Forecast of sugar demand in retail using SARIMA and decomposition models case study: a retail store in Indonesia

Abstract

This study discusses forecasting demand in a retail store, focusing on sugar, which is a staple food in Indonesia, as the research object. Despite its importance and forecast challenge, there is no research has been done on sugar at the retail level. This study aims to find the most suitable forecast model that can capture data patterns well to give a good prediction of sugar sales in a retail store in Indonesia by comparing SARIMA and decomposition models. This study uses a stationary test and ACF pattern analyses to prepare the data, a residual test to avoid forecast bias, cross-validation to check the forecast model performance, and MAPE as the performance indicator. SARIMA (0,0,0)(0,1,1)8 and multiplicative decomposition with 3 periods of double-moving average models are chosen. Both models have similar patterns but different slopes because the decomposition model is more sensitive to data patterns, resulting in different MAPEs, which are 15.22% and 13.64%.  Despite the popularity of SARIMA, decomposition can be an interesting alternative to use since it can capture trend data patterns better. However, the short forecast period is preferable for the decomposition model to avoid high trend slope prediction in the long run, leading to more frequent forecast activity and higher resources compared to SARIMA.