Threshold bounds in SVD and a new iterative algorithm for order selection in AR models

摘要

The problem of order determination of AR (autoregressive) models using singular value decomposition (SVD) is reexamined from a statistical point of view. Thresholds for distinguishing between significant and nonsignificant singular values are derived, and a novel iterative algorithm for order selection in AR models is presented. Simulation results show the technique to be very effective when a small number of samples is available.