Peer prediction markets to elicit unverifiable information - Cem Peker

Abstract: This paper introduces an incentive mechanism to elicit subjective judgments. We consider binary questions where responses cannot be verified for accuracy. Self-reported information may not reflect the truth due to lack of cognitive effort or motives to lie. In our formal framework, agents choose whether to receive a costly private signal, which leads them to endorse either "yes” or “no” as an answer. Then, they either buy or sell a single unit of an asset, whose value is determined by the endorsement rate of “yes” answers. The price of the asset is set at the prior expectation of the endorsement rate. We obtain a separating equilibrium, where agents choose to receive a costly private signal and buy or sell the asset as a function of their signal. Trades reflect agents’ true unverifiable information. Two experimental studies test the theoretical results. The first study shows that peer prediction markets motivate agents to seek costly information and reveal it in a simple prediction task. The second study implements an online field experiment to demonstrate feasibility in a natural setting. We find that agents in a peer prediction market are more likely to truthfully self-report socially stigmatized behavior that they can easily deny without being caught