Every loan above €25k, every counterparty, ~95 attributes, monthly. For a decade it was an ECB statistical return. Under CRR3 — and under the 2026 wave of JST deep-dives — it is the dataset supervisors triangulate against FINREP and COREP when they want the truth.
Three companion pieces cover what actually changes in each of the three reporting frameworks that matter most to supervisors. Read them in order.
35+ template changes under CRR3, Output Floor, FRTB, CVA. The headline rebuild — already in most 2026 plans.
IFRS 9 stage-migration, three-way reconciliation, ESG overlay, and the EBA data-quality regime.
~95 attributes per loan, ECB/2016/867. How supervisors triangulate it against COREP and FINREP.
AnaCredit went live in September 2018 as a statistical reporting obligation under Regulation ECB/2016/867. For several years it was read mostly by the ECB's statistics function. That changed quietly. By 2023 AnaCredit had become a primary dataset for SSM supervision; by 2026 it is the dataset that lets JSTs answer questions COREP and FINREP cannot.
Commitments of €25k and above to legal-entity counterparties are in-scope — so essentially the corporate and SME book, most commercial real estate, and a substantial share of specialised lending. Retail lending to natural persons and intra-group exposures are out of scope. For most G-SIBs, AnaCredit captures the bulk of where credit risk actually lives.
Three linked datasets. The instrument record captures the loan (origination date, currency, interest rate, maturity, PD, LGD, accounting treatment). The counterparty record captures the obligor (NACE, size, relationship). The protection record captures the collateral. The three join into a supervisory view of every large-corporate credit line in the euro area.
Unlike FINREP and COREP (quarterly), AnaCredit arrives at NCBs monthly. The JST has a rolling view of the book — migration, origination, prepayment, default. When a supervisor says “we see that your corporate origination slowed in March”, that statement is based on AnaCredit, not on FINREP delivered two quarters later.
AnaCredit feeds the SSM supervisory analytics tools, the EBA EU-wide stress test data requests, risk-indicator production at the EBA and ECB, and — increasingly — the ECB macroprudential function and the climate risk analytics team. Your AnaCredit return is not a single-consumer dataset.
The ~95 attributes cluster into five groups. Data-lineage gaps tend to be concentrated in groups 3 and 4 — collateral and accounting overlays — where the AnaCredit logic sits downstream of multiple source systems.
| Cluster | Examples | Typical source system | Lineage risk |
|---|---|---|---|
| 1 · Counterparty reference | Legal Entity Identifier, NACE code, institutional sector, size class, parent | Counterparty master / KYC / CRM | MEDIUM |
| 2 · Instrument terms | Origination date, currency, interest rate type, legal final maturity, repayment frequency, purpose | Core banking / loan engine | LOW |
| 3 · Protection & collateral | Protection type, allocated value, third-party guarantor, date of valuation, type of collateral location | Collateral management system — often the weakest link | CRITICAL |
| 4 · Accounting & risk overlay | Accounting classification, impairment stage, IFRS 9 transfer flags, PD, LGD, default status | Credit risk system + accounting engine — two feeds that must reconcile | CRITICAL |
| 5 · Financial data | Outstanding amount, accrued interest, arrears, write-off history, off-balance amount | GL feed / loan servicing | MEDIUM |
Clusters 3 and 4 carry the majority of supervisory pushback. Collateral attributes require downstream logic that is rarely owned by one team; accounting-and-risk overlays require two engines to agree. Most data-quality findings we see in AnaCredit trace back to one of three failure modes: collateral valuation staleness, IFRS 9 stage attribution mismatch with the credit risk system, or default flag divergence between the two.
When a supervisor sends an ad-hoc request for a portfolio view your reporting team cannot reconstruct in under a week, the supervisor is almost certainly comparing it to what AnaCredit already tells them. Five uses matter most.
AnaCredit is the most granular view of single-name and group-of-connected-counterparties concentration. The JST can re-derive your large-exposure return independently and compare. Divergence triggers a reconciliation request.
