Assessing regional econometric models: A discussion and application

Abstract

Economic literature shows it is possible to conduct reasonably complete model assessment tests on small-area regional econometric models, even though several highly informative tests are not commonly reported. The paper demonstrates and encourages the use of Theil-type “U” statistics. These embody implicit naive models to serve as standards of accuracy, facilitate cross-model comparisons, and further facilitate model testing and adjustment. Tests were run on the MAP econometric model of Alaska, widely used for assessment of economic policies in that state. Root mean square percent error, variations of Theil's U statistic, and experiments with the model's block structure were used to test model performance. Use of the family of U-type statistics demonstrated certain structural problems exist in the A84.2 version of MAP, even though this model generally passed the commonly reported tests in the regional literature. For the most part, MAP also passed the additional historical accuracy and impact assessment tests used in this paper; however, areas for potential improvement were found.