2017 Puebla Earthquake: USD 2,300M in insured losses that could have been modeled better
How modern probabilistic analysis would have transformed seismic risk management for insurers in Mexico — and what we can learn 9 years later.
April 10, 2026
8 min readDynamis Associates × Appgile
On September 19, 2017, a magnitude Mw 7.1 earthquake struck central Mexico. With its epicenter 12 km southeast of Axochiapan, Puebla, at a depth of 57 km, the quake was felt violently in Mexico City — 120 km from the epicenter — causing the collapse of 44 buildings and damage to more than 180,000 homes.
Beyond the human tragedy (369 fatalities), the earthquake exposed critical failures in how the Mexican insurance sector modeled seismic risk. Insured losses far exceeded prior estimates, triggering a solvency crisis at several companies and a hardening of the reinsurance market that lasted years.
Seismic rupture at 57 km depth. Intraslab earthquake within the subducted Cocos plate. Normal faulting mechanism.
13:14:40 – 13:15:00
No early warning. Unlike coastal earthquakes, the intraslab origin prevented the SASMEX system from providing advance warning to Mexico City.
13:15 – 13:45
44 structures collapsed in Mexico City. Buildings of 5-8 stories made of unreinforced masonry and pre-code concrete (built before 1985) accounted for 78% of all collapses.
Weeks 1 – 4
Claims avalanche. Mexican insurers received more than 52,000 claims. The CNSF (National Insurance Commission) activated emergency protocols.
Q4 2017 – 2018
Reinsurance crisis. Swiss Re reported USD 680M in claims. Catastrophe reinsurance premiums for Mexico rose 15-30%.
Why losses exceeded estimates
Insurance portfolios in Mexico in 2017 relied on risk models with three fundamental deficiencies that this earthquake brutally exposed:
Issue 1
Incomplete seismic catalog
Models based only on post-1985 data. They ignored pre-instrumental historical seismicity and deep intraslab sources.
Issue 2
Generic fragility curves
Building typologies poorly classified. Pre-1985 buildings treated the same as post-1985 despite radically different building codes.
Issue 3
Underestimated site effects
The lakebed soil of Mexico City (Vs30 < 180 m/s) amplified seismic waves by 2x to 5x. Many models used a single generic Vs30 for the entire city.
Seismic amplification by soil type in Mexico City
PGA amplification factor relative to bedrock (Vs30 = 760 m/s)
Sources: Singh et al. (2018), Mexico National Seismological Service. Representative values for fundamental period ~2s.
What modern PSHA would have revealed
A full probabilistic analysis — integrating a historical seismic catalog, 3D fault geometry, regionally calibrated GMPEs, and site-specific Vs30 — would have produced risk indicators significantly different from those insurers were using in 2017.
EP Curves: Traditional model vs. Full PSHA
Exceedance Probability for a C1-L building (concrete, 5 stories, pre-code) in Mexico City Lakebed Zone. Replacement value: USD 5M
Illustrative values based on actual parameters from the 2017 earthquake and HAZUS-MH fragility curves for C1-L pre-code typology.
Key indicators: Traditional model vs. Full PSHA
For a C1-L building (reinforced concrete, low-rise, pre-code design) located in the Mexico City Lakebed Zone, with a replacement value of USD 5,000,000:
Indicator
Traditional Model (2017)
Full PSHA
Difference
PGA 475yr
0.18g
0.32g
+78%
PML 475yr
USD 1,400,000
USD 2,280,000
+63%
PML 2475yr
USD 2,100,000
USD 3,850,000
+83%
AAL
USD 18,200
USD 34,500
+90%
AAL Ratio
0.36%
0.69%
+92%
Loss Ratio 475yr
28%
45.6%
+63%
Direct implication for insurers
Under the traditional model, the technical premium for this asset was calculated at approximately USD 18,200/year. A full PSHA would have indicated a technical premium of USD 34,500/year — a 47% premium shortfall that, multiplied across a portfolio of hundreds of assets, explains the magnitude of unanticipated losses.
