BIMSA >
BIMSA Digital Economy Lab Seminar
Textual Narratives and Volatility Forecasting for Chinese Banks: Evidence from High-Frequency News and the CSI Banks Index
Textual Narratives and Volatility Forecasting for Chinese Banks: Evidence from High-Frequency News and the CSI Banks Index
Speaker
Time
Friday, December 12, 2025 3:00 PM - 4:00 PM
Venue
A3-2-303
Online
Zoom 435 529 7909
(BIMSA)
Abstract
The CSI Banks Index is typically calm but experiences sharp volatility spikes around property and LGFV stress, which standard HAR-RV and macro–financial models fail to predict. This paper builds four daily Chinese textual narrative factors—Bank micro, Policy, Property/LGFV, and Liquidity/Reg—using a weakly supervised word-embedding approach on more than eight million bank-related newspaper articles. Embedding these indices into HAR-type volatility regressions and VaR/ES forecasting, we show that text adds little to mean-squared or absolute forecast errors for average volatility once economic controls are included. In contrast, the Property/LGFV factor has strong, state-dependent effects on the right tail: it predicts upper quantiles of log realized volatility, the probability of extreme-volatility days, and improves tail-risk measures such as QLIKE and Fissler–Ziegel loss for VaR/ES, especially during property-stress episodes like Evergrande. The results suggest that property-related news acts as a state-dependent tail risk amplifier for Chinese banks rather than a generic sentiment indicator.
Speaker Intro
Chen Liang is a PhD student at BIMSA and RUC.His research interests focus on digital economy, liquidity risk, and portfolio investment.