What is IBNR? Incurred but not reported reserves explained with a worked example

IBNR stands for incurred but not reported. It is the part of a non-life insurer’s claims reserve that pays for claims which have already happened but which the insurer does not yet know about, or does not yet know the full cost of. This article defines the term, puts it next to case reserves and IBNER, works through a chain-ladder calculation on a small paid triangle, and shows where the figure ends up under Solvency II: inside the best estimate of the claims provision, in template S.17.01 and in the S.19.01 claims triangles. The example is the same kind of calculation that TP Tool runs on your own claim transactions, only much smaller. Actuaries will know the arithmetic. For a CFO or an auditor who signs off on the number, the point is to see how few inputs it rests on.

IBNR and its place in the claims reserve

At any valuation date a non-life insurer owes money on claims in three states. Some claims have been reported and are still open; a claim handler has set a case reserve on each, also called the reported but not settled (RBNS) reserve. Some claims have happened but no file exists yet. And some reported claims will cost more, or less, than the case reserve, because the injury turns out worse or a closed claim is reopened.

The Institute and Faculty of Actuaries’ Claims Reserving Manual defines IBNR as claims not yet known to the insurer but for which a liability is believed to exist at the reserving date. That is the pure, or “true”, IBNR. The manual uses IBNER for the second item, the development on claims already reported, and expands it as “incurred but not enough reserved”; most practitioners say “incurred but not enough reported”, which means the same thing. In practice the two are rarely estimated separately. A triangle method projects the total ultimate cost of each accident year and does not know why the case reserves fell short, so the IBNR figure it produces is the broad one:

Outstanding claims provision = case reserves + IBNR (including IBNER)

Under Solvency II this whole amount, together with the expenses of settling it, is the best estimate of the claims provision. Delegated Regulation (EU) 2015/35 keeps it separate from the premium provision: the claims provision relates to claim events that have already occurred, whether or not they have been reported, and the premium provision relates to claim events that have not yet happened on contracts the insurer is already bound to. IBNR belongs to the first.

Why IBNR exists

Two things create it. The first is reporting lag. A motor property damage claim is usually reported within days. A liability claim, an occupational disease or a bodily injury claim can surface months or years after the accident, and a valuation at 31 December always misses the December accidents nobody has reported by the time the books close. The second is late development. Case reserves are set with the information available on the day; litigation runs its course, and a claim closed for 5,000 is reopened at 50,000.

For short-tail lines such as property the IBNR is a small fraction of the case reserves. For long-tail liability lines it can exceed them, sometimes by a wide margin, which is why the IBNR is the number an auditor or a supervisor asks about first.

A worked example: chain ladder on a paid triangle

The chain-ladder method takes the historical pattern of payments and assumes the open years will follow it. Here is a cumulative paid loss triangle for one homogeneous risk group, four accident years and four development years, in EUR thousand. Row 2022 is fully developed, row 2025 has only its first year.

The development factors are volume weighted: for each step, the sum of the later column divided by the sum of the earlier column, using only the rows that have both values.

Factor from year 1 to year 2 = (1,500 + 1,800 + 1,650) / (1,000 + 1,200 + 1,100) = 4,950 / 3,300 = 1.500

Factor from year 2 to year 3 = (1,800 + 2,160) / (1,500 + 1,800) = 3,960 / 3,300 = 1.200

Factor from year 3 to year 4 = 1,980 / 1,800 = 1.100

We assume no payments after development year 4, so the tail factor is 1.000. On a real liability portfolio a tail would be fitted; here it keeps the example short.

Each open accident year is then carried from its latest diagonal to ultimate by multiplying through the remaining factors.

2022: fully developed, ultimate 1,980

2023: 2,160 × 1.100 = 2,376

2024: 1,650 × 1.200 × 1.100 = 2,178

2025: 1,300 × 1.500 × 1.200 × 1.100 = 2,574

The outstanding provision is the ultimate less what has been paid so far. To get from there to IBNR we need the case reserves the claim handlers hold at the valuation date, which do not come from the triangle. Here they are 150 for 2023, 400 for 2024 and 900 for 2025; 2022 has no open claims.

So this risk group needs an outstanding claims provision of 2,018, of which 1,450 is already sitting in the case reserves and 568 is IBNR. The youngest year carries about two thirds of it (374 out of 568), which is normal: it has been multiplied by three factors, its cumulative factor to ultimate is 1.98, and only half of its final cost has been paid.

