Causal Coherence ∗ Job Market Paper

Agents with the same information and same preferences can make different choices. Agents differ not only with respect to their preferences and information, but their causal interpretations of that information. This can lead to what agents with the correct causal model would perceive as “irrational mistakes” committed by others. I apply an axiomatic representation to develop the causally coherent agent, who has a causal model about a causally ambiguous phenomenon that is consistent with data, makes choices rationally, but is unaware of alternative models. In essence, her model is not identified so she hazards a guess. The causal model is a causal bayesian network. In this framework, I show how agents with the same information and the same preferences will make different choices. Moreover, with this framework, I can construct a set of reasonable theories that emerge from data the agents see. This provides a framework for constructing agents’ conjectures in a general setting. I apply this framework to an auction to show that agents with wrong models suffer a ‘causal curse’ similar in kind to the winner’s curse. ∗I would like to thank Daniel Benjamin, Joshua Cherry, Peter DiCola, Michael Elsby, Robert Gazzale, David Greenstreet, Patrick Kline, Greg Lewis, Dmitry Lubensky, Yusufcan Masatlioglu, Daisuke Nakajima, Emily Shelton, Doug Smith, Charles Taragin, Seminar participants at Michigan, and especially Emre Ozdenoren, Scott Page, and John DiNardo for helpful comments. †University of Michigan, Department of Economics