Catastrophe Model
Software that runs a large simulated catalog of catastrophe events to estimate portfolio losses. It gives probabilities, not a forecast of next year.
A catastrophe model is, in the NAIC’s words, “a computerized process that simulates thousands of plausible catastrophic events scenarios,” using meteorology, geology, and geography to estimate what those events would cost an insurance portfolio. The models were built for hurricanes and earthquakes, perils too rare for ordinary historical experience to price, and have since been extended to wildfire, severe convective storm, and non-weather risks including terrorism and cyber.
The NAIC’s own primer draws a line that matters when a model output shows up in a rate filing: catastrophe models “represent plausible event scenarios that could happen in the future,” and the simulated event catalog is meant to cover what is possible rather than to forecast what will happen. That distinction applies to frequency, location, and severity alike. A model that says a loss of a given size has a one-percent annual probability has not said anything about next year.
Structurally these are four connected modules. A hazard module generates events, a vulnerability module converts intensity at a location into damage, an exposure module holds what is actually insured there, and a financial module applies policy terms to produce metrics such as average annual loss and probable maximum loss. Almost all of it is licensed from a small number of vendors, which makes catastrophe modeling a third-party dependency as much as a technical one. Whether a given catastrophe model belongs inside an AI systems program is not a question the NAIC answers for you: the evaluation tool leaves risk classification to the insurance company, so a carrier that treats a physics-based simulation and a machine-learned roof-condition score as the same category should be able to say why. Our guides cover catastrophe modeling in reinsurance pricing and property pricing in homeowners.