Staatswissenschaftliche Fakultät. Faculty of Economics, Law and Social Sciences. Discussion Paper No.: E

Transcription

1 Staatswissenschaftliche Fakultät Faculty of Economics, Law and Social Sciences Discussion Paper No.: E Are Non-Binding Contracts Really not Worth the Paper? Bernd Irlenbusch Für den Inhalt des Diskussionspapiers sind die jeweiligen Autoren/innen allein verantwortlich. ISSN (Print) ISSN X (Internet)

2 Are Non-Binding Contracts Really not Worth the Paper? BERND IRLENBUSCH Abstract This paper experimentally investigates behavior in sequential one-shot transactions which are governed by non-binding contracts. In a second, incomplete information treatment, contracts are binding for some players. While according to traditional game-theoretical analysis no trade is expected in the first treatment, full trade should result in the latter. However, we find that trade is even higher in the non-binding contract treatment. On the one hand, non-binding contracts although they are cheap talk do guide behavior, especially at the beginning of a business relationship, while reciprocal reactions prevail later on. On the other hand, in the treatment with binding contracts cooperative behavior appears to be crowded out. Keywords Incomplete Contracts, Cheap Talk, Promises, Trust, Reciprocity, Experiments JEL Classification Codes C72, C91, K12 Acknowledgements Many thanks to three anonymous referees and to Klaus Abbink, Heike Hennig- Schmidt, Rosemarie Nagel, Elke Renner, Bettina Rockenbach, Abdolkarim Sadrieh, Lars Schade, Patrick Schmitz and Reinhard Selten for valuable comments. Suggestions from Ron Harstad helped to give the paper a more concise shape. Financial support from the Deutsche Forschungsgemeinschaft through Sonderforschungsbereich 303 at the University of Bonn and grant KR2077/2-1, the Ministerium für Wissenschaft und Forschung des Landes Nordrhein-Westfalen and the European Union through the EU- TMR Research Network ENDEAR (FMRX-CT ) is gratefully acknowledged. Address University of Erfurt Department of Economics Nordhäuser Strasse 63 D Erfurt, Germany phone +49 (0) fax +49 (0) bernd.irlenbusch@uni-erfurt.de

3 1 Introduction "Non-binding contracts are not worth the paper they are written on!" At first glance this advice appears to be quite plausible. However, in daily life it has often been observed that business partners refer to non-binding agreements rather than to legal sanctions in order to adjust their relationships or to settle disputes. Legal sanctions are often unnecessary and may even have undesirable consequences. This is the case not only because of their costs but rather because reference to legal norms might be interpreted as a betrayal of trust or even a form of unkind action. It is argued that insisting upon a strict legal analysis of the relation or the dispute is likely to close down any further reference to norms of goodwill or cooperation. In a remarkable study MACAULAY (1963), for example, finds overwhelming support for these observations in interviews he conducted with businessmen and lawyers on noncontractual relations in everyday business life. The fact that parties often abstain from using legal sanctions is closely related to the observation that most contracts are vague or silent on a number of key features, i.e., they are incomplete. If the enforcement of contracts by legal means is often considered not to be a valid option, it might be too costly to be precise and complete on every contingency that might occur. TIROLE (1999) emphasizes the importance of the understanding of incomplete contracting for a wide range of economic issues and argues that the topic is left largely unexplored and poorly understood. Important questions are: When and why are incomplete contracts a suitable choice? Are there circumstances under which complete contracts might even be outperformed by incomplete contracts? This paper introduces an experiment which is designed to gain insights into the effectiveness of contracts that are incomplete and which are therefore to a large degree non-binding. One of the most important questions in this study is whether non-binding contracts affect the behavior of the contracting parties. Emphasis is put on efficiency aspects, especially with respect to promise-keeping, as well as trust and the social norm of reciprocity. A key research issue of this paper is the influence of the use of legal institutions on the behavior of contracting parties. Do legal sanctions undermine the effectiveness of social norms which are likely to guide behavior especially in the absence of legal enforcement measures. The influence of non-binding contracts on behavior is contrasted with behavior induced by partially binding contracts. The experiment follows a model introduced by HART and HOLMSTRÖM (1987). In a simple buyer/seller relation units of a fictitious commodity and money are exchanged in a sequential manner, i.e., payments and deliveries are performed in an alternating order. First the seller proposes a contract to the buyer by specifying how much money he desires in return for a fixed number of units. In our first treatment the contract is non-binding for both parties, i.e., the buyer does not have to make the specified payments, nor is the seller forced to deliver the units. Since in reality there is very often uncertainty as to whether the parties can be forced to adhere to the contract, we performed a second, incomplete information treatment in which the contract is binding for some buyers. There are at least two main reasons why a contract is non-binding: (i) parties cannot enforce the obligations that are written in the contract or (ii) business partners do not really want to go to court, for example, because they do not want to break a social norm or damage their reputation. In a game-theoretical sense our non-binding contract is mere cheap talk. The contract is a non-binding, costless, non-verifiable message that may affect the listeners beliefs but not necessarily players payoffs. On 1

