Why Alignment Comes Before an Indication
An indication can be arithmetically correct and still be misleading if its inputs describe different things. For example, dividing losses for one coverage by premium for a broader package can produce a precise ratio that does not measure the selected coverage. The same problem arises when exposure is assigned by one time basis, premium by another, and losses by a third without a reconciliation.
The practical task is to build a data map before calculating an indication. For each experience period, the map identifies the selected coverage, the eligible population, the time basis, the exposure measure, the premium measure, and the loss measure. This is a data-control step, not a substitute for selecting the experience period, on-leveling premium, or projecting losses using the methods established in the earlier material.
Specify the Target Before Extracting Records
Write down the target in operational terms. A useful specification answers four questions:
- Coverage: Which coverage or coverages are included?
- Population: Which policies, risks, locations, or insured items qualify?
- Time basis: Is the period defined by accident date, policy period, calendar period, or another selected basis?
- Measurement basis: What exposure unit, premium measure, and loss measure will be used?
The time basis deserves particular attention. An accident-year loss record is assigned according to when the accident occurred. An earned exposure or earned premium record accrues as coverage is provided. A policy-year record is associated with the policies’ effective periods. These labels are not interchangeable. If the selected analysis is by accident year, records must be mapped to accident years using the appropriate dates; matching the text “2025” in two files does not establish that both files represent the same period.
Premium and loss amounts also need not be at the same valuation stage. For example, the selected premium base may be earned and on-level, while the selected loss measure may be ultimate. Those preparations do not change the need to align coverage, population, and period. State the basis for each input explicitly so that a reviewer can distinguish a deliberate adjustment from a data mismatch.
Record the coverage, eligible population, experience-period definition, and exposure unit before reviewing totals.
Use the date and scope fields relevant to that measure. For example, use accident date to assign a claim to an accident period.
Separate earned from written premium and identify the chosen loss valuation basis. Do not treat a measure’s preparation as a change to its target scope.
Compare included record counts and totals with source reports, explain exclusions, and check that all three measures refer to the same target.
A compact map makes the choices visible. The following symbols describe a single selected coverage and population for period \(t\):
The superscripts identify the chosen measurement bases; they do not imply that every analysis must use those exact measures. A data map should instead name the measures actually selected. If loss adjustment expense is included, the map should state that and use the corresponding loss basis consistently.
Worked Examples of Building and Checking the Map
Worked Example: Reconciling Quarterly Records to an Accident Year
A hypothetical insurer is preparing an accident-year 2024 data extract for one homeowners coverage and one defined policy population. The exposure and premium team has divided earned records into calendar quarters. Exposure is in house-years, and premium has already been earned and adjusted to the current rate level. The selected loss file contains ultimate losses assigned by accident date. Determine the annual totals and calculate the loss cost per exposure unit and loss ratio as alignment checks.
Method. Confirm that every quarter belongs to the same target coverage and population, then add the quarterly exposure and premium. The loss file is already organized by the selected accident-year basis, so use its stated total rather than trying to assign losses by policy effective date.
| Quarter of 2024 | Earned exposure (house-years) | On-level earned premium |
|---|---|---|
| Q1 | 300 | $48,000 |
| Q2 | 320 | $51,200 |
| Q3 | 280 | $44,800 |
| Q4 | 300 | $49,200 |
Step 1: Add exposure. \(E_{2024}=300+320+280+300=1{,}200\) house-years.
Step 2: Add premium. \(P_{2024}^{\mathrm{OL}}=\$48{,}000+\$51{,}200+\$44{,}800+\$49{,}200=\$193{,}200\).
Step 3: Use the selected accident-year loss total. The loss file reports \(L_{2024}^{\mathrm{ult}}=\$116{,}000\) for this same coverage and population.
Step 4: Calculate the checks. Loss cost is \(\$116{,}000/1{,}200=\$96.666\ldots\) per house-year. The loss ratio is \(\$116{,}000/\$193{,}200=0.600414\ldots\), or \(60.04\%\). Rounded to cents for loss cost, the result is \(\$96.67\) per house-year.
