The Intelligent Quarterly from the publishers of The Insurance Insider

Summer 2013

Search archive

Advanced search

IQ: Recent high profile model changes at your rivals have had a significant impact on the property (re)insurance marketplace. What lessons can be learned about the use of catastrophe models and could those changes have been better managed?

Bill Keogh: We have a process for managing model change and in the case of our US hurricane model we've updated eight times in 10 years. But not everyone does it that way, and volatility of the changes has not been great for the industry.

Change can be difficult to manage, but if a model change becomes so disruptive to the market I suggest you should look more to the market than the model for the solution to that.

The real solution is that you have to have an independent view of risk. That view of risk should be informed by models because that's a robust framework for analysing things. But if you're blindly taking results from a single model and it's informing all your capital decisions you're exposing yourself to model volatility.

So we encourage people to use a multi-model approach. When you look at more than one model the first thing you'll notice is the results are different, and that leads you to think about why they're different. That exploration educates you on all of the different parameters that make up models and then you're able to form your own view of risk.

You can't outsource that to anybody. I would hate to think that people are outsourcing 100 percent of their view of risk because catastrophe risk is such a complicated problem.

IQ: Much of last year's near-record industry cat burden was driven by what some people are calling unmodelled losses. How would you characterise that experience and what does it say about the usefulness of cat models?

Bill Keogh: The subject of unmodelled loss is an important one. In the case of tornado hail there might have been models, but if people didn't believe them they might not have used them - therefore it's unmodelled.

In the case of New Zealand and Japan it was unmodelled because perhaps the source was completely unidentified. That's a scientific failing as much as it might be considered a model failing. The area source in Christchurch was just not known. And the consensus for maximum magnitude of earthquake in Northern Japan was much lower than what actually happened in Tohoku.

What's really interesting is to look at Thailand as an unmodelled event. In some ways we think people might have been deluded into thinking that if there isn't a model for it there isn't a risk there. Nobody has a Thai flood model but we sure know that there's Thai flood risk now.

This gets back to the idea of owning your view of risk and understanding what you are insuring.

IQ: What about the contingent business interruption (CBI) element of those losses?

Bill Keogh: Think of all the CBI that was exposed in Thailand - you simply can't model that. It's not like the insurance industry has tons of data that tells us about the interconnection between risks everywhere in the world that can then be used in the modelling process.

And if you can't quantify it in a model how can you quantify it from an underwriting perspective? The silver lining to all this might be that it's exposed some business practises that are not very robust and it will result in a change of behaviour.

IQ: Eqecat has a new global platform launching later this year. How will it help you make up ground on your bigger rivals?

Bill Keogh: Each modelling firm has very different approaches to modelling - for better or worse. There are always trade-offs to building models, and I think the trade-off Eqecat made in the early days was that we wanted to have a very robust way of characterising uncertainty and risk, which meant the methodology was quite complex. I think we did that to the detriment of usability and integration. Roll that forward 20 years and easy-to-use models won out.

But because we have so much more computational power today we believe we can now deliver the most robust, easy-to-use and integrated platform with the release of RQE this fall.

We think we've been able to liberate the models and make them accessible to 100 percent of our addressable market. In the past the percentage we could have reached was maybe 30 percent, so for us this is a game changer.

IQ: Why is robustness key to the underwriting process?

Bill Keogh: Precisely because the more robust treatment of uncertainty you have the better characterisation of risk you get. If you have a model that can capture 10 parameters of uncertainty and the distributions around those versus one that might only capture say three, you can begin to understand that would be better.

And you'd rather have distributions than point estimates. If you look at Tohoku, the scientific consensus was that the largest magnitude quake that could happen in that region was an 8.2 but then a 9.0 occurred.

Users of cat models are interested in capturing some of the uncertainty around that maximum magnitude - which we now know has enormous implications for expected losses. We model the uncertainty of event magnitude and frequency.

On the hazard side, the question of what happens to a building when subjected to 140mph winds is not a point estimate but a distribution. If a model only did point estimates you'd get a very different answer that would ultimately underestimate risk in the tail of the distribution.

And the biggest concern for (re)insurers is in the tail of the distribution where they're making capital decisions, pricing decisions and significant risk transfer decisions. Having a more robust treatment of uncertainty translates into better-informed decision making.