Quantile regression and variable selection of single-index coefficient model

From MaRDI portal
Publication:2409394





The single-index coefficient model \[ Y=g(X^{\top}\theta)^{\top}Z+\epsilon \] is studied, where \(Y\) is the response variable, \(X\) and \(Z\) are two covariates random vectors, \(\epsilon\) is the model error, \(g\) is an unknown coefficient function vector, and \(\theta \in \mathbb{R}^p\) is the index parameter. For identifiability, \(\|\theta\| =1\) and \(\theta_1>0\) are assumed. A minimizing check loss estimation procedure (MACLE) is proposed in the framework of quantile regression. The estimators are asymptotically normal and achieve the best convergence rate. A variable selection method is investigated by combining the MACLE method with the adaptive Lasso penalty, and the corresponding oracle property is established. Two simulations are conducted with different error settings to assess the finite sample performance of the proposed method. A real-data application is presented.











This page was built for publication: Quantile regression and variable selection of single-index coefficient model

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2409394)