Information about Sellers’ Past Behavior in the Market for Lemons∗ Kyungmin Kim† May 2012

Abstract In dynamic markets under adverse selection, uninformed players’ (buyers’) inferences on the quality of goods rely on the information they have about informed players’ (sellers’) past behavior. This paper examines the roles of different pieces of information about sellers’ past behavior. Agents match randomly and bilaterally, and buyers make take-it-or-leave-it offers to sellers. It is shown that if market frictions are small (low discounting or fast matching), the observability of time-on-the-market improves efficiency, while that of number-of-previous-matches deteriorates it. If market frictions are not small, the latter may improve efficiency. The results suggest that market efficiency is not monotone in the amount of information available to buyers but depends crucially on what information is available under what market conditions. The results have implications for policy makers and market designers who can control information about participants’ trading histories and for a general modeling issue in the literature on dynamic adverse selection. JEL Classification Numbers: C78, D82, D83. Keywords : Adverse selection; decentralized market; information flow; time-onthe-market; number-of-previous-matches; bargaining with interdependent values.

∗ I am grateful to Yeon-Koo Che, Pierre-Andr´ e Chiappori, In-Koo Cho, Jay Pil Choi, Jan Eeckhout, Srihari Govindan, Ayca Kaya, Stephan Lauermann, Ben Lester, Jin Li, Marco Ottaviani, Santanu Roy, Wing Suen, Charles Zheng, and Tao Zhu for helpful comments. I also thank seminar audiences at Columbia, CUHK, HKU, HKUST, Midwest Theory conference, North America Winter Meeting of the Econometric Society, Southern Methodist, UIUC, Western Ontario, and Yonsei. † University of Iowa. Contact: [email protected]

1

1 Introduction Consider an investor (he, buyer) who is interested in acquiring a company of an entrepreneur (she, seller).1 He understands that the entrepreneur has superior information about her own company, and thus there is an adverse selection problem. A low price may be rejected, while a high price runs the risk of overpaying for a lemon. To mitigate this problem, the investor can rely on the information he has about the entrepreneur’s past behavior. If the entrepreneur rejected a good price in the past, it would indicate that her company is even more worthwhile. However, access to the information about the entrepreneur’s past behavior may be limited. Regulation or market practice may not allow it, or relevant records simply may not exist. There are several possibilities. The investor may not get any information or may find out how long the company has been on sale. Or, he may know how many other investors have approached the entrepreneur or even what prices they have offered to her. The investor’s inference will depend crucially on what information he has. What makes the problem non-trivial is the entrepreneur’s strategic incentive. She has an incentive to reject an acceptable price today if, by doing so, she can extract an even better offer tomorrow. The investor must take into account that the entrepreneur has behaved strategically in the past and will respond strategically to his offer. This paper examines the roles of different pieces of information about informed players’ (sellers’) past behavior in the market for lemons. In particular, it investigates the relationship between market performance and the amount of information available to uninformed players (buyers): Does allowing buyers to get access to more information about sellers’ past behavior improve efficiency? Intuitively, there are two opposing arguments. On the one hand, as in reputation games, information about informed players’ past behavior provides information about their intrinsic types, and thus more information would enable uninformed players to better tailor their actions. On the other hand, as in signalling games, if informed players’ current behavior is better observable by future uninformed players, then informed players have a stronger incentive to signal their types, which may cause efficiency losses. The problem is relevant to market designers and policy makers. When setting up or running a marketplace or an industry, they must decide how much and what information to disclose about sellers (and perhaps also about buyers). The decision is crucial, because the information-disclosure policy affects participants’ incentives in the market. In reality, different marketplaces adopt different policies. For example, Craigslist does not provide 1 This

is just an example. The same story applies to any other context where adverse selection matters.

2

any information about sellers, other than that provided by sellers themselves.2 To the contrary, the Multiple Listing Service (MLS) keeps track of the history of each posted property. In particular, it exhibits how many days each property has been on sale. Why do different marketplaces adopt different information policies? Which policy is optimal? This paper sheds some light on these questions. In addition, it has general implications for the literature on dynamic adverse selection. The literature is growing fast, partly due to the recent financial crisis and partly due to theoretical developments within economics.3 However, it is silent about one crucial modeling issue: information flow about informed players’ past behavior. Most studies in the field take a particular form of information flow, typically no information flow in a fullblown market setting and full information flow in a reduced-form strategic setting, but do not provide enough justification for their use. A natural question arises: Would (and, if so, how) the predictions obtained under one information-flow assumption carry over or extend to the environments with different assumptions? This paper offers a unique perspective on this question. The model is a dynamic decentralized version of Akerlof’s market for lemons (1970). In each unit of time, equal measures of buyers and sellers enter the market for an indivisible good. Buyers are homogeneous, while there are two types of sellers. Some sellers possess a unit of low quality, while the others own a unit of high quality. A high-quality unit is more valuable to both buyers and sellers. There are always gains from trade, but type (quality) is private information to each seller. Agents match randomly and bilaterally according to a Poisson rate. In a match, the buyer offers a price and the seller decides whether to accept it or not. If an offer is accepted, trade takes place and the pair leave the market. Otherwise, they stay and wait for the next trading opportunity. The following three information regimes are considered: • Regime 1 (no information) : Buyers do not receive any information about their partners’ past behavior. • Regime 2 (time on the market) : Buyers observe how long their partners have stayed on the market. • Regime 3 (number of previous matches): Buyers observe how many other buyers 2 Craigslist

shows the posting date of each item. But the information does not convey any real information, because sellers can freely repost their ads and, in fact, they have an incentive to do so, in order to be exposed as much as possible. 3 A non-exclusive list of recent works includes Camargo and Lester (2011), Chang (2010), Chari, Shourideh and Zetlin-Jones (2010), Chiu and Koeppl (2011), Daley and Green (2011), Guerrieri and Shimer (2011), Horner ¨ and Vieille (2009a), Kurlat (2010), and Moreno and Wooders (2010).

3

have approached their partners, that is, how many offers their partners have rejected before. A priori the regimes are not fully ordered in the amount of information available to buyers. In particular, the ranking between Regime 2 and Regime 3 is not straightforward.4 It will be shown later that number-of-previous-matches dominates time-on-the-market in the sense that Regime 3 outcome is essentially independent of whether time-on-themarket is jointly observable or not (that is, Regime 3 outcome obtains if both time-on-themarket and number-of-previous-matches are observable). Therefore, the three regimes are de facto fully ordered. I fully characterize steady-state market equilibrium of each regime,5 and compare the performance of the regimes in terms of market efficiency. I first show that if market frictions are small (agents are patient or matching is fast), then the observability of numberof-previous-matches reduces efficiency. There is a cutoff level of market frictions below which Regime 1 weakly Pareto dominates Regime 3: Buyers weakly prefer Regime 1 to Regime 3, while (low-type) sellers strictly prefer Regime 1 to Regime 3. The underlying reason for this result is as follows. In any regime, high-type sellers, due to their higher cost, stay on the market relatively longer than low-type sellers. Therefore, the proportion of high-type sellers in the market is higher than the corresponding proportion among new sellers. In Regime 1, all sellers look identical to buyers, that is, all cohorts of sellers are completely mixed. Therefore, a higher proportion of high-type sellers in the market makes buyers more willing to offer a high price and, therefore, facilitates trade. In Regime 3, the observability of number-of-previous-matches prevents mixing of different cohorts of sellers. Therefore, a higher proportion of high-type sellers in the market does not facilitate trade. This result is in accordance with the result of Horner ¨ and Vieille (2009a) and seems to suggest that less information could always be preferable. Horner ¨ and Vieille study a reduced-form game in which a single seller faces a sequence of buyers and show that the observability of previously-rejected-prices can cause bargaining impasse.6 Precisely, provided that adverse selection is severe and discounting is low, only the first buyer offers 4 On

the one hand, in dynamic environments, time-on-the-market enables informed players to credibly signal their types and/or uninformed players to effectively screen informed players. Therefore, the role of number-of-previous-matches might be just to provide an estimate of time-on-the-market. On the other hand, number-of-previous-matches purely reflects the outcomes of informed players’ decisions, while timeon-the-market is compounded with search frictions. 5 The characterization of each regime is interesting itself, because it can be used to address other substantive questions and also shows how crucially market outcome depends on buyers’ information about sellers’ past behavior. 6 They also study the case where buyers do not observe past offers (private offers). Their setup is in discrete time without search frictions. Since time-on-the-market and number-of-previous-matches are in-

4

a price that can be accepted with a positive probability by the seller. In the two-type case, the low-type seller trades with probability less than 1 at her first match. All subsequent buyers make only losing offers, so trade never occurs afterward. If their game is embedded in the market setting of this paper, no agent obtains a positive expected payoff.7 Therefore, the observability of previously-rejected-prices unambiguously impedes market efficiency. The observability of time-on-the-market, however, enhances efficiency. I show that if market frictions are sufficiently small, realized market surplus is strictly higher in Regime 2 than in Regime 1. To understand this, consider a new seller who quickly meets a buyer. In Regime 3, the opportunity cost of accepting a current offer is rather large, as the seller’s rejection would be observable by future buyers, who would update their beliefs accordingly. In Regime 2, the corresponding opportunity cost is smaller, as future buyers would not know that the seller rejected a price. This difference yields the following consequence: in Regime 2 a low-type seller who quickly meets a buyer trades with probability 1, while in Regime 3 she does not. In Regime 1, the opportunity cost of accepting a current offer is independent of a seller’s private history, and thus a low-type seller often rejects a low price at her first match. Consequently, within a cohort, the proportion of high-type sellers increases faster in Regime 2 than in Regime 1, which induces buyers to offer a high price relatively quickly in Regime 2. In Regime 2, as in Regime 3, different cohorts of sellers are not mixed, which negatively affects efficiency. When market frictions are small, it turns out that the former positive effect more than offsets the latter negative effect, and thus Regime 2 is more efficient than Regime 1. This result demonstrates that efficiency is not monotone in the amount of information available to buyers. It also shows that the results on number-of-previous-matches and previously-rejected-prices stem from the nature of such pieces of information, and not from any general relationship between efficiency and information. In addition, depending on market conditions, the observability of number-of-previousmatches may contribute to efficiency. I show that if market frictions are not sufficiently small, there are cases where Regime 3 is more efficient than the other two regimes. This distinguishable and always observable by buyers, their private case corresponds to the reduced-form game of the regime in which both time-on-the-market and number-of-previous-matches are observable (so, effectively the reduced-form game of Regime 3). 7 The intuition behind the payoff result is as follows. Given that high-type units never trade, buyers offer only the reservation price of low-type sellers, and thus low-type sellers do not obtain a positive expected payoff. Buyers’ zero expected payoff is due to the endogenous distribution of sellers in the market. A buyer gets a positive expected payoff if and only if he meets a new seller. However, some low-type sellers and all high-type sellers stay in the market forever, and thus the probability of a buyer meeting a new seller is negligible.

5

reinforces the argument that efficiency is not monotone in the amount of information. It also points out that what matters is what information is available under what market conditions. The implications of the results go beyond the narrow confines of this paper. Positively, the results illustrate the importance of making the correct information-flow assumption or obtaining more robust results. The predictions obtained under one information-flow assumption may not generalize to the environments with other assumptions. Therefore, when analyzing a market and discussing relevant policy issues, it is crucial to identify true information flow or provide the results that are robust to the potential misspecification of information flow. Normatively, the results suggest that a policy maker can indirectly influence market performance by adjusting the disclosure policy or controlling the information about agents’ trading histories, but they must be done with caution. Market efficiency depends not on the amount of information available to buyers, but on the nature of each piece of information and market conditions. Therefore, if a policy maker can control several pieces of information, then he should examine the role of each piece of information and also understand the characteristics of the market, instead of simply deciding how much information to reveal.

