Research On EEMD-XGBoost Decomposition Integrated Oil Price Forecasting Model Based On Financial Market Index And Macroeconomic Index

In recent years,the financial attributes of international oil price and the volatility of international oil prices has increased significantly.For countries,companies and investors who carry out statistical arbitrage based on crude oil prices,a reasonable model which can predict international crude oil prices more accurately is of great significance.This article focuses on WTI crude oil prices,combines the domestic and foreign literature of WTI crude oil price forecasting methods,and finds that existing studies mainly use the decomposition-integration model to characterize the nonlinear and nonstationary characteristics of crude oil prices.The decomposition-integration methods are commonly used following three steps: Firstly,use decomposition algorithm to decompose the crude oil price.Secondly,use machine learning algorithms to predict the obtained sub-sequences separately.Thirdly,integrate the predicted values of the obtained sub-sequences to obtain the final prediction result.There are three main types of defects in this type of model.First is the large amount of calculation,because the decomposition algorithm usually obtains more than ten subsequences which takes more time.The second is that these forecasts are only based on the original sequence and its decomposition sequence.However,the fluctuation of crude oil prices is also affected by other market and macroeconomic factors.The third is that most of the integrated algorithms currently used are SVR and neural network.SVR can only handle the case with a small number of features,while the neural network requires a large data set,which does not apply to the situation in this article.In order to overcome the above shortcomings,this paper proposes an EEMD-XGBoost decomposition integrated model that considers crude oil fundamental indicators,financial markets and macroeconomic indicators.Firstly,directly putting the lag value of the decomposed sub-sequence as a feature into the integrated model can greatly reduce the complexity of the model to overcome defect one.Secondly,select indicators that reflect the supply and demand of the crude oil market,financial markets,and macroeconomic indicators into the model to improve the accuracy of crude oil price forecasts to overcome the second defect.Finally,the XGBoost method is used as an integrated algorithm,which can handle small data sets with multiple features to overcome defect three.This paper selects WTI crude oil price data and 15 indicators derived from crude oil fundamentals,financial markets and the macro economy,and uses the proposed EEMD-XGBoost model to carry out the following empirical research.Firstly,the prediction results of the XGBoost model,which only uses the lag value of the original crude oil price series as the feature,are compared with the prediction results of the SVR and LSTM methods commonly used in the existing literature.It is found that the performance of the XGBoost model is better than that of the SVR and LSTM models in the data set used in this paper.Secondly,compared the prediction results of XGBoost model with the mixed original sequence lag value and the decomposition sequence lag value(i.e.,EEMD-XGBoost model)and the XGBoost model with the original crude oil price sequence lag value as the feature only,it is found that the delay value of the EEMD decomposition sequence can indeed improve the prediction effect of the model.Finally,the prediction results of the model adding crude oil fundamentals,financial market and macroeconomic indicators one by one on the basis of the EEMD-XGBoost model are compared with the prediction results of the EEMD-XGBoost model,and it is found that different indicators have different effects on improving the prediction accuracy of the model when predicting in single step and multi-step.In the end,the thesis builds the investment strategy based on the proposed model.The empirical results show that the crude oil price prediction results based on the proposed model can effectively guide the buying and selling of USO fund,improve the return rate of the fund and reduce the maximum retracement.