SMU SOE Seminar Series (February 20, 2025): Accuracy and Selection in Prediction Markets

TOPIC:

ABSTRACT

The paper investigates under which conditions a market selection argument supports the accuracy of prediction markets. In a prediction market traders can bet on the realization of a binary outcome, such as a presidential election, any other economic or political event, or the result of a sport competition. We assume that risk-averse traders learn about the probability of the binary outcome by observing the same public signals, use heterogeneous models to interpret the signals, and can trade bets without frictions whenever they observe a signal in order to maximize their subjective expected utility at the final outcome date. First, the model is generically consistent with the empirical evidence about volume and momentum in prediction markets. Second, provided agents' utilities satisfy the Inada condition,  as signals become more and more frequent, the price of the bet on an outcome converges to its probability as computed by a Bayesian econometrician who learns the signal generation process by mixing the model used by all the agents. As an implication, a market selection argument supports price accuracy only when signals and outcomes have similar likelihoods, as determined by an underlying unknown state.

Click here to view the CV.