November 29th, 2012

Taking Control of Quantifying Your Natural Catastrophe Risk: Part II

Posted at 1:00 AM ET

Elizabeth Cleary, Managing Director, Valerie Kloepfer, Managing Director, Imelda Powers, Ph.D., Global Chief Cat Modeler, Sherry Thomas, Head of Catastrophe Management - Americas and James Waller, Ph.D., Research Meteorologist
Contact

Why Do Results for North Carolina Differ So Dramatically Between ALPHA and GAMMA?

  • Overall frequency is similarly modeled for North Carolina in both ALPHA and GAMMA. However, GAMMA has slightly higher overall modeled frequency (historical view), while ALPHA has a measurably higher modeled frequency of Cat 3-5 storms for North Carolina, and therefore fewer lower severity Cat 1-2 storms versus GAMMA.
  • In the ALPHA model, North Carolina storms generally have larger footprints, pushing the storms further north and west, even producing losses in areas where GAMMA does not generate any loss. For a state like North Carolina, where there are high exposure values inland, this is generally a key driver for larger ALPHA losses in the state. Factors that could reverse the aforementioned ALPHA and GAMMA comparison are usually not strong enough to reverse the general statewide observations. Such factors include a larger GAMMA damage ratio especially at higher wind speeds (impacting coastal counties) and larger wind deductible impact modeled in ALPHA.
  • Inland portfolios modeled in ALPHA are impacted by ALPHA’s slower wind decay for storms making landfall in North Carolina, typically resulting in larger losses for those portfolios versus GAMMA. The example in Figure 2  from Hurricane Isabel shows losses going all the way up to northwest Pennsylvania, which seems to be consistent with the reported impacts of the storm. Historical storm loss footprints for Hugo (1989), Fran (1996) and Floyd (1999) show similar patterns, with GAMMA losses typically truncated geographically more so than observed for these actual events.

Figure 2

takingf2

Isabel made landfall on September 18, 2003, and was downgraded to a tropical storm near the North Carolina/Virginia border, before crossing Pennsylvania on September 19. Storm force gusts were later reported in Buffalo, New York and well into southern Ontario, Canada. This was a large storm, with impacts well away from the center of circulation.

What Is Guy Carpenter’s Current View Toward U.S. Hurricane Landfall Frequency?

In 2011, Guy Carpenter suggested that it was time to revisit the use of Atlantic Multi-decal Oscillation (AMO)-influenced rates in catastrophe models. AMO-influenced rates reflect anticipated frequency of U.S. hurricane landfall relative to long term (historical) averages; however, experience is pointing towards AMO-influenced rates being unpredictable.

At this time, our view is that there is a more compelling reason to use the long term historical rate of U.S. landfall frequency. While the heightened rate of modeled hurricane frequency was reasonably accepted in 2006 and for several years afterward, the passage of time now causes us some pause. In fact, when 1995 to 2012 is viewed as a whole, the overwhelming majority of those years are very simply “average” or even “less than average” for U.S. landfall frequency. This throws great doubt to the direct correlation between Atlantic Basin activity to the number of U.S. landfalling events. The distinction between hurricane generation frequency and landfalling U.S. hurricanes becomes even sharper when you consider that Hurricane Andrew (1992) made landfall during a season of record low hurricane activity in the Atlantic basin.

Figure 3

takingf3

Does Data Quality Remain An Important Factor in Catastrophe Modeling?

As data collection becomes more granular and sophisticated, interesting differences regarding data usage within each model can be observed. For example, number of stories has no impact on single-family dwelling losses in ALPHA or BETA, while GAMMA distinguishes them. Knowing what truly affects losses (and what doesn’t) helps guide clients to focus not just on collecting more data, but on collecting meaningful data. While many users of cat models strive for completeness in cat modeling data, understanding how that data can impact losses helps insurers make better choices about investments in data capture and reporting.

Summary

Each model has its strengths and weaknesses. While many of these strengths and weaknesses have a direct effect on market share or acceptability of the model, a particular strength is sometimes outweighed by a weakness or vice versa. For example, the ease of using GAMMA, both in terms of user experience and the detailed and open nature of the background databases, historically led to high market penetration despite weaknesses in some components of the model. Meanwhile, the difficulty in using BETA has traditionally led to low market penetration, outweighing many strengths of that model’s science. It is only through understanding the properties of each model that we can recommend one fitting a portfolio better than another, or if a combination approach is superior. As a reinsurance broker and a proponent of (re)insurers using catastrophe models to help quantify their risk potential, it is our mission to help (re)insurers use models to their fullest potential and to advocate for better model performance for that purpose.

If clients would like to discuss issues specific to their portfolios, including options for analysis settings and model blending, and to gain insights and advice for their approach to utilizing catastrophe models, please contact a GC Analytics® representative.

Click here to read Part I of “Taking Control of Quantifying Your Natural Catastrophe Risk” >>

Click here to register to receive e-mail updates >>

AddThis Feed Button
Bookmark and Share


Related Posts