Assessing Granger non-causality using nonparametric measure of conditional independence.

Area of Science:

• Statistics
• Econometrics
• Neuroscience
• Engineering

Background:

• Linear Granger causality is widely used but insufficient for non-linear or higher-order stochastic processes.
• Analyzing complex dynamics requires advanced methods beyond traditional linear approaches.

Purpose of the Study:

• To discuss discovering Granger non-causality using conditional independence.
• To propose a novel conditional independence measure for robust causality analysis.

Main Methods:

• Utilizes conditional independence to detect Granger non-causality.
• Proposes a novel measure estimating conditional distribution via kernel-based least squares regression.

Main Results:

• The proposed method offers a new way to analyze Granger non-causality in complex systems.
• Comparative analysis highlights strengths and weaknesses against existing methods.

Conclusions:

• Conditional independence provides a powerful, assumption-free framework for Granger non-causality.
• The novel kernel-based measure enhances the reliability of causality detection in diverse stochastic processes.