On a Distribution Arising in the Context of Comparative Model Performance Evaluation Problems (original) (raw)

This paper presents a methodology for evaluating the performance of predictive models through a distribution derived from recursive residuals. The proposed approach utilizes a scoring rule based on the average of squared recursive residuals for model comparison, particularly in cases where models exhibit dependent residuals. The findings suggest that the derived statistics follow an F distribution under appropriate circumstances, providing a framework for assessing the adequacy of competing linear models.