Empirical Best Linear Unbiased Prediction in Misspecified and Improved Panel Data Models with an Application to Gasoline Demand

We emphasize using our solutions to the problems of omitted variables, measurement errors, and unknown functional forms to improve model specification, and to estimate the mean square error of an empirical best linear unbiased predictor of an individual drawing of the dependent variable of an improved model. We illustrate using data to compare the forecasting performances of misspecified and improved models of the U.S. gasoline market. The performance criterion used is the tightness of the distribution of the absolute relative errors in out-of-sample multi-step-ahead forecasts around zero. The results show that significant improvements in forecasting accuracy can be obtained by improving the specifications of misspecified models. Numerical algorithms for generating forecasts from a rolling forecast method are presented