Methodology & scenario
- CRR3 starting point on Day-1 — output-floor uplift bites.
- FRTB-aligned market risk — IMA/SA bifurcation applies.
- NII technical refinement on re-pricing, behavioural caps.
- Geopolitical-heavy adverse, climate stays thematic.
Spring 2027 submission
- T-0 is 31-Dec-2026 — COREP then EBA-restatement.
- Q1–Q2 2027: template completion, JST challenge.
- H2 2027: SREP letter, P2G anchor set.
- 2028-2030: three SREP cycles under the same anchor.
Where banks lose P2G
- Day-1 CRR3 bridge not reconciled — top-down floors it.
- ICAAP milder than EBA adverse — reads as under-estimation.
- IRRBB outlier without a pre-written rationale.
- MB engagement thin — qualitative SREP downgrade.
Six plays for 2026
- Build the FY2025 → Dec-2026 bridge (Q2 2026).
- Refresh NII engine to 2025 methodology (Q3 2026).
- Shadow the ECB top-down challenger (Q3 2026).
- Recalibrate ICAAP adverse ≥ EBA (Q3 2026).
The stress test is not a modelling exercise. It is a piece of supervisory evidence that the JST carries into the next SREP letter, the next OSI, and the next capital-planning conversation. The bank that treats the 2027 EBA submission as a templates-and-reconciliations problem is preparing to lose somewhere between 50 and 150 basis points of P2G that a better-run neighbour will not lose. The machinery that separates those two outcomes is already in motion — and it is in motion this year, in 2026.
This essay is a walk-through of that machinery. It is a practitioner view, built on public EBA, ECB and ESRB documentation, FY2025 Pillar 3 disclosures, and the standardised carry-forward of methodology from the 2021–2023–2025 cycles. Every number on this page is one of three things: a public-source citation, a clearly-labelled estimate with stated inputs, or a qualitative anchor. Nothing is client data, nothing is mandate-specific, nothing is reverse-engineered from an advisory engagement. If you want the proprietary view, we run that in a mandate, under NDA, against your own portfolio.
What follows is the public layer — the part that every CRO, ICAAP head, and Group Treasury function should be running inside their own team in Q2 and Q3 of this year. If the answers do not line up, the number is already moving.
"The 2027 stress test is being priced right now. The submission the bank will close next spring is governed by the data architecture, internal adverse and management-body engagement it puts in place this quarter. By the time the templates land, the outcome is already anchored."
— Ezelman briefing, Group ICAAP office, Q1 20261 · Why 2026 matters — cadence, not calendar
The EBA EU-wide stress test runs on a biennial cadence. The formal publication cycle over the last decade runs 2014 · 2016 · 2018 · 2021 (deferred from 2020) · 2023 · 2025, with the next regular exercise scheduled for 2027.[3] The 2025 exercise was published in early August 2025 and remains, at the time of writing, the most recent reference point. The 2027 exercise is already being scoped — the ESRB is finalising the adverse macro-financial scenario, the EBA is consolidating methodology feedback, and the ECB's horizontal functions are mapping the bank-level perimeter for the Significant Institutions review.
What this means operationally is simple: the next stress-test submission will close in spring 2027; the next SREP letter that embeds its output will land in late 2027 or early 2028; and the P2G number anchored to that letter will influence the 2028, 2029 and 2030 capital-planning cycles. That is a three-cycle dwell window. The bank that submits a weak 2027 exercise is signing up for thirty-six months of supervisory posture before the anchor resets.
The practitioner conclusion: 2026 is not a quiet year in stress testing. It is the preparation year in which the 2027 submission is built. Infrastructure, methodology alignment, management-body engagement, internal adverse design, and data-quality remediation all have to be in place before the first template request arrives. Banks that treat 2026 as a dormant year lose optionality they cannot recover inside the submission window itself.
2 · The methodology rails — what carries forward, what changes
The EBA methodological note that governs the EU-wide stress test has been deliberately stable across the 2021, 2023 and 2025 cycles. The rails the 2027 exercise will run on are, in the working assumption of every credible preparatory programme, the same four that have governed the last three cycles — with a small number of focused adjustments flagged in the EBA's published consultation feedback.[2]
Balance-sheet composition, business mix and geographic footprint held constant over the 3-year horizon. Maturing positions replaced like-for-like.
Only mechanical or contractual actions reflected — automatic MDA, coupon deferral, pre-approved cost plans. No ad-hoc rights issues.
Credit, market, NII, operational, non-interest income, sovereign revaluation projected separately over a common macro overlay.
Bottom-up projections floored where they diverge from ECB top-down without justification. Floors travel with reputational and P2G cost.
