. By contrast. we will show that not all the results remain the same. the s state is realized and the posterior becomes either 1 (if = H) or 0 (if = L).e. The absolute impact 2 2 on beliefs between t = 1 and t = 2 of the realization of the state is greater for the low state = L than for the high state = H. in order to generate stock return implications. if the speculator is always present.1 but studying prices rather than beliefs. This result is stated in Corollary 1 below: Corollary 1 (Asymmetric e¤ect of high and low state realization on beliefs at t = 2). While this analysis is similar to Section 4. the manager has no such inference problem and there is no asymmetry. At t = 2.spec y = abs q L. abs 0 q L.e.spec > 0.. The above analysis considered the change in the manager’ posterior at t = 1.spec y . Just as < 1 was a necessary condition for the asymmetric feedback equilibrium to be the only equilibrium in the …rst place. surprisingly this result need not hold when we examine the e¤ect of news on prices rather than posteriors. The source of the result is that.
The asymmetry is monotonically decreasing in the frequency of the speculator’ presence . negative ow information has a smaller e¤ect on his belief. 0. in which case the average posterior equals the prior of 2 and so we have abs q H.

The price at t = 0 is an unbiased expectation of the t = 1 expected price in the high state and the t = 1 expected price in the low state. there is feedback): 2 2 (i) If = H and the speculator is present.spec 1 p0 = 1 (1 3 ) (p (2) p ( 1)) > 0. the average return between t = 0 and t = 1 is L. unconditionally. the absolute e¤ect of the high state must equal the absolute e¤ect of the low state. is positive: pspec 1 p0 = 11 3 2 (p (2) 26 p ( 1)) > 0: (17)
. This is an inevitable consequence of market e¢ ciency.1
Short-Run Stock Returns
Even though the speculator trades asymmetrically.
Proof.e. We have pH p0 = pL p0 : the negative e¤ect of the low state equals the positive 1 1 e¤ect of the high state.
1 Lemma 4 Consider the BN S equilibrium where < 2 and 1 < (i. (iv) The average return. conditional on the speculator being present.) Consider the BN S equilibrium where < 1 and 1 < (i.e. See Appendix A. the stock price increase upon good news exceeds the stock price decrease upon bad news. before the state has been realized. there is feedback): 2 (i) The stock price impact of the high state being realized is pH p0 = 6 [p (2) p ( 1)] > 0.spec p1 p0 = 6 (p ( 1) p (2)) < 0.spec p0 = 1 1 2 (p (2) p ( 1)) > 0.2. this need not imply that realizations of the high and low states will have a di¤erential price impact. Lemma 4 is s analogous to Lemma 3 and shows that.e. Proposition 6 is analogous to Proposition 5 and shows that. This di¤erence is decreasing in . 1 3 (ii) If = L and the speculator is present. An uninformed investor cannot trade the stock at t = 0 and expect a non-zero average return at t = 1.. the good and bad states have the same absolute impact on prices. (16)
i. Let p0 denote the “ex ante” stock price at t = 0..spec 1 p0 abs pL. good news has a greater e¤ect than bad news: Proposition 6 (Asymmetric e¤ect of positive and bad news on returns between t = 0 and t = 1. the average return between t = 0 and t = 1 is pH. since the market maker takes into account the speculator’ trading strategy when devising his pricing function. 1 (ii) The stock price impact of the low state being realized is pL p0 = 6 [p ( 1) p (2)] < 1 H 0= p1 p0 . Since both states are equally likely. Thus. the unconditional expected return is zero. (iii) The di¤erence in the absolute average returns between the speculator learning = H and = L is given by: abs pH. conditional on the speculator being present.4.

Note that the positive average return given in part (iv) is not inconsistent with market e¢ ciency.spec = 1 1 i R 3 H
i RL
1 3
1 2
n (RH
n RL ) . 1 Consider the BN S equilibrium where < 2 and 1 < (i. even though the market maker is rational. conditional on the speculator being present.spec 2 pL. i. Corollary 2 (Asymmetric e¤ect of positive and bad news on returns between t = 1 and t = 2). to analyze the stock return analog of Corollary 1. equations (16) and (17) become zero and there is no asymmetry. the stock price increase upon positive information exceeds the stock price decrease upon negative information (part (iii)). which is private information.spec 2 pH.e.
27
. Proposition 6 states that the average return. the average return between t = 1 and t = 2 is i i pH. See Appendix A. If = 1.spec = 1 1 3 1 2
n (RL n RH ) . 2 1 3 (ii) If = L and the speculator is present.e.spec 1 abs pL. no information).spec 2 pL. there is feedback): 2 (i) If = H and the speculator is present. conditional on the speculator being present. Corollary 2 below shows that this need not be the case for returns: it is possible for bad news to have a smaller e¤ect than good news at both t = 1 and t = 2.2 Long-Run Drift
We now move from short-run returns to calculating the long-run drift of the stock price.spec pH. the average return between t = 1 and t = 2 is pL. the key to this result is that. An uninformed investor cannot buy the stock at t = 0 and expect to earn a positive return at t = 1 because she will not know whether the speculator is present.. (iii) If (18) < 0. positive information is impounded into prices to a greater degree than negative information.
(18)
which is negative in Case 1. because it is conditional upon the speculator being present.spec = 1 (RH RL ) > 0.. is positive (part (iv)).e. the asymmetry is monotonically decreasing in . As with Proposition 5. 4. is positive –i. Proof.This di¤erence is decreasing in . this means that the average return. Corollary 1 showed that the smaller e¤ect of bad news on beliefs at t = 1 is counterbalanced by a larger e¤ect on beliefs at t = 2. the di¤erence in the absolute average returns between the speculator learning = H and = L is given by: abs pH. Since good and bad news are equally likely. but can be positive or negative in Case 2. he is unable to distinguish the case of a negatively-informed speculator from that of an absent speculator (i. the impact of the state realization on prices between t = 1 and t = 2. Put di¤erently.2.e.

