Let $X_1, \cdots, X_n$ be a sample from the $N(\mu, \sigma^2)$ density, where $\mu, \sigma^2$ are unknown. I want to find a lower bound $L_n$ which is valid for all sample-sizes $n$ for the variance of unbiased estimators of $\theta = \mu/\sigma$.

I know that for one-dimensional parameters the Cramer-Rao lower bound for unbiased estimators is given by $\frac{1}{I(\theta)}$, where $I(\theta)$ is the fisher information. Since $\mu, \sigma^2$ are unknown the Fisher information becomes a $2\times 2$ matrix. So how can the Fisher information matrix help me to find the Cramer-Rao lower bound?

Any hints will be appreciated.


1 Answer 1


Suppose $\theta=(\mu,\sigma^2) \in \Omega$ is the two-dimensional parameter where $\Omega=\mathbb R\times (0,\infty)$.

You are interested in the parametric function $g:\Omega \to \mathbb R$ where $g(\theta)=\frac{\mu}{\sigma}$.

Then $g$ is differentiable and its gradient at $\theta$ is denoted by $\nabla g(\theta)$.

The Cramér-Rao lower bound (see this or this) for the variance of any unbiased estimator of $g(\theta)$ is the inverse of Fisher information about $g(\theta)$, which is given by

$$\text{CRLB}(g(\theta))=\nabla g(\theta)^T(I(\theta))^{-1}\nabla g(\theta)$$

  • $\begingroup$ This is very useful. @StubbornAtom many thanks! $\endgroup$
    – Oscar24680
    Commented Jul 28, 2023 at 23:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.