A new method using coherence to obtain the model order in the evaluation of partial directed coherence

Recently partial directed coherence has been introduced to study interrelations in multivariate time series, using the vector autoregressive model. In this procedure the choice of the model order is not clear. The use of spectral density has been suggested [B. Schelter, M. Winterhalder, B. Hellwig, B. Guschlbauer, C.H. Lucking, J. Timmer, Direct or indirect? Graphical models for neural oscillators, J. Physiol. (Paris) 99 (2006) 37–46]. This was examined along with a new method which employs coherence. The studies indicate that coherence provides an accurate estimate of the order unlike the spectral density, which underestimated the value of the order leading to inaccurate values for the partial directed coherence. On the other hand the use of coherence gave the correct value for the order, leading to accurate values for the partial directed coherence.