Enhanced Coal Price Forecasting Using Time Series and Regression Models: A Data-Driven Approach

Accurate coal price forecasting is essential for optimising procurement and managing costs in the energy sector as well as in the steel manufacturing department. This study develops a robust forecasting model by combining time series analysis and regression techniques, using business days data from April 2, 2020, to July 28, 2024. The forecasting approach involves two main methods. First, a univariate Holt-Winters multiplicative seasonality with trend model is employed to directly forecast coal prices[3][7]. Second, a regression model is developed by incorporating external factors that influence coal prices. Through correlation analysis, key external factors such as global oil prices, exchange rates, and economic indicators were identified[2]. Forecasts for these external factors were generated using the Holt-Winters model, and these predictions were used as inputs for the regression model, with actual coal prices as the target variable. Model performance was assessed using Mean Absolute Percentage Error (MAPE) for both training and test datasets. A selection of regression models was evaluated, and the best-performing model was chosen based on the lowest test MAPE from the three months leading up to the forecast start date. Once the best model was identified, it was trained on the entire dataset to predict coal prices for the specified forecast period. The results demonstrate that integrating external factors with regression models significantly improves forecast accuracy[5]. This study highlights the value of combining advanced time series methods with regression techniques to support more informed decision-making in coal procurement. Future research could focus on incorporating additional variables and exploring machine learning models to further enhance forecast precision.