Analysts Sale and Distribution of Non-Fundamental Information

Edwige Cheynel Carolyn B. Levine Tepper School of Business Carnegie Mellon University February 2009
Abstract We examine the sale and distribution of information in a setting where analysts possess non-fundamental information, or information related to short-term price movements, but unrelated to underlying rm value. Although a risk neutral agent would not directly sell fundamental (value-relevant) information, an analyst will sell non-fundamental information. Increasing the number of non-fundamental traders increases the competition on that signal, but it does not follow that overall non-fundamental prots shrink. Prices become more sensitive to net order ow, thus leading to a simultaneous increase in the value of the non-fundamental signal. The more precise the analysts information, the more widely he distributes it. In the limit, a perfect non-fundamental signal will be provided to an arbitrarily large number of investors (i.e., near-public disclosure, like a newsletter) for a arbitrarily small fee. Consistent with empirical ndings, analysts recommendations can be protable, even if they do not contain information about long-term cash ows. Analysts with non-fundamental information do not contribute to greater price efciency.

Introduction

In this paper, we describe the optimal sale and pricing of non-fundamental information, or information on contemporaneous liquidity demand. We nd that risk-neutral analysts will always sell non-fundamental information and will sell it without noise or bias, unlike fundamental information which would not be directly sold. By selling information to clients, the analyst affects the value of the signal in two offsetting ways. As is standard, competition reduces total prots since the clients trade more aggressively than a monopolist would. However, increased non-fundamental trading reduces the residual variance in net orderow which in turn increases price sensitivity and increases the value non-fundamental trading, a novel benet of competition. The analyst weighs the costs and benets, and the optimal distribution of information is increasing in its precision while the price of information is decreasing in the number of clients to whom it is sold. Although they can be protable to follow, recommendations based on non-fundamental information do not contribute to price efciency. Responding to the call to integrate research on earnings forecasts information (inputs) with stock recommendations (the nal product), several studies nd a surprising inconsistency. Analysts recommendations cannot be traced to valuation methods using the earnings or growth forecasts that are expected inputs (Block (1999) and Bradshaw (2002, 2004)). For example, Bradshaw (2002) nds that analysts opinions are not explained by per-period earnings forecasts despite the fact that earnings forecasts can predict one year ahead stock returns.1 The SEC has interpreted recommendations that differ from analysts internal (supporting) memos as smoking gun evidence of the analyst having succumbed to a conict of interest. For example, in SEC vs. Henry Blodget, the SEC states Blodget issued research reports that expressed views inconsistent with
Barniv, Hope, Myring, and Thomas (2009) conrms these ndings, but nds a dampened negative relation between recommendations and returns following Regulation FD.
1

privately expressed negative views.2 Are analysts ignoring their value-relevant expectations in making recommendations? Is it always to the detriment of their clients? Several papers show that recommendations have investment value, particularly in short horizons (e.g., Walther (1997), Gleason and Lee (2003), Frankel, Kothari, and Weber (2006), Ertimur, Sunder, and Sunder (2007)). Stickel (1995) and Womack (1996) show that changes in analysts recommendations are accompanied by abnormal returns in the direction of the change, both on the announcement date and in the months thereafter. Barber, Lehavy, McNichols, and Trueman (2001) show that a hedge portfolio created using publicly available consensus forecasts leads to an average abnormal gross return of 75 basis points per month. If analysts recommendations are based only on the underlying value of the asset (fundamental information), there should be no prospects for abnormal returns following their public disclosure. Lee (2001) notes that market prices are a product of the interplay between noise traders and rational arbitrageurs and Jegadeesh, Kim, Krische, and Lee (2004) demonstrates a correlation between recommendations and momentum. Perhaps recommendations convey price-relevant (but not value-relevant) information. We take this set of evidence as an invitation to reconsider the basic assumption that analysts recommendations are based on fundamental analysis alone. In this paper, sell-side analysts possess superior information about short-term demand shocks. For example, in an integrated nancial services rm, traders and sales teams collect orders, brokers provide investment advice that leads to orders and investment bankers make presentations at road shows. Research analysts may cull together the information from these groups to determine market color, market sentiment, supply, prevailing prices or relative value of an issuers security.3 Alternatively, analysts may have information that forms the basis for predictions about liquidity trade. For example, if an analyst foresees a rm missing its earnings targets, or has superior knowledge of the degree of earnings
http://www.sec.gov/litigation/complaints/comp18115b.htm Remarks before the SIA Compliance and Legal Division Member Luncheon by Annette L. Nazareth, Director, Division of Market Regulation, U.S. Securities and Exchange Commission, New York City, NY, July 19, 2005
3 2

management, the analyst can forecast the (over-)response to earnings news.4 , 5 The projected liquidity trades provide an information set, holding aside the information generating that projection. Supporting our information assumption, Lee (2001) writes in making security selections. . . investors need to consider the behavior of noise traders. The analyst then uses his non-fundamental information to recommend stocks, resulting in recommendations that are related to price movements, but unrelated to future earnings or cash ows, and possibly not supported by reasonable fundamental analysis. In a market with privately-held value-relevant information, net demand affects the market price of the asset. Because the market maker cannot determine the source of the demand, uninformed demand also moves prices, but in a direction unrelated to the true value of the asset. Knowledge of this uninformed demand (nonfundamental information) permits the analyst (and his clients) to trade in the opposite direction, capturing some of the incorrect price movement created by uninformed trades. For example, according to various testimonies collected in Lowenstein (2002), intermediaries trading with Long Term Capital Management learned about the funds short-term liquidity needs, allowing them to trade against LTCMs portfolio. As its positions worsened, LTCMs managers revealed more and more about their (otherwise highly secretive) planned trading to help nd offsetting trades. Releasing non-fundamental information leads to greater competition that reduces the value of the signal to each individual in a standard way. However, competition over non-fundamental information has the offsetting effect of increasing the sensitivity of price to order ow, and thus increases the value of the signal. Our paper demonstrates that the optimal amount of distribution is strictly positive. Whereas a risk-neutral analyst with fundamental information would never share his signal, he will always share non-fundamental information with clients. Despite the distribution,
We thank Gus DeFranco and Stephannie Larocque for helpful discussions about analysts information sets. Although their focus is on incentives to manage earnings, the division of information into two components (fundamental earnings and earnings management) in Fischer and Stocken (2004) provides another example of fundamental vs. non-fundamental information.
5 4

