The Price Prediction Of International Crude Oil

Crude oil plays an important role in the world economy and national security, it is taken for granted that the analysis and prediction of its price has been a hot topic all the time.This article mainly uses statistical theories and methods, and combines the achievement of natural science(neural networks and genetic algorithms), to analyse the volatility characteristics of international crude oil price—take WTI(West Texas Intermediate Crude Oil) for example. At the same time, we explore forecasting models with better predicting results, and during the analysis process, we take the historical spot and futures price into account.Firstly, the article illustrates the background of choosing the theme, summarizes some related researches at home and abroad, and for what reason that we think futures price as essential in the forecast process. Then, we use cointegration test and error correction model (ECM) to verify whether there is causal relationship between the spot price and futures price, the conclusion tell us that there is indeed causal relationship between them, the error correction coefficients are significantly less than 0, indicating that when the system deviates from the equilibrium state, the adjustments of the spot price and futures price in the next time have direct impacts on the repair of non-equilibrium state, and both the adjusted effects are negative. On the other hand, the absolute value of adjusted coefficient of spot price is more than that of the futures price, which means that the spot price can return to equilibrium state more quickly. Specifically, when short-term fluctuation deviates from the long-run equilibrium, the spot price will go back to equilibrium with the adjusted strength of(-0.68), and the adjusted force of the futures price is (-0.151).Furthermore, we take use of information share model to find out to what extent do the futures market and spot market make contributions to price discovery of crude oil, which was proposed by Hasbrouck (1995). It is resulted that the spot market and futures market play almost the same roles in price discovery (both close to 50%), many scholars such as Wang qunyong, Zhang xiaotong (2005) used monthly time-series data, they concluded that the futures market contributes 54.27% to price discovery, accordingly, the spot market make a contribution of 45.73%,which means that the futures market makes greater contribution to price discovery, the reason for such a difference is that we use high-frequency data, the difference of daily data among both time series is smaller than that of monthly data.Our ultimate goal is to forecast crude oil price, however, the prediction methods depend on the features of the time series. So before making prediction, it is essential to analyze the characteristics of the time series of spot crude oil price, which includes analyzing the characteristics of the residual series and whether the time series is long memory or short memory. In this section, we make use of the parameter method (GARCH type models) and non-parametric methods (R/S analysis) respectively, both of which reflect the nonlinear fluctuation of the time series, but the R/S method can further determine its memory length, to make preparation for the coming forecast, and only within the memory length, the prediction can be more precise. Some previous studies, during the process of establishing varieties of GARCH models on the crude oil price time series, generally considered the residual series obeying normal distribution, some researches established GARCH-class models on the basis of normal distribution. According to the analysis, we concluded that not only the spot price series but the return series exhibit volatility clustering phenomenon, and the latter one takes on ARCH effects, which will result that the return series no longer subject to normal distribution. In order to describe the distribution characteristics of residuals in the GARCH models more accurately, we use Generalized Error Distribution (GED) to develop various GARCH models, which was proposed by Nelson's. The results of GARCH-class models test show that there is volatility clustering phenomenon on the return series of crude oil spot price, all kinds of GARCH models are significant, indicating that the rise and fall information of oil price have asymmetric impacts on the future price, that is, there is obvious asymmetry phenomenon on the volatility of oil price. The results of R/S analysis show us that the memory length of the spot price time series is 19 days, however, that of the futures price time series is 14 days, which provides a more accurate price basis for the next step-To make prediction using the genetic algorithm and neural network.Next, we will forecast the spot price of crude oil, the forecasting model combines neural networks and genetic algorithms(GA-BP model), this model will be compared with the simple prediction model and error correction model, and the predictive index are the root mean square error (RMSE, also called standard error) and percentage of the average relative error (MAPE). Futher more, I think out-of-sample forecast is important, so I did it. The result tell us the GA-BP model is the best one of predicting oil price, and for oil price, we had better predict it in the short run.Finally, I point out some problems in the prediction of rhe crude oil price, and make a summary of this paper. In the end, I come up with some suggestions in the process of developing the futures market about crude oil in China.