Robust multiplicity with (transfinitely) vanishing naiveté (Q2280050)

A very interesting development in game theory has been the idea of {\em global games} [\textit{H. Carlsson} and \textit{E. van Damme}, Econometrica 61, No. 5, 989--1018 (1993; Zbl 0794.90083)]. They conveyed the possibility of getting rid, thanks to slight perturbations to the signals received by the players, of the usual multiplicity of equilibria of incomplete information games. Further developments (for instance [\textit{J. Weinstein} and \textit{M. Yildiz}, Econometrica 75, No. 2, 365--400 (2007; Zbl 1132.91324)]) have shown that this result is fairly robust: slight perturbations of the types of players lead to unique rationalizable actions and thus the prediction of a single outcome. Two assumptions are needed to prove this result: \begin{itemize} \item There is \textit{common belief of rationality} among the players. \item There exists common knowledge of the players having \textit{infinite depth of reasoning}. \end{itemize} The last condition, contested once and again in experiments is the focus of interest in this paper of Prof. Heifetz. Here he weakens the assumption of infinite depth of reasoning to something he describes as a \textit{transfinitely ``remote suspicion''} of finite depth of reasoning. In natural language, this would mean that at least one player suspects that a player suspects (\(\ldots\)) that somebody may reason only up to a finite depth. The interesting feature is that \dots is assumed here to cover a transfinite range. The formalization applies here to any ordinal \(\alpha < \omega^{\omega^{\omega ^{\dots}}}\). The main result of the paper is that there exists a threshold level of noise \(\bar{\sigma}\) such that for \(\sigma < \bar{\sigma}\) the rationalizable actions are \textbf{not} unique. This shows how sensitive is the uniqueness result to the knowledge about the reasoning abilities of the other players.