Show expectation is infinite

176 Views Asked by At

Let $X_1,\ldots,X_n$ be independent, identically distributed with expectation 1 and finite variance. Find the limit distribution of $\sqrt{n}(\bar{X}_n^{-1}-1)$. If the random variables are sampled from a density $f$ that is bounded and strictly positive in a neighborhood of zero, show that $\text{E}|\bar{X}_n^{-1}|=\infty$ for every $n$.

My attempt:

First question is fairly simple. From the central limit theorem we have $$\sqrt{n}(\bar{X}_n-1) \overset{D}{\to} \text{N}(0,\sigma^2)$$ and using Delta Method, where $g(x)=1/x$ is continuously differentiable at $1$ we have $$\begin{align*} \sqrt{n}(g(\bar{X}_n) - g(1)) &\overset{D}{\to} \text{N}(0, g'(1)^2\sigma^2)\\ \sqrt{n}(\bar{X}_n^{-1}-1) &\overset{D}{\to} \text{N}(0,\sigma^2) \end{align*}$$ I am struggling with the second part.

I can see that in a neighborhood around $0$, $1/x$ is going to tend to be very large, and potentially pull the whole integral toward infinity, but I'm not exactly sure how to formally write this.

1

There are 1 best solutions below

2
On BEST ANSWER

If the random variables are sampled from a density $f$ that is bounded and strictly positive in a neighborhood of zero, show that $\text{E}|\bar{X}_n^{-1}|=\infty$ for every $n$.

This is wrong if the density $f$ is allowed to be zero on the left of $0$ (and bounded and strictly positive on the right of $0$): try $f$ the standard exponential density, then the density of $S_n=X_1+\cdots+X_n$ is proportional to $x^{n-1}\mathrm e^{-x}\mathbf 1_{x\gt0}$ hence $E(\bar X_n^{-1})$ is finite for every $n\geqslant2$. More generally, $E(\bar X_n^{-\alpha})$ is finite for every $n\gt\alpha$.

If one insists that the density $f$ should be bounded and strictly positive on a (bilateral) neighborhood of $0$ and if one actually means that $f\geqslant\varepsilon$ uniformly on $(-\varepsilon,\varepsilon)$, then it is enough to prove the result when $f$ is the uniform density on $(-1,1)$.