clients can trade on it protably. The number of clients to whom the analyst sells non-fundamental information is increasing in the precision of the non-fundamental signal, and becomes arbitrarily large (i.e., equivalent to nearly-public disclosure) when the non-fundamental information is perfect. Substantial demand-based trading reduces the value of fundamental (hereafter, proprietary) trading. Therefore, a proprietary trader might be willing to offer an analyst fundamental information to create conicting revenue incentives for him. To maximize prots from the fundamental information, the number of demand-based traders should be as low as possible whereas to maximize prots from non-fundamental information, the number of speculators may be high. We show that a proprietary trader is never willing to give up his monopolist power for free, although the price at which he is willing to sell is decreasing in the analysts information precision. There is no mutually acceptable price for fundamental information at which fundamental information sharing takes place. In fact, when the analysts information is very precise, he is strictly better off without access to the fundamental information. We interpret our ndings in the broader context of non-fundamental information circulation and nancial services. The SEC argued in a recent ling that sharing information about condential institutional customer order ow allowed traders to take advantage of price movements caused by execution of these orders to the detriment of the institutional customers.6 If institutional trades are motivated by private information, front-running can certainly be detrimental to them. However, our paper shows that knowledge of large trades reduces costs for institutional traders when demand is liquidity driven.7 For example, J.P. Morgan Chase & Co. estimated there would be about $100 billion in redemption requests for funds of funds in the fourth quarter of 2008. Redemption requests prompt order ow that may be unrelated to future value but is likely to move stock price.
www.sec.gov/litigation/admin/33-8673.pdf Following Rule 10b5-1, regulators allow certain corporate insiders to pre-commit and, possibly, disclose in advance trades motivated by liquidity needs (Henderson, Jagolinzer, and Muller (2008)). However, it may be undesirable for mutual funds to precommit to a long term trading strategy, and preannouncing a single trade cannot credibly signal that it is not information based.
7 6

As well, there has been signicant concern over conicts of interest in which analysts provide fraudulent recommendations to support investment banking or other department revenues. While this problem cannot be ignored, there may be instances in which recommendations and beliefs about underlying value justiably differ. In fact, the results of our paper point to a complementarity across departments in an integrated nancial services rm. Sell-side analysts may be in a better position to acquire non-fundamental information if they work for a rm that also provides brokerage services or investment banking. For example, squawk boxes are the conduit for the internal morning call . . . during which analysts and traders communicate with a rms brokers about changing opinions on stocks. The systems also carry so-called trader calls in which information about block trades can be disseminated (Goldstein 2005). They can, in turn, sell the information to clients who trade on it. Both the brokerage clients and the demand-based traders stand to benet from the joint provision of services. Whether analysts disclose selectively to privileged clients has also been a source of debate. In our model, information is priced based on how widely it is distributed leaving demand-based traders indifferent between receiving information and not receiving it. In other words, a client excluded from the analysts disclosure is not worse off than a client with access to it. Additionally, we provide conditions under which the analyst prefers wide distribution to selective disclosure. Our model also addresses issues of information ow within rms. A common claim is divisions in an integrated rm face conicts of interest, whereby they will exploit client information, beneting the rm and harming the clients. While such conicts of interest undoubtedly exist, the setting in this paper is one in which the rms itself prefers to limit information ows across divisions. If it is not optimal for the proprietary trading desk to exchange information with analysts in the rm, it is also not in the interest of the (centralized) rm to require such exchanges to take place.8 This indicates that rms would endogenously limit information ows across divisions.
8

For example, Lehman Brothers writes: The Strategic/Proprietary Trading desk leverages the Firms capital to

We note, however, that endogenous limits to information ows must be credible; rms may benet from regulation requiring the separation of divisions to the extent it increases market makers condence in the integrity of Chinese walls. We contribute to the literature on non-fundamental information and market microstructure. Roell (1990) shows that brokers will trade against the demands of their liquidity motivated clients, reducing the clients transaction costs. In Madrigal (1996), speculators learn about non-fundamental information (after the fact) that allows them to decompose order ow into informed and uninformed demand, and consequently learn about (long-lived) fundamental information. If the nonfundamental information is available before the rst round of trading, it can be used for both speculation in the rst period, and information based trading in the second (see the discussion on page 567). Our paper contributes to this literature by describing the way that non-fundamental information distribution affects the prots of each market participant, the spillover effects on the value of the fundamental information, and optimal dissemination choices by analysts. Brunnermeier and Pedersen (2005) examine a setting in which the liquidity needs of a distressed trader are exploited by predatory traders. Because of the limits on trading quantities, it can be in the interest of the predators to trade initially in the same direction as the distressed investors, and after price has overadjusted, in the opposite direction. In our model, with short lived non-fundamental signals, demand-based trade is in the opposite direction of the liquidity, and is not predatory but benecial. Carlin, Lobo, and Viswanathan (2007) show that cooperation, where traders allow the distressed investor to trade at more favorable prices, is generally sustained, with (rare) periods of episodic illiquidity. In our model, the analysts facilitate cooperation by providing recommendations such that demand-based traders supply the offsetting trades. Daniel, Hirshleifer, and Subrahmanyam (2001) show the benets to (otherwise uninformed traders) of exploiting heuristic (overly aggresinvest in the global markets. As a buy-side desk within a predominately sell-side Firm, this desk does not directly interact with Lehman Brothers clients. Instead, it relies on other sell-side rms in the market to provide incoming research and facilitate trade executions, acting like an internal hedge fund within the Firm.

