Local Walsh-average-based estimation and variable selection for single-index models (Q2010424)

The single-index model (SIM) is a natural extension of the linear regression model to cope with nonlinear relationships. In the paper under review, the estimation and variable selection problem of SIM are studied by combining local Walsh-average regression (LWR) with the smoothly clipped absolute deviation (SCAD) penalty in SIM. The authors propose an effective estimation procedure to obtain the estimate of the single-index parameter and of an unknown link function in the SIM model. The paper is organized as follows. In Section 2, an adaptive estimation procedure (based on the LWR) to generate the effective estimate for the single-index parameter is proposed. In Section 3, theoretical properties and the asymptotic relative efficiencies of the proposed estimates for the parametric and the nonparametric parts versus those of LS estimates are derived. In Section 4, the authors present a variable selection procedure by combining the proposed estimation approach with SCAD penalty, and the oracle property is established under some regularity conditions. Moreover, to select the tuning parameter, they propose a Bayesian information type criterion, which can consistently identify the true model. In Section 5, numerical simulation examples and a real data analysis are conducted to illustrate the finite sample performance of the proposed methods. A short discussion (about ways in which the proposed method can be extended) follows in Section 6. Proofs of the main results are collected in the Appendix.