Entrepreneur: Start & Grow Your Business

Contract enforcement, social efficiency, and distribution: some experimental evidence.


by Wu, Steven Y.^Roe, Brian

Recent controversies surrounding contract production in agriculture have raised concerns among policy makers that vertically coordinated industries might favor large agribusiness at the expense of growers. Among the issues that have received attention include the apparent lack of bargaining power possessed by growers when negotiating and performing under contracts, and the lack of transparency concerning how performance and payments are determined (Pierce and Stewart 1997). As an example of the latter, Schrader and Wilson (2001) analyze data from a survey of broiler growers and suggest that many growers were concerned about the quality of inputs they received from contractors and the timeliness and accuracy of bird weighing. Because both input quality and the accuracy of weighing can affect performance outcomes, which in turn, determine compensation, there might be a lack of transparency in the determination of pay. (1) Along similar lines, Hamilton (2001) points out that many broiler contracts contain a provision that allows a contractor to unilaterally change payment methods or rates, which provides integrators with the option of making discretionary ex post adjustments in pay. The lack of transparency can make it difficult for third parties, such as courts or regulators, to enforce contracts. This means contracts may be rendered incomplete by limits to enforcement.

While enforcement problems appear to be key issues for policy makers, there have been few efforts to clarify the consequences of contractual enforcement on efficiency and distribution. Textbook principal-agent models are inadequate for studying enforcement problems because an assumption of these models is that third-party enforcement of contracts is perfect. Thus, law and economics scholars focus on contracting environments where there are limits to third-party enforcement so that contracts are incomplete.

An analysis of the impact of third-party contract enforcement is complicated by the fact that traders can devise informal enforcement mechanisms to self enforce contracts. Incomplete contracts combined with the possibility of repeated trading, which is common in some agricultural subsectors, (2) enable private parties to devise informal enforcement mechanisms (i.e., form relational contracts) to mitigate hold-ups and other distortions (Telser 1980; Klein 1996; Macleod and Malcomson 1989; Levin 2003). Relational contracts fit many of the stylized facts of agricultural contracting because even if explicit contracts exist, some dimensions of performance (e.g., quantity commitments, timing of deliveries, contract renewal) are frequently omitted from explicit agricultural contracts, opening the door for informal incentives.

Apart from theoretical issues, there is a paucity of data for studying policy issues in vertically coordinated industries. Explicit terms in contracts are often buttressed by unwritten rules, informal incentives, and tacit expectations so that it is difficult to find observational data that captures every important aspect of the contracting environment. There is also a lack of historic precedent for many of the contracting issues that have emerged recently in U.S. agriculture. Thus, there is little data available for assessing the impact of alternative enforcement regimes on trading outcomes.

Given the lack of theoretical and empirical research, it is not surprising that economists and policy makers hold numerous opinions about how government regulations of agricultural contracts would affect efficiency and distribution. Some economists suggest that too much regulation spurs unintended consequences that, for example, may cause livestock industries to leave states in which regulations are binding (Boehlje et al. 2001). Others suggest that regulations and enforcement are necessary to enhance both fairness and competitiveness in contract agriculture (Harl et al. 2001; Taylor 2002).

This study uses experimental economics to investigate the effects of third-party enforcement on contractually based markets with buyer concentration. Our experimental setup is based on the design of Brown, Falk, and Fehr (2004); two treatments are identical to theirs while a third treatment unique to our study facilitates examination of issues pertinent to agricultural contracting. An experimental approach allows for exogenous variation in enforcement regimes, which facilitates estimation of causal relationships. We view this study as an initial step in understanding the microeconomic forces that shape trading patterns in a contractually based economy where buyers have bargaining power and third-party contractual enforcement might be imperfect. The insights generated in this study can shed light on how policies that increase transparency and/or third-party enforcement of contracts can affect efficiency and the distribution of rents.

