Tail Factors And Runoff Security

Why Tail Factors Matter

A large-deductible claim can continue to generate reimbursement obligations long after the initial claim has been reported. Medical treatment, litigation, and claim administration may extend beyond the development periods available in historical data. A projection based only on observed development can therefore omit a meaningful part of the expected obligation.

A tail factor extends a development projection beyond its last observed age. The resulting estimate can affect both the amount of security held and how long it is needed. A reliable estimate should answer two separate questions: how much remains unpaid at the valuation date, and how that amount is expected to run off over future periods.

Definition: Let $C_t$ denote cumulative paid reimbursement amounts through development age $t$. An age-to-age factor from age $t$ to age $t+1$ is the selected ratio $f_{t,t+1}=C_{t+1}/C_t$, or a portfolio estimate of that ratio. A cumulative development factor (CDF) from age $t$ to ultimate is the product of the selected factors from age $t$ onward, including any tail factor. The projected ultimate amount is $U_t=C_t \times \mathrm{CDF}_t$, and the projected unpaid amount is $U_t-C_t$.

These definitions assume the cumulative amounts and development factors refer to the same type of obligation, valuation basis, and portfolio. In this tutorial, the amounts are hypothetical cumulative reimbursements in the large-deductible program. A paid-reimbursement triangle can be useful when its historical development is a reasonable guide to the obligations being projected. It may not be appropriate without adjustment if claim handling, coverage, reimbursement rules, or the portfolio have changed materially. Recoveries or other negative adjustments can also make simple paid-development patterns unreliable.

From Observed Development to Ultimate

When historical observations extend only to a selected age, first multiply the observed age-to-age factors to obtain the CDF to the end of observed development. Then multiply by the selected tail factor. The tail factor represents development after that observed endpoint; it should not include development already represented by the observed factors.

\mathrm{CDF}_t = \left(\prod_{j=t}^{n-1} f_{j,j+1}\right) F_{\mathrm{tail},n}, \qquad U_t=C_t\mathrm{CDF}_t, \qquad \mathrm{Unpaid}_t=U_t-C_t

The product is essential. Development percentages compound on the amount already developed. Adding age-to-age factors does not produce a cumulative factor. Once an ultimate amount is estimated, the unpaid amount is the difference between that ultimate and the cumulative amount already paid, on the same basis.

Worked Example: Projecting Unpaid Reimbursements With a Tail Factor

A hypothetical program has $4,800,000 of cumulative paid reimbursements at 12 months. Selected factors are 1.20 from 12 to 24 months, 1.10 from 24 to 36 months, and a 1.05 tail factor after 36 months. Estimate ultimate reimbursements and the unpaid amount at 12 months. Treat all values as nominal dollars: no discounting or interest is applied ($i=0$). This is a deterministic projection, not a mortality model, and no probability distribution is assumed.

Method. Multiply the selected development factors to obtain the 12-month-to-ultimate CDF:

\mathrm{CDF}_{12} =1.20 \times 1.10 \times 1.05 =1.386

The first multiplication gives $1.20 \times 1.10=1.32$; then $1.32 \times 1.05=1.386$. The projected ultimate is:

U_{12} =\$4{,}800{,}000 \times 1.386 =\$6{,}652{,}800

Subtract the cumulative amount already paid to obtain the projected unpaid amount:

\mathrm{Unpaid}_{12} =\$6{,}652{,}800-\$4{,}800{,}000 =\$1{,}852{,}800

Check. The projected unpaid amount is 38.6% of the 12-month paid amount because $1.386-1=0.386$, and $0.386 \times \$4{,}800{,}000=\$1{,}852{,}800$. The common mistake is to add the factors, giving $1.20+1.10+1.05=3.35$. That is not a cumulative factor: the correct factor is their product, 1.386.

Estimating Development Beyond Observed Ages

A tail factor may be selected from longer historical data, external information judged suitable for the portfolio, or an explicit extrapolation of the observed pattern. A transparent extrapolation is preferable to silently assuming that the most recent observed factor applies forever. One simple approach is to assume that the excess of future age-to-age factors above 1.00 declines over successive periods. This is a modeling assumption, not a universal law of claim development; the selected decay pattern should be tested for reasonableness.

Model assumption: If future selected factors are estimated as $1+e_1,1+e_2,\ldots,1+e_k$, the finite tail factor through those periods is $\prod_{j=1}^{k}(1+e_j)$. The tail factor is greater than 1 when these periods add positive cumulative development. A finite tail horizon does not imply that every claim is closed by that horizon unless that is separately supported.

