Convergence in distribution and in moment... and in central moments?

1.1k Views Asked by At

In Fumio Hayashi's Econometrics, Lemma 2.1 (convergence in distribution and in moments) states that

Lemma 2.1 $\quad$ Let $\alpha_{sn}$ be the $s$-th moment of $z_n$ and $\lim_{n\to\infty}\alpha_{sn}=\alpha_s$ where $\alpha_s$ is finite. Then $$ z_n \to_d z \implies \alpha_s \text{ is the $s$-th moment of $z$}$$

Then the author goes on to comment that

For example, if the variance of a sequence of random variables converging in distribution converges to some finite number, then that number is the variance of the limiting distribution.

My question is, Why is this comment true? I don't think this lemma about moments can be directly applied to the variance, which is a central moment. If this is not a direct application, it remains unclear to me how this lemma can lead to the author's claim about the variance.

Call the sequence of our concern $y_n\to_d y$. To apply the lemma to the variance, we need to convert $y_n$ into $z_n:=y_n-\mathbb{E}[y_n]$, so that the convergence $\operatorname{Var}[y_n]\to\alpha$ for some $\alpha\in\mathbb{R}$ is translated into $\mathbb{E}[(z_n)^2] \to \alpha$. The obstacle to apply the lemma is, we do not know $z_n:=y_n-\mathbb{E}[y_n]$ converges in distribution to $y-\mathbb{E}[y]$ or not (for one thing, $\{\mathbb{E}[y_n]\}$ is not known to converge). If this were known, then by the lemma $\mathbb{E}[(y-\mathbb{E}[y])^2]=\alpha$, as desired.

1

There are 1 best solutions below

0
On BEST ANSWER

As pointed out in the comments to the question, both Lemma 2.1 and Hayashi's comment are incorrect. The issue with the Lemma is in Hayashi's errata.

Counterexample to Lemma 2.1: Let $z_n$ be such that $\mathbb{P}\{z_n=n\} = \frac{1}{n}$ and $\mathbb{P}\{z_n=0\} = \frac{n-1}{n}$. Then $\mathbb{E}[z_n] = 1$. $z_n \to_d 0 =: z$. Clearly, $1=\lim_{n\to\infty}\mathbb{E}[z_n] \ne \mathbb{E}[z]=0.$

Counterexample to Hayashi's comment: Let $z_n$ be such that $\mathbb{P}\{z_n = \sqrt{n}\} = \mathbb{P}\{z_n = -\sqrt{n}\} = \frac{1}{2n}$ and $\mathbb{P}\{z_n = 0\} = 1-\frac{1}{n}.$ Then $\operatorname{Var}(z_n) = 1 \to 1$ as $n\to\infty.$ However, $z_n\to_d 0 =: z,$ so $\operatorname{Var}(z) = 0 \ne 1.$