Modelos Vetoriais de Correção de Erros Aplicados à Previsão de Crescimento da Produção Industrial

Abstract

In this paper we implement and evaluate several forecast econometric vectorial autoregressivemodels for quarterly Industrial GDP. We have built co-integration vectorrestriction for several sets of variables (Industrial GDP, long interest rates, short interestrates, spread, inflation) in order to assess the improvement on forecast performance.A expectative variable of growth of industrial production among firms was alsoutilized and was proved to be valuable. The predictive power of different modelswas evaluated from diverse loss functions evaluated on out-of sample invariable sizerolling-window method. Additionally, we have considered also combining forecastmethods, following the subject literature which has been indicating this approach asone the most efficient (BATES e GRANGER, 1969).We have concluded, the use of co-integration vector may improve substantiallythe forecast performance. Specialy, the interest rate spread has been proved to be aimportant leading indicator of industrial activity as well as the expectative variable(FGV). Furthermore, the c combining forecast models over-performed generally,apart others very good results of individual models.