A Study On Alpha Strategy's Quantitative Investment

With the rapid development of Internet,big data and artificial intelligence technology in China,artificial intelligence has been widely applied in biological gene sequencing,automatic driving,commercial retail and other aspects.As the core technology of artificial intelligence,machine learning is gradually known and recognized.Machine learning can automatically mine important information from a large amount of data and carry out automat ic analysis,providing users with a problem solving algorithm,which plays a very important role in solving various personal biases in handling problems : affective bias and cognitive bias.Therefore,in the field of finance,machine learning is increasingly favored and valued by researchers and practitioners.From the perspective of fundamental analysis and research,this paper mainly selects six fundamental factors as the target factor and PB as the prediction target to establish a regression model,and then makes a more accurate prediction of the theoretical price-to-book ratio of the selected stocks in the stock pool.By examining the comparison of training results in different stock pools and comparing with the traditional training methods with return rate as the target variable,the feasibility and applicability of this training method based on PB as the target factor is investigated.On this basis,this paper also studies the applicability of different weight allocation methods to different learners and the improvement of different weight allocation methods to performance indicators by improving the risk function and constraint conditions of markowitz model,and finally obtains the comprehensive performance of the strategy combination.Based on in this paper,the results obtained by machine learning training performance analysis and strategy combination available: based on the city net rate as the objective factor which a machine learning training method has certain feasibility and its suitable stock pool,for the csi 300 stock pool and targeted yield for the traditional factor of training method has superiority,to reduce the dimension of input factors get relatively good learning;The fitting of fundamental factors with machine learning learner can obtain better fitting degree of learner than traditional linear fitting learner to a great extent.For all learning,macovei hereby model based on Var risk function with macovei hereby model based on risk Cvar function can obtain better than combination weights allocation and so on performance,but the two kinds of weight optimization method based on the two different risk function superiority and inferiority of some form of strategy is not obvious,for different learning,there are risks of different function weight optimization method can make the performance better;For different learners,whether based on Var risk function or Cvar risk function,the addition of constraints based on the optimal performance can improve the performance of the constructed alpha strategy.