An improved divergence information criterion for the determination of the order of an ar process

Abstract

In this article we propose a modification of the recently introduced divergence information criterion (DIC, Mattheou et al., 2009) for the determination of the order of an autoregressive process and show that it is an asymptotically unbiased estimator of the expected overall discrepancy, a nonnegative quantity that measures the distance between the true unknown model and a fitted approximating model. Further, we use Monte Carlo methods and various data generating processes for small, medium, and large sample sizes in order to explore the capabilities of the new criterion in selecting the optimal order in autoregressive processes and in general in a time series context. The new criterion shows remarkably good results by choosing the correct model more frequently than traditional information criteria. © Taylor & Francis Group, LLC.