Development Factor Sensitivity by Maturity

Why Maturity Matters in Development-Factor Sensitivity

A selected age-to-age factor affects only cohorts that still need that development step projected. A change to an early factor may affect many cohorts, while a change to a later factor affects only cohorts that have not yet reached that maturity. The size of the effect also depends on the amount of loss entering the link and on the selected factors after it.

This tutorial measures those effects using a deterministic chain-ladder projection. It focuses on sensitivity to development-factor selections by maturity, not on selecting a reserving method or estimating a tail factor. All examples use hypothetical loss data. No interest or mortality assumptions apply, and no probability distribution is assumed: the calculations are deterministic.

Definition: Let $C_{i,j}$ be cumulative loss for cohort $i$ at development age $j$, and let $f_j$ be the selected age-to-age factor from age $j$ to the next age. For a cohort observed at age $m$, its projected ultimate is $U_{i,m}=C_{i,m}\prod_{j=m}^{J} f_j$, where $J$ is the final projected link. A factor sensitivity is the change in that projected ultimate when one or more selected factors change, with the observed cumulative loss and other specified assumptions held fixed.

The maturity convention is important. A cohort already observed at 36 months does not use the 12-to-24 or 24-to-36 factors in its projection: those links have already occurred. It may still use the 36-to-48 factor and subsequent selected factors. A cohort at 12 months may use all of those links.

Deriving the Effect of One Factor Change

Suppose the selected factor $f_k$ is used in the projection for a cohort observed at age $m$, with $m\leq k$. Hold all other selected factors fixed, and change $f_k$ by an absolute amount $\Delta f_k$. Write the product of all projected factors other than $f_k$ as $Q_{i,m,k}=\prod_{j=m,\,j\ne k}^{J} f_j$. The baseline and alternative ultimates are then:

U_{i,m}=C_{i,m}Q_{i,m,k}f_k,\qquad U'_{i,m}=C_{i,m}Q_{i,m,k}(f_k+\Delta f_k).

Subtracting the first expression from the second gives the exact change:

\Delta U_{i,m} =U'_{i,m}-U_{i,m} =C_{i,m}Q_{i,m,k}\Delta f_k =U_{i,m}\frac{\Delta f_k}{f_k}.

This is exact, not an approximation, provided only one factor changes and the other factors and observed loss are held fixed. The last equality shows why the same absolute factor-point change does not usually produce the same dollar effect at different maturities: each cohort has a different observed loss and a different product of remaining factors. If $m>k$, the factor is already part of the cohort’s observed development and is not used in its remaining projection; changing the selected factor therefore has no effect on that cohort’s projected ultimate under this setup.

Key result: For a one-factor change, include only cohorts whose projections use that factor. For each included cohort, multiply its observed cumulative loss by the changed factor amount and by every other projected factor. Sum cohort-level changes to obtain the aggregate change.

Worked Examples

Worked Example: One Factor Change Across Two Maturities

A hypothetical liability portfolio has a 12-month cohort with cumulative loss of $500,000 and a 24-month cohort with cumulative loss of $800,000. The selected factors are $f_{12}=1.40$, $f_{24}=1.20$, and $f_{36}=1.10$, projecting to 48 months. Consider an alternative selection that raises $f_{24}$ from 1.20 to 1.23. Find the change in ultimate for each cohort and in total. Interest is not applied, mortality is not applicable, and the projection is deterministic with the stated factors.

Method. Apply the one-factor sensitivity formula. The factor $f_{24}$ is in the remaining projection for both cohorts. For the 12-month cohort, the other projected factors are $f_{12}=1.40$ and $f_{36}=1.10$; for the 24-month cohort, the other projected factor is $f_{36}=1.10$. The absolute factor change is $1.23-1.20=0.03$.

For the 12-month cohort, the baseline ultimate is $500{,}000\times1.40\times1.20\times1.10=\$924{,}000$. The alternative ultimate is $500{,}000\times1.40\times1.23\times1.10=\$947{,}100$. Its increase is $\$947{,}100-\$924{,}000=\$23{,}100$. Equivalently, the sensitivity formula gives $500{,}000\times1.40\times0.03\times1.10=\$23{,}100$.

For the 24-month cohort, the baseline ultimate is $800{,}000\times1.20\times1.10=\$1{,}056{,}000$. The alternative ultimate is $800{,}000\times1.23\times1.10=\$1{,}082{,}400$. Its increase is $\$26{,}400$, also given by $800{,}000\times0.03\times1.10$.

