Why the Triangle Layout Depends on the Time Interval
A loss development triangle organizes observations by two dates: the period in which the underlying business originates and the valuation date at which its experience is observed. The origin period may be a year or a quarter. The interval chosen for origin periods determines both the row labels and the steps between development-age columns. A quarterly layout therefore has more closely spaced ages than an annual layout; it does not simply divide annual values into smaller pieces.
This tutorial focuses on the grid: how to label its rows and columns, which cells can be observed by a stated valuation date, and how to count those cells. The amount recorded in an observed cell depends on the measure being tabulated, a distinction treated separately from the layout itself. No interest, mortality, or probability distribution is assumed in the date-indexing examples below. The layout conventions are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Annual and Quarterly Age Columns
Use period-end valuation dates and measure age from the beginning of the origin period. Under this convention, a full year of development for an annual origin period is age 12 months. For a quarterly origin period, one full quarter is age 3 months. Thus the annual columns are typically 12, 24, 36 months, and so on, while quarterly columns are typically 3, 6, 9 months, and so on.
For example, the annual origin period calendar year 2022 begins on January 1, 2022. Its 12-month valuation date is December 31, 2022, and its 24-month valuation date is December 31, 2023. For origin period Q1 2022, the 3-month date is March 31, 2022, the 6-month date is June 30, 2022, and the 12-month date is December 31, 2022. The dates correspond to elapsed time from the start of each origin period, not to a count of how many row labels appear before the current row.
A triangle is often displayed with origin periods down the rows and increasing development ages across the columns. When the origin and valuation periods use the same interval and the data are valued at a period end, the latest origin is at the first development age, the preceding origin has two possible ages, and so on. This familiar triangular shape is a consequence of the date alignment; it is not a separate rule about the values in the cells.
Worked Example: Building an Annual Grid Through Year-End
A hypothetical insurer has annual origin periods 2020 through 2023 and a valuation date of December 31, 2023. Construct the grid’s available development ages. Use \(Z_{i,j}\) to denote the generic observation for origin period \(i\) at age \(j\); the symbol does not specify a particular loss measure.
Step 1: Set the age columns. Annual periods advance in 12-month intervals, so the possible columns are 12, 24, 36, and 48 months. For example, age 12 for 2020 is December 31, 2020; age 48 for 2020 is December 31, 2023.
Step 2: Test each row against the valuation date. The 2020 origin has four dates through year-end 2023. The 2021 origin has three, the 2022 origin has two, and the 2023 origin has one.
| Origin period | 12 months | 24 months | 36 months | 48 months |
|---|---|---|---|---|
| 2020 | \(Z_{2020,12}\) | \(Z_{2020,24}\) | \(Z_{2020,36}\) | \(Z_{2020,48}\) |
| 2021 | \(Z_{2021,12}\) | \(Z_{2021,24}\) | \(Z_{2021,36}\) | — |
| 2022 | \(Z_{2022,12}\) | \(Z_{2022,24}\) | — | — |
| 2023 | \(Z_{2023,12}\) | — | — | — |
Step 3: Count and check. The observed-cell count is \(4+3+2+1=10\). The most recent origin, 2023, has only its 12-month cell because its 24-month valuation date would be December 31, 2024, after the stated valuation date. The dashes mark unavailable cells, not zero observations.
Method and trap avoided. This is an annual period-end layout. A common mistake is to label the 2020 row’s last cell “2023” rather than age 48 months. The row label identifies the origin year; the column identifies elapsed development age. Keeping those roles distinct prevents a cell from being assigned to the wrong date.
Constructing a Quarterly Triangle
For a quarterly layout, each row represents a quarter, and each successive development-age column advances by three months. The period labels should make the year and quarter unambiguous, such as Q1 2023 or 2023 Q1. If the valuation date is a quarter-end, the row for the latest completed origin period has one observed cell, at age 3 months. A quarter that has not ended by the valuation date is not a completed origin period under this convention.
The quarterly table can have more columns than an annual table covering the same number of years. That is because its columns distinguish more valuation points. A 12-month age appears as one annual step in an annual layout, but as the fourth quarterly development age in a quarterly layout.
Worked Example: A Quarterly Grid at September 30
A hypothetical portfolio has quarterly origins Q1 2023 through Q3 2023 and is valued at September 30, 2023. Construct the observable portion of a quarterly grid, showing ages in months.
Step 1: List the quarterly ages. The first three possible ages are 3, 6, and 9 months. For Q1 2023, these dates are March 31, June 30, and September 30, respectively.
Step 2: Determine the available ages by row. Q1 has three observed valuation dates by September 30. Q2 has two, at June 30 and September 30. Q3 has one, at September 30.
| Origin period | 3 months | 6 months | 9 months |
|---|---|---|---|
| Q1 2023 | \(Z_{Q1\,2023,3}\) | \(Z_{Q1\,2023,6}\) | \(Z_{Q1\,2023,9}\) |
| Q2 2023 | \(Z_{Q2\,2023,3}\) | \(Z_{Q2\,2023,6}\) | — |
| Q3 2023 | \(Z_{Q3\,2023,3}\) | — | — |
Step 3: Check the dates and the count. For Q2 2023, age 3 is June 30 and age 6 is September 30, both no later than the valuation date. Age 9 would be December 31, so that cell is unavailable. The total is \(3+2+1=6\) observable cells.
