California's Public Wildfire Model Enters Its RFP Stage

California's Department of Insurance released a Request for Expertise on March 25, 2026 under SB 429, the first formal step toward building a state-owned wildfire catastrophe model to compete with the Verisk, Karen Clark, and Moody's models it just finished certifying for rate filings, with university consortium proposals due June 22, 2026 (California DOI).

Having followed the Sustainable Insurance Strategy's vendor model approvals since Commissioner Ricardo Lara finalized the framework, the emergence of a state-funded competitor model is the kind of methodological fork that rarely gets actuarial scrutiny before it becomes policy. The department spent 2025 running three proprietary catastrophe models, Verisk's Wildfire Model for the United States, Karen Clark and Company's US Wildfire Reference Model, and Moody's wildfire model, through its Pre-Application Required Information Determination (PRID) review, clearing all three for use in prior-approval rate filings (California DOI, July 2025). SB 429, chaptered on October 10, 2025 as Chapter 541 of the Statutes of 2025, was moving through the legislature at the same time, and it does not ask whether cat models belong in ratemaking. It asks whether the state, not three private vendors, should own the model that decides who gets to keep a homeowners policy.

What SB 429 Actually Builds, and What "Black Box" Means Here

SB 429 creates the Wildfire Safety and Risk Mitigation Program inside the Department of Insurance and directs it to fund, through a competitive grant to a university-led consortium, the development of a "public wildfire catastrophe model," defined in the statute as a computerized process using the best available science to simulate wildfire property damage, with full public access to its documentation, programs, underlying data, and algorithms (SB 429, Chapter 541, Statutes of 2025). That last clause is the entire point of the bill. The Verisk, Karen Clark, and Moody's models cleared PRID review through a process that included public comment and a technical webinar, but the models' internal fire-spread algorithms, ignition probability surfaces, and vulnerability functions remain proprietary. Regulators, consumer advocates, and competing carriers can see the model's output and a description of its methodology; they cannot see the code.

Consumer Watchdog and other intervenors who pushed for SB 429 frame the objection as a due-process problem: a homeowner whose premium doubles because a proprietary model reclassified their parcel's wildfire hazard has no practical way to audit the reclassification, and neither, in full, does the department that approved the model. That is a real limitation of PRID review, which validates a vendor's methodology and calibration on a sampled basis rather than re-deriving every parameter from source data. It is also a limitation every catastrophe-modeling regime in the country shares, including Florida's decades-old Commission on Hurricane Loss Projection Methodology process for hurricane models. SB 429 does not amend PRID review or claim the vendor models are wrong. It builds a parallel, fully transparent alternative and lets the market, and eventually rate filings, decide whether transparency is worth the years it will take to build.

The RFP Stage: What Happens Next and On What Clock

The department's public wildfire model page lays out a sequence that moves faster on paper than most state IT procurements. Cal Poly Humboldt led a strategy group that delivered consortium-structure recommendations in May 2025, ahead of the bill's own passage. The department then released its Request for Expertise to prospective university leads on March 25, 2026, with responses due June 22, 2026 (California DOI, Public Wildfire Model). A grant to the winning consortium is anticipated in October 2026, at which point actual model development, not procurement, begins. The department wants a consortium spanning fire science, wildfire mitigation, actuarial science, data science, and computational modeling, which is a meaningfully broader mandate than a single vendor's in-house modeling team; Verisk, KCC, and Moody's each built their wildfire models primarily with catastrophe modeling and geospatial staff, not a standing academic consortium with quarterly public deliverables.

That structural difference is worth sitting with for a moment before the timeline math, because it explains why this project looks nothing like Florida's public hurricane model. Florida's Public Hurricane Loss Model, built at the University of Florida and maintained under state contract, has run for over two decades as a genuine public alternative that competes directly with the Florida Commission's roster of approved vendor models in Citizens Property Insurance Corporation ratemaking. California's project is explicitly modeled on that precedent, but wildfire behavior, driven by fuel load, topography, and human ignition patterns that shift year to year with drought and vegetation management, has a shallower validated data record than Atlantic hurricane climatology. Building a credible public wildfire model from scratch is a harder scientific problem than the Florida precedent suggests, which is precisely why the timeline runs in years, not months.

