Fifteen catastrophe events analysed in depth, what happened physically, why the losses were what they were, what cat models got wrong, and what changed as a result. The real-world complement to the Intermediate course.
Blind faults beneath a city no one knew were there, a Mw 6.2 aftershock triggered the most extensive liquefaction in a modern urban area and cost 20% of New Zealand's GDP.
→ Liquefaction as the dominant damage mechanism, the gap between ground shaking models and soil failure realityA Mw 9.1 megathrust event generated a tsunami that overtopped Japan's seawall system, triggered the Fukushima nuclear accident, and exposed supply chain CBI as a major unmodelled loss driver.
→ Maximum magnitude underestimation, and the emergence of supply chain CBI as an unmodelled loss categoryTwo earthquakes nine hours apart on the East Anatolian Fault killed 53,000 people. Modern building codes existed, enforcement did not. The direct reference for Marmara Fault risk in the Intermediate course.
→ Code existence vs code compliance, and BPT vs Poisson frequency debate for Turkish seismic hazardThe storm that created catastrophe modelling as a profession. Andrew's losses far exceeded everything insurers had reserved for, 11 companies went insolvent and AIR was founded in the weeks that followed.
→ The founding event of the cat modelling industry, why deterministic scenarios gave way to probabilistic modelsNot primarily a wind event, a catastrophic engineering failure. Katrina's storm surge overwhelmed a levee system built to a lower standard than publicly stated, killing 1,833 people and exposing three fundamental cat model failures.
→ Levee reliability assumed not modelled, infrastructure dependency as a non-modelled loss amplifierA hybrid meteorological event, neither tropical nor extratropical at landfall, that made an unprecedented hard left turn into New Jersey and overwhelmed New York City with a record 13.88-foot storm surge, exposing every assumption the industry had made about hurricane tracks, surge modelling, and flood coverage.
→ Unprecedented track, hybrid storm structure, surge geometry, why Saffir-Simpson category is an inadequate loss predictorThree major Atlantic hurricanes in four weeks, USD 92 billion in combined insured losses, 2017 the costliest hurricane season in history, and three entirely different model failures across three events.
→ Three storms, three model failures: Harvey's inland flood gap, Irma's track sensitivity, Maria's vulnerability underestimationThe fastest-intensifying hurricane ever recorded in the eastern North Pacific struck a city of one million with no warning time, models gave 24 hours' notice of a tropical storm; it made landfall as a Category 5.
→ Rapid intensification in an emerging market, the limits of hurricane modelling outside the Atlantic basinA quasi-stationary low stalled over Western Europe and delivered unprecedented rainfall into small, steep-sided catchments. The Ahr Valley became a catastrophic torrent, exposing the largest flood protection gap in a wealthy European economy.
→ Small catchment flash flood, the gap between large-river hydraulic models and the events that actually kill peopleA displaced polar vortex froze every fuel type in ERCOT's generation fleet simultaneously, cascading through the natural gas supply chain and bringing Texas to within four minutes of a complete grid collapse.
→ Demand-surge coupled with supply-collapse, a loss type standard winter storm models were never built to captureTemperatures climate scientists called "virtually impossible" killed more than 1,400 people and produced near-zero insured losses, because no insurance product for heat as a standalone catastrophe peril exists.
→ Heat as a non-insured catastrophe, the largest unmodelled, uninsured, unpriced peril in the climate risk landscapeA rare jet-forced Santa Ana wind event drove simultaneous fires through the densest, most expensive urban interface in American wildfire history, destroying more than 16,000 structures across Pacific Palisades, Eaton, and surrounding communities.
→ Urban conflagration not forest fire, wildfire models calibrated to vegetation, not how fire moves through dense residential blocksThree consecutive years of more than USD 45 billion in annual insured losses from hail, tornadoes, and straight-line winds, eclipsing hurricanes as the dominant US catastrophe loss driver and forcing a fundamental reclassification of SCS from secondary to primary peril.
→ Secondary peril becomes primary, why SCS models lagged a decade behind the risk they were supposed to measureA Russian military cyberweapon disguised as ransomware escaped its intended Ukrainian target and propagated across global corporate networks, destroying data at Maersk, Merck, and FedEx with no insurance coverage for most victims.
→ Silent cyber, unpriced, unmodelled cyber exposure sitting inside property books that insurers did not know they had writtenA single defective configuration file pushed to 8.5 million Windows endpoints without staged testing crashed airlines, hospitals, and banks, 78 minutes of deployment, days of recovery, and a coverage gap that exposed how far cyber insurance lags the actual risk.
→ Technology systemic risk, a software update with no malicious intent produced a larger insured loss than most natural catastrophes* Inflation-adjusted to 2025 dollars