Finite foresight in Chomp

We experimentally study sequential rationality in the Game of Chomp, a two-player win-lose game whose rules are simple but whose subgame-perfect Nash equilibria vary in complexity across initial states. Subjects frequently deviate from SPNE: the probability of making an SPNE-consistent move falls with the distance to the end of the game, and most can correctly implement SPNE only three to five moves ahead. Experience mainly improves play at intermediate distances. We also document a common opening heuristic that, although often theoretically suboptimal, is still associated with higher chances of winning.