So Where Is Tornado Alley Exactly?
March 10, 2014| By Tony Iafrate |
Region: North America
Occasional op-eds in the insurance press point out the risk of being overly reliant on the output of risk models. The points made in such articles are often accurate and insightful - but they make actuaries like me, who build such models, wince.
Over the past few years, my colleagues and I have spent some time studying tornado incidence statistics. Recently, we were fortunate to come across a research paper written by Chris Broyles and Casey Crosbie of the Storm Prediction Center in Norman, Oklahoma.
The authors pieced together over 120 years of tornado data (1880-2003) to develop relative frequency rates for 4,000-plus counties in tornado-susceptible states. It led us to ask how 120 years of data could be compared to indications from one of the widely used catastrophe model vendors. So we ran it with identical risks in each of the counties and asked what would be the 99th percentile loss.
The idea was to compare the ranking of counties from the Broyles-Crosbie study with the ranking we would develop from the catastrophe model output. Notwithstanding some (known) uncertainties, we felt the comparison would be a useful “first pass” in assessing the model’s accuracy in capturing county-by-county frequency differences.
The maps below can be used to make the comparison. In both maps the more intense the shading, the more highly ranked the county in terms of tornado frequency. The first map, based upon the model output, displays a peak in an oval centered over western Oklahoma. The exposure seems to drop off in a fairly smooth pattern from there.
The second map, based upon the Broyles-Crosbie statistics, displays peaks in central Mississippi and Northern Alabama. There is practically no commonality in the top 5% ranked counties using the two criteria.
What does it tell us? With respect to valuing one’s exposure to tornado, it seems that there may be sufficient history to form a reliable view that’s distinctly different from model indications. In the context of recognizing the danger of excessive reliance on model indications, we seem to have identified a case where model indications diverge from the historical record. That doesn’t mean the model is not useful. Rather, it means that one should understand and be confident in the modeling techniques and assumptions used.
The most fundamental risk management discipline is tracking your exposure. If you are interested in knowing your exposure to the "perfect storm," a robust accumulation management tool (such as Gen Re Intermediaries’ MapPML tool) is essential.
To find out more about our analyses of tornado data, contact Tony Iafrate, email@example.com, +1 203 328 5406.
If you’re interested in learning more about how Gen Re can help you measure and reduce your exposure to tornado risk, contact your Gen Re representative.