摘要：This paper illustrates the usefulness of resampling-based methods in the context
of multiple (simultaneous) tests, with emphasis on econometric applications. Economic
theory often suggests joint (or simultaneous) hypotheses on econometric models;
consequently, the problem of evaluating joint rejection probabilities arises frequently in
econometrics and statistics. In this regard, it is well known that ignoring the joint nature of
multiple hypotheses may lead to serious test size distortions. Whereas most available multiple
test techniques are conservative in the presence of non-independent statistics, our
proposed tests provably achieve size control. Specifi cally, we use the Monte-Carlo (MC)
test technique to extend several well known combination methods to the non-independent
statistics contexts. We fi rst cast the multiple test problem into a unifi ed statistical framework
which: (i) serves to show how exact global size control is achieved through the MC
test method, and (ii) yields a number of superior tests previously not considered. Secondly,
we provide a review of relevant available results. Finally, we illustrate the applicability of
our proposed procedure to the problem of moments-based normality tests. For this problem,
we propose an exact variant of Kiefer and Salmon’s (1983) test, and an alternative
combination method which exploits the well known Fisher-Pearson procedure. Our simulation
study reveals that the latter method seems to correct for the problem of test biases
against platikurtic alternatives. In general, our results show that concrete and non-spurious
power gains (over standard combination methods) can be achieved through our multiple
Monte-Carlo test approach.
