Essay on some identification problems in economics.

Summary This thesis presents three independent research topics, linked nevertheless by the question of the identification of economic models. The first chapter is devoted to non-parametric instrumental models. I first study the completeness condition, which is recently used to identify instrumental non-parametric regressions for instance. This essay considers an additively separable non-parametric model with a broad support condition. In this framework, different versions of the completeness condition are obtained. I then consider a new method to deal with endogenous selection, based on the independence between instruments and selection variable, and the completeness condition. In addition to the identification, an estimation method and an application are proposed. The second chapter focuses on two industrial economy models. The first test considers the non-parametric identification of the common value auction model. The identifying assumption is that the support of the distribution of signals conditional on the value of the good varies with this value. The interest of this approach is that apart from this condition, it does not rely on functional restrictions. The second essay studies the adverse selection model. It shows that in the absence of exogenous contract changes, the identification of the model requires the knowledge of at least one of the model parameters. However, in the presence of such changes, the model is partially or completely identified. An estimation method and an application are also proposed. The third chapter, finally, focuses on peer effect models. While these are considered as unidentified, a slight modification of the standard linear model allows us to find the structural parameters thanks to the variations in group size. These results are extended to a binary model of interactions.