What IBNR is, and why it is the largest number on an insurer's balance sheet
IBNR stands for incurred but not reported, and in practice it covers two distinct things. There are claims that have occurred and have not yet been reported to the insurer at all — pure IBNR. And there is the amount by which claims already reported will develop beyond their current case reserves, which is properly called IBNER, incurred but not enough reported. Most reserving methods, including the one here, estimate the two together, because the data cannot separate them.
The reserve matters more than almost anything else an insurer reports. Loss reserves are the largest liability on a property-casualty balance sheet, they are an estimate rather than a measurement, and every dollar of reserve is a dollar of reported profit. An insurer that under-reserves looks profitable for several years and then does not.
The chain ladder is the oldest and most widely used method for producing that estimate. Its premise is simple: whatever proportion of ultimate losses an accident year has reported by a given age, the next year will report a similar proportion by the same age. Measure that pattern on mature years, then apply it to immature ones.
This calculation feeds the rest of the actuarial chain on this site. Developed losses are what a rate indication needs before it can be built with the pure premium calculator, they are the incurred values that drive the workers comp experience modifier, and they are the losses in the loss ratio that decides whether a rate is adequate.
From age-to-age factors to a cumulative development factor
Start with a loss triangle: rows are accident years, columns are ages at evaluation. An age-to-age factor — also called a link ratio — is the ratio of cumulative losses at one age to cumulative losses at the previous age, calculated across the years that have reached both. A 12-to-24 factor of 1.55 means the average accident year reported 55% more by 24 months than it had by 12.
Multiply the remaining age-to-age factors together, plus the tail, and you get the cumulative development factor from any age to ultimate. For a year at 12 months with factors 1.55, 1.18, 1.08 and 1.03 and a 1.02 tail, that is 1.55 × 1.18 × 1.08 × 1.03 × 1.02 = 2.075271. Multiply reported losses by that and you have the estimated ultimate; subtract reported losses and you have the IBNR.
Two properties of the cumulative factor are worth internalising. It falls steeply with maturity — 2.075 at 12 months, 1.339 at 24, 1.135 at 36 — because most of the development happens early. And the fraction of ultimate still unreported is 1 − 1/CDF, so a CDF of 2.075 means 51.8% of that year's ultimate losses are not yet on the books. Half the answer for the newest year is coming from the factors rather than from the data.
The tail factor covers everything beyond the oldest age in your triangle. It is invisible in the data by construction, which is why it is estimated by fitting a curve to the observed factors, by reference to industry development patterns, or by extending the triangle with older data. It applies to every accident year, including the mature ones, so a tail that is 2% too low understates the entire reserve by roughly 2%.
Worked example: five accident years at 16,000,000 reported
An insurer has five accident years on its triangle. Reported losses are $4,200,000 at 60 months, $3,900,000 at 48, $3,600,000 at 36, $2,800,000 at 24 and $1,500,000 at 12. Selected age-to-age factors are 1.55, 1.18, 1.08 and 1.03, with a 1.02 tail.
- Build the cumulative factors. At 60 months only the tail remains: 1.02. At 48 months: 1.03 × 1.02 = 1.0506. At 36: 1.08 × 1.0506 = 1.134648. At 24: 1.18 × 1.134648 = 1.33888464. At 12: 1.55 × 1.33888464 = 2.075271192.
- Develop each year. $4,200,000 × 1.02 = $4,284,000. $3,900,000 × 1.0506 = $4,097,340. $3,600,000 × 1.134648 = $4,084,732.80. $2,800,000 × 1.33888464 = $3,748,876.99. $1,500,000 × 2.075271192 = $3,112,906.79.
- Total. Estimated ultimate losses are $19,327,856.58 against $16,000,000 reported.
- IBNR. $19,327,856.58 − $16,000,000 = $3,327,856.58.
- Share unreported. $3,327,856.58 ÷ $19,327,856.58 = 17.22% of ultimate losses are not yet on the books.
Now look at where the reserve comes from. The most recent year alone carries $3,112,906.79 − $1,500,000 = $1,612,906.79 of IBNR, which is 48.5% of the total reserve from a single year holding 9.4% of the reported losses. That concentration is the defining feature of chain-ladder estimates and the reason they are volatile: the newest year has the largest factor applied to the smallest, least stable base.
Test the leverage. Change the 12-to-24 factor from 1.55 to 1.60 — a 3.2% change in one selected factor — and the cumulative 12-month factor rises to 2.142215, so the newest year's ultimate becomes $3,213,323.14 and total IBNR rises to $3,428,272.93. A 3.2% move in one factor produced a 3.0% move in the whole reserve, and every dollar of that came from one year.
How to read the reserve
The share-of-ultimate-unreported figure is the first thing to look at, because it tells you how much of your answer is estimate rather than observation. Below about 15% the reserve is largely confirming what is already reported. Above 40% the method is heavily leveraged on the earliest development factors, and small changes in a selection you made by judgement move the whole answer.
