Model selection and estimation of a component in additive regression
Abstract
Let be a random vector with mean and covariance matrix where is some known -matrix. We construct a statistical procedure to estimate as well as under moment condition on or Gaussian hypothesis. Both cases are developed for known or unknown . Our approach is free from any prior assumption on and is based on non-asymptotic model selection methods. Given some linear spaces collection , we consider, for any , the least-squares estimator of in . Considering a penalty function that is not linear in the dimensions of the 's, we select some in order to get an estimator with a quadratic risk as close as possible to the minimal one among the risks of the 's. Non-asymptotic oracle-type inequalities and minimax convergence rates are proved for . A special attention is given to the estimation of a non-parametric component in additive models. Finally, we carry out a simulation study in order to illustrate the performances of our estimators in practice.
Cite
@article{arxiv.1209.6534,
title = {Model selection and estimation of a component in additive regression},
author = {Xavier Gendre},
journal= {arXiv preprint arXiv:1209.6534},
year = {2012}
}