Essays on identification and estimation of structural parametric and semiparametric models in microeconomics

• Main Author: Martinez-Sanchis, Elena
• Published: University College London (University of London) 2005
• Subjects: 338.544
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.422278

This thesis focuses on identification and estimation of structural parametric and semi-parametric models in microeconometrics. The analysis of the conditions under which in the context of an econometric model-data can be informative about the parameters of interest of an economic process is essential and must be of high priority in any econometric work. When considering models with which to identify interesting features, emphasis should be placed on imposing the minimum set of restrictions in order to achieve identification, since inappropriate restrictions may lead to inconsistent estimates of the parameters of interest. For this reason in the literature one finds that some attention has been paid to relaxing parametric distributional assumptions on the unobservables or functional forms of the relationships between observables and unobservables. To begin with, I examine how the parameters of interest of a general class of models can be identified and then estimated when not all of the relevant variables are jointly observed in the same dataset. To do so, the existence of an additional data set with information on both the missing variables and on some common variables in the original data set is necessary. I then move on to an analysis of the identification of the preference parameters in a discrete choice demand model in which individuals only derive utility from the characteristics of the goods they consume. I discuss how this particular model makes the estimation of these parameters feasible without imposing distributional assumptions in the errors even if the number of goods in the choice set is very large. Finally, I consider the comparison of nonparametric regression curves between different samples. I propose to estimate the parameters that explain these differences between the conditional mean functions by using an estimator developed in the semiparametric literature which avoids the computational problems faced by the previously proposed estimators.