A recurring problem with any kind of data work is credibility: does the quality of the data warrant the conclusions being drawn from it?
With insurance losses, credibility problems take a number of forms. One of the most common is having too few years of data. When we can only look back to one or two years of losses, it’s hard to tell good years from bad. More technically, it’s hard to establish a credible estimate of likely future losses.
Standard methodologies might, numerically speaking, produce a result. But the meaning of that result may not be credible. If the business’s operations (and risk-management practices) have been consistent for five or ten years, however, it is easier place some trust in those calculations.
A related problem arises when there are too few losses. Practically speaking, of course, fewer losses is always better, but it makes the numerical work harder. If a typical year has only one or two losses—even if there are many years—it’s hard to say what a typical year would look like. In other words, the data may reflect those individual losses more than they reflect the underlying risk.
Both of these problems arise frequently with captive insurance companies and self-insurance programs, especially those of small or mid-sized firms. These companies may have few losses and a limited number of years of data. Though the captive or self-insurance arrangement may make business sense, it may nonetheless be hard to pin down credible actuarial estimates.
A number of approaches can be used in the face of credibility problems. I’ll take up some of these approaches in coming posts.