Introducing Appropriate Models to Forecast Gas-oil Price

The aim of this study is to introduce appropriate models for forecasting the gas oil prices in Singapore market which influence the gas-oil price in the Middle East. The used data are on weekly basis and covering the period of (1997-2010).The forecasts were made for 10, 20 and 30 percent ages of the data, separately. The models have been employed for prosecutions included four models of Neural Network and ARIMA model. The 4 selected Neural Networks are Feed-Forward Back Propagation, Cascade Back Propagation, Elman Back Propagation and Generalized Regression. The training functions are also Levenberg- Marquardt and Quasi-Newton BFG. The results that the least forecasting error belongs to the network in the Levenberg-Marquardt training function has been used. The results also reveal that for forecasts of 20 and 30 percent of data, Elman Back Propagation and for the 10 percent of the data Feed-Forward Back Propagation networks have the least forecast error. Moreover, the findings also reveal that Generalized Regression network and ARIMA have the largest forecasting error as compared to the other models. However, the findings of Diebold-Mariano statistics showed no statistically significant difference among the networks with the least forecast error, in respect to forecasting accuracy for the gas-oil price using a combination of 80 percent for training and 20 percent of day a for forecasting, as compared to the data combination, how less forecasting error. Finally, forecasting error of in the case of the best model is around 2 percent.