From Observed Triangle to Reserve
A chain-ladder reserve estimate follows a sequence: check the observed triangle, select age-to-age development factors, project each incomplete accident period to an ultimate loss, and subtract the amount already paid. Keeping these steps separate makes the result easier to audit and helps expose errors before they flow into the total reserve.
The examples use hypothetical cumulative paid losses, stated in thousands of dollars. No interest discounting, mortality assumption, or severity distribution is used: this is a development-based projection of aggregate losses. Where a triangle is treated as complete at its final age, that is an explicit modeling assumption.
The age-to-age observations must be defined consistently. If \(C_{i,j}\) is the cumulative loss for accident period \(i\) at development age \(j\), an observed link ratio is \(C_{i,k}/C_{i,j}\), where the same accident period has observations at both ages. A common selection is the volume-weighted factor: the sum of the later-age amounts divided by the sum of the corresponding earlier-age amounts. It gives more weight to accident periods with larger earlier cumulative losses.
Here \(I_{j,k}\) is the set of accident periods with observed cumulative amounts at both ages, \(j_i\) is the latest observed age for accident period \(i\), \(\widehat U_i\) is its projected ultimate, and \(\widehat R_i\) is its estimated unpaid reserve. The product contains only the factors needed to move from that period’s latest age to ultimate.
Check the Triangle Before Selecting Factors
First confirm that the entries are cumulative amounts, that the development ages are correctly ordered, and that each accident period’s observations align with those ages. For ordinary positive paid-loss data, cumulative amounts should not decrease as development age increases. If incremental payments are supplied instead, accumulate them within each accident period before calculating cumulative age-to-age factors. A triangle may have different numbers of observed ages by accident period; that is expected when the valuation date cuts across the data.
A useful check is the valuation diagonal: identify the latest observed cell for every accident period. Those cells represent the cumulative paid losses available at the valuation date. Do not use an earlier cell as the current paid amount for one period when a later cell is observed. Also distinguish row checks from diagonal comparisons. Cumulative amounts within a row should generally be nondecreasing; amounts across different accident periods need not be, because periods can have different exposure and claim levels.
Worked Example: Convert Incremental Payments and Reconcile the Triangle
A hypothetical insurer has the following incremental paid losses, in thousands of dollars, at development ages 12, 24, and 36 months. Accident period A has increments 80, 40, and 20; period B has 90 and 45; period C has 100. Assume no payments are negative and use the volume-weighted factor at each available transition. Treat age 36 as ultimate for this example; there is no tail factor.
Method and data check. Convert increments to cumulative amounts by adding within each accident period. Period A gives \(80\), \(80+40=120\), and \(120+20=140\). Period B gives \(90\) and \(90+45=135\). Period C has \(100\) at age 12. The cumulative triangle is therefore A: \(80,120,140\); B: \(90,135\); C: \(100\). The observed rows do not decrease.
Factor from 12 to 24 months. Both A and B have observations at both ages. The factor is \((120+135)/(80+90)=255/170=1.5\).
Factor from 24 to 36 months. Only A has both observations. The factor is \(140/120=7/6=1.166666\ldots\).
Project each period. A is observed at age 36, assumed ultimate, so its ultimate is \(140\) and its reserve is \(140-140=0\). B is observed at age 24, so its ultimate is \(135(7/6)=157.5\) and its reserve is \(157.5-135=22.5\). C is observed at age 12, so its ultimate is \(100(1.5)(7/6)=175\) and its reserve is \(175-100=75\).
Reconcile. The total projected ultimate is \(140+157.5+175=472.5\). The latest observed cumulative paid total is \(140+135+100=375\). Their difference is \(472.5-375=97.5\), which agrees with the sum of accident-period reserves, \(0+22.5+75=97.5\). Amounts are in thousands of dollars; no rounding is needed.
Common mistake avoided: Treating the supplied increments as if they were cumulative would give incorrect link ratios. The factors and reserve calculations require cumulative amounts at the stated development ages.
Why the Projection Formula Works
The chain-ladder calculation applies the selected development relationship successively. If a cumulative amount \(C_{i,j}\) is projected from age \(j\) to age \(k\), the selected factor estimates its later cumulative amount as \(C_{i,j}\widehat f_{j,k}\). If another development step remains, that projected amount is multiplied by the next factor. Repeating this process produces the product of factors in the ultimate formula.
