Information geometry of estimating functions in semi-parametric statistical models (Q1361123)

A semi-parametric statistical model treats a family of probability distributions \(\{p(x,\theta, k)\}\) specified by a finite-dimensional parameter \(\theta\) of interest and an infinite-dimensional nuisance parameter \(k.\) The present paper aims to elucidate the differential geometrical structure underlying the semi-parametric model from the point of view of the dual differential geometry of statistical inference and its extension to Hilbert fibre bundles. The paper studies the following problems: when does an estimating function exist and what is the set of all the estimating functions; what is the best estimator derived from estimating functions; when does the best estimating function give an efficient estimator; in the case where the best one is not efficient, what is the amount of loss of information caused by using the estimation function method; how to choose a good (quasi-) estimating function based on the observed data. The concept of \(m\)-curvature freeness plays a fundamental role in solving the above problems.