Hits: 5797
This paper describes the way uncertainty is treated in today’s catastrophe loss models and how it could be improved in the future to provide more credible estimations of risk.
The paper argues there are major deficiencies in the treatment of uncertainty in catastrophe loss models with inappropriate simplifications that mislead rather than inform. It explains how and why we need an evidential basis for models, an honesty about what we don’t know, and expression of this lack of knowledge in the rigorous use of probability.
After a brief introduction to uncertainties in catastrophe loss modelling, the paper will focus on just two of the many elements of uncertainty:
1. The representation of vulnerability.
2. The handling of probability in catastrophe loss models.
From this analysis three conclusions will emerge:
1. Vulnerability (or fragility) curves should allow discontinuities and expose the associated damage distributions, not just assume a parametric continuous distribution.
2. Uncertainty due to lack of knowledge (“epistemic” uncertainty) should be represented using probability distributions, not ignored.
3. The handling of probability in catastrophe loss models should be empirical, not based on the convenience of closed form parametric distributions or the false god of a presumed central tendency.