Keeping Coverage Definitions Consistent

Why Consistent Definitions Matter

A loss development triangle is useful only if its cells measure a comparable population of claims. If one cell includes collision claims while another includes all auto physical damage, or if one origin period excludes reopened claims while another includes them, a difference between cells may reflect a change in data scope rather than claim development. The triangle can still look orderly while answering the wrong question.

Earlier material on accident-year and other origin-period rows, development ages, and incurred cells establishes how records are assigned to cells. The focus here is the definition of which records belong in those cells. A triangle’s target scope should specify the relevant coverage, line of business, and claim-inclusion criteria, then apply those rules consistently to every applicable cell.

Definition: A coverage definition states which insured coverage or peril is included, such as collision but not comprehensive. A line-of-business definition states which business grouping is included, such as commercial auto liability. A claim-inclusion definition states which claims qualify for the triangle, including any rules for claim status, catastrophe or other exclusions, and claim identification. Together, these rules define the triangle’s target population.

For a cumulative incurred triangle, let $C_{i,j}$ be the cumulative incurred amount for origin period $i$, evaluated at development age $j$. Let $Z_{k,j}$ denote the incurred amount for claim $k$ at that age, and let $I_k$ equal 1 when the claim meets the triangle’s stated eligibility rules and 0 otherwise. In a simple claim-level representation,

C_{i,j}=\sum_{k\in i} I_k Z_{k,j}.

In practice, claims may be split into coverage-level records or require an approved allocation when a claim spans coverages. The essential requirement is that the same target definition governs all cells being compared. The eligible claims observable at different valuation dates will naturally differ as claims emerge; that is not a reason to change the eligibility rule.

Specify the Target Before Summing

A usable scope definition is concrete enough that two analysts applying it to the same source records would select the same items. For example, “auto losses” may be too broad if one analyst includes collision and comprehensive while another includes liability. A more precise definition could be “commercial auto collision losses, excluding comprehensive claims, for covered occurrences assigned to the stated origin period.”

Claim-inclusion rules also need care. A triangle might include all claims for the selected coverage, including closed and reopened claims, while excluding a defined catastrophe segment. That rule is different from including only claims currently open, claims with a payment in the valuation period, or claims that have ever had a positive payment. Those status- or payment-based filters can change the population from one age to another. Do not use them unless they are deliberately part of the intended analysis and are applied consistently.

State the target scope.
Name the coverage, line of business, and claim-inclusion rules. Specify any exclusions and how mixed-coverage claims will be handled.

Map source records to that scope.
Use the coverage and line represented by the underlying risk or claim, not an unexamined label that may have changed over time.

Apply the same rules in every cell.
Retain eligible claims at each valuation and exclude ineligible claims, regardless of origin period or development age.

Check the result.
Review representative included and excluded records, and confirm that each cell’s total reflects the same intended scope.

Worked Applications

Worked Example: Separating Collision From Comprehensive

A hypothetical commercial auto triangle is intended to measure collision incurred losses only. For accident year 2022 at 12 months, two collision claims have incurred amounts of $18,000 and $27,000; a comprehensive claim has incurred of $15,000. At 24 months, the collision claims have incurred amounts of $20,000 and $30,000, and a newly reported collision claim has incurred of $12,000. The comprehensive claim is now at $18,000. For accident year 2023 at 12 months, two collision claims are at $31,000 and $14,000, and a comprehensive claim is at $11,000. Construct the available cells on the stated basis.

Method. Apply the coverage filter first: include collision only. The line of business is commercial auto, and the stated claim scope includes covered collision claims whether newly reported or previously known.

For accident year 2022 at 12 months, include the two collision claims: $18{,}000+27{,}000=\$45{,}000$. Exclude the comprehensive claim.

At 24 months, include the updated incurred amounts for the two collision claims and the newly reported collision claim: $20{,}000+30{,}000+12{,}000=\$62{,}000$. Again, exclude the comprehensive claim.

