Geodesic estimation in elliptical distributions (Q1372215)

We investigate the class of multivariate elliptical distributions. We find that a change of coordinates reduces the system of geodesic equations to a very simple form admitting an appealing solution that is nothing but a straight line. The distance between two elliptical distributions with equal location and different scatter matrices is then calculated. This geodesic distance which is based on the information metric, comes as an addition to the list of geodesic distances given by \textit{C. R. Rao} [Diff. Geom. Stat. Inference, 217-237 (1987)]; see also \textit{J. Burbea} and \textit{C. R. Rao} [Probab. Math. Stat. 3, 241-258 (1982; Zbl 0562.62003)]. We then derive a geodesic discrepancy function for use in covariance structure analysis. The estimator of the structure parameter is shown to have desirable properties. A test statistic is then built upon this discrepancy function and shown to be asymptotically distributed as \(\chi^2\).