The multiplication follows directly from the definition of successive development. If the current cumulative amount is $C$, one period later it is $C(1+e_1)$. After the next period it is $C(1+e_1)(1+e_2)$, not $C[1+e_1+e_2]$. Continuing this step for $k$ periods yields $C\prod_{j=1}^k(1+e_j)$. Dividing the projected amount by $C$ gives the stated tail factor.

Worked Example: Extrapolating a Finite Tail

A hypothetical reimbursement portfolio has $2,000,000 paid through 60 months. The latest observed age-to-age factor is 1.05. For the three future annual periods, an analyst assumes that the excess over 1.00 halves each year: the projected factors are therefore 1.025, 1.0125, and 1.00625. Estimate the tail factor, projected ultimate, and additional development beyond 60 months. Amounts are nominal; assume $i=0$, no discounting, and no mortality model or probability distribution.

Method. Multiply the three future factors. The successive products are:

1.025 \times 1.0125=1.0378125

1.0378125 \times 1.00625=1.044298828125

Thus, the selected finite tail factor is approximately 1.0443. Apply it to the cumulative amount at 60 months:

U_{60} =\$2{,}000{,}000 \times 1.044298828125 =\$2{,}088{,}597.65625

The projected additional development is:

U_{60}-C_{60} =\$2{,}088{,}597.65625-\$2{,}000{,}000 =\$88{,}597.65625

Rounded to the nearest cent, the projected ultimate is $2,088,597.66, and additional development is $88,597.66. The result depends on the stated halving assumption and three-period horizon. It is not evidence that the true tail follows that pattern. The common mistake is to add the excesses $0.025+0.0125+0.00625$ and use 1.04375 as the tail factor; compounding gives 1.044298828125 instead.

Runoff Schedules, Security Duration, and Release

A total unpaid estimate alone does not describe the timing of future cash needs. A runoff schedule allocates the projected unpaid amount to future periods. The schedule can support an estimate of how long security may be needed, but it is a projection rather than a promise that actual payments will follow the percentages exactly.

For a simple security illustration, let $R$ be the estimated unpaid obligation at the valuation date and let $q_j$ be the share expected to be paid in future year $j$. If the shares cover the entire projected runoff, they sum to 1. The amount expected to be paid in year $j$ is $Rq_j$; the amount remaining after year $m$ is $R(1-\sum_{j=1}^{m}q_j)$. A stated margin can be applied to the remaining amount to illustrate a security target. Actual collateral requirements depend on program terms and applicable controls; the schedule does not determine release by itself.

R_m=R\left(1-\sum_{j=1}^{m}q_j\right), \qquad K_m=(1+\lambda)R_m

Here $R_m$ is the projected nominal obligation remaining after year $m$, $K_m$ is an illustrative security target, and $\lambda$ is a proportional margin. This calculation assumes no discounting and a stable schedule for the illustration. If the security basis uses discounted values, the timing of payments and the chosen discount assumptions must be reflected consistently.

Worked Example: Translating a Runoff Schedule Into a Release Path

A hypothetical program has an estimated $2,400,000 of unpaid reimbursements. The projected payment shares are 50% in year 1, 30% in year 2, 15% in year 3, and 5% in year 4. For this illustration, security is set at the projected remaining nominal obligation plus a 10% margin. Assume no discounting ($i=0$); no mortality assumption or probability distribution is used.

Method. First calculate expected payments by year:

\begin{aligned} \text{Year 1} &: \$2{,}400{,}000 \times 0.50=\$1{,}200{,}000\\ \text{Year 2} &: \$2{,}400{,}000 \times 0.30=\$720{,}000\\ \text{Year 3} &: \$2{,}400{,}000 \times 0.15=\$360{,}000\\ \text{Year 4} &: \$2{,}400{,}000 \times 0.05=\$120{,}000 \end{aligned}

The shares sum to $0.50+0.30+0.15+0.05=1.00$, and the scheduled payments sum to $2,400,000. The projected remaining amounts and illustrative security targets are:

\begin{aligned} R_0&=\$2{,}400{,}000, &K_0&=1.10R_0=\$2{,}640{,}000\\ R_1&=\$1{,}200{,}000, &K_1&=1.10R_1=\$1{,}320{,}000\\ R_2&=\$480{,}000, &K_2&=1.10R_2=\$528{,}000\\ R_3&=\$120{,}000, &K_3&=1.10R_3=\$132{,}000\\ R_4&=\$0, &K_4&=\$0 \end{aligned}

If the opening security were exactly $2,640,000 and every year’s actual payment matched the projection, the arithmetic would indicate potential reductions of $1,320,000 after year 1, $792,000 after year 2, and $396,000 after year 3, leaving $132,000 until year 4. These are illustrative differences between successive targets, not automatic release instructions. Re-estimation, payment verification, and the program’s release conditions still matter. The common mistake is to use the amount paid in a year as the amount of security to release; the target is based on the remaining projected obligation, including the stated margin.

