The Logistic Regression Model Applied To Credit Risk Measurement

The credit risk is the most primary risk faced by commercial banks, and has critical influence on the sustained and sound operation of banks. Since 1990s, financial liberalization and financial innovation develope vigorously and credit derivatives have been widely adopted, which makes the commercial banks face more severe credit risk. Therefore, how to effectively manage the credit risk based on accurate measurement is a most challenging issue to the commercial banks.Through analyzing the status of the measurement and management of credit risk in our country, we can find that the means of credit risk measurement are still backward, and our commercial banks mainly adopt qualitative and simple quantitative method, which cause the weak credit risk management and make our commercial banks face serious credit risk. For this situation, this article firstly reviewed the international widely used credit risk measurement models, and then analyzed the applicability of models combined with the actual situation in our country at this stage. By analyzing the application scope, data requirements, forecast accuracy and stability of each model, we find the logistic regression model is more suitable for the actual situation of our country at this stage in terms of data input and assumptions. So, this article finally chooses the logistic regression method to establish a credit risk measurement model and carries out an empirical test to the effects of this model. Due to lack of relevant theoretical guidance, academician ususally use their subjective judgments to select the model input indicators when they use logistic regression method to establish credit risk measurement model, resulting in the appear of redundancy indexes or missing the crucial indexes. To solve this problem, based on the information entropy theory, this article puts forward a method to objectively select the input parameters of logistic model.This paper takes all the listed companies in our country as the research sample in the empirical research, abandoning the usual pairing method in previous research, and based on this basis, this paper uses the information entropy theory to objectively measure the prediction ability of financial indicators, and then selects the first 11 indicators which have the most strongest prediction ability as input variables to construct the credit risk measurement model based on the logistic regression, and finally carries out an empirical test to the model. The results showed that the net assets per share, operating profit margin and other 11 absolute financial indicators have strong predictive ability on whether the companies breach a contract; This credit risk prediction model based on logistic regression is more stable and has higher forecast accuracy. The overall accuracy was about 90%. Therefore, the model can be used to assess and predict the credit risk.In addition, in order to test whether there is an improvement between credit risk measurement models based on the method used in this paper and traditional methods, this paper established the Fisher discriminant model by use of stepwise discriminant method, and carried out a comprehensive comparison on the prediction results of the credit risk measurement models based on two methods, then we found that whether on the overall prediction accuracy rate or the probability of two types of errors, the credit risk measurement model established on the method in this paper is comprehensively better than the Fisher model established on traditional methods.