The reserve identity follows directly from the meaning of unpaid losses: ultimate losses consist of amounts already paid plus amounts still unpaid. Under the projection, \(\widehat U_i=C_{i,j_i}+\widehat R_i\), so rearranging gives \(\widehat R_i=\widehat U_i-C_{i,j_i}\). Summing this identity across accident periods also gives a reconciliation check: total reserve equals total ultimate less total latest observed cumulative paid losses.
The calculation is a projection, not a guarantee that each accident period will develop exactly according to the selected factors. It relies on the selected development pattern being suitable for the periods being projected. Sparse observations, changes in claims handling, or non-comparable data can make a mechanically correct calculation a poor estimate.
Complete Chain-Ladder Projection by Accident Period
The next example uses a four-age cumulative paid triangle. It shows how the selected factors, current valuation diagonal, ultimate losses, and reserves fit together. The factors are volume-weighted, and the final observed age of 48 months is assumed to be ultimate, so no additional development beyond that age is included.
Worked Example: Project a Paid Triangle to Ultimate
A hypothetical insurer reports cumulative paid losses in thousands of dollars at 12, 24, 36, and 48 months:
| Accident period | 12 months | 24 months | 36 months | 48 months |
|---|---|---|---|---|
| A | 100 | 150 | 180 | 198 |
| B | 120 | 180 | 216 | |
| C | 140 | 210 | ||
| D | 160 |
Assume losses are positive, the displayed amounts are cumulative paid losses, and age 48 is ultimate. Select volume-weighted age-to-age factors and calculate each period’s ultimate and unpaid reserve. No interest discounting, mortality assumption, or severity distribution is needed.
Step 1: Check the observed data. Each accident-period row is nondecreasing: A has \(100\leq150\leq180\leq198\); B has \(120\leq180\leq216\); and C has \(140\leq210\). The latest observed cells are A at 48 months with 198, B at 36 months with 216, C at 24 months with 210, and D at 12 months with 160. These are the valuation-diagonal amounts.
Step 2: Calculate the 12-to-24 factor. The paired observations are A, B, and C. The sum at 24 months is \(150+180+210=540\), and the corresponding sum at 12 months is \(100+120+140=360\). Thus \(\widehat f_{12,24}=540/360=1.5\).
Step 3: Calculate the 24-to-36 factor. The paired observations are A and B. The sum at 36 months is \(180+216=396\); the corresponding sum at 24 months is \(150+180=330\). Thus \(\widehat f_{24,36}=396/330=1.2\).
Step 4: Calculate the 36-to-48 factor. Only A has both ages. Therefore \(\widehat f_{36,48}=198/180=1.1\). This factor is based on one observed pair, so its calculation is straightforward but the limited basis should be kept in mind when assessing whether it is representative.
Step 5: Project ultimates and reserves. A is already at the assumed ultimate age: \(\widehat U_A=198\), and \(\widehat R_A=198-198=0\). B needs the 36-to-48 factor: \(\widehat U_B=216(1.1)=237.6\), so \(\widehat R_B=237.6-216=21.6\). C needs two factors: \(\widehat U_C=210(1.2)(1.1)=277.2\), so \(\widehat R_C=277.2-210=67.2\). D needs all three factors: \(\widehat U_D=160(1.5)(1.2)(1.1)=316.8\), so \(\widehat R_D=316.8-160=156.8\).
Step 6: Reconcile the totals. The projected ultimate total is \(198+237.6+277.2+316.8=1029.6\). The latest observed cumulative paid total is \(198+216+210+160=784\). Therefore the total unpaid reserve is \(1029.6-784=245.6\). The period reserve sum is \(0+21.6+67.2+156.8=245.6\), agreeing exactly. The values are in thousands of dollars; displayed results are exact to two decimal places.
Common mistake avoided: The 24-to-36 denominator is the sum of the paired 24-month amounts, \(150+180=330\), not the sum of all 24-month observations, which would incorrectly include C even though C has no corresponding 36-month amount. Use the same accident periods in a link-factor numerator and denominator.
Factor Selection and Its Effect
Volume-weighted factors are convenient because they use the combined experience of all eligible accident periods. A simple arithmetic average of individual link ratios is another possible selection, but it gives each eligible accident period equal weight regardless of its size. The methods need not agree. Neither should be applied without checking whether the underlying periods and data are sufficiently comparable for the intended projection.
