This chapter addresses the application of causal inference techniques to temporal data and systems where dynamics play a significant role. Analyzing cause and effect in time series presents unique difficulties, including autocorrelation, non-stationarity, and feedback mechanisms, which require specialized consideration.
We will start by reviewing the formulation and shortcomings of Granger causality. You will then learn to apply Structural Vector Autoregression (SVAR) models to identify and estimate causal effects in multivariate time series. The discussion extends to sequential decision-making problems through the study of Dynamic Treatment Regimes (DTRs) and methods for their estimation. We will also cover algorithms designed specifically for causal discovery in temporal datasets and explore connections to reinforcement learning, focusing on causal perspectives for off-policy evaluation. The objective is to equip you with methods for sound causal reasoning when dealing with data collected over time.
5.1 Challenges in Causal Inference with Time Series
5.2 Granger Causality: Formulation and Shortcomings
5.3 Structural Vector Autoregression (SVAR)
5.4 Dynamic Treatment Regimes and Estimation
5.5 Time Series Causal Discovery Methods
5.6 Causal Reinforcement Learning and Off-Policy Evaluation
5.7 Practice: Temporal Causal Analysis
© 2026 ApX Machine LearningEngineered with