1.1 Related Literature Various dynamic versions of Akerlof’s market for lemons have been developed.8 To my knowledge, none of the previous studies concerns the central issue of this paper, namely information about informed players’ past behavior. One paper that deserves mention is Moreno and Wooders (2010), who study a discrete-time version of Regime 1. The equilibrium characterization of Regime 1 is similar to theirs, but the continuous-time setting of this paper has two advantages. First, it allows a complete characterization under a mild assumption on the discount factor (Assumption 2).9 The complete characterization is important, because some of my main substantive results build upon it. Second, it uncovers the main driving force for their main welfare result (that social welfare is higher in the decentralized market than in the competitive benchmark). In the continuous-time setting 8 To name a few seminal contributions,

Janssen and Karamychev (2002) and Janssen and Roy (2002, 2004) examine settings, with constant inflow of agents or one-time entry, where a single price clears each spot market. Inderst and Muller ¨ (2002) study competitive search equilibrium with constant inflow of agents. Wolinsky (1990), Serrano and Yosha (1993, 1996), Blouin and Serrano (2001), and Blouin (2003) consider various settings, with constant inflow of agents or one-time entry and with two-sided uncertainty or onesided uncertainty, in which agents meet bilaterally and play a simple bargaining game with only two possible transaction prices. Hendel and Lizzeri (1999, 2002) and Hendel, Lizzeri, and Siniscalchi (2005) study dynamic durable goods markets where units are classified according to their vintages. 9 Moreno and Wooders (2010) restrict attention to the case where agents are sufficiently patient.

6

of this paper, as long as market frictions are small, social welfare in Regime 1 is exactly the same as that of the competitive benchmark, independently of the discount rate and the matching rate (see Proposition 1). This shows that their result is driven by their assumption that matching occurs only at the beginning of each discrete-time period. If matching occurs at the end of each period, the difference between the decentralized market and the competitive benchmark also disappears in the discrete-time setting. A few papers study related issues in reduced-form game settings. In a dynamic version of Spence’s signalling model, Noldeke ¨ and Van Damme (1990) show that, although there are multiple sequential equilibria, there is an essentially unique sequential equilibrium outcome that satisfies the never a weak best response requirement (Kohlberg and Mertens (1986)). As the offer interval tends to zero, the unique equilibrium outcome converges to the Riley outcome. Swinkels (1999) points out that the result depends crucially on information flow. He shows that if offers are not observable by future uninformed players (which are observable in Noldeke ¨ and van Damme), then the unique equilibrium outcome is complete pooling with no delay. Using the terminology of this paper, full histories of informed player’s past behavior are observable by future uninformed players in Noldeke ¨ and van Damme, while only time-on-the-market and number-of-previousmatches are observable in Swinkels. Horner ¨ and Vieille (2009a) can be interpreted as considering the two cases (public and private offers) in the bargaining-with-interdependentvalues context. In a related two-period model, Taylor (1999) shows that the observability of previous reservation price and inspection outcome is efficiency-improving, that is, more information about past trading outcome is preferable. The key difference is that buyers’ herding, rather than sellers’ signaling or buyers’ screening, is the main concern in his paper.10 Two papers study the effects of decreasing information asymmetry in the market for lemons. Levin (2001) considers a static setting and shows that less information asymmetry may or may not improve trade. Daley and Green (2011) consider a dynamic setting in which a single seller faces a competitive market and news (public signals about the seller’s type) arrives over time. They show that increasing the news quality may or may not improve efficiency. Despite apparent similarities, the question and the result of this paper are fundamentally different from those of the two papers. They study the effects of exogenously changing the level of information asymmetry given an information struc10 In his model, buyers have no incentive to trade with a low-type seller (no gains from trade of a low-type

unit) and winners conduct an inspection prior to exchange. These cause buyers to get more pessimistic over time, that is, the probability that a seller owns a high-type unit is lower in the second period than in the first period. The better observability of past outcomes improves efficiency by weakening negative buyer herding.

7

ture,11 while this paper examines the effects of changing the information structure. As a result, they are mainly concerned with quantitative aspects of sellers’ incentives in the market for lemons, while this paper is more about qualitative aspects.12 Lastly, the setting of this paper can be interpreted as bargaining taking place in a market. In this regard, this paper is also related to the literature on bargaining with interdependent values. Evans (1989) and Vincent (1989) offer early results and insightful examples. Deneckere and Liang (2006) provide a general characterization for the finite type case and explain the source and mechanics of bargaining delay due to adverse selection. They find an equilibrium structure that is quite similar to the ones in this paper. I explain similarities and differences in Section 7. The remainder of the paper organizes as follows. Section 2 introduces the model. The following three sections analyze each regime. Section 6 compares the welfare consequences of the regimes. Section 7 concludes by discussing a few relevant issues.

2 The Model 2.1 Physical Environment The model is set in continuous time.13 In each unit of time, unit measures of buyers and sellers enter the market for an indivisible good. Buyers are homogeneous, while there are two types of sellers. A fraction qb of new sellers possess a unit of low quality (low type), while the others own a unit of high quality (high type). A unit of low (high) quality costs

c L (c H ) to a seller and yields utility v L (v H ) to a buyer. A high-quality unit is more costly to a seller (c H > c L ≥ 0) and more valuable to a buyer (v H > v L ). There are always gains from trade (v H > c H and v L > c L ), but the quality of each unit is private information to each seller. Agents match randomly and bilaterally according to a Poisson rate λ > 0. In a match, the buyer offers a price and the seller decides whether to accept it or not. If an offer is accepted, then they receive utilities and leave the market. All agents are risk neutral. If 11 In

Daley and Green (2011), buyers observe the seller’s time-on-the-market and number-of-previousmatches. 12 Borrowing from Horner ¨ and Vieille (2009a), “it is important to distinguish between how much information can possibly be revealed given the information structure and the information that is actually revealed in equilibrium.” Levin (2001) and Daley and Green (2011) exogenously control the amount of information that is revealed, while this paper (and Horner ¨ and Vieille (2009a)) examines the relationship between the information structure and the information that is revealed in equilibrium. 13 This is only for tractability. All the results can be established in the corresponding discrete-time setting, but only with rather significant notational complexity.

8

a price p is accepted by a low-type (high-type) seller, then the buyer’s utility is v L − p (v H − p) and the seller’s is p − c L ( p − c H ). Otherwise, they stay in the market and wait for the next trading opportunity. The common discount rate is r > 0. It is convenient to R∞ define δ ≡ λ/ (r + λ) = 0 e−rt d(1 − e−λt ). This is the effective discount factor in this environment that accounts for search frictions as well as discounting. I focus on the case where (1) adverse selection is severe (so high-type units do not trade in the static Walrasian benchmark) and (2) market frictions are small (so agents have non-trivial intertemporal considerations). Formally, I make use of the following two assumptions. Assumption 1 (Severe adverse selection) qbv L + (1 − qb) v H < c H .

This inequality is a familiar one in the adverse selection literature. The left-hand side is a buyer’s willingness-to-pay to a seller who is randomly selected from an entry population. The right-hand side is a high-type seller’s reservation price. When the inequality holds, no price can yield non-negative payoffs to both buyers and high-type sellers, and thus high-type units cannot trade. For future use, let q be the value such that qv L + (1 − q)v H = c H ⇔ q =

vH − cH . vH − vL

A necessary condition for a buyer to be willing to offer c H to a seller is that he believes that the probability that the seller is the low type is less than or equal to q. Assumption 1 is equivalent to qb > q.

Assumption 2 (Small market frictions)

v L − c L < δ (c H − c L ) =

λ (c H − c L ) . r+λ

This assumption states that a low-type seller never accepts any price that a buyer is willing to offer to her (at most v L ) if she expects to receive an offer that a high-type seller is willing to accept (at least c H ) at her next match. Given Assumption 1, this assumption is satisfied when δ is rather large (r is small or λ is large).

9

2.2 Strategies and Equilibrium In order to highlight the economic role of each piece of information, I focus on steadystate equilibrium in which agents of each type employ an identical strategy. In particular, buyers’ offers condition only on sellers’ observable types, and sellers’ acceptance strategies condition only on their intrinsic and observable types.14 Formally, let Ξ be the information set of buyers, that is, the set of sellers’ feasible states that are observable by buyers.15 A buyer’s strategy is a Lebesgue-measurable function B : Ξ × R+ → [0, 1] where B(ξ, p) denotes the probability that the buyer offers p to a seller whose observable type is ξ. A seller’s strategy is a function S : { L, H } × Ξ × R+ → [0, 1] where S(z, ξ, p) denotes the probability that a type (z, ξ ) seller accepts a price p. Finally, define a Lebesgue-measurable function G : { H, L} × Ξ → R+ so that G (z, ξ ) is the steadystate measure of type z sellers whose observable types are less than or equal to ξ. A steady-state equilibrium consists of an offer strategy B(·, ·), an acceptance strategy S(·, ·, ·), and a measure G (·, ·) that satisfy the following requirements: (i) Given G (·, ·) and S(·, ·, ·), any p such that B(ξ, p) > 0 maximizes the expected payoff of the buyer who is facing a type ξ seller. (ii) Given B(·, ·), S(z, ξ, p) > 0 only when p − cz is not less than the continuation payoff of a type (z, ξ ) seller. (iii) Given constant inflow of new agents, B(·, ·) and S(·, ·, ·), G (·, ·) is time-invariant.

2.3 Preliminary Observations I restrict attention to the equilibrium in which each buyer offers the reservation price of a low-type seller or that of a high-type seller. Formally, denote by VS (z, ξ ) a type (z, ξ ) seller’s expected payoff in the market. Consider a buyer who is facing a seller whose observable type will become ξ if she rejects the buyer’s offer.16 I assume that the buyer offers either c L + VS ( L, ξ ) or c H + VS ( H, ξ ). This restriction incurs no loss of generality. First, it is strictly suboptimal for each buyer to offer more than c H + VS ( H, ξ ) or between c L + VS ( L, ξ ) and c H + VS ( H, ξ ).17 Second, in all the three regimes, each seller’s future 14 In

other words, each buyer’s offer is independent of his own history. Each seller’s action depends on her own history, but only through its effect on her observable type. For example, suppose in Regime 2 there are two sellers who have met different numbers of buyers or been offered different prices. If they have stayed on the market for the same length of time, then they are assumed to behave identically. 15 The set Ξ is a singleton in Regime 1. It is isomorphic to the set of non-negative real numbers R in + Regime 2 and to the set of non-negative integers N0 in Regime 3. 16 Rejecting an offer does not affect the seller’s type in Regimes 1 and 2, but increases the seller’s type by 1 in Regime 3. 17 c > c implies c + V ( H, ξ ) ≥ c + V ( L, ξ ) for any ξ ∈ Ξ in all the regimes. H L H L S S

10

type does not depend on the current price.18 Therefore, whenever a buyer makes a losing offer (offer that will be rejected for sure), the offer and the corresponding acceptance probability can be set to be equal to VS ( L, ξ ) and 0, respectively. The following result, which is a straightforward generalization of the Diamond paradox, greatly simplifies the subsequent analysis. Lemma 1 In equilibrium, buyers never offer strictly more than c H . Therefore, high-type sellers’ expected payoffs are always equal to 0 (VS ( H, ξ ) = 0 for any ξ ∈ Ξ in any regime), and c H is always accepted by any seller. From now on, for notational simplicity, I denote by VS (ξ ) a type ( L, ξ ) seller’s expected payoff and by VS a low-type seller’s expected payoff in the market. In addition, I denote by VB a buyer’s expected payoff.

