Profitability and the evaluation of economic forecasts
Description
The intent of this dissertation is to reconcile the observed actions of profit-maximizing agents who buy analysts' or econometric forecasts. This will entail a discussion of the shortcomings of magnitude error measures as indicators of profits, and the development of a more important role for directional accuracy as a forecast criterion. The case against the literature's use of magnitude errors, and in favor of directional accuracy, proceeds from consideration of how forecast purchasers use the forecasts: to prepare for change, so as to make profits (or avoid losses). Essentially the argument is that the predicted direction of change represents more important information to the purchaser than the predicted magnitude of change The plan of investigation is to measure the ex post performance of actual interest rate and GNP forecasts of different types, not only in terms of magnitude errors (Root Mean Squared Error, Average Absolute Error, and the Theil U statistic) and in terms of directional accuracy, but where possible to calculate an actual profit rule from using forecasts. Following this, the correlations among profits, directional accuracy, and the magnitude error measures are determined. The statistical significance of the correlations, and the probabilities of the observed significant correlations for different evaluators will provide evidence regarding how well the various evaluation criteria reflect actual use profitability The results of the investigation are that profitability is not strongly associated with the magnitude error measures, but profitability is strongly associated with directional accuracy. This outcome casts serious doubt on the validity of nearly all the existing literature on forecast evaluation. Once the evidence of a strong profits & directional accuracy link is coupled with the results indicating analyst and econometric forecasts have greater directional accuracy than other forecasts, it becomes apparent that the profits of analyst and econometric forecasts (despite their worse performance in magnitude error terms) must be greater as well. The superior profitability of the analyst and econometric forecasts is consistent with their on-going purchase by economic agents. As a consequence it appears that the traditional view in the literature is misguided