Files

Action Filename Size Access Description License
Show more files...

Abstract

A variety of asymptotically valid tests for orthogonality, serial correlation, predictive failure, and of coefficient restrictions are presented, and their rejection probabilities are assessed in linear structural models with lagged-dependent and (possibly) jointly dependent variables by Monte Carlo methods. For all test procedures the small-sample distribution under the null usually deviates substantially from the asymptotic distribution; this impedes their use in a reliable model selection strategy for econometric time-series analysis. Despite the harassing dependence of type I errors on factors generally unknown to the practitioner, inconsistencies originating from specification errors or from disregarded simultaneity may be detected by particular tests in particular situations. From this study some clues emerge on how to interpret (in)significant values of the various test statistics.

Details

Downloads Statistics

from
to
Download Full History