Transparency and Deliberation Within the FOMC: A Computational Linguistics Approach*

Abstract

How does transparency, a key feature of central bank design, affect the deliberation of monetary policymakers? We exploit a natural experiment in the Federal Open Market Committee in 1993 together with computational linguistic models (particularly Latent Dirichlet Allocation) to measure the effect of increased transparency on debate. Commentators have hypothesized both a beneficial discipline effect and a detrimental conformity effect. A difference-in-differences approach inspired by the career concerns literature uncovers evidence for both effects. However, the net effect of increased transparency appears to be a more informative deliberation process.Keywords: Monetary policy, deliberation, FOMC, transparency, career concerns JEL Codes: E52, E58, D78 * We would like to thank Francesco Amodio, Andrew Bailey, Francesco Caselli, Gilat Levy, Rick Mishkin, Emi Nakamura, Tommaso Nannicini, Bryan Pardo, Cheryl Schonhardt-Bailey, Jón Steinsson, Dave Stockton, Thomas Wood, and Janet Yellen for insightful discussions. We are particularly grateful to Omiros Papaspiliopoulos for numerous helpful discussions on MCMC estimation and Refet S Gürkaynak for sharing the monetary policy surprise data. We have also benefited from comments and suggestions by seminar attendees at the San Francisco Federal Reserve, University of Warwick, University of Manchester, INSEAD, Bank of England, LSE, the New York Federal Reserve, Columbia University, the ESRI, Universitat Pompeu Fabra and the CEP conference. We thank Eric Hardy for excellent research assistance in gathering biographical data, and the Bank of England's Research Donations Committee for seed corn financial support. Any errors remain ours alone.

Shocking language: Understanding the macroeconomic effects of central bank communication

“…In this subsection we outline the basic steps and intuition for the algorithm. Hansen, McMahon, and Prat (2014) provide a full description along with the statistical foundations.…”

“…We believe that the approach of using computational linguistics to create measures of communication from large databases of text has broader applications beyond monetary policy analysis and can help bringing economics into the increasingly important world of "Big Data". Existing work using computational linguistics tools to analyse monetary policy data include Bailey and Schonhardt-Bailey (2008) and Schonhardt-Bailey (2013) who focus on arguments and persuasive strategies adopted by policymakers; Fligstein, Brundage, and Schultz (2014) who apply LDA to the FOMC transcripts in order to examine the concept of "sense-making" on the FOMC; Acosta (2015) looks at how the FOMC responded to calls for greater transparency; and our own recent work examining the effect of transparency on the deliberation of the FOMC using LDA applied to FOMC transcripts (Hansen, McMahon, and Prat 2014). Hendry and Madeley (2010) and Hendry (2012) are closely related papers focusing on Canada.…”

Forward or Backward Looking? The Economic Discourse and the Observed Reality

“…Larsen and Thorsrud examine the impact of "news" on the business cycle, based on a Norwegian business newspaper followed over a period of 9000 days. In contrast to our analysis, both articles consider a rather short 3 time span and, in the case of Hansen Hansen, McMahon, and Prat (2014) are based on a narrowly defined text corpus.…”

“…However, establishing a link between topic weights and real data appears to be a novel contribution. To the best of our knowledge, only the recent contributions by Hansen, McMahon, and Prat (2014) and Larsen and Thorsrud (2015) follow a similar approach. Hansen, McMahon, and Prat analyze the impact of increased transparency on the functioning of central banks and ultimately on monetary policy using the minutes and transcripts of the Federal Open Market Committee.…”

A Structural Topic Model of the Features and the Cultural Origins of the Baconian Program

“…Topic modeling has become part of the economists' toolkit only very recently. The Latent Dirichlet Allocation (LDA) model (Blei et al 2003, Blei 2012 in particular has been applied by scholars to understand, or develop new quantitative measures of, monetary policymaking (see, e.g., Fligstein et al 2014, Hansen and McMahon 2016, Hansen et al 2015, financial market performance (see, e.g., Larsen andThorsrud 2015, Huang et al 2016), and corporate behavior (e.g., Bandiera et al 2016, Bellstam et al 2016. We move beyond the application of LDA, and use the Structural Topic Model (Roberts et al 2014(Roberts et al , 2016a, a novel topic-modeling framework that incorporates document-level covariate information into the analysis.…”