On the non-existence of a rationality definition for extensive games (Q920856)

This paper shows by example that there exists no solution to extensive games that satisfies four axioms. The example is a variant of a game introduced by \textit{R. Rosenthal} [J. Econ. Theory 25, 92-100 (1981; Zbl 0467.90084)]. The analysis is conducted for two player extensive games with perfect information. The axioms place restrictions on the solution concept and an irrationality map defined in terms of the solution. The first axiom requires that the solution concept be a Cartesian product of subsets of the players' sets of strategies. The second axiom requires, roughly, that any player who fails to satisfy backward induction rationality be classified as irrational. The third axiom implies that a player who chooses strategies that lead to higher payoffs must be classified as rational. The fourth axiom states that if a player is considered to be irrational when a fixed node in the game is reached, then the solution permits, whenever possible, strictly more predictions of the player's behavior following the irrational node than from the beginning of the game.