Cat modelling firms open platforms as insurers demand flexibility

In a changing environmental and regulatory climate insurers are demanding greater knowledge of and control over every component of their catastrophe modelling software. And software vendors are responding to this, increasing transparency and sharing data. There are even moves to create an open catastrophe modelling platform. Clive Davidson reports

Catastrophe modelling is undergoing a transformation. The hazards remain as threatening as ever – witness the recent Oklahoma tornado, Hurricane Sandy in 2012, or the Tohoku earthquake and tsunami in 2011. But the industry’s attitude to the risks and the models they use to help manage them is changing fast.

Unlike market risk or non-catastrophe insurance risk modelling, catastrophe risk modelling has traditionally been outsourced by much of the industry. There are good historical reasons for this – largely to do with the cost and complexity of building the models. But now insurers and reinsurers want to take back the ownership of their view of catastrophe risk.

New regulations such as Solvency II are contributing to this change by requiring companies to demonstrate that they understand the models they use and the assumptions that are incorporated into them – something that has been difficult to do in the past given the ‘black box’ nature of these outsourced models. 

The major model vendors have recognised the changing attitude of the insurance industry, and are taking steps to address it. Meanwhile, an industry-sponsored project is creating an open modelling platform that could radically transform the nature of catastrophe risk modelling (see below).

Catastrophe modelling for re/insurance purposes has several basic elements: a stochastic model of an event (for example, hurricane, earthquake, flood); a hazard module assessing the level of physical hazard across a geographic area of risk; a model of the vulnerability of types of properties to damage from the hazard; and finally, a financial module translating the physical damage into a total financial loss using policy information such as deductibles and limits.

This is a somewhat simplified view, but even so it is clear that it needs a wide range of expertise, including geographical, geological, meteorological, engineering, statistics, insurance and software development knowledge.

Few companies can bring all these resources together and, as a consequence, the market has been dominated by three pioneering companies: Boston-based AIR Worldwide; and California-based firms Risk Management Solutions (RMS) and Eqecat. There has been no significant, surviving new entrant since 1994.

There is now a strong awareness of the science that is available and model users are seeking to access a range of expertise beyond that which can be found in a single company,” says Jane Toothill, director at JBA Risk Management, a natural hazard modelling company, based in Skipton, North Yorkshire.

“When catastrophe modelling first started, the focus was on working out what a model needed do and how to put the software together. We know a lot more about that now, but meanwhile the insurance industry has become more complex and science has moved on,” she adds.

However, when these insurers come to place their reinsurance, they tend to use one, or perhaps a combination, of the major modelling firms – AIR, RMS or Eqecat – which employ different data sets that can be less detailed than the specialised models. As a result, industry experts say, companies can lose some of the benefits of their careful underwriting and end up paying more for their reinsurance.

The trio of traditional catastrophe model providers argue that their strengths lie in the breadth of their coverage and their tried and tested platforms that they have been evolving for 20 years or more. Furthermore, they say they are not standing still and are responding to the insurance industry’s demands. 

A key feature of the traditional catastrophe models is that they provide a complete process for re/insurers, from the detailed scientific understanding of a range of physical phenomena, to a detailed financial estimate of losses for particular portfolios of policies.

To do this, the modelling companies have created integrated platforms whereby the various components – the hazard and vulnerability models, and financial components such as coverage conditions – are tightly coupled, not only for efficiency and consistency, but also simply to get the whole highly complex, multi-part, computationally demanding process to work in a reasonable time frame.

This tight integration also has other significant benefits, says Paul Little, president of Eqecat, for example, by enabling correlation assumptions to be applied consistently all the way through the simulation process of RQE, Eqecat’s catastrophe modelling platform.

Mix and match
The tight integration means it is not possible to simply decouple the components and mix and match the best elements of AIR, Eqecat and RMS’s models.  However, a growing number of re/insurers want to be able to consider alternative views in their catastrophe modelling and not be locked into a single vendor’s version of the world.

“With the current technologies, if through your research you concluded that model X had the best representation of a hazard, but model Y had the best all-round understanding of the levels of damage for the types of buildings you actually insured, it can be very challenging or impossible to harness both of these components to build an improved view of risk,” says Alan Calder, head of group catastrophe risk management at London-based insurer Aspen Group.

