The Risks of Hazard

In August 2015, Lloyd’s and JBA Consulting published a white paper called California Flood – Central Valley Risk Analysis. At the time of publication, it was received with raised eyebrows as the state was in an historic drought. Characteristically, Lloyd’s took the long view and they released the study, knowing El Niño was on the way.

The report is very much focused on the accumulated flood risk in the Central Valley, including property, crop, and business interruption, and the figures it cites are as huge as expected – a potential loss exceeding $24B.

Last week, during the lull of Thanksgiving, AM Best published one of their most interesting articles of the year (mind the subscription wall). Regular readers of the Risks of Hazard will know that a study called Systemic Risk of Modelling in Insurance: Did your model tell you all models are wrong? is a winning topic. Thank you to the authors at Amlin and Oxford for initiating this three-year study (which is just getting going).

There is a great article over at Insurance Thought Leadership by Deb Smallwood describing the traditional interaction between agents and insurers. Or, to be precise, the lack of interaction between them (It’s not a good sign when the metaphor for the link is a trench!). The issue she explores is the difficulty in exchanging the necessary information between agents (or brokers) and insurers to ensure smooth business processes. The same problem could be expanded to include any interface in the entire insurance industry, including reinsurance brokers working with their customers and the reinsurers, and even the reinsurers working with the markets.

Big data is big news, and rightly so. The ability to glean answers from huge datasets is enabling previously impossible innovation in insurance, just like these gents mentioned a few months ago. There are datasets that comprise decades of building history for most houses in the United States. There are geospatial models of the Earth’s surface, with elevations every 15’ or so for the lower 48 states and Hawaii. Then there are risk models and hazard maps that combine all sorts of scientific data. And then there are historical records of losses and loss-causing events. There are tons and tons of data — to no end.

However, there is an unspoken premise for solutions that excavate answers out of data: the right data is in there somewhere. Even though there are “predictive analytics”, “intelligent algorithms”, “virtual learning” and other software tricks to deliver answers, there is nothing quite like having the right data involved. Sometimes there is no “right” data – wouldn’t it be nice if there was a catalog of future floods or earthquakes? Without some predictive modeling, underwriters would be dealing with the Turkey Problem. But using the right information in predictive models or algorithms is, again, essential.

The Associated Press recently ran a story on new research that tries to ascertain the return period (or annual likelihood) of the 2013 Colorado floods. Here is a reminder from the Denver Post of what the flood was like; since it was over two years ago (it seems like yesterday!).

As any decent summary should, let’s start with the conclusion of the research: “…the best way to characterize this storm is a very unusually widespread 100-year event with some small pockets of a 500-year event thrown in there.”

A very unusually widespread event? A 100-year event, except where it was a 500-year event?