Let $X_i,X_j,X_k$ be IID random variables with finite moments. In particular we denote $E[X_i] := \mu$, $E[(X_i - \mu)^4] := \mu_4$ and $Var[X_i] := \sigma^2$.

Why does the expected value of $$\left[{1\over2}(X_i-X_j)^2-\sigma^2\right] \left[{1\over2}(X_i-X_k)^2-\sigma^2\right]$$

equal $(\mu_4-\sigma^4)/4$? This question comes from this answer (point 2) to a question about the variance of the sample variance.

My attempt

The only route I see is to expand the product, I obtain $$ \frac{1}{4} E[X_i^4] + E[X_i^2]^2 - E[X_i^3]E[X_i] -\frac{5}{2} \sigma^2 E[X_i^2] +3\sigma^2E[X_i^2] + \sigma^4$$ which is not correct.


Calculating it along your route leads to:$$\frac14[\mu_4-4\mu_1\mu_3+3\mu_2^2]-\frac12\sigma^2[4\mu_2-4\mu_1^2]+\sigma^4\tag1$$

(so if you did not make a mistake, then I did)

WLOG you may assume that $\mu_1=0$ and consequently $\mu_2=\sigma^2$.

This because the expression does not change if $X_i,X_j,X_k$ are replaced by $X_i-\mu_1,X_j-\mu_1,X_k-\mu_1$ respectively.

Substituting this in $(1)$ leads to: $$\frac14\mu_4-\frac14\sigma^4$$


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.