Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

For a Cauchy distribution with density $f(x) = \frac{1}{\pi(1 + x^2)}$ , it doesn't have well-defined moments. Therefore both the law of large numbers and the central limit theorem can't apply to the distribution. For example, Given iid sample points $X_1, \dots, X_n$ of the distribution, as $n\to \infty$, $\bar{X}$ will not converge to a constant in probability.

However a note says under the above Cauchy distribution, "asymptotic variance of $\sqrt{n} \bar{X}$ is $\infty$", which seems like a version of CLT?

The definition of the asymptotic variance of a statistic is defined in Casella's Statistical Infernce:

Definition 10.1.9 For an estimator $T_n$, suppose that $k_n(T_n - \tau(\theta)) \to n(0, \sigma^2)$ in distribution. The parameter $\sigma^2$ is called the asymptotic variance or variance of the limit distribution of $T_n$.

So I wonder if the above definition applies to the case here, for example, with $\tau(\theta) =0$ and $k_n = \sqrt{n}$? If yes, the limit distribution of $\sqrt{n} \bar{X}$ is $N(0, \infty)$, the asymptotic mean of $\sqrt{n} \bar{X}$ is $0$ and its asymptotic variance is $\infty$? If no, how shall we understand "the asymptotic variance of $\sqrt{n} \bar{X}$ is $\infty$"?

Btw, for each $n$, $\sqrt{n} \bar{X}$ doesn't admit mean and variance, right?

Thanks and regards!

share|improve this question
add comment

1 Answer

For $X_i \sim \operatorname{Cauchy}(0,1)$, their linear combination is also Cauchy distributed, specifically $$ Z = \sum_{k=1}^n a_k X_k \sim \operatorname{Cauchy}\left(0, \sum_{k=1}^n |a_k| \right) $$ which is easily proven by considering a characteristic function: $$ \varphi_Z(t) = \prod_{k=1}^n \phi_X\left( a_k t \right) = \exp\left( - |t| \sum_{k=1}^n |a_k| \right) $$ which is the characteristic function of the claimed distribution.

With this say, $\bar{X} \sim \operatorname{Cauchy}\left(0, 1\right) $ for all $n$, and neither $\mathbb{E}(\bar{X})$ nor $\mathbb{Var}(\bar{X})$ is defined.

share|improve this answer
    
Thanks, Sasha! Why is "the asymptotic variance of n√X¯ is ∞"? –  Tim Mar 27 '13 at 20:41
    
I am not certain what is meant by this. $\bar{X}$ does not admit variance, and neither does $\sqrt{n} \bar{X}$, thus I would say $\mathbb{Var}(\sqrt{n} \bar{X}) = \infty$, but I see no need to bring up "asymptotic" here. Asymptotic would be appropriate if $\sqrt{n} \bar{X}$ had finite variance, which would grown boundlessly as $n$ were increasing. –  Sasha Mar 27 '13 at 20:57
    
Thanks! (1) the variance of $\bar{X}$ is actually not well defined, which can have different values ($0$ or $\infty$ or ...) when the improper integral is interpreted in different ways. If $\sqrt{n} \bar{X}$ is also such case, why would its variance is $\infty$ instead of $0$? –  Tim Mar 27 '13 at 21:09
    
(2) "Asymptotic would be appropriate if $\sqrt{n} \bar{X}$ had finite variance, which would grown boundlessly as n were increasing." Casella's book says, which I quoted in my post, the asymptotic variance is the variance of the limit distribution of $\sqrt{n} \bar{X}$. So the asymptotic variance is not the same as the limit of the variance of $\sqrt{n} \bar{X}$. –  Tim Mar 27 '13 at 21:09
    
By $\bar{X}$ is equal in distribution to the standard Cauchy, thus $\sqrt{n} \bar{X}$ is distributed as $\operatorname{Cauchy}(0,\sqrt{n})$. The scale parameter of it grows unboundedly, but the scale parameter has nothing to do with variance. –  Sasha Mar 27 '13 at 21:31
show 1 more comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.