3 Regime 1: No Information In this section, the set Ξ is a singleton. Denote by 0 its only element. Since buyers cannot differentiate different cohorts of sellers, by definition, buyers cannot screen sellers, and sellers cannot signal their types. Under severe adverse selection, one may think that no information flow (and the consequent impossibility of screening and signaling) would cause no trade of high-type units. Such intuition does not apply to the current dynamic setting. Suppose only low-type units trade. Due to constant inflow of agents, the proportion of high-type sellers would keep increasing and eventually the market would be populated mostly by high-type sellers. A buyer would then be willing to offer a high price, which is a contradiction.19 On the other hand, it cannot be that buyers always offer c H and any match turns into trade. If so, the proportion of low-type sellers in the market would be equal to qb, which implies, together with Assumption 1, that buyers’ expected payoff would be strictly negative. The discussion above suggests that in equilibrium, buyers randomize between p∗ ≡

c L + VS and c H , which are the reservation prices of low-type and high-type sellers, respectively. Intuitively, an equilibrium is sustained as follows. High-type sellers accept only c H , while low-type sellers accept both c H and p∗ . Since high-type sellers stay relatively longer than low-type sellers, the proportion of high-type sellers in the market is greater 18 This

claim does not hold if buyers observe previously rejected prices. In fact, the dependence of future offers (and, consequently, sellers’ continuation payoffs) on the current price is the key to the bargaining impasse result of Horner ¨ and Vieille (2009a). 19 Technically, trade of high-type units is a necessary condition for the existence of a steady-state equilibrium.

11

than 1 − qb. This provides an incentive for buyers to offer c H and trade also with high-type sellers. In equilibrium, buyers offer p∗ and low-type sellers accept p∗ with just enough probabilities so that, with the resulting proportion of high-type sellers in the market, buyers are indifferent between p∗ and c H . Formally, the steady-state equilibrium requirements are translated as follows. For notational simplicity, I suppress sellers’ unique observable status and simply let B( p) ≡ B(0, p), S(z, p) ≡ S(z, 0, p), and G (z) ≡ G (z, 0). In addition, let q∗ ≡ G ( L)/(G ( L) + G ( H )) be the proportion of low-type sellers in the market. Buyer optimality:

A buyer must be indifferent between offering p∗ and c H . Therefore,

q∗ S( L, p∗ ) (v L − p∗ ) + (1 − q∗ S( L, p∗ )) VB = q∗ v L + (1 − q∗ ) v H − c H .

(1)

The left-hand side is a buyer’s expected payoff by offering p∗ . The offer is accepted only by a low-type seller, whose probability is q∗ , and conditional on that, with probability S( L, p∗ ). If the offer is not accepted, which occurs with probability 1 − q∗ S( L, p∗ ), the buyer returns to the market and receives market utility VB . The right-hand is his expected payoff by offering c H , which is accepted for sure by both types of sellers. To pin down VB , it suffices to consider a buyer who would offer c H at the next match. Therefore, VB = δ (q∗ v L + (1 − q∗ ) v H − c H ) . Seller optimality:

Since p∗ is the reservation price of low-type sellers, p∗ − c L = VS = δ ( B( p∗ ) p∗ + (1 − B( p∗ )) c H − c L ) .

(2)

The left-hand side is a low-type seller’s payoff by accepting p∗ , while the right-hand side is her expected payoff by rejecting it. If she rejects, then she needs to wait for a new buyer (thus δ), who would offer p∗ with probability B( p∗ ) and c H with the complementary probability. Steady-state condition:

A high-type seller leaves the market if and only if she meets

a buyer and receives the offer c H and, therefore, her market exit rate is λB(c H ). Since a low-type seller accepts p∗ with probability S( L, p∗ ), her exit rate is λ( B( p∗ )S( L, p∗ ) + B(c H )). In steady state, as usual, the inflow of agents of each type must be identical to the

12

corresponding outflow. Therefore, qb = G ( L)λ( B( p∗ )S( L, p∗ ) + B(c H )),

and

1 − qb = G ( H )λB(c H ).

Using q∗ = G ( L)/(G ( L) + G ( H )), the two steady-state conditions can be reduced to qb q∗ B( p∗ )S( L, p∗ ) + B(c H ) = . 1 − qb 1 − q∗ B (c H )

(3)

Intuitively, the left-hand side is the ratio of low-type sellers to high-type sellers among new sellers, while the right-hand side is the corresponding ratio among leaving sellers. The proportion of low-type sellers in the market is invariant only when the two ratios are identical. A steady-state equilibrium in Regime 1 can be described by a tuple (p∗ , B( p∗ ), S( L, p∗ ), q∗ ) that satisfy (1), (2), and (3).20 The following proposition characterizes the equilibrium. Proposition 1 If δ≥

qb (v L − c L ) , qb (v L − c L ) + (1 − qb) (v H − c H )

(4)

then p∗ = v L , VB = 0, and VS = v L − c L . Otherwise, p∗ < v L , VB > 0, and VS < v L − c L . The equilibrium is unique as long as (4) holds or δ ≤ qb.21

Proof. There are 4 equilibrium variables (p∗ , B( p∗ ), S( L, p∗ ), q∗ ), but only 3 equilibrium conditions ((1),(2), and (3)). This problem can be resolved by considering an additional

obvious constraint that, since p∗ is accepted only by low-type sellers, p∗ ≤ v L . The two cases in the proposition correspond to the case where this condition is binding (p∗ = v L ) and the case where it is not (p∗ < v L ). In the former case, it is straightforward that 20 The

full description of the equilibrium is as follows:

• B( p) = 0 if p 6= p∗ , c H (buyers offer only p∗ or c H ), and B(c H ) = 1 − B( p∗ ). • S( L, p) = 0 if p < p∗ and S( L, p) = 1 if p > p∗ (p∗ is low-type sellers’ reservation price). • S( H, p) = 0 if p < c H and S( H, p) = 1 if p ≥ c H (c H is high-type sellers’ reservation price). 21 The

uniqueness is not guaranteed only when qb < δ <

qb (v L − c L ) , qb (v L − c L ) + (1 − qb) (v H − c H )

which can happen only when v H − c H < v L − c L .

13

VB = 0 and VS = v L − c L . The other equilibrium variables can also be easily found. In the latter case, if p∗ is not accepted by low-type sellers with probability 1 (that is, if S( L, p∗ ) < 1), then a buyer has an incentive to deviate and offer a slightly higher price, which is a contradiction. Therefore, S( L, p∗ ) = 1, and all the other variables solve the three equations. See the Appendix for a formal proof. Intuitively, if δ is large, then low-type sellers are willing to accept p∗ only when it is sufficiently high. However, buyers know that p∗ would be accepted only by low-type sellers and, therefore, p∗ ≤ v L . In equilibrium, the inequality binds, and all other results follow from there. If δ is relatively small, sellers are eager to trade quickly and buyers can exploit such incentive. As a result, low-type sellers obtain less than v L − c L , while buyers receive a strictly positive expected payoff.

4 Regime 2: Time on the Market In this section, buyers observe how long their matched sellers have stayed on the market. The set Ξ is isomorphic to the set of non-negative real numbers R+ . A typical element t ∈ Ξ represents the length of time a seller has stayed on the market. Since, unlike in Regime 1, a direct characterization of market equilibrium is very involved, I divide the analysis into the following two stages, as is often done in the search literature. First, I consider the problem between a single seller and a sequence of buyers who sequentially contact the seller. At this stage, I assume that buyers’ continuation payoff (outside option) VB is exogenously given and characterize buyers’ optimal offer strategies B(·, ·) and the seller’s optimal acceptance strategy S( L, ·). Second, I embed the outcome into the market and characterize market equilibrium. In particular, I deduce the steady-state measure G (·) and endogenize buyers’ market expected payoff VB .

4.1 Reduced-form Game The first-stage problem is formally equivalent to the following game. There is a seller whose type is L with probability qb and H with probability 1 − qb. Starting from time 0,

buyers arrive at Poisson rate λ. Each buyer offers a price, conditioning only on the length of time the seller has stayed in the game (equivalently, the time the buyer arrives). If a

price is accepted, then the seller and the buyer receive utilities (p − cz and vz − p, respectively, where z is the seller’s type), and the game ends. Otherwise, the buyer receives his outside option VB , and the seller waits for a new buyer. I refer to the buyer who arrives

14

at time t as time t buyer and the seller who has stayed in the game for t length of time as time t seller. I assume that VB ∈ [0, min { v L − c L , v H − c H }). This is necessary for the existence of a steady-state market equilibrium. Using the same notation as in the original problem, denote by B(t, p) the probability that a buyer who arrives at t offers p and by S(z, t, p) the probability that a type z seller accepts p at time t. In addition, denote by q(t) the probability that the seller is the low type at time t.22 Obviously, q(0) = qb. As in other dynamic games with asymmetric information, time-on-the-market serves

as a screening device. The reservation price of the low type, due to her lower cost, is strictly smaller than that of the high type. Therefore, the low type finishes the game relatively faster than the high type. Buyers offer a low price initially and a high price if the seller has stayed long. Of course, in equilibrium, the seller must wait long enough to receive a high price. Otherwise, the low type would mimic the high type. A class of simple equilibria. I start by constructing a class of equilibria that implement the aforementioned dynamic structure in the simplest way. Let t be the value that satisfies VB =

qbe−λt 1 − qb v + vH − cH . L qbe−λt + (1 − qb) qbe−λt + (1 − qb)

Such t is well-defined, because VB ≥ 0 > qbv L + (1 − qb)v H − c H . In addition, let t be the value such that v L − c L − VB = e−r(t−t)

Z ∞ t

  λ e − r ( t − t) ( c H − c L ) d 1 − e − λ ( t − t ) = e − r ( t − t ) ( c H − c L ). r+λ

Assumption 2 guarantees that t > t. Lastly, for any t˜ ∈ [t, t], let α(t˜) be the value such that v L − c L − VB = e

−r (t˜− t)

Z ∞ t˜

  λα(t˜) ˜ ˜ ˜ ˜ ( c H − c L ). e−r(t−t) (c H − c L )d 1 − e−λα(t)(t−t) = e−r(t−t) r + λα(t˜) r (v − c −V )

It is straightforward to show that α(t˜) is strictly increasing in t˜ from λ(c L −vL +VB ) to 1. H L B For each t˜ ∈ [t, t], consider the following strategy profile (see Figure 1 for the structures of the two extreme cases): 22 When

the reduced-form game is embedded into the market, q(t) corresponds to the proportion of lowtype sellers among those whose observable type is t.

15

Price

Price cH

p(t) vL

VB

p(t)

cL t 0

Low type

t

No trade

t˜ = t

0

Both types

t˜ = t Low type

t Both types

Figure 1: The left panel shows the structure of the equilibrium in which all buyers offer c H for sure after some time (t˜ = t in the example), while the right panel shows that of the equilibrium in which all buyers play an identical strategy after t (t˜ = t in the example). • Let  

  R∞ λα(t˜) c L + t e−r(s−t) (c H − c L )d 1 − e−λα(t˜)(s−t) = r+λα(t˜) (c H − c L ), if t ≥ t˜,   p (t) ≡ R ∞ −r (s − t ) ˜  cL + ˜ e (c H − c L )d 1 − e−λα(t˜)(s−t˜) = e−r(t˜−t) r+λαλα(t()t˜) (c H − c L ), if t < t˜. t

If t < t˜, then the buyer offers p(t) with probability 1. If t ≥ t˜, then the buyer offers c H with probability α(t˜) and p(t˜) with the complementary probability.

