Essays on Machine Learning Methods and Household Dynamics

Abstract

This dissertation consists of three independent chapters on questions surrounding household dynamics and machine learning methods developed to address them. In the first chapter, which is co-authored with Jesus Fernandez-Villaverde, Galo Nuno, and Maximilian Vogler, we develop a deep-learning algorithm to globally solve high-dimensional dynamic programming problems that arise in macroeconomic models. This approach allows us to address household dynamics in a rich environment driven by both aggregate and idiosyncratic uncertainty. We evaluate our methodology in a standard neoclassical growth model and then demonstrate its power in two high-dimensional applications -- a model of dynamic capital allocation and a model of migration and labor mobility. In the second chapter I propose a new machine learning approach to understanding consumption insurance of households, a central issue in the context of household dynamics. I draw on a state-of-the-art machine learning method, gradient boosted trees, to predict consumption in the Panel Study of Income Dynamics. With the resulting panel data set in hand, I adopt Blundell, Pistaferri, and Preston's (2008) assumptions about the underlying permanent-transitory income process, which allows me to estimate insurance coefficients. Importantly, I find qualitative and quantitative differences to their insurance predictions. In the third chapter, which is co-authored with Riccardo Cioffi, and Maximilian Vogler, we investigate how different theories of wealth inequality interact with important policy experiments. To this end, we calibrate four models, each incorporating a different inequality-generating channel emphasized in the theoretical literature, and compare their predictions regarding different policy experiments. We find stark quantitative and qualitative differences in predictions across channels for a given policy experiment, indicating that analyzing their relative importance is crucial to our understanding of wealth inequality.