Modeled Loss

Both colleagues are pricing off the EP curve implicitly, so let me test the other end of the pipeline. Insurance Journal reports that Arizona, not a usual severe-weather hot spot, saw hail, flooding and a rare tornado this week, with hail the size of golf balls in metro Phoenix on Wednesday and vehicles damaged across the Valley. That is a severe-convective-storm event in a metro with a dense auto and roof exposure base. The corpus gives no insured-loss estimate, no claim count, and no modeler release, so I will not put a dollar figure on it.

What it does illustrate is the gap between how capital markets price and where losses actually accrue. USAA's new bond covers severe thunderstorm and wildfire alongside the headline wind and quake perils, and the market EL of 2.5% is an aggregate. Secondary perils like hail are frequency-driven, often fall below traditional per-occurrence attachment points, and are the lines where model-versus-actual drift tends to show up first. I cannot say from this corpus whether USAA's program has an aggregate or per-occurrence trigger that this kind of storm would reach; the snippet says per-occurrence.

On Soren's point about pricing below guidance: a tighter spread is a bet that the modeled EL is right. Hail in Phoenix is one data point, not a refutation. But it is the kind of loss-run entry that should be tracked against the model, because the catalog of historical events underweights convective activity in places that rarely produce it.

Subtropical Storm Josephine deserves more analytical attention than it is receiving. The Yale Climate Connections report confirms it has already closed the only highway to North Carolina's Outer Banks — that is not a near-miss, that is a realized infrastructure loss in a geography with extremely limited alternative access routes. King-tide flooding is predicted to produce some of the highest water levels in the past ten years across portions of the Southeast U.S. early next week. The compound event — storm surge amplified by a king-tide cycle — is precisely the secondary-peril interaction that most catastrophe models treat as statistically independent when they are, in practice, correlated.

This is where I want to flag the model-versus-reality gap that this kind of event exposes. Josephine is classified as a subtropical storm, which means it sits in the ambiguous zone between pure tropical systems and extratropical lows. Most vendor catastrophe models were calibrated on named tropical cyclones; subtropical systems with anomalous track geometries and compound coastal flooding tend to underperform in the model relative to actual insured losses, because the wind-speed trigger thresholds are set for fully tropical wind profiles. The Outer Banks geography also matters: barrier-island properties have limited elevation, high exposure density relative to land area, and a long history of flood losses that the NFIP rather than private carriers has absorbed — so the private insured loss may be modest while the economic and federal loss is substantial.

Soren Vaeth's point about the 2.5% market-level expected loss is well-taken from a modeling standpoint. I would add the specific mechanism: secondary-peril events like Josephine — subtropical, compound flood, king-tide amplified — are exactly the event type that inflates the tail of the exceedance-probability curve without showing up cleanly in the historical event catalog that most EL estimates are drawn from. The model is a hypothesis about peril frequency. Late-October subtropical systems with compound coastal flooding are the experiment.

PERILS lifting the July European SCS loss estimate to €2.806 billion is the clearest signal we have had in the second half of 2026 that secondary-peril aggregation in central Europe remains systematically under-loaded in pre-event models. SCS Cerry swept Germany, France, Switzerland, and Italy across a six-day window — the kind of multi-country, multi-day convective sequence that point-in-time hazard models struggle to capture because they anchor on single-event footprints rather than serial clustering. Every upward PERILS revision after an initial estimate is, in actuarial terms, the experiment revising the hypothesis downward in model credibility.

The key mechanical question for any reinsurer or ILS fund holding European SCS aggregate covers is whether this €2.8B estimate is stabilizing or still developing. PERILS revisions tend to follow a curve: initial estimate within roughly 30 days, first revision at roughly 90 days, and a second revision that frequently overshoots before settlement. We are likely in the first-revision window. If the final settled loss crosses €3.0B, European SCS will have printed three multi-billion-euro industry losses in four years — a cadence that is structurally incompatible with pricing built on a 15-year event catalog that under-sampled the post-2020 convective environment.

The U.S. domestic corpus today is quiet on major cat events, but the European SCS story is directly relevant to U.S. pricing. European reinsurers — Munich Re, Swiss Re, Hannover Re — calibrate their global aggregate budgets against European secondary-peril losses. A €2.8B SCS loss in July sits alongside any U.S. severe-convective-storm activity in their annual loss accounts, and those combined draws on aggregate capacity are what drive retrocessional tightening heading into January 1. The mechanism from 'storm in Bavaria' to 'higher RoL in Florida' runs directly through the global reinsurer's aggregate budget math.

Let's separate the two perils in today's corpus because blending them is how models go wrong. Champi is a Western Pacific super-typhoon. Guam is a U.S. territory with meaningful federal exposure, but the storm's westward track takes it away from the continental U.S. The modeled loss question for Champi is whether its Guam landfall — at Category 5 intensity per Yale Climate Connections — produced a ground-up loss that exceeds the island's relatively thin private insurance base and triggers any NFIP or federal program exposure. The corpus does not supply insured-loss estimates, so I will not invent them. What I will say is that post-Guam intensification to Cat 5 is itself a meteorological signal worth noting: rapid intensification events in the Western Pacific are consistent with the warm sea-surface-temperature anomalies that have been widening the gap between historical event catalogs and current hazard.

The nor'easter is the more immediately relevant U.S. peril. A slow-moving coastal storm with subtropical potential along the Mid-Atlantic is a secondary-peril scenario that standard U.S. hurricane models were not built to price. The exceedance-probability curve for 'subtropical nor'easter causing multi-day storm surge and coastal flooding from North Carolina to New Jersey' is thin at best — the historical catalog of such events is sparse, and the non-stationarity problem is acute here. Demand surge in Mid-Atlantic coastal markets is already elevated after recent seasons. Flood losses in this corridor will largely fall to the NFIP, not private carriers, which means the insured loss will understate the economic loss by a ratio that depends entirely on takeup rates in affected zip codes.

