Why Loss Development Selections Matter
Losses for recent accident periods are often incomplete: claims may not yet have been reported, case reserves may change, and payments may continue after the evaluation date. A rate indication based on those losses therefore needs an estimate of their eventual cost. Loss development uses observed experience at earlier maturities to estimate losses at ultimate, while a filing should make the data, method, and judgment behind that estimate understandable.
Development is not a mechanical correction that always increases losses. Paid losses commonly accumulate over time, but reported or incurred losses can increase or decrease as claims develop, recoveries are recognized, or reserves are revised. A selected factor must be appropriate to the loss basis and maturity of the observation to which it is applied.
Triangles, Factors, and Ultimate Losses
Let \(C_{a,t}\) denote cumulative losses for origin period \(a\), evaluated at age \(t\). The origin period might be an accident year or another consistently defined period. “Cumulative” means losses accumulated through age \(t\), not the losses arising only during the latest interval. A triangle has fewer observations at older ages because recent periods have not yet matured.
For a particular interval from age \(t\) to age \(t+1\), an individual observed link ratio is \(C_{a,t+1}/C_{a,t}\), when the denominator is nonzero. A common volume-weighted estimate is the sum of later-age cumulative losses divided by the sum of earlier-age cumulative losses for the eligible periods. A simple average of individual ratios is another possible selection; it gives each period equal weight rather than weighting by its earlier cumulative losses.
Here \(T\) is the last observed age in the selected development pattern, \(\widehat{f}_{T,\text{ult}}\) is the selected tail factor from that age to ultimate, and \(\widehat{U}_a\) is the projected ultimate loss. If no further development is expected after age \(T\), the tail factor can be 1.000. Otherwise, its selection needs support, such as sufficiently mature historical experience or a documented external basis. Factors must be consistent with the loss definition, treatment of recoveries, evaluation dates, and valuation basis used for the triangle.
Worked Example: Selecting Factors From a Reported-Loss Triangle
A hypothetical insurer has the following cumulative reported losses, in thousands of dollars. Use volume-weighted age-to-age factors, select a 1.020 tail factor after 48 months, and estimate ultimate losses for a new period with 160 at 12 months.
Method. Calculate each link ratio from the eligible periods, select factors to three decimals, multiply them to obtain the 12-month CDF, and apply that CDF to the 12-month reported amount.
| Origin period | 12 months | 24 months | 36 months | 48 months |
|---|---|---|---|---|
| 1 | 100 | 150 | 180 | 198 |
| 2 | 120 | 174 | 209 | |
| 3 | 140 | 203 | ||
| 4 | 160 |
12-to-24-month factor. The eligible later-age total is \(150+174+203=527\); the corresponding earlier-age total is \(100+120+140=360\). Thus the volume-weighted ratio is \(527/360=1.463888\ldots\), selected as 1.464.
24-to-36-month factor. The later-age total is \(180+209=389\), and the earlier-age total is \(150+174=324\). The ratio is \(389/324=1.200617\ldots\), selected as 1.201.
36-to-48-month factor. There is one eligible observation, so the ratio is \(198/180=1.100\). The tail factor from 48 months is given as 1.020.
12-month CDF. Multiplying the selected factors gives \(1.464\times1.201=1.758264\); then \(1.758264\times1.100=1.9340904\); and \(1.9340904\times1.020=1.972772208\).
Ultimate estimate. The 12-month reported amount is 160, so the estimate is \(160\times1.972772208=315.64355328\) thousand dollars. Rounded only at the final step to two decimals, selected ultimate losses are $315.64 thousand. This is a projection, not a claim that the observed losses have already reached that amount. The key trap avoided is including a link ratio whose periods do not match the factor’s stated maturity.
Choosing and Applying a Development Method
The chain-ladder method applies selected development factors to the observed cumulative amount. It is most directly supported when development patterns are reasonably stable across periods and the observed amount is a useful indicator of ultimate loss. A sparse or unusually volatile triangle can make a factor unstable; a large link ratio from a small denominator, a change in claim handling, or a shift in the mix of business may require investigation rather than automatic acceptance.
