11.2 Data Set

200 triangles from Sch P 1997 and reviewed 10 years later

• Paid and incurred

• 50 triangles from 4 LoB (Commercial Auto, Personal Auto, WC, Other Liab)

Potential pitfalls in sample triangle selections

1. Insurer with significant changes to their books over the exposure period would violate the assumptions of the model and should be excluded

   • Solution: Use consisteny of NEP and Net:Gross Premium to establish stability in book

   • Use CoV to establish consistency

   • Pick triangles with CoV below a threshold

2. Need to avoid selecting datasets that best suit the model e.g. removing “outliers” from the data

   • Solution: Choose the company in an automated and well defined manner

Losses are considered fully developed at 10 years so in practice the paid and incurred ultimate is slightly different