PTA Futures Price Prediction Research Based On GRU-LSTM Hybrid Model

In economic and social development everywhere is closely related to commodities in every areas,commodity price changes will not only directly affects the industry chain in the enterprise’s economic benefit,but also become the mainstream trend of globalization and increasingly fierce competition in the international community,holding on commodity pricing power and reasonable price is expected to be in a country’s foreign trade plays a significant effect.Therefore,futures will play a more and more important role in the process of economic and financial development.There are numerous and complex factors affecting futures prices.In the era of big data,in order to extract more valuable information effectively,it is necessary to adopt certain technical means,and deep learning model provides the possibility for this demand.In this paper,PTA was selected as the research object.The sample data came from the Wind database,including the fundamental factors of PTA including five subcategories of supply,demand,price,inventory and macroeconomic factors,a total of29 indicators and 22 technical indicators.Time series data of 1734 days from October30,2013 to December 7,2020.In terms of prediction methods,this paper selects four basic neural network models: BP neural network,RNN neural network,long and short memory neural network(LSTM)and gated recurrent unit neural network(GRU)were built on this basis to improve the prediction accuracy,and Mean Squared Error(MSE)was used to represent the prediction accuracy of the model.Finally through the empirical analysis of the advantages and disadvantages of different models to predict results,this paper found out the optimal prediction model and got the following main conclusions: PTA futures price there is a certain degree of predictability,can make use of its goods through appropriate modeling unique fundamental information for a rational forecast of its future price.GRU-LSTM hybrid neural network model has the best performance in predicting the price of PTA futures.RNN model and LSTM model have a small mean square error in predicting the price of PTA futures,but there is always the problem of time delay.The hybrid neural network model designed in parallel with GRU and LSTM effectively solves this problem.And the prediction effect of the model will become worse after the add of technical indicatorsData selection,processing and model construction are all very important steps in forecasting PTA futures price.In empirical test,the research of this paper has certain reference value for processing these links.On the model construction,this paper mainly proposed the construction of GRU-LSTM hybrid neural network to choose the single-step prediction network structure applied to the PTA futures price prediction and achieved good forecasting effect,so it has a certain practical significance for the industry chain customers to use futures commodity futures speculative hedging.