A0501
Title: Enhancing dynamic credit scoring with splines specification
Authors: Viani Djeundje - University of Edinburgh (United Kingdom) [presenting]
Abstract: Credit scoring models constitute a major instrument used by financial institutions to evaluate the risk associated with a loan. At its core, a credit scoring model involves predicting the probability that an account will default over a future time period based on a number of observed variables or attributes that characterize account holders or applicants. Traditional scoring methods were based essentially on the attributes of the applicants measured at the time of application. Yet, many characteristics of the applicants change with time. Survival analysis techniques provide an attractive platform to address the limitations of traditional methods. However, most applications of survival models encountered in the literature assume that the impact of each risk factor on the probability of default remains constant over the business cycle. The purpose is to investigate the validity of such an assumption in the context of retail banking using a large portfolio of credit card loans from a major UK bank. Specifically, a class of flexible models is considered in which the relative impacts of the risk factors are free to vary. A parametric formulation and a spline specification are then proposed to capture the dynamic patterns of the impacts of these risk factors over time. Finally, the varying coefficient approach is shown to outperform a standard model in terms of overall model quality and prediction accuracy.
