06 · Loss data regime

Art 316–323 CRR

Purpose & history

The dataset that outlived its model

Under AMA, loss data fed capital models; the EU's ILM = 1 severed that link — and kept the dataset anyway. Deliberately: the loss register is the management-information core of operational risk (what actually goes wrong, where, how expensive), the disclosure raw material that lets markets compare conduct records, the Pillar 2 evidence base, and the option value of ever rejoining Basel's loss-sensitive mainstream. The regime's craft is in its details: events are recorded at ≥ €20,000 across ten years; each carries three dates (occurrence, discovery, accounting) because the misconduct wave's defining feature was lag — LIBOR conduct from 2005–09 hit P&L in 2012–15; related losses from one root cause are grouped as one event; and the boundary with credit risk (fraud in lending is credit RWA but op-risk registry) is a standing classification dispute with real consequences either way.

Requirements

Completeness (BI ≥ €750m)Ten years, ≥ €20k threshold, all entities and event types; capture wired into finance, legal and complaints processes, not a survey.

Three-date disciplineOccurrence, discovery, accounting dates per event; provisions and subsequent adjustments tracked to the originating event.

Grouping & boundary rulesRoot-cause grouping policy; credit/op boundary classifications documented and consistently applied.

Recoveries & insuranceRecognised when received, netted transparently; insurance dependence visible rather than buried.

ExclusionsRemoving anything (divested businesses, settled one-offs) needs supervisory permission with an evidence file.

Disclosure & reviewAnnual loss disclosure per templates; independent periodic review of the dataset's quality — it is examined like a financial statement.

Calculations

Net loss of an event Mis-selling settlement: gross €30m − direct recoveries €5m − insurance received €10m = net €15m recorded (gross and recoveries both visible). Provisions booked earlier attach to the same event, not the booking year.

Grouping Three settlements, three years, one flawed sales process → one grouped event, €45m — the dataset should count causes, not invoices.

What the register drives (EU) Pillar 1: nothing — ILM = 1
Disclosure: annual totals and largest events → public
Pillar 2 / SREP: scenario anchoring, conduct-risk capital
Basel comparison: LC = 15 × avg losses; €80m avg → hypothetical ILM ≈ 1.4 → the ≈ €230m of Pillar-1 capital the EU chose not to charge (see SMA panel).

Changes & sources

Sources: Art 316–323 CRR; EBA RTS/GL on the loss data set and ITS on disclosure; Basel OPE25 for the ILM the EU switched off.
Watch: the EBA's ILM review — the dataset is the ready-made bridge back to loss-sensitivity if the EU ever crosses it.

Independent analysis by Elderlake Partners, as at August 2026. Not legal, accounting or supervisory advice.