The IBNR on the most recent year, taken against the total, tells you the same thing from a different angle. Where a single immature year drives most of the reserve, the chain ladder is at its least reliable, and the standard response is the Bornhuetter-Ferguson method, which anchors the immature years to an expected loss ratio applied to premium rather than extrapolating from a thin reported figure. The two methods agree closely on mature years and diverge exactly where it matters.
Compare the estimated ultimate for each year against each other as a sanity check. Absent a change in exposure or in rate adequacy, ultimate losses should be broadly stable from year to year. A jump in the newest year's estimated ultimate that is not explained by growth in exposure is more likely to be an artefact of the development factor than a real deterioration.
Finally, hold the reserve against premium. The estimated ultimate divided by earned premium is the ultimate loss ratio, and it is the figure that actually decides whether the business was written profitably — the reported loss ratio on an immature year is meaningless by comparison.
How the cumulative factor and the unreported share fall with maturity
| Age at evaluation | Remaining factors | Cumulative LDF | Reported as % of ultimate | Unreported % |
|---|---|---|---|---|
| 12 months | 1.55 × 1.18 × 1.08 × 1.03 × 1.02 | 2.075271 | 48.19% | 51.81% |
| 24 months | 1.18 × 1.08 × 1.03 × 1.02 | 1.338885 | 74.69% | 25.31% |
| 36 months | 1.08 × 1.03 × 1.02 | 1.134648 | 88.13% | 11.87% |
| 48 months | 1.03 × 1.02 | 1.050600 | 95.18% | 4.82% |
| 60 months | 1.02 (tail only) | 1.020000 | 98.04% | 1.96% |
Reported as a percentage of ultimate is 1 ÷ CDF, and the unreported share is its complement. The 12-month row is why a single year's reported losses tell you almost nothing about how that year will finish.
The chain ladder breaks precisely when you most want an answer
Every age-to-age factor is an average of the past, and the method assumes that past pattern continues. Three things break it, and all three tend to happen at once during a period of change. A change in case reserving practice — adjusters told to reserve more conservatively — raises reported losses early and lowers subsequent factors, so applying historical factors to the new pattern overstates the reserve. A change in claim settlement speed does the same in reverse. A change in mix or in limits alters the shape of development entirely. The tell is a triangle whose link ratios drift consistently in one direction down a column rather than fluctuating around a level. When you see that, the selected factor should not be a simple average of the column, and the reserve deserves a second method alongside it.
Assumptions and limits of this calculation
- One triangle, one set of factors. Real reserving runs paid and incurred triangles separately and reconciles them. Divergence between the two is one of the most useful diagnostics available and is invisible in a single-triangle calculation.
- Factors are inputs, not derived. This calculator applies factors you select. Deriving them from a triangle involves choosing between simple averages, weighted averages, medians and excluding outlier years, and that selection is where most of the actuarial judgement lives.
- No allowance for large or catastrophe losses. A single very large claim distorts a link ratio badly. Standard practice is to develop losses net of large claims and add a separate provision for them.
- No discounting. Reserves under US statutory accounting are generally carried undiscounted for most lines. If you need a present value, discount the payout pattern separately.
- No allocated loss adjustment expense. ALAE is usually developed on its own triangle with its own factors and added to the loss reserve.
- No variability estimate. The chain ladder produces a point estimate. A range or a distribution requires a stochastic method such as Mack's model or a bootstrap, and regulators increasingly expect one.
Chain ladder, Bornhuetter-Ferguson and where each belongs
The chain ladder trusts the data. It takes what an accident year has reported and scales it, which is exactly right for a mature year with a stable pattern and exactly wrong for a year at 12 months where a random early claim can move the estimate by millions.
The Bornhuetter-Ferguson method fixes that by splitting the difference. It estimates ultimate as reported losses plus expected losses multiplied by the unreported percentage, where expected losses come from premium times an a priori loss ratio. The immature year is therefore anchored to what you expected to happen rather than extrapolated from what has happened so far, and the estimate converges to the chain ladder as the year matures and the unreported percentage falls towards zero. The unreported percentage this calculator reports, 1 − 1/CDF, is exactly the weight the BF method applies to expected losses.
The expected loss ratio method is the limiting case: ignore reported losses entirely and set ultimate equal to premium times the expected loss ratio. That is the right answer for a brand-new line with no credible data at all, and a common practice is to move from expected loss ratio through Bornhuetter-Ferguson to chain ladder as each accident year matures.
Actuarial Standard of Practice No. 43 governs unpaid claim estimates in the United States and requires, among other things, that the actuary consider the sensitivity of the estimate to the methods and assumptions chosen. That is a formal way of saying what the leverage test in the worked example above shows: know how much of your reserve is coming from a factor you selected rather than from a loss that has actually been reported. For a business deciding how much of this volatility to keep rather than transfer, the self-insured retention break-even calculator works the other side of the same question, and the credibility weighting calculator formalises how much to trust thin data.