Verisk's 2026 benchmark sets global insured catastrophe losses at $171 billion in an average year, up $19 billion from $152 billion a year earlier and the highest figure the firm has published (Verisk, September 2026). It is a modeled long-run mean rather than a forecast for any single year.

The mean rose 12.5% in twelve months during which the market moved the other way. Guy Carpenter's global property catastrophe rate-on-line index is down 16% across the 2026 renewals, the steepest annual decline in the index's 25-year history (Artemis, July 2026). Rob Newbold, president of Verisk's Catastrophe and Risk Solutions unit, named the tension directly: "A quiet hurricane season can lead markets to respond as if risk has eased" (Insurance Journal, September 2026).

Key Takeaways

  • $117 billion of the $171 billion, or 68% of global modeled insured catastrophe risk, sits in the United States. The concentration means a US-only view of the benchmark is close to a global one, and a US-only reinsurance panel is not diversified.
  • Severe thunderstorm carries 40% of the global insured mean, ahead of tropical cyclone at 27%, earthquake at 10%, winter storm at 9%, flood at 7% and wildfire at 6%. That implies roughly $68 billion a year from the peril least likely to reach a treaty.
  • The 12.5% increase outruns the exposure drivers Verisk names. Property exposure has grown about 7% a year since 2021 and US reconstruction costs about 5%, leaving a residual attributable to expanded model coverage and updates to the risk view.
  • $477 billion at the 100-year return period and $606 billion at 250 years put the one-in-a-hundred aggregate at about 2.8 times the mean, a ratio that matters to any layer priced as a multiple of modeled AAL.
  • Realized 2025 insured losses were $107 billion, roughly 60% below the new mean, in a year with no US hurricane landfall (Swiss Re Institute, March 2026).

What the $19 Billion Increase Contains

The benchmark is the mean of Verisk's full simulated event set, so it is a distributional statistic and not an expectation for 2026. It has moved in steps: $151 billion in the 2024 edition, $152 billion in 2025, $171 billion now, against $59 billion when the firm first published the series in 2012 (Artemis, September 2026). The 2012 comparison is the one most widely quoted and the one that carries the least information, for reasons taken up below.

Verisk attributes the current increase to three things at once: property exposure growth of roughly 7% a year since 2021, US residential repair and rebuild cost increases of roughly 5% a year over the same period, and updates to model coverage and to its own view of risk. Only the first two are quantified in the release.

The peril mix is where the benchmark becomes usable. Severe thunderstorm is the single largest contributor at 40% of global insured AAL, a share that reflects the peril's dominance in the US book carrying two thirds of the total.

PerilShare of global insured AALImplied annual mean
Severe thunderstorm40%$68.4bn
Tropical cyclone27%$46.2bn
Earthquake10%$17.1bn
Winter storm9%$15.4bn
Flood7%$12.0bn
Wildfire6%$10.3bn

Shares are Verisk's, September 2026. The dollar column is computed by applying each share to the $171 billion global insured AAL and is not published in that form.

The insured mean sits inside a modeled economic mean of more than $450 billion, so about 38% of modeled annual catastrophe loss carries insurance. Verisk's companion European release puts the continent at $110 billion economic against roughly $24 billion insured, a 22% take-up well below the global figure.

Why the Largest Modeled Peril Is the One Treaties Rarely Pay

Severe thunderstorm's implied $68 billion arrives as many moderate events rather than one large one, which is the shape that sits under an occurrence attachment. Gallagher Re counted $26 billion of US severe convective storm insured loss inside a $46 billion global first-half total, and recorded a fifth consecutive quarter with no single event costing insurers more than $10 billion (Gallagher Re, July 2026).

Attachment points did not move to meet that. Reinsurers held the elevated structures set in the 2023 reset through the 2026 renewals, leaving primary carriers retaining the same share of secondary-peril loss they retained last year, a pattern AM Best has tracked across the big four European reinsurers. Price came down; the point of attachment did not.

For a primary carrier building a catastrophe load, those two facts do not offset. The gross modeled cost of the book rose 12.5% on Verisk's view, and the portion of that increase concentrated in severe thunderstorm lands almost entirely inside the retention, because a $2 billion hail outbreak does not reach a treaty structured for a $10 billion hurricane. The ceded cost fell instead: rate-on-line down 16%, and catastrophe bond spread multiples averaging 2.29 times expected loss against 3.14 a year earlier (Artemis, Q2 2026). A cheaper program does not reduce a net catastrophe load when the growth in modeled loss sits below the attachment.

The tail sets the other boundary. Verisk's modeled aggregate insured loss reaches $477 billion at the 100-year return period and $606 billion at 250 years, so the one-in-a-hundred aggregate is about 2.8 times the mean. A reinsurer pricing a high layer as a multiple of modeled AAL inherits the full 12.5% move in that mean, including on the layers that repriced downward this year.

A Benchmark That Cannot Be Differenced

The awkward property of the $19 billion is that it blends a change in the world with a change in the measuring instrument, and Verisk says so. Exposure growth near 7% and reconstruction cost growth near 5% are real inputs, but they do not add to 12.5% in any way the release makes reproducible, and the residual belongs to model coverage and risk-view updates the user did not choose.

The same problem compounds over the fourteen-year comparison. Moving from $59 billion to $171 billion is not a loss trend, because part of the 2026 figure covers perils and territories the 2012 model did not represent at all. Differencing two vintages of a model returns the change in the model plus the change in the risk, with no published split between them.

Realized experience does not settle it either, though it leans one way. Severe convective storm produced $51 billion of the $107 billion insured in 2025, roughly 48%, in a year Swiss Re Institute recorded secondary perils at a record 92% of global insured catastrophe loss. Through the first half of 2026 the peril ran at 57% of the global total. Two observations against a long-run mean prove nothing, but both sit above the modeled 40%.

So the 2026 renewal and the 2026 benchmark answer different questions in the same units. The market repriced a year in which $46 billion of loss emerged against a $64 billion ten-year first-half average. Verisk repriced a distribution. Reconciling the two requires knowing how much of the $19 billion was the world and how much was the model, and that split is not in the release.

Further Reading

Sources