MODEL DISTANCE CRITERION FOR ORDER DETERMINATION OF ARMA MODELS

摘要

This paper addresses the issue of order determination for ARMA models. A novel concept called the model distance is presented to determine the distance between two possible models when the prediction error is the primary concern of these models. Through a comparison between the model distance from a lower order model to a higher order model (which is regarded to be a measure of the improvement of the modelling accuracy due to an increase in the number of parameters) and an estimate of the modelling accuracy associated with the higher order model, a model distance criterion is proposed. It is shown that the models selected by this criterion is either optimal or nearly optimal for purpose of prediction. Simulations are performed to verify the effectiveness of the model distance criterion.