The Impact of Different Data Frequency on Prediction Powers of Various Short- and Long Memory Models: an application to Oil Market Volatility

Many researchers have used different methods to forecast the volatility of goods and capital markets. However, only few have taken into account the importance of data frequency on their predictions. However, none of them have considered the possibility of long run memory in predicting the volatility of oil market. In order to fill this gap in the literature, we have estimated a class of ARFIMA and GARCH models (long- and short-run memory models) with different data frequency to predict oil market volatility. Based on Root Mean Square Error (RMSE) criterion, irrespective of model's type, all models with high frequency data outperform the low frequency data models. The result also shows that at each frequency level, the prediction powers of both ARFIMA and GARCH models are the same. To sum up, we suggest the use of short-run memory model with high frequency data to forecast volatility of oil market. Hence, it seems a proper GARCH model can do the job and there is no need to use ARFIMA model for this purpose.