Reflections on Recent Events and Implications for Catastrophe Risk Modeling

At the recent Rendez Vous de Monte Carlo there was a lot of discussion about catastrophe risk models. How will the recent events in Japan and New Zealand impact risk models in the future? How will these events impact the way we think about and use models? A series of catastrophe events with significant surprises can create tension between modelers and their clients. But it also creates an opportunity to have a more open dialogue about models and how they fit in to a robust risk management strategy.

What we learn from Japan and New Zealand will be the result of a rigorous examination of lots of new data from these events. In the case of Japan, the Tohoku event produced a rich set of data on ground motion, acceleration and attenuation and tsunami risk. The physics of the mega-thrust fault rupture is also being studied. The many aftershocks are providing us with greater insight in to the implications of aftershocks and clustering. This will take some time to understand and implement in models, perhaps several years. What we learn may have broad implications in how we perceive earthquake risk not only in Japan, but elsewhere.

In New Zealand the shallow depth of the events, the aftershocks, the significant liquefaction and the discovery of a heretofore unknown fault system will expand our learning about earthquake risk in New Zealand, and also improve our understanding of this hazard in other regions.

Despite that future models will better reflect our improved understanding of earthquake risk, it’s important to remember that even after all this new information is incorporated in to risk models, they will still have a significant degree of uncertainty in the loss estimates they produce.

Which brings us to the second topic: how do these events impact the way we think about and use models? Here are some thoughts based on what we are observing as best practices in the market.

• Have a policy for vetting risk models, and acceptance criteria. Not just catastrophe risk models, but any model. Vetting for risk models should be conducted in the same way that an employer performs due diligence before hiring a key executive. Rip a model apart before you put it into production. Understand its assumptions, strengths and weaknesses and where the greatest uncertainties lie. This requires employing top notch talent in the physical sciences.
• Understand what a model does and what it doesn’t do. There is significant “basis risk” between a catastrophe risk model and an “all risk” insurance policy. How are you quantifying and accounting for that basis risk?
• Modeling captures our understanding of the physical world and how frequently rare, but high consequence events will occur and what impact they will have on physical structures. This understanding is, and always will be incomplete.
• Exposure management and exposure analysis is an essential component of catastrophe risk management. Understand what’s going on with your underlying exposures — independent of models — provides insight into changes and developments in risk portfolios. How do changes in exposure compare to changes in modeled results? Does it make sense?
• Scenario/stress testing can account for non-modeled losses and help think through correlation across lines of business that might not be captured by risk models.
• Employ multiple models to access diverse views of risk. There is more than one way to view a risk and models are very sensitive to assumptions. When models differ that leads to an exploration of why they differ, and a deeper understanding of the risk.

Mutual Trust

The best relationships between modelers and risk practitioners are based on mutual trust, honesty and transparency. Given all the uncertainties in quantifying and characterizing catastrophe risk, we strive to share our insights through a robust treatment of uncertainty.

As models evolve based on recent events, the level of dialogue and transparency will only increase as we engage with the market on these important developments. We’ll always be surprised by events. With a commitment to better understand them, we should never be surprised by risk models.