The focused adjustments for 2027
The public trajectory — as visible from the EBA's 2025 feedback statement and the ESRB's 2026 scenario work — points to incremental refinements rather than a structural rewrite. Expectation by risk class:
CRR3 risk-weight framework applies; Day-1 output-floor uplift becomes the starting-point lever. High-end uplifts enter adverse already short of buffer.[5]
FRTB in force as implemented in the EU; IMA/SA bifurcation where applicable. Simplified approach retained for most SIs per CRR3 transitional.
2025 carried 2023 treatment with technical refinements on re-pricing assumptions and behavioural caps; 2027 expected in the same direction.
Projection methodology stable across cycles. ICT / cyber tested thematically outside the Pillar-1 computation, not as a mandatory add-on.
Credit-spread-driven revaluation on sovereigns continues; peripheral-stress channel expected to be heavier under the geopolitical narrative.
Climate and ICT treated qualitatively rather than as Pillar-1 add-ons; 2027 adverse narrative expected to lean more on geopolitical escalation than climate.
The net effect: the 2027 submission is best prepared as a CRR3-consistent, Basel IV-integrated extension of the 2025 exercise, not as a new exercise built from a blank methodology. The banks that over-invest in bespoke 2027-specific infrastructure in 2026 are building for a problem that does not exist. The banks that under-invest in CRR3 output-floor integration and NII modelling quality are building for a submission that will be floored by the top-down.
3 · The ESRB scenario — what public signal tells us
The adverse macro-financial scenario is designed by the ESRB in coordination with the ECB's macroprudential function and delivered to the EBA for overlay onto the bank-level exercise. The scenario is narrative-led — a coherent storyline of macroeconomic and financial shocks — translated into a set of variable-level paths (GDP, unemployment, HICP, house prices, equity indices, sovereign and corporate spreads, swap rates) over the 3-year horizon.[6]
The 2023 scenario was constructed around a materialised geopolitical escalation leading to a renewed supply shock, sustained inflation and monetary-policy divergence.[1] The 2025 scenario carried that architecture forward with a tightening of the financial-market stress channel. The public ESRB material from 2025–2026 points to a 2027 adverse scenario anchored on three themes, in indicative terms:
Ezelman reading of ESRB public material and EBA methodology signals. Theme weights are indicative; the authoritative scenario is the ESRB's final release at the launch of the exercise.
2023 and 2025 figures from ESRB adverse scenarios as published.[1] 2027 paths are Ezelman's indicative reading of ESRB public material, not authoritative; validate against the final ESRB release at launch.
What banks can do with the scenario before it is published
3b · Peer benchmarks — French G-SIB vs EU G-SIB median
Lessons from the 2023 and 2025 cycles are encoded in the public EBA transparency data. The benchmarks below are extracted from the Ezelman EU G-SIB Stress-Test Dataset v1 — 18 banks, 4 cycles, CC BY 4.0. French G-SIB aggregate: BNPP, SG, Crédit Agricole, BPCE.
BNPP peak-to-trough depletion −235 bps[12] vs sector aggregate −370 bps[3] — and improved +163 bps from the 2023 cycle (−398 → −235). Leverage low-point 3.9% (vs 3.4%). Concrete regulatory outcome: BNPP reclassified into Bucket 1 of the ECB P2G framework (0–300 bps depletion → P2G range 0–100 bps, down from 50–200).
Starting CET1 (FR avg 13.65% vs EU median 13.9%) and NII stability (moderate re-pricing sensitivity; dispersion across FR cohort). Op-risk containment roughly on EU median. The FR cohort is narrow (four banks) — inter-bank dispersion visible on capital resilience and credit quality where BNPP pulls the top end of the range.
Where the French G-SIBs sit on the headline metrics
PD & LGD shock multipliers — adverse scenario calibration
Source: EBA transparency Q4 2024 · asset-class averages, IRB approach weighted by exposure. Multipliers are practitioner calibration to the 2023 & 2025 adverse methodology notes — indicative, not authoritative. Full detail in the open dataset.
4 · Starting points — what the FY2025 Pillar 3 disclosures show
The starting CET1 ratio for each bank in the EU-wide exercise is the December Pillar 3 position. For the 2027 submission, that is the December 2026 position. A credible 2026 preparation programme reconciles the bank's projected December 2026 CET1 to the FY2025 disclosed position, adjusted for the Day-1 CRR3 uplift and the staged output-floor phase-in.[5]
The public FY2025 Pillar 3 filings from the major EU G-SIBs provide the market-wide anchor for this reconciliation. The figures below are drawn from the published FY2025 regulatory reports of the respective institutions, accessed via their investor-relations disclosures. They are the public market anchor, not a proprietary sample.
