Tail Factor Selection and Reserve Sensitivity

Why Tail-Factor Selection Matters

A chain ladder projection can depend materially on development expected after the latest age-to-age factor that can be estimated from observed data. The factor that carries losses from that final modeled age to ultimate is the tail factor. Since little or no direct development data may be available for this period, alternative reasonable tail selections can produce different ultimate-loss and unpaid-liability estimates.

The important comparison is not just whether a larger tail factor produces a larger estimate. An actuary should quantify the dollar change, identify which accident years contribute most to it, and distinguish a change in ultimate losses from a change in unpaid liabilities. This tutorial is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: Let $C_{i,k}$ be the latest observed cumulative loss for accident year $i$, at development age $k$. Let $G_{i,k}$ be the product of the selected age-to-age factors from age $k$ to the age at which the tail factor applies. Let $T$ be the selected tail factor from that age to ultimate. The projected ultimate loss is $U_i=C_{i,k}G_{i,k}T$. For a paid-loss projection, the estimated unpaid liability is $R_i=U_i-C_{i,k}$. The tail factor is a multiplier, not an amount added to cumulative loss.

Set Up the Sensitivity Comparison

The selected age-to-age factors and the tail factor have distinct roles. The age-to-age factors project the observed cumulative loss to the age where the tail begins; the tail factor carries that projected amount from the tail-entry age to ultimate. For each accident year, define the tail-entry base as $B_i=C_{i,k}G_{i,k}$. Then $U_i=B_iT$. Under a paid chain ladder, if the latest cumulative paid loss is $P_i$, the unpaid estimate is $R_i=B_iT-P_i$.

This comparison treats the observed data, selected age-to-age factors, and paid amounts as fixed. It changes only the selected tail factor. It also assumes that the same definition of ultimate and the same loss basis are used in every scenario. If age-to-age factors are also changed, the resulting difference cannot be attributed to the tail selection alone.

Project to the tail-entry age.
Multiply the latest cumulative loss by the selected age-to-age factors that remain before the tail begins. Record this result as $B_i$.

Apply each alternative tail factor.
For each scenario $s$, calculate $U_i^{(s)}=B_iT^{(s)}$.

Calculate unpaid liabilities.
For paid losses, subtract the latest observed cumulative paid amount: $R_i^{(s)}=U_i^{(s)}-P_i$.

Compare results consistently.
Calculate differences between scenarios for each accident year and for the aggregate. State whether a comparison is in dollars or as a percentage of a specified baseline.

Why the Sensitivity Formula Holds

With the data and age-to-age selections fixed, the tail-entry base $B_i$ is constant as the tail factor changes. For two tail selections $T_a$ and $T_b$, the ultimate-loss difference is therefore:

$$ U_i(T_b)-U_i(T_a)=B_iT_b-B_iT_a=B_i(T_b-T_a). $$

For paid losses, the same observed paid amount $P_i$ is subtracted under both selections. Thus:

$$ R_i(T_b)-R_i(T_a) =[B_iT_b-P_i]-[B_iT_a-P_i] =B_i(T_b-T_a). $$

So the dollar change in ultimate loss equals the dollar change in unpaid liability when only the tail factor changes. The percentage changes need not be equal: ultimate loss and unpaid liability have different baselines. For a small change in $T$, the dollar sensitivity is $B_i$ dollars for each 1.0 change in the tail factor. Equivalently, changing the tail factor by 0.01 changes the estimate by $0.01B_i$.

Worked Example: Quantifying a Tail-Factor Range

A hypothetical paid-loss triangle has cumulative paid losses of $8.400 million at 24 months for one accident year. The selected 24-to-36-month factor is 1.40, and the selected 36-to-48-month factor is 1.15. At 48 months, the actuary is considering tail factors of 1.03, 1.06, and 1.09. Calculate the ultimate loss and unpaid liability under each selection, and compare the low and high scenarios.

Method. Project to the tail-entry age of 48 months, then apply each tail factor. The calculation assumes the two age-to-age factors and observed paid loss are fixed; the tail factor is the only changing selection.

First calculate the projected 48-month cumulative paid loss:

B=8.400(1.40)(1.15)=11.760(1.15)=13.524\text{ million}.

For $T=1.03$, the ultimate loss is $13.524(1.03)=13.92972$ million. The unpaid liability is $13.92972-8.400=5.52972$ million.

For $T=1.06$, the ultimate loss is $13.524(1.06)=14.33544$ million. The unpaid liability is $14.33544-8.400=5.93544$ million.

