A Study Of Quantitative Timing And Quantitative Stock Selection Based On Machine Learning

Since China's sound financial market,quantitative investment with large data statistics and scientific investment management has been more and more attention.But compared to the Western market,China's quantitative investment is still in its infancy,the development prospects are huge.The traditional quantitative investment model mainly relies on the researcher's financial knowledge and experience to design the indicators and with the economic model or strategy,which makes the low efficiency of the development of the quantitative model and the more impact of human factors.However,machine learning is a major data-driven modeling approach and has a natural advantage in dealing with high noise nonlinear problems.With the explosion of financial data,applying the machine learning to the field of quantitative investment has important significance and prospects.Based on the theory of machine learning,this paper studies two types of the quantitative investment:quantitative timing and quantitative stock selection.And the corresponding quantitative investment models are proposed.By testing the models,it is verified that they have better profitability and risk Control ability,which can provide guidance to investors.The main contributions and innovations of this paper are as follows:Firstly,aiming at the shortcomings of the traditional feedforward artificial neural network in dealing with financial timing data,this paper proposes a quantitative timing model based on recurrent neural network.It is verified that it has better profitability and risk control ability in different periods of stock markets compared with neural network.Secondly,based on the classical cooperative filter in the field of recommendation system,an Alpha cooperative Kalman filter model is proposed for financial asset modeling.This model solves two problems when the cooperative filter models financial assets:One is that it lacks the ability to model dynamic problems;the other is that it can't model the excess returns of financial assets.Thirdly,aiming at the shortcomings of the traditional linear regression method in mining the potential factor variables of stock price,this paper proposes a factor variable mining method based on Alpha cooperative Kalman filter model.And combined with SVR model,a new quantitative stock selection model is proposed.In the seven years of testing,it is verified that it has better profitability and risk control ability.