Noussair and Plott (1995) argue that experiments need not replicate field situations and all institutional details to retain relevance for policy analysis. Instead, experiments are valuable in that they allow economists to examine general theories that should apply more broadly. If a theory does not apply in simple, controlled environments, one must question whether the theory is appropriate for explaining behavior or predicting responses in more complex environments. Moreover, abstracting from reality is not unique to experiments; indeed, most economic studies incorporate simplifying assumptions and abstractions.

One main finding is that perfect third-party enforcement promotes social efficiency. However, we also find that a regime devoid of third-party enforcement can generate comparable social efficiency. In such regimes, many subjects use contractual renewal and discretionary adjustments in contract terms to provide informal enforcement that can be nearly perfectly substitutable with formal enforcement in terms of generating social surplus. This finding is consistent with the case studies provided by Anderson (1999) and Gow, Streeter, and Swinnen (2000), which show that self-enforcing agreements can substitute for inefficient third-party enforcement. However, we find that a nontrivial fraction of trades resulted in sellers (i.e., growers) making ex post profits that fell below reservation payoffs due to opportunistic buyer behavior. Another interesting result is that one-sided formal enforcement (buyers' obligations only) constrains subjects' ability to use informal enforcement, which resulted in significant efficiency losses.

Experimental Design

Motivation

Agricultural contracts facilitate vertical coordination because processors can provide incentives for performance factors that enhance profits. In a typical contract, a grower agrees to produce crops or animals in a manner consistent with conditions in a written or verbal agreement. (3) Growers are then paid according to a price established in the agreement.

In practice, a processor may care about a range of performance factors, including quality, on-time delivery, product consistency, and efficient input use. While there is considerable heterogeneity in performance factors across contracts, the key point is that contracts are essentially incentive devices for increasing productivity and the value of trade. Thus, in designing our experiments, we were primarily interested in a design that would allow us to assess the incentive effects of contracts.

In complete contracting environments, where third-party enforcement is assumed perfect, the crucial question is how variations in explicit incentive structures affect productivity. However, if there are impediments to complete contracting, such as limits to enforcement, then parties may be constrained in their ability to structure explicit incentives and must also rely on implicit or informal incentives. In practice, contractors use discretionary adjustments in pay and contract termination policies. Such discretionary ex post adjustments in contract terms can affect future trading equilibria.

Given that there often are barriers to third-party enforcement of agricultural contracts, implicit incentives cannot be ignored. Moreover, even if explicit incentives exist in an incomplete contract, informal incentives will still supplement explicit incentives in motivating performance. For example, although broiler contracts contain explicit incentives for efficient input use, processors may supplement these incentives with discretionary adjustments in pay and/or the number and frequency of flock placements, which affect future pay. Some have suggested that when there are limits to enforcement, it may be optimal to omit explicit incentives entirely even if some dimensions of performance are enforceable (Bernheim and Whinston 1998). Thus, the focus of our study is on informal incentives and how interventions (e.g., new policies or institutions) that impact third-party enforcement will affect trading outcomes.

Details of the Design

We follow the basic design of Brown, Falk, and Fehr (BFF) and two of our treatments are identical to theirs. In addition, we add a third treatment, which is unique to our study. In this treatment, we let buyers make discretionary ex post adjustments in compensation. Hamilton (2001) points out that some contractors can make unilateral adjustments in payments ex post, so this allowance is consistent with agricultural contracting. Further, parties can track the reputations of past trading partners and have discretion to renew or dissolve existing relationships based on past performance. This creates the possibility of relational contracting where the promise of future relationship-specific gains from trade can be used to provide informal incentives to discipline current behavior.

In each experiment subjects are partitioned into two groups: buyers and sellers. (4) All trading takes place on networked computers enclosed in cubicles to eliminate between-subject visual contact. Moreover, anonymity is preserved by assigning all subjects identification (ID) numbers. Each experiment has two trading sessions featuring distinct contract enforcement regimes. Each session has 17 trading rounds--two practice rounds and 15 "live" rounds that may determine eventual cash payment. ID numbers are fixed across rounds, which allows subjects to develop and track reputations.

