Introduction
A paid-loss triangle organizes payments by two dates: the period in which a claim originates and the period in which a payment is made. In an incremental paid-loss triangle, each cell contains the payments made during one development interval for claims from one origin period. This differs from recording a running total through a development age: the cell is an amount for an interval, not an amount accumulated to date.
The key construction decision is the interval convention. In this tutorial, development interval 1 is the origin period. That convention determines how payments are indexed and how the valuation date limits the observable cells. The definitions and methods here are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Indexing the Cells
Suppose origin periods and development intervals are annual, and the data are valued at the end of calendar period \(n\). Number the origin periods \(1,2,\ldots,n\) in chronological order, and use \(i\) for the origin-period index and \(j\) for the development-interval index. Under the stated convention, the calendar period associated with cell \((i,j)\) is
The first interval is the origin period itself: when \(j=1\), \(t=i+1-1=i\). The next interval is the following period: when \(j=2\), \(t=i+1\). Therefore a cell is observable by the end of period \(n\) only when
This condition describes the triangle's boundary. A cell beyond that boundary is not yet observable at the valuation date; it is not a zero. A cell within the boundary can be an observed zero if no payments occurred in that interval.
Why the Triangle Has Its Shape
The indexing and boundary rule follow directly from counting periods. The origin period for row \(i\) is calendar period \(i\). Development interval 1 is that origin period, so interval \(j\) is \(j-1\) periods later. Adding those \(j-1\) periods to \(i\) gives the calendar-period index \(t=i+j-1\). Since the valuation period is \(n\), a payment interval can be observed only if its index does not exceed \(n\):
Solving for the largest observable development interval in row \(i\) gives \(j\le n-i+1\). Thus earlier origin periods have more observable development intervals, while later origin periods have fewer. This is a consequence of the common valuation date, not evidence that later-origin claims have fewer payments in total.
Worked Example: Constructing Cells From Payment Records
A hypothetical insurer records the following payments, in dollars, for claims from four annual origin periods. The data are valued at the end of 2025. Assign each payment to an incremental paid-loss cell, aggregate transactions that fall in the same cell, and verify the total.
- Origin 2022: payments of $12,000 and $4,000 in 2022; $5,000 in 2023; $2,000 and $3,000 in 2024; and $1,000 in 2025.
- Origin 2023: payments of $14,000 and $6,000 in 2023; $8,000 in 2024; and $4,000 in 2025.
- Origin 2024: payments of $17,000 in 2024 and $6,000 in 2025.
- Origin 2025: a payment of $9,000 in 2025.
Step 1: Assign development intervals. For origin 2022, calendar 2022 is interval 1, 2023 is interval 2, 2024 is interval 3, and 2025 is interval 4. For origin 2023, the corresponding intervals are 2023, 2024, and 2025. Apply the same rule to each origin period.
Step 2: Aggregate payments within each cell. For origin 2022, interval 1 is $12,000 + $4,000 = $16,000; interval 2 is $5,000; interval 3 is $2,000 + $3,000 = $5,000; and interval 4 is $1,000. For origin 2023, the first interval is $14,000 + $6,000 = $20,000, followed by $8,000 and $4,000. For origin 2024, the first two intervals contain $17,000 and $6,000. Origin 2025 has $9,000 in interval 1.
Step 3: Display the observed triangle. A dash marks a cell not yet observable at the end of 2025; it does not represent a zero payment.
| Origin period | Interval 1 | Interval 2 | Interval 3 | Interval 4 |
|---|---|---|---|---|
| 2022 | $16,000 | $5,000 | $5,000 | $1,000 |
| 2023 | $20,000 | $8,000 | $4,000 | — |
| 2024 | $17,000 | $6,000 | — | — |
| 2025 | $9,000 | — | — | — |
Step 4: Check the totals. The four origin-period totals represented by the observed cells are $16,000 + $5,000 + $5,000 + $1,000 = $27,000; $20,000 + $8,000 + $4,000 = $32,000; $17,000 + $6,000 = $23,000; and $9,000. Their sum is $27,000 + $32,000 + $23,000 + $9,000 = $91,000. Directly adding all listed transactions also gives $91,000, so the aggregation reconciles.
Method and mistake avoided. Each transaction was classified by its origin and payment period before aggregation. The common error is to put the 2022-origin payment made in 2023 into interval 1 because it is the first payment after origin. It belongs in interval 2: interval 1 is 2022 itself.
Reading Incremental Amounts
An entry answers a specific question: how much was paid during this development interval on claims assigned to this origin period? A row therefore describes payments over successive intervals, with each observed cell contributing only its own interval's amount. Adding the observed cells in a row gives the amount paid through the valuation date for that origin period. It does not, by itself, give an estimate of payments that may occur after the valuation date.
Likewise, a zero in an observed cell means no payment was recorded for that origin and interval under the data definition. It is different from a dash or blank beyond the triangle boundary. An incremental cell may be zero even when other intervals in the same row contain payments.
Worked Example: Interpreting a Zero and a Row Total
A hypothetical liability portfolio has the following incremental paid losses, stated in thousands of dollars, as of the end of 2024. For origin 2021, the observed cells are $30, $0, $7, and $3 in intervals 1 through 4. For origin 2022, the observed cells are $28, $5, and $2 in intervals 1 through 3. For origin 2023, the observed cells are $32 and $6 in intervals 1 and 2. Origin 2024 has $35 in interval 1.
Step 1: Check the boundary. There are four origin periods, 2021 through 2024, so \(n=4\). For origin 2021, \(i=1\), and the maximum observable interval is \(4-1+1=4\). All four listed cells are observable. For origin 2022, the maximum is \(4-2+1=3\); for origin 2023, it is 2; and for origin 2024, it is 1. The stated row lengths are therefore consistent with the valuation date.