sive) traders; the noise component of the heuristic trade will move prices in the wrong direction. Similarly in our model, clients can earn prots by following analysts recommendations that exploit the noise component of orderow, even if the recommendations are unrelated to underlying asset value. Our paper is linked to work on imperfect trading heuristics (e.g., overcondence or other nonBayesian updating heuristics) and their relation to price, efciency and the trading strategies of other rational market participants. In a market with overcondent traders, Bayesian traders in Fischer and Verrecchia (1999) regulate their trades; this is parallel to the proprietary trader in our market who adjusts demand in the presence of non-fundamental traders. Hirshleifer and Luo (2001) show that overcondent traders can survive in a market with rational traders because they can better exploit the mispricing caused by noise traders. Our demand-based traders survive, despite having a fundamental information disadvantage for the same reason. Lundholm (2008) examines a setting where the long-short strategies of hedge traders can be implicitly exploited by fundamental traders who trade more when hedge traders push price in the wrong direction due to their rank-order strategy. Therefore, the traders with fundamental information implicitly benet from the noise component in hedge traders, where in our model, analysts clients directly use the demand-based signal. Finally, we relate our paper to studies on the effect of information release on the trading decisions of informed agents (e.g., Baiman and Verrecchia (1996), Huddart, Hughes, and Levine (2003), Bertomeu, Beyer, and Dye (2008)). Our proprietary traders must alter their trading strategy (i.e., trade less intensely) in the presence of non-fundamental traders as the residual variance, which provides them disguise, is diminished. The more widely non-fundamental information is disclosed, the more limited the prots of proprietary traders. Earlier analytical literature describes price informativeness in the presence of endogenous disclosure where the disclosure is value rel-

evant and non-selective (e.g., Penno (1987), Gigler and Hemmer (2001), Jorgensen and Kirschenheiter (2003)) Because proprietary traders moderate their trades in response to the presence of analysts (and their chosen optimal distribution of information), nal price efciency in our model is unaffected by non-fundamental disclosure. Section 2 describes the basic model, where a single proprietary trader holds fundamental information and a single analyst holds non-fundamental information. Section 3 demonstrates the optimal sale and use of information and Section 4 examines the possibility for information sharing within a rm. Section 5 concludes.

The Model

We analyze a Kyle security market for a single asset, whose terminal cash ow is initially unknown. The (fundamental) value of the asset, v , is a normally distributed random variable with mean zero and variance . A proprietary trader (e.g., an insider or fund manager) receives a noisy signal, s = v + , on the security, where is normally distributed with mean zero and variance, 2 . The proprietary trader can submit a trade, based on his signal. Uninformed traders submit their demands that derive from exogenously determined liquidity needs. Specically, liquidity demand
2 is u, which is normally distributed with mean zero and variance u .

An information seller (hereafter, the analyst) receives a signal, = u + , on the order ow originating from these liquidity traders, where represents the noise in the signal, with normally distributed with mean zero and variance 2 . Thus, there are two types of information in the model, fundamental and non-fundamental. Fundamental information is a signal about the underlying asset value (v ) and non-fundamental information is a signal about contemporaneous liquidity demand (u). The non-fundamental signal is an abstract representation of the activities of nancial analysts that lead to superior knowledge about the demand for a security (e.g., road-shows, the analysis

of various data feeds, relationships with institutional investors, orders over a squawk box, general market data, etc). The intermediary offers his information for fee c (publicly known). A set of competitive investors (hereafter, the clients) decide whether or not to buy the non-fundamental information, and then submit demands conditional on their information. Let m be the number of clients that purchase information. We assume that clients cannot resell their information. In order to focus on the essential aspects of the model, we assume that the analyst does not have fundamental information.
9

Finally, traders submit market orders for shares. A competitive market-maker observes total order ow (i.e., the sum of the orders of the proprietary trader, the liquidity traders and the analysts clients) and sets a price for the asset. The market price is equal to the expected value of the asset conditional on the (observed) net order ow. All informed traders trade strategically, anticipating the impact of their trades on order ow and consequently on price. All agents in our model are risk-neutral, and thus this paper highlight non-risk based incentives for selling or exchanging information. Specically, the proprietary traders demand function, X , will depend on his information, s, and the number of analysts clients, m. A representative clients demand function, Z , will depend on his information, , and the number of clients, m. Following convention, we denote actual demand of the proprietary trader and client, given realized information sets as x X (s, m) and z Z (, m), respectively. The market maker observes aggregate order ow, y = x + mz + u, and the asset is traded according to the price schedule, P (y ), where the price is set so that the market makers expected prots, conditional on y , are zero. The expectation of v given the noisy signal s is E (v |s) = s; the expectation of u given the
Extending the analysis to an environment where the analyst also possesses fundamental information yields similar results with respect to the sale of non-fundamental information (See section 4 for additional details).
9

noisy signal is E (u| ) = , where

2 2 + 2 ). /(u /( + 2 ) and u

Hereafter, we refer to and as the precision of the fundamental and non-fundamental information, respectively. Trading prots are computed as the difference between price and intrinsic value multiplied by quantity (purchased or sold). A clients net prots are equal to his trading prots less the cost of information, c. The analysts prots are the sum of the fees he collects from each client, or m c. A nancial market equilibrium is a set of trading rules that maximize traders expected prots, given their information sets and the number of other market participants, and a pricing rule so that the market maker breaks even in expectation. To the standard nancial market equilibrium, we add two additional conditions to dene an information sales equilibrium. Denition: An information sales equilibrium (ISE) is a fee, an information purchase decision, and demand and pricing schedules such that (A) Purchase. There are exactly m clients willing to purchase at price c . (B) Optimal Sharing. The analysts revenues are maximized at (c , m ). (C) Financial Market Equilibrium (FME). Demand and price schedules satisfy: (i) Client prot maximization: For all trading strategies Z , Z = argmax E (T (X , Z, P )|, m), (ii) Proprietary trader prot maximization: For all trading strategies X , X = argmax E (I (X, Z , P )|s, m) and, (iii) Market efciency: Price satises P = E (v |y, m). Parts (A) and (B) together allow the analyst to extract all of the prots from his clients, leaving clients indifferent between purchasing and not purchasing information (i.e., they earn zero expected 10

prots either way).10 Part (C) involves pricing schedule and trading decisions of all informed market participants, both those with fundamental information (the proprietary trader) and those with non-fundamental information (the clients). We adopt the convention that the analyst does not trade, consistent with the observed separation between sell-side and buy-side analysts. However, for analytical purposes, this assumption is without loss of generality: if the analyst were to trade, then the information would have been sold to m 1 speculators, leaving the predictions of the model unchanged. Second, we assume that the price c at which the information is sold is public; given that the value of c is in a one-to-one mapping with the number of speculators buying the information (by the zero-prot condition), this is equivalent to assuming that the number of clients m is public - thus one can directly condition trading decisions in Part (C) on m. Third, as is common in entry games (here, the clients purchase decision is analogous to an entry decision), we approximate the number of clients m as a continuous variable. Although in practice, the number of clients is discrete and will set to either [m] or [m + 1] (where [m] is the integer part) the directional predictions are unaffected.

