Handling Unobserved Confounders

Background: Causal inference methods, when applied to time series, often rely on assumptions about the data-generating process (DGP) such as stationarity and causal sufficiency, which may not hold in complex, real-world systems.

Question / Future Work: Extend the Causal-INSIGHT framework to scenarios where unobserved confounding variables exist, which violates the current assumption of causal sufficiency. This would require exploring how the model-implied structure changes under the presence of hidden common causes and potentially developing techniques to assess the robustness of the extracted structure to such violations.

Why It Matters: Handling unobserved confounders is a major challenge in structural causal modeling; extending a model-probing technique to address this gap would significantly broaden its practical applicability beyond idealized settings.

Evidence: We assume throughout this work: (3) causal sufficiency, i.e., the absence of unobserved confounders. These assumptions are standard in time-series causal analysis and Granger-based discovery frameworks