Dynamic Modeling On Credit Risk Evaluation With Incremental Support Vector Machine

Commercial banks, as the hub of national economy, play irreplaceable important roles in many aspects. In order to prevent a new financial crisis with credit risk as its core, commercial banks have enhanced risk management in recent years, especially credit risk management. How to make scientific credit risk evaluation is significant. Modeling research on credit risk evaluation has become a hot topic in academic and practice.Actually, process of bank credit management belongs to a type of evaluation or classification on default risk of loan customers. So far, lots of researches have used various theories and technologies from financial engineering to establish credit risk evaluation models. However, all the existing researches focuse on static modeling on sample dataset in a certain period of time without dynamic updating mechanism.Therefore, based on financial theory, artificial intelligence and mathematical statistics theorys, this paper adopts qualitative analysis, quantitative analysis and ensemble methods to study on dymanic modeling for commercial banks’credit decision-making.Firstly, according to the five principles of comprehensive, general, sensitive, measureable and priori, several financial ratios are selected to compose a system of features for modeling, which respectively belong to solvency ratio, profitability ratio, operating capability ratio, and development capacity ratio, and then branch and bound algorithm is employed to select signi(?)cant indicators from above candidate financial ratios to generate final index system.Secondly, in this paper, we address the integration of branch and bound algorithm with modeling of credit risk evaluation using incremental support vector machine ensemble to generate a dynamic evaluation model. This new model hybridizes support vectors of old data with incremental financial data of corporate in dynamic ensemble modeling of bagged support vector machine. In the incremental stage, multiple base support vector machines are dynamically adjusted according to bagged new updated information for credit risk evaluation. These updated base models are further combined to generate a dynamic evaluation.Finally, in the empirical experiment, the new method is compared with the traditional model of non-incremental support vector machine ensemble for credit risk evaluation.The experimental results demonstrate that the new model proposed in this paper is able to adjust itself according to incremental information of credit risk evaluation, and produce superior performance to the non-incremental SVM ensemble. Thus, the new model offers a new effective method for bank credit evaluation as well. Specifically, the research in this paper provides a new ideal for the academic study on credit risk evaluation methods, and has played an important role in the development of social and economic. In additon, it also coordinates long-standing conflicts about the loan problem between banks and enterprises. Therefore, the research in this paper is of great significance not only in theory but also in practice.