Non-Fundamental Information Sales

We solve for the optimal information sales equilibrium by holding xed the number of clients at arbitrary m, and solving for a nancial market equilibrium (Part (C) of the denition above). Then, we solve for the analysts prot maximizing combination of intermediation fee and number of clients, (c , m ) such that the value of information to a client is non-negative.
Mikhail, Walther, and Willis (2004) show that analysts can produce protable recommendations, but after accounting for transactions costs, returns are no longer reliably positive.
10

11

Financial Market Equilibrium

Suppose the market order of each type of informed trader is a linear function of his information. Then, with the normality of all random variables, the market makers pricing function can be written as P (y ) = y , where y is net order ow and represents the sensitivity of price to the order ow. Because the clients are identically informed, we can solve for the optimal linear trading strategy of a single client and scale up by the number of clients. Then, given the trading strategies of the proprietary trader and demand-based traders, we solve for the market makers breakeven pricing function. Lemma 1. With m demand-based traders and a single proprietary trader, there is a unique linear nancial market equilibrium. Let rm = (m(m + 2)(1 ) + 1)/(m + 1)2 and dene constants
2 2 1/2 m , m and m by m = (rm u /)1/2 , m = /(m + 1), and m = (/(4rm u )) . Then,

equilibrium demand and price schedules are

Xm = m s Zm = m Pm = m (x + mz + u)

The analysts clients each trade against the liquidity demand as one of an m-trader oligopoly. The sensitivity of the clients demand to the non-fundamental signal ( ) depends only on its accuracy, . The effect of demand-based trading is to reduce the liquidity available for the proprietary trader to exploit, and leads to a higher price sensitivity parameter, m . To see this, consider the impact of m clients observing . Initially, the liquidity traders provide an order imbalance of u. Then, the clients trade m shares, driving order ow (in expectation) back towards zero. The residual liquidity variance, when there are m clients, is the variance of the net non-fundamental trades, calculated as m(m + 2)(1 ) + 1 (m + 1)2 12

V ar(u + m ) =

2 2 u rm u

and rm 1 for all m 1. Since the market maker incorporates all of the sources (proprietary, demand-based, and liquidity traders) of order ow in calculating his conditional expectation, break-even prices are more sensitive to the order imbalance than they would be in the absence of speculation. Price sensitivity is increasing in the precision of both fundamental and non-fundamental information and the number of informed clients. The proprietary desks trading intensity on its signal (m ) has the same relation to the exogenous parameters that it would in a model without demandbased trading, but original liquidity variance is replaced by ex-post variance. A large liquidity realization moves the price along the linear pricing function, but demand-based trading reverses the move through trades in the opposite direction of the liquidity demand. The proprietary desk trades based on the remaining disguise (i.e., after netting liquidity and anti-liquidity trades). Consequently, total information content of prices does not increase with demand-based trading (i.e., more fundamental information is not revealed through trades), where information content is the inverse of the conditional variance of v , given price. Observation: An analyst releasing non-fundamental information does not contribute to price efciency. Specically, the information content of prices V ar(v |P ) = (1 ) does not depend on 2 the number of clients m or the precision of the non-fundamental information.

Optimal Information Sales and Purchases

The larger the number of clients, the more each individual client must moderate his individual demand, reducing individual prots. However, there is a novel offsetting benet to competition in non-fundamental trading. Increasing the number of demand-based traders moves net demand closer to zero, and observed order imbalances are more likely to come from fundamental (proprietary) trading. Consequently, there is an endogenous increase in price sensitivity that increases per

13

unit non-fundamental trading prots. Figure 1: Optimal Non-Fundamental Information Sales

m
10 8 6 4 2

m
10

isoprofits

(c,m) frontier

8 6 4 2

(c,m) frontier

(c*,m*)

Panel A:The (c,m) Frontier

Panel B: The Information Sharing Equilibrium

The shaded region in Panel A of Figure 1 shows feasible combinations of fees and number of clients. At each combination (c, m) in that region, m clients would be willing to purchase the information at fee c, using the nancial market equilibrium price sensitivity for the particular number of clients. The solid boundary line is the (c, m)-frontier representing the largest fee the analyst can charge such that m clients are willing to purchase.11 The analyst chooses the fee along the frontier that maximizes revenues. The prot maximizing fee, c , is on the analysts iso-prot curve tangent to the (c, m) frontier (depicted in Panel B of Figure 1). The optimal fee is unique and the complete information sales equilibrium is presented in Proposition 1. Proposition 1. i. For (0, 1), there exists a unique information sales equilibrium (ISE) where
Equivalently, the frontier can be interpreted as the maximum number of clients m willing to purchase information at every given price c.
11

14

0 = m 1 m m2 (1 ) + 1 c
2 2 (rm u ) = 2 (m (m + 2)(1 ) + 1)
1

and trading strategies and prices follow Lemma 1 at m . ii. The analyst always discloses his information, or m > 1. iii. The number of clients is increasing in the precision of the non-fundamental signal; if 1, m is arbitrarily large. Part (ii) indicates that non-fundamental information is always sold, whether or not the analyst trades directly on it.12 The sale of non-fundamental information occurs without risk-sharing motives and without a reduction in the precision of the sellers information. Higher precision leads to greater distribution and higher expected prots; therefore an analyst has no incentives to add noise to the signal. Contrast this with the results on fundamental information-sharing. Admati and Peiderer (1988) and show that direct information sales will not occur (i.e., the optimal number of informed traders is 1) with a risk-neutral information owner. Additional informed traders would generate competition that would always decrease surplus. If information owners are risk-averse instead, they may sell information to diversify away their liquidity risk. Although the case of risk-aversion is an interesting one for an individual seller, it is less likely to apply to sell-side analysts employed by large diversied investment rms.13 The conclusions about information sales in the presence of risk-aversion also seem less likely to apply to sell-side analysts. The information widely distributed by sell-side analysts tends to be in the form of a generic (i.e., non-personalized)
12 Allowing the number of clients to be a continuous variable indicates, conceptually, that some degree of information sale is optimal. If m is required to be an integer, there may be levels of precision such that m = 1 is preferred to m = 2. 13 Most large investment rms have proprietary desks that do analysis, but do not give direct access to their information sets. If risk sharing motives were signicant at the rm or individual analyst level, we would expect direct sales of information by these buy-side analysts as well.

