Guides
Catastrophe models: how insurers price hurricanes, quakes and fires
How catastrophe models turn simulated storms and quakes into loss estimates, why your building data moves your premium, and where models fall short.
A catastrophe model is software that simulates tens of thousands of hurricanes, earthquakes, wildfires or storms, estimates the damage each would do to a specific list of buildings, and applies the insurance terms to produce a loss figure. Insurers use the output to set the catastrophe part of the rate, to decide how much exposure they can write in a zone, and to buy reinsurance. For a business owner the practical point is that the model sees your building as a row of data, so the accuracy of that row can move your premium and your capacity.
The four parts of a model
Every commercial model, whatever the vendor, is built from four linked modules.
| Module | What it does | What can go wrong |
|---|---|---|
| Hazard (event set) | Generates thousands of synthetic events with a frequency, intensity and path, calibrated to history and science | Short historical records; climate change shifting the odds |
| Exposure | Takes the insurer's list of insured locations, values and construction details | Missing, stale or default data (an unknown roof is usually modeled as a blend of roof types, often skewed to poorer performance) |
| Vulnerability | Converts hazard intensity at each location into a damage ratio for that construction type | Few claims data for some building types and perils |
| Financial | Applies deductibles, limits, sublimits and reinsurance to give gross and net loss | Policy wording that is more complex than the model assumes |
The vendors are a short list. Florida's accepted hurricane models in 2025 came from Karen Clark & Company, Cotality, Moody's RMS and Verisk, among others, and California's Department of Insurance has reviewed wildfire models from Verisk, Karen Clark and Moody's (below). Insurers also run in-house views and blend them with vendor output. When a vendor releases a new version, modeled losses for the same portfolio can move, which is one reason a carrier's appetite changes even in a year without a storm.
What the output means
Models do not forecast a particular storm. They describe a distribution of possible annual outcomes.
- Average annual loss (AAL). The long-run average of simulated yearly losses, including years with none. It is the baseline for the catastrophe component of the technical rate.
- Return period. A "1-in-100" loss is one with a 1% chance of being exceeded in any given year. It does not mean it happens once a century; two such years can occur back to back.
- Occurrence vs aggregate exceedance. Occurrence exceedance (OEP) looks at the largest single event in a year, which drives per-occurrence reinsurance. Aggregate exceedance (AEP) sums all events in a year, which matters for aggregate covers and cat bonds.
- Probable maximum loss (PML). The loss at a chosen return period, for example the 1-in-100 or 1-in-250 PML. See probable maximum loss. Regulators, rating agencies and reinsurers all ask for these figures.
A simplified view of how a technical cat price is built: AAL, plus a risk load for the capital needed to hold against the tail, plus reinsurance cost and expenses. Reinsurers and bond investors price the layers the same way, which is why Catastrophe bonds and ILS: how they work and who they affect quote an expected loss and a spread above it, and why layer prices are measured as rate on line.
Why your building data matters
The exposure data an insurer submits is the largest controllable input. Your broker's statement of values feeds it directly. The fields that tend to move modeled hurricane loss the most are:
- Exact address, geocoded to the building rather than the ZIP code (distance to coast, elevation and terrain all depend on it).
- Year built, because building codes and wind-resistance requirements changed over time.
- Construction class, number of stories and square footage.
- Roof shape, covering, age and the way the roof deck is attached.
- Opening protection: impact-rated windows or shutters on glass.
- Replacement values. If they are low, the model understates both the loss and the premium base; see insurance to value.
- Occupancy and contents, which drive contents vulnerability.
Add sprinklers and distance to hydrants for fire, retrofits and soft-story status for earthquake, and defensible space and ember-resistant vents for wildfire. The checklist for assembling it is in the statement of values: what property underwriters need, and the discounts a carrier may give for reinforcing a roof or opening protection are covered in mitigation credits and resilient construction for commercial property.
