Why Independence Matters
When two losses are modeled together, their individual distributions do not by themselves determine how often particular pairs of outcomes occur. For example, knowing the probability distribution of a property loss and the distribution of a cyber loss does not tell us whether large values tend to occur together. Independence is a precise assumption about joint event probabilities: it says that knowing an event about one loss does not change the probability of an event about the other.
The definition applies to the losses as random variables, not just to one pair of thresholds. To establish independence, the required product relationship must hold for every relevant pair of events. This distinction is important in practical modeling: matching a single joint probability to a product is not enough.
The condition is stronger than a statement about one outcome or one selected pair of events. It must hold across all such pairs. In particular, the probabilities of joint events can be calculated as products only after independence has been established or explicitly assumed.
Testing Independence With Discrete Losses
For discrete losses, the joint probability mass function gives the probability of each possible pair of values. The marginal probability of a value of \(X\) is found by summing the joint probabilities across all possible values of \(Z\); the marginal for \(Z\) is found by summing down the possible values of \(X\).
If this equality holds for every cell of the joint probability table, probabilities of more general events can be obtained by adding the relevant cells. Equivalently, independence can be checked directly for every pair of event sets. A failure in even one cell is enough to show that the losses are dependent.
Worked Example: Testing a Discrete Loss Table
A hypothetical insurer models two annual losses, \(X\) and \(Z\), in dollars. The only possible values of \(X\) are \(0\) and \(100\), and the only possible values of \(Z\) are \(0\) and \(200\). The joint probabilities are shown below. Determine whether the losses are independent, and calculate \(P(X>0,\ Z=200)\).
| \(Z=0\) | \(Z=200\) | Marginal for \(X\) | |
|---|---|---|---|
| \(X=0\) | 0.30 | 0.20 | 0.50 |
| \(X=100\) | 0.10 | 0.40 | 0.50 |
| Marginal for \(Z\) | 0.40 | 0.60 | 1.00 |
Method. Compare each cell with the product of its row and column marginal probabilities. The marginal probabilities are obtained by summing the table: \(P(X=0)=0.30+0.20=0.50\), \(P(X=100)=0.10+0.40=0.50\), \(P(Z=0)=0.30+0.10=0.40\), and \(P(Z=200)=0.20+0.40=0.60\).
For the cell \((X=0,Z=0)\), the product is \(0.50(0.40)=0.20\), but the table gives \(0.30\). Since these are unequal, the independence condition fails. One failed cell is sufficient; there is no need to test the remaining cells.
The event \(X>0\) means \(X=100\), so the requested joint event is the cell with \(X=100\) and \(Z=200\): $$ P(X>0,\ Z=200)=0.40. $$ If the losses were independent, this probability would be \(P(X>0)P(Z=200)=0.50(0.60)=0.30\). The table’s value, \(0.40\), confirms the difference.
Check. The four joint probabilities sum to \(0.30+0.20+0.10+0.40=1.00\), and the requested event probability is the appropriate table cell. The common mistake avoided here is treating one event pair as a complete test of independence; the unequal cell already disproves independence.
Testing Independence With Continuous Losses
For continuous losses with a joint density, independence can be checked by comparing the joint density with the product of the marginal densities. The joint density must factor into the two marginal densities almost everywhere over the possible values. In particular, the support matters: a product expression that works only on part of the plane does not establish independence if the joint density is zero elsewhere while the product of the marginals is positive there.
An equivalent test uses the joint distribution function. Write \(F_{X,Z}(x,z)=P(X\leq x,\ Z\leq z)\). Independence holds if and only if \(F_{X,Z}(x,z)=F_X(x)F_Z(z)\) for every \(x,z\). This is the event definition applied to the particular events \(X\leq x\) and \(Z\leq z\).
Why the Product Formula Holds
The density test follows from the definition by integrating over event regions. Suppose the losses are continuous and have joint density \(f_{X,Z}(x,z)=f_X(x)f_Z(z)\). For intervals or other suitable sets \(A\) and \(B\), the probability of both losses falling in their respective sets is
The separation into two integrals is possible because the integrand is a product of a function of \(x\) and a function of \(z\), and the event region is the rectangle \(A\times B\). Thus, the density factorization gives the defining event-probability condition. Conversely, independence requires the joint probabilities to factor for all such regions, which gives the corresponding product form for the joint distribution and, when densities exist, the joint density.
Worked Example: Checking a Rectangular Joint Density
A hypothetical model assigns continuous losses \(X\) and \(Z\) a joint density equal to \(1/6\) when \(0<x<2\) and \(0<z<3\), and equal to zero elsewhere. Determine whether the losses are independent, then calculate \(P(X\leq 1,\ Z>1)\).
Method. First find each marginal density by integrating the joint density over the other loss’s full range. For \(0<x<2\), $$ f_X(x)=\int_0^3 \frac{1}{6}\,dz=\frac{3}{6}=\frac{1}{2}. $$ It is zero outside \(0<x<2\). For \(0<z<3\), $$ f_Z(z)=\int_0^2 \frac{1}{6}\,dx=\frac{2}{6}=\frac{1}{3}. $$ It is zero outside \(0<z<3\).
Within the rectangle, the product of these marginals is \((1/2)(1/3)=1/6\), which equals the joint density. Outside the rectangle, at least one marginal density is zero, so their product is also zero. The factorization therefore holds over the whole support, and \(X\) and \(Z\) are independent.
