Stephen Hansen

Plugging measures derived from unstructured data into regression models leads to biased inference. Our proposed solution is to jointly model informational retrieval and regression, which modern computational methods make possible.

Inflation uncertainty leads to more hawkish policy stances on the Federal Open Market Committee. This appears to arise from concerns for tail risk than model parameter uncertainty.

Using novel daily measures of consumption, firm sales, and employment, we show that the economy reacts within days to monetary policy shocks. Time aggregation of the measures to quarterly frequency masks these short-run dynamics.

We use a large language model trained on tens of thousands of human labels to measure remote and hybrid work adoption with unprecedented granularity. We document large heterogeneity in adoption across narrow occupation categories, cities, and firms.

We show how observed connections among variables in multiple networks can improve estimation of the precision matrix by targeting regularization and develop a Bayesian model for inferring pairwise correlations.

We use comprehensive financial transactions from one of Europe's largest banks to build a large-scale consumption survey using national accounting principles. We use this to build aggregate and distributional accounts for consumption, as well as a detailed individual consumption panel.

We measure the demand for executive skills aross firms using a unique, large-scale corpus of job specifications. The importance of social skills is growing over time and is related to firm size, firm scope, and the information intensity of worker skills.

We combine elements of supervised machine learning and dictionary methods to uncover narrow, interpretable risk exposures from 10-K filings that account for firm-level reactions to aggregate shocks. We apply this idea to study equity returns during the COVID-19 pandemic.