Research On The Delisting Risk Warning Methods Of Listed Companies Based On SVM

The delisting risk warning of listed company is the core content of the securities market’s delisting mechanism. Building and designing a scientific and effective method of delisting risk warning is very important for improving the securities market exit mechanism and protecting stakeholders’legitimate rights and interests, improving the quality of listed companies and reducing the market risk.Support vector machine as a new technology in the field of machine learning have good performance in solving Pattern recognition of small sample size, nonlinear and multi-dimensional, and also have learning ability as well as generalization ability.therefore, this paper attempts to construct and design a delisting risk warning method that based on support vector machine (SVM) which is suitable to China’s securities market, aims at studying the delisting risk warning of China’s listed CorporationThe article takes China’s main board and small board of listed manufacturing companies for example and uses the data of164*ST sample groups which was first exchanges in delisting risk warning and164groups of paired samples which is normal in2004to2013to make matlab simulation experiments based on SVM.Finally,we found that t-2years’delisting risk Warning precision reached between90.625%to93.75%, t-3years’delisting risk warning accuracy is between66.41%to75.781%, t- 4years’delisting risk warning accuracy is between67.9688%to71.875%. The risk warning method of SVM for all three-year overall haves a good performance. At the same time, we also study the impact on the risk warning results of multi-collinearity problem of initial warning indicator. The experimental results show that principal component analysis to the early warning index will decrease the predictive accuracy of SVM method, the multi-collinearity of warning index does not adversely affect to the SVM method. In the end, the paper studys the influence to the risk warning results of two heuristic parameter optimization method—GA genetic algorithm and particle swarm algorithm PSO. Through the comparison of the experimental results, we found that the GA genetic algorithm has the best performance in parameters optimization of SVM delisting risk warning methods.