Sparse Quantile Regression
Abstract
We consider both -penalized and -constrained quantile regression estimators. For the -penalized estimator, we derive an exponential inequality on the tail probability of excess quantile prediction risk and apply it to obtain non-asymptotic upper bounds on the mean-square parameter and regression function estimation errors. We also derive analogous results for the -constrained estimator. The resulting rates of convergence are nearly minimax-optimal and the same as those for -penalized and non-convex penalized estimators. Further, we characterize expected Hamming loss for the -penalized estimator. We implement the proposed procedure via mixed integer linear programming and also a more scalable first-order approximation algorithm. We illustrate the finite-sample performance of our approach in Monte Carlo experiments and its usefulness in a real data application concerning conformal prediction of infant birth weights (with and up to ). In sum, our -based method produces a much sparser estimator than the -penalized and non-convex penalized approaches without compromising precision.
Cite
@article{arxiv.2006.11201,
title = {Sparse Quantile Regression},
author = {Le-Yu Chen and Sokbae Lee},
journal= {arXiv preprint arXiv:2006.11201},
year = {2023}
}
Comments
51 pages, 3 figures, 3 tables