Forecasting Final Energy Consumption in the European Union with Exponential Smoothing

Optimising final energy consumption is the key to reducing the adverse impact of human activity on the environment and limiting climate change. The capability of forecasting the energy consumption of the economy of each of the Member States of the European Union is a prerequisite for the development of policies for adequate behaviour in the energy market. This paper aims to evaluate exponential smoothing forecasts of final energy consumption for 27 EU countries using annual time series data from Eurostat’s Simplified Energy Balances for the period 1990–2022. We consider exponential smoothing in its evolution from methods to state-space models, outlining its advantages with respect to forecasting a large number of indicators in a relatively accurate, automated and easy-to-interpret way. We estimate non-seasonal exponential models for 244 time series with maximum likelihood and select the best models based on the corrected Akaike Information Criterion. Further, we diagnose the model residuals with the Box-Ljung test. We compare the forecasting performance of the exponential smoothing models to mean, naïve, drift and auto arima models. We evaluate forecast accuracy using cross-validation on the basis of root mean square error.