The Actuarial Certification Problem No One Has Modeled Yet

Assume, for a moment, that the consortium delivers a working model on schedule and California eventually authorizes it for rate filings alongside Verisk, KCC, and Moody's. An appointed actuary certifying a homeowners rate indication for a carrier writing across wildfire-distressed counties would then face a genuinely new problem: reconciling loss cost outputs from a public model and a proprietary model that were built with different fuel-load assumptions, different ignition-probability calibrations, and, almost certainly, materially different expected annual losses for the same portfolio of addresses.

Under current PRID practice, a carrier picks one approved model (or blends models it has separately justified) and documents that choice in its rate filing exhibits under ASOP No. 38's catastrophe modeling standard. That works because every currently approved model is proprietary and unauditable to roughly the same degree, so the actuary's job is choosing among comparably opaque alternatives on the basis of validation studies and peer use. A public model changes the comparison's character entirely. If the state's own model produces a materially lower expected annual loss for a given territory than the vendor model a carrier has filed on, the carrier faces a documentation problem it has never had before: explaining, in a rate filing an intervenor group can now audit against a fully open alternative, why it selected the model that produces the higher indicated rate. Regulators reviewing the filing gain a reference point they never had under an all-proprietary regime, and appointed actuaries lose the shelter of "every available model is similarly opaque" as a justification for model choice. The certification standard for wildfire ratemaking in California does not currently contemplate this scenario, because until SB 429's model exists, it could not.

The CMS Precedent: What Happens When a Public Model Enters a Vendor's Market

Actuaries in health and Medicare Advantage practice have already lived through a version of this fork. CMS built its Hierarchical Condition Category risk-adjustment model as the mandatory federal standard for Medicare Advantage risk scoring starting in 2004, developed through CMS's own actuarial and clinical staff rather than licensed from a commercial vendor. Commercial risk-scoring vendors did not disappear once the CMS-HCC model became mandatory; they repositioned around it, building analytics, coding-optimization, and calibration tools that operate on top of the public model rather than competing with a rival proprietary score. The public model became the shared, auditable substrate; the vendor economics shifted from selling the score itself to selling services that interpret and act on it.

The parallel for California is instructive but imperfect. CMS's model was mandatory from the outset, which forced the market's hand immediately. SB 429's public wildfire model would enter a market where three proprietary models are already approved and in active use, competing for adoption rather than displacing an incumbent by regulatory fiat. Verisk in particular has built a multi-peril catastrophe modeling franchise, reporting that its underwriting-facing revenue lines, which include catastrophe and risk solutions alongside forms and rules services, grew across 2025 (Verisk, Q4 2025 results). A state model that insurers can adopt for free, with no license fee and full algorithm access, is a direct threat to that revenue line if California ever mandates or even strongly incentivizes its use, the way the 85% wildfire-distressed writing commitment already steers carrier behavior under the existing framework. Vendors' more likely path, following the CMS playbook, is repositioning around the public model, offering validation services, portfolio-level analytics, and blended methodologies that layer proprietary refinement on top of the state's open baseline rather than competing model-for-model on a fully transparent product they cannot out-transparency.

Quantifying the Stakes: What Any New Model Must Be Validated Against

Two figures anchor what a credible wildfire model has to get right in California, and both come from the market's most recent and most expensive test case. The January 2025 Palisades and Eaton fires produced an estimated $28 billion to $35 billion in insured property losses, with Verisk's Extreme Event Solutions group splitting that range into roughly $20 billion to $25 billion for the Palisades fire and $8 billion to $10 billion for Eaton (Verisk, January 2025). Milliman's independent estimate ran higher, at $25.2 billion to $39.4 billion, and Swiss Re Institute later characterized the pair as the costliest wildfire event globally on record. Any model, public or proprietary, that cannot reproduce loss severity in that range for a comparable event configuration has already failed its most important validation test before it reaches a single rate filing.

The second figure is the California FAIR Plan's own reckoning with that loss experience. The Plan filed for a 35.8% statewide rate increase in October 2025, its largest requested hike in seven years and the first FAIR Plan filing built on forward-looking wildfire catastrophe models and net cost of reinsurance rather than smoothed historical losses. The department approved a lower 29% increase rather than the full request (Insurance Business Magazine, 2026). That gap, roughly seven points of requested rate the department declined to grant, is itself a data point about how much scrutiny cat-model-driven wildfire indications already draw before a public alternative even exists to contest them. A state-built model that eventually produces a lower loss cost for the same territories the FAIR Plan just repriced would not simply offer competition; it would hand every future intervenor a specific, itemized number to litigate the FAIR Plan's methodology against.

