Economic Forecast Evaluation: Profits versus the Conventional Error Measures.

Economists are often puzzled as to why profit-maximizing firms buy professional forecasts when statistics such as the root-mean-squared error or the mean absolute error often indicate that a naive model will forecast about as well. This paper argues that the reason is that these traditional summary statistics may not be closely related to a forecast's profits. Using profit measures, the authors find only very weak relationships between such summary error statistics and forecast value. If these results are robust, then least-squares regression analysis may not be appropriate for many studies of economic behavior. Copyright 1991 by American Economic Association.