Essays in economic analysis using text data
OA Version
Citation
Abstract
This thesis explores the application of text data in economic analysis and demonstrates its potential to provide insights that tranditional data sources may overlook. The first essay, coauthored with Tarek Hassan, Jesse Schreger, and Ahmed Tahoun, and published in the Review of Economic Studies, uses textual analysis of earnings conference calls held by listed firms around the world to measure the amount of risk managers and investors at each firm associate with each country at each point in time. Flexibly aggregating this firm-country-quarter-level data allows us to systematically identify spikes in perceived country risk (“crises”) and document their source and pattern of transmission to foreign firms. We also apply our measures to show that elevated perceptions of a country’s riskiness, particularly those of foreign and financial firms, are associated with significant falls in local asset prices, capital outflows, and an increased likelihood of a sudden stop. The second essay, coauthored with Tarek Hassan, Stephan Hollander, Laurence van Lent, and Ahmed Tahoun, and published in the Review of Financial Studies, constructs text-based measures of the primary concerns listed firms associated with the spread of COVID-19 and other epidemic diseases. We identify which firms perceive to lose or gain from a given epidemic and textually decompose the epidemic's effect on the firm's demand and supply. We find that the effects of COVID-19 manifest as a simultaneous shock to demand and supply, with both shocks affecting firms’ market valuations in equal measure on average. By contrast, demand-related impacts appear more important in accounting for the observed collapse in firm-level investment during the COVID-19 crisis. The third essay, coauthored with Tarek Hassan, Stephan Hollander, Aakash Kalyani, Laurence van Lent, and Ahmed Tahoun, solicited by the Journal of Economic Perspectives, presents a novel strategy for economic surveillance, leveraging computational linguistics to analyze unstructured corporate texts such as earnings conference calls and firms’ job postings. We showcase the efficacy of this ‘text-as-data’ approach in extracting insights that traditional data sources may overlook, particularly in response to economic shocks. Our analysis not only provides a more nuanced understanding of market and firm-reactions, but also suggests practical implications for policymaking and corporate decision-making.
Description
2024