Some of America's largest home insurers are pricing wildfire risk at dramatically different levels of precision - and the gap is costing both carriers and consumers.
That is the central finding of "How Are Insurance Markets Adapting to Climate Change? Risk Classification and Pricing in the Market for Homeowners Insurance," a June 2024 National Bureau of Economic Research working paper (NBER Working Paper No. 32625) by Judson Boomhower of UC San Diego, Meredith Fowlie of UC Berkeley, Jacob Gellman of the University of Alaska Anchorage, and Andrew J. Plantinga of UC Santa Barbara. Using proprietary parcel-level wildfire risk data combined with insurers' regulatory filings across approximately 100,000 California households, the paper documents what the authors call a "winner's curse" at the heart of the US wildfire insurance market.
"There's tons of heterogeneity in wildfire loss risk, even within zip codes or even within neighborhoods," Boomhower said. "The wildfire insurance industry hasn't converged on a single model. There are striking differences in both granularity and modeling approaches across players."
The mechanism works as follows. Carriers using coarse geographic segmentation - pricing at the zip-code level, for instance, rather than at the parcel or grid-cell level - cannot distinguish between high- and low-risk properties within a given area. Granular pricers can. They attract the lower-risk homes at competitive prices, leaving the coarser pricer with a disproportionate share of higher-risk properties at rates that do not reflect the true exposure. Expected losses run above projections. The insurer either exits the market or raises rates further - which accelerates the cycle.
"Our research highlights the key role of information limitations in the pricing and availability of insurance for wildfires and other climate risks," Boomhower said. "Firms that rely on coarser measures of wildfire risk are exposed to potentially severe adverse selection as a result of their information disadvantage relative to insurers with richer models."
The paper found that firms relying on simpler pricing methods respond in one of two ways: they charge relatively high prices in high-risk market segments, or they choose not to serve those areas at all. Both outcomes reduce availability. Both raise costs. Neither solves the underlying problem.
Carriers with more sophisticated pricing infrastructure rely on stochastic catastrophe models - from vendors including Verisk and Moody's RMS - that simulate hundreds of thousands of wildfire scenarios, integrating vegetation data, slope, wind patterns, and construction characteristics at the individual parcel level. Leading underwriters license multiple models and triangulate across outputs, treating divergence between models as a signal of uncertainty that informs both pricing and capacity decisions.
Peggy Brinkman, a principal actuary at Milliman, tracks the methodology evolution closely. The actuarial toolkit has advanced from basic univariate techniques through generalized linear models to gradient boosted machine models and, most recently, explainable boosting machines - a class of model that combines gradient boosting accuracy with the interpretability that state regulators require for rate filings.
"New modeling techniques can extract more value from the same data in terms of risk understanding," Brinkman said. "Advances in methodologies are as important as bringing in new data sources."
On the data side, satellite and aerial imagery now enable automated assessment of property characteristics - roof condition, vegetation proximity, construction materials - at a scale and frequency that on-the-ground inspection cannot match. Internet of Things sensors represent the next frontier, though Brinkman notes these applications remain in an early, low-sample-size phase.
Boomhower's research identifies regulatory approval burden as a compounding factor. California's prior approval regulations limit the rate increases insurers can request, and approval timelines can stretch for months. "Real costs are involved in implementing sophisticated pricing systems," Boomhower said, "both from a backend technology and regulatory perspective."
Those costs tip the investment calculation against improvement for smaller regional carriers. For large national insurers with the actuarial capacity and legal resources to engage regulators across multiple states, the investment is justified. For a regional mutual carrier operating in a single state, the payback period may not be. The result is a two-tier market in which pricing sophistication increasingly tracks company size rather than the actual risk profile of the book.
Recent California reforms - making it easier to incorporate frontier catastrophe model outputs in rate filings and to include reinsurance costs in pricing - represent a shift in that environment. The California Department of Insurance has published updated guidance on catastrophe model use in rate cases. Whether those changes are sufficient to bring large admitted carriers back at scale remains to be seen; the California FAIR Plan's enrollment growth through 2025 suggests the market has not yet reversed course.
Carriers with finer-grained pricing technology hold a risk-selection advantage that compounds over time. The coarser pricer's loss ratio problem does not announce itself immediately - it surfaces after several fire seasons of accumulating adverse selection. By the time it is visible in the results, the gap in modeling infrastructure between the carrier and its competitors has typically widened further.
"There would be a lot of competition among insurance companies to find the low-risk homes in these designated high-risk areas," Boomhower projected. The carriers best positioned to win that competition are those that have already made the investment in granular pricing infrastructure. For those that have not, the question is whether the cost of catching up is lower than the cost of continuing to price blind.
