This thesis describes two multivariate statistical based approaches to generate unconditional monthly forecasts for the brazilian industrial electricity demand covering the lead time spanning from Jan/2010 to Dec/2010. For that, it was first checked the causality among the series involved followed by stationarity tests. It was also carried out cointegration tests to check the existence of long range trend among the series. The two approaches adopted were, respectivelly, the Classical Error Correction Vector Model (VAR/VEC) and the Bayesian counterpart (BVAR/BVEC); both modelling simultaneously the series involved in the study as a vector of time series that follow a kind of vector autoregressive structure. The results obtained with both, were compared, and, a main conclusion of the thesis, the Bayesian model produced better results, in terms of accuracy, them the Classical counterpart.