Elderlake Partners

02 · Output floor

Art 92(3)–(7) · Art 465 CRR

Purpose & history

The failure it answers

By the early 2010s the Basel Committee's hypothetical-portfolio exercises had produced an uncomfortable result: identical portfolios, handed to different IRB banks, came back with risk-weighted assets differing by a factor of three to four. Some dispersion is legitimate — different data, different workout experience — but the studies showed a large residual that could only be modelling choice. Once that was public, the credibility of every model-based ratio was in question: 13% CET1 at one bank was simply not the same claim as 13% at another.

Bank A
34
Bank B
55
Bank C
76
Bank D
118

One hypothetical portfolio, four IRB banks — model RWA, indexed. Illustrative of the BCBS 2013 banking-book RCAP findings.

Intent, in one line

The floor does not fix the models; it caps how much they can matter. Internal approaches stay worth up to a 27.5% saving — enough to reward genuine risk management, not enough to make modelling the business model. The price is risk-sensitivity at the very low end, and that trade was made knowingly.

How it was fought over

2013
RCAP hypothetical-portfolio studies publish the dispersion — the floor's origin document.
2016
Negotiations stall on the calibration: the US pushes high, France and low-RWA Europe resist. The package misses its deadline by a year.
Dec 2017
72.5% agreed — the single hardest number in the finalisation.
2021
CRR3 proposal: "parallel stacks" (floor only in the Basel ratio) vs single stack. Single stack wins; the transitional package is the price — mortgages, unrated corporates, p-factors, pushed by NL, DK, SE.
2024
CRR3 adopted (Reg. (EU) 2024/1623).
2025
Phase-in begins at 50%.
2030
72.5% steady state.
end-2032
Transitionals lapse — the cliff.

Requirements

What a bank must actually operate

Shadow standardised calculationFull S-TREA every quarter across all risk types — SA-CR, SA-CCR (also for IMM banks), BA-/SA-CVA, SMA, standardised market risk. Same cut-off and scope as the internal run, reconciled to it. A production system, not a spreadsheet.
SA input data at IRB banksData IRB banks never maintained to production standard: ratings coverage and ECAI mapping, ETV at origination with the revaluation cap, transactor flags, SCRA assessments. Lineage and controls are a supervisory expectation.
Dual-stack disclosurePillar 3 templates show U-TREA and S-TREA side by side. The floor's bite is public; analysts track the ratio.
Capital planning through the phase-inICAAP projections must show the binding constraint switching as x steps up — and the 2032 cliff where the transitionals matter. Plans that ignore the cliff read as findings.
P2R double-count hygieneAnalysis demonstrating that risks capitalised by the floor are not re-charged in Pillar 2 — prepared by the bank, relied on by the supervisor.
Level of applicationFloor computed at each consolidation level (member-state discretion for sub-levels) — group and solo functions must agree on data and method.

Calculations

When the floor binds — the switch year, visually

IRB mortgage bank, U-TREA = 60, S-TREA = 100 (both stable). Track = S-TREA; the black marker is U-TREA; the bar is x · S-TREA. The larger of the two is that year's TREA.

2025 · x=50%
50
TREA €60bn · models
2026 · x=55%
55
TREA €60bn · models
2027 · x=60%
60
TREA €60bn · knife edge
2028 · x=65%
65
€65bn · floor +8%
2029 · x=70%
70
€70bn · floor +17%
2030 · x=72.5%
72.5
€72.5bn · floor +21%

The switch year depends only on the ratio U/S — here 60%, so the floor binds from the first year x exceeds it.

The cap, and the cliff

Transitional cap — Art 465(2), where exercised Floored TREA ≤ 1.25 × unfloored. 2030: min(72.5, 1.25 × 60) = €72.5bn — not binding here. It protects only banks with U/S below 72.5/1.25 = 58%, i.e. the very lowest-RWA models.
What the mortgage transitional is worth Dutch book €50bn; average SA RW 37% (loan-splitting) vs 21% under Art 465(5).
S-TREA: €18.5bn vs €10.5bn → €8bn difference → 72.5% × 8 = €5.8bn of TREA once the floor binds. That is the 2033 cliff in one number.
Reading it as capital €5.8bn TREA × a 13% all-in CET1 stack ≈ €750m of CET1 — the order of magnitude a €50bn mortgage book must plan to raise, retain or price in before 2033, absent an extension.

Changes & sources

CRR2 → CRR3

BeforeNow
Basel I floor (Art 500, largely spent) — no meaningful RWA floorOutput floor on total TREA, phase-in 50→72.5%
IRB benefit unlimited in EU-specific requirementsSingle stack: floor propagates into P2R, buffers, MREL
—Transitional S-TREA calibrations to 2032 (mortgages, unrated corporates, p-factors)

Where the law lives

Art 92(3)–(7) and Art 465 CRR; disclosure in the revised Pillar 3 ITS; EBA monitoring reports track the bite per country. Review clauses cover the transitionals — extension requires new legislation, not a delegated act.

Open items

  • The end-2032 cliff — the largest unpriced parameter in low-RWA capital plans.
  • Leverage interaction: as floored RWA rise, some banks switch binding constraint from leverage to floor — plan both.

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