Belief convergence and divergence under rationally inattentive learning

Abstract

Conventional Bayesian rationality predicts that heterogeneous beliefs among agents will always converge as agents keep learning. In this paper I show that common learning will fail if agents have limited learning capacity. One agent's constrained optimal learning depends on his prior belief. Namely, one will choose to learn those states which he believes most likely to happen. Hence, heterogeneous beliefs result in heterogeneous learning, even when the natural of information is public. More interestingly, in this model we can obtain belief divergence, which is very hard to explain in traditional frameworks.