Check and trap avoided. The quarterly exposure sums to the stated annual exposure, and the premium sum is $193,200. A common mistake is to match a policy-year premium total labeled “2024” directly to the accident-year 2024 loss total. The quarter-by-quarter earned records and the stated accident-year loss basis make the intended period explicit. These checks do not, by themselves, project losses to a prospective period.
Worked Example: Splitting Policy Records Across Accident Years
A hypothetical insurer selects accident year 2025 for a small commercial coverage. Assume coverage is earned uniformly over each policy term, there are no midterm cancellations or changes in exposure, and the annual premiums below are earned pro rata. All three policies belong to the selected coverage and population. Map the policies’ exposure and earned premium to accident year 2025, then compare them with $7,000 of ultimate losses assigned by accident date to that year.
| Policy term | Annual exposure | Annual premium | Portion in calendar 2025 |
|---|---|---|---|
| July 1, 2024–June 30, 2025 | 1.00 unit | $1,200 | 6 of 12 months |
| January 1–December 31, 2025 | 1.00 unit | $2,400 | 12 of 12 months |
| October 1, 2025–September 30, 2026 | 1.00 unit | $1,600 | 3 of 12 months |
Method. Allocate earned exposure and premium to the selected accident-year calendar period using the stated uniform earning assumption. Assign losses by accident date, not by policy start date.
Step 1: Allocate exposure. The first policy contributes \(1.00(6/12)=0.50\) unit. The second contributes \(1.00(12/12)=1.00\) unit. The third contributes \(1.00(3/12)=0.25\) unit. Thus \(E_{2025}=0.50+1.00+0.25=1.75\) units.
Step 2: Allocate earned premium. The first policy contributes \(\$1{,}200(6/12)=\$600\). The second contributes \(\$2{,}400(12/12)=\$2{,}400\). The third contributes \(\$1{,}600(3/12)=\$400\). The 2025 earned premium is \(\$600+\$2{,}400+\$400=\$3{,}400\).
Step 3: Compare on the mapped basis. Ultimate losses assigned to accident year 2025 are $7,000. The loss cost is \(\$7{,}000/1.75=\$4{,}000\) per exposure unit, and the loss ratio is \(\$7{,}000/\$3{,}400=2.058823\ldots\), or \(205.88\%\).
Check and trap avoided. The exposure allocation totals \(1.75\) units and the premium allocation totals $3,400. Assigning the entire first and third policies to 2025 because each overlaps that year would overstate both measures. The pro-rata allocation is valid here because the problem explicitly supplies uniform earning and no midterm changes; it should not be assumed where those conditions do not hold.
Worked Example: Matching Claim Coverage to the Exposure and Premium Scope
A hypothetical insurer is reviewing accident year 2026 for scheduled-drone physical damage coverage. The eligible population is scheduled drones covered by that coverage. The exposure file reports 800 drone-years, and the earned premium adjusted to the current rate level is $240,000. The ultimate claim register contains three physical damage losses of $24,000, $36,000, and $20,000, plus two liability losses of $18,000 and $42,000. Calculate the aligned loss cost and loss ratio, and show the effect of including the wrong claims.
Method. Filter the claim register to the selected coverage and eligible population before summing. The exposure and premium already describe scheduled-drone physical damage, so liability losses do not belong in their numerator.
Step 1: Sum eligible losses. \(L_{2026}^{\mathrm{ult}}=\$24{,}000+\$36{,}000+\$20{,}000=\$80{,}000\).
Step 2: Calculate aligned measures. Loss cost is \(\$80{,}000/800=\$100\) per drone-year. The loss ratio is \(\$80{,}000/\$240{,}000=0.3333\ldots\), or \(33.33\%\).
Step 3: Diagnose the tempting error. Adding the two liability losses would produce a loss total of \(\$80{,}000+\$18{,}000+\$42{,}000=\$140{,}000\). Dividing that by the physical-damage premium gives \(\$140{,}000/\$240{,}000=58.33\%\). This is not a valid physical-damage loss ratio: its numerator includes a different coverage from its denominator.
Check and trap avoided. The physical damage claim total is $80,000, matching the selected coverage. The incorrect $140,000 total is arithmetically correct for all listed claims but belongs to a broader coverage scope. A matching calendar or accident year alone does not make records comparable.
