How does catastrophe (CAT) modelling work in commercial property insurance?

CAT modelling is the actuarial engine that lets commercial property carriers price catastrophic risk and set their treaty programmes. Without it, no carrier could underwrite Florida coastal, California wildfire, or Texas convective-storm exposure.

The four modules of a CAT model

1. Stochastic event catalogue. A simulated set of 10,000–100,000 plausible event years, statistically representative of the long-run frequency / severity distribution. For US hurricane: anywhere from 0 to 8 named storm landfalls per year, with tracks, intensities, and surge characteristics drawn from NOAA HURDAT2 history.

2. Hazard module. For each simulated event, what physical intensity reaches each location? For hurricanes: peak gust wind speed, surge height, surge inland penetration. For earthquakes: peak ground acceleration, spectral acceleration, liquefaction susceptibility. For wildfires: fire spread, ember exposure, burn duration.

3. Vulnerability module. Given the physical hazard at a location, what fraction of the building / contents / business-interruption insured value is damaged? Damage curves are conditioned on construction type (wood frame, steel frame, reinforced concrete), year built, code adoption, occupancy, sprinkler protection, roof age, distance to coast.

4. Financial module. Apply policy deductibles, sublimits, limits, treaty cessions, and the carrier's reinsurance programme. The output is net retained loss per event and per year.

The two main outputs

OEP (Occurrence Exceedance Probability) — the probability that a single event causes a loss above a given threshold in a year. "OEP 1-in-100" means there's a 1% annual probability of one event causing this much loss.

AEP (Aggregate Exceedance Probability) — the probability that the sum of all events in a year exceeds a threshold. Always higher than OEP at the same return period because multiple events can pile up.

Standard return periods used in carrier capital models: 1-in-100, 1-in-250, 1-in-500.

The three vendor models

• Moody's RMS (Risk Management Solutions). Largest market share, especially in commercial property.
• Verisk AIR (Atmospheric and Environmental Research). Strong in personal lines and severe convective storms.
• Karen Clark & Company (KCC). The Clark family historically pioneered the field. Strong in independent validation work.

Most carriers run two models and reconcile. Material disagreement between RMS and AIR on Florida AEP 1-in-250 is a known industry pain point.

What CAT modelling does NOT capture

• Tail events outside the catalogue. A Cat 6 hurricane (winds > 200 mph) isn't in the standard catalogues — but it's no longer impossible.
• Demand surge. Post-event, construction labour and materials prices spike. Most vendor models capture this in aggregate but not at the policy level.
• Litigation amplification. AOB litigation in Florida added 30%+ to ultimate losses on Hurricane Irma. Vendor models pick this up only after data accumulates.
• Correlated cyber + property. A cyber attack during a CAT event isn't modelled jointly.

How AI underwriting platforms use CAT modelling

Modern AI platforms don't replace CAT models — they integrate them at the right step: - The flood / CAT agent scores each individual submission against the carrier's stored CAT footprint - The portfolio agent runs the pre-bind concentration check that maps to the carrier's OEP / AEP thresholds - During a live CAT event, the platform re-scores every active bound risk against the actual storm track (using NOAA HURDAT2 live feed) to produce a 18-minute net-retained-loss estimate

Reference sources

• Verisk AIR
• Moody's RMS
• Karen Clark & Company
• NOAA HURDAT2