Predicting financial stability with TopicGPT: Insights from corporate and central bank communications
Loading...
genai.pdf (981.9 KB)
0000-0001-7148-2092 Date
2025-12-10
Journal Title
Journal ISSN
Volume Title
Publisher
Elsevier
License
Abstract
Can generative artificial intelligence (GenAI) help us predict financial stability? To address this question, we employ TopicGPT, a prompt-based framework for topic modeling powered by large language models. By analyzing over 238,000 corporate earnings calls and 4300 Federal Reserve speeches over the period from 2002 to 2023, we combine microeconomic and macroeconomic perspectives to forecast key measures of financial stability. TopicGPT’s ability to generate interpretable and tailored topics improves predictions for systemic risk measures, such as the National Financial Conditions Index and a capital shortfall, outperforming traditional models, particularly for long-term horizons. The two data sources complement each other: earnings calls provide dynamic, firm-specific insights critical for short-term forecasts, while Fed speeches highlight systemic risks, offering a long-term perspective. Together, they identify critical themes – such as economic conditions, debt management, and the housing market – and enable real-time risk assessment.
Description
Adam Zaremba acknowledges the support of the National Science Center of Poland , Grant No. 2022/45/B/HS4/00451.
Keywords
Citation
Fieberg, C., Hesse, M., Liedtke, G., & Zaremba, A. (2025). Predicting financial stability with TopicGPT: Insights from corporate and central bank communications. Journal of Banking & Finance, 183, 107598.