Why a Loss Triangle Has Its Shape
A loss triangle is a compact way to organize how claim information for different groups of business becomes available over time. Its shape is not just a formatting choice: at a given valuation cutoff, older groups have had more time to develop than newer groups. Reading the layout correctly is a prerequisite for interpreting the loss measurements it contains.
In this tutorial, a paid-loss triangle means an arrangement of paid-loss measurements by origin period and development age. The exact meaning of the measurement in each cell must be specified with the data; the triangle’s layout alone does not say whether a value represents payments during an interval or payments accumulated through an age.
For example, a cell in the third row and second column refers to the third origin period shown and its second listed development age. The row and column together identify the observation. They do not, by themselves, identify a particular calendar date; that depends on how the origin periods, ages, and valuation cutoff are defined.
A Basic Triangle and Its Boundary
Suppose origin periods are listed from oldest to newest, and development ages are listed from earliest to latest. At a common valuation cutoff, a measurement is available only if enough time has elapsed for that origin period to reach the corresponding age. The oldest origin period will generally have the most observed ages; each later origin period will have fewer.
The following hypothetical table gives paid-loss measurements in thousands of dollars. A dash marks a cell that is not available at the displayed cutoff. It does not mean that paid losses are zero.
| Origin period | Age 1 | Age 2 | Age 3 | Age 4 |
|---|---|---|---|---|
| O1 (oldest) | 410 | 690 | 820 | 900 |
| O2 | 380 | 640 | 780 | — |
| O3 | 450 | 720 | — | — |
| O4 (newest) | 390 | — | — | — |
The observed region is triangular: it begins at the upper left and steps inward as the origin periods become newer. A rectangular table may be used to display the data, but its unobserved cells do not become observations merely because the table has space for them.
Worked Example: Reading Rows, Columns, and Missing Cells
Use the table above. Identify the measurement for O3 at Age 2, determine whether O3 at Age 3 is observed, and count the observed and unavailable cells.
Step 1: Read the coordinate. O3 is the third origin-period row, and Age 2 is the second development-age column. Their intersection contains 720, so the measurement is $720,000, because the table is in thousands of dollars.
Step 2: Check the next age. The cell in row O3 and column Age 3 contains a dash. It is unavailable at this cutoff; the table does not report a zero for that cell.
Step 3: Count the observed cells by row. O1 has 4, O2 has 3, O3 has 2, and O4 has 1. Thus the observed count is \(4+3+2+1=10\).
Step 4: Count the unavailable cells. The rectangular display has \(4\times4=16\) possible positions. Therefore \(16-10=6\) positions are unavailable.
Method and mistake avoided. Read the row and column first, then distinguish a number from a dash. Treating each dash as zero would incorrectly assert that a measurement was observed and equaled zero.
Why the Observed Cells Form a Triangle
The ordinary triangle boundary follows from a simple availability condition. To express it precisely, suppose there are \(n\) equally spaced origin periods, listed oldest to newest, and \(n\) ordered development ages. Assume the periods and ages use a consistent cadence and the data are assembled at one common valuation cutoff. Under this aligned layout, the oldest row has \(n\) observed ages, the next has \(n-1\), and so on.
Number rows \(i=1,\ldots,n\), with \(i=1\) oldest, and columns \(j=1,\ldots,n\), with \(j=1\) earliest age. The observed-cell condition is:
This rule describes the usual triangular boundary under the stated alignment assumptions. It is not a universal substitute for checking how a particular data set was constructed. If periods have unequal spacing, a different cutoff, or incomplete reporting, the available cells may not follow this simple pattern.
Why the boundary follows
Under the aligned setup, the oldest origin period has had time to reach all \(n\) listed ages by the cutoff. Each successive origin period begins one period later, so it has one fewer age available. Consequently, row \(i\) has \(n-i+1\) observed cells. The observed column indices are \(1\) through \(n-i+1\), which is exactly the condition \(j\leq n-i+1\).
For instance, with \(n=4\), row 1 has \(4-1+1=4\) observed ages, row 2 has \(4-2+1=3\), row 3 has \(4-3+1=2\), and row 4 has \(4-4+1=1\). Adding the row counts gives \(4+3+2+1=10\) observed cells. More generally, the total is \(n+(n-1)+\cdots+1=n(n+1)/2\); the remaining \(n^2-n(n+1)/2=n(n-1)/2\) positions are outside the observed triangle.
Worked Example: Checking the Boundary in a Five-Period Triangle
A hypothetical data set has five aligned origin periods and five development ages, ordered as above. Determine the expected number of observed cells, the number unavailable in the rectangular display, and whether row 3, Age 3 is within the observed region.
