← Back to Frameworks
Methodology · Technical Note
ANCHOR™ — the 7-pillar starting-point protocol for EBA stress testing
ANCHOR is the Ezelman methodology for ensuring that every EBA EU-wide, SSM thematic or ICAAP stress-test submission opens on a reconciled, defensible starting position. It is the sub-framework inside PRISM™ that governs the T-0 balance sheet, the classification locks and the supervisor-facing defence. This page is the external technical note — a 4-6 page appendix we release so that a Head of Model Risk, CFO or ex-ECB reviewer can read the logic without needing a mandate.
7
Pillars — Assemble, Normalise, Classify, Harmonise, Own, Review, Defend
3
Independent challenge rounds before any EBA submission leaves the building
5
Validation checkpoints from T-0 sign-off to post-submission archive
<24h
Written clarification SLA for live EBA challenge queries
1. Purpose
The starting point is the single most under-weighted part of a stress-test programme and, mechanically, the single largest lever on the P2G outcome. A 30-bp error at T-0 becomes a 70-90-bp error at the adverse end-point after three years of scenario depletion, RWA migration and dividend mechanics.
ANCHOR is the seven-step protocol we apply before any scenario is run. It exists because, in our experience across multiple EBA cycles — including the two FTA-grade exercises of the past decade (IFRS 9 in 2018, CRR3 in 2025) — roughly 60-70% of all methodology challenges from the EBA or JST trace back to a weakness at T-0: an unreconciled exposure, a classification inconsistency, an IFRS-vs-prudential gap, an owner who cannot sign. ANCHOR closes those off one at a time, in a fixed order, with named accountability.
ANCHOR is written to be used. Every pillar in the sections below has explicit inputs, explicit outputs, and a sign-off owner. This is not a maturity model; it is a delivery protocol.
2. The seven pillars
Deployed in sequence. A pillar cannot be opened until the prior pillar is closed and signed. The letters spell ANCHOR + the defence step.
Framework · Ezelman ANCHOR™
Seven pillars that carry the starting point into the scenario engine
A
ASSEMBLE
T-0 balance-sheet assembly. One source of truth across counterparties, trading book, NII book, off-balance-sheet and intragroup.
N
NORMALISE
IFRS vs prudential, solo vs consolidated, COREP vs FINREP reconciliation — every delta tagged and explained.
C
CLASSIFY
NPE, IFRS 9 stages, SICR triggers, forbearance, LGD-defaulted — all locked at T-0, with downward-migration rules pre-agreed.
H
HARMONISE
Consistent calibration across credit, market, operational, NII, liquidity. No siloed assumptions. No double counting.
O
OWN
Named accountability: who approves T-0, who signs scenario translation, who defends to EBA / JST. One person per deliverable.
R
REVIEW
Three independent challenge rounds before submission: methodology, numbers, narrative. Each round chaired by a non-involved partner.
+
DEFEND
Live EBA / JST challenge sessions. Written clarifications within 24h. Post-submission archive within 10 working days.
3. Pillar detail — inputs and outputs
Each pillar is specified below with the source systems and data feeds it consumes, and the deliverables and sign-offs it produces. The list is deliberately concrete: if a client cannot identify where the input lives, the pillar cannot be opened.
A
ASSEMBLE
T-0 balance-sheet assembly · one source of truth
Build a single, fully attributed T-0 balance sheet that survives every subsequent pillar without rework. In practice, this means pulling the reference-date exposure stack from the general ledger, the credit data mart, the ALM cube and the trading sub-ledger — and reconciling them before any scenario work begins.
Inputs
- General ledger extract at reference date (typically 31-Dec, audited)
- Credit data mart — exposure, collateral, rating, stage, NPE flag
- Trading book sub-ledger with FRTB-aligned sensitivities
- ALM cube (NII book) — re-pricing ladder, behavioural profile
- Off-balance-sheet register (commitments, guarantees, derivatives gross notional)
Outputs
- Single T-0 exposure stack — one row per counterparty / position, fully attributed
- GL-to-risk reconciliation table with <1 bp unexplained residual
- Data-quality dashboard (completeness, lineage, freshness) signed by the CDO
- T-0 sign-off memo — partner-level, dated, archived
- Frozen baseline snapshot in the secure archive (tamper-evident)
N
NORMALISE
IFRS vs prudential · solo vs consolidated · COREP vs FINREP
The EBA templates demand a particular view of the bank. That view is neither the audited IFRS financial statements nor the pure COREP return — it is a reconciled composite, and the reconciliation must be traceable. Normalisation tags every delta, explains every scope difference, and documents every adjustment.
