Robust Estimation and Shrinkage in Ultrahigh Dimensional Expectile Regression with Heavy Tails and Variance Heterogeneity

TL;DR

This paper introduces a robust expectile regression method tailored for ultrahigh dimensional, heavy-tailed, and heterogeneous data, combining asymmetric loss, divergence-tuned parameters, and folded concave penalties for improved estimation and model selection.

Contribution

It develops a novel robust expectile regression framework with divergence-tuned parameters and folded concave penalties, addressing heavy tails and heterogeneity in high-dimensional data.

Findings

Method performs well in coefficient estimation.

Effective in model selection and heterogeneity detection.

Demonstrates robustness across various distributions.

Abstract

High-dimensional data subject to heavy-tailed phenomena and heterogeneity are commonly encountered in various scientific fields and bring new challenges to the classical statistical methods. In this paper, we combine the asymmetric square loss and huber-type robust technique to develop the robust expectile regression for ultrahigh dimensional heavy-tailed heterogeneous data. Different from the classical huber method, we introduce two different tuning parameters on both sides to account for possibly asymmetry and allow them to diverge to reduce bias induced by the robust approximation. In the regularized framework, we adopt the generally folded concave penalty function like the SCAD or MCP penalty for the seek of bias reduction. We investigate the finite sample property of the corresponding estimator and figure out how our method plays its role to trades off the estimation accuracy against the heavy-tailed distribution. Also, noting that the robust asymmetric loss function is everywhere differentiable, based on our theoretical study, we propose an efficient first-order optimization algorithm after locally linear approximation of the non-convex problem. Simulation studies under various distributions demonstrates the satisfactory performances of our method in coefficient estimation, model selection and heterogeneity detection.