How catastrophe models shape your premium

A catastrophe model is software that simulates thousands of possible hurricanes, wildfires, earthquakes or storm seasons and estimates what each would cost an insurer. The average of those simulated losses, spread across the homes an insurer covers, becomes the catastrophe part of your premium, and the same output tells the insurer how much reinsurance to buy. You rarely see the model, but you see its effects: a rate that varies street by street, a credit for a hip roof, or a renewal that jumps after a vendor updates its model.

What is inside a catastrophe model

The NAIC's overview of property catastrophe models describes four modules, though vendors label them differently.

• Hazard. A catalog of simulated events, each with a strength, a location or path and an annual probability. For hurricanes that means thousands of synthetic storms with wind fields; for wildfire, simulated ignitions and spread over vegetation, slope and wind.
• Vulnerability. Damage functions that turn the hazard at a site into physical damage, based on features such as construction type, roof shape and age, and number of stories.
• Exposure. The insurer's own portfolio: addresses, rebuilding values, building characteristics and policy terms such as deductibles and limits.
• Financial. Converts damage into insured loss after deductibles, limits and reinsurance, then totals the results into a loss distribution.

Two outputs matter for pricing. The average annual loss is, in the NAIC's words, the loss amount averaged across all years in the event set. The exceedance probability curve shows how likely a loss of a given size is to be reached or passed. A "1-in-100-year" loss means a 1% chance of being exceeded in any one year, not a loss that arrives once a century. Insurers call the corresponding dollar figure the probable maximum loss.

The main vendors are Verisk, Moody's, Karen Clark & Company and Cotality (formerly CoreLogic). Large insurers often run more than one model and some build their own, since the answers differ.

From model output to your premium

The model produces a loss estimate. A pricing actuary does the rest, in two places.

In your rate. The expected annual catastrophe loss for your territory becomes a "cat load" built into the base rate, along with expected non-catastrophe losses, expenses, and a return on the capital the insurer must hold against bad years. A high-wind coastal ZIP Code or a wildfire-prone canyon gets a large cat load; an inland neighborhood a small one. Your own house then adjusts that number up or down based on the vulnerability inputs the insurer collected about it.

In the insurer's reinsurance. The same model estimates the loss at various return periods, which tells the insurer where its reinsurance layers need to attach. What reinsurers charge for those layers is built into rates too. reinsurance costs and your premium covers that step. Public programs use the same logic: the Texas Windstorm Insurance Association set a 1-in-100 probable maximum loss of $6.227 billion for the 2025 season, then a 1-in-50 figure of $4.3 billion for 2026 under a new state law, which shows how the choice of return period moves the target (TWIA, February 2026).

A hypothetical build-up

Say a model estimates your house's expected annual hurricane loss at $900 after the deductible. The numbers below are invented to show the structure, not any insurer's pricing.

If a model update raises the expected hurricane loss from $900 to $1,100, that piece alone adds $200, and the reinsurance and capital loads, which usually track modeled loss, rise with it. The premium could climb by well over $200 while nothing about your house changed.

Who checks the models

Because a model drives so much of the rate, regulators have started to look inside it. The approach differs sharply by state.

Florida has done this longest. Section 627.0628 of the Florida Statutes creates a Commission on Hurricane Loss Projection Methodology that sets actuarial standards and reviews models. Insurers filing rates must use accepted methods and, in the statute's words, "may not modify or adjust" them when determining hurricane loss factors and probable maximum loss levels (Fla. Stat. 627.0628). Portions of commission meetings that discuss trade-secret model details are closed.

California historically priced wildfire using a catastrophe factor based on past losses and did not allow reinsurance costs in rates. Under the Sustainable Insurance Strategy, a regulation operative on December 12, 2024 lets insurers use forward-looking models, which must reflect mitigation by homeowners and communities, if they commit to write in distressed areas (10 CCR 2644.4.5 and 2644.4.8). The Department of Insurance said on August 1, 2025 that it had finished reviewing wildfire models from Verisk, Karen Clark & Company and Moody's, making all three available to every insurer. See California's Sustainable Insurance Strategy.

