Nonparametric causal inference for bivariate time series.

Area of Science:

• Statistics
• Time Series Analysis
• Causal Inference

Background:

• Causal inference from time series data is crucial for understanding dynamic systems.
• Existing methods often rely on complex models or embedding procedures, limiting interpretability.
• There is a need for computationally straightforward and model-agnostic causal inference tools.

Purpose of the Study:

• To introduce novel quantities, penchants and leanings, for exploratory causal inference in bivariate time series.
• To provide a clearer interpretation of causal relationships compared to existing methods.
• To develop a computationally efficient and model-independent approach.

Main Methods:

• Development of new probabilistic quantities: penchants and leanings.
• Computation based on a structured method for calculating probabilities.
• Application to bivariate time series without assuming underlying data-generating models or using embedding.

Main Results:

• Penchants and leanings are computationally straightforward to apply.
• These quantities directly follow from probabilistic causality assumptions.
• They do not depend on time series models or embedding procedures, potentially enhancing result interpretability.

Conclusions:

• Penchants and leanings offer a promising new avenue for exploratory causal inference in time series.
• The model-agnostic and computationally simple nature facilitates broader application and understanding.
• These metrics provide a valuable alternative to existing, more complex time series causality tools.