IBNR Triangles: A Practical Guide for Actuaries
Incurred-but-not-reported reserves start with a grid of numbers most people call a triangle. Here's how to build one, read it, and know when to distrust it.
IBNR — incurred but not reported — is the reserve an insurer holds for claims that have already happened but haven't yet been reported, plus the further development expected on claims that have been reported but aren't fully settled. It sits alongside the case reserve on outstanding claims to make up the full liability for incurred claims, and for long-tail lines it can be the larger of the two. Almost every method for estimating it starts with the same tool: a loss triangle.
Building the triangle
A loss triangle lays claims out on two axes. Rows are accident periods (or underwriting periods) — usually years or quarters, depending on volume. Columns are development periods — how many months or years have passed since the start of that accident period. Each cell holds a cumulative value, either paid claims or incurred claims (paid plus case reserves), as of that development point.
The shape is a triangle rather than a rectangle because older accident periods have had longer to develop and so have more columns filled in, while the most recent period only has one or two data points. That asymmetry is the whole reason estimation is needed in the first place — the newest periods are, by construction, the least mature.
Reading development factors
From the triangle, actuaries calculate age-to-age development factors: for each development period, the average (or selected) ratio of claims at the next stage of maturity to claims at the current stage, across all accident periods that have reached both points. A factor of 1.15 at the 12-to-24-month mark means claims typically grow by 15% between those two points in their life.
Multiplying an accident period's most recent observed value by the chain of remaining development factors projects it out to an assumed ultimate. Because very long-tail lines may still be developing beyond the observed data, a tail factor is often selected to capture development past the edge of the triangle — usually the least certain part of the whole exercise, since by definition there's no direct data to calibrate it against.
The triangle is a summary, not a fact. Every cell is the sum of individual claims that behaved very differently from one another. A triangle that looks smooth can still be masking a handful of large claims that happen to offset each other in aggregate — until they don't, in the next accident period.
What distorts a triangle
Development factors assume that history is a reasonable guide to the future. Several things quietly break that assumption:
- Large losses. A single large claim can dominate a whole accident period's development pattern. Many actuaries separate large losses out and reserve them individually, running the triangle on the "attritional" remainder to keep the bulk of the data statistically well-behaved.
- Changes in claims handling. A faster or slower claims process, a change in case reserving philosophy, or a new claims system can shift how quickly claims move through the triangle — and a development-factor method has no way to distinguish "claims are settling faster" from "claims are actually smaller."
- Reinsurance recoveries. Gross and net triangles can develop differently if recovery timing lags claim settlement, so it matters which one is being reserved and reported.
- Inflation, including claims inflation. Rising repair costs, medical costs, or legal costs can make older accident periods look like they're developing "normally" in nominal terms while actually understating the real trend, particularly across periods of unstable inflation.
From triangle to IBNR estimate
The IBNR reserve itself is simply the projected ultimate for each accident period, less what's already been paid and less the case reserve already held — chain ladder and Bornhuetter-Ferguson both ultimately produce an ultimate loss estimate, and IBNR falls out as the residual. What differs between methods is how much weight goes on the triangle's own development pattern versus an external prior, particularly for the least mature periods.
The problem most insurers actually have
None of the mechanics above are unusual — they're standard actuarial technique. What causes real problems in practice is timing: triangles that only get rebuilt at quarter-end, from a manually pulled claims extract, mean the reserve view in front of management is already weeks old by the time it's reviewed, and months old by the time it's acted on. A large claim reported the week after the last extract simply isn't in the triangle yet.
Keeping the triangle current as claims data changes — rather than rebuilding it from scratch each period — means development factors, large-loss splits, and IBNR estimates can reflect this week's claims activity, not last quarter's. It also makes it far easier to spot early when a period's development is deviating from pattern, while there's still time to investigate why.