15

newsletter or press release. For for wide distribution to occur with risk-averse information owners, the signal must be sold with individualized noise (Garcia and Sangiorgi 2007). Increasing the underlying variance of fundamental information or the precision fundamental signals ( and respectively) is benecial to both the proprietary trader and the analyst. Although the number of clients is independent of the fundamental information precision, the analysts fee is increasing in and . Although the prediction is the same, the reason the analysts prots are increasing in underlying variance is very different when analysts have fundamental vs. non-fundamental information. When little is currently known about the rm, an analyst with fundamental information will have a greater impact on prices. With non-fundamental information, rather than improving market efciency, the value of the information is simply higher when there is a greater degree of information asymmetry (between proprietary and uninformed traders). An increase in the precision of the non-fundamental signal () increases the analysts prots and decreases the liquidity traders losses at the expense of the proprietary desk.14 An increase in
2 the underlying variance of liquidity trades, u , which can serve as a proxy for volume, leads to an

increase in and therefore an increase in non-fundamental prots.15 Lang and Lundholm (1996) write analysts provide different forecasts primarily because of differences in non-rm provided information, rather than differences in interpretation of common information. Although their focus is on earnings forecasts, the statement applies equally well to non-fundamental value. It is easy to imagine different analysts having different signals on short-term demand (due to interactions with different sets of liquidity clients). It is the difference in information that creates the
Although a single individual with both fundamental and non-fundamental information would prefer perfect (to noisy) non-fundamental information, a proprietary trader with fundamental information alone prefers that nonfundamental information is noisy when in the hands of a (second party) analyst who can sell it. As non-fundamental precision increases, the number of clients gets large and the prots of the proprietary desk are driven down. In the limit, the number of clients would be innite and the proprietary desks prots would be zero. If fundamental is held (and traded on) by a corporate insider, our results suggest that laws that prohibit insider trading become less essential when analysts provide non-fundamental information to a large set of clients. 15 Our conclusions are consistent with Bhushan (1989) and OBrien and Bhushan (1990), which document a relation between trading volume and analyst following.
14

16

recommendation differences rather than different interpretation of common signals.

Precision Costs When information is costless, the analyst prefers the most precise information (or, = 1). However, if the cost of information is increasing in its precision, it may be optimal for the analyst to acquire less than perfect information. See Figure 2, where the thick and dashed lines represent the analysts prots with costless and costly precision respectively. When the overall protability of the market is low (u , and are low), the analyst is less willing to expend resources to improve the precision of his signal. To see this, compare the left and right panels with high and low respectively. Taking this further, with precision costs, low liquidity rms, rms with low underlying uncertainty, and rms with low information asymmetry will have less protable and less widely distributed recommendations. With a focus on earnings forecasts (i.e., fundamental information), Das, Levine, and Sivaramakrishnan (1998) nd greater bias in earnings forecasts for rms with unpredictable earnings indicating the demand for private information is greater with rms are hard to forecast. This model relates underlying variance (rather than predictability) to the information acquisition decision. This model suggests that recommendations for these rms will be less profitable, not only because information is less likely to move prices signicantly, but also because analysts will not expend the resources to improve the quality of their non-fundamental information if the costs of precision are high. In a model where insiders have both proprietary and liquidity motives for trade, pre-announcement cannot credibly communicate the rationale for trading. Huddart, Hughes, and Williams (2004) show ex ante commitments to better public disclosure reduces the insiders incentives to distort trade but imposes price risk if the information is unfavorable. In our model, managers have conicting incentives for disclosure, even if they trade for liquidity reasons alone, and this is publicly known. Reducing variance through disclosure, reduces price sensitivity, but at the same time de17

Figure 2: Costless vs. Costly Precision

Analyst Profits Analyst Profits 80 200 60 40 100 20 20 100 40

0.2

0.4

0.6

0.8

1.0

0.2

0.4

0.6

0.8

1.0

(a) u = 100, = 100, = 0.95

(b) u = 100, = 10, = 0.95

creases the analysts incentives to gather information about the liquidity signal. The benets to liquidity traders (or, costs to proprietary traders) of a reduction in underlying information variance () are offset by the costs of a reduction in the analysts information quality (). Therefore, even when managers trade solely for liquidity motives, it may not be in their best interest to increase disclosure (reduce uncertainty) of underlying value.

Fundamental Information Exchange

When non-fundamental information is very precise, the resulting wide distribution of information will signicantly reduce the prots from proprietary trading. If the proprietary desk could nd a way to change the analysts incentives to distribute information, would it wish to do so? That is, suppose the proprietary desk shares fundamental information with the analyst. By doing so, it creates conicting interests for the analyst. As before, the analyst wishes to prot on the nonfundamental information, but doing so has spillover effects on his ability to prot from the (newly acquired) fundamental information. Lemma 2. With m clients and two fundamentally informed traders (the proprietary desk and m , analyst), there is a unique linear nancial market equilibrium. Dening constants m and 18

 m by m = (rm 2 /(2))1/2 , m = (2/(9rm 2 ))1/2 , equilibrium m = /(m + 1), and u u demand (for each trader with each type of information) and price schedules are

m s m = X

m = Z m

m (2x + mz + u), m = P

(1)

In the nancial market equilibrium, the proprietary desk and analyst act as duopolists in the fundamental information; the demand-based clients demands are unaffected by the number of fundamentally informed traders. However, because prices are more sensitive to trades with increased competition in fundamental trading, the non-fundamental signal becomes more valuable. Proposition 2. i. For (0, 1), there exists a unique information sales equilibrium (ISE) where

0 = 1m 2 (3 + 2m )(1 ) 1 2 2 2 (rm u ) c = 3 (m (m + 2)(1 ) + 1) and trading strategies and prices follow Lemma 2 at m . ii. Holding exogenous parameters constant, non-fundamental information is less widely sold when the analyst also has access to the fundamental signal (m m ). Possession of fundamental information indeed reduces the extent of distribution, and for imprecise non-fundamental information, the optimal number of clients may be less than one (i.e., no sharing). Although possessing the fundamental information induces the analyst to reduce the degree of distribution, the cost to the proprietary desk of the increased competition on the fundamental signal is too great, giving Corollary 1 below.

