Resumen
This paper examines exponential smoothing constants that minimize summary error measures associated with a large number of forecasts. These forecasts were made on numerous time series generated through simulation on a spreadsheet. The series varied in length and underlying nature no trend, linear trend, and nonlinear trend. Forecasts were made using simple exponential smoothing as well as exponential smoothing with trend correction and with different kinds of initial forecasts. We found that when initial forecasts were good and the nature of the underlying data did not change, smoothing constants were typically very small. Conversely, large smoothing constants indicated a change in the nature of the underlying data or the use of an inappropriate forecasting model. These results reduce the confusion about the role and right size of these constants and offer clear recommendations on how they should be discussed in classroom settings.