A Study Of Early-Warning System Of The Stock Market Crisis:Based On Text Mining

The early warning of stock market crisis has always been a hot issue of common concern in academia and industry.With the integration of the global economy and the liberalization of financial markets,the contagion of foreign financial crises has increased.Compared with other countries,the current economic situation at home and abroad is complex,and the downward pressure on China's economy is increasing.Once the stock market crisis occurs,it will cause huge losses to the society.Therefore,in the face of severe domestic environment and complex international environment,it is particularly important to predict and handle the stock market crisis.From the perspective of behavioral finance,this paper introduces some variables to measure investors' psychology in traditional stock market crisis model,such as investor attention,investor sentiment and social interaction based on Internet information.Using BP neural network model based on particle swarm optimization algorithm,this paper constructs an early warning model for stock market crisis based on text mining of Internet information,and proves that it is a better crisis prediction model.Firstly,this paper constructs the signal variables of stock market crisis,and divides the dependent variables into five categories: green light,red light,orange light,yellow light and blue light,which are replaced by 0,1,2,3,4 respectively.Then,according to the theoretical analysis and literature review,we construct four kinds of indicators: macro-economy,financial market,confidence index and Internet information.In particular,this paper uses text mining technology to analyze large amounts of data to constructs behavioral financial indicators such as investor concern,investor sentiment and social interaction based on Internet information,and adds them to the traditional stock market crisis model.Finally,this paper builds four stock market crisis early warning models.Through the comparative analysis of the models,we can draw three conclusions:(1)compared with BP neural network,the early warning effect of BP neural network model based on particle swarm optimization has been greatly improved;(2)the Internet information indicators significantly improve the early warning effect of the stock market crisis warning model;(3)the stock market crisis early warning model based on text mining of network information and PSO-BP neural network algorithm has the best ability of crisis early warning.The accuracy of crisis month is as high as 80%,that of non-crisis month is as high as 100%,and that of the overall sample is as high as 96.3%.