Your Bank's Physical Risk Provisions Are Probably Wrong. And Almost Certainly Too Low

Most banks' physical risk provisions aren't just imprecise. They're structurally too low. Current models capture direct asset damage only, while evidence from MSCI, S&P Global, and our own analysis shows that indirect impacts (revenue loss, supply chain disruption, business interruption) run 6–14× larger. Your provisions could be underestimated by an order of magnitude. And the uncertainty in how those losses are modelled compounds the problem: the error bars on current estimates almost exclusively point upward.

Regulators have stopped hinting. In December 2025, the PRA released SS5/25, replacing SS3/19 outright and mandating that banks assess physical risk at property-level granularity. Postcodes are no longer sufficient. Firms have until June 2026 to present gap-closure plans. The statement treats physical risk as a first-order determinant of financial soundness, not a sustainability reporting exercise.

The ECB has gone further. In November 2025, it issued its first-ever penalty against a bank for failing to assess climate risk materiality, with a second enforcement action reportedly in preparation. The ECB's supervisory report flags climate risk provisions as a “prioritized vulnerability”, and the distance between what banks are provisioning and what regulators expect is growing, not shrinking.

The problem isn't just that provisions may be wrong. It's that they're systematically too low. EY's analysis of UK and European bank disclosures found that only a minority of UK banks had any ECL (Expected Credit Loss) provision for climate risk. Those that did focused primarily on transition risk, with amounts that were marginal relative to overall ECL. Physical risk provisions were particularly limited. The banks that have provisioned for physical risk have almost certainly done so based on direct damage estimates that miss the majority of the financial exposure. Every model that omits indirect impacts produces a number that is too small.

Swiss Re and Sustainalytics frameworks consistently put business interruption at 2–3× direct property damage in major events. Climate X's own Spectra analysis, combining business disruption, supply chain modelling, and revenue transmission channels, shows approximately 6× as a cross-sector average. MSCI's analysis of ~11,000 companies found the ratio reaches 14× when you include full revenue chain impacts. A subsequent MSCI study, “The Bill is Due”, estimates $1.3 trillion in total annual physical-hazard revenue losses, with the gap between recognised risk and actual exposure continuing to widen. Their Macroeconomic Physical Risk Climate VaR model confirms that spillover losses for global equity portfolios can be up to 9× greater than direct damage alone.

The implication for provisions is stark. If your model only captures direct asset damage, you aren't missing a margin of error. You're missing the majority of the loss. First-generation physical risk models were built for hazard scoring, not financial loss quantification. They answer “this asset is in a flood zone” but not “this borrower loses 18 days of revenue per year from business disruption, costing $X million in annual cash flow impact.” Banks can't provision for what they haven't quantified, and the numbers they have quantified are systematically too small.

You can't provision for what you haven't mapped."

Banks initially assumed they could map borrower assets from client surveys and postcode lookups. That turned out to be wildly optimistic. Physical climate hazards propagate through supply chains via two primary channels: direct degradation of production capacity, and interruption of enabling infrastructure including energy, logistics, and transport systems. Bruegel's 2025 analysis found that while extreme weather events are localized, their economic impacts reverberate through global supply chains in ways that are still poorly quantified. McKinsey's supply chain research reaches similar conclusions: supply chains designed for a stable climate become increasingly fragile as hazards evolve.

Corporate borrowers, especially those in industrial, energy, and logistics sectors, operate asset hierarchies with subsidiaries, operational sites, and supply chain nodes spread across dozens of countries. Mapping a single large corporate's full physical exposure means identifying thousands of discrete assets, geolocating each one, and running multi-hazard models against every node. Climate X's Carta product, integrated with FactSet, automates corporate asset hierarchy mapping and eliminates reliance on client surveys, moving from months of manual asset discovery to weeks of automated mapping.

The broader point: we recently published an analysis showing that for oil and gas infrastructure alone, business disruption represented 4.4× the financial exposure of direct physical damage across 380+ facilities. The same dynamics apply to any sector with concentrated, high-value infrastructure, which describes most bank lending books. Every asset your model hasn't mapped is an exposure your provision hasn't captured.

The gap between first-generation and second-generation physical risk models isn't incremental. It's structural.

