CSAA Insurance Group refused to renew its policy this year on the home that Ronnie de Supinski owns with her husband, Bronis, in Livermore, Calif.
The reason given? Wildfire risk. The wildfire-risk score sent by CSAA? Zero.
As insurers suffer rising losses from wildfires and thunderstorms, they are using scores and models of natural perils to cherry pick which properties they insure, and set their rates.
Insurers say this helps keep home insurance available in disaster-prone areas, by allowing them to predict and manage risks. Consumers and industry analysts question how fair, accurate and reliable the tools really are.
Scoring of individual homes is one of two main methods insurers use to reduce losses from storms, wildfires and other risks. The other is catastrophe models that try to estimate the likelihood and cost of thousands of possible disasters. They are used by companies to justify rate increases and to reduce the risk that a single catastrophe, such as a hurricane, could blow them up.
The models and scores all ask broadly the same question: What is the likely risk of future losses? But they can spit out very different answers, particularly at the level of an individual home or even neighborhood, according to industry insiders.
After de Supinski complained, CSAA wrote to her that wildfire risks can still be present at homes with low-risk scores. The score looks only at the area up to a quarter of a mile from the property, but strong winds can carry embers up to 5 miles, the insurer said.
The 2017 Tubbs fire in California, the 2021 Marshall Fire in Colorado and last year’s Lahaina Fire in Hawaii all devastated neighborhoods containing homes with very low wildfire-risk scores, the letter added.
De Supinski said an insurance broker found her better coverage at a lower price than her original policy. But concerns persist. “My local friends…are all worried about [whether] they will get similar letters,” she added.
An analysis by Andrew Siffert, senior meteorologist at reinsurance broker BMS Re, found stark differences in wildfire-risk scores from leading hazard-analytics firms. A home close to de Supinski’s, in a similar neighborhood, was rated high risk by two hazard-scoring firms, moderate risk by one, and low risk by two others, the analysis showed.
One example of a widely used score is FireLine from risk-assessment firm Verisk Analytics. It rates buildings from zero to 30, with 30 being the highest risk of suffering a wildfire loss. Insurers often have internal guidelines, rejecting any home that scores above a certain—often undisclosed—number. The score can also have a dramatic impact on premiums.
Models go back to the 1980s, but their use for wildfires, storms and other common-but-costly perils is more recent—and, many say, less reliable. “Models are still a work in process…and every event tends to expose a gap,” said Martha Bane, a practice leader at insurance broker Arthur J. Gallagher. “They are better at predicting portfolio or industry losses than an individual client’s loss expectancy.”
Regardless, these tools are taking on increasing importance for insurers. Rising losses from hurricanes, storms and wildfires are prompting insurers to hike rates, curtail coverage and cull customers.
The most recent example came from Travelers, the big insurer, which was hit by Midwestern hail storms. It said this month it was making customers pay a bigger share of losses from wind, hail and tornadoes in 21 storm-prone states.
Wildfire risk is hard to predict accurately, in part because it is affected by everything from the dryness of local vegetation to the slope of the land and how the property is built, BMS Re’s Siffert said.
The risk can vary markedly within a relatively small area. Risk-scoring service Guidewire’s HazardHub, for example, rates more than half the properties in de Supinski’s ZIP Code as a very low or low wildfire risk, but almost a fifth as high or very high risk.
Insurers say they need models and scores to keep offering coverage in disaster-prone areas, as climate change, inflation and population shifts push extreme-weather costs ever higher. Global insured losses from natural disasters are increasing by 5% to 7% annually, and last year topped $100 billion for the fourth time in a row, according to reinsurer Swiss Re.
Models have been at the heart of the fight between insurers and California state regulators. Insurers have been fleeing the state in part because current rules say they can only base requests for rate increases on historic losses, rather than the expected future increase in wildfire costs.
Industry organizations told California regulators this month that models improve “the accuracy of assessing and pricing of risk for wildfire, flood” and other disasters. That is “critical in helping improve…[the] availability” of insurance in the state, the industry said.
California has been the only state not to allow catastrophe models to be used to set rates. That is changing as state Insurance Commissioner Ricardo Lara plans to allow insurers to use models as part of a package designed to end California’s continuing home-insurance crisis.
Matthew Nielsen, a senior vice president at ratings giant Moody’s, said allowing the models should help smooth rates, which could otherwise need to increase following a catastrophe. “It allows us to understand and quantify the magnitude of the wildfire problem,” he said. “In many ways, it helps us take the blindfolds off.”
There is an urgent need for insurers to improve their models. Insured losses from U.S. thunderstorms topped $10 billion from January to March, second only to last year’s tally for the same period, according to reinsurance broker Gallagher Re.
There is less reliable historic data on hailstorms, tornadoes and wildfires—known as secondary perils—making it harder for models to make good predictions. Models that rely on past events are being upended by climate change.
Barely a quarter of U.S. insurers have advanced predictive modeling capabilities, a report from consulting firm Capgemini said this month. “A huge portion of the industry is relying on outdated models,” that don’t take into account how fast catastrophes are changing, the report added.
Write to Jean Eaglesham at Jean.Eaglesham@wsj.com