Inputs
- FINREP submission at reference date (bank-audited)
- COREP submission at reference date (prudential)
- Scope-of-consolidation register — solo, sub-consolidated, full consolidated
- IFRS 9 ECL model outputs at T-0 (stages 1, 2, 3)
- Prudential filters and regulatory deductions (CRR Art. 32-35)
Outputs
- FINREP-to-COREP reconciliation walk, line by line
- Solo-to-consolidated bridge with attribution to entity
- IFRS-to-prudential ECL adjustment note (transitional filter treatment)
- Normalised T-0 dataset loaded into the stress engine
- Normalisation sign-off by Group CFO and Head of Prudential Reporting
C
CLASSIFY
NPE, IFRS 9 stages, SICR triggers — locked at T-0
Classification drift is the most frequent single finding in EBA challenge. Banks lose the argument when the definition of Stage 2 at T-0 differs from the definition used in scenario year 1, or when forbearance rules migrate mid-exercise. ANCHOR locks every classification at T-0 and pre-agrees the downward-migration rule with the EBA methodological note of the cycle.
Inputs
- IFRS 9 staging output (stage 1 / 2 / 3) at T-0
- SICR trigger rule-book — quantitative (PD) and qualitative (watchlist) criteria
- NPE register under EBA ITS definition (past-due, unlikely-to-pay, forborne)
- Collateral valuation file — haircuts, LTV, indexation
- Defaulted-LGD register with cure-rate history
Outputs
- Frozen T-0 classification file with row-level stage and NPE flag
- Pre-agreed SICR migration rule for scenario years 1, 2, 3
- Forbearance treatment protocol aligned to EBA methodological note
- Classification challenger — independent team replicates 20% sample
- Sign-off by Head of Credit Risk and Chief Risk Officer
H
HARMONISE
Consistent calibration across risk classes
Banks routinely run credit stress, market stress, operational stress and NII stress as four independent islands. The EBA doesn't accept that. Harmonisation ensures every risk class shares the same macro driver, the same FX path, the same rates path, and the same sovereign-spread trajectory — and that the aggregation doesn't double-count.
Inputs
- EBA / ESRB adverse scenario variables (GDP, unemployment, HPI, rates, FX, equity)
- Internal macro-to-PD translation models (credit)
- Rates / FX / equity shock grid (market)
- Conduct / cyber / external fraud loss distributions (operational)
- Re-pricing gap and deposit beta assumptions (NII)
Outputs
- Single scenario-translation matrix shared across all risk classes
- Double-count register (e.g. sovereign shock in market AND credit)
- Cross-risk sensitivity tests (rates +100bp → NII, credit, market simultaneously)
- Harmonised second-round effects note (credit losses → RWA density → CET1)
- Sign-off by Head of Stress Testing and Head of Risk Analytics
O
OWN
Named accountability for every deliverable
Under pressure, accountability diffuses. ANCHOR-OWN pins one name to every artefact: the T-0 approver, the scenario signer, the EBA defender, the Board presenter. Names are fixed in writing at mandate kick-off and do not change mid-cycle without partner-level escalation.
Inputs
- Mandate governance charter — steering committee, working groups
- RACI matrix by deliverable (Responsible / Accountable / Consulted / Informed)
- Delegation-of-authority register (who may sign on behalf of the CRO)
- Escalation path to CEO and Risk Committee Chair
Outputs
- Named owner per pillar, per sign-off, per EBA challenge topic
- Signed deliverables log — one row per artefact, partner-countersigned
- Escalation log — issues resolved, open, escalated to Board
- Post-cycle lessons-learned note attributed to named owners
R
REVIEW
Three independent challenge rounds before submission
Every material assumption is challenged three times before submission, in three separate rounds, by three independent teams. Round 1 interrogates methodology. Round 2 reruns the numbers in a challenger model. Round 3 pressure-tests the narrative against the EBA methodological note and the written clarifications archive.
Inputs
- Full methodology pack (incl. data lineage) for Round 1
- Challenger model (simplified, independent build) for Round 2
- EBA methodological note of the cycle — all paragraphs cross-referenced
- Prior-cycle EBA written clarifications to the bank and to peers
Outputs
- Round 1 methodology review memo — open / closed issues
- Round 2 challenger-vs-production reconciliation — variance explained
- Round 3 narrative dry-run transcript and red-team scorecard
- Residual-risk register signed by the CRO before submission
+
DEFEND
Live EBA challenge sessions · <24h written SLA
Submission is not the end. The EBA / JST challenge cycle runs for weeks after the templates are filed, with live calls, follow-up queries and the constant risk of being told the bank is “an outlier.” ANCHOR-DEFEND runs this phase as a war-room: written responses within 24 hours, every answer signed by a named owner, every clarification logged for the post-submission archive.
Inputs
- EBA / JST challenge queries (verbal and written)
- Full pillar output library (searchable in the war-room)
- Peer-benchmark dataset for outlier-response framing
- Partner on-call rota for escalation
Outputs
- Written responses within 24h, owner-signed
- Live-session briefing pack for CRO / CFO on each call
- Clarifications archive — one file per query, searchable for the next cycle
- Post-submission lessons-learned pack delivered to the Risk Committee
4. Validation checkpoints
Five mandatory gates. A submission cannot advance past a gate without the named signature. Gates are not rubber-stamped; each generates an artefact retained in the post-submission archive.