The Bornhuetter-Ferguson method combines observed losses with an expected ultimate estimate. With a CDF from the current age, the estimated proportion of ultimate losses already emerged is \(1/\mathrm{CDF}\), under the selected development pattern. The method uses actual observed losses for that emerged portion and the expected ultimate for the not-yet-emerged portion. It is useful when early observed experience is not yet sufficiently credible or stable to drive the entire estimate; it depends on having a defensible expected ultimate.
Here \(E[U_a]\) denotes the expected ultimate loss estimate used as the prior basis for the undeveloped portion. This notation represents an actuarial estimate, not necessarily a mathematical expectation derived from a probability model. The formula presumes a meaningful CDF and a suitable expected ultimate on the same loss and period basis as the observed losses. It should not be used without explaining where that expected ultimate came from.
Worked Example: Estimating Paid Ultimate at Two Maturities
A hypothetical insurer selects paid-loss factors of 1.500 from 12 to 24 months, 1.250 from 24 to 36 months, and 1.080 from 36 months to ultimate. A new period has cumulative paid losses of 70 at 12 months and 126 at 24 months, in thousands of dollars. Estimate ultimate losses from each observation.
Method. Paid losses and all selected factors must refer to the same payment basis. For the 12-month observation, the CDF is \(1.500\times1.250\times1.080=1.875\times1.080=2.025\). Its chain-ladder estimate is \(70\times2.025=141.75\) thousand dollars.
For the 24-month observation, the applicable CDF begins at 24 months, not 12 months. It is \(1.250\times1.080=1.350\), so the estimate is \(126\times1.350=170.10\) thousand dollars. Rounded to two decimals, the respective estimates are $141.75 thousand and $170.10 thousand.
The two estimates need not agree: each uses a different observed amount and maturity, and the example supplies no reason to assume that the observations have identical experience. A common error is to apply a CDF from the wrong age. That is incorrect, but its direction is not universal. If an earlier-age CDF is larger than the CDF for the observation’s actual age, using it overstates the estimate; if it is smaller, using it understates the estimate.
A Short Justification of the Bornhuetter-Ferguson Formula
The Bornhuetter-Ferguson calculation follows by splitting ultimate loss into the portion already emerged and the portion not yet emerged. Under the selected development pattern, the estimated emerged proportion at age \(t\) is \(1/\mathrm{CDF}_t\). The corresponding un-emerged proportion is therefore \(1-1/\mathrm{CDF}_t\). Applying that proportion to expected ultimate gives an estimate of the losses still to emerge. Adding observed cumulative losses produces the formula:
For a CDF of 2.000, the emerged share is \(1/2.000=0.500\), and the un-emerged share is \(1-0.500=0.500\).
Multiply expected ultimate by the un-emerged share. This allocates the prior estimate to the part of ultimate that has not yet appeared.
The method uses actual reported or paid losses for the emerged portion, rather than replacing them with an expected amount.
This calculation does not assert that the observed amount is exactly the emerged share of ultimate. It is an estimation method whose assumptions concern the selected development pattern and the expected ultimate for the remaining portion.
Worked Example: Applying Bornhuetter-Ferguson to Early Reported Losses
A hypothetical coverage has reported losses of 180 at 12 months, in thousands of dollars. The selected 12-month CDF is 2.000, and the expected ultimate for that period is 500 thousand dollars. Estimate ultimate using Bornhuetter-Ferguson and compare it with chain ladder.
Bornhuetter-Ferguson method. The estimated emerged proportion is \(1/2.000=0.500\), so the un-emerged proportion is \(1-0.500=0.500\). Estimated un-emerged losses are \(500\times0.500=250\) thousand dollars. Adding the observed 180 gives \(180+250=430\) thousand dollars. The Bornhuetter-Ferguson estimate is $430.00 thousand.
Chain-ladder comparison. Applying the same CDF directly to observed losses gives \(180\times2.000=360\) thousand dollars. Thus, the chain-ladder estimate is $360.00 thousand, which is $70.00 thousand below the Bornhuetter-Ferguson estimate. The difference arises because chain ladder projects the observed amount, while Bornhuetter-Ferguson uses the expected ultimate to estimate the un-emerged portion. The trap avoided is treating the expected ultimate as an amount to add in full: only its estimated un-emerged share, \(250\), is added.
