Variable selection in high-dimensional linear model with possibly asymmetric or heavy-tailed errors

TL;DR

This paper introduces an adaptive LASSO expectile method for variable selection in high-dimensional linear models with asymmetric or heavy-tailed errors, addressing challenges of non-differentiability and diverging parameters.

Contribution

It develops a new penalized expectile estimation approach with proven convergence rates and oracle properties for high-dimensional settings, including cases where parameters exceed sample size.

Findings

The estimator achieves oracle properties in high-dimensional models.

Simulation studies show competitive performance compared to quantile-based methods.

Application to genetic data demonstrates practical utility in real-world scenarios.

Abstract

We consider the problem of automatic variable selection in a linear model with asymmetric or heavy-tailed errors when the number of explanatory variables diverges with the sample size. For this high-dimensional model, the penalized least square method is not appropriate and the quantile framework makes the inference more difficult because to the non differentiability of the loss function. We propose and study an estimation method by penalizing the expectile process with an adaptive LASSO penalty. Two cases are considered: the number of model parameters is smaller and afterwards larger than the sample size, the two cases being distinct by the adaptive penalties considered. For each case we give the rate convergence and establish the oracle properties of the adaptive LASSO expectile estimator. The proposed estimators are evaluated through Monte Carlo simulations and compared with the adaptive LASSO quantile estimator. We applied also our estimation method to real data in genetics when the number of parameters is greater than the sample size.