Optimal Forecast Combination with Univariate Models for Natural Gas Prices in Spain

The forecast combination for energy prices has generally focused on the combination of neural network algorithms. This paper extends the literature by combining conventional models with neural network models. This work analyzes the point forecast combination applied to the daily prices of natural gas in Spain during 2023. To do this, a model with a deterministic trend and ARIMA process in the residuals (RMSE = 1.40) is combined with a neural networks model (RMSE = 1.65) offering excellent results in terms of predictive ability (RMSE = 0.88) in the short term. The best forecasting performance was obtained by applying the Bates and Granger combination method, estimating very similar weights to the two estimated models: 0.58 and 0.42, respectively. The presented combined model outperforms the models reported in the literature. The result of the forecast exercise indicates that the fall in the price of natural gas will continue.