Pricing in General Insurance

231Severity Modelling: Selecting and Calibrating a Severity Model

Parameter bias. When the data is sparse, not only is there a large error on the param-

eter, but there is also a strong bias on the parameters themselves: ξ is underesti-

mated, and σ (which is negatively correlated to ξ) is overestimated.

16.4 A Simple Strategy for Modelling Ground-Up Losses

In Section 16.2.1, we have given some ‘general rules for model selection’: in this section,

we present a specic method for modelling the severity of losses, which is consistent with

these rules and is adequate for losses from the ground up.

The method presented here, although somewhat general, ...