Data Sciences Institute - Data Speaker Series

Prof. Timothy Christensen, Yale University

Unstructured data in economics: Opportunities and challenges
Researchers across economics and related social sciences increasingly use machine learning and AI to generate new variables from unstructured data. These generated variables are typically used as inputs in downstream models. However, naively treating the generated variables as regular numerical data can lead to biased estimates and invalid inference. This talk discusses methods to debias estimates and restore valid inference when validation data are not available. We focus on two key economic applications: measuring “soft” variables for macroeconomic forecasting, and demand estimation for online platforms. We will also relate the approach prediction-powered inference, highlighting challenges that arise when, as is often the case in economics, complete validation datasets are unavailable.

Biography:
Prof. Christensen’s research interests lie broadly across theoretical and applied econometrics, financial econometrics, and statistics/data science. His most recent research is at the intersection of econometrics and machine learning, where he works on the integration of unstructured data into quantitative economic modelling. Before joining Yale, he was a Professor of Economics at University College London.

This talk is co-sponsored by the Data Sciences Institute and the Ontario Regional Centre of the Canadian Statistical Sciences Institute (CANSSI Ontario) , University of Toronto.

For more information, please visit https://datasciences.utoronto.ca/dsi-home/data-sciences-speaker-series/.