Comparação de métodos estatísticos para previsão da demanda de café no Brasil

Abstract

The Brazil is the world's largest producer and exporter and the second in terms of consumption and this product is of great importance for the country in historical, economic and cultural terms. It is also understood the need to forecast demand as a tool to help managers in relation to the structuring of strategic planning, PCP, marketing, sales, as well as several other areas to be used directly or indirectly in practically all the coffee production chain. The main purpose of this study was to carry out forecasts for coffee demand in Brazil through the making, interpretation and evaluation of several mathematical models. The models were based on statistical methodologies explored in the literature and applied the historical series of coffee consumption in Brazil between 90s and 2016s. The evaluation was done by means of a direct comparison between their Mean Absolute Deviation (MAD) and the interpretation of the Theil U coefficient to select which methods excelled in relation to the trivial prediction. By means of these measures of accuracy and the coherence analysis to the natural behavior of the data, the ARIMA model (1,1,5) was chosen, with which, it was generated projections for the 5 periods after the last of the real series of consumption. It should be noted that, as the projection for the following year is more relevant, due to the annual characteristics of the data and possible readjustments with the value of the current year in the models for the next year. Thus, a forecast for the 2017s with the adjusted model and 95% confidence interval was 21,438,900 bags of 60 kg, which was considered as estimates of the reference organizations in the area.