4 the one hand, cheap talk can be valuable for signaling players private information 1. On the other hand, cheap talk is found to be effective in signaling players intentions regarding future decisions. In our setting the non-binding contracts may be regarded as promises. Although, in general, promises are not seen as enforceable, FRIED (1981) interprets legal contracts as mutual promises. SCHELLING (1989) argues that promises are often used instead of legal provisions because the agreement cannot be enforced anyhow 2. The experiment we analyze in this paper spans two lines of experimental research on incomplete contracts. The first line consists of experimental studies that have demonstrated that trade is possible even under non-binding agreements. This finding is to a large degree attributed to social preferences exhibited by a considerable number of contractors. Social preferences incline persons not only to care about their own material well-being but also about the material well-being of members of a reference group. The social norm of reciprocity induces social preferences, which are conditioned on the behavior and on the intentions of the reference persons. 3 For example, if a buyer perceives an action of the seller as friendly, he values the seller s payoff positively. Analogously, the seller s payoff is valued negatively by the buyer if the seller s action is regarded to be unfriendly. The importance of the norm of reciprocity for the behavior of economic actors is confirmed in numerous studies. In their seminal experimental work, FEHR, KIRCHSTEIGER, and RIEDL (1993) find that "firms" who pay more than they have to are reciprocally rewarded with voluntarily higher effort contributions by their "workers." BERG, DICKHAUT, and MCCABE (1995) introduce the two-player, two-stage investment game. During the first stage the investor can invest an arbitrary fraction of a given amount. The tripled investment is given to the trustee, who in turn decides whether to send some of this money back to the investor. A view of man as a rational money-maximizer predicts that the trustee has no incentive to return anything. This will be anticipated by the investor, which implies that neither the investor nor the trustee will send anything. But in fact a considerable number of subjects invest a positive amount some even everything. In return most of the trustees reward trust by sending money back. Only few trustees who received a positive amount return nothing. It is hypothesized that this result arises from the expectation among some investors that the trustees will behave reciprocally 4. FEHR, 1 For overviews see FARRELL and RABIN (1996), CRAWFORD (1998), BROSIG, OCKENFELS, and WEIMANN (2003). PALFREY and ROSENTHAL (1991) report on a three-person public goods experiment with simultaneous, non-binding announcements. Subjects had considerable success in efficiently eliciting the contributions required for provision of the public good from the subjects whose privately known costs were lowest. Interestingly enough, more rounds of announcements performed even better. COOPER, DEJONG, FORSYTHE, and ROSS (1989), for example, study the effect of explicit communication in the Battle of the Sexes game. When both subjects make announcements they perform significantly better than they do without communication. 2 SCHELLING (1989) characterizes a promise as follows: (i) what is promised must appear to the second party, the one to whom the promise is made, as being in his interest; (ii) it should be something that one would not ordinarily be expected to do without the promise; (iii) it should be something that the second party perceives to be within the promisor s control; 3 For the influence of the reciprocity norm on behavior and its modeling see RABIN (1993), DUFWENBERG and KIRCHSTEIGER (1999), and FALK and FISCHBACHER (2000). Two important models of other-regarding preferences that deal with inequity aversion are FEHR and SCHMIDT (1999) and BOLTON OCKENFELS (2000). 4 Evidence for the importance of the reciprocity norm is reported in several other studies, for example, GÜTH, OCKENFELS, and WENDEL (1994), JACOBSEN and SADRIEH (1996), GÜTH, KLOSE, KÖNIGSTEIN, and 2

5 GÄCHTER, and KIRCHSTEIGER (1997) test for the reciprocity norm in a market setting with more sellers than buyers. They find that the sellers deliver higher quality if buyers offer them higher payments beforehand. In contrast to these observations DUFWENBERG and GNEEZY (2000) do not find a correlation between the amount given to the trustee and the amount returned to the trustor. Instead, their study shows that the amount which is sent back by the trustee correlates with his belief about the trustor s expectations of what will be returned. FEHR and SCHMIDT (2000) and FEHR, KLEIN, and SCHMIDT (2002) find that fairness considerations have a significant impact on the optimal provision of incentives in a moral hazard context. Incentive contracts that are optimal in the presence of merely selfish actors become inferior to incomplete contracts when some agents are concerned about fairness. The second body of literature relevant to our study originates from the field of social psychology (e.g., DECI 1975, DECI and RYAN 1985) and argues that explicit incentives, for example, monetary incentives or threats of punishment, might have substantial countervailing effects on motivation. The importance of the hidden costs of incentives has recently also been investigated by an increasing number of economists (FREY 1997, GNEEZY and RUSTICHINI 2000, FEHR and GÄCHTER 2002, BOHNET, FREY, and HUCK (2001), FEHR and ROCKENBACH 2003, IRLENBUSCH and SLIWKA 2003). FREY (1993) and SLIWKA (2002) offer models that predict the crowding out of voluntary cooperative behavior by incentive contracts. The paper proceeds as follows. The next section describes the model which serves as the basis for the experiment. Section three deals with some game-theoretical aspects of the model presented. The experimental design is reported in section four; section five outlines the experimental results and section six concludes the paper. 2 A Simple Model of Non-Binding Contracts The model used in the experiment described below is introduced in HART and HOLMSTRÖM (1987). Consider a buyer B and a seller S who wish to trade two units of an item which has the value v for the buyer. Also delivery causes costs c for the seller. If the item is not delivered it causes no costs and is of no value either. Assume that v > c > 0, i.e., it is profitable to trade the item. Both the buyer and the seller know the value v and the costs c. One single buyer is matched with one single seller. We assume that the trade is performed in a sequence, i.e., delivery and payment are made sequentially over five stages 5. The buyer has to make a first payment, then the seller has to decide whether she wants to deliver the first unit of the item. Thereafter, again, it is the turn of the buyer to make a second payment, followed by a second delivery decision by the seller and a third payment by the buyer, i.e., the number of payments exceeds the number of deliveries by one. The i th payment from the buyer to the seller is denoted by r i, r i 0, i = 1, 2, 3. The j th delivery from the seller to the buyer is denoted by d j, d j {0,1}, j = 1, 2. If the seller delivers the j th unit we have d j = 1 otherwise d j = 0. The sequence of trade can be depicted as follows: SCHWALBACH (1998), ABBINK, IRLENBUSCH, and RENNER (2000), CLARK and SEFTON (2001), VAN DER HEIJDEN, NELISSEN, POTTERS, and VERBON (2001), GÄCHTER and FALK (2002), CHARNESS forthcoming. 5 Similar sequential experimental games are also analyzed, for example, in SELTEN and STOECKER (1986), CAMERER and WEIGELT (1988), OCHS and ROTH (1989), MCKELVEY and PALFREY (1992), and NERAL and OCHS (1992). 3

