Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $X,Y$ are two random variables which are not necessarily independent. It is easy to get $\mathbb{E}(X)$ ann $\mathbb{E}(Y)$. I want to know: is there some approximation to $\mathbb{E}(\frac{X}{Y})$?

[Update] The background is that I want to calculate the expectation of Pearson product-moment correlation coefficient - $\mathbb{E}(\rho_{xy})$.

The Pearson product-moment correlation coefficient is $\rho_{xy} = \frac{Cov_{xy}}{\sigma_x\sigma_y}$

It is easily to get $\mathbb{E}(Cov_{xy})$ and $\mathbb{E}(\sigma_x\sigma_y)$, so I want to know a approximation to get $\mathbb{E}(\rho_{xy})$ from the above two value.

share|improve this question
    
The expected value of a constant is the constant itself. For random variables $X$ and $Y$, $Cov_{xy}$ by which I assume you mean $E[(X-E[X])(Y-E[Y])]$ is a constant. So, when you write $E[Cov_{xy}]$, its value is just $Cov_{xy}$, unless there is some other random variable with respect to which you are computing the correlation. That is, are there three random variables $X,Y,Z$ and $Cov_{xy}$ is the conditional covariance of $X$ and $Y$ conditioned on $\{Z = \alpha\}$, say, and you are now trying to find the unconditional covariance? –  Dilip Sarwate Jan 15 '12 at 13:29
    
@DilipSarwate Please see the second answer of stats.stackexchange.com/questions/10981/… –  Fan Zhang Jan 15 '12 at 13:43
    
I have looked at the question and answer on stats.SE. There the issue was estimating the variance, covariance, correlation coefficient, etc. given $n$ i.i.d. samples from $(X,Y)$ with unknown joint distribution, and you got answers for that. Here you are not talking about $n$ i.i.d. samples, or estimating $Cov_{xy}$ from such samples, etc. but just about two random variables $X$ and $Y$. I ask again, what is the meaning of $E[Cov_{xy}]$? –  Dilip Sarwate Jan 15 '12 at 15:34
add comment

1 Answer

up vote 2 down vote accepted

If $X$ and $Y$ have a joint density $f(x,y)$, $E[X/Y] = \int_{-\infty}^\infty \int_{-\infty}^\infty \frac{x}{y} \ f(x,y)\ dx\ dy$ (assuming that converges absolutely). Similarly, if they have a joint probability mass function $p(x,y)$, $E[X/Y] = \sum_x \sum_y \frac{x}{y} p(x,y)$.

Let $\mu = E[Y]$. If you can treat $Y - \mu$ as small compared to $\mu$ (in particular if there is $c>0$ such that $c < Y < 2 \mu - c$ almost surely) then $$E[X/Y] = \sum_{j=0}^\infty (-1)^j \frac{E[X (Y - \mu)^j]}{ \mu^{j+1}}$$ so you could use a partial sum of that as an approximation.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.