Based On LSTM Neural Network Research On Arbitrage Strategy Of Non-ferrous Metal Futures

With the rapid development of the futures market,the structure of participants in the futures market and the correlation between varieties are becoming more and more complex.It is increasingly difficult for investors to obtain stable and considerable income through cross variety arbitrage trading.It is often difficult to rely on the traditional measurement model to obtain more accurate prediction data,so as to obtain stable income.The long-term and short-term memory model can effectively deal with the problem of massive non-stationary data and achieve good prediction accuracy.It is suitable for the non-ferrous metal futures market with excessive speculation,high turnover rate and frequent intra day fluctuations.Therefore,this paper introduces the LSTM neural network model into the cross variety arbitrage field of non-ferrous metal futures,so as to design effective arbitrage strategies and obtain stable and considerable profits.At present,the main theories of futures cross variety arbitrage strategy design are divided into two categories: traditional econometric theory and neural network theory.The model used in traditional econometrics can fit a stable time series,but there are some problems in practical application,such as over fitting,weak generalization ability and too ideal assumptions.Considering the characteristics of high frequency,large amplitude of volatility and the simultaneous development of both long and short in the futures market,the futures price,as a time series data,does not have the stability required by traditional econometrics in practice.The neural network model has the advantages of high precision,self adaptation and massive data processing in the field of non-stationary time series prediction,and the data characteristics of futures market also highly meet the requirements of neural network.Therefore,based on the relevant theory of cross variety arbitrage strategy design,it is feasible to use LSTM neural network model to design cross variety arbitrage strategy.The design of cross variety trading strategy is mainly divided into two parts: inter variety spread prediction and trading signal formulation.Spread prediction not only needs to ensure the cointegration between varieties,but also needs to fully consider the investment characteristics of the futures market in the model construction stage.This paper makes a comparative analysis on the correlation and fundamental correlation between non-ferrous metal varieties.The results show that Shanghai copper futures and Shanghai aluminum futures conform to the principle of cross variety arbitrage.This paper selects the closing price of daily K-line data of Shanghai copper futures and Shanghai aluminum futures from january4,2011 to december31,2021,determines the ratio coefficient of arbitrage portfolio according to the cointegration theory,and constructs an LSTM neural network model to predict the price difference of arbitrage portfolio,At the same time,adjust the forgetting gate structure and loss function of the LSTM model to make the model more suitable for the actual investment in the futures market,gradually traverse and determine the super parameters of the model,and obtain the prediction data that performs better than the traditional measurement model.When the forecast data is obtained,the threshold breaking principle is more applicable to the formulation of arbitrage strategy trading signals than other methods,and the degree of strategic risk preference can be flexibly controlled through the threshold adjustment.Taking the difference between the predicted value of the LSTM model and the actual price difference as the starting point,when the difference exceeds the threshold,the transaction corresponding to the signal is carried out,and the settings of the latent signal and the exposed signal are used to screen out the occasional fluctuations so as to reduce the loss.The back test results show that the cross variety arbitrage strategy of Shanghai copper,Shanghai aluminum based on the LSTM model is feasible.It can control the maximum retreat within 9% while obtaining 43%of the income,and the winning rate is close to 50%.Considering the shortcomings of single model input and lack of fundamental analysis support,this paper optimizes the arbitrage strategy of single variable LSTM neural network model from the perspective of technical analysis and fundamental analysis.The minimum price,maximum price,position and trading volume equivalent factors of the main linked contract of Shanghai copper and Shanghai aluminum futures and the fundamental factors such as registered warehouse receipt,inventory,GDP growth rate and industrial production growth rate are selected,and the multivariable LSTM model is constructed with the price factor,fundamental factor and closing price as the model inputs respectively to obtain the corresponding predicted value of price difference,and the back test is carried out using the trading strategy of threshold breakthrough,The results show that the price volume factor optimization strategy can slightly increase revenue.According to the neural network model and the investment characteristics of the futures market,the strategy designed in this paper is applicable to the stable and disciplined investors with stable supply and demand of the trading target.At the same time,investors need to have a certain understanding of the trading target and the LSTM model when applying this strategy,and can adjust the model parameters with time and personal risk preference.