As I have studied underwriting natural catastrophe, one of the more interesting things I’ve learned is that it’s widely done with software built for something else: accumulation. Even though many accumulation solutions (including some cat models) offer an underwriting function, it is not suited to the actual task of underwriting.
To illustrate the point and avoid confusion, here are the definitions of accumulation and underwriting from the IRMI glossary (a very handy tool for an industry with so much jargon):
- Accumulation: In property-casualty (P&C) insurance, refers to the total combined risks that could be involved in a single loss event (involving one or more insured perils).
- Underwriting: The process of determining whether to accept a risk and, if so, what amount of insurance the company will write on the acceptable risk, and at what rate.
The definitions above make it clear that these are two very separate functions, with one central and critical difference: accumulation is concerned with a collection of risks, while underwriting is concerned with a single risk. It should be obvious that different solutions are needed to perform each task.
However, the reason that accumulation tools are used for underwriting is also clear: they use the same risk models.
To use flood as an example, a high quality flood model can be used (and is well suited) to both underwriting and accumulation. Underwriters can consult the model to see the risk at a location, and accumulation managers can use that same model to determine the accumulation of a portfolio. But, it must be remembered that flood models are not perfectly accurate – they have biases, they statically represent dynamic events, they are based on varying qualities of elevation data, and, in some cases, they are not suited to the purpose. In fact, there is even risk embedded within flood models that can be modeled. This is not to dismiss flood models, because they are necessary and critical for flood insurance – this is to illustrate the limitations.
One way to overcome the limitations of a flood model is to use it with a large number of locations. The law of big numbers can improve the reliability of a flood model by ensuring a significant portion of the locations are rated well, while relegating the locations rated not-so-well to a small, less-significant subset. The more locations evaluated at once, the better the results should be. Since accumulation evaluates large numbers of locations, a single flood model is well suited to the task at hand.
Underwriting, though, can’t leverage the law of big numbers. It is concerned with one point at a time. Using an accumulation tool (with one flood model) to underwrite a location means the chances of that single point being mis-rated are significant. Even worse, it’s not possible to know if it’s rated well or not, because there is nothing with which to validate the rating.
The way an underwriter can overcome the limitations of flood models is to use multiple flood models along with different types of data. An underwriter’s rating improves because they can rate a risk based on convergent results from the flood models, or use alternate datasets (e.g. height above water, distance to water, localized claims history) to augment contradictory results from the flood models. Even better, they can get an idea of the reliability of the rating, too, based on the agreement or disagreement between the different models and datasets.
Many underwriters use accumulation software because it is already on hand, and they get access to it for a small incremental cost. However, it’s not the right solution for their work. Using the right solution is a sure way to increase underwriting profitability and reduce costly underwriting leakage.