Optimal climate policy with fat-tailed uncertainty : what the models can tell us

We present a modification of the most commonly used integrated assessment model (IAM) of climate change (DICE-2016), AD-DICE2016, which is designed to address three key aspects of climateeconomy models: treatment of uncertainty, the use of more appropriate utility functions, and including adaptation policies to climate change. These modifications ensure that two of the key difficulties identified with IAMs, the choice of the risk aversion parameter and the underestimation of damages, are also directly addressed. The use of a bounded (Burr) utility function ensures that the model is able to appropriately assess the effects of parameters whose distributions have "fat tails". Uncertainty is accommodated via the state-contingent approach enabling us to include more state (seven) and control variables (four) than recursive derivatives of DICE. Our approach to uncertainty ensures that the optimal climate policies account for outcomes in every possible state, unlike the Monte Carlo approach. Our treatment of uncertainty is extensive: eight parameters are allowed to be random, with distributions - many "fat tailed"- identified using current knowledge. Our model suggests that uncertainty regarding damages and climate sensitivity are key drivers of climate policy. We also find that uncertainty leads to increases in both optimal mitigation and adaptation, with adaptation and mitigation reacting differently to uncertainty over different parameters. Finally, our estimates of the social cost of carbon are larger when uncertainty is allowed for and significantly affected by adaptation.