First order difference; fractionally difference; overdifference; stationary; forecast
The Box-Jenkins model assumed that the time series is stationary. Generally, researchers will conduct the first order difference as a necessary procedure of stationarity data. The first or second order difference seems to be a good solution towards nonstationarity counterparts, but this effort might lead into the possible over difference. Thus, alternative procedure of fractionally difference can be considered as a solution towards the over difference, since it permits the non-integer value of . However, the fractionally difference has been proved by several researchers to produced poor out-sample forecast as compared to its rival models. Therefore, we investigate the over differenceâ€™s effect on five of the selected world edible oil prices that observed to have long memory behavior. Besides, we compare the performance of two difference models which are the autoregressive integrated moving average (ARIMA) and autoregressive fractionally integrated moving average (ARFIMA) models using the time series data that observed with the over difference and long memory behavior. The forecasting show mixed results and the addressed over difference seems not to give a significant effect neither ARIMA nor ARFIMA models. We also found that the ARFIMA model does not demonstrate poor out-sample forecasting.
Full Text : PDF