Catastrophe model validation

Catastrophe model validation is the documented work through which an insurer or reinsurer shows that the model it uses, bought from a vendor or built in-house, reasonably reproduces its own risk. The case that changed everything came in 2011. Version 11 of RMS's US hurricane model pushed modeled losses sharply higher for many portfolios, and boards found out that their capital requirement hinged on a vendor update. Solvency II gave the response a framework. Article 124 of Directive 2009/138/EC sets validation standards for any internal model, and Article 126 makes clear that using a third-party model or third-party data is no exemption. Lloyd's, which aggregates its syndicates' internal models, requires each managing agent to validate its cat models and justify its choices. In practice the model gets taken apart. The hazard module is checked against observations: landfall frequencies, measured winds, footprints of historical events. The vulnerability module is tested against the insurer's actual claims, by construction type and damage level. Major past events are replayed on today's portfolio, and the outputs are stress-tested on the assumptions that matter: exposure data quality, demand surge, secondary uncertainty. A serious validation often ends in adjustments, in other words in an own view of risk. That leaves the tail. A 1-in-200 PML (probable maximum loss) cannot be validated against forty years of claims: you check the building blocks, the consistency of the whole, and argue the rest.