Severe Convective Storms Overtake Hurricanes as the Costliest Insured Peril: What P&C Actuaries Need to Recalibrate

Aon's January 2026 Climate and Catastrophe Insight recorded a crossing point: since 2000, secondary perils including severe convective storm, flood and wildfire have produced at least $1.56 trillion of industry insured losses against roughly $1.04 trillion for tropical cyclone, earthquake and European windstorm combined.

Severe convective storm is now the costliest insured peril of the century, and not because of one outlier year. The five-year average annual U.S. SCS loss stands at roughly $32.5 billion, inflation-adjusted, against approximately $2.5 billion in the early 1980s.

Key Takeaways

• $50 billion-plus of U.S. SCS insured losses for a third consecutive year in 2025, with global SCS insured losses of $61 billion, the third-highest annual total on record.
• Up to 90% of SCS loss growth since 2000 is non-weather on Gallagher Re's analysis: development into exposed areas up 20%, roof replacement costs, litigation trends and insurance-to-value gaps.
• Model event sets were calibrated on data running through the 2010s, before the three consecutive $50 billion years, and Guy Carpenter measured a 4.6% variance between expected and actual losses over 1990 to 2023.
• Allstate booked roughly $925 million from fifteen March wind and hail events, with three of them producing about 80% of that, inside $1.24 billion of Q1 catastrophe losses.
• Gallagher Re puts the reinsurance repricing threshold at $115 billion to $125 billion above expected annual catastrophe losses, which SCS at current levels does not approach as a single event.

A Peril That Accumulates

The crossing happened without a single defining event, which is most of why it took a cumulative study to notice.

Year	U.S. SCS Insured Losses	Billion-Dollar SCS Events (U.S.)	Notable Context
2023	~$55B	Record high	Previous all-time SCS record
2024	~$41B insured / $57B total	Second-highest count	Second costliest SCS year
2025	$51B insured / $68B+ total	Third consecutive $50B+ year	EF5 tornado in North Dakota
Q1 2026	~$16B (all perils, 79% of global)	5 events above $1B insured	Single $4B SCS event (March 10-12)

SCS produced approximately $55 billion of global insured losses in 2023, beating the 2011 record by almost $20 billion. In 2024, U.S. severe thunderstorms alone caused $57 billion of total losses with $41 billion insured, the second costliest SCS year. In 2025, global SCS insured losses reached $61 billion, and U.S. SCS insured losses passed $50 billion for a third consecutive year against total economic damage above $68 billion.

A Category 4 or 5 hurricane can generate $30 billion to $100 billion in days, episodically. SCS works through the loss budget quarter after quarter, across a geographic spread concentration management cannot avoid.

Q1 2026 shows the pattern inside a quiet quarter. Global economic losses of $37 billion sat well below the 21st-century Q1 average of $64 billion, while insured losses exceeded $20 billion, 6% above the historical average. The March 10-12 SCS outbreak alone produced $5 billion economic and $4 billion insured.

Carrier disclosures locate it. Allstate reported $1.24 billion of pretax catastrophe losses for Q1, of which fifteen March wind and hail incidents produced roughly $925 million, three events supplying about 80% of that. State Farm had taken over 53,000 home and auto claims by March 16, concentrated in Illinois, Ohio and Missouri. A carrier weighted toward Texas, Illinois and Ohio cannot diversify away from this peril without rebuilding its book.

Why the Models Are Behind, and by How Much

The pricing problem is that the tools carriers rate this peril with were fitted to a period that ended before the losses arrived.

The observational base carries a known bias. The Storm Prediction Center database, the foundation of U.S. severe weather climatology since the 1950s, collects far more reports where people are: storms over the Plains and upper Midwest go underreported, so apparent frequency is inflated near cities and total event counts understated. Moody's RMS documents greater undercounting in low-density Plains areas during summer.

The calibration vintage compounds it. Research presented at the CAS Spring Meeting found that the data behind leading SCS model event sets ran only through the 2010s, before the three consecutive $50 billion-plus years of 2023 to 2025. Guy Carpenter measured a 4.6% variance between expected and actual losses across 1990 to 2023, and that gap has widened in the most recent years. Hazard resolution adds a third layer: a hailstorm dropping 3-inch stones on one subdivision can leave rain two miles away, and coarse footprints miss percentage deductibles for wind and hail.

Vendors have moved. Moody's RMS launched North America Severe Convective Storm HD models in December 2025 with a re-engineered vulnerability module calibrated against over $55 billion of location- and policy-level claims and more than 2,700 validated damage curves, using machine learning on densely observed regions to infer undercounting nationally. Cotality's 2026 report puts 43.5 million U.S. properties at moderate or greater hail risk, and counts 142 damaging hail days in 2025 against a 20-year average of 122.

The actuarial consequence sits in the transition rather than the models. As carriers adopt updated event sets, indicated rates for SCS-exposed territory rise, and the difference between legacy and updated output is the accumulated underpricing from the prior model generation arriving in one filing. That is a regulatory conversation as much as a technical one, and the NAIC has noted it expects these models to appear in rate filings more frequently.

Recalibrating the weather does not close the whole gap. Gallagher Re attributes up to 90% of SCS loss growth since 2000 to non-weather factors: urban and suburban development in SCS-prone regions up 20% since 2000, roof replacement costs rising on materials, labor and code upgrades, assignment-of-benefits and attorney representation surging in Texas, Florida and Colorado, and replacement cost values that have not kept pace with construction costs. An event set fitted to a better hazard history still prices those through exposure and severity assumptions the model does not set.

The Peril the Reinsurance Market Does Not Discipline

The complication is structural: the market that normally forces primary rate adequacy barely touches this peril.

Property catastrophe excess-of-loss treaties are built for low-frequency, high-severity events. SCS produces frequent moderate losses that erode retained earnings and rarely reach high attachment points. Gallagher Re's April First View put the loss level needed to move property reinsurance pricing at $115 billion to $125 billion above expected average annual catastrophe losses. SCS at current elevated levels does not approach that as a single event or series.

So the loss lands almost entirely on the primary carrier. Reinsurers collect higher ceding premium as SCS-driven rate increases flow through, while bearing limited loss exposure absent aggregate stop-loss or quota share structures. The reinsurance pricing signal, historically the strongest external pressure on primary rate adequacy, is muted for the peril that now contributes the largest share of insured losses.

The concentration inside the peril makes that worse rather than better. Hail accounts for as much as 80% of SCS claims in a given year, with roof damage an estimated 70% to 90% of insured residential catastrophe losses.

Cotality puts 1-in-100-year hail at roughly $30 billion of insured losses, comparable to a major hurricane, and 1-in-500-year hail at about $58 billion, roughly 80% of the $71 billion total SCS loss at that return period. Texas carries more than $2.2 trillion of reconstruction cost value at moderate or greater hail risk across the Dallas-Fort Worth, Houston, Austin and San Antonio corridor, and more than 235,000 Texas homes took damaging hail in 2025.

That is hurricane-scale tail exposure sitting on the side of the balance sheet where the treaty does not respond.