Wednesday, May 9, 2012

Ends and Means of Catastrophe Modeling and Citizens Bonds



It was announced recently that Citizens Property Insurance Corp has sold $750 million in catastrophe bonds.  It is not only the largest sum of bonds that they have sold, but the largest to have ever been sold. The bonds will mature in 2014 which is also the deadline that the Citizen's Board of Governors has set for reducing Citizens' policies by about half.

AIR did the modeling for the bond.  They seem to do the modeling for many of the bonds being sold now- at least based on the list from the link above.  S&P graded the bonds at a B+, which is slightly higher than most other catastrophe bonds. To arrive at this rating, S&P had to answer the question "What is the risk?"  of more specifically, "What is the modeled risk?"

The raters chose to use AIR's catastrophe model that is "conditioned" on warmer than average sea surface temperature (SST) rather than the standard catalog because the conditioned model
 generated a more conservative (higher probability of attachment) result.

AIR conditioning methodology results in elevated estimates of risk 
On average, the increase in the mean frequency of the conditioned catalog over AIR’s standard catalog is approximately ten percent, although regional differences can be higher.


AIR highlighted their SST conditioning method in a report earlier this year.  The firm argued that their conditioning methodology is better than other firms' SST forecasting methodology because
There are currently severe limitations in forecasting sea surface temperatures (SSTs), and so-called “medium term” models based on a forecast of SSTs are beset by immense uncertainty.
AIR explained that their
WSST catalog is based on the standard catalog, but is modified based on the frequencies and intensities (by coastal region) of tropical cyclone activity during those years since 1900 when sea surface temperatures have been higher than the long-term average. It does not attempt to predict, or forecast, either SSTs or hurricane activity.
The company proposes that end users should choose models based on considerations of the long term average.
Ultimately, one critical benchmark by which a model should be judged is whether the loss estimates it produces are consistent with historical information, including trended industry and company reported losses. Models should not contain any inherent bias such that they consistently produce results above or below the trended historical loss data available.

If there is no unique scientific basis for choosing a model that deviates from the risk generated from the historical average, then any choice to deviate is a decision to value something beyond calculations of risk (eg. risk avoidance).

To further illustrate, consider that for the purposes of windstorm ratemaking in Florida, warm (or cool) SST is not used as a predictor of hurricane activity or losses.  Although, in principle at least, it could be.  

1 comment:

  1. Great Post! This is why reinsurers tend to use the AIR models. They make more money, not because the risk is greater, but because they check the box on the model that makes them more money.

    ReplyDelete

Note: Only a member of this blog may post a comment.