Suppose we have multivariables ${\boldsymbol {\theta }}=\left[\theta _{1},\theta _{2},\dots ,\theta _{d}\right]^{T}\in {\mathbb {R}}^{d}$, and we want to estimate the function of parameters $\boldsymbol \phi(\boldsymbol {\theta })$ with unbiased estimator $\boldsymbol T(X)$, CRB states that we have a lower bound as: $${\displaystyle \operatorname {cov} _{\boldsymbol {\theta }}\left({\boldsymbol {T}}(X)\right)\geq \boldsymbol\phi (\theta )^TI\left({\boldsymbol {\theta }}\right)^{-1}\boldsymbol\phi (\theta )}$$ where $I(\boldsymbol \theta)$ is the fisher information matrix. I wonder what's the attainability of this bound, I find the bound in Wikipedia link but I can't find the attainability in the wekipedia link. Is this bound always attainable?

Thanks in advance!


1 Answer 1


Under some conditions, the Cramer-Rao bound is attained if and only if we deal with an exponential family.

See e.g.

Lehmann, E. L. (1983). Theory of Point Estimation. Springer-Verlag, New York. p.123


Shao, Jun. Mathematical statistics. Springer Science & Business Media, 2003. p.171

  • $\begingroup$ Thank you very much, but why I see in Wikipedia that the bound is always attainable by maximum likelihood estimator asymptotically? $\endgroup$
    – narip
    Commented Oct 16, 2022 at 10:54
  • 1
    $\begingroup$ @narip The statement in Wikipedia is about asymptotics, $n\to\infty$, but the attainability theorem I cited above is for finite $n$. $\endgroup$
    – frank
    Commented Oct 16, 2022 at 11:27

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.