Adaptive elastic-net selection in a quantile model with diverging number of variable groups

TL;DR

This paper introduces an adaptive elastic-net group estimator for quantile models with diverging variable groups, improving parameter estimation accuracy and automatic group selection in high-dimensional settings.

Contribution

It develops a novel adaptive elastic-net approach for quantile models with diverging groups, including theoretical properties and practical algorithms.

Findings

Estimator achieves consistent group selection with probability approaching one.

Non-zero parameter estimators are asymptotically normal.

Simulation results show superior accuracy over existing methods.

Abstract

In real applications of the linear model, the explanatory variables are very often naturally grouped, the most common example being the multivariate variance analysis. In the present paper, a quantile model with structure group is considered, the number of groups can diverge with sample size. We introduce and study the adaptive elastic-net group estimator, for improving the parameter estimation accuracy. This method allows automatic selection, with a probability converging to one, of significant groups and further the non zero parameter estimators are asymptotically normal. The convergence rate of the adaptive elastic-net group quantile estimator is also obtained, rate which depends on the number of groups. In order to put the estimation method into practice, an algorithm based on the subgradient method is proposed and implemented. The Monte Carlo simulations show that the adaptive elastic-net group quantile estimations are more accurate that other existing group estimations in the literature. Moreover, the numerical study confirms the theoretical results and the usefulness of the proposed estimation method.