EBA 2025 Benchmarking of Internal Approaches: What the IRB and Market Risk Findings Mean for Prudential Reporting Teams

ByOfficer

Last updated: June 2026

The EBA IRB benchmarking exercise is a supervisory data collection where your bank’s internal-model outputs are benchmarked against defined portfolios, supervisory clusters and peer submissions from the institutions included in the exercise. If your probability of default sits well below the peer range on a corporate portfolio that everyone else prices higher, that gap does not stay quiet. It surfaces in your Joint Supervisory Team’s file, it can trigger a question on calibration, and in the worst case it sits behind a request to revisit a risk parameter that drives your own funds requirements. Getting the benchmarking submission wrong, or treating it as a back-office reporting chore rather than a model-risk signal, is how a reporting team ends up explaining a deviation it never saw coming.

On 18 June 2026 the European Banking Authority published its annual assessment of banks’ internal approaches for calculating capital requirements, covering both the 2025 credit risk benchmarking exercise on the Internal Ratings Based (IRB) approach and the 2025 market risk benchmarking exercise across the Internal Model Approach (IMA) and the Alternative Standardised Approach (ASA). The credit risk report carries the reference EBA/REP/2026/15 and rests on data as of 31 December 2024, collected between April and September 2025. This is not a new rule. It is the EBA telling supervisors, and indirectly telling you, where unwarranted variability in capital outcomes still hides.

For a prudential reporting team the practical question is narrow. What does the EBA’s reading of the 2025 benchmarking results change about how you populate the benchmarking templates, how you reconcile them to your COREP credit risk returns, and how you prepare for the supervisory conversation that follows a flagged deviation. That is what this change note works through.

Related reading: COREP Reporting Explained

What the EBA IRB benchmarking exercise actually is

The legal anchor is Article 78 of the Capital Requirements Directive, Directive 2013/36/EU. Article 78 requires competent authorities to monitor the range of risk-weighted exposure amounts or own funds requirements that result from the internal approaches of the institutions they supervise, using defined benchmark portfolios, and to assess the quality of those approaches. The EBA produces an annual report to assist competent authorities in that assessment. The 2025 credit risk report states the mandate plainly: Article 78 of the CRD provides for the monitoring and assessment of risk-weighted exposure amounts that determine the own funds requirements for IRB banks.

Commission Implementing Regulation (EU) 2016/2070 defines the benchmark portfolios and reporting templates, and Commission Implementing Regulation (EU) 2025/379 amended them for the 2025 exercise. For credit risk, the framework is not a pure hypothetical-portfolio test. It uses a clustering approach, under which the credit risk portfolio is decomposed into sub-portfolios with roughly similar risks across institutions, covering exposures to corporates, credit institutions, central governments, retail SMEs, corporate SMEs, exposures secured by residential mortgages and exposures to the construction sector, with further clustering by counterparty residence, collateralisation, default status or industry sector. The credit risk templates include low-default and high-default exposure or portfolio reporting as well as low-default hypothetical transactions. That means differences in reported outcomes do not always come from model choices alone; residual portfolio composition and risk drivers still have to be controlled, reconciled and explained.

Where teams get this wrong is in assuming the benchmarking portfolio is a backtesting exercise. It is not. The EBA is explicit that comparing a bank’s PD against the observed default rate in the same year is meant to compare trends, not to check an ex ante estimate against an ex post realisation. A PD that looks high against last year’s default rate is not automatically wrong, because a through-the-cycle PD is supposed to be a long-run average. The reporting officer who maps a benchmarking figure straight to a single-year default rate and concludes the model is miscalibrated has misread what the exercise measures.

What CRD VI changed about the assessment

Article 78 did not stand still. Directive (EU) 2024/1619, the sixth Capital Requirements Directive known as CRD VI, amended the benchmarking provision. The amended text keeps the obligation on competent authorities to monitor the range of risk-weighted exposure amounts or own funds requirements for the exposures in the benchmark portfolio, but it sharpens what supervisors look for. The amended Article 78(3) point (b) now directs attention to approaches where there is particularly high or low variability, and also where there is a significant and systematic under-estimation of own funds requirements. The phrase significant and systematic under-estimation is new emphasis in the monitoring paragraph, and it tells you what a supervisor is primed to find.

