Necessary and sufficient conditions for causal feature selection in time series with latent common causes

Abstract

We study the identification of direct and indirect causes on time series and provide necessary and sufficient conditions in the presence of latent variables. Our theoretical results and estimation algorithms require two conditional independence tests for each observed candidate time series to determine whether or not it is a cause of an observed target time series. We provide experimental results in simulations, where the ground truth is known, as well as in real data. Our results show that our method leads to essentially no false positives and relatively low false negative rates, outperforming the widely used Granger causality, and two more methods. * Preprint. Under review.

Cause-effect inference through spectral independence in linear dynamical systems: theoretical foundations

“…On the other hand, causal relations between time series are typically explored via Granger-causality type methods (Granger, 1969), which require causal sufficiency. To overcome this limitation, more recent approaches (Mastakouri et al, 2021) (and references therein) employ Markov condition and causal faithfulness to identify characteristic patterns of conditional (in)dependences that witness causal influence even in the presence of hidden common causes.…”

“…These methods can successfully estimate causal relationships when empirical data is generated according to the assumptions, but the results can be misleading when the model is misspecified. In particular, Granger causality may fail to infer the true direction of causation when the sampling of the time series is not fast enough to capture the dynamical interactions precisely (Geweke, 1982;Gong et al, 2015), an issue that also spoils approaches like (Mastakouri et al, 2021), which also -like Granger-relies on observations that refer to measurements at well-defined points in time.…”