CHOICE OF REGRESSORS IN NONPARAMETRIC-ESTIMATION

It is well-known that pure nonparametric techniques perform less and less well when the dimension of the regression function increases, because of the sparsness of the data. We attack this curse of dimensionality by proposing a method to select a set of regressors. Optimality with respect to quadratic loss function is shown. Then, simulated examples are presented to illustrate both curse of dimensionality and behaviour of the variables selection procedure.