Study On Alpha Portfolio Based On AdaBoost Algorithm

This thesis mainly studies the Alpha protfolio.We use seven broad factors,which are basic factor,mometum factor,technogy factor,flow factor,quality factor,valuation factor and growth factor to train AdaBoost.Then we build a factor system with the factor passing validity test.Since people are more sensitive to the loss than the same amount of income,a muti-classification cost sensitive AdaBoost algorithm is used.And,it is improved in the process of practice.We use the entire A share stocks as the stock pool and divide it according to the Wind first level industry standard.Then,AdaBoost algorithm is used to select high-quality stocks in each industry and form portfolio to hold.At the same time,selling the same market value of China Stock Exchange 500 stock index future.The initial amount of funds is determined to be ”1” in the process of analog holding.In order to reduce the deviation of hedging,the proportion of funds in various industries keep the same with the China Stock Exchange 500 stock index.Equities in the same industry are allocated equally and changed every month.In order to test the rationality of stock selection by AdaBoost algorithm,we analog holding the quality stocks and inferior stocks selected by AdaBoost alorithm.And holding the China Stock Exchange 500 stock index to make a contrast.When 1 month for stock training cycle,the basic factor has the best performing.But its performance is still inferior to stock selection in the same period.Therefore,considering increasing the holding period.When the training period is 9 months,it achieved an annual yield of 18% in all A shares,which is relatively outstanding in the same period of stock selection level.The construction of factor system and the application of AdaBoost algorithm in Alpha stock selection at this thesis have reference significance.And it fully illustrates the necessity and superiority of combining machine learning with financial investment.