Order selection for the multirate burg algorithm

In many applications one is interested in the behavior of a process at various time scales or intervals T. The key issue in accurate time series analysis at interval T is order selection, rather than parameter estimation. Order selection at interval T uses knowledge of the statistal behavior of estimated AR parameters at interval T. A finite sample expression for this behavior is derived and verified in simulations. Spectral details which are not considered significant with standard modeling are accurately described with modeling at interval T. As a result, order selection at interval T yields better models than application of standard order selection.