0
$\begingroup$

Consider a $W \sim \text{Pareto}(\beta)$ and let $\mathbf{X}=(X_1,X_2)^T$ be a bivariate vector whose components are given by:

$$X_1 = \sqrt{W}(Z_1+Z_2) \quad \text{and} \quad X_2 = \sqrt{W}(Z_1-Z_2)$$

where $Z_1$ and $Z_2$ iid standard normal, independent of $W$.

Now since it is obvious that the margins of $\mathbf{X}$ are not independent, I would like to show that formally. Natural way of showing this would be to show that $\mathrm{E}\left[X_1X_2\right] \neq \mathrm{E}[X_1]\mathrm{E}[X_2]$, however this approach doesn't produce anything useful.

How does one go about proving random variables are $\textit{not}$ independent in general?

Showing that their joint PDF/CDF do not factorize into a product form seems difficult in general.

$\endgroup$
3
$\begingroup$

Partial answer: compute $EX_1^{2}X_2^{2}$ and $EX_1^{2}EX_2^{2}$. If these two are equal you will see that $EW^{2}=(EW)^{2}$ or $var (W)=0$ which is not true. However this works only for $\beta > 2$ since $W$ has infinite variance for $\beta \leq 2$.

$\endgroup$
2
$\begingroup$

Hint:

If $X,Y$ are independent then $X^2,Y^2$ are independent.

or more generally then $f(X),f(Y)$ are independent (for suitable $f$).

$\endgroup$
  • $\begingroup$ Where can I find the proof of this claim? $\endgroup$ – Žan Žurič Aug 3 '18 at 10:15
  • $\begingroup$ Nevermind. I found the proof here. Thank you for the hint. $\endgroup$ – Žan Žurič Aug 3 '18 at 10:27

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.