Exploring Financial Relationships Using Probabilistic Topic Models (Demonstration Paper)

Abstract

Understanding relationships among financial entities can provide insight into the behavior of complex financial eco-systems. In this demonstration paper, we consider datasets of financial documents that describe the activity or role played by a financial institution (FI), typically with respect to a financial product or another financial entity. We develop community models based on financial institutions (FI) and their behavior or activity described by their roles (Role). Our models are based on an intuitive assumption that FIs will form communities, and FIs within a community are more likely to collaborate with other FIs in that community, and to play the same role, in other communities. Inspired by the Latent Dirichlet Allocation (LDA) and topic models, we develop several probabilistic financial community models and we use those models to identify interesting financial communities in two datasets.