0
$\begingroup$

I'm trying to prove that the covariance of a bivariate normal distribution, $Cov(X,Y)$, is equal to $\rho\sigma_X\sigma_Y$. I'm getting stuck.

The approach I have to use is to apply a change of variables to standardize the variables X and Y, and apply the formula for covariance $Cov(X,Y) = E[(X-\mu_X)(Y-\mu_Y)]$, which is equal to:

$$Cov(X,Y)=E[(X-\mu_X)(Y-\mu_Y)]=\int_{-\infty}^\infty\int_{-\infty}^\infty(x-\mu_X)(y-\mu_Y)f(x,y)dxdy$$

Where $f(x,y)$ is the nasty pdf for a bivariate normal. Then, when you standardize and do a change of variable ... and complete the square... you get:

$$Cov(X,Y)=\sigma_X\sigma_Y\int_{-\infty}^\infty\int_{-\infty}^\infty \frac{uv}{2\pi\sqrt{1-\rho^2}}\exp\left(-\frac{(v-\rho u)^2}{2(1-\rho^2)} - \frac{u^2}{2}\right) dudv$$

So, this can be separated into two integrals that look like the expected value of normal random variables...

$$Cov(X,Y)=\sigma_X\sigma_Y\int_{-\infty}^\infty \frac{u}{\sqrt{2\pi}} \exp{\left(\frac{-u^2}{2}\right)} \int_{-\infty}^\infty\frac{v}{\sqrt{2\pi}\sqrt{1-\rho^2}}\exp\left( \frac{-(v-\rho u)^2}{2(1-\rho^2)} \right)dvdu$$

But that looks like $\sigma_X\sigma_Y*E[N(0,1)]*E[N(\rho u, 1-\rho^2)]$ which is ... zero! Because the expected value of $N(0,1)$ is 0. But it really should come out to $\rho\sigma_X\sigma_Y$ What am I missing?

$\endgroup$
0
$\begingroup$

But that looks like $σ_X σ_Y \cdot \mathsf E[N(0,1)]\cdot \mathsf E[N(ρu,1−ρ^2 )]$ which is ... zero! Because the expected value of $N(0,1)$ is $0$. But it really should come out to $ρσ_X σ_Y$.

What am I missing?

You don't have a product of two separate integrals.

$$\int_\Bbb R g(u) \left(\int_\Bbb R h(u,v)\operatorname d v\right)\operatorname d u \quad\neq\quad \left(\int_\Bbb R g(u) \operatorname d u \right)\cdot \left(\int_\Bbb R\int_\Bbb R h(u,v)\operatorname d v\operatorname d u \right)$$

$\endgroup$
  • $\begingroup$ Oh man, after looking at it again and evaluating it properly, it all falls into place now! Thank you! $\endgroup$ – user19650 Apr 21 '16 at 7:03

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.