Transfer Entropy on Rank Vectors

A number of measures have been proposed for the direction of the coupling between two time series, and transfer entropy (TE) has been found in recent studies to perform consistently well in different settings. Symbolic transfer entropy (STE) has been very recently proposed as a variation of the transfer entropy operating on the ranks of the components of the reconstructed vectors rather than the reconstructed vectors themselves. Here, an improvement of STE is proposed. Specifically, the ranks of the samples of the response system for given time steps ahead are computed with regard to the current reconstructed vector. The grounds of this modification are given and the new measure, called Transfer Entropy on Rank Vectors (TERV), is compared to STE and TE on different settings of state space reconstruction, time series length and observational noise. The results on two simulated systems have shown that the detection of the direction and strength of coupling is improved with TERV over both...

• Type: Journal
• Year: 2010
• Where: CORR
• Authors: Dimitris Kugiumtzis
• Keywords: CORR 2010, Education, Entropy, Reconstructed Vector, Transfer Entropy