Out-of-Sample Bootstrap Tests for Non-Nested Models

When testing non-nested models, the asymptotic distribution theory of the ordinary likelihood ratio statistic is not valid anymore. Several test statistics, some of them based on information criteria, have been proposed in order to test such non-nested hypotheses. Concerning bootstrap approaches to simulate goodness-of-fit measures such as the likelihood ratio value, have been elaborated as well. Based on these methods, we extend existing bootstrap simulations towards out-of-sample bootstrap evaluation. As an application, a parametric bootstrap on simulated regression data is provided.