Sorry for asking a stupid question, but I was not able to find answer online. I have two iid normal random variables: $X_1, X_2$ and I want to find the expectation of $e^{X_1X_2}$. Is it $\int e^{x^2_1}f(x_1) dx_1$ (since their distributions are the same) or $\iint e^{x_1x_2}f(x_1)f(x_2)dx_1 dx_2$?

  • $\begingroup$ You mean convolution of $x_1$ and $x_2$? $\endgroup$ – GNUSupporter 8964民主女神 地下教會 Feb 19 '17 at 1:15
  • $\begingroup$ Sorry, used wrong notations. I mean just their multiplication. $\endgroup$ – user0347284 Feb 19 '17 at 1:20
  • $\begingroup$ Nice question. This is a fairly common point of confusion, and it's good to get it settled when it first arises. $\endgroup$ – BruceET Feb 19 '17 at 1:35

Not the same, second is correct. Even if $X_1$ and $X_2$ have the same distribution, that does not mean that $X_1^2$ and $X_1X_2$ have the same distribution. For example, $P(X_1^2>0)=1,$ but no so for the product.

Moreover, $Cov(X_1, X_2) = E(X_1X_2) - E(X_1)E(X_2),$ but $Var(X_1) = E(X_1^2) - [E(X)]^2.$ Your theory would equate these two.

Note: As an experiment, here is a sample of a million observations $X_{1i}$ from $\mathsf{Norm}(0,1)$ and, independently, another million $X_{2i}$'s from the same distribution. Results suggest that $E(X_1^2) = 1,$ which makes sense because $X_1^2 \sim \mathsf{Chisq}(1),$ a distribution known to have mean $1$. But results also suggest that $E(X_1X_2) = 0.$ Can you prove that?

x1 = rnorm(10^6);  x2 = rnorm(10^6)
mean(x1^2);  mean(x1*x2)
## 1.00017         # aprx 1
## -0.0004455942   # aprx 0

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.