MVAR model-order selection by AIC / BIC

Fits MVAR models over a range of orders and returns the order minimising the Akaike (AIC) or Bayesian (BIC) information criterion. For an m-channel model of order p with n effective samples the criteria are n log det(Sigma_p) + 2 m^2 p (AIC) and n log det(Sigma_p) + log(n) m^2 p (BIC).

Usage

mvarOrderSelect(
  X,
  max_order = 20L,
  criterion = c("bic", "aic"),
  method = c("ols", "yulewalker", "nuttall-strand")
)

Arguments

X

A numeric matrix (time x channels).

max_order

Maximum order to consider (default: 20).

criterion

"bic" (default) or "aic".

method

Estimator passed to .fitMVAR (default "ols").

Value

A list with the selected order, the chosen criterion, and a data.frame criteria of order, aic, bic.

See also

.fitMVAR, mvarFit