The unprecedented insured loss caused by Hurricane Andrew in 1992 was a watershed for the fledgling catastrophe modelling profession, whose `dark arts' were suddenly in big demand. It woke reinsurers up to the fact that they were exposed to increasing insurance concentrations in catastrophe-prone areas. The 1980s economic boom had lead to a significant influx of wealth into coastal areas - areas inherently more exposed to windstorm, floods and, particularly on the US west coast, earthquakes.
But what started off as relatively simple tools to calculate probable maximum losses (PML) and reinsurance costs have now become an integral part of the insurance business, and are used for everything from primary insurance pricing and underwriting to portfolio optimisation, dynamic financial analysis and enterprise-wide risk management.
Early modellers still lead
One of the first catastrophe modellers was Karen Clark, an economist by background who initially worked with EW Blanch in an attempt to bring their different disciplines into one model. She went on to set up her own company, Applied Insurance Research (now AIR Worldwide Corp, owned by Insurance Services Office) in 1987 and is arguably still leading the way - at least in technology terms.
The market leader, however, is now Risk Management Solutions (RMS), owned by UK media group Daily Mail and General Trust plc. Robert Muir-Wood, the company's head of global risk modelling, claims to have at least a 50% share of the global market for cat modelling. This is estimated by some to be worth $100m-$150m and now includes the primary insurance and investment sectors in addition to reinsurers and brokers. "We are growing at around 20% to 25% a year and currently employ over 280 staff," he said.
The company's top-selling product is its US hurricane model, first launched in 1993 and re-certified in May this year by the Florida Commission on Hurricane Loss Projection Methodology. This allows the model to be used by primary insurers for setting residential insurance rates and is effectively a high-level endorsement of the product's accuracy and reliability.
No single answer
The Florida Commission also recently re-certified the equivalent US hurricane models from AIR and EQECAT, the other two major players in the market. Both these models form a part of global models or `platforms' called CATRADER and WORLDCAT respectively, which cover all natural perils in most major countries (over 80 in EQECAT's case). Depending on the number of perils and number of countries required to be factored in, annual licences will cost anything from around $100,000 to over $1m.
But if the Florida Commission has certified all three products, does that mean any one will do? Not so, said Dickie Whitaker, manager of Guy Carpenter's Instrat catastrophe services unit. "No one model provides the right answer - and it is impossible for anyone to claim they produce the best model," he said.
He should know, since Guy Carpenter used to have a 50% share in EQECAT (now wholly owned by the American Bureau of Shipping, ABS). The broker now licences software from all three leading suppliers and adds value to the modelling process by being aware of their respective strengths and weaknesses in any given situation.
Blended results better
Employers Reinsurance Corp (ERC) has taken the process even further and has blended the RMS, AIR and EQECAT products into a `supermodel' based in Bangalore, India. "This was no easy task," said Samira Barakat, head of ERC underwriting and risk management. "Fortunately we had a highly skilled operations team in India able to do this work and, using the internet, we now get all our risks analysed there to ensure consistency," she explained. Before June 2001 the modelling was done in a dozen R&D centres all over the world.
Oliver Peterken of Willis Re's consulting and analytical services division also licences all three leading products, but his firm produces its own more detailed models as required. "For example, we recently produced a earthquake and flood model for the Turkish government which went right down to postcode sector level." Avia, the UK general insurer until recently known as CGNU, has also recently commissioned the firm to produce street-level flood risk maps for all parts of the UK.
Benfield Group is another broker which produces its own models. John Gascoigne, risk analyst at its modelling unit Remetrics, said that meteorological and loss data from the 1999 windstorms Anatol, Lothar and Martin storms in Denmark, France, Germany, Spain and Switzerland helped to improve the accuracy of continental European models significantly."Models are only as good as the data you can put into them," he said. "In the US and UK there is plenty of readily available data, but the importance of recording such information is now more appreciated in continental Europe since the 1999 events."
However, no matter how good a catastrophe model is, the output is totally dependent on the risk data that insurers and brokers feed into them. For example, actuary David Anderson, senior vice president at Guy Carpenter, is currently working with the Danish insurance industry to establish an agreed standard for defining insured property.
