EVENTSET is the first globally consistent insurance loss model. Magnitude 6 to 9 events are drawn using machine learning from the Global Earthquake Activity Rate (GEAR) model, supplemented by the highest quality national seismic catalogs available. We make no assumptions about characteristic earthquakes; we assign no maximum magnitudes, fault lengths, or fault slip rates; and we impose no area sources. Shaking is propagated from each earthquake source and is corrected for site amplification at each building location. To transform shaking into mean damage ratios, our machine learning model furnishes robust loss ratio curves from the USGS PAGER ExpoCat of 5,000 earthquakes (which, crucially, includes incidences of zero loss).


Comparison of M≥6 earthquakes for 50 random years from EVENTSET (top) with the USGS catalog of M≥6 shocks for the past 50 years (bottom). Without imposing any rules on maximum magnitude or source type, the two distributions are remarkably similar. Here, the portfolio is 10,000 globally distributed buildings. Much larger portfolios, and longer time spans, can be generated.


EVENTSET2FMKat is compatible with the KatRisk® sampling and financial module, FMKat. When processed in FMKat, all standard EP curves, year loss tables, and summary statistics are generated in either csv or binary files. FMKat computes all insurance contracts (ground up, gross, net, and reinsurance losses at site, policy, account, and portfolio level) with transparent rules based on sampled location coverage losses by event. Thus, EVENTSET2FMKat is compatible with your existing workflow, without any customization needed: It’s turnkey.


Upper panel: Losses reflect proximity to sites of large or frequent earthquakes. Lower panel: The ‘cliffs’ in the exceedance curves at $1000 occur when damage is suffered in a second earthquake in one year. Ground up is total loss; Gross Loss is after deductibles and limits are removed. OEP is loss exceeded by any event in any given year; AEP is the aggregated losses in any given year.


Losses determined from EVENTSET compared to hazard in PUSH. Losses reflect the shaking hazard.