Within each trading round, buyers offer contracts to sellers specifying a price--quality combination for a unit of an abstract good, where quality can be thought to embody all costly performance factors. Sellers can only accept or reject offers. A buyer can make as many offers as desired in each round, but once one offer is accepted, all other offers are withdrawn and no additional offers can be made. Similarly, once a seller accepts an offer, no other offers can be entertained. In short, each buyer and seller can conclude at most one trade per round. No buyers (sellers) are obligated to make (accept) offers in any round.

Buyers can extend two types of offers: public and private. Public offers are displayed on the computer screens of all sellers and buyers; any seller can accept any public offer. Private offers are extended by entering a specific seller's ID number into the computer. Only the seller identified sees the offer and only he can choose to accept it. Private offers enable long-term relationships, which lie at the core of relational contracting theory. For example, if a buyer predicts benefits from contracting with a specific seller (i.e., can earn relationship-specific rents) and wants to establish a long-term relationship, the buyer can make a single, private offer to that seller in each round rather than venturing into the open market and hoping that that seller is the first to accept the offer. (5) Had we not incorporated private trading, it would have been difficult for parties to establish relational agreements that persisted over time as buyers would have had to hope that their targeted sellers accepted their public contracts before any other sellers did so.

Every round features the same five buyers and seven sellers. Fewer buyers than sellers create buyer concentration because at least two sellers do not trade in each round. This forces sellers to compete for a limited number of contracts, which tilts bargaining power in favor of buyers. (6) Excess demand for contracts is consistent with stylized facts in some sectors. For example, according to Mitchell (2004), most chicken processors have waiting lists for those who want to become growers and some existing growers want to add capacity to expand contract production. Our subjects maintained the same role (e.g., buyer) during both sessions in an experiment and no subject participated in more than one experiment (two sessions).

In order to test the impact of third-party enforcement on efficiency and surplus, we examine three enforcement regimes. First, in the "complete contract" (C) treatment, the computer fully and perfectly enforces all contractual components. That is, sellers must supply the quality stipulated in the contract, and buyers must pay the agreed upon price. In the second treatment, called "Relational Contract 1" (RC1), the computer only enforces contractual price, so that discretionary price adjustments are made illegal. Quality is unenforceable, that is, sellers can supply quality that differs from contractual specifications. This treatment mimics situations where neither input quality nor output measurement (e.g., weighing of birds) is verifiable by a third-party. In this case, measured performance is unenforceable by a third-party because when quality is low, both parties can blame each other (e.g., growers can claim poor quality inputs and processors can claim grower torpor) and verifiability is costly. Our C and RC1 treatments are identical to BFF's. The third treatment, called "Relational Contract 2" (RC2), is unique to our study. This treatment is similar to RC1 except that the buyer can make expost adjustments to the promised price; that is, after observing the seller's delivered quality, the buyer can choose a price that deviates from contract specifications. Our RC2 treatment allows for the examination of a broader range of informal incentive mechanisms used to regulate trading. We show that our RC2 treatment can provide significant new insights into the nature of relational contracting relative to the findings of BFF.

Round specific payouts are determined for buyers as follows:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where [[pi].sub.b] is the buyer's profit, Q is the actual quality chosen by the seller, and P is the actual price received by the seller. All payments are given in experimental points where subjects earn 1 dollar for 70 points.