The amendment also restates that the EBA shall produce a report to assist competent authorities in assessing the quality of the approaches based on the benchmark portfolios. So the annual report you are now reading the findings of is the formal output the directive contemplates. For a reporting team the takeaway is not a new template field. It is a change in supervisory posture: the directive has put under-estimation of capital, rather than mere dispersion, at the centre of the assessment.

The headline credit risk findings, in reporting terms

Three findings from EBA/REP/2026/15 matter most for how you read your own numbers.

First, the share of exposure under the IRB approach keeps shrinking. The EBA reports that the share of performing exposure at default under the IRB approach has continued its gradual decline and now sits below 50% of total performing EAD. It remains higher for asset classes like corporates, and it is far higher among the largest institutions, around 57.6%, than among smaller banks at roughly 13.3%. The report also notes that exposures under the Foundation IRB approach increased materially starting from 2025, reflecting the transition to the new Basel III standards. If your bank is moving exposures from Advanced IRB to Foundation IRB, or from IRB to the standardised approach under your roll-out and permanent partial use arrangements, your benchmarking population shifts with it, and your reconciliation to COREP credit risk templates needs to track that move rather than assume a static scope.

Second, variability is falling for PD but not clearly for LGD. Over the 2015 to 2024 period the EBA finds a decreasing trend in PD variability for several asset classes, including corporates, institutions, and retail exposures secured by immovable property. For loss given default the picture is flatter: the report says LGD variability does not present a clear trend, or is only slightly decreasing for most asset classes. The EBA attributes part of the persistent LGD spread to differences in collateralisation, measured through loan-to-value, and to margin of conservatism. Practically, this means an LGD deviation on a secured portfolio is more likely to be explained by genuine collateral and recovery differences than a PD deviation is, but you still have to be able to show the driver rather than assert it.

Third, supervisors judged most estimates adequate but found a stubborn tail of under-estimation. The EBA reports that, on average, almost 50% of the estimates assessed are considered adequate. Unjustified negative deviations, meaning estimates that are too low, typically affect around 10 to 13% of estimates across several portfolios, while unjustified positive deviations are limited and rarely exceed 5%. That asymmetry is the whole point of the CRD VI emphasis on under-estimation. A reporting team should expect that a parameter sitting low against the benchmark draws a sharper question than one sitting high.

The IRB roadmap is still being closed out

The EBA frames the entire exercise against its IRB roadmap, the package of guidelines and technical standards it set out in 2016 to reduce unwarranted variability by harmonising modelling practice. The roadmap runs in phases. Phase 1 is the assessment methodology, delivered through the regulatory technical standards on the assessment methodology under Articles 144(2), 173(3) and 180(3b) of the CRR, adopted as Commission Delegated Regulation (EU) 2022/439 and applied since the second quarter of 2022. Phase 2 is the definition of default. Phase 3 covers risk parameter estimation, including the Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures, applicable since January 2022. Phase 4 covers credit risk mitigation for IRB banks using own LGD estimates.

The 2025 report makes clear the roadmap is not finished in practice even though most deadlines have passed. For the 2025 exercise the EBA received completed supervisory questionnaires covering 87 institutions, down from 97 the previous year, so the year-on-year compliance figures are not directly comparable. A significant share of material IRB models has reached broad compliance with the Guidelines on PD and LGD, but a non-negligible proportion remains classified as having a material model change planned or ongoing. The EBA is careful to say that category does not always mean a lack of progress. It often captures models where the on-site inspection is finished but final supervisory authorisation to use the validated model for own funds calculation has not yet been granted. Supervisory feedback indicates the IRB repair programme is still running in several jurisdictions, with final compliance for some portfolios expected only in the course of 2026.

The common misreading here is to treat a model classed as compliant in the benchmarking questionnaire as fully approved for capital. The two are not the same. A model can be substantively compliant with the Guidelines and still be waiting for the formal permission that lets you use its outputs in your own funds requirements. Your reporting of risk parameters has to follow the permission, not the compliance status.

How benchmarking actually feeds supervision

One number in the report deserves attention because it reframes what a flagged deviation means. When the EBA asked how benchmarking data were used, the largest share of responses, 49%, indicated the results were not used to identify new potential under-estimations. That sounds like the exercise does little. The EBA’s own reading is the opposite. It reflects that such under-estimations are, in many cases, already being addressed through existing supervisory measures and through the ongoing IRB roadmap implementation. In other words, by the time a benchmarking deviation appears, the supervisor often already knows about the weak model and has a remediation track running.

