Smart Homeowners Insurers Use More Than Cat Models

Implications: A recent news story quotes the insurance modeling pioneer Karen Clark as saying, with keen insight, that insurers rely too much on catastrophe models.  This analysis provides background on cat models and how homeowners insurers use them.  The analysis also offers suggestions for evaluating a property insurer's skill at applying modeled results in conjunction with other information.

Analysis: Catastrophe models, especially hurricane models, have revolutionized property insurance underwriting in the last ten years.  Catastrophes are so difficult for insurers not because they are so damaging, but because they are so infrequent.  Low frequency, high severity events like hurricanes don't generate the many observations that traditional actuarial science needs to quantify risk.  Using simulation, a cat model attempts to turn hurricanes into high frequency events whose risk can be more easily predicted.  A cat model uses years of weather observations to simulate storm patterns and then subjects a hypothetical population of houses and commercial buildings to those storms.  A typical modeling analysis runs 50,000 to 100,000 years of simulated weather across the same population of buildings and accumulates the resultant damage.

The output from cat models appears very precise and objective.  It's quantitive, well organized, and can be reduced to a single number -- the Probable Maximum Loss to be expected from a 50 year or 100 year storm.  So it's incredibly tempting to evaluate and act on that single number.

But like any model, a cat model is a collection of approximations -- equations that describe our best estimates of how atmospheric conditions produce severe weather, how much damage that weather will inflict on buildings and how much the repairs will cost.  Models are riddled with assumptions -- some implicit and some introduced by the modeler -- and highly dependent on the precise description the buildings in the population.  Different models and different modeling analysts often produce a wide range of estimated loss on the same set of properties.  Insurers that rely on a single model may or may not be getting the "right" answer, and insurers that consider several models may be faced with an error range of 50% to 100%.

So the real competitive advantage an insurer brings to property insurance is not buying or building the best model, but rather integrating the results of one or many models with other knowledge about the insurer's data quality, its sales agents, local building codes and construction techniques, and other non-catastrophe risks.

Investors evaluating an insurer's catastrophe exposure should consider not only the reported Probable Maximum Loss for a 100 year storm, but also the company's ability to accurately describe its portfolio of insured properties and its ability to articulate and execute an underwriting and distribution strategy that marries sophisticated number crunching with profound local knowledge.  While it may be useful for an analyst to know that an insurer is reducing its PML from $500 million to $300 million, it's far more informative to know HOW the company is doing it.