Central limit theorem: where is the martingale in this proof?

111 Views Asked by At

Yet another question from the depths of Durrett. Again in the proof of Theorem 8.8.3, the author notes that "by the orthogonality of martingale increments," $$ E \left( \sum_{m=1}^{[nt]} t_{n,m} - E(t_{n,m} \mid \mathcal{F}_{n,m-1}) \right)^2 = E \sum_{m=1}^{[nt]} \big[ t_{n,m}-E(t_{n,m}\mid\mathcal{F}_{n,m-1}) \big]^2 $$ I understand the orthogonality concept, but... where is the martingale? At first I tried making a martingale out of $Y_m = t_{n,m} - E(t_{n,m} \mid \mathcal{F}_{n,m-1})$, but it doesn't seem to work out.

I suspect this has to do with his observation that $$E(t_{n,m} \mid \mathcal{F}_{n,m-1}) = E(X_{n,m}^2 \mid \mathcal{F}_{n,m-1})$$ since $X_{n,m}$ is a martingale difference array, but it's not clear how this alone creates a martingale in terms of the $t_{n,m}$.

1

There are 1 best solutions below

0
On

Define $Y_{n,m}=\sum_{k=1}^m\left[t_{n,k}-\mathbb{E}(t_{n,k}\mid\mathcal{F}_{n,k-1})\right]$. $Y_{n,m}$ is a MTG because

$$\mathbb{E}[Y_{n,m}\mid\mathcal{F}_{n,m-1}]=\sum_{k=1}^{m-1}\left[t_{n,k}-\mathbb{E}(t_{n,k}\mid\mathcal{F}_{n,k-1})\right]=Y_{n,m-1}$$

and

$$t_{n,m}-\mathbb{E}(t_{n,m}\mid\mathcal{F}_{n,m-1})=Y_{n,m}-Y_{n,m-1}$$

By Theorem 5.4.6 (Durrett) for $l\ne m$

$$\mathbb{E}[Y_{n,m}-Y_{n,m-1}][Y_{n,l}-Y_{n,l-1}]=0$$

so that

$$\mathbb{E}\left(\sum_{m=1}^{[nt]} t_{n,m}-E(t_{n,m} \mid \mathcal{F}_{n,m-1}) \right)^2=\mathbb{E}\left(\sum_{m=1}^{[nt]}[Y_{n,m}-Y_{n,m-1}] \right)^2$$ $$\mathbb{E}\sum_{m=1}^{[nt]}\left[Y_{n,m}-Y_{n,m-1} \right]^2+\sum_{m=1}^{[nt]}\sum_{l\ne m}\mathbb{E}\left[Y_{n,m}-Y_{n,m-1} \right]\left[Y_{n,l}-Y_{n,l-1} \right]$$ $$=\mathbb{E}\sum_{m=1}^{[nt]} \big[ t_{n,m}-E(t_{n,m}\mid\mathcal{F}_{n,m-1}) \big]^2$$