Cat models under pressure after a year of global shocks

Global insured catastrophe losses remain below the decadal average this year, yet 2025 has delivered a series of events that challenged underwriting confidence and revealed emerging gaps in model behaviour.

The Webinar, titled “Lessons from 2025’s major catastrophe events” was hosted by insurance technology events provider InsTech, with risk consultant Claire Souch in the role of moderator.

Nigel Winspear, head of natural catastrophe analytics research at insurer Sompo International, opened the discussion by examining Myanmar’s 28 March earthquake. He described it as “a relatively small event in terms of insured loss, but the human cost was significant”.

The 7.7 magnitude earthquake had its epicentre just south of Mandalay and with far-afield impacts extending more than 1,000km. It caused a high-rise building under construction in Bangkok to collapse, showing “this potentially could happen in future” in other basins with similar characteristics.

He used the Burmese event as an analogue for California’s a San Andreas fault, noting “similarity in scale” and warning that vulnerability functions for existing California ‘quake cat models do not take similar events into account, such as for super-shear behaviours and long-period amplification affecting high-rise risks.

Wildfire model gaps and growing urban exposure

Turning to the year’s headline US wildfire losses, Nalan Senol Cabi, head of catastrophe model research at Arch Insurance Group, said the Los Angeles fires were not a Black Swan event but rather “a harsh confirmation of two specific trends… one on the exposure side and the other one on the modelling side”.

On exposure, she highlighted rapid expansion of the wildland-urban interface. “We now have 44 million homes sitting in that zone… we are literally fuelling the risk,” she said.

Urban conflagration represent “the biggest gap” for wildfire models, she said.

Current frameworks, she warned, are underweighting the tail risk of structure-to-structure spread.

Wildfire models are “quite sophisticated at predicting how fire spreads through vegetation”, Cabi observed.

She added: “There is still room to improve with the transition phase, when the fire leaves the trees and enters the city”.

The scale of risk is being worsened due to the pace of urban development and rising exposure values within cities, she stressed.

Cabi added that new tools, particularly with artificial intelligence (AI) could accelerate progress “exponentially”.

AI can be used to ingest dynamic fuel changes or atmospheric signal changes, she suggested, providing solutions these modelling challenges faster than previous challenges, such as for flood models.

“We will solve the urban conflagration problem much faster than we have solved the floodplain problem,” she added.

Convective storm escalation

Cameron Rye, director of natural catastrophe analytics at Willis Re, warned that the US is on course for another large loss year from US severe convective storms (SCS).

He estimated insured losses of around $50bn in 2025 after even larger years in 2023 and 2024.

He emphasised that claims severity has risen each year by around 7-8% for SCS events.

This has been driven by exposure growth but also “the cost to repair buildings [has] increased much faster than the rate of general inflation”, he observed.

Asphalt roof replacement costs have risen 250% since 2000 and by 45% in the past five years alone, he added.

Rising flood footprints

Cabi said more than 200 newsworthy floods were documented globally in 2025, adding that “we cannot mistake a lack of insured loss for a lack of risk”.

Flash floods in Texas, river flooding in Monmouth and severe inundation from Hurricane Melissa and Cyclone Alfred collectively illustrated the spread of events and divergent vulnerabilities.

She argued that tropical cyclone flood components “are not as sophisticated or high resolution as the inland flood models” and said the industry should “demand better solutions… for the tropical cyclone induced flood component and storm surge component”.

She also highlighted widening protection gaps.

The impacts of Cyclone Dido in Sri Lanka predictably demonstrated a low insured loss.

“The lack of insured loss is not a success; it is a major protection gap”, she said.

Even in the US, she suggested the National Flood Insurance Program (NFIP) remains hostage to longer-term uncertainties, receiving only “short term reauthorisations”.

Looking at trends, she said she expected surface water flooding caused by sudden rains to intensify fastest.

“Pluvial floods could be worse, infrastructure is already overwhelmed”, she warned, amplified by urban expansion into areas that cannot cope with short-duration intense rainfall.

Her practical message for re/insurers and their cat modelling providers was clear.

“This is the best time to fix your geocoding…if you are sitting on just zip-code geocoding in 2025 you are underwriting blind,” she warned.

Rye added that a rapid intensification in tropical cyclones leading to increased rainfall was also under-represented in modelling updates.

Discipline and stress testing

From a Lloyd’s perspective, Robert Stevenson, director of cat risk and reinsurance analytics, emphasised a need for market discipline and operational readiness.

He said the key market message was “maintaining discipline in the exposure management space” because without timely, accurate and complete data “you are going to be struggling”.

Stevenson warned that forthcoming market stress test exercises would test firms’ ability to produce robust exposure intelligence.

He ended by highlighting the rising relevance of manmade cat perils.

“We are pretty good at the nat cat stuff, but the things that may well start to surprise us are secondary elements,” he added, listing business interruption (BI) and strikes riots and civil commotion (SRCC) as rising supply chain threats.

To listen to the webinar, click here.