Research On Multi-factor Quantitative Investment Strategy Based On Dynamic Neural Network

Quantitative investment refers to the method of trading profits by using mathematical methods,establishing models on financial data,and then issuing trading orders through programming It has been nearly 70 years since it was first proposed.Compared with the past securities investment methods,quantitative investment is more rational,and faster.Nowadays,with the development of deep learning,quantitative investment theory is also developing at an accelerating pace,because there are many factors affecting the security market,which are not completely rational,and it is a non-linear system.However,neural network has unique advantages in nonlinear prediction,so it is more accurate than traditional linear regression prediction.This thesis mainly discusses the application of dynamic neural network in stock market forecasting and quantitative investment strategy.This thesis firstly explain the concept of quantitative investment and a brief introduction to the advantages and development trend is made,and then expounds the rational and irrational factors in the securities market,and then by getting the data of A shares pingan bank from May 2018 to November 2020 and the data is processed to generate ten representative technical analysis indicators,the data were verified for authenticity,outliers were removed and standardized,and then used as the input of multiple linear regression,BP neural network and long and short-term memory network respectively to predict the rise and fall of stocks.By comparing the performance of the three models in the prediction accuracy,it is proved that the neural network is better in the multi-factor model,and LSTM network is better than BP neural network.Although the BP neural network and LSTM neural network prediction accuracy reached 60.5% and 65% respectively,but through the simulated trading after the annualized yield is still not enough to make it into practice,so this thesis proposed an improved prediction method based on the turning point and trading strategy,further improve the prediction accuracy and simulated trading annualized yield.Then,the model is applied to Shanghai Pudong Development Bank,Datong Coal Industry and CSI 300 Index,and the results show that the model has universal application value.Through the prediction comparison experiment and improvement experiment of the three models and the extended application,the following conclusions are finally drawn: 1.The prediction effect of BP neural network with multiple index factors is better than that of multiple linear regression.2.LSTM model has better effect,and the effect is the best among the three models.Through the construction of trading strategies based on the prediction results of the three models,LSTM model has the best annual return rate in the backtest,but it still needs to be further improved.3.LSTM was used for model classification,and the prediction accuracy was further improved to 83.33%.A quantitative investment strategy was developed according to the classification results,and the annual return rate was increased to11.23%.4.The prediction model based on turning points has universal applicability.