For a reporting team that changes the stakes of a clean submission. A benchmarking figure that is internally inconsistent, or that cannot be reconciled to your COREP credit risk numbers, is not just a data quality lapse. It can introduce a new, unexplained deviation into a picture the supervisor thought it understood, and that is exactly the kind of surprise that generates an information request. The ECB’s Targeted Review of Internal Models, TRIM, already worked through the major SSM banks’ models, so a fresh anomaly in a benchmarking return for a previously reviewed model invites the question of what changed.

Where this bites for cross-border groups is that competent authorities run their own assessment with at least the same frequency as the EBA exercise. A deviation visible to the EBA at the consolidated EU level is also visible to each national authority for the entities it supervises. There is no single audience for the submission.

Market risk: IMA and ASA in the same package

The 2025 assessment is not only about credit risk. The EBA published two market risk reports alongside the credit risk one: the results of the 2025 market risk benchmarking exercise for the Internal Model Approach, and the results for the Alternative Standardised Approach. The IMA exercise covered 43 EU banks across 13 jurisdictions. The EBA reports that variability in Value-at-Risk remains at historically low levels, and that dispersion in the Sensitivities-Based Method has declined, to around 8%.

The reason the ASA report now sits next to the IMA report is the Fundamental Review of the Trading Book transition. As the Basel III market risk framework phases in, the alternative standardised approach becomes the comparison point that the benchmarking exercise has to cover properly, and the EBA expanded the ASA validation portfolios for the 2025 exercise compared to 2024 through Commission Implementing Regulation (EU) 2025/379. If your trading desk is in the FRTB transition, the market risk benchmarking population you report against is not the one you reported two years ago. For the broader trading book capital picture, our EU Basel III market risk and FRTB guide sets out where the standardised and internal model figures land in your returns.

A point teams get wrong on the market risk side is conflating the benchmarking VaR figures with your regulatory VaR for own funds. The benchmarking VaR is computed on hypothetical instruments under prescribed conditions to compare models. It is not your capital number. Treating a low benchmarking dispersion as evidence your live market risk capital is well calibrated is a category error.

Reconciling benchmarking to your COREP returns

The benchmarking submission does not live in isolation from your regular supervisory reporting. The credit risk variability analysis in the EBA report is drawn from COREP templates C 08.02, with the IRB and standardised exposure split pulled from C 08.02 and C 07.00. That is the same template family your team populates for the credit risk leg of COREP. The risk parameters you report for benchmarking, the PD, the LGD, the conversion factors, and the resulting risk-weighted amount, have to be consistent with the parameters behind your C 08 returns for the corresponding exposures.

This is where a practical trap sits. The benchmarking templates do not ask only for hypothetical transactions. They include low-default and high-default portfolio or exposure reporting and, separately, low-default hypothetical transactions. Where the 2016/2070 instructions tie a field to the supervisory reporting framework, apply the same definitions and secured/unsecured splitting logic that your production IRB reporting uses. If your bank splits secured and unsecured exposure for its real portfolios but reports a blended LGD in a benchmarking field where separate secured and unsecured reporting is required, you create an artificial deviation that has nothing to do with model quality. The fix is to apply the same production logic to the relevant benchmarking exposure, portfolio or hypothetical transaction.

The interaction with the CRR3 output floor adds a layer. As the floor phases in, the relationship between IRB-derived risk-weighted amounts and the standardised floor changes the relative weight of your internal models in the capital stack. The detail of that phase-in sits in our CRR3 output floor phase-in guide, and it is worth reading alongside the benchmarking results because the floor is what limits how far a low IRB parameter can actually reduce your capital.

What is coming under CRR3 and the new assessment RTS

The benchmarking findings land in the middle of a live regulatory change to the IRB framework itself. Regulation (EU) 2024/1623, CRR3, amended the IRB chapter of the CRR, including the mandate for the EBA to deliver an updated regulatory technical standard on the assessment methodology that competent authorities follow when assessing compliance with the requirements to use the IRB approach. Under the amended Article 144(2), the EBA is to submit those draft RTS to the Commission by 10 July 2026. CRR3 also amended Article 173(3), which concerns the methodologies competent authorities use to assess the integrity of the assignment process and the regular and independent assessment of risks, with the same 10 July 2026 submission date. Article 143(5) carries a 10 January 2026 date for the RTS on the materiality of rating-system changes.

