Variable-Lag Granger Causality for Time Series Analysis

Abstract

Granger causality is a fundamental technique for causal inference in time series data, commonly used in the social and biological sciences. Typical operationalizations of Granger causality make a strong assumption that every time point of the e ect time series is in uenced by a combination of other time series with a xed time delay. e assumption of xed time delay also exists in Transfer Entropy, which is considered to be a non-linear version of Granger causality. However, the assumption of the xed time delay does not hold in many applications, such as collective behavior, nancial markets, and many natural phenomena. To address this issue, we develop Variable-lag Granger causality and Variable-lag Transfer Entropy, generalizations of both Granger causality and Transfer Entropy that relax the assumption of the xed time delay and allow causes to in uence e ects with arbitrary time delays. In addition, we propose a method for inferring both variable-lag Granger causality and Transfer Entropy relations. We demonstrate our approaches on an application for studying coordinated collective behavior and other real-world casual-inference datasets and show that our proposed approaches perform be er than several existing methods in both simulated and real-world datasets. Our approaches can be applied in any domain of time series analysis. e so ware of this work is available in the R-CRAN package: VLTimeCausality.

Time-varying effects of fuel prices on stock market returns during COVID-19 outbreak

“…Inferring causal correlations from data, however, has been suggested to be a basic difficulty by Amornbunchornvej et al (2019) . They use Granger's definition of causality, which they call " predictive causality," in their work.…”

Causal Inference for Time series Analysis: Problems, Methods and Evaluation

“…While the original BTS method includes all lags up to the selected order p k for each time series, X k , the proposed modification includes only the lags of each X k that are selected at each step of the algorithm. In [5] authors claim that most of the existing works in inferring causal relations from time series data using Granger causality assume that the lag between a cause and an effect is at a fixed time point. To address this problem, they propose a novel method that uses Dynamic Time Warping (DTW) which is a distance measure between two time series along with Granger causality to identify the variable lagged based causality in time series.…”

Beyond Predictions in Neural ODEs: Identification and Interventions

“…The general principles readily extend to more complex model classes. Among several methods developed for causal inference based on Granger causality (Tank et al, 2021;Hyvärinen et al, 2010;Runge et al, 2019;Amornbunchornvej et al, 2019), we compare our method to Dynotears (Pamfil et al, 2020). This choice is due to the fact that our datasets mostly contain cyclic dependencies, which cannot be represented via DAGs.…”