A dynamic epistemic characterization of backward induction without counterfactuals (Q1941238)

The author provides a new characterization of the backward induction solution in perfect-information extensive form games. First, such games are transformed into Kripke frames where possible worlds are state-instant pairs \((w, t)\). This also allows the formal representation of agents' beliefs in the game. A player is rational at \((w, t)\) if she is either inactive or the action she takes is optimal given her beliefs. With \(\mathbf{T}_t\) being the set of true beliefs of the player acting at time \(t\) and \(\mathbf{R}_t\) the set of state-instant pairs at which the player acting at time \(t\) is rational, the backward induction solution is characterized by the proposition \[ S^\infty := (\mathbf{T}_0 \cap \mathbf{R}_0) \cap B_0 (\mathbf{T}_1 \cap \mathbf{R}_1 ) \cap B_0 B_1 (\mathbf{T}_2 \cap \mathbf{R}_2 ) \cap \cdots \] This characterization differs from existing ones in that it relies only on the active agent's ``pre-choice'' or ``deliberation-stage'' beliefs. No use of counterfactuals is required.