See Appendix A. conditional upon the speculator being present. which means that there is a large amount of bad news to come out between t = 1 and t = 2. there is an additional e¤ect of the speculator on prices that s does not exist in the analysis of beliefs: not only does she convey information about the state. thus. but also this information improves the manager’ decision-making and enhances …rm value – s n n the essence of the feedback e¤ect. this feedback e¤ect is su¢ ciently strong to turn the average return between t = 1 and t = 2 positive. However. little bad news emerges about the state between t = 0 and t = 1. Indeed. 2 2 3 6 and the average return between t = 1 and t = 2 if the speculator is present is: pspec 2 pspec = 1 11 62
n (RL n RH ) . analogous to Corollary 1. By contrast. good news has a larger e¤ect on beliefs at t = 1 than bad news. i. This is because the stock price depends not only on the beliefs about the state. the expected change in beliefs between t = 0 and t = 1 is positive.which is positive in Case 2 and negative in Case 1. if the speculator is present. it must have a larger e¤ect at t = 2 (since the truth about the state comes out at t = 2). is : 1 i 1 n 1 i pspec = RH + RL + RL . if the n n feedback e¤ect is su¢ ciently strong. Since state L is bad for …rm value regardless of whether the manager takes the corrective action or not. In state L. this in turn leads to the large downward revision in beliefs in Corollary 1. in Case 1. The magnitude of the di¤erence is decreasing in .
which is positive in Case 2 and negative in Case 1. the long-run drift to the low state is larger in magnitude. Proof. but also the manager’ action. Corollary 1 showed that. Since bad news has a smaller e¤ect at t = 1. Thus. the e¤ect on prices in Corollary 2 is muted because the damage to …rm value caused by state L can be mitigated by taking the corrective action. RL is much higher than RH . (iv) Expected …rm value at t = 2. because she trades on the former but not the latter. the return to bad news n n between t = 1 and t = 2 can be positive ((18) > 0). the realization of state L at t = 2 leads to a large decrease in prices.e. the negative e¤ect of bad news is smaller than the positive e¤ect of good news between t = 1 and t = 2 as well as between t = 0 and t = 1 due to the feedback e¤ect. Thus. Corollary 2 shows that this need not be the case when we study prices rather than beliefs: the speculator’ presence can lead to positive average returns s in both the short-run (between t = 0 and t = 1) and also in the long-run (between t = 1 and t = 2).
28
. (RH > RL ). In Case 2 (RH < RL ). The magnitude of the di¤erence is decreasing in . and so the average increase in beliefs between t = 0 and t = 1 is reversed by an average decrease in beliefs between t = 1 and t = 2.

to a greater degree at t = 1) than the returns to a bad investment. See Appendix A. the price impact of a good investment is more front-loaded than for a bad investment. Parts (i) and (ii) of Proposition 7 show that investing in the high state leads to both positive short-run returns between t = 0 and t = 1 and also positive long-run drift between t = 1 and t = 2. compared to a bad investment.e. the returns to a good investment manifest more rapidly than the returns to a bad investment. Thus.H = 1
i i RH RL > 0: 2+
(20)
(ii) If investment is undertaken in state L: (iia) The average return between t = 0 and t = 1 is 1 [(1 6
i ) RH n i RH + RL n RL ] < 0:
(21)
(iib) The average return between t = 1 and t = 2 is
i RL
pi.e. Proposition 7 (Faster incorporation into prices of good investment than bad investment.H = 1
1 i (R 2 H
i RL ) < 0:
(22)
(iii) The returns to a good investment manifest more rapidly (i.2. Proof. are more front-loaded. i...The analysis thus far has considered the impact of news on prices at t = 1 and t = 2. Part (iii) demonstrates that the returns to a good investment are realized to a greater extent at t = 1 rather than t = 2. In other words. To our knowledge. speci…cally.
29
.e.. We now consider the impact of investment (a real variable) on prices. While Section 4.1 showed that good news received by the speculator has a greater short-run price impact than bad news. the extent to which it is impounded into prices at t = 1 or at t = 2. this prediction has not yet been tested. Investing in the low state leads to negative short-run returns and negative long-run drift. Proposition 7 now demonstrates a related result: the proportion of the total returns to an investment that is realized in the short-run (at t = 1) rather than the long-run (at t = 2) is greater for a good investment ( = H) than a bad investment ( = L)..H 1 p0 = 1 [(2 + 2 6(2 + )
2 i )RH + (2 i 2 )RL
(2
2
n )RH
n (2 + )RL ] > 0: (19)
(ib) The average return between t = 1 and t = 2 is
i RH
pi. abs ((22) (21)) > abs ((20) (19)).) 1 Consider the BN S equilibrium where < 2 and 1 < (i. i. there is feedback): 2 (i) If investment is undertaken in state H: (ia) The average return between t = 0 and t = 1 is pi.e.

In Proposition 6. if investor sophistication and information improve over time. Here. investor irrationality or investors’limited information on the quality of a portfolio manager. instead. the negative returns must manifest predominantly at t = 2. and this in s turn a¤ects the speculator’ decision to trade. the intuition is as follows. which will likely diminish with the development of …nancial markets.
5
Summary of Implications
This section discusses several implications of our model. the valuedestructiveness of the investment seeps out ex post. Put di¤erently. Note that the long-term drift in returns does not violate market e¢ ciency. which means that the market maker cannot distinguish the case of a negatively-informed speculator from that of an uninformed speculator. There is no exogenous friction preventing the arbitrageur from trading. this force will also diminish. which leads to mutual funds avoiding arbitrage trading that will only converge in the long run (Shleifer and Vishny (1997)). the market maker knows that the speculator is s pursuing an asymmetric trading strategy. In both Propositions 6 and 7. the asymmetry occurs because the low state has a lesser impact on prices than the high state. This is a fundamental force that does not rely on short-sale constraints. Either way. If anything. If the investment is bad. (The only market imperfection that our model requires is trading costs. the limit to arbitrage may increase with investor sophistication. this arises from the fact that < 1. as this augments the extent to which speculators have value-relevant 30
. there are bad investments that do not lead to a sharply negative reaction at t = 1 because the speculator did not trade on the bad news. or rational if investors have limited information on the fund manager’ quality but instead must infer it s imperfectly from short-run performance. the limit to arbitrage analyzed by this paper stems from …rm value being endogenous to the act of arbitrage. If speculators sold aggressively in response to a bad investment. A second is that investors in professional money managers make their allocation decisions based on short-run measures of performance. is likely to persist over time even as markets evolve and investors become more sophisticated. she endogenously chooses not to trade because of the feedback e¤ect. By contrast.The intuition behind the asymmetry is di¤erent from Proposition 6. and so may continue to persist over time. the decline in the stock price will lead to the investment being cancelled. the negative returns cannot manifest too strongly at t = 1. The key to reconciling this result with market e¢ ciency is that …rm value is endogenous to trading. The …rst is that this paper identi…es a limit to arbitrage which. in contrast to alternative explanations. Thus. otherwise the decline in the stock price will have led to the investment being canceled. Instead. One existing source of limited arbitrage is market frictions such as short-sales constraints. the negative returns must manifest predominantly at t = 2. Such behavior can either be irrational over-extrapolation. All agents in the model act with full rationality: the market maker takes into account the manager’ learning when setting the price. which exist even in developed …nancial markets). Thus.

