Abstract
Exponential smoothing is one of the most successful naive forecasting schemes, but it suffers from the inability to select the smoothing constants quantitatively without making crucial assumptions about the time series. The method presented, called SAFT, utilizes a modified evolutionary operation first to evolve the smoothing constants directly from the data and then to monitor them to account for changing influences in the data. SAFT provides good forecast accuracy and a more advantageous dynamic response than other forecasting schemes. Because of the modest information storage and computation time, SAFT is recommended when a large number of forecasts are needed on a routine basis.