Representing Paired Discrete Losses
An insurer may face two losses during the same period, such as a property loss and a related business-interruption loss. When each loss can take only a finite or countable set of values, a joint probability table records the probability of each possible pair. The table preserves information about which values can occur together; a list of possible values for each loss alone is not enough to answer every question about the pair.
Let \(X\) and \(Y\) denote two discrete losses. Suppose their possible values are \(x_1,x_2,\ldots\) and \(y_1,y_2,\ldots\). The pair \((X,Y)=(x_i,y_j)\) is one outcome of the joint model. The probability assigned to that outcome is an atom of the joint law.
The probability of an event involving both losses is found by identifying every table cell whose pair satisfies the event, then adding the probabilities in those cells. A single cell gives a probability for one exact pair. A region, such as \(X+Y\leq c\), usually contains several pairs.
Checking a Joint Probability Table
A table is a probability model only if every entry is nonnegative and all entries together sum to one. Row and column subtotals can help find arithmetic errors, but the total of all cells is the condition that matters. A subtotal below one is not automatically invalid: the remaining mass may be in other rows or columns. Conversely, a table with a negative cell is invalid even if its entries sum to one.
When a table includes an unknown cell, the total-probability condition can determine it, provided the listed cells are all possible outcomes. The resulting value must still be nonnegative. Once the table is valid, probabilities for pairs or regions can be calculated directly from its cells.
Worked Example: Summing Probability Over a Loss Region
A hypothetical insurer models a property loss \(X\) and a related interruption loss \(Y\), both in thousands of dollars. The complete joint probability table is below. Find \(P(2<X+Y\leq 9)\).
| \(X\backslash Y\) | 0 | 4 | 7 |
|---|---|---|---|
| 0 | 0.1000 | 0.0800 | 0.1200 |
| 3 | 0.1500 | 0.2000 | 0.0500 |
| 8 | 0.1000 | 0.0500 | 0.1500 |
Method. First verify the table, then check each possible pair against both bounds. The row totals are \(0.3000\), \(0.4000\), and \(0.3000\), so the total probability is \(0.3000+0.4000+0.3000=1.0000\). All entries are nonnegative.
For \(X=0\), the sums with \(Y=0,4,7\) are \(0,4,7\). The qualifying pairs are \((0,4)\) and \((0,7)\), contributing \(0.0800+0.1200\).
For \(X=3\), the sums are \(3,7,10\). The qualifying pairs are \((3,0)\) and \((3,4)\), contributing \(0.1500+0.2000\). For \(X=8\), the sums are \(8,12,15\); only \((8,0)\) qualifies, contributing \(0.1000\).
Therefore, \(P(2<X+Y\leq 9)=0.0800+0.1200+0.1500+0.2000+0.1000=0.6500\). The strict lower bound excludes sums of 2 or less, and the inclusive upper bound allows a sum of exactly 9 (there is no such pair in this table). The result is between zero and one, as required.
Why Region Probabilities Are Sums
A discrete joint model breaks the outcome space into individual pairs. Distinct pairs cannot occur simultaneously: for example, \((X,Y)=(3,0)\) and \((X,Y)=(3,4)\) are mutually exclusive outcomes. An event described by a region is the union of the individual pair outcomes inside that region. Because those outcomes are disjoint, the probability of the union is the sum of their probabilities.
Here \(A\) is any specified set of pairs and \(\mathbf{1}\{\cdot\}\) equals 1 when its condition is true and 0 otherwise. The indicator notation expresses the practical procedure: include an atom if its pair satisfies the event, and omit it otherwise. For a finite table, this is simply a finite sum.
This also explains why the boundary symbols matter. If the event contains \(X+Y<c\), pairs with sum exactly \(c\) are omitted. If it contains \(X+Y\leq c\), those pairs are included. With a discrete model, probability can be concentrated exactly on a boundary, so changing a strict inequality to an inclusive one can change the answer.
Worked Example: Applying Strict and Inclusive Boundaries
In a hypothetical portfolio, \(X\) and \(Y\) are losses in thousands of dollars with this complete joint law. Find \(P(X>0,\ 4\leq X+Y<8)\).
| \(X\backslash Y\) | 0 | 3 |
|---|---|---|
| 0 | 0.1200 | 0.1800 |
| 2 | 0.2000 | 0.1500 |
| 5 | 0.1000 | 0.2500 |
Method. Check validity, then apply every condition to each pair. The row totals are \(0.3000\), \(0.3500\), and \(0.3500\), summing to \(1.0000\); all cells are nonnegative.
