Robust prediction in games with parameter uncertainty

Abstract

In empirical studies about strategic interactions, often it is assumed that all the parameters of interest are commonly known by the players of the interactions (and only the analyst does not know them). However, in some cases, such a ``complete information'' assumption may not seem so reasonable. For example, if one of the parameters is about the effect of a new policy in an oligopoly market, then it may be more reasonable to assume that not only the analyst but also the firms may face similar uncertainty. More importantly, in such a case, it is not trivial how to model the firms' knowledge/beliefs about each other's information, a crucial element to make a sensible prediction. In this project, I propose a model of incomplete information about the parameters (more formally, a ``type space'' a la Harsanyi), which enables us to obtain a robust prediction in a certain sense, even if the analyst's knowledge of the true information structure is limited.