Standardising sums insured
The major Danish primary insurers (four companies have 75% of the market) are still licking their wounds from Anatol. During analysis of their portfolios last year, Guy Carpenter found each was getting very different results from the leading catastrophe models.
"All houses in Scandinavia are full-value insured, but the problem was that each company was using different methods of calculating value," explained Anderson. "This meant there was a chance that all the leading models were not properly calibrated and it was dangerous for reinsurers to rely on them."
In an attempt to create a level playing field for buying reinsurance, the insurers have agreed to value all houses by multiplying the living area by a specific value per square metre, which is indexed by an agreed amount each year. The insurers will also provide all their historic portfolios and losses to the major modellers so that vulnerability curves for Denmark can be recalibrated on a more equitable basis.
"We hope to have new versions of all three major Danish models available by the 2003 reinsurance renewal, which should enable fairer pricing for everyone," said Mr Anderson. A similar exercise is planned in Sweden and Norway.
According to Willis' Mr Peterken, the cat-modelling sector is if anything enjoying a welcome break from the frenetic pace of technological development over the past ten years. "We are waiting for insurance data to catch-up with us," he said. No doubt the increased premiums imposed by reinsurers at the latest year's renewal for inadequate risk information will help spur on improved data collection.
Technological innovation in catastrophe modelling continues nevertheless, thanks primarily to the increasing availability of computing power. The latest development is `physical' modelling, in which the physical laws of nature are modelled to create events, rather than simply modelling events themselves. Fortunately, governments and academics have already invested billions of dollars and man-hours in developing physical modelling technology; the trick was simply to adapt it for use in the insurance industry. Willis and AIR claim to lead the field in this respect, with Willis producing the first physical model for UK windstorm in 1998, followed by AIR's European windstorm model based on Numerical Weather Prediction (NWP) technology in 2000.
According to AIR president and CEO Ms Clark: "It requires a step-change in research and development investment to produce physical models, but the result is a step-change in realism."
The company is currently developing NWP for use in hurricane models and a similar approach is being adopted for earthquake products. "We have gone from general damage functions, which were based on expert opinion combined with loss data, to modelling the failure of specific components with structural engineering equations," she said.
The new probabilistic UK fluvial flood model launched by RMS in October last year is also considered to be a major advance. This took 12 man-years to develop and allows reinsurers to analyse for the first time the aggregate risks they face from extreme multiple river events, as happened in Britain in autumn 2000. "It is a big step forward," said Guy Carpenter's Mr Whitaker. Similar models are being developed for Europe and the US.
RMS is also developing a significant upgrade to its US hurricane model, which is expected to provide far more realistic variation in events due to its ability to synthesise multiple hurricane tracks from start to finish over the entire Atlantic basin. It is scheduled for launch in February 2003 with version 4.3 of its RiskLink cat-model platform.
Terrorism - the new hazard
However, following the catastrophic events of 11 September 2001, the hazard that everyone would now like to model is terrorism. RMS, AIR and EQE are all developing models for launch this year covering at least some aspects of terrorism risk in the US. RMS's development team hopes to have the first stage of a probabilistic model available this summer. According to EQECAT managing director Rick Clinton, his company has already created a model for assessing the impact of terrorism and earthquakes on US workers' compensation insurance, reflecting increasing concerns about large concentrations of employees in potentially vulnerable buildings.
AIR is creating a probablistic structured approach to property damage taking into account different terrorist groups, their objectives, the kinds and location of likely targets, types of attacks and the probability of achievement. To do this, the company has drawn on the knowledge of security experts and used what is known as the Delphi method to assign probabilities to their views. Starting with studies done in 1944 and developed by the Rand Corp, the Delphi method is used to collect and distill knowledge from a group. It has been widely used to generate forecasts in technology and education, among others.
However, Ms Barakat of ERC is sceptical. "Unlike natural perils, there is very little history on which to model terrorism," she said. "We are keen to seen what the new models can do, but I suspect it will be little more than identifying potential concentrations of risk."
By Simon Fullalove
Simon Fullalove is an engineer and writer specialising in engineering, shipping and insurance topics. He is editor of professional journals for the UK Institution of Civil Engineers.