The second main category of applications stems from the fact that the limit to arbitrage is asymmetric. but by trading on it. leading to negative long-run returns as documented by Agrawal. leading to overinvestment –even though there is an agent who knows
Note that our paper focuses on the e¤ect of news on stock prices. Lim. s the manager can take a corrective action to mitigate the negative impact of the state on …rm value. the feedback e¤ect means that the lack of negative information in prices will have further consequences on real decisions. While the speculator buys on good information. and Stein’ results are consistent with the Hong and Stein (1999) model s that news travels more slowly in small …rms with low analyst coverage. who disseminate information not s through public news releases. Key information is held by a …rm’ investors. Our theoretical model o¤ers a potential alternative explanation.information which the manager attempts to learn by observing the price. Hong and Stein do not predict an asymmetry between good and bad news. in some cases. Even though less bad news is transmitted to the market at t = 1. and Stein speculate that the asymmetry arises because key information is held by the …rm’ managers. and they dissems inate favorable information more enthusiastically than unfavorable information because they are evaluated according to the stock price. While Hong. it must be counterbalanced by a larger long-run drift. In particular. Lim. By contrast. stock returns depend not only on the state. Proposition 6 shows that negative information will enter into prices more slowly.8 Hong. even though the market maker takes the asymmetric trading volume into account. Moreover. if bad news has a smaller e¤ect on short-run returns than good news. in an underreaction model. Lim. because the speculator trades more readily on good news than bad news. While the above results are unconditional on investment occurring. Corollary 2 shows that. and Mandelker (1992) and Rau and Vermaelen (1998).g. Karpo¤ (1987)). Proposition 7 shows that the returns to good investment are more front-loaded than the returns to a bad investment.. so the e¤ect of bad news on returns is lower at t = 2 as well as at t = 1. meaning that there is more to come out at t = 2. e. Thus. but also the manager’ decision. This prediction has implications for trading volume. and Stein (2000) …ndings. some negative-NPV projects will not s be optimally abandoned. Lim. the model also generates implications for the short. the value-destructiveness of a bad investment seeps out to a greater extent ex post. Another di¤erence is that. and Stein (2000). if speculators choose not to trade on negative information. Moreover. Such a relation is consistent with the well-documented positive correlation between trading volume and stock returns (see.and long-run returns to investment. Thus. which is another component of the Hong. but due to the limit to arbitrage created by the feedback e¤ect. then such negative information does not become incorporated into stock prices and fails to in‡ uence the manager’ behavior. Ja¤e. suggesting that volume should be higher upon good investments (such as M&A or capital expenditure) than bad investments. In our feedback model.
8
31
. It does not address the predictability of future returns from past returns. as found empirically by Hong. bad news generates a smaller e¤ect on returns in both the short-run and long-run. she does not sell on bad information. Their reluctance to disseminate bad news is not because they are evaluated according to the stock price.

While he …nds that some transactions are canceled in equilibrium.with certainty that the investment is undesirable.g. Interestingly. The manager may then take a corrective action that improves …rm value but reduces the pro…ts from her short position below the cost of trading. The asymmetry of our e¤ect has implications for both stock returns and real investment. she may strategically refrain from trading on it. Critically. The above overinvestment result can apply to M&A as well as organic expansion. but is unable to do so since speculators refrain from impounding their information into prices. Trading in either direction impounds information into prices. even though the market maker is rational and takes the speculator’ trading strategy into s account when devising his pricing function. since the manager can take a corrective action to attenuate the negative e¤ect of the state on …rm 32
. unlike limits to arbitrage based on short-sale constraints. the smaller short-run reaction to bad news may also coincide with smaller long-run drift. which improves the manager’ decision-making and increases fundamental value. Even if a speculator has negative information on the state. as in the standard theories of Jensen (1986). overinvestment does not occur because the manager is pursuing private bene…ts. This increases the pro…tability of s a long position but reduces the pro…tability of a short position. we have d = i if the noise trader does not sell.)
6
Conclusion
This paper has modeled a limit to arbitrage that stems from the fact that …rm value is endogenous to the act of exploiting the arbitrage. In contrast. the manager is fully aligned with …rm value and there are no agency problems. thus encouraging buying on good news but discouraging selling on bad news. our model does not rely on exogenous forces or agency problems. holding costs or portfolio delegation. Overinvestment occurs even though the manager is fully aware that the speculator does not trade on negative information and takes this into account.. the e¤ect is generated endogenously as part of the arbitrage process. The manager wishes to maximize …rm value by learning from prices. In the model. noise trader risk. our model suggests that there are other negative-NPV deals that should optimally be canceled but are not because speculators do not impound their negative views into prices. instead. Luo (2005) shows that managers sometimes use the market reaction to announced M&A deals to guide whether they should cancel the acquisition. Stulz (1990) and Zwiebel (1996). even if = L. bad news has a smaller e¤ect on short-run prices than good news. unlike limits to arbitrage based on fundamental risk. or holding costs. in contrast to underreaction models. This may explain why a large proportion of M&A deals destroy value (see. First. Andrade. it still takes place. and the limits to arbitrage identi…ed by prior literature. our e¤ect is asymmetric. Mitchell and Sta¤ord (2001). Second. because doing so conveys her information to the manager. e. and sometimes causes her to realize a loss. There are several important di¤erences between the feedback-driven limit to arbitrage that we study. In terms of stock returns.

and the manager is aware of the speculator’ asymmetric trading strategy. this information is not s conveyed to the manager and so the project is not abandoned. even though there are no agency problems and he is attempting to learn from the market to take the e¢ cient decision.value. the manager may overinvest in negative-NPV projects.
33
. the valuedestructiveness of a bad investment seeps out ex post. Even though there is an agent in the economy who knows with certainty that the investment is undesirable. the returns to a good investment are more front-loaded than the returns to a bad investment –since the speculator does not trade on negative information. In terms of real investment. In addition.

