Dynamic Investment Strategy Based On LSTM Neural Network And CPT

With the improvement of material level and the increasing convenience of financial investment channels,the willingness of ordinary people to invest has become stronger and stronger.However,there are many factors that affect the fluctuation of stock price,it is difficult for ordinary investors to identify the high-quality stocks from asset pool.This paper is aim to design a set of trading strategies including stock selection and portfolio optimization for individual investors,and to construct an optimal factors collection for investors.This paper comprehensively selected 205 candidate factors,covering macro economy,market,corporate behavior,corporate financial status,investor sentiment and institutional investor attention,and prediction error and so on.Firstly,this paper screened the efficient factors by comparing the correlation coefficient between stock return and factors,and the return difference between the head portfolio and tail equity portfolio.Then,we used the clustering method to delete the redundant factor.The best stock-selecting factors is made up of 30 factors,including the gold price yield,the S&P 500 index volatility,liquidity index,beta and research attention,PS,long-term capital yield et al.Secondly,this paper established a stock-selecting model based on LSTM neural network.By comparing the prediction accuracy,error loss,the percentage of recall and precision,we selected the optimal model under different parameters.The empirical results showed that the model trained in this paper has higher accuracy and lower error loss than the existing literatures.Finally,this paper studied the portfolio optimization under the CPT.This paper also considered the influence of dynamic reference point,transaction cost and multi-period investment on investors’ decision-making behavior,which is closer to the actual market situation than the existing research.In this paper,we selected the shares of the Shanghai stock exchange from 2016 to 2019 as samples to test this investment strategy.The optimal stock-selecting factors collection is input as the characteristic variable into the model,and the portfolio optimization method proposed in this paper is used to construct the investment portfolio of this trading strategy.By comparing the return rate between this investment portfolio and the benchmark portfolio,it is found that the annualized return rate of the portfolio constructed is more than 50%,which higher than market portfolio.Trading strategies have earned excess returns for investors in more than 75% of trading months.