Banks' Climate Stress Tests Face a Credibility Test

A climate stress test can produce a precise-looking estimate of future loan losses.

That precision depends on a chain of assumptions. Where a borrower's assets are located, how exposed they are to heat or flooding, what climate policies may change, and how each effect translates into revenue and credit quality all feed into the final number, and a weak link anywhere in that chain quietly becomes a weak link in the output.

One part of that chain has come under direct scrutiny. The Network for Greening the Financial System, or NGFS, says the academic paper underpinning certain physical-damage estimates in its Phase V long-term climate scenarios, Kotz et al. (2024), has been retracted from the journal Nature. The network has identified the specific affected outputs and cautioned users about applying them, while its short-term scenarios and other specified outputs remain unaffected.

For bank boards and executives using climate analysis to inform lending and capital planning, the question is practical. Can they trace a result back to the assumptions that produced it, and would the decision still hold if those assumptions changed?

Bank of England Exercise Found Gaps in Asset-Location and Emissions Data

Climate scenarios are designed to explore possible futures, not predict which one will occur. They can help a bank examine what a sharp change in energy policy or worsening physical hazards might do to its portfolio, but they cannot turn incomplete borrower information into a dependable forecast.

The Bank of England's Climate Biennial Exploratory Scenario exposed that difficulty. Participating banks and insurers faced gaps in information about the locations of corporate assets, which firms need to assess physical hazards, and in standardized data on borrowers' value-chain emissions, a limitation that mirrors the scenario-selection gaps in corporate transition planning. The exercise also found that participating firms' loss estimates for the same corporate customers varied substantially from one institution to the next.

Those findings matter at the individual loan level. A model may identify a borrower's industry and headquarters but miss a flood-exposed production site or a supplier on which that site depends. A bank could then present a portfolio-wide loss figure that appears settled while material exposures remain only partly measured, a version of the facility-level blind spot already reshaping how corporate asset owners price physical climate risk.

NGFS Confirms Which Phase V Outputs the Retraction Actually Touches

NGFS scenarios give institutions a common starting point for examining climate risk, which makes the network's notice consequential. An error in research used by a widely available scenario can travel into many separate bank and insurer analyses at once.

The NGFS has specified which Phase V outputs relied on the retracted paper and says its scenarios do not relieve banks of responsibility for their own risk-management frameworks. A reasonable response, though not one the notice itself mandates, is to identify whether a bank used an affected output, examine how strongly its conclusions depend on it, and rerun material decisions with other defensible assumptions where necessary.

The retraction does not establish that physical climate losses are negligible, and it does not invalidate every climate scenario a bank might run. What it demonstrates is why a single modeled loss number needs its source, its limitations and its sensitivity to alternative methods attached to it, not just the headline figure.

PRA's SS5/25 Gap-Analysis Deadline Has Already Passed

In the United Kingdom (UK), the Prudential Regulation Authority's December 2025 supervisory statement, known as SS5/25, addresses governance, data and climate scenario analysis as parts of managing financial risk rather than as a separate climate-specific exercise. The statement took effect December 3, 2025 and gave firms until June 3, 2026 to complete an internal gap analysis against its expectations and agree a remediation plan. That window has already closed. Supervisors can now request evidence of those internal reviews and action plans as part of routine engagement, which shifts the practical question for a board from whether the work has started to whether it would hold up if a supervisor asked to see it. The authority says its approach is proportionate to a firm's exposures and business model, and these remain UK supervisory expectations rather than rules for banks in every jurisdiction.

For executive teams anywhere, a useful test of their own analysis has several parts. Which borrower and asset data are observed, and which are estimated? What outside models feed the results? Do the findings change materially when a key assumption changes? Most of all, did the exercise reveal something that altered a lending decision, a request for borrower information, or the bank's view of a concentration?

A climate exercise earns its credibility through those answers rather than through the precision of its output. Faking those answers gets harder as banks face growing outside pressure over how seriously they treat climate risk, including from investors watching which institutions have stepped back from voluntary climate commitments altogether. Its real value lies in making uncertainty visible enough to manage, while giving decision-makers a clear account of what the model can and cannot support, now that UK supervisors can ask to see that account directly.