The condition \(X>0\) removes both pairs in the \(X=0\) row. In the \(X=2\) row, the sums are \(2+0=2\) and \(2+3=5\); only \((2,3)\) satisfies \(4\leq X+Y<8\), contributing \(0.1500\). In the \(X=5\) row, the sums are \(5+0=5\) and \(5+3=8\); only \((5,0)\) qualifies, contributing \(0.1000\).
Thus the probability is \(0.1500+0.1000=0.2500\). The pair \((5,3)\) is excluded because its sum is 8, which fails the strict upper bound.
Using an Unknown Cell
An unknown entry can be inferred from the probability constraint only when the table is intended to describe the complete joint distribution. After solving for it, check that its value is nonnegative and verify the total. Then use the completed table in the same way as any other joint law.
Worked Example: Completing and Using a Joint Table
A hypothetical insurer records two discrete losses, \(X\) and \(Y\), in thousands of dollars. Their only possible pairs are represented by the table below. The cell marked \(c\) is unknown. Find \(c\), then calculate \(P(X+Y\geq 8)\).
| \(X\backslash Y\) | 0 | 2 | 6 |
|---|---|---|---|
| 1 | 0.0800 | 0.1200 | \(c\) |
| 4 | 0.2000 | 0.1500 | 0.2700 |
Method. Add the known entries: \(0.0800+0.1200+0.2000+0.1500+0.2700=0.8200\). Since the table lists the complete support, the missing cell must make the total equal to one: \(c=1.0000-0.8200=0.1800\).
This value is nonnegative, and the completed table sums to \(0.8200+0.1800=1.0000\), so it satisfies both probability constraints.
Now check the pairs whose sums are at least 8. In the \(X=1\) row the sums are 1, 3, and 7, so none qualifies. In the \(X=4\) row the sums are 4, 6, and 10; only \((4,6)\) qualifies, with probability \(0.2700\). Therefore \(P(X+Y\geq 8)=0.2700\). The completed cell \((1,6)\) has sum 7 and is not included.
Common Mistakes and Practical Checks
The main calculation risk is not usually the addition; it is selecting the wrong cells. A disciplined table scan helps: list the candidate pairs, compute the quantity in the event for each pair, apply all conditions, and only then add the selected probabilities.
- Including a boundary that is excluded. For \(X+Y<8\), a pair summing to 8 does not qualify. Do not treat \(<\) as \(\leq\).
- Forgetting another condition. In \(X>0\) together with a sum restriction, a pair can satisfy the sum condition and still fail \(X>0\).
- Adding row totals or column totals instead of selected cells. A region may include only part of a row or column. Use the individual joint probabilities for the included pairs.
- Assuming an incomplete table is valid. A total below one means probability is missing unless an unknown cell or unlisted outcomes have been specified.
- Using a table entry as a region probability. One entry is the probability of one exact pair; an event containing several pairs requires summing all their atom probabilities.
A useful final check is to confirm that the answer lies between 0 and 1, that no selected pair violates any condition, and that every pair satisfying the event has been included. These checks are especially valuable when the event uses both a loss threshold and a separate restriction on one loss.
Check Your Understanding
For each problem, treat the displayed table as the complete joint law unless an unknown cell is shown. Check the event conditions pair by pair before adding probabilities.
- A hypothetical pair of losses has the following joint probabilities:
Find \(P(X+Y\leq 4)\). (Answer: 0.5000)\(X\backslash Y\) 1 4 0 0.1000 0.1500 2 0.2500 0.2000 5 0.1000 0.2000 - For a complete law with \(X\in\{1,3,7\}\), \(Y\in\{0,2\}\), the rows of probabilities for \(Y=0,2\) are respectively \((0.0500,0.1500)\), \((0.2000,0.1000)\), and \((0.2500,0.2500)\). Find \(P(X\geq3,\ X+Y<8)\). (Answer: 0.5500)
- A two-row, two-column joint table has \(X\in\{0,4\}\), \(Y\in\{1,5\}\), and entries \((0.1200,c)\) for \(X=0\) and \((0.1800,0.2500)\) for \(X=4\), in the stated order of \(Y\). Find \(c\), then \(P(X+Y\geq9)\). (Answer: \(c=0.4500\); probability \(0.2500\))
- For \(X\in\{0,3,6\}\) and \(Y\in\{1,4\}\), the rows of a complete joint table are \((0.0800,0.1200)\), \((0.2000,0.1000)\), and \((0.2500,0.2500)\). Find \(P(4\leq X+Y<10,\ X\geq3)\). (Answer: 0.5500)
- A complete joint table has \(X\in\{2,8\}\), \(Y\in\{0,3,6\}\), with rows \((0.1000,0.1500,0.2000)\) and \((0.0500,0.2500,0.2500)\). Find \(P(X+Y>10\text{ or }(X=2\text{ and }Y=6))\). (Answer: 0.7000)