Prediction Of Shanghai Copper Futures Trading Price Based On SV Model And Artificial Neural Network Model

Copper occupies an important position in industrial production because of its high quality.However,China’s copper natural resources are not rich and high-quality enough,so the stable supply of copper is of great significance to China’s industrial production.In order to promote the research of Shanghai copper futures price prediction,the 5-minute high-frequency trading data of Shanghai copper main force from 9: 05 on June 5th,2020 to 1: 00 on March 3rd,2021 and the daily trading data from June 2nd,2006 to June 30 th,2021 are selected to forecast the price of Shanghai copper futures in this paper.The sample data are divided into training set,verification set and test set according to 70%,15% and 15% of the sample length respectively.Selecting the training set and the verification set of the sample data for statistical analysis,it is found that both of the two kinds of data belong to stationary series with the characteristics of peak and thick tail.After random run test,correlation coefficient test and variance ratio test,it is considered that the two markets types have not yet reached weak efficiency,while the correlation of high-frequency data is not very obvious.After that,SV model and artificial neural network models including LSTM,GRU and Self-Attention mechanism are established respectively,and the price of two kinds of data is predicted then.The fluctuation functions of Shanghai copper yield model are estimated by combining SV model with MCMC method here.Comparing the prediction results with the real transaction data in the test set,this paper finds that the prediction ability of SV model is much higher than that of artificial neural network models in high frequency data.For artificial neural network models,the prediction ability and training efficiency of the newly proposed GRU model are higher than those of LSTM model,while the Self-Attention mechanism has ability to promote the stability of them.However,when it comes to low-frequency data,the performances of SV model and artificial neural network models are at the same level.Limited by data length,Self-attention mechanism can’t get good training results for daily price,and can’t give full play to its advantages of stabling forecasting ability then.