Settlement-Speed Changes in Paid Chain Ladder

Why Settlement Speed Matters in Paid Chain Ladder

Paid chain ladder uses the development of cumulative paid losses to project ultimate losses. Its central practical vulnerability is that paid development reflects not only how much loss an accident year will ultimately produce, but also when payments are made. If claims are paid more quickly or slowly than in the experience used to select development factors, the historical pattern may no longer describe the current accident years.

A faster settlement process can cause more of an accident year’s ultimate loss to appear in early paid evaluations. Applying an older, slower paid pattern to those larger early payments can project too large an ultimate. Conversely, slower payment can leave less paid at a given age than the historical pattern anticipates, potentially producing an ultimate that is too low. The effect depends on the direction and size of the pattern change, and on how the selected factors reflect it.

This tutorial is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus. The focus is diagnosing changes in paid development caused by settlement speed, rather than treating every movement in a paid link ratio as a change in the underlying ultimate loss.

Definition: Let $P_{i,j}$ be cumulative paid loss for accident year $i$ through development age $j$, where $j$ increases as more time passes. Let $U_i$ be that accident year’s ultimate loss. When $U_i>0$, define $q_{i,j}=P_{i,j}/U_i$ as the proportion of ultimate paid by age $j$. The paid link ratio from age $j$ to age $j+1$ is $P_{i,j+1}/P_{i,j}$, when $P_{i,j}>0$.

The proportion $q_{i,j}$ is a useful way to reason about settlement speed: a larger proportion at the same age means more of the ultimate loss has been paid by then. In practice, ultimate loss is not known for an immature accident year, so its paid proportion cannot simply be observed. Instead, analysts infer development from historical paid triangles and assess whether the historical payment pattern remains applicable.

How Stable Settlement Patterns Support Chain Ladder

For a simple explanation, suppose paid losses follow the model $P_{i,j}=U_iq_j$. Here, $q_j$ is the same cumulative paid proportion at age $j$ for every accident year, while $U_i$ may vary by accident year. This is a stylized model: it assumes a common settlement pattern, positive cumulative paid amounts at the ages used for ratios, and no material change in the way the loss amounts are measured. It does not require all accident years to have the same ultimate.

Under that model, the expected relationship between paid amounts at adjacent ages is determined by the common pattern. For a volume-weighted link ratio using accident years with observations at both ages, the ratio is the sum of later-age paid amounts divided by the sum of earlier-age paid amounts. If every included accident year has the same $q_j$ and $q_{j+1}$, the factor reduces to their ratio.

f_j = \frac{\sum_i P_{i,j+1}}{\sum_i P_{i,j}} = \frac{\sum_i U_iq_{j+1}}{\sum_i U_iq_j} = \frac{q_{j+1}}{q_j}.

Thus, when the pattern is stable, a chain of paid development factors estimates the movement from the observed paid amount toward the ultimate. A change in settlement speed changes the $q$'s. The historical factors may then be inappropriate for the current accident years, even if the underlying ultimate losses have not changed.

Why Faster Payment Can Overstate an Ultimate

The direction of the basic bias follows directly from the paid-proportion model. Suppose a current accident year has ultimate loss $U$, and its current payment pattern has paid proportion $q_j^{\text{new}}$ at age $j$. Its observed paid amount is $P_j=Uq_j^{\text{new}}$. If the selected historical pattern has proportion $q_j^{\text{old}}$, the corresponding age-to-ultimate factor is $1/q_j^{\text{old}}$. Applying it gives:

\widehat{U}_j = \frac{P_j}{q_j^{\text{old}}} = U\frac{q_j^{\text{new}}}{q_j^{\text{old}}}.

If settlement is faster and $q_j^{\text{new}}>q_j^{\text{old}}$, then $\widehat{U}_j>U$ under these assumptions. If settlement is slower and $q_j^{\text{new}}<q_j^{\text{old}}$, then $\widehat{U}_j<U$. This result isolates the pattern effect; it does not establish that a real estimate is biased whenever a paid factor changes. The ultimate loss, claim mix, and other relevant conditions must also be considered.

Worked Example: Faster Payments with an Unchanged Historical Factor

A hypothetical insurer has a stable historical pattern under which 40% of ultimate loss is paid by 12 months. A new process leads to faster payments, and a separate, comparable assessment indicates that 60% is paid by 12 months for a particular accident year. At 12 months, cumulative paid loss is $60,000. Estimate ultimate and unpaid loss using the old pattern, then compare with the estimate using the updated pattern. Assume the accident year’s actual ultimate is $100,000 for the purpose of checking the estimates.

Method and assumptions. This is a paid-proportion projection, equivalent to applying an age-to-ultimate factor $1/q_{12}$. Amounts are deterministic, with no probability distribution, interest discounting, or mortality model. The comparison assumes the stated proportions are applicable and that the underlying ultimate is unchanged.