Noise Trading and Stock Price Behavior.” Working paper. 187– 206. [6] Campbell. James. Reed (2002). [2] Agrawal. David K. Philippe and Jeremy C. Stein (2008): “Growth versus Margins: Destabilizing Consequences of Giving the Stock Market What It Wants. Handbook of Economics of Finance (Elsevier/North-Holland. Milton Harris and René Stulz (eds). New York).”Journal of Finance. 781– 820. 619– 650. Alex.”Journal of Political Economy 98. and Alexander Guembel (2010): “Incentives for Information Production in Markets where Prices A¤ect Real Investment. and Wei Jiang (2011): “The Real E¤ects of Financial Markets: The Impact of Prices on Takeovers. Itay Goldstein. 1– 34. [7] Chakraborty. [5] Bond.”Journal of Finance 63. [13] Geczy. [3] Andrade.”Journal of Financial Economics 66. and Robert Waldmann (1990): “Noise Trader Risk in Financial Markets. and Edward S. Je¤rey F. Bradford. Kyle (1993): “Smart Money. In-Koo and David M. 179– 221 [10] DeLong.”Journal of Finance 47.References
[1] Aghion. 703– 738. Anup.”Review of Financial Studies 23. [8] Chen. Kreps (1987): “Signaling Games and Stable Equilibria. Musto. Prescott (2010): “Market-Based Corrective Actions. 241– 269. Gregor.”Review of Economic Studies 60. [4] Barberis. University of Pennsylvania. and Gershon N.”Review of Financial Studies 20. forthcoming. Mark Mitchell.”Journal of Economic Perspectives 15. Itay Goldstein. Lawrence Summers. Itay Goldstein. and Adam V. [12] Edmans. John and Albert S. Andrei Shleifer. Philip. Qi. Christopher. 1025– 1058. 1605– 1621. [11] Dow. Nicholas and Richard Thaler (2003): “A Survey of Behavioral Finance.”Journal of Financial Markets 7.
34
. [9] Cho.” in George Constantinides. and Erik Sta¤ord (2001): “New Evidence and Perspectives on Mergers. “Stocks Are Special Too: An Analysis of the Equity Lending Market. Ja¤e. Itay Goldstein. 103– 120. Archishman and Bilge Yilmaz (2004): “Manipulation in Market Order Models. Mandelker (1992): “The Post-Merger Performance of Acquiring Firms: A Re-Examination of an Anomaly.”Quarterly Journal of Economics 102. and Wei Jiang (2007): “Price Informativeness and Investment Sensitivity to Stock Price.

(1985): “Continuous Auctions and Insider Trading. [15] Goldstein. [20] Kondor.”Journal of Finance 55.”Review of Economic Studies. and Kathy Yuan (2011): “Learning and Complementarities in Speculative Attacks. Analyst Coverage and the Pro…tability of Momentum Strategies. (1987). [16] Hong. Owen. [18] Jensen.”Journal of Political Economy 111. Itay. [17] Hong. Stein (2000): “Bad News Travels Slowly: Size. Stein (1999): “A Uni…ed Theory of Underreaction. Emre Ozdenoren. Harrison and Jeremy C.”Journal of Financial and Quantitative Analysis 22.”Journal of Financial Economics 13.” Econometrica 53. [22] Lamont.[14] Goldstein. 133– 64. Je¤rey (1996): “Costly Arbitrage: Evidence from Close-End Funds.”Journal of Finance 64. Harrison. and Richard Thaler (2003): “Can the Market Add and Subtract?: Mispricing in Tech-Stock Carve-Outs. [21] Kyle. [19] Karpo¤. 187– 221 [27] Ponti¤.
35
. 1135– 1151. M. forthcoming. Yuanzhi (2005): “Do Insiders Learn from Outsiders? Evidence from Mergers and Acquisitions. Itay. 631– 655. [23] Luo. Momentum Trading and Overreaction in Asset Markets.”Journal of Finance 57. and Erik Sta¤ord (2002): “Limited Arbitrage in Equity Markets. 323– 329. 1315– 1336. Peter (2009): “Risk in Dynamic Arbitrage: The Price E¤ects of Convergence Trading. (1986): “The Agency Costs of Free Cash Flow: Corporate Finance and Takeovers. 551– 584. Albert S. 2135– 2176. Stewart C. Mark. Michael C. 227– 268. Todd Pulvino. and Alexander Guembel (2008): “Manipulation and the Allocational Role of Prices.”Review of Economic Studies 75. “A Relation between Price Changes and Trading Volume: A Survey.” Quarterly Journal of Economics 111. Terence Lim and Jeremy C. Jonathan.”Journal of Finance 60. [25] Mitchell. and Todd Pulvino (2001): “Characteristics of Risk and Return in Risk Arbitrage. 1951– 1982 [24] Mitchell. [26] Myers. 109– 126. Mark. 2143– 2184. and Nicolas Majluf (1984): “Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have.”Journal of Finance 54.”American Economic Review 76.”Journal of Finance 56. 265– 296.

[30] Stulz. P.
36
.” Journal of Finance 52. 1197– 1215. Value and the PostAcquisition Performance of Acquiring Firms. Avanidhar and Sheridan Titman (2001): “Feedback From Stock Prices to Cash Flows. [31] Subrahmanyam. 3– 26. 223– 253. 2389– 2413. [32] Zwiebel.[28] Rau.” Journal of Financial Economics 49. Raghavendra and Theo Vermaelen (1998): “Glamour. Je¤rey (1996): “Dynamic Capital Structure Under Managerial Entrenchment.”Journal of Finance 56. Andrei and Robert W.” Journal of Financial Economics 26.” American Economic Review 86. [29] Shleifer. Vishny (1997): “The Limits of Arbitrage. René M. (1990): “Managerial Discretion and Optimal Financing Policies. 35– 55.

i i W. No Trade Equilibrium NT. X = 1 and she pays 1 RH + 1 RL per share. her expected gross pro…t is given by: 11 i R 32 H
i RL +
11 i R 32 H
i RL =
1 i R 3 H
i RL =
3:
(23)
A similar calculation shows that.p. Thus. trading pro…ts are zero. The order ‡ ows of X 2 f 1. the no-trade equilibrium is sustainable. her gross pro…t is also given by (23). Thus. 2 and q (0) and q (1) are calculated in exactly the same way. i i W.) 3 . The order ‡ ows of X = 2 and X = 2 are o¤ the equilibrium path and the posteriors are given by 0 and 1.A
Proofs
Proof of Proposition 1 This proof only provides supplementary material to what is in the text. X = 0 and she pays 1 RH + 1 RL per share. as these are the only posteriors that satisfy the Intuitive Criterion (as stated in the main proof). If the positively-informed speculator chooses to deviate from not trading to buying:
1 With probability (w. Sequential rationality leads to the decisions d and prices p as given by the Table. and so her pro…t is 2 (RH RL ) > 0.p.
Thus. 0.p. The fundamental value of each share 3 2 2 1 i i i is RH . The order ‡ of X = 2 is o¤ the equilibrium path and ow
37
. We now turn to calculating trading pro…ts. 1 . if a negatively-informed speculator sells. 1g are observed on the equilibrium path and so the posteriors can be calculated by Bayes’rule: q(X) = Pr(HjX) = We thus have: q( 1) = (1=3) + (1 )(1=3) (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) Pr(XjH) : Pr(XjH) + Pr(XjL)
1 = . The fundamental value of each share 3 2 2 1 i i i is RH . X = 2 and she is fully revealed. if and only if 3 . respectively. Partial Trade Equilibrium SNB. and so her pro…t is 2 (RH RL ) > 0. 1 .

