Comparative Analysis of Arima, MLPNN and Hybrid Models for Forecasting Natural Gas Spot Prices From Henry Hub

This study aims to forecast natural gas spot prices (NGP) using three modeling approaches: ARIMA, Multi-Layer Perceptron Neural Network (MLPNN), and a hybrid ARIMA-MLPNN model. The analysis is based on monthly data from June 1986 to August 2024. Given the volatility of gas prices caused by geopolitical tensions and market dynamics, improving forecast accuracy is crucial for planning and policymaking. The ARIMA model, suitable for linear patterns, and the MLPNN model, effective in capturing nonlinear relationships, were evaluated alongside a hybrid model that integrates both approaches to address their individual limitations. To evaluate the models, statistical performance metrics were used: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The results showed that the hybrid ARIMA-MLPNN model outperforms both individual models, providing higher forecast accuracy and better handling of structural changes in the data. The hybrid model demonstrated superior performance during recent periods characterized by global instability, confirming its robustness in capturing both linear and nonlinear components of the time series. This research highlights the importance of adopting hybrid forecasting techniques in energy economics, especially for strategic decision-making under uncertainty. The findings indicate that combining traditional statistical models with machine learning enhances predictive capabilities, making the hybrid ARIMA-MLPNN model a reliable tool for forecasting natural gas prices.