The Study On Financial Crisis Prewarning Model Based On Kernel Fisher Discriminant

While the dog-eat-dog market economy is pregnant with the opportunity for development, there are countless risk and crisis hidden in it. As for the listed companies, it grows more and more seriously that the ones are titled with"ST"plate and even forced to quit listing due to financial crisis. It is not only the threat to survival and development of the companies in the financial crisis, but also brings enormous loss to investors and creditors. Therefore it has already not merely been an academic problem to establish an effective practical pre-warning model on financial crisis with the content of interests'increasingly urgent needs, but also become realistically significant and influenced sound development of capital market of our country.This paper proposes a non-linear classification method—Kernel Fisher Discriminant to resolve the problem, which is the non-linear extension of the linear FLD. The sample data is the selection of financial statistics three years before of listed companies titled with ST in 2005 and 2006 as well as the matching sample. First of all, the K-S test and Mann-Whitney U test is used to choose the financial indexes which have a significant difference between the two groups. Base on this analysis, PCA is applied on the chosen indexes and then forms the principal factor as the input of the KFD. By the discriminant built by the KFD in the high dimension space, it can solve the non-linear problem in the original space, which has an approving effect. Then as a contrast, this paper uses the same sample to build FLD and ANN model respectively; and rough set is used as another way to choose indexes for the KFD. The result indicates that the non-linear methods is better, and based on the view of total error cost, the combined model of PCA and KFD has a better effect than other methods. At last, the applied cost and foreground of this model is analyzed.