| 'New Basel Capital Accord'is a guidance documents for commercial bank risk management. Credit risk management is the important content. According'Baselâ…¡'requirements, base on a comprehensive summary of existing measurement methods of credit management, writer used a new measurement technology available for commercial banks credit rating system, a Adequacy test. Meanwhile, writer proposed a new prediction method for measuring the probability of default, the core parameter of credit risk management.The main contents of this paper and important conclusion:1. This article develops a new model called the ordinal responsive panel data model which can be used to test and analyze the adequacy of credit rating system. In the separation the factors occur together of the sample individividual during the same period, the article found that the ordinal responsive panel data model, which contains a random intercept and coefficient to random coefficients, is the most reasonable test model. After testing, we have concluded:First, the existing system indeed includes redundant indicators. The existence of these indicators, in large part because subjective factors. Their presence affects the accuracy of credit ratings. Removing these indicators will be better for enhance the objectivity of the credit rating system, making the credit rating results more adequacy and reasonable.Secondly, model tested the weight of each index. According to test results, we found out some weights set unreasonable. On the one hand we can re-set the importance of financial ratios to make really important ratios to play important role. The other hand we can make unimportant financial ratios in the rating system to adjust the weight to match its actual ratings impact.Finally, adding common time effect of macroeconomic factors at different time points on the various enterprises, the model can descript bank credit rating system characteristics from two aspects, cross-sectional and time series. This model will be more reasonable for the existing rating system test.2. Writer used the binary responsive panel data models for credit default probability estimating and prediction of commercial bank loans. We build three different models, using real data on commercial bank lending to predict the probability of customer default, we have concluded:Firsty, according to results of the estimation, we picked out financial ratios having a material impact on the corporate default. Financial strength and the importance of these were classified order. This conclusion can provide for banks more comprehensive and more flexible basis for arrangement of commercial loans. Secondly, after tested each model using the'0.5'standard static test and ROC curve dynamic test method, we proofed that the binary response panel data model with a random intercept and random coefficient has better predictive ability. Regardless of sample prediction, forecast sample, or population projection, the model shows a more perfect prediction effect. Therefore, this model can accomplish better the bank's customers default probability forecasting, and more effectively implement risk control of commercial banks.According the'Baselâ…¡'requirements, Thesis integrated a inspection to the existing credit rating system and a forecast loan default probability together, using a new measurement technology. It has a very strong theoretical and practical significance. Comparing to relevant literature, this innovative and unique with the following:Firstly, the thesis successfully sorts panel data model and multiple responses model together. Writer further developed the work by Wang Heng and Shen Lisheng in 2006. The new model will be used in the test work for existing bank rating system. Time effect of the samples individual is fully taken into account by new model. The new model relaxes the independence assumptions of the old method and is more in line with reality. Therefore, the new model can test better existing credit rating system. Secondly, writer separated out common factors of individual at the same point from a random error term, and established a corresponding binary response panel data model. Because the relaxation of independence assumption, it makes the model can fit better with the reality, and get more objective estimates and prediction. Thesis uses this method in commercial bank loan default probability forecasts, to meet better the needs of commercial bank risk management.Thirdly, writer did not simply adopt the "0.5" standard usually using by the existing literature for testing of default forecast models. The results of the model predictions were test separately by the "0.5" standard static test and ROC curve dynamic test. ROC curve dynamic test overcomes the deficiencies and shortcomings of the static test. It is a more scientific method.Fourthly, the thesis is not simply the empirical analysis of banking data. This thesis summarized the shortcoming of various existing methods and extracted the advantages of existing methods. Furthermore, writer bound the discrete response model and panel data analysis together, and applied it to the rating practice. The new model not only solves difficult of lacking of data, which shows in the practice of existing methods, and more important is integrates random effects result of the different observation period into the model. It should be said that this idea is a important breakthrough of credit risk assessment. On the other hand, this new method is combined into the traditional logit model. It has the advantages which includes the visual interpretation and the conclusion simplicity. Therefore, it is a theoretical and practical of the method.Fifthly, the thesis establishes a comprehensive framework of discrete response panel data model. It can be said that this framework shows a more comprehensive combination of the discrete response model and panel data. It riches the analysis work of the discrete response panel data model. For research on combination of discrete variables and panel data model, it may be able to play a role for promoting in a certain degree.
|
PDF Full Text Request