For $T=1.09$, the ultimate loss is $13.524(1.09)=14.74116$ million. The unpaid liability is $14.74116-8.400=6.34116$ million.

The high-minus-low ultimate difference is $14.74116-13.92972=0.81144$ million. The unpaid-liability difference is $6.34116-5.52972=0.81144$ million, as the derivation predicts. Rounded to the nearest dollar, both differences are $811,440.

Common mistake avoided: Applying the 0.06 difference between the high and low tail factors to the observed $8.400 million gives $504,000. That is wrong because the tail factor applies to the projected 48-month base, not the 24-month observed amount. The correct sensitivity is $13.524(0.06)=0.81144$ million.

Immaturity, Absolute Sensitivity, and Percentage Sensitivity

An accident year with little observed development may have a large tail-entry base relative to its latest observed paid amount, because the selected age-to-age factors project substantial future payments before the tail is applied. A given change in the tail factor then produces a larger dollar change when its tail-entry base is larger. However, being immature does not, by itself, prove that the percentage change in unpaid liability will be greater. That percentage also depends on how much liability is already implied by the projected ultimate relative to the observed paid amount.

The next example isolates that distinction. The tail-entry bases are treated as outputs of the selected age-to-age projections; they need not be produced by identical development patterns for the two accident years.

Worked Example: Comparing a Mature and an Immature Accident Year

For a hypothetical mature accident year, the projected tail-entry base is $5.0 million and observed cumulative paid loss is $4.4 million. For a hypothetical immature accident year, the corresponding amounts are $18.0 million and $7.2 million. Compare unpaid liabilities under tail factors of 1.04 and 1.08, and calculate the percentage increase from the lower-tail reserve to the higher-tail reserve for each year.

Method. For each year, calculate $U=BT$, then $R=U-P$. Calculate the percentage reserve increase as $(R_{\text{high}}-R_{\text{low}})/R_{\text{low}}$. The comparison assumes each year’s tail-entry base and observed paid amount remain fixed.

For the mature year, at $T=1.04$, ultimate loss is $5.0(1.04)=5.20$ million and unpaid liability is $5.20-4.40=0.80$ million. At $T=1.08$, ultimate loss is $5.0(1.08)=5.40$ million and unpaid liability is $5.40-4.40=1.00$ million. Its reserve increase is $1.00-0.80=0.20$ million, or $0.20/0.80=25\%$.

For the immature year, at $T=1.04$, ultimate loss is $18.0(1.04)=18.72$ million and unpaid liability is $18.72-7.20=11.52$ million. At $T=1.08$, ultimate loss is $18.0(1.08)=19.44$ million and unpaid liability is $19.44-7.20=12.24$ million. Its reserve increase is $12.24-11.52=0.72$ million, or $0.72/11.52=6.25\%$.

The immature year has the larger dollar change: $720,000 rather than $200,000. But the mature year has the larger percentage change in its reserve: 25% rather than 6.25%. This is not a contradiction. The percentage denominator—the lower-tail reserve—is much smaller for the mature year.

Common mistake avoided: Do not conclude that the least mature year must have the greatest percentage reserve sensitivity. Its larger tail-entry base does imply greater dollar sensitivity here, but the percentage comparison also depends on the starting unpaid liability.

Aggregate Sensitivity Across Accident Years

When the same tail factor is applied to multiple accident years on the same basis, aggregate sensitivity can be calculated by summing the accident-year tail-entry bases. If $B_{\mathrm{agg}}=\sum_i B_i$, then a change from $T_a$ to $T_b$ changes aggregate ultimate losses and aggregate unpaid liabilities by $B_{\mathrm{agg}}(T_b-T_a)$, provided the observed paid amounts and all other selections are held fixed. Calculating by accident year as well as in aggregate helps identify which projections account for the largest dollar impacts.

Worked Example: Aggregating Sensitivity for Unevenly Developed Years

A hypothetical portfolio has three accident years. Their projected paid-loss amounts at the tail-entry age are $12.0 million, $9.0 million, and $6.0 million. The corresponding latest observed cumulative paid amounts are $10.5 million, $5.4 million, and $1.8 million. Compare the total unpaid liability under tail factors of 1.02, 1.05, and 1.08.

Method. Apply each common tail factor to each year’s tail-entry base and subtract that year’s observed paid amount. The selected bases already reflect any age-to-age development before tail entry. All amounts below are in millions of dollars.

