11 March 2009

Published in: Capital - models, Insurance risk, Cat risk - ILS

Trying to find the sting in the tail

Tail risk is hard to understand and very difficult to mitigate. But the effects can be seen, disastrously, in the banking industry. Jessica Baylis talks to those looking for answers.

The phenomenon of tail dependency has brought the banking industry crashing down and insurers should take heed. Risk in the tail is highly complex, little understood and potentially devastating, experts say. With climate change and globalisation pushing the boundaries of risk, insurers who do not invest in the latest measures to tame the tail look set to suffer.

Tail dependency is the risk that one extreme event will trigger risks normally assumed to be independent. Such events can be fatal: "Tail dependent risks could lead to catastrophic losses, which could threaten the solvency of insurers" states a report produced last month* which looks into the challenges facing the insurance industry from climate change. It describes research into tail dependencies as "much needed".

"It is acknowledged that very large events can trigger multiple lines of business," explains Steve Mathews, actuary and director at EMB. "The attack on the World Trade Center attack was the classic example because it cut across insurance lines: property policies, business interruption, marine, workers compensation, life insurers." What made the attack even more devastating, he says, was the economic effects, "It paralysed the New York financial district causing problems with stock market fluctuations. Then it started hitting the asset side of insurers' balance sheets as well."

Hurricanes, earthquakes, pandemics and terrorist attacks all have the potential to be tail events. They are a problem for the insurance industry for two reasons. Firstly, they undermine the assumed diversification benefits built into insurers' models. Secondly, risks in the tail by their very nature are hard to foresee and their policy implications for insurers are even harder to decipher. "When you get deeper into the tails, you get greater dependencies," says Robert Muir-Wood, chief research officer at the London office of Risk Management Solutions (RMS), "It gets to a point where it's pretty impossible to track all the consequences."

"A really big cat can have an impact on the economic activity," he adds. "It can start running across all sorts of other things that may actually affect insurers' bottom line. There's a perception that investment decisions and insurance underwriting are totally unrelated but actually in the extremes they're probably not."

Modelling problems

Tail dependencies, however, are so unpredictable and interlinked that they're extremely hard to measure, "Trying to model this structurally is very complicated," says Muir-Wood. The two most common techniques for dealing with the risk are modelling and building scenarios. "Copulas are a modelling technique which allows for a loose correlation most of the time," explains Mathews, "but when you get more towards your extreme scenarios [the model] actually pulls the two loss classes together and forces a very high degree of dependency. So it's a way of trying to model a tail dependency."

But the problem with modelling is the lack of data. "Model output will only be as good as model input," explains August Pröbstl, executive manager in the department responsible for non-life accumulation risks at Munich Re. "We are talking about complex accumulations and in some areas there is almost no data available."

The more "intuitive" method is building scenarios, says Mathews. This is particularly useful when considering risks that rarely materialise. When building a scenario for a nuclear power plant failure, for example, says Pröbstl, "we look into risks where Munich Re is highly exposed and we define those locations, we define potential wind fields, we make assumptions on what the market loss will be and we translate this into an insured loss. Then, based on our portfolio, we translate this into a reinsurance loss for Munich Re."

This is one area where there has already been change as a result of the financial crisis. "It has certainly brought the issue to the surface," says Mathews' colleague, Richard Millns, senior consultant and director at EMB. The worst-case scenarios being considered are even worse than before. "How severe things can get has been modified in light of the recent events," he comments.

Tom Mount of AM Best: "Tail dependencies will become more important in the terms of rating the company."

Certainly rating agencies, under fire in recent months for their role in the sub-prime crisis, are taking tail risk seriously. Tom Mount, chief actuary in the property / casualty ratings area at AM Best's New Jersey office, thinks "tail dependencies will become more important in the terms of rating the company. I think that will be the case in the medium- to longer-term."

So what can insurers do beyond stress testing and modelling? At RMS they are developing means of gaining a greater understanding of tail events. This is key, says Muir-Wood, because "you can't be prudent until you understand what are the structural reasons for correlation."

One area where RMS is doing a lot of research is the "super cat agenda," catastrophes where the secondary effects are more damaging than the catastrophe itself. "Hurricane Katrina [in 2005] was a ‘super cat' because the flooding of New Orleans which happened because the defences failed, was as much of a loss as the actual hurricane."

In California about 12% of residents have earthquake coverage whereas about 80% have fire coverage. An earthquake which causes large fires, therefore, will trigger claims from more than just the 12% with earthquake policies: "That's why it's very important to understand all the secondary consequences," stresses Muir-Wood.

