Multi-Step Crude Oil Future Price Forecasting Based On Deep Leaming

As the material basis for human survival,crude oil is one of the most important energy and bulk commodities in the world.As the most populous country in the world,China is also the largest crude oil importer in the world.The rapid economic development is accompanied by the rapid consumption of oil.Therefore,the domestic crude oil supply and demand relationship,as well as the dependence and risk of oil prices on foreign energy,are increasing.The rise in international crude oil prices will lead to an increase in the cost of China’s industrial and manufacturing industry,which in turn will lead to an increase in commodity prices,trigger inflation,and lower the level of real national income.On this basis,the tax revenue of the government of China will decline,while foreign exchange expenditure will rise,which will increase the government’s budget deficit,and even more,may threaten national energy security.Therefore,accurate prediction of crude oil futures prices is an important means to avoid the significant negative impact of severe fluctuations in crude oil prices on the economic operation of countries and enterprises.However,due to the many and complex factors affecting crude oil prices,accurate forecasting of crude oil prices is still an extremely difficult and complex issue.The fluctuation of international crude oil prices is not only affected by the well-known market supply and demand relationship,but also by many other factors,such as bad weather,regional military conflicts,market speculation,international major events,and so on.With the maturity of the futures market and the continuous development of big data technology,a large amount of information will emerge every day,which will have a certain impact on investors’ investment sentiment and thus affect the crude oil futures market.Therefore,this thesis obtains cumulative sentiment scores sequence through sentiment analysis of crude oil futures news headline data and uses them as influencing factors to predict crude oil futures prices.In addition,this thesis proposes a hybrid oil price prediction model based on decomposition-ensemble strategy and deep learning.This thesis mainly uses the idea of decomposition-ensemble to predict the crude oil futures price by combining the sentiment analysis with the deep learning model optimized by the swarm intelligence algorithm.First,through preprocessing the news headline data and performing sentiment analysis on it,a cumulative sentiment score sequence is obtained.Secondly,an adaptive signal decomposition method,ensemble empirical mode decomposition(EEMD),is used to decompose historical crude oil futures price data into several components to reduce the impact of noise.Third,a seagull optimization algorithm(SOA)is introduced to tune the hyperparameters of gated recurrent unit(GRU).The optimized GRU model is established to acquire the predicting values of each component integrated with the cumulative sentiment score sequence.Subsequently,multiple linear regression(MLR)is then introduced as the ensemble approach that integrates the forecasting results of each component.The empirical results show that,in general,EMD,SOA and GRU have better performance and higher prediction accuracy than EEMD,WOA and LSTM.Multiple linear regression can further improve the prediction accuracy than simple linear summation.However,there is no significant difference between the prediction accuracy of GRU and LSTM.When only the historical data of crude oil futures prices are used for prediction,the conclusion of the data decomposition method is just the opposite,that is,EEMD has better performance than EMD.In addition,this thesis also uses the WTI crude oil futures price data from 2000 to June2021 to qualitatively analyze the impact of the black swan event on crude oil price fluctuations since the 21 st century.Each sharp increase or decrease in the low-frequency component corresponds to the impact of international remarkable events and black swan events.The trend of the high-frequency components has a small amplitude,which describes the market’s shortselling effect and short-term fluctuations.The residual slowly varies around the long-term mean;thus,it is regarded as a long-term trend of crude oil price evolution and represents the price changes caused by supply and demand in the economic sense.In order to statistically test the effectiveness of the hybrid model based on decompositionensemble and deep learning,the Diebold–Mariano(DM)test is carried out at the end of this thesis.The test results show that EMD,SOA,and GRU generally have more advantages and higher prediction accuracy compared to EEMD,WOA,and LSTM when combined with cumulative emotional score sequences,and multiple linear regression can further improve prediction accuracy compared to simple linear aggregation.The proposed model is a reasonable and effective tool for nonlinear and unstable time series data such as crude oil futures price.