X = 1 and she receives 1 RH + 2 RL per share. 3 .
1 i i W.p.p. 2
Thus. X = 0 and she receives 1 RH + 1 RL per share. and so her pro…t is 2 (RH RL ) > 0. 2 i i W. The order ‡ of X = ow
2
<
3. 1 . X = 1 and she pays 2 1 RH + 1 RL per share. 1 . The posteriors of the other order ‡ ows are given as follows: q( 2) = q ( 1) = 0 =0 (1=3) 1 2 1 2
(1=3) + (1 )(1=3) = (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) (1=3) + (1 )(1=3) = q (0) = (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) (1=3) + (1 )(1=3) 1 q(1) = = : (1=3) + (1 )(1=3) + (1 )(1=3) 2
Under this equilibrium. Thus. The fundamental value of each 3 2 i i i share is RL . and so her pro…t is 1 (RH RL ) > 0.p. X = 2 and she is fully revealed.p. 2 i i W. 1 . 3 .the posterior is given by 1.
Thus.
Thus. and so her pro…t is 1 (RH RL ) > 0. 1 i i W. Thus. the negatively-informed speculator sells. trading pro…ts are zero. X = 3 2 and she is fully revealed.p.p. trading pro…ts are zero. her expected gross pro…t is given by: 1 i R 3 H
i RL = 3:
If the positively-informed speculator deviates to buying:
1 W. The fundamental value of each share 3 2 2 1 i i i is RH . X = 0 and she pays 1 RH + 1 RL per share.
2 is o¤ the equilibrium path and
38
. The fundamental value of each 2 i i i share is RH . 1 . W. The fundamental value of each 3 2 2 i i i share is RL . the SNB equilibrium is sustainable if and only if Partial Trade Equilibrium BNS. her expected gross pro…t is given by: 1 3 1 1 + 2 2
i RH i RL = 2. and so her pro…t is 1 (RH RL ) > 0.

this equilibrium is sustainable if and only if
1
<
Trade Equilibrium T. buying on good information) yields pro…ts of 3 . Hence. her expected gross pro…t is given by: 1 3 1 i R 2 H
i RL +
1 2
n (RH
n RL )
=
3. The fundamental value 3 n of each share is RL because the manager is now taking the corrective action. the manager now takes decision d = n upon observing order ‡ X = 1. 1 . For the sub-case of feedback ( 2 < ). All order ‡ ows are on the equilibrium path and so the posteriors are
39
.p. 1 . The fundamental value of each 3 2 2 i i i share is RL . X = 0 and she receives 1 RH + 1 RL per share. In the sub-case of no feedback ( 1 > ). Thus. analogous to the SNB equilibrium. The pro…ts from deviating to selling on negative information are now given as follows: W. The posteriors of the other order ‡ ows are given as follows: q ( 1) = (1 )(1=3) 1 = (1 )(1=3) + (1=3) + (1 )(1=3) 2 (1=3) + (1 )(1=3) 1 q (0) = = (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) 2 1 (1=3) + (1 )(1=3) = q(1) = (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) 2 (1=3) = 1: q(2) = (1=3)
There are two sub-cases to consider. trading pro…ts are zero. 1 . 2
Thus. and so her pro…t is 1 (RH RL ) > 0. 2 i i W.p.the posterior is given by 0.
1. informed trading (in this case.p. X = 1 and she receives 2 RH + 2 1 RL per share. this 1 equilibrium is sustainable if and only if 2 < 3 . and so her n n pro…t is 1 (RH RL ) > 0. X = 3 2 and she is fully revealed. Thus. if the negatively-informed speculator deviates to selling. decision 2 d = i is taken for all of the order ‡ ows on the equilibrium path.
1 n n W. she earns pro…ts of 2 .
Thus. The pro…ts from ow trading on positive information are unchanged.

2 Proof of Proposition 3 This proof only provides supplementary material to what is in the text. we require 1 3
i RH i n RL + (RL n RH ) < :
40
. the pro…ts are given by 1 and so the equilibrium is sustainable if and only if < 1 . No Trade Equilibrium NT.p. 2 1 i i i W.given as follows: q( 2) = q ( 1) = 0 =0 (1=3)
(1 )(1=3) 1 = (1 )(1=3) + (1=3) + (1 )(1=3) 2 1 (1=3) + (1 )(1=3) = q (0) = (1=3) + (1 )(1=3) + (1=3) + (1 )(1=3) 2 (1=3) + (1 )(1=3) 1 q(1) = = (1=3) + (1 )(1=3) + (1 )(1=3) 2 (1=3) = 1: q(2) = (1=3)
The pro…ts from buying on positive information are given by 2 .p. as in the SN B equilibrium.
Thus. With feedback ( 1 > ). to derive the necessary conditions to prevent such a deviation. X = 1 and she receives 2 RH + 1 RL for a share that is worth RH . X = 0 and she receives 1 RH + 1 RL for a share that is worth RH . As discussed in the main text. 1 i i i W. 1 . which yields a pro…t n n of (RL RH ) > 0.
For the positively-informed speculator not to deviate to selling. the 2 pro…ts are given by 2 and so the equilibrium is sustainable if and only if < 2 . 3 . If the positively-informed speculator deviates to selling:
1 n n W. which correspond to the two sub-cases in the BN S equilibrium.p. which yields a 2 2 1 i i pro…t of 2 (RL RH ) < 0. Without feedback ( 1 > ). which yields 3 2 i i a pro…t of 1 (RL RH ) < 0. Here we calculate the pro…ts made if the positively-informed speculator deviates to selling. it is straightforward to show that the negatively-informed speculator will not deviate to buying. X = 2 and she receives RL for a share that is worth RH . 3 . For the pro…ts from selling on negative information. her overall pro…ts are given by 1 n (R 3 L
n RH ) +
1 i R 3 L
i RH . there are two sub-cases to consider.

