Back to Blog
- Thirteen physical risk vendors assessed the same properties under the same scenario, and still disagreed.
- The disagreement comes from a chain of choices: warming assumptions, where the asset is placed, what each metric measures, and how uncertainty is reported.
- We set our own methodology against each of those choices.
In October 2025, the GARP Risk Institute published a benchmarking study of 13 physical risk data vendors on behalf of the Climate Financial Risk Forum. As far as its authors are aware, it is the first study of its kind: a systematic comparison of how vendors assess physical climate risk at the level of an individual asset.
A regulator-adjacent body benchmarking vendors is itself a signal. Physical risk data has become a standard input to underwriting, lending, and valuation, and the market has grown quickly. That makes it hard for risk teams to judge which data source fits their use case.
As a relatively new entrant to the market, we were not among the 13 vendors surveyed. In this article, we examine GARP’s methodology and assessment criteria, and consider how our own approach compares against the framework used in the study.
The report is less a verdict on any single provider than a map of the choices that separate one physical risk assessment from another. This piece walks through how the study was built, why those choices matter, and where our approach sits against each of them.
A three-pronged approach
There were three components to the study’s benchmarking exercise. Select one to read more.
The benchmarking exercise
No two vendors are the same
Given the same properties and the same prescribed scenario, the vendors still produced a significant spread of results at the asset level. Some of the gaps came from the models; others came from steps that happen before any modelling begins. In the most extreme case, one vendor placed a property 1,507 kilometres from the median of the others: given the name of a well-known store in Boston, it located a similarly named road in Atlanta.
Where the spread comes fromSame assets · same prescribed scenario (RCP 8.5)
1
Illustrative: the widening shows how each factor adds to the divergence between vendors; it is not drawn to scale.
The study’s most useful contribution is its breakdown of why this dispersion exists. Here are four of the reasons the vendors differed, and where our approach sits in comparison:
Metrics, outputs, and comparability
What the study found
The study found that standardisation between hazards existed only in pockets. Flood depth (in metres) was fairly consistent and cyclone risk mostly converged on wind speed, but even that agreement was superficial: it could mean sustained 1-minute winds, sustained 10-minute winds, or 3-second gusts, and one vendor measured something different again, the probability of hitting a given category on the Saffir-Simpson scale.
Beyond these two hazards, there was little shared language. Heat measures ranged from wet bulb readings to cooling degree days to simple threshold-day counts, and wildfire risk could be a fire weather index, a burn probability, or a raw count of past fires. Drought was the extreme case: definitions varied so much that the study excluded it from comparison entirely. The result is that even when two vendors assess the same asset for the same hazard, their outputs may not be measuring the same thing.
Our approach
We use two consistent outputs across every hazard: the Climate Exposure Rating and the Climate Risk Rating. Both ratings are expressed on an A to G scale combining physical and transition exposure scores at two horizons: 2035 and 2050.
Underlying each rating, we also calculate the same type of detailed, hazard-specific metrics other vendors report, such as flood depth or heat threshold days, but rather than leaving clients to reconcile these individually, we blend them into a single rating. Every rating is measured against the same benchmark, so a rating means the same thing regardless of the hazard, sector, or geography. Each rating is then broken down into financial metrics, such as the percentage impact on asset value or enterprise value, to give a quantitative view.
Climate scenarios
What the study found
The study called for the participants to adopt just one climate scenario for this exercise: RCP 8.5. In theory, this should have ensured consistency, yet the vendors still assumed warming of between 3.2°C and 5.7°C by 2100. Part of the gap came from different baseline periods, and one vendor did not run RCP 8.5 at all. The same scenario name, in other words, did not mean the same climate.
Our approach
We consult all eight NGFS scenarios, from Net Zero 2050 to Climate Breakdown, each paired with a consistent SSP climate pathway for physical risk. Each scenario carries an assigned probability, so our ratings reflect a probability-weighted expectation, not just a range of potential outcomes.
