Events not in data (ENID)

Article 77 of the Solvency II Directive defines the best estimate as the probability-weighted average of future cash flows. All outcomes, then, and not only those already seen. A twenty-year liability triangle contains neither a court ruling that retroactively widens cover, nor a long-latency disease that has yet to surface, nor the once-a-century claim. A chain ladder or Bornhuetter-Ferguson projection calibrated on that data returns an average of what has happened, which is a different thing from an average of what can happen. The London market named the gap: events not in data, or binary events in the early texts. Lloyd's gives it a section of its technical provisions guidance, which expects any ENID load to be explicit. Three ways of putting a number on it are in use. A percentage of the best estimate, quick and contested. A list of scenarios with frequency and cost, heavier, which forces you to write down what you fear. Or fitting a distribution and reading off the part that lies beyond what the data could have observed. The loading often comes out small relative to reserves, and that is where the argument starts. Small does not mean negligible on a long-tail line, and a figure carried forward year after year with no scenario behind it proves nothing to an auditor. It covers the mean, not the tail in the capital sense. Not to be confused with the risk margin, which pays for the capital held, or with the tail factor, which extends observed development patterns.