tail factor

A triangle only goes as far back as your oldest data — maybe 10 years. But some claims keep developing long after that: an asbestos claim, a lifetime medical award, a slow-settling lawsuit. The link ratios you can read from the triangle stop at the edge of the data, yet the losses do not stop growing there. The tail factor is the extra development factor you bolt on beyond the last observable column to capture all that future growth the triangle cannot see.

Precisely, the tail factor is the loss development factor from the oldest age in your triangle out to true ultimate (sometimes thought of as 'to infinity'). Because there is no data past the triangle's edge, it cannot be read off directly and must be estimated some other way: by fitting a curve (exponential, inverse-power, or other) to the declining late link ratios and extrapolating, by using industry benchmark tails, or by comparing paid-to-incurred ratios at the latest age. For example, if the last observed link ratio is 1.03 and a fitted curve implies the remaining factors multiply to 1.05, then 1.05 is the tail factor, and it gets multiplied into every accident year's LDF. A 1.05 tail on a line whose triangle otherwise looks nearly done can still add several percent to total reserves.

Tail factors matter because they hit the oldest accident years — exactly the ones thought to be nearly settled — and on long-tail lines a seemingly small tail can dominate the reserve. They are also among the least verifiable numbers in reserving: by definition the data to test them does not exist yet, so they rely heavily on judgment, benchmarks, and curve assumptions. A common error is to ignore the tail entirely because the visible triangle 'looks finished,' which systematically under-reserves long-tail business.