Default Risk Management And Default Risk Modeling For Micro And Small Enterprises Of Commercial Bank

As the most vitality and innovative part of national economy, MSEs(Micro and Small Enterprises) played an important role in economic growth stimulation, employment stabilization and expansion. Businesses cannot develop without financial support, However, on contrary with their important positions, MSEs cannot get sufficient financial support from commercial banks,which are their most important source of external financing. Despite the fact that, in the face of the intense competition in large corporations' financial services market, the pressures from interest rate liberalization and financial disintermediation, MSEs have become the "Blue Ocean"for banks in the future. But in reality, due to the high default risk and cost, uncontrollable actual risk and weak profitability, this "Blue Ocean " has difficulties in both profitability earning and regulatory compliance achievement. Therefore, banks are very cautious when doing business with MSEs.The fundamental reason for why it is so difficult to approve line of credits or offer loans to MSEs is because banks lack the appropriate risk management capabilities. Banks are a kind of institutions who operate risk and earn risk adjustment profit through effectively identifying,measuring, releasing, hedging and properly pricing risk. In the traditional credit business, banks estimate and determine the appropriate risk premium for each loan. However, because of the small size of production of MSEs, the imperfection financial statements, the lack of effective collateral, the information asymmetry between the banks and MSEs are serious, which cause banks cannot effectively identify the real probability of default and make reasonable pricing.Thus two extremes in practice happened: credit rationing or losing control of MSEs' loans risk,both are difficult to achieve sustainable development in the long term.So, the key to solve the problem is to develop suitable default risk management methods and techniques for MSEs, to achieve effective identification and measurement of their defaultrisk in order to support the pricing. Among them, an accurate estimate of the probability of default is the most essential one.Although the causes of MSEs' default risk, control strategy, moral hazard, adverse selection,credit rationing behavior and other related issues have been extensively studied, but there is little attention focuses on the characteristics of MSEs' default risk and the corresponding control mechanism, the study on modeling the probability of default based on the default risk control mechanisms for MSEs is even less. Therefore, based on the review of previous literatures,theories and practice experiments, this article focuses on the research of default risk management mechanism of MSEs and develops proper default probability prediction models for them. Through abstraction and quantification of the default trigger mechanism, we make an in-depth characterization of the default mechanisms for MSEs, construct three kinds of defaults prediction models, two of them are under the assumption of perfect data condition, the third one is under the assumption of imperfect data condition.Results of this study will contribute to a better understanding of default risk characteristics and default risk control mechanism of MSEs; to change the traditional risk control methods rely on assets and collateral which are unsuitable for MSEs, to enhance the accuracy, effectiveness and feasibility of default risk prediction. Also the models develop in this study will help banks to establish the internal rating system suitable for MSEs and play an important role in customer assessment, credit approval, risk monitoring, tracking and economic capital allocation.The article consists three parts, seven chapters:First, based on the literature review and successful practices ,we conduct a study on the default risk characteristics and default control mechanism of MSEs. Firstly, summarize four risk characteristics of MSEs: the lack of effective collateral; more severely information asymmetry;the size of single loan is small but large-scale of the asset pool as well as more sensitive to changes in the external environment. Then combine with the industry practices and the results of previous studies, we propose three MSEs default control mechanisms: the default trigger mechanism based on cash flows; information asymmetry reduction based on the relationship lending; risk management based on loan asset pool with both quantitative and qualitative methods. And accordingly refine the characteristics required to MSEs default risk modeling. All these results are providing a theoretical basis and modeling ideas for the main part of the article.Second, based on the default risk control mechanisms of MSEs, we build two kinds of models under the assumption of perfect data condition. One is a default prediction model using the framework of incomplete information, another is belongs to statistics and machine learning class models.Based on framework of incomplete information model, we construct suitable theoretical models for MSEs under the condition of information asymmetric. By abstraction and depiction the core elements of cash flow default trigger mechanism: default boundary and distribution of real cash flow, we build the theoretical models which can effectively estimate the probability of MSEs and can applies to the condition of changing information asymmetry level. We establish three models, each model has a more realistic assumption: for the first model we assume banks can observe complete initial information and the loan application by MSEs will not influence the default probability, then we relax the first assumption by assuming banks cannot observe the complete initial information to build the second model, while we relax both assumptions to build the third model.Based on real data, establish a statistical and machine learning class model for MSEs. At first, filter the existing predictors for MSEs' default risk evaluation to obtain the best predictors,which are mainly variables represent the cash flow and relationship lending, this result verify the theoretical analysis of the foregoing conclusions. Meanwhile, empirical test find that the model integrated the Logistic regression method and support vector machine method is most suitable for our MSEs data, which can achieve the best accuracy and stability of the prediction with lowest prediction error cost.Third, aiming at solving the problem of data missing. This article integrates the expert prior information and data information by using Bayesian method to obtain a more effective posterior estimation. Conclusions have shown that this method can effectively solve the problem of insufficient historical data, and provide a smoother and more accurate estimate of default. Robust test results also demonstrate the robustness of the method. Furthermore, we consider the existence of default correlation and multiple periods estimation to improve the estimation accuracy. Meanwhile, the evaluation method based on Proper Scoring Rules can judge the effectiveness of expert information and change the weights of each expert dynamically.