Escaping the curse of dimensionality in estimating multivariate transfer entropy

Jakob Runge, Jobst Heitzig, Vladimir Petoukhov, Juergen Kurths

Research output: Contribution to journalArticle

137 Citations (Scopus)

Abstract

Multivariate transfer entropy (TE) is a model-free approach to detect causalities in multivariate time series. It is able to distinguish direct from indirect causality and common drivers without assuming any underlying model. But despite these advantages it has mostly been applied in a bivariate setting as it is hard to estimate reliably in high dimensions since its definition involves infinite vectors. To overcome this limitation, we propose to embed TE into the framework of graphical models and present a formula that decomposes TE into a sum of finite-dimensional contributions that we call decomposed transfer entropy. Graphical models further provide a richer picture because they also yield the causal coupling delays. To estimate the graphical model we suggest an iterative algorithm, a modified version of the PC-algorithm with a very low estimation dimension. We present an appropriate significance test and demonstrate the method’s performance using examples of nonlinear stochastic delay-differential equations and observational climate data (sea level pressure).
Original languageEnglish
Article number258701
Number of pages5
JournalPhysical Review Letters
Volume108
Issue number25
Early online date21 Jun 2012
DOIs
Publication statusPublished - 22 Jun 2012

Fingerprint

estimating
entropy
estimates
sea level
climate
differential equations

Cite this

Escaping the curse of dimensionality in estimating multivariate transfer entropy. / Runge, Jakob; Heitzig, Jobst; Petoukhov, Vladimir; Kurths, Juergen.

In: Physical Review Letters, Vol. 108, No. 25, 258701, 22.06.2012.

Research output: Contribution to journalArticle

Runge, Jakob ; Heitzig, Jobst ; Petoukhov, Vladimir ; Kurths, Juergen. / Escaping the curse of dimensionality in estimating multivariate transfer entropy. In: Physical Review Letters. 2012 ; Vol. 108, No. 25.
@article{c386e303b0484c52907eac2ccbebbfd9,
title = "Escaping the curse of dimensionality in estimating multivariate transfer entropy",
abstract = "Multivariate transfer entropy (TE) is a model-free approach to detect causalities in multivariate time series. It is able to distinguish direct from indirect causality and common drivers without assuming any underlying model. But despite these advantages it has mostly been applied in a bivariate setting as it is hard to estimate reliably in high dimensions since its definition involves infinite vectors. To overcome this limitation, we propose to embed TE into the framework of graphical models and present a formula that decomposes TE into a sum of finite-dimensional contributions that we call decomposed transfer entropy. Graphical models further provide a richer picture because they also yield the causal coupling delays. To estimate the graphical model we suggest an iterative algorithm, a modified version of the PC-algorithm with a very low estimation dimension. We present an appropriate significance test and demonstrate the method’s performance using examples of nonlinear stochastic delay-differential equations and observational climate data (sea level pressure).",
author = "Jakob Runge and Jobst Heitzig and Vladimir Petoukhov and Juergen Kurths",
year = "2012",
month = "6",
day = "22",
doi = "10.1103/PhysRevLett.108.258701",
language = "English",
volume = "108",
journal = "Physical Review Letters",
issn = "0031-9007",
publisher = "American Physical Society",
number = "25",

}

TY - JOUR

T1 - Escaping the curse of dimensionality in estimating multivariate transfer entropy

AU - Runge, Jakob

AU - Heitzig, Jobst

AU - Petoukhov, Vladimir

AU - Kurths, Juergen

PY - 2012/6/22

Y1 - 2012/6/22

N2 - Multivariate transfer entropy (TE) is a model-free approach to detect causalities in multivariate time series. It is able to distinguish direct from indirect causality and common drivers without assuming any underlying model. But despite these advantages it has mostly been applied in a bivariate setting as it is hard to estimate reliably in high dimensions since its definition involves infinite vectors. To overcome this limitation, we propose to embed TE into the framework of graphical models and present a formula that decomposes TE into a sum of finite-dimensional contributions that we call decomposed transfer entropy. Graphical models further provide a richer picture because they also yield the causal coupling delays. To estimate the graphical model we suggest an iterative algorithm, a modified version of the PC-algorithm with a very low estimation dimension. We present an appropriate significance test and demonstrate the method’s performance using examples of nonlinear stochastic delay-differential equations and observational climate data (sea level pressure).

AB - Multivariate transfer entropy (TE) is a model-free approach to detect causalities in multivariate time series. It is able to distinguish direct from indirect causality and common drivers without assuming any underlying model. But despite these advantages it has mostly been applied in a bivariate setting as it is hard to estimate reliably in high dimensions since its definition involves infinite vectors. To overcome this limitation, we propose to embed TE into the framework of graphical models and present a formula that decomposes TE into a sum of finite-dimensional contributions that we call decomposed transfer entropy. Graphical models further provide a richer picture because they also yield the causal coupling delays. To estimate the graphical model we suggest an iterative algorithm, a modified version of the PC-algorithm with a very low estimation dimension. We present an appropriate significance test and demonstrate the method’s performance using examples of nonlinear stochastic delay-differential equations and observational climate data (sea level pressure).

U2 - 10.1103/PhysRevLett.108.258701

DO - 10.1103/PhysRevLett.108.258701

M3 - Article

VL - 108

JO - Physical Review Letters

JF - Physical Review Letters

SN - 0031-9007

IS - 25

M1 - 258701

ER -