Fit a multivariate autoregressive (MVAR) model

Fits X(t) = sum_{k=1}^{p} A_k X(t-k) + E(t), with E(t) white noise of covariance Sigma, to a multichannel time series. This is the shared MVAR estimator used across the ecosystem for connectivity spectra (DTF, PDC, spectral Granger causality). The series is mean-centred before fitting.

Usage

.fitMVAR(X, order, method = c("ols", "yulewalker", "nuttall-strand"))

Arguments

X

A numeric matrix (time x channels).

order

Integer model order p.

method

Estimator: "ols" (ordinary least squares, the default), "yulewalker" (block-Toeplitz normal equations), or "nuttall-strand" (Nuttall-Strand / Vieira-Morf multichannel lattice).

Value

A list with the coefficient array A (channels x channels x order), the residual covariance Sigma, the order, and the method.

References

Barnett, L., & Seth, A. K. (2014). The MVGC multivariate Granger causality toolbox. Journal of Neuroscience Methods, 223, 50-68. doi:10.1016/j.jneumeth.2013.10.018

See also

mvarFit, mvarOrderSelect, .mvarSpectral