Reasearch On Stock Price Prediction And Quantitative Stock Selection Based On CNN-LSTM

With the rapid development of China's capital market and the gradual improvement of residents' income,the stock market has attracted more and more investors to participate in it.Especially under the influence of the COVID-19 in 2020,the relatively loose monetary policy to boost the economy has made the stock market perform well,and the stock market has attracted a large number of new investors.Data show that as of December 2020,China A share investors total177 million households,the average market value of the household reached 450,000,the new investors in the whole year 18021,100 households,an increase of 36.02%.The large number of participants in the stock market indicates that the market trading is active.However,individual investors have the tendency of irrational speculation in the trading,and blindly chasing after the rise and killing the fall is not conducive to the healthy development of the stock market.If we can use technical means to predict the trend of stock prices and provide investment suggestions for investors,then this will promote the sound development of the capital market and improve the return level of investors,which has a strong theoretical and practical significance.In this paper,deep learning neural network algorithm is introduced into the trend prediction of stock price,and two neural network architectures with different principles,CNN and LSTM,are combined to predict the rise and fall of stock price.First of all,to the Shanghai 50 index from 2011 to 2020,the frequency of the original transaction data for data preprocessing,the sliding window scroll way to generate the data samples to train neural network model,and through the comparison on the accuracy of the prediction model under different input variables to determine the final input variables in this paper,we study model,are including basic trading stocks index,the Shanghai 50 index,index and technique index stocks,a total of 20 characteristic factors.Secondly,several control experimental groups were set.By changing the internal parameter combinations of the model,the influence of different parameters on the prediction effect of the model was studied,and the parameter combination with the best prediction accuracy was selected,and the prediction effect of the CNN-LSTM model under the parameter combination was compared with that of BP model,CNN model and LSTM model.Finally,the market state during the backtest period is divided into bull and bear markets,and the rise and fall probability predicted by CNN-LSTM model is used as the condition for stock selection to construct quantitative stock selection strategy and conduct backtest.Moreover,the performance of the strategy under the bull and bear market conditions is analyzed.The empirical conclusions of this paper prove that:(1)CNN-LSTM prediction model has a good performance in predicting the rise and fall of Shanghai Stock Index components.Compared with the single model with other neural network architectures,the combination of the two architectures can improve the prediction ability of the model.(2)The quantitative stock selection strategy based on CNN-LSTM model can achieve significantly higher returns than the benchmark under different market conditions,indicating that the model output probability of rise and fall can be regarded as a stock selection factor to obtain excess returns,and the probability of rise and fall can be used as an effective method of quantitative stock selection,which has investment reference value for investors in stock selection.(3)China's stock market has not yet reached the weak efficient market.Since the input variables of the model studied in this paper include individual stock trading data and related technical indicators,the trading strategy based on this model can obtain excess returns in the market,indicating that stock prices do not fully reflect all historical information,and profits can be made in the market by mining and processing the historical data of stock prices.