Using the historical pattern. The old 12-month age-to-ultimate factor is $1/0.40=2.5$. The projected ultimate is:

\widehat{U}_{\text{old}} = \$60{,}000 \times 2.5 = \$150{,}000.

The implied unpaid loss is $\$150{,}000-\$60{,}000=\$90{,}000$. The known ultimate for this check is $100,000, so actual unpaid loss is $\$100{,}000-\$60{,}000=\$40{,}000$. The old pattern overstates ultimate by $50,000 and unpaid loss by $50,000.

Using the updated pattern. The updated age-to-ultimate factor is $1/0.60=1.6667$ before rounding. Therefore:

\widehat{U}_{\text{new}} = \frac{\$60{,}000}{0.60} = \$100{,}000, \qquad \widehat{R}_{\text{new}} = \$100{,}000-\$60{,}000 = \$40{,}000.

The updated estimate matches the stated ultimate and unpaid amount. This check confirms the direction: applying a slower historical pattern to payments that have arrived more quickly overstates the estimate in this example. The common mistake is to interpret higher early paid loss as automatically meaning a higher ultimate, rather than asking whether more of the loss has simply been paid sooner.

How the Change Appears in Link Ratios

A settlement-speed change can also affect the link ratios selected from a triangle. The effect may be diluted when a factor combines older and newer accident years. As a result, a selected factor can move in the direction of faster settlement without fully representing the current pattern. The next example shows how that partial movement can still leave a biased projection.

Worked Example: A Volume-Weighted Factor Blends Different Speeds

A hypothetical paid triangle is reported in thousands of dollars. Accident year 1 has cumulative paid losses of $40 at 12 months, $70 at 24 months, and $90 at 36 months. Accident year 2 has $50 at 12 months and $80 at 24 months. Accident year 3 has $60 at 12 months. Assume the first two accident years each ultimately produce $100, and the third accident year also ultimately produces $100. The second and third years have faster early payment than the first. Use volume-weighted link ratios to project accident year 3 ultimate through 36 months.

Method and assumptions. Use the available observations at each pair of ages to calculate volume-weighted paid link ratios, then multiply accident year 3’s 12-month paid amount by the selected factors. No interest discounting or mortality model is used, and no probability distribution is assumed. The known $100 ultimate for accident year 3 is used only to assess the projection.

12-to-24-month factor. The numerator is the sum of 24-month paid losses for accident years 1 and 2: $70+80=150$. The denominator is their sum at 12 months: $40+50=90$. Thus:

f_{12\text{--}24} = \frac{150}{90} = \frac{5}{3} = 1.6667.

24-to-36-month factor. Only accident year 1 has observations at both ages in the supplied data. Its link ratio is:

f_{24\text{--}36} = \frac{90}{70} = \frac{9}{7} = 1.2857.

Projected ultimate. Applying both factors to accident year 3’s $60 at 12 months gives:

\widehat{U}_3 = 60\times\frac{5}{3}\times\frac{9}{7} = \frac{900}{7} = \$128.57 \text{ thousand}.

The projected unpaid loss is $\$128.57-\$60=\$68.57$ thousand. The known ultimate is $100 thousand, so actual unpaid loss is $40 thousand. In this constructed example, the chain ladder projection overstates ultimate and unpaid loss by $28.57 thousand. The 12-to-24 factor reflects some faster payment through accident year 2, but combining it with the older 24-to-36 pattern does not reproduce accident year 3’s full payment pattern. The common mistake is to assume that a factor that has moved in the right direction must have fully corrected the pattern change.

Using Updated Pattern Information Carefully

A revised payment pattern may be more appropriate when there is credible evidence that recent accident years are settling at a different pace. Potential evidence includes payment-process changes, operational analyses of comparable claims, or observed patterns in sufficiently mature recent experience. Such evidence should be evaluated for comparability: a change in claim mix or coverage may alter both timing and ultimate amounts, while a process change may primarily alter timing.

An updated pattern should not be treated as automatically correct just because it produces a more plausible estimate. It needs support, and the analyst should consider whether the observed change persists across accident years and development ages. If the evidence is limited, estimates under both patterns can help show sensitivity. The important diagnostic is to separate the question “Are payments arriving sooner?” from “Has the expected total loss changed?”

Worked Example: Comparing Historical and Updated Patterns at Two Ages

A hypothetical accident year has $72,000 cumulative paid at 12 months and $102,000 at 24 months. The historical pattern indicates that 40% of ultimate is paid by 12 months and 70% by 24 months. A separate analysis of comparable recent claims supports updated proportions of 60% and 85% at those ages. Estimate ultimate and unpaid loss at each age using both patterns, and check the updated estimates for consistency. Assume the ultimate is $120,000 for checking purposes.

