Toward a More Precise Definition of Forecastability

The idea of forecastability is valuable because it allows attention to shift from forecasting methods to the series that are being forecast, but the concept of forecastability needs sharpening. • Coefficients of variation and approximate entropy metrics assess the stability of a time series. Stability is an important idea, but it should be distinguished from forecastability. • Forecastability should refer to the range of forecast errors that are achievable, on average, in the long run. A forecastability metric should supply an upper and lower bound of forecast error. Any method producing greater errors (less accurate forecasts), on average, than the upper bound should be discontinued. • There are conceptual and practical difficulties in determining “the lowest achievable level of forecast error.” Recourse to automatic method-selection features of forecasting software, combining forecasts, and use of analogous series can point to the effective lower bound. • Greater attention should be given to identifying more forecastable series. These can be used to assist the forecasting of less forecastable series. They may, in some instances, be used instead of the less forecastable series to guide decision. Copyright International Institute of Forecasters, 2009