Previsão de demanda de crédito na região centro-oeste do Brasil

Abstract

In this assignment, forecasts were made of the series of credits in the Center-West region of Brazil. The methods of Box-Jenkins and Holt-Winters, Additive and Multiplicative, as well as the linear combination of these two were used he models were obtained by minimizing the RMSE error, with the exception of the ARIMA model, which was based on the Box-Jenkins methodology, and its predictions were compared to each other by means of the mean absolute error (MAPE) for both training sample and sample. In the training sample, the MAPE obtained with the linear combination model was 4,62%. With individual Box-Jenkins, HW Additive and HW Multiplicative methods was 4,58%, 4,71% and 5,61%, respectively. While for the test sample, the MAPE obtained with the linearly combined model was 6,11%. With the individual Box-Jenkins, HW Additive and HW Multiplicative methods the was 7,66%, 6,18% and 5,4%, respectively. The composite model provided better predictions than the individual ones in the training sample.