Imelda Powers, Global Chief Catastrophe Modeler
Model Suitability Analysis
There is a repertoire of tests, at both the macro and micro levels, to evaluate the merits of a CAT model. Regardless of resources, insurers should analyze how each test’s result informs their view of risk. Here we review the macro level.
Each model is built from multiple engineering sources and from insurers with different underwriting and loss adjusting practices. An “off-the-shelf” model is unlikely to be a perfect fit for any company unless its claims dominated the vendor’s product development methodology. Customization is almost always desirable, especially at the bottom of the loss curve, where there is more claims data. The following exercises strengthen the development of an internal model:
(1) Undertaking a comparison of modeled losses and observed claims from historical events; done carefully to ensure the modeled losses and the actual claims are stated on the same basis. Adjustments for loss inflation, exposure changes and non-modeled perils are needed to ensure an apples-to-apples comparison.
(a) For small and medium events, loss estimates can be checked against the insurer’s claim history.
(b) For extreme events, an insurer may not have sufficient claim experience, in which case they must turn to relevant industry data. In some cases it may be necessary to expand the analysis to include data from other countries, adjusting to reflect local conditions.
(2) It is helpful to check the reasonability of return periods for the insurer’s and the industry’s adjusted historical losses against both the portfolio and industry occurrence and aggregate exceedance-probability curves.
These exercises are challenging, but fundamental to building resilient models that provide insights users can leverage to customize a model for their own portfolio of risks.