Why Consistent Aggregation Matters
When a portfolio is divided into groups, aggregation should preserve the same scope and measurement basis. Suppose every included group \(j\) has positive exposure \(E_j>0\), and the total exposure and loss are \(E=\sum_j E_j\) and \(L=\sum_j L_j\). Then the portfolio loss cost is an exposure-weighted average of the group loss costs:
To see why, expand each term on the right: \(\frac{E_j}{E}\frac{L_j}{E_j}=\frac{L_j}{E}\), where cancellation is valid because \(E_j>0\). Summing gives \(\sum_j L_j/E=(\sum_j L_j)/E=L/E\). The weights \(E_j/E\) sum to one because \(\sum_j E_j=E\).
The condition \(E_j>0\) is essential for every group represented by a group loss cost. A zero-exposure group makes \(L_j/E_j\) undefined, even when total portfolio exposure is positive. Handle such a group separately. In particular, if it has nonzero loss, that loss cannot be represented by the stated weighted-average identity; investigate the records and their scope before using the aggregate.
This identity is also an audit tool: if group and portfolio totals do not reconcile, possible causes include omitted records, duplicate records, mismatched coverage filters, inconsistent period assignment, or different measurement bases. Resolving the data issue is preferable to forcing the totals to agree through an unexplained adjustment.
Common Alignment Traps
- Matching year labels instead of period definitions. “2025” might mean policy year, calendar year, or accident year. Confirm the date rule behind each label.
- Using written premium as if it were earned premium. Written premium is associated with policies written; earned premium reflects coverage provided. Use the premium measure selected for the analysis and document it.
- Including losses from a broader coverage. A claim file may contain multiple coverages. Filter to the target scope before summing, and confirm the exposure and premium use that same scope.
- Assuming an allocation rule without support. Prorating by months requires an appropriate earning assumption. If exposure does not accrue uniformly, a simple month fraction may not represent the actual earned amount.
- Using an undefined group loss cost. The weighted-average formula requires positive exposure in every group whose loss cost appears. Zero-exposure groups need separate investigation or treatment.
Check Your Understanding
For each problem, first identify the target scope and time basis, then perform the calculation. Answers use the assumptions stated in each problem.
- A hypothetical accident-year 2027 extract has quarterly exposure of 90, 110, 100, and 100 units; current-rate-level earned premium of $18,000, $22,000, $20,000, and $20,000; and ultimate losses of $64,000. Find annual exposure, premium, loss cost, and loss ratio. (Answer: 400 units; $80,000; $160 per unit; 80.00%.)
- A policy earns exposure and premium uniformly, with no cancellations. It runs from April 1, 2026, through March 31, 2027, with annual exposure 1.00 unit and annual premium $1,800. How much belongs to calendar 2026 under a pro-rata allocation? (Answer: 0.75 exposure unit and $1,350.)
- For a selected coverage, earned exposure is 250 units and on-level earned premium is $75,000. The aligned ultimate losses are $21,000 and $9,000. A separate $12,000 claim is for another coverage. Calculate the aligned loss cost and loss ratio, and state the loss total to use. (Answer: use $30,000; loss cost $120 per unit; loss ratio 40.00%.)
- Two included groups have exposures and losses of \(E_1=100,L_1=\$8{,}000\) and \(E_2=300,L_2=\$36{,}000\). Find the portfolio loss cost both from totals and as an exposure-weighted average. (Answer: $44,000/400 = $110 per unit; \((100/400)(\$80)+(300/400)(\$120)=\$110\) per unit.)
- A portfolio has one group with exposure 200 and losses of $10,000, and another group with exposure zero and losses of $2,000. Can the portfolio loss cost be written as the exposure-weighted average of both groups’ loss costs? Explain and calculate the overall total-loss cost if the records are confirmed to belong to the same target scope. (Answer: No; the second group loss cost is undefined because its exposure is zero. The total-loss cost is \($12,000/200=\$60\) per unit, but the zero-exposure group’s loss requires separate investigation and cannot be expressed in the weighted-average identity.)