Climate risk has moved from stand-alone ESG overlay to credit-risk driver, and the practical frontier in 2026 is the explicit translation of physical and transition risk into PD and LGD. A UNEP FI survey found 61% of respondents already incorporating climate risk into PD modelling and 43% into LGD measurement. ECB supervisory work points the same direction: the strongest practice has moved past generic qualitative references toward hazard-specific physical-risk models and climate-conditioned PD and LGD frameworks, with implementation still uneven across banks.
The engineering question is no longer whether climate belongs in credit models. It is which transmission channels to model, how scenarios feed them, and which integration pattern survives both validation and a supervisory review.
How climate enters PD
For physical risk, the mechanism is exposure to impairment. Hazard exposure is converted into expected borrower cash-flow damage, operational disruption or asset impairment, and default propensity is re-estimated at obligor or segment level. In practice this runs on geospatial exposure mapping, hazard intensity metrics and vulnerability proxies: location, asset type, insurance cover, and adaptation or resilience measures. At portfolio level, a BIS working paper shows banks using Vasicek-type models can introduce physical risk as a new independent systemic risk factor for unexpected losses, and a Cambridge framework goes further by adjusting credit risk explicitly for insurance adequacy and resilience investment.
Transition risk enters through the income statement: revenue pressure, cost inflation, limits on passing carbon prices through, asset stranding and deteriorating market access. ECB good practice includes carbon-tax shocks in corporate stress tests, linked to borrower-specific financials and sectoral sensitivities rather than applied as a flat industry haircut.
The survey evidence on method is sobering and useful. The most common PD approaches are adjusting existing models to include climate factors, using stress-testing outputs to inform PD adjustments at portfolio or obligor level, and manually overriding model outputs for highly exposed counterparties. Overlays and qualitative modules dominate over full redevelopment, which reflects data and calibration constraints more than ambition. For corporates, the defensible route is usually sector-region segmentation with climate variables layered into scorecards or satellite models, preserving the IRB and IFRS 9 infrastructure underneath.
LGD runs through collateral
LGD integration is more concrete because the channel is physical. The working methods are collateral haircuts, asset-level damage estimates, loan-to-value deterioration and stressed recovery assumptions. In mortgage and real estate portfolios, the dominant path is damage to collateral value from floods, heat, wildfire or chronic degradation, feeding through to revised recoveries. For corporate lending, the channel is stranded-asset risk and reduced resale value of pledged assets in sectors facing carbon-cost shocks.
ECB good practice is specific here: EPC-linked collateral revaluation, carbon-tax impacts on property cash flows, emission-intensity parameters for corporate collateral, and, critically, a conditional PD-LGD relationship. That last point is the one models most often miss. A borrower that becomes more likely to default under climate stress usually also shows weaker recovery prospects, and treating PD and LGD adjustments as independent understates expected loss precisely in the stressed states the exercise exists to measure.
Scenarios need a transmission layer
NGFS scenarios are the backbone of climate stress testing because they provide comparable transition and physical pathways over long horizons. The recurring implementation error is feeding NGFS outputs directly into credit models. Good practice interposes a transmission layer: an intermediate model that converts carbon prices, macroeconomic deterioration, flood damage and property price paths into borrower-level and collateral-level impacts before anything touches PD or LGD. ECB reviews assess exactly this, whether institutions carried climate shocks into PD and LGD through channels such as carbon-price shocks, gross value added deterioration and flood impacts.
The credible 2026 setup is a multi-horizon stack: short-term acute physical shocks, medium-term transition repricing, and long-term chronic hazard and adaptation effects, all mapped to portfolio segmentation. Reporting is converging on the same grain. EBA ESG templates now include 12-month PD, lifetime PD, LGD and LTV, built to monitor credit deterioration from climate events over time, which means the model outputs and the regulatory reporting need to reconcile.
Data, evidence, and the greenwashing trap
The binding constraint is data: geolocation, EPCs, emissions, borrower-level resilience measures and reliable asset-level hazard data. ECB-related commentary keeps flagging the same gaps, and they explain why physical-risk PD and LGD sensitivity is still limited in scope at many banks. Vendor data helps only when outputs link to counterparty identifiers, property coordinates, sector classifications and scenario horizons. A standalone climate score that cannot be traced to hazard exposure or cash-flow mechanics is decoration, and increasingly a liability.
That liability has a name. The greenwashing risk in climate-credit modelling is overstating how far climate variables are embedded in decisions when they exist only in disclosures or loose overlays. Supervisory scrutiny is moving toward evidence that climate metrics affect credit approval, limit-setting, pricing, monitoring, capital planning and stress testing, not only reporting.
Operationally, the choice of integration pattern should follow materiality and data, not fashion: a minimal viable setup that geocodes priority exposures and applies simple overlays where hazard is material; scorecard augmentation that adds climate predictors while preserving the rating architecture; a scenario-overlay engine that leaves base models untouched; a collateral-led LGD revaluation for property-heavy books; or a two-layer design with a statistical baseline and a climate overlay on top. Whichever pattern wins, the audit trail carries the weight. Every climate adjustment to PD or LGD needs its transmission channel, model dependency, override policy and governance documented, because the supervisory question has stopped being whether you considered climate and become how exactly it changed the number.
FAQ
Do we have to rebuild our IRB models to include climate?
No, and most banks are deliberately not doing so. The survey evidence shows overlays and adjustments to existing models are the dominant approach, and supervisory good practice accepts scorecard augmentation and scenario overlays that preserve the core IRB architecture. What is expected is that the adjustment is documented, calibrated as well as the data allows, and consistent between PD and LGD, not that the rating system is redeveloped from scratch.
Which exposures should be geocoded first?
Start where collateral carries the risk: mortgage and commercial real estate books in hazard-prone regions, because the transmission from flood or heat damage to collateral value to LGD is the most direct and the most supervisable. Then move to corporate exposures in carbon-intensive sectors where transition costs hit cash flows. Full-portfolio geocoding can follow; materiality-first sequencing gets defensible results while the data effort continues.
How do we defend a climate overlay to model validation?
Document the transmission channel before the adjustment. Validation needs to see how a hazard or carbon-price variable becomes a borrower cash-flow or collateral-value impact, which model produced the sensitivity, what override policy applies, and who approved it. An overlay with a named mechanism, a stated calibration basis and a review cycle passes; a percentage add-on justified by expert judgement alone increasingly does not.