Research On Financial Crisis Early Warning Model Forlisted Companies Based On GA-RBF

Until July in 2009, the number of listed companies in China has been more than 1,600, and the accumulation of financing for businesses in the domestic capital market has been more than 54,000 billion Yuan. However, study form an overall perspective, the business performance of listed companies in China can not be optimistic.The outstanding problem is that there is a rising trend in the number of deficit listed companies and the loss is larger. As to listed company, if it is falling into the financial crisis, not 'only endangers their own survival and development, but also brings huge losses to investors and creditors. Therefore, the financial crisis early-warning model research is a crucial work.This paper selects 120 listed companies as sample, including 40 Special treatment (ST) companies and 80 non-ST companies. Collectes the T-2 year data and T-3 year data of these companies, totally 240 records. Then divides them into two groups:180 (including 120 non-ST and 60 ST) as a training sample, the remaining 60 (40 non-ST and 20 ST) as the test sample. The main work of this paper are:Firstly, researches the data characteristics of the financial indicators of the sample companies with SPSS 17 statistical analysis software, makes a significance testing on the 21 financial indicators of the ST and non-ST companies, and gets 12 key indicators. Secondly, simplifies the wariables and avoides multi-collinearity effects by factor analysis, and finally selectes six factors to establish Logistic regression model. Thirdly, establishes and achieves the financial early warning model based on Back Propagation (BP) neural network with Matlab9.0 software. Fourth, by analyzing and comparing the performance of BP neural network and Radial Basis Function (RBF) neural network, finds the training speed and generalization ability of RBF neural network are better than BP neural network. Fifth, introduces RBF neural network into the financial early-warning model, and optimizes RBF neural network using genetic algorithm.This thesis first proposes the financial early-warning model for listed companies based on genetic algorithm optimization RBF neural network. The test results with the same sample showes that:the sample forecast accuracy rates tested by Logistic regression model, BP neural network model, and GA-RBF model respectively are 73.3%,80%,91.7%; and the number of iterations trained by GA-RBF neural network is 45, is lower than that 116 by the BP neural network. The results showed that: compared with the financial early-warning model based on traditional Logistic regression, or the financial early-warning model based on BP neural network, the model based on genetic optimization RBF neural network has a higher discriminant accuracy, reliability and application value. Finally, this paper discussed the limitations and further research possibilities of this study.