In the Beginning was the Word: LLM in Risk Forecasts.

Financial risk management measures quantify potential losses in investment portfolios under different market conditions. A common practice is to employ Value at Risk (VaR) and Expected Shortfall (ES) as metrics. These are often estimated with limitations in terms of accuracy, computational demand, and assumptions about market behaviour. Machine learning tools, like large language models (LLMs), have emerged as innovative alternative. We address a critical research gap by investigating the use of LLMs to estimate market risk indicators. Our empirical analysis demonstrates that LLMs can significantly improve the precision of tail risk forecasts. These findings suggest that LLM-based approaches hold substantial potential for financial forecasting, offering a competitive edge in predicting market risk indicators and enhancing overall risk management strategies.

Speaker bio

Prof. Wolfgang Karl Härdle is the Ladislaus von Bortkiewicz Professor of Statistics in Humboldt -Universität zu Berlin. He is currently a visiting professor at the University of Edinburgh hosted by Prof. Tiejun Ma. Wolfgang Karl Härdle is an internationally renowned senior academic with ongoing contributions to the field of dimension flexible data analytic technology and machine learning for economics and finance. Prof. Härdle has over 500 publications with high citations (citation profile). Prof. Härdle is the founding director of the BRC Blockchain Research Centre.