01
T-0 sign-off
Gate: must-passPartner-level confirmation that the reference-date balance sheet is reconciled, classified, normalised and frozen. No scenario work can start until this gate closes.
02
Scenario sign-off
Gate: must-passThe scenario-translation matrix — macro drivers to PD, LGD, market shocks, NII paths — signed by Head of Stress Testing and countersigned by CRO.
03
Bottom-up vs top-down reconciliation
Gate: must-passThe client's bottom-up projection reconciled against the EBA top-down benchmark and against the internal challenger. Variance explained line by line.
04
JST challenge briefing
Gate: recommendedPre-brief with the Joint Supervisory Team on the anticipated outlier points. Not all cycles permit this; where allowed, we recommend it.
05
Post-submission archive
Gate: must-passComplete, tamper-evident archive of everything filed plus every clarification exchanged. Delivered to the Risk Committee within 10 working days of the EBA close.
5. Deliverables tree
The concrete artefacts shipped to the Steering Committee at mandate inception. Each is a working document, maintained live through the cycle, and archived at close.
01
Master technical note — T-0 dataset, classification rules, normalisation bridges. Typically 80-120 pages, partner-signed.
02
Macro drivers translated to risk parameters. Baseline, adverse, reverse paths fully documented with EBA methodological-note cross-references.
03
Simplified independent re-build of the bank's production engine. Typically converges within 5-8% of production on adverse CET1.
04
Reconciliation pack
SupportGL-to-risk, FINREP-to-COREP, solo-to-consolidated walks. One row per reconciling item, fully attributed.
05
ICAAP integration note
SupportHow the EBA stress feeds the bank's ICAAP — buffer calibration, P2G dialogue input, MDA-distance narrative.
06
One-page outlier cards, written-clarification log, anticipated-question matrix. Used live in the EBA / JST challenge sessions.
6. Where ANCHOR sits vs other Ezelman frameworks
ANCHOR is one of four proprietary frameworks we operate. It is the sub-framework inside PRISM™ that governs the T-0 and the classification locks; it sits alongside but independently of the frameworks we run for on-site inspections, regulatory programmes and transformation.
The 5-step stress-testing discipline: Pressure, Reverse, Iterate, Scenario, Map. ANCHOR sits inside step P (starting-point build); PRISM governs the delivery sequence.
Applies to: every EBA EU-wide, SSM thematic, ICAAP, reverse and ORSA-adjacent stress test.
FORTRESS™
On-site inspectionsOur OSI framework — preparation, contestation, remediation under ECB / JST inspection. Different mandate type, different pressure curve, different protocol.
Applies to: ECB on-site inspections on IRB, ICAAP, credit risk, liquidity, governance.
DELIVERY™
Regulatory programmesOur programme framework for multi-year regulatory mandates — Basel IV, CRR3, CCF RTS, ITS on supervisory reporting. End-to-end with audit-ready trace.
Applies to: multi-year prudential transformation programmes.
Our target-operating-model framework for Risk & Finance transformation. Cost, data, architecture. Not invoked on stress-test mandates.
Applies to: Risk / Finance TOM redesign, data platform programmes, cost-out.
In short: ANCHOR governs what happens before any scenario is translated into numbers. PRISM governs what happens when it is. The two run together on every stress-test mandate; the other two frameworks address different mandate types.
7. Data-quality commitments
Every input that enters a pillar passes five data-quality tests. The tests are run at T-0 and re-run at every subsequent gate. A failure at any gate halts progression until remediation is signed.
01
Every counterparty, position and cashflow reconciled to the GL with zero unexplained residuals. Sample-based audit on 5-10% of rows.
02
Lineage-to-source trace
GateEvery field traces back to the source system of record. No manually-keyed overrides without partner countersignature.
03
The designated owner of each input (per ANCHOR-OWN) signs the dataset at T-0 and re-signs at every update cycle.
04
Challenger comparison
GateIndependent challenger rebuilds each input from source on a sample basis. Variance >2% triggers full re-review.
05
Post-submission archive
GateFrozen snapshot retained in tamper-evident archive for the longer of 7 years or the next full cycle, whichever is later.
These commitments are written into every ANCHOR engagement letter. They are not aspirational — they are contractually binding on the mandate and on the partner-in-charge.
Licence and corrections
ANCHOR™ is an Ezelman proprietary framework. This technical note is released for reader reference; the framework may be cited with attribution but not re-used as the house methodology of another adviser.
If you spot an error, a misreference to an EBA paragraph or a calibration we should update, write to contact@ezelman.com. Substantive corrections are acknowledged in a dated revision log. Figures in this note are sourced to EBA 2018 / 2021 / 2023 / 2025 transparency disclosures and to the Ezelman EU G-SIB stress-test dataset; methodological statements are sourced to the EBA methodological note of the cycle in question.