Estimating energy savings model for an energy-efficiency program: A semiparametric analysis

This dissertation work undertakes a serniparametric analysis of an empirical model in the field of energy-efficiency program evaluation. This empirical model falls into the widely known class of 'single index' selectivity models. Using a data set for. an energy-efficiency program for residential customers, energy impacts that are attributable to the program are estimated using a serniparametric analogue of a popular 'two-step' method for such models. Given the dissatisfaction pointed out in literature about this popular strategy, various semiparametric approaches have been proposed in literature. However, there is a dearth of empirical applications to put these approaches to test. This dissertation has found one new application where semiparametric techniques proposed in theory can be used successfully. It is shown that these techniques are workable tools for the energy-efficiency program evaluation. The analysis shows that given a simple energy savings model specification, serniparametric model proves to be the correct model specification versus the popular parametric model mostly used in such applications to correct for self-selectivity. The semiparametric model estimates statistically significant net energy savings for the residential space conditioning rebate program versus the model estimates of the traditional 'two-step' approach. The estimated savings are determined to be about 78% of the ex ante engineering estimate of the first-year impact savings and hence could be regarded quite credible based on observations on past evaluation estimates for such kinds of programs.