Um modelo econométrico + insumo-produto para a previsão de longo prazo da demanda de combustíveis no Brasil

Abstract

This work developed a model for long run forecasting the demand for fuels in Brazil based on the integration of a time series econometric model with a hybrid input-output model. The data used to estimate the econometric model were the annual time series of of final demand components (GDP, household consumption, investment, exports, government spends and imports) produced by the System of National Accounts of the IBGE. The development of the hybrid input-output module made use of the national input-output matrix of Brazil for the year 2005, estimated by IBGE, and the data on petroleum derivatives (gasoline, diesel oil, combustible oil and alcohol, measured in “tep”) consumption available in the National Energy Balance of 2008. In the construction of the econometric model, it was verified that the time series in addition to not being stationary are also cointegrated. Thus, Vector Error Correction models were estimated for the original variables and, in an alternative version, for those variables transformed in log. To select the best model, a predictive test was performed on the models for years 2004 to 2007 and the conclusion was that the model with variables in log presented the best performance. In a similar fashion, it was also undertaken a predictive test at the same period to evaluate the forecasting ability of the integrated model with regard to fuels consumption. Afterwards, the integrated model was then used to generate future forecasts in the period 2008 to 2017. Two scenarios for the exogenous variables were considered: one featuring a short duration for the current world crisis and another featuring a long duration. In both cases, the projections indicated an increase in the fuels demand for the next 10 years.