Time Varying Parameters Bayesian Forecasting of Electricity Demand: the Italian Case

by Margherita Grasso

Electricity demand is modeled as a time-varying parameters (TVP) vector autoegression with or without imposing cointegration. The paper applies Bayesian strategies where all or a part of the parameters are allowed to vary, and compares their forecasts performances with alternative time series models, namely a seasonal ARIMA (SARIMA) specification and a vector error correction model (VECM). Considering Italian data, the appropriate diagnostic tests and estimation results are in favour of non-stability of the parameters. However, the forecasts abilities of the models do not show significant differencies when measured by RMSE and MAE, and compared trough the Diebold Mariano statistic. On the other hand, forecast intervals of Bayesian models show higher empirical coverage rates.