The seller's profit is

(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where r is a reservation payoff in the absence of trade (5 or 10 depending on the session). We impose positive reservation payoffs for sellers to ensure that they earn some money during the experiment. Varying reservation payoffs should only induce buyers to change their price offers to ensure that sellers' reservation payoffs are covered, but efficiency should not be affected. During the experiment, Q is an integer in the set {1, 2, ...,10} and P is an integer in the set {0, 1, ..., 100}. The cost function, c(Q), is fully represented by the following schedule of quality-cost combinations: {1, 0}, {2, 1}, {3, 2}, {4, 4}, {5, 6}, {6, 8}, {7, 10}, {8, 12}, {9,15}, {10, 18} so that c(Q) is increasing and convex. Note that marginal cost never exceeds 3 and that a buyer's marginal benefit always exceeds 3 so that first best is achieved if the seller delivers maximum quality. At the end of each round, each subject is informed of her own payoff and her trading partner's payoff.

We conducted thirteen experiments. Each experiment included two of the treatments (C, RC1, or RC2). Each subject participated in both treatment sessions during an experiment, which means that order effects might arise. Hence, the design counterbalances the order of appearance for each of the three enforcement regime sessions. (7) Before subjects were paid, they provided demographic information. (8) There were a total of 156 subjects in twenty-six sessions (two in each of the thirteen experiments) where thirteen were C treatments, seven were RC1 treatments, and six were RC2 treatments. Subjects made 1,903 out of a total of 1,950 possible trades across all sessions. Table 1 provides a summary of treatments, sessions, participants, rounds, and trades. The experimental economy was programmed using Z-TREE software (Fischbacher 1999). For five C treatments, r = 10, and for the others, r = 5; r = 5 for all RC1 and RC2 sessions. See Wu and Roe (2006) for additional details of the experiment, including subject instructions.

Predictions

In this section, we provide theoretical predictions for each treatment. Like BFF, we use the theory of repeated games to generate predictions. We focus on the case where the seller's reservation payoff is set at r = 5, but the case where r = 10 is analyzed similarly.

In treatment C, the sequence of events within a round is as follows. First the buyer offers a contract, which specifies the desired P and Q. If a seller accepts, payoffs are determined and the round ends. Note that Q and P specified in an agreement are third- party enforced, that is, neither party can deviate from contracted Q and P. Thus, the buyer's profit-maximizing contract choice is determined by solving the problem

(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].

Substituting the binding constraint into the objective function yields

(4) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

which gives the first-order condition

(5) 10 - c'(Q) = 0.

As noted above, marginal cost never exceeds 3 so the buyer should implement maximum quality, [Q.sup.*] = 10. With [Q.sup.*] in hand, it is easy to solve for [P.sup.*] = 23 from the participation constraint to ensure that the seller will accept. Because both parties can earn at least as much when trading, we predict that all five buyers will make offers and that five sellers will accept them. Hence, in equilibrium, five trades occur in the round and joint surplus is given by

(6) [S.sup.C] = [[pi].sub.b] + [[pi].sub.s] - r = 77.

Due to perfect third-party enforcement, sellers and buyers have no incentive to deviate from this prediction in any round of the session.

Under RC1, the sequence of events within a round are the same as under C except that after a seller accepts, Q is unenforceable by a third party so the seller can deviate from the quality specified in the contract. After Q is chosen by the seller, payoffs are determined and the round ends. To determine what Q the seller will choose and what contract the buyer will offer, note that it is common knowledge that each live session ends after fifteen rounds. Thus, we must use backward induction and begin analysis in round 15 of the finitely repeated game. Within a round, the seller is the last mover. Therefore, consider what the seller should do given that she has accepted a contract and is guaranteed a payment P. Note from her objective function (2) that profit is maximized when Q = 1 so that production cost is zero. Hence, equilibrium quality is [Q.sup.RC1] = 1. The buyer anticipates that the seller will deliver the lowest quality and offers just enough to ensure participation ([p.sup.RC1] = 5). Again, both parties earn at least as much under trade. Thus, in equilibrium, five trades occur during the round and joint surplus is given by [S.sup.RC1] = 5. Unenforceable quality leads to substantially lower joint surplus and quality, which, via backward induction, will obtain in all previous rounds.