For accident year 2023 at 12 months, include: $31{,}000+14{,}000=\$45{,}000$. The available triangle entries are therefore $45,000 and $62,000 for accident year 2022 at 12 and 24 months, and $45,000 for accident year 2023 at 12 months.

Check. For accident year 2022, the two existing collision claims increase by $2,000 and $3,000, and the newly reported collision claim contributes $12,000. Thus the total increase is $2{,}000+3{,}000+12{,}000=\$17{,}000$, matching $\$62{,}000-\$45{,}000$. This checks the arithmetic without adding comprehensive loss.

Common mistake avoided: Adding the comprehensive values would give $60,000 at 12 months and $80,000 at 24 months for accident year 2022. Those totals answer a broader coverage question, not the stated collision-only question.

Worked Example: Mapping Line Codes Across a System Change

A hypothetical triangle is intended to contain commercial auto collision losses. A source-system change has left claims under several codes. For accident year 2021 at 12 months, a legacy collision code contains $90,000 and a comprehensive code contains $40,000. At 24 months, the collision amount is $100,000 and a newly reported collision claim, verified from its claim details, is $20,000; comprehensive is $45,000. For accident year 2022 at 12 months, a new “COLL” code contains $70,000, another “vehicle damage” code contains $25,000 verified as collision, and comprehensive is $35,000. Calculate the cells using a consistent collision definition.

Method. Map records to the underlying coverage, not merely to a code’s name. Include both verified collision codes, and exclude comprehensive in every origin period and age.

For accident year 2021 at 12 months, the eligible collision amount is $90,000. At 24 months it is $100{,}000+20{,}000=\$120{,}000$. For accident year 2022 at 12 months it is $70{,}000+25{,}000=\$95{,}000$. The comparable cells are $90,000, $120,000, and $95,000, respectively.

Check. The accident year 2021 change is $\$120{,}000-\$90{,}000=\$30{,}000$, equal to the $10,000 increase on the existing collision amount plus the $20,000 newly reported collision claim. If the “vehicle damage” code were omitted, the accident year 2022 cell would be understated by $25,000. If comprehensive were included as well, the cells would instead mix coverages.

Common mistake avoided: Treating a code label as a permanent coverage definition. Code mappings can change; verify and apply a consistent business definition.

Worked Example: Keeping Claim Eligibility Stable

A hypothetical triangle is defined as ordinary wind losses under a specified property coverage, excluding named-storm losses and theft. The inclusion rule retains eligible claims whether open, closed, reopened, or newly reported. At 12 months for accident year 2020, ordinary wind claims have incurred amounts of $12,000 and $8,000; named-storm incurred is $30,000 and theft incurred is $5,000. At 24 months, the first two eligible claims are at $13,000 and $9,000, a newly reported ordinary wind claim is at $4,000, and a reopened eligible claim is at $2,000. Named-storm incurred is then $35,000, and theft is $5,000. For accident year 2021 at 12 months, ordinary wind claims are at $15,000 and $7,000, and named-storm incurred is $20,000. Calculate the cells.

Method. Apply the same peril and claim-status criteria to all three cells. Include all ordinary wind claims, including the newly reported and reopened claims; exclude named-storm and theft claims.

Accident year 2020 at 12 months is $12{,}000+8{,}000=\$20{,}000$. At 24 months it is $13{,}000+9{,}000+4{,}000+2{,}000=\$28{,}000$. Accident year 2021 at 12 months is $15{,}000+7{,}000=\$22{,}000$.

Check. For accident year 2020, the existing ordinary wind claims increase by $1{,}000+1{,}000=\$2{,}000$; the newly reported and reopened eligible claims add $4{,}000+2{,}000=\$6{,}000$. The total increase is $8,000, which agrees with $\$28{,}000-\$20{,}000$. Neither excluded peril enters the calculation.

Common mistake avoided: Excluding a reopened claim because it was closed at an earlier valuation, or excluding a claim solely because it had not yet been reported. Those are not the stated eligibility rules.

