The Research Of International Crude Oil Price Forecasting Model

Oil is an important chemical raw materials and strategic resources, which is regarded as energy security in supporting the national economy sectors running normally. The development of world economic has a close contact with global crude oil market, so the oil price volatility has become an unstable factor that constraining the economy development. Mining the intrinsic elements of oil price fluctuation and forecasting the future trend are very important for the national development, enterprise operation and the people's daily life. This paper established the forecasting model of WTI oil price and forecast the price trend from May 2010 to February 2011, basing on wavelet analysis and support vector machine model. The multi-scale analysis feature of wavelet can decompose price signals into long-term trends and short-term fluctuations. The support vector regression model, which was built on the structural risk minimization principle, has excellent learning ability and generalization ability. The combination of wavelet analysis and support vector regression performs well in simulating nonlinear characteristics of oil market. In empirical research of forecasting oil price long-term trend, the factors such as supply, demand, inventory and economics factors were considered. We combine the qualitative and quantitative analysis to determine the input variables of the SVR model. When forecasting the short-term fluctuations of oil price, the different lagged values were considered as input variables of the model due to the difficulty in quantifying the unexpected events. The nuclear parameters and hyper-parameters have a great impact on the model's predict performance in the process of support vector model building. This paper used multi-grid search method to find the optimal parameters, and constantly trained the original data to establish the final forecasting models. The result shows that the prediction accuracy of the SVR model used in forecasting the long-term price trend increased by 10.5% and 35.7% compared to traditional regression model and neural network model. The accuracy of the oil price direction judgment reached 70%. When forecasting the short-term of the oil price, the SVR model didn't perform as well as time series model.