NlinTS: An R Package for Causality Detection in Time Series

Authors

Youssef Hmamouche, Université de Toulon

Date of this Version

6-2020

Document Type

Article

Citation

The R Journal (June 2020) 12(1); Editor: Michael J. Kane

Comments

Copyright 2020, The R Foundation. Open access material. License: CC BY 4.0 International

Abstract

The causality is an important concept that is widely studied in the literature, and has several applications, especially when modelling dependencies within complex data, such as multivariate time series. In this article, we present a theoretical description of methods from the NlinTS package, and we focus on causality measures. The package contains the classical Granger causality test. To handle non-linear time series, we propose an extension of this test using an artificial neural network. The package includes an implementation of the Transfer entropy, which is also considered as a non linear causality measure based on information theory. For discrete variables, we use the classical Shannon Transfer entropy, while for continuous variables, we adopt the k-nearest neighbors approach to estimate it.