• The low-type seller accepts any price strictly greater than p(t) and rejects any price strictly less than p(t). She accepts p(t) with probability 1 if t < t˜ and rejects it with probability 1 if t ≥ t˜. • The high-type seller accepts a price if and only if it is greater than or equal to c H . The strategy profile is an equilibrium for the following reasons: • p(t) is the low-type seller’s reservation price. Each buyer offers the reservation price of the low type or that of the high type. Therefore, for the purpose of calculating the low type’s reservation price, she can be assumed to accept only c H . Since all the buyers after t˜ offer c H with probability α(t˜), the low-type seller’s reservation price at λα(t˜) t ≥ t˜ is p(t) = c L + r+λα(t˜) (c H − c L ). Since no buyer offers c H before t˜, the low-type λα(t˜) seller’s reservation price at t < t˜ is p(t) = c L + e−r(t˜−t) r+λα(t˜) (c H − c L ).

• Buyers’ beliefs evolve as follows: q(t) = max



 qbe−λt qbe−λt . , qbe−λt + 1 − qb qbe−λt + 1 − qb 16

The low-type seller finishes the game whenever a buyer arrives before t and, therefore, the probability that the seller is the low type decreases according to the matching rate. The decrease stops once it reaches t. After t, either no type trades (between t and t˜) or both types trade (after t˜). Therefore, q(t) is constant after t. • Given buyers’ beliefs, it is straightforward that buyers’ strategies are also optimal. If t < t, then the buyer strictly prefers offering p(t) to c H , because q(t)(v L − p(t)) + (1 − q(t))VB > VB > q(t)v L + (1 − q(t))v H − c H . If t ≥ t, then the buyer is indifferent between offering c H and p(t), because q(t )v L + (1 − q(t ))v H − c H = VB , while p(t) is not accepted for sure (in which case the buyer simply gets his outside option VB ). Characterization. The equilibrium strategy profiles above share several notable properties. In all equilibria, the buyers before t offer p(t) = c L + e−r(t−t) (v L − c L − VB ), which is accepted by the low-type seller with probability 1. Buyers’ beliefs about the seller’s intrinsic type evolve exactly in the same way. The buyers after t behave differently, but are indifferent between offering c H and taking their outside option. The following proposition shows that the properties also hold in any other equilibrium. Proposition 2 Given VB , let t be the value such that VB =

1 − qb qbe−λt v L + −λt vH − cH . − λt qbe qbe + (1 − qb) + (1 − qb)

A strategy profile ( B(·, ·), S(·, ·, ·)) is a sequential equilibrium of the reduced-form game if and only if the following properties hold: • Let p (t) ≡ c L +

Z ∞ t

  Rs e−r(s−t) (c H − c L )d 1 − e− t λB(x,c H )dx .

p(t) − c L is the expected payoff of the low-type seller at time t when she accepts only c H . • Buyers: B(t, p(t)) =

(

1, 1 − B(t, c H ),

if t < t, otherwise, and

B(·, c H ) satisfies p(t ) = v L − VB , and 17

(5)

p(t) ≥ v L − VB , ∀t ≥ t.

(6)

In words, buyers offer p(t) with probability 1 until t and randomize between p(t) and c H after t. • Low type: S( L, t, p) =

(

1, if p > p(t) or p = p(t) and t < t, 0, otherwise.

That is, the low type accepts a price above p(t) for sure and rejects a price below p(t) for sure. She accepts p(t) if and only if t < t. • High type: S( H, t, p) =

(

1, if p ≥ c H , 0, otherwise.

That is, the high type accepts a price if and only if it is greater than or equal to c H . Proof. See the Appendix. One subtle equilibrium requirement is (6). It states that the low-type seller’s reservation price p(t) should not fall below v L − VB . The condition is necessary, because if p(t) < v L − VB for some t > t, then the buyer would strictly prefer offering p(t) to c H (because q(t )(v L − p(t)) + (1 − q(t ))VB > VB = q(t )v L + (1 − q(t ))v H − c H ). In all the equilibria in the example above, p(t) is strictly increasing between t and t˜ and constant after t˜. Therefore, the condition is never violated. However, depending on buyers’ offer strategies, it is possible that p(·) decreases over some regions. The two extreme equilibria depicted in Figure 1 are of particular interest in this regard. In the case where t˜ = t, p(t) is constant after t. Therefore, (6) binds for any t ≥ t. This means that buyers cannot offer c H more quickly than in the equilibrium: If buyers offer c H with greater probabilities than α(t ) near t, then (6) would be violated. Now consider the case in which t˜ = t (or α(t) = 1). In this case, p(t) increases fastest. Therefore, the equilibrium satisfies (6) most effectively. This means that buyers cannot offer c H more slowly than in this equilibrium. These properties are not particularly notable in the analysis of the reduced-form game, but turn out to be crucial in the characterization of the set of market equilibria. Although there is a continuum of equilibria, all of them are payoff-equivalent. This is because the evolution of buyers’ beliefs is common across all equilibria. It uniquely determines the low-type seller’s reservation prices before t. Since the expected payoff of the buyer before t depends only on his belief about the seller’s type and the low-type 18

seller’s reservation price, it is also common across all equilibria. Finally, buyers’ beliefs evolve so that once they reach a critical level q(t ), they are indifferent between offering c H and taking their outside option. Therefore, although they behave differently in different equilibria, their expected payoffs are identical. Corollary 1 In any equilibrium of the reduced-form game, the low-type seller’s expected payoff is p(0) − c L , and time t buyer’s expected payoff is q(t)(v L − p(t)) + (1 − q(t))VB (> VB ) if t < t and VB otherwise, where p(t) − c L = e−r(t−t) (v L − c L − VB ), ∀t ∈ [0, t], and q(t) = max



qbe−λt qbe−λt , qbe−λt + (1 − qb) qbe−λt + (1 − qb)



.

Despite the fact that all equilibria are payoff equivalent, the choice of a reduced-form game equilibrium still matters for market equilibrium. As will be explained shortly, it is because different reduced-form equilibria induce different distributions of sellers in the market. Consequently, agents’ payoffs in market equilibrium turn out to differ, depending on which reduced-form game equilibrium is adopted in the market setting.

4.2 Market Equilibrium Now I characterize market equilibrium by embedding a reduced-form game outcome into the stationary market environment. I first deduce the measure G (·, ·) and then endogenize VB . Steady-state distribution. Fix an equilibrium of the reduced-form game. A low-type seller stays (does not trade) until t with probability e−λt if t < t and with probability e−λt e

−

Rt t

λB ( x,c H )dx

if t ≥ t. A high-type seller stays until t with probability 1 if t < t and −

Rt

λB ( x,c )dx

H if t ≥ t. This means that considering a cohort of sellers with probability e t (who enter the market at the same time), the fraction of low-type sellers who stay until

t is equal to e−λt if t < t and e−λt e

−

Rt t

λB ( x,c H )dx if t ≥ t. Rt − t λB ( x,c H )dx

The corresponding fractions of

high-type sellers are equal to 1 if t < t and e if t ≥ t. Now consider the market in which all cohorts of sellers are present simultaneously. Using the results in the previous paragraph, in symmetric steady-state equilibrium, the

19

measure of low-type sellers who have stayed shorter than t is equal to

Similarly,

 R  qb t e−λs ds, if t < t, Rs 0R  R G ( L, t) = t − λs t − λB ( x,c ) dx H  qb 0 e ds + e−λt t e t ds , otherwise.  R  (1 − qb) t 1ds, if t < t, 0R R t − R s λB(x,c H )dx  G ( H, t) = t  (1 − qb) 0 1ds + t e t ds , otherwise.

The function G (·, ·) depends on the equilibrium of the reduced-form game. A vast multiplicity of reduced-form game equilibria implies that G (·, ·) takes various forms after t.23 For example, if the equilibrium depicted in the left panel of Figure 1 (the one in which t˜ = t) is employed, then   −λt   qb 1−e + e−λt (t − t) ,    λ G ( L, t) =  qb 1−e−λt + e−λt t − t + 1−e−λ(t−t) , λ λ

and

G ( H, t) =

(

(1 − qb)t,

(1 − qb) t +

1− e−λ( t−t) λ



if t ≤ t < t. if t ≥ t,

if t < t, ,

if t ≥ t.

If the one in the right panel of Figure 1 (the one in which t˜ = t) is adopted, then whenever t ≥ t, ! − λα(t)(t− t) 1 − e−λt − λt 1 − e G ( L, t) = qb , +e λ λα(t )

and

where

1 − e−λα(t)(t−t) G ( H, t) = (1 − qb) t + λα(t ) α (t ) =

!

,

r (v L − c L − VB ) . λ(c H − v L + VB )

As will be shown shortly, it turns out that only the total measure of sellers in the 23 In

any equilibrium, if t < t, then G ( L, t) = qb

1 − e−λt and G ( H, t) = (1 − qb)t. λ

20

market is crucial for the characterization of market equilibrium. Let G (z) be the total measure of type z sellers in the market, that is, G (z) = limt→∞ G (z, t). The following result shows that G (z) is bounded by those for the two extreme equilibria depicted in Figure 1. Lemma 2 Let     1 − e−λt 1 1 − λt Gt ( L) ≡ qb , Gt ( H ) ≡ (1 − qb) t + , +e t−t+ λ λ λ 

and



1 − e−λt 1 Gt ( L) ≡ qb + e−λt λ λα(t )





1 , Gt ( H ) ≡ (1 − qb) t + λα(t )



.

The total measure of type z sellers associated with any equilibrium of the reduced-form game cannot be smaller than Gt (z) and cannot be larger than Gt (z), that is, G (z) ∈ [ Gt (z), Gt (z)] for both z ∈ { L, H }. Proof. See the Appendix. To understand this result, suppose B(t, c H ) marginally increases for some t > t. This increases the low-type seller’s expected payoff at t. Since the equilibrium requires that her reservation price at t be always equal to v L − VB (see (5) in Proposition 2), this means that B(s, c H ) must decrease at some s 6= t. Suppose s > t. The increase of B(t, c H ) decreases G (z) (because the seller finishes the game faster now), while the decrease of B(s, c H ) increases G (z). The net effect on G (z) is, however, positive. This is because sellers discount future payoffs, while the discount rate plays no role for G (z). If the seller’s discount rate is zero, then B(t, c H ) and B(s, c H ) are perfect substitutes. Therefore, the decrease of B(s, c H ) necessary to keep the seller indifferent between the changes of B(·, c H ) is exactly the same as that to keep G (z) constant. When the discount rate is positive, the former is larger than the latter. Therefore, the changes of B(·, c H ) always increases G (z). Since B(·, c H ) increases fastest in the equilibrium with t˜ = t and slowest in the equilibrium with t˜ = t, G (z) ∈ [ Gt (z), Gt (z)]. Endogenizing VB .

Now I close the model by endogenizing VB . Fix an equilibrium in

the reduced-form game, and let G (t; VB ) ≡ G ( L, t; VB ) + G ( H, t; VB ) be the measure of the sellers who have stayed shorter than t. In addition, let G (VB ) be the measure of all sellers in the market. I include VB as an argument of all these functions and other equilibrium variables (such as t(VB ) and p(t; VB )) in order to make explicit their dependence on VB .

