Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

In the version of central limit theorem for strictly stationary but weakly dependent (for instance $\alpha$-mixing with fast decaying mixing coefficient) random variables $X_1, X_2, \cdots$, the theorem in this Wikipedia page states (see also Billingsley 1995 Theorem 27.4):

Theorem. Suppose that $X_1, X_2, \cdots$ is stationary and $\alpha$-mixing with $\alpha_n = O(n^{−5})$ and that $\mathbb{E} X_n = 0$ and $\mathbb{E} X_n^{12} < \infty$. Denote $S_n = X_1 + \cdots + X_n$, then the limit $\sigma^2 = \lim_{n\to \infty} \mathbb{E} S_n^2/n$ exists, and if $\sigma \ne 0$ then $S_n/(\sigma \sqrt{n})$ converges in distribution to the standard Gaussian distribution $\mathcal{N}(0, 1)$.

Here $\alpha_n$ is the mixing coefficients defined as e.g. in this Wikipedia page.

My question is: Is there a proof for the necessity of $\sigma > 0$? What can we conclude (if possible) when $\sigma = 0$?

In the wikipedia page, there is a little remark trying to explain the necessity of $\sigma > 0$ (note that $\sigma^2 = \mathbb{E} X_1^2 + 2 \sum_{j=1}^\infty \mathbb{E} X_1 X_{1+j} \ge 0$). Namely, if we take $Y_0, Y_1, \cdots$ to be an i.i.d. sequence, and $X_i = Y_i - Y_{i-1}$ for all $i = 1, 2,\cdots$. Then $X_1, X_2 \cdots$ is stationary and $\alpha$-mixing with $\alpha_n = 0$ for all $n \ge 2$. However, $\sigma^2 = 0$, violating the condition $\sigma^2 > 0$. I don't think this example prohibits us from concluding: $$ \frac{S_n}{\sqrt{n}} \xrightarrow[n\to \infty]{distribution} \ \sigma \mathcal{N}(0,1). $$ In the above example though the right hand side becomes the Dirac measure at zero, the left hand side is $S_n/\sqrt{n} = (Y_n - Y_0)/\sqrt{n}$, which indeed converges to zero in probability (hence in distribution), say by applying Chebyshev and assuming $\mathbb{E} Y_0^{12} < \infty$.

share|improve this question

1 Answer 1

up vote 6 down vote accepted

How is the general case different than this example?

If $\sigma=0$ then the variance of $S_n/\sqrt{n}$ goes to 0 so $S_n/\sqrt{n} \to 0$ in distribution.

share|improve this answer
    
... and therefore $S_n / \sqrt{n} \to 0$ in probability as well, by this fact, with no further moment assumptions needed. –  Nate Eldredge Apr 10 at 7:04

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.