On Bayesian Epistemology of Myerson Auction

Bayesian Epistemology bases its analysis of the objects under study on a prior, a (jointly known) probability distribution, which is in turn the subject matter in statistical learning, and that of machine learning at least implicitly. Our interests are in data related to bidder behavior in a game theoretical setting where the learner and the owners of data to be learned are affected by each other’s strategies of data reporting and data utilization. More specifically, the learner takes actions according to the probability governing the data distributions. The owners of data may reveal false probability distributions of their private data for their better utilities. We focus on the Myerson auction for such an issue of learning and optimization of the seller who wants to gain information on bidders’ value distributions to achieve the maximum revenue. We study bidders’ best responses against each other. We make a special effort to understand the possibility of truthful reporting of agent value distributions. For cases this is not possible, we study the dynamic properties of bidder value distribution reporting and the convergence toward a local Nash equilibria.