R robustified additive nonparametric regression

Additive modelling is known to be useful for multivariate nonparametric regression as it reduces the complexity of problem to the level of univariate regression. This usefulness could be compromised if the data set was contaminated by outliers whose detection and removal are particularly difficult to achieve in high dimension. We propose an estimation procedure for the additive component of the regression function , less sensitive to possible outliers in the sample. Our procedure is based on marginal integration of conditional R-estimators. In addition to univariate rate of convergence and asymptotic distribution, we also obtain robustness results for our estimator. All of our results are valid for a broad class of ß mixing processes. Monte Carlo findings confirm the theoretical results in finite sample.