Multivariate credit portfolio management using cluster analysis

Purpose – This paper aims to improve decision making in credit portfolio management through analytical data-mining methods, which should be used as data availability and data quality of credit portfolios increase due to (semi-)automated credit decisions, improved data warehouses and heightened information needs of portfolio management. Design/methodology/approach – To contribute to this fact, this paper elaborates credit portfolio analysis based on cluster analysis. This statistical method, so far mainly used in other disciplines, is able to determine “hidden” patterns within a data set by examining data similarities. Findings – Based on several real-world credit portfolio data sets provided by a financial institution, the authors find that cluster analysis is a suitable method to determine numerous multivariate contract specifications implying high or, respectively, low profit potential. Research limitations/implications – Nevertheless, cluster analysis is a statistical method with multiple possible settings that have to be adjusted manually. Thus, various different results are possible, and as cluster analysis is an application of unsupervised learning, a validation of the results is hardly possible. Practical implications – By applying this approach in credit portfolio management, companies are able to utilize the information gained when making future credit portfolio decisions and, consequently, increase their profit. Originality/value – The paper at hand provides a unique structured approach on how to perform a multivariate cluster analysis of a credit portfolio by considering risk and return simultaneously. In this context, this procedure serves as a guidance on how to conduct a cluster analysis of a credit portfolio including advices for the settings of the analysis.