One of the many reasons casualty catastrophe is a challenge to model is that the insured’s exposure is driven by its past and present business activities: the industrial processes it employs, the ingredients it consumes, and the products and services it sells. While taxonomies of business activity exist (and Praedicat employs several of them), they are designed for use in economic analysis, and the individual components are typically far less precise than that needed to determine whether a company has exposure to a potential casualty catastrophe event.
Contrast this with property catastrophe. The principal driver of an insured’s exposure to property catastrophe, geolocation, is known precisely for most physical assets. Uncertainty in modeled losses is driven by uncertainty in the location and intensity (e.g., shaking, wind-speed, rainfall) of disasters rather than uncertainty in where insured property is located on a map of potential disasters.
At Praedicat we’ve drawn a high-resolution map of casualty catastrophe based upon tens of thousands of hypothetical mass litigation events involving thousands of distinct business activities. It is a map insofar as it describes a landscape of commercial litigation risk, wherein certain products and business practices represent potential sources of casualty accumulation risk. Like a flood map, the casualty catastrophe map shows which products and practices are exposed to a potential “flood” of litigation.
But drawing and maintaining the map is only the beginning. We also must place individual businesses on the map. While standard taxonomies indicate approximate location – analogous to political subdivisions such as state and county on a physical map – there is no geocode of business activities and products; no single database exists that describes an insured’s business activities in a manner that can be readily placed on a casualty catastrophe map in the way an insured’s properties can be placed on, say, a flood map.
Until now that is…
Earlier this month, we released in CoMeta® data describing the business activities of more than 39,000 business groups – any business entity with at least $100 million in U.S. revenue – and their connection to our casualty catastrophe event set. This nearly 10-fold increase in company-level data means that we can now locate companies representing 85 percent of the U.S. economy.
How did we do this, you ask? Well, it wasn’t entirely by hand.
Over the years, Praedicat company analysts have reviewed thousands of possible connections between companies and the business activities that could expose them to future casualty catastrophe events. Our data scientists have incorporated these training data along with externally-available data on companies in machine-learning and natural language processing models that generate a probability that any one company is connected to any one potential casualty catastrophe event. Nearly 300,000 company-business activity connections in all!
And the beauty of this approach is that it is designed with continuous improvement in mind. Each model run prioritizes company-business activity connections for analyst review based upon how uncertain the model is that the connections are true. These newly-generated analyst data then form the training data for the next model run improving both our assessment of ground-truth and the accuracy and precision of the model itself. Statistics on model uncertainty also help prioritize incorporation of external data sources in terms of their likely impact on reducing model uncertainty in the future.
Thus, while we can’t yet claim to know the precise “latitude and longitude” of every company’s business activities, we know that we’ve made a quantum leap in coverage with this latest data release and, just as significantly, laid the groundwork for making improvements in accuracy with every subsequent release.
We’re eager to work with you to help you manage your risk of exposure to “litigation flood.” Interested? Send me an email at loughran@praedicat.com.