Forecasting Indonesian Coal Price Using Hybrid Transfer Function-Machine Learning with Exogenous Variables

An accurate forecasting model for future coal prices is crucial because coal is the critical raw material for energy production and the primary energy source. Forecasting coal prices will assist the government in formulating energy policy and achieving national energy security. The pattern of Indonesian coal prices data is occasionally tricky because coal, the principal energy commodity, is highly susceptible to being influenced by various factors. This research used a hybrid model that combined two methods. The hybrid approach is expected to improve forecast accuracy by capturing data patterns. We also used exogenous variables in the model. We combine Transfer Function (TF) as a linear model with Machine Learning. Machine learning methods used are Support Vector Regression (SVR), Neural Network (NN) and Long Short-Term Memory (LSTM). The model's performance is measured by Root Mean Square Error (RMSE) and Mean Average Error Percentage (MAPE). We compared the hybrid model to another competitive model, including classical Transfer Function and Transfer Function with outlier as dummy variable. We revealed that hybrid Transfer Function-NN is the best model to forecast the Indonesian coal prices with the lowest RMSE and MAPE in out-sample data. We also provide that the significant exogenous variable that influence Indonesian Coal Price is world natural gas prices (X2,t) and Monthly Average Rates USD to IDR (X5,t). Therefore, the proposed approach is an effective and promising technique for forecasting future fluctuations in Indonesian coal prices.