Uncertainty in the National Energy Modeling System. I: Method Development

This work documents a major part of a large effort to characterize the uncertainty present in key elements of the U.S. Department of Energy's new National Energy Modeling System (NEMS). It is expected that NEMS will be used to estimate the impact that Federal policy initiatives, such as taxes, subsidies, and regulations, will have on energy markets well into the 21st century. But, the effective use of NEMS requires that the uncertainties in its predictions be well understood. So, the Energy Information Administration, the Washington Consulting Group, and the School of Information Technology and Engineering of George Mason University collaborated on the development and testing of methods for quantifying uncertainty in the three most important NEMS submodel types, namely, linear-optimization, econometric, and heuristic or balance equations. We constructed assessment procedures for each of these classes, and used them on the Energy Information Administration transportation sector demand model, petroleum market model, and electricity fuel dispatch model. The methods used include special sampling procedures applied to computer-based experiments, response surface techniques, and Taylor series methods, and they are described in this first of a two-part paper on the complete study. The important numerical results of our experiments and their subsequent analyses are presented in the companion paper.