Why Stable Eligibility Supports Comparison

The basic reason for fixing the definition can be seen by representing eligibility with a single indicator $I_k$ for each claim. If the eligibility rule is stable across two development ages, then $I_k$ is the same at both ages. For a fixed origin period,

C_{i,j+1}-C_{i,j} =\sum_{k\in i} I_k\left(Z_{k,j+1}-Z_{k,j}\right).

This follows by subtracting the two cell sums: both sum over the same eligible claims, so each claim contributes its later incurred amount less its earlier incurred amount. Claims not meeting the target definition have $I_k=0$ at both ages and contribute nothing. If the inclusion rule changes with age, the sums no longer represent the same eligible population; an apparent change can then combine changes in claim amounts with changes in which claims were selected.

The same principle applies across origin periods. Consistent definitions make the cells represent the same type of business, but do not make different years identical in exposure, claim mix, or underlying risk. The purpose is narrower and essential: prevent differences caused solely by inconsistent selection or classification from being mistaken for meaningful loss development or origin-period variation.

Common Definition Traps

Using broad labels without specifying coverage. “Property” or “auto” can include distinct coverages. State precisely what is in scope.
Applying a source code as if its meaning never changed. Verify code mappings across system changes and use a consistent underlying definition.
Filtering by claim status at each valuation. If the intended rule includes eligible claims regardless of open, closed, or reopened status, do not drop claims merely because their status changed.
Confusing emergence with eligibility. A covered claim may appear in a later valuation because it was reported later. Apply the same qualification rule; do not treat “not yet observed” as “ineligible.”
Counting a mixed-coverage claim twice or assigning it inconsistently. Use a consistent coverage-level split or allocation, and ensure the same loss is not duplicated across target populations.

Key takeaway: Before constructing or comparing triangle cells, state the target coverage, line of business, and claim-inclusion rules. Map each record to those definitions and apply them consistently across origin periods and development ages. Stable scope does not eliminate real differences in experience; it prevents avoidable differences in scope from obscuring them.

Check Your Understanding

For each problem, apply one consistent target definition to every stated cell.

A hypothetical renters-property triangle includes water-damage claims only and includes newly reported eligible claims. For accident year 2024 at 12 months, two water claims are $11,000 and $16,000, while a liability claim is $9,000. At 24 months, the water claims are $13,000 and $19,000, a new water claim is $4,000, and liability is $12,000. What are the two cell values? (Answer: $27,000 at 12 months and $36,000 at 24 months.)
A hypothetical commercial auto triangle includes collision only. At 12 months for one origin year, a verified collision code is $52,000, a second code verified as collision is $18,000, and comprehensive is $20,000. At 12 months for a second origin year, collision codes are $46,000 and $27,000, and comprehensive is $31,000. What are the comparable cell values? (Answer: $70,000 and $73,000.)
A hypothetical fire-coverage triangle includes fire claims and excludes flood. For one origin year, fire incurred is $6,000 and $9,000 at 12 months, while flood is $4,000. At 24 months, the first two fire claims are $8,000 and $12,000, a newly reported fire claim is $3,000, and flood is $5,000. What are the cells, and what is the increase? (Answer: $15,000 at 12 months, $23,000 at 24 months, and an $8,000 increase.)
A hypothetical equipment-breakdown triangle includes eligible claims whether closed or reopened and excludes equipment theft. At six months, eligible claims are $14,000 and $0, while theft is $7,000. At 12 months, the claims are $16,000 and $3,000, a newly reported eligible claim is $5,000, and theft is $9,000. What are the eligible cell totals? (Answer: $14,000 at six months and $24,000 at 12 months.)
A hypothetical liability triangle includes liability claims only. For accident year 2025 at 12 months, a legacy liability code is $40,000, a verified newly mapped bodily-injury code is $12,000, and property damage is $8,000. At 24 months, those liability amounts are $44,000 and $15,000, a newly reported liability claim is $6,000, and property damage is $9,000. For accident year 2026 at 12 months, a new bodily-injury code is $30,000, a legacy liability code is $11,000, and property damage is $10,000. What are the three cell values under the consistent definition? (Answer: $52,000, $65,000, and $41,000, respectively.)

What You'll Learn