Last September's attacks in the US shed a new light on catastrophic workers' compensation exposures.

Recent figures indicate that 2001 was a difficult year for providers of workers' compensation insurance. With a combined ratio of 121%, the industry is struggling to make up for years of underpricing. It is also adjusting to a new reality, one in which workers' compensation losses from a single event can reach catastrophic proportions. Before the events of September 2001, workers' compensation reinsurance was relatively inexpensive, readily available and virtually unlimited in scope. This meant that primary insurers did not need to assess carefully or manage closely the potential for catastrophic workers' compensation losses. The attack on the World Trade Center (WTC) changed all that.

Current estimates are that workers' compensation losses from the WTC attack total $2bn to $3.5bn, though there is speculation that this figure will grow. This was the first catastrophic event, either man-made or natural, in which workers' compensation claims played so prominent a role. In the face of the rapidly increasing workers' compensation rates that have resulted, companies have begun to rethink their positions not only with respect to possible future terrorist activity, but also with respect to natural catastrophes. Workers' compensation losses from the WTC attack were unprecedented; they also brought the realization that the more widespread damage caused by earthquakes could result in even higher claims.

As the insurance industry revisits existing risk assessment processes with the goal of improving knowledge of their accumulated exposures and potential losses from extreme events across multiple lines, catastrophe modelers are making available new tools for quantifying the risk.

Earthquake exposure
Simulation models that provide probabilistic estimates of future property losses from natural catastrophes, such as hurricane, earthquake, tornado, hail and flood, have been available since the mid-1980s. Initially greeted with polite scepticism, they have since become the standard global risk assessment technology and provide essential support to pricing, underwriting and risk transfer decisions. The extension of the methodology to estimating workers' compensation losses from earthquakes was a natural one.

To do so, one must begin with an assessment of the hazard, with estimates of the frequency, severity and location of future earthquake events. Catastrophe modelers are well equipped to do this. Detailed historical databases of actual events are supplemented with auxiliary information from paleoseismological studies and geodetic surveys, scientific expertise and sophisticated statistical techniques, to produce large catalogues of simulated earthquakes.

Once the hazard is identified, the vulnerability of the building stock must be assessed. Critical to estimating the workers' compensation losses that result from earthquakes is a rigorous methodology for estimating building damage. It is extensive structural and non-structural damage to buildings that causes the casualties which ultimately result in large workers' compensation losses. Models that incorporate rigorous engineering-based techniques for estimating building vulnerability can provide a much more accurate estimate of building response to ground shaking.

The number of injuries and fatalities, as well as the severity of injuries, is a direct function of building damage. In the case of earthquakes, a large percentage of injuries occur as a result of damage to non-structural elements, such as ceilings and windows (typically the first to go), and the failure of individual building components (fallen beams). Construction type is also a consideration. The collapse of heavy members of concrete buildings, for example, will typically cause more severe injuries than the collapse of structural components of wood frame buildings. Also, highly brittle structures, such as reinforced concrete and unreinforced masonry (URM), are fatal during earthquakes. Structural failure or collapse accounts for most fatalities and serious injuries.

The number of injuries and deaths is also a function of the number of workers occupying the building at the time of the earthquake. Probabilistic loss estimates are therefore typically provided under various assumptions about the time of day and the day of week that an event occurs.

Ultimately, estimates of workers' compensation loss are based upon the number of people injured, the severity of their injuries, and the cost of the injuries given their severity. Model output consists of complete probability distributions of workers' compensation losses and injury counts for the total number of injuries and fatalities or broken down by injury severity level.

Loss dimensions
The ground shaking caused by earthquakes can cause severe building damage over many hundreds of miles. This is particularly true, for example, of the New Madrid Seismic Zone (NMSZ) in the central US. The amplitude of seismic waves attenuates very slowly in that region due to the density of the underlying geological materials. AIR Worldwide (AIR) estimates that a repeat of the 1811-1812 series of large magnitude earthquakes (the largest of which is estimated at M8.1) would cause workers' compensation losses exceeding those that would result from a similar magnitude event in California.

As figure 1 shows, the 1811-1812 earthquake series caused significant ground motion in nine states, covering an area of about 350,000 square miles. While the region was sparsely populated in the early 1800s, today the NMSZ is home to millions of people and includes major metropolitan areas like St Louis, Missouri, and Memphis, Tennessee.

