Title: A consistent gradient-based nonparametric test for regression structures
Authors: Taining Wang - Capital University of Economics and Business (China) [presenting]
Feng Yao - West Virginia University (United States)
Abstract: A consistent nonparametric test is considered for the relevant variables in the gradient function of the regression model, which can be used to detect the interaction among regressors and nonlinearity of a single regressor in a nonparametric regression. Our test statistics are based on the first-order gradient obtained by local quadratic estimation and we obtain its empirical distribution via bootstrap. Regarding the contribution in empirical studies, we show that it can be applied to identify regression structures, including additive, varying coefficient, and partially linear models, thereby providing statistical evidence for which semiparametric structure should be implemented in practice.