6 Trading phase: B S B S B p 1 d 1 p 2 d 2 p 3 Before trading the seller offers the buyer a contract. The contract specifies the payments which have to be made in order for units to be delivered. Hence, the contract consists of individual payments. We denote the proposed payments as p i, p i 0, i = 1, 2, 3 and restrict the values for the i th payment r i as well as for the i th proposed payment p i to non-negative integers out of an interval of [0,..., M]. We are assuming a world of voluntary contracts, i.e., nobody is forced to trade. Therefore, we model the proposed contract as a take-it-or-leave-it-offer 6, i.e., the buyer is asked whether he accepts the contract or not. We denote this decision by a. If the contract is accepted (a = 1) the trade is performed successively, otherwise (a = 0) no trade occurs. If there is no trade, both the seller and the buyer receive a payoff of zero. The contract is binding for some players but not binding for others. A seller is never bound to a contract. At each delivery stage she can freely decide whether or not to deliver. We are assuming there are two types of buyers. Whereas a restricted buyer has to abide by a contract, an unrestricted buyer does not have to do so 7. An unrestricted buyer can simply refuse to pay or can deviate from the payment agreed upon in the contract. Abiding by a contract means that a buyer makes a payment at least as large as the payment specified in the contract, i.e., p i r i, i = 1, 2, 3. However, even a restricted buyer does not have to adhere to a contract if his trading partner has already previously broken the contract. We are using a model of incomplete information, i.e., a buyer knows his own type, but the corresponding seller does not. However, it is common knowledge that the fraction of restricted buyers in the population is θ (0 = θ = 1). 3 Game-Theoretical Analysis This section deals with some game-theoretical considerations of the described model 8. We assume that agents are rational and aim to maximize their expected payoff. 3.1 Unrestricted Buyers For the moment let us concentrate on the case θ = 0, i.e., it is common knowledge that all players are unrestricted. By a simple backward induction argument one can see 6 We aimed to simplify the bargaining process. For a more detailed discussion of take-it-or-leave-it-offers in bargaining contexts, see the ultimatum game literature, for example, GÜTH, SCHMITTBERGER, and SCHWARZE (1982), CAMERER and THALER (1995). 7 We distinguish between restricted and unrestricted buyers to capture the inherent uncertainty of incomplete contracts whether they are exogenously enforceable or not. HART and HOLMSTRÖM (1987) call the two types of buyers to be honest or dishonest. Recent studies on social preferences provide a justification for this naming. They find that there is considerable heterogeneity among humans with respect to selfish behavior, for an overview see FEHR and FISCHBACHER (2002). Note that traders who are non-selfish might additionally be restricted, for example by guilt or reputation. 8 HART and HOLMSTRÖM (1987) provide a brief equilibrium analysis. For details see IRLENBUSCH (1999). 4

