An Empirical Study Of Dynamic Financial Early Warning Model Under The Concept Of Discrimination

At present, our country’s economy is facing severe pressure and challenge. Chinese listed companies get losses and face the financial trouble, even get into survival crisis constantly because of the comprehensive factors including the industry’s internal management and external business environment and macroeconomic. If a company can’t predict financial crisis and adjust management strategy timely, after the loss of enterprise, it will not only reduce the income from investment of investors and creditors inevitably, but also affect the economic interests and the future development of the enterprise itself. Therefore, the research of listed companies’ financial status and development trend has practical significance.The study for the financial early warning has been last for decades, the early warning model which can adapt to different assumptions is established by many scholars from different angles during these years. But in how to improve the accuracy of the early warning model by making full use of financial data time continuity, there still has so many research gaps, so finding a reasonable approach to use the information factors into the model in order to building the dynamic financial early warning model is the core of the research in recent years. Obviously from a loss of listed companies to eventually be special treatment even to be forced to retreat city is not formed overnight, but a process of accumulate and deteriorate over a long period. Only using a period of financial situation to predict the crisis is too arbitrary,and the manifestation of different crisis stage is different, we can’t use a model accurately detected samples of all the crisis stage, thus building a early warning system that can distinguish between different levels of crisis and dynamically analyze the crisis situation is necessary.The 71 crisis samples and the 410 normal samples which is four years’ panel data is used in the empirical study in this article. A pure financial indicators dynamic early warning system and the comprehensive indicators dynamic early warning system contained non-financial factors is established based on the CHAID(Chi-Square Automatic Interaction Detection)model. By comparing the model regression judgment accuracy in the two systems, it’s foundthat the warning system after the introduction of market information and ownership structure has a higher prediction accuracy. In order to enhance the practicability of the early warning system and make sure that the enterprises can identify its crisis degree as soon as possible,this paper puts forward the concept of differentiation model, and according to the characteristics of the CHAID model to improve the algorithm of the distinguish degree.By comparing the model regression judgment accuracy in the two systems, it is found that the early warning system after the introduction of market information and ownership structure has a better prediction accuracy. So the introduction of the non-financial factors in the model is necessary and meaningful, it can improve the predicted precision of the early warning model, especially when the long-term prediction is done, the optimization effect is larger. In the comprehensive indicators early warning systems, the number of crisis sample that mistakenly divided into the normal sample is totally 11 when classifying the test set according to the concept of the degree of differentiation, decreased by 26.67% comparing with the conventional direct classification; the classification effect of the early warning system which is improved by the degree of differentiation is the best, crisis samples are divided into the normal sample, the total number of crisis sample that mistakenly divided into the normal sample is 2, decreased by 86.67% compared with the conventional direct classification.The characteristics of dynamic early warning system established in this paper is showed as follows:(1) using CHAID model to classify, this algorithm can analyze the complex interaction between variables, from multiple variables automatically select the most significant difference variable combination,on the basis of statistical significance to determine dividing point, deal with the col-linearity data and does not require sample data normally distributed; CHAID results in a tree, it is a simple and intuitive description of a crisis situation and more likely to be understand by those enterprise decision makers who do not have professional knowledge and analysis skill;(2) from the perspective of practical application to put forward the concept of the degree of differentiation between the model for determining the final classification, besides of improving the accuracy of the crisis prediction, we can get a sample of the crisis degree, which is advantageous for the enterprise managers to improve its’ business strategy as soon as possible, it also improve the accuracy and practicability of the early warning model.