In RC2, the sequence of events within a round is the same as RC1 except that after the seller chooses quality, the buyer chooses a payment that can differ from the price in the contract. To determine equilibrium in RC2, first consider how the buyer, who is the last mover, will respond to a given quality, [Q.sup.0] chosen by the seller. The buyer's payoff is [[pi].sub.b] = 10[Q.sup.0] - P, which is maximized by setting [p.sup.RC2] = 0. The seller, anticipating that the buyer will renege on any contracting agreement with positive payment, will expect to earn [[pi].sub.s] = [p.sup.RC2] - c([Q.sup.0]) = 0 - c([Q.sup.0]) < 5 from the contract, which is lower than reservation earnings. Thus, the seller accepts no contract. In equilibrium, no trade takes place and no surplus is earned in round 15. A similar logic follows for all earlier rounds. With two-sided discretion, the contracting market is completely destroyed by opportunism.

The above analysis provides us with several hypotheses:

HYPOTHESIS 1: When Q is unenforceable by a third party (RC1 or RC2), total surplus will be lower than the case when quality is enforceable (C).

HYPOTHESIS 2: Whether quality is enforceable or not by a third party, all surplus goes to the buyer and trading sellers earn reservation payoffs.

HYPOTHESIS 3: If both quality and payment are unenforceable by a third party (RC2), the contracting market collapses and no trade occurs in equilibrium.

Certain aspects of Hypotheses 1 and 2 are similar to the equilibrium predictions of BFF in that only minimal quality trading is anticipated in RC1 and all surplus goes to buyers. Our prediction about RC2 is even more stark-the market collapses. However, these predictions rely on strong assumptions about the self-interested motivations of all subjects. When these assumptions are violated, such as when a subset of subjects has social preferences (Fehr and Gachter 2000; Charness and Rabin 2002; Engelmann and Strobel 2004, among others), then different outcomes may emerge. BFF show that when enough subjects place a value on fairness, there is a perfect Bayesian equilibrium in which trade and high-quality levels emerge, even in a finitely repeated game. The key mechanism supporting this "cooperative" equilibrium is that subjects who care about equity will choose strategies that are "fair" and support high relationship-specific rents, even in the final period. The existence of these rents in the final period will incentivize even selfish subjects to cooperate in earlier periods so that the perfect Bayesian equilibrium outcomes may resemble what might occur in an infinitely repeated game with strictly selfish types. Moreover, multiple equilibria are possible depending on subjects' beliefs and the nature of subjects' social preferences. Given that multiple equilibria are possible and the potential for equity-minded subjects, it becomes even more important to study subjects' actual behavior in an experimental setting, rather than to rely purely on theory. Nonetheless, our theoretical predictions provide a useful organizing framework.

Results

Trading

We now provide an overview of trading outcomes from each treatment. Focusing on rows 8 and 9 of table 1, a first notable result is that, under RC2, subjects executed 449 out of the 450 possible trades, which contrasts the dire prediction from hypothesis 3 that the market will collapse. Of the three treatments, RC2 may most resemble agricultural markets given barriers to third-party enforcement and the ability of processors to make ex post adjustments in contract terms. One can see that such markets can remain active as subjects can use informal enforcement mechanisms to replace third-party enforcement.

One informal enforcement mechanism is private trading. Figure 1 graphs the evolution of private trading across periods. The share of private trades under RC1 increases over time before settling between 60% and 70% by the ninth round. The share of private trades under C is significantly lower than under RC1 and never exceeds 30%. These patterns are similar to BFF's results, which corroborate our experimental procedures and results. Under RC2, the share of private trades is significantly lower than under RC1 but consistently higher than under C. Kruskal-Wallis (KW) tests indicate significant differences in private trading between RC1 and RC2 (p = 0.0001), C and RC2 (p = 0.004), and C and RC1 (p = 0.0001). Our results suggest that buyers rely more on private trading when other enforcement instruments are missing.

