LAUSR

This paper investigates the dynamic relationship between steam coal price and its drivers sampling mixed frequencies to improve the prediction of weekly steam coal price. A novel hybrid method, combining the mixed data sampling (MIDAS) model with eXtreme Gradient Boosting (XGBoost) algorithm, is proposed to perform forecast of weekly steam coal prices by applying the latest mixed factors with high frequencies. The empirical evidences indicate that the daily natural gas prices, temperatures, and air quality index (AQI) have better predictive abilities for steam coal prices than the A-share index and crude oil prices. It’s shown that the hybrid model has approximately 23.27% and 78.39% accuracy improvement over the combination-MIDAS and other benchmark models, respectively. The empirical results are helpful for the government to effectively capture the fluctuation and uncertainty of steam coal prices from the energy market and environmental conditions to make reasonable strategies in China.