7 that for every accepted contract there is a unique way of behaving in a subgame perfect equilibrium. The buyer does not pay anything at the final stage. This is anticipated by the seller and she does not deliver the final unit. Again, this is anticipated by the buyer and he will not make the last but one payment. This argument can be extended back to the beginning of the game. Hence, in a subgame perfect equilibrium no trade occurs. This is true independently of the proposed contract. Regardless of whether the seller delivers or not, in a subgame perfect equilibrium the buyer does not pay anything. Regardless of whether the buyer pays anything or not in a subgame perfect equilibrium, the seller does not deliver. Note, however, that given this behavior, the buyer is indifferent between accepting and rejecting the contract. To sum up, if it is common knowledge that the buyer is unrestricted in a subgame perfect equilibrium the buyer does not pay anything nor does the seller deliver any unit. Note that this is true irrespective of the values for v and c. 3.2 Restricted and Unrestricted Buyers Let us now consider the case 0 < θ < 1, which means that the buyer is restricted with a strictly positive probability. Figure 1 sketches the corresponding game tree, with truncated branches 9. The root of the game tree symbolizes the chance move with a known probability of θ which determines whether the buyer is restricted or not. The branches shown in Figure 1 essentially represent sequences of actions in the event that the seller S cannot distinguish between a restricted and an unrestricted buyer B. The fact that the seller does not know whether her trading partner is restricted or unrestricted is indicated by the three information sets with two decision nodes which belong to the seller 10. The first of these information sets represents the contract proposal. The following two denote the delivery decisions. The truncated branches have to be labeled with payoff-pairs, which result if the agents play the subgame perfect equilibrium strategies in the subsequent subgames. For example, if the buyer pays less than agreed on in the contract he reveals himself to be unrestricted and thus by a backward induction argument one can see that in the following subgames no deliveries and payments will be performed. If the seller does not deliver even a restricted buyer is not bound to the contract anymore and, thus, it is dominant not to trade afterwards. In the following we give an idea of the analysis of the setting used in the experiment. Since we now have a game of imperfect information we are focussing on the concept of sequential equilibrium (KREPS and WILSON 1982) when referring to an equilibrium. For simplicity we will concentrate on those equilibria in which on the equilibrium path the restricted buyer does not pay more than the amount prescribed in the contract. 11 DEFINITION: Let β be a sequential equilibrium in pure behavioral strategies and let p be the contract which is proposed on the equilibrium path of β. We call β a proposal- 9 At first glance the game tree looks rather complicated. However, as described below, in the experiment the game is not introduced to the subjects by showing them the game tree, but by explaining the sequence of actions. 10 In fact, these information sets are the only ones in which the seller cannot distinguish between a restricted and an unrestricted buyer if we concentrate on those strategies where buyers do not pay more than is prescribed by the contract (see below for the definition of a proposal-confirming equilibrium). 11 The omitted equilibria essentially result in equal efficiency and payoffs for both players. 5

8 confirming equilibrium if on the equilibrium path the restricted buyer does not pay more than what is proposed in p. B is restricted θ B is unrestricted (1 -θ) reject B (a = 0) B S (p 1, p 2, p 3 ) (p 1, p 2, p 3 ) B accept (a = 1) accept (a = 1) B r 1 = p r 1 1 = p 1 reject (a = 0) r 1 = 0 d 1 = 0 B d 1 = 1 r 2 = p 2 S d 1 = 1 r 2 = p 2 d 1 = 0 B r 2 = 0 d 2 = 0 B d 2 = 1 S d 2 = 1 d 2 = 0 B r 3 = 0 r 3 = p 3 r 3 = p 3 Seller s payoff: Π S = a (r 1 + r 2 + r 3 (d 1 + d 2 )c) Buyer s payoff: Π B = a ((d 1 + d 2 )v r 1 r 2 r 3 ) FIGURE 1 GAME TREE FOR n=3 WITH TRUNCATED BRANCHES AFTER NO PAYMENT AND NO DELIVERY. Let us first assume that there exist contracts which in a proposal-confirming equilibrium result in a strictly positive payoff for the seller. In fact - as we will see below - this is the case for the parameter values chosen in the experiment. It is immediately evident that if a contract is acceptable for a restricted buyer, then an unrestricted buyer will also accept the contract, since in any case he is able to imitate the restricted buyer by choosing the same actions. Thus, it is not possible for the seller to initially discriminate between a restricted and an unrestricted buyer by making different contract proposals. After the buyer has accepted the contract either both players follow their agreement or one of the players is the first to break the contract. An initial contract violation can only occur either if the seller does not deliver or an unrestricted buyer makes a payment which is lower than the payment agreed upon in the contract. Note, however, that after a contract is broken even a previously restricted buyer is no longer bound to the contract. This means that after an initial contract violation, it is common knowledge that both players are unrestricted afterwards. Knowing this, a simple backward induction argument reveals that in equilibrium trade becomes impossible after a contract violation (the argument is analogous to the one in section 3.1). Thus, in equilibrium, trade (i.e., deliveries and payments according to the contract) is only possible in an uninterrupted sequence from the outset. Once trade has been broken off, there is no way to take it up again. 6

