A new paper in the Journal of Catastrophe Risk and Resilience by MS Amlin researchers Luke Wedmore and William Sturgeon puts a number on a gap the industry has known about and not closed: supershear earthquakes account for 66% of insured earthquake losses since 2016, roughly $13.2 billion, and appear in no commercial catastrophe model.

The gap is not in the hazard science. It is in the tooling every carrier, cat bond sponsor and rate filing depends on.

Key Takeaways

  • 66% of insured earthquake losses since 2016, about $13.2 billion, came from events where supershear rupture was the dominant propagation mode.
  • Losses rise 5% to 10% at the 200-year return period and 30% to 60% at the 500-year when supershear effects are modelled into representative portfolios.
  • Roughly 36% of major strike-slip earthquakes since 2010 involved supershear propagation, and it favors long, straight, mature faults like the San Andreas.
  • $425 million of Sutter Re 2026-1 priced in May 2026 on expected losses of 2.30% and 4.05%, generated by a model that does not carry supershear.
  • None of the three major vendors has incorporated supershear rupture dynamics into a standard earthquake module as of June 2026.

What the Paper Found

In a conventional earthquake the rupture front travels below the shear wave velocity of the surrounding rock, around 3 kilometers per second. A supershear rupture breaks that barrier and can approach compressional wave speeds near 6 km/s, producing a Mach cone.

Three consequences separate it from a subshear event for loss purposes. Energy travels in a focused corridor rather than spreading radially, so intensities stay high well beyond the distances conventional attenuation predicts. Successive waves arrive in rapid sequence, loading a structure again before it has dissipated the first pulse, against design codes that assume one attenuating event. And rotational ground motion twists structures about their vertical axis, a mechanism with a moment arm that grows with building height and no representation in current vulnerability functions.

The event record is the evidence. The Mw 7.8 Kahramanmaras earthquake in February 2023 showed supershear characteristics and produced insured losses estimated at $5.8 billion by CRESTA. Myanmar's Mw 7.7 event in March 2025 ruptured 475 km, roughly 230 km further than standard models predicted, transitioning to supershear at about 5.3 km/s and holding it for more than 200 km. And the 1906 San Francisco earthquake, which anchors much of California's historical loss scenario work, has since been reclassified as a supershear event over roughly 470 km of the San Andreas.

Approximately 36% of major strike-slip earthquakes globally since 2010 have involved supershear propagation. It is not a curiosity.

The Two Return Periods Do Different Jobs

MS Amlin modelled the effect on representative insurance and reinsurance portfolios, and the asymmetry in the result is the whole point.

Incorporating supershear raises modelled losses 5% to 10% at the 200-year return period and 30% to 60% at the 500-year. The first is a margin adjustment. The second is not, because the 500-year band is where catastrophe excess-of-loss attachments and cat bond attachment points get calibrated.

Take the capital side first. Solvency capital derives from the 200-year value at risk, so the smaller adjustment still moves the charge. For a carrier holding $500 million of earthquake PML at the 250-year return period, a 7.5% upward revision is $37.5 million of additional capital against the same book, with no change in exposure or in the underlying rate.

The treaty side is where the larger number lands. A programme structured to attach above a modelled 500-year loss, on a PML that is 30% to 60% low, is not attaching where it was priced to attach. The layer sits at a materially shorter return period than intended, and the cedent retains risk the programme was bought to remove.

The same mechanism reaches the ILS market through expected loss. The California Earthquake Authority priced Sutter Re Ltd. Series 2026-1 in May 2026, raising $425 million across two tranches.

Tranche Size Expected Loss Spread (Final) Initial Guidance
Class C Notes $325M 2.30% 3.50% 4.25% - 5.00%
Class F Notes $100M 4.05% 5.50% 6.50% - 7.25%

Investor demand let the CEA upsize twice, from $300 million to $400 million and then to $425 million, while tightening spreads well inside initial guidance. Both expected losses come from a model without supershear. On the Class F notes, a revision toward the 4.25% to 4.45% range against a final 5.50% spread is a tighter multiple than the book was priced on, and there is more than $6 billion of purely earthquake-exposed cat bonds outstanding in the 144A market plus $17.6 billion of multi-peril bonds carrying quake exposure.

Nobody Has Shipped the Fix

The science is not contested, the loss attribution is published, and as of June 2026 no major vendor module carries it.

Moody's RMS previewed a North America Earthquake HD update at Exceedance 2026 promising new occurrence and hazard science, with no public mention of supershear rupture. Verisk is migrating its platform from Touchstone to Synergy Studio across more than 110 risk models, a transition consuming the development capacity that a hazard rebuild would need. CoreLogic's earthquake models are among those approved for NAIC and state regulatory use.

That last one is the structural problem rather than a scheduling one. California's approval of third-party cat models for property rate filings makes the approved vendor model the regulatory standard, so a filed catastrophe load inherits the gap by design. Approval does not correct a model; it makes its output the reference point.

MS Amlin has updated its own internal view of risk. Most carriers cannot: the in-house seismology to build a supershear overlay is not a common capability, and the interim step the paper suggests, running alternative shaking patterns through existing models, still needs a vendor to produce the scenario.

Meanwhile the pricing environment moves the other way. Property catastrophe rates are falling at the fastest pace since 2014, which means reinsurers are accepting less rate for a peril the research says may be larger than modelled, on the same mid-year renewals where earthquake programmes are being restructured.

California is where that concentrates: stresses on the San Andreas and San Jacinto are at their highest in 1,000 years, the southern section has been locked for over a century, and the CEA alone runs a risk transfer programme of roughly $9.15 billion including $2.875 billion of catastrophe bond coverage. It is the same sequence secondary perils ran before they reached 92% of $107 billion in 2025 nat cat losses: undermodelled until the loss experience forced the recalibration.

Further Reading

Sources