Mexico's M7.3 Quake Missed Its Cat Bond Trigger by One Tenth of a Magnitude

The magnitude 7.3 earthquake that ruptured 58 kilometers off Puerto Madero on July 17 landed inside a live trigger box of the IBRD CAR Mexico 2024 catastrophe bond, in a cell where the $70 million Class B notes activate at M7.4 (USGS, July 2026; Artemis, July 2026). One tenth of a magnitude unit kept $17.5 million of principal in investors' hands. Mexico's own seismological service measured 7.4. The bond did what its map said it would do, and the map is where the actuarial content sits.

Key Takeaways

One tenth of a magnitude unit decided $17.5 million. The USGS settled the July 17 rupture at Mww 7.3; the cell's Class B notes activate at 7.4.
M7.9 is what the same cell required to touch the $225 million Class A tranche, against 6.7 in the grid's most sensitive cells. The threshold prices distance to exposure before any event occurs.
About $16.7 million a year of earthquake risk margin buys $295 million of quake limit, so Mexico had spent roughly $38 million and recovered nothing in the 27 months to July.
$260 million is what Mexican triggers have paid in two decades, against roughly $218 million of risk margin on the current four-year program alone.
4.4 times expected loss is what Class A's 1-in-85 layer earns, against 1.9 times for Class B and 3.9 times on the identical 2020 vintage tranche.

How the 2024 Grid Reads an Earthquake

The earthquake classes of IBRD CAR Mexico 2024 run on what the World Bank calls a cat-in-a-grid trigger. Covered territory is divided into boxes, each box carries its own minimum magnitude for each class of notes, and a qualifying event pays a share of principal that steps up with intensity. Verisk's AIR unit built the risk model and serves as calculation agent, reading each event's magnitude, depth and location against the grid. The 2024 renewal redrew that geometry deliberately, adding an explicit depth condition, smaller boxes and smoother transitions between adjacent cells (World Bank, February 2025).

The grid prices geography the way an exposure-rating actuary would. The published Class A payout map shades each cell by the magnitude that pays 100 percent of principal, in bands running from 6.7 at the low end to 8.8 and above at the high end. The offshore cell where July's rupture landed required 7.9 for Class A and 7.4 for the riskier Class B. A subduction-zone event off the Chiapas coast, however energetic, projects modest modeled loss onto the federal asset base the bond protects; a smaller, shallower event under a population center clears a much lower bar.

Attenuation, depth and distance to exposure are all priced into the thresholds in advance, which is what lets the payout decision collapse to a table lookup afterward.

The payout function is stepped rather than binary, running from 30 to 70 percent and from 70 to 100 percent of principal as intensity climbs, with Artemis reading Class B's first activation in this cell at 25 percent of principal for an M7.4 (Artemis, July 2026). On $70 million of notional, that first rung is worth $17.5 million.

The nominated reporting agency then decided the event. The USGS catalog carries a spread of solutions beneath its single decimal: one contributing network's Mwp solution read 7.4, the W-phase inversion 7.3, a body-wave method 7.0. Mexico's Servicio Sismologico Nacional published magnitude 7.4, locating the event 135 kilometers southwest of Ciudad Hidalgo (N+, July 2026). The bond does not insure the earthquake. It insures the nominated agency's estimate of it, and moment magnitudes routinely differ by a tenth because they invert different waveform sets through different velocity models.

The Premium Ledger Behind the Near-Miss

Mexico pays the risk margin component of the coupon, with the World Bank's funding rate riding on top for investors, and the disclosed terms let anyone reconstruct the ledger. Margins run 4.0 percent on Class A, 11.0 percent on Class B, 13.5 percent on the Atlantic hurricane Class C and 12.0 percent on the Pacific hurricane Class D. Applied to the tranche sizes, the $595 million program costs roughly $54.6 million a year, of which the two earthquake classes account for about $16.7 million against $295 million of quake limit.

