Um modelo de combinação de previsões para arrecadação de receita tributária no Brasil

This study aimed to the application of forecasts combination of model to predict tax revenues in Brazil. Here we combine the predictions obtained from three models: dynamic factor model (DFM), seasonal autoregressive integrated moving average (Sarima) and model of Holt-Winters smoothing. We adopted five criteria for combine predictions: optimal combination, performance, simple regression, simple average and median. We work with monthly data for a total of nine federal taxes for the period from January 2001 to December 2013. The out of sample forecast are done for the year 2014. Considering the results, it can be seen that the combined predictions proved generally superior to those derived from genuine models with the exception of a few taxes. Notwithstanding there is no specific method of combination that provide better forecasts concern the others.