Detection of direct and delayed causal effects in time series

(Investigators: A. Krakovská, J. Jakubík, M. Chvosteková)


In cooperation with the Institute of Informatics of the ASCR in Prague, we focused on the detection of causal relationships between time series. The standard approach is the Granger causality test. In addition, we analyzed methods that estimate conditional mutual information between signals, or work with a degree of predictability and mutual mappability in reconstructed state spaces. We have shown that the classical Granger test is suitable for use only in cases allowing autoregressive modeling. Modern methods are more effective in complicated cases, such as interconnections between complex nonlinear systems or apparent causality due to external influences. However, none of the methods has been shown to be generally applicable in determining the delay of a causal effect. For tested continuous chaotic systems e.g. the method of evaluating mutual information dominated, while in discrete systems the methods were more reliable in reconstructed spaces. However, if the studied systems contain a strong oscillating component in their dynamics, then the results of all methods become ambiguous. We have therefore published a study which has shown that the detection of delayed causality is generally a problem which has not yet been solved.

Fig. 1: An example of a discrete population dynamics model in which the reconstructed space mapping method (CCM) correctly detects delays (0, 2, 4) in a causal effect, while the conditional mutual information (CMI) method fails.


Foreign partner:
• Institute of Informatics AS CR, Prague, CR.
Related projects:
• MAD – Bilateral Mobility Project, SAS-AS CR 2015-2018, Synchronization and causality in complex systems: Synchronization and causality in complex systems: time series methods.
• VEGA 2/0011/16, Analysis of causal relationships in complex systems with emphasis on biomedical applications.

 

Publications:

  1. COUFAL, D. – JAKUBÍK, J. – JAJCAY, N. – HLINKA, J. – KRAKOVSKÁ, A. – PALUŠ, M. Detection of coupling delay: A problem not yet solved. In Chaos, 2017, vol. 27, no. 8, p. 083109. ISSN 1054-1500. (2.283-IF2016).
  2. KRAKOVSKÁ, A. Predictability improvement as a tool to detect causality. In MEASUREMENT 2017 : Proceedings of the 11th Int. Conference on Measurement. Editors: J. Maňka, M. Tyšler, V. Witkovský, I. Frollo. – Bratislava, Slovakia: Institute of Measurement Science, Slovak Academy of Sciences, 2017, 39-42. ISBN 978-80-972629-0-7.
  3. KRAKOVSKÁ, A. – HANZELY, F. Testing for causality in reconstructed state spaces by an optimized mixed prediction method. In Physical Review E, 2016, vol. 94, no. 5, p. 052203. ISSN 2470-0045. (2.252-IF2015).
  4. KRAKOVSKÁ, A. – JAKUBÍK, J., CHVOSTEKOVÁ, M. COUFAL D., JAJCAY N., PALUŠ M. (2017): Comparing five methods for detection of causality in bivariate time series, submitted to Physical Review E.