2023 US Bank Failures: Predicting Insolvency using Text Mining
Authors
Malit, Robenille
Issue Date
2024-05
Degree
HDip in Data Analytics
Publisher
Dublin Business School
Rights holder
Rights
Abstract
Failure of banks is one of the significant risks in the financial system that regulators critically need to manage and look after. Using sentiment results of annual reports filed by US commercial banks to the Securities and Exchange Commission (SEC) from fiscal years 2016 to 2020, this paper attempts to create a predictive classification model that would act as a supplement to existing regulatory oversight of bank insolvency risks.
Ensemble models (Random Forest, AdaBoost and Light GBM) have been trained which achieved significant results during testing phase (around 60 to 80% accuracy across the models). When back validated to predict the 2023 SVB bank failure using sample data, the best model (Light GBM) returned results with 81% accuracy.