A modification of the Mellows — Akaike criterion for choosing the order of a regression model

The article is concerned with choice of the order of a regression model whose forecasting error would be minimal. The well-known Mellows — Akaike criterion is shown to be asymptotically optimal. A choice procedure, more satisfactory for small samples, is proposed and proved to be asymptotically equivalent to the theoretical criterion. Computation results are cited and the procedure is shown to be efficient.