Pairing Trading Strategy Of CSI 300 Constituent Stocks Based On Deep Learning

With the vigorous development of my country’s economy,the development of the securities market has become increasingly mature and various systems have been gradually improved.After a series of demonstrations and experimental preparations,the margin financing and securities lending system officially entered the market operation stage in my country’s securities market in 2010,which has changed my country’s For many years,the securities market can only "one-sided transactions",securities investors have more investment options.With the development of margin financing and securities lending business,various quantitative investment strategies have gradually begun to attract attention.Among them,pairing trading strategy is one of the hot spots.Pairing trading began to be applied by Wall Street in the 1980s.It is a neutral trading strategy that can be applied to bull markets and bear markets.The main idea of this strategy is to find two underlying assets that are closely related and tend to have the same long-term trend.After the price deviation of the pair of underlying assets,long the undervalued asset and short the overvalued asset.When the price of the underlying asset returns to the long-term trend again,the position is closed,and the price difference generated between transactions is used to obtain profits.The realization of this strategic process is based on the analysis,calculation and prediction of financial time series data,and financial time series data usually show significant nonlinear and non-stationary characteristics,so it has strong processing ability for financial nonlinear data.Deep learning methods are gradually applied to the processing of financial time series data with complex features.Based on the comparative analysis of various methods of statistical arbitrage in pair trading,this paper selects cointegration theory to screen paired stocks,and selects deep learning with strong adaptive processing ability for nonlinear data according to the significant nonlinear and non-stationary characteristics of financial time series.The LSTM neural network is combined with the trading strategy to form a paired trading strategy model based on deep learning.The research takes the constituent stocks of my country’s CSI 300 Index as the research object,and predicts the price difference of paired stocks by constructing an LSTM neural network model.Combined with the output results of the neural network,the empirical analysis of the pairing trading strategy is carried out,and a control group is set up with the BP neural network as the model for prediction.The results show that the prediction results of the LSTM network are better than the BP neural network model,and the pairing trading strategy combined with the prediction of the LSTM neural network is feasible and effective.At the end of this paper,the effectiveness and feasibility of using high-frequency data to implement pair trading strategy are studied and discussed.