Research On IRB Modelings And Informatization For China’s Commercial Banks

With the further deepening of China’s financial system reform, the gradual implementation of the Basel Capital Accord, and the vigorous development of the Internet Financial, there has put forward higher requirements for credit risk quantitative management. The models of IRB elements, such as PD, LGD, EAD and M are the foundation to achieve credit risk quantitative management. Because of data, the complexity of the model and other reasons, the IRB elements modeling and their subsequent applications need to be supported by information technology. This is a valuable research about how to achieve IRB Modelings and Informatization for Chinese Commercial Banks according to the research results at home and abroad. The main innovative points and achievements are summarized as follows.1. PD predictive models The impact of the obligors and the loans to PD is analyzed. The variable selection strategies and sample ratio are given. The criteria using hit rate, coverage rate and accuracy rate is put forward to evaluate the pros and cons of a model performance. Three categories including six models are established using Logistic, SVM and DT and a better model in each category is selected as the final model. Using MDA approach, an efficient combination model is built based on the three categories final models.2. LGD prediction models The distribution characteristic of LGD is given and the impact factors are analyzed. The full sample discriminant model is established using the DT method. According to the situation of non-extreme cases, the point estimate models are built using Beta transformation, Logit transformation and WOE transformation and distribution estimate model is built using generalized beta distribution. A combination forecasting model including point estimate models and distribution model is built combining the advantages of each model.3. EAD prediction models The distribution characteristic of EAD is given and the influencing factors are analyzed. The primary prediction model is established using linear regression method. According to the specialties of EAD distribution, a combine model containing DT and SVM is built to distinguish the full use situation. In the light of the non-full-utilization samples, the point estimate model and distribution estimate model are built with WOE transformation and generalized beta.4. M and regulatory capital measurement simulations based on the accurately measure of M, the impact of the obligors and the loans to M is analyzed. A linear regression predictive model of M is established. The regulatory capital is calculated respectively using the standard method, the confused method and the pool method, and the effects of these ways are compared. The influence of M on regulatory capital is also analyzed.5. IRB system prototype design The ideas of "three unified" are given to construct IRB system. The job scheduling optimization algorithms and data processing optimization scheme are proposed.In the background of the actual situation of China, this paper researches the data construction, elements modeling and informatization of IRB. The conclusions can be of great importance for China’s commercial banks, and helpful to the foreign fellows.