Betting with the planet: Uncertainty and global warming policy

This dissertation presents a theoretical, analytical, and methodological examination of the implications of extensive scientific and economic uncertainty for global warming policy. The purpose is to seek a method to best inform decision-makers in a dynamic systems/policy analysis environment where critical tension exists between the comprehensive representation of the decision and the uncertainty aspects of the issue. To accomplish this purpose, the theoretical importance of the representation of uncertainty in decision problems is investigated, then William Nordhaus's prominent climate/economy model (DICE) is modified to explore the relationship between parametric uncertainty representation and preferred policy, and finally an alternative non-parametric uncertainty assessment methodology is developed for decision systems with feedback.;The dissertation pursues a comprehensive treatment of parametric uncertainty within the DICE model. Since the curse of dimensionality prevents a full treatment of parametric uncertainty, the decision aspect is limited to choice among emissions scenarios while the uncertainty representation aspect is simultaneously expanded in a manner entirely consistent with the model's parameter probability assumptions. Monte Carlo experiment results reveal that more restrictive emissions policies are preferred to the standard DICE model "optimal" policy under more comprehensive uncertainty representations. An additional extension to include the possibility of systemic uncertainty through the incorporation of probabilistically neutral feedbacks strengthens this result. Finally, an examination of the impact of objective function "preference" parameters reveals that the coefficient of relative risk aversion used in the DICE model fails to meaningfully express aversion to long-term trend uncertainty in the model.;Even though extensive parametric uncertainty analysis can reveal critical variable relationships, the entirety of uncertainty is not represented by model parameters; much uncertainty remains within the description of causal relationships by simplified functional forms. Therefore, the final part of the dissertation develops a general non-parametric structure of a scenario-based uncertainty assessment methodology. As a model developed for policy input, it is constructed to incorporate the best available uncertain knowledge across many disciplines into a decision analysis formulation modified to incorporate the systems concept of feedback. The proposed model is nonparametric in the sense that no parametric mathematical form is supplied to express cause and effect between any two variables; only the mathematical form of conditional probability, commonly used in decision analysis to express the logical concept of causation, is imposed by the structure of the model. Finally, the computational mathematics of probabilistic feedbacks are developed yielding the composite probability distribution for a system with an uncertain input, an uncertain transfer function to the output, and an uncertain feedback from the output to the input.