[FIGURE 1 OMITTED]

Because buyers in RC2 resort to private trading less than they do in RC1, presumably due to the availability of discretionary ex post adjustments in price as a second informal enforcement mechanism, we examine this second instrument further. While the latitude to make ex post adjustments in payment terms is a controversial aspect of agricultural contracts as it can increase the probability of opportunistic behavior by buyers, our results suggest that these adjustments also provide incentives. That is, if sellers anticipate that buyers will reward (deduct) for high (poor) performance, then sellers may increase quality. Such price adjustments are consistent with discretionary bonuses of the sort discussed by Levin (2003), which can be used to provide incentives in relational contracts. If buyers are indeed making discretionary adjustments to incentivize sellers, then we ought to observe some correlation between performance and price adjustments.

Table 2 reports price adjustments made by buyers cross tabulated against performance for all RC2 trades. Buyers do seem to make adjustments contingent on performance. Conditional on good performance (Q > [Q.sup.*]), rewards are used 33% (2%/6% [approximately equal to] 33%) of the time under public trading and 67% (4.4%/6.6% [approximately equal to] 67%) under private trading whereas conditional on poor performance, sellers are rewarded only 1.1% (0.4%/35.2%) of the time under public trading and 2.5% (0.2%/8%) under private trading. Conditional on good performance, deducts are observed 55% and 19.6% of the time under public and private trading respectively, but these shoot up to 83.5% and 66.3% conditional on bad performance. Moreover, when sellers deliver quality that equals agreed upon quality, no price adjustment is most frequently observed. These patterns suggest that (1) discretionary price adjustments do appear to be contingent on quality, which is consistent with incentive provision, and (2) opportunistic behavior, independent of incentive provision, still occurs; that is, some buyers impose price deductions in trades where sellers meet or exceed performance obligations. This opportunistic behavior is more common in public trading. For example, when sellers meet performance obligations, buyers make downward adjustments in 41% (11.4%/27.6%) of public trades and in only 9.6% of private trades.

Surplus and Profits

We now discuss efficiency and the distribution of profits that result for each treatment. Hypothesis 1 predicts that total surplus under imperfect third-party enforcement will be lower than under perfect third-party enforcement. Figure 2(a and b) provides views of total surplus in each treatment under private and public trading. Under private trading, surplus under RC1 is consistently lower than under the other two treatments. In contrast, surplus under RC2 appears similar to surplus under C; we show formally in subsequent econometric analysis that there is no statistically significant difference. The combination of private trading combined with ex post price adjustments appears to provide powerful incentives, even in the absence of third-party enforcement. In contrast, under public trading, surplus under RC2 appears consistently lower than under C.

[FIGURE 2 OMITTED]

Turning now to distributional issues, note that hypothesis 2 predicts that, in a concentrated market, buyers obtain all surplus while sellers earn reservation payoffs. Figure 3 illustrates how the ratio of the sellers' surplus to total surplus evolves across periods. In both C and RC2, sellers consistently captured less than 50% of total surplus; in fact, in round 15, sellers earned negative surplus (sellers earn less than reservation payment). At the other extreme, under RC1, sellers collectively never earned less than 50% of total surplus and, in some cases, earned more than 100%, especially in early and late rounds when relationships were being established or were ending.

[FIGURE 3 OMITTED]

One logical explanation for high seller surplus in RC1 is that Q is unenforceable; thus, the power to engage in opportunism belongs to sellers. Figure 4(a) illuminates this by contrasting the ratio of seller to total surplus under RC1 in private and public trading. Note that the ratio tends to be lower under private trading. Private trades indicate that relational trading is taking place, making it less likely that parties will renege on their agreements for fear of breaching self-enforcement constraints. If sellers or buyers shirk on their agreements for short-term gains, they risk being terminated by the other party and foregoing long-term relationship-specific gains. Nonetheless, one can see that even under private trading, the ratio increase dramatically in the final periods when relationships are about to end.

