主 题：Transformer-based CoVaR: Textual Information in Systemic Risk基于Transformer的CoVaR模型：文本信息在系统性风险度量中的应用
主讲人：布里斯托大学 王玮宁教授
主持人：小妲己直播
兰伟教授
时间：2025年12月12日（周五）上午10：00-11：00
地点：柳林校区 诚正楼1320会议室
主办单位：小妲己直播
科研处
主讲人简介：
Dr Weining Wang is currently a professor in economics in University of Bristol. Dr Weining Wang was a chair professor in econometrics at the University of Groningen, faculty of economics and business. Dr. Weining Wang was a chair professor of financial econometrics in department economics and related studies at the University of York, UK. She received a Doctor Degree in Economics from Humboldt University in Berlin. Her research fields mainly include non-parametric and semi-parametric econometrics, high-dimensional econometrics, network models, time series. He published in several top journals in the areas, including “Annals of Statistics", "Journal of Business and Economic Statistics" ", Journal of Econometrics", "Journal American Statistics Association", "Econometric Theory", and others. Her research mainly focuses on panel data, high-dimensional time series models, and other applied econometrics methods. The goal is to address specific economic and financial research questions, such as system risk model analysis, financial derivatives asset pricing, and social network analysis.
王玮宁博士现任布里斯托大学经济学教授。此前，她曾担任格罗宁根大学经济与商业小妲己直播计量经济学讲席教授，以及英国约克大学经济及相关研究系金融计量学讲座教授。王博士在柏林洪堡大学获得经济学博士学位，主要研究领域涵盖非参数与半参数计量经济学、高维计量经济学、网络模型及时间序列分析。她在多个顶级学术期刊发表过重要成果，包括“Annals of Statistics", "Journal of Business and Economic Statistics" ", Journal of Econometrics", "Journal American Statistics Association", "Econometric Theory"等。其研究主要集中于面板数据、高维时间序列模型及其他应用计量经济学方法，旨在解决具体的经济与金融研究问题，如系统性风险模型分析、金融衍生品定价以及社会网络分析等。
内容提要：
Conditional Value-at-Risk (CoVaR) quantifies systemic financial risk by capturing the expected loss of one asset conditional on another experiencing significant distress. We develop a Transformer-based methodology that integrates financial news articles directly with market data to improve CoVaR estimates. Unlike approaches that utilize predefined sentiment scores, our method incorporates raw text embeddings extracted from a large language model (LLM). We prove explicit error bounds for our Transformer CoVaR estimator, indicating accurate learning of CoVaR even with smaller datasets. Using U.S. market returns and Reuters news from 2006–2013, our out-of-sample results show that textual information impacts the CoVaR forecasts. In particular, we identify a pronounced negative dip during the crisis period across several equity assets when comparing the Transformer-based CoVaR to both the CoVaR without text and the CoVaR using traditional sentiment measures. Collectively, our results show that textual data can be used to effectively model systemic risk without requiring prohibitively large data sets.
条件风险价值（CoVaR）通过捕捉某一资产在另一资产陷入显著困境时的预期损失，来量化系统性金融风险。本文开发了一种基于Transformer架构的方法，将财经新闻文本与市场数据直接融合以优化CoVaR估计。与传统使用预定义情感评分的方法不同，本方法直接采用从大语言模型提取的原始文本嵌入向量。我们证明了Transformer-CoVaR估计器具有明确的误差界限，表明即使在较小数据集下也能准确学习CoVaR参数。基于2006-2013年美国市场收益与路透社新闻数据的实证研究表明，文本信息对CoVaR预测具有显著影响。特别地，通过比较基于Transformer的CoVaR、无文本基础的CoVaR及传统情感分析方法驱动的CoVaR，我们发现金融危机期间多个权益资产的风险指标呈现明显的负向低谷特征。总体而言，研究结果表明文本数据能够有效建模系统性风险，且无需依赖超大规模数据集。
