Property Cat Risk Management – Understanding Correlation
September 30, 2014| By Tony Sammur | Property | English
Strong property catastrophe risk management is a core competency for any professional reinsurer. Because we’re a global group, it’s important for Gen Re to understand the potential impact of Cat risk correlation at every level. Locations within an account, accounts within a portfolio, portfolios within a business unit, and business units within the group; these are all perspectives of catastrophe model output generated by our Cat modeling team.
We thoroughly analyze the tails of the different loss distributions for each subset. This is where inappropriate correlation assumptions can produce nasty surprises, especially when aggregating different segments of the portfolio. While it might not impact the ground-up average annual losses, correlation can make a material difference to PMLs, TVaRs and excess layer losses.
We try to examine correlation from every angle. How strong is the relationship of loss from one business segment to another where both are exposed to a specific Cat event? Should the correlation assumptions used to answer this question be the same for every modeled event for that peril? Should they vary by region, line of business, construction, occupancy, or any other variable? How granular do these assumptions need to be?
Different catastrophe models take different approaches to the problem of correlation, but they all make similar simplifying assumptions.
Intuitively, individual risks in local areas with common features - such as elevation or soil characteristics - will have higher correlation, especially if they are similarly constructed. Think about the risk to buildings in flood zones exposed to severe storm surge. Superstorm Sandy (2012) and Hurricane Katrina (2005) - in both cases flood containment failed - are great examples of this correlation.
Actual hurricane events show that windstorms don’t abruptly stop at borders. The geographic footprint of a Cat event can create links between location losses across multiple regions. Hurricane Ike (2008) demonstrated how inland areas can exhibit correlation with coastal areas in the same event, especially as inland construction tends to be less wind resistive.
Globally, the risks being aggregated can behave even more unpredictably. We’ve learned that geographic diversification alone is not enough to prevent or even mitigate the potential for unanticipated correlation.
In my experience, major catastrophes tend to exhibit unique correlations that are based on the specific characteristics of each event. The Japanese Tohoku earthquake and tsunami (2011) as well as the Thailand floods (2011) showed that correlation can be local, regional and global as a result of supply chain disruption. Yet at the time, these particular events and their impacts were not anticipated by models. Also, a distant major catastrophe or war can drive up the local costs of reconstruction materials, such as plywood or drywall.
I’ll stay with the global theme in my next blog on correlation, and focus on assumptions and uncertainty.