Calendar
DR@W Forum: Erik Mohlin (University of Oxford)
Erik Mohlin (University of Oxford)
Learning by Imitation in Games: Theory, Field and Lab
We study how players learn to play the lowest unique positive integer (LUPI) game. Although the game is novel, and the completely mixed unique symmetric equilibrium is difficult to compute, people quickly learn to play close to it both in the field and laboratory. We argue that a simple model of global cumulative imitation (GCI) of previous winning numbers, and numbers that are close to winning numbers, can explain the data, while standard models of belief-based learning and reinforcement learning cannot satisfactorily account for the quick movement toward equilibrium. By means of stochastic approximation we show that GCI asymptotically approaches the replicator dynamic in LUPI.