Um estudo empírico do desempenho de combinações de previsões

Abstract

The combination of forecasts is a procedure that attempts to improve their precision and which takes advantage of the availability of many individual forecast methods. This paper intends to apply the acquired theoretical knowledge throughout the course and thus make a study about the performance of different forecast combina- tions. We compared individual forecast methods and combination methods, using only one performance measure, the Mean Square Error (MSE). The individual forecasts were made through the Holt Exponential Smoothing, as well as through ARIMA models and structural models of local tendency.We did six types of combination: simple average, me- dian, extreme average, regression, optimal combination and optimal combination with independency. To check the best performance of forecast combinations, we fitted several methods in ten real series, stationary and non-seasonal.