Method and trap avoided. The quarterly layout uses a three-month age step. Calling the Q1-to-September observation “three development periods” would count intervals, but the development age in the table is 9 months. Conversely, calling it 3 months confuses the number of elapsed quarters with the length of one quarter.
Why the Cell Count Forms a Triangle
The triangular count can be derived directly from the dates. Suppose there are \(N\) consecutive, completed origin periods, all of the same length, and the final origin period is valued at its first period-end age. Number the rows \(i=1,\ldots,N\), starting with the oldest origin. The oldest row has \(N\) available cells, the next has \(N-1\), and the newest has one. Therefore the number of available cells is:
The sum identity follows by pairing terms from opposite ends: \((N+1)\) appears in each pair \(1+N\), \(2+(N-1)\), and so forth. There are \(N/2\) such pairs when \(N\) is even, giving \(N(N+1)/2\); for odd \(N\), the middle term completes the same expression. For four aligned origins, for instance, the count is \(4(5)/2=10\), as in the annual example.
This count assumes a full aligned triangle: consecutive origin periods, period-end valuation, and one initial observed age for the latest origin. It does not apply unchanged if the data begin or end partway through a period, if origins are missing, or if the valuation date does not coincide with the relevant period end. In those cases, determine each cell’s valuation date and test it individually rather than forcing the grid into the triangular count.
Worked Example: Counting Cells in a Quarterly Grid
A hypothetical data extract contains origins Q1 2023 through Q2 2024 and is valued at June 30, 2024. The analyst needs the quarterly grid through the maximum observed age. Determine the age columns, the observable cells in each row, and the total cell count.
Step 1: Find the maximum age. Q1 2023 begins January 1, 2023. From that date to June 30, 2024 is 18 months, so the age columns are 3, 6, 9, 12, 15, and 18 months.
Step 2: Count each row’s eligible ages. The six origins are Q1 2023, Q2 2023, Q3 2023, Q4 2023, Q1 2024, and Q2 2024. Their maximum ages at the valuation date are 18, 15, 12, 9, 6, and 3 months, respectively.
| Origin period | Maximum age at June 30, 2024 | Number of observable cells |
|---|---|---|
| Q1 2023 | 18 months | 6 |
| Q2 2023 | 15 months | 5 |
| Q3 2023 | 12 months | 4 |
| Q4 2023 | 9 months | 3 |
| Q1 2024 | 6 months | 2 |
| Q2 2024 | 3 months | 1 |
Step 3: Sum and verify. The total is \(6+5+4+3+2+1=21\). The same result follows from \(N(N+1)/2=6(7)/2=21\). The earliest origin’s 18-month age is six quarterly intervals, not six years or six months. No interest, mortality, or probability assumptions enter this date-based calculation.
Method and trap avoided. Counting the ages row by row is a reliable check on the formula. The common error is to use the number of years in the experience window as the number of quarterly columns. The age steps are quarters, so this grid has six development-age columns, even though the origins span only a year and a half.
Reading the Layout Carefully
A grid’s row interval and column interval answer different questions. The row label tells which business period an observation belongs to. The column tells how much time has elapsed from that period’s start at the cell’s valuation date. A cell at age 12 months can therefore occur in an annual layout and a quarterly layout, but those layouts have different intermediate ages and may have different origins.
Always state the valuation convention before constructing a grid. Here, ages are measured from the start of the origin period, and dates fall at period ends. If a problem instead specifies a different origin or age convention, use its definitions consistently. Do not infer the convention from a row label alone.
- Keep the units visible. Label quarterly ages as months or quarters; do not silently switch between the two.
- Separate origin from age. An origin year or quarter is a row identity, not a development age.
- Respect the valuation date. A cell whose valuation date is later is unavailable in the stated triangle.
- Do not fill unavailable cells with zero. A dash indicates no observation at that date; zero would be an observed numerical value.
- Use the triangle-count shortcut only when its assumptions hold. Missing periods and nonaligned valuation dates require cell-by-cell date checks.
Check Your Understanding
For each problem, use period-end valuation dates and measure development age from the start of the origin period. Treat unavailable cells as blank, not zero.
- A hypothetical annual grid contains origins 2019 through 2022 and is valued at December 31, 2022. How many cells are observable, and which ages are available for origin 2020? (Answer: 10 cells; ages 12, 24, and 36 months for 2020.)
- A quarterly grid contains origins Q2 2022 through Q1 2023 and is valued at March 31, 2023. What is the maximum age for Q2 2022, and how many total cells are observable? (Answer: 12 months; 10 cells.)
- A complete aligned annual triangle has seven origins and is valued at the latest origin’s first year-end. How many cells are observable? (Answer: 28 cells.)
- At December 31, 2024, what are the ages of origins Q1 2024 and Q3 2024 in a quarterly layout? How many age columns are needed to include the oldest of those origins? (Answer: 12 and 6 months; 4 columns.)
- A quarterly triangle has origins Q1 2023 through Q2 2024 and is valued at June 30, 2024. How many cells are at ages 12 months or older? (Answer: 6 cells: 3 for Q1 2023, 2 for Q2 2023, and 1 for Q3 2023.)