Empirical Analysis Of Prediction For Enterprise Financial Crisis Based On Logistic Model

Modern China is now in the middle of a golden period for development, facing chances and dangers at the same time. Under circumstances of unpredictable internal and external economic environment, as well as fierce competition among enterprises, how to effectively monitor financial risks in case crisis happens is very important for self-protection of companies, virtuous cycle of bank loan, and macro-control of government organization. Based on such thought, prediction models for enterprise financial crisis established from empirical data by Logistic regression were introduced in this paper so as to promote research of related topics.Firstly, manufacture companies suffered from financial crisis were defined as the main body for sample design, referred to the research requirement and the industrial feature among domestic listed companies. Secondly, an indicator system of financial ratios was established from previous research and authoritative resources. After that, new data disclosed from annual reports of 150 specially treated (ST) stocks and their comparative Non-ST stocks between 2005 and 2007 were collected and calculated, which covered 22 financial indicator items and dated from the past two and three years prior to the specially treated time of ST stocks. As was followed, indicators with remarkable capacity to tell ST stocks from Non-ST ones were screened out through a between-group difference analysis. Then, principal component analysis (PCA) and a method which practically modified PCA were comparatively applied to eliminate multicollinearity effect among indicators and reduce dimension, respectively resulting in principal components and typical indicators to be processed by Logistic Regression. After treated by Backward Method, the best modelling variables were selected out, and after one more regression, the prediction model was obtained. Test showed that, the model derived from 2-year before ST by PCA had a forecast accuracy of 95.8% for internal samples and 93.3% for external samples, while the one by modified method 97.5% and 90.0%; the model derived from 3-year before ST by PCA had a forecast accuracy of 76.7% for internal samples and 70.0% for external samples, while the one by modified method 76.7% and 66.7%.Results displayed that, the forecast models established in this paper were authentic and effective; moreover, the model derived from 2-year before ST had excellent prediction effect, while that from 3-year manifested poor accuracy, which showed the heavy influence of time factor over models; then some financial ratio indicators were found to have more outstanding forecast capacity than other ones; finally, the over-laboring PCA method could be substituted to some extent by the modified method which is simple to apply and worth further promotion.