Why Separate Loss-Cost and Exposure Sensitivity?
An exposure-based expected loss estimate depends on two selections: the number of exposures being measured and the expected loss cost per exposure. If the estimate changes, it matters whether the change came from one input or both. Separating these effects helps an actuary explain a change, test alternative assumptions, and avoid attributing an exposure-volume change to the loss-cost selection.
For example, an exposure might be one earned vehicle-year, one insured location-year, or another consistently defined unit. The selected expected loss cost must be expressed per that same unit. An estimate based on 10,000 vehicle-years and a cost per vehicle-year cannot be compared directly with one based on policy counts and a cost per policy unless the units are reconciled.
Here, \(U\) is an expected ultimate loss estimate, not a guarantee of the losses that will occur. The sensitivity calculations below treat the selected inputs as fixed point estimates. They do not require a claim-frequency distribution, severity distribution, mortality assumption, or interest rate: no stochastic distribution or discounting is used in these examples.
Changing One Input at a Time
A one-at-a-time sensitivity isolates an input by holding the other input fixed. If the exposure count changes from \(E_0\) to \(E_1\), while the expected loss cost remains \(L_0\), the corresponding estimate change is \(L_0(E_1-E_0)\). If the expected loss cost changes from \(L_0\) to \(L_1\), while exposure remains \(E_0\), the change is \(E_0(L_1-L_0)\).
These are conditional comparisons, not claims that the fixed input is known without error. They answer a specific question: how much would the estimate change if this one selection changed and the other did not? The answer is in dollars when \(E\) is an exposure count and \(L\) is dollars per exposure.
Worked Example: Isolating a Loss-Cost Change
A hypothetical portfolio has 8,400 earned vehicle-years. The selected expected loss cost changes from $142.50 to $151.20 per vehicle-year. Calculate the estimate under each selection and isolate the effect of the loss-cost change. No interest discounting or mortality model is used; the selections are deterministic point estimates.
Method. Hold the exposure count fixed and multiply each loss-cost selection by 8,400 vehicle-years.
Original estimate. \(8{,}400\times \$142.50=\$1{,}197{,}000.00\).
Revised estimate. The cost increase per vehicle-year is \(\$151.20-\$142.50=\$8.70\). Thus, \(8{,}400\times \$151.20=\$1{,}270{,}080.00\).
Isolated effect. \(8{,}400\times \$8.70=\$73{,}080.00\). Subtracting the estimates gives \(\$1{,}270{,}080.00-\$1{,}197{,}000.00=\$73{,}080.00\), which checks the result.
The sensitivity is $8,400 for each $1.00 change in selected loss cost, because there are 8,400 exposures. A common mistake is to multiply the $8.70 change by a revised exposure count from a different scenario. That would mix two effects instead of isolating the loss-cost change.
Worked Example: Isolating an Exposure-Count Change
A hypothetical equipment program has 6,750 insured unit-years at a selected expected loss cost of $236.40 per unit-year. A revised count is 7,020 unit-years, with the loss-cost selection unchanged. Calculate both estimates and the exposure effect. No interest discounting or mortality model is used; the selections are deterministic point estimates.
Method. Hold the $236.40 loss cost fixed. The exposure increase is \(7{,}020-6{,}750=270\) unit-years.
Original estimate. \(6{,}750\times \$236.40=\$1{,}595{,}700.00\).
Revised estimate. \(7{,}020\times \$236.40=\$1{,}659{,}528.00\).
Exposure effect. \(270\times \$236.40=\$63{,}828.00\). The difference between the revised and original estimates is also \(\$1{,}659{,}528.00-\$1{,}595{,}700.00=\$63{,}828.00\). The two calculations agree.
The unit check is useful: unit-years multiplied by dollars per unit-year produce dollars. A common mistake is to use the exposure percentage change as though it were a dollar change. The revised estimate is about 4.00% higher because \(270/6{,}750=0.04\), but the dollar increase is $63,828, not 0.04 dollars.
When Both Selections Change
If both inputs change, the total difference compares \(U_0=E_0L_0\) with \(U_1=E_1L_1\). The difference is not generally the sum of the two one-at-a-time effects calculated at the original inputs. There is also an interaction: the change in exposure applies to a changed loss cost, or the change in loss cost applies to changed exposure.
To see why the formula holds, substitute \(E_1=E_0+\Delta E\) and \(L_1=L_0+\Delta L\) into the new estimate:
Subtracting \(U_0=E_0L_0\) leaves the three terms in the decomposition. The interaction term is positive when both changes have the same sign and negative when their signs differ. It can be material even when each individual change looks modest.
