Optimal climate policy when damages are unknown

Integrated assessment models (IAMs) are economists' primary tool for analyzing the optimal carbon tax. Damage functions, which link temperature to economic impacts, have come under fire because of their assumptions that may be incorrect in significant, but a priori unknowable ways. Here I develop recursive IAM frameworks to model uncertainty, learning, and concern for misspecification about damages. I decompose the carbon tax into channels capturing state uncertainty, insurance motives, and precautionary saving. Damage learning improves ex ante welfare by $750 billion. If damage functions are misspecified and omit the potential for catastrophic damages, robust control may be beneficial ex post.