Abstract: | In this paper we consider estimators that (asymptotically) admit a so called linear representation. Using a parametrization of the model, that has been defined in a previous paper 1], and a certain notion of smoothness of the parametrization, it is possible to define a concept of optimality for these estimators and to characterize the optimal estimators. In contrast with the situation in 1], only the compensator is fully parametrized by the parameter we want to estimate. Embedding the problem under consideration in the previously developed framework then requires the introduction of several nuisance parameters, that are needed to describe certain stochastic integrals with respect to the compensator of the jump measure |