14 - Anticipated Utility and Dynamic Choice (1997)

Summary

Imagine you are about to play a repeated strategic-form game against a randomly selected Stanford MBA student. You may not talk to her before or during the play of the game - interactions are computer-mediated. To the extent that it matters, her exposure to game theory has consisted of two lectures in the autumn quarter of her first year (which prominently featured an informal account of the folk theorem, based on the prisoners' dilemma); subsequently, she has specialized in courses in finance and accounting.

The game is played repeatedly, with a constant termination probability: You play once and learn (to an extent specified below) the outcome of the first round of play. Then a two-digit random number is drawn: If it is 00, the interaction is over; anything else and you play again. Hence after each round there is a 0.01 chance of the interaction ending and a 0.99 chance of it going on for at least one more round.

The stakes per round are on the order of tens of dollars, enough (I hope) so you and she will take the game seriously, but not so large that risk aversion or liquidity constraints play an appreciable role. Your name and hers will be shielded so that reputation effects (outside of this encounter) can be ignored.