The aggregate baseline ultimate is $\$924{,}000+\$1{,}056{,}000=\$1{,}980{,}000$; the alternative aggregate is $\$947{,}100+\$1{,}082{,}400=\$2{,}029{,}500$. Thus the aggregate increase is $\$49{,}500$, which checks against $\$23{,}100+\$26{,}400$. The common mistake avoided is applying the factor change to a cohort’s observed loss without including its other projected factors.

Worked Example: Several Factor Changes and Their Interaction

A hypothetical cohort has cumulative loss of $200,000 at 12 months and selected factors $f_{12}=1.40$, $f_{24}=1.20$, and $f_{36}=1.10$, projecting to 48 months. In an alternative scenario, $f_{12}=1.50$ and $f_{24}=1.25$, while $f_{36}$ remains 1.10. Calculate the baseline ultimate, the alternative ultimate, and the change. Interest is not applied, mortality is not applicable, and no probability distribution is assumed.

Method. Calculate both complete factor products. The baseline ultimate is $200{,}000\times1.40\times1.20\times1.10$. Since $1.40\times1.20=1.68$, and $1.68\times1.10=1.848$, the baseline is $\$369{,}600$.

The alternative product is $1.50\times1.25\times1.10$. First, $1.50\times1.25=1.875$; then $1.875\times1.10=2.0625$. The alternative ultimate is $200{,}000\times2.0625=\$412{,}500$. The increase is $\$412{,}500-\$369{,}600=\$42{,}900$.

The interaction can be seen by changing the factors in sequence. First change $f_{12}$ while leaving the later factors at baseline: the increase is $200{,}000\times(1.50-1.40)\times1.20\times1.10=\$26{,}400$. Next, with $f_{12}$ already at 1.50, change $f_{24}$: the increase is $200{,}000\times1.50\times(1.25-1.20)\times1.10=\$16{,}500$. Together these sequential changes are $\$26{,}400+\$16{,}500=\$42{,}900$, matching the full scenario calculation.

The common mistake avoided is adding isolated factor effects calculated at the baseline as if they were independent. The baseline-only effect of changing $f_{24}$ would be $200{,}000\times1.40\times0.05\times1.10=\$15{,}400$. Adding that to the $f_{12}$ effect gives only $41,800; the missing $1,100 is the interaction $200{,}000\times0.10\times0.05\times1.10$. For multiple simultaneous changes, compare full scenarios or state the order used to allocate sequential effects.

Worked Example: Ranking Factor Sensitivity by Development Period

A hypothetical portfolio has cumulative losses of $300,000 at 12 months, $500,000 at 24 months, $900,000 at 36 months, and $700,000 at 48 months. The selected factors are $f_{12}=1.50$, $f_{24}=1.20$, $f_{36}=1.10$, and $f_{48}=1.05$, projecting through 60 months. Compare the aggregate dollar effects of these separate alternative selections: $f_{12}$ increases by 0.02, $f_{24}$ by 0.03, $f_{36}$ by 0.04, or $f_{48}$ by 0.02. These are absolute factor-point changes, not relative percentage changes. Interest is not applied, mortality is not applicable, and the projection is deterministic.

For a change to $f_{12}$, only the 12-month cohort uses that factor in its projection. Its increase is $300{,}000\times0.02\times1.20\times1.10\times1.05=\$8{,}316$.

For a change to $f_{24}$, the 12- and 24-month cohorts use the factor. Their increases are $300{,}000\times1.50\times0.03\times1.10\times1.05=\$15{,}592.50$ and $500{,}000\times0.03\times1.10\times1.05=\$17{,}325$. The aggregate increase is $\$32{,}917.50$.

For a change to $f_{36}$, the 12-, 24-, and 36-month cohorts use the factor. Their increases are $300{,}000\times1.50\times1.20\times0.04\times1.05=\$22{,}680$, $500{,}000\times1.20\times0.04\times1.05=\$25{,}200$, and $900{,}000\times0.04\times1.05=\$37{,}800$. The aggregate increase is $\$85{,}680$.

For a change to $f_{48}$, all four cohorts use the factor. The increases are $300{,}000\times1.50\times1.20\times1.10\times0.02=\$11{,}880$, $500{,}000\times1.20\times1.10\times0.02=\$13{,}200$, $900{,}000\times1.10\times0.02=\$19{,}800$, and $700{,}000\times0.02=\$14{,}000$. The aggregate increase is $\$58{,}880$.

