Modified simple exponential smoothing

Abstract

In this study, we propose a new exponential smoothing method, modified simple exponential smoothing (MSES) as an alternative to simple exponential smoothing (SES). Despite its success and widespread use in many areas, SES has some shortcomings that negatively affect the accuracy of forecasts made using this method. For example, there is no agreed upon concensus on choosing an initial value and determining an optimum smoothing parameter and these decisions greatly affect the forecasting accuracy of SES. The proposed method will help cope with these shortcomings. It is compared to SES on popular metrics that are commonly used for evaluating performance of forecasting techniques and is shown to have better performance. The two models are applied to the 1001 time series of the M-competition data simultaneously and their prediction accuracies are compared under various settings.