In the 27 months between settlement and July's event, Mexico had spent about $38 million of earthquake risk margin and recovered nothing. That is the unremarkable base case for remote-layer protection, and still the first number a budget hearing hears.

Class	Size	Peril	Trigger	Expected loss	Attachment prob.	Risk margin	Margin / EL
A	$225m	Earthquake, low risk	Cat-in-a-grid	0.90%	1.17%	4.0%	4.4x
B	$70m	Earthquake, high risk	Cat-in-a-grid	5.84%	8.30%	11.0%	1.9x
C	$125m	Atlantic hurricane	Cat-in-a-gate	5.69%	7.96%	13.5%	2.4x
D	$175m	Pacific hurricane	Cat-in-a-gate	4.09%	6.26%	12.0%	2.9x

The shape is the standard cat bond result, with remote layers earning the richest compensation per unit of modeled risk: Class A's 1-in-85 attachment probability commands 4.4 times its expected loss while Class B's roughly 1-in-12 layer earns 1.9 times. The vintage comparison prices the hard market the sponsor renewed into: the March 2020 bond's equivalent Class A carried the identical 0.90 percent expected loss at a 3.5 percent margin, a 3.9x multiple.

Pricing has since rolled over, with secondary spreads compressing to 5.71 percent in late June. A 2028 renewal at current multiples buys the same grid materially cheaper, an option the sponsor holds and the 2024 investors have already been paid to write.

The cross-cycle ledger is blunter. Mexico's triggers have paid $260 million ever: $50 million for Patricia, $150 million for the 2017 Chiapas earthquake, $60 million for Otis. The government has since doubled its separate traditional parametric program to roughly $575 million for the June 2026 to May 2027 term, carrying about $1.17 billion of parametric protection in total. It prices speed well above expected-value breakeven.

Where the 0.1 Lands in Basis Risk Pricing

The reflex is to read July 17 as basis risk made visible, and the classification runs the other way on the facts. Basis risk is the mismatch between payout and need, and a program's history sorts into four quadrants: payout with loss, payout without loss, loss without payout, and neither. July produced no reported structural damage and no payout, which files it in the quiet quadrant where trigger and need agree. Had the SSN's 7.4 been the nominated datum, investors would have funded $17.5 million against near-zero damage, a false positive on the other side of the ledger.

The genuine sponsor-side failure in the record is June 2020, when a USGS M7.4 near La Crucecita struck inside the prior bond's trigger zone, killed at least six people and damaged buildings across Oaxaca, and paid nothing because that cell's minimum sat higher (NBC News, June 2020). Six years apart, the same printed number lands on opposite sides of the payout line depending on which cell contains it.

The other compromise is scale rather than accuracy. Otis produced a $60 million payout against the NHC's $12 to 16 billion damage estimate (NHC, March 2024). The bond is a liquidity layer scaled to first response, and it pays while emergency-phase spending is still under way: the M8.1 Chiapas event of September 2017 exhausted a $150 million tranche with confirmation on October 11, five weeks after the rupture, with no adjuster involved. A sovereign cannot get that from indemnity, because its covered exposure is a federal asset base spread across ministries, states and municipalities.

Each generation of the grid has been redrawn against accumulated events rather than left static: smaller boxes, the depth condition, smoothed cell transitions, and storm payouts moved onto NHC B-deck data available about two weeks after the final advisory instead of the full report that pushed Otis's payout to April 2024. Trigger design is experience rating on a four-year update cycle.

What that cycle cannot correct is the denominator underneath it: if vendor earthquake models systematically understate tail hazard, as the supershear research suggests, the expected-loss figures anchoring parametric spreads inherit the understatement, and the 4.4x multiple is computed against a number that may be too small. A near-miss tests the map, not the model behind it.

Mexico's M7.3 Quake Missed Its Cat Bond Trigger by One Tenth of a Magnitude

Key Takeaways

How the 2024 Grid Reads an Earthquake

The Premium Ledger Behind the Near-Miss

Where the 0.1 Lands in Basis Risk Pricing

Further Reading

Sources