BOOTSTRAP-BASED CRITICAL-VALUES FOR THE INFORMATION MATRIX TEST

In finite samples, the true size of White's information matrix test often differs greatly from its nominal size based on asymptotic critical values. This paper shows how the bootstrap can be used to obtain improved finite-sample critical values. The results of Monte Carlo experiments show that for the cases investigated, the bootstrap largely eliminates the problem of incorrect finite-sample size. Moreover, when size-corrected critical values are used, forms of the test that have small size distortions with asymptotic critical values can have much lower power than forms that have large size distortions with asymptotic critical values.