Normalized Multivariate Time Series Causality Analysis and Causal Graph Reconstruction

Causality analysis is an important problem lying at the heart of science, and particular importance in data science machine learning. An endeavor during past 16 years viewing causality as a real physical notion so to formulate it from first principles, however, seems have gone unnoticed. This study introduces community this line work, with long-due generalization information flow-based bivariate time series causal inference multivariate series, based on recent advance theoretical development. The resulting formula transparent, can be implemented computationally very efficient algorithm for application. It normalized tested statistical significance. Different previous work along where only flows are estimated, here also quantify influence unit itself. While forms challenge some inferences, comes naturally, hence identification self-loops graph fulfilled automatically causalities edges inferred. To demonstrate power approach, presented two applications extreme situations. network processes buried heavy noises (with noise-to-signal ratio exceeding 100), second nearly synchronized chaotic oscillators. In both graphs, confounding exist. reconstruct given these easy application immediately reveals desideratum. Particularly, been accurately differentiated. Considering surge interest community, timely.