Apr 3 • JULIUS ATULINDE
Translating IPCC Hazard Data into Credit Risk | PD & LGD Modelling
Physical climate risk is increasingly recognized as a material driver of financial risk, yet most institutions struggle to translate climate hazard data into actionable credit metrics. While IPCC AR6 and NGFS scenarios provide probabilistic hazard projections, they do not directly inform probability of default (PD) or loss given default (LGD)—the core parameters underpinning credit underwriting, pricing, and capital adequacy.
This article presents a structured framework for bridging that gap, outlining how hazard probability curves can be translated into credit risk outcomes using damage functions, exposure modelling, and scenario-based financial analysis.
The Translation Problem Not Fully Resolved
When regulators began pressing financial institutions to integrate climate risk into capital planning, the immediate response was largely procedural. Risk teams obtained climate scenario data — mostly from the NGFS pathways, occasionally IPCC AR6 regional hazard outputs — and mapped them to exposure categories. The result, in most cases, was a qualitative heat map: high, medium, and low exposure designations across asset classes. What this process does not produce is an adjustment to the probability of default (PD) or the loss given default (LGD) of specific counterparties.
Credit pricing depends on PD and LGD, the same parameters capital adequacy under Basel III uses. If physical climate risk cannot be translated into PD/LGD adjustments, it remains a disclosure exercise rather than a risk management discipline. Climate science provides useful hazard information — in increasing resolution — but there still a methodological gap to translate hazard probability into accurate credit risk assessments financial risk frameworks demand.
What IPCC AR6 Actually Provides and What It Does Not
The IPCC Sixth Assessment Report (AR6) provides probabilistic hazard outputs at regional scale such as likelihood distributions for sea-level rise, shifts in 100-year flood return periods, changes in heat stress event frequency, drought intensity, and storm surge exposure across warming scenarios from 1.5°C to 4°C. These outputs are expressed as exceedance probability curves — the probability that a hazard intensity exceeds a given threshold within a specified period.
What AR6 does not provide is property-level exposure data, economic damage functions that are calibrated to specific asset types, or any direct linkage to financial distress.
The damage function problem — translating a flood depth at a given location into a specific reduction in asset value or revenue — requires additional modelling layers that physical climate science does not supply. Representative Concentration Pathways (RCPs) and the newer Shared Socioeconomic Pathways (SSPs) describe emissions trajectories and their physical consequences, not economic outcomes. The financial translation layer requires separate assumptions about asset sensitivity to physical hazard, business interruption, insurance coverage, and the pace of adaptation.
The damage function problem — translating a flood depth at a given location into a specific reduction in asset value or revenue — requires additional modelling layers that physical climate science does not supply. Representative Concentration Pathways (RCPs) and the newer Shared Socioeconomic Pathways (SSPs) describe emissions trajectories and their physical consequences, not economic outcomes. The financial translation layer requires separate assumptions about asset sensitivity to physical hazard, business interruption, insurance coverage, and the pace of adaptation.
The Damage Function Problem in Credit Assessment
Consider a bank underwriting a ten-year term loan to a coastal commercial property owner in a low-elevation urban area. The AR6 outputs can tell the underwriter that under SSP2-4.5, the probability of a flood event exceeding the current 100-year return level at this location increases from roughly 1% per year to 8–12% per year by 2050. That shift in exceedance probability is meaningful but insufficient.
To translate this into a PD adjustment, the underwriter needs:
To translate this into a PD adjustment, the underwriter needs:
-
An estimate of the flood damage to the property at different inundation depths, typically expressed as a depth-damage function derived from engineering surveys or insurance actuarial data
-
An estimate of the borrower's revenue sensitivity to flood disruption, which varies by business type, asset redundancy, and supply chain structure
-
An assessment of insurance adequacy — whether flood coverage is maintained, at what deductible levels, and how coverage gaps affect recovery capacity; and
-
A view on the trajectory of adaptation — whether the locality is investing in flood barriers, and whether regulatory requirements for construction standards will shift.
These inputs come from disciplines — actuarial science, engineering, public finance — that are rarely integrated into credit origination workflows.
Common Approaches and Their Limitations
Three methodological approaches have gained traction among more advanced financial institutions.
The first is adaptation of actuarial depth-damage functions developed by the reinsurance industry. AIR Worldwide, RMS, and the Flood Hazard Assessment Tool (FHAT) maintained by the European Environment Agency all provide damage function libraries that can be anchored to specific hazard scenarios. The limitation is that these functions are calibrated to historical events and do not capture forward-looking shifts in hazard intensity without explicit adjustment.