21

In market equilibrium, buyers’ outside option VB in the reduced-form game must coincide with their market utility. Therefore, VB must satisfy VB = δ

Z

t(VB ) 0

dG (t; VB ) (q(t; VB )(v L − p(t; VB )) + (1 − q(t; VB ))VB ) + G (VB )

 dG (t; VB ) . VB G (VB ) t(VB )

Z ∞

If a buyer meets a seller who has stayed shorter than t, trade occurs, at price p(t; VB ), if and only if the seller is the low type, whose probability is q(t; VB ). In all other cases, the buyer obtains VB , whether trade occurs or not. Rearranging the terms with VB = v L − p(t; VB ), Z t(VB ) δ qb e−λt ( p(t; VB ) − p(t; VB ))dt. VB = 1 − δ G (VB ) 0

Define a correspondence Φ : [0, min {v L − c L , v H − c H }) ⇒ [0, min {v L − c L , v H − c H }) so that   Z t(VB ) δ qb − λt Φ(VB ) = e ( p(t; VB ) − p(t; VB ))dt : G (VB ) ∈ [ Gt (VB ), Gt (VB )] . 1 − δ G (VB ) 0 Then it is a market equilibrium if and only if VB is a fixed point of the correspondence Φ.

Proposition 3 There exists a continuum of market equilibria in Regime 2. Buyers’ expected payoff is maximized (minimized), while low-type sellers’ expected payoff is minimized (maximized) when the reduced-form game equilibrium in which t˜ = t (t˜ = t) is embedded into the market environment. Proof. t(VB ) always satisfies

qbe−λt(VB ) v L +(1− qb)v H qbe−λt(VB ) +(1− qb)

≥ c H . Therefore, t(VB ) is bounded away

from 0 even if VB tends to zero. In addition, Gt (VB ) is finite for any VB . Therefore, the minimum value of Φ(0) > 0. On the other hand, if VB is sufficiently close to min{v L − c L , v H − c H }, then the maximum value of Φ(VB ) < VB , because no buyer can obtain more than min{v L − c L , v H − c H } from any seller, but matching takes time. The existence of a continuum of equilibria follows from the fact that Φ(VB ) is an interval (convex) for any VB 24 and Φ(VB ) changes smoothly in VB . The results regarding the equilibria that maximize and minimize agents’ payoffs follow from the fact that the maximum (minimum) of Φ(VB ) is achieved in the equilibrium in which t˜ = t (t˜ = t). 24 It

is straightforward that the class of simple equilibria I constructed in the reduced-form game spans the interval.

22

5 Regime 3: Number of Previous Matches In this section, the set Ξ is isomorphic to the set of non-negative integers, N0 . A typical element n ∈ Ξ represents the number of matches a seller has experienced in the market. As in Regime 2, I first analyze the reduced-form game between a seller and a sequence of buyers and then characterize market equilibrium.

5.1 Reduced-form Game The reduced-form game is almost identical to that of Regime 2. The only difference is that now buyers can condition only on the number of previous buyers, that is, on how many offers the seller has rejected until then. Using the same notation as in the original market setting, denote by B(n, p) the probability that a buyer offers p to a seller who has rejected n offers in the past. Also, denote by S(z, n, p) the probability that a type z seller accepts p in her (n + 1)-th match. As in Regime 2, I first present a class of simple equilibria and then provide a complete characterization of the set of all equilibria. A class of simple equilibria. Let α be the value such that ∞

v L − c L − VB =

δα

∑ δk (1 − α)k−1 α(c H − c L ) = 1 − δ + δα (c H − c L ).

k=1

α uniquely exists, because Assumption 2 ensures v L − c L − VB ≤ v L − c L < δ(c H − c L ). For any α ∈ [α, 1], let n(α) and β(α) be the values such that n(α) is an integer, β(α) < α, and ! ∞

v L − c L − VB = δn(α)−1

= δ

n (α)−1

β(α) + (1 − β(α))



∑ δ k (1 − α ) k − 1 α

k=1

δα β(α) + (1 − β(α)) 1 − δ + δα



(c H − c L )

( c H − c L ).

The pair of n(α) and β(α) uniquely exist: As α increases, β(α) continuously decreases. When β(α) reaches 0, n(α) increases by 1, β(α) jumps to α, and this process continues. Fix α ∈ [α, 1], and consider the following strategy profile (see Figure 2 for the structures of the two extreme cases):

23

Price

Price cH

cH p(n)

vL

vL

cL

cL

p(n)

n

n 0

1

2

3

4

0

1

2

3

4

Figure 2: The left panel shows the structure of the equilibrium in which all buyers offer c H for sure once the seller experiences a certain number of matches (α = 1 in the example), while the right panel shows that of the equilibrium in which all buyers offer c H with probability α after the first match. • Let p (n) ≡

(

δα if n ≥ n(α) − 1,  c L + 1−δ+δα (c H − c L ), δα n ( α )− n − 1 β(α) + (1 − β(α)) 1−δ+δα (c H − c L ), if n < n(α) − 1. cL + δ

If n < n(α) − 1, then the buyer offers p(n) with probability 1. If n = n(α) − 1, then the buyer offers p(n) with probability β(n) and c H with the complementary probability. If n > n(α) − 1, then the buyer offers p(n) with probability α(n) and c H with the complementary probability. • The low-type seller accepts any price strictly above p(n) and rejects any price below 1− qb − c H −VB p(n). She accepts p(0) with probability 1 − qb vc HH − v L +VB , while rejects p(n) for any n > 0.

• The high-type seller accepts a price if and only if it is greater than or equal to c H . The strategy profile is an equilibrium for the following reasons: • p(n) is the low-type seller’s reservation price when her observable type is n. For the same reason as in Regime 2, her reservation price can be calculated by assuming that she would accept only c H . p(n) is constructed exactly that way. • Buyers’ beliefs about the seller’s intrinsic type are q(0) = qb and q(n) =

1− qb v H − c H −VB qb c H − v L +VB 1− qb v H − c H −VB qb qb c H −v L +VB + (1 − qb)

qb

24

=

v H − c H − VB , ∀n > 0. vH − vL

q(n) is constant after the seller’s first match, because the seller accepts only c H thereafter. • The first buyer obtains VB if he follows the strategy. If he offers c H , then his expected payoff is negative, because qbv L + (1 − qb)v H < c H . Therefore, he has no incentive to deviate. All the other buyers are indifferent between offering c H and p(n): Observe

that q(n) is such that q(n)v L + (1 − q(n))v H − c H = VB .

Characterization. As in Regime 2, various properties of the simple equilibria generalize to any other equilibrium. Proposition 4 A strategy profile ( B(·, ·), S(·, ·, ·)) is an equilibrium of the reduced-form game if and only if the following properties hold: • Let ∞

p (n) ≡ c L +

∑ k= n +1





!

−1 δk−n Πkl = n +1 (1 − B(l, c H )) B(k, c H ) (c H − c L ).

• Buyers: B(n, p(n)) =

(

1, 1 − B(n, c H ),

if n = 0, otherwise, and

B(·, c H ) satisfies p(0) = v L − VB , and

(7)

p(n) ≥ v L − VB , ∀n ≥ 1.

(8)

In words, the first buyer offers p(0) with probability 1, while all the other buyers mix between p(n) and c H . • Low type:     1, S( L, n, p) = 1−    0,

if p > p(n), 1− qb v H − c H −VB qb c H − v L +VB ,

if n = 0 and p = p(n), otherwise.

That is, the low type accepts a price above p(n) for sure and rejects a price below p(n) for sure. She accepts p(n) with a positive probability only at her first match (n = 0).

25

• High type: S( H, n, p) =

(

1, if p ≥ c H , 0, otherwise.

That is, the high type accepts a price if and only if it is greater than or equal to c H . Again, as in Regime 2, all the equilibria are payoff equivalent. Corollary 2 In any equilibrium of the reduced-form game, the low-type seller’s expected payoff is δ(c L − v L − VB ), and the expected payoff of any buyer is VB . Proof. The low-type seller is indifferent between accepting p(0) = v L − VB at her first match. Therefore, her market expected payoff is δ(v L − c L − VB ) (note that she needs to wait until the first buyer arrives). If p(0) is accepted, then the first buyer’s payoff is VB . Since he obtains VB even if the offer is not accepted, his expected payoff is VB . It is optimal for any other buyer to make a losing offer. Therefore, their expected payoff is also VB . One difference from Regime 2 is that even the first buyer does not obtain more than VB . To see why, suppose the first buyer can obtain more than VB . Since c H is obviously suboptimal for the first buyer (whose belief about the seller’s intrinsic type is qb), this

means that the low-type seller’s reservation price must be strictly less than v L − VB . This implies that p(0) must be accepted with probability 1: Otherwise, the buyer has an incentive to offer a slightly higher price, because it would be accepted with probability 1 then. This in turn implies that the second buyer would believe that the seller is the high type with probability 1 and, therefore, offer c H . But then, by Assumption 2, the low-type seller would strictly prefer rejecting p(0), which is a contradiction. It will be shown shortly that this difference has significant implications for market equilibrium. In particular, unlike in Regime 2, the choice of the reduced-form game equilibrium does not affect agents’ payoffs in market equilibrium.

5.2 Market Equilibrium Now I characterize market equilibrium by embedding a reduced-form game equilibrium into the market. As in Regime 2, the steady-state measures of sellers’ types associated with each reduced-form game equilibrium can be deduced and then buyers’ market utility VB can be endogenized. I skip the first step, because its procedure is similar to that of Regime 2, while the following result implies that agents’ payoffs can be determined, independently of the distribution of sellers’ types (which depends on the choice of a reducedform equilibrium). 26

Lemma 3 In any market equilibrium, buyers’ expected payoff is 0. Proof. As shown above, a buyer’s expected payoff is VB , no matter which seller he meets. Yet, matching takes time. Therefore, VB = δVB , which implies that VB = 0. Given this result and Proposition 4, the following result is straightforward. Proposition 5 There exists a continuum of market equilibria in Regime 3, but all of them are payoff equivalent. In any equilibrium, buyers’ expected payoff is 0, and low-type sellers’ expected payoff is δ(v L − c L ). I conclude this section with two remarks on the information structure. First, in the current two-type case, less information suffices to produce essentially the same outcome. The only necessary information is whether sellers have been matched before or not. If only that information is available, the equilibrium described in the right panel of Figure 2 (the one in which α = α) is the unique equilibrium. Second, the results do not change even if time-on-the-market is also observable. In particular, the result that no buyer obtains more than VB in the reduced-form game, which is crucial for the characterization of agents’ payoffs in market equilibrium, is independent of the observability of time-onthe-market. The additional information may be used as a public randomization device to enlarge the set of equilibria, but does not affect agents’ payoffs.

6 Welfare Comparison This section compares the welfare consequences of the regimes.

6.1 Regime 1 vs. Regime 3: the role of number-of-previous-matches Suppose market frictions are relatively small (r is small or λ is large). In particular, qb(v − c )

for simplicity, assume that δ ≥ qb(v −c )+(L1−qbL)(v −c ) . It then immediately follows from L L H H Propositions 1 and 5 that Regime 1 weakly Pareto dominates Regime 3: Buyers are indifferent between the two regimes, while low-type sellers are strictly better off in Regime 1 (v L − c L ) than in Regime 3 (δ (v L − c L )).25 Why is Regime 1 more efficient than Regime 3? In Regime 1, different cohorts of

sellers are completely mixed. Such mixing helps relax the incentive constraint for buyers to offer a high price, because high-type sellers, due to their higher cost, stay on the market 25 There is another cutoff, say δ, such that if δ falls between δ and

sellers and buyers strictly prefer Regime 1 to Regime 3.

27

qb( v L −c L ) , then both low-type qb( v L −c L )+(1−qb)( v H −c H )

relatively longer than low-type sellers and, therefore, the proportion of high-type sellers is larger in the market than among new sellers. In Regime 3, buyers’ access to numberof-previous-matches does not allow such mixing. Consequently, a higher proportion of high-type sellers in the market does not facilitate trade. In particular, buyers never offer c H to new sellers in Regime 3, while they often do in Regime 1. One can easily check that this is exactly the reason why low-type sellers’ expected payoff is higher in Regime 1 than in Regime 3.

