Machine Learning Techniques in Applied Econometrics

The main focus in econometrics is to provide an explanation of various observed outcomes. Structural econometricians obtain reliable estimates of parameters that describe an economic system to provide an understanding of the underlying processes that determine equilibrium outcomes. The estimation process is based on conditions implied by economic theory. On the other hand, the main focus in machine learning is to provide accurate predictions of the variables of interest. While these techniques are extremely powerful for forecasting, it can be very hard to interpret the underlying structure implied by them. As machine learning techniques become more popular and computers become capable of storing and processing large quantities of data, there have been some recent eorts to incorporate such techniques into structural econometric models. My research aims to extend this literature. In my thesis I investigate whether it is possible to incorporate machine learning techniques in econometric models in a meaningful way. I explore two different approaches for doing so - first, I generalize the idea of regularization from machine learning to the Generalized Method of Moments framework; second, I apply pre-existing classification techniques from machine learning to the Propensity Score framework. Finally, I employ the empirical techniques developed in the second chapter to address a public policy question { that of the effectiveness of India's Safe Motherhood Scheme.