The first is adaptation of actuarial depth-damage functions developed by the reinsurance industry. AIR Worldwide, RMS, and the Flood Hazard Assessment Tool (FHAT) maintained by the European Environment Agency all provide damage function libraries that can be anchored to specific hazard scenarios. The limitation is that these functions are calibrated to historical events and do not capture forward-looking shifts in hazard intensity without explicit adjustment.
The second approach uses satellite and geospatial data to assign physical risk scores at the asset level, then maps those scores to historical default and recovery rates for assets with similar exposure characteristics. Moody's Climate Solutions, Four Twenty Seven, and Climate X have developed proprietary asset-level scoring systems along these lines. The methodological challenge is that historical default data for high climate-risk assets is sparse — most severe physical risk events are, by definition, tail events with limited historical observations.
The third approach, more common in sovereign and project finance, uses integrated assessment models (IAMs) that link economic damage functions to physical hazard scenarios. The DICE and PAGE models, for instance, produce GDP damage estimates under different warming pathways. The limitation here is model uncertainty whereby damage function parameters in IAMs have extremely wide confidence intervals, and the functional form assumed — typically a polynomial relationship between temperature and GDP loss — is contested in the economic literature.
What Supervisors Expect or Require Now
There is a supervisory expectation toward quantification. The Basel Committee on Banking Supervision's Principles for Effective Management and Supervision of Climate-Related Financial Risks (2022) call for climate risk integration into credit risk frameworks, stress testing, and capital planning — but stop short of specifying the methodological standard. This leaves financial institutions with the obligation to quantify but without a standardized methodology, creating significant variation in how physical risk translates (or fails to translate) into credit metrics across the system.
The European Central Bank's thematic review on climate risk (2022–2023) found that the majority of supervised institutions had not translated climate risk exposures into quantified impacts on credit metrics. The Bank of England's Climate Biennial Exploratory Scenario (CBES) required participating banks to estimate losses under physical and transition risk scenarios, but explicitly acknowledged that methodologies were exploratory and that comparability across institutions was limited.
The European Central Bank's thematic review on climate risk (2022–2023) found that the majority of supervised institutions had not translated climate risk exposures into quantified impacts on credit metrics. The Bank of England's Climate Biennial Exploratory Scenario (CBES) required participating banks to estimate losses under physical and transition risk scenarios, but explicitly acknowledged that methodologies were exploratory and that comparability across institutions was limited.
What Needs to Change in Credit Underwriting
For physical climate risk to become genuinely integrated into credit underwriting, three operational changes are required.
First, data sourcing needs to be restructured. Climate hazard data — at sufficient geographic resolution to be useful at the property or facility level — must be incorporated into origination workflows, not added as a post-hoc overlay. This requires vendor relationships, data standards, and staff trained to interpret probabilistic hazard outputs rather than single-point estimates.
Third, loan covenants and pricing should reflect differentiated climate exposure. A ten-year loan to a coastal property with high flood exposure and no insurance should carry a different spread than an equivalent loan to an inland property with comparable financial characteristics. This pricing differentiation will not happen until PD/LGD adjustments are credible enough to be defensible in credit committee review — which requires the first two changes to be in place.
The methodological bridge is buildable. The components exist in adjacent disciplines. What is missing, at most institutions, is the organisational decision to integrate them.
First, data sourcing needs to be restructured. Climate hazard data — at sufficient geographic resolution to be useful at the property or facility level — must be incorporated into origination workflows, not added as a post-hoc overlay. This requires vendor relationships, data standards, and staff trained to interpret probabilistic hazard outputs rather than single-point estimates.
Second, credit models need to incorporate climate-sensitive variables explicitly. The standard PD model inputs — leverage, coverage ratios, revenue volatility — need to be supplemented with measures of climate exposure and adaptive capacity. This is methodologically tractable; the actuarial literature on catastrophe modelling provides the frameworks. The implementation challenge is that most credit model validation frameworks are not designed to accommodate forward-looking scenario variables with multi-decade time horizons.
Third, loan covenants and pricing should reflect differentiated climate exposure. A ten-year loan to a coastal property with high flood exposure and no insurance should carry a different spread than an equivalent loan to an inland property with comparable financial characteristics. This pricing differentiation will not happen until PD/LGD adjustments are credible enough to be defensible in credit committee review — which requires the first two changes to be in place.
The methodological bridge is buildable. The components exist in adjacent disciplines. What is missing, at most institutions, is the organisational decision to integrate them.
Empty space, drag to resize
***************************End of Article******************************
Copyright © 2026