The Four Axes of RWA Optimisation Under CRR3 — And Where Banks Are Leaving Capital on the Table

CRR3 Is Not a Compliance Event. It Is a Capital Repricing.

In January 2025, the Capital Requirements Regulation 3 (CRR3, EU 2024/1623) went live. Four months in, most European banks are still running firefighting mode: model recalibrations, validation updates, FRTB tooling deployments, CVA framework migrations. These are necessary. They are not sufficient.

The institutions that will emerge with the greatest capital headroom—and the lowest cost of risk-taking—are those treating CRR3 as a strategic capital problem, not a regulatory filing problem. That distinction matters. A bank running axis-by-axis workstreams (Axis 1: output floor; Axis 2: IRB models; Axis 3: SA-CCR; Axis 4: FRTB) in isolation will almost certainly leave 15–20% of available RWA reduction on the table. The interactions matter far more than the sum of the parts.

This article lays out the four axes explicitly. It describes where the levers are. And it names the most common mistake: treating each axis as an independent problem.

Axis 1: Managing the Output Floor

The output floor is non-negotiable. Under CRR3, a bank's total RWA cannot fall below 72.5% of what it would have been calculated under the Standardised Approach (SA). This is the binding constraint for most large IRB banks operating in mortgages, corporate credit, and specialised lending.

In practice, the floor hits hardest on portfolios where IRB-modelled capital is materially lower than SA-modelled capital. A low-PD corporate portfolio (predominantly AAA-to-A rated names) will see IRB RWA at perhaps 25–35% of exposure; the SA equivalent sits at 50–100%. The delta is where the floor bites. For a EUR 10bn portfolio of EUR-denominated AAA-rated corporates:

• IRB RWA: roughly EUR 2.5–3.5bn (25–35% RW)
• SA RWA: roughly EUR 5bn (50% RW)
• Output floor impact: IRB RWA is locked to 3.6bn minimum (72.5% of EUR 5bn)

So even though the bank's IRB model says the capital charge is EUR 2.5bn, CRR3 forces it to hold EUR 3.6bn. The bank has three responses:

Lever 1: Portfolio Mix. Shift capital away from low-PD corporates into higher-PD, higher-equity-weighted segments (subordinated lending, certain SME portfolios). Where the IRB RWA is closer to SA-equivalent, the floor is not binding. This lever is crude—it means exiting profitable business—but it is real.

Lever 2: Collateral Optimisation. Secured lending (mortgages, reverse-repo, pledge-backed facilities) has lower LGD under both IRB and SA. A EUR 5bn mortgage portfolio might sit at a 15% RW in SA and 10–12% in IRB; the floor (10.875%) barely bites. Shift marginal capital from unsecured to secured: it lowers both IRB and SA RWA, so the floor stays non-binding. This is operationally expensive but increasingly common.

Lever 3: Hybrid Approach. Voluntarily exit IRB for specific portfolios. This is provocative, but it is not irrational. If a bank's low-PD corporates are hitting the output floor anyway, switching them to SA and applying the (now-identical) SA RWA across both sides of the floor eliminates the binding constraint. The bank regains strategic capital optionality. The trade-off: it sacrifices IRB benefit on those positions permanently and faces supervisory pushback on backsliding. But for some banks, this is the cost-benefit winner.

Axis 2: IRB Model Recalibration and Input Floors

CRR3 introduced new, tighter input floors. These are not advisory; they are binding constraints on model parameters. A bank's estimated PD for a corporate obligor cannot fall below 0.05% (previously 0.03% under CRR2). LGD floors: 25% for unsecured corporate, 10% for mortgage collateral, 15% for other secured exposures. EAD floors under IRBA (the IRB Advanced Approach) are now tighter, and the scope for collateral haircuts under the IAA (Internal Assessment Approach) is narrower.

These floors eliminate some of the "optimisation" that banks were doing under CRR2. A bank with a large portfolio of AAA-rated corporates that had calibrated PDs of 0.015% is now forced to use 0.05%. That is a roughly 3.3x increase in capital charge on that segment. For a EUR 5bn portfolio, that might mean EUR 150–250m of incremental RWA.

But there is still discretion within the bounds. Here are the levers:

Recalibration of Foundation IRB (FIRB) transition matrices. CRR3 allows—even encourages—banks to recalibrate their state-of-the-art probability-of-default (PD) estimates using the most recent data. If a bank has overestimated PDs on certain cohorts (e.g., mid-market manufacturing firms in stable jurisdictions), a data-driven recalibration can defensibly lower the median PD towards the floor but not below it. The supervisory tolerance is high for recalibrations backed by rigorous backtesting.

LGD granularity and collateral modelling. The 25% unsecured-corporate LGD floor is a macro-floor. A bank can still estimate LGDs by collateral type, seniority, and recovery lag if the methodology is defensible. Many banks are finding EUR 200–400m of headroom by disaggregating their LGD models and proving that certain tranches (e.g., senior-secured trade finance) warrant LGDs of 15–20%, not 25%.

