Breaking the Curse of Dimensionality in Nonparametric Testing

For tests based on nonparametric methods, power crucially depends on the dimension of theconditioning variables, and specifically decreases with this dimension. This is known as the"curse of dimensionality." We propose a new general approach to nonparametric testing inhigh dimensional settings and we show how to implement it when testing for a parametricregression. The resulting test behaves against directional local alternatives almost as if thedimension of the regressors was one. It is also almost optimal against classes of onedimensionalalternatives for a suitable choice of the smoothing parameter. A simulationstudy shows that it outperforms the standard test by Zheng (1996).