19

Corollary 1. The proprietary desks expected prots are lower with fundamental information sharing than without sharing for all combinations of exogenous parameters. Figure 3 graphs Corollary 1. The prot function for the proprietary desk labeled no sharing is calculated using the optimal number of clients when the analyst does not possess fundamental information (m ), and the prot function labeled sharing is calculated using the optimal number of clients when the analyst has fundamental information (m ). The no sharing prot curve is everywhere above the sharing prot curve for all values of exogenous parameters. The required fee is increasing in the variance of underlying value and the precision of fundamental information. Figure 3: Proprietary desks prots: Shared Fundamental Information vs. No Sharing
profits
0. 5 0. 4 0. 3 0. 2 Sharing 0. 1 0. 2 0. 4 0. 6 0. 8 1. 0 0.05 No Sharing

profits
0.25 0.20 0.15 0.10 Required Fee

0. 2

0. 4

0. 6

0. 8

1. 0

If the proprietary desk could extract a portion of the analysts fundamental trading prots in exchange for sharing his fundamental information, might it then be optimal to share?16 The difference between the shared information and no sharing curves determines the required fee, or the minimum transfer payment the proprietary desk would demand from the analyst in exchange for information, plotted on the right hand side of Figure 3
An alternative way to think about the resource transfer from the analyst to the proprietary desk would be for the analyst to commit to sell to a smaller number of clients than is optimal in exchange for the information. Whether such commitments are feasible may be debatable, however we can imagine an unmodeled repeated relationship between the proprietary desk and analyst where a failure to stick to the agreed upon distribution would lead to no sharing in all subsequent periods.
16

20

Figure 4: Analysts prots: Shared Fundamental Information vs. No Sharing

profits
0. 5 0. 4 0. 3 0. 2 0. 1 No Sharing 0. 2 0. 4 0. 6 0. 8 1. 0 Sharing 0.20 0.15 0.10 0.05 Maximum Fee

profits

0. 2

0. 4

0. 6

0. 8

Figure 4 graphs the analysts prots under no sharing and sharing of fundamental information, assuming the proprietary desk provides the information for free. In possession of fundamental information, the analyst generates prots from both fundamental trading and the sale of nonfundamental information. Without fundamental information, the analysts prots are exclusively from the sale of non-fundamental information. We assume the analyst cannot resell fundamental information to other clients. The analysts prots with the fundamental information are actually lower than without it for high values of . Rather than pay for the information in these cases, the analyst would have to be paid to accept it. The difference between the two curves determines the analysts maximum willingness to pay for fundamental information, plotted on the right hand side of Figure 4. Proposition 3. i. If the price sensitivity is consistent with the number of fundamentally informed traders, there is no fee at which both the proprietary desk is willing to sell his fundamental information and the analyst is willing to buy it. ii. If the analyst and proprietary desk work for separate divisions in a common rm, overall rm prots cannot be enhanced by sharing fundamental information. 21

The minimum price at which the proprietary desk is willing to sell is everywhere above the analysts maximum willingness to pay (solid vs. dashed line, respectively on the left hand side of Figure 5). Therefore, the proprietary desk will never provide the analyst access to his private signal (part (a) of Proposition 3). Since the analyst and proprietary desk do not share fundamental information voluntarily, does this extend to information sharing within a rm? Suppose there is an integrated nancial services rm that provides nancial analysis, advisory services, investment banking, brokerage, etc. A common criticism of providing all of these services under one roof is that it creates a conict of interest: departments interact (e.g., share information) to benet some clients at the expense of others. Suppose the integrated rm could force information exchange to either benet the analyst (when is low) or the proprietary desk (when is high). Part (b) of Proposition 3 shows that the integrated rm will never require sharing among groups. Figure 5: Impossibility of Fundamental Sharing Equilibrium

profits
0.25 0.20 0.15 0.10 0.05

profits
0.50 0.49

___ ----0. 2

0.48 Required Fee Maximum Fee 0. 4 0. 6 0. 8 0.47 0.46

___ ----0. 2 0. 4 0. 6

No Sharing Sharing

Insider does not share

Common firm prefers no sharing

To see this, Figure 5 plots the sum of the analyst and proprietary desks prots with (solid line) and without (dashed line) sharing. Notice that the prots of the overall rm are higher without sharing. This suggests a natural division among groups in the rm. Even without the requirement that the rm erect a Chinese wall, this is a setting where one would, de facto, arise endogenously. 22

However, the market maker must believe in the efciency of the Chinese wall (i.e., information ows can genuinely be avoided when it is benecial for the rm). If the absence of a formal Chinese wall was taken as prima facie evidence that information leakage (sharing) would occur, it would be self fullling. That is, if the market maker assumes the information will be shared, and will price the asset accordingly, the proprietary desk would indeed share its information as long as > 0.502. Therefore, regulation requiring Chinese walls may be benecial in that it enhances the credibility that divisions are not sharing, and that information does not ow freely across departments.

