Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $(\Omega, \mathcal{F}, P)$ be a probability space, $X$ an integrable random variable, $\mathcal{G} \subset \mathcal{F}$ a $\sigma$-field. The conditional expectation of $X$ given $\mathcal{G}$ is by definition the unique random variable $Y$ which is $\mathcal{G}$-measurable and satisfies $E[Y;A] = E[X;A]$ for all $A \in \mathcal{G}$. Proving the uniqueness of $Y$ is easy, but existence is harder. I am looking for a nice existence proof with minimal prerequisites.

The traditional proof is to invoke the Radon-Nikodym theorem: the signed measure $\nu(A) = E[X;A]$ on $(\Omega, \mathcal{G})$ is absolutely continuous to $\mu = P|_\mathcal{G}$, so take $Y$ to be the Radon-Nikodym derivative, and it clearly has the desired properties. But the proofs I know of the Radon-Nikodym theorem, while elementary, are somewhat involved (at least 2 pages, even if you only do the absolutely continuous case).

Another proof is to first take $X$ with finite variance, and note that $K = L^2(\Omega, \mathcal{G}, P)$ is a closed subspace of the Hilbert space $H = L^2(\Omega, \mathcal{F}, P)$; then take $Y$ to be the orthogonal projection of $X$ onto $K$. Again, it is then easy to see that $Y$ has the desired properties. But this is not as suitable for students with no functional analysis background. You can develop the necessary facts from scratch but it's a little tedious.

So I am wondering if anyone knows of a simple proof, preferably using only basic measure theory and probability facts.

share|cite|improve this question

For the basic case:

Assume that $X$ and $Y$ are random variables on a probability space $(\Omega, \mathcal{F}, P)$ where $E[|Y|] < \infty$. Further assume that $X$ and $Y$ have joint probability distribution $f_{X, Y}(x,y)$. Define:

$$g(x) = \int_{\mathbb{R}} \frac{f_{X,Y}(x,y)}{f_X(x)} \, dy$$

where $f_X$ is the marginal density of $X$. now $g$ is the conditional expectation $E[Y|X = x]$ from elementary probability theory. Now we can see that $E[Y|X] = g(X)$. Now $g$ is $\sigma(X)$-measurable so now we need to check:

$$\int_A g(X) \, dP = \int_A Y \, dP \textrm{ for $A$ in $\sigma(X)$}$$

This is the partial-averaging-property, so we get the conditional expectation. Well, this is just some syntax-manipulation so I'll skip that. I can add it if you want.

share|cite|improve this answer
Thanks, but I am really interested in the general version where we may condition on any $\sigma$-field and the random variables need not be absolutely continuous. – Nate Eldredge Oct 6 '10 at 18:37
Yes, I thought so. But maybe someone else can use it somewhere in the future. For that I only know the two methods you've sketched. – Jonas Teuwen Oct 6 '10 at 21:55

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.