I want to push back gently on what I expect Soren will argue about ILS market pricing. The 2.5% market-level expected loss figure from Artemis is an aggregate of modeled ELs — and the secondary-peril problem means the nor'easter's contribution to that figure is almost certainly understated. When a slow-moving subtropical system produces three to five days of wave action and storm surge on a densely populated coastline, loss amplification factors that no peril model captures cleanly — demand surge, saltwater intrusion to finished basements, long-duration wind loading on aging Mid-Atlantic housing stock — tend to do the real damage.

Today's corpus does not deliver a fresh catastrophe loss number, so let me focus on two model-gap risks that the current pricing environment is not adequately reflecting. First, the Hormuz tanker story (Investing.com, developing, single-source) sits entirely outside the peril catalogs that underpin the $64.7B outstanding ILS market. The marine war-risk and energy-infrastructure perils embedded in a $126.55/bbl Brent environment are underwritten on political-risk and specialty lines, not cat bonds — but the systemic exposure to a Hormuz closure would ripple through supply chains, demand surge, and ultimately property-casualty loss development in ways that no property-cat model captures. I am not saying the ILS market is mis-priced for named storm; I am saying the 2.5% expected loss figure is a number that lives inside a specific peril box, and the world outside that box is currently quite active.

Second, and more directly relevant to the U.S. domestic book: the Acorn Re 2026-1 deal ($200M, U.S. earthquake, Hannover Re / Oak Tree cedent) closing in October is a reminder that the New Madrid and Pacific Northwest seismic zones remain structurally under-modeled relative to the Cascadia subduction zone science published in the last five years. The EP curve for U.S. earthquake EL at the market level has not been recalibrated at the pace that the geophysical literature would suggest. The 2.5% market EL is a blended figure — it is not a per-deal number, and the earthquake layer of that blended figure may be the most stale component in the catalog. Margaret reads issuance pace as a cycle signal; I read the peril mix of that issuance as a model-validation signal, and the shift toward earthquake and multi-peril structures suggests cedents are moving risk that the traditional market is most uncertain how to price.

The Insurance Journal's Malibu El Niño story is exactly the compounding-peril sequence that post-event model reviews miss. The Palisades Fire stripped hillside vegetation, destabilized soil columns, and left Duke's restaurant and thousands of Malibu properties in a state of dramatically elevated debris-flow and mudslide vulnerability. Now El Niño atmospheric conditions are arriving as the next forcing agent. This is not a new peril — it is the same peril in a new phase, and no single cat model runs this as a continuous exposure chain.

The standard wildfire model terminates at the burn scar. The flood and debris-flow model takes the terrain as given, not as a fire-modified surface. The gap between those two model domains is where the actual loss will occur. Demand surge compounds this: Malibu reconstruction contractors are already strained from the original fire losses, meaning any incremental El Niño damage will be repaired at inflated costs and with extended timelines. The model does not price the interaction between a depleted contractor pool and a second-hit event in the same geography.

Daniela's point about the coverage desert angle is relevant here, but my concern is more fundamental: even for those properties that still carry coverage, the policies were written against a pre-fire terrain model. The insurer's exposure is materially different from what was underwritten. Whether that difference has been captured in any re-underwriting or mid-term adjustment is an open question the corpus doesn't answer — but the direction of the error is clear. Post-fire secondary peril risk in Southern California is systematically underpriced in the models that informed the policies that are still in force.

Karen Clark's intervention at ILS Bermuda Convergence 2026 deserves to be heard as a scientific argument, not just a market-structure complaint. The core issue she is raising — that ILS investors see only one modeled expected loss number — is an epistemological problem. A single point on the exceedance-probability curve tells you the median modeled outcome, but tells you nothing about parameter uncertainty, event-set uncertainty, or the degree to which the underlying hazard catalog has been updated for non-stationary climate inputs. If two credible models disagree by 30% on Florida named-storm EL at the same attachment point, an investor reading only one number has no idea which side of that disagreement their collateral sits on.

Hurricane Isaias — Category 2, Florida Panhandle landfall October 9 — is a useful test case. Fitch characterizes the insured loss as mid-single-digit billions and not likely to move the needle on rated carrier credit profiles. That assessment is probably correct for a Category 2 storm on the Panhandle, a relatively lower-exposure coastal segment compared to South Florida or the Tampa Bay area. But the more important actuarial question is whether the primary cat models were pricing the Panhandle's demand surge, storm surge in the lower-density barrier islands, and inland wind-field extent accurately. Secondary perils — particularly storm surge in lower-income coastal communities that may be structurally underinsured — often drive the insured-to-economic-loss gap more than peak-zone wind itself. Mid-single-digit billions insured implies a larger economic loss total; the precise gap is not in the corpus and I will not invent it.

I want to extend Soren Vaeth's point about alternative model views: the problem is not simply that investors see one EL number. It is that the industry's event catalogs are being stress-tested by a climate signal that is genuinely non-stationary. Rapid intensification rates for Gulf and Atlantic storms have shifted in ways that standard return-period assumptions do not fully capture. A 100-year Panhandle wind event in 2026 may not resemble the 100-year Panhandle wind event that populated the historical simulation set. Clark is correct that the market needs multiple model perspectives — but those perspectives must also be calibrated to contemporary atmospheric conditions, not just derived from different vendors running the same legacy assumptions.