There is growing appreciation of the importance of low-probability, high-consequence climate change outcomes given the speed and scale at which greenhouse gas (GHG) emissions are increasing. Deep uncertainty in equilibrium climate sensitivity (ECS), the eventual temperature increase in response to changes in GHG concentrations, has persistently been a critical issue in determining an effective yet feasible hedging strategy against such tail events. In this paper, we examine how learning efforts to reduce uncertainty in ECS affects the optimal policy when the downside of climate change is indeed catastrophic. Using stochastic dynamic programming, we investigate the role of learning under a “triple worst-case” scenario, in which climate damages are catastrophic at high temperatures; the true ECS, unknown to the world, exceeds the range of current scientific beliefs; and negative emissions technologies do not materialize at scale for deployment. Under these assumptions, we find that learning primarily contributes to welfare improvements through corrections in short-run emission abatement and long-run capital investment decisions. In the short-run, a more stringent carbon abatement policy accelerates the transition to a zero-carbon economy. In the long-run, increased capital investment in anticipation of higher climate damages increases economic output and smoothens consumption. Strong precautionary measures are often infeasible due to political constraints. Our results imply that modest carbon abatement policies coupled with efforts to reduce climate uncertainty can be a feasible and effective mitigation strategy.

We present a mixed complementarity problem (MCP) formulation of continuous state dynamic programming problems (DP-MCP). We write the solution to projection methods in value function iteration (VFI) as a joint set of optimality conditions that characterize maximization of the Bellman equation; and approximation of the value function. The MCP approach replaces the iterative component of projection based VFI with a one-shot solution to a square system of complementary conditions. We provide three numerical examples to illustrate our approach.

How does risk and uncertainty in climate thresholds impact optimal short-run mitigation? This paper contrasts the near-term mitigation consequences of using an expected value, stochastic programming, and stochastic control model to capture the policy effects of uncertain climate thresholds. The risk of threshold outcomes increases expected climate damages. The passive learning associated with stochastic programming creates an extra incentive to mitigate promptly by reducing the damages from remaining threshold hazards. The active learning associated with stochastic control creates yet another incentive to do near-term mitigation, through the delaying of potential threshold effects.