1
$\begingroup$

Let $x_1, \ldots, x_n$ and $y_1, \ldots, y_n$ be two independent random samples from $X$ and $Y$. We have $\mu_X = E (X ) > 0, \mu_Y = E (Y ) > 0$ and $\sigma^2_X = Var (X )$ and $\sigma^2_Y = Var (Y )$. Derive the asymptotic distribution of $\frac{\overline x_n+ \overline y_n}{\overline x_n- \overline y_n}$, where ${\overline x_n}$ is the sample average of the $x_i$s.

So I looked at the following first:

It is known as the central limit theorem that the sample mean of any distribution has an asymptotic Gaussian distribution. So both Xn & Yn have a Gaussian distribution.

second: the sum of 2 Gaussian random variable is a gaussian random variable, the mean is just the sum of each Xn and Yn, the variance is also the sum of the individual variances since theres independence of each Xi third: the distribution of the ratio is more complicated since it involves doing an integral I guess.

So this would be the intuition could someone please help me formalize it especially step 3, I don’t have the mathematical capacity yet to do it.

Really need the help.

$\endgroup$
  • 1
    $\begingroup$ First establish the bivariate CLT, then use multivariate Delta method. $\endgroup$ – Zhanxiong Nov 18 '19 at 4:02
  • $\begingroup$ Could you break it down for me, that would be super helpful. $\endgroup$ – James Nov 18 '19 at 6:54
  • $\begingroup$ Asked and answered at math.stackexchange.com/q/3439657/321264. $\endgroup$ – StubbornAtom Nov 18 '19 at 13:12
  • $\begingroup$ It would be interesting to see a different approach, reason it being on a different thread as the one in math assumes a lot of factors known predominantly to mathematicians $\endgroup$ – James Nov 18 '19 at 13:28
  • $\begingroup$ The question is technically mathematical, so unlikely you would get a very different response here. Nevertheless, you should cross-link your posts if you choose to cross-post. $\endgroup$ – StubbornAtom Nov 18 '19 at 16:40
1
$\begingroup$

Assuming $(X_n)$ and $(Y_n)$ are two i.i.d sequences with $E|X_1|<\infty$ and $E|Y_1|<\infty$.

By the law of large numbers, we have the following convergence in probability:

$$T=\frac{\frac1n\sum\limits_{i=1}^n X_i+\frac1n\sum\limits_{i=1}^n Y_i}{\frac1n\sum\limits_{i=1}^n X_i-\frac1n\sum\limits_{i=1}^n Y_i}\stackrel{P}\longrightarrow\frac{E(X_1)+E(Y_1)}{E(X_1)-E(Y_1)}\,,$$

provided $E(X_1)\ne E(Y_1)$.

This only uses that $(X_n)$ and $(Y_n)$ are (separately) i.i.d, not the fact that they are independent.

As a result $T$ also converges in distribution to $\frac{E(X_1)+E(Y_1)}{E(X_1)-E(Y_1)}$.

I am not sure if we can say anything about a non-degenerate asymptotic distribution of $T$ in general.

$\endgroup$
  • $\begingroup$ I wonder what happens when the $X_i$ are iid Cauchy and so are the $Y_i.$ $\endgroup$ – whuber Nov 18 '19 at 19:11
  • 1
    $\begingroup$ @whuber I have assumed $E(X_1),E(Y_1)$ both exist. $\endgroup$ – StubbornAtom Nov 18 '19 at 19:19
0
$\begingroup$

Don't know the answer but by the CLM you know that:
$\sqrt{n}({\overline x_𝑛}-µ_x) \rightarrow_p N(0,σ^2_X)$

Maybe someone else can chime in, as to how to proceed.

$\endgroup$
  • $\begingroup$ Thank you for trying $\endgroup$ – James Nov 18 '19 at 10:54

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.