Assessing Dependences within Multivariate Time Series Partializing the Knowledge of Thirds
A method to estimate from multivariate measurements the dependences within a network of coupled dynamical systems is proposed. The method is non-parametric and resorts to a statistics of the eigen-spectrums of the time series partial correlation matrices. The method is successfully validated on numerically generated data, demonstrating its capability to distinguish between direct and indirect dependences.
Record created on 2008-03-06, modified on 2016-08-08