Research On Stock Data Mining Based On Event-Study Method

The stock has already become an important component of modern investment with the development of economy and transformation of people's investment consciousness. it has significant theory meaning and application value on analysis of the inherent law of the stock market.This paper will analysis the stocks' information which security analysts recommend based on the data mining and analysis the event that security analysts recommend stocks by event-study. The event-study method was used to research the impact of events , such as stock trading volume and price. When a certain incident influenced notably,the abnormal rate of return and abnormal trading volumes have specific distribution at the moment of incident. A certain incident is judged to occur while the abnormal value unconventionality change before the incident. According to different tendencies of abnormal value curve before the incident ,the different directions of the stock price can be judged after the incident. This paper will achieve these two goals by using classification method and clustering method .The main jobs of the thesis are as followed: Firstly,research on the the characteristic of the stocks that security analysts recommend by event-study.Two models are used to estimate the normal values of recommended stocks.According to the characteristics of the recommended stocks,the goals of classification and clustering analysis are proposed. One object is finding out the stocks that accord with recommended stocks' characteristic by the classification before the recommendatory information issue.The other is finding out the stocks that price rise continuously by clustering the stocks using the data before the information issue. Seconedly,two-pattern classifiers are constructed by using two kinds of classifying methods according to abnormal trading volume and abnormal earning ratio of recommended stocks.They are used to find out the stocks that accord with recommended stocks' characteristic 2 days before the recommended information publishing. The experimental results indicate that this methoe of stock-selection could get better income.Finally, proposed the method of clustering the data of abnormal trading volume and abnormal earning ratio of recommended stocks before the recommended information issue .Results show that the price of some stocks in one cluster keep growing after recommended information issue. This information is significant to stock investor.