AnaCredit carries your PD and LGD at loan level. The ECB's centralised model-landscape analytics (SSM Supervisory Analytics) use AnaCredit to test whether model outputs are stable, whether overrides are material, and whether your default rates match what your PDs predict at portfolio segment. This is one source of on-site inspection targets.
For the EBA EU-wide stress test and ECB climate stress exercises, AnaCredit is the primary source for the starting-point book — exposure by segment, sector, geography (EBA ST Methodological Note; ECB climate stress-test framework). Banks whose AnaCredit does not match their internal stress-test book spend the first three months of the exercise reconciling, not projecting.
Financed-emission estimation, sector-transition exposure, physical-risk mapping — all use AnaCredit as the granular anchor because FINREP alone is too aggregated. Any material mismatch between the AnaCredit NACE tagging and the FINREP sector breakdown becomes a supervisory question.
The ECB benchmarks banks against peer groups on origination spreads, collateralisation patterns, forborne-exposure trajectories, and IRB parameter distributions — all at AnaCredit granularity. When a bank is an outlier, the conversation starts: “explain why your pattern differs”.
A loan reported in AnaCredit must appear in FINREP exposure and in COREP RWA. The three are calculated by different systems, aggregated differently, and reviewed by different teams. When they do not triangulate, the JST notices.
Sum of in-scope AnaCredit instrument outstandings should reconcile to the corresponding FINREP exposure class. Scope differences (threshold, natural-person exclusion) explain most, but not all, of the gap. An unexplained residual signals a data-feed issue.
The AnaCredit default flag must be consistent with the CRR Art. 178 default definition (EBA/GL/2016/07) as implemented in COREP. Divergences at loan level propagate into IRB parameter estimation and RWA. A diverging pair is a common finding in recent OSIs on credit risk.
The IFRS 9 stage on the AnaCredit record must match the stage reflected in FINREP F 18. Mismatch usually means the accounting engine and the credit risk system have drifted on SICR triggers or on the Stage-3 (default) boundary.
The AnaCredit allocated protection value must be consistent with CRM-recognised collateral in COREP. Discrepancies are often explained by CRM eligibility rules — but the bank needs to be able to explain each one on demand, not retrospectively.
AnaCredit is no longer a statistical return managed by a reporting team in the finance function. It is a supervisory dataset. That shift has five concrete consequences for how banks manage the data.
Every AnaCredit attribute needs a traceable source-system lineage with a named owner. The typical “AnaCredit engine reads from three upstream systems” story is not sufficient when a JST asks where a single collateral value came from and why it changed in the last monthly snapshot.
The credit risk system's default flag, the accounting engine's Stage-3 flag, and the AnaCredit default status must be one flag, not three. Where they diverge, establish a reconciliation control with clear escalation — not a quarterly investigation.
Collateral is the weakest data link at most banks. AnaCredit requires attributes (valuation date, valuation method, location) that many legacy collateral systems do not cleanly carry. Plan the collateral data-uplift as a multi-year programme, not a spreadsheet workaround.
AnaCredit-FINREP-COREP reconciliation should be a monthly attested control with a named owner, not a spreadsheet on demand. If the JST sends an ad-hoc request, the bank should be able to answer in days, not weeks.
Any forthcoming on-site inspection on credit risk, IRB, or IFRS 9 will start from an AnaCredit-anchored data pack. The bank's own data pack to the inspection team needs to come from the same source. A divergence between the two packs is the first week of the inspection wasted.
Most internal audit functions review financial reporting. Few review AnaCredit with the same rigour. That should change in 2026. Supervisors are treating AnaCredit as supervisory; internal audit should match the classification.
Dates, paragraph numbers and document versions verified at publication. Where consultations are still open, we flag draft status in text. Corrections: research@ezelman.com.
Partner-led, senior-only, no audit conflict. If the question in this piece is live at your institution, the first conversation is with a partner — not a BD team.
Talk to a partner →