Unlocking Supervisory Effectiveness in Credit Risk: Insights from the ECB's Supervisory Findings and Measures (European Central Bank)

Summary

• The analysis covers 8,106 supervisory measures issued and closed between 2016 and 2025, using confidential data from ECB Banking Supervision.p.2
• The Findings and Measures database holds more than 75,000 observations since the establishment of the SSM, much of it in unstructured narrative fields.p.6
• An alternative model (o4-mini) replicated the LLM classification with an agreement rate of approximately 90%.p.12
• Measures on data, reporting and IT deficiencies are less likely to achieve full remediation and need longer remediation periods.p.2
• Measures originating from on-site inspections are associated with higher remediation rates and faster closure than SREP or ongoing monitoring measures.p.4
• In the duration model, SREP-originated measures have a hazard ratio of 0.66 and a 52% longer time to closure than on-site inspection measures.p.25

Key insights

1. More severe supervisory measures are less likely to achieve full remediation and generally remain open for longer.p.4
2. Recommendations are more likely to achieve full remediation than legally binding obligations.p.20
3. Weaker governance, proxied by a higher SREP governance score, is associated with a lower likelihood of full remediation.p.20
4. Legal enforceability does not appear to affect how quickly remediation is achieved.p.24
5. Since 2022, ECB Banking Supervision has strengthened its approach to findings and measures through escalation frameworks and additional guidance for Joint Supervisory Teams.p.22
6. The duration model for fully remediated measures uses 3,641 observations.p.26
7. The outcome indicator classifies measures as effective, somewhat effective or ineffective using a large language model on measure descriptions and closure narratives.p.18