Using Benchmarks with Sparse Data
If a reserving class does not contain enough data to allow for a credible projection, it may be appropriate to use a suitable benchmark instead. This may be done in two ways:
• Firstly, a benchmark development pattern may be employed if the data does not allow for a meaningful pattern of its own to be projected. For example, a benchmark liability pattern may indicate that claims are typically 50% developed after 2 years. This pattern can be used in a Development Factor Model (see elsewhere on this website).
• Secondly, a benchmark loss ratio will give an indication of the typical final loss level that might be expected. This may then be combined with the Development Factor Model (either data-driven or benchmark-driven) in a Bornhuetter-Ferguson model to project ultimate losses.
Some combination of these methods and a pure data method may also be used. For example, the benchmarks may simply be used as a check that the data is giving reasonable estimates.
Benchmarking - Introduction
Using Benchmarks with Credible Data
