AI indicator SPOT quantifies financial stability risk triggers
Researchers from the European Central Bank have developed an AI-based indicator called SPOT to systematically extract and quantify potential financial stability trigger events from news texts. The model analyzes over one million articles from 2005 to 2026 to improve downside economic risk assessments.
Mapping news narratives to financial stress
Researchers Domenic Kellner, Jan Hannes Lang, Lukas Joseph Nagy, and Marek Rusnák introduced the Severity and Probability Of Triggers (SPOT) indicator using GPT-4o-mini across 1.065 million Financial Times articles.
The three-stage prompting pipeline filters economic relevance, identifies forward-looking events over a 1-to-12-month horizon, and rates probability, severity, time horizon, and trigger source on ordinal scales.
Across 146,000 classified trigger articles, financial market events accounted for 35.1 percent and macroeconomic events represented 30.9 percent.
The benchmark indicator increases ahead of major stress periods, including the 2008 global financial crisis, euro area debt crisis, COVID-19 pandemic, and geopolitical escalation in 2026.
Superior accuracy in downside growth projections
Evaluating model performance via euro area panel growth-at-risk estimations over 2005 to 2024 reveals that incorporating SPOT improves one-quarter and one-year ahead tail risk forecasts by over 6 percent and 10 percent respectively.
SPOT consistently outperforms established benchmarks such as the Composite Indicator of Systemic Stress (CISS), Geopolitical Risk Index (GPR), and Economic Policy Uncertainty (EPU).
Interacting SPOT with systemic financial vulnerability indicators further enhances predictive power by two percentage points, underscoring critical amplification mechanisms.
Illuminating risks, but not a standalone cure
Quantifying qualitative risk narratives via LLMs marks a real advance for surveillance.
Yet relying solely on the Financial Times introduces publisher bias into systemic risk signals.
Central banks should pair text AI indicators with hard supervisory data rather than relying on automated media scanning alone.