Optimal Detection of a Change Point in a Poisson Process for Different Observation Schemes (Q4677100)

Detection of a change of a Poisson process is considered in three cases: sequential observations, ex post analysis, and a combined case. In the third case, the process is observed up to a fixed point \(t^*\) and at this point \(t^*\) it is decided whether the change has already taken place and when. If it is decided that there were no change, then the process is observed from \(t^*\) on. The optimal stopping rule maximizes the expectation of the gain \[ Z_t=c_0(t\wedge\sigma)+c_1(t-\sigma)^{+}-k_0(1-Y_t)-k_1 Y_t, \] where \(\sigma\) is the true change-point, \(t\) is the stopping time, \(Y_t={\mathbf 1}\{\sigma\leq t\}\), and \(c_i\), \(k_i\) are some fixed constants. The authors derive a smooth semimartingale decomposition of \(Z\) and construct infinitesimal-look-ahead stopping rules for these three cases.