03 · Credit risk
Art 111–191 · 271–311 CRR
Purpose & history
The Basel II bargain — granted, then constrained
Basel II granted (2004)
- Own PD, LGD, EAD estimates for approved banks
- Unlimited capital benefit vs the standardised answer
- Bank-wide approvals; supervision of the rating system
- The theory: risk sensitivity aligns capital with truth
Basel IV constrains (2017 → CRR3)
- A-IRB withdrawn where data cannot support it: large corporates, institutions; equity out entirely
- Floors under every remaining input (PD 5bp, LGD 25%…)
- Benefit capped at portfolio level by the output floor
- The lesson: incentives beat validation when data is thin
Basel I's four flat buckets invited arbitrage — keep what the buckets underprice, securitise the rest. Basel II's answer created the opposite problem: every optimistic parameter saved real capital, and the discipline meant to prevent it proved weaker than the incentive. Post-2008 evidence, and in the euro area the ECB's TRIM on-site inspection of internal models, confirmed unjustified variability, chronic low-default-portfolio problems and downturn parameters that were not downturn at all. The finalisation is best read as a partial withdrawal of the bargain: models stay where defaults are plentiful enough to learn from, and the standardised approach was rebuilt — LTV real estate, SCRA, transactors — to be a credible yardstick for the floor.
The lineage
1988
Basel I: the 8% Cooke ratio, four risk buckets. Simple, comparable, arbitrageable.
2004
Basel II: the IRB bargain — own estimates in exchange for supervised rating systems.
2008
The estimates meet a downturn. Low-default portfolios, optimistic LGDs, procyclicality.
2016–21
ECB TRIM: multi-year on-site review of internal models across significant institutions; findings become binding model decisions.
Dec 2017
Finalisation: constrained IRB + rebuilt SA + output floor.
2019–22
EBA IRB repair: PD/LGD estimation and downturn guidelines — the approval standard in practice.
2025
CRR3 applies: new SA live, IRB constraints live, floor phase-in begins.
Requirements
Standardised approach — what "no model" still requires
Rating due diligenceDocumented review of rated counterparties; the external rating may only be overridden upward. A process with evidence, not a policy sentence.
Real-estate valuation governanceIndependent valuation at origination, the revaluation cap, indexation policy, ETV lineage into the capital engine.
Classification enginesTransactor identification, IPRE vs general, ADC criteria, SCRA grading of unrated banks — each a data flag someone must own.
CRM eligibility filesCollateral enforceability opinions, guarantee documentation, haircut application, maturity-mismatch tracking.
IRB — the approval burden, grouped
Data history≥ 5 years retail PD, ≥ 7 years corporate PD/LGD/CCF — in practice a full cycle including genuine downturn observations.
Rating system standards≥ 7 non-default grades, meaningful differentiation, annual re-rating, governed overrides.
Estimation standardsLong-run default-weighted averages, downturn LGD add-ons, margin of conservatism per identified deficiency — the EBA repair GL are effectively part of the approval.
Use testRatings demonstrably drive approval, limits, pricing, provisioning — not capital alone.
ValidationAnnual, independent: discrimination, calibration, stability; findings tracked to closure.
Roll-out & partial useA governed map of which classes stay SA — liberalised under CRR3, but a choice, not drift. TRIM decision letters remain binding on top.
Calculations
The same mortgage, three ways — the floor debate in one chart
€300k loan · property €375k · ETV 80% · retail obligor
1 · SA, loan-splitting
Secured 55% × €375k = €206.25k @ 20% → €41.3k · rest €93.75k @ 75% → €70.3k · RWA €111.6k, avg ≈ 37%
2 · SA, floor transitional (Art 465(5), S-TREA only)
€206.25k @ 10% + €93.75k @ 45% = €62.8k, avg ≈ 21%
3 · IRB, formula worked through
PD = 1.0%, LGD = 10% (both above their floors), R = 0.15, retail → no maturity adjustment.
K = 0.10·N( (N⁻¹(0.01) + √0.15·N⁻¹(0.999)) / √0.85 ) − 0.10×0.01
= 0.10·N( (−2.326 + 0.387×3.090)/0.922 ) − 0.001 = 0.10·N(−1.225) − 0.001 = 0.0100
RW = 12.5 × K ≈ 12.5% → RWA ≈ €37.7k
K = 0.10·N( (N⁻¹(0.01) + √0.15·N⁻¹(0.999)) / √0.85 ) − 0.10×0.01
= 0.10·N( (−2.326 + 0.387×3.090)/0.922 ) − 0.001 = 0.10·N(−1.225) − 0.001 = 0.0100
RW = 12.5 × K ≈ 12.5% → RWA ≈ €37.7k
Corporate, counterparty & funds
Unrated corporate €10m — three treatments
a · plain: €10m × 100% = €10.0m
b · IG + floor transitional (S-TREA): × 65% = €6.5m
c · SME: × 85% = €8.5m, then factor: (2.5×0.85×0.7619) + (7.5×0.85×0.85) = €7.04m
b · IG + floor transitional (S-TREA): × 65% = €6.5m
c · SME: × 85% = €8.5m, then factor: (2.5×0.85×0.7619) + (7.5×0.85×0.85) = €7.04m
CIU — mandate-based
€100m fund, mandate permits 100% unrated corporates → MBA: RWA €100m regardless of holdings · no information → FBA 1250% = €1.25bn.
SA-CCR — unmargined 7y IRS
N = €100m · V = +€2m · C = €0
RC = €2.0m · SD = (1−e^(−0.35))/0.05 = 5.91 · adj. N = €590.6m · δ = 1 · MF = 1 · SF = 0.5% → AddOn €2.95m · mult = 1 → PFE €2.95m
EAD = 1.4 × 4.95 = €6.93m
RC = €2.0m · SD = (1−e^(−0.35))/0.05 = 5.91 · adj. N = €590.6m · δ = 1 · MF = 1 · SF = 0.5% → AddOn €2.95m · mult = 1 → PFE €2.95m
EAD = 1.4 × 4.95 = €6.93m
Cleared with a QCCP
EAD €50m × 2% = RWA €1.0m vs ~€30–40m bilateral; KCCP-based default-fund charge on top. The saving migrated into margin liquidity.
Changes & sources
CRR2 → CRR3, the working list
| Area | Change |
|---|---|
| SA real estate | Loan-splitting & ETV whole-loan tables; IPRE and ADC own classes; valuation capped at origination |
| SA counterparties | SCRA; CQS2 institutions 30%; due-diligence override; equity 250/400% (phase-in); subordinated 150%; UCC CCF 10% (phased) |
| IRB scope | A-IRB withdrawn for large corporates & institutions; equity out; partial use liberalised |
| IRB parameters | Input floors; F-IRB senior unsecured LGD 45→40%; 1.06 scalar deleted |
| CRM | Own-estimate volatility adjustments abolished; F-IRB secured LGD* recalibrated |
| New | Retail FX multiplier ×1.5 (Art 123a); crypto transitional (Art 501d) |
Where the law lives
Art 111–191 CRR plus the Level 2 layer: EBA PD/LGD estimation and downturn GL (IRB repair), RTS on economic downturn, connected-clients GL, ECAI mapping ITS. For ECB banks, TRIM decisions remain binding model-by-model.
Common findings
- ETV data quality; revaluation-cap breaches in mortgage books.
- Transactor flags unavailable → the 45% weight forfeited.
- Downturn LGD and MoC below EBA-GL standard — the recurring TRIM finding.
Independent analysis by Elderlake Partners, as at August 2026. Not legal, accounting or supervisory advice.