Application of information criteria to AR-order estimation.

R. Moddemeijer

University of Groningen, Department of Computing Science,
P.O. Box 800, NL-9700 AV Groningen, The Netherlands,
phone: +31.50.363 3940 - fax: +31.50.363 38005 - e-mail: rudy@cs.rug.nl

Abstract

Information criteria, like the Akaike criterion, can be applied to AR-order estimations. In recent research we have indicated that the search strategy by the application of information criteria to Auto-Regressive (AR) order estimation is essential. There are essentially two different strategies: search the optimal order by the first local minimum of the criterion or search the order by the global minimum given a maximum candidate order. Although the second strategy is mostly used, we will argue and verify by simulations that the first strategy can give significantly better results.

Keywords

AIC, Akaike criterion, AR, ARMA, autoregressive process, composite hypothesis, GIC, maximum likelihood, model order, system identification, time series analysis.

Published

Application of information criteria to AR order estimation, BibTeX

Related work


other publications