The backtesting risk. A critical caution: over-optimising models to run below CRR3 floors often triggers backtesting failures when live P&L diverges from estimates. If a bank recalibrates PDs downward and then sees obligor defaults spike in an economic downturn, the model fails the 1-in-100 test (traffic light framework). The supervisory penalty—higher RW multipliers, potential IRBA downgrade to FIRB—is severe and may exceed any benefit from the initial optimisation.

Axis 3: SA-CCR and CVA Reform — The Underestimated Drag

CRR3 replaced the Credit Exposure Method (CEM) with the Standardised Counterparty Credit Risk (SA-CCR) approach for derivatives exposure. This change is often treated as a technical update. It is not. For banks with large derivatives books, it is a capital shock.

SA-CCR calculates exposure as a weighted sum of: (i) replacement cost (RC), i.e., mark-to-market; plus (ii) potential future exposure (PFE), a function of notional, maturity, and asset class volatility. The formula is more granular than CEM, but it is also more exposed to notional size. A EUR 100m interest-rate swap, under CEM, had PFE of roughly EUR 2–3m. Under SA-CCR, the same swap now has PFE of roughly EUR 4–6m, depending on maturity. The RWA impact: +15% to +40% on the aggregate derivatives portfolio.

Lever 1: Netting agreements. SA-CCR allows bilateral netting on derivatives under a master agreement only if the netting agreement has been stress-tested against the relevant jurisdiction's bankruptcy law. Most large banks have documented this, but secondary counterparties and certain emerging-market relationships may not. A bank can identify EUR 50–100m of RWA headroom by formalizing netting documentation with partners where it has not yet been done.

Lever 2: Collateral optimisation. SA-CCR reduces RC by the amount of eligible collateral held. VA (variation margin) also reduces PFE. A bank with bilateral margin agreements on a major share of its derivatives portfolio can reduce SA-CCR RWA by 10–20%. The leverage: negotiate term extension and frequency of margin calls on legacy trades to bring more collateral into scope.

Lever 3: Trade compression and novation. Compressing redundant trades (particularly long-dated swaps that offset each other) reduces both notional and PFE. A bank's EUR 2tn derivatives notional might compress to EUR 1.8tn with minimal P&L impact. The RWA reduction: 5–8%.

Lever 4: CVA framework expansion. CRR3 tightened CVA (Credit Valuation Adjustment) capital requirements. Fewer banks now qualify for the Advanced CVA approach; most are pushed to the Standardised CVA method, which is more conservative. But the threshold for CVA exemption (NCVA designation) is still available for banks with gross derivative notional below EUR 100bn. If a bank can defend an NCVA claim through documentation, it avoids CVA capital entirely. This is worth EUR 50–150m RWA for many counterparty portfolios.

Axis 4: FRTB — IMA vs SA and the Market Risk Boundary

The Fundamental Review of the Trading Book (FRTB) went live in January 2025. Banks can model market risk using either the Internal Model Approach (IMA) or the Standardised Approach (SA). The IMA is more capital-efficient but requires supervisory approval through a grueling 18–24 month validation process. Most banks are running SA indefinitely—and quietly accepting 10–25% higher market risk RWA.

But the real lever is not IMA vs SA. It is the trading/banking book boundary.

The boundary problem. Under CRR3, a position is in the trading book if it is: (i) actively traded; (ii) managed on a trading desk for the purpose of profit-and-loss (P&L) realisation; and (iii) subject to market-risk backtesting. Many banks have positions that are technically tradeable but are held for structural reasons (e.g., a large corporate lending portfolio with embedded optionality, or a bond portfolio held for ALM). If these can be moved to the banking book, they use credit risk RWA (which may be lower) instead of market risk RWA. Under FRTB SA, bond RWA can be 40–60% higher than under credit risk models on the same position.

P&L attribution testing. The mechanism for defending the banking book boundary is P&L attribution testing (PLAT). The bank must show that the trading P&L can be explained by risk factors in the model with high accuracy (R-squared target: >75%). If PLAT fails, the position is reclassified to the trading book, and RWA jumps. Conversely, if PLAT passes and the desk is certified as non-trading, it stays in banking book capital. A large bank working systematically through PLAT can recover EUR 200–500m of RWA by boundary reclassification.

IMA approvals for selected desks. For desks with high volumes, tight market liquidity, and disciplined risk governance, IMA approval is worth the investment. A equity market-making desk that qualifies for IMA might see 15–25% RWA reduction vs SA. For a major investment bank with EUR 10bn notional, this is EUR 150–250m. The approval process is 2 years; the payoff horizon is 5–10 years; the ROI is strong. But it requires operational excellence and supervisory credibility. Few banks will achieve this in 2025–2026.

The Mistake Banks Are Making Right Now

The winners are running an integrated capital model. They ask: given the output floor constraint, which portfolios should stay in IRB vs SA? Given that decision, what recalibrations are possible without triggering backtesting failures? Given those parameters, where does it make sense to optimize collateral and netting (Axis 3)? And given all of that, which desks should be pushed towards the banking book boundary and which should pursue IMA? That is a systems problem, not a checklist.

Compliance Approach vs Strategic Approach

FAQ: CRR3 RWA Optimisation