Research On The Selection Of Ratios In The Distress Prediction Of Listed Companies

This paper systematically studied the selection of ratios in Distress Prediction Model.27 ST companies between 2007 and 2009 were selected as training samples and 20 ST companies in 2010 as prediction samples, control samples were randomly selected by 1:3 and expanded the period of datasets to T-4. The whole modeling process was divided into two phrases, including baseline model and ratios selection model and especially in the ratios selection model, we used four main methods to choose the proper ratios and built distress prediction model separately, finally we compared and analyzed the effect and influence of different prediction models.Through the two-stage experimental design, the process of ratios selection was proved to be necessary and important. On the basis of experiment, we compared four main methods of ratios selection, including T test, VIF test, Stepwise regression and PCA, the improvement on the prediction of early-warning model of the four methods were testified. This paper analyzed and summarized the pros and cons of each method, screened the most suitable ratios selection method to Logistic Regression Model-VIF test. Otherwise this paper discussed the necessity of ratios selection in different stages of financial crisis. Through the ratios selection in different stages, this paper found the different driving factors in different stages of crisis and consequently built the multi-stages dynamic Distress Prediction Model.The main contribution of this paper was providing the experimental evidences on the necessity of ratios selection and guidelines of method selection, especially this paper used forecast rate of accuracy instead of classification rate of accuracy which was different from most of other papers. This paper also focused on the research to find the proper ratios selection method to the main modeling method-Logistic Regression and provided the evaluation standards. Finally we discussed the problem of ratios selection in different stages and built the multi-stages dynamic Distress Prediction Model.