Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 70 - Multivariate Statistical Methods
Type: Contributed
Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #314007
Title: Envelope Huber Regression
Author(s): Le Zhou* and Dennis Cook and Hui Zou
Companies: University of Minnesota and University of Minnesota and University of Minnesota
Keywords: Envelope; Huber regression; Dimension reduction

We introduce a newly proposed envelope huber regression. It is known that huber regression is robust in the scenario where there are outliers and error follows heavy tail distribution, while maintaining acceptable efficiency loss compared to OLS when error follows normal distribution. In this paper we generalize huber regression using envelope model. Assuming the true signal lies in a reducing subspace of the covariance matrix of predictors, it is known that envelope model could produce estimator with smaller standard error compared to OLS. We borrow this intuition, propose the formulation of envelope huber regression, and introduce a non-likelihood based estimation procedure through Generalized Method of Moments (GMM). The asymptotic normality of our estimator is established. Moreover, we show that envelope huber regression is more efficient than usual huber regression, namely, it could produce estimator with smaller standard error compared to usual huber regression. This helps us achieve better performance in hypothesis testing and statistical inference, especially in the case where error distribution is heavy-tailed other than just normal.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program