Behavioral decision studies of individuals who make repeated decisions with feedback reveal robust evidence that people behave as if they believe “it won’t happen to me”, a phenomenon coined underweighting of rare events. We experimentally show that the tendency to underweight rare events persists in 2-person repeated games with stochastic payoffs, and that other agents can learn to exploit it. In a simultaneous-move asymmetric 2×2 game, most row players consistently choose a stochastically dominated action that provides a better payoff most of the time but on average leads to a large loss over the equilibrium prediction, behavior consistent with underweighting of the rare event. In response, most column players learn to exploit this bias by choosing a strictly dominated action that is worse for the row players on average but better most of the time. That is, due to underweighting of rare events by the row players, most dyads converge to a profile of two dominated strategies. A second study rules out different explanations like boredom, altruism or risk seeking.
Joint work with Yefim Roth (University of Haifa)