Research On International Crude Oil Price Forecast And Driving Mechanism Under Multiple Predictors

Crude oil is the core part of the energy issue,not only as an important material material for social production and life in agriculture,industry and transportation,but also as an important strategic resource for guaranteeing national energy security,political security and maintaining national discourse.However,in response to the current accelerated global energy transition trend,the emergence of alternative innovative energy sources,the frequent trading of crude oil derivatives,the high-speed flow of financial capital and constant geopolitical friction,the market supply and demand balance is constantly broken and readjusted,the structure of the international crude oil market shows new characteristics,and the oil price sequence once again enters the mode of uncertain fluctuations and significant ups and downs.As the international oil price movements become more and more closely related to macroeconomic and financial markets,it is the most interesting and complex hot research topic in the financial and economic academia at present to grasp the fluctuation pattern of the international oil price system more finely and precisely,to analyze the driving mechanism of the crude oil market in depth,and especially to pay attention to the time-varying characteristics of oil price forecast indicators in different periods.This not only provides directional indications for market participants’ production and investment,and improves the early warning and foresight ability to deal with the risk of shocks,but also has important reference value for policy makers to adjust energy strategies more scientifically in order to maintain the sustainable and stable development of China’s oil industry and even the economy and society.This paper theoretically proposes a novel combined framework for driving mechanism and price system analysis and forecasting of crude oil market.First,considering the emergence of new structural features in crude oil market,more complete and comprehensive oil price forecasting information is integrated in order to construct a screening set of oil price forecasting indicators.Second,an innovative efficiency-based two-stage model is proposed to summarize the driving mechanism laws of the international crude oil market in depth.In particular,the dynamic transmission mechanism between various risk factors and crude oil prices is exposed under the time-varying environment.Finally,feature selection methods with different advantages are adopted as comparative models,which are incorporated into forecasting algorithms to perform oil price forecasting respectively,focusing on breaking through the difficult problem of oil price change trend forecasting.A study of multivariate analytical framework of international crude oil price forecasting indicators.In response to the new structural characteristics of the crude oil market gradually,the overly simple fundamental factors of supply and demand are no longer sufficient to explain the complexity of the crude oil market,and new influencing factors are highlighted to bring drastic impact and influence to the international crude oil market.This paper further analyzes and examines in detail the forecasting indicators for the integrated crude oil market by systematically combing previous studies and constructs a multivariate analytical framework covering eight different dimensions(demand,supply,inventories,financial markets,macroeconomics,economic policy uncertainty,geopolitical risks and technical indicators),integrating a total of 92 forecasting indicators.By incorporating as many predictors as possible into the indicator filter set,it is possible to include more complete and comprehensive oil price forecasting information,and more interpretable for abnormal crude oil price fluctuations.Exploration of crude oil price driving mechanism based on GCT-SFA model.This paper innovatively proposes a two-stage model based on efficiency from the dual perspectives of statistical efficiency and technical efficiency-combining Granger Causality Test(GCT)and Stochastic Frontier Analysis(SFA)in two stages to comprehensively evaluate and filter many indicators,and finally obtain the optimal and most efficient subset of indicators for prediction.Meanwhile,in order to verify the superiority of the two-stage model in variable selection,six feature selection methods recursive feature elimination(RFE),lasso regression(LASSO),elastic net regression,stepwise regression,Bayesian model average(BMA)and dynamic model average(DMA)are used as competitive models.First,the GCT-SFA model is used to detect and analyze the predictive ability of the drivers of the eight dimensions on the future trend of oil price system evolution.It is empirically found that among the crude oil market driving mechanisms financial market factors,macroeconomic factors and demand factors have outstanding predictive ability and are the dominant drivers affecting the trend of oil price movement during the current sampling period,and the inclusion of these indicators can significantly improve the accuracy of oil price prediction.In addition,the time-varying characteristics of crude oil market driving mechanisms are explored through the quarterly selection percentages of each category throughout the sampling period by an efficiency-based two-stage model.It is found that different forecast indicators provide different perspectives of forecast information at different points in time,and their forecasting power is enhanced or weakened accordingly depending on the specific period,and the time-varying characteristics of the crude oil market driving mechanism are significantly present.Crude oil price forecasting based on the context of driving mechanism analysis.Based on the different subsets of indicators generated by each of the two-stage model and the competing model after evaluating and screening the set of forecast indicators,they are incorporated into four machine learning forecasting algorithms,linear regression(LR),artificial neural network(ANN),support vector regression(SVR)and random forest(RF),which are combined to construct a combined research framework for oil price forecasting to verify that the two-stage oil price forecasting model is useful for portraying oil price The significant superiority of the two-stage oil price forecasting model to characterize the evolutionary trend of oil prices.Three loss functions,mean absolute error(MAE),root mean square error(RMSE)and mean absolute percentage error(MAPE),are provided and combined with two statistical tests,ANOVA and Friedman’s test,for a comprehensive and systematic comparative analysis.The empirical results show that the efficiency-based two-stage oil price forecasting model can obtain more interpretable and high-precision forecasting results compared with the competing models,confirming that the two-stage model is good at analyzing the key drivers of the international crude oil market,uncovering clear and specific price driving mechanisms,and improving the input structure of the machine learning forecasting model by effectively reducing the set of forecasting indicators to reveal the evolution trend and volatility of the oil price system.The model is designed to improve the input structure of machine learning forecasting models and reveal the trend and volatility of oil price system.