On the robust estimation in Poisson processes with periodic intensities (Q2640291): Difference between revisions

Let X(t), \(t\geq 0\), be a Poisson process with a parametrized intensity function \(\lambda\) (t,\(\theta\)) where \(\theta\in \Theta\) is an open interval of \(R^ 1\). The general problem is to estimate \(\theta\) on the basis of the observations (X(t), \(0\leq t\leq T)\). One of the frequently used methods is the maximum likelihood method. However, as the author indicates, this approach is not always appropriate due to different reasons. Instead, he suggests to look for M-estimators. Recall that given functions h(t,\(\theta\)) and H(t,\(\theta\)), the solution \({\hat \theta}{}_ T\) of the equation \[ C(T,\theta)=\int^{T}_{0}h(t,\theta)dX(t)- \int^{T}_{0}H(t,\theta)dt=0 \] is called the M-estimator of \(\theta\) based on the observations (X(t), \(0\leq t\leq T)\). (If \(h(t,\theta)={\dot \lambda}(t,\theta)/\lambda (t,\theta)\) and \(H(t,\theta)={\dot \lambda}(t,\theta),\) the M-estimator is exactly the MLE.) The author concentrates his attention to the detailed study of the M- estimators for a Poisson process with a periodic intensity function. First he has described conditions under which the M-estimator \({\hat \theta}{}_ T\) exists and obeys the properties of consistency and asymptotic normality as \(T\to \infty\). Further the robustness property is analyzed. Finally, a nice illustrative example is presented.