Commentary on a research call issued by the Casualty Actuarial Society in January 2026, seeking Canada-focused work on potential bias in geographical ratemaking influenced by climate risks. Details are available from the CAS.
In January 2026 the CAS issued a call for Canada-focused research on potential bias in geographic ratemaking influenced by climate risk. It is a short notice about a research topic. It also identifies, precisely, what I expect to become the defining fairness problem in property insurance over the next decade.
The problem is this: as climate risk becomes better measured, territorial rating becomes simultaneously more accurate and more socially contested. Those two things are not in tension by accident. They are in tension for a structural reason that is worth working through carefully, because most commentary on it goes wrong in the first paragraph.
Why Territory Is a Different Kind of Rating Variable
Every rating variable is a proxy for something. Vehicle type proxies for repair cost and driver behaviour. Building age proxies for construction standard and maintenance. Credit-based scores, where permitted, proxy for something contested enough that jurisdictions differ on whether to allow them at all.
Territory is unusual in that it proxies for almost everything at once.
A postcode carries flood exposure, wildfire exposure, wind exposure, building stock characteristics, construction era, local building code enforcement, emergency service response times, crime rates, repair market pricing, litigation propensity — and, unavoidably, the income, ethnic composition and housing tenure of the people who live there.
That last set is not a defect in the data. It is a fact about how people are distributed across space, and it exists in every country with any history of residential sorting. It means territorial rating is always correlated with protected and quasi-protected characteristics, whether or not the actuary intends it, and whether or not those characteristics carry any causal relationship to loss.
This is the definition of proxy discrimination, and territory has always been the hardest case.
What Climate Change Does to This
Climate change intensifies the problem through three distinct mechanisms. They are frequently conflated and they have different implications.
Mechanism 1 — Resolution. Climate analytics are pushing territorial rating from broad zones to individual addresses. Higher resolution genuinely improves risk assessment. It also destroys the pooling that coarse territories provided. When your territory contained 50,000 properties, the high-risk and low-risk properties within it shared a rate. At address-level resolution, they do not. Nobody decided to remove that cross-subsidy; it was dissolved by better measurement.
Mechanism 2 — Correlation with disadvantage. Physical climate exposure is not randomly distributed with respect to income. Floodplain land is systematically cheaper. Wildland-urban interface development includes both affluent exurban housing and lower-cost fringe development. Urban heat exposure concentrates in areas with less tree canopy and older building stock — which correlate strongly with historic disinvestment. The correlation direction varies by peril and country, but the correlation is rarely zero.
Mechanism 3 — Forward-looking uncertainty. This is the mechanism that makes climate territorial rating legally distinctive, and it is the one most often missed. A conventional rating factor is calibrated on observed experience. A climate-adjusted territorial factor is partly calibrated on projections — model outputs about future hazard frequency that cannot be back-tested and that, as I have written elsewhere, diverge materially between vendors for the same address.
Charging a customer more today on the basis of an unverifiable projection about 2050 is a materially different act, legally and ethically, from charging them more on the basis of observed loss experience. Regulators have not yet worked out what to do about this. They will have to.
The Fairness Definitions Are Genuinely Incompatible
Here is the part practitioners must understand, because it explains why these debates never resolve.
There are several defensible definitions of fair pricing, and they cannot all be satisfied simultaneously. This is a mathematical result, not a policy disagreement.
Actuarial fairness — premium proportional to expected loss. On this definition, charging a high-flood-risk property more is not merely permitted, it is required, and any deviation is a cross-subsidy from low-risk to high-risk policyholders.
Demographic parity — outcomes not differing systematically across protected groups. On this definition, a rating structure producing much higher premiums in areas of concentrated ethnic minority residence is problematic regardless of the loss data supporting it.
Equality of opportunity / calibration by group — the model equally accurate for all groups. Achievable simultaneously with actuarial fairness only under restrictive conditions that rarely hold.
The impossibility results in the fairness literature establish that where base rates genuinely differ across groups, you cannot have calibration and equalised error rates at once. A model cannot be simultaneously actuarially fair and demographically neutral when the underlying risk genuinely differs by geography and geography correlates with demography.
This means the choice is not technical. It is a policy choice about which fairness definition governs, and it belongs to regulators and legislatures rather than to actuaries. What actuaries owe is clarity about the trade-off — quantifying what each definition costs and who bears it — rather than a claim that the tension can be engineered away.
