As the rains of tropical storm Allison lashed Texas, wreaking damages in excess of $2bn, Professor William Gray, guru of the inexact science of hurricane forecasting, issued stern instructions to insurers. Bolster reserves for catastrophic weather events. Enforce building codes. Expect a massive storm to smash through insured property on the south-east US coast. The hurricane holiday enjoyed by southern Florida and the Gulf and East Coasts cannot go on. Speaking to the Catastrophe Insurance Working Group of the US National Association of Insurance Commissioners, Gray tersely admitted: “I would not recommend writing insurance in south Florida.”

Such a dramatic withdrawal from one of the most capacity-hungry catastrophe-exposed regions in the world is not an option, of course. But the professor's warning should serve as a heads-up to the reinsurance industry to take a second, much more technical look at wind exposures, how and when they occur, how they are connected, and the damage potential that swirls in their destructive landfalls and torrential releases.

Martin Bisping, tropical cyclone expert in the risk and reinsurance division of Swiss Re, cites the possibility of a hurricane making landfall in both Puerto Rico, with its masses of pharmaceutical industry manufacturing facilities, and continuing on to land again in Florida. “The potential for a concentration of losses from Puerto Rico and the US is something that has been missed,” he contends. “If your risk management is based on 50-or 100-year loss events, the contribution of such a combined loss may not be significant. Yet at a low probability level these dependencies play a much more important role, and normal risk management procedures might begin to fail.” He compares the phenomenon to that experienced by the financial markets, where, at very low probabilities, all events appear to be correlated.

Thus, he says, return periods are an insufficient factor for measuring reinsurance and reserving needs. “We use, as a measure of risk, the quantity of so-called shortfall – the average impact of all low probability events, including 500-year and 1,000-year events,” Mr Bisping says. Swiss Re calculates its risked-based capital requirements based on this technique, rather than the more conventional return-period methodology, which, by definition, does not take into account very rare losses. The reinsurer also uses shortfall for risk-based capital allocation to other major exposures such as credit risk.

Shortfall methodology
Using a starting point of, say, 100-year events, the shortfall methodology measures all events with a likelihood of 1% or less, and assesses their average cost. That average is the shortfall – a potential loss amount used to ensure that risk-based capital allocations are sufficient to encompass such unlikely events. “Based on shortfall measurement, the risk of a low-probability, high-severity event can be more appropriately captured in our risk assessment,” says Edouard Schmid, a senior manager in Swiss Re's reinsurance and risk division. “Sticking to single return periods for these events does not make sense. This provides a true picture of the loss distribution.”

Low probability events such as a Puerto Rico/Florida hurricane landfall can have a significant impact on reinsurance, risk capital and reserving. Not just the amount of reserves is affected, but also the geographical methodology for defining them. “On the shortfall measure it is no longer valid to look just at Florida or the Caribbean. You have to incorporate the interdependencies,” Mr Bisping says. Swiss Re's research indicates that, on a market basis, catastrophic loss events in the south-eastern US and the northern Caribbean islands show a much higher interdependence than exists between losses in Florida or Texas and other parts of the US. “It depends on the portfolio spread,” he concedes, “but on a market-wide basis it is important.”

Damage potential
Data from Benfield Greig, the reinsurance broker, highlights one recent example of a storm which had the potential to cause damage in both the Caribbean and the US mainland. Its 2000 Tropical Cyclone Season Commentary and Catalogue shows that Tropical Storm Helene, which came made landfall on the Florida Gulf coast on September 21 and 22, 2000 (fortunately causing just $16m in damage and one death, since its intensity was low), had passed over the Lesser Antilles, Hispaniola and the western tip of Cuba before hitting the US. It passed within miles of Puerto Rico, where many US corporations have operations covered by nationwide insurance policies, cover which was priced without a premium loading to take into account the additional catastrophic exposure.

“Hurricane Georges is the primary example of a storm that caused damage to both areas,” says Helen Dyke, a member of the ReMetrics Natural Hazards team at Benfield Greig. Georges, which swept through the Atlantic basin over 17 days in the 1998 season, made seven landfalls over a zone extending from the north-eastern Caribbean to the coast of Mississippi. “There are upper limits to the intensity of hurricanes, but there is the possibility that intense storms could make landfalls in both the Caribbean and the US,” she says. “Georges is a good example; Mitch is another. Although these storms were travelling over land, they gathered more energy from the sea surrounding the islands as they travelled through the Caribbean.”

Swiss Re admits that the chance of an extremely destructive storm hitting Puerto Rico and urban Florida is relatively low, and that the degree of dependence between the two areas is extremely difficult to calculate, but Mr Bisping maintains that these factors are no reason to ignore what could cause a market-wide calamity. “There exists only a limited number of hurricane tracks, but from them we can calculate the likelihood that a major hurricane might hit more than one area.” Historical tracks give clues to other possible tracks, and it is possible that a weak historical storm – such as Helene – could recur as an intense future storm. “We calculated a loss of $50bn for the industry for an event hitting Puerto Rico and Miami, at category 4 on the island and category 5 in Florida. It is a rare event, but possible at any time. About 20% of the total losses stems from the Caribbean landfall.”

