Group Lasso for high dimensional sparse quantile regression models

TL;DR

This paper analyzes the statistical properties of the group Lasso estimator in high-dimensional sparse quantile regression models, providing bounds, practical tuning methods, and demonstrating near-oracle convergence rates.

Contribution

It extends the analysis of group Lasso to high-dimensional quantile regression, proposes a data-driven tuning parameter, and applies the method to additive models with theoretical guarantees.

Findings

Non-asymptotic $\, ext{l}_2$-error bounds established.

Data-dependent tuning parameter improves practicality.

Simulation results support theoretical claims.

Abstract

This paper studies the statistical properties of the group Lasso estimator for high dimensional sparse quantile regression models where the number of explanatory variables (or the number of groups of explanatory variables) is possibly much larger than the sample size while the number of variables in "active" groups is sufficiently small. We establish a non-asymptotic bound on the $\ell_{2}$-estimation error of the estimator. This bound explains situations under which the group Lasso estimator is potentially superior/inferior to the $\ell_{1}$-penalized quantile regression estimator in terms of the estimation error. We also propose a data-dependent choice of the tuning parameter to make the method more practical, by extending the original proposal of Belloni and Chernozhukov (2011) for the $\ell_{1}$-penalized quantile regression estimator. As an application, we analyze high dimensional additive quantile regression models. We show that under a set of suitable regularity conditions, the group Lasso estimator can attain the convergence rate arbitrarily close to the oracle rate. Finally, we conduct simulations experiments to examine our theoretical results.