Modeling risk, or maybe not…

With the rising trend of North American natural catastrophe losses, a great deal of attention is being turned to the development of computer simulated cat risk programs. Cat risk models are hardly a new technological advancement for the insurance industry, they have been around since the devastation inflicted on the industry by Hurricane Andrew. Forced by the legislators to provide ongoing coverage against hurricane losses, U.S. insurers looked to the technology industry for a means of rating the risk and reserving accordingly.

There is no doubt that cat risk modeling software has improved significantly since the early models post of Hurricane Andrew. However, as this area of technology expands with new competitors entering the market with alternative products, a growing concern within the insurance industry is the widespread results produced by the different models in rating what appears to be the same risk.

If insurer A is rating its exposure based on what risk model B indicates while insurer C is rating according to completely different results produced by another model, the issue brought under the spotlight is that one of the insurers in question is either over-exposed or competing at a disadvantage in the market. The concern over the variance produced by the various risk models is accentuated at the financial rating and regulation level — particularly as both rating agencies and regulators move more toward individual company risk rating in their assessments.

The Insurance Bureau of Canada (IBC) has been actively pushing the federal and provincial regulators to apply greater emphasis on individual risk rating. This naturally implies heavier reliance on computer risk modeling. In particular, the single largest unknown cat risk in Canada is earthquake which, to some extent, was addressed last year with the introduction of the new earthquake loss provision regulations.

However, reserving for earthquake exposure and rating the risk still depends largely on what the computer risk models say. In that regard, an agreement was recently signed between U.S.-based EQECAT and Canada’s Automated Information Services Inc. (IAO) to introduce the CANADAQUAKE risk modeling software. Although there are currently four cat risk models available in Canada, the CANADAQUAKE program is likely to become the accepted cat risk model for rating earthquake exposure.

Interestingly, and perhaps alarmingly, a spokesperson from within this technology field admits that the industry is really shooting in the dark with regard to modeling potential earthquake losses, primarily due to lack of historical loss experience. Out of the last 100 earthquake events over the past century, the risk models have been fashioned on data from about 20 incidences where the industry has been in a position to carry out analysis.

One of the biggest problems, the spokesperson points out, is that analysis of a geographical area relies on surface signs of an earthquake threat. In many cases, such as the Northridge Earthquake, there was no obvious ground disturbance to indicate that the area was high risk. As such, the accuracy of a cat risk model depends on the accuracy of the data computed. The different variables applied to risk models is the where and why results differ from one model to another, an EQECAT spokesperson says. That, of course, is not the full story, he admits that cat risk model developers apply different value weightings to the risk factors employed by the software. This, another commentator notes, is the competitive edge used by developers to market their wares. My concern with this so-called “competitive edge” is whether the rating of a mega risk such as earthquake should be subject to marketing vagaries. As an observer, and far from an expert on the topic of cat risk modeling, my comment would be that specific rating parameters should be set out by the insurance industry or by the legislators to ensure that each model compares apples with apples.

As one investment banker involved with structuring capital market risk deals points out, a significant difference in the projected risk and potential incidence of say, an earthquake, and that of the actual loss and probability of the incidence, could cause financial havoc. The markets would most likely accept a minor variance in the projected risk of a major catastrophe incidence. However, should an earthquake occur, and it turns out to be a one in ten year event rather than what a risk model may have projected as being a one in 100 year event, “all hell is going to break out on the litigation front”.