Text-Based Identification of Monetary Policy Shocks

This paper identifies monetary policy shocks for Sweden. Based on the written language in documents prepared by staff head of Riksbank policy meetings, I construct text-based regressors that are shown to systematically predict shocks identified with standard methods such as the Romer and Romer (2004) Greenbook approach and high-frequency identification. Defining text-based monetary policy shocks as the part of the standard shock that is uncorrelated with this rich nonstandard source of information, I study the effect of monetary policy on key monetary aggregates using local projections. The dynamic responses to my text-based policy shocks are in line with theoretical expectations for a majority of variables at both quarterly and annual frequency. Text-based Romer and Romer shocks are more in line with ’textbook’ responses compared to high-frequency based shocks. All results are robust to alternative text-analytic implementations, model specifications and lag length selection in local projections.