A simple measure of runoff duration is the payment-weighted average time. If $q_j$ is paid at the end of year $j$, the average time is $\sum_j jq_j$. In the example, it is $1(0.50)+2(0.30)+3(0.15)+4(0.05)=1.75$ years. This summarizes the schedule but does not mean that all obligations disappear after 1.75 years. A small share may remain for several years, and that late share can determine the final security release date.

Key distinction: A tail factor estimates the total amount expected to develop beyond an observed age. A runoff schedule allocates the projected unpaid amount across future periods. The tail factor informs the size of the projected obligation; the runoff pattern informs its expected timing. Applying both to the same projected amount is appropriate only when the runoff shares allocate that amount, not when they add a second tail estimate.

Practical Checks and Common Traps

Tail and runoff estimates are sensitive to definitions and timing. Before using a result to inform security, check that the numerator and denominator of each factor are consistent; the selected tail begins after the last observed development age; and the estimated unpaid amount is not confused with ultimate reimbursements. A factor applied to the wrong base can materially distort a projection even when the arithmetic is correct.

Adding factors instead of multiplying them. Successive development is compounded. A chain of factors 1.10 and 1.05 has a combined factor of $1.10 \times 1.05=1.155$, not 2.15.
Applying a tail factor to the wrong amount. If a factor is defined from the current cumulative paid amount to ultimate, apply it to that cumulative amount. Do not apply it to an already calculated unpaid estimate.
Double-counting tail development. Establish the final age represented by the observed factors. A tail factor beginning at that age should not include development already reflected in the last link ratio.
Treating the average duration as the final release date. Weighted-average time summarizes payment timing; it does not show when the last material obligation is expected to be settled.
Releasing solely from a historical schedule. A runoff schedule is an estimate. Actual payments, updated valuations, and the applicable security terms may support a different remaining target.

A useful control is to reconcile the projected ultimate to the valuation-date components: cumulative paid plus projected unpaid should equal ultimate. Separately reconcile the runoff schedule: expected future payments should sum to the projected unpaid amount, and remaining obligations should decline by the projected payments. These checks catch many unit, timing, and double-counting errors before a security recommendation is made.

Check Your Understanding

Work each projection on a consistent nominal-dollar basis, and distinguish ultimate from unpaid obligations.

A hypothetical program has $3,200,000 paid at 24 months. The selected 24-to-36-month factor is 1.15, and the tail factor after 36 months is 1.04. Find the CDF to ultimate, projected ultimate, and unpaid amount. Assume no discounting. (Answer: CDF $=1.196$; ultimate $=\$3{,}827{,}200$; unpaid $=\$627{,}200$.)
A portfolio has $1,500,000 paid at 48 months. The 48-to-60-month factor is 1.02, and the tail factor after 60 months is 1.03. Calculate the ultimate and unpaid amounts. (Answer: ultimate $=\$1{,}575{,}900$; unpaid $=\$75{,}900$.)
Estimated unpaid reimbursements are $900,000, with runoff shares of 40%, 35%, 20%, and 5% in years 1 through 4. With a 8% margin on remaining obligations and no discounting, find the security target after year 2. (Answer: remaining obligation $=\$225{,}000$; target $=\$243{,}000$.)
A hypothetical portfolio has $750,000 paid at the last observed age. Selected future factors are 1.04, 1.02, and 1.01. Find the tail factor, ultimate, and additional development, rounding money to cents. (Answer: tail factor $=1.071408$; ultimate $=\$803{,}556.00$; additional development $=\$53{,}556.00$.)
A runoff schedule assigns 30% of the unpaid amount to year 1, 40% to year 2, 20% to year 3, and 10% to year 4. Calculate the payment-weighted average time and explain why it is not the final release date. (Answer: $2.10$ years; it is an average, while 10% remains scheduled for year 4.)

What You'll Learn