Conclusion

This paper analyzes the sale of non-fundamental information by a single analyst to clients who can trade on that information in a Kyle (1985) market. The analysis highlights the differences between fundamental and non-fundamental information. While a single, risk neutral agent with fundamental information always prefers to be a monopolist, an individual with non-fundamental information increases the informations value by sharing it. The demand-based traders issue orders in the opposite direction of noise traders, driving expected order ow towards zero and reducing the variance of the disguise for proprietary trading. Order imbalances are more likely to come from the proprietary desk possessing fundamental information, leading to an increase in price sensitivity. With more sensitive prices, demand-based traders can better capitalize on incorrect price swings stemming from liquidity trades. In our model, the analyst, who possesses non-fundamental information, extracts the total surplus from his clients. With this feature, the optimal number of clients is always greater than one. The more precise the non-fundamental information, the more widely it is sold. The analyst has no incentives to add noise or distort his signals, as lower precision reduces clients willingness to

23

purchase. Proprietary traders in our model are strictly worse off with an analyst who possesses and optimally sells non-fundamental information to clients than in a market absent demand-based trading. Despite the costs of demand-based trading, the proprietary desk will not exchange fundamental information with analysts to limit their willingness to sell non-fundamental information. The reduction in the number of clients that would result is insufcient to compensate the proprietary desk for competition on the fundamental signal. Our results suggests complementarities between services in a nancial services rm despite the typically one-sided criticism that they lead to conicts of interest. By offering brokerage services, analysts gain access to non-fundamental information that can be subsequently sold to potential clients. While this may reduce the value of proprietary desk trading, such a reduction would occur as long as there is demand-based trading (i.e., regardless of if it occurs within the rm itself). Because a rm with divisions producing (or gaining access to) both fundamental and nonfundamental will not increase its overall value by sharing within the rm, naturally arising limits to intra-rm information ow are possible.

24

References
A DMATI , A. R., AND P. P FLEIDERER (1988): Selling and Trading on Information in Financial Markets, American Economic Review, 78(2), 96103. BAIMAN , S., AND R. E. V ERRECCHIA (1996): The Relation Among Capital Markets, Financial Disclosure, Production Efciency, and Insider Trading, Journal of Accounting Research, 34(1), 122. BARBER , B., R. L EHAVY, M. S. M C N ICHOLS , AND B. T RUEMAN (2001): Can Investors Prot from the Prophets? Security Analyst Recommendations and Stock Returns, Journal of Finance, 56(2), 10831124. BARNIV, R., O.-K. H OPE , M. M YRING , AND W. T HOMAS (2009): Do Analysts Practice What They Preach and Should Investors Listen? Effects of Recent Regulations, Accounting Review, forthcoming. B ERTOMEU , J., A. B EYER , AND R. A. DYE (2008): Capital Structure, Cost of Capital, and Voluntary Disclosures, Working Paper, Northwestern University. B HUSHAN , R. (1989): Firm characteristics and analyst following, Journal of Accounting and Economics, 11, 255274. B LOCK , S. (1999): A study of nancial analysts: practice and theory, Financial Analysts Journal, 55(4), 8695. B RADSHAW, M. T. (2002): The Use of Target Prices to Justify Sell-Side Analysts Stock Recommendations, Journal of Finance, 16, 2741. (2004): How Do Analysts Use Their Earnings Forecasts in Generating Stock Recommendations?, Accounting Review, 79(1), 2550. B RUNNERMEIER , M. K., AND L. H. P EDERSEN (2005): Predatory Trading, Journal of Finance, 60, 18251863. C ARLIN , B., M. L OBO , AND S. V ISWANATHAN (2007): Episodic Liquidity Crises: Cooperative and Predatory Trading, Journal of Finance, 62, 22352274. DANIEL , K., D. H IRSHLEIFER , AND A. S UBRAHMANYAM (2001): Overcondence, arbitrage, and equilibrium asset pricing, Journal of Financial Markets, 56, 921965. DAS , S., C. B. L EVINE , AND K. S IVARAMAKRISHNAN (1998): Earnings Predictability and Bias in Analysts Earnings Forecasts, Accounting Review, 73, 277294. E RTIMUR , Y., J. S UNDER , AND S. V. S UNDER (2007): Measure for Measure: The Relation between Forecast Accuracy and Recommendation Protability of Analysts, Journal of Accounting Research, 45(3), 567606. 25

F ISCHER , P., AND R. V ERRECCHIA (1999): Public Information and Heuristic Trade, Journal of Accounting and Economics, 27, 89124. F ISCHER , P. E., AND P. C. S TOCKEN (2004): Effect of Investor Speculation on Earnings Management, Journal of Accounting Research, 42(5), 843870. F RANKEL , R., S. KOTHARI , AND J. W EBER (2006): Determinants of the informativeness of analyst research, Journal of Accounting and Economics, 41(1-2), 2954. G ARCIA , D., AND F. S ANGIORGI (2007): Information Sales and Strategic Trading, Working Paper, University of North Carolina. G IGLER , F. B., AND T. H EMMER (2001): Conservatism, Optimal Disclosure Policy, and the Timeliness of Financial Reports, Accounting Review, 76(4), 471493. G LEASON , C. A., AND C. M. C. L EE (2003): Analyst Forecast Revisions and Market Price Discovery, Accounting Review, 78(1), 193225. G OLDSTEIN , M. (2005): SEC Probes Squawk Box Front-Running, TheStreet.Com. H ENDERSON , M. T., A. D. JAGOLINZER , Working Paper, Stanford University.
AND

K. A. M ULLER (2008): Scienter Disclosure,

H IRSHLEIFER , D., AND G. Y. L UO (2001): On the survival of overcondent traders in a competitive securities market, Journal of Financial Markets, 4, 7384. H UDDART, S., J. S. H UGHES , AND C. B. L EVINE (2003): Public Disclosure and Dissimulation of Insider Trades, Econometrica, 69(3), 665 681. H UDDART, S., J. S. H UGHES , AND M. W ILLIAMS (2004): Pre-announcement of Insiders Trades, Working Paper, Pennsylvania State University. J EGADEESH , N., J. K IM , S. D. K RISCHE , AND C. M. C. L EE (2004): Analyzing the Analysts: When Do Recommendations Add Value?, Journal of Finance, 59(3), 10831124. J ORGENSEN , B. N., AND M. T. K IRSCHENHEITER (2003): Discretionary Risk Disclosures, Accounting Review, 78(2), 449469. K YLE , A. (1985): Continuous Auctions and Insider Trading, Econometrica, 53(6), 13151335. L ANG , M., AND R. L UNDHOLM (1996): Corporate Disclosure Policy and Analyst Behavior, Accounting Review, 71, 467492. L EE , C. M. (2001): Market efciency and accounting research: a discussion of capital market research in accounting by S.P. Kothari, Journal of Accounting and Economics, 31(1-3), 233 253.