[FIGURE 4 OMITTED]

Figure 4 (b) provides the same analysis for treatment RC2. Note that unlike RC1, the ratio is higher under private trading. Intuitively, in RC2, buyers may also engage in opportunistic behavior so that private trading appears to discipline buyer opportunism thereby increasing sellers' surplus. Nonetheless, in the last few rounds, buyer opportunism increases so that sellers' surplus falls significantly, even under private trading.

So far, we have suggested that opportunistic behavior may explain some of the patterns observed in figures 3 and 4. To better understand the role of opportunism, turn to figure 5, which graphs sellers' propensity to renege on quality across periods in the incomplete contract treatments. Note that the percentage of trades in which sellers reneged was lowest under private RC2 trading and highest under public RC1 trading. It is particularly interesting that sellers reneged less often in private RC1 trading than in public RC1 trading, except in the early and late rounds. It appears that sellers were more likely to renege on quality before relationships were well established (in the early rounds) and when they were about to end (in the late rounds). However, in RC2 sellers renege less often in early and late rounds, presumably because buyers can still use discretionary price adjustments to discipline sellers even when relationships are fragile.

[FIGURE 5 OMITTED]

Rows 1 through 4 of table 3 report sellers' propensity to renege on quality after buyers retaliate; that is, after being terminated. One might anticipate that sellers would renege less often after being punished (terminated) but there was little evidence of this. Under RC1, sellers renege more if they were just terminated regardless of whether post-termination trade is public or private. This might be because terminated sellers have to switch to partners with whom they have no relationship-specific capital. The evidence is more mixed for RC2. Under private (public) trading, sellers renege more (less) if they have just been terminated.

Rows 5 to 8 compare sellers' promised profit, which is what they would earn if all parties honor the terms of the contract, to realized profits. On average, sellers earn more than promised in RC1 in both public and private trading, which is no surprise considering that sellers can engage in opportunistic behavior. In contrast, in RC2, sellers earn less profit than promised perhaps because buyers can behave opportunistically by making price deductions even when sellers meet obligations (recall that conditional on good performance, deducts were observed 55% of the time in public trades). This is consistent with claims made by some growers that they earn less than expected under contract production. For instance, Hamilton (2001) reports that, in a survey of broiler growers, 43% indicated that they earned less than they had anticipated under contract production.

Rows 9 and 10 reveal that in the C and RC1 treatments, sellers rarely earn profits below reservation levels so that, on average, sellers make good contract choices. However, in RC2, buyer opportunism results in sellers earning less than reservation payoffs, ex post, in 24% of public trades and in 7.9% of private trades. Thus, although sellers in RC2 accept the best contracts ex ante, in nearly one in four public trades, sellers end up with profits below reservation payoffs. This is mitigated by private trading but sellers still earn profits that fall below reservation payoffs in nearly 8% of trades.

We now turn to an econometric investigation of our data, which allows for a more complete analysis of surplus and profits. We estimate regressions where the dependent variables are surplus, buyer profit and seller profit. We are interested in the impact of RC1 and RC2 on each of the dependent variables so we include dummy variables for each regime as explanatory variables. Moreover, because our experiment is a repeated game where subjects can establish cooperative agreements by making the continuation equilibrium conditional on past play, the observations are history dependent. Thus, surplus/profits in a given period may depend on the underlying equilibrium, which in turn depends on past strategies and actions. We account for this dependence by including lagged quality chosen by sellers, lagged discretionary price deviations by buyers, and a measure of the length of private relationships. Additional control variables include demographic information and linear and quadratic time trends to capture learning and search. The "reservation" variable represents the seller's reservation payoff to account for the fact that reservation payoffs varied from 5 to 10 in some C sessions. Unobserved experiment-specific effects may cause the composite error term of observations within an experiment to be correlated. Thus, the robust standard errors were adjusted for clustering on experiments. (9)

