Large exposure estimation through automatic business group identification

Large exposure rules are considered critical for financial institutions, as they directly restrict the lending activity of banks to clients. However, empirical evidence suggests that those rules are difficult both for regulators to enforce and for financial institutions to implement. We present a data-driven analytical model that automatically and algorithmically creates groups of related parties based on ownership information, financial dependencies, business associations, and family ties. We develop a methodology based on linear algebra and networks to group clients, highlight missing critical information, and identify unreported business partners. The approach can be used both prospectively by banking institutions analyzing credit risk and by regulators. We include a case study, applying the methodology retrospectively to highlight large exposure violations and systemic risk leading up to the 2008 banking crises in Iceland.