Why reserves exist and why IBNR is the hard part
Insurance inverts the normal production cycle: the premium is collected first and the true cost — claims — emerges over months, years, or for casualty lines, decades. At any balance date the insurer owes money for claims in progress and for losses that have already happened but not yet surfaced. Reserves are the accounting recognition of that debt.
Case reserves on known claims are set by adjusters with facts in hand. IBNR has no facts yet: the fire has burned, the injury has occurred, the negligence has happened, but no claim has arrived. It must be estimated statistically from how the book has behaved historically — which is why IBNR is where actuarial method and judgement carry the balance sheet.
Development triangles: the raw material
Reserving starts with a triangle. Claims are grouped by accident year (the year the loss occurred) and their cumulative paid or incurred amounts are recorded at successive ages — 12 months, 24, 36 and so on. Reading across a row shows one year's losses maturing; reading down a diagonal shows the latest snapshot of every year.
The pattern in the triangle — how losses at age 12 typically grow by age 24, and so on — is the empirical engine of most reserving methods. Long-tail lines like general liability develop for many years; short-tail lines like property settle quickly. The triangle makes that behaviour measurable.
Chain Ladder, expected loss ratio and Bornhuetter-Ferguson
The Chain Ladder method computes age-to-age development factors from the triangle and projects each accident year to its ultimate cost. It works well for mature, stable books but is dangerously leveraged for recent years: a small random fluctuation in early data gets multiplied through the whole projection.
The expected loss ratio method ignores emerging experience and reserves to the pricing assumption — sensible when a year is too green to trust. Bornhuetter-Ferguson blends the two with credibility weights that shift from the prior expectation toward actual experience as the year matures. It is the standard choice for immature accident years precisely because it tempers the Chain Ladder's leverage.
Ranges, not points: stochastic reserving
A single best estimate hides how uncertain reserves are. Stochastic methods quantify the spread: the Mack model derives a standard error for the Chain Ladder estimate analytically, and bootstrapping resamples the triangle's residuals to simulate thousands of alternative outcomes, producing a full distribution of possible ultimates.
Ranges matter for more than intellectual honesty. Capital requirements, risk adjustments under IFRS 17, management margins and reinsurance decisions all hang off reserve uncertainty, and boards increasingly expect to see the distribution, not just the point.
Reserving as a governed, continuous process
Reserve estimates move markets and careers, so the process around them matters as much as the mathematics. Assumptions — development factors, loss ratio priors, tail factors — should be versioned with a documented rationale; every run should be reproducible; and actual-versus-expected monitoring should compare emerging claims against what the reserves implied, every quarter, so deterioration is caught early rather than discovered at year end.
The direction of travel is continuous reserving: triangles that rebuild from live claims data instead of quarterly extracts, methods re-run automatically as diagnostics, and the actuary's time spent on judgement and communication rather than data assembly. The result is not just faster closes — it is reserves the appointed actuary can defend line by line.