Title: An iterative approach for model selection in single-index varying coefficient models
Authors: Efang Kong - University of Electronic Science and Technology of China (China) [presenting]
Abstract: The penalised least squares estimation based model selection has many advantages over the traditional ones. Much literature has been devoted to this area in recent years. The single-index varying coefficient models (SIVC) have proved to be a class of very useful models in data analysis. Model selection in such class is important but challenging due to the complicated structure of SIVC. We take on this challenge and develop an iterative approach for model selection in SIVC. The proposed approach is easy to implement as there is a closed form for estimators obtained in each step. Asymptotic properties of the proposed iterative approach are also established, which provide theoretical justification for the proposed approach. Comprehensive simulation studies show that the proposed iterative approach works very well even with modest sample size. Finally, we apply the SIVC and the proposed model selection method to an environmental set from Hong Kong and the Boston housing dataset from Boston, which lead to some interesting findings.