Now \(P(X\leq 1)=1/2\), because the interval from 0 to 1 has length 1 within a uniform range of length 2. Also \(P(Z>1)=2/3\), because the interval from 1 to 3 has length 2 within a range of length 3. Independence gives $$ P(X\leq 1,\ Z>1)=\frac{1}{2}\frac{2}{3}=\frac{1}{3}. $$ Direct integration checks the result: the relevant rectangle has area \(1\cdot2=2\), so its probability is \(2(1/6)=1/3\).
Check. The density integrates to \((2)(3)(1/6)=1\), and the product and direct calculations both give \(1/3\). The common mistake avoided is checking factorization only inside part of a support without checking where the joint and marginal densities are zero.
A Support That Rules Out Independence
A joint density need not occupy a rectangular region. If the possible values of one loss restrict the possible values of the other, the joint support can reveal dependence. The next example checks both the density factorization and a specific pair of event probabilities.
Worked Example: Dependence on a Triangular Support
Suppose two nonnegative losses have joint density \(f_{X,Z}(x,z)=2\) for \(0<z<x<1\), and zero elsewhere. Determine whether they are independent by comparing a joint event probability with the product of its marginal probabilities.
Method. The possible pairs lie below the diagonal \(z=x\). Integrating the density over this triangle gives \(2\int_0^1 x\,dx=1\), so it is a valid joint density. For \(0<x<1\), the marginal density of \(X\) is $$ f_X(x)=\int_0^x 2\,dz=2x. $$ For \(0<z<1\), the marginal density of \(Z\) is $$ f_Z(z)=\int_z^1 2\,dx=2(1-z). $$
Consider the events \(X\leq 1/2\) and \(Z\leq 1/2\). Since \(Z<X\), whenever \(X\leq 1/2\), \(Z\) is also less than \(1/2\). Thus their intersection has probability $$ P(X\leq 1/2,\ Z\leq 1/2) =\int_0^{1/2}\int_0^x 2\,dz\,dx =\int_0^{1/2}2x\,dx =\left[x^2\right]_0^{1/2} =\frac{1}{4}. $$
The separate event probabilities are $$ P(X\leq 1/2)=\int_0^{1/2}2x\,dx=\frac{1}{4} $$ and $$ P(Z\leq 1/2)=\int_0^{1/2}2(1-z)\,dz =\left[2z-z^2\right]_0^{1/2} =1-\frac{1}{4} =\frac{3}{4}. $$ Their product is \((1/4)(3/4)=3/16=0.1875\), whereas the joint probability is \(1/4=0.2500\). These differ, so the losses are dependent.
Check. Both marginal densities integrate to 1: \(\int_0^1 2x\,dx=1\) and \(\int_0^1 2(1-z)\,dz=1\). The event probabilities use the triangular support, not the full square. A common mistake is to multiply the two marginal probabilities before establishing independence; here that would incorrectly give the joint probability as \(0.1875\).
Common Exam Trap: One Equality Is Not Enough
A common error is to test one convenient pair of events, find that its joint probability equals the product, and then conclude that the losses are independent. Independence is a universal condition: it must hold for every pair of event sets. A single mismatch disproves independence, while a single match does not prove it.
For example, two events can be disjoint and still happen to have joint probability equal to the product for that particular pair. That coincidence says nothing about all other pairs of events. For discrete losses, check every joint cell against its marginal product, or find one cell that fails. For continuous losses, check the joint-density factorization across the support, including regions where the density is zero.
Check Your Understanding
For each problem, state the independence conclusion and show the probability calculation requested.
- A discrete pair of hypothetical losses has \(P(X=0,Z=0)=0.15\), \(P(X=0,Z=100)=0.35\), \(P(X=50,Z=0)=0.25\), and \(P(X=50,Z=100)=0.25\). Find \(P(X>0,Z>0)\) and decide whether the losses are independent. (Answer: \(P(X>0,Z>0)=0.25\); not independent, since \(P(X=0,Z=0)=0.15\ne P(X=0)P(Z=0)=0.50(0.40)=0.20\).)
- Suppose independent discrete losses satisfy \(P(X=0)=0.70\), \(P(X=10)=0.30\), and \(P(Z=0)=0.50\), \(P(Z=20)=0.25\), \(P(Z=40)=0.25\). Calculate \(P(X>0,Z\geq20)\). (Answer: \(0.30(0.25+0.25)=0.1500\).)
- Let independent continuous losses be uniform on \(0<X<4\) and \(0<Z<5\). Calculate \(P(X\leq1,Z>2)\), using the interval lengths and independence. (Answer: \((1/4)(3/5)=0.1500\).)
- A joint density is \(1/2\) on \(0<z<x<2\) and zero elsewhere. Find \(P(X\leq1,Z\leq1)\), and compare it with \(P(X\leq1)P(Z\leq1)\). (Answer: joint probability \(0.2500\); product \(0.25(0.75)=0.1875\), so the losses are not independent.)
- A discrete pair has \(P(X=0)=0.20\), \(P(X=100)=0.50\), \(P(X=200)=0.30\), and \(P(Z=0)=0.60\), \(P(Z=500)=0.40\). The joint probabilities for \(Z=0\), \(Z=500\), respectively, are \((0.12,0.08)\) when \(X=0\), \((0.30,0.20)\) when \(X=100\), and \((0.18,0.12)\) when \(X=200\). Are the losses independent, and what is \(P(X\geq100,Z=500)\)? (Answer: independent, since every cell equals its marginal product; \(P(X\geq100,Z=500)=0.20+0.12=0.32\).)