Increased frequency
Dr Mark DeMaria, regional and mezzoscale meteorology team leader at the US National Environmental Satellite, Data and Information Service (NESDIS, an agency of the National Oceanic & Atmospheric Administration, or NOAA), cites Hurricane Hugo, which affected Puerto Rico and went on to a further landfall in South Carolina. “There are a number of examples of very strong hurricanes that have affected the Caribbean and moved on to the US.” He suggests that the frequency of large, devastating hurricanes is set to remain high for the coming several years, and that in the last six the damage caused by Atlantic storms has been relatively light. “We have had an upturn in Atlantic hurricanes since 1995. It is kind of amazing that there hasn't been more damage, given the number of major hurricanes that have occurred since then.”

He explains that the period of high activity is part of a well-documented pattern, which shows natural oscillations in Atlantic storm activity that last several decades. “You have up to 25 years where it is quiet, then it becomes more active again.” The period from 1970 to 1995 was quiet, he says, and the upturn since then is not about to end soon. “Looking at past history, these things tend to last several decades. There is no reason to believe this time it will be any different.”

He says a double-landfall event is more likely over the next few years, as the period of heightened activity continues. “The systems that have the potential for the double hits tend to come out of the deep tropics,” he says. “They are the ones that have showed the most dramatic change in number in the recent period – category 3, 4 and 5 events. The change is pretty dramatic.” The high intensity storms are also the ones that do the most damage, he adds. Finally, he warns that the usual law-of-averages does not apply to the frequency of intense hurricanes. “If you think because you had a bunch of storms last year that this year will be quiet, you're probably got it wrong.”

Dr Mark Saunders, a climate physicist and head of the tropical storm risk team at the Benfield Greig Hazard Research Centre, University College London, said forecasting double-landfall events should be possible. “There are a number of examples in the historical data,” he confirms. “We could look at return periods for events of different strengths hitting different territories.” He notes, however, that to date companies have been more concerned with the exposure of individual territories or islands to a single event, rather than the possibility of a single event impacting multiple locations. “It is a valid point,” he agrees.

EQECAT, the US catastrophe modelling company, has found that there is correlation between Caribbean and US landfalling tropical storms, and the potential for multiplied damages. Mahmoud Khater, senior vice-president and chief technology officer of EQECAT, believes it is essential to take into account the possibility in any probabilistic risk assessment. “These low frequency, high severity events significantly affect the loss exceedance probability curves at the low probability levels,” he says. “Some US primary insurers as well as many global reinsurers have exposures in both regions, and hence it is crucial for good risk management to include the correlation between these regions. Some of these companies may not be aware of this significant correlation.”

He says EQECAT's pure analysis of historical data shows about a 40% chance that an event which hits Puerto Rico will continue and hit the mainland US between Texas and Maine. “There's about 25% chance that this event will hit Florida,” he reveals. “If you look at the severity of historic events, about half of them are moderate to severe.” But the assessment should go further, he says, than a simple analysis of historic tracks and their impact. “Probabilistic risk assessment is essential.” He argues that a stochastic event set including hypothetical events provides an even more potent illustration of the potential for one event to have two impacts. More evidence that the next big Atlantic Basin hurricane could be a double whammy for reinsurers.

SR Wind cat bond hedges dual landfall risk
Swiss Re has laid off some of the risk of a dual-landfall wind event in the Atlantic Basin through the issue of catastrophe bonds. In spring this year, SR Wind Ltd issued $120m of notes and preference shares which will pay out if specified weather events occur in France and/or Florida and Puerto Rico before May 2005. Swiss Re has calculated that of its exposure to hurricane risk in Florida and Puerto Rico, the latter contributes 23% of the total.

“When structuring this catastrophe bond we considered both investor response and Swiss Re's exposures,” said David Colarossi, vice-president of Swiss Re Capital Markets, the lead manager for the bond placement. “This is the first cat bond with Puerto Rico property risk explicitly included. Some investors had the perception that ‘every hurricane hits Puerto Rico'. In actuality, in the last 100 years, 36 hurricanes have passed within two degrees of Miami, while 20 have passed near San Juan. Nine of those hurricanes passed near both cities, including the 1928 hurricane, which was a category 5 when it hit San Juan and a category 4 when it hit Florida near Palm Beach. If that event happened today, industry losses have been estimated in excess of $55bn. SR Wind protects Swiss Re against such an event.”

The notes for the US hurricane are triggered when an event exceeds a value of 2,775 on SR Wind's US Hurricane Index. The index is calculated using peak gust windspeed data from every ZIP code in Florida and Puerto Rico. Uniquely, the notes offer a reinstatement. The issue is spit into two tranches, one for French events and one for US hurricanes. Should one layer of protection be depleted, the other provides cover for both events, assuming it too has not been exhausted.