Why Development Pattern Stability Matters
Chain ladder projects losses using development factors selected by development age. That approach is most persuasive when accident years have broadly similar development patterns: for a given pair of development ages, their age-to-age link ratios should be reasonably consistent. If ratios differ, the variation may be ordinary fluctuation in observed losses—or it may indicate that the underlying pattern has changed. The distinction matters because applying one common factor to materially different accident years can misstate projected ultimates.
This tutorial focuses on interpreting variation across accident years. It does not make a formal statistical test of link ratios or identify the cause of a pattern shift. An observed ratio is evidence to evaluate, not proof that the underlying process has changed.
The comparison must be made at the same development age. A ratio from 12 to 24 months describes a different stage of claim development from a ratio from 24 to 36 months. Their values need not be similar, even when every accident year follows the same development pattern.
A Model for a Common Pattern
A simple representation of an exact common development pattern is \(C_{i,j}=U_iG_j\), where \(U_i\) is the scale of accident year \(i\), and \(G_j\) is the cumulative development factor through age \(j\). Assume \(U_i>0\) and \(G_j>0\) at the ages considered. In particular, \(U_i\ne0\), so the cumulative amount in the denominator is nonzero and the ratio below is defined.
The accident-year scale cancels, leaving an age-specific factor that does not depend on \(i\). This is why the model implies a common link ratio across accident years at a given development age. In practice, cumulative losses do not follow this exact equation perfectly. Random variation, changes in claim handling, changes in business mix, and other influences can produce observed ratios that differ from one another.
It is useful to distinguish the model’s underlying factor from an observed link ratio. The model explains what a common pattern would imply; it does not guarantee that every observed ratio equals the underlying factor. Nor does a ratio near the others prove stability. The assessment is a reasoned judgment based on the amount and structure of variation in the available data.
Choose one age-to-age interval, such as 12 to 24 months. Do not mix it with another interval.
For each accident year with both cumulative amounts available, divide the later-age cumulative loss by the earlier-age cumulative loss.
Consider whether the ratios are scattered around a broadly similar level or whether they form a sustained higher or lower group across accident years.
Describe a common factor as plausible, questionable, or poorly supported by the observed pattern. Do not claim that the ratios alone identify a cause.
Worked Comparisons
Worked Example: Scattered Variation at One Development Age
A hypothetical insurer reviews cumulative losses, in thousands of dollars, from 12 to 24 months for four accident years. Assess whether a common factor is plausible from these ratios. No interest or mortality adjustment is being made, and no probability distribution is assumed; the task is a descriptive comparison of the stated triangle values.
| Accident year | 12-month cumulative losses | 24-month cumulative losses | 12-to-24 link ratio |
|---|---|---|---|
| A | 1,000 | 1,200 | To calculate |
| B | 2,000 | 2,460 | To calculate |
| C | 1,500 | 1,740 | To calculate |
| D | 800 | 1,008 | To calculate |
Method: calculate each ratio at the same development age, then inspect their spread.
Accident year A: \(1{,}200/1{,}000=1.20\). Accident year B: \(2{,}460/2{,}000=1.23\). Accident year C: \(1{,}740/1{,}500=1.16\). Accident year D: \(1{,}008/800=1.26\). The ratios range from 1.16 to 1.26.
As a compact description, their simple average is \((1.20+1.23+1.16+1.26)/4=4.85/4=1.2125\). The values are not identical, but they do not show a sustained increase or decrease by accident-year order. A common factor near this group’s level may be plausible, subject to the amount and quality of the data and other relevant context.
Check. Multiplying each 12-month amount by its calculated ratio recovers its 24-month amount: \(1{,}000(1.20)=1{,}200\), \(2{,}000(1.23)=2{,}460\), \(1{,}500(1.16)=1{,}740\), and \(800(1.26)=1{,}008\). The variation is real, but this small set alone does not establish a persistent change. The common mistake is to treat any difference among ratios as proof that the common-pattern assumption fails; the correction is to assess the overall pattern of differences, not demand exact equality.
Worked Example: A Sustained Shift Between Accident-Year Groups
A hypothetical insurer calculates 12-to-24-month ratios for six accident years. The cumulative losses are in thousands of dollars. Decide whether the ratios look like scattered variation around one common level or suggest a persistent shift. No interest or mortality adjustment is made, and no probability distribution is assumed.
| Accident year | 12-month cumulative losses | 24-month cumulative losses |
|---|---|---|
| 1 | 1,000 | 1,200 |
| 2 | 1,500 | 1,830 |
| 3 | 2,000 | 2,360 |
| 4 | 900 | 1,215 |
| 5 | 1,100 | 1,463 |
| 6 | 1,400 | 1,904 |
Method: calculate the ratios and compare the earlier and later groups at this one age interval.
For accident years 1 through 3, the ratios are \(1{,}200/1{,}000=1.20\), \(1{,}830/1{,}500=1.22\), and \(2{,}360/2{,}000=1.18\). Their average is \((1.20+1.22+1.18)/3=3.60/3=1.20\).
