Bootstrap Tests Based on Goodness-of-Fit Measures for Nonnested Hypotheses in Regression Models

This paper utilizes the bootstrap to construct tests using the measures for goodness-of-fit for nonnested regression models. The bootstrap enables us to compute the statistical significance of the differences in the measures and to formally test on nonnested regression models. The bootstrap tests that this paper proposes are expected to show better finite sample properties since they do not have accumulated errors in the computation process. Moreover, the bootstrap tests remove the possibility of inconsistent test results that the previous tests suffer from. Because the bootstrap tests only evaluate if a model has a significantly higher explanatory power than the other model, there is no possibility for inconsistent results. This study presents Monte Carlo simulation results to compare the finite sample properties of the proposed tests with the previous tests such as Cox test and J-test.