The Vuong test for nonnested models

Comparing nonnested models is challenging, especially in duration data, where there are not many available methods. This presentation discusses a transformed Vuong's test, which is applicable to models for duration data, in particular to hazard models that are not directly comparable. The proposed method adapts the Vuong's likelihood-ratio closeness test to compare parametric, semiparametric, and discrete hazard models. The test is built in Stata and shows good performance on ranking different nonnested hazard models. An empirical example presents the performances of different hazard models used to model the duration of shipbuilding firms. The results suggest that models that model the unobserved heterogeneity using finite mixtures perform better than other hazard models that model unobserved heterogeneity using different parameterizations.