2
$\begingroup$

The bounds of correlation coefficient $\rho$ is shown to be $\pm 1$ in class. In many situations the bounds are sharper, i.e. they stay away from $+1$ or $-1$. Consider the random variables $X_1,\dots,X_n$ such that $E(X_i)=\mu$, $Var(X_i)=\sigma^2$, and for each pair $i \neq j$, $Corr(X_i,X_j)=\rho$. Such equi-correlation structure is sometimes referred to as exchangeable or compound symmetry. Show that in this case $$\rho \geq -\frac{1}{n-1}$$

$\endgroup$
2
$\begingroup$

Using that $$\begin{align*}0\le\operatorname{Var}(\overline{X}_n)&=\frac{1}{n^2}\left(\sum_{i=1}^{n}\operatorname{Var}(X_i)+\sum_{i=1}^{n}\sum_{i\neq j=1}^{n}\operatorname{Cov}(X_i,X_j)\right)=\\\\&=\frac{1}{n^2}\left(nσ^2+\sum_{i=1}^{n}\sum_{i\neq j=1}^{n}\frac{\operatorname{Cov}(X_i,X_j)}{\sqrt{σ^2}\sqrt{σ^2}}\cdot σ^2\right)=\\\\&=\frac{1}{n^2}\left(nσ^2+\underbrace{\sum_{i=1}^{n}\sum_{i\neq j=1}^{n}}_{\begin{matrix}n(n-1)\\\text{ summands}\end{matrix}}\underbrace{\operatorname{Corr}(X_i,X_j)}_{=ρ}\cdot σ^2\right)=\\\\&=\frac{1}{n^2}\left(nσ^2+n(n-1)ρσ^2\right)=\frac{σ^2}{n}\left(1+(n-1)ρ\right)\end{align*}$$ and since $\frac{σ^2}{n}>0$ you obtain that $$1+(n-1)ρ\ge 0 \implies ρ\ge-\frac{1}{n-1}$$

$\endgroup$

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.