Catastrophe model blending
A weighted combination of the outputs of several catastrophe models, to reduce reliance on a single vendor and absorb version changes.
Model blending combines the loss curves of several catastrophe models, vendor or in-house, through a weighting the insurer chooses and documents. Two methods coexist and do not give the same result. Loss blending takes the weighted average of amounts at the same return period; frequency blending takes the weighted average of exceedance probabilities at the same amount, which generally produces a thicker tail. The practice spread after version changes in major models, notably in 2011 for US hurricane, shifted the capital of many portfolios at a stroke. It has a limit that is easily forgotten: blending two models biased in the same direction does not produce a correct model, and a weighting chosen after seeing the results is only an elegant way of keeping the figure that suits.
Hypothetical case: at the 200-year return period, model A gives a 400 million euro loss and model B 250 million. An equal-weight loss blend retains 325 million. The choice of weights, rather than either model, then becomes the assumption that weighs most on capital, and it is the one the supervisor asks to have justified.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.