Location, location, location

One way RMS is increasing understanding of tail events and super cats is by looking at causes of the dependencies, "We ask what is actually physically driving the process itself. It's going beyond the historical patterns." Location of policy coverage is one focus of current research at RMS. When an event strikes, seemingly unconnected insurance lines are linked by being, for example, in the path of a hurricane. One of the lessons drawn from the World Trade attacks was that location of risks covered by all classes of policies such as life and liability matters - insurers were hit by multiple claims across different lines by virtue of underwriting policies in one area. "Before 9/11 this wasn’t considered very important," Muir-Wood explains.

He continues, "The first lessons is to start mapping the spatial location of other lines of business. Professional liability of an architect or engineer, for example, maybe affected by the failure of a building they have designed, even in a different city during an earthquake."

Jim Webber, chief risk officer at Aviva, agrees that location is important and says it plays a large part in underwriting. "You manage tail dependencies by looking at risk retentions and avoiding concentrations of risk," he says, "We look at things like geographical concentrations and with storm and flood risk, we look at which part of the country could cause us the most problems. We try and make sure that we are managing our exposures in those areas and avoiding excessive concentrations of risk."

At Munich Re, they believe expertise is the key. As a reinsurer, they take tail dependencies, or correlation risks as they call them, very seriously, "Being able to assume and manage accumulation risk is one of the key success factors of a leading professional reinsurer like Munich Re," says Pröbstl.

Munich Re has an integrated risk management team which deals with accumulation risks across its life and non-life reinsurance, health and primary operations, as well as asset management. In non-life reinsurance alone, the accumulation risks unit employs around 30 experts. Their backgrounds range from geoscientists like seismologists, meteorologists and hydrologists to actuaries and other relevant disciplines, the kinds of skills needed, for instance, to analyse the accumulation risk stemming from nuclear incidents, gradual pollution or terrorism. These people work closely with the experts in insurance and reinsurance.

"Whenever we define scenarios and the means of controlling those, we ensure that we have put all the expertise available at work to capture all the lines of businesses which could contribute to a single event," explains Pröbstl.

Frontiers of insurability

Pröbstl notes the process is a continuous evolution, "We are also widening the list of scenarios we model and control and we continuously evaluate whether there could be correlated lines of business that we haven't yet included." Our strategy is to make use of a collaborative network, thus spurring initiatives and sharing new developments and findings, Pröbstl continues. One such example is "the frontiers of insurability" - the rather secretive research being conducted by Munich Re and RMS into new and complex areas of risk where models are not yet available.

Research is continually expanding, therefore, and the tools available to insurers are increasing in sophistication. However, this picture seems to hold true only for the largest insurers. According to Mount, "Right now, only the larger companies are actually studying tail dependencies. Some of the modelling actuarial consulting firms are heading down this path, but it's still early days for this area of research."

The problem, it seems, is not just lack of resources. Mount's colleague, Raj Guttha, explains that even in firms where they have the appropriate modelling, "at least 20% of them" do not conduct a proper application of the model. A common problem is the delay involved in running lengthy simulations and getting the results to management, he says.

Another more general problem is that the risk management functions do not have the weight behind them to see through the results of modelling and stress tests, "Senior management is not very heavily involved in risk management. Even though they have risk managers and chief risk officers, they don't have any teeth; the senior management would not pay any heed to the risk manager."

Jim Webber of Aviva: "You manage tail dependencies by looking at risk retentions and avoiding concentrations of risk."

Perhaps a more serious problem is that companies seem unwilling to choose the most appropriate model for tail dependencies if it means raising economic capital. By way of example, Guttha describes one company which deliberately chose a model which would require lower capital. "They chose not to use the t copula and instead used the Gaussian copula. The difference is that the Gaussian copula is based on the normal distribution which has thin tails but the T copula is a distribution that has fat tails. Because they have fat tails, the tail correlation will be much higher and they'd have to hold much more capital than they have right now."

The bad news for anyone wanting to sidestep this risk, however, is that tail dependencies are only becoming a more pressing issue. Climate change is increasing the frequency and severity of natural catastrophes and globalisation makes the effects of disasters more complex and further reaching.

The issue is not going away and whilst there are many improvements in the field, bigger steps will need to be taken. "It's an interesting and difficult topic," says Muir-Wood. The insurance industry are "sort of aware of it" but "probably not" doing enough about it. The danger, warns Millns, "is that if they think the risks are hard to quantify, then they don't quantify them."

"People can be very critical," agrees Mathews. "They say, well, you're just crystal ball gazing. But you're really trying to use a whole range of statistical tools, with as much experience as you can possibly get from the insurance industry and from other sectors as well." Nothing will create perfect results, but history shows that the consequences of ignoring tail dependencies are too big." You've got to put all of this together to make the best educated guess you can," concludes Mathews.

Back to top