Tests non paramétriques optimaux pour le modéle autorégressif d'ordre un

This paper proposes nonparametric inference methods for the first-order autoregressive process. The problem studied is to test any hypothesis stating that the autocorrelation coefficient has a given value . We consider the family of tests based on rank autocorrelations from the model transformed under. Within this class, we derive asymptotically optimal tests against local alternatives, and compute their asymptotic relative efficiencies with respect to several available procedures (parametric and non-parametric).