Conditional Local Independence Testing for Dynamic Causal Discovery
By: Mingzhou Liu, Xinwei Sun, Yizhou Wang
Potential Business Impact:
Finds how things affect each other in moving systems.
Inferring causal relationships from dynamical systems is the central interest of many scientific inquiries. Conditional Local Independence (CLI), which describes whether the evolution of one process is influenced by another process given additional processes, is important for causal learning in such systems. However, existing CLI tests were limited to counting processes. In this paper, we propose a nonparametric CLT test for It\^o processes. Specifically, we first introduce a testing statistic based on the Local Covariance Measure (LCM) by constructing a martingale from the conditional expectation of the process of interest. For estimation, we propose an efficient estimator based on the optimal filtering equation, which can achieve root-N consistency. To establish the asymptotic level and power of the test, we relax the restrictive boundedness condition to a moment bound condition, which is practical for It\^o processes. We verify the proposed test in synthetic and real-world experiments.
Similar Papers
Conditional independence testing with a single realization of a multivariate nonstationary nonlinear time series
Methodology
Finds hidden patterns in changing data.
Meta-Dependence in Conditional Independence Testing
Machine Learning (CS)
Finds hidden causes by checking how things relate.
A Fast Kernel-based Conditional Independence test with Application to Causal Discovery
Machine Learning (CS)
Makes finding causes in big data much faster.