1
$\begingroup$

Given that $X_1,\ldots,X_n$ are $n$ independent and identically distributed random variables. We know that they have finite moments up to third order e.g. $EX_i=0$, $EX_i^2<\infty$ and $EX_i^3<\infty$. The question is to find accurate probability bound for the quantity $\frac{n^2\sum_{i=1}^n X_i}{\sum_{i=1}^n X_i^3}$?

I have two results for this quantity. First is that since $\sum_{i=1}^n X_i = O_p(n^{1/2})$ based on central limit theorem (CLT) and by law large number we have $\sum_{i=1}^n X_i^3=O_{p}(n)$, then I have $$\frac{n^2\sum_{i=1}^n X_i}{\sum_{i=1}^n X_i^3} = \frac{n^2O_p(n^{1/2})}{O_p(n)}=O_p(n^{3/2}).$$ Second is that since $\sum_{i=1}^n X_i=O_p(n^{1/2})$ based on central limit theorem (CLT) and by CLT we have $\sum_{i=1}^n X_i^3=O_p(n^{1/2})$, thus

$$\frac{n^2\sum_{i=1}^n X_i}{\sum_{i=1}^n X_i^3} = \frac{n^2 O_p(n^{1/2})}{O_p(n^{1/2})} = O_p(n^2).$$

which one is correct probability bound? What's more, could I find exact order or more accurate order of the quantity?

$\endgroup$
  • $\begingroup$ The central limit theorem does not apply in the way you state unless $E[X_i]=0$. Is the mean zero? $\endgroup$ – Michael Feb 26 '16 at 4:52
  • $\begingroup$ Thx, Michael, I could add this assumption. $\endgroup$ – lzstat Feb 26 '16 at 4:54
  • $\begingroup$ It makes the problem harder if $E[X_i]=0$. What about $E[X_i^3]$? Is that zero? $\endgroup$ – Michael Feb 26 '16 at 4:55
  • $\begingroup$ we only know $EX_{i}^3$ is finite. $\endgroup$ – lzstat Feb 26 '16 at 4:57
  • $\begingroup$ Your CLT application to $X_i^3$ does not work unless $E[X_i^3]=0$. $\endgroup$ – Michael Feb 26 '16 at 5:04
1
$\begingroup$

Let $c = E[X_i^3]$ and assume $c\neq 0$. Then for large $n$ we have $$\frac{n^2\sum_{i=1}^nX_i}{\sum_{i=1}^nX_i^3} = \frac{n^{1.5}\frac{1}{\sqrt{n}}\sum_{i=1}^nX_i}{\frac{1}{n}\sum_{i=1}^n X_i^3} \approx \frac{n^{1.5}G_n}{c}$$ where $G_n$ is Gaussian with mean $0$ and variance $Var(X_1)$.

$\endgroup$
  • $\begingroup$ Note that, in this problem, we apply the LLN when the mean is nonzero (denominator), and the CLT when the mean is zero (numerator). $\endgroup$ – Michael Feb 26 '16 at 5:13
  • $\begingroup$ Yea, I understand. It makes sense. thank you $\endgroup$ – lzstat Feb 26 '16 at 5:16
  • $\begingroup$ You need to specify that $\approx$ is in probability only. $\endgroup$ – A.S. Feb 26 '16 at 6:07

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.