Predicting taxation on consumption of goods for the state of Minas Gerais: a comparation among Arima and Arfima models

Abstract

The aim of this work was to predict the taxations on consumption of goods of the state of Minas Gerais from January 1998 to August 2011. The methodology used was ARIMA and ARFIMA models. The analyzes showed that the most suitable models to model the series were the ARIMA (1,0,1) and ARFIMA (1,0.36,1). In evaluating of the models, the ARIMA model was superior to the ARFIMA in the criteria for Root Mean Squared Error (RMSE), Mean Absolute Prediction Error (MAPE) and Theil's Coefficient of Inequality . On the other hand, the ARFIMA model was superior in the criteria (MAPE). In general, the models fitted well to the data and have proved to be useful tools in assisting decision making by public managers.