1

I have been wondering how to calculate $E[X\mid E[X\mid Y]]$. The lecturer gave this question after explaining how the error $X-\widehat{X}$ will always be perpendicular to any vector $Z-\widehat{X}$, which is in the linear space $S$ built from any linear combination of functions $Y_{1},Y_{2},\cdots,Y_{n}$.

I know that $E[X\mid Y]$ will be a function of $Y$, so we are basically calculating the mean of $X$ given the optimal estimator of $X$. I'm just stuck here with getting a final answer, and I can't draw this in a graphical way. I hope if someone can shed some light on this.

Ѕᴀᴀᴅ
  • 34,263
Firas Abd El Gani
  • 2,136

1 Answers1

1

The answer is $E(X|Y)$. To show that we have to show that $$\int_E XdP= \int_E E(X|Y) dP$$ for all $E \in \sigma (E(X|Y))$. But note that $\sigma (E(X|Y)) \subseteq \sigma (Y)$. Hence the equation holds by deinition of $E(X|Y)$.

Kavi Rama Murthy
  • 311,013
  • what is $\sigma$ in $\sigma (E(X|Y))$? I'm just not familiar with the notation. – Firas Abd El Gani May 08 '20 at 08:44
  • @FirasAliAbdelGhani Do you know the general definition of conditional expectation given a sigma algebra? Notation I used: $\sigma (Z)$ is the sigma algebra generated by $Z$ which is nothing but the collection of sets $Z^{-1}(E)$ where $E$ is a Borel set in $\mathbb R$. – Kavi Rama Murthy May 08 '20 at 08:48
  • I can read about this but that's not something in the scope of my knowledge. – Firas Abd El Gani May 08 '20 at 08:57