Worked Example: Reconciling Two Simultaneous Changes
A hypothetical liability segment has 4,800 exposures and an expected loss cost of $310 per exposure. A revised selection uses 5,100 exposures and $298 per exposure. Calculate the original and revised estimates, then decompose the change. No interest discounting or mortality model is used; the selections are deterministic point estimates.
Original estimate. \(U_0=4{,}800\times \$310=\$1{,}488{,}000.00\).
Revised estimate. \(U_1=5{,}100\times \$298=\$1{,}519{,}800.00\). Therefore, the total change is \(\$1{,}519{,}800.00-\$1{,}488{,}000.00=\$31{,}800.00\).
Calculate the input changes. \(\Delta E=5{,}100-4{,}800=300\) exposures, and \(\Delta L=\$298-\$310=-\$12\) per exposure.
Loss-cost effect at original exposure. \(E_0\Delta L=4{,}800\times(-\$12)=-\$57{,}600.00\).
Exposure effect at original loss cost. \(L_0\Delta E=\$310\times300=\$93{,}000.00\).
Interaction effect. \(\Delta E\Delta L=300\times(-\$12)=-\$3{,}600.00\).
Check. \(-\$57{,}600.00+\$93{,}000.00-\$3{,}600.00=\$31{,}800.00\), matching the direct difference. Exposure growth raises the estimate, while the lower loss cost reduces it; the net change is positive.
The attribution depends on the order used for a sequential explanation. If exposure changes first, its effect is \(300\times\$310=\$93{,}000\), then the loss-cost effect at 5,100 exposures is \(5{,}100\times(-\$12)=-\$61{,}200\). These sum to $31,800. If loss cost changes first, its effect is \(-\$57,600\), then exposure changes at $298 per exposure, adding \(300\times\$298=\$89{,}400\); again the sum is $31,800. The total is fixed, but the allocation of the interaction across sequential effects differs.
Relative Sensitivity and Interpretation
When both inputs are positive, a proportional comparison can complement the dollar decomposition. Since \(U=EL\), the ratio of revised to original estimate is the product of the input ratios:
Thus, relative changes do not simply add for finite changes. If exposure changes by a proportion \(e\) and expected loss cost by a proportion \(l\), then the estimate changes by the factor \((1+e)(1+l)=1+e+l+el\). The product term \(el\) is the relative interaction. This comparison is useful for communicating scale, while the dollar decomposition shows the monetary effect.
A sensitivity result is conditional on the selected inputs and the defined basis. It does not establish that a revised selection is more accurate, nor does it measure the uncertainty around an ultimate loss estimate. It answers how the arithmetic estimate responds to the specified input changes.
Common Exam Traps and Basis Checks
- Changing both inputs while calling the result one-at-a-time sensitivity. State which input is fixed. If both change, show the interaction or clearly state the sequential order.
- Mixing exposure definitions. Earned vehicle-years, written policies, and insured locations are not interchangeable without a conversion or reconciliation.
- Using a loss cost on a different scope. The exposure count and expected loss cost must match for coverage, period, and included loss components. For example, a cost including allocated loss adjustment expense should not be combined with an exposure count paired with a loss-only selection without recognizing the basis difference.
- Confusing a relative change with a dollar change. A percentage change in exposure scales the estimate; it is not itself an amount of loss.
- Omitting the interaction term. Adding \(E_0\Delta L\) and \(L_0\Delta E\) alone does not generally equal the total finite change. The missing amount is \(\Delta E\Delta L\).
Check Your Understanding
For each problem, use consistent exposure and loss-cost units and calculate the requested estimate or sensitivity.
- A hypothetical portfolio has 2,400 earned location-years and a selected expected loss cost of $87.50 per location-year. What is its exposure-based expected ultimate loss estimate? (Answer: $210,000.00)
- A segment has 9,200 exposures. Its expected loss cost increases from $64.00 to $68.50 per exposure, with exposure fixed. What is the change in the estimate? (Answer: increase of $41,400.00)
- A hypothetical program’s exposure count increases from 3,150 to 3,360, while its loss cost remains $412 per exposure. What is the change in the estimate? (Answer: increase of $86,520.00)
- A segment changes from 2,000 exposures at $525 per exposure to 2,150 exposures at $510 per exposure. Calculate the total estimate change and its decomposition into original-basis loss-cost, exposure, and interaction effects. (Answer: increase of $46,500.00; effects are \(-\$30,000.00\), \(+\$78,750.00\), and \(-\$2,250.00\), respectively.)
- A hypothetical portfolio’s exposure count falls by 4% while selected loss cost rises by 6%. If the original expected ultimate estimate is $3,750,000, what is the revised estimate and percentage change? (Answer: $3,816,000.00, an increase of 1.76%.)