0
$\begingroup$

Let $\displaystyle X_1,X_2,...X_n$ be identically distributed such that $\displaystyle Pr\{a \leq X_i\leq b\}=1$ for bounded constants $\displaystyle a,b$. Further Let $\displaystyle cov(X_i,X_j)=\alpha$, $\displaystyle E(X_i^2)=\mu_2<\infty$. What do we then know about the distribution of $$\frac{1}{\delta n}\sum_{i=1}^n(X_i-EX_i)$$ Is there a proper $\delta$ that we could choose to have some "known" (or "anything close to known") asymptotic distribution?

$\endgroup$
1
$\begingroup$

Every square integrable centered distribution can be a limit.

To see this, assume that, for every $i$, $$X_i=B_iZ+(1-B_i)Y_i,$$ where $Z$ is centered, $(Y_i)$ is i.i.d. and centered, $(B_i)$ is i.i.d. Bernoulli with $P(B_i=1)=p$ and $P(B_i=0)=1-p$ for some $p$ in $(0,1)$, and $Z$, $(Y_i)$ and $(B_i)$ are independent.

Then, for every $i\ne j$, $$E(X_i)=0,\qquad\mathrm{var}(X_i^2)=pE(Z^2)+(1-p)E(Y^2),\qquad\mathrm{Cov}(X_i,X_j)=p^2E(Z^2).$$ Furthermore, $$S_n=\frac1n\sum_{i=1}^n(X_i-E(X_i))=R_nZ+T_n,\qquad R_n=\frac1n\sum_{i=1}^nB_i,\quad T_n=\frac1n\sum_{i=1}^n(1-B_i)Y_i.$$ The usual strong law of large numbers applies to $(R_n)$ and $(T_n)$ and shows that $R_n\to p$ and $T_n\to0$ almost surely, hence $$S_n\to pZ\quad\text{ almost surely}.$$ Another example is to pick some $Z$ and $(Y_i)$ with $(Y_i)$ i.i.d. and independent of $Z$ such that $E(Z)=0$ and $E(Y_i)=1$, and to consider $$X_i=Y_iZ.$$ Then $E(X_k)=0$, $E(X_k^2)=E(Y_1^2)E(Z^2)$, $\mathrm{Cov}(X_i,X_j)=E(Z^2)$, and $$S_n\to Z\quad\text{ almost surely}.$$ Edit: More generally (and this might answer a comment), assume that $X_i=u(Y_i,Z)$ for some $Z$ and $(Y_i)$ with $(Y_i)$ i.i.d. and independent of $Z$, then $$\frac1n\sum_{i=1}^nX_i\to v(Z)\quad\text{almost surely},$$ where the function $v$ is defined as $$v(z)=E(u(Y_1,z)).$$ Furthermore, for every $i\ne j$, $\mathrm{Cov}(X_i,X_j)=E(v(Z)^2)$.

$\endgroup$
  • $\begingroup$ Did: These are very nice examples. But I had some other sequence in mind. When establishing a CLT, in general, "X_i"'s admitting a suitable decomposition, as your examples clearly show, simplifies the math. But sometimes it is not so clear how to do this. For example let $X_i=\frac{aY_i+bZ}{cY_i+dZ}$, where $Y_i$'s are iid, and we know that moments of $X_i$ exist. Then what would think? I appreciated your hints. $\endgroup$ – Math-fun Dec 4 '14 at 11:49
  • $\begingroup$ See Edit. $ $ $ $ $\endgroup$ – Did Dec 4 '14 at 21:49
  • $\begingroup$ its been quite some time since you kindly helped me with this result, could you please explain how to establish the general result? A reference would be most helpful. Many thanks again. $\endgroup$ – Math-fun Jun 2 '15 at 12:37

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.