she receives 2 RH + 2 1 RL for a share that is worth RH . her pro…ts must be greater under the former. In the case of no feedback. her overall pro…ts are given by 1 1 1 1 n n n n (RL RH ) + (RL RH ) + 3 32 3 1 1 i 1 3 n n (RL RH ) R = 3 2 3 2 H 1 i R 2 L
i RL i RH
. which yields a 2 2 1 i i pro…t of 2 (RL RH ) < 0. X = 2 and she receives RL for a share that is worth RH .
i RH
41
.
In the case of feedback.p.p.p. which 2 i i yields a pro…t of 2 1 (RL RH ) < 0. 3 . Partial Trade Equilibrium BNS. X = 0 and she receives 1 RH + 1 RL for a share that is worth RH . which yields a pro…t of 2 1 (RL RH ) > 0. ow 6 3 i i This yields the condition 6 2 (RH RL ) > 1 in the Proposition. 1 . This requires: 1 i R 3 H 3 i R 2 H
i RL > i RL
1 3 3 > 2
3 2
n (RL n (RL
n RH )
1 3
1 i R 2 H
i RL
n RH ) :
The …rst term is the “fundamental” e¤ect. 1 i i i W. she receives 1 RH + 2 1 RL for 3 2 2 n n n a share that is worth RH . In the case of feedback ( 1 < ). n n W. 1 . which arises s because selling manipulates the order ‡ and causes the manager to take the wrong decision. her overall pro…ts are given by 1 n 1 1 n i (RL RH ) + RL 3 32 1 n 1 4 n = (RL RH ) 3 3 4 2
i RH + i RH
1 3
i RL
1 i R 2 L . which represents the pro…ts from trading in the direction of one’ private information. In the case of no 1 i i i feedback ( 1 > ).
For the positively-informed speculator to choose buying over selling. If the positively-informed speculator deviates to selling:
n n W. X = 1.The calculations for the Partial Trade Equilibrium SN B are identical. The second term is the “feedback” e¤ect. which yields a pro…t 3 n n of (RL RH ) > 0.

if = H. In the case of feedback. 6 3 3 3 ) 1 1 1 q (0) + q (1) + q (2) 3 3 3
(24)
42
. we require: 1 3 1 1 + 2 2
i RH i 3 2 RH n 3 RL i RL >
1 3
3 2
n (RL
n RH )
1 3
1 i R 2 H
i RL
i RL > 1. we require: 1 3 1 1 + 2 2
i RH i RH 2 n RL i RL >
1 n (R 3 L
n RH )
1 3
4 4 2
i RH
i RL
i RL > 1: n RH
Proof of Lemma 3 For part (i). for the positively-informed speculator to choose buying over selling.For the positively-informed speculator to choose buying over selling. for the positively-informed speculator to choose buying over selling. the pro…ts from buying (that we need to compare against the pro…ts from selling) are di¤erent. However. we require: 1 i 1 n i RH RL > (RL 3 3 i i 8 3 RH RL > 1: n n 4 2 RL RH
n RH )
1 3
4 4 2
i RH
i RL
Trade Equilibrium T. the calculations are exactly the same as in the Partial Trade Equilibrium BN S. the expected posterior is given by: q H = (1 1 1 1 q ( 1) + q (0) + q (1) + 3 3 3 1 1 1 = q ( 1) + q (0) + q (1) + q (2) 3 3 3 3 (1 )2 1 = + + . n RH
In the case of no feedback. If the positively-informed speculator deviates to selling.

the …rst channel is ruled out: conditional on the speculator being present and = H. The variable can a¤ect the expected posterior in two ways: …rst. there is a 1 probability 3 that the order ‡ is X = 1. it can change the relative likelihood of the di¤erent order ‡ ows. and second. X 2 f0. her information is still partially incorporated into prices. 43
. it is not consistent with the speculator observing = H. we have: qL = 1 (q ( 1) + q (0) + q (1)) 3 1 1 + : = 6 3 3
(25)
This is decreasing in . she is more likely to impound her information into prices by trading. and thus the greater the decrease in the market maker’ posterior. the greater the likelihood that X = 1 stems from = L.We have: @q H 1 1 = + @ 3 3 " 1 = 1+ 3 = 1 1 3 2(1 1 2 1 2 )(2 (2
2
2
2
1 2
) + (1 )2 #
)2
> 0: The expected posterior is increasing in : if the speculator is more likely to be present. The greater the likelihood that the speculator is present.spec is independent of . 1.spec = 1 (q (0) + q (1) + q (2)) 3 2 = . Even though the speculator does not trade upon = L if she is present. Moving to part (ii). 3 1 = (q ( 1) + q (0) + q (1)) 3 1 1 = + . we have: q H. Since we are conditioning on the speculator being present. Part (iii) follows from simple calculations. 2g with uniform probability regardless of .spec is decreasing in . This is consistent with the speculator being absent (in which ow case the state may be either H or L) or her being present and observing = L. With = L. but q L. 6 3 3
(26)
q L.spec
(27)
Note that q H. it can change the actual posterior given a certain order ‡ ow. if = L. s Proof of Proposition 5 For parts (i) and (ii).

This is because. which guarantees that X 0 and the investment is undertaken. This is intuitive. 1g is completely uninformative and so the posterior is again independent of . Moreover. the state will be = L and there is no trade. 2g. the …rm @ values in the low state. X 2 f0. 0. 0. 2 regardless of whether the speculator is present. 1. 1 With probability 2 . Part (iii) follows from simple calculations. If the speculator is present. the only posterior that depends on is q ( 1): since X = 1 is inconsistent with the speculator being present and seeing = H. If she is absent. there is a possibility that X = 1. If the speculator is present.spec depends on but q L. @p0 is independent of RL and RL . since she does not trade. This leads the manager to take the suboptimal corrective action. the state will be = H. 1g with uniform probability regardless of . the speculator’ presence s does nothing to help the manager’ decision.conditional on the speculator being present and = L. n i i n reducing …rm value by (RH RH ). the increase in …rm value from taking the e¢ cient continuation @ decision in the high state. there is no trade and we again have X 2 f 1. With probability 1 . and is given as follows: p0 = 2 1 1 = 3 1 = (1 6 1 1 1 p (0) + p (1) + p (2) + 1 3 3 3 2 2 1 1 1 p ( 1) + p (0) + p (1) 3 3 3
p ( 1) + p (0) + p (1) + p (2) 2 (28)
n n i i )RH + RL + (2 + )RH + 2RL :
Even though the initial belief y is independent of . X 2 f0. Thus. the initial stock price p0 is increasing in . Letting p (X) denote the stock price set by the market maker after observing order ‡ X at t = 1. 1g. 1g with equal probability. because the speculator provides information to improve the manager’ decision. the expected price at t = 1 is given by: pH = (1 1 1 1 1 1 1 1 p ( 1) + p (0) + p (1) + p (0) + p (1) + p (2) 3 3 3 3 3 3 1 1 1 = p ( 1) + p (0) + p (1) + p (2) : 3 3 3 3 )
(29)
44
. 2g is fully revealing and so the posterior is independent of . X 2 f 2.p. s For part (i). X 2 f 1. If the speculator is absent (w. only q L. (1 )). it has a particularly negative impact on the likelihood of = H if the speculator is more likely to be present. s @p0 n i is increasing in (RH RH ). ow the price at t = 0 will be the expectation over all possible future prices at t = 1. Since X = 1 can only occur in the presence of a speculator if she has received bad news. if = H is realized. By contrast. she always buys in state H. Turning to the second channel. By contrast. 0. Proof of Lemma 4 We start by calculating p0 .spec does not. if the low state is realized. X 2 f 1.

