The competition of robust methods in linear and nonlinear regression with heavy-tailed stable errors

Document Type : Original Article

Authors

Department of Mathematics and Computer Science, Amirkabir University of Technology (Tehran Polytechnic), Iran

Abstract

Robust regression methods including Least Trimmed Squares are one of the most important methodologies to compute exact coefficient estimators when data is polluted with outliers. There is interest to generalize Least Trimmed Squares for regression models with heavy-tailed stable errors. In this manuscript, we compare estimating coefficients methods with the robust Least Trimmed Squares method in stable errors case. Therefore, we propose Stable Least Trimmed Squares and Nonlinear Stable Least Trimmed Squares methods for linear/nonlinear regression models with stable errors, respectively. The joint distribution of ordered errors is used with the finite variance property of ordered stable errors, whose indexes are defined by cut-off points. We make many comparisons using simulated and real datasets.

Keywords

Main Subjects

AUT Journal of Mathematics and Computing

Articles in Press, Accepted Manuscript
Available Online from 05 May 2024

Files

Share

How to cite

Statistics