1 Background and context

This Climate Financial Risk Forum (CFRF) study, undertaken by the GARP Risk Institute, benchmarks asset-level climate physical risk assessments from 13 third-party vendors. It responds to growing regulatory expectations that financial institutions embed climate risk into day-to-day risk management. The study focuses on how vendor methodologies differ and the scale of variation in estimated hazards and damages for individual assets, rather than identifying a “best” provider.

2 Physical risk assessments: Introduction and overview

The report defines physical risk through the interaction of hazard, exposure and vulnerability. Vendors use diverse modelling stacks, combining global and regional climate models, catastrophe models, downscaling techniques, vulnerability curves and financial loss modules. Even under a single high-emissions scenario (RCP 8.5), modelling choices, assumptions and metrics vary substantially.
Nine of the 13 vendors provide some measure of uncertainty, most commonly standard deviations or percentile ranges, but approaches are inconsistent. The report highlights limited standardisation of hazard metrics beyond flooding and wind, complicating comparisons. Financial institutions are encouraged to engage with vendors to understand modelling choices, uncertainty treatment and metric definitions.

3 Asset location and characteristics

Asset geolocation quality is shown to be a material driver of risk estimate dispersion. In a test of 20 properties, vendor-supplied coordinates diverged by up to 1,507 km in extreme cases, though higher-quality address data generally reduced dispersion. Even with full address information, meaningful variation remained. Vendors also differ in how they define the assessment point for large sites, using single points, buffers, polygons or site-wide meshes. Property characteristics such as construction type, age, number of floors, basements and occupancy are incorporated unevenly across providers. The report stresses that improving asset data quality and understanding vendor geolocation methods can materially improve assessment reliability.

4 Physical risk quantification study

Vendors assessed hazards for a common portfolio of 100 properties across the U.K., mainland Europe, the U.S. and Asia. Hazards included combined and coastal flooding, cyclones, windstorms, heat and wildfire. Results were provided for multiple time horizons, return periods and damage metrics.
Even under RCP 8.5, assumed global warming by 2100 ranged from 3.2°C to 5.7°C, reflecting different baselines and scenario interpretations. This alone introduced variation before hazard modelling. The study standardised metrics where possible (for example, flood depth in metres, wind speed in km/h), but wildfire and heat remained difficult to compare due to divergent definitions.

5 Physical risk benchmarking results

Across all hazards, the study finds large dispersion in both hazard severity and damage ratios. For flooding at a one-in-200-year return period in 2030, most properties were affected, but flood depth and damage estimates varied widely between vendors, even when defences were assumed. Damage ratio correlations between vendors ranged from 0.2 to 0.9, indicating materially different views on vulnerability. Coastal flooding affected only 22 of 100 properties, yet dispersion remained significant. Cyclone and windstorm results showed strong non-linear relationships between wind speed and damage, with substantial variation in thresholds and loss severity. Heat metrics exhibited extreme dispersion for some locations, with zero exposure reported by some vendors where others estimated over 190 days above 35°C. Wildfire results could not be meaningfully benchmarked due to incompatible metrics. Dispersion generally increased for longer time horizons and more severe return periods when measured by standard deviation, reinforcing uncertainty in forward-looking risk estimates.

6 Checklist for financial institutions

The report integrates lessons throughout, culminating in a checklist rather than prescriptive recommendations. Financial institutions are encouraged to clarify relevant perils, scenarios, time horizons and metrics; assess asset data quality; scrutinise vendor modelling assumptions and uncertainty measures; and ensure outputs align with intended use cases, regulatory needs and internal model risk management frameworks.

7 Conclusions

The study concludes that wide variation in vendor physical risk assessments is inherent given methodological complexity and deep uncertainty. Transparency, robust due diligence and stronger internal capability are essential for interpreting results and using them effectively in financial decision-making.