| At present,the financial capital market is growing along with China’s economic development regardless,and the soft power of finance is an important part of our national comprehensive national power.As a result,the field of securities investment is the object of attention,whether in academic research or in investment practice.And with the development of computer theory and technology,quantitative investment as an emerging investment approach began to enter the focus of investors,quantitative investment combined with big data and financial theory,in China’s stock market has become a popular investment approach.And having a more high-yield,low-risk investment strategy,both in the private and public sectors,can gain wider and longer-term support.This thesis constructs a quantitative investment strategy using the CSI Pension Index constituents as the research object.The main research is to use machine learning and deep learning algorithms to build stock selection models to select and improve multi-factor quantitative stock selection strategies,and to achieve considerable and stable strategy returns.Multi-factor stock selection is a common strategy in the field of quantitative investment.The core of the strategy is to find the indicators that affect the return and build a portfolio based on the indicators in order to expect to achieve the return of outperforming the index.Its theoretical explanation has evolved with the introduction of machine learning,and multi-factor regression combined with machine learning is widely used in the practice of the strategy.What machine learning does is to sift through existing research for factor datasets and build models to fit them.Specifically,the initial pool of factors for a stock is first screened by factor importance and correlation analysis and the final valid factors are identified to form the final pool of factors.Based on the factor data,stock selection models are then selected.There are two main bodies in the construction of the stock selection model strategy.One is the selection of a model by a multi-factor machine stock selection model in terms of prediction accuracy,and then the screening of stocks.The second is a deep network that learns historical data to make predictions on stock price data and make stock selection model screenings.Then based on the stock selection model,a multi-factor rotation stock selection is made each month to screen out the top performing stocks.The selected outstanding stocks are eventually traded and the output is evaluated by forming trading backtest results through the quantitative platform statistics and calculating relevant evaluation indicators.Based on the above research,this paper uses the data from 2014 to 2021 for factor screening and stock selection model construction,and the data from 2021 to 2022 for trade backtesting to verify the effectiveness of the strategy.The backtest results show that the multi-factor stock selection constructed in this paper is able to obtain good excess returns in a robust manner,and the CNN-LSTM stock selection model performs better under the evaluation of each backtest indicator,and is able to achieve a strategy return of 32% in 2022 despite the general downturn in the financial market,carrying the turbulence of the general environment.In addition,this paper also wanted to further improve the returns based on the stock picking model,so it added broad market timing,and the results showed further improvements in all its backtest indicators. |
