Robust regression with asymmetric loss functions.

Area of Science:

• Statistics
• Robust Regression Analysis

Background:

• Traditional robust regression methods assume symmetric error distributions or contamination, which often fails in practice.
• Violating symmetry assumptions leads to reduced efficiency in parameter estimation using standard robust techniques like Tukey's biweight and Huber's method.

Purpose of the Study:

• To develop an asymmetric Tukey's biweight loss function to address non-symmetric error distributions in robust regression.
• To propose a data-driven approach for selecting optimal tuning parameters for the asymmetric loss function.
• To enhance parameter estimation robustness and efficiency under asymmetric conditions.

Main Methods:

• Construction of an asymmetric Tukey's biweight loss function with two tuning parameters.
• Development of a data-driven method for selecting appropriate tuning parameters.
• Implementation of an adaptive algorithm for robust and efficient parameter estimation.

Main Results:

• Simulation studies demonstrate superior performance of the proposed asymmetric method over symmetric methods.
• The method shows improved efficiency when error terms are asymmetric or asymmetrically contaminated.
• The asymmetric approach provides more reliable parameter estimates in practical scenarios.

Conclusions:

• The proposed asymmetric Tukey's biweight method effectively handles non-symmetric error distributions in robust regression.
• The data-driven tuning parameter selection and adaptive algorithm enhance estimation accuracy and robustness.
• The method offers a valuable alternative for analyzing real-world data, such as cardiovascular risk factors.