A new method of evaluating the accuracy of economic forecasts

Conventional measures of forecasting accuracy reflect the view that forecast evaluation should concentrate on all large disturbances and ignore turning-point errors. Many forecasters, however, believe missed turns are the most grievous of all forecasting errors. Despite this consensus, no generally acceptable measure of this type of forecasting error exists. In this paper, such a measure—the probability of correctly forecasting directional change—is introduced. Values of this measure are computed for eleven well-known macroeconometric forecasting models. An inequality-type index of relative directional accuracy based on this measure is also presented and used to evaluate the models in terms of their relative accuracy.