By 
The Bank for International Settlements (BIS) is exploring how artificial intelligence (AI) can enhance financial stability oversight, unveiling a new tool designed to forecast and explain potential market dysfunction.
In a recently published working paper, BIS researchers introduced a two-step AI framework that combines advanced predictive modelling with contextual analysis. The system relies on a recurrent neural network (RNN) trained on more than 100 daily market indicators. Its task: to predict deviations in so-called triangular arbitrage parity — the difference between direct euro-yen trades and indirect euro-dollar-yen transactions.
Under normal conditions, these arbitrage gaps disappear almost instantly. Persistent or widening gaps, however, signal growing frictions in financial markets.
The model goes beyond prediction by highlighting the specific indicators driving its stress signals on any given day. These high-importance signals then feed into a large language model (LLM), which scans recent news coverage to provide timely, real-world context for the emerging risks.
According to BIS, the system can flag vulnerabilities up to 60 business days in advance. In back-testing with data from 2021 to 2024, it accurately identified stress points that later aligned with real-world events, such as the March 2023 banking turmoil.
Case studies showed that when alerts were triggered, the tool’s indicator-driven news searches pointed to discussions of underlying vulnerabilities days before disruptions became visible in the market.
“The tool not only detects risk early but also explains it in accessible terms, helping authorities sharpen their surveillance and prepare timely responses,” the BIS researchers concluded.
The findings reflect growing interest in how AI can augment financial market monitoring, equipping regulators with more proactive and transparent approaches to managing systemic risk.
Comments