Three Essays on Innovation: A Text-Analysis Approach

Abstract

Innovation plays a central role in tackling modern economic challenges, from stagnant firm productivity growth to climate degradation and worsening health outcomes. Understanding how and why inventors innovate is therefore of vital public policy importance. This thesis examines innovation through patent texts, leveraging recent advances in text analysis and machine learning. These methods are transforming daily life, popular culture, and science-including the study of innovation. By integrating these techniques with rigorous economic theory, this thesis improves our ability to measure the knowledge held by inventors, teams, and firms and to understand how they produce the innovations needed to tackle tomorrow's challenges. In Chapter 1, Teams and Text: Modelling Collaboration Through Patent Documents, I introduce a novel methodology that integrates inventor teams and their patent texts into a unified framework for studying collaborative innovation. I develop a Bayesian model of Natural Language Processing that captures the scientific division of labour within teams. By combining high-dimensional patent data with a statistical model of teamwork the method developed allows me to infer each team member's contribution to a patent's knowledge content. Building on this framework, Chapter 2, Catalyst or Constraint: The Dual Role of Prior Innovation for Breakthroughs, examines how prior innovations shape a team's ability to push the innovation frontier forward. I once again apply the model from Chapter 1 to patent text data. In this case mapping inventors, teams, and research fields into a structure known as the knowledge space. I combine this with data on premature team member deaths to provide a quasi-random shock to the research potential of the team. Through a continuous treatment model, I identify how team innovations change as they pivot to more or less advanced research areas. This framework offers a flexible and tractable approach to studying the creation of new research fields, an area largely overlooked in the literature due to a lack of suitable models and data. In Chapter 3, From Shares to Machines: How Common Ownership Drives Automation, we examine three increasingly important economic phenomena: the rise of common ownership in public firms, monopsony power, and the shift toward automated production processes. This chapter is co-authored with Dennis C. Hutschenreiter, Felix Noth, Stefano Manfredonia, and Tommaso Santini. We propose a theory that greater overlap in the stockholders of local labour market competitors drives automation-related innovation. We measure automation using a classification derived from the text of each patent produced at a firm. To estimate a causal effect, we exploit exogenous increases in common ownership due to institutional investor mergers, which provide a quasi-experimental setting. Our findings confirm that when common ownership among local competitors increases, firms expand automation and reduce employment.

Date of Award	20 Jun 2025
Original language	English
Supervisor	Hannes Felix Muller (Director)