Efficiency of estimators for partially specified filtered models (Q2640277)

Let \(X_{n1}(t),...,X_{nn}(t)\) be counting processes admitting a Doob- Meyer decomposition \(X_{ni}=M_{ni\nu}+A_{ni\nu}\), \(1\leq i\leq n\), in which the compensator \(A_{ni\nu}\) has a representation \[ A_{ni\nu}(t)=\int^{t}_{0}a_{ni\nu}(s)ds. \] The predictable intensity process \(a_{ni\nu}\) is assumed to depend on the unkown parameter \(\nu\) and a vector process of covariates. A Hájek-type convolution theorem is derived for regular estimates of \(k(\nu_ 0)\), where k is a smooth functional of the true parameter \(\nu_ 0\). The result is applied to obtain efficient estimates in the additive risk and proportional hazards model.