Greatly exacerbating the risk in this region is the fact that most of the structures were not built to withstand earthquake shaking. An earthquake of 6.0 or larger in this region would cause significant damage, especially to unreinforced masonry construction, which is still the predominant building type in the region's urban centres. While the NMSZ has received more attention in recent years, efforts at mitigation are not widespread.

Loss estimates obtained using the AIR model indicate that if the 1811 New Madrid earthquake struck the Central Mississippi Valley midday today, workers' compensation losses could reach $33bn.

Terrorist attacks
Terrorist activity in the US is not new - the first WTC bombing in 1993 comes to mind. Most attacks, however, have been more limited in scope and carried out by domestic groups or individuals whose agenda has not typically included the deaths of large numbers of people. The 1995 bombing of the Murrah Federal Building in Oklahoma City, which resulted in the deaths of 168 people, was a shocking exception. But it is the magnitude of the losses from events of September 2001 that has the insurance industry scrambling for answers to questions regarding coverage, exclusions and price.

Without doubt, the most challenging aspect of modeling workers' compensation losses from terrorist attacks is estimating the frequency and severity of potential future attacks. Unlike earthquakes and other natural disasters, whose occurrence has a physical basis that can be understood by the scientists who study them, terrorist attacks are a function of the malicious intent of groups of varying size and with varying agendas that may change over time. Complicating things further is the fact that terrorist attacks, unlike earthquakes, can be prevented. The potential effectiveness of a Department of Homeland Security adds a dynamic aspect to the problem of estimating frequency. In any case, the uncertainty surrounding the frequency, location and severity of future terrorist activity is clearly higher than for natural catastrophes.

Modelers have put forward a variety of possible approaches to deal with this problem. The approach taken by AIR, in its recently released terrorism model, is based on the Delphi Method, which was developed by the RAND Corp at the start of the Cold War and has been used to generate forecasts in many subjects, including intercontinental warfare and technological change. For this application, a team of counter-terrorism specialists was put together. These experts have national, high-level operational and analytic expertise and a detailed knowledge of databases of both domestic and international terrorist activity from such sources as the Federal Bureau of Investigation (FBI), the US Department of State, and the Centre for Defence and International Security Studies (CDISS). Repeated cycles of formal questions were posed and answers statistically combined to achieve a matrix of rates of occurrence by attack type. As with any probabilistic loss model, whether for natural or man-made events, estimates of frequency and severity can be updated as new information becomes available.

Probabilities are also assigned to databases of possible target locations, which include high-profile and therefore high-probability trophy targets. The type of attack perpetrated is a function of the type of potential target.

As with earthquake hazard, a rigorous engineering-based approach to estimating building damage from weapons effects is critical to understanding the potential for injuries and fatalities to result. These effects, on both the target and surrounding buildings, are multiple and can include pressure waves, shock waves, fire, and both falling and projectile debris. AIR's weapon effects models are applied to property databases that contain structural and square footage detail.

Loss dimensions
In general, the effects of terrorist attacks are highly localized, at least in cases that involve conventional weapons. (The AIR model currently covers the effects of blast from a wide range of bombs and of airplane crash, though research on the effects of chemical, biological, radiological or nuclear (CBRN) attack is underway.) AIR estimates that if a truck bomb approximately two and a half times the size of the one used in Oklahoma City were to be detonated somewhere on Sixth Avenue in midtown Manhattan in the middle of a weekday, workers' compensation losses would be in the range of $4bn to $5bn. Based on estimates of building occupancy, some 25,000 casualties could result, of which as many as 10% would be fatalities. But this is not a worst-case scenario; larger bombs and therefore larger losses are possible.

Managing risk from terrorism has, unfortunately, become an essential part of today's business decision-making process. Probabilistic models will help quantify that risk to insurers, reinsurers and corporations. A well-structured model can shed light on the various attributes of the risk and the factors that are the drivers of the risk. The modeling framework also forms the basis to conduct uncertainty and sensitivity analyses which are key to building confidence in the modeled results.

While catastrophe models will continue to evolve as the hazards they deal with become better understood, a probabilistic approach to assessing potential catastrophic workers' compensation losses, whether from earthquakes or terrorist attack, is currently the most appropriate way to handle the abundant sources of uncertainty inherent in such phenomena.

By Beverly Porter
Beverly Porter is director of communications at AIR Worldwide Corp (AIR). She holds an MAPE Degree in Economics from Boston University.