Algorithms, Data, and Computing Power: Development and Application from an Economics Perspective

Abstract

This dissertation comprises a collection of research articles exploring various aspects of artificial intelligence (AI) from the perspective of economics. With recent advances in computer algorithms, data accessibility, and computing power, machine learning and deep learning techniques have found extensive applications in both industrial and academic activities. The five articles in this dissertation explore this duality. On the one hand, the diffusion of AI and ICT-related knowledge and its broader economic implications are examined. On the other, it explores novel approaches to indicator development in economics and the broader social sciences, using big data and machine learning techniques. Chapter 1 estimates the causal effect of local academic research on industrial technological development by exploiting natural variation in the regional onset of academic activity. Using text-based methods, I combine public research output with corporate patent portfolios into a long-run panel. Difference-in-difference estimators reveal robust positive effects of local scientific research on industrial activity, which diminish with distance. However, network analysis of science-to-industry knowledge flows highlight the importance of remote connections between top performing scientific and industrial hubs. The findings offer insights into the dynamics of knowledge flows, spillovers, and distance, as well as the technological inequality between regions. Chapter 2, a collaboration with Samad Sarferaz and Martin Wörter, studies international dynamics in the development of digital technologies and their effect on local economic performance. We proposes a structural dynamic factor model to decompose trends in digital innovations across countries and technological fields into global and country-specific components, using patent data. Our findings confirm the existence of local and international digital technology cycles. Using the resulting indicators, we estimate the impact of international and country-specific development cycles on local economic performance and technological specialisation. Our empirical results show that international technology dynamics have a significant impact on the performance of a country's digital industries. Specifically, we find that countries benefit from international efforts to develop digital technologies if their ICT sector encompasses a broad range of digital technologies and is not specialised in only specific technologies. Chapter 3, a collaboration with Martina Jakob, evaluates the usage of social media data to generate indicators for economic studies. In response to persistent gaps in the availability of survey data, a new strand of research leverages alternative data sources through machine learning to track global development. While previous applications have been successful at predicting outcomes such as wealth, poverty or population density, we show that educational outcomes can be accurately estimated using geo-coded Twitter data and machine learning. Based on various input features, including user and tweet characteristics, topics, spelling mistakes, and network indicators, we can account for ~70 percent of the variation in educational attainment in Mexican municipalities and US counties. Chapter 4 investigates the potential of indicators derived from corporate websites to measure technology related concepts. Using artificial intelligence (AI) technology as a case in point, I construct a 24-year panel combining the texts of websites and patent portfolios for over 1,000 large companies. By identifying AI exposure with a comprehensive keyword set, I show that website and patent data are strongly related, suggesting that corporate websites constitute a promising data source to trace AI technologies. Chapter 5, a collaboration with Florian Seliger, and Martin Wörter, presents a database that classifies all patent applications filed at either the United States Patent and Trademark Office (USPTO) or the European Patent Office (EPO) as being either product patents, process patents or 'mixed patents'. We use the share of claims that refer to either product or process inventions which allows to classify all patent applications along a continuum of pure process patents and pure product patents. We find that process-oriented patents draw more on previous knowledge, are more original and more radical than product patents. Lower breadth of protection is positively associated with pure process patenting, whereas product and mixed variants can be protected more broadly. This characterisation uncovers heterogeneity of patented inventions that allows for a more sophisticated use of patent statistics. It can improve the accuracy of analyses, but also reveal new aspects related to property rights.