Identification and Inference in Structural Models

Massachusetts Institute Of Technology, Cambridge MA

Investigators

Abstract

Structural estimation is vital in empirical economics. The need for structural estimation arises from confounding factors due to individual choice making and market forces. A classical application is estimation of the effect of tax changes in a particular market, where structural estimation can be used to separate supply and demand factors. This research will study how to do structural estimation without functional form assumptions, which will help avoid misleading inferences that do occur in applications. A menu of models and methods will be developed. The research will also develop statistical tools to help in selecting among the methods. These tools will evaluate the performance of the different methods using higher-order approximations, and help determine which method is best. Previous work shows that avoiding functional form assumptions in structural modeling can lead to more accurate inferences in important applications such as evaluating the effect of tax changes on labor supply or consumer well being. This research will add significantly to our ability to do structural estimation by developing methods for some of the most important models. Also, the study will show that certain of the methods have particularly good statistical properties. This research will show that among methods that are currently widely used for estimating models in both microeconomics and macroeconomics the one called empirical likelihood has particularly attractive properties. These results suggest that using this particular method may lead to more accurate inference in empirical research.