Forecasting carbon emissions from electricity generation: classical vs functional methods

This study investigates short-term forecasting of carbon emissions from electricity generation in the northern and southern Italian electricity markets. The analysis considers linear parametric models, such as seasonal autoregressive integrated moving average (SARIMA); functional parametric models, including seasonal functional autoregressive (S)FAR and functional autoregressive with exogenous input (SFARX); and nonparametric (possibly nonlinear) models, such as generalised additive models (GAM) and the TBATS model. In addition, forecast combination techniques are explored, incorporating predictions from multiple models through simple averaging, the Bates and Granger (1969) method, and a selection-based approach. Forecast performance is evaluated using hourly root mean squared error (RMSE) over a one-year test period. The results indicate that the functional models consistently achieve the lowest relative RMSE during the early morning hours, whereas the GAM models outperform all others throughout the day, afternoon, and evening. These findings highlight the benefits of forecast combinations, particularly the selection-based approach, which effectively integrates functional and GAM models to improve predictive performance.