88
15

Model selection and estimation of a component in additive regression

Abstract

Let YRnY\in\R^n be a random vector with mean ss and covariance matrix σ2Pn\traPn\sigma^2P_n\tra{P_n} where PnP_n is some known n×nn\times n-matrix. We construct a statistical procedure to estimate ss as well as under moment condition on YY or Gaussian hypothesis. Both cases are developed for known or unknown σ2\sigma^2. Our approach is free from any prior assumption on ss and is based on non-asymptotic model selection methods. Given some linear spaces collection {Sm, m\M}\{S_m,\ m\in\M\}, we consider, for any m\Mm\in\M, the least-squares estimator s^m\hat{s}_m of ss in SmS_m. Considering a penalty function that is not linear in the dimensions of the SmS_m's, we select some m^\M\hat{m}\in\M in order to get an estimator s^m^\hat{s}_{\hat{m}} with a quadratic risk as close as possible to the minimal one among the risks of the s^m\hat{s}_m's. Non-asymptotic oracle-type inequalities and minimax convergence rates are proved for s^m^\hat{s}_{\hat{m}}. 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.

View on arXiv
Comments on this paper