IMPROVED ESTIMATORS OF ENERGY MODELS

It has been suggested that Stein-rule estimators of econometric models can be used to improve inferences in those models by combining sample information with uncertain non-sample information in the estimation p rocess. Pretest estimators are often used for model specification, eve n though it is well known that they possess poor sampling properties. The Stein-rule contains the same elements used by the classical pretes t estimator but performs much better than the pretest estimator in rep eated sampling. Based on Monte Carlo experiments the Stein-rule estima tor is shown to offer significant quadratic risk improvements over max imum likelihood, restricted maximum likelihood, and pretest estimators of energy models.