Empirical Research On Bank Credit Risk During The Transitional Period In China

In the economic transitional period in China, commercial banks face with kinds of severe risks, particularly the credit risk that is the most prominent one, therefore, the research on the credit risk of commercial banks becomes the problem that many scholars want to focus on. Credit Risk assessment is a very complicated process, the measurement and estimation of credit risk is the primary work and key process of credit risk management. The prevalent method to evaluate credit risk is to convert the measurement of credit risk into measurement of financial condition in certain enterprise.Credit Risk identification model can also known as the enterprise's financial distress model, the two are essentially the same but the different research purposes. It is called credit risk identification from the bank's point of view, and called the enterprise's financial distress from the perspective of enterprises. Enterprise's financial distress not only has foreboding but also is predictable. Correctly forecasting the financial distress has important practical significance for the protection of the interests of investors and creditors, for the operators to prevent a financial crisis, for the government's management of monitoring the quality of listed companies and stock market risks.According to the above understanding, this paper does some empirical study from many aspects on the credit risk of banks, or enterprise's financial situation. The selection methods and the theoretical basis for the selection of the various indicators used by the establishment of credit risk assessment models were discussed so as to determine the various indicators of risk assessment model more rationally. For that, correlation coefficient, cluster analysis, discriminate analysis, primary component analysis etc. are adopted to select the evaluative indicators of bank credit risk quantitatively. Besides, this paper carried on overview of expert method, neural network method, Z and ZETA score model, CreditMetrics model, KMV model, Credit Portfolio View model, GA and GP model, Credit Risk model etc, and made comments on these methods and models.This paper elaborated on the previous research methods of the financial distress and compared the merit and shortcoming of various methods, investigated the present condition of domestic and oversea scholars' research on financial distress problem and did comparative analysis to each scholar's research method and conclusion. Firstly the various financial indicator data of 1,350 listed companies in both Shanghai and Shenzhen securities exchange station were used, in which we selected 121 companies that had been marked with ST because of "unusual financial situation" and at the same time had six years' financial indicator data before being ST. In accordance with the principle of same period and same size, I selected 121 matching samples by the ratio of 1 to 1 and selected 23 traditional financial ratio indicators classified into four categories. Secondly, Fisher discriminate analysis and Logistic regression model were used in this research to build financial distress predicting model during the period of ST T-1 to ST-6. And used multiple cross-validation to test accuracy rates of the 12 models. After testing, the two kinds of models have better estimate ability in the period of ST T-l to T-4, the accuracy rate is above 65%. But the accuracy rate are all low in the period of ST T-5 to T-6, can not estimate effectively. Thirdly, did comparative analysis for prediction variables. According to the use frequency and status of the variables in the predicting model, we can say that the yield of assets, the ratio of asset turnover, the ratio of debt capital play the most important role in enterprise's financial distress early warning model. The debt-asset ratio, equity ratio and net asset growth rate, and other indicators also have strong predictive ability.The above study is essentially a "static" use of the available financial data. Originally, for every company and every financial indicator,there are a number of data and consistent time-series data can be used, but the static method only use one of these data, we can then imagine that how much useful information would be lost. Based on this consideration, the paper used a method that can study dynamic information implied in the time-series data of financial indicator variables, and the two new variables of the linear change rate and average value are generated by time-series data, both the two new variables are used to substitute for the original data for further statistic analysis, such as discriminate analysis and Logistic regression analysis. In order to distinguish it from the aforesaid "static" method, I called them "dynamic" discriminate analysis and "dynamic" logistic regression analysis. The word "dynamic" is actually referring to use the data "dynamically". Empirical results indicate that this method is successful, having ideal effect.This "dynamic" study firstly used the database of financial information of general listed companies provided by Sinofin Information Services Limited, 241 companies being ST because of "financial situation abnormal" and 1109 un-ST companies before 2006 are selected from A stock market in Shenzhen and Shanghai stock exchange stations. Then selected 122 companies which had six years' financial data before being ST as the financial distress samples from the 241 ST companies; this paper deleted such un-ST companies that had less than six years' financial indicator data due to the difference of periods of available data, the longest 13 years and the shortest only one year, the data of the last 6 years before ST of the remaining 829 companies were chosen, so the total of ST and un-ST are 951 companies, selected 23 