Twitxy Saved

Sara Casella

@SaraCasella9

AP @UniLUISS and Research Affiliate @EIEF_Rome (2024-). PhD from UPenn @penneconomics. Macroeconomics 👩🏻‍💻
saracasella.com Joined April 2019
893 Following 987 Followers 103 Posts

Latest posts 5

Sara Casella @SaraCasella9
Very proud to see this draft out. Working on this project and learning from Jesús, Stephen, Ryohei, and Minchul has been a real privilege. Comments and thoughts very welcome!
Jesús Fernández-Villaverde Verified account @JesusFerna7026 Jul 9
I am extremely happy to circulate a new paper, “Structural Estimation with Unstructured Data,” joint with Sara Casella (@SaraCasella9), Stephen Hansen, Ryohei Oishi, and Minchul Shin:

sas.upenn.edu/~jesusfv/Text_…

This has been a long time in the making: I presented a first set of slides (no draft!) at the ASSA meetings in San Diego in 2020. But COVID and life in general slowed down progress. Fortunately, we finally pushed the paper across the (working paper) finish line over the last few weeks.

The idea, I believe, is very cool, and I am proud of how we implemented it.

Can we use unstructured data (e.g., text, images, sound, ...) to estimate dynamic equilibrium models and their structural parameters? The answer is yes.

More concretely, we augment the standard state-space representation of a dynamic equilibrium model with a noncore measurement block that links structural shocks to time series derived from unstructured data, and then use standard filtering and McMc to draw from the posterior. We rely on a spike-and-slab prior to let the data select which series are informative.

To illustrate how well this works, we take a medium-scale New Keynesian model for the U.S. and estimate it using conventional data (i.e., output, inflation) together with FOMC transcripts (text).

We document how incorporating text improves predictive performance and materially alters structural inference: the new model estimates a lower response of the policy rate to inflation, higher price stickiness, and lower price indexation, implying a flatter and less backward-looking price Phillips curve.

The intuition is simple: the FOMC transcripts tell us a great deal about the Fed’s true policy function because they reflect the expectations and preferences of policymakers, even if only in verbal terms. The likelihood function appreciates such information.

The idea generalizes to all sorts of dynamic equilibrium models. For example, one could use Amazon reviews alongside standard data on prices and quantities to estimate structural models in IO. Or one could use images to expand the information set of DSGE models. The possibilities of the methodology are endless.

As always, comments are much appreciated.
View quoted post on X
1 replies 8 reposts 61 likes 7.9K views View on X
Sara Casella @SaraCasella9
Such great group of people and JMPs! Looking forward to the roman leg of the tour!!
📢📢We now know the Tourists and Hosts for the 2026 REStud European Tour!📢📢

The REStud European Tour recognizes the most promising graduating doctoral students in economics and finance and introduces them and their research to audiences in Europe.

Let's Meet the Hosts and the Tourists in this thread!

1/16
View quoted post on X