What this means for a reporting team is that the assessment methodology your supervisor uses to judge your models is itself being rewritten on a 2026 timeline, on top of the 2022 RTS that is in force today. The benchmarking exercise is the running measurement; the RTS is the yardstick. Both are moving. Anchoring your model documentation only to the 2022 assessment methodology RTS, without tracking the CRR3-mandated update, leaves you exposed to a standard that is about to shift.

There is also a parameter-level point that the EBA flags directly. The 2025 report notes that default rates have continued to increase since 2022 for most retail exposures, and that in some cases the EAD-weighted average PD increased less than the observed default rate, signalling a possible decrease in PD conservatism for some retail portfolios. The report reminds supervisors that institutions must ensure the long-run average default rates used for PD calibration reflect the likely range of variability of default rates, as required in Article 46(3) of the RTS on IRB assessment methodology. If your retail PD has not kept pace with rising default rates, that is precisely the kind of low-side deviation the benchmarking exercise is designed to surface.

Common errors when preparing the benchmarking submission

Several mistakes recur, and each produces a deviation that costs explanation later.

Mapping benchmarking parameters to the wrong COREP column. The 2016/2070 instructions reference specific template columns for LGD, maturity, and RWA. Using a parameter from a different reporting context, even a closely related one, manufactures a mismatch.

Reporting a model that is compliant but not yet authorised as if its outputs were live. The benchmarking questionnaire status and the supervisory permission are distinct, and the report makes that distinction explicit.

Treating the construction-sector and residential-mortgage clusters as interchangeable. The 2016/2070 clustering separates exposures secured by residential mortgages from exposures to the construction sector, and they carry different risk drivers. Collapsing them distorts the comparison.

Assuming the IRB scope is static. With FIRB exposures rising under the Basel III transition and the IRB EAD share falling below 50%, the population you benchmark this year may not match last year’s. Reconcile the scope before you reconcile the parameters.

Frequently Asked Questions

Is the EBA 2025 benchmarking report a new regulatory requirement we have to implement?

No. It is the EBA’s annual assessment under Article 78 of the CRD, not a new rule. It reports the results of the 2025 credit risk and market risk benchmarking exercises and helps competent authorities assess model quality. The reporting obligation it relies on, the submission of benchmark portfolio data, already exists under Commission Implementing Regulation (EU) 2016/2070, as amended by Commission Implementing Regulation (EU) 2025/379 for the 2025 exercise.

What reference date does the 2025 credit risk benchmarking exercise use?

The credit risk report, EBA/REP/2026/15, is based on data as of 31 December 2024, collected between April and September 2025. The supervisory questionnaire information on model compliance reflects the situation as of 30 September 2025.

How many banks took part in the 2025 exercise?

For credit risk, the EBA received completed supervisory questionnaires covering 87 institutions, compared with 97 the previous year, so the EBA cautions that the year-on-year compliance figures are not directly comparable. The 2025 market risk IMA exercise covered 43 EU banks across 13 jurisdictions.

Does a benchmarking deviation mean our capital will be increased?

Not automatically. A deviation is a signal for the competent authority to investigate. Article 78 requires corrective action only where it can be clearly identified that an institution’s approach leads to an under-estimation of own funds requirements that is not attributable to differences in the underlying risks. The EBA reports that almost half of assessed estimates are considered adequate, and that unjustified low-side deviations affect roughly 10 to 13% of estimates across several portfolios. We do not speculate on supervisory outcomes for any individual bank.

Why does the report cover the Alternative Standardised Approach alongside the Internal Model Approach for market risk?

Because of the Fundamental Review of the Trading Book transition under Basel III. As the standardised approach becomes the relevant comparison point in the new market risk framework, the EBA expanded the ASA validation portfolios for the 2025 exercise compared with 2024 through Commission Implementing Regulation (EU) 2025/379, and now publishes a dedicated ASA report alongside the IMA report.

How does the benchmarking exercise relate to our COREP credit risk reporting?

The benchmarking parameters you report, PD, LGD, conversion factors and the resulting risk-weighted amount, must be consistent with the parameters behind your COREP credit risk returns, principally the C 08.02 templates the EBA uses for its variability analysis. An inconsistency between the two creates an unexplained deviation that can prompt a supervisory question.

What changes are coming to the IRB assessment methodology?

