An Empirical Study Of Commodity Futures Cross-species Arbitrage Based On GARCH-LSTM Combination Model

With the rapid development of China's economy,the trading scale of bulk commodities is expanding day by day,which not only promotes the domestic and foreign trade,but also makes the futures market more prosperous.Commodity futures arbitrage has been widely concerned by investors and scholars.As more and more commodity futures are open to market,there are more and more varieties with correlation,so we can carry out arbitrage according to the long-term equilibrium relationship between different futures.In addition,the development of computer and machine learning in recent decades has opened up a new way for futures investment strategy.The characteristics of the futures market,such as high investment risk and leverage trading,make it different from the common commodity market,and the futures price is usually non-linear and non-stationary.If we use the traditional time series model to predict it,we will inevitably encounter various difficulties.In recent years,deep learning is a relatively new research direction in the field of machine learning.It has a good performance in fitting a variety of highly nonlinear data,which greatly promotes the application of artificial intelligence technology in the financial market.Firstly,this paper takes soybean and soybean oil futures as an example,selects the sample interval from January 4,2010 to November 22,2019,and takes the daily closing price data of soybean and soybean oil futures listed in Dalian Commodity Exchange as the research object,combining the traditional time series model GARCH model and deep learning LSTM neural network to carry out empirical analysis on the cross species arbitrage strategy of commodity futures.Firstly,the long-term equilibrium relationship between soybean and soybean oil futures is verified by co integration test.Because the residual of co integration regression equation has conditional heteroscedasticity,GARCH model is used to fit the data.According to the average equation of GARCH model,the portfolio is established,and the conditional heteroscedasticity sequence of GARCH model is used as an input feature of LSTM neural network to predict the price difference sequence between soybean and soybean oil futures,and the trading strategy is empirically studied by combining the prediction results with statistical arbitrage theory.Two arbitrage strategies are designed based onfixed threshold and mean reversion of price difference,and the results are compared with other cross commodity arbitrage strategies.Finally,the cross breed arbitrage strategy based on GARCH-LSTM combination model is applied to corn,corn starch,coke and screw steel futures to verify the general applicability of arbitrage strategy.The first innovation of this paper is to take into account the characteristics of volatility clustering of futures prices,use GARCH model to describe the variance time-varying of the residual of cointegration model,and use the obtained conditional variance sequence as an input feature of LSTM neural network to predict the price difference sequence between cross varieties of futures,in order to improve the prediction ability of LSTM neural network;the second is to combine GARCH model to build Based on the GARCH-LSTM portfolio model,an arbitrage strategy is designed by using the forecast results of GARCH-LSTM portfolio model.The experimental results show that the prediction effect of the LSTM neural network with GARCH estimation is better than that of the general LSTM neural network,and also better than the traditional BP neural network and RNN cyclic neural network.Through the empirical analysis of the arbitrage strategy,it is proved that the arbitrage strategy based on GARCH-LSTM portfolio model has a good performance in mining the arbitrage opportunity and obtaining the income,and has a good general applicability in the different commodity futures' cross arbitrage.Finally,through the comparison with the cross commodity arbitrage strategy based on co integration,GARCH model and LSTM neural network,it is found that the arbitrage strategy based on GARCH-LSTM model is feasible and effective,which shows advantages in both winning rate and investment return.