BIRS Workshop Lecture Videos
Quantile regression in Variable Screening Kong, Linglong
We introduce a quantile regression framework for linear and nonlinear variable screening with high-dimensional heterogeneous data. Motivated by success of various variable screening methods, especially the quantile-adaptive framework, we propose to combine the information from different quantile levels to provide more efficient variable screening procedure. In particular, there are two ways to do so: one is to simply take (weighted) average across different levels of quantile regression; the other one is to use (weighted) composite quantile regression. Asymptotically, these two approaches are equivalent in terms of efficiency. Numerical studies confirm the fine performance of the proposed method for various linear and nonlinear models. Joint work with Qian Shi.
Item Citations and Data
Attribution-NonCommercial-NoDerivs 2.5 Canada