26

L OWENSTEIN , R. (2002): When Genius Failed: The Rise and Fall of Long-Term Capital Management. Fourth Estate. L UNDHOLM , R. (2008): Fundamental Analysis and Quantitative Hedge Trading in a Disagreement Model, Working Paper, University of Michigan. M ADRIGAL , V. (1996): Non-Fundamental Speculation, Journal of Finance, 51(2), 553578. M IKHAIL , M. B., B. R. WALTHER , AND R. H. W ILLIS (2004): Do security analysts exhibit persistent differences in stock picking ability?, Journal of Financial Economics, 74, 6791. OB RIEN , P. C., AND R. B HUSHAN (1990): Analyst Following and Institutional Ownership, Journal of Accounting Research, 28, 5576. P ENNO , M. C. (1987): A Note on the Value of Information Given Asymmetric Information and Self-Reporting, Contemporary Accounting Research, 3, 368374. ROELL , A. (1990): Dual-capacity Trading and the Quality of the Market, Journal of Financial Intermediation, 1(2), 105124. S TICKEL , S. E. (1995): The Anatomy of the Performance of Buy and Sell Recommendations, Financial Analyst Journal, 51, 2539. WALTHER , B. R. (1997): Investor Sophistication and Market Earnings Expectations, Journal of Accounting Research, 35(2), 157179. W OMACK , K. L. (1996): Do Brokerage Analysts Recommendations Have Investment Value?, Journal of Finance, 51(1), 137167.

27

Proof of Lemma 1: Suppose for constants m , m and m , linear functions X, Z and P are given by Xm = m s Zm = m Pm = m y Given the linear rules, the proprietary desk and a single informed client choose Xm and Zm to maximize prots, that can be written as E {(v Pm )Xm |s, m} = E (s(v (s + m + u))|s) = E (v 2 2 s2 ) E {(v Pm )Zm |, m} = E ( (v (s + + (m 1) + u))| ) 2 2 2 2 = E ( u ) where is the trading strategy of each of the other m 1 clients. Taking rst order conditions and replacing with since all clients are identical, we have m = P = m y = E {v |y } implies m =
2 / m

and m = . 2m m+1

(A-1)

m +
2 u
2 m2 m

. + 2mm + 1

(A-2)

Solving Equations A-1 and A-2 subject to the second order condition m > 0 gives the equilibrium. 2 Proof of Proposition 1: The expected prot of an individual client in a market with a total of m clients, m is: m m m =
(m1)m 2 + 1 u

(A-3)

Replacing the expression of m and simplifying m =

2 u (m + 1)2

(A-4)

An informed client is indifferent between buying and not buying information at c: c=

2 u (m + 1)2

(A-5)

28

Replacing by m , we get cm = 2(m + 1)

2 u m(m + 2)(1 ) + 1
1 2

(A-6)

The analysts prots are equal to A = mcm . To maximize the analysts prots we endogenize the number of clients. We take the FOC of A in m:
2 (m (1 m m2 ) (1 ) + 1) u =0 2(m + 1)2 (m(m + 2)(1 ) + 1)3/2

(A-7)

Therefore m and c are implicitly dened by the following equations: 0 = m 1 m m2 (1 ) + 1 c = 2(m + 1)

2 u rm
1 2

(A-8) (A-9)

Write equation (A-8) as (, m ) = 0 where (, m) = m 1 m m2 (1 ) + 1. (, m) is a polynomial expression of degree three with domain m [0, +) and ranges from 1 to . Therefore there exists a unique real solution to (, m ) = 0. Thus there exists a unique function () such that m = (). (0) solves (0, (0)) = 0 or (1 (0))((0) + 1)2 = 0, which gives (0) = 1. 1 m 2 2 (u ) , which is increasing in m; thus the prot is maximized m, lim1 A = 2(m +1) at m . Applying the implicit function theorem, () = (, m)/ (, m) m m (m2 m + 1) = (3m2 2m + 1) (1 )

Given equation (A-8), (m2 m + 1) < 0 and also (3m2 2m + 1) < 0. Therefore () is positive and m is increasing in . 2

29

Proof of Proposition 2: From Lemma 2, the price sensitivity with two fundamentally informed traders and m clients is equal to m = 2(m + 1) 3 2 (m(m + 2)(1 ) + 1)u
1 2

As before, to make an informed client indifferent between buying the information and not having it, we set c =
2 u (m + 1)2 2 = 3(m + 1)

2 u m(m + 2)(1 ) + 1

1 2

Taking the FOC of the analysts non-fundamental prots, mcm , with respect to m: (1 m2 (2m + 3)(1 )) u =0 3 2 ((m + 1)3 m(m + 1)(m + 2))2 m(m+2)(1 )+1 (A-10)

Therefore the optimal number of clients, m , and fees, c , are implicitly dened by the following equation: 0 = m m m 2 + 1 (1 ) + 1 m (m + 1)2 (1 ) 1 2 2 2 u c = 3(m + 1) m (m + 2)(1 ) + 1 , m Write equation (A-11) as ( ) = 0 where , m) = m 1 m m2 (1 ) + 1 m(m + 1)2 (1 ). ( , m) is a polynomial of degree three decreasing with domain m (0, +) and ranges from 1 ( ). to . Thus, there is a unique real solution given by A-11 such that m = ( , (0)) = 0 or 1 (0) 2 (2(0) (0) solves (0 + 3) = 0, which gives (0) = 1/2. m, lim1 A =
(2m+1) 3 2(m+1) 2 2 (u ) , which is increasing in m; the optimal m is innite.
1

(A-11) (A-12)

, m) is decreasing in m. At m = m , ( , m ) = m (m + 1)2 (1 ) < 0 and at m = m ( , (, m ) = 0 respectively by equations (A-8) and (A-11). Thus (, m ) < (, m ) and m m . 2 Proof of Proposition 3: The separability of information implies that the analyst and proprietary F F trader have the same strategy on fundamental information, or P () = A (). Therefore, we 30

eliminate subscripts in the remainder of the proof. The maximum fee the proprietary desk can extract from the analyst is = N F (m f ) + F (m ) N F (m ), where the superscripts are used to denote prots from each type of information use (NF =non. fundamental, F =fundamental). Simple comparison shows F (m ) > F (m )+f NF F NF F Since (m )+ (m ) > (m )+2 (m ), interdivisional sharing cannot increase overall prots. 2

31