Let's say we are tossing two fair coins. Let $X$ be the random variable representing the number of heads.

$X$ is defined on $(\Omega, F, P)$ where $\Omega = \{HH,HT,TH,TT\}$.

Say if we want to find $E[X|G]$ where $G$ is a $\sigma$-algebra which contains the information about the first coin toss. It is clear $G=\{\emptyset, \Omega,\{HH,HT\},\{TT,TH\} \}$, where $G \subset F$. By definition, we know that $E[X|G]$ is a random variable mapping $\Omega \rightarrow \mathbb{R} $

Intuitively we know that the conditional expectation of X given the first toss is a $H$ is $1.5$, and the conditional expection of X given the first toss is a $T$ is $0.5$

However, when we are evaluating this quantity formally, using $E[X|G](\omega)$, wouldn't this mean we already know $\omega$? Why isn't $E[X|G](HH)$ simply $2$?

I do not understand why $E[X|G]$ takes the input $\omega \in \Omega$, instead of a set $s \in G$?

Any help is greatly appreaciated!

Just a thought (can someone confirm if this is correct?): By evaluating $E[X|G](HH)$ we do not actually know $\omega = HH$, but simply that $\omega \in \{HH,HT\}$ since all we are given is $G$


Let $Y$ denote the number of heads thrown at the first toss.

$\mathbb E[X\mid G]:\Omega\to\mathbb R$ has the following properties:

  • it is measurable wrt $G$ or equivalently some $f:\mathbb R\to\mathbb R$ exists with $\mathbb E[X\mid G]=f(Y)$
  • for every $B\in G$ we have: $\mathbb E[\mathbb E[X\mid G]\mathbf1_B]=\mathbb EX\mathbf1_B$

Combining the bullets we find that $$\mathbb Ef(Y)\mathbf1_B=\mathbb EX\mathbf1_B\text{ for every }B\in G$$

Substituting $B=\{HH,HT\}$ in the second bullet we find:$$\frac14f(1)+\frac14f(1)+\frac140+\frac140=\frac142+\frac141+\frac140+\frac140$$ telling us that $f(1)=\frac32$.

Substituting $B=\{TT,TH\}$ in the second bullet we find:$$\frac140+\frac140+\frac14f(0)+\frac14f(0)=\frac140+\frac140+\frac140+\frac141$$ telling us that $f(0)=\frac12$.

$\mathbb E[X\mid G]$ is defined for every $\omega\in\Omega$ and is prescribed as follows:

  • $HH\stackrel{Y}{\to}1\stackrel{f}{\to}\frac32$
  • $HT\stackrel{Y}{\to}1\stackrel{f}{\to}\frac32$
  • $TT\stackrel{Y}{\to}0\stackrel{f}{\to}\frac12$
  • $TH\stackrel{Y}{\to}0\stackrel{f}{\to}\frac12$
  • $\begingroup$ Thanks a lot for the answer, this was very helpful! I have a quick question regarding $f(Y)$. Is there a way to evaluate this quality without setting $E[X|G] = f(Y)$? $\endgroup$ – Putter Lertplakorn Oct 17 '18 at 15:59
  • $\begingroup$ The essence is that $\mathbb E[X|G]$ is constant on the sets $\{HH,HT\}$ and $\{TT,TH\}$. That comes to same as that it factors through $Y$. I used that in my answer but you can also focus on "being constant on..." itself. $\endgroup$ – drhab Oct 18 '18 at 8:51

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.