Aplicação de um modelo fatorial dinâmico para previsão da arrecadação tributária no Brasil

The aim of this article is to estimate a Bayesian factorial dynamic model for the analysis and forecasting of the Brazilian tax burden (BTB) using monthly data from 1996 to 2007. Twenty taxes are responsible for about 80% of the BTB, each of which with a distinct seasonal pattern The factorial model has no problems accommodating the high dimensionality of the data-contrarily to what happens, for instance, with VARs-while simultaneously allowing the identification of a short number of factors responsible for the joint dynamics of the various taxes. Therefore, this procedure allows one to obtain relevant insights about the public revenues in Brazil. Moreover, due to the fact that seasonality is a remarkable feature of the series of government receipts, the seasonal component is modeled endogenously using a Fourier form representation that is an unrestricted and flexible way to assess seasonality. Finally, we forecast the future path of the public receipts separately for the period of 2008.