Colorado passed HB 25-1182, effective July 1, 2026, which requires insurers using wildfire models or scores to share model information with the Division of Insurance and to publish how policyholders can earn discounts and appeal (Colorado; wildfire risk scores and how insurers use them).

In many other states regulators review a model case by case when an insurer files a rate, and can question it if something looks off. The NAIC's Catastrophe Risk Management Center of Excellence, a unit of its research arm, trains regulators on how the models work and maintains vendor model documentation for state insurance departments (NAIC). It is a resource for regulators, not a rate approver.

Where models fall short

Models are estimates with wide error bars, and the honest limits matter for how much weight to give a surprising rate.

• Data gaps. Models are only as good as the property data fed in. Moody's said in May 2026 that year built, roof age and roof covering are often missing from insurers' files, so models fall back on average-property assumptions (Carrier Management). A wrong roof age can cost you real money.
• Secondary perils. Hurricanes and earthquakes have decades of modeling behind them. Hail, tornado outbreaks, wildfire and inland flood are younger problems. The NAIC notes that severe convective storm models are under ad hoc regulatory review, and that Hurricane Katrina showed models had not fully captured flood, demand surge and evacuation effects, which modelers then added. Swiss Re counted secondary perils at a record 92% of global insured losses in 2025, so the weaker models cover a growing share of what insurers pay.
• Model change. The NAIC says models are "continually advanced as new events occur." After a large event, a new version can move expected losses for a whole region at once.
• Mitigation credit. A model only rewards what its vulnerability module recognizes. A feature that helps in real storms but is not an input may earn no discount.

The honest case on each side

Insurers and modelers argue that models are the only way to price a rare, enormous loss. A company with 30 years of claims in your ZIP Code may never have seen a Category 4 hurricane or a canyon fire, so a pure history understates risk, as California's old method did. Forward-looking models, they say, also let regulators demand credit for mitigation, which history cannot do.

Critics answer that vendor models are proprietary, that reputable models can disagree widely, and that trade-secret protection means a rate increase can rest on numbers consumers cannot check. The disagreement is real: within days of the January 2025 Los Angeles fires, Moody's RMS put insured losses at $20 billion to $30 billion while CoreLogic (now Cotality) put them at $35 billion to $45 billion (Artemis; Artemis). Those were early event estimates, not rate models, but they show how far apart good models can start. Others point out that when an entire industry buys from three or four vendors, a shared blind spot becomes a market-wide one. Regulators like California's try to answer that by trading model use for market commitments; whether that lowers prices or only improves availability is still being tested.

What you can do about it

1. Find out what the insurer thinks your house is. Ask for the roof age, roof material, year built, construction type and any wildfire score or tier on file. In California and Colorado you can demand some of this in writing.
2. Correct errors and send proof. A roof replacement invoice or permit, a photo of a clean defensible-space perimeter, or a corrected square footage can change an input. Keep the dated copy.
3. Document the features models reward. A wind mitigation inspection in Florida, a FORTIFIED designation (The FORTIFIED Home standard and insurance discounts), impact-rated openings, roof-deck attachment and secondary water resistance are the kinds of items vulnerability modules and state credit schedules recognize.
4. Ask which discounts apply and what proof they need before you pay for upgrades. The roof and mitigation upgrade payback calculator compares the credit to the cost.
5. Read the rate filing if your state publishes it. Many states post filings through the SERFF Filing Access portal, and how insurers file rate increases with regulators shows what to look for.
6. Shop with the same facts. Different insurers use different models, so a quote that looks punishing at one company may be ordinary at another. Use the home insurance renewal increase checker to see whether your increase is out of line.

Key takeaways

• A catastrophe model simulates thousands of events to estimate average annual loss and the odds of very large losses; the output feeds both your rate and your insurer's reinsurance purchases.
• The catastrophe share of your premium can change when a vendor updates a model or an insurer switches models, without any change to your home.
• Regulation varies: Florida has a state commission that accepts hurricane models, California reviewed three wildfire models in 2025, Colorado requires disclosures from July 1, 2026, and most states review case by case.
• Models are weakest on hail, wildfire and inland flood, and on missing property data such as roof age; you can often fix the data.
• Ask your insurer which features it credits, document your mitigation, and compare quotes from companies that use different models.