
321Experience Rating for Non-Proportional Reinsurance
20.4.2 Selecting and Calibrating a Severity Model
We are interested only in losses above a certain threshold, and therefore we normally use
‘threshold’ distributions only. There is no formal denition of a threshold distribution,
but the idea is that a threshold distribution assumes that only losses above a certain value
are possible, and that the probability density is decreasing (which makes intuitive sense).
Examples of threshold distributions are the exponential, Pareto (single- or two-parameter),
and GPD. Occasionally, a shifted lognormal distribution, that is, a distribution such t ...