Model	Status as of mid-2026	Transparency posture
Verisk Wildfire Model (US)	PRID review completed July 2025, first model cleared	Proprietary; methodology disclosed, algorithms confidential
Karen Clark & Company US Wildfire Reference Model v3.0	PRID review completed August 2025	Proprietary; methodology disclosed, algorithms confidential
Moody's wildfire model	PRID review completed 2025	Proprietary; methodology disclosed, algorithms confidential
California Public Wildfire Model (SB 429)	RFP stage; consortium responses due June 22, 2026, grant anticipated October 2026	Fully open by statute: data, algorithms, and code public

The Multi-Year Gap and the 85% Writing Commitment

Nothing about the RFP timeline suggests the public model will exist, let alone be validated and available for rate filings, before the early 2030s. An October 2026 grant funds the start of development, not a finished product; the department's own page frames the initial phase as development and demonstration, with deployment as a later milestone. In the interim, the entire weight of Commissioner Lara's Sustainable Insurance Strategy rests on the three proprietary models the department already certified. Under that framework, admitted carriers that use department-reviewed catastrophe models and incorporate net cost of reinsurance into their rate filings must, in exchange, write at least 85% of their statewide voluntary market share in wildfire-distressed ZIP codes and counties (California DOI, Sustainable Insurance Strategy).

That commitment is the entire mechanism keeping the admitted market from ceding more ground to the FAIR Plan while the public model is still years from existing. If a carrier's 85% writing commitment rests on rate adequacy calculated from a Verisk, KCC, or Moody's model that a public alternative later shows to be materially miscalibrated for a specific set of high-hazard territories, the interim years are not a neutral waiting period. They are years in which coverage commitments, pricing, and depopulation of the FAIR Plan all proceed on the assumption that today's three certified models are correct, an assumption SB 429 was written specifically because some intervenors do not fully accept it. Actuaries at carriers relying on the 85% framework should treat the public model's eventual arrival as a future audit of decisions being locked in now, not a distant academic exercise.

Will Other Wildfire States Follow?

Colorado and Oregon are the two most-watched candidates for parallel legislation, and both have the regulatory infrastructure to move quickly if SB 429 produces a usable model. Colorado's Division of Insurance has already built out a wildfire risk model use framework tied to its own catastrophe-modeling rate filing rules, adopted in the wake of the 2021 Marshall Fire, and its legislature has shown a pattern of importing California framework language on climate-linked insurance regulation with a one-to-two-year lag. Oregon's insurance division has similarly signaled interest in catastrophe-model transparency following its own wildfire seasons, though neither state has introduced SB 429-equivalent legislation as of mid-2026. The more likely near-term outcome is that Colorado and Oregon regulators watch California's RFP and initial model architecture closely, and that any legislative follow-on waits until the consortium produces a demonstrable methodology rather than committing state funding to an unproven concept. A failed or indefinitely delayed California build would chill replication elsewhere; a credible demonstration model within two to three years would likely trigger genuine legislative interest in both states.

Why This Matters for Actuarial Practice

The immediate actuarial work is not about the public model itself, which does not yet exist in any usable form. It is about the certification and disclosure practices that will need to exist before it does. Appointed actuaries at carriers with material wildfire-distressed exposure should begin documenting, now, the specific reasons a given proprietary model was selected over available alternatives, in language that would survive comparison against a future transparent baseline rather than language that only needed to survive comparison against equally opaque competitors. Reserving and pricing teams tracking the FAIR Plan's 29% approved increase and the 35.8% it did not get should treat that seven-point gap as a preview of the scrutiny a public model, once it exists, will bring to every wildfire rate filing that follows it. And capital and ERM actuaries modeling multi-year wildfire exposure should build the public model's eventual arrival into their planning horizons as a known, if distantly timed, source of model risk: not because the state model will necessarily prove more accurate, but because its mere existence changes what regulators, intervenors, and courts will expect a proprietary model's defenders to prove.