Systemic Risk and Bank Networks: The Use of a Knowledge Graph with Generative Artificial Intelligence

In this paper, we study the systemic risk and networks of top financial institutions using textual data (i.e., news). In particular, we draw knowledge graphs after the textual data are processed through various natural language processing and embedding methods, including the use of the most recent version of ChatGPT (via the OpenAI API). We also compare knowledge graphs drawn from the textual data with those from the numeral data. We test a wide collection of models (i.e. knowledge) with both textual and numeral data for the networks of the top financial firms. Given that systemic risk is crucial in crisis times, we compare networks for the periods of the 2008 crisis (2007 bubble, 2008 bust, 2009 post-crisis). In particular, we focus on the troubled banks (bankrupt and bailed-out) and try to discover any early warning signs of these firms in terms of their networks (i.e. systemic risk). Although the models yield different knowledge graphs, the ensemble results consistently reveal a strong network of interconnections among the troubled firms and their closest counterparties.