Comparação de técnicas de previsão em análise de séries temporais: uma proposta via wavelets e modelos tradicionais aplicadas à série de arrecadação do ICMS no RS

Abstract

The main purpose of this research is to compare and explore traditional ARIMA and exponential smoothing models, likewise to propose a combined use of these forecasting techniques with the wavelets decomposition. The comparison is made by mean a empirical study applied to the monthly tax revenue of ICM of the Rio Grande do Sul state. The series period is from January, 1998 till October, 2008 using the date deflated at constant price of October, 2008 using the IGP-DI. Among the exponential smoothing models the additive Holt- Winter model was used, the class of ARIMA and the SARIMA models, because the series shows a seasonal behavior. These same models were used to modeled and forecast each of sub-series of the decomposition in wavelets. The forecasting techniques were compared using different forecasting horizons and was possible to observe that the proposed use of combine algorithmic of Holt-Winters and wavelets decomposition gave a better results to the forecasted values, principally to the horizons with steps of forecasting equal h = 2 e h = 12.