Why Development May Continue Beyond the Triangle
A loss triangle contains observations only through the ages represented in the available data. Selected link ratios can project losses from one represented age to another, but they do not by themselves establish that losses have stopped developing at the final age in the triangle. Claims may continue to be reported, settled, or revised after that point. This is especially relevant for hypothetical portfolios with long settlement periods, such as liability claims that can take many years to resolve.
A tail factor represents the development from the latest age for which a cumulative value is available to the assumed ultimate value. It is needed when the selected link ratios stop before development is reasonably considered complete. If no tail is included, the projection effectively assumes that the cumulative amount at the last selected age is already ultimate.
The symbol \(\infty\) is a modeling convention for the point at which further development is taken to be negligible. It does not necessarily mean that every claim has literally closed. A tail factor is therefore an estimate based on a selected development pattern or other justified information, not a directly observed ratio when the triangle ends before ultimate.
How a Tail Factor Completes the Development Pattern
Suppose \(f_j\) is a selected age-to-age link ratio from age \(j\) to age \(j+1\). If the selected pattern continued through all future ages, the tail from age \(m\) would be the product of the future link ratios:
In practice, the available triangle usually does not contain every future ratio. The actuary must therefore estimate the remaining development. One approach is to extrapolate a pattern in the latest observed link ratios. Another is to model the rate of development continuously and integrate that rate beyond the final selected age. A tail factor can also be selected using relevant external information and actuarial judgment; the rationale should make clear why development beyond the observed ages is expected to be material or negligible.
The tail is distinct from the cumulative development factor from an earlier age to age \(m\). If selected link ratios are available from age \(k\) through age \(m\), their product develops from \(k\) to \(m\). The tail starts at \(m\) and extends beyond it. Keeping those intervals separate prevents double counting.
Deriving the Tail as Remaining Development
The product definition follows directly from successive ratios. For a finite final age \(n\), the link ratio from age \(j\) to age \(j+1\) is \(f_j=C_{j+1}/C_j\). Multiplying the ratios after age \(m\) cancels all intermediate cumulative amounts:
Thus the product is precisely the multiplier from age \(m\) to age \(n\). If the sequence has a finite limit as \(n\) increases, that limiting multiplier is \(C_{\infty}/C_m\), the tail factor. A finite tail requires the cumulative product to converge. An extrapolation that implies an indefinitely growing product is not a plausible finite ultimate-development model.
This argument also explains the effect of omitting a positive tail. If the expected ultimate is \(C_{\infty}=T_m C_m\) with \(T_m>1\), using \(C_m\) as ultimate omits the amount \((T_m-1)C_m\). The relative understatement compared with the ultimate is \(1-1/T_m\), not \(T_m-1\); the latter is the omitted amount as a proportion of the age-\(m\) cumulative value.
Worked Example: Identifying Development After the Last Selected Age
A hypothetical liability portfolio has a cumulative loss amount of $8,000,000 at the final age represented by selected link ratios. A separate actuarial projection of post-age development suggests additional cumulative increases of $80,000, $40,000, and $20,000 in three later periods, with no further development assumed thereafter. Find the tail factor from the final represented age and quantify the amount omitted if the tail is ignored. Assume amounts are nominal dollars with no discounting.
Method. Add only the projected increases after the final represented age to the age-\(m\) cumulative amount. Divide the resulting ultimate amount by the age-\(m\) amount.
The projected increase after age \(m\) is:
The assumed ultimate amount is:
Therefore the tail factor is:
Ignoring the tail would omit $140,000. As a proportion of the amount at age \(m\), that is \(140{,}000/8{,}000{,}000=0.0175\), or 1.75%. As a proportion of the assumed ultimate, it is \(140{,}000/8{,}140{,}000\approx0.0172\), or 1.72% to three significant figures.
Check. Multiplying the age-\(m\) amount by the tail gives \(8{,}000{,}000(1.0175)=8{,}140{,}000\), matching the constructed ultimate. The common mistake avoided is treating the last selected age as ultimate merely because it is the last age in the observed pattern.
Estimating a Tail by Extrapolating a Development Rate
A convenient continuous model describes the remaining development through a nonnegative rate \(g(t)\), where \(t=0\) is the final selected age. If the instantaneous proportional development rate is \(g(t)\), the cumulative multiplier from \(t=0\) to \(t=b\) is \(\exp(\int_0^b g(t)\,dt)\). Extending to ultimate gives:
This follows because a small interval \(dt\) multiplies the cumulative amount by approximately \(1+g(t)dt\); adding the corresponding log increments over time gives \(\log T_m=\int_0^\infty g(t)\,dt\). The rate model must be appropriate for the remaining development and its integral must be finite. For example, a positive constant rate continuing forever would give an infinite integral and no finite tail.
