What is it about?

Financial crises and periods of severe market stress can develop quickly, but many traditional early-warning systems rely on low-frequency indicators and linear models that may react too slowly. This study develops a high-frequency machine-learning framework for monitoring systemic financial stress in Europe using daily financial, banking, macro-financial, and sentiment indicators. Financial stress is identified using the European Central Bank’s Composite Indicator of Systemic Stress. The framework uses a strictly time-ordered modeling process to avoid look-ahead bias and applies stability-based feature selection, time-series cross-validation, and evaluation measures suited to rare stress events. The results show that models combining financial and sentiment information outperform linear benchmarks, particularly when systemic stress is intensifying.

Featured Image

Why is it important?

Financial instability can spread rapidly across markets and institutions, making timely detection essential for regulators, central banks, and risk managers. Our results show that high-frequency machine-learning models can identify emerging systemic stress more effectively than conventional linear approaches, especially when financial indicators are combined with non-financial information such as sentiment. This can provide earlier signals of deteriorating financial conditions and help complement existing macroprudential monitoring tools, giving decision-makers more timely evidence when assessing risks to financial stability.

Perspectives

Early-warning systems are only useful if they provide signals soon enough to support action. In fast-moving financial markets, relying mainly on monthly or quarterly indicators can leave important changes undetected until stress has already intensified. Our work shows that combining daily financial, banking, macro-financial, and sentiment information with machine learning can provide a more responsive view of systemic risk. An important aspect of the framework is methodological discipline: the data are processed in strict chronological order, and models are evaluated using time-series validation and rare-event metrics. This helps ensure that the apparent predictive performance reflects information that would actually have been available at the time. We see these models as complementary tools for financial-stability monitoring rather than substitutes for expert judgment or established macroprudential indicators.

Prof. Afshin Ashofteh
Universidade Nova de Lisboa

Read the Original

This page is a summary of: High-Frequency Early Warning of Systemic Financial Stress in Europe Using Financial and Non-Financial Data with Machine Learning, July 2026, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/compsac69091.2026.00296.
You can read the full text:

Read

Contributors

The following have contributed to this page