Investigation of the Effects of Some Statistical Data Components on the Selection of Optimum Smoothing Constant
Keywords:Simple exponential smoothing, optimum smoothing constant, trial and error, demand pattern, number of observations
Simple exponential smoothing is one of the best forecast methods, especially for time series data. Its efficacy depends on a parameter called smoothing constant (α) which, if optimally determined, minimises the mean square error (MSE), the mean absolute error (MAE) and the mean absolute percentage error (MAPE). The widely used method for selecting the optimum smoothing constant is to conduct a grid search within a wide range of possible values of α using the trial-and-error method. Not only that this method involves the knowledge of advanced statistical processes, but it is also time-consuming, and its results are limited to the data being analysed. In order to eliminate these limitations, there is a need to develop a benchmark that will guide the users of simple exponential smoothing to select the optimum α without necessarily repeating the trial-and-error method once a value has been established for data of similar statistical components. This study investigated some statistical components (mean, standard deviation, range, number of observations and pattern) of data to determine which components could aid in the quick and easy determination of optimum smoothing constant. The study determined the optimum smoothing constants for 16 different data of varying statistical components, and found that mean, standard deviation, range and the number of data observations are not related to the optimum smoothing constants. However, the demand pattern is an excellent precursor to determining the optimum smoothing constant. The study recommends further study in developing a classification model for demand patterns in job shops.
Keywords: Simple exponential smoothing; optimum smoothing constant; trial and error; demand pattern; number of observations.