Worked Example: Exponential Decay in the Remaining Development Rate
For a hypothetical portfolio, suppose the post-age proportional development rate is modeled as \(g(t)=0.003986742328e^{-0.2t}\), where \(t\) is measured in years after the final selected age. The model is assumed to apply continuously, with no discounting. Calculate the tail factor, rounding the final factor to six decimal places.
Method. Integrate the assumed rate over all future years, then exponentiate the accumulated log development.
Because \(\int_0^\infty e^{-0.2t}\,dt=1/0.2=5\), the total log development is:
Therefore:
Rounded to six decimal places, the tail factor is 1.020134. The factor implies approximately 2.0134% development relative to the amount at the final selected age, since \(1.020134-1=0.020134\).
Check. The integrated rate is positive and finite, so the resulting factor is finite and greater than 1, as expected under this positive-development model. The common mistake avoided is exponentiating the rate at \(t=0\) alone; the tail depends on the integral of the rate over all future time.
Combining Successive Parts of a Tail
Sometimes an extrapolation is represented in stages: for example, one factor covers a near-term interval and a separate factor covers the more remote remaining development. These factors multiply because each starts where the preceding interval ends. Their sum is not the total multiplier.
Worked Example: Combining a Near-Term Tail and a Remote Tail
A hypothetical development analysis produces a multiplier of 1.010531527 from the final selected age through an intermediate age. A fitted decay model estimates the remaining log development after that intermediate age as 0.0024989591. Find the full tail factor, reporting intermediate multipliers to nine decimal places and the final factor to six decimal places. Assume the two intervals are consecutive and do not overlap.
Method. Convert the remaining log development to a multiplier, then multiply that multiplier by the near-term factor.
The remaining multiplier is:
The full tail factor is consequently:
To six decimal places, the full tail factor is 1.013060. The remote segment adds a smaller amount of development than the near-term segment, but it is not zero.
Check. Both interval multipliers exceed 1, so their product must exceed 1.010531527. The calculated product, 1.013059962, satisfies that check. The common mistake avoided is adding the two factors, which would not represent consecutive proportional development.
Selection, Interpretation, and Exam Traps
A calculated tail is only as credible as the assumptions used to extend development beyond observed ages. When the latest observed link ratios are still materially above 1, assuming no further development needs particular justification. Conversely, a tail close to 1 can be reasonable when late development is negligible, but should not be selected only because the data end at a convenient age. Sparse late-age observations can make extrapolation especially uncertain.
Identify the final age at which the selected pattern ends. The tail must start at this age, not at an earlier age already covered by selected link ratios.
State whether the tail comes from projected post-age increments, extrapolated link ratios, a continuous rate model, or a justified selection.
For a positive-development model, the remaining log development should be nonnegative and finite, producing a finite factor of at least 1.
Verify that no development interval is omitted or counted twice when combining selected link ratios with the tail.
A frequent exam trap is confusing the tail factor with the full cumulative development factor from an early age. If selected link ratios already develop from age 12 to age 120, and the tail begins at age 120, multiplying a factor from age 12 directly to ultimate by the tail would count development from age 12 to 120 twice. Another trap is calling an extrapolated amount “observed”; ultimate development beyond the last represented age is an assumption unless later data actually establish it.
Check Your Understanding
For each problem, identify the tail interval before calculating. Assume no discounting unless stated otherwise.
- A hypothetical portfolio has cumulative losses of $5,000,000 at its final selected age. A justified projection indicates $50,000 of development after that age. Find the tail factor to six decimal places. (Answer: 1.010000)
- A development rate after the final selected age is \(g(t)=0.002e^{-0.25t}\), with \(t\) measured in years. Find the tail factor to six decimal places. (Answer: 1.008032)
- A hypothetical tail has a multiplier of 1.012 from the final selected age to an intermediate age, followed by a further multiplier of 1.004. Find the combined factor to six decimal places. (Answer: 1.016048)
- At the final selected age, cumulative losses are $12,000,000. The assumed ultimate is $12,300,000. Find the tail factor and the omitted amount if the tail is ignored. (Answer: tail factor 1.025000; omitted amount $300,000)
- An analyst selects factors through age 120 and then uses a tail factor from age 96 to ultimate. State the error and the correction. (Answer: Development from age 96 to 120 is counted twice; use a tail factor from age 120 to ultimate, or otherwise redefine the preceding factors so the intervals do not overlap.)