Enhanced Causal Discovery for Autocorrelated Time Series via Adaptive Momentary Conditional Independence

ICLR 2026 Conference Submission20164 Authors

Keywords: Causal Discovery, Time Series, Autocorrelation, Conditional Independence Tests, Momentary Conditional Independence, PC-algorithm

Abstract: Discovering causal relationships from time series data is essential for understanding complex dynamical systems across a range of domains. However, strong autocorrelation often limits the detection power of existing algorithms and increases the risk of false positives. Moreover, when both lagged and contemporaneous links are considered, existing algorithms are prone to generating false positives in lagged link detection due to indirect causal effects induced by contemporaneous mediators. To address these challenges, the Adaptive Momentary Conditional Independence (aMCI) method is introduced to mitigate the masking effects of autocorrelation and maintain control over false discovery rates. The aMCI adaptively modifies the conditioning set while aggregating conclusions from multiple conditional independence tests. In addition, a multi-phase algorithm is proposed to robustly learn the causal graph by effectively applying the aMCI. The algorithm is designed to be hyperparameter-insensitive, order-independent, and provably consistent under oracle conditions. Extensive evaluations on simulated and benchmark datasets demonstrate that the proposed algorithm substantially improves the accuracy of causal discovery from time series, especially in detecting lagged links.

Primary Area: causal reasoning

Submission Number: 20164