Travail de recherche/Working paper
| Résumé : | Locally asymptotically optimal (in the Hajek-Le Cam sense) pseudo-Gaussian and rank-based procedures for detecting randomness of coefficients in linear regression models are proposed. The parametric and semiparametric efficiency properties of those procedures are investigated. Their asymptotic relative efficiencies (with respect to the pseudo-Gaussian procedure) turns out to be be huge under heavy-tailed and skewed densities, stressing the importance of an adequate choice of scores. Simulations demonstrate the excellent finite-sample performances of a class of rank-based procedures based on data-driven scores. |