Method and assumptions. Divide cumulative paid loss at each age by the relevant cumulative paid proportion. This deterministic comparison assumes each stated pattern is applicable to the accident year; it uses no probability distribution, interest discounting, or mortality model.

Historical pattern at 12 months. The estimated ultimate is $\$72{,}000/0.40=\$180{,}000$, and estimated unpaid loss is $\$180{,}000-\$72{,}000=\$108{,}000$.

Historical pattern at 24 months. The estimated ultimate is $\$102{,}000/0.70=\$145{,}714.29$, and estimated unpaid loss is $\$145{,}714.29-\$102{,}000=\$43{,}714.29$. The two historical-pattern projections differ because the observed payments are more advanced relative to that older pattern.

Updated pattern at 12 months. The estimated ultimate is $\$72{,}000/0.60=\$120{,}000$. Estimated unpaid loss is $\$120{,}000-\$72{,}000=\$48{,}000$.

Updated pattern at 24 months. The estimate is $\$102{,}000/0.85=\$120{,}000$, with unpaid loss of $\$120{,}000-\$102{,}000=\$18{,}000$. Both updated estimates agree with the stated $120,000 ultimate, and the unpaid amounts reconcile to the known ultimate at each age. This is an internally consistent check, not independent proof that the updated pattern is right. The common mistake is to divide by the percentage of loss still unpaid; an age-to-ultimate projection divides paid loss by the proportion already paid.

Diagnosis and Common Traps

A change in the paid triangle does not by itself identify its cause. Before concluding that settlement speed has changed, check whether paid amounts are comparable across accident years and development ages. Consider whether the claim mix, coverage, limits, deductibles, or loss measurement has changed. A change in claim closure administration without a corresponding change in the timing of payments need not change the paid development pattern. Conversely, a change in the timing or size of payments can affect paid emergence even when the underlying ultimate is unchanged.

When the evidence does support a speed change, establish the direction at comparable ages. If a higher proportion of ultimate is paid earlier than under the selected historical pattern, an old age-to-ultimate factor will generally project too high under the simple model. If a lower proportion is paid, it will generally project too low. This is a diagnostic result, not a guarantee about a complex real triangle: selected factors may combine different cohorts, and more than one underlying change may be present.

Confusing paid dollars with ultimate loss. More paid at an early age can reflect quicker payment rather than a larger ultimate.
Ignoring factor composition. A volume-weighted factor can blend accident years with different payment speeds; its value need not match the newest cohort’s pattern.
Assuming settlement speed means claim closure speed. The paid triangle responds to payment emergence; an administrative change matters only if it changes paid amounts or their timing.
Using a revised pattern without support. A plausible adjustment still requires evidence that the new pattern applies to the accident years being estimated.

Key takeaway: Paid chain ladder relies on a sufficiently stable relationship between paid loss and ultimate loss by development age. Faster settlement raises the proportion paid at early ages; applying an unchanged slower pattern can overstate projected ultimate loss. Slower settlement can have the opposite effect. Diagnose the pattern change, its evidence, and its relevance before relying on historical paid factors.

Check Your Understanding

For each question, use the stated deterministic pattern and assumptions; no interest discounting, mortality model, or probability distribution is involved.

A historical pattern has 45% of ultimate paid at 12 months. A hypothetical current accident year has $54,000 paid at that age, and independent evidence indicates that 60% of its ultimate is paid. What ultimate does the historical proportion project, what ultimate follows from the updated proportion, and what is the historical-pattern overstatement? (Answer: Historical estimate $120,000; updated estimate $90,000; overstatement $30,000.)
For two accident years, 12-month cumulative paid amounts are $30,000 and $50,000, and their 24-month amounts are $54,000 and $80,000. What is the volume-weighted 12-to-24-month paid link ratio? (Answer: $134/80=1.675$.)
A supported updated pattern indicates that 60% of ultimate is paid by 12 months. A hypothetical accident year has $72,000 paid at 12 months. What are its projected ultimate and unpaid loss under that pattern? (Answer: Ultimate $120,000; unpaid loss $48,000.)
An insurer changes the administrative process for closing claim files, but the timing and amounts of payments are unchanged. Based on that information alone, should a change in the paid development pattern be assumed? (Answer: No; a closure-process change alone does not establish a change in paid emergence.)
At 24 months, a hypothetical accident year has $84,000 paid. The historical pattern indicates 60% paid by that age; a comparable updated analysis supports 70%. Calculate both projected ultimates and the difference, and state the direction of the old-pattern effect relative to the updated estimate. (Answer: Historical estimate $140,000; updated estimate $120,000; the historical-pattern estimate is $20,000 higher.)

What You'll Learn