Abstract
Static, simultaneous-move finite games of complete and incomplete information in the presence of multiple equilibria; network formation models; and auction models under weak assumptions on bidding behavior or on bidders’ information, are examples of strategic interaction models that deliver set-valued predictions for the optimal action that each agent can take. In the absence of (typically untestable) assumptions specifying a mechanism to select one prediction out of the set implied by the model, point identification of agents’ preferences based on observing the joint distribution of actions taken (and covariates) cannot be achieved. Rather, there is a set of observationally equivalent preferences that can rationalize the observed data. This talk presents a methodology based on tools from random set theory to characterize the set of observationally equivalent preferences through a collection of moment inequalities, and describes existing approaches for statistical inference based on finite data.