What is the Economic Value of Information about Uncertain Climate Thresholds? or How a Parallel Stochastic Optimization Algorithm Solves a Nonconvex, Non-di erentiable, and Global Problem BY ADAM SCOTT LIEBER

We consider the economic optimal control problem of emissions abatement with uncertain climate thresholds. The basis of study is Nordhaus' (1994) DICE model augmented for parameter uncertainty, uncertain climate thresholds [Keller et al. 2000], and parameter learning. The method of optimization is an improved version of Storn & Price's (1995) Di erential Evolution. The algorithm is modi ed to be parallel in execution and to incorporate line-scan polishing techniques. The results demonstrate that optimal policy is highly sensitive to parameter speci cation and the date at which uncertainty is resolved. Depending on the model used, decisions during times of uncertainty can call for more or less abatement than if the policy-maker had perfect information. As the model is augmented with climate thresholds, and di ering from Nordhaus & Popp (1997), the value of learning information about climate systems surpasses that of learning about economic parameters.