A novel STL-based hybrid model for forecasting hog price in China
Huamin Zhu,Ru Xu,Hongyao Deng
DOI: https://doi.org/10.1016/j.compag.2022.107068
IF: 8.3
2022-07-01
Computers and Electronics in Agriculture
Abstract:Forecasting hog prices is challenging and of great importance for hog producers and administrators due to its role in supporting decision-making. Considering that the price of hogs dynamically changes as a result of potential cyclical growth, seasonal variation and irregular fluctuation, this study proposes a novel hybrid model for forecasting the price of hogs in the next breeding cycle by combining the seasonal-trend decomposition procedure based on loess (STL), support vector regression (SVR) and autoregressive moving average (ARMA). Specifically, the subseries of rising, descending and transitional phases are first obtained based on the identification of hog-price cycles and the corresponding three typical phases. The subseries of the current phase and historical identical-type phases are subsequently combined and designed to estimate the price of hogs in predicted periods so as to capture its trend over time. Then, the hog price series extended by the estimated prices is decomposed by the STL into the cyclical-trend, seasonal and remaining components, which are truncated by the extended length and subsequently predicted by the SVR, seasonal-naïve and ARMA models, respectively, and the resulting forecasts are finally aggregated. Our main contributions are to propose a novel method of extending hog price series to reduce the caudal decomposition deviation caused by the end effect, and provide a tailored prediction method for each hog-price component. The weekly price series of pork and hogs from Jan. 2006 to Dec. 2018 in China are used to evaluate the forecasting performances of the proposed model and the selected counterparts. The empirical results indicate that the proposed hybrid model is a promising alternative for the hog price forecasting.
agriculture, multidisciplinary,computer science, interdisciplinary applications