B0547
Title: A significance test for feature variables through deep neural networks
Authors: Yue Zhao - University of York (United Kingdom) [presenting]
Jianqing Fan - Princeton University (United States)
Weining Wang - University of York (United Kingdom)
Abstract: The focus is on testing the statistical significance of the feature variables in nonparametric regression. The test statistic is constructed as the moment-generating function of the partial derivative with respect to the variable of interest of the estimated regression function. This estimate is constructed through deep neural networks and is smoothed and debiased to ensure proper coverage. To tackle the case of high-dimensional feature variables, we also consider the case in which the feature variables arise from a factor model, and only the lower-dimensional but latent factors have a direct impact on the underlying regression function. The asymptotics of the test statistic are derived under both the null and alternative.