The Stock Selection Factor Design Based On The Announcement Information Of Listed Companies

In the stock market,stock prices are always affected by some information or events which result to some abnormal volatility.This phenomenon is called "event shock" in the stock market.Among them,the more frequent and more influential ones are the announcements issued by listed companies.Digital information only takes up a very small part,but text information accounts for the vast majority of the space in the announcements,and managers have a higher discretion when arranging text information,so that the text has more information content,Therefore,combining the event analysis and text mining related to announcements can fully discover the incremental information of the company's operating status,which helps investors to quickly and effectively understand the announcements event and make adjustments to investment decisions promptly.This research comprehensively analyzes the impact and the sentiment of listed company announcements by the method of event analysis and text mining.Analyzing several major types of announcements(including: performance forecast,performance express,shareholder changes,private placement,equity incentives,dividends,employee shareholdings),determining the event date,testing the abnormal returns and accumulative abnormal returns,Distinguish different impact types and according to the effects to classify them into three types: information leakage,short-term impact,continuous impact,Design an event factor considering to the event effects and stock realized returns comprehensively.Then crawling,storing,and extracting keywords of the announcement texts,contrast the sentiment index by three methods: naive Bayes method,word frequency method and word vector method,design a sentiment factor by the cosine similarity of the word vectors.We can incorporate event effects into the multi-factor system by the event factor and emotion factor extracted from the announcements.This paper select the constituent stocks of CSI 500(000905.SH),Perform a single factor test of each period in 975 trading days from 2015 to the end of 2018,Under the weekly adjustment frequency,the IC of the event factor and the emotion factor is 0.009 and 0.006,and it is significant at the 0.05% level,the annualized alpha of the event factor and emotion factor is 19.3% and 5%,Under the monthly adjustment frequency,the IC of the event factor and the emotion factor is 0.013 and 0.008,the annualized alpha of the event factor and emotion factor is 11.3% and 4%.In general,both factors can provide positive alpha returns,which will decrease when the adjustment period increases,this indicates that the shortterm effects of such factors are stronger than the long-term continuous effects,this phenomenon is consistent with the effect of most announcement events.Use the equal weight method to combine these two factors and set trading parameters to make backtest closer to real situation.We construct a portfolio holding stocks by regular adjustment,during the backtest period,the volatility is reduced about half comparing with the benchmark,which beta is 0.48 and annual alpha is 5%,by analyzing the style factor attribution of this portfolio: the correlation with SMB(scale factor)was high and the correlation with HML(value factor)and CMA(investment factor)was negative before July 2016,after a change of market style,the correlations are gradually approaching zero,the style exposure is gradually reduced.Finally we point out the improvements and prospects in accordance with the deficiency of this paper.