9 Furthermore, if an unrestricted buyer breaks a contract, paying nothing dominates making a positive payment. From the rules of the game it follows that only the seller and an unrestricted buyer can break a contract first. Thus, in equilibrium an accepted contract may induce the seller or the unrestricted buyer to stop trading at different points in time. These two points in time essentially characterize the exchange phase which follows an accepted contract. If the contract induces the seller not to deliver the first unit, no buyer would be willing to make a positive first payment because trade would stop thereafter. Thus, the contract proposed in equilibrium should induce the seller either to deliver both units or to deliver the first but not the second unit. If it is profitable for a restricted buyer to make the first payment, it is also profitable for an unrestricted buyer. However, for an unrestricted buyer it is always dominant not to make the last payment. Thus, it is not immediately clear whether the contract proposed in equilibrium induces the unrestricted buyer to make the second payment or not. This leaves us with three possible types of contract which each induce different behavior in the trading phase. In the case of the first type of contract in equilibrium all players abide by the contract except that the unrestricted buyer does not make the last payment. In the case of the second type of contract, in equilibrium the contract is completely fulfilled if the buyer is restricted. But if the buyer is unrestricted, he has an incentive to break the contract after the first delivery. The third type of contract gives the seller incentives to deliver only the first unit and, thus, the second payment is only made by a restricted buyer. These three combinations of behavior are the only ones which in equilibrium may follow a contract that is accepted and that guarantees a strictly positive payoff for the seller. In our second experimental treatment we use the following parameters: c = 5, v = 20, and θ =.5. Given this parameter constellation one can show that every contract of the first type results in a higher seller payoff than a contract of the second or third type. Thus, in a proposal-confirming equilibrium only contracts of the first type are proposed 12. To summarize, these equilibrium contracts provide the seller with incentives to deliver both units. Therefore, a restricted buyers is forced to pay exactly what is written in the contract. This is true for all three installments. An unrestricted buyer imitates a restricted buyer in the beginning of trade, i.e. he pays the first two installments as they are agreed upon in the contract. However, in equilibrium an unrestricted buyer does not pay anything after he has received the second unit. To simplify experimentation we restrict the payments which are proposed in the contract as well as those amounts which are actually paid to integers. From the constraints described in the Appendix one can derive all contracts which might be proposed in a proposal-confirming equilibrium. These are shown in Table 1. In every contract almost the total surplus from trading is transferred to the seller, i.e., the sum of the three payments essentially corresponds to the buyer s value of the two units. This result is due to the take-it-or-leave-it structure of the present bargaining situation. In equilibrium buyers of both types pay the first two installments as they are specified in the contract. Interestingly, within a certain range the amount of payment can be shifted between the first two installments. In an equilibrium contract a third payment is proposed which is high enough to provide the seller in expectation with an 12 The proof is given in the Appendix. 7

10 incentive to deliver the second item, i.e., the expected third payment has to cover the cost of delivery 13. Note that an unrestricted buyer does not pay the last installment. Therefore, the third payment should be as small as possible, given the abovementioned constraint. The variation in the sellers expected payoffs from various equilibrium contracts results from different actions on the equilibrium path in case of indifference. TABLE 1 PROPOSAL-CONFIRMING EQUILIBRIUM CONTRACTS FOR θ =.5 Expected seller s payoff = p 1 {8,..., 23} p 1 {7,..., 23} p 1 {9,..., 24} p 1 {8,..., 24} p 1 {9,..., 24} p 1 {10,...,25} p 2 = 28 - p 1 p 3 = 11 p 2 = 27 - p 1 p 3 = 13 p 2 = 29 - p 1 p 3 = 10 p 2 = 28 - p 1 p 3 = 12 p 2 = 29 - p 1 p 3 = 11 p 2 = 30 - p 1 p 3 = 10 The following section describes the experimental setup. 4 Experimental Design and Experimental Procedure Two different treatments were conducted. In the first treatment U the contract is completely non-binding (θ = 0), i.e., all participants are unrestricted. As we have seen in the third section, in equilibrium there is no trade at all, if the subjects are rational and merely money maximizing. In the second treatment RU a fraction of buyers is bound to the contract. As a first step towards our investigation into legal binds, we chose to implement a setting in which half of the buyers are restricted, i.e., θ =.5. From a traditional game-theoretical point of view this modification dramatically changes the usefulness of contracts, since both units are delivered in a proposalconfirming equilibrium as shown above. The set of business relations with restricted buyers in treatment RU is indicated by the symbol RU R. Analogously, the set of business relations with unrestricted buyers in treatment RU is denoted by RU U. The experiment was conducted at the Laboratorium für experimentelle Wirtschaftsforschung at the University of Bonn. The subjects were recruited by advertising the experiment on posters. Most of them were students from various disciplines, although the majority were law and economics students. In total 120 students participated in the experiment 60 students per treatment. It is important to note that one person was allowed to participate only once, i.e., in only one business relationship 14. The treatments were performed as a Mensa Experiment, i.e., due to the short decisionmaking period in the one-shot setting the lab was taken to a convenient location for the participants: Two teams of laboratory staff members were placed in separate university buildings - one in the canteen (called Mensa) and one in a lecture building. The distance between the buildings and the large number of people milling around 13 This characteristic of the equilibrium contracts is in line with the notion of relational contracts which is used to describe (informal) agreements sustained by the value of future transactions within the same relationship (MACAULAY 1963, MACNEIL 1978, 1980, GRANOVETTER 1985, BERNSTEIN 1992, GREIF 1993, BAKER, GIBBONS, MURPHY 2002). These contracts are sometimes also called self-enforcing or implicit contracts (TELSER 1981, BULL 1987, MACLEOD and MALCOMSON 1989). 14 The influence of non-binding contracts on behavior in a repeated setting is investigated in IRLENBUSCH, forthcoming. 8

