FGLS Using Machine Learning 2017-05-12T23:51:42+00:00

Project Description

More efficient estimation through better modeling of noise.
Co-author: Dick Startz (UCSB)

In this project we examine whether support vector regression (and other machine learning methods) can be effectively used to model the relationship between error variance and explanatory variables in regression models. As part of Feasible Generalized Least Squares (FGLS), these methods offer substantial efficiency gains over ordinary least squares with weaker assumptions than parametric FGLS and better coverage than nonparametric FGLS.