Probable Maximum Loss (PML) / Estimated Maximum Loss (EML)

Probable Maximum Loss (PML) is the estimate of the largest loss an insurer can sustain on a risk or a portfolio under realistic adverse conditions, short of the absolute worst case. Estimated Maximum Loss (EML) is the underwriter's version of the same idea, worked out site by site. That absolute worst case, with no regard to probability, is the Maximum Possible Loss (MPL). Two conventions coexist for measuring PML. Actuaries and modelers tie PML to a quantile of the annual loss distribution, and the 250-year loss is the 99.6th percentile. The scale comes from solvency regulation. Solvency II, Article 101: the SCR is calibrated to a 99.5% VaR over one year, a 1-in-200 return period. Cat models work on the same return-period scale. The 200-year PML is the usual quote, and cedants also look at the 1-in-100 and 1-in-250 losses. Underwriters tend to talk about EML instead, a deterministic figure built from an engineering scenario for a given site, with maximum realistic fire spread and minimum structural resistance, stated as a percentage of total insured value. The two get mixed up all the time. A probabilistic PML estimate is a portfolio quantile, an EML is a site scenario, and comparing two submissions means knowing which one you are reading, all the more since each market has its own assumptions about whether fire protection holds. Either way, PML is the sizing metric and the common language between cedants and reinsurers in treaty negotiations. It sets the reinsurance capacity bought, the attachment point and exhaustion point of the cat program, and the capital. Overstate it and the cedant buys cover that will never be hit. Understate it and a cat loss runs straight through the top of the program. On August 12, 1953, a welding torch spark set fire to General Motors' Hydra-Matic plant in Livonia; the fire took the whole building and cost at least 50 million dollars. That is the maximum possible loss made real, the one the PML sets aside by assuming the protections hold. The notion travels badly to cyber, where the maximum loss depends on digital dependencies that shift from one quarter to the next.