The goal of this article is to model the desagregated series of taxes in Brazil. We use monthly data of a sample of taxes in charge for 80% of the Brazilian gross tax burden in the 1995-2010 years. To estimate the model we employ a Dynamic Linear Model (DLM) with variable parameter (West & Harrison 1997). The choice of this particular model was motivated by the constant changes made in the Brazilian tax system during these years. The conditional forecast is performed a year ahead out of the sample. The main conclusions of the paper are the following. In general the results seem strongly satisfactory. The forecasts fall inside the error bands and the predicted error is bellow 10% until six steps ahead. Above this horizon the forecast lose efficiency. Although the model performed quite well for some taxes, further efforts are required for others. Finally, the elasticity appears to have fluctuated below the unity or the majority of taxes
Brazilian gross tax burden; Dynamic linear model with parameter variable; Bayesian methods; Forecasting
