| For four decades of the economic reform and opening,trade between China and abroad has been active,capital and liquidity has been enhanced and people’s living standards have been improved,which has provided favorable conditions for the financial market.In recent years,a large number of institutional and individual investors investing in stocks,and analysts’ analysis reports on the industry and listed companies have gradually aroused extensive attention.Along with the development of market economy and growing,Securities analyst provide relevant information of industry and listed companies to all investors by publishing reports,at the same time give investment advice for decision-making.They also maintain a good relationship with the listed company to catch the relevant information on time,and update the investment advice,increase earnings with lower risk.However,the level of Chinese analysts is uneven,and the content of the analysis report does not have corresponding supervision and restrictions,which cannot play an effective and active role in guiding individual investors who blindly follow the trend of stocks and do not know much about financial transactions.Reasonable calculation with the analyst report,optimizing the analysis report data and summarizing the consensus of analysts,can provide more objective and comprehensive reference basis for investors and help them to understand market information more comprehensively and objectively.In this paper,earning per share in analyst reports from 2014 to 2018 are taken as sample data,combined with the widely popularized machine learning method,focus on calculating,optimizing and seeking more accurate prediction results on the earning per share estimates.In this paper,estimates were classified with 4 characteristics: updating time,the number of following up brokers,brokers updating times,and historical prediction accuracy.In the different situations,comparing the impacts of the accuracy in four kinds of classification of data.By using the multivariate regression,neural network,random forests,support vector regression(SVR)for listed companies in 2018 earning per share forecast,compared with the actual results,the prediction method of multiple regression equation and support vector machine have the highest accuracy.The prediction results of random forest and arithmetic mean are similar,but the neural network is not stable.The conclusion is that the predicted results obtained by machine learning method are better than the arithmetic mean of all the analyzed data. |
