Graphical models of applicants for credit

Graphical models are a class of statistical tools which have recently undergone extensive theoretical development. They allow one to build models representing the relationships between large numbers of variables, helping to identify paths by which different variables are influenced by others. They look particularly promising for credit-scoring and credit-control problems, since they allow the construction of a holistic applicant model. They can be used in an investigative way, displaying the major influences between variables, or dynamically, allowing statistical prediction of the likely behaviour of individual applicants. They can also be used ‘in reverse’ to identify the characteristics of individuals demonstrating certain kinds of behaviour. This paper describes an initial investigation into the value of graphical models for bank loans. In particular, we describe the graphical models we constructed for a large set of unsecured personal loan data, and we draw some general conclusions.