For accident years 4 through 6, the ratios are \(1{,}215/900=1.35\), \(1{,}463/1{,}100=1.33\), and \(1{,}904/1{,}400=1.36\). Their average is \((1.35+1.33+1.36)/3=4.04/3=1.3467\), rounded to four decimal places.
All three later ratios exceed all three earlier ratios. The difference between the group averages is \(1.3467-1.20=0.1467\). This ordered separation is more suggestive of a persistent shift than the scattered variation in the first example. It makes a single common factor less convincing unless there is a reason to treat the groups as comparable despite the difference.
Check. Each ratio reproduces the stated 24-month amount when multiplied by its 12-month amount; for example, \(900(1.35)=1{,}215\) and \(1{,}400(1.36)=1{,}904\). The data support concern about stability, but the ratios alone do not establish why the shift occurred or prove that it will continue. The common mistake is to average all six ratios and report only one number, hiding the separation between groups. The correction is to preserve the accident-year ordering when interpreting variation.
Worked Example: Stability Is Age-Specific
For three hypothetical accident years, the observed ratios are 12-to-24 months: 1.10, 1.12, and 1.09; and 24-to-36 months: 1.04, 1.03, and 1.05. Determine what can be said about stability. These are descriptive factors only: no interest or mortality assumptions or probability distribution are used.
Method: summarize each development age separately. The 12-to-24 average is \((1.10+1.12+1.09)/3=3.31/3=1.1033\), rounded to four decimal places. Its range is \(1.12-1.09=0.03\). The 24-to-36 average is \((1.04+1.03+1.05)/3=3.12/3=1.0400\), and its range is \(1.05-1.03=0.02\).
Within each interval, the three accident-year ratios are close together, so a common factor at each age is plausible from these values. The two averages differ because they refer to different development ages; that difference is not evidence of instability across accident years.
Check and trap. An incorrect calculation pools all six ratios: \((3.31+3.12)/6=6.43/6=1.0717\), then treats 1.0717 as a common factor. This mixes distinct development ages and erases the age-specific pattern. The corrected comparison keeps the intervals separate: 1.1033 for 12 to 24 months and 1.0400 for 24 to 36 months. Each original ratio remains within its own interval’s reported range, as required.
Interpreting Stability Carefully
A common factor is an assumption about the underlying development process, not a claim that every historical ratio will match exactly. Variation that is scattered across accident years may be consistent with one common pattern, especially when the differences are modest relative to the data being examined. A sequence of ratios that stays systematically higher or lower for a group of accident years is a stronger reason to question whether one factor describes all the years.
The pattern should be examined separately for every development age used in a projection. Stability from 12 to 24 months does not guarantee stability from 24 to 36 months. Likewise, several stable-looking ratios at one age do not prove that the whole development pattern is stable at every age.
The amount of loss underlying a ratio also matters to interpretation. Ratios based on small cumulative amounts can be more volatile in practice than ratios based on larger amounts. But a large amount does not make a ratio automatically representative, and a numerical summary such as an average can conceal a persistent grouping. Review the accident-year order and the underlying values before treating a summary as a sufficient description.
This is a judgment about plausibility, not a causal diagnosis. If a pattern appears to shift, further investigation may be needed before selecting a factor or deciding how to reflect that shift. The observed ratios alone do not establish whether the reason is random fluctuation or a change in the underlying conditions.
Check Your Understanding
For each problem, calculate or compare the stated age-to-age ratios before drawing a conclusion. No interest or mortality adjustments or probability distributions are assumed.
- A hypothetical insurer has 12-month and 24-month cumulative losses of 700 and 840, 1,000 and 1,230, and 1,600 and 1,888 for three accident years. Calculate the three 12-to-24 ratios and their simple average. (Answer: 1.20, 1.23, and 1.18; average \(3.61/3=1.2033\), rounded to four decimals.)
- At 24-to-36 months, four accident-year ratios are 1.06, 1.08, 1.05, and 1.07. State the range and whether this set alone shows a sustained shift by accident year. (Answer: Range \(1.08-1.05=0.03\); no sustained shift is shown by these values alone.)
- For 12-to-24 months, the first three accident-year ratios are 1.14, 1.16, and 1.15; the next three are 1.29, 1.31, and 1.28. Calculate each group average and interpret the pattern. (Answer: First average \(3.45/3=1.15\); second average \(3.88/3=1.2933\); the ordered separation suggests a persistent shift and makes one common factor less convincing.)
- Ratios are 1.20 and 1.22 from 12 to 24 months, and 1.04 and 1.06 from 24 to 36 months, for the same two accident years. Should all four ratios be averaged to assess stability across accident years? (Answer: No. Compare within each age: the averages are 1.21 and 1.05, respectively; pooling different development ages would be misleading.)
- For one development age, accident-year ratios in chronological order are 1.10, 1.12, 1.09, 1.25, 1.27, and 1.26. What feature makes this more concerning than a similarly wide but unordered set, and what can the ratios not establish? (Answer: The last three ratios form a sustained higher group; this raises concern about a common factor, but the ratios alone cannot establish the cause of the shift.)