St · Statistics & Probability

Is the Conditional Expectation of a Positive Random Variable Positive?

Published By the 9to5Science editorial team

At a glance

  1. 1. Fix the meaning

    Assume positive means strictly greater than zero almost surely, and that the random variable is integrable.

  2. 2. Use nonnegativity

    Conditional expectation preserves nonnegativity, so the conditional expectation is at least zero almost surely.

  3. 3. Test its zero set

    On the event where the conditional expectation is zero, its defining integral identity forces the original variable to be zero almost surely.

  4. 4. Conclude strict positivity

    Since the original variable is strictly positive almost surely, that zero event must have probability zero.

Advertisement

Yes—if “positive” means strictly greater than zero almost surely and the random variable is integrable, then its conditional expectation is strictly positive almost surely as well. This does not mean it has a fixed positive lower bound: it may take values arbitrarily close to zero.

The distinction between strictly positive and merely nonnegative matters. A nonnegative random variable can be zero on an event of positive probability, and its conditional expectation can also be zero there.

The statement

Let XX be an integrable random variable with X>0X>0 almost surely, and let G\mathcal{G} be a sigma-algebra representing the information you condition on. Then

E[X∣G]>0almost surely.\mathbb{E}[X\mid\mathcal{G}]>0\quad\text{almost surely}.

Conditional expectation is defined up to changes on events of probability zero, so conclusions about it are normally stated almost surely—not necessarily at every individual outcome.

Why the result holds

Set Y=E[X∣G]Y=\mathbb{E}[X\mid\mathcal{G}]. Since X≥0X\geq 0 almost surely, the conditional expectation satisfies Y≥0Y\geq 0 almost surely. The only remaining possibility that would contradict strict positivity is that Y=0Y=0 on an event of positive probability.

Consider the event A={Y=0}A=\{Y=0\}. This event belongs to G\mathcal{G} because YY is G\mathcal{G}-measurable. By the defining property of conditional expectation, for every event in G\mathcal{G},

E[X1A]=E[Y1A].\mathbb{E}[X\mathbf{1}_A]=\mathbb{E}[Y\mathbf{1}_A].

But Y=0Y=0 on AA, so the right-hand side is zero. Thus E[X1A]=0\mathbb{E}[X\mathbf{1}_A]=0. Since X1AX\mathbf{1}_A is nonnegative, it can have expectation zero only if it is zero almost surely. That would require X=0X=0 almost surely on AA. Because X>0X>0 almost surely, this forces P(A)=0\mathbb{P}(A)=0. Therefore Y>0Y>0 almost surely.

Worked example

Suppose the sample space has four equally likely outcomes, and XX takes the values 1,3,2,81,3,2,8 on them, respectively. Let G\mathcal{G} reveal which pair of outcomes occurred: either the first pair or the second pair. Conditional expectation averages XX within each pair:

E[X∣G]={(1+3)/2=2,on the first pair,(2+8)/2=5,on the second pair.\mathbb{E}[X\mid\mathcal{G}]= \begin{cases} (1+3)/2=2, & \text{on the first pair},\\ (2+8)/2=5, & \text{on the second pair}. \end{cases}

Every value of XX is positive, and both conditional averages are positive. The result does not require the conditional expectation to equal XX or to be constant; it says only that it cannot be zero on a set of positive probability.

What changes if “positive” means nonnegative?

Some texts use “positive” to mean X≥0X\geq 0, while others reserve it for X>0X>0. If you know only that XX is nonnegative, you can conclude only that

E[X∣G]≥0almost surely.\mathbb{E}[X\mid\mathcal{G}]\geq 0\quad\text{almost surely}.

For example, if XX is identically zero, its conditional expectation is zero. More generally, conditional expectation may be zero on parts of the sample space where XX is zero almost surely given the information in G\mathcal{G}.

A related distinction: strict positivity does not imply a uniform positive lower bound. If XX is uniform on (0,1)(0,1) and G\mathcal{G} contains all the information about XX, then E[X∣G]=X>0\mathbb{E}[X\mid\mathcal{G}]=X>0 almost surely, even though XX can be arbitrarily close to zero. There is no fixed c>0c>0 such that X≥cX\geq c almost surely.

Common mistakes

  • Claiming pointwise positivity everywhere: Conditional expectations are defined only up to almost-sure equality. The correct claim is positivity almost surely.
  • Using only nonnegativity to claim strict positivity: From X≥0X\geq 0 you get E[X∣G]≥0\mathbb{E}[X\mid\mathcal{G}]\geq 0, not necessarily a strict inequality.
  • Assuming conditioning cannot change values: Conditional expectation averages according to the information in G\mathcal{G}. It need not equal XX, even though strict positivity is preserved.
  • Assuming a positive lower bound: The conclusion rules out zero with positive probability; it does not say the conditional expectation is bounded away from zero.

For a nonnegative random variable that is not integrable, conditional expectation can be defined in the extended nonnegative sense, and the same zero-set argument shows that E[X∣G]=0\mathbb{E}[X\mid\mathcal{G}]=0 only on events where X=0X=0 almost surely. If X>0X>0 almost surely, this again gives strict positivity of the conditional expectation almost surely. The standard integrable version above is the usual setting for the theorem.

References

  1. Probability with Martingales — David Williams
  2. A First Look at Rigorous Probability Theory — Jeffrey S. Rosenthal

Spotted an error? Send a correction. Read how we write and check articles in our editorial policy.