6.2 Regime 1 vs. Regime 2: the role of time-on-the-market When market frictions are relatively small, Regime 1 and Regime 2 are not Pareto ranked. Buyers always obtain a positive expected payoff in Regime 2 and, therefore, strictly prefer Regime 2 to Regime 1. To the contrary, low-type sellers strictly prefer Regime 1 to Regime 2. Their expected payoff in Regime 2 is p (0) − c L , which is strictly smaller than v L − c L . The two regimes can still be compared in terms of total market surplus, as the model is of transferable utility. In addition, the environment is stationary and I have focused on steady-state equilibrium. Therefore, attention can be further restricted to total market surplus of a cohort, that is, qbVS + VB .26

The following proposition shows that (at least) when market frictions are sufficiently small, Regime 2 outperforms Regime 1.27

Proposition 6 When market frictions are sufficiently small, total market surplus of a cohort is greater in any equilibrium of Regime 2 than in the equilibrium of Regime 1. Proof. When market frictions are small, VS = v L − c L and VB = 0 in Regime 1, and thus qbVS + VB = qb (v L − c L ). In the Appendix, I prove that if λ is sufficiently large or r is sufficiently small, then qbVS + VB > qb (v L − c L ) in Regime 2. The proof proceeds in two steps. First, I show that VS and VB approach v L − c L and 0, respectively, as λ tends to infinity or r tends to 0. I then show that qbVS + VB is decreasing at the limit. Why does the observability of time-on-the-market improve efficiency, while that of

number-of-previous-matches deteriorates it? As with number-of-previous-matches, the observability of time-on-the-market prevents mixing of different cohorts of sellers, which 26 Recall

that high-type sellers’ expected payoff is always zero and the measures of low-type sellers and buyers in each cohort are qb and 1, respectively. 27 Sufficiently small market frictions are only a sufficient condition. Numerical simulations show that Regime 2 outperforms Regime 1 even when market frictions are not so small. The difficulty of characterizing agents’ expected payoffs in Regime 2 does not allow me to perform a more comprehensive welfare comparison.

28

is negative for efficiency. However, there is an offsetting effect. To see this, consider a new seller who quickly meets a buyer. In Regime 3, the seller has an incentive to reject a low price, because doing so will convince future buyers that she is the high type with a high probability. In Regime 2, the same incentive is present but weaker than in Regime 3. The length of time a seller has to endure to receive a high price is independent of whether, and how many times, the seller has rejected offers. Therefore, the opportunity cost of accepting a current offer is smaller in Regime 2 than in Regime 3 (recall that a low-type seller accepts a low price with probability 1 if she matches before t in Regime 2, while she does with probability less than 1 at her first match in Regime 3). Consequently, the proportion of high-type sellers within a cohort increases fast in Regime 2, which induces buyers to offer c H relatively quickly. In Regime 1, sellers cannot signal their types by rejecting offers or waiting for a certain length of time. Still, they have an incentive to wait for a high price. This incentive is constant over time in Regime 1, while it is increasing in Regime 2 (and Regime 3). For sellers who are quickly matched, this incentive is stronger in Regime 1 than in Regime 2. As a result, low-type sellers often reject their reservation price in qb(v − c )

Regime 1 (recall that B( p∗ ) < 1 whenever δ > qb(v −c )+(L1−qbL)(v −c ) ). This directly slows L L H H down trade of low-type units and indirectly discourage buyers to offer c H , by increasing the proportion of low-type sellers in the market. When market frictions are small, this offsetting effect turns out to dominate, and thus Regime 2 outperforms Regime 1.

6.3 The role of market frictions The following proposition shows that depending on market conditions, the observability of number-of-previous-matches may improve market efficiency. Proposition 7 For a fixed δ, if v H − c H is sufficiently small, then realized market surplus of a cohort is close to zero in Regimes 1 and 2, while it is equal to qbδ (v L − c L ) in Regime 3.

Proof. Recall that in Regime 3, VS = δ (v L − c L ) and VB = 0 as long as Assumptions 1 and 2 are satisfied. It suffices to show that in Regimes 1 and 2, both VS and VB converge to zero as v H tends to c H . The result for Regime 1 is immediate from the characterization in Section 3. In Regime 2, VB < v H − c H , and thus obviously VB converges to 0 as v H tends to c H . For VS , observe that when v H is close to c H , q is close to 0. Since q (t) ≤ q (otherwise, buyers would never offer c H ), t must be sufficiently large. This implies that p (0) is close to c L . The intuition for this result is as follows. In any regime, buyers offer c H to some sellers. Their benefit of offering c H depends on v H − c H , while their opportunity cost 29

is independent of v H − c H . If buyers obtain a positive expected payoff, they are less willing to offer c H as v H − c H gets smaller. This is exactly what happens in Regimes 1 and 2.28 In the limit, buyers offer c H with probability 0. Then, as in Diamond (1971), buyers offer c L with probability 1 and low-type sellers do not obtain a positive expected payoff. Buyers’ expected payoff would be also zero. A buyer would get v L − c L if he meets a lowtype seller, but the probability of a buyer meeting a low-type seller would be zero, as the market would be populated mostly by high-type sellers. In Regime 3, the effectiveness of number-of-previous-matches as a signalling device leaves no surplus to buyers in any circumstance and, therefore, makes their incentive to offer c H independent of v H − c H . This prevents all the surplus from disappearing even as v H − c H approaches zero. Rather informally, number-of-previous-matches serves as sellers’ signalling device (while time-on-the-market as buyers’ screening device). Using it for signalling purpose is socially wasteful in itself, as in other signalling games. However, it also allows informed players (sellers) to secure a certain payoff. When market frictions are small, the former (negative) effect dominates and thus the observability of number-of-previous-matches reduces efficiency. When market frictions are not small, the latter effect could be significant. In particular, when all agents’ expected payoffs may be driven down close to zero, the observability of number-of-previous-matches can contribute to market efficiency by preserving informed players’ information rents.

7 Discussion Liquidity and Information The presence of low-type units and information asymmetries between sellers and buyers slow down trade of high-type units. Several recent studies utilize this phenomenon to model the (il)liquidity of high-type units (assets) and study relevant policy issues.29 Here, I illustrate that the liquidity of high-type units depends crucially on information flow. The liquidity of high-type units can be measured as the discounted average time to sell a high-type unit: E[e−rτ ] where τ is a random (stopping) time that a seller is offered c H . This seems to be appropriate in the current dynamic environment, because it reflects not only the length of time necessary to sell a high-type unit, but also discounting. Notice that VS = E[e−rτ ](c H − c L ), that is, a low-type seller’s expected payoff is equal to the measure of liquidity times c H − c L . This is because a low-type seller’s incentive qb(v −c )

28 Buyers

obtain a positive expected payoff in Regime 1 whenever δ < qb(v −c )+(L1−qbL)(v −c ) L L H H 29 See, e.g., Camargo and Lester (2011), Chang (2010), and Guerrieri and Shimer (2011).

30

compatibility is binding whenever she is offered a lower price than c H : Each buyer offers either the reservation price of a low-type seller or c H . Therefore, a low-type seller can still obtain VS by simply waiting for c H . It then follows that Regime 1 is most liquid for high-type units if market frictions are qb(v − c ) relatively small, but may not be the case otherwise. Precisely, if δ ≥ qb(v −c )+(L1−qbL)(v −c ) , L

L

H

H

then low-type sellers’ expected payoff and, therefore, the liquidity of high-type units are highest in Regime 1 (v L − c L > p(0) − c L , δ(v L − c L )). However, as shown in Proposition 7, for a fixed δ, if v H − c H is sufficiently small, then low-type sellers’ expected payoff

and, therefore, the liquidity of high-type units are higher in Regime 3 than in Regime 1. The underlying reasons for these results are essentially identical to those for the welfare results in Section 6.

More than Two Types Equilibrium characterization can generalize beyond the tractable two-type case, with significant technical difficulties but without any conceptual difficulties. In Regime 1, buyers would play a mixed bidding strategy over the set of sellers’ reservation prices. The bidding strategy and sellers’ acceptance strategies would be jointly determined so that the distribution of seller types in the market is invariant and buyers are indifferent over the prices. In Regimes 2 and 3, the equilibrium structures would extend similarly to Deneckere and Liang (2006). Welfare comparison, however, would be significantly more involved. Closed-form solutions, which greatly simplify the welfare comparison in the previous section, would not be available in any regime, unless some strong restrictions are imposed on the parameter values.

Comparison with Deneckere and Liang (2006) The equilibrium structures of Regimes 2 and 3 resemble that of Deneckere and Liang (2006) (DL, hereafter). In particular, DL’s limit outcome (as the offer interval tends to zero) exhibits the same qualitative properties as in the equilibria described in the left panels of Figures 1 and 2: trade occurs either at the beginning of the game or only after some real-time delay. There are two important differences. First, it is only the limit outcomes that have the same qualitative properties. Away from the limit, trade occurs with a positive probability in every period in DL, while there is always an interval of time or a set of matches during which trade does not occur in the equilibria of Regimes 2 and 3. Second, in the

31

comparable limit case (λ is arbitrarily large in Regimes 2 and 3), the equilibrium delay is half as much as that of DL. To be more precise, let τ be the value that satisfies (v L − c L ) = e−rτ (c H − c L ). The equilibrium delay is equal to 2τ in DL, while it is equal to τ in Regimes 2 and 3.

Further Questions The results in the paper raise several further questions. A question particularly germane to this paper is what information flow would be most efficient. When market frictions are sufficiently small, is it possible to improve upon Regime 2? Also, what information flow is optimal for buyers?30 More generally, one may ask what is the constrained-efficient benchmark in dynamic markets under adverse selection with constant inflow of agents (with or without search frictions). Different from static settings, with constant inflow of agents, as shown in Regime 1, subsidization across different cohorts of sellers is possible. Would the social planner exploit such possibility or completely separate different cohorts of sellers? Would the constrained-efficient outcome be stationary, cyclical (for example, high-type units trade every n periods), or non-stationary? One potentially interesting extension, which is also suggested by Horner ¨ and Vieille (2009a), is to allow buyers to conduct inspections before or after bargaining and with or without cost. With inspections, buyers’ beliefs can evolve in any direction. When a unit remains on the market for a long time, it might be because the previous offers have been rejected by the seller (signalling effect), or because the previous buyers have observed bad signals about the unit (herding effect). The former inference shifts buyers’ beliefs upward, while the latter does the opposite. It might be interesting to examine a setting in which both effects are present and study which effect dominates and what consequences it would have on the relationship between efficiency and information about past trading outcomes.

Appendix: Omitted Proofs Proof of Proposition 1: (1) Existence when δ ≥

qb(v L − c L ) . qb(v L − c L )+(1− qb)(v H − c H )

30 Low-type sellers’ expected payoff is bounded by v − c in any circumstance, and thus Regime 1 is the L L one (of possibly many) that is optimal for sellers, as long as market frictions are relatively small.