Step 2: Interpret the zero. The $0 in origin 2021, interval 2 is an observed zero: no payment was recorded in that interval. The $7 in interval 3 remains a separate payment amount. The zero does not mean that the claim group had no payments at other development intervals.
Step 3: Calculate the observed row sum. For origin 2021, the total represented through the valuation date is $30 + $0 + $7 + $3 = $40 thousand. For origin 2022, it is $28 + $5 + $2 = $35 thousand. These are paid amounts through the valuation date, not statements about the ultimate amount.
Check. The origin 2021 sum is $40 thousand, and it includes all four observable intervals. Treating its zero as an unavailable cell would incorrectly omit a valid observation; treating a future, unobservable interval as zero would make the opposite error.
Valuation Boundaries and Consistent Conventions
The triangle's apparent blank upper-right region is created by the valuation date. For \(n\) periods of data, row \(i\) has \(n-i+1\) observable cells under the indexing convention used here. The number of observed cells is consequently \(n+(n-1)+\cdots+1=n(n+1)/2\), assuming a complete triangular data set with no additional missing records. This count is a useful structural check, not a claim that every cell must contain a positive payment.
The mapping also works when periods are not annual, provided origin periods and development intervals use a consistent, defined time unit. For quarterly data, for example, interval 1 is the origin quarter, interval 2 is the next quarter, and each increase in \(j\) advances one quarter. Do not mix quarterly development intervals with annual index increments.
Incremental values are often recorded as gross paid amounts in a cell. If the data definition includes recoveries, reversals, or accounting corrections as net adjustments, an incremental cell could be negative. The triangle's indexing still identifies the relevant origin and interval; the analyst must understand the payment basis before interpreting the amount.
Determine which origin-period row the claim or payment belongs to, using the stated data definition.
Use the period in which the payment was made, not the period in which it was reported or later reviewed.
With interval 1 equal to the origin period, the interval index is the payment-period index minus the origin-period index plus 1.
Confirm that the cell is not beyond the valuation boundary, then sum all qualifying payments in that origin-interval combination.
Worked Example: Testing a Cell Against the Valuation Date
A hypothetical insurer has five annual periods of data, 2020 through 2024, valued at the end of 2024. Determine the calendar period and observability of (a) origin 2022, interval 3; (b) origin 2023, interval 3; and (c) origin 2024, interval 1. Also explain the status of a reported zero for origin 2022, interval 2.
Step 1: Number the periods. Assign index 1 to 2020, 2 to 2021, 3 to 2022, 4 to 2023, and 5 to 2024. Thus \(n=5\).
Step 2: Map each cell. Origin 2022 has \(i=3\). For interval 3, \(t=3+3-1=5\), which is calendar 2024. Since \(5\le5\), this cell is observable. Origin 2023 has \(i=4\); its interval 3 maps to \(t=4+3-1=6\), or 2025. Since \(6>5\), it is not observable at the valuation date. Origin 2024 has \(i=5\); interval 1 maps to \(t=5+1-1=5\), or 2024, and is observable.
Step 3: Classify the reported zero. Origin 2022, interval 2 maps to \(t=3+2-1=4\), or 2023. This is within the valuation boundary. If the data explicitly report zero for that cell, it is an observed zero, not an unavailable cell.
Check. The row for origin 2022 can have three observed intervals through 2024. The row for origin 2023 can have only two. This agrees with \(n-i+1\): \(5-3+1=3\) and \(5-4+1=2\).
Common Indexing Trap
A tempting but incorrect rule is to calculate the development interval as the payment-period index minus the origin-period index, without adding 1. For origin period 2023, a payment made in 2023 would then be assigned to interval \(i-i=0\), even though the convention defines the origin period as interval 1. A payment in 2025 would be incorrectly called interval 2 instead of interval 3.
For example, if 2023 is index \(i=2\) and 2025 is \(t=4\), the correct interval is \(j=4-2+1=3\). The cell maps back to 2025 because \(i+j-1=2+3-1=4\). The mistaken rule would give interval 2, mapping to 2024 under this convention. Always check both directions: the interval formula should assign the cell, and \(t=i+j-1\) should recover its payment period.
Another trap is to treat each incremental cell as a cumulative amount through the end of that interval. That misreads what the triangle records. A cell contains only payments in its interval; any cumulative view requires a separate calculation and careful handling of the same origin-period and valuation conventions.
Check Your Understanding
For each question, use the convention that development interval 1 is the origin period. Amounts are hypothetical.
- A portfolio valued at the end of 2025 records two payments for origin 2023: $11,400 in 2023 and $2,600 in 2023, then $4,500 in 2024 and $1,500 in 2025. Give the incremental amounts in intervals 1, 2, and 3 and their observed sum. (Answer: $14,000, $4,500, and $1,500; observed sum $20,000.)
- There are six annual periods of data. If origin period \(i=4\) has a payment in interval \(j=2\), which numbered calendar period is it, and is it observable? (Answer: Period 5; yes.)
- Five periods are available through period 5. For origin \(i=4\), determine whether interval 2 and interval 3 are observable. (Answer: Interval 2 is observable; interval 3 is not observable.)
- An observed row contains incremental amounts of $8,000, $3,000, $0, and $2,000 through the valuation date. What is the amount paid through that date, and how should the zero be classified? (Answer: $13,000; it is an observed zero.)
- Calendar periods 2020 through 2024 are numbered 1 through 5. A payment for origin 2022, index \(i=3\), is made in 2024, index \(t=5\). Find its development interval and check whether it is within the valuation boundary. (Answer: Interval 3; it is observable because \(3+3-1=5\).)