The Application Of Extreme Value Theory In The Personal Credit Risk Assessment

It is an important research topic in the financial field that the assessment aboutpersonal credit risk. In this paper, we try to establish a model to assess the individualcredit risk rely on the extreme value theory and it will make contributions to theprediction of the development of personal credit risk.With the extreme value theory and the datas from a commercial bank in a secondtier city to explore the problem that to assess the personal credit risk. By analysisingthe factors of affecting the personal credit evaluation system of the financialinstitutions in domestic and foreign, to select the factors of impacting the credit ofperson in our country, and set up indexes and fuctions. The generalized extreme value(GEV)distribution, distribution of type Ⅰ (Gumbel) distribution and the generalizedPareto (GPD)distribution of extreme value theory are quoted to the personalcredit risk assessment. Different parameter estimation methods, Matlab software andR language who build models of individual credit risk. It can be seen the developmenttrend of the models by the method of the least squares fitting.This paper established eight models about the assessment of individual creditrisk,which under different parameter estimations of diffetent extreme valuedistributions, such as: maximum likelihood, Probability weighted distance, L momentestimation of GEV distribution; moment estimate, maximum likelihood estimation ofGumbel; maximum likelihood, Probability weighted distance, L moment estimation ofGPD distribution. Among them, the model of GEV and Gumbel distribution dependon the block maximum method, the model of GPD distribution relies on the superthreshold method. By the test of the least squares fitting of Matlab, it can be knownthat the best of models to evaluate personal credit risk is that the moment estimationand the maximum likelihood estimation of Gumbel distribution. They will give somepractical values: The innovation of this paper:Building different models of extreme value distributionswhich under various methods of Parameter estimation with the use of factors whatinfluencing personal credit.