Asset location
What the study found
One vendor placed a property 1,507 kilometres from the median of the other vendors’ coordinates. More subtly, even with identical geocoding, vendors differed in the precise location at which they chose to assess the property’s exposure to a hazard; the centre of the area might be used, or a radius, or a point that could be the most vulnerable.
Our approach
We use geolocation and detailed asset boundaries to compute an asset’s precise exposure to a range of physical risks. In this way, we can ensure that our ratings reflect the risk profile of the specific asset, not just the hazards associated with a general area.
Transparency, uncertainty, and data quality
What the study found
Four of the thirteen vendors did not provide any measure of uncertainty to their clients. Of the nine that did, the methods varied so significantly, ranging from standard deviations of the modelled outputs to accuracy scores to probabilistic distributions, that comparing confidence levels across vendors would be an impossible task.
Our approach
Every one of our ratings carries data quality scores aligned with the Partnership for Carbon Accounting Financials (PCAF) reference scale. These scores provide a clear measure of the reliability and accuracy of all the input data points used to calculate our climate ratings.
Data quality is only part of the picture. The three NGFS models behind our scenarios can disagree, so the Climate Risk Rating is reported for each model as well as for their average. Clients can see how much the choice of model moves the answer, rather than receiving a single number.
The principles behind our ratings
Our approach incorporates eight probability-weighted NGFS scenarios, verified and precise asset boundaries, peer-reviewed damage functions, and PCAF-aligned data quality scores. As physical risk data becomes embedded in underwriting, lending, and valuation decisions, this degree of transparency and rigour should be the baseline the market demands, not a differentiator.
Against the sources of divergence that the GARP study identified, three principles anchor every rating that we produce:
Precision
We use exact geolocation and detailed asset boundaries, so a rating reflects the specific site being assessed, not just an approximation of its surroundings.
Comparability
Every asset is scored on the same A to G scale, against the same benchmark, using the same two outputs (the Climate Exposure Rating and the Climate Risk Rating). You can be confident that a rating means the same thing whatever the hazard, sector, or geography.
Peer-reviewed and auditable
Our damage functions and scenario probabilities are founded on the EDHEC Climate Institute’s years of peer-reviewed research, and every rating carries a PCAF-aligned data quality score, so clients can see exactly how reliable the underlying inputs are.
The key takeaways
The GARP benchmark does not crown a winner among the vendors it surveyed. Rather, its value lies in showing that a physical risk rating is not a single objective fact but the output of a chain of choices, each of which can move the answer.
We assessed our methodology against the GARP framework. Here is how it compares on three of the study’s headline findings.
9 of 13 vendors
could offer clients any measure of uncertainty at all.
Scientific Climate Ratings
Every input carries a PCAF-aligned data quality score from 1 to 5, and the Climate Risk Rating is also reported for each of the three NGFS models, so the spread between them is visible.
8 of 13 vendors
described identifying a property by its building footprint.
Scientific Climate Ratings
We start from a detailed asset boundary as standard.
4–5 of 13 vendors
modelled the individual NGFS pathways, while all thirteen ran the three headline SSP-RCP scenarios (SSP5-RCP8.5, SSP2-RCP4.5, SSP1-RCP2.6).
Scientific Climate Ratings
We run all eight NGFS scenarios, each paired with an SSP climate pathway, blended into a probability-weighted expected outcome.
On each of the dispersion factors the report identifies, our approach sits at the more rigorous end of the range.
For risk teams choosing a data provider, the practical lesson is to put the same four questions to any vendor, us included: which climate and warming level sits behind the numbers, exactly where the asset has been placed, what each metric actually measures, and how confident the result is.
Source: GARP Risk Institute for the Climate Financial Risk Forum, A risk professional’s guide to physical risk assessments: A GARP benchmarking study of 13 vendors (October 2025).
Want to see how our ratings answer those questions for your assets?