Essays on Granularity and Machine Learning in Macroeconomics

Abstract

This dissertation consists of three independent chapters on granularity in macroeconomics and machine learning methods designed to solve the methodological challenges imposed by granular models, i.e. models with many different individual agents, firms or countries. In the first chapter, which is co-authored with Jesus Fernandez-Villaverde, Galo Nuno and George Sorg-Langhans, we develop a deep-learning algorithm to globally solve high- dimensional dynamic programming problems that result from granular macroeconomic mod- els. We evaluate our methodology in a standard neoclassical growth model and then demon- strate its power in two high-dimensional granular applications – a model of dynamic capital allocation and a model of migration and labor mobility. In the second chapter, which is co-authored with Cecile Gaubert and Oleg Itskhoki, we focus on the importance of considering a granular firm distribution for government policy in an international trade setting, highlighting three implications: (i) In antitrust regulation, governments face an incentive to be overly lenient towards domestic mergers in comparative advantage sectors. (ii) In trade policy, targeting individual foreign exporters shifts the burden of tariffs from domestic consumers towards foreign producers. (iii) In industrial policy, while generally suboptimal in a closed economy subsidizing ’national champions’ can be unilaterally welfare improving in an open economy. In the third chapter, I demonstrate the importance of considering granularity at the country level by demonstrating that economic crises drive cross-country migration. I show in an event study setting that net migration caused by the Euro crisis accounts for roughly 21% of all migration from EU countries into Germany between 2010-2019. In addition I highlight three salient facts: (i) This increase is mostly due to inflows rather than outflows. (ii) The migration response evolves gradually and achieves its maximum size only after five years. (iii) There is strong evidence for hysteresis.