Catastrophe modelling has come a long way since it was first introduced. However, not every peril is modelled. Mr Tobias Pfau of Gen Re said (re)insurers must always remember to take all relevant exposures into account.
The number and severity of catastrophic events has been increasing over the last five years. According to the latest Sigma publication, 172 events hit the region in 2015, which is a 50% increase or 58 events more than in 2011. Insured losses are also on the rise and reached US$9.1 billion in 2015. While the verdict is still out on whether this is driven by short-term weather patterns, long-term climate change, demographic changes or simply better tracking of events, the trends in severity and frequency are simply that, facts.
At the same time the regulatory environment has put increased pressure on management to understand catastrophe modelling. Catastrophe models are not perfect and contain assumptions, limitations and unknown shortcomings that create uncertainty in their output. An increasing number of regulators in Asia Pacific expect insurance company boards to have a sound understanding of those issues.
Quality of models have improved tremendously, yet…
Catastrophe modelling is therefore evolving with increasing speed. The industry has come a long way from a few modelling providers that covered a few territories with relatively simple models, to an increasing number of vendors that cover more territories and perils with increasingly sophisticated models. The choices are more than ever.
The quality of models improves with ongoing updates that reflect the latest research in science, computing power and loss experience. Despite all these efforts to capture exposures proactively, it seems that almost every event presents new challenges: each event provides a new insight or something that has been neglected or not considered in the past.
The Christchurch Earthquakes in 2011 reflected a formerly undetected fault line that had not been considered in most models. Seeing three large earthquakes in the span of about a year was unprecedented. The 2012 Earthquake in Japan showed the tsunami exposure, which had not been considered previously, to be a driving peril in earthquake exposure in Japan as it came on top of an earthquake with a magnitude not considered possible for that region, even by Japan’s own scientists.
Additional perils often are added with any new modelling release. Unfortunately, most models play catch up in the sense that only actual experience triggers new development in them. While this might seem a logical step, it leaves insurance and reinsurance companies with not only the known unknowns but the unknown unknowns. Although the perils as such might be known, their severity or frequency is often underestimated.
Adding new business lines to models
Adding other lines of business into existing models is another dimension where models are constantly evolving. Originally focused on property exposures, more and more vendors include other lines, such as motor or marine exposures, in their models.
Arguably, these exposures are more difficult to model but simply ignoring them is not a solution. This evolution also reflects the way companies are required to manage their portfolio—with probable maximum loss (PML) determined across the entire portfolio rather than individual lines of business.
Increase in territorial coverage
Another factor affecting the models is vendors’ increases in territorial coverage, particularly in Asia. As insurance markets grow and change, so does the need for catastrophe management and modelling.
Most pointedly, in Southeast Asia the modelling landscape is scarce and white spots need to be filled, even for the main perils and lines of business.
Some perils are not modelled; non-natural perils
Despite all the efforts to improve, enhance and extend the coverage and quality of models, some exposures and perils still slip under the radar. They are not modelled for a number of different reasons: they do not happen often enough to generate sufficient data for a model (eg, snowstorms); they are so dependent on local characteristics that it is very difficult to arrive at a confident result (eg, flood); or their severity is generally low enough not to be a main concern (eg, localised rainstorms).
Recent losses in the region also outline the exposure of non-natural perils such as conflagration, strike riots or civil commotion. Of the top five insured losses in 2015, the Tianjin explosion was not only the deadliest with 173 victims but also the most expensive insured loss.
Casualty exposure after Nat CAT events
Another often underestimated area of CAT exposure is casualty exposure in natural catastrophes.
The current catastrophe models usually include first-party motor within their assumptions, but there are further casualty classes with accumulation potential that can contribute to a natural catastrophe loss. Obvious examples are workers’ compensation and personal accident. Some model vendors include these classes in their models to an extent. Some insurers have developed their own, relatively simple, in-house assessment techniques.
It should not be forgotten that other casualty classes may occasionally contribute to the overall loss as well, such as resultant professional liability claims and public liability claims.
We observed professional liability claims against engineers and insurance brokers following earthquakes in New Zealand in 2011. In Japan we have read reports of public liability claims against nuclear reactor operators following the tsunami in 2011. Public liability claims were made against authorities responsible for dam operations after floods in Australia in 2011.
Over recent decades, we have also seen many public liability claims instigated against contractors and authorities responsible for forestry management following wildfires in Australia.
Best practice to use scenarios that include all factors
While any non-modelled exposures are difficult to quantify, in a risk assessment exercise they cannot simply be ignored. Best practice has moved towards the use of scenarios based upon plausible but extreme events causing significant capital impact.
These scenarios should include multiple factors, incorporating correlated stresses and events designed to replicate the wider impact of actual catastrophes. These scenarios should consider modelled and non-modelled factors, span multiple periods and factor in management action or decisions.
India – Lack of flood models
A good example of the often-underestimated exposures is flood in India. The 2015 Chennai Floods in India claimed nearly 500 lives and displaced approximately two million people. Although this level of flooding was attributed to a 1-in-100-year level, further analysis showed several factors contributing to the event, such as poor drainage and overflow of dam water.
India is a developing market and the main modelling companies focus on non-flood perils, mainly typhoon and earthquake. While this might be the normal evolution of catastrophe models, and one has to start somewhere, the exposures for other perils cannot be ignored. The experience shows that more than 50% of insured natural catastrophe losses are attributable to floods.
There are probably reasons for the fact that flood modelling in India is in its early stages. Fast-growing exposures, lack and quality of flood mapping data and manmade contributing factors are a few examples outlining the difficulty in modelling flood exposures accurately. However, this is not an argument to wait for a perfect model. Insurers and reinsurers need to address exposures adequately in the underwriting process to provide a sustainable outcome.
Take all relevant exposures into account
One can only spend a dollar once, and taking the modelled exposures away from overall loss costs leaves hardly anything for non-modelled exposures. In terms of the overall experience, they should represent a significant part.
Especially in such markets as China and India, we see that reinsurance pricing is based mainly on modelled perils, which leaves nothing for unmodelled perils, let alone expenses or the need to pay shareholders for the higher volatility of CAT business, which accumulates across many insurers in a market.
Overall, the risk landscape is developing quickly with increasing exposures, new perils and regulatory requirements. While the coverage of current models constantly improves in terms of perils, exposures and territories, there are remaining areas that are not captured by current models.
Experience shows that those exposures cannot be neglected. Reinsurance is only one way to deal with them. However one should start with the gross portfolio, and it is an obligation and prudent risk management to take all relevant exposures into account.
Mr Tobias Pfau is the Regional Chief Underwriter for Gen Re’s Treaty department in the Asia-Pacific region.