
226 Pricing in General Insurance
The results are shown in Figure 16.5, which demonstrates that if we rank the distri-
butions according to AIC, the lognormal comes up on top, as it should (third column) –
although not by much. This table also shows what the effect of punishing complexity is:
the second column shows the goodness-of-t as assessed by the likelihood of the data only
(the rst term of the AIC). As you can see, if you used that criterion only, the Burr would
come on top, but once you add the penalty for the number of parameters (3 × 2 = 6) the Burr
distribution falls into second place, behind the lognormal.
16.3 Modelling Large (‘