Energy modeling, numerical modeling based on economic principles, has become the dominant analytical tool in U.S. energy policy. Energy models are widely used by researchers across the public and private sectors. However, the widespread application of these models in policy analysis poses challenges to decision-makers. We are developing a framework and analysis that demonstrate how non-Bayesian decision rules can address fundamental model uncertainty in the domain of energy policy, technological change, and greenhouse gas abatement.

Numerical modeling based on economic principles has become the dominant analytical tool in U.S. energy policy. Energy models are now used extensively by public agencies, private entities, and academic researchers, and in recent years have also formed the core of integrated assessment models used to analyze the relationships among the energy system, the economy, and the global climate. However, fundamental uncertainties are intrinsic in what has become the typical circumstance of multiple models embodying different representations of the energy-economy, and producing different policy-relevant outputs that model users are compelled to interpret as equally plausible and/or valid. Because the policy implications of these outputs can diverge substantially, policy-makers are confronted with a significant degree of model-based uncertainty and little or no guidance as to how it should be addressed.

This problem of “model uncertainty” has recently been the focus of work in macroeconomics, where scholars have studied the problem of how a decision-maker should proceed in the face of uncertainty re- garding the correct model of an economic system that is the object of policy. We focus on analyzing a low-dimensional numerical integrated assessment model using the “minimax regret” metric. Specifically, we have demonstrated that deep uncertainty regarding energy-related technological change can be addressed using this approach. Our findings include comparison with expected cost minimization, to show how the interaction of solution methods and model structure affect the influence of this form of deep uncertainty on low-run CO2 emissions abatement policies. We also examined other methods assuming some prior distribu- tions over uncertain parameters for analyzing the difference between our robust solution and the non-robust solution from those methods.

We demonstrate that the fundamental model uncertainty can be represented and analyzed in the context of energy policy problems determining optimal CO2 abatement strategies. The robust solution from min-max regret method is significantly different with any solutions from sensitivity analysis over uncertain parameters or those methods assuming prior distributions over uncertain parameters. The following figure shows the difference of the robust min-max regret solution over all three uncertain parameters (the red line) and others with min-max regret solution over only one uncertain parameter, Technical Change level, while the other two parameters are used for sensitivity analysis.