Learning and climate feedbacks: Optimal climate insurance and fat tails

We study the effect of potentially severe climate change on optimal climate change policy, accounting for learning and uncertainty in the climate system. In particular, we test how fat upper tailed uncertainty over the temperature change from a doubling of greenhouse gases (the climate sensitivity), affects economic growth and emissions policy. In addition, we examine whether and how fast uncertainties could be diminished through Bayesian learning. Our results indicate that while overall learning is slow, the mass of the fat tail diminishes quickly, since observations near the mean provide evidence against fat tails. We denote as "tail learning" the case where the planner rejects high values of the climate sensitivity with high confidence, even though significant uncertainty remains. Fat tailed uncertainty without learning reduces current emissions by 38% relative to certainty, indicating significant climate insurance, or paying to limit emissions today to reduce the risk of very high temperature changes, is optimal. However, learning reduces climate insurance by about 50%. The optimal abatement policy is strongly influenced by the current state of knowledge, even though greenhouse gas (GHG) emissions are difficult to reverse. Once the mass of the fat tail diminishes, the remaining uncertainty is largely irrelevant for optimal emissions policy.