Worked example (hypothetical)
Say two warehouses, each valued at $10 million, stand in the same coastal county. Warehouse A was built in 1980, has an older roof with unknown deck attachment and no opening protection. Warehouse B was built to a modern wind code, with a reinforced roof deck and impact-rated openings. The numbers below are invented to show the mechanism.
| Building | Modeled AAL | AAL as % of value | Comment |
|---|---|---|---|
| A (older, unprotected) | $140,000 | 1.4% | Unknown roof data may be treated more conservatively |
| B (modern, reinforced) | $60,000 | 0.6% | Documented features lower the vulnerability curve |
A carrier loading its price to cover AAL plus a charge for tail risk would price A much higher. If the owner of A documents a roof replacement and adds shutters, the vulnerability inputs change and so does the quote. If the data stays "unknown", the model often assumes a poorer construction, and the cost falls on the insured. The same logic drives named storm deductibles: how percentage wind deductibles work: the higher the modeled loss, the more of it the carrier shifts to a percentage deductible.
How regulators use models
State regulators decide how far models can drive rates, and the rules differ.
- Florida. The Florida Commission on Hurricane Loss Projection Methodology reviews and accepts hurricane models for use in residential rate filings. Insurance Journal reported in June 2025 that Karen Clark & Company and Cotality were the first to satisfy the commission's newest standards, and Moody's reported that its RMS North Atlantic Hurricane Models version 25 was approved on June 12, 2025. The commission revises its standards every odd-numbered year. The commission's work is aimed at residential rate filings, so commercial buyers feel it indirectly. See florida's commercial property insurance market.
- California. The Department of Insurance announced enforcement of its catastrophe modeling regulation on December 13, 2024, and a companion rule on the cost of reinsurance followed on December 30, 2024. On August 1, 2025 the Department said it had completed review of wildfire models from Verisk, Karen Clark and Moody's and made all three available to insurers. Insurers using them in rate filings must commit to writing more policies, including commercial, in distressed wildfire areas (California Department of Insurance, August 1, 2025). The Department's 2026 filing instructions carry that commitment requirement forward. See california's commercial property and wildfire market and wildfire and commercial property insurance: cover, cost and mitigation.
Rate approval rules, filing mechanics and eligibility for these models change, so confirm the current state of a filing with your broker or the state department of insurance. Commercial lines in many states are not subject to prior approval, and surplus lines carriers generally do not file rates.
Where models fall short
- Secondary perils. Hail, severe convective storms, flood outside the big zones and some wildfire behavior are harder to model than hurricanes and quakes, and recent loss years have been dominated by them. See secondary perils: the storms, floods and fires that are not hurricanes and severe convective storms: hail, tornadoes and rising insured losses.
- Demand surge and claims behavior. Labor and material shortages after a regional event raise repair costs beyond the model's baseline.
- Model-to-model differences. Two vendors can differ materially on the same portfolio. Insurers usually blend or adjust.
- Non-stationarity. Models calibrated to past decades may not reflect current conditions; vendors update, but with a lag.
- Data quality. A model cannot correct a schedule that lists the wrong address or a replacement value that has not been updated in years.
What to ask your broker at renewal
- Which perils were modeled for my locations, and which were not?
- Did the carrier change a modeled zone, deductible or sublimit this year, and why?
- Which of my building features are being credited, and what is missing from the file?
- How would a named storm deductible apply to my schedule? See hurricane season and commercial property insurance: what to do before a storm.
Key takeaways
- A catastrophe model simulates events, applies them to your building data and the policy terms, and outputs average annual loss and return-period losses.
- The inputs you control, mainly accurate addresses, values, roof and opening details, directly change the modeled loss and the price.
- A 1-in-100 loss means a 1% annual chance, not a once-a-century event.
- Florida and California regulate how models are used in residential and wildfire rating, and the rules differ by state.
- Models are weakest on secondary perils such as hail and convective storms, where losses have grown.