Policy preference at central banks: Quantifying monetary policy signals using keyword topic models

Abstract

This work analyzes central banking information flow and proposes a novel strategy to estimate individual-level policy preferences, toward monetary policy objectives, by quantifying the narratives within its different communication channels for the case of the United States. While most of the literature related to central banking corpora used unsupervised topic models to quantify narrative signals, we propose a semi-supervised, keyword-based approach built upon groups of words linked to monetary objectives in order to have a coherent, dynamic estimation of topic prevalence, whose scores could determine individual policy preferences for inflation and unemployment rates. The corpus of Federal Reserve governors' speeches (1996-2020) identified three non-keyword topics matching financial stability, financial innovation and the banking regulation, whose dynamics follow the Chairman's tenure, considered as informative policy signals