
162 Pricing in General Insurance
So how does experience and prior knowledge ‘add value’ to the data so that the resulting
model has more predictive power than the data itself? There is no unique recipe for this,
in the same way in which there is no algorithm for scientic discovery. However, here are
a few examples of how this happens in practice.
• Experience might tell us, for example, that a lognormal model is usually quite
good to model loss amounts, as it has proven its worth in many circumstances,
perhaps with large data sets.
• Theoretical results may tell us that, under certain conditions, large losses should
be modelled with a general ...