Asymptotic efficiency in semi-parametric models with censoring (Q1083159)

The paper considers models with censoring of the following form: \[ Y=X\beta_ 0+U,\quad W=1\;if\;X\gamma_ 0-V\geq 0,\quad W=0\quad otherwise. \] The error (U,V) is independent of the vector X. \(Z=(W,YW,X)\) is observable. It deals with the case of nonnormality of g, the density of (U,V). More precisely, g is restricted to lie in a set \(\Gamma\). Since standard theory delivers (in the parametric case g is bivariate normal) an information matrix bound on the asymptotic variance of a consistent estimator of \(\theta_ 0=(\beta_ 0,\gamma_ 0)\), the author copes here with the semi-parametric case minimizing the partial information over all ''paths'' of \(\Gamma\) satisfying a smoothness condition. He also considers relaxing the assumption that X is independent of the error distribution.