Ratios as disclosed in the respective institutions' FY2025 Pillar 3 filings. Footnote links above each bank name resolve to the primary disclosure. Cited for market-anchor purposes; verify at source before use in your own submission materials. Corrections: research@ezelman.com.
EBA-published aggregate CET1 FL at end-2024. The exercise opens on this ratio — transitional CRR3 basis.
Per the EBA 2025 results publication: the impact of CRR3 at end-2024 is negligible on a transitional basis. The output-floor phase-in is only just beginning.
Net depletion − 370 bps — the 2025 exercise result, over 100 bps milder than 2023.
Fully-loaded aggregate impact concentrated on DE −284, FR −170, NL −140 bps. Adverse end-point on fully-loaded basis falls above 11%. This is where the 2027 cycle starting-point trap lies — not 2025.
Per BNP Paribas' own press release (1-Aug-2025), BNPP posted a fully-loaded CET1 peak-to-trough depletion of −235 bps in the 2025 exercise (vs −398 bps in 2023 — a 163 bps improvement) and a leverage-ratio low point of 3.9% (vs 3.4%). The concrete regulatory outcome: BNPP reclassified into Bucket 1 of the ECB Pillar 2 Guidance framework — P2G range narrowed from 50–200 bps (Bucket 2) to 0–100 bps (Bucket 1). That is the kind of outcome preparation is supposed to deliver.
The 2027 cycle opens at end-2026 with more of the output floor phased in. Fully-loaded aggregate impact is −129 bps, but DE banks face −284 bps, FR −170 bps, NL −140 bps. A 2026 preparation programme must model this jurisdiction-specifically — the 2025 "all-clear" is not a 2027 forecast. Build the FY2025 → Dec-2026 bridge with a fully-loaded CRR3 overlay.
Sources — EBA 2025 EU-wide stress-test results publication[3] (sector aggregate 16.0% → 12.06%, −370 bps, losses €547 bn, NII +1,047 bps gross). BNP Paribas press release[12] corroborated by Euronext filing[13] (−235 bps 2025 vs −398 bps 2023, leverage 3.9% low-point, P2G reclassified to Bucket 1). Country-level fully-loaded CRR3 output-floor impacts from Generation Impact analysis[14].
From FY2025 to 2027 starting point: three levers
The bank's 2027 submission will open on a ratio that is the FY2025 position adjusted for three items. None of these three is an input to the stress-test shocks themselves — they are pre-scenario calibrations that determine how close to the MDA trigger the bank starts. This is the single most underweighted part of a 2026 preparation programme.
CRR3 Day-1 RWA uplift
The output-floor phase-in started in January 2025 at 50% and steps up on an annual basis. A bank whose Day-1 uplift (Dec-2024 to Jan-2025) was in the upper band of the EBA Basel III monitoring range will carry more of that uplift into the December 2026 starting position than a bank whose position was already IRB-conservative. The public EBA Basel III monitoring exercise — and the FY2025 Pillar 3 disclosures that replay it — show meaningful dispersion across the EU G-SIB and Tier-1 perimeter, with individual institutions landing anywhere in an illustrative 50–150 bps RWA-uplift range over the phase-in horizon.[5]
Undistributed earnings & capital generation
The FY2026 retained earnings contribution is not a scenario variable — it is a capital-plan assumption that the JST reads in the ICAAP submission. Banks running aggressive distribution policies enter the exercise on tighter starting buffers than their peers, for identical economic performance. Several institutions publicly run payout ratios in the mid-50s-to-70s percent range as part of stated capital-return policies; banks at the aggressive end of that band should plan for the JST to read the resulting buffer compression as a signal about risk appetite.
AT1 & Tier-2 recognition in transition
The CRR3 transitional provisions for grandfathered AT1 and Tier-2 instruments run on a defined schedule. The 2027 submission is expected to apply the instrument recognition prevailing at the December 2026 cut-off, which for some banks means a lower fully-loaded eligible stack than the current position. This is a reconciliation issue as much as a modelling one — the ECB top-down model applies the CRR3 rules uniformly; a bank whose internal model carries forward a higher eligibility will trigger a floor.