The vendors acknowledge this and are beginning to respond, but they accept there are limitations. As a first step along this road to cooperation, in 2012 RMS and AIR entered into an agreement to share exposure data schemes, with Eqecat joining in June this year. Under these agreements, the model vendors will share such things as the coding schemes they use for identifying roof types or building materials, which will allow re/insurers to convert exposure data more or less automatically for inputting to the three vendor models.

“We recognise that as an increasing number of insurers and reinsurers are using multiple models, they have been struggling with the workflow inefficiencies of moving exposure data between models,” says Milan Simic, senior vice-president and managing director of AIR Worldwide, in London.

He concedes, however, that options for addressing this situation will remain “limited and imperfect” until the catastrophe model providers cooperate directly to come up with a better solution.

Meanwhile, re/insurers have been clamouring for more openness from the vendors about the workings of their models. One reason is the discrepancy in results that many users have found when they upgrade to new versions of vendor models. This could be because, for example, the vendors have upgraded the science or adjusted assumptions in the models, which leads to different outcomes in simulation processes). Many companies also suffered severe losses from recent events that were poorly modelled, such as Hurricane Sandy, the Japanese tsunami and New Zealand earthquake, and want to understand why.

Finally, Solvency II has a stark imperative: if an insurer uses a model, it must be able to demonstrate it understands how it works. As a result, the major model vendors have been slowly removing some of the veils around their black boxes, opening them up to an unprecedented degree of scrutiny, as well as allowing some adjustment and customisation. Furthermore, companies using internal models are required by Solvency II to take their own view of catastrophe risk.

“The need to better understand models has been encouraged by the perceived failures in modelling for recent losses,” says David Simmons, managing director, analytics at reinsurance intermediary Willis Re in London. “It no longer suffices for companies to simply licence a model from a third-party vendor and to use that data uncritically. They are required to look at the individual model and decide whether it is appropriate,” he adds.

Race to transparency
As a result, the software vendors are providing more detailed documentation on their models, revealing their assumptions and methodologies, and enabling users to show regulators they understand the science and engineering behind them. Modelling companies are also allowing clients to adjust assumptions and other parameters and do things like test the models’ sensitivity to adjustments.

“Solvency II has spurred a race to transparency,” says Little of Eqecat. “There is an effort among all modelling firms to become more transparent in order to help companies comply with Solvency II standards.”

In addition, intermediaries such as Willis Re and other advisory service providers are doing a significant amount of research for themselves and are asking far more penetrating questions about the vendors’ models in order to advise their clients on the suitability of models. The same goes for some of the new institutions, such as pension funds, that are providing capital for the re/insurance market.

This year, all three vendors have released new software platforms – Touchstone from AIR, Risk Quantification & Engineering (RQE) from Eqecat and RMS(one) from RMS (currently a demonstration version with the production version promised for 2014). All three offer upgraded models of key hazards, far more transparency of assumptions and methodologies than previous platforms, and exploit high-performance technologies for greater speed and depth of modelling.

Touchstone and RQE give users a choice of how they take advantage of these technologies, including in-house implementation or through cloud computing, while RMS(one) is only available via the cloud.

“We designed our system to be highly scalable, manage very large data sets and to be an open platform, and the only way to achieve that was via the cloud,” says Bobby Soni, chief platform and services officer for RMS, based in  the San Francisco Bay Area. ?].

The other innovation of RMS(one) is that for the first time one of the major catastrophe modelling providers allows users to run third-party models, analytics and applications on its platform. The first three companies to implement models on RMS(one) are JBA, Latin American hazard specialist ERN and Australia and Asia-Pacific specialist Risk Frontiers.

Insurers say there are significant benefits to having multiple views of risk. This is particularly the case when internal research and models are incorporated to produce an in-house view of risk, according to Calder of Aspen, which runs all three major catastrophe models. “The benefits include the ability to address non-modelled perils, understand model differences and improve the confidence level in downstream usage such as capital modelling,” he says.

