11.5 Non-Bayesian Model

This paper doesn’t go into much details about them, they are mostly presented to demonstrate their shortcomings and as a motivation to develop the Bayesian methods

11.5.1 Mack Model

See procedure from Mack-1994 on the log-normal CI

Recall the Chainladder assumptions

Incurred

Light on both tail \(\Rightarrow\) Does not have enough variability in it’s predicted distribution

Paid

Similar to ODP, biased high on personal auto and light left tail on WC

11.5.2 ODP Bootstrap

England & Verall ODP forecasts log incremental losses \(\Rightarrow\) Only suitable for paid losses

• Can handle occasional negative losses as long as the \(\sum\) column is positive

Same procedure as Shapland paper but without all the residual adjustments

Overall shows biased high