What Practitioners Should Actually Do
Six things, in rough order of urgency.
1. Measure the disparate impact of your existing territorial structure. Not to change it necessarily, but to know. Most insurers cannot currently state the demographic gradient of their territorial relativities. Being asked this by a regulator and not knowing is a much worse position than knowing and having a rationale.
2. Separate observed from projected components. Split territorial relativities into the part supported by observed loss experience and the part driven by forward-looking climate model output. These two components have different evidential status and should be defended differently. Presenting them as a single number forfeits your strongest argument on the observed part.
3. Quantify the vendor uncertainty in the projected component. If your climate-adjusted territorial factor rests on a single vendor's hazard score, and that score diverges materially from a competitor's for the same address, you are charging a customer a differential you cannot substantiate. Run a second provider on the material segments.
4. Model the availability consequence, not just the rate. The social harm from climate territorial rating usually shows up as unavailability rather than expensive availability. Track where your rate structure is producing effective withdrawal, and put that in front of the board separately from the pricing analysis. It is the number regulators and legislators will eventually ask about.
5. Distinguish mitigable from unmitigable exposure. A flood factor a homeowner can reduce through defensible action is different in kind from one they cannot. Pricing structures that reward mitigation convert a static penalty into a behavioural incentive, and are far easier to defend on fairness grounds. This is also where the insurance industry has the most genuine social value to add.
6. Document the fairness definition you are using. Explicitly, in the rate filing rationale. "Rates are calibrated to actuarial fairness as defined by expected loss proportionality" is a defensible position stated plainly. An undocumented implicit choice is not.
Why Canada Is a Sensible Testbed
The CAS call is Canada-focused, and that is a well-chosen jurisdiction. Canada has high and rising flood and wildfire exposure, a concentrated insurance market, significant regional variation in climate peril, and a regulatory environment that has engaged seriously with both climate adaptation and equity questions. It also has population geography where the correlation between climate exposure and socioeconomic characteristics differs meaningfully from US patterns — which makes it useful for separating what is a general structural feature from what is a US-specific artefact of historic policy.
Findings there will travel.
The Honest Position
Actuaries did not create the correlation between climate exposure and disadvantage. They are, increasingly, the profession that measures it — which means they will be the profession asked to answer for it.
The defensible position is not that risk-based pricing is beyond criticism because the mathematics is correct. It is that the profession can state precisely what the trade-offs are, quantify who bears them, and distinguish clearly between what the data supports and what a projection assumes. That is a contribution only actuaries can make, and it is more valuable than winning the argument.
Key Takeaways
- The CAS is funding Canada-focused research into potential bias in geographic ratemaking influenced by climate risk — a well-targeted call at an emerging structural problem.
- Territory proxies for nearly everything, including protected characteristics, making it the hardest case in proxy discrimination.
- Climate change intensifies the issue through finer resolution dissolving cross-subsidy, correlation between exposure and disadvantage, and reliance on unverifiable forward projections.
- Actuarial fairness and demographic parity are mathematically incompatible where base rates genuinely differ — the choice between them is a policy question, not a technical one.
- Separate observed from projected components of territorial relativities; they have different evidential status and should be defended differently.
Frequently Asked Questions
Is territorial rating discriminatory? Territorial rating correlates with protected characteristics in virtually every market, because people are not randomly distributed across space. Whether that constitutes unlawful discrimination depends on jurisdiction and on which definition of fairness the regulatory framework adopts. It is a live and unsettled question that climate-driven refinement of territory is making more pressing.
Can insurance pricing be both actuarially fair and demographically neutral? Generally not, where underlying risk genuinely differs across groups. Formal impossibility results establish that calibration and equalised error rates cannot both hold when base rates differ. This means the choice between fairness definitions is a policy decision rather than a technical problem that better modelling can resolve.
How should insurers handle climate projections in territorial rating? By separating the component of the relativity supported by observed loss experience from the component driven by forward-looking model projections, quantifying the uncertainty in the projected component including divergence between data vendors, and documenting both explicitly in the rate filing rationale.
Jonas Osman Abdelghafour is the founder of Quantica Risk, which builds pricing, climate and fairness-assessment models for insurers. This article is commentary on a publicly available CAS research call.