| The question on how to avoid and control loan risk has long been the major concern for financial institutions and their monitoring departments all over the world. Along with the trend of financial globalization as well as the fluctuation in the financial market, banks and financial institutions have been confronted with unprecedented credit risks. The World Bank's research on the crisis of global bank industry indicates that loan risk is the major causes for banks' bankruptcy. Nowadays banking plays a leading role in China's financial system. As the core business, credit line is the major source of commercial bank's revenue, while credit risk becomes the paramount risk confronting commercial banks. Furthermore, overall banking is being confronted with increasingly complicated external operating environment which entails great challenge on risk management institution, technology and effects of China's bank industry, opening to the outside world on December 11th, 2006, which should inevitably push China's banking to be involved in fierce international competitions. Therefore, it has been an important topic debated upon in China's banking industry as well as the academia that how to achieve effective control over loan risks, improvement of loan quality, strengthened post-loan management, conversion of the drawbacks into advantages, and loan risk prevention under the warning system of loan risks. Meanwhile, since the features of banking risks include diffusibility, concealment and hysteresis quality, it is relatively difficult to precisely grasp the risks in time, and it is especially complicated to carry out analysis both quantitatively and qualitatively. Hence, the way to establish an integrated and appropriate early warning system on banking risks, as well as the evaluation methods selected to exactly reflect the fact, has been a key point in the warning of banking risks.Reference to the achievements and experiences of studies on warning system of loan risks both domestic and abroad as the basis, while theories of loan risks and the warning system as the starting point, this article analyzes several faults of China's existing warning index system of loan risks in commercial bank industry. For the elementary selection, Delphi Method is adopted, while for the further selection, adjustment and amendment, both mathematical statistics and qualitative analysis is employed. As a result, an early warning index system of loan risks is established, consisting of 3 levels and 28 terms.For traditional ways to evaluate and warn the banking risks, Fuzzy Integrated Evaluation Method has the defects including the unavoidable randomicity and subjective uncertainty in the decision process, and the lack of capability of self-updating; the once adopted RBF Neural Network Evaluation Method is blamed for the stability of hidden notes in the network, and local extreme. Although the latter has the advantage of self-study, there is a crucial defect in the self-study training that, for the lack of precise experimental design caused by limited samples, the trained capability of the network can hardly cover most risks, resulting in the failure of the artificial neural network evaluation to meet the expected demand.This article intends to combine the Uniform Design Method, developed by Professor Fang Kaitai and Wang Yuan, with the Self-adaptive RBF Neural Network, so that the defects referred above could be overcome. Designing the neural network training samples with the Uniform Design Method can arrange multifactor and multilevel reason analyzing experiments with fewer experimental activities, which is the best design method of reason analyzing experiments under uniform measurement. Self-adaptive RBF Neural Network Method can utilize the capabilities of self-adaptive and nonlinear approaching. The combination of the Uniform Design Method and the Self-adaptive RBF Neural Network can overcome the performance defects of traditional neural network and the inexactitude of experimental design, as well as the subjective randomicity and uncertainty of traditional evaluation.With the Uniform Design Method U1000(528)Table and DSP software, we gain large-scale standardized Uniform Design samples. Then typical samples that are able to cover the whole sample space is chosen for the self-adaptive RBFNN training, so as to gain model experiment results with an error whose value is less than the set error's, 0.0001 on one hand, and, on the other hand, to gain application results with an error whose value is less than the set error's, 0.2, warning gradation and color consistent with the practice.In order to establish warning & evaluating system of banking risks proposed in this article, and to adopt corresponding risk warning models and evaluation methods, we also give suggestions on improvement of China's warning system of loan risks in commercial bank industry. |
PDF Full Text Request