While our main explanatory variables of interest are RC1 and RC2, we want to interact these treatment variables with a "private" dummy because it is likely that the incremental impact of RC1 and RC2 depends on whether trades are private or public. Because the decision to trade privately may be endogenous, a two-stage endogenous switching model is used. The first-stage probit model is used to predict the probability of private trading and then the inverse Mill's ratio calculated from this probit is included as a regressor in the second-stage regressions. In the second stage, the sample is split into public and private trades and the regression equation is estimated for each subsample. The exclusion restrictions used for the first-stage probit are lagged variables for "positive surprise" (Q > E(Q)) and "negative surprise" (Q < E(Q)), where E(Q) is the quality level the buyer expected to receive. In each round of RC1 and RC2 each buyer was asked to enter their quality expectation immediately after the seller accepted the offer (and prior to delivery). These variables are attractive exclusion restrictions because they should affect the probability that a buyer will continue to contract privately in the following period, but should have no direct impact on surplus and profits in a future period; that is, these variables should only affect future surplus and profits through the private variable.

Table 4 presents the results of the first-stage regression. (10) The regression reveals a strong, positive relationship between RC1 and the probability that private trading is used, which is consistent with figure 1. The previous length of a private relationship between a buyer and seller also significantly increases the probability of private trading. Thus, buyers/sellers locked in relationships are highly likely to use private trading.

Table 5(a) reports the second-stage regression explaining total surplus. The explanatory variables are the same as those used in the first-stage probit except for positive and negative surprise, which are excluded. A dummy variable for treatment C was omitted so that C represents the base treatment against which comparisons are made. The coefficient for RC1 is negative and significantly different from zero for both subsamples, suggesting that surplus under RC1 is lower than surplus under C whether trades were conducted publicly or privately. However, the difference in coefficients across the subsamples is estimated to be 26.48 and is significant at the 5% level according to a Chow test, suggesting that private trading reduces efficiency loss (relative to C) by an average of 77%. Turning to RC2, it is interesting that the coefficient for the private sample is not significantly different from zero--RC2 does not appear to reduce surplus relative to C, provided that subjects trade privately. This result is in stark contrast to RC1 and it suggests that relational contracting combined with a buyer's ability to make discretionary adjustments in price can provide informal enforcement that can nearly perfectly substitute for formal enforcement in producing social surplus.

Table 5(b and c) report results for the second stage regressions for buyer and seller profit. In table 5(b), the coefficient for RC1 is negative and significantly different from zero for both subsamples, meaning that buyers clearly profit less under RC1 than C. The same coefficients in table 5(c) are positive, though only the private subsample coefficient is significant. Thus, sellers may benefit from RC1 (relative to C) under private trading. While this appears to contradict figure 4(a), which shows that sellers earn a larger fraction of surplus under RC1 in public trading, recall from figure 2(a and b) that surplus is much higher under private trading; thus, a higher fraction of surplus may not necessarily mean higher absolute profits.

Our results indicate that, on average, the impact of RC1 under public trading is to reduce buyer profit by 45.18 compared to profit under C; seller profits are the same. Under private trading, both buyers and sellers benefit compared to public trading as buyer losses drop from 45.18 to 16.20, while seller profits are 7.97 higher than they are under C. Hence, under private trading, RC1 reduces buyer profit and increases seller profit, which contradicts hypothesis 2. In contrast, RC2 has very different effects on buyers and sellers. None of the coefficients for RC2 in table 5(b or c) are significantly different from zero. The only significant finding regarding RC2 is that buyers profit more under RC2 when engaged in private trades, though this result is significant at only the 10% level.

Overall, the results suggest that neither social efficiency nor profits will change in the absence of formal enforcement of contracts provided that the trading parties have a sufficient range of informal enforcement mechanisms available. If the number of informal instruments is limited (e.g., RC1) so that buyers are constrained in their ability to use informal mechanisms to enforce quality, then significant surplus losse