This paper intends to model time series with the aim of perform forecast using integer and fractional differencing for agricultural commodities future’s price. Time series models of the ARMA/ARIMA type (integer differencing) will be estimated and compared to ARFIMA type models (fractional differencing). In both cases errors are modeled assuming the occurrence of volatility. The forecast power of each model will be compared using the criterion of the mean squared error (MSE). The estimation of fractional term (d) will be also used to examine the long run dependency properties of the series. The results showed that, for all series, returns are stationary. The sugar series, however, showed anti-persistency, while all other series showed to be long memory. The ARFIMA models showed, in general, a better forecasting performance.
ARIMA models; fractional differencing; ARFIMA models
