A Multi-factor Stock Pricing Model Based On Neural Network

Along with the continuous evolution of Chinese financial market,the role of quantitative investment in China has become more prominent.The multi-factor stock selection model has been playing a important role in the quantitative investment,because of its controllable risk,large strategic capacity,and stable excess returns.The foundation of the multi-factor selection model is that there is an approximately linear relationship between the factors and the expected return.However,short-selling method in China is scarce and costly.And the approximate linear relationship cannot guarantee the accuracy of selected high-yield stocks in the portfolio.In short,the traditional multifactor stock selection model has a large problem in how to construct a stable income portfolio in the Chinese market.Neural network model,as a classifier's algorithm in machine learning,has the advantages of non-linear fitting,strong adaptability,and good universality.This makes it possible to improve the multi-factor selection model through the neural network model.This paper first analyzes the empirical and flaws of the traditional multi-factor stock selection model,and then determines that the main evaluation method is accuracy and return of the high-yield group.In the second part,it uses the neural network model for single factor(priceearnings ratio)and multi-factor(relative value).And then it analyzed and determined the specific training process,and model feasibility of the neural network model.Finally,the entire network of financial data is processed using a neural network model through classification and integration.The results show that the neural network model can effectively improve the utilization rate of factor data,improve the correct rate of high-yield groups,enhance the profitability of strategies,and have certain guiding significance for actual investment.