11 increased anonymity. The two teams communicated via telephone. 15 Instruction sheets were handed out to those who registered to participate. 16 The instructions framed the decision situation as a trading situation in which the seller could propose a contract to the buyer. The buyer could reject or accept the contract and could make payments. The seller could make deliveries. Participants were given enough time to read the instructions carefully. The experimenters answered any remaining questions. A business relation was established by matching two participants from different buildings - one participant in the canteen and one in the lecture building. The matching was done randomly and anonymously and was fixed for the whole business relationship. Participants took a seat in a cubicle. First of all each participant had to throw a die. The numbers thrown by two matched participants determined their roles. Both subjects were equally likely to become the seller or the buyer. After the roles were determined in treatment RU, the buyer had to throw the die again in order to define his type. If the die showed an even number the buyer was restricted, otherwise he was unrestricted. Note that the type of buyer was not revealed to the seller. In treatment U it was not necessary to determine the buyer s type because it was common knowledge that all buyers were unrestricted. Sellers as well as buyers received an initial endowment of 20 units of the fictitious currency unit, called "talers." The endowment was common knowledge. After the roles and the buyer s type were determined, the seller had to offer a contract by simultaneously specifying p 1, p 2, and p 3 on a decision sheet. The contract offered (as well as all following decisions) was transmitted via telephone by one of our team members to the other site. On the other site the experimenter who was in charge of the telephone filled in a copy of the decision sheet. Afterwards, he passed it over to another experimenter, who gave it to the matched participant. The buyer had to decide whether he accepted the contract or not. If the buyer rejected the contract, the business relation was over. Both participants buyer and seller were paid their endowment. If the buyer accepted the contract he immediately had to make his first payment r 1. Now it was the seller s turn to decide whether she wanted to deliver the first unit. If the unit was delivered, the seller's capital balance was reduced by c = 5 talers and the buyer's capital balance increased by v = 20 talers. In the following the decisions concerning r 2, d 2, and r 3 took place analogously. Note that in treatment RU it was the duty of our teams to check that restricted buyers did not break a contract first. In the end buyer and seller were paid their final capital balances. All talers were changed into DM at an exchange rate of 1 DM ( 0.51 ) for 5 talers. On average a business relation lasted for approximately 15 minutes. In both treatments we had 30 business relationships each, i.e., we collected 30 statistically independent observations in each treatment. In the next section we provide the results obtained from the experiment. 15 Participants were informed that there was another site on the campus at which the trading partners were recruited, i.e., in principle participants from the Mensa could go to the other site in the lecture building to verify that there indeed was another recruitment site and vice versa. Although participants could not listen to the telephone calls, they could see the experimenter who was handling their decision sheet and the telephone on their site. 16 Instruction sheets and the decision sheet can be found in the Appendix. 9

12 5 Experimental Results A first impression of the experimental results can be obtained from a brief look at the data collected in both treatments, which is provided in the Appendix 17. We start our analysis with a short discussion of rejections and payments. Afterwards, we pose one of the central questions of the present study, namely whether the efficiencies obtained in the two treatments reflects the results from the traditional game-theoretical prediction. We investigate the influence of non-binding contracts and outline the observed evidence for reciprocal behavior. In order to look for potential crowding out effects induced by binding contracts we conclude our analysis by investigating behavioral differences between both treatments. All of our statistical tests preserve the independence of observations. If one compares the installments proposed in the contracts to the installments that are actually paid, one obtains the following intuitive result: One can reject the null hypothesis that the sum of the installments proposed in the contract is as high as the sum of the actual payments in favor of the alternative hypothesis that actual payments are lower (Wilcoxon signed-rank test, one-tailed, p = for both treatments). Whereas in treatment U two contract proposals are rejected we find seven rejections in treatment RU five by restricted buyers and two by unrestricted ones. From a traditional game-theoretical point of view it is difficult to justify the four contract rejections by unrestricted buyers because a rejection is weakly dominated. A reason for rejection could be that the contract proposal is somehow interpreted by the buyer as a signal that the seller is not trustworthy. If one compares those contracts in both treatments which are accepted by an unrestricted buyer with the four rejected ones, one recognizes that the third installment is significantly lower in the rejected contract proposals (p = 0.016, Mann-Whitney U test, one-tailed). Thus, in those contracts the seller gives herself very low incentives to deliver the second unit because the contract promises her only a small third installment. The buyer may interpret this to mean that the seller does not really want to deliver the second unit and therefore it may be taken as a reason for distrust and rejection. Similarly, if one compares the four contracts which are rejected by a restricted buyer with those that are accepted, we see that the sums of all three payments are significantly higher in the rejected ones (p = 0.001, Mann-Whitney U test, one-tailed). Delivery of one unit creates a surplus of (v c) = 15 talers. The surplus can be divided between seller and buyer by payments, for example, if both units are delivered, a total payment of 25 talers would equally divide the surplus. In fact this division is proposed by five contracts in treatment U and by three contracts in treatment RU. However, most of the contracts proposed by the seller promise a lower payoff to the buyer. In fact one can reject the hypothesis that the seller s and buyer s payoffs resulting from contracts are equal in favor of the alternative hypothesis that 17 In the following we exclude pair 21 of treatment U and pair 16 of treatment RU from the analysis of the contracts. The proposed contract (50, 80, 120) in treatment U is the only one which cannot be fulfilled. It is not clear why the seller proposes a contract with such high installments. Maybe by this she wants to express her right to claim a high profit. Another explanation would be that she does not care what she proposes since the contract is non-binding anyway. The proposed contract (25, 5, 5) in treatment RU is not feasible because the initial endowment of the (restricted) buyer is too low to make the first payment. Thus, the contract has to be rejected. Figure A.1 and Figure A.2 in the Appendix show the decisions taken by the participants in treatment U and treatment RU respectively. For an additional investigation of treatment U see IRLENBUSCH and SCHADE (1999). 10

