Correlation Analysis And Prediction On Crude Oil Prices Series In Different Oil Markets Based On Empirical Mode Decomposition

The crude oil, as a special strategic resources, has been playing a more and more important role in Chinese ever increasingly-open economy.With the rapid and sustained growth of Chinese economy, the increasing dependence on foreign oil and the strong upheaval of oil market have made a big challengen for China.The international price of crude oil has influenced Chinese domestic crude oil price more tremendously ever since 1998 when domestic crude oil pricing menchanism was internationalized. In recent years, international oil price fluctuates violently and frequently, which has tremendously impacted the development of such a big oil-consuming country as China and has become one of those factors for instability of macroeconomic of the country. Therefore, it is very important to monitor, compare and predict the fluctuations of oil prices in both domestic and foreign markets.This thesis proposes a different crude oil market's price correlation analysis method based on EMD decomposition. First, the crude oil price series of different crude oil market are decomposed into several components (including a set of intrinsic mode functions and a residual component) with different frequencies by EMD, extracting market fluctuation,major events and trend components respectively. Secondly, the short-term dependency and interaction of the extracted three components are analyzed, applying the integration theory, Granger causality test based on vector autoregressive and error correction model, to further study the long-term equilibrium relationship and short-term fluctuation model of the financial crisis impact on different crude oil market. And finally, the results of quantitative analysis are summarized and the policy implications of this study are briefly analyzed.It is indeed challenging to predict crude oil price because crude oil price series are non-linear and non-stationary. The traditional statistical and economic models are built on the linear assumption, so it is difficult to capture the non-linear model hidden in the crude oil price sequence, thus it is difficult to get the accurate prediction of crude oil price. Computational intelligence methods such as Artificial Neural Network (ANN), Support Vector Machines (SVMs) and Genetic Programming (GP) are applied to predict crude oil price so as to overcome the limitations of traditional statistical and econometric models. Experimental results show that the computational intelligence method is superior to the traditional statistical and econometric models in the prediction accuracy.To predict crude oil price accurately, the thesis proposes a prediction method to combine EMD(Empirical Mode Decomposition) and SVMs(Support Vector Machine). This method applies EMD to decompose the crude oil price series into several components with different frequencies, which are then composed into three sub series according to their frequency level. The three new sequences represent the market fluctuation price, major event price and trend price respectively. Different SVMs models are constructed for the three new sequences mentioned above to predict the final value of each sequence respectively. And the final predictive value can be obtained by constructing combined model using the predictive values of the three sequences with SVMs. The validity of the method is verified by WTI and Brent crude oil price data, and the result shows that this method has higher prediction accuracy by comparing to SVMs model or artificial neural network model.