e.e. pH is increasing in . since the speculator impounds information about the high state into 1 prices. she will buy one share and so the expected price becomes: pH. for the same reasons that q H.spec = 1 1 (p (0) + p (1) + p (2)) : 3 (31)
Unlike pH (equation (29)). @p( 1) @ 1 1 (2 )2
n RH
i. 1 1
Proof of Proposition 6 For part (i). The calculations of pH p0 and pL p0 follow automatically. if the speculator receives positive information.spec (equa1 tion (26)) is independent of . if = L is realized.spec 1 p0 = 1 1 (p (0) + p (1) + p (2)) 3 3 1 = 1 (p (2) p ( 1)) 3 2 1 1 i n = 1 RH RH 3 2 2 @ pH.spec 1 1 p ( 1) + p (0) + p (1) + p (2) 2
2
1 2
n RL
> 0. If the speculator is more likely to be present. the price is higher if and only if …rm value is higher in this n n state. Thus. this is independent of . Turning to part (ii). The stock return realized when the speculator receives good information is thus given by: pH.. whereas the stock return not conditioning on the speculap0
45
. i. then X = 1 is more 2 likely to result from = L. the expected price at t = 1 is given by: pL = 1
@pL Rn Rn
1 (p ( 1) + p (0) + p (1)) : 3
(30)
L We have @ 1 = 3(2 )H . RL > RH (Case 2).Note that: @pH 1 1 1 1 = p(2) p( 1) + @ 3 3 3 1 1 i = RH Rn 3 (2 )2 L 1 i n > [RH RL ] 3 > 0.
(32)
and we have 1 n i = [RH RH ] < 0: @ 6 Equation (32) is decreasing in .

The project will i n be cancelled if the noise trader sells. This yields: 2 pH.spec = RH 1
1 (p (0) + p (1) + p (2)) 3
i RL .
=
1 i R 3 H
Moving to part (ii).spec 1 p0 = = 1 (p ( 1) + p (0) + p (1)) 3 6 (p ( 1) p (2)) = pL 1 1 3 p0 < 0. We thus have pH. she will not trade. For part (ii).spec = RH . (32). if the speculator receives bad news. which is smaller that RH . else it will be continued. We thus have pL. 1 3 (33)
Parts (iii) and (iv) follow from simple calculations.spec = 1 and pL.tor’ presence. If the speculator receives good news. both equations (16) and (17) are decreasing in . since RH > RH and RH > RL . we have a convex n i n i i n n combination of RH and RL . conditional on the speculator being present) become: 1 1 i n n (1 ) RH RH RL : 2 2 Di¤erentiating with respect to
i RH +
gives: 1 2
n RL + (1
1 2 3
n RH +
)
1 (2
)
n 2 RH
1 (2 )2
n RL
=
i RH +
4 + 2 n 1 n 2 RH + 2 RL (2 ) (2 )
n n The coe¢ cients of RH and RL are positive and add up to one. 2 3 3
46
. This is because p0 is increasing in .spec is s 1 independent of . was increasing in . Dropping constants. if the speculator is present and receives negative information. both equation (16) (the asymmetry between the price impact of good and bad news) and equation (17) (the average return. That is. we have: pL. 1 2 p ( 1) + p (0) + p (1) + p (2) 2 1 (p ( 1) + p (0) + p (1)) = pL . she will buy and so the project i will always be undertaken. but pH.spec 2
i pH.
Proof of Corollary 2 We start with part (i). Thus.spec = 2 RL + 1 RL .

the speculator is present and buys. we …rst calculate pi. 2 2 3 6 and so we have pspec 2 pspec = 1 11 62
n (RL n RH ) . we …rst calculate the expected …rm value at t = 2 if the speculator is present. Expected …rm value at t = 2 is thus given by: 1 i 1 n 1 i pspec = RH + RL + RL .H . If = H. If = L. We have = H w. 3 3 2
which can be positive or negative. Note that this argument did not apply to 47
. she will trade 1 on her positive signal and impound it into prices. (34)
~i min )RH
~i + RL :
For part (i).H is increasing in : if the speculator is present.
which is positive if we are in Case 2 and negative if we are in Case 1. we have d = n so we exclude this case. we have d = i and so v = RL . We therefore have: pi. For part (iv). 1). 1 which implies that 2 (1 1 i ~i ~i ~n De…ne [RH RL RL ] = [RH Equation (5) thus yields: . not conditioning on the state. ow i and so …rm value v = RH . 1. Proof of Proposition 7 We start with some preliminary results that will be useful in the main proof.H 1 (p(0) + p(1) + p(2)) + 1 2 1 (p(0) + p(1)) 3 = 1 + (1 )2 2 3 p(0) + p(1) + p(2) = . so there is no trade. Part (iii) follows from simple calculations. W. whether the project is undertaken depends on the order n i ‡ ow: if X = 1. the project is always undertaken.p. if X 2 f0.This yields: pL.spec = RL + RL (p ( 1) + p (0) + p (1)) 1 3 3 3 1 i 1 1 i n n = RH RL (RH RL ) . 1 the speculator is absent. 0. ~i Note that RH > 0 from equation (1).p. 1g. Let min = 1 n i n n RH RL RH RL ~n RL = (1
1 n RH ].
1 2
<
and max = 1. . From we have >1 . regardless of the order ‡ at t = 1. so X 2 f0. If X = 1. 1g. which yields X 2 f 1. 2g with uniform probability and the investment is always undertaken.spec 2 2 i 1 n 1 pL. W. the expected stock price at t = 1 if investment has been 1 1 undertaken and the state is good. 2+
1 23
Simple calculations show that pi.p. we have d = n and so v = RL . 2 .

