Experimental Design
This study augments the lottery experiment developed by Tanaka, Camerer, and Nguyen (2010) to include the negotiation failure risk present in agricultural markets. Participants were presented with five series of lotteries, each a sequence of lottery questions, where potential payoffs from each lottery were presented in terms of “tokens” (100 tokens=$1). For each question in a series, participants were tasked to choose between two lottery options, Lottery A and Lottery B, where Lottery A is a “safer” option and Lottery B is “riskier.” Both lotteries had a known probability of a “high payout” and a known probability of a “low payout.” Within each series of lottery questions, the probabilities associated with Lottery A and Lottery B remained the same. At the beginning of the series, the expected value of Lottery A is higher than Lottery B. As the questions in the series progress, the relative expected value of Lottery A to Lottery B decreases, and by the end of the series, the expected value of Lottery A is less than that of Lottery B.
Subjects were asked to report the question in a given series in which they switch their preference from entering into Lottery A to entering into Lottery B (from switching at the first question through never switching) (see Appendix for survey instructions provided to participants). Subjects were informed that they could only choose one question in which they would switch from Lottery A to Lottery B in a given series (that is, they could not switch back to Lottery A once they chose Lottery B) in order to enforce monotonicity. Subjects that choose to switch at any of the first few questions are more risk seeking, and those that never switch, or switch only at the last few questions, are more risk averse.
The first four lottery series had 14 questions each, and the last series had 7 questions (63 questions total). Subjects were instructed to indicate their switching point for each of these five lottery series, generating five data points for each subject. Subjects were informed that they would earn tokens based on their response to only one randomly selected question. After participants made all of their choices, the experimenter drew a random chip from a jar of 63 labeled chips to determine the specific lottery question on which all subjects’ earnings would be based. Then the experimenter drew a ball from a jar, where the color of the ball indicated which outcome prevailed (“low payout” or “high payout”) for the selected question. Following Tanaka, Camerer, and Nguyen (2010), subjects earned tokens based on their choice of Lottery (either A or B) for the specific question and outcome that were randomly selected.
The first two lottery series (Series 1 and 2) are identical to those developed by Tanaka, Camerer, and Nguyen (2010). In these series, the decision context is a simple lottery. The payoffs for Lottery A remain constant throughout the series, while those for Lottery B change.
We add two new series to the original design of Tanaka, Camerer, and Nguyen (2010) to represent the fact that risky decisions are typically not presented to agricultural producers in terms of traditional lotteries, but rather in the context of trading commodities in a private negotiation setting. In Series 3 and 4 are in the context of forward delivery and Series 5 (comparable to Series 3 of Tanaka, Camerer, and Nguyen [2010]) reflects spot delivery by adding advance production risk to sellers. In Series 3 and 4, participants were told that they were either a buyer or a seller of a product, and that they would earn tokens based upon successfully negotiating a trade. To mimic negotiation failure risk in forward markets, each question in a lottery series had a given probability that a specific asking price was accepted, leading to positive profits, and a given probability that the asking price was rejected, leading to no trade and zero profits. The expected value of these questions was identical to that of Series 1 and 2, but since Series 3 and 4 include a risk of zero profit, the token values associated with the high and low outcomes are different.
To measure the influence of role (buyer or seller) on behavior, we varied assignment of buyer and seller roles across series. In Series 3, participants were told that they were buyers bidding to purchase a product from a seller. To make a profit and earn tokens in the experiment, buyers were told they would be able to resell any product purchased to the experimenter, known as the resale value. Participants were informed that tokens were earned by purchasing the product at a price below this resale value. In Series 4, participants were informed that they were now sellers contracting a product for sale to a buyer. Participants were given a production cost and told that if they successfully negotiated a sale, their profit would be the negotiated price minus the production cost. If no sale was negotiated, participants were informed that they would neither incur production costs nor earn any tokens, consistent with forward delivery.
To elicit risk preferences across a broader range of probabilities, we created two versions of the survey. In Version 1, the probability of making a positive profit in Series 3 as a buyer was relatively low (30% and 10% for Lottery A and B, respectively, representing high negotiation failure risk) but relatively high in Series 4 as a seller (90% and 70% in Lottery A and B, respectively, representing low negotiation failure risk). In Version 2, the probability of making a positive profit in Series 3 as a buyer was relatively high (90% and 70% for Lottery A and B, respectively), but relatively low in Series 4 as a seller (10% and 30% in Lottery A and B, respectively). Because the probabilities changed between the two versions, the resale value, production cost, and potential positive profits changed as well.
In Series 5, participants were assigned the role of a seller in a spot market (in both survey versions). Since production happens before a sale in spot markets, producers risk losing their cost of production from negotiation failure. Even if a trade is agreed upon in this market, the price may be below a seller’s production cost, leading to profit loss. Series 5 used the same probabilities and outcomes as designed by Tanaka, Camerer, and Nguyen (2010), yet participants were informed that they were a seller offering to sell a product that they had already produced. Thus, in both Lottery A and Lottery B there was a 50% chance of a successful trade price that generated positive profits, and a 50% chance of either no trade (represented by a buyer counteroffer of 0 tokens) or a price below the seller’s production cost.
Series 1 and 2 are used to measure σ and α (consistent with Tanaka, Camerer, and Nguyen [2010]) as defined in equations 1 through 3. Because Series 3 and 4 were designed with a zero possible payoff, they cannot be used jointly to measure individuals’ level of risk aversion (σ) and curvature of weighting function (α). However, Series 3 and 4 are used along with the level of α from Series 1 and 2 to determine σ30_10 and σ90_70. The parameter σ30_10 is an individual’s level of risk aversion based on the series with high negotiation failure risk (i.e., probability of positive outcomes of 30% and 10% for Lottery A and B, respectively), and σ90_70 is an individual’s level of risk aversion based on the series with low negotiation failure risk (i.e., probability of positive outcomes of 90% and 70% for Lottery A and B, respectively). Lastly, Series 5 was used to measure loss aversion (λ) in the context of spot delivery to capture the advance production risk environment where loss aversion may affect seller behavior.
Market Experiment
To test the influence of prior market experience on behavior observed in the CPT survey, and the influence of σ, α, and λ on market outcomes, half of the sessions had subjects participate in a laboratory market experiment prior to the survey. Market experiments were conducted immediately before the CPT survey, with trading conducted over a computer network using software that simulated a privately negotiated market with forward delivery. During each trading period, buyers and sellers were randomly matched and negotiated trades one-on-one. Buyers made bids and sellers made offers until they reached an agreed price and a trade was made. The buyer and seller could then begin negotiations to trade the next unit (up to 8 units) until the end of the 1-minute trading period. Then, buyers and sellers were again randomly re-matched to begin trading units in the next period, where each buyer and seller pair would start over on their production cost and redemption value schedules. By doing this, matched buyers and sellers were on the same unit in their respective schedule in order to reduce the potentially confounding influence of matching risk (i.e., matched partner having a different incentive to trade because they are on a different unit in their schedule) and negotiation failure risk.