The Hours Between the Model and the Calculation Agent

In March we published a post that said most catastrophe bonds use parametric triggers. That was wrong, and we have corrected it in place with a note saying so. Of the 245 outstanding bonds Calchis tracks from the Artemis Deal Directory, 162 settle on the sponsor's own claims, 76 settle on an industry loss index, and 7 are parametric. By principal it is roughly $48 billion, $15 billion and under $2 billion. Anyone who has priced a cat bond knew that already. We should have.

We are writing this rather than quietly editing because the correction is the product. A monitoring tool for parametric positions is only worth anything if the people using it believe the numbers it shows them, and the fastest way to lose that is to be caught overstating the market you serve. So: the market for what we do is 7 bonds by count today, plus every parametric position, industry-loss warranty and private deal that never appears in a deal directory. Small in count. Large in consequence per position, because the trigger is binary.

Two authorities, on either side of the event

The cat bond market already has its authorities, and it is worth being exact about what they do. The modeling agent prices the risk before issuance. Verisk, formerly AIR Worldwide, says it has modeled more than $63 billion of cat bond principal and supported over 80 percent of public property-catastrophe issuances since 2011; its 2026 tropical cyclone model puts 65 percent of the expected loss on outstanding 144A principal on U.S. hurricane. That work is done when the offering circular is printed.

The calculation agent works after a potential trigger event. In Verisk's own description, that means "exposure growth calculations, determining related earthquake events to a denoted principal earthquake, reconciling event reports to parametric trigger requirements, or, for Industry Loss Index triggers, disaggregating state-level industry loss reports to county level." For the 76 industry-loss bonds, the index itself is Property Claim Services: a designated catastrophe of $25 million or more in insured loss, estimated from roughly 150 contributing insurers, published as a preliminary figure after the event and resurveyed until it is final.

Both are authoritative. Both are on either side of the event. Neither is a measurement, during the event, against a threshold a fund manager set.

What happens in between

A hurricane advisory cycle is six hours. A fund manager holding a parametric wind bond written on a box off the Louisiana coast gets, in each of those six hours, an NHC advisory with a maximum sustained wind, a forecast track, and a broker's email summarizing both. The bond's threshold is in a spreadsheet. The question — is this storm going to cross 96 knots inside the box, and when do I have to tell my investors — gets answered by a person reading two documents against each other at whatever hour the advisory lands.

Some of that question can be answered from public federal data while the event is running, and some of it cannot. NHC publishes maximum sustained wind and position every advisory. USGS publishes magnitude and epicenter within minutes. NWS gauges publish stage. NIFC publishes containment. Those are the quantities parametric triggers are written on, and they are public. Industry loss is not measurable from any public feed; it is PCS's number or PERILS' number, weeks later, and we do not derive one. Indemnity cannot be measured from outside the sponsor's books at all.

What we do in that window, and what we refuse to do

Calchis lets a fund manager define the threshold — wind speed, magnitude, surge, gauge stage, containment — and the geography, as a point and radius or a drawn polygon, and evaluates it against the live federal feed every 5 minutes. Each evaluation records the measured value, the source it came from, the threshold and the percentage reached. When a reading crosses the threshold, the record is a content-addressed determination: the reading and the specification it was evaluated against are each named by the hash of their content, so anyone holding the same public document can recompute the same identifier and check it. A warning level, 80 percent of threshold by default, fires before the breach does, by webhook or email.

Here is what that is not. It is not a settlement. The calculation agent's determination stands, and nothing we publish changes it. It is not an industry-loss estimate; we carry none, derived or otherwise, and we will not offer that threshold type until we carry a licensed one. And it is not a model of the bond's own terms: the catalog says, for every one of the 245 bonds, that zero are evaluated against live events, because their payout terms live in offering circulars we have not been given, and we do not guess terms. The number moves when a cited structure is transcribed and reviewed by a person. Not before.

Why an honest zero is the feature

It would be easy to make the catalog look busier. An earlier version of this product did exactly that, seeding plausible thresholds for a set of bonds, and we deleted it, because a wrong attachment point behind a confident interface is worse than a blank field. A blank field sends the reader to the document. The same discipline is why this post opens with a correction. If a tool that exists to tell you when a number crossed a line cannot say plainly when its own number was wrong, it is not a tool you can put a position behind.

Sources: Verisk, "Cat Bond Fundamentals" (modeling and calculation agent roles, $63 billion, 80 percent); Verisk, "A sharper view of U.S. tropical cyclone risk," 2026 (65 percent of expected loss on outstanding 144A principal); Artemis, "How PCS integrates Contributor, Verisk, and Public Data to define industry losses" (designation criteria). The catalog counts are ours as of September 23, 2026.