This paper proposes an alternative method for applying the Moran's I test in exploratory analyses for spatial autocorrelation. An application of this new method found evidence that regional electricity consumption in Brazil is spatially dependent, presenting a spatial pattern of dissimilarity among regions. Such dependence suggests that the space dimension must be included in the specification of the forecast model to be used so as to ensure consistent, unbiased and efficient estimates. To achieve a higher energy security in the Brazilian Electricity Sector, it is essential to have accurate forecasts of the electricity consumption, and so a forecasting method that considers the spatiotemporal dynamics was proposed. The Spatial ARIMA model (ARIMASp) presented in this paper shows a better predictive performance measured by a reduction of the Mean Absolute Percentage Error of forecasts - as compared to the ARIMA model. These results confirm that spatiotemporal models can improve forecasts of electricity demand in Brazil and demonstrate that considering spatial correlations is paramount to achieving the Brazilian Electricity Sector goals of security of electricity supply, affordability of tariffs and universalization of access to the Brazilian population. (C) 2017 Elsevier Ltd. All rights reserved.