A colleague in the reinsurance industry emailed me this morning to ask my thoughts on the return period for the magnitude of losses from US tornadoes in 2011, based on a preview of our tornado normalization work found in this just-published paper:
Simmons, K., D. Sutter, R.A. Pielke, Jr. (2012), Blown away: monetary and human impacts of the 2011 U.S. tornadoes. Extreme events and insurance: 2011 annus horribilis (Edited by C. Courbage and W.R. Stahel) The Geneva Reports: Risk and Insurance Research , Published March 2012.
Here is how I responded:
On your question, a few thoughts.
1. When using our dataset, it is best to use the damage numbers as tabulated by the US NWS as they are consistent over time
2. That said, 2011 damage is qualitatively indistinguishable from 1974 and 1953 at >;$20B
3. That would give a simple baseline expectation of 1 in 20 for 2011, but half or twice that would not be implausible given the uncertainties, so between 1 in 10 and 1 in 40
4. For 2012 and looking ahead there are two big question marks, one more certain than the other. Urbanization is increasing, which means that the chance of large losses increases (somewhat at the expense of smaller and medium losses of course). And there has been a notable and significant decline in the incidence of strong tornadoes in recent decades
5. So, depending upon how you view projections in development and tornado incidence, I'd venture that a somewhat clever but justifiable statistical approach could half and double the estimated return period once again.So that leaves use with somewhere between 1 in 5 and 1 in 80
In short, as first best guess -- 1 in 20 based on a short 60 years of record with 1 in 10 and 1 in 40 unlikely to be significantly different, and with 1 in 5 and 1 in 80 as plausible and defensible. Anyone professing greater certainty is likely making some key assumptions that make uncertainties go away ;-)
I'd welcome any comments.