32

Suppose p∗ = v L . Then from the equilibrium conditions, vH − cH , vH − vL δ (c H − c L ) − (v L − c L ) , B( p∗ ) = δ (c H − v L ) 1 − B( p∗ ) qb(1 − q∗ ) − (1 − qb)q∗ . S( L, p∗ ) = B( p∗ ) (1 − qb)q∗ q∗ = q =

Under Assumption 2, B( p∗ ) is always well-defined. S( L, p∗ ) is well-defined if and only if qb(1 − q∗ ) − (1 − qb)q∗ B( p ) ≥ . qb(1 − q∗ ) ∗

Applying the solutions for q∗ and B( p∗ ) and arranging terms, δ≥ (2) Uniqueness when δ ≥

qb(v L − c L ) . qb(v L − c L ) + (1 − qb)(v H − c H )

qb(v L − c L ) . qb(v L − c L )+(1− qb)(v H − c H )

It suffices to show that there cannot exist an equilibrium with p∗ < v L . Suppose p∗ < v L . Then, q∗ < q and S( L, p∗ ) = 1. From sellers’ indifference condition,

(1 − δ)( p∗ − c L ) (1 − δ)(v L − c L ) > 1− . ∗ δ (c H − p ) δ (c H − v L )

B( p∗ ) = 1 −

On the other hand, from the steady-state condition and the fact that q∗ < q, B( p∗ ) = 1 −

(1 − qb)q∗ (1 − qb)q < 1 − . qb(1 − q∗ ) qb(1 − q)

There exists B( p∗ ) that satisfies both inequalities if and only if

Arranging terms,

(1 − δ)(v L − c L ) (1 − qb)q < . δ (c H − v L ) qb(1 − q) δ<

qb(v L − c L ) . qb(v L − c L ) + (1 − qb)(v H − c H ) qb(v − c )

(3) Existence when δ < qb(v −c )+(L1−qbL)(v −c ) . L L H H In this case, S( L, p∗ ) = 1. Then, from the steady-state condition and sellers’ indiffer-

33

ence condition, B( p∗ ) =

qb − q∗ δ(1 − B( p∗ )) ∗ , and p = c + ( c H − c L ). L qb(1 − q∗ ) 1 − δB( p∗ )

Applying these expressions to buyers’ indifference condition, q∗ (1 − δ + δq∗ )



δ(1 − qb)q∗ vL − cL − (c − c L ) (1 − δ)qb(1 − q∗ ) + δ(1 − qb)q∗ H

The right-hand side strictly decreases from v H − c H to 0 as

q∗



= q ∗ v L + (1 − q ∗ ) v H − c H .

(9) increases from 0 to q. The

left-hand side is 0 if q∗ = 0. If q∗ = q, then the second term on the left-hand side is δ(1 − qb)(v H − c H ) (c − c L ) (1 − δ)qb(c H − v L ) + δ(1 − qb)(v H − c H ) H (1 − δ)qb(c H − v L )(v L − c L ) − δ(1 − qb)(v H − c H )(c H − v L ) (c H − c L ) = (1 − δ)qb(c H − v L ) + δ(1 − qb)(v H − c H ) (1 − δ)qb(v L − c L ) − δ(1 − qb)(v H − c H ) (c − c L )(c H − v L ). = (1 − δ)qb(c H − v L ) + δ(1 − qb)(v H − c H ) H vL − cL −

This expression is strictly positive if and only if δ <

qb(v L − c L ) . qb(v L − c L )+(1− qb)(v H − c H )

Therefore, there

exists q∗ that solves (9). Once q∗ is found, the other two variables are straightforward. qb(v − c ) (4) Uniqueness when δ < qb(v −c )+(L1−qbL)(v −c ) and δ ≤ qb. L L H H I show that if δ ≤ qb then the left-hand side in (9) is strictly concave. Since the righthand side is linear in q∗ , the result immediately follows. Let f (q) = and g (q) = v L − c L −

q > 0, 1 − δ + δq

δ(1 − qb)q (c − c L ) > 0. (1 − δ)qb(1 − q) + δ(1 − qb)q H

34

Then, for q ∈ (0, q), 1−δ > 0, (1 − δ + δq)2 2δ(1 − δ) f 00 (q) = − < 0, (1 − δ + δq)2 (1 − δ)δb q (1 − qb) (c H − c L ) < 0, g 0 (q) = − ((1 − δ)qb(1 − q) + δ(1 − qb)q)2 2(1 − δ)δb q (1 − qb)(−qb + δ) g00 (q) = (c H − c L ) ≤ 0. ((1 − δ)qb(1 − q) + δ(1 − qb)q)2 f 0 (q) =

The second derivative of the left-hand side in (9) is f 00 (q) g(q) + 2 f 0 (q) g0 (q) + f (q) g00 (q) and, therefore, is strictly negative. Q.E.D.

Proof of Proposition 2: Since the sufficiency part is essentially identical to the arguments made for the class of simple equilibria, I prove only the necessity part here. (1) Fix any (sequential) equilibrium of the reduced-form game. Let t∗ be the first moment at which buyers become indifferent between offering c H and taking their outside option VB . Formally, t∗ is the infimum value such that q(t∗ )v L + (1 − q(t∗ ))v H − c H = VB . (2) t∗ is finite, and t∗ ≥ t. Suppose not. Then, buyers would never offer c H . Then, by the same reasoning as in the Diamond paradox, all buyers would offer c L , and the low type would accept it for qbe−λt sure. This means that q(t) = qbe−λt +1−qb and, therefore, t∗ = t, which is a contradiction. Also, since q(t) ≥

qbe−λt

qbe−λt +1− qb

for any t in any equilibrium, t∗ ≥ t.

(3) After t∗ , trade occurs only at c H and, therefore, buyers’ beliefs q(t) do not change after t∗ , that is, q(t) = q(t∗ ) for any t ≥ t∗ . Suppose not. Let t1 (≥ t∗ ) and t2 be the first and last moments at which trade occurs at a different price from c H with a positive probability.31 Since buyers after t1 can ensure strictly more than VB by offering c H (q(t)v L + (1 − q(t))v H − c H > q(t∗ )v L + (1 − 31 t is obviously finite, while t is finite for the following reason: Suppose not. Since no buyer makes a 2 1 losing offer after t1 , if t2 is not finite, then q(t) converges to 0 as t tends to infinity. But, then for sufficiently large t, q(t)v L + (1 − q(t))v H − c H > q(t)(v L − p(t)) + (1 − q(t))VB = q(t)(v L − c L ) + (1 − q(t))VB , that is, buyers strictly prefer offering c H to p(t) and, therefore, would offer only c H .

35

q(t∗ ))v H − c H = VB ), they would never make a losing offer. This implies that all buyers after t2 would offer c H for sure. This in turn implies that the low-type seller would accept only a price close to δ(c H − c L ) right before t2 . Obviously, buyers would never be willing to offer such a high price and, therefore, trade cannot occur at a lower price than c H right before t2 . This contradicts to the fact that buyers never make a losing offer after t2 . (4) It then follows that p (t) − c L =

Z ∞ t

  Rs e−r(s−t) (c H − c L )d 1 − e− t λB(x,c H )dx ≥ v L − c L − VB ,

for any t ≥ t∗ and that p(t∗ ) − c L = v L − c L − VB . If the first condition is violated, then the low type’s reservation price would fall short of v L − VB , in which case trade would occur at a different price from c H after t∗ . If p(t∗ ) − c L > v L − c L − VB , then there cannot be any trade right before t∗ , because if so the buyer would obtain strictly less than VB . But, then q(·) does not change right before t∗ , which contradicts the definition of t∗ . (5) t∗ = t. This stems from the fact that the low-type seller’s reservation price strictly increases before t∗ . At any point in time before t∗ , buyers offer the reservation price of the low-type seller. Therefore, to calculate the low-type seller’s continuation payoff, it can be assumed that the low type simply rejects all the offers until t∗ . This implies that the low type’s reservation price at t < t∗ is equal to p (t) = c L + e −r (t

∗ −t)

( p (t∗ ) − c L ) = c L + e −r (t

∗ −t)

(v L − c L − VB ) < v L − VB .

The low type is indifferent between accepting and rejecting p(t). However, in equilibrium, it must be accepted with probability 1, because otherwise a buyer would have an incentive to slightly increase his offer. Therefore, t∗ = t. Q.E.D.

R ∞ − R s λB(x,c H )dx Proof of Lemma 2: It suffices to show that t e t ds is minimized when B(·, c H ) increases fastest (this corresponds to the case where B(t, c H ) = 0 if t < t and B(t, c H ) = 1

if t ≥ t), while maximized when B(·, c H ) increases slowest (this corresponds to the case where B(t, c H ) is constant). The constraints on B(·, c H ) are p (t ) − c L =

Z

∞ t

e

−

Rs t

(r + λB ( x,c H ))dx



λB(s, c H )ds (c H − c L ) = v L − c L − VB ,

36

and p (t) − c L =

Z

∞

e

t

−

Rs t

(r + λB ( x,c H ))dx



λB(s, c H )ds (c H − c L ) = v L − c L − VB ≥ 0, ∀t > t.

Since the latter constraint binds for all t > t only in the equilibrium with t˜ = t (depicted in the right panel of Figure 1), it can be ignored for local variations. R ∞ − R s λB(x,c H )dx ds Consider a marginal increase of B(t, c H ). Then, the objective function t e t increases at rate

−λ

R

while the constraint 

λ e

−

Rt

Rs

∞ − t e

t

Z ∞

e

t

−

(r + λB ( x,c H ))dx

t (r + λB (t,c H ))dx

−

Z ∞ t

e

−

Rs t

Rs

λB ( x,c H )dx

t

ds,

 λB(s, c H )ds (c H − c L ) increases at rate

(r + λB ( x,c H ))dx



λB(s, c H )ds (c H − c L ).

These marginal changes evolve over time at rates −

−λ and

λe R∞

Rt

e

t

λB ( x,c H )dx Rs − t λB ( x,c H )dx t

ds

= −R ∞ t

e−

Rs t

1 λB ( x,c H )dx

ds

Rt

−λre− t (r+λB(t,c H ))dx (c H − c L )  R  t R ∞ − R s (r+λB(x,c H ))dx − t (r + λB (t,c H ))dx t λ e − t e λB(s, c H )ds (c H − c L )

= −

1−

R∞ t

e−

Rs t

r

(r + λB ( x,c H ))dx

λB(s, c H )ds

,

respectively. The absolute value of the latter is strictly larger than that of the former, because r

> r

Z ∞ t

Z ∞

t Z ∞

e− e

−

Rs t

Rs

Rs

t

λB ( x,c H )dx

ds +

(r + λB ( x,c H ))dx

Z ∞ t

ds +

e−

Rs

Z ∞ t

t

e

(r + λB ( x,c H ))dx

−

Rs t

λB(s, c H )ds

(r + λB ( x,c H ))dx

λB(s, c H )ds

e− t (r+λB(x,c H ))dx (r + λB(s, c H ))ds t Z ∞   Rs d 1 − e− t (r+λB(x,c H ))dx = 1. =

=

t

This means that when B(t, c H ) increases, the decrease of B(t + dt, c H ) necessary to keep 37

the constraint hold is larger than that to keep theR objective function. Therefore, the higher R ∞ − s λB(x,c H )dx ds is. The result then immediB(t, c H ) is than B(t + dt, c H ), the higher t e t ately follows from the fact that B(t, c H ) is highest near t in the equilibrium with t˜ = t, while lowest in the equilibrium with t˜ = t.Q.E.D.

Proof of Proposition 4: As in the proof of Proposition 2, I prove only the necessity here. (1) Since all buyers offer either c H or the reservation price of the low type, the reservation price of the low type who has rejected n(≥ 0) offers is ∞

p (n) ≡ c L +

∑ k= n +1





−1 δk−n Πkl = n +1 (1 − B(l, c H ) B(k, c H )

!

( c H − c L ).

