Solution
= Solution
If $a>1$, compare with
$$
\widetilde\theta=(2-a)\overline X_n-b.
$$
Its bias is the negative of the original bias, so the squared-bias terms agree, while
$$
(2-a)^2<a^2.
$$
Its risk is therefore strictly smaller for every $\theta$. Thus
$\widehat\theta_{a,b}$ is not an <admissible estimator>.