5 · Depletion bands — what the public data actually shows
The single most cited number from the EU-wide stress test is the aggregate 3-year adverse CET1 depletion. In the 2023 exercise, across the 70 banks in scope, the aggregate fully-loaded CET1 ratio fell from 15.02% at starting point to 10.38% at the adverse end-point — a depletion of approximately 479 basis points.[1] The 2025 exercise improved materially: starting CET1 16.0%, adverse end-point 12.06%, depletion of 370 bps across the 64-bank sample — over 100 bps milder than 2023. The improvement was largely driven by net interest income as a positive tailwind (+1,047 bps gross contribution under adverse) offsetting losses (credit −437 bps, market −108 bps, operational −61 bps), combined with a higher starting position.[3]
The aggregate number hides a dispersion that is operationally more important. Across the 2023 perimeter, individual bank depletions ranged from under 100 basis points for the most resilient institutions to more than 1,000 basis points for the most affected. The dispersion tracks three factors: business-model exposure (retail-heavy vs. market-sensitive), geographic concentration (core vs. peripheral sovereign exposure), and the quality of the bottom-up model infrastructure. The last of these three is the only one a bank can influence in the 2026 preparation window.
Bands are Ezelman's calibration to the published 2023 and 2025 EBA EU-wide stress test dispersion; they are practitioner anchors, not EBA-defined thresholds. The official result will always supersede indicative banding.
2023 aggregate from the EBA 2023 EU-wide stress-test results publication.[1] 2025 sector figures from the EBA 2025 EU-wide stress-test results publication & press release of 1 Aug 2025.[3] BNPP-specific figures from BNP Paribas press release of 1 Aug 2025[12] corroborated by Euronext filing.[13] Country-level fully-loaded CRR3 output-floor impacts from Generation Impact analysis.[14]
Three practitioner notes on these bands. First, where the bank sits in the band is largely a function of business-model exposure that cannot be re-engineered inside the submission window. Second, the quality of the projection — whether the JST can reproduce it, whether the ECB top-down floors it, whether the ICAAP corroborates it — is entirely inside the bank's control and is the single largest driver of whether a mid-range depletion translates into a low or a high P2G number. Third, the banks that end the cycle in the resilient band did not arrive there by optimising for the EBA scenario — they arrived there by running internal stress tests more severe than the EBA for several years running, and carrying the embedded conservatism into the submission.
"The depletion number is the bank's. The band the supervisor places you in is a function of the story you submitted to support that number. Two banks with an identical 450 bps can end the SREP cycle 100 bps apart on P2G."
— Ezelman, Q1 2026 CRO-office briefing note6 · Management actions — what the methodology allows and what it doesn't
The EBA methodology places a tight constraint on management actions under adverse. The working rule is: actions that are either mechanical (triggered by contractual terms) or that have been formally approved in a pre-submission capital plan may be reflected; actions that would require a new Board decision conditional on the scenario materialising cannot be.[2] This distinction is where inexperienced submissions lose the most ground.
- Dividend pay-out per automatic MDA. National CRR MDA formula limits distributions without further Board action.
- AT1 coupon deferral per contractual terms. Mechanical coupon cancellation on contractual trigger.
- Prudential filters and transitional arrangements. CRR filters (e.g. Article 468) that apply mechanically during defined windows.
- Cost adjustments within a pre-approved cost plan. Not conditional on the scenario; subject to JST scrutiny of assumptions.
- Ad-hoc rights issues or capital increases. Require a Board decision conditional on adverse — excluded.
- RWA optimisation conditional on the scenario. Mitigating actions triggered only by stress are not permitted.
- Non-core disposals or balance-sheet reshaping. Not permissible unless pre-submission Board-approved and documented.
- Unrealised capital gains and subjective revaluations. Governance-driven reclassifications not pre-agreed with the JST are out.
The practitioner summary: management actions are the wrong place to find comfort in a stress test. The bank that arrives in the exercise planning to manage through adverse via a hypothetical post-event rights issue has already lost the methodology argument. The disciplined preparation is to project under strict constraint, and to use the ICAAP narrative — not the stress-test template — as the vehicle for explaining the bank's real mitigating capacity to the JST.
7 · How supervisors read the output — from templates to P2G
The stress-test result does not land at the JST's desk as a number. It arrives as a set of templates, a bottom-up narrative, a set of methodological notes, and an internal-consistency review against the ECB top-down challenger. The supervisor reads that package in three consecutive steps, each of which can move the resulting P2G recommendation by 25 to 75 basis points.
Step 1 — reconciliation. The ECB compares the bank's bottom-up projections to the top-down challenger for each risk class. Where the bank is materially more optimistic and cannot justify the divergence, the top-down floor applies. Large divergences in credit-loss projections (PD and LGD paths), NII trajectories and sovereign-spread impacts are the most common sites of floor application.
Step 2 — horizontal triangulation. The bank is compared to its peers on a risk-by-risk basis. A bank whose IRRBB sensitivity is materially milder than the peer median — without a justifiable business-model rationale — is flagged. A bank whose credit-loss projection on, for example, unsecured retail is materially below the peer distribution is similarly flagged. The flag does not always move the number directly; it triggers a written follow-up and a quality-assurance conversation that can still move the P2G.
