Adapted for Insurance Nerds from "Viewpoint: When Andrew Returns – A Look Back at a Pivotal Insurance Event," Insurance Journal, August 25, 2026.
If you came into property insurance in the last decade, Hurricane Andrew is something you inherited rather than something you lived. It shows up as an event in a model's historical catalog, a line in the rate history, the reason the Florida Building Code exists. That is a thin version of it, and it leaves out the part that still matters to your book.
Start with the numbers. Andrew came ashore near Homestead on August 24, 1992 as a Category 5 Hurricane. It produced about $15.5 billion in insured losses in 1992 dollars and drove at least 11 insurer insolvencies, with Triple-I counting 16 failures across 1992 and 1993. Run the same storm on the same track into today's South Florida and the major modelers land between $70 and $100 billion insured. RMS and Verisk are near $100 billion. Karen Clark & Company is at $70 billion. Florida's population is up roughly 72% since 1992, and Miami-Dade, Broward, and Palm Beach now hold about $1.2 trillion in property.
The storm did not get bigger. What we built underneath the storm track did.
Here are some lessons that we need to learn from that storm.
Andrew crossed the coast about 20 miles south of downtown Miami. Shift the track 50 miles north into a direct Miami hit and modeled insured losses run past $200 billion. Nothing about the meteorology changes in that scenario. Only the position of the eyewall does.
The historical hurricane database is a sample of what has happened, not a limit on what can happen. Andrew was also primarily a wind event, while Katrina thirteen years later was dominated by surge and flooding. Same Saffir-Simpson vocabulary, different loss mechanics, different policy forms responding. Forward speed, radius of maximum winds, angle of approach, and rainfall all move the answer.
So when someone says worst case, find out whether they mean the worst thing observed or the worst thing plausible. Those are not the same thing, and the gap between them is where carriers can get hurt.
Andrew exposed how little insurers knew about what they were actually covering. Thirty-four years later I still see portfolios with stale values, missing secondary characteristics, placeholder business interruption limits, and buildings coded to the wrong construction or occupancy class. A model fed bad exposure will return a confident, precise, wrong answer, and it will not tell you it did. Garbage in...garbage out.
Andrew's losses were concentrated, and the tri-county corridor is denser and more correlated now than most portfolios recognize. Individually reasonable accounts sum into a peak that nobody underwrote on purpose. Reading the file(s) after the fact and no single underwriting decision looks wrong. Put them all together, and it can get pretty bad.
Pricing to the mean while the tail grows is precisely how carriers went insolvent in 1992. Secondary uncertainty, demand surge, and correlation deserve the attention that usually goes to the headline expected loss.
Model output becomes useful when it turns into underwriting appetite, portfolio limits, reinsurance structure, and a capital story a board can act on. That means catastrophe analytics has to live inside the underwriting workflow, before the risk is bound, not in a backward-looking review after the exposure is already on the books.
An anniversary is more useful as a stress test than as a memorial. Andrew is well documented and easy to describe. Run it through your current book. Then move the track 50 miles north and run it again. The second answer is the one worth the discussion.