Detection of Loss Aversion in Auctions Against Pre-Programmed Computers

We run experiments of first-price auctions with two groups, by which we directly detect the presence of bidders’ loss-aversion. Each human bidder bids against three preprogrammed computer bidders. The computer bidders independently draw their values from the uniform distribution and bid their values in one group and 75 percent of their values in the other group. We show that loss-aversion differentiates human bidders’ bids between the two groups, unlike other behavioral influences, e.g., risk-aversion, ambiguity aversion, biased beliefs, level k rationality, etc. We present statistical evidence of loss-aversion in our data. We then identify and estimate the joint distribution of bidder-specific loss and risk-aversion coefficients. We also document that loss-aversion and risk-aversion coexist with a strong positive correlation and high heterogeneity. Finally, 38 to 46 percent of overbidding is attributable to loss-aversion