Maximum likelihood parameter estimation from incomplete data via the sensitivity equations: the continuous-time case

This paper is concerned with maximum likelihood (ML) parameter estimation of continuous-time nonlinear partially observed stochastic systems, via the expectation maximization (EM) algorithm. It is shown that the EM algorithm can be executed efficiently, provided the unnormalized conditional density of nonlinear filtering is either explicitly solvable or numerically implemented. The methodology exploits the relationships between incomplete and complete data, log-likelihood and its gradient.