Title: Backfitting tests in generalised structured models
Authors: Stefan Sperlich - University of Geneva (Switzerland) [presenting]
Enno Mammen - Heidelberg University (Germany)
Abstract: Bootstrap tests for generalised structured models are introduced. They can be applied for testing interaction terms or the impact of certain covariates in a large family of semiparametric models. This implies, for example, additivity testing, variable selection, linearity, constant returns or checking for endogeneity. The test is based on a comparison of non- and semiparametric alternatives, i.e both the null hypothesis and the alternative are non- or semiparametric. The test statistic is a weighted L2-distance of the two fits, where both estimation procedures make use of the smooth backfitting technique. Asymptotic theory for the test is developed, implementation and computational issues are discussed. Simulations and comparison studies show an excellent performance of the test procedures. We apply the method to additive and varying coefficient models.