With the trend towards more frequent and extreme hazardous events, re/insurers are hoping the opening up of the catastrophe modelling world to a wider range of scientific, engineering, mathematical modelling and technological contributions will give them better insight and measurements of the risks they are taking on board.

Oasis loss modelling framework

Understanding model uncertainty, managing the perils not captured in existing vendor models and providing model access to a wider community are three of the key challenges faced by the industry today. 

To help address these issues, 20 leading insurers, reinsurers and advisers, including Lloyd’s of London, Cathedral Underwriting, Aspen, Scor and Guy Carpenter, are sponsoring the Oasis Loss Modelling Framework (LMF). Initiated by the UK’s Financial Services Knowledge Transfer Network, Oasis LMF is a London-based, not-for-profit project to create a catastrophe modelling framework and an open platform for third-party catastrophe data, models and services.

At the centre of the Oasis LMF is a simulation engine and calculator that can take hazard and vulnerability models and exposure and insurance data from various sources and calculate potential losses. Around this simulation kernel, Oasis LMF provides an architecture for plugging in and integrating the models and data. Oasis LMF does not actually provide the hazard and vulnerability models, or exposure and insurance data – all these will come from the insurers themselves or third-party providers.

Nor does the framework provide the actual computer capability to carry out the simulations and loss calculations. Again, this will have to come from insurers or a third-party services provider. However, the framework and its calculation kernel are designed and coded to operate on a range of systems and to take advantage of modern high-performance computing technologies, such as parallel processing.

To facilitate the integration of components, the Oasis framework will accept input data in any format. The industry uses a wide range of formats for things like building types and vulnerabilities, and it would be a major constraint and an uphill battle to try to impose standardisation on these. On the other hand, Oasis LMF is creating standards for model development to encourage more consistency in the way models are built.

“We have worked with the industry and academics on what standards we should determine for building catastrophe models,” says Dickie Whitaker, project director of Oasis LMF. The standards will cover such things as sampling techniques and the way in which models handle uncertainty. The sponsor firms will get first access to the standards, following which Oasis LMF will publish them publicly in 2014.

Taken all together – the simulation and calculation kernel, along with the third-party hazard and vulnerability models and exposure and insurance data, all plugged into the framework – Oasis LMF and components are conceptually equivalent to the catastrophe models of AIR, Eqecat and RMS. A key difference is Oasis LMF is open and, therefore, has the potential to create a broad market in plug-in components, with the added advantage that the models and data can be readily compared with one another.

“One of the key objectives for Oasis is the development of transparency, openness and availability in catastrophe modelling,” says Trevor Maynard, head of exposure management and reinsurance at Lloyd’s of London. “By providing the simulation kernel component of the cat model, it enables organisations that hold vulnerability and/or hazard data to make available their data to a wider community. The simple [framework] structure that facilitates plug-and-play use, allows users to truly get an understanding of the inherent uncertainty in the model through sensitivity testing of each component.”

Many universities, research institutes and commercial organisations have a wide range of expertise in all sorts of hazards and regions, with especially valuable knowledge of less common events and less modelled regions, but this information can be difficult to access for insurers. However, the lack of an open platform has been an obstacle to delivering their models and data.

“By building a community and defining modelling standards we are stimulating innovation, supported by a viable commercial environment,” says Whitaker. “Vulnerability and hazard specialists are joining Oasis, along with exposure, run time, visualisation and satellite specialists.”

Associate members of Oasis (who support but do not sponsor the project) include JBA Risk Management, Risk Frontiers, Columbia University, Imperial College London and the Met Office. High-performance computing specialists, such as IBM and Maxeler, are also associate members.

So far, Oasis LMF has developed a demonstration version of its platform, with four third-party hazard models adapted for it, with a production release of the platform and more models planned for 2014.

Insurers say the Oasis framework has the potential to alter the current rather rigid structure of catastrophe modelling software. “It has the ability to transform the market into a more dynamic proposition, enabling plug-and-play capabilities for catastrophe model components from a huge variety of sources in order for entities to better understand the totality of their portfolio risk,” says Stephen Gentili, London-based head of exposure management at Cathedral Underwriting, a Lloyd’s insurer.