Seasonal forecasting of agricultural commodity price using a hybrid STL and ELM method: Evidence from the vegetable market in China

Abstract In view of the importance of seasonal forecasting of agricultural commodity price, particularly vegetable prices, and the limited research attention paid to it previously, this study proposes a novel hybrid method combining seasonal-trend decomposition procedures based on loess (STL) and extreme learning machines (ELMs) for short-, medium-, and long-term forecasting of seasonal vegetable prices. In the formulation of the proposed method (termed STL-ELM), the original vegetable price series are first decomposed into seasonal, trend, and remainder components. Then, the ELM is used to forecast the trend and remainder components independently, while the seasonal-naive method is used to forecast seasonal components with a 12-month cycle. Finally, the prediction results of the three components are summed to produce an ensemble prediction of vegetable prices. In addition, an iterated strategy is used to implement multi-step-ahead forecasting. In terms of two accuracy measures and the Diebold-Mariano test, the experimental results show that the proposed method is the best-performing method relative to the competitors listed in this study, indicating that the proposed STL-ELM model is a promising method for vegetable price forecasting with high seasonality.