Simulation-based Tests for Non-nested Regression Models

Abstract

It has been argued in previous chapters that the proper use of bootstrap methods can often produce better control of finite sample significance levels than can be obtained from asymptotic theory. Moreover, an appropriate bootstrap approach can sometimes be used to derive a valid large sample test when none is available from standard asymptotic theory. However, all of the previous discussions and recommendations have been based upon the assumption that the model of the null hypothesis is a special case of the model of the alternative hypothesis; it has, therefore, been possible to refer to the former as restricted and the latter as unrestricted.