joint avec le Séminaire de Statistique et d’Econométrie
Jean-Pierre Florens (Toulouse School of Economics IDEI and GREMAQ), Anna Simoni (Universita Bocconi IGIER)
Résumé
The problem of identification and partial identification in econometrics is considered from a Bayesian point of view. We carry out a comprehensive analysis of this topic and we also stress the links with classical identification. We extend results about identification by the prior distribution to infinite dimensional parameters that lack identification in the sampling model. For the partially identified case, we provide nonparametric Bayes estimators of the identified set based on Dirichlet process priors. Moreover, we show how parameters that are non identified or partially identified in the complete model can be identified in the marginal model, where the marginalization is done with respect to the prior distribution of the identified parameter. Finally, the paper provides several examples that show how to implement in practise our techniques.