and so its minimum occurs at either min or max . it reduces to ~i ~i 3 min (RH RL ) > 0.H 1 p0 = p(2) + p(1) + p(0) 1 1 p ( 1) + p (0) + p (1) + 2+ 3 2 2 1 1 1 4 p(2) + p(1) + p(0) p( = 6(2 + ) 3(2 + ) 3(2 + ) 3 6 1 2 i i 2 n = [(2 + 2 )RH + (2 2 )RL (2 )RH 6(2 + ) p (2)
2
1)
n (2 + )RL ]. If X = 1. We therefore have: pi.L 1 =
1 2 1 p (0) 3 1 2
+ 1 p (1) 3
2 3
=
1 (p(0) + p(1)) . For part (ia). it is su¢ cient to prove that it is positive at both min ~n ~i and max .
(2
2
n ) (RH
n RH )
n (2 + ) (RL
n RH )
(36)
Equation (36) is quadratic and concave in . 2
This is independent of since the presence of the speculator does not change the order ‡ ow. Thus. to prove that (36) > 0. the expected stock price at t = 1 if investment has been 1 undertaken and the state is bad. Here.spec (equation (26)) and pH. we have d = n so we exclude ow this case.q H.
(35)
The sign of (35) is the same as the sign of (2 + 2 = (2 + 2
2 i n i n ) RH RH + (2 2 ) RL RH 2 ~i ~i ~n )RH + (2 2 )RL (2 + )RL . equation (36) reduces to 3RH 3RL > 0. 0. the short-run return to an investment (d = i) in the high state is given by: pi.spec L Unlike in the earlier cases of p1 and p1 which did depend on . Thus. At = max . For part (ib). This rules out the case of p ( 1) which is the only price that depends on . equation (35) is positive. the long-run drift is given by:
i RH i pi.H = RH 1
=
i RH 2+
p(2) + p(1) + p(0) 2+ i RL > 0:
(37)
For part (ii). 2g were all equally likely. X = 2 is particularly likely if the speculator is present. Regardless of whether the speculator is present.
48
. At = min .spec (equation (31)) because those quantities are conditional on 1 the speculator being present. 1g with uniform probability.L . 1. so X = f0. the order ‡ will be X 2 f 1. we …rst calculate pi. here we are conditioning upon the investment being undertaken. L.

(37)-(35). i. we must prove that ~i ~i ~n (4 2 + 2 )RH (8 2 )RL + (2 + )RL (41) ~i is positive.e. n RH n (RL n RH )
(39)
~i Equation (39) is decreasing in . s For part (iib). This implies that (1 ~i ~i )RH + RL ~n RL < (1 = 0. 6(2 + ) ~n (2 + )RL ]
In order to calculate abs ((37) (35)). and p0 is increasing in (because the speculator improves 1 the manager’ decisions). Thus. Suppose RL > 0.L = RL 1
~i min )RH
~i + RL
~n RL
1 (p(0) + p(1)) 2
i RL ) < 0
=
1 i (R 2 H
(40)
and independent of .L is independent of . the absolute return is decreasing in . the long-run drift is given by:
i RL i pi. equation (38) is negative. Since pi. from equation (34). it is automatic that (41) > 0. we …rst calculate the di¤erence between the long-run drift and the short-run return to a good investment.For part (iia). If RL ~i 0.L 1 p0 = 1 1 1 (p(0) + p(1)) p(2) + p(1) + p(0) + 2 6 3 3 1 1 1 p(2) + p(1) + p(0) ( )p( 1) = 6 6 6 3 6 1 i n i n = [(1 ) RH RH + RL RL ]. the short-run return to an investment in the low state is: pi. 1 For part (iii). we must …rst sign (37)-(35). This yields: 1 1 2 ~i ~i ~i ~i (RH RL ) [(2 + 2 )RH + (2 2 )RL 2+ 6(2 + ) 1 ~n ~i ~i = [(4 2 + 2 )RH (8 2 )RL + (2 + )RL ]. since pi. To prove this is positive. Evaluating (41) at
min
49
. since RH > 0.L is independent of . 6 1 3 p( 1)
6
(38)
The sign of (38) is the same as the sign of: (1
i n n n i ) RH RH (RH RH ) + RL ~n ~i ~i = (1 )RH + RL RL .

we must …rst sign (40) (4 ~i )RH ~i 2RL ~n RL
(38). max ). The sign of abs ((40) (38))
50
. This yields: 1 i (R 2 H 1 = [(4 6 In order to calculate abs ((40) negative. To prove this is (42)
~i ~i ~n is positive.
together with (34). we must prove that
i RL ) i )RH
1 [(1 6
i 2RL
i i )RH + RL n RL ]. Suppose that (41) is not monotonic in . At = max .e. it follows that (41) > 0. ~ 2R i
H
The condition
0
>
min . yields
min
= =
:(4
2
min
+
2 ~i min )RH
(8
2
~i min )RL
+ (2 +
min )[(1
~i min )RH
~i + RL ]
max
~i = (6 3 min )RH (6 3 ~i ~i ~n :3RH 6RL + 3RL > 0:
~i min )RL
> 0:
If (41) is monotonic in in its feasible range. we wish to show that abs ((40) (38)) > abs ((37) (35)). The derivative of (41) is equal to ~i ~i ~n (2 2 )RH + 2RL + RL : Then
0
=1
~i ~n 2RL + RL . (40) (38). Then its derivative with respect to must have a root 0 2 ( min .
n RL ]
(38)). i. we have (42) > 0: Finally.
respectively. Since (42) is decreasing in lambda. equation (42) becomes 3RH 2RL RL > 0.and
max . which implies (41) > (4 ~i ~i ~i )RH (8 2 )RL + 4(2 + )RL ~i ~i ~i = (4 2 + 2 )RH RH + 6 RL ~i > 0 for RL > 0: 2 +
2
We now calculate the di¤erence between the long-run drift and the short-run return to a good investment. yields
1
~n ~i ~i ~n RL RL 2RL + RL >1 ~i ~i 2RH RH ~ ~i ~ ~n Rn Ri 2RL + RL < L i L ) ~ ~ 2Ri R
H H
~i ~n ) 4RL < RL .

For = min . and so abs ((40)
(38))
abs ((37) > (35)). (43) is always positive.
51
.
or
max .
At
~i min )RL
2(2 +
min )[(1
~i min )RH
~i + RL ]
Thus. Then its minimum occurs at either ~i ~n = max .abs ((37)
(35)) is equal to the sign of: ~i ~i ~n )RH 2RL RL ] 2 ~i ~i )RH (4 + 2 )RL 2 ~i ~i )RL )RH + 4(1 ~i ~i ~n 2 + 2 )RH (8 2 )RL + (2 + )RL ] ~n ~i ~i ~n (2 + )RL ] [(4 2 + 2 )RH (8 2 )RL + (2 + )RL ] ~n (43) 2(2 + )RL : [(4
min
(2 + )[(4 =[(8 + 2 =2(2 + 2
(43) is quadratic and concave in .
2 ~i (43) = 2(2 + 2 min min )RH + 4(1 ~i ~i = 6 min (RH RL ) > 0. (43) reduces to 6(RH RL ) > 0.