On the penalty factor for autoregressive order selection in finite samples

The order selection criterion that selects models with the smallest squared error of prediction is the best. The finite sample theory describes equivalents for asymptotic order selection criteria that are better in the finite sample practice. This correction for finite sample statistics is the most important. Afterwards, a preference in order selection criteria can be obtained by computing an optimal value for the penalty factor based on a subjective balance of the risks of overfitting and underfitting.