Worked Example: Compare Two Selections for an Early Development Factor
A hypothetical cumulative paid triangle, in thousands of dollars, is A: 100 at 12 months and 150 at 24 months and 180 at 36 months; B: 120 at 12 months and 168 at 24 months; C: 140 at 12 months only. Assume age 36 is ultimate. Use the simple average of observed 12-to-24 link ratios for one estimate, and the volume-weighted factor for a second estimate. Use the observed 24-to-36 factor for both.
Check the relevant observations. The 12-to-24 ratios are \(150/100=1.5\) for A and \(168/120=1.4\) for B. The volume-weighted factor is \((150+168)/(100+120)=318/220=1.4454545\ldots\). The simple average is \((1.5+1.4)/2=1.45\). The 24-to-36 factor is \(180/150=1.2\), based on A.
Project using the volume-weighted factor. A is at age 36, so its ultimate is 180 and reserve is zero. B is at age 24, so its ultimate is \(168(1.2)=201.6\), with reserve \(201.6-168=33.6\). C is at age 12, so its ultimate is \(140(318/220)(1.2)=242.83636\ldots\), or 242.84 to two decimals; its reserve is \(242.83636\ldots-140=102.83636\ldots\), or 102.84. The total reserve is \(0+33.6+102.83636\ldots=136.43636\ldots\), or 136.44.
Project using the simple average. A remains at ultimate 180 with zero reserve, and B remains at ultimate 201.6 with reserve 33.6. C’s ultimate is \(140(1.45)(1.2)=243.6\), so its reserve is \(243.6-140=103.6\). The total reserve is \(0+33.6+103.6=137.2\).
The two total reserve estimates differ by \(137.2-136.43636\ldots=0.76363\ldots\) thousand dollars, or 0.76 thousand dollars to two decimals. The difference arises from the selected early factor, not from a change in the observed latest paid amounts. This comparison is a calculation check, not a rule that one selection is always preferable. No discounting, mortality assumption, or severity distribution is used.
Common mistake avoided: Averaging the ratios and calculating the ratio of sums are not interchangeable. State the selection method, calculate it from the matching observations, and apply it consistently to each projected period.
Practical Checks and Exam Traps
- Use cumulative losses for link ratios. If the input is incremental, accumulate it first. Do not mix increments in one part of the calculation and cumulative amounts in another.
- Use paired observations. A factor from one age to the next uses only periods observed at both ages; incomplete pairs cannot contribute to that link ratio.
- Start each projection at the latest observed age. A period already at the assumed ultimate age needs no additional factor. Applying an earlier factor again double-counts development.
- Subtract the current cumulative paid amount. The reserve is not the ultimate itself. It is ultimate less the latest paid cumulative amount for that accident period.
- Check the total twice. Sum the period reserves and compare with total projected ultimate less total latest observed paid. A mismatch usually signals a missing or extra factor, an incorrect valuation-diagonal amount, or an arithmetic error.
The method also depends on the selected development pattern being appropriate for the accident periods being projected. A factor supported by only one pair, as in the final-age link in the complete example, is numerically calculable but rests on little observed experience. The chain-ladder arithmetic does not remove the need to assess data consistency and factor credibility.
Check Your Understanding
Work each problem by identifying the latest observed cumulative amount before projecting. Amounts are hypothetical and stated in thousands of dollars unless indicated otherwise.
- A cumulative paid triangle has accident period A at 12 and 24 months: 50 and 75; B at 12 and 24 months: 70 and 105. Using the volume-weighted 12-to-24 factor, what is the factor? (Answer: 1.5000)
- A period has cumulative paid loss 240 at 24 months. Selected factors are 1.25 from 24 to 36 months and 1.10 from 36 to ultimate. Calculate its ultimate and reserve. (Answer: Ultimate 330; reserve 90)
- A triangle’s paired observations for a link are 12-month amounts 80 and 120, with corresponding 24-month amounts 100 and 150. Calculate the volume-weighted factor. (Answer: 1.2500)
- At a valuation date, three periods have latest cumulative paid amounts 100, 160, and 200. Their projected ultimates are 100, 200, and 260. Calculate the reserve for each period and reconcile the total. (Answer: Reserves 0, 40, and 60; total reserve 100)
- A period has cumulative paid loss 90 at 12 months. Selected factors are 1.4 from 12 to 24 months and 1.25 from 24 months to ultimate. A candidate applies only the second factor. Calculate the correct ultimate and reserve, then identify the candidate’s error. (Answer: Ultimate 157.50; reserve 67.50; the candidate omitted the 12-to-24 factor)