Case-Study Of Model-based Early Warning Listed Companies In Financial Stress

Firstly, summary and analysis of classical researches on early warning of listed companies in financial stress are made in this thesis. Secondly, detailed research is made on sample design, selection of prediction variables and statistical methods. On the basis of prediction abilities by different models, support vector machine is used to analyze the internal mode of the data in prediction of financially stressed companies and applied to predict financial distress of Chinese listed companies. The software of NeuroSolutions is used to build SVM model with three layers in which kernel Adatron algorithm is chosen and default of parameters is used in the software.In case study, 30 financially stressed companies and 30 healthy companies are included in training samples and finally 6 financial ratios are chosen as input variables among 14 financial ratios to predict financial stress. The precision of SVM is compared with those of BP neural network and Logistic regression in case study. The result shows that the effectiveness of Logistic model and BP neural network is generally equal. The precision of support vector machine is generally higher than those of BP neural network and Logistic regression by 2.77% and 3.57%. The conclusion also shows that as the training sample becomes smaller, the precision of support vector machine still keeps a high level.