Estimation and test of linearity for a class of additive nonlinear models (Q1305278)

Let us consider an additive separable nonlinear model which has the following form \(Y_n=\sum^d_{i=1}f^i(X^i_n)+\xi_n\), where \(Y\) is the dependent variable and \(X^1,\dots X^d\) are the explanatory variables. This paper deals with the estimation and test of linearity of this model under the following main assumption: the explanatory variables \(Xi\) are supposed to be mutually independent but each of them is allowed to be a dependent sequence. Before focusing on this framework, let us make a few comments about additive modelling. Such nonparametric regression models have been introduced in order to combine the flexibility of nonlinear modelling with the structure of additivity. This gives a compromise between the fully nonparametric model of the form \(Y_n=f(X^1_n,\dots,X^d_n)+\xi_n\) for which the curse of dimensionality occurs and linear parametric models. The paper is organized as follows. In Section 2, we introduce the model and we present the result of joint asymptotic normality for the nonparametric estimators. In Section 3, a test of linearity is proposed and asymptotic results (needed for the test) for parametric estimators are derived. Finally, Section 4 provides some simulation studies and Section 5 contains all the proofs.