Heuristics approach for the Holt-Winters multiplicative method with new initial values

Abstract

When trend and seasonality are detected, the Holt-Winters multiplicative approach is one of the most commonly used methods for forecasting time series data. Choosing the proper initial values for level, trend, and seasonality plays a vital role in this method. In this paper, a new and efficient procedure to choose the initial values for the Holt-Winters multiplicative method is developed. A total of 12 types of agricultural satellite backscatter values are used for analysis, estimated, and compared with the existing Hansun and Holt-Winters methods and the proposed initial setting method with the best smoothing constants. According to the analysis of the mean absolute percentage error, symmetric mean absolute percentage error, Theil-U statistics, and root mean squared error, the proposed approaches outperformed the existing methods in this experiment.