scientific article; zbMATH DE number 1124619

From MaRDI portal
Publication:4378648





The aim of this paper is to give consistency results for \(L_1\)-norm estimators in semiparametric regression models, where the parameter of interest is of fixed dimension and the nuisance parameter is either given as an unknown function from a nonparametric function space or has a dimension which increases with the sample size. We study simultaneously three models, namely nonlinear regression, nonlinear functional relation, nonlinear semiparametric regression, all to be introduced in Section 2. This allows us to demonstrate the line of proving the \(L_1\)-norm consistency results and to emphasize the underlying problems. For illustration reasons the \(L_1\)-approach is also embedded in the minimum contrast context. The proof consists mainly of two steps given in Section 4: first the approximation of the empirical \(L_1\)-norm by its expected value, which will be done in Subsection 4.1, and second the identification of the parameter by using the expected values of the difference of empirical \(L_1\)-norms in Subsection 4.2. The identification problem is specific for the individual models. This is not characteristic for the \(L_1\)-norm approach, since the same problem also occurs in the \(L_2\)-theory. The technique of approximation is more or less standard and is based on results of the increments of sub-Gaussian processes. In Section 5 the consistency results for each model are separately given. For the nonlinear regression model this is a known strong consistency result in form of an exponential probability inequality. The consistency result for the nonlinear functional relation models is new. Under this kind of entropy condition on the nuisance parameter space only results are known for the least squares estimator. The nonlinear semiparametric model is of a special structure, because of the constraints imposed by the identification problem. \textit{O. Linton} [Econometrika 63, No. 5, 1079-1112 (1995; Zbl 0836.62050)] studied this model with a linear parametric part. The result given here for this model seems to be new as well. For the entire collection see [Zbl 0882.00044].











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4378648)