All the major catastrophe modeling companies have reset their view of U.S./North American hurricane risk in the past year. Revisions to storm characteristics have been the key component of change for AIR, EQECAT and RMS. Many users have been frustrated by the lack of regard for risk management implications as some vendors packaged several significant U.S. hurricane changes into their releases. In subsequent GC Capital Ideas posts we will provide an overview of the significant changes to U.S./North Atlantic models over the last 12 months and give our view on each update.
AIR CLASIC/2 v12
AIR’s v12 was the first of the new hurricane models, released in the second half of 2010. The updated model introduced a new basin-wide stochastic catalog that included the United States, Caribbean and Mexico. It also expanded its historical catalog to include the 2008 storms of Ike, Gustav and Dolly, while the three states of Missouri, Illinois and Indiana were added to allow for the extension of hurricane tracks further inland. Several changes were made to the model’s vulnerability module, and the year of construction was used in vulnerability calculations for all hurricane states. There were marked changes in personal lines vulnerability (almost uniformly upward) for structures built prior to 1995 (except in the Florida counties of Broward and Miami-Dade). Vulnerability functions were updated for various construction types, particularly commercial structures of reinforced concrete, steel and reinforced masonry.
AIR CLASIC/2 v13
AIR only made minor updates to its U.S. hurricane product in 2011, but its tropical cyclone hazard and vulnerability model for the Caribbean underwent more significant change. This, the first Caribbean update since 2003, added 17 new countries and enabled users to model wind and precipitation-induced flood separately. The changes resulted in an estimated 20 percent to 50 percent increase in insurable losses across the new countries.
Guy Carpenter Perspective on AIR Hurricane Model Update
The release of AIR v12 was expected to result in modest loss decreases in personal lines, while increasing commercial lines in lower return periods and average annual losses (AAL). However, changes in loss results were not uniform across portfolios and were often seen to move in directions unexpected by the AIR guidance. For residential classes in Florida and Texas, for example, AIR guidance indicated that double-digit decreases in AAL would be expected. However, based on actual client data, there were very few instances of pure model change decreases for residential lines in either state.
The model change impact of v12 for commercial portfolios was more consistent with AIR guidance, in that the direction of loss estimates (almost always upward) did prove to be correct. But, client portfolios saw increases that were considerably higher than the AIR guidance had suggested. Such inconsistencies suggested that AIR’s industry database was not adequately representative to anticipate how the model update would impact insurer portfolios. When AIR’s industry exposure data was updated from v11 to v12, the assumption that the newer building stock in v12 would result in some industry loss decreases did not materialize for the majority of insurers.
RMS RISKLINK V11
RMS released their much anticipated North American hurricane model in February 2011. This new model takes into account observed inland losses from Hurricane Ike (2008) that were not available previously. Other significant changes included:
• Decreased filling rates at which storms dissipate after landfall
• Reset of frequency rates by region and storm category
• Increased vulnerabilities at lower wind speeds
• Increased vulnerabilities of commercial risks
• Decreased impact of secondary modifiers
• Updated regional building practice considerations
• Decreased hazards for some coastal areas
• Remodeled storm surge with revised hazard and take-up rate assumptions for the National Flood Insurance Program
Guy Carpenter Perspective on RMS Hurricane Model Update
Since 2006, RMS has recommended their medium term rate (MTR) view of risk for near-term risk management. However, recent Guy Carpenter research and sensitivity testing calls this recommendation into question. Our advice to clients is to review each component of model change and make educated choices regarding best settings for their portfolios rather than a wholesale adoption of vendor recommended best practices. Rating agencies are careful not to require a specific view of risk when modeling but expect all companies to provide supporting documentation validating their choice of model settings.
Losses in the new model are generally much higher than expected for most users, based on the RMS guidance. The RMS changes also develop losses that are generally higher than AIR v12. Guy Carpenter believes there are a number of factors leading to a disconnect between actual insurer portfolios and the RMS industry loss summaries provided pre-release. One such reason is the use of an industry exposure database that apparently does not adequately represent the insured industry, as demonstrated by the low number of insurers that observed changes in loss estimates anywhere close to the RMS guidance.
This leads us to question the accuracy of the industry databases typically used to derive industry change numbers between models. For Florida in particular, the amount of coastal exposure skewed the “industry” guidance, even though it was also well known that inland losses would be increasing. Many Florida insurers maintain some inland exposure to balance their coastal writings, so guidance geared towards an “industry loss” view in Florida was virtually worthless to most insurers.
The storm surge component of the model can also have a substantial impact on results, with estimates varying widely due to minor differences in the data on number of building stories. Furthermore, insurers with inspection data for coastal risks that include base flood elevation often have documented elevation heights that vary significantly from the RMS hazard lookup data.
Across the U.S. insurance industry, there has consistently been more interest in reviewing two or more perspectives of modeled loss estimates, and blended model answers are becoming more common.
EQECAT Worldcat Enterprise V3.16
EQECAT’s North American hurricane updates were the last to be released and, generally, had the least impact on loss results. The new version contains hazard and vulnerability updates and an upgrade to its storm surge and demand surge components.
Guy Carpenter Perspective on EQECAT Hurricane Model Update
Both long-term and near-term losses have decreased about 15 percent as a result of changes in EQECAT windfield modeling. Commercial and industrial lines experienced larger decreases than residential lines. As EQECAT had already adjusted its hurricane model prior to 2010 to take into account claims data from hurricanes Katrina and Rita, along with updated frequency and demand surge modeling that had resulted in loss increases, the relatively minor 3.16 update is not surprising.
Some increases were observed, however, with rises in the District of Columbia, New Jersey, Pennsylvania and Vermont. The Bahamas had the greatest increase in loss estimates at 15 percent to 20 percent, while losses for Caribbean hurricane ranged between down 10 percent to up 10 percent.