O uso do método de combinação de previsões aplicadas aos gastos no consumo da cesta básica de Porto Alegre - RS

Abstract

Making forecasts of the average consumption spending of the Basic Basket of Porto Alegre, RS, allows us to have a brief approximation of the possible values to be charged for this product. Given this situation, the objective of this research was to adjust models and combine forecasts. For this, the ARIMA, ARFIMA, SARIMA, SARMA and Holt-Winters Exponential Smoothing models were found, which underwent statistical tests. For the prediction combinations of two, three, four, five and six models, the methods used were Simple Arithmetic Average, Median, Minimum Variance, Ordinary Least Squares Regressions, Pruned Least Squares Robusta, MM Robusta and Quantile, as well as the Reverse Rank and Auto Vector corrected by Bias. Thus, through the MAPE, Theil U, RMSE and MAE Accuracy measurements, the best predictions for combining predictions resulted from the Quantum Regression and Ordinary Least Squares methods.