The arithmetic is consistent: chain ladder treats the 180 as half of ultimate, implying a further 180; Bornhuetter-Ferguson adds 250 based on the expected ultimate, for a total of 430. The method choice requires an explanation of the expected ultimate and why observed early losses should or should not dominate the estimate.
Explaining Selections in Filing Support
A useful filing exhibit should allow a reviewer to trace the estimate from observed losses through selected factors to ultimate losses. The narrative should identify the evaluation date, origin periods, loss basis (such as paid or reported), and any adjustments needed to make periods comparable. It should also show the eligible observations, selected link ratios or factors, tail treatment, and the resulting CDFs. Explain material deviations from a mechanical average; a selection is more persuasive when its rationale is visible than when it is presented as an unexplained judgment.
Selection considerations include the stability of factors across periods, the amount of data supporting each factor, the relevance of older experience, and whether changes in claim practices or coverage affect comparability. A volume-weighted factor gives greater influence to periods with larger earlier-age losses; a simple average gives equal influence to each eligible period. Neither choice is universally correct. The filing should explain why the method is suitable for the data and disclose significant limitations, such as sparse older-age observations or a tail factor based on limited experience.
Reported and paid development patterns should not be interchanged without justification. Reported losses include amounts recognized as incurred under the selected reporting basis; paid losses reflect cash payments to date. Their timing and development can differ. Any expected ultimate used in Bornhuetter-Ferguson should also be defined on a consistent basis. Development adjustments estimate ultimate losses at the experience period’s valuation basis; they do not, by themselves, project losses to a different future cost level.
Common Mistakes and Review Checks
- Using incremental losses as if they were cumulative. Age-to-age ratios and CDFs are ordinarily applied to cumulative losses at a stated maturity. Confirm the triangle’s basis before calculating.
- Using a factor from the wrong maturity. Identify the age of the observed amount first, then use the CDF beginning at that age. A mismatch can overstate or understate ultimate, depending on the selected factors.
- Assuming development must be positive. Reported losses may decrease. A factor below 1.000 can be valid when supported by the data and the underlying loss basis.
- Presenting a mechanical result without selection rationale. Check unusual ratios, sparse cells, changes in data or claim handling, and the basis for the tail. Explain material judgment rather than implying that the data determine a unique answer.
- Rounding too early. Retain precision through the factor product and ultimate calculation; round the reported result only at the end.
As a final check, recalculate each selected CDF from its component factors, confirm that its starting maturity matches the observed loss amount, and verify that the resulting ultimate has the stated units and loss basis. These checks make the estimate both more reliable and easier to review.
Check Your Understanding
Work each problem using the stated loss basis and maturity. Answers are rounded only where indicated.
- A reported-loss observation is 250 at 24 months. Selected factors are 1.180 from 24 to 36 months, 1.060 from 36 to 48 months, and 1.020 from 48 months to ultimate. Calculate the 24-month CDF and estimated ultimate. (Answer: CDF \(=1.275816\); ultimate \(=318.95\) in the same units.)
- Cumulative paid losses are 45 at 12 months. Selected paid factors are 1.400 from 12 to 24 months, 1.200 from 24 to 36 months, and 1.050 from 36 months to ultimate. What is the chain-ladder ultimate? (Answer: CDF \(=1.764\); ultimate \(=79.38\) in the same units.)
- Two eligible periods have cumulative losses of 100 and 200 at an earlier age, and 130 and 240 at the later age. Calculate the volume-weighted factor and apply it to a current cumulative amount of 300 at the earlier age. (Answer: factor \(=370/300=1.2333\); projected later-age amount \(=370\).)
- At 12 months, reported losses are 220, the selected CDF is 2.500, and expected ultimate is 800, all in thousands of dollars. Calculate the Bornhuetter-Ferguson estimate. (Answer: \(220+800(1-1/2.500)=700\) thousand dollars.)
- A 24-month reported amount is 400. The selected 12-to-24-month factor is 0.950 and the selected 24-month-to-ultimate factor is 1.100. Calculate the correct ultimate and the result from mistakenly applying the 12-month CDF to the 24-month amount. State the direction of the error. (Answer: correct \(=400(1.100)=440\); incorrect \(=400(0.950\times1.100)=418\), an understatement of 22.)