STRUCTURAL ESTIMATION WITH UNSTRUCTURED DATA

The following information was released by the Federal Reserve Bank of Dallas:

Research Department Working Papers

No. 2620

Sara Casella, Jess Fernandez-Villaverde, Stephen Hansen, Ryohei Oishi and Minchul Shin

July 31, 2026

Abstract: Standard macroeconomic data do not cleanly separate the systematic and nonsystematic components of monetary policy. We show that incorporating unstructured text data into the structural estimation of a DSGE model can sharpen this distinction. We augment a standard state-space model with a non-core measurement block that links structural shocks to time series derived from FOMC transcripts, using a spike-and-slab prior to let the data select which series are informative. In a medium-scale New Keynesian model for the U.S., 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, implyinga flatter and less backward-looking price Phillips curve.

DOI: https://doi.org/10.24149/wp2620

Read paperhttps://www.dallasfed.org/~/media/documents/research/papers/2026/wp2620.pdf