What can we learn from the 2004 hurricane season? asks Bill Churney

Now that the 2004 US hurricane season is finally over, the industry is turning its attention to the season's potential impact on reinsurance practices and the catastrophe models on which re/insurers rely to manage their hurricane risk. While the season has certainly been an active one, the questions on many companies' minds are just how unusual was it? and can companies take full advantage of their modelling software to generate reliable estimates of the probability of multiple-event seasons, assess their financial impact, and evaluate a variety of reinsurance contracts?

Putting it in perspective

Four hurricanes made landfall in the US in 2004: Charley as a Category 4, Frances as a Category 2, and Ivan and Jeanne, both as Category 3 storms. Hurricane Charley was the most intense storm to make landfall in the US since Hurricane Andrew, and was responsible for the largest insured loss. Current estimates from ISO's property claims services (PCS) unit indicate that, together, the four storms will cost insurers at least $20bn.

While 2004 does rank as one of the costliest and more active hurricane seasons, it was not, however, a record-breaking year either in terms of loss or the number of landfalling storms. The record for the largest insured loss from a natural catastrophe in the US is still held by 1992's Hurricane Andrew. Combined with losses from Hurricane Iniki, which devastated parts of Kauai that year, insurers paid out the equivalent of some $30bn in today's dollars.

In terms of frequency, the most active year since 1900 was 1985, when six hurricanes made landfall. Over this timeframe, 2004 is one of six years in which four or more hurricanes made landfall. Thus a season with four landfalling hurricanes, while not a common occurrence, is still not particularly unusual (see figure 1).

While only three of the four storms made landfall in Florida - Hurricane Ivan technically made landfall near Gulf Shores, Alabama - that state clearly bore the brunt of all four hurricanes. While not a frequent occurrence, four hurricanes have been known to impact a single state. In 1886, four hurricanes made landfall in Texas and three additional hurricanes made landfall in Florida along the Gulf coast.

In terms of insured losses, 2004 was above average, but not highly unusual. Many published reports have been issued comparing this season's losses with historical losses. However, those comparisons use the original losses or the original losses updated for inflation only and are therefore misleading since they do not account for growth in the number and value of properties affected. By re-simulating all historical hurricanes of the past 104 years using current property exposures and values, insured losses can be estimated for each of the historical storms were they to recur at today's property exposures and values. Using this approach, 2004 is one of eight years in which losses would have exceeded $20bn. The largest losses would have occurred in 1926, when a Category 4 storm hit the Miami area (see figure 2).

The largest-loss storm in each year (the occurrence loss) is also shown relative to the total, or aggregate, loss for each year. What is more unusual than either the number of storms or the total loss is that the losses this past season were much more evenly divided between the hurricanes than in previous years, in which one storm tended to dominate total hurricane losses.

Managing the risk from multiple-event seasons

Since Hurricane Andrew, most companies have tended to focus on the potential loss from individual hurricanes, rather than the accumulation of losses from multiple events. In 2004, this approach left some companies looking to purchase extra coverage mid-season.

However, tools have been available that enable companies to analyse the potential impact and manage the risk from multiple event seasons. The following case study illustrates how a representative company can use AIR's CATRADER and CLASIC/2 systems to analyse a wide range of multiple-event scenarios and contracts, including:

- multiple-event probabilities;

- multiple industry and company-specific losses;

- short-term covers that recognise seasonality;

- additional reinstatements;

- aggregate covers.


The AIR hurricane model incorporates a large catalogue of simulated storms with a range of meteorological parameters, including simulated year, day of the year, central pressure, landfall location, radius of maximum winds, forward speed and landfall angle. This approach makes it extremely easy for users to determine the probability of multiple-event seasons, either for the industry as a whole or for individual company portfolios.

For example, to determine the probability of exceeding any given number of events in a year - say, four - the user need only enter that number and the software will return only years in the catalogue with four or more hurricanes. From the exceedance probability curve that is generated, the user can identify the percentile for the first year that meets the criteria. The AIR hurricane model estimates that a single year with four or more hurricane losses in the US has a return period of only about 12 years. A similar calculation for Florida only reveals that the expected frequency of four hurricane losses in that state in a single season is about once every 150 years - still within the range to which most insurance companies manage their catastrophe risk.

Switching from an industry perspective to an individual company perspective, suppose that a particular company is interested in determining the probability of experiencing three or more losses of $50m to its portfolio (see figure 3). The user would simply specify the minimum amount of loss the company would retain for each event and the minimum number of such events in any year - in this case three. In this example, our company faces a 0.54% chance of being hit with three or more $50m hurricane losses, which represents a 182-year return period. Depending on how the company is managing its portfolio, this may or may not be of concern.

If, however, the company also writes in other parts of the country, it might be useful to consider the possibility of three $50m losses from multiple perils. CATRADER and CLASIC/2 handle this easily by providing the ability to run multi-peril analyses. By considering losses from earthquakes, severe thunderstorms and hurricanes, our company's exceedance probability for three $50m losses increases to 1.28%, which represents an 80-year return period - well below the level to which this company manages its catastrophe risk.

CATRADER and CLASIC/2 can also help with real-time decision making regarding short term covers. Suppose Hurricanes Charley, Frances and Jeanne exhausted our company's cat cover with two months left in the hurricane season. In this case it is important that any short term cover the company secures is priced in accordance with the probability that an event will impact the layer, given that the season is already half over. To reflect historical hurricane seasonality, every simulated hurricane in the AIR model is identified by the day of the year on which it occurs. Thus, by simply entering the dates of inception and expiration - for example, from 1 October to 31 December - the software will only consider hurricanes likely to occur between those two points in time.

This kind of analysis is useful not only in real-time situations, but also to prepare for the year ahead. Companies can easily determine the probability that any number of events - whether from hurricanes or a mix of perils - will exceed their retention. As demonstrated in our case study, considering multiple perils may produce a much lower return period and prompt a review of catastrophe cover. In the course of that review, AIR software provides the capability to compare different options, such as second reinstatements and aggregate excess of loss covers.

Will hurricane models change as a result of this year's events? Certainly Charley, Frances, Ivan and Jeanne will be added to the historical catalogues on which the models are based. However, the AIR hurricane model already accounts for multiple event seasons and the return periods generated are consistent with the past 100+ years of historical data and meteorological expertise.

AIR will also collect detailed claims data for each storm to allow our engineers to review damage rates by location, construction type and building code level. This analysis may result in some fine-tuning of the model's damage functions. Therefore, users should expect some small changes at a detailed level, but no fundamental or major changes to the model or model results. Thus, while it was unquestionably an active year, AIR does not anticipate any significant change to the model's estimated hurricane frequency and severity distributions, or to the modelled loss estimates.

All of AIR's catastrophe models provide the probabilities of multiple-event seasons and enable users to easily estimate the losses to their portfolios from individual events from one or more perils in a multiple-event season. The 2004 hurricane season has shown that re/insurers should consider the probability of multiple event seasons and evaluate a wide variety of reinsurance terms and conditions to prepare for multiple event seasons in the future.

Bill Churney is responsible for providing support and service to AIR Worldwide clients in the reinsurance and intermediary markets.