Step 3 — ICAAP corroboration. The stress-test result sits beside the ICAAP in the supervisor's package. If the ICAAP internal adverse is milder than the EBA adverse, the ICAAP is read as a signal about the bank's internal risk culture and management-body engagement. If the ICAAP is more severe, with a documented rationale, the bank is signalling conservatism — which the JST rewards.
Step 4 — bucket placement. The P2G bucket system, covered in detail in the companion essay on Pillar 2G formation logic, takes the stress-test CET1 depletion as the primary anchor and adjusts for the qualitative overlay from steps 1 to 3. Same headline number; different supervisory outcome:
- Bottom-up vs top-downReconciles cleanly — no floors triggered
- Peer triangulationSits within the peer distribution — unremarkable
- ICAAP corroborationInternal adverse more severe than EBA adverse
- Bottom-up vs top-downTop-down floors triggered on credit & NII
- Peer triangulationIRRBB sensitivity flagged horizontally vs peers
- ICAAP corroborationInternal adverse milder than EBA — underestimation signal
8 · The operational test — ten questions for Q2 and Q3 2026
The practitioner version of this walk-through is a preparation checklist. If the answers to all ten questions below are "yes, with documented evidence," the bank is positioned for a credible 2027 submission. Three or more "no" answers correlate, in our experience, with a higher-band P2G outcome relative to the bank's underlying business-model severity — an illustrative 50–150 bps incremental add-on, consistent with the dispersion visible in published SREP outcomes across the SSM perimeter and with the bucket taxonomy the ECB publishes annually in the SREP methodology booklet. Ranges are practitioner estimates, not EBA or ECB-disclosed ranges.
Ten preparation questions · run them in Q2 and Q3 2026
- Does your projected December 2026 CET1 position reconcile to your FY2025 Pillar 3 disclosure with a fully-documented bridge covering CRR3 Day-1 effects, output-floor phase-in, and retained earnings?
- Has your bottom-up credit-risk infrastructure been benchmarked against the ECB top-down challenger on at least one recent horizontal exercise — and is the divergence profile documented?
- Is your NII projection capability aligned with the 2023 and 2025 EBA methodology on re-pricing assumptions, behavioural caps and volume-constant projections?
- Does your internal ICAAP adverse produce a CET1 depletion equal to or greater than the 2023 and 2025 EBA adverse shapes, with bank-specific overlays?
- Is your Management Body engaged on the stress-test preparation with documented minutes, challenge questions and capital-plan review — from the 2026 preparation phase, not the 2027 submission phase?
- Is every open Severity-3+ OSI finding that affects stress-test infrastructure mapped to a remediation plan closing before the 2027 submission window?
- Does your IRRBB shock sensitivity sit within a justifiable band versus peers — and do you have a pre-written JST briefing explaining the position?
- Are your sovereign-concentration and credit-concentration exposures stress-testable against the 2023 and 2025 adverse shapes without remedial model work?
- Is your stress-test data architecture documented to the reproducibility, data-lineage and model-validation standard of a Targeted Review of Internal Models submission?
- Does your CRO own the stress-test preparation end-to-end, with monthly technical engagement with the JST from Q2 2026, rather than a one-shot engagement at submission?
Every engagement is founder-led through submission close. The supervisor-facing arguments are prepared before the templates. The ICAAP is rebuilt in support of the stress test, not as a parallel workstream. The work is signed by the partner who will defend it.
9 · Closing — the long preparation
The EU-wide stress test is not an event. It is a three-year cycle of infrastructure, evidence, governance and relationship, with a submission window that is the last mile of a much longer road. The bank that approaches 2027 as a submission problem is already pricing the consequence of doing so — a higher P2G, a tighter management buffer, a more constrained distribution policy, and a more aggressive JST posture for three cycles.
The bank that approaches 2027 as a 2026 preparation problem has the opposite set of options available. It can reconcile its Day-1 CRR3 position, rebuild its NII infrastructure, re-calibrate its internal adverse, re-paper its governance, close its open findings, and re-anchor its JST relationship — all before the first template request arrives. None of this is expensive relative to the capital cost of getting the submission wrong. All of it is easier to do in 2026 than in 2027.
The single sentence we leave with every CRO we brief this year: the cheapest stress test is the one you rehearsed for eighteen months in advance. The conversation worth having in Q2 and Q3 of 2026 is not whether the bank is ready for 2027. It is whether the bank is using 2026 to build the submission it will not need to repair.