| With the rapid development of computer technology and the rapid improvement of computing power and automation level,the financial industry has received tremendous development assistance and greatly promoted the development of quantitative investment.Compared with traditional investment methods,quantitative investment will pay more attention to the influence of mathematical measurement on investment behavior,and integrate these influential parameters into factors to form a factor library.In recent years,a variety of factors with better effects have emerged on the market,such as Alpha factors,CTA factors,etc.Strategies based on these factors are also emerging in endlessly.However,after years of dynamic games between strategies and the market,artificial quantitative strategies have exposed more and more limitations.In recent years,reinforcement learning has made significant progress in fields such as electronic games,chess,and decision control.Considering the game nature and regularity of the financial market,more and more researchers apply reinforcement learning to the field of quantitative investment.Based on the development status of the financial industry,this article uses reinforcement learning to carry out quantitative investment.By building an excellent portfolio model,it aims to seize opportunities in massive data,advance and retreat freely in different markets,and allow investors to obtain more substantial returns.This thesis mainly completes the following work:1.In view of the large amount of noise and redundant data in financial data,this thesis uses financial transaction indicators to represent,effectively realizing the training and learning of the model.The stock feature indicators and macro market indicators are used for auxiliary characterization,and these indicators are processed through the LSTM neural network and then applied to the investment model.Experimental results show that LSTM-based eigenvalues can greatly increase the sensitivity of investment models to financial market transactions,thereby improving profitability and anti-risk capabilities.2.For the investment portfolio model of quantitative trading,this thesis uses the traditional equal-weight allocation model for model training,and finds that it has certain limitations in controlling the size of funds,range of action,and stock correlation.Therefore,on this basis,use the mean-variance model to build a weight-based portfolio model.The experimental results show that the weight-based portfolio model has a stronger ability to control the size of funds and action range,and is more scalable,which helps to further improve the model’s income and anti-risk capabilities.3.In view of the frequent fluctuations in the trading environment,this thesis uses the research on the value function in reinforcement learning to realize the adaptive improvement of the model to effectively respond to different market environments.In this thesis,two improvement schemes,continuous and discrete,are proposed,both of which can effectively increase the rate of return,so that the trading model can balance the benefits and risks in different environments,and explore better trading strategies.This thesis applies the idea of deep reinforcement learning,cyclic neural network and modularization to the quantitative trading of investment portfolios,represents the market state through a more scalable portfolio model,optimizes the value function,and makes it choose more Good strategy for maximum profit.The experimental results show that this investment portfolio model has achieved an annualized return of more than 30%,which has good practical value and research significance. |