Expected damage distribution
HAZUS-MH fragility curves applied to the actual PGA recorded in the Lakebed Zone (~0.30g) show that a C1-L pre-code building had a 28% probability of extensive damage or collapse — a risk that traditional models underestimated by failing to incorporate site amplification.
Damage probability distribution | PGA = 0.30g
C1-L building type (concrete, low-rise, pre-code), Mexico City Lakebed Zone
12% — No damage
25% — Slight damage (cosmetic cracks)
35% — Moderate damage (repairable)
18% — Extensive damage (irreparable)
10% — Complete collapse
Insurance market impact
Consequences for the Mexican insurance sector (2017-2019)
Key indicator evolution post-earthquake
Sources: CNSF Annual Report 2017-2018, Swiss Re Sigma, AM Best Mexico Insurance Report 2018.
The 2017 earthquake triggered a cascading effect across the market:
Catastrophe premiums rose 35% on average for seismic risk in Mexico, with peaks of up to 60% in Mexico City's lakebed zone.
The sector's combined ratio reached 112% — for every peso collected, insurers paid out $1.12. Several companies entered technical losses.
International reinsurers (Swiss Re, Munich Re, Hannover Re) raised premiums by 25% and tightened contract terms.
The CNSF mandated a 42% increase in technical reserves for seismic risk, compressing margins even further.
“The problem was not the earthquake. The problem was that our risk models said an event like this would generate losses of USD 1,200M — and the reality was double that.”
— Chief Risk Officer, international reinsurer (2018)
How Xpectral addresses these deficiencies
Our seismic intelligence platform is specifically designed to solve the three problems the 2017 earthquake exposed:
Complete global catalog
USGS + ISC + GEM from 1900
We automatically integrate the USGS FDSN catalog, the ISC Bulletin (historical events), and GEM's active fault database (GAF-DB). Magnitude homogenization to Mw via ObsPy.
Site-specific Vs30
1 km² resolution
USGS Vs30 data at 1 km² resolution (not regional averages). For Mexico City's lakebed zone, our model uses Vs30 of 90-180 m/s vs. the generic 360 m/s value used by many models in 2017.
Calibrated HAZUS fragility
36 typologies × 4 levels
HAZUS-MH fragility curves with 36 building typologies and 4 seismic design levels. We properly differentiate between pre-code and post-code buildings.
Insurance-grade metrics
PML, AAL, EP, SCR
Direct output in the metrics the industry needs: PML at multiple return periods, AAL, EP curves (OEP/AEP), damage distribution, and inputs for Solvency II SCR calculation.
The lesson: proper modeling is cheaper than paying for not doing it
The 2017 Puebla earthquake was not an unprecedented event. Mexico has experienced destructive earthquakes repeatedly — 1985 (M8.0), 1999 (M7.0 Oaxaca), 2017 (M7.1), 2022 (M7.6 Michoacan). Intraslab seismicity in central Mexico is a well-documented scientific risk.
What failed was not the science. What failed was the transfer of scientific knowledge to the insurance sector. The models available in 2017 were costly black boxes, slow to update, and did not incorporate advances in open data (USGS, GEM) or computational power.
Today, with Xpectral, an insurer can obtain a full PSHA analysis in hours, not months. With up-to-date data, calibrated fragility, and metrics directly integrable into solvency models.
By the numbers
Had Mexican insurers used a full PSHA model with site-specific Vs30 in 2017, technical premiums would have been 47-90% higher for lakebed zone assets — generating adequate reserves to absorb the loss. The cost of PSHA analysis per asset is a fraction of the 0.1% premium shortfall that materialized.
Disclaimer: The numerical values presented in the comparisons are illustrative, calculated with real event parameters but on a hypothetical asset. Each building requires an individual analysis with its specific coordinates, typology, and site conditions.
