Empirical best linear unbiased prediction in misspecified and improved panel data models with an application to gasoline demand

Misspecifications in econometric models can result in misestimated coefficients. An improved method for specifying econometric models is presented. The mean square error of an empirical best linear unbiased predictor of an individual drawing for the dependent variable of an improved model is derived. These ideas are illustrated using certain misspecified and improved models of the demand for gasoline in the US. It is shown that the forecasting gains from using the improved instead of the misspecified version of the gasoline demand model are very large. A description of a computational algorithm for combining iteratively re-scaled generalized least-squares estimation with out-of-sample multistep-ahead forecast generation is included.