Transformer-based CoVaR: Systemic Risk in Textual Information

2026-02-16 · Source: stat.ML updates on arXiv.org · Field: Finance & Economics — Risk Management, Econometrics, FinTech & Digital Financial Services · Depth: Expert, quick

Summary

A new Transformer-based methodology integrates financial news articles with market data to enhance Conditional Value-at-Risk (CoVaR) estimates, a metric for systemic financial risk. This approach, developed by Junyu Chen, Tom Boot, Lingwei Kong, and Weining Wang, directly uses raw text embeddings from a large language model (LLM) instead of relying on predefined sentiment scores. The researchers proved explicit error bounds for their Transformer CoVaR estimator, demonstrating its accuracy even with small datasets. Utilizing U.S. market returns and Reuters news from 2006 to 2013, out-of-sample results indicate that textual information significantly influences CoVaR forecasts. The Transformer-based CoVaR exhibits superior predictive performance, revealing a distinct negative dip during market stress periods for several equity assets compared to CoVaR models without text or those using traditional sentiment measures.

Key takeaway

For AI Scientists developing financial risk models, this research indicates that incorporating raw textual data via Transformer-based LLM embeddings can significantly improve CoVaR forecast accuracy. You should consider integrating such textual information into your systemic risk assessments, especially when analyzing market stress periods, as it offers better predictive performance than traditional sentiment-based or text-agnostic models.

Key insights

Integrating raw LLM text embeddings with market data improves systemic financial risk (CoVaR) estimation.

Principles

• Textual data enhances systemic risk modeling.
• Accurate CoVaR learning is possible with small datasets.

Method

The method uses a Transformer-based approach to integrate raw text embeddings from an LLM with market data for CoVaR estimation, bypassing predefined sentiment scores.

In practice

• Apply LLM embeddings for financial risk modeling.
• Use Reuters news for market stress analysis.

Topics

• Systemic Risk
• Conditional Value-at-Risk
• Transformer Models
• Large Language Models
• Financial Econometrics

Best for: AI Scientist, AI Researcher, Data Scientist, Research Scientist

Related on AIssential

• LLM-Based Examination of Eligibility Criteria from Securities Prospectuses at the German Central Bank — LLMs can automate complex financial document eligibility checks with high precision and semantic evaluation.
• Forecast collapse of transformer-based models under squared loss in financial time series — Highly expressive models like Transformers degrade on financial time series due to increased variance from noise reuse.
• PortBench: A Correlation-Aware, Full-Pipeline Benchmark for LLM-Driven Portfolio Management — LLMs struggle with real-world portfolio management, failing to leverage correlation and exhibiting fragile performance under stress.
• Evaluating Financial Intelligence in Large Language Models: Benchmarking SuperInvesting AI with LLM Engines — Evaluating financial intelligence in LLMs requires a multi-dimensional framework beyond simple factual accuracy.
• Data Machina #261 — Generative AI and LLMs are being adapted to achieve state-of-the-art performance in diverse time-series tasks.
• Enhancing Brazilian Inflation Forecasts through Sentiment Analysis Using Large Language Models — LLM-derived sentiment from policy minutes significantly enhances inflation forecasting accuracy when aligned with human assessments.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.