Multi-factor Quantitative Trading Strategies Combined With Machine Learning Algorithms

Since 2014,following the officially launch of Shanghai-Hong Kong Stock Connect,Shenzhen-Hong Kong Stock Connect and Shanghai London,the stocks from China mainland market successively have been included in the MSCI index and the FTSE Russell Index and repeatedly increased the percentage of inclusion factors.With the accelerated pace of China's capital market opening up and along with the changes in the structure of A-share market investors,the overall investment style of the A-share market will gradually mature.On the other hand,with the increase quantity of institutional investors,the proportion of individual investors in the A-share market continues to decline.On the other hand,for individual stocks,they should pay close attention to the flow of the main funds and actively adjust their investment strategy based on tracking the institutional investors' weight shares in order to share the benefits of institutions' investment research results.Based on the above investment thinking,this paper attempts to verify the effectiveness of the quantitative trading model combined of fundamental factor and the capital flow factors in the A-share market.Through the fundamental factor to adhere to the value investment strategy,through the capital flow factor to share the investment research results of professional investment institutions,to build a quantitative trading model and back-test the historical data and exploring the performance of the model in the A-share market.In the framework of the empirical model of this paper,the first step is to select the fundamental factor and the capital flow factor.The main work content is to determine the set of candidate factors in order to obtain sufficient factor data for the model to use.Secondly,the data of the factor values are preprocessed and then focus on raw data processing,de-extreme processing,normalization processing and neutralization processing.The purpose is to avoid excessive interference of the noise in the original data.The third step is to use the income analysis and IC analysis to test the validity of each single factor.The purpose is to judge the quality of each factor,select a factor that has a logical meaning and can effectively distinguish individual stocks,so that the factor value has a certain predictive ability for the future earnings of individual stocks.Last but not least,to use multiple linear regression and support vector regression algorithms to construct a multi-factor quantitative trading model combining the fundamental factor and the capital flow factor.Combine the selected fundamental factor and capital flow factor as input variables,meanwhile,use the market value of individual stocks as output variables to form a training set.By inputting the training set data to the machine learning algorithm,thereby outputting the theoretical market value.When the actual market value of individual stocks is lower than the theoretical market value,it can be considered that the stock market value does not reach the expected value.The more the actual value deviates downward from the theoretical value,the more serious the undervalued stock is,and the more likely it is to return to a reasonable market value in the future which means greater arbitrage.Therefore,purchasing the stocks that market value with most serious deviation in each adjustment cycle.The results of this paper is to take the fund flow factor into consideration.The empirical results show that whether it is a multiple linear regression or a support vector regression algorithm,the strategy of adding the capital flow factor back testing results is better than the strategy without the capital flow factor,and the capital flow factor can provide significant excess returns.The combination of the capital flow factor and the fundamental factor,two different styles of factors,can explain the market value of individual stocks in more dimensions and improve the overall return of the multi-factor strategy model.In addition,the machine learning algorithm is introduced in the quantitative trading model.By back testing the real historical data of the CSI,the model using the support vector regression algorithm has achieved a good rate of return.The results show that machine learning can be well used in quantitative transactions.