Independence and Product Measures

"These two stocks are unrelated" is a sentence traders say all the time, and it needs a precise mathematical meaning before you can build a portfolio model out of it. The imprecise version — "knowing one tells you nothing about the other" — is correct in spirit, but it needs to be turned into an equation before a computer can use it. That equation is a multiplication rule, and the space it lives on is called a product measure.

The analogy: two separate spinners

Picture two carnival spinners, spun independently — the result of one has no physical effect on the other. If spinner A lands on red with chance $\tfrac13$ and spinner B lands on blue with chance $\tfrac14$, common sense says the chance of "A red and B blue" should be $\tfrac13 \times \tfrac14 = \tfrac1{12}$ — you multiply, you don't add, and you don't need to know anything about how A landed to answer the question about B. That multiplication instinct is exactly the formal definition of independence; nothing more mysterious is going on.

Writing it down

Two events $A$ and $B$ are independent if

$$P(A \cap B) = P(A) \, P(B) .$$

In words: the probability that both happen equals the product of their individual probabilities. This is not something you assume by eyeballing a situation as "unrelated" — it is a testable equation, and events can fail it even when they feel unrelated.

Two random variables $X$ and $Y$ are independent if this multiplication rule holds for every pair of events they can generate — equivalently, their joint distribution factors as the product of the two marginal distributions:

$$f_{X,Y}(x,y) = f_X(x) \, f_Y(y) \quad \text{for all } x, y .$$

In words: the combined density (or probability, for discrete variables) at any point splits cleanly into "how likely is $X$ alone" times "how likely is $Y$ alone" — nothing about $x$ changes how likely $y$ is, and vice versa. When the underlying sample space is literally built by pairing up two separate spaces — one for $X$, one for $Y$ — with probabilities assigned exactly this multiplicatively, that combined space is called a product measure, and Fubini-Tonelli is precisely the theorem that lets you integrate over a product measure by integrating each coordinate separately.

Worked example 1: checking independence with a joint table

Roll a fair die ($X$) and flip a fair coin ($Y$) together. $P(X=3) = 1/6$ and $P(Y=\text{heads}) = 1/2$. Are "$X=3$" and "$Y=\text{heads}$" independent? Compute directly: there are $6 \times 2 = 12$ equally likely combined outcomes, and exactly one of them is "$X=3$ and $Y=\text{heads}$," so $P(X=3, Y=H) = 1/12$. Now check the rule: $P(X=3) \times P(Y=H) = \tfrac16 \times \tfrac12 = \tfrac1{12}$. It matches, so the events are independent — as the physical setup (a die and a coin have no way to influence each other) suggested they should be.

Worked example 2: two variables that look unrelated but aren't

Let $X$ be uniform on $\{-1, 0, 1\}$ (each with probability $1/3$) and let $Y = X^2$. Compute $E[X] = 0$ and $E[XY] = E[X^3] = \tfrac13(-1)^3 + \tfrac13(0)^3 + \tfrac13(1)^3 = 0$, so $\text{Cov}(X,Y) = E[XY]-E[X]E[Y] = 0$ — zero covariance, which might tempt you to call them independent. But check the actual definition: $P(X=0) = 1/3$ and $P(Y=0) = P(X=0) = 1/3$, so if independent, $P(X=0, Y=0)$ should equal $\tfrac13 \times \tfrac13 = \tfrac19$. In reality, $X=0$ forces $Y=X^2=0$, so $P(X=0,Y=0) = 1/3 \neq 1/9$. The multiplication rule fails badly — $X$ and $Y$ are strongly dependent (knowing $X$ tells you $Y$ exactly), despite zero covariance.

What this means in practice

Portfolio variance formulas, joint default-probability models, and every Monte Carlo simulation that draws separate random numbers for separate risk factors all rely on the independence multiplication rule to justify treating draws as combinable via simple products or sums of variance. Getting it wrong in either direction is costly: assuming independence when there is hidden dependence (as in worked example 2) is exactly the failure mode behind the 2008 CDO mispricing of correlated mortgage defaults.