MLE of $\mu$ when $X_1,\ldots,X_n$ are i.i.d $N(\mu,\mu^2)$

353 Views Asked by At

The Question:

Given independent random variables $X_1,\dots,X_n \sim N(\mu,\mu^2)$ where $\mu>0$ is unknown, find the MLE $\hat \mu$ of $\mu$.


My Attempt:

Yes, I know this is a really standard question, but I got a really strange answer so I am not sure. We go through the usual drill:

$$L(\mu)=(2\pi \mu^2)^{-n/2}\exp\Bigl(\frac{-1}{2\mu^2} \sum_{i=1}^n(x_i-\mu)^2 \Bigr)$$

$$\ell (\mu)=-\frac n2 \ln(2\pi\mu^2)-\frac{1}{2\mu^2}\sum_{i=1}^n(x_i-\mu)^2 $$

\begin{align} \ \frac{d\ell}{d\mu} & = \Bigl(-\frac n\mu\Bigl)+\Bigl(\frac{1}{\mu^2}\sum_{i=1}^n(x_i-\mu)\Bigl)+\Bigl(\frac{1}{\mu^3}\sum_{i=1}^n(x_i-\mu)^2\Bigl) \\ \ & = \Bigl(-\frac n\mu \Bigl) -\Bigl(\frac n\mu +\frac{1}{\mu^2}\sum_{i=1}^nx_i \Bigl)+ \Bigl(\frac{1}{\mu^3}\sum_{i=1}^nx_i^2-\frac{2}{\mu^2}\sum_{i=1}^nx_i+\frac n\mu \Bigl) \\ \ & = -\frac n\mu - \frac{1}{\mu^2}\sum_{i=1}^nx_i+\frac{1}{\mu^3}\sum_{i=1}^nx_i^2 \end{align}

Thus

\begin{align} \ & \frac{d\ell}{d\mu}=0 \\ \ \implies & n\mu^2+\mu \sum_{i=1}^nx_i-\sum_{i=1}^nx_i^2=0 \\ \ \implies & \mu = \frac{-\sum_{i=1}^nx_i + \sqrt{\bigl(\sum_{i=1}^nx_i \bigr)^2+4n\sum_{i=1}^nx_i^2}}{2n} \end{align}

... and I feel that I have done something wrong, since there is no way I can use this disgusting thing to do the next part of the question.

Any help would be much appreciated!

1

There are 1 best solutions below

0
On

OK, so seems like there is nothing wrong with that disgusting expression for $\hat \mu$. The next part of the question is to show that the asymptotic variance is $\mu^2/3n$ which actually does not require the expression for $\hat \mu$:

\begin{align} \ \frac{d^2 \ell}{d \ell ^2} & = \frac{d}{d \ell}\Bigl(-\frac n\mu-\frac{1}{\mu^2}\sum_{i=1}^nx_i+\frac{1}{\mu^3}\sum_{i=1}^nx_i^2\Bigl) \\ \ & = \frac{n}{\mu^2}+\frac{2}{\mu^3}\sum_{i=1}^nx_i-\frac{3}{\mu^4}\sum_{i=1}^nx_i^2 \\ \end{align}

Hence, the asymptotic variance is the reciprocal of the Fisher Information, which is:

\begin{align} \ I(\mu)^{-1} & =\biggl(\Bbb E\Bigl[-\frac{d^2 \ell}{d \ell ^2}\Bigr]\biggr)^{-1} \\ \ & =\biggl(-\frac{n}{\mu^2}-\frac{2}{\mu^3}\sum_{i=1}^n\Bbb E(x_i)+\frac{3}{\mu^4}\sum_{i=1}^n\Bbb E(x_i^2) \biggr)^{-1} \\ \ & = \biggl(-\frac{n}{\mu^2}-\frac{2}{\mu^2}(n\mu)+\frac{3}{\mu^4}(2\mu^2)\biggr)^{-1} \\ \ & = \frac{\mu^2}{3n} \end{align}

noting that $\Bbb E(x_i^2)=Var(x_i)+[\Bbb E(x_i)]^2=\mu^2+\mu^2=2\mu^2$