Predicting one Shot Play in 2x2 Games Using Beliefs Based on Minimax Regret

Abstract

We present a simple procedure that selects the strategies most likely to be played by inexperienced agents who interact in one shot 2x2 matching pennies games. As a first step we axiomatically describe players' beliefs. We find the minimax regret criterion to be the simplest functional form that satisfies all the axioms. Then we hypothesize players act as if they were best responding to the belief their opponent plays accordino to minimax regret. When compared with existing experimental evidences about one shot matching pennies games, the procedure correctly indicates the choices of around 80% of the players. Applications to other classes of games are also explored