Modern Approach to Early Warning Signals in Banking Risk Management

For banking institutions, Early Warning Signals (EWS) are crucial in successful risk management. This thesis assesses the importance of monitoring signals in a timely manner to mitigate risks in corporate credit, and proposes a solution to leverage unstructured data for enhancing warning systems. In collaboration with a leading provider of EWS solutions, this research emphasizes the value of such systems in the era of large language models (LLMs). The theoretical framework includes a comprehensive definition of early warning systems, corporate banking risk management, and their intersection, emphasizing the value of unstructured data in effective risk mitigation. The practical section presents a detailed design and description of an EWS system, outlining the process from data inputs to potential outputs, with the objective of meeting all requirements from future users. The second part is dedicated to the development of a machine learning model for signal prediction. Various architectures are implemented, and an extensive grid search is performed to select the optimal model. The thesis also discusses several ideas for improving the model, recognizing that research in this area is still in its early stages.