2
$\begingroup$

Suppose $\{X_1,X_2,Z\}$ is a vector of 3 real valued continuous random variables with compact support, $f_1(X_1,X_2)$, $f_2(X_1,X_2)$, and $g(X_1,X_2)$ are measurable functions with at least 2 continuous partial derivatives all of which are integrable. Are the following equations valid ways of manipulating each equation? If not why not and if you have a pertinent reference I would appreciate it.

$1.) \;\;E\bigr[\; f_1(X_1, X_2) \; E[g(X_1,X_2) Z \; |X_1 ] \; \big|X_1 \bigr] = E\big[ E[f_1(X_1,X_2)g(X_1,X_2)Z|X_1] \big|X_1 \big] $

$2.) \;\;E\bigr[\; f_1(X_1, X_2)f_2(X_1,X_2) \; E[g(X_1,X_2) Z \; |X_1 ] \; \big|X_1 \bigr] = E\big[f_2(X_1,X_2) E[f_1(X_1,X_2)g(X_1,X_2)Z|X_1] \big|X_1 \big] $

Your help will be much appreciated.

$\endgroup$
2
$\begingroup$

1) Note that $E[g(X_1,X_2)Z \mid X_1]$ is only a function of $X_1$ and thus conditional on $X_1$ is a constant. Similarly, $E[f_1(X_1,X_2)g(X_1,X_2)Z \mid X_1]$ is only a function of $X_1$ and thus conditioned on $X_1$ is a constant. \begin{align*} E\left[f_1(X_1,X_2) E[g(X_1,X_2)Z \mid X_1] \mid X_1\right] & = E[g(X_1,X_2)Z \mid X_1] \,E\left[f_1(X_1,X_2) \mid X_1\right]\\ & \ne E\left[ f_1(X_1,X_2)g(X_1,X_2)Z \mid X_1\right]\\ & = E \left[ E\left[ f_1(X_1,X_2)g(X_1,X_2)Z \mid X_1\right] \mid X_1\right] \end{align*}

The last equality is due to the inside term being constant (as mentioned before) and the $\ne$ is because it is unknown because it is not certain than $f_1(X_1, X_2)$ and $g(x_1, X_2) Z$ are conditionally independent (when conditioned on $X_1$). The $\ne$ will be a $=$ is they were conditionally independent.

2) Similarly, you can show that the second statement is also untrue in general.

$\endgroup$
  • $\begingroup$ The conditioning here appears to be with respect to the random variable (essentially the sigma algebra generated by it), not on a specific value of it. Then, these are functions, not constants. $\endgroup$ – Alecos Papadopoulos May 4 '16 at 22:58
  • $\begingroup$ I guess, what I meant was that these are constants for the sake of the expectation. $\endgroup$ – Greenparker May 4 '16 at 23:25
2
$\begingroup$

Let $f_1(X_1, X_2) = \frac {X_1}{X_2},\; g(X_1,X_2) = \frac {X_2}{X_1}$. Then looking at the proposed equation $1).$, do you think that the right-hand side

$$ E\left(\; \frac {X_1}{X_2} \; E\left [\frac {X_2}{X_1} Z \; \mid X_1 \right] \; \big|X_1 \right) = E\left(\; \frac {1}{X_2} \; E\left [X_2 Z \; |X_1 \right] \; \big|X_1 \right)\\=E\left(\; \frac {1}{X_2} \; \; \big|X_1 \right)\cdot E\left [X_2 Z \; |X_1 \right] $$

is equal to the left hand side ?

$$ \;E\left( E\left [\frac {X_1}{X_2}\frac {X_2}{X_1}Z|X_1\right] \big|X_1 \right) = E\big[ Z \mid X_1 \big] $$

$\endgroup$

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.