RBI paper
Neural networks using geopolitical risk predict volatility better
This story was originally published at 21:27 IST on 22 December 2025
Register to read our real-time news.Informist, Monday, Dec. 22, 2025
MUMBAI – A study in the Reserve Bank of India's monthly bulletin for December, titled 'Decoding Safe Asset Volatility Amid Geopolitical Risks Using Neural Networks', shows that neural network models, particularly non-linear frameworks incorporating country-specific geopolitical risk indices, outperform traditional econometric models in forecasting volatility of major asset classes to geopolitical shocks.
The study incorporated geopolitical risks into asset volatility forecasting models by employing the news-based Geopolitical Risk Index, developed by Caldara and Iacoviello in 2022. It addressed three key questions: how safe-haven assets respond to geopolitical risk, whether non-linear neural network models provide superior volatility forecasts compared with linear econometric benchmarks, and how sensitive each asset is to escalating geopolitical tensions.
Among the four major asset classes--gold, silver, crude oil, and US Treasury securities--crude oil was the most sensitive to geopolitical shocks, consistent with its exposure to supply disruptions and regional conflicts. Gold remained most stable, reaffirming its traditional role as a safe-haven asset, according to the study.
Silver lay in between. It proved more volatile than gold owing to its exposure to industrial demand but less sensitive than crude oil. US Treasury securities exhibited a steady rise in volatility with increasing geopolitical risks, reflecting their role as a flight-to-safety asset during global stress, the study found.
"These results demonstrate that safe haven assets exhibit heterogeneous volatility responses and that incorporating geopolitical risk within nonlinear frameworks significantly enhances forecast accuracy," according to the study. "By incorporating country specific geopolitical risk indices, the nonlinear autoregressive neural network model with exogenous inputs consistently outperforms traditional econometric benchmarks, thus offering a more reliable tool for volatility forecasting in volatile macro-financial conditions and reinforcing the central result that nonlinear approaches outperform linear models in geopolitical stress environments." End
Reported by Ashutosh Pati
Edited by Rajeev Pai
For users of real-time market data terminals, Informist news is available exclusively on the NSE Cogencis WorkStation.
Cogencis news is now Informist news. This follows the acquisition of Cogencis Information Services Ltd. by NSE Data & Analytics Ltd., a 100% subsidiary of the National Stock Exchange of India Ltd. As a part of the transaction, the news department of Cogencis has been sold to Informist Media Pvt. Ltd.
Informist Media Tel +91 (22) 6985-4000
Send comments to feedback@informistmedia.com
© Informist Media Pvt. Ltd. 2025. All rights reserved.
To read more please subscribe
