A comparison of forecasting methods between Winters’ multiplicative exponential smoothing, Box-Jenkins, and combined methods, for forecasting monthly prices of white shrimp Litopenaeus vannamei

Abstract

          The purpose of this research was to compare three forecasting methods: 1) Winters’ multiplicative exponential smoothing, 2) Box-Jenkins, and 3) combined methods, for forecasting monthly prices of white shrimp Litopenaeus vannamei, size of 70 shrimp per kilogram. The set of time series data of 168 values used in this research was monthly price of the white shrimp from January 2005 to December 2018. The first category of 156 values from January 2005 to December 2017, was used for modeling time series models, and the second category of 12 values from January to December 2018, was used for finding the most suitable forecasting method by the criterion of the lowest mean absolute percentage error (MAPE). The results showed that;

          1. The Winters’ multiplicative exponential smoothing model was  , where  and  were forecasting values and seasonal indices respectively,  was natural numbers,  was number of time series data, and   =  + . The correlation coefficient of this model was 0.98.

          2. The Box-Jenkins model was SARIMA (1, 1, 0), (0, 1, 1)₁₂ (no constant), presented by  , where  and  were forecasting and error values, respectively. The correlation coefficient of this model was 0.98.

          3. The combined forecasting model was  , where   ,   and  were forecasting values form this model, Winters’ exponential smoothing model, and Box-Jenkins model, respectively. The correlation coefficient of this model was 0.98.

          4. The most suitable forecasting method for this research was Winters’ multiplicative exponential smoothing method because it had the lowest mean absolute percentage error (MAPE = 16.40), which was less than that from Box-Jenkins method (MAPE=19.10) and combined method (MAPE=17.50).