That last observation is also the weakness of the method. If the first factor had been 1.550 instead of 1.500, the 2025 ultimate would move from 2,574 to 2,660 and the total IBNR from 568 to 654, a rise of 15 per cent from a change in one factor that the data can barely distinguish. This is why reserving actuaries look at the individual age-to-age ratios before accepting the weighted average, and run a Bornhuetter-Ferguson estimate next to the chain ladder for the youngest years. Bornhuetter-Ferguson blends the development pattern with an expected loss ratio, so a single large early payment is not multiplied all the way to ultimate.

IBNR under local GAAP and under the Solvency II best estimate

The 568 above is a management figure. It is undiscounted, it holds no explicit expense loading, and depending on the local accounting rules the insurer may add a margin of prudence on top before it books the provision. Many local GAAP frameworks allow or require exactly that.

Solvency II asks for something different. Article 77 of Directive 2009/138/EC sets the technical provisions equal to a best estimate plus a risk margin, and defines the best estimate as the probability-weighted average of future cash flows, taking account of the time value of money using the relevant risk-free interest rate term structure. The consequences for the IBNR are these:

The provision is discounted. The 2,018 outstanding is spread over future payment years, and each year is discounted at the EIOPA risk-free curve for the currency. For a short-tail group the effect is small; for a long-tail liability group it is not, and the discount unwinds every year.

There is no prudence margin inside the best estimate. Any margin sits in the risk margin, which is calculated separately on a cost-of-capital basis. An actuary who is used to “reserving conservatively” has to strip that habit out of the best estimate and let the risk margin carry it.

The cash flows must include everything needed to settle the claims, so claims handling expenses are added, and the projection must allow for events not in the data (ENID): outcomes that have never appeared in the triangle, such as a new type of latent claim, which a projection of history cannot produce. The Bank of England’s supervisory statement SS5/14 says so for UK firms and the EIOPA Guidelines on the valuation of technical provisions for the EU; a flat percentage uplift without a rationale does not pass.

The best estimate is gross; reinsurance recoverables are shown separately, adjusted for the reinsurer’s default risk.

In our example the Solvency II claims provision would therefore start from the 2,018 undiscounted outstanding, add the expected claims handling expenses and an ENID allowance, and discount the result. The split into case reserves and IBNR still matters for management and for the local accounts, but the Solvency II balance sheet shows one number: the best estimate of the claims provision.

Where IBNR lands in the Solvency II templates

Two templates carry it. Template S.17.01, Non-life Technical Provisions, is reported quarterly and annually per line of business and shows the gross best estimate split into premium provisions and claims provisions, the reinsurance recoverables, the net figures, and the risk margin. The IBNR has no row of its own; it is inside the best estimate of the claims provisions. The template is also reported separately for each ring-fenced fund and for the remaining part, so an insurer with a ring-fenced fund under Solvency II files it more than once. The annual solo variant is S.17.01.01.

Template S.19.01, Non-life insurance claims, is the annual template with the triangles themselves. For each material line of business, by accident year or underwriting year, it asks for three triangles going back up to 15 years plus prior: gross claims paid, the gross undiscounted best estimate of the claims provisions, and the gross RBNS. The triangle cells are undiscounted; the “year end” column is the latest diagonal on a discounted basis. Because the RBNS triangle explicitly excludes IBNR, the difference between the best estimate diagonal and the RBNS diagonal is the closest thing to an IBNR line in the whole QRT set. The annual solo variant is S.19.01.01, and validation rules cross-check its best estimate totals against S.17.01.

The claims provision also feeds the capital calculation. The best estimate of the net claims provisions per line of business is the volume measure for reserve risk in the non-life underwriting risk module, which is one of the larger blocks of the SCR standard formula calculation. An IBNR that is set too high therefore costs capital twice: once in the technical provisions and once in the SCR.

Where this lands in the software

TP Tool starts one step earlier than the triangle. Claim and premium transactions are loaded as they are, checked for completeness and inconsistencies, and grouped in the Risk Group Designer into the homogeneous risk groups the actuary wants to reserve. The Statistical Engine then builds the run-off triangles for each group and runs Chain-Ladder, Bornhuetter-Ferguson and fixed loss ratio estimates side by side, with the age-to-age factors and diagnostics visible, so the sensitivity we saw in the example above is on screen rather than in a spreadsheet. Discounting, inflation and claims handling expenses are applied in the same run, and the results populate S.17.01 and S.19.01 together with S.05.01, S.18.01, S.20.01, S.21.01, S.28.01, S.29.02 and S.29.03, with ECB currency conversion as at the reference date.