Vendor models significantly underbaked the chances of a ~$40bn California wildfire, with components of the risk, including some climate change impacts, not adequately reflected in modelled outcomes, sources have told Insurance Insider ILS.
While each vendor model takes a different approach, sources canvassed by this publication suggested that model outputs generally equate a $40bn California wildfire with a return period in the range of 1-in-100 years to 1-in-200 years.
The range reflects the outcomes after vendors released updates to their wildfire models in 2024, which addressed some areas of shortfall and cut return periods.
Verisk updated its wildfire model last summer, taking its return period for a $40bn California wildfire to 1-in-150 years, down from 1-in-multiple-hundreds of years, according to sources.
Meanwhile, a consensus of ILS managers indicates that their in-house view of a $40bn California wildfire equates to a return period of around 1-in-50 years, to as low as 1-in-30 in one case.
Wildfire modelling has come into focus following the Palisades and Eaton fires that tore through Southern California in January, destroying around 16,000 structures and racking up industry losses currently estimated to be around $35bn-$40bn for the two events combined.
Several senior re/insurance executives foregrounded their concerns about the issue on Q4 earnings discussions with analysts.
Kevin O’Donnell, CEO at RenaissanceRe, told analysts that, while its own models had performed well in its assessment of return period, “a loss of this magnitude implies that models will need to steepen the curve in the tail to better reflect the higher frequency of severe events.”
RenRe has estimated its own loss at $750mn and the industry loss at $50bn, from the California fires.
Meanwhile Peter Zaffino, CEO at AIG, said the wildfires “demonstrate the increased loss from secondary perils and the magnitude of tail events that are not captured well in the modelling.”
AIG is expecting to incur a loss of roughly $500mn from the fires.
Climate change impacts on wildfires
Sources said the main driver of the shortening tail in wildfire risk is climate change, with drought conditions and changes to rainfall patterns contributing to the increasing frequency and severity of wildfires.
Firas Saleh, director, North American wildfire models, at Moody’s RMS, said: “Greater precipitation allows for the rapid growth of vegetation. When this is followed by severe droughts, as was the case in the Southern California wildfires, vegetation dries out and becomes a more flammable fuel to wildfire, thereby increasing wildfires risk.”
From June to December last year, almost no rain fell in the Southern California Coastal region, according to the US National Oceanic and Atmospheric Administration (NOAA) -- resulting in unusually dry conditions during the winter months.
This came after 20 months of elevated rainfall, with 37.1 inches falling in the year to September 2023 and a further 19.5 inches in the eight months to May 2024, supporting vegetation growth.
In January this year, the emergence of strong Santa Ana winds of up to around 70 miles per hours at their peak then acted to accelerate the spread of the fires.
Sources said one way that wildfire models could better account for these risks would be to improve data collection on defensible space. This is the area of vegetation and/or trees surrounding a property that can impact on the spread of fires.
Sources noted that data about defensible space collected by primary insurers is often limited, leaving reinsurers to fill in the gaps with assumptions.
Alex Allman, head of corporate underwriting – geo-risks, at Munich Re said: “We have high quality exposure data but not for every client, so in a few cases we have to make assumptions.”
Ignition source data and subrogation impacts
The ignition source can also significantly impact on the size of a wildfire insured loss, and therefore how this factor is modelled is also under the spotlight following Palisades and Eaton.
Multiple sources said model outcomes would be improved by incorporating data on ignition sources.
The three main ignition sources are man-made fires, naturally occurring fires and fires sparked by utility company issues.
The impact of adding ignition source data would be particularly significant in the case of utility-caused fires, because subrogation can materially reduce the overall loss to carriers, a source noted.
Utility-caused fires in the past five years include Echo Mountain Fire (2020), Marshall Fire (2021), Dixie Fire (2021), Lahaina Fire (2023) and Smokehouse Creek Fire (2024).
“The vendor models don’t explicitly state the ignition source, so it’s harder to apply how you might think of subrogation in the overall loss distribution,” a source said.
Another source noted that utility-caused fires tend to result in large losses, because powerlines, transformers and other electrical equipment are built near populated areas.
Andrew Siffert, senior VP and senior meteorologist, at BMS Group, added that the changing profile of wildfire risk has resulted in utilities taking more risk mitigation measures such as implementing public safety power shutoff.
Other factors that worsened the loss outcome for Palisades and Eaton but are difficult to model for included issues with low water pressure and availability as well as fire-fighting resources.
The Santa Ynez Reservoir in Pacific Palisades was emptied for repairs at the time of the fires, while the fire-fighting response has been criticised as inadequate by some commentators.
Sources also raise costs of smoke damage to properties and the expense associated with re-housing policyholders as having the potential to inflate losses beyond the modelled numbers.
Wildfire model updates
Model vendors KCC, Moody’s RMS and Verisk all updated their wildfire models during the past year.
Saleh said Moody's RMS's update “refreshed many of the underlying fuel data layers and introduced a comprehensive array of modifiers.”
This includes allowing landscape-scale fuel modifications, letting users make alterations to account for vegetation expanding or disappearing, or becoming drier or wetter.
The latest RMS wildfire model also features an urban conflagration module, “specifically calibrated to assess the vulnerability of built environments to wildfires”, Saleh said.
The module lets users account for movement of embers between one property and another, including how this can enable a fire to spread.
One characteristic of the Palisades and Eaton fires was the rapid and widespread movement of embers across the LA landscape, carried on the powerful Santa Ana winds.
Verisk’s model update included focusing on “understanding the eco-system and how that interacts with the climate at hand”, as this is one of the key elements of wildfire modelling, according to Dr Julia Borman, director of regulatory and rating agency client services, at Verisk Extreme Event Solutions.
The update brought in changes around how weather conditions impact on eco-systems and land areas, including how droughts, intense rains and humidity can impact a forested area versus a grassy area.
Karen Clark & Company (KCC)’s wildfire model update of early 2024 included a focus on incorporating drier conditions.
Karen Clark, co-founder and CEO at KCC, said: “Our scientists implement climate change impacts through a variable called vapour pressure deficit (VPD). There is scientific consensus that as the temperature warms and the air gets drier, the VPD increases and our models can determine this at a high resolution across different regions.”
The update did not greatly alter the model’s view of wildfire risk, which was already “conservative,” Clark added. The model’s return period for a $40bn California wildfire is below 1-in-100.
Meanwhile, CoreLogic’s return period for a $40bn California wildfire is approximately 1-in-150 years.
The firm’s latest update to its wildfire model incorporated the Coupled Model Intercomparison Project 5 (CMIP5) dataset, which includes conditions beyond those observed in historical data.
CMIP is a project of the World Climate Research Programme providing climate projections to understand past, present and future climate changes.
The update allows the model to run simulations that show expected climate conditions in the future and the impact this would have on wildfire.
Jamie Knippen, senior product manager, at CoreLogic, said the update had not changed the return period for a $40bn California wildfire, because “at the state level, the exceedance probability curve in the relevant range did not change significantly.”