Abstract (Translated):
This paper applies the dynamic linear model (DLM) estimated based on the Bayesian approach to project electricity consumption for five Brazilian regional units. The appeal for the use of DLM in the case of energy consumption is due to the fact that in this model the adjustment occurs in the unit of time ensuring more accurate projections in series with a high degree of variability, such as energy consumption. The results corroborated the expectation regarding the adequacy of DLM for the purpose of making projections. The different forecast validation criteria calculated for a 12-month horizon showed very satisfactory results. In all cases, MLD had a forecast error within the 3% range, taken as a reference for the utilities. In the case of the Center, Northeast and South regions, this indicator was even lower. We tested the robustness of the MDL using a model panel data with random coefficients (MRC) that allows to obtain a set of distinct coefficients for each region, but within a common structure. Regarding forecasting, MRC performance was also reasonable, even underperforming DLM. With the exception of the Southeast region, MRC's Mape was below or in the 3% range for the other regions.