The present thesis provides a general asymptotic theory for preliminary test estimators (PTEs). PTEs are typically used when one needs to estimate a parameter having some uncertain prior information about it. In the literature, preliminary test estimation have been applied to some specific models, but no general asymptotic theory was already available to the best of our knowledge. After a study of PTEs in a multisample principal component context, we first provide a general asymptotic theory for PTEs in uniformly, locally and asymptotically normal (ULAN) models. An extensive list of statistical and econometric models are ULAN making our results quite general. Our main results are obtained using the Le Cam asymptotic theory under the assumption that the estimators involved in the PTEs admit Bahadur-type asymptotic representations. Then, we propose PTEs involving multiple tests and therefore multiple constrained estimators; we call them preliminary multiple test estimators. For the latter, we also derive a very general asymptotic theory in ULAN models. Our theoretical results are illustrated on problems involving the estimation of covariance matrices both via simulations and a real data example. / Doctorat en Sciences / info:eu-repo/semantics/nonPublished