Empirical Study On Financial Crisis Prediction Of Listed Companies Based On Support Vector Machines

Forcasting of enterprises'financial crisis is one of the oldest problems in money market, and is also one of the hottest researches worthy of being discussed. the research on financial distress is of practical importance and is also a good reference for managers,operation, loaners,stockholders and other beneficiaries to assess credit and investment of the company. With the increasing number of distress companies in the stock markets in China, this paper, based on the financial distress situation in Chinese market, finds a better financial distress model for the ST companies by using support vector machines and principal components analysis, aiming to find if this kind of model without special hypothesis has a better performance when compared with the statistic methods such as MDA, Logit and so on.At first, this paper introduces relevant literature in and abroad, presents the meaning to analyze financial crisis of listed companies, introduces and compares existing relevant theory and model on financial crisis alarming. On the base of those, together with information of listed companies in the year 2004 which is marked as ST at the first time in 2006 in Shanghai and Shenzhen, this paper at first uses PCA to map the original dataset to a lower feature space omitting some information unuseless, and then uses the SVM model to map the dataset to a higher feature space to find a separating hyper plane to classify the 2 kinds of listed companies.According to the demonstration of experiment, the Support Vector Machines model has a higher forecast precision of warning the financial crisis.The utilization of PCA to map the riginal dataset to a lower feature space can shorten the SVM model's length of computing time, and comtributes a lot to increase the forcast precision of this model.The result is that any special hypothesis are not needed in this method. So SVM model permances better than other traditional statistic models.