Relative error prediction in nonparametric deconvolution regression model

Baba Thiam (1)

1 LEM - Lille économie management - UMR 9221

In this paper, we studied an alternative estimator of the regression function when the covariates are observed with error. It is based on the minimization of the relative mean squared error. We obtain expressions for its asymptotic bias and variance together with an asymptotic normality result. Our technique is illustrated on simulation studies. Numerical results suggest that the studied estimator can lead to tangible improvements in prediction over the usual kernel deconvolution regression estimator, particularly in the presence of several outliers in the dataset.

Domaines: Sciences de l'Homme et Société; Méthodes et statistiques

Baba Thiam. Relative error prediction in nonparametric deconvolution regression model. Statistica Neerlandica, 2018, ⟨10.1111/stan.12135⟩. ⟨hal-01824274⟩