Tilman Börgers, University of Michigan

"Similarity-Based Learning and Similarity Equilibria"

Abstract

We consider non-cooperative games in which a boundedly rational learning process leads to equilibrium play, and, moreover, learning is based on recollecting past situations that were similar, but not identical to the current situation. We begin by introducing an equilibrium notion called "similarity equilibrium" that captures the basic idea. We relate this equilibrium notion to more standard equilibrium notions in games. We then formulate a learning model, and show that its rest points are similarity equilibria. Finally, we study whether the learning process provides underpinnings for a theory of selection among multiple similarity equilibria.

Contact person: Peter Norman Sørensen