ON THE PENALTY FACTOR FOR AUTOREGRESSIVE ORDER SELECTION IN FINITE SAMPLES

The order selection criterion that selects models with the smallest sq uared error of prediction is the best. The finite sample theory descri bes equivalents for asymptotic order selection criteria that are bette r in the finite sample practice. This correction for finite sample sta tistics is the most important. Afterwards, a preference in order selec tion criteria can be obtained by computing an optimal value for the pe nalty factor based on a subjective balance of the risks of overfitting and underfitting.