Frequently asked questions
Key data for insurance professionals and catastrophe risk modelers
Market What were the insured losses from the 2017 Mexico earthquake?
Insured losses from the 2017 Puebla earthquake (M7.1) reached USD 2,300 million, while total economic losses were estimated at USD 4,700 million. The insurance penetration ratio was 48.9%, significantly higher than the Latin American average.
Technical What PGA was recorded in Mexico City during the 2017 earthquake?
In the lakebed zone of Mexico City (Vs30 < 180 m/s), the recorded PGA reached approximately 0.30g, with site amplification factors between 3x and 5x relative to bedrock. This value far exceeded predictions from traditional models that used a generic Vs30 of 360 m/s.
Market How did the 2017 Puebla earthquake affect the reinsurance market?
Catastrophe reinsurance premiums for Mexico rose 15-30% post-event. Swiss Re reported USD 680M in claims. The Mexican insurance sector's combined ratio reached 112%, and the CNSF mandated a 42% increase in technical reserves for seismic risk.
Cat Modeling What is the 475-year PML for a pre-code concrete building in Mexico City's lakebed zone?
For a C1-L building (reinforced concrete, low-rise, pre-code) in the lakebed zone of Mexico City with a replacement value of USD 5M, the 475-year PML calculated with full PSHA is USD 2,280,000 (loss ratio of 45.6%). Traditional 2017 models estimated only USD 1,400,000, a 63% shortfall.
Cat Modeling What is AAL (Average Annual Loss) and what was it for Mexico in 2017?
AAL (Average Annual Loss) is the expected average annual loss, calculated by convolving the seismic hazard curve with fragility curves. For a C1-L pre-code building in Mexico City's lakebed zone, the AAL with full PSHA is USD 34,500/year (ratio 0.69%), 90% higher than the AAL of USD 18,200 estimated by traditional models.
Pricing Why did seismic risk models underestimate losses from the 2017 earthquake?
Three main factors: (1) incomplete seismic catalog that ignored pre-instrumental intraslab seismicity, (2) generic fragility curves that did not differentiate between pre-1985 and post-1985 buildings, and (3) underestimated site effects from using an average Vs30 instead of site-specific values for Mexico City's lakebed zone (actual Vs30 < 180 m/s vs. generic 360 m/s).
Technical What percentage of buildings collapsed in Mexico City's lakebed zone in 2017?
78% of the 44 buildings that collapsed in Mexico City were located in the lakebed zone, where soft soil (Vs30 < 180 m/s) amplified seismic waves by 3x to 5x. The most affected buildings were 5-8 stories tall, built with unreinforced masonry and pre-code concrete (prior to 1985).
Pricing What is the difference between a traditional model and full PSHA for seismic risk assessment?
Full PSHA integrates historical seismic catalogs from 1900 (USGS + ISC), 3D fault geometry (GEM GAF-DB), regionally calibrated GMPEs, and site-specific Vs30 at 1 km resolution. For the 2017 Mexico earthquake, this difference meant traditional models underestimated PGA by 78% and PML by 63-83% depending on return period.
Pricing How much would it have cost to properly model seismic risk in Mexico before 2017?
The cost of a full PSHA analysis per asset represents less than 0.1% of the premium shortfall that materialized. Had insurers used PSHA with site-specific Vs30, technical premiums would have been 47-90% higher for lakebed zone assets, generating adequate reserves to absorb the loss without a solvency crisis.
Cat Modeling Which fragility curves should be used for concrete buildings in Mexico?
For concrete buildings in Mexico, HAZUS-MH fragility curves with the correct typology classification (C1, C2, C3) and seismic design level (high/moderate/low/pre-code) are recommended. For a C1-L pre-code building subjected to PGA of 0.30g, the combined probability of extensive damage or collapse is 28%, based on calibrated HAZUS fragility parameters.