Tuesday, December 17, 2013

Florida's Hurricane Risk is Overdetermined

The above images shows Wordles or word clouds of the first authors listed in the Meteorological references section of individual modeling groups submissions to the FCHLPM under 2011 standards.  You can find the same sort of word clouds for 2009 standards here.

Unique mixtures of science result in different hypothetical storms resulting in different hypothetical interactions with society.  

The above graph shows the differences in the radius of maximum winds between modeling groups.  The radius of maximum winds (denoted as Rmax) is a measure of the distance from the center of a hurricane to its maximum winds.  The mathematical derivation of Rmax depends upon central pressure.  Central pressure is calculated differently by company and judged reasonable by  respective company... often by the same statistical metrics.  

The graph depicts the range between a company's minimum Rmax and maximum Rmax for a given central pressure (CP). The difference is indexed to the average of the ranges for all companies for a given central pressure.  So, for example, at a central pressure of 920 mb, storms created by EQECAT have an Rmax anywhere from 4 miles to 62 miles (see data below), a range of 58 miles.  The average range for all the models at 920 mb is 35.42.  So, EQECAT's Rmax range is 60% greater than the average range at 920mb of all companies.  
Rmax values submitted to FCHLPM (All distances in Miles)
No matter.  The FCHLPM considers all of these accurate and reliable.  So too, the output.  

The table above shows the uncertainty interval of each company's 100 PML and 250 PML.  Given the exposure data, EQECAT estimates a 1% annual chance of a loss of within the range of $35B and $104B.  They have a 95% statistical confidence about this.  

Yet under 2009 standards and the same data set they had 95% confidence that the loss would fall between $21B and $122B. You can see 2009 standards data in the below table (I make no adjustments for inflation).  You can see this type of variation for any of the companies- I'm just using EQECAT as an example.

All of the data here is considered accurate, reliable and my favorite "reasonable."  Consider further that outside of Florida, insurers and others use totally different numbers and catalogs that they find to be accurate, reliable and reasonable. 

So my point here is that the risk is overdetermined.  This means that there are more scientific reasons and acceptable estimates than observations needed for determining the scientific quality of risk estimates.  

Given the substantial variation in accurate estimates of risk, the right or best estimate is determined by external factors such as, market and political conditions. Determining hurricane risk is, as this analysis shows, well suited for negotiation.