traditional financial ratio indicators. Secondly, established and tested the "dynamic" discriminate analysis model and the "dynamic" logistic regression model. The result is that the two modes both have relative high accuracy, accounting for the usefulness of the two dynamic methods and the practical value of the models. Furthermore, from the view of accuracy, the discriminate accuracy for ST in the "dynamic" logistic regression model is higher than the "dynamic" discriminate analysis model no matter whether classified or unclassified. As to the discriminate accuracy for normal companies and overall companies , the two models have no major differences.Our data is actually a panel data, so the various analysis methods can be used to forecast financial distress. Panel data are three-dimensional data due to dependent (be interpreted) variables and the independent (interpretation) variables have one more time factor (axis) to be taken consideration than usually classical multiple regression statistical data, so its statistical analysis is more complicated than the usual methods of multivariate statistical analysis. In the past 20 years, the study on the panel data statistical analysis have developed rapidly because of the need of econometric analysis, which has become indispensable area in econometrics. Therefore, we also tried to establish logistic regression model and probit model both based on panel data, explained and analyzed the empirical results of each model, compared the two models. There are 242 listed companies section members and 6 time section members in Panel Data Analysis Model. 23 financial indicators which were classified into 4 categories were selected as the discriminate parameters of the models so as to study on the dynamics of financial distress .The conclusion is that the estimate probabilities of the two models are very close. In this way, we took logistic regression model based on panel data as an example to test the merits and shortcoming of the model, and did a comparative analysis to the annual logistic regression model. Results can be seen: the forecasting accuracy of the two types are reduced as time flow back, it is because the information content of various financial indicators had declined. The first four years of distress are 65% above the forecasting accuracy rate, in the rest of years which are besides the first four years, the overall accuracy rate in the logistic regression model based on panel data is better than annual logistic regression model; considering the cost of misjudgment, the forecasting accuracies in both models are basically the same; the information content of forecasting indicators in logistic regression model based on panel data is relatively fixed, and the annual one has large randomness.The main innovative research of this paper:1. The premise of Linear Discriminate model assumption is forecasting variables subject to the joint normal distribution, but in reality most of the financial ratio indicators do not meet this requirement, what's more, once having virtual forecasting variables, the assumptions of normal distribution are totally unsubstantiated, hi order to overcome this limitation, some scholars have introduced Logit and the probability ratio (Probit) regression methods. Dealing with the application of the models and research methods, this paper uses linear discriminate analysis, logistic regression model, panel data model, the dynamic discriminate analysis model and other methods.2. Some of the research papers on the financial distress were based on part of(even a small amount) and the same quantity of financial data of typical un-ST and ST companies to establish discriminate model, I think this is not perfect, it is firstly because of the inadequate use of information and secondly because of the distortion of the rules extracted from the data of typical companies ,having no universal significance, hi reality, un-ST companies are eight or nine times as many as ST companies, choosing a few un-ST companies and relatively large quantity of ST companies to build a data prediction model is obviously not universally applicable. So , part of this empirical research used data of all stocks both in the Shanghai and Shenzhen stock exchanges, one ST matches with one un-ST, making the application space as widely as possible.3. The forecast of the financial distress in the past were based on two-dimensional data of different financial indicators in certain year to establish and verify models, this paper considers time-series factors for the establishment of models. The specific approach is to do linear regression for years of data of financial indicators, and use the slope of regression model as a new indicator to establish linear discriminate model. Another approach is to average the years of data of financial indicators to build discriminate model, these two methods effectively solve the problems of establishing dynamic discriminate models.4. The research samples are new, having a long time length and huge capacity. Covering ST companies before 2006,the collection period of sample data has a large span, including the last 6 years' data before the companies being ST. The primary financial indicators cover profitability indicator, liquidity and solvency indicators, asset management capacity indicators, and business development capacity indicators etc, totally 23 financial indicators.5. Using a triple cross-validation method to test predicting accuracy rate of the models, this paper took the average accuracy rate got from cross-certification as the standard of the accuracy of evaluation model, making the accuracy rate we can get more reliable.6. This research established logistic regression model and probit model both based on panel data, explained and analyzed the empirical results of each model and compared the two models.