For me it is pretty clear how the conditional expectation

$$ \mathbb{E}\left[X\left|\right.\mathcal{G}\right]\quad (1) $$

of a random variable $X$ with respect to a $\sigma$-algebra $\mathcal{G}$ is defined. In particular, $(1)$ is the almost-unique $\mathcal{G}$-measurable random variable such that

$$ \mathbb{E}[(X-\mathbb{E}\left[X\left|\right.\mathcal{G}\right])^2] = \inf_{Z\text{ is }\mathcal{G}\text{-measurable}}\mathbb{E}[(X-Z)^2]. $$

This said, how the conditional expected value

$$ \mathbb{E}\left[X\left|\right.Y=a\right], $$ where $Y$ is another random variable and $a\in\mathbb{R}$, is formally defined?


If you understood conditional expectation with respect to a sub-$\sigma$-algebra (actually your definition only work for $X\in L^2$, for $L^1$ you should use Radon-Nikodym or limiting argument), then the conditional expectation $\mathbb{E}[X\mid A]$ with respect to an event $A$ is simply as the value of the conditional expectation with respect to the sub-$\sigma$-algebra $\{\varnothing,A,A^c,\Omega\}]$ evaluated at points of $A$ (since $\mathbb{R}$ is Hausdorff every $a\in A$ give the same answer). Note that since $\mathbb{E}[X\mid\mathcal{G}]$ is only defined as an element of $L^1(\mathcal{G})$ (i.e. modulo almost sure equivalence) and not the set of all integrable $\mathscr{G}$-measurable random variable $\mathscr{L}^1(\mathcal{G})$, this is ill-defined for null event $A$, and is well-defined if $A$ is non-null. Conventionally, if $A$ is null we define $\mathbb{E}[X\mid A]=0$ (similar to how we construct $\mathbb{E}[X\mid\mathcal{G}]$) for definiteness). In particular, we have $$ \mathbb{E}[X\mid A]\mathbb{P}(A)=\mathbb{E}[X1_A]. $$ Note that $Y=a$ is an event $Y^{-1}(a)$, so this gives $\mathbb{E}[X\mid Y=a]$.

There are authors who insist on promoting nonsense such as $\displaystyle\mathbb{E}[X\mid Y=a]=\int_{\mathbb{R}} x\frac{f_{X,Y}(x,a)}{f_Y(a)}\,\mathrm{d}x$ even when $\mathbb{P}(Y=y)=0$. If you encounter it in any book, please do the good deed and burn the book along with the author.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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