Climate Risk in Credit Risk Models

Climate risk must enter the credit model through an economic mechanism

Adding a climate score to a borrower rating does not automatically create a climate-sensitive credit model. A credible approach requires an economic transmission mechanism. The question is not whether a borrower is exposed to climate risk in the abstract. The question is how a climate driver changes the borrower's capacity and willingness to repay, the value of collateral, the bank's exposure at default, or the timing and severity of loss.

The credit-risk study included in the source compilation proposes a sector-based transition framework built around carbon intensity, policy exposure, rating adjustments and expected loss. It is useful as an illustrative architecture, but its simulated results should not be treated as universal calibration. The broader lesson is the important one: climate variables need to be linked to credit parameters through observable and testable financial channels.

Schematic of a credit risk parameter chain linking exposure, default probability and loss given default, illustrating "Climate risk must enter the credit model through an economic mechanism" in Integrating Climate Risk into Credit Risk Models: From Physical Hazards to PD, LGD and Lending Decisions — Figure 1. Schematic of a credit risk parameter chain linking exposure, default probability and loss given default, in the context of climate risk must enter the credit model through an economic mechanism.

PD: connect climate drivers to borrower cash flow and solvency

Probability of default should respond to climate risk only where there is a defensible causal link. Physical hazards may reduce revenue, interrupt production, destroy assets, increase operating expenses or require reconstruction spending. Transition shocks can reduce demand, increase energy or carbon costs, require capital expenditure, strand existing assets or raise refinancing costs.

A bank can reflect these effects through stressed financial statements, sectoral scenario variables, rating factors or satellite models. For example, a transition scenario may change energy prices, carbon costs and sector output. Those variables flow into EBITDA, interest coverage, leverage and liquidity. The resulting financial ratios then affect the rating or PD model. This is preferable to imposing an arbitrary climate add-on because the effect can be explained, challenged and back-tested where data permit.

LGD: climate risk is often a collateral and recovery problem

Loss given default may be at least as important as PD for physical climate risk. Flood, wildfire, storm or chronic sea-level exposure can reduce collateral values, increase repair costs, lengthen recovery periods and affect insurance availability. A property that remains economically usable but becomes expensive to insure can still produce a meaningful change in recovery value.

LGD analysis should therefore consider hazard intensity, collateral location, property characteristics, insurance coverage, seniority, cure rates and time to recovery. The HomeEquity Bank material in the compilation provides a practical example: the bank focuses on the long-term valuation and insurability of residential real estate collateral as part of its climate risk strategy.

Overlaid physical hazard intensity layers for flood, wind and heat exposure, illustrating "LGD: climate risk is often a collateral and recovery problem" in Integrating Climate Risk into Credit Risk Models: From Physical Hazards to PD, LGD and Lending Decisions — Figure 2. Overlaid physical hazard intensity layers for flood, wind and heat exposure, in the context of lgd: climate risk is often a collateral and recovery problem.

EAD and lending behaviour can also change

Exposure at default should not be assumed constant in every climate scenario. Acute physical events can create immediate liquidity demand. The ECB working paper in the compilation finds that lending to flood-affected firms increased by roughly 3.5 to 5 percent in the quarter of the event before contracting by a similar magnitude in the following quarter. That pattern is consistent with firms drawing credit for working capital and disruption costs.

This matters for stress testing. A model that increases PD but leaves utilisation and EAD unchanged may understate short-term funding demand. Climate credit modelling should therefore consider contingent drawdowns, maturity changes, credit-line utilisation and restructuring behaviour.

Expected loss is the final output, not the starting point

Expected Loss = PD x LGD x EAD. The formula is familiar; the difficult part is making each component respond consistently to the scenario. The 2019 credit-risk paper in the compilation shows, in a simulated portfolio, how sector transition scores and rating migration can increase expected loss in high-emission sectors. The precise percentages are model-specific, but the methodology illustrates why forward-looking climate variables can materially alter credit profiles.

A stronger implementation would avoid a single sector score as the only risk driver. It would combine borrower-level financial sensitivity, geography, technology, collateral, insurance, adaptation, policy exposure and scenario variables. It would also distinguish between direct emissions exposure and the indirect exposure of financial or service companies to climate-sensitive clients.

Prior and posterior densities illustrating Bayesian parameter updating, illustrating "Expected loss is the final output, not the starting point" in Integrating Climate Risk into Credit Risk Models: From Physical Hazards to PD, LGD and Lending Decisions — Figure 3. Prior and posterior densities illustrating Bayesian parameter updating, in the context of expected loss is the final output, not the starting point.

Model governance is part of the credit architecture

Climate credit models create model risk because data are incomplete, relationships are non-stationary, scenarios are not forecasts and historical default data may not contain the future climate states being modelled. Independent validation should therefore examine conceptual soundness, data relevance, parameter stability, scenario dependence, overrides, conservatism and sensitivity to key assumptions.

A well-designed climate credit framework should be able to answer five questions. What is the climate driver? Through which financial mechanism does it affect the borrower? Which credit parameter changes? How large is the change under each scenario? What decision changes because of the result? If those questions cannot be answered, the model is not yet ready for credit decision use.

Conclusion

The key requirement is decision usefulness. Climate analysis adds value when it improves risk identification, pricing, capital allocation, portfolio management or governance, while making uncertainty and model limitations explicit.

Sources

Saadu-Ayuba, M. et al. Integrating Climate Risk into Credit Risk Models. 2019. DOI: 10.36349/EASJEBM.2019.v02i12.017

Monnin, P. Integrating climate risks into credit risk assessment. https://www.cepweb.org/wp-content/uploads/2019/02/CEP-DN-Integrating-climate-risks-into-credit-risk-analysis.pdf

Ackerer, D. and Filipovic, D. Linear credit risk models. https://doi.org/10.1007/s00780-019-00409-z

European Central Bank. Working Paper Series No. 3224, Physical climate risk, credit risk and lending activity. https://www.ecb.europa.eu

Disclaimer: This article is professional risk-management analysis and is not investment, legal or regulatory advice.

Integrating Climate Risk into Credit Risk Models: From Physical Hazards to PD, LGD and Lending Decisions