The nature, frequency and widespread occurrence of tornado and hail risks pose a significant threat to today's re/insurers, further reinforcing the need for an enhanced and comprehensive risk assessment model.

Individually and collectively, hail events have caused a disproportionate amount of the insured loss associated with thunderstorms. In 2001, approximately 50% of the property damage associated with tornadoes, hail and straight-line winds through July was caused by a single hail event in Missouri on 10 April (National Climatic Data Center). Insured losses from this hail event, affecting the northern suburbs of St Louis, Missouri, have been estimated at $1.2bn by the Insurance Services Office (ISO). The magnitude of this event is comparable with the estimated $1.5bn and $1.2bn in insured losses caused by the 1995 Fort Worth, Texas and 1990 Denver, Colorado hailstorms,1 respectively. Additionally, hailstorms have also caused property damage in excess of $200m in Omaha, Nebraska and Bismarck, North Dakota in April and June 2001.

Although tornado losses during 2001 have been relatively small, some significant events have occurred that serve as reminders of the scope of tornado risk. The loss of over $70m (Washington Post, 30 September) caused by the 24 September tornadoes in the Washington, DC area has underscored the fact that significant tornado risk exists outside of ‘tornado alley'. History further supports this notion as the 1953 Worcester, Massachussetts tornado accounted for between $850m and $1.2bn in losses, and 1979's Windsor Locks, Connecticut tornado caused additional losses estimated at approximately $600m.

Early autumn 2001 has also been marked by many tornadoes, including a record of 83 reported in the first half of October. Tornado outbreaks are often viewed as a surprise at this time of year, when attention is more focused on hurricane hazard. However, tornado and hail risks are prevalent throughout the year, with the secondary maximum in tornado and hail probability occurring in early October in the central US. The tornado that moved through downtown St Louis, Missouri on 29 September 1927 – an estimated $1.9bn loss – further demonstrates the tornado risk in autumn.

In fact, at least four weak tornadoes were confirmed by Environment Canada near Toronto, Ontario on 4 July. Although these tornadoes did not cause catastrophic losses, their proximity to Canada's most populated city illustrates that tornado risk extends outside the US. Approximately 450 tornadoes have been reported in southern Ontario since 1919. In 1985 a tornado produced F4 damage in Barrie, causing more than C$150m in loss (Environment Canada). The risk due to hail in Canada is also substantial, particularly in central Alberta near Calgary and Edmonton.

Risk in metropolitan areas
Although there is no evidence of long-term increases or decreases in tornado or hail hazard, economic losses have and will likely continue to increase with time as the population and wealth increases. The concentration of exposure in metropolitan areas makes these locations particularly susceptible to large losses. The Oklahoma City tornado of May 1999 is another example of a high loss event in a suburban metropolitan area, causing over $900m in damages. However, consideration of other historic events reveals that the loss potential from tornadoes can be much greater. When adjusted for wealth, St Louis, Missouri has sustained the two largest losses from individual tornadoes in US history. The most damaging tornado, estimated as F4 on the Fujita Scale, formed in St Louis on 27 May 1896 and was near maximum intensity as it crossed the Mississippi River into East St Louis, Illinois. In total, the single tornado caused between $2bn and £3bn in losses.

Tornadoes downtown
Unlike the Oklahoma City tornado of May 1999, the 1896 St Louis tornado caused substantial losses in the city's central business district. While the specific effects of an urban area on the initiation and/or maintenance of tornadoes are not fully understood, it is indisputable that tornadoes can and do cause losses in downtown metropolitan areas. The Storm Prediction Center has compiled a catalogue of tornadoes that have caused damage in the central business district of major cities. Since 1871, at least 21 tornadoes have impacted central business districts in major US cities, and downtown St Louis has been hit at least four times since then. Most recently an F2 tornado moved through downtown Fort Worth, Texas in March 2000 causing an estimated $500m in property damage. A number of high-rise office buildings were impacted, including the 37-storey Bank One Tower. Business interruption can also become a substantial component of the total loss in these situations.

In the last 50 years, nearly every major urban area east of the Rocky Mountains has been impacted at some point by golf ball or greater size hail. This hazard is somewhat different than that associated with tornadoes; hailstorms are both more frequent and in most cases produce larger areas of damage than tornadoes. Although the level of damage at a spot location is often less than that from tornadoes, losses for a single event can accumulate to substantial levels. In addition to the Denver and Fort Worth hailstorms mentioned previously, a hailstorm in Calgary, Alberta in September 1991 caused insured losses of approximately C$500m.

The tornadoes and hailstorms of 2001 have reinforced the need for accurate quantification of risk in major metropolitan areas. Released on 2 October 2001, RMS's US and Canada tornado/hail models incorporate an innovative methodology for capturing these extreme loss events, accurately and efficiently.

Tornado/hail model
The high loss tornado and hail events in the last decade have created a growing demand for accurate and efficient assessments of tornado and hail risk. Risk assessment models require years of production time; for instance, approximately six person-years have been invested at RMS in the development its new US and Canada tornado/hail models. In this development effort, specific focus was placed on accurate simulation of high return period losses, such as events impacting metropolitan areas.

Multiple thunderstorms
One of the challenges associated with modelling tornado and hail is representing the hazard of each tornado and/or hail event. New research, including extensive hand analysis of historic events, was required to reproduce the multiple thunderstorm outbreaks that control the damage patterns on the ground. This process involved recognition of tornado and hail ‘clusters', or areas of focused hazard within the larger scale event. This method of event description is essential for modelling tornado outbreaks, where the accumulation of losses from multiple tornadoes can be significant. Understanding the orientation of individual tornadoes and hailstreaks within an event has enabled simulation of damage patterns consistent with different meteorological sources, such as lines of thunderstorms along cold fronts or multicellular systems with more random patterns of damage.

Set of stochastic events
The small scale of individual tornado and hail damage areas and the large geographic areas susceptible to these perils presents a significant modelling challenge. RMS invested substantial resources into the optimisation of the stochastic event set size, yielding diversity in event intensity and geographic coverage, without significantly compromising the time required to analyse a portfolio. Because the most damaging tornadoes are rare, stratified sampling should be used in the generation of tornado events to ensure a robust set of even the most intense events. In total, over 44,000 tornado/hail events are included in the US model and nearly 9,000 events are included in the Canada model.

Variable resolution grid
In creating the tornado/hail model, RMS developed a variable resolution analysis grid to further optimise both the accuracy and the efficiency of loss analyses. Larger grid cells have been implemented in areas with relatively low risk of tornado and hail and low exposure concentration such as the western US and northern provinces of Canada. Smaller grid cells have been implemented in areas of greater risk to tornado and hail including the central and eastern US and the southern provinces of Canada. This advanced modelling methodology yields improved accuracy of modelled high-return period losses by differentiating between areas of varied exposure concentration. By using grid cells that are comparable in area to the exposure data resolution, the computer time required to analyse a portfolio is minimised.

The need for efficient and accurate assessments of tornado and hail risk continues to grow. The models available to primary insurers, brokers and reinsurers worldwide are great demonstrations of advanced catastrophe modelling, designed to solve the real-world underwriting and portfolio management problems of US and Canadian insurers.

1 Loss values throughout this article have been adjusted to year 2000 dollars by accounting for changes in exposure. Pre-1950 losses are based on the published results of Brooks and Doswell (2000) and ‘Significant Tornadoes, 1680-1991' by Thomas P Grazulis. Post-1950 losses are based on Property Claim Services (PCS) data unless specified otherwise.