Research On Quantitative Stock Selection And Timing Strategies Based On Genetic Programming And Transforme

The quantitative trading strategy is becoming popular in China’s investment market owing to its advantages such as intelligent decision,automation and interpretability in the background of China’s financial reform continues to deepen.Many researchers are devoted to building quantitative trading strategies with higher returns and greater stability.In the real-world investment,the general investment logic of investors is to select high-quality stocks in the pool according to the fundamental and technical indicators.And then they choose a time to make decisions such as buying and selling according to the information of the trend of the stock,so that to obtain higher and more stable excess returns.According to the logic of real-life investment,this paper proposes the quantitative stock selection strategy based on genetic planning,the quantitative timing strategy based on Transformer and the combination strategy based on both of them.Their effectiveness is verified through theoretical analysis and empirical research.The quantitative stock selection strategy based on genetic programming is divided into two steps:factor digging and strategy construction.The factor digging is modeled as a symbolic regression problem.And the relationship between the current factors,factors series and the reporting period’s return series is fitted by the genetic programming algorithm,so as to construct new factors with strong correlation with the reporting period’s return.The new multi-factor stock selection strategy is constructed and back-tested by these new factors.The Transformer-based quantitative timing strategy is also divided into two steps:stock price time series fitting and trading signal recognition.After verifying that Transformer has better generalization ability than other time series models,the trading signal is recognized by predicting the closing price trend.Selling if the stock price is predicted to fall in the future,and buying otherwise.On this basis,a combination strategy of the two is constructed,completing the study of the whole process of trading strategy from stock selection to timing.The backtesting experiments show that the stock selection strategy is better than the FF five-factor model;the timing strategy has excellent ability to obtain excess return but has higher risk;the combination strategy can well avoid the high risk problem of timing strategy and obtain excess return better than both of the two strategies mentioned above.Through the above research,we provide theoretical basis and guidelines for investors’ real investment in the whole perspective of stock selection,timing and portfolio strategies.