All methodology pages

Ratings and loss metrics

How the loss distribution becomes a probability, severity measures, and a Cork rating.

The rating starts with the estimated chance of permanent loss over one year and then accounts for how severe the loss can become.

Reading the loss distribution

All simulation results
Loss-event threshold count impairment events
Percentile boundary locate a distribution boundary
Average tail loss average outcomes beyond the boundary
The same simulation results determine the loss-event count, percentile boundaries, and the average loss in the tail beyond a selected boundary.

Probability used for the letterLink to this section: Probability used for the letter

The observed P(loss) is the share of simulation paths that finish above the 1% loss threshold.

observed P(loss) = loss-event paths / total simulation paths
Observed probability of loss The share of simulation paths that finish beyond the permanent-loss threshold.

The rating uses a conservative 95% upper estimate to allow for uncertainty from a finite number of simulation paths. The estimate remains above zero when a run observes no loss events.

The rating scaleLink to this section: The rating scale

RatingMeaningEstimated P(loss)
AAAExceptionally low riskLess than 0.02%
AAVery low risk0.02% to less than 0.2%
ALow risk0.2% to less than 0.5%
BBBModerate risk0.5% to less than 1%
BBElevated risk1% to less than 5%
BMaterial risk5% to less than 10%
CCCSubstantial risk10% to less than 30%
CSevere risk30% or more

Severity adjustmentsLink to this section: Severity adjustments

The first letter comes from P(loss). Two severity checks can then lower it by up to two categories.

The expected-loss budget for a rating is half of that rating's upper probability boundary. Reaching the budget lowers the letter once. An average loss above 25% among loss-event paths lowers it once more.

Loss measuresLink to this section: Loss measures

Expected lossLink to this section: Expected loss

Expected loss is the average across every simulation path. It combines the frequency and size of loss.

Conditional lossLink to this section: Conditional loss

Conditional loss is the average among paths that cross the loss threshold. It shows the typical depth after a loss event occurs.

Ninety-fifth percentile lossLink to this section: Ninety-fifth percentile loss

The 95th-percentile loss is the level that 95 percent of paths stay at or below.

Worst-five-percent average lossLink to this section: Worst-five-percent average loss

This is the average loss among the most severe five percent of paths.

Ninety-ninth percentile lossLink to this section: Ninety-ninth percentile loss

The 99th-percentile loss is the level that 99 percent of paths stay at or below.

Loss-adjusted yieldLink to this section: Loss-adjusted yield

Loss-adjusted yield subtracts the model's one year expected loss from a comparable one year net yield. It provides one common-unit view of return and estimated loss. The calculation is shown only when the yield and loss use compatible periods and units.

Rating flagsLink to this section: Rating flags

Flags identify an important assumption, limited cross-check, or sensitivity in the result. The letter remains visible when the model has a supported basis for the assessment.

Continue with Protection pricing.