A robust, adaptive M-estimator for pointwise estimation in heteroscedastic regression (Q396020)

The authors introduce a robust method for pointwise estimation in heteroscedastic regression. The method is adaptive with respect to the noise and design distribution (D-adaptive) and the smoothness of the regression function (S-adaptive). The noise and design distributions considered are unknown and fulfill weak assumptions -- moment conditions on the noise distribution are not imposed and positive density for the design distribution is not required. In a first step, the authors derive general properties of M-estimators such as pointwise risk bounds, including S-minimax results for degenerate designs over unidimensional Hölder spaces. After, they choose the contrast and the kernel of the estimators that minimize an empirical variance term and demonstrate that the corresponding M-estimator is adaptive with respect to the noise and design distributions and adaptive Huber minimax for contamination models. Choosing a data-driven bandwidth via Lepski's method leads to an M-estimator that is adaptive with respect to the noise and design distributions as well as with respect to the smoothness of an isotropic, multivariate, locally polynomial target function. The results are also extended to anisotropic, locally constant target functions. This approach provides also a level of robustness that adapt to the noise, contamination, and outliers.