Robust approximation in a filtering problem with real state space and counting observations (Q1573572)

Let \((X_t,Y_t)\) be a pure jump Markov process, where \(X_t\) takes values in \(\mathbb{R}\) and \(Y_t\) is a counting process. Sensitivity with respect to the model for the filter of this system is studied. For this end, the filters of two similar systems are compared. The authors compute a bound for the distance in the so-called bounded Lipschitz metric between the two filters. As an application a filter \(\pi_t^\varepsilon\) is constructed, which converges to the filter \(\pi_t\) and has discrete support. The results are based on a coupling technique. Similar results for the case where \(X_t\) takes values in a discrete state space have been given by the authors [Stochastics Stochastics Rep. 57, No. 1/2, 71-87 (1996; Zbl 0897.60046) and ibid. 61, No. 1/2, 1-19 (1997; Zbl 0880.93054)].