CRR3, Regulation (EU) 2024/1623, mandates the EBA to submit updated regulatory technical standards on the assessment methodology under the amended Article 144(2) and, under the amended Article 173(3), on the methodologies for assessing the integrity of the assignment process and the regular and independent assessment of risks, both to the Commission by 10 July 2026. The current in-force assessment methodology RTS is Commission Delegated Regulation (EU) 2022/439, applied since the second quarter of 2022.

Related Articles

Key Takeaways

  • The EBA’s 18 June 2026 assessment covers the 2025 credit risk benchmarking exercise (EBA/REP/2026/15, data as of 31 December 2024) and the 2025 market risk IMA and ASA exercises, all under Article 78 of the CRD.
  • CRD VI, Directive (EU) 2024/1619, amended Article 78(3) to focus supervisors on approaches with high or low variability and on significant and systematic under-estimation of own funds requirements.
  • The share of performing EAD under the IRB approach has fallen below 50%, with Foundation IRB exposures rising under the Basel III transition, so your benchmarking scope likely shifted year on year.
  • PD variability is trending down across several asset classes; LGD variability shows no clear trend and is driven partly by collateralisation and margin of conservatism.
  • Almost half of assessed estimates were considered adequate; unjustified low-side deviations affect around 10 to 13% of estimates, while high-side deviations rarely exceed 5%.
  • The credit risk benchmarking framework includes low-default and high-default exposure or portfolio templates as well as low-default hypothetical transactions; differences in reported outcomes reflect portfolio composition and risk drivers as well as model choices, and each must be reconciled to your COREP C 08.02 credit risk returns with the same production logic applied.
  • The IRB roadmap is not fully closed in practice; a model can be broadly compliant with the Guidelines on PD and LGD yet still await final supervisory authorisation for capital use.
  • CRR3 mandates updated assessment methodology RTS under Articles 144(2) and 173(3) by 10 July 2026, on top of the in-force Commission Delegated Regulation (EU) 2022/439; Commission Implementing Regulation (EU) 2025/379 already amended the benchmark portfolios and templates for the 2025 exercise.

Sources and References

What to put on your benchmarking checklist this cycle

The 2025 assessment does not hand reporting teams a new template field to build. It hands them a sharper supervisory lens. CRD VI has named under-estimation of capital as the thing supervisors hunt for, the IRB EAD share is falling while Foundation IRB rises, and the assessment methodology RTS is being rewritten on a 2026 timeline. The practical work is unglamorous and specific: reconcile your benchmarking scope to the exposures actually under IRB this year, identify which of your template fields relate to low-default or high-default portfolio reporting and which to hypothetical transactions and apply your production secured and unsecured splitting logic to all of them, tie every benchmarking parameter back to the right COREP C 08.02 column, and make sure your retail PD calibration has kept pace with rising default rates so you are not the bank explaining a low-side deviation that Article 46(3) already told you to avoid. Do that, and a flagged deviation becomes a conversation you can win rather than a surprise you have to survive.

Disclaimer: The information on RegReportingDesk.com is for educational and informational purposes only. It does not constitute legal, regulatory, tax, or compliance advice. Always consult your compliance officer, legal counsel, or the relevant supervisory authority for guidance specific to your institution.

Similar Posts

  • MiFIR Transaction Reporting: Complete Guide to Article 26 Compliance

    Last updated: March 2026 What Is MiFIR Transaction Reporting and Why It Matters MiFIR transaction reporting is one of the most operationally demanding requirements in European financial regulation. Every day, thousands of investment firms, trading venues, and systematic internalisers in the EU submit transaction data to national competent authorities (NCAs) under MiFIR Article 26. If…

  • CESOP: What Payment Service Providers Need to Report

    Last updated: March 2026 Introduction CESOP reporting is a mandatory quarterly obligation for European payment service providers handling cross-border transactions above the 25-payment threshold – missing the deadline or submitting inaccurate data can result in supervisory action and penalties. If you work in payments or compliance at a European financial institution, CESOP likely sits on…

  • CARF: The New Global Tax Reporting Framework for Crypto

    Last updated: March 2026 Introduction The Crypto-Asset Reporting Framework (CARF) represents a fundamental shift in how crypto-asset transactions are reported across borders for tax purposes. Developed by the OECD and endorsed by the G20, CARF establishes a common standard for automatic exchange of information about crypto-asset transactions – similar in scope to the Common Reporting…