Combining Machine Learning and Econometric Forecasts for Hog Markets: A Decision Framework for Price Prediction and Production Planning

Abstract The extraordinary fluctuations in China's hog prices, particularly since the outbreaks of the African swine fever in 2018, have brought tremendous risks to hog producers' production planning. This study proposes a decision framework for hog price prediction and production planning. Specifically, we use six forecast combination strategies to integrate multi‐step‐ahead forecasts of the hog price generated by 11 econometric and machine learning methods. Furthermore, we design a practical production plan based on the obtained hog price forecasts to determine the number of piglets purchased and fattened pigs sent to slaughter every month. Our results show that the forecast combinations consistently generate more accurate forecasts across six prediction horizons and three accuracy measures than individual forecasting methods. No model consistently performs best among the 11 individual forecasting models. The least absolute deviation strategy is particularly well‐suited for predicting hog prices. Concerning economic performance, the cumulative returns of hog producers who use the proposed production plan are consistently higher than those who use a naïve production plan. Experimental results indicate that our decision framework is a promising and practical tool for price forecasting and production planning in the hog industry.