Filtering of finite-state time-nonhomogeneous Markov processes, a direct approach (Q1596349)

This paper is about a cadlag process \(\{\underline W_t,\;t\geq 0\}\), built as follows: \[ \underline W_t={U_t\brack V_t},\;t\geq 0, \] where: 1. \(U\) is a (unobservable) ``signal process'' and \(V\) is an ``observation process'', 2. \([1:p]\) and \([1:q]\) are finite intervals of integers, and \(\{U_t,\;t\geq 0\} \subseteq [1:p]\) and \(\{V_t,\;t\geq 0\}\subseteq [1:q]\), 3. \(\underline W\) is a Markov process which can be non-homogeneous. The authors of the paper want and do show how one can obtain the filtering equations for \[ P\bigl( U_t=i\mid \sigma_t(V) \bigr), \] in an elementary way, avoiding in particular recourse to stochastic calculus. The existence of infinitesimal intensities for \(\underline W\) is assumed. The derivation that is given, though elementary, is not that simple, and even the proposed heuristic arguments require close attention: the look, if not the substance, of stochastic calculus is still there! The paper contains an example which may help in a first reading. As a bonus, the last section of the paper contains some properties for the \(\underline W\) process.