Does complexity compensate for accuracy in annual final energy demand forecasting? A multi-methods case study in G7 countries

A significant gap exists regarding the method that should be used to accurately forecast the annual final energy demand. This paper exploits divergent forecasting approaches, ranging from time series and regression-based methods to a global dynamic-recursive Computable General Equilibrium (CGE) macroeconomic model, to forecast the annual final energy demand of the G7 countries over the 2000-2009 and 2010-2019 periods, at 5-year and 10-year time steps. The basic socio-economic indicators of the G7 countries over the 1972-2019 period are harvested or calculated to train or feed up the models utilized. Results suggest that there is no single best method to forecast the annual final energy demand, with the best-performing method depending on the forecasting horizon and characteristics of the country in question. Linear regression models should be utilized in "perfect foresight exercises" that hinge on a "what-if" rationale (e.g., scenario analyses), whereas time series methods should be prioritized over their counterparts in "myopic foresight cases".