Say we have $X \sim N(0,1)$. I was wondering how we can use Stein's Lemma to derive the moments of the r.v. $X$ by calculating the first few moments.

How would you summarize it in the form $EX^n$ if the moments we calculated end up being quite different in form? Or are they?

So what I've tried to do so far is that I used the formula:

$E\bigl(g(X)(X-\mu)\bigr)=\sigma^2 E\bigl(g'(X)\bigr)$

And from there, I tried to see if I can make any substitutions but I don't think you can. Furthermore, I really don't see how you can even calculate moments from this thing. From class, I think you have to use Taylor expansion but I really don't see how you can get to that here either. I really need some help on this.


Let $X \sim N(\mu,\sigma^2)$.

  1. Take $g := 1$. Then Stein's lemma gives $$\mathbb{E}(1 \cdot (X-\mu)) = 0, \tag{1}$$ i.e. $\mathbb{E}(X) = \mu$.
  2. For $g(x):= (x-\mu)^{n-1}$, $n \geq 2$, we have, by Stein's lemma, $$\mathbb{E}((X-\mu)^n) = \sigma^2 (n-1) \mathbb{E}((X-\mu)^{n-2}).$$ If $n$ is odd, i.e. $n=2k+1$, then we get by iteration $$\begin{align*} \mathbb{E}((X-\mu)^n) &= (\sigma^2 (n-1)) (\sigma^2 (n-3)) \cdots (\sigma^2 \mathbb{E}(X-\mu)) \\ &\stackrel{(1)}{=} 0. \end{align*}$$ On the other hand, if $n=2k$ is even, then $$\begin{align*} \mathbb{E}((X-\mu)^n) &= (\sigma^2 (n-1)) (\sigma^2 (n-3)) \cdots (\sigma^2 \cdot 1) \\ &= (\sigma^2)^{\frac{n}{2}} (n-1) (n-3) (n-5) \cdots 1. \end{align*}$$
  3. If we are not interested in the centered moments $\mathbb{E}((X-\mu)^n)$, but in $\mathbb{E}(X^n)$, then we can use the identity $$X^n = ((X-\mu)+\mu)^n = \sum_{k=0}^n {n \choose k} (X-\mu)^k \mu^{n-k}$$ to express $\mathbb{E}(X^n)$ in terms of $\mu$ and $\sigma^2$ using the results from step 2.

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.