Skip to contents

McMillan degree by Bauer's SVC criterion

Source: R/low_level.R
svc_order.Rd

Estimates the state dimension from the past-future sample canonical correlations (CCA weighting), minimising SVC(n) = sigma_{n+1}^2 + C(T) 2n / T (Bauer 2001). Structurally immune to observation noise, which changes the ARMA orders but not the McMillan degree.

Usage

svc_order(x, penalty = "logt", window_factor = 1, window = NULL)

Arguments

x

A numeric vector (a stationary series).

penalty

"logt" (default, Bauer's practical choice) or "fplogt" (his consistent penalty; strongly underestimates in finite samples).

window_factor

The window is f = p = floor(window_factor * p_AIC); Bauer recommends values in [1, 2].

window

Explicit window overriding the rule.

Value

A list with order, the canonical correlations, the criterion values and the window f, p.