Title: Counterfactual prediction in complete information games: point prediction under partial identification
Speaker: Joris Pinkse, The Pennsylvania State University
Host: Zaichao Du, Professor, RIEM
Time: 14:30-16:00, Jan.9, 2017, Monday
Venue: Gezhi 1211, Liulin Campus
Abstract: We study the problem of counterfactual prediction in discrete decision games with complete information, pure strategies, and Nash equilibria. We show that the presence of multiple equilibria poses unique challenges for the problem of counterfactual prediction even if the payoff structure is known in its entirety. We show that multiple types of counterfactuals can be defined and that the prediction probabilities are not generally pointidentified. We establish the sharp identified bounds of the prediction probabilities. We further propose, compare, and contrast various decision methods for the purpose of producing a point prediction, namely midpoint prediction, a decisiontheoretic possibility using a Dirichletbased prior, and a maximumentropy approach. On balance, we conclude that the maximumentropy approach is the least of several evils. Our results have implications for counterfactual prediction in other models with partial identification.