catastrophe risk and cat modeling
/ kuh-TAS-truh-fee; cat /
Most of insurance relies on bad luck being spread out: your car crash has nothing to do with your neighbour's, so a pool of thousands of drivers is steady and predictable. Catastrophes break that comfort. A single hurricane, earthquake, or wildfire can wreck tens of thousands of homes in one stroke, all at once, all from one event. That clustering — many losses from a single cause, hitting one region together — is catastrophe risk, and it is the nightmare scenario insurers most need to plan for.
Because catastrophes are rare and huge, ordinary 'look at last few years of claims' methods fail — you might have zero great earthquakes in your data and then one that bankrupts you. So the industry uses catastrophe models: large simulations that combine science and engineering rather than just history. A cat model generates thousands of hypothetical events (storms of various tracks and strengths), maps each onto the insurer's actual exposures (where the insured buildings are, how strongly they are built), estimates the damage, and produces a distribution of possible total losses. From that you read off numbers like the 1-in-100-year or 1-in-250-year loss, and the average annual loss used in pricing.
Cat modeling shapes pricing (a cat load added to premiums in exposed areas), reinsurance buying (how much cover to buy and at what attachment), and capital (how much surplus to hold against a mega-event). It is essential, but honesty matters: models are not crystal balls. They embed assumptions about climate, building codes, and rare physics that may be wrong; different vendors' models disagree, sometimes by a lot; and '1-in-100-year' is an annual probability of about 1 percent, not a promise it happens only once a century — you can suffer two such years back to back.
An insurer with 80,000 coastal homes runs a hurricane cat model. It reports an average annual loss of 40,000,000 (used in pricing) and a 1-in-100-year loss of 900,000,000 (used to decide how much reinsurance and capital to hold). No single year of past claims could have revealed that 900,000,000 tail.
Cat models simulate many hypothetical events on real exposures to estimate rare, region-wide losses.
A '1-in-100-year' loss is roughly a 1 percent chance each year, not a guarantee of one per century — two such years can land in a row. Cat models are indispensable but uncertain; different vendors disagree, and their assumptions (especially about a changing climate) can be wrong.