Essays on the Use of Unstructured Data in Macroeconomics

Abstract

Economists need data to find patterns and test theories, but many concepts are hard to measure. This work contains three chapters showing how new unstructured data sources, such as text corpora, can be used to overcome this data scarcity. First, I analyze the effects of automation on employment by training a machine learning algorithm to classify patents as automation. Second, I create a new indicator of financial stress from historical newspapers. Third, I analyze whether errors in expectation formation might explain credit cycles, for which I build on a world-wide survey of banks and research institutes.