Among these stated scenarios, the $f_{36}$ change produces the largest aggregate increase: $85,680. This ranking is conditional on the particular factor changes chosen. Since the shocks differ in size, it is not a normalized comparison of underlying sensitivity. To compare sensitivities on a common basis, use the same absolute factor change, the same relative change, or report the derivative $U/f_k$ alongside the scenario results. The common mistake avoided is concluding that a link is intrinsically most influential without stating the size and type of its shock.

Interpreting and Reporting the Results

A useful sensitivity analysis reports both the cohort-level effects and the aggregate. Aggregation shows the total change under the specified scenario; cohort detail identifies which maturities contribute to it. In general, an early factor may affect more cohorts, but that does not guarantee its aggregate dollar effect is largest. The observed losses, factors remaining after the changed link, and size of the assumed factor change all matter.

When comparing valuation periods, state the change being applied. An increase of 0.02 to a factor of 1.20 is an absolute factor-point change. A 2% relative increase instead gives $1.20\times1.02=1.224$, an absolute increase of 0.024. Those changes produce different projected ultimate effects. Also state the projection endpoint and whether the table includes all relevant cohorts; a factor cannot affect a cohort whose remaining projection does not use it.

A one-way sensitivity table changes one factor at a time while holding the others fixed. This isolates each factor’s effect under the stated baseline. It does not show the full effect of a scenario in which several factors change together; for that, calculate the entire alternative factor product. Report dollar amounts to cents when the inputs and purpose support that precision, and avoid implying that a deterministic scenario difference is a probability or a range of possible outcomes.

Common Mistakes

Using the wrong maturity set. A factor applies only to cohorts whose projected development includes that link. Exclude cohorts already beyond the link.
Omitting other projected factors. The effect of a changed factor is multiplied by the remaining applicable factors, not just by the observed cumulative loss.
Confusing factor points with relative change. Adding 0.02 to a factor is generally not the same as increasing it by 2%.
Adding isolated multi-factor effects without interaction. If more than one factor changes, calculate the complete alternative ultimate or use a clearly stated sequential decomposition.
Overgeneralizing a scenario ranking. The factor with the largest dollar effect under one set of shocks is not necessarily the most sensitive under another.

Key takeaway: Development-factor sensitivity depends on both the size of the factor change and the cohort’s remaining projection. Identify which maturities use the changed link, calculate each cohort’s exact dollar effect, and then aggregate. Interpret any ranking in light of the specified shocks.

Check Your Understanding

For each hypothetical deterministic projection, identify the applicable cohorts and use the stated factor changes. Interest is not applied, mortality is not applicable, and no probability distribution is assumed.

A 12-month cohort has cumulative loss of $250,000. Its factors to 48 months are $f_{12}=1.30$, $f_{24}=1.20$, and $f_{36}=1.10$. If $f_{24}$ rises by 0.04 and the other factors are unchanged, what is the ultimate increase? (Answer: $14,300.)
A 24-month cohort has cumulative loss of $640,000, with $f_{24}=1.15$ and $f_{36}=1.08$ projected to 48 months. If $f_{36}$ rises by 0.03, what is the ultimate increase? (Answer: $22,080.)
A portfolio has $350,000 at 12 months and $620,000 at 24 months. The factors are $f_{12}=1.45$, $f_{24}=1.18$, and $f_{36}=1.07$, projecting to 48 months. If only $f_{36}$ rises by 0.04, calculate the increase for each cohort and in aggregate. (Answer: 12-month increase $23,954; 24-month increase $29,264; aggregate increase $53,218.)
A 12-month cohort has cumulative loss of $150,000 and baseline factors $1.30$, $1.15$, and $1.08$. In an alternative scenario, the first factor is 1.34 and the second is 1.20, while the third is unchanged. Calculate the baseline ultimate, alternative ultimate, and increase. (Answer: baseline $242,190; alternative $260,496; increase $18,306.)
A portfolio has cumulative losses of $240,000 at 12 months, $400,000 at 24 months, and $600,000 at 36 months. Its factors are $f_{12}=1.40$, $f_{24}=1.20$, $f_{36}=1.10$, and $f_{48}=1.05$, projecting to 60 months. Separately increase $f_{24}$ by 0.02, $f_{36}$ by 0.03, or $f_{48}$ by 0.01. Which change gives the largest aggregate increase, and what is that increase? (Answer: increasing $f_{36}$ by 0.03, for an aggregate increase of $46,720.80.)

This sensitivity framework is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

What You'll Learn