Optimal model averaging estimator for semi-functional partially linear models (Q2227201)

The work here is related to the article of \textit{X. Zhang} and \textit{W. Wang} [Stat. Sin. 29, No. 2, 693--718 (2019; Zbl 1422.62245)], where the Mallows-type model averaging criterion for partially regression models is developed. The aim here is on one side to extend the Mallows-type model averaging criterion to semi-functional partially linear models (SFPLM) and on the other side to develop a generalized cross-validation (CGV) model-averaging weight choice criterion to select weights. The authors consider an SFPLM of the form \[ Y_i = \mu _i +\epsilon _i,\; i=1,\dots,n, \] \[ \mu_i= \sum_{j=1} ^{\infty}X_{ij}\beta_j + m(T_i). \] Here $X_i$ are (i.i.d) vectors of countable infinite random real explanatory variables, $T_i$ are (i.i.d) explanatory variables of functional nature, \(m(\cdot)\) an unknown smooth real function and $\epsilon_i$ random errors. The goal is to estimate $\mu_i$. One introduces candidate models to approximate the considered SFPLM, one estimates them and defines the model averaging estimator for $\mathbf{\mu}$. In the second section, the SFPLM Mallows-type weight choice criterion and the GCV model averaging weight choice criterion estimators are introduced. A Monte-Carlo simulation to evaluate finite-sample performances of both estimators is presented in the third section. The fourth section is devoted to the presentation of an application of the developed strategy to an actual data set of air pollution measurements and mortality counts for U.S. cities, year 2000. Proofs are given in the Appendix.