Recent emerging-risk survey work across the actuarial profession has reclassified climate from an "emerging" risk to an embedded one, and supervisory climate reviews report sharply increased actuarial involvement. This article is about what that reclassification means in practice for model validation.
Climate risk model validation standards have lagged a change of status that the profession itself has now formally recognised: climate is no longer an emerging risk. Practitioner surveys have reclassified it as embedded - a permanent feature of the risk landscape rather than a novel arrival - and supervisory thematic reviews find most actuarial functions now involved in climate work. The label matters, because "emerging" was doing quiet regulatory work: emerging risks get scenario exercises, qualitative overlays and exploratory workstreams. Embedded risks get validation standards.
What "embedded" changes
An emerging risk is allowed to live in a side process: a dedicated climate team, an annual scenario exercise, a board paper. An embedded risk lives inside the models - pricing, reserving, capital, planning - and therefore inside the validation of those models. The practical consequence is a shift in the null hypothesis. For an emerging risk, the default is exclusion and the burden is on inclusion. For an embedded risk, the default reverses: a material model that ignores climate now needs a documented justification for the omission, exactly as it would for ignoring inflation.
This is not an argument that every model needs a climate module. Materiality still governs. It is an argument that the materiality assessment itself must exist, be evidenced, and be refreshed - model by model, not enterprise-wide by assertion.
What validators should now ask of every material model
Is the climate materiality assessment current and specific? Not "climate considered" but: which perils, which channels (physical, transition, liability), which horizon, and why the conclusion follows for this model's use.
Are historical calibrations still representative? Climate non-stationarity undermines the foundational assumption that the calibration window predicts the projection window. Validators should expect trend analysis on climate-sensitive experience - weather claims frequency and severity, subsidence, business interruption - and explicit treatment where trends are detected, rather than silent averaging over a shifting record.
Where do external climate inputs enter, and how are they governed? Vendor hazard scores, scenario pathways, transition assumptions. Each is a third-party model component and attracts the standard obligations: understand it, benchmark it, bound it. I have written separately on vendor divergence in physical climate risk - the inter-vendor spread is the natural uncertainty estimate validators should demand to see.
Is the dependency treatment defensible in the tail? Climate risks correlate with each other and with market risks precisely in adverse states. Gaussian aggregation assumptions deserve specific challenge here, for the reasons set out in my piece on transition risk tail dependency.
Are the projected components separated from the observed ones? Validation evidence differs in kind: observed components can be backtested; projected components can only be benchmarked, bounded and sensitivity-tested. Reports that blend the two overstate the evidential status of the projection.
Does the documentation date-stamp its climate assumptions? Scenario vintages, vendor model versions, policy assumptions. Climate inputs age quickly, and undated assumptions cannot be re-validated on trigger.
Standards are converging on the same expectation
The direction is consistent across frameworks. UK supervisory expectations on climate risk management have for some years required embedding in governance and risk management, and thematic feedback has pushed firms from qualitative narrative toward quantified, decision-useful analysis. Actuarial standards on both sides of the Atlantic - technical actuarial standards in the UK, standards of practice in the US - require consideration of relevant risks and disclosure of material assumptions, which straightforwardly captures climate once it is conceded to be embedded rather than exotic. And model risk frameworks' proportionality principles cut the other way from how firms sometimes read them: proportionality tunes the depth of validation, not whether an embedded material risk is in scope at all.
The consequence for validation functions is capability, not just checklists: enough climate-science literacy to challenge a downscaling choice, enough familiarity with scenario frameworks to spot a stale pathway, and enough statistical footing to interrogate non-stationarity claims. That is a resourcing decision boards should make consciously.
A pragmatic sequence for the next validation cycle
First, sweep the model inventory with a one-page climate materiality triage per model - peril, channel, horizon, conclusion. Second, for models flagged material, add the six questions above to the validation scope. Third, standardise the treatment of vendor climate inputs once, centrally, rather than model by model. Fourth, report the residual: the list of models where climate is material but untreated, with owners and dates. That list - not a glossy climate report - is what an embedded risk's governance actually looks like.
The reclassification of climate from emerging to embedded is the profession marking its own homework honestly. Validation standards that still treat climate as a special topic are now behind the profession's own stated view - and behind is not where a validation function can afford to stand.
Key Takeaways
- Professional surveys and supervisory reviews now treat climate as an embedded risk, ending its "emerging" status - and embedded risks belong inside model validation, not beside it.
- The null hypothesis flips: material models omitting climate need documented justification for the omission, model by model.
- Core validation questions: current materiality assessment, non-stationarity of calibrations, governance of vendor climate inputs, tail dependency treatment, separation of observed from projected components, and date-stamped assumptions.
- UK and US standards frameworks already capture embedded climate risk through existing requirements; proportionality tunes depth, not scope.
- The deliverable of embedded-risk governance is the residual list - material-but-untreated models with owners and dates - not a standalone climate report.
Frequently Asked Questions
Is climate risk still considered an emerging risk? Increasingly not. Recent professional emerging-risk survey work has reclassified climate as an embedded risk - a permanent feature of the landscape - and supervisory reviews report widespread actuarial involvement in climate work. The practical consequence is that climate moves from exploratory exercises into routine model validation scope.
What should model validation cover for climate risk? A current, model-specific materiality assessment; testing of historical calibrations for climate-driven non-stationarity; governance of third-party climate inputs including vendor divergence; challenge of tail dependency assumptions; separation of backtestable observed components from projected ones; and date-stamped climate assumptions with re-validation triggers.
Do actuarial standards require climate risk to be considered? Existing standards already capture it: UK technical actuarial standards and US actuarial standards of practice require consideration of relevant material risks and disclosure of material assumptions. Once climate is recognised as embedded and material, those general requirements apply without needing a climate-specific standard.