(2) Every buyer’s expected payoff is VB . Let n be the minimal number of previous matches such that the (n + 1)-th buyer obtains more than VB . Then, it must be that (i) p(n) < v L − VB (the buyer obtains more than VB by offering the low type’s reservation price) or (ii) q(n)v L + (1 − q(n))v H − c H > VB (by offering c H ). Suppose (i) is the case. Then the low-type seller must accept the offer for sure: Otherwise, the buyer would slightly increase his offer. Given this, the next buyer ((n + 2)-th buyer) is certain that the seller is the high type32 and, therefore, would offer c H . But then, due to Assumption 2, the low-type seller in her (n + 1)-th match would not accept p(n), which is a contradiction. Suppose (ii) is the case. Due to Assumption 1, certainly n ≥ 1. Suppose trade does not occur at the n-th match of the seller. Then the probability that the seller is the low type does not change between n-th and n + 1-th matches (q(n − 1) = q(n)). This implies that the n-th buyer can obtain strictly more than VB as well, which contradicts the definition of n. Now suppose trade occurs with a positive probability at the n-th match of the seller. If it occurs only at c H , then the same contradiction as before arises. On the other hand, by Assumption 2, trade cannot occur at v L − VB with a positive probability, because the

(n + 1)-th buyer offers c H with probability 1 (notice that (i) is already ruled out). (3) The low-type seller accepts p(n) only in her first match, and p(0) = v L − VB . Suppose p(n) is accepted with a positive probability. (2) implies that p(n) = v L − VB . If n > 0, then p(n − 1) = δp(n) < v L − VB . This cannot be the case because if so, then the 32 Note

that this is true even if the (n + 1)-th buyer randomizes between p(n) and c H .

38

n-th buyer can offer slightly above p(n − 1) and obtain more than VB , which contradicts (2). 1− qb c H −VB (4) The low-type seller accepts p(0) with probability 1 − qb vc HH − − v L +VB . After the first match, buyers’ beliefs about the seller’s intrinsic type become q (1 ) =

qb(1 − S( L, p(0))) . qb(1 − S( L, p(0))) + (1 − qb)

(3) implies that q(n) = q(1) for any n ≥ 1. Since each buyer must be indifferent between offering c H and taking his outside option (for the reason explained in (2)), q(1)v L + (1 − q(1))v H − c H = VB . Solving this equation, S( L, p(0)) = 1 −

1 − qb v H − c H − VB . qb c H − v L + VB

(5) p(n) ≥ v L − VB for any n ≥ 1. If this condition is violated, then the (n + 1)-th buyer strictly prefers offering p(n) to c H and this offer should be accepted by the low-type seller, which contradicts (2) and (3). Q.E.D.

Proof of Proposition 6: Step 1: Given VB , Z

t(VB ) δ qb Φ(VB ) = e−λt ( p(t; VB ) − p(t; VB ))dt 1 − δ G (VB ) 0 Z t(VB ) λ qb ≤ e−λt ( p(t; VB ) − p(0; VB ))dt r G (VB ) 0 qb (1 − e−λt(VB ) )( p(t; VB ) − p(0; VB )). = rG (VB )

(10)

(1) VB approaches 0 as λ tends to infinity.

The proof differs depending on whether v L − c L < v H − c H or not. (i) v L − c L < v H − c H Suppose λ is sufficiently large. I prove that in this case t is close to 0, which immediately implies that VB is close to 0 and VS is close to v L − c L (see Figure 1). Suppose t is

39

bounded away from 0. Then q(t ) is close to zero. Buyers then strictly prefer offering c H to making losing offers after t, because q (t) v L + (1 − q (t)) v H − c H ' v H − c H > v L − c L ≥ v L − p (t) . This violates the structure of the equilibrium. (ii) v L − c L ≥ v H − c H (ii-1) Given VB , Φ (VB ) approaches 0 as λ tends to infinity (pointwise convergence). Fix VB < v H − c H . If λ is large, by the same reasoning as in (i), t must be sufficiently small. Since G (VB ) is bounded away from zero, (10) implies that Φ (VB ) is close to zero. (ii-2) For a fixed λ, Φ (VB ) approaches 0 as VB tends to v H − c H . Together with (ii1), this implies that if λ is sufficiently large, then equilibrium VB is close to 0 (uniform convergence). Fix λ and suppose VB is sufficiently close to v H − c H . For q (t) v L + (1 − q (t)) v H − c H = VB , q (t) must be sufficiently small, and thus t must be sufficiently large. Since G (VB ) > (1 − qb) t, (10) implies that Φ (VB ) is close to 0. (2) VB approaches 0 as r tends to zero. Since

p (t) − c L = e−r(t−t) ( p (t) − c L ) = e−r(t−t) (v L − c L − VB ) , (10) is equivalent to

VB ≤

qb G (VB )



 1 − e−rt(VB )  r

 1 − e−λt(VB ) (v L − c L − VB ) .

(11)

(i) v L − c L < v H − c H Suppose r is sufficiently small. Then t must be bounded from above. Otherwise, q (t) would be close to 0, and then the same contradiction as in (1-i) arises. In (11), the term
1 − e−rt /r is bounded from above (as r tends to zero, the term approaches t). On the other hand, for low-type sellers’ incentive compatibility (see (5) in Proposition 2), B(·, c H ) must be low overall (more concretely, in the equilibrium with t˜ = t, t must be sufficiently large, and in the equilibrium with t˜ = t, α(t˜) must be sufficiently small). This implies that for r sufficiently small, G (VB ) is sufficiently large, and thus VB is close to zero. (ii) v L − c L ≥ v H − c H The proof is essentially identical to the one in (1). First, use the argument in (2-i) to show that given VB , Φ (VB ) approaches 0 as r tends to zero (pointwise convergence). Then, apply (ii-2).

40

Step 2: Recall that in Regime 2, VS = p (0) − c L and VB = v L − p ( t). From the characterization of reduced-form game equilibrium,

Therefore,


p ( t) − p (0) = 1 − e−rt ( p (t) − c L ) .


dp (t) d ( p (t) − p (0)) dt = re−rt ( p ( t) − c L ) + 1 − e−rt . dλ dλ dλ As shown above, for λ sufficiently large, t is close to 0, and thus the second-term does not have a first-order effect. Similarly, when λ is sufficiently large, dVB dq (t) = − (v H − v L ) , dλ dλ and

Using all the results,


2 qbe−λt + (1 − qb) dq ( t) dt =− . dλ dλ λb q (1 − qb) e−λt

d (qbVS + VB ) dV d ( p (0)) d ( p ( t)) = qb − qb + (1 − qb) B dλ dλ dλ dλ d ( p (t) − p (0)) dVB = −qb + (1 − qb) dλ dλ
2 − λt qbe + (1 − qb) dq (t) dq (t) − (1 − qb) (v H − v L ) = qbr ( p (t) − c L ) − λt dλ dλ λb q (1 − qb) e !
2 qbe−λt + (1 − qb) dq ( t) − (1 − qb) (v H − v L ) = qbr ( p (t) − c L ) . − λt dλ λb q (1 − qb) e

This is negative because for λ sufficiently large, the first term is negative, while the second one is positive (as λ tends to infinity, VB approaches zero, and thus q (t) converges to q from the left). The proof for the case where r tends to zero is essentially the same. Q.E.D.

References Akerlof, G. A. (1970), ‘The market for “lemons”: Quality uncertainty and the market mechanism’, Quarterly Journal of Economics 84(3), 488–500. Blouin, M.R. (2003), ‘Equilibrium in a decentralized market with adverse selection’, Economic Theory 22(2), 245-262. 41

Blouin, M.R. and R. Serrano (2001), ‘A decentralized market with common values uncertainty: non-steady states’, Review of Economic Studies 68, 323-346. Camargo, B. and B. Lester (2011), ‘Trading dynamics in decentralized markets with adverse selection,’ mimeo. Chang, B. (2010), ‘Adverse selection and liquidity distortion in decentralized markets,’ mimeo. Chari, V.V., A. Shourideh, and A. Zetlin-Jones (2010), ‘Adverse selection, reputation and sudden collapses in secondary loan markets,’ mimeo. Chiu, J. and T.V. Koeppl (2011), ‘Market freeze and recovery: trading dynamics under optimal intervention by a market-maker-of-last-resort,’ mimeo. Daley, B. and B. Green (2011), ‘Waiting for news in the dynamic market for lemons,’ Econometrica, forthcoming. Deneckere, R. and M.-Y. Liang (2006), ‘Bargaining with interdependent values’, Econometrica 74(5), 1309–1364. Diamond, P.A. (1971), ‘A model of price adjustment’, Journal of Economic Theory 3, 156–168. Evans, R. (1989), ‘Sequential bargaining with correlated values’, Review of Economic Studies 56(4), 499–510. Guerrieri, V. and R. Shimer (2011), ‘Dynamic adverse selection: a theory of illiquidity, fire sales, and flight to quality,’ mimeo. Hendel, I. and A. Lizzeri (1999), ‘Adverse selection in durable goods markets’, American Economic Review 89(5), 1097-1115. Hendel, I. and A. Lizzeri (2002), ‘The role of leasing under adverse selection’, Journal of Political Economy 110(1), 113-143. Hendel, I., A. Lizzeri, and M. Siniscalchi (2005), ‘Efficient sorting in a dynamic adverseselection model’, Review of Economic Studies 72(2), 467-497. Horner, ¨ J. and N. Vieille (2009a), ‘Public vs. private offers in the market for lemons’, Econometrica 77(1), 29-69.

42

Horner, ¨ J. and N. Vieille (2009b), ‘Supplement to ’Public vs. private offers in the market for lemons’ ’, Econometrica Supplement Material 77, http://www.econometricsociety.org/ecta/Supmat/6917 extensions.pdf. Inderst, R. and H.M. Muller ¨ (2002), ‘Competitive search markets for durable goods’, Economic Theory 19(3), 599–622. Janssen, M.C.W. and V.A. Karamychev (2002), ‘Cycles and multiple equilibria in the market for durable lemons’, Economic Theory 20(3), 579-601. Janssen, M.C.W. and S. Roy (2002), ‘Dynamic trading in a durable good market with asymmetric information’, International Economic Review 43(1), 257-282. Janssen, M.C.W. and S. Roy (2004), ‘On durable goods markets with entry and adverse selection’, Canadian Journal of Economics 37(3), 552-589. Kohlberg, E. and J.F. Mertens (1986), ‘On the strategic stability of equilibria’, Econometrica 54, 1003-1039. Kurlat, P. (2010), ‘Lemons, market shutdowns and learning’, mimeo. J. Levin (2001), ‘Information and the market for lemons’, RAND Journal of Economics 32(4), 657–666. Moreno, D. and J. Wooders (2010), ‘Decentralized trade mitigates the lemons problem’, International Economic Review 51(2), 383–399. Noldeke, ¨ G. and E. Van Damme (1990), ‘Signalling in a dynamic labor market’, Review of Economic Studies 57, 1-23. Rubinstein, A. and A. Wolinsky (1990), ‘Decentralized trading, strategic behaviour and the Walrasian outcome’, Review of Economic Studies 57, 63-78. Serrano, R. and O. Yosha (1993), ‘Information revelation in a market with pairwise meetings’, Economic Theory 3, 481-499. Serrano, R. and O. Yosha (1996), ‘Welfare analysis of a market with pairwise meetings and asymmetric information’, Economic Theory 8, 167-175. Spence, A.M. (1973), ‘Job market signaling’, Quarterly Journal of Economics 90, 225-243. Spence, A.M. (1974), ‘Market signaling, information transfer in hiring and related processes’, Cambridge: Harvard University Press. 43

Swinkels, J.M. (1999), ‘Education signalling with preemptive offers’, Review of Economic Studies 66, 949-970. Taylor, C. R. (1999), ‘Time-on-the-market as a sign of quality’, Review of Economic Studies 66, 555-578. Vincent, D. R. (1989), ‘Bargaining with common values’, Journal of Economic Theory 48(1), 47–62. Wolinsky, A. (1990), ‘Information revelation in a market with pairwise meetings’, Econometrica 77(1), 29-69.

44