Thinking Through Categories

(Research Seminar, March 13, 2003)

Sedhil Mullainathan

Department of Economics,

Massachusetts Institute of Technology

Abstract

I present a model of human inference in which people use coarse categories to make inferences. Coarseness means that rather than updating continuously as suggested by the Bayesian ideal, people update change categories only when they see enough data to suggest that an alternative category better fits the data. This simple model of inference generates a set of predictions about behavior. I apply this framework to produce a simple model of financial markets, where it produces straightforward and testable predictions about returns predictability, comovement and volume.