Hi I am trying to solve the problem 5.13 of the book Statistical inference by George Casella and Roger L. Berger. The problem is

Formulate and prove a version of the WLLN with a weaker assumption than a finite variance. Use the Markov inequality to prove it.

I don't what it means by $\textbf{weaker assumption than a finite variance}$.

By Markov's inequalty we have for nonnegative random variable X $$P(X\ge\epsilon)\le\frac{E(X)}{\epsilon}$$

We need to prove $$P(|\bar{X}_n-\mu|\ge\epsilon)\to 0$$

So we have $P(|\bar{X}_n-\mu|\ge\epsilon)=P((\bar{X}_n-\mu)^2\ge\epsilon^2)\le\frac{E(\bar{X}_n-\mu)^2}{\epsilon^2}=\frac{\text{Var}(\bar{X}_n)}{\epsilon^2}=\frac{\sigma^2}{n\epsilon^2}$

Since I don't know what would be the weaker version I am stuck here. Any kind of help would be highly appreciated.

  • $\begingroup$ Any response to my answer below? $\endgroup$ – John Feb 13 '16 at 3:01
  • 1
    $\begingroup$ @JohnZHANG Sorry for not replying any. I wanted a different one. Yours answer has also given me a different aspect. $\endgroup$ – mint Feb 13 '16 at 8:52

Since there are several versions of WLLN with weaker assumptions, I am not sure which one you want. So I think your question is not well-defined. But one direct generalization of WLLN is that

Let $X_n$ be sequence of r.v, $\mu_n=ES_n$, $\sigma_n^2=var(S_n)$. If $\frac{\sigma_n^2}{b_n^2}\to 0$, then $$\frac{S_n-\mu_n}{b_n}\to 0$$ in probability.

Note we don't assume each $var(X_n)\le C<\infty$ and uncorrelated $X_n$.

The proof is easy by Markov inequality.

$P(|S_n-\mu_n|>\epsilon |b_n|)\le \frac{1}{\epsilon^2}\frac{E(S_n-\mu_n)^2}{b_n^2}=\frac{1}{\epsilon^2}\frac{\sigma_n^2}{b_n^2}\to 0$


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.