Quantitative stock selection: A Comparative Study of Model and Strategies under Multiple Economic Objectives

In this paper, we are going to build an investment portfolio that mainly use the linear regression model, neural network model and XGboost model combine with the Markowitz Mean Variance theory to predict the rotation forecasting return rate and volatility rate via machine learning algorithm, thereby obtaining the strategy of adjust weights of portfolio. We found that the return rate of strategy of minimum risk will be maximized by Neural Network algorithm, the portfolio's possible losses are also minimal, making it suitable for a stable investment approach. XGboost model resulting maximum Sharpe ratio strategy has a large yield, while its volatility is also large so that it may suffer losses. Therefore, it is suitable for risky investment. The rate of maximum return strategy and the volatility in the Linear Regression model have outstanding advantages over the other two models, making it becoming a helpful way to gain the main income source. By analyzing the yield and volatility results brought by different models under different market cycles, this article aims to provide certain reference suggestions for investors to make cyclical rotation decisions when constructing their investment portfolios.