Bagging Cross-Validated Bandwidths in Nonparametric Regression with Application to Large Sample Sizes

• Activity Number: 386 - Nonparametric Modeling II
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Daniel Barreiro Ures* and Ricardo Cao Abad and Mario Francisco Fernández
• Companies: Universidade da Coruña and Universidade da Coruña and Universidade da Coruña
• Keywords: bagging; subsampling; cross-validation; bandwidth selection; nonparametric regression; big data

Abstract

Cross-validation is a well-known and widely used bandwidth selection method in nonparametric regression. However, it has two important drawbacks: (i) the large variability of the selected bandwidths, and (ii) the inability to provide results in a reasonable time for very large sample sizes due to its high computational complexity. To overcome these problems, the use of subsampling and bagging is proposed and the asymptotic properties of the resulting bagged bandwidth are derived. Theory shows that, for appropriate choices of the number and size of the subsamples, the use of bagging can lead to much better rates of convergence. Furthermore, simulation studies show the behavior of the proposed bandwidth selector to be superior to ordinary cross-validation in terms of both statistical precision and computational agility.