Research On Returns Prediction Of Crude Oil Market From The Big Data Perspective

The changes of Crude oil price have important influence on financial markets and entity economies.Particularly,the central bank regard crude oil prices as one of the important variables to evaluate macroeconomic risks and generate macroeconomic inference.A sharp drop in oil prices will exacerbate market panic and stock market volatility.Soaring oil prices will lead to an increase in the cost of industrial production and bring about an international financial crisis.Therefore,the academia has always investigated more advanced statistical models to predict crude oil returns.There are many literatures related to crude oil forecasts,but most literatures are confined to traditional model which can only contain the same frequency data.However,the availability of various types of high-frequency and mixed-frequency data in data-rich environment brings new challenges to crude oil forecasts.Thus,this paper further discusses whether the Lasso-MIDAS model can help improve the predictive accuracy of crude oil returns in data-rich environment.In empirical study,this paper employees68 mixed-frequency predictors to model and predict oil returns of WTI based on the Lasso-MIDAS model,and introduces the ridge-MIDAS model,the PCA-MIDAS model,the ASPCA-MIDAS model,the Group-Lasso-MIDAS model,the standard MIDAS model,and two combined prediction methods to compare the out-of-sample forecasting performance with the Lasso-MIDAS model.Additionally,we evaluate the out-of-sample forecasting performance of the Lasso-MIDAS model and different competing models by out-of-sample~2,test of Direction-of-change,forecast encompassing test,the cumulative sum of the squared predictive error difference and economic performance.The results show that the Lasso-MIDAS model has higher predictive accuracy in both statistical and economic significance.The robustness check and further analysis also suggest that the Lasso-MIDAS model exhibits the highest predictive accuracy among all forecasting models.In terms of theoretical innovation,this paper designs a new algorithm to estimate the Lasso-MIDAS model based on maximum a posteriori estimation(MAP).The new algorithm makes up for the deficiency of existing algorithms in efficiency and application.In terms of practical innovation,this paper further extends the application field of Lasso-MIDAS model,and enrich the literature related to oil forecasts.Particularly,we consider a large-scaled mixed frequency data in modelling,which can help select more valuable information,thus improve the short-term predictive accuracy of WTI returns.