13 seller s payoffs are higher (p = for both treatments, Wilcoxon signed-rank test, one-tailed). Note that there are no apparent differences between treatments in payoffs intended by the contracts Buyer s payoff [taler] d 1 + d 2 = 2 d 1 + d 2 = 1 d 1 + d 2 = 0 [taler] Treatment U Treatment RU restricted buyers Treatment RU unrestricted buyers Seller s payoff FIGURE 2 SELLER/BUYER ACTUAL PAYOFFS Figure 2 shows the payoffs achieved in those business relations in which the contract is accepted 18. The three lines represent the different possible efficiency levels, i.e., no unit, one unit, or two units are delivered. The dashed line is the equal payoff line. Interestingly, in treatment RU the payoff of unrestricted buyers is positive only in two business relationships, while in treatment U buyers normally earn positive amounts. Observation 1: In both treatments sellers tend to earn more than buyers. (p = for both treatments, Wilcoxon signed-rank test, one-tailed). The most frequent observations in treatment U are four combinations of 15/15 taler payoffs for buyers and sellers. In treatment RU there is no equal payoff division. Interestingly, the payoff differences (seller s payoff buyer s payoff) in those business relations in which both units are delivered are lower than those of the others in the same treatment (p = for treatment U, p = for treatment RU, Mann-Whitney U test, one-tailed). 5.1 Do Non-Binding Contracts Prevent Trade? In the third section we saw that from a traditional game-theoretical point of view trade is impossible in treatment U because the contract is mere cheap talk. The situation is different in treatment RU. Here in a proposal-confirming equilibrium an unrestricted buyer hides his type by behaving exactly like a restricted buyer - at least except for the last payment. Thus, whereas in the first treatment in equilibrium no delivery is obtainable, delivery of both units should be observed in the second treatment. Since surplus is created only by deliveries, we define efficiency as the fraction of actually delivered units in relation to potentially delivered ones, which means that in a single 18 A more detailed payoff comparison between treatments can be found in the Appendix (Table A.4). 11

14 business relation efficiencies of.0,.5, and 1 can be achieved depending on how many units are delivered: zero, one, or two. As one can see from Table 2 on average the efficiency obtained in treatment U is even higher than in treatment RU. In fact one can weakly reject the null hypothesis that efficiencies obtained in treatment U and treatment RU are equal in favor of the alternative hypothesis that efficiencies in treatment U are higher (p = 0.091, Mann-Whitney U test, one-tailed): Observation 2: Efficiency tends to be higher in treatment U than in treatment RU. The difference in efficiency between the treatments is even more pronounced if one concentrates on those business relations of treatment RU in which buyers are unrestricted, i.e., if one compares the 30 business relations of U with the 14 business relations of RU U. One can clearly reject the null hypothesis that efficiencies are equal in the setting U and the setting RU U in favor of the alternative hypothesis that efficiencies in treatment U are higher (p = 0.012, Mann-Whitney U test, one-tailed). TABLE 2 EFFICIENCIES OBSERVED IN THE TWO TREATMENTS Efficiency Average (a, d 1, d 2 ) = (0, -, -) (1, 0, 0) (1, 1, 0) (1, 0, 1) (1, 1, 1) # in U # in RU # in RU U RU U denotes those business relations of treatment RU in which the buyer is unrestricted. The acceptance decision is denoted by a = 1 and a rejection is denoted by a = 0. If the ith unit is delivered, this is denoted by d i = 1 and if delivery of the ith unit is refused, this is denoted by d i = 0 (i = 1, 2). Since the experimental findings described in this section cast serious doubts on the predictive power of the theoretical analysis given above, in the following we provide a deeper analysis of the observed behavior. 5.2 Do Non-Binding Contracts Influence Actual Behavior? In treatment U relatively high efficiency is achieved. Although from a game-theoretical point of view the non-binding contracts are cheap talk, this section analyses whether, nevertheless, the contracts have some influence on actual behavior. If this is the case, the availability of non-binding contracts might be one explanation for the unexpected high number of deliveries. This section and section 5.3 therefore concentrate on treatment U. A first hint that non-binding contracts do systematically influence behavior is evident from the comparison of total payments proposed in the contract and those amounts actually paid: Observation 3: In treatment U total proposed and actual payments are correlated. (p = 0.035, Spearman's rank correlation, one-tailed). Interestingly, an analogous correlation cannot be found in the other treatment, neither for RU nor for RU U 19. Especially the fact that we do not observe a correlation between proposed and actual payments in the business relations of treatment RU, in which the buyer is unrestricted, 19 The relations between proposed payments and actual payments in all three settings are illustrated in the Appendix (Figure A.3). 12

