MODEL SELECTION CRITERIA USING LIKELIHOOD FUNCTIONS AND OUT-OF-SAMPLE PERFORMANCE

Model selection is often conducted by ranking models by their out-of-sample forecast error. Such criteria only incorporate information about the expected value, whereas models usually describe the entire probability distribution. Hence, researchers may desire a criteria evaluating the performance of the entire probability distribution. Such a method is proposed and is found to increase the likelihood of selecting the true model relative to conventional model ranking techniques.