A Bayesian Approach to the Evaluation of a Forecasting System

It is proposed that forecasting systems should be implemented on a trial basis and evaluated in terms of accuracy and/or economic benefits prior to full‐scale implementation. The conventional method of evaluating a forecasting system is to compute one or more error terms. Problems occur when no error term can be calculated. A methodology for evaluating a forecasting system under such conditions, based on Bayesian analysis, is put forward. The forecasting system subjected to the evaluation process was intended to improve the catch volume of a salmon fishery. Data for analysis were derived from the activities of two groups, one using the proposed forecasting analysis and one not using it, and the economic consequences associated with each.