Semiparametric Credit Scoring Model

Credit scoring is one of most successful applications of statistics and opera-tional research in finance and banking industry as well as one of the earliest tools of risk controlling and managing in finance industry. There are many parametric and nonparametric statistical methods of establishing credit scoring models such as Linear Regression Models, Logistic Regression Models, k-Nearest Neighbor and so on. However there are few research of semiparametric methods in credit scoring. Semiparametric is a kind of important statistical model which developed from 1980s. This kind of model contains both parametric and nonparametric components and can be used to describe many practical problems. In this paper we introduce the development and methods of credit scoring, then we consider a semiparametric approach which generalizes the traditional method. We use Generalized Partial Linear Modelto describe the probability of a client belonging to the group of good, here f(·) is a known function,βis an unknown parameter vector, g(·) is an unknown function. A quasi-likelihood method is used to estimate the model. We introduce relative estimate theory, calculation steps, bandwidth choosing methods, and through simulation we compare semiparametric model with traditional models in order to show the difference of identification rate.Main results: We generalized the probability of a client belonging to the group of good from P(Y=1|X)=(?)(X~τβ) to a partial linear argument P(Y = 1|X,T)=f(X~τβ+g(T)). Thus semiparametric credit scoring model is more is more adaptive than traditional models, it can also avoid some disadvantage of nonparametric methods. Through simulation we show that semiparametric credit scoring model has better separation of clients than traditional models.