15 supports the conjecture that the existence of binding contracts may reduce cooperation. Trading behavior in treatment U seems to be more tightly connected to the contracts than in treatment RU, when both contractors are unrestricted like in treatment U. The possibility to be bound to the contract seem to induce unrestricted buyers in treatment RU to ignore the contract to a greater extent than this is the case with unrestricted buyers in treatment U. This suggests that the mere existence of contract enforcement measures might lead to some crowding out of the motivation to voluntarily follow a contract. Let us have a look at the first installments (see Figure A.2). In 14 out of 27, i.e., in over 50% of the business relations in which the contract is accepted, the first payment is performed as proposed, i.e., p 1 = r 1. Additionally, on the basis of all 27 business relations, p 1 and r 1 are highly positively correlated (p = 0.000, Spearman's rank correlation, one-tailed). It is interesting that analogous claims do not hold for the other two installments. Apparently, the buyers tie their decisions concerning later payments more loosely to the contract than it is the case with the first payment. In almost three quarters of all business relations with accepted contracts, namely 20 out of 27, the first unit is delivered as agreed in the contract. Thus, at least at the beginning of the business relation the seller adheres to the contract. Delivery of the second unit is made less frequently. But even at the second delivery stage there are indications that the contract has some influence on the chosen action: the second unit is delivered in those business relations in which the proposal for the third installment in the contract p 3 is higher (p = 0.002, Mann-Whitney U test, one-tailed). It is possible that a seller prefers to deliver the second unit if the contract promises her a higher third payment. Interestingly, the sums of the three proposed installments are lower in those business relations in which no unit is delivered (p = 0.043, Mann-Whitney U test, one-tailed). We use the term "overpayment" if a buyer pays more than what is agreed upon in the contract, i.e., p i < r i. Overpayments do not occur at the first and third installments but there are four overpayments (up to five talers) in the second installment. In three of these cases overpayment is successful in the sense that the second unit is delivered in one case even though the first delivery is refused. This observation might be a first hint of reciprocal behavior. Maybe the overpayments are made precisely because the buyers trust the sellers will perform a similarly friendly act afterwards. An alternative explanation is the let-down-aversion observed by DUFWENBERG and GNEEZY (2000). It could be that a seller delivers the second unit because she does not want to disappoint a buyer who overpaid 20. Further indications of a strong influence of reciprocity are given below. If p i > r i this is termed an "underpayment." If one assumes that the contract has an influence on actual behavior, one might conjecture that underpayments affect subsequent decisions concerning delivery. This is exactly the case for the second delivery. In business relations in which the second unit is delivered we observe a lower previous underpayment (p 2 r 2 ) (p = 0.037, Mann- Whitney U test, one-tailed). It is interesting that a comparable effect of underpayment for the first installment is not significant. 20 I am grateful to an anonymous referee for pointing out this explanation. 13

16 5.3 The Impact of Trust and Reciprocal Behavior in Treatment U In our view it is quite astonishing how unambiguously the traditional game-theoretical prediction under the standard assumptions of rationality and self-interest is rejected at the beginning of a business relation: 100% of the buyers pay strictly positive amounts as first installments. This is a clear indication that buyers trust they will receive a unit from the sellers if they have previously paid a certain amount. On the one hand, trust is possibly based on the promise made in the contract. On the other hand, buyers seem to trust that sellers will follow the norm of reciprocity, i.e., if a buyer voluntarily pays a considerable amount the seller should feel obliged to deliver. In almost three quarters of all business relations with accepted contracts the first unit is delivered. But, surprisingly, no apparent significant influence of the amount of the first installments on the delivery decisions can be detected. The same is true for the size of the underpayments. Thus, we have to assume that the delivery decisions of the first unit are not primarily regarded as reciprocal reactions following the first installments. Instead, the sellers might be led by their promises in the contract. Trust in future reciprocal behavior of the buyers may be an additional driving force. If the first unit has been previously delivered, buyers pay a positive figure significantly more often in the second installment, i.e., r 2 > 0, (p = 0.036, Fisher exact test, onetailed). We interpret this observation as a reciprocal reaction of the buyer to the experienced behavior of the seller. The buyers seem to be more willing to pay a positive amount if the sellers adhere to the contract. The second installments r 2 are higher in those business relations in which the second units are delivered (p = 0.002, Mann-Whitney U test, one-tailed). Again this must be seen as evidence for reciprocal behavior this time by the sellers. It seems to be easier for a seller to deliver the second unit if she has been treated generously by the buyer beforehand. Additionally, there is another indication of reciprocal behavior of the buyer at the third installment. In five business relations a positive third installment is made, i.e., r 3 > 0. In four of these cases both units are previously delivered. According to the Fisher exact test this means that the third installment is significantly more often positive if both units are delivered (p = 0.024, one-tailed). A buyer who receives both units seems to feel more obliged to pay something positive at the end than a buyer who receives fewer units. To sum up, in treatment U we find several indications that at the beginning of a business relation participants prefer (more or less unconditionally) to adhere to the contract. They might do so because they want to keep their promise or trust in positive future reactions 21. In the course of the business relation they tend to make their actions reciprocally dependant on the behavior of their counterparts. 5.4 Binding Contracts Reduce Cooperation In the introduction we mentioned that MACAULAY (1963) argues that business partners often close down reference to social norms of goodwill and cooperation if legal arrangements are used. Several observations in our study are in line with his findings. In treatment U buyers more often pay what is proposed in the contract at the first installment, i.e., r 1 = p 1, compared to unrestricted buyers in treatment RU (p = 0.009, Fisher exact test, one-tailed). Furthermore, in treatment RU buyers who 21 I am grateful to an anonymous referee for clarifying that an additional explanation for the fulfillment of the contract at the beginning of the trading phase might be reputation building (KREPS and WILSON 1982, SELTEN and STOECKER 1986). 14