I've been slowly working my way through problems in Casella and Burger's Statistical Inference and I stumbled upon this problem:

don't worry about problem 7.19--it won't help here

I haven't been able to get past (a). I've tried using the approximation formulas derived from the Delta Method:

$$E[\frac{U}{V}] \approx \frac{E[U]}{E[V]}$$ $$\mathbb{and}$$ $$Var(\frac{U}{V}) \approx (\frac{E[U]}{E[V]})^{2}\cdot(\frac{Var(U)}{E(U)} - \frac{2Cov(U,V)}{E[U]\cdot E[V]} + \frac{Var(V)}{E[V]})$$ where U,V are random variables. Before I go on, observe that $$\frac{\sum(X_i Y_i)}{\sum X^{2}_i} = \beta + \frac{\sum(X_i \epsilon_i)}{\sum(X^{2}_i)}$$ Approximating our expectation, we obtain $$E[\beta + \frac{\sum(X_i \epsilon_i)}{\sum(X^{2}_i)}]=\beta$$ since $X_i$ and $\epsilon_i$ are independent of each other and $E[\epsilon_i]=0$. Now, if I proceed the following, I get variance of 0 using the approximation formula I have up top: $$Var(\beta + \frac{\sum(X_i \epsilon_i)}{\sum(X^{2}_i)})=Var(\frac{\sum(X_i \epsilon_i)}{\sum(X^{2}_i)})\approx 0$$ which seems remarkably unlikely. Thus, I am wondering if it is possible to have the following approximation given the independence of our samples:

$$Var(\frac{\sum(X_i \epsilon_i)}{\sum(X^{2}_i)})\approx\frac{Var(\sum(X_i \epsilon_i))}{Var(\sum(X^{2}_i))}$$

If this is indeed possible, would someone be kind enough to give an explanation as to why?


$Var( X_i\epsilon_i)= E[(X_i\epsilon_i)^2]-(E[X_i\epsilon_i])^2 =EX_i^2 \times E\epsilon_i^2-0=(\mu^2+\tau^2)\sigma^2$

Corrected formula:

$$Var(\frac{U}{V}) \approx (\frac{E[U]}{E[V]})^{2}\cdot(\frac{Var(U)}{(E(U))^2} - \frac{2Cov(U,V)}{E[U]\cdot E[V]} + \frac{Var(V)}{(E[V])^2})$$

and hence in this case(since $EU=0$):

$$Var(\frac{U}{V}) \approx \frac{Var(U)}{(E(V))^2} = \frac{n(\mu^2+\tau^2)\sigma^2}{(n(\mu^2+\tau^2))^2} = \frac{\sigma^2}{n(\mu^2+\tau^2)}$$

  • $\begingroup$ Oh god. So I was fried and wrote the formula out wrong. -_- Thank you, rightskewed! $\endgroup$ – Savage Henry Sep 26 '15 at 6:50

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.