DEMAND SYSTEMS FOR ENERGY FORECASTING: PRACTICAL CONSIDERATIONS FOR ESTIMATING A GENERALIZED LOGIT MODEL

Generalized Logit models of demand systems for energy and other factors have been shown to work well in comparison with other popular models, such as the Almost Ideal Demand System and the TransLog model. The main reason is that the derived price elasticities are robust when expenditure shares are small, as they are for electricity and fuels. A number of different versions of the Generalized Logit model have been applied in the literature, and the primary objective of the paper is to determine which one is the best. Using annual data for energy demand in the USA at the state level, the final model selected is similar to a simple form that was originally proposed by Considine. A second objective of the paper is to demonstrate that the estimated elasticities are sensitive to the units specified for prices, and to show how price scales should be estimated as part of the model.