Granger Causality: A Review and Recent Advances.

Area of Science:

• Time Series Analysis
• Econometrics
• Neuroscience
• Genomics

Background:

• Granger causality, introduced over 50 years ago, is widely used for time series analysis.
• Its validity for inferring causal relationships remains a subject of ongoing debate.
• Computational limitations historically restricted Granger causality to bivariate vector autoregressive processes.

Purpose of the Study:

• To review early developments and debates surrounding Granger causality.
• To discuss recent advancements addressing the limitations of traditional Granger causality methods.
• To highlight new models applicable to complex and diverse time series data.

Main Methods:

• Review of historical Granger causality frameworks and their limitations.
• Exploration of recent methodological advancements for time series analysis.
• Discussion of models accommodating high-dimensional, nonlinear, and non-Gaussian data.

Main Results:

• Recent advances have overcome limitations of early Granger causality models.
• New methods enable analysis of high-dimensional, nonlinear, and non-Gaussian time series.
• The framework now supports subsampled and mixed-frequency time series analysis.

Conclusions:

• Granger causality has evolved significantly beyond its original bivariate application.
• Modern approaches enhance the robustness and applicability of Granger causality for causal inference.
• The reviewed advancements broaden the utility of Granger causality across various scientific domains.