Bayesian Inference in Partially Identified Models

• Activity Number: 337
• Type: Topic Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Paul Gustafson*+
• Companies: The University of British Columbia
• Keywords: asymptotics ; Bayesian inference ; nonidentified model

Abstract

Partially identified models involve the large-sample limiting posterior distribution over the target parameter having support (or identification region) bigger than a single point but smaller than the corresponding prior support. This talk will review the determination of the limiting posterior distribution in such contexts, paying particular attention to the shape of the distribution. This provides a point of departure with non-Bayesian approaches that focus on inferring the identification region without attempting to speak to relative plausibilities of values inside the region. The utility of the shape of the limiting distribution is investigated in several partially identified models.