Detection, Identification, and Estimation of Loss Aversion: Evidence from an Auction Experiment

We provide a novel experimental auction design, in which (i) an exogenous decrease in the probability of winning, conditional on the bid, reduces the optimal bid of a loss averse agent whose reference point is expectations based; (ii) observed bid distributions uniquely identify the participants' latent value distribution and loss-aversion parameter. Experimental evidence affirms the presence of such reference points. We show that at the estimated magnitudes of loss aversion, (a) conventional Becker, DeGroot, and Marschak (1964) experiments may lead to large biases in estimated willingness to pay (which our design can correct for); and (b) first-price auctions may fetch moderately higher revenue, compared with second-price auctions.