At $T=1.02$, the three ultimate losses are $12.0(1.02)=12.24$, $9.0(1.02)=9.18$, and $6.0(1.02)=6.12$. Their unpaid liabilities are $12.24-10.5=1.74$, $9.18-5.4=3.78$, and $6.12-1.8=4.32$. Total ultimate loss is $12.24+9.18+6.12=27.54$, total observed paid is $10.5+5.4+1.8=17.7$, and aggregate unpaid liability is $27.54-17.7=9.84$.

At $T=1.05$, the ultimate losses are $12.60$, $9.45$, and $6.30$; the unpaid liabilities are $2.10$, $4.05$, and $4.50$. Aggregate ultimate loss is $28.35$, so aggregate unpaid liability is $28.35-17.7=10.65$.

At $T=1.08$, the ultimate losses are $12.96$, $9.72$, and $6.48$; the unpaid liabilities are $2.46$, $4.32$, and $4.68$. Aggregate ultimate loss is $29.16$, so aggregate unpaid liability is $29.16-17.7=11.46$.

The high-minus-low aggregate unpaid change is $11.46-9.84=1.62$ million. As a check, the sum of tail-entry bases is $12.0+9.0+6.0=27.0$ million, and $27.0(1.08-1.02)=27.0(0.06)=1.62$ million. The calculation reconciles.

Common mistake avoided: The portfolio’s change is based on the sum of the projected tail-entry amounts, $27.0 million, not the total observed paid amount, $17.7 million. The observed paid amount affects the reserve level, but cancels when comparing two tail scenarios.

Interpreting and Reporting the Results

A useful sensitivity summary states the alternatives, the fixed assumptions, and the impact in both dollars and context. For example, an actuary might report the low, central, and high tail selections; the resulting ultimate and unpaid estimates by accident year; and the aggregate dollar range. If relative sensitivity is important, the report should name its denominator, such as “percentage change from the low-tail unpaid estimate.” Without that qualifier, percentage comparisons can be ambiguous.

A sensitivity calculation does not establish which tail factor is most appropriate. It shows the consequence of alternatives under the stated selections. The factor still requires actuarial judgment supported by the available development information and other relevant evidence. A sensitivity range is not automatically a probability interval, and its endpoints should not be described as confidence bounds unless a separate statistical model justifies that interpretation.

Keep these checks in view when interpreting a result:

A larger positive tail factor should not reduce projected ultimate losses when the tail-entry base is positive.
With observed paid losses fixed, the change in ultimate loss must equal the change in unpaid liability.
A larger tail-entry base means greater dollar sensitivity for the same change in tail factor.
Percentage sensitivity depends on the denominator and can rank accident years differently from dollar sensitivity.
If the tail factor is applied at a different development age, first ensure the tail-entry base and factor definition match that age.

Key takeaway: With other selections held fixed, the dollar impact of changing a tail factor is the tail-entry base multiplied by the change in that factor. Immature years may have large dollar sensitivity because of substantial projected development, but percentage reserve sensitivity must be evaluated against the reserve baseline—not inferred from maturity alone.

Check Your Understanding

For each problem, treat the stated paid amounts, tail-entry bases, and age-to-age selections as fixed. Report money in dollars unless the problem states otherwise.

A hypothetical accident year has cumulative paid loss of $2.5 million at 18 months. The selected factors to the tail-entry age are 1.60 from 18 to 36 months and 1.20 from 36 to 60 months. What is the change in ultimate loss and unpaid liability between tail factors of 1.05 and 1.08? (Answer: $144,000 increase in both ultimate loss and unpaid liability.)
A paid-loss projection has a tail-entry base of $7.2 million and latest cumulative paid loss of $3.6 million. Calculate the unpaid liability under tail factors 1.04 and 1.07, then find the change. (Answer: $3.888 million and $4.104 million; increase of $216,000.)
Three accident years have tail-entry bases of $4 million, $8 million, and $10 million, and latest paid amounts of $2 million, $5 million, and $9 million. Find the aggregate unpaid liability at tail factors 1.03 and 1.06. (Answer: $6.66 million and $7.32 million, respectively; increase of $660,000.)
A year has tail-entry base $6.0 million and observed paid loss $5.4 million. Compare unpaid liabilities under tail factors 1.02 and 1.06, and express the increase as a percentage of the lower-tail reserve. (Answer: $720,000 and $960,000; increase of $240,000, or 33.3% of the lower-tail reserve.)
A hypothetical year has cumulative paid loss of $1.5 million at 12 months, with selected factors of 1.50 from 12 to 24 months and 1.20 from 24 to 36 months. The tail begins at 36 months. Compare the unpaid liabilities under tail factors 1.04 and 1.07. (Answer: $1.308 million and $1.389 million; increase of $81,000.)

What You'll Learn