AI could strengthen financial stability monitoring

Artificial intelligence could significantly improve the monitoring of financial stability risks by identifying potential economic and financial threats before they materialise through the analysis of millions of financial news reports, according to a European Central Bank (ECB) working paper. The research, titled “A SPOT in the dark: using AI to assess financial stability risks“, introduced SPOT, an artificial intelligence-based indicator designed to measure the Severity and Probability Of potential Trigger events, addressing what the authors describe as a longstanding weakness in financial stability analysis. According to the researchers, financial stability risks are made up of two separate components, namely underlying vulnerabilities within the financial system and the trigger events capable of exposing those weaknesses. While economists and central banks have made considerable progress in measuring vulnerabilities, they argue that assessing potential trigger events has remained largely dependent on qualitative judgement rather than systematic analysis. To address this gap, the researchers developed SPOT using large language models (LLMs) to analyse a vast database of financial news articles covering the period between 2005 and 2026. The AI system evaluates whether individual news reports describe events that could adversely affect future euro area economic activity or financial stability before carrying out a structured assessment across four dimensions. These include the probability that an event will occur, the likely severity of its macro-financial impact, the expected timing of that impact and the principal source of the trigger. Those individual article assessments are then aggregated into a single indicator designed to capture both the frequency and intensity of potential trigger events over time. The authors found that the resulting SPOT indicator consistently increased ahead of major historical episodes of financial stress, suggesting that artificial intelligence can identify emerging risks before they fully materialise. According to the paper, SPOT rose before the global financial crisis, the euro area sovereign debt crisis, the Covid-19 pandemic and periods of heightened geopolitical tensions. The indicator also increased before Russia’s invasion of Ukraine, highlighting its ability to detect emerging threats across a wide range of economic and geopolitical events. Beyond identifying periods of elevated risk, the framework also distinguishes between different types of trigger events. The researchers said SPOT can be broken down according to the source of potential risks, as well as by their estimated probability, severity and expected timing. This provides policymakers with what the paper describes as a structured and comprehensive picture of evolving financial stability risks and the factors driving them. The study also found that SPOT captures information that is not fully reflected in existing measures of financial stress, geopolitical risk, economic policy uncertainty or market volatility. Although the new indicator is correlated with these established measures, the authors argue that it incorporates broader forward-looking narratives contained within financial news coverage. Importantly, SPOT also includes explicit assessments of both the probability and the likely economic impact of adverse events. The researchers tested whether this additional information could improve economic forecasting. They found that incorporating SPOT into a growth-at-risk framework significantly enhanced estimates of downside risks to economic growth. According to the paper, predictive performance improved further when the indicator was combined with measures of existing financial vulnerabilities. The authors said this demonstrates that vulnerabilities and trigger events reinforce one another in shaping economic tail risks. In their conclusion, the researchers argued that the paper provides evidence that AI-based text analysis can strengthen financial stability surveillance by systematically extracting forward-looking signals from large volumes of unstructured information. They said the benchmark SPOT indicator consistently spiked ahead of major historical episodes of financial instability, including the global financial crisis, the euro area sovereign debt crisis, the Covid-19 pandemic and Russia’s invasion of Ukraine. The researchers also said that decomposing the indicator according to trigger source helps policymakers better understand the underlying causes of emerging risks and develop clearer risk narratives. They added that sector-specific SPOT indicators focusing on individual trigger categories, including geopolitical or fiscal risks, make those underlying drivers even more visible and can complement the headline indicator when monitoring specific threats. According to the paper, this allows the framework to provide a comprehensive view of the dynamics, drivers and characteristics of perceived financial stability risks. The research also concluded that different versions of the SPOT indicator contain valuable information about future downside risks to the economy. Moreover, the paper found that the indicator outperformed a range of commonly used measures of financial stress, geopolitical risk, policy uncertainty and market volatility when incorporated into euro area panel growth-at-risk models. The explanatory power of those models improved further when SPOT was combined with indicators measuring existing financial vulnerabilities. According to the authors, this provides evidence that vulnerabilities and trigger events amplify one another when generating risks to economic growth. The paper places the findings within a broader trend of applying artificial intelligence to economics and finance. The authors said recent research has shown that machine learning and large language models can extract valuable information from unstructured data to construct indicators covering sentiment, uncertainty and macro-financial risks. Building on those advances, they argued that SPOT demonstrates how large language models can be applied to financial news to create a systematic and interpretable measure of potential trigger events, complementing existing indicators used to monitor financial stability. Looking ahead, the researchers said similar techniques could be applied to a much wider range of information sources. These could include financial disclosures, supervisory information, market intelligence findings and even social media posts, potentially providing central banks and regulators with earlier warning signals of emerging threats. However, the paper also acknowledged that important challenges remain before artificial intelligence becomes fully integrated into financial stability frameworks. The authors pointed to issues including model reliability, interpretability and data governance, which they said will require continued attention as AI tools become more widely used. Even so, the researchers concluded that AI-based approaches have the potential to become an increasingly important component of financial stability surveillance, helping central banks improve their assessment of future risks by systematically analysing information that would otherwise be impossible to process manually.