Step 1: State the method and assumption. Use the common-cutoff boundary rule \(j\leq n-i+1\), assuming equal spacing and aligned origin and age periods.
Step 2: Count observed cells row by row. With \(n=5\), the row counts are \(5,4,3,2,1\). Their sum is \(5+4+3+2+1=15\).
Step 3: Count unavailable positions. The rectangular display has \(5\times5=25\) positions. Thus \(25-15=10\) positions are unavailable.
Step 4: Check the requested coordinate. For row 3 and Age 3, \(i=3\), \(j=3\), and \(n=5\). The maximum observed age index in row 3 is \(5-3+1=3\). Since \(3\leq3\), row 3, Age 3 is within the expected observed region.
Check. The row counts total 15, and the unavailable count is 10, so \(15+10=25\), the full number of displayed positions.
Method and mistake avoided. Apply the boundary using the row’s position, not just the age label. A common error is to assume every row has the same number of observed ages because every row appears in the same rectangular display.
Unavailable Cells Are Not Zeroes
A zero is a recorded measurement: it says that the relevant paid-loss measure equals zero under the data convention. An unavailable cell says that the measurement is not present in the triangle at that cutoff. Those statements are different and should not be substituted for each other in a table, a calculation, or an explanation.
The distinction matters even before any development method is applied. Replacing a dash with zero changes the recorded data, can distort averages or totals, and falsely implies that the later-age experience is known. Conversely, a blank inside what should be the observed region deserves investigation. It may signal a recording problem, a reporting limitation, or a layout that does not satisfy the simple aligned-triangle assumptions.
Worked Example: Diagnosing an Interior Gap
A four-row table is intended to follow the common-cutoff pattern. Its row entries, in order from Age 1 to Age 4, are: R1 has 100, 170, 230, 260; R2 has 95, 160, 215, then a dash; R3 has 110, 185, then two dashes; R4 has 105, a dash, 140, then a dash. All numerical values are hypothetical and in thousands of dollars. Decide whether R4’s pattern fits the ordinary triangle boundary.
Step 1: Find the expected observed ages for R4. R4 is row \(i=4\) in a four-row table, so the maximum observed age index under the aligned rule is \(n-i+1=4-4+1=1\).
Step 2: Compare the actual entries with the rule. Age 1 has a measurement, which is consistent with the boundary. Age 2 is unavailable, and Age 3 has a measurement. But Age 3 does not satisfy \(j\leq1\), since \(3\leq1\) is false.
Step 3: State the conclusion. R4’s Age 3 value is outside the expected observed region, so the row does not fit the ordinary common-cutoff triangle. The layout or data should be checked before interpreting that value.
Method and mistake avoided. Check the entire observed pattern against the boundary. It is not enough to note that a newer row has at least one blank. Also, an interior blank followed by a later observed age cannot be explained by the ordinary rule that later ages become unavailable after the cutoff.
What a Triangle Tells You—and What It Does Not
A basic paid-loss triangle tells the reader how observations are organized: which origin period each row represents, which development age each column represents, and where measurements are available at the displayed cutoff. It also makes the uneven maturity of the origin periods visible. Older rows contain more observed ages, while newer rows are more incomplete.
The shape alone does not explain why losses differ between rows, determine the appropriate origin-period definition, or establish what numerical convention is used in each cell. Those details must be specified separately. In particular, do not infer a claim outcome from a missing cell, and do not infer the cell’s payment convention from its location.
Correctly identifying rows, columns, and the observed boundary is the foundation for later work with loss development. This layout is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Check Your Understanding
For each problem, use the stated layout assumptions and distinguish unavailable positions from recorded zeroes.
- A three-row, three-age table follows the aligned common-cutoff pattern. How many cells are observed, and how many positions are unavailable? (Answer: 6 observed and 3 unavailable.)
- In a six-period aligned triangle, determine the observed-cell count and unavailable-cell count in the \(6\times6\) display. (Answer: 21 observed and 15 unavailable.)
- A four-row triangle is ordered oldest to newest. What is the latest expected observed age in row 2 under the common-cutoff rule? Is row 2, Age 4 observed? (Answer: Latest expected age is Age 3; Age 4 is unavailable.)
- A five-row aligned triangle has a recorded value in row 4, Age 2. Is that coordinate inside the expected observed region? Show the boundary comparison. (Answer: Yes; \(5-4+1=2\), and \(2\leq2\).)
- In a four-row aligned triangle, row 4 has a measurement at Age 1, a blank at Age 2, and a measurement at Age 3. Is that pattern consistent with the ordinary triangular boundary? (Answer: No; row 4’s maximum observed age is \(4-4+1=1\), so Age 3 lies outside the expected observed region.)