A Technique for Evaluating Forecasting Models

This paper presents a new methodology for evaluating the performance of a forecasting model. The evaluation-criterion utilizes a “credibility interval” centered at the model prediction. Given predicted and observed values, the length of the “credibility interval” is increased (or decreased) according as an observed value of the dependent random variable falls out of (or into) the interval. Based on that, various ways of assessing the rating of the model are discussed and illustrative examples are given