Tracking a random walk first-passage time through noisy observations (Q691106)

Given a Gaussian random walk (or a Wiener process), possibly with drift, observed trough noise, the problem of estimating its first-passage time \(\tau_l\) of a given level \(l\) with a stopping time \(\eta\) defined over the noisy observation process is considered. Non-asymptotic upper and lower bounds on the minimum mean absolute deviation \(\inf_{\eta}\operatorname{E}|\eta-\tau_l|\) have been derived that coincide in certain asymptotic (as \(l\to\infty\)) regimes. In these regimes the estimation error does not get smaller if \(\eta\) is an arbitrary function of the entire observation process, not necessarily a stopping time. In the particular case where there is no drift, it is shown that it is impossible to track \(\tau_l:\) \(\inf_{\eta}\operatorname{E}|\eta-\tau_l|^p=\infty\) for any \(l>0\) and \(p\geq 1/2\). In the paper, it is mentioned that extensions to other loss functions or non-Gaussian settings may be envisioned.