Estimation Model Construction Of Loss Given Default For Commercial Bank

Loss given default (LGD) is an important parameter for the calculation of regulatory capital, also the parameter must be estimated by banks of their own according the implementation of Basel II internal ratings-based (IRB) Advanced law. With the domestic banking sector to promote the implementation of Basel II to speed up the process, the development of LGD model becomes the focus of tackling domestic banks. However, due to complex factors of LGD, lack of qualified data and modeling requirements of advanced factors, either domestic academia, regulatory authorities or the bank’s internal, both lack of proven methods and techniques.By LGD home and abroad measured research, combining with the relevant requirements of the new Capital Accord and the actual situation of China’s banking industry, this paper try to construct the quantitative modeling method for domestic banks’LGD, based on the Bank of China Jiangsu Branch from2002to July2010of default loans debt information.Around the development of LGD quantitative models, on the base of system analysis of the relevant article in the LGD theory, literature, the existing modeling methods and the domestic development of LGD models such as the difficulties and constraints of research and analysis, this paper propose a construction method of LGD mixed-model system, with the establishment of quantitative models as the core, supplemented by experts.In order to construction the LGD quantitative models, this paper focus on two key aspects of a breakthrough:The first is to complete calculation and conversion of history debt LGD; and the second is to determine the model variables.Around the first link, from the six aspects, including the concept of default and loss from the definition of the end of the clear closing time determination, clear recovery of part of the calculation of income, the cost of clearing some of the calculation of income, LGD setting the discount rate calculation and the calculation of EAD, this paper break the technical problems exist to complete the historical default LGD calculation of the debt for LGD calculation; On this basis, by observing and analyzing the distribution of LGD samples, it found that the distribution was similar with the Beta distribution, based on this, through the distribution of LGD Beta fitting, and by switching between Beta distribution and normal distribution, the data sequence into a subject LGD Normal distribution, then it can be used for statistical modeling of the Y sequence, also the Y value and the LGD value can be interchangeable, thus the problem of the dependent variable for LGD statistical model was solved.Around the second aspect, firstly, this paper teases out the underlying factors that may affect the LGD; Secondly, the potential fators was caome out one by one by single factor analysis. Given the potential influence factors are type variables, this paper set the dummy variable approach to the type variable handling, in the course of a single factor analysis to statistical significance, economic implications such as the constraints on the part of the variables do.not meet the conditions to give up processing to filter out options for variable statistical model LGD:Third, through the use of stepwise regression approach to model optimization and variable selection, the final LGD quantitative models was identified; Fourth, checking the quantitative model from the three dimensions of the fitting results, the reliability and accuracy. Through the above steps, this paper builds the LGD virtual statistical model including11independent variables, such as egion, industry, ownership, firm size, loan accounting, loan source, type of risk mitigation factors.Base on the LGD statistical model, this paper propose a "hybrid model" program on the base of adjusted by experts. The program includes two aspects, the first, taking into accounting the important factors that do not considered by expert adjustment, including six fators such as macroeconomic cycle factors, the borrower’s repayment order, the borrower’s debt ratio, mortgage debt situation, pledge debt situation, debt guarantees. By this way. the predictive value of LGD can contain as much as possible factors affect LGD information. The second, comparing the predictive value of LGD mixed model with the traditional expert model of LGD, once breaking the boundary, then starting the whole review experts Program, thus avoiding the predicted results to the case of larger deviations. Statistical measurement model through the introduction of expert judgments on the basis of adjustment and manual review to make up the limitations of purely statistical models LGD, ensuring the accuracy and reasonableness of the LGD model, thus, providing a realistic feasible method of theory to pratice.The main contribution of this paper is:1. On the base of historical data and empirical research, this paper constructs a LGD mixed-model system, with the establishment of quantitative models as the core, supplemented by experts, the model’s construction ideas and build technology for the development of domestic banks provide LGD models useful reference and specific examples.2. For the calculation of LGD are explored in this key link, based on the actual situation of domestic banks, this paper give the definition of relevant concepts and specific LGD calculation scheme, with a strong operational; the same time, LGD of the Beta fit and normality Distribution conversion method, for domestic LGD Model of a breakthrough technical problems.3. As to the type of variable factors of LGD, according to the different types of the average LGD to be re-classified, and on this basis, setting the dummy variable approach is the first time in the domestic banking industry, this method effectively solves the problems of more types of factors and the difficult to carry out the virtual variable for LGD facts, and the method is simple and proven results is higher.The main inadquance of this paper is:1. The data used in model is the information of public loan defaults on dabt from Bank of China, Jiangsu Branch between2002and July2010. During this period, as to the stocks were listed, part of non-performing assets were stripped, resulting some data absence, it may have some impact on the result of model.2. This paper used expert adjustment method on the base of LGD statistical model, when constructing the hybrid LGD model. Since this method requires exploration on practice, therefore, this paper just proposes an adjustment program by expert, but how to fall in the real world do not systematically discussed.