The Bayesian formulation of incomplete information. The non-compact case (Q2366096)

If several decision makers are involved in a situation with incomplete information, these decison makers may hold beliefs not only about the aspects of their environment about which they lack information, but also about the other agents' beliefs, about their beliefs about beliefs, etc. This paper concerns the question how such sequences of beliefs can be described in a mathematically consistent and simple way. The author builds on an approach due to Mertens and Zamir, which in turn formalises an idea of Harsanyi. According to this approach, it is possible to define for each of the agents involved a ``type space'' such that each agent's infinite hierarchy of beliefs can be identified with a simple belief concerning the true state of the environment and the types of the other players. This argument has been formalized for the case that the set of possible states of the agents' environment is a compact Hausdorff space or a Polish space, and agents' beliefs are general Borel probability measures. In the current paper, the argument is extended to the case that the set of possible states of the agents' environment is a general Hausdorff space, and that the agents' beliefs are regular Borel probability measures.