Note: We've been working with several banks on accelerated paths to Gen 2 capabilities. In some cases, we've delivered the model two years earlier than planned, at lower total cost, while also enabling Adaptation Finance and counterparty risk scoring that most current approaches miss entirely.

There's a common assumption that uncertainty in climate modeling cuts both ways, that losses could be lower as easily as higher. In practice, for physical risk provisions, the uncertainty almost entirely points in one direction: upward.

Much of the current regulatory discussion around climate uncertainty focuses on tipping points: ice sheet collapse, permafrost thaw, coral reef die-off. These are real and warrant attention. At the current warming of ~1.1°C, we already sit within the lower bounds of several tipping point ranges, and at 1.5–2°C, multiple tipping elements become increasingly likely. The NGFS's 2025 tipping point analysis acknowledges these risks are profound but notes their direct financial impacts on shorter timescales remain subject to high uncertainty, with cascading effects between tipping elements still “largely unknown.”

Tipping points matter. But the uncertainty already baked into your model is probably larger, and it's already biasing your provisions downward."

The more decision-relevant problem for banks is what we call the full uncertainty cascade. Climate projection uncertainty, spanning emissions scenarios, model structural differences, and internal variability, is well recognized. What receives far less attention is that hazard characterisation and loss modelling uncertainties compound on top of projection uncertainty and are often substantially larger at the asset and portfolio level.

Consider the chain: a climate projection feeds into a hazard model (flood depth, wind speed, heat intensity), which feeds into a vulnerability function (damage curve), which feeds into a financial translation (revenue loss, repair cost, business interruption days). Each step introduces its own error distribution. Research on uncertainty in climate risk assessments shows that varying hazard, exposure, and vulnerability modeling choices produce wide confidence intervals well beyond what climate model spread alone would suggest. In our own modeling experience across thousands of asset-level assessments, these downstream uncertainties frequently dominate the total error budget. This is especially true at regional and asset-level resolution, where climate model agreement may be relatively high, but hazard characterization and financial translation introduce substantial additional variance.

Here's the critical point: the uncertainty in the indirect transmission pathways, the same channels that produce the 6–14× multiplier, is itself a major contributor to tail risk dispersion. It compounds rather than merely adds to climate projection uncertainty. A framework that builds tail risk narratives primarily around tipping point scenarios, while ignoring the larger and more proximate modeling uncertainties, creates a false precision that obscures the broader risk picture.

For banks, the practical consequence is clear. Tail risk is already poorly captured in capital allocation, not primarily because of tipping points, but because the full uncertainty cascade from hazard-to-financial-impact is inadequately propagated through risk models. The provisions that emerge from these models aren't just point estimates that happen to be uncertain. They're point estimates drawn from a distribution skewed toward underestimation because each layer of the cascade tends to omit or underweight the tail.

We help banks move from hazard scores to defensible financial loss estimates, with the infrastructure mapping and indirect impact modeling that regulators are now expecting.

• MSCI, “Hidden in Plain Sight: Physical Risk in Asset Owners' Portfolios” (2025)
• MSCI, “The Bill is Due: Physical-Hazard Revenue Losses Mount, Plans Lag” (September 2025)
• MSCI, “The Macroeconomic Side of Physical Risk” (2025)
• S&P Global Sustainable1, “Climate physical risks have a $1.2 trillion annual price tag by the 2050s” (2025)
• PRA Supervisory Statement SS5/25 (December 2025)
• Green Central Banking, “ECB's green supervision grows teeth” (February 2026)
• Swiss Re Institute, sigma No. 1/2025: Natural catastrophes
• Bruegel, “Climate Risks to Global Supply Chains,” Working Paper 20/2025
• McKinsey & Company, “Could Climate Become the Weak Link in Your Supply Chain?”
  Armstrong McKay et al. (2022),
• “Exceeding 1.5°C global warming could trigger multiple climate tipping points,” Science, 377(6611)
• NGFS, “Tipping Points in the Earth System” (2025)
• Dawkins et al. (2023), “Quantifying uncertainty and sensitivity in climate risk assessments,” Climate Risk Management, Elsevier
• EY, “Impact of climate risk on banks and ECL” (2024)
• Climate X Spectra analysis (internal, 2025–2026)