Abstract:
We suppose that we observe a process $y(t)$ when $t\in [-T,T]$,
$$
y(t)\;=\;s(t)\;+\;x(t) \qquad (t \in [-T,T]),
$$
where $s$ is an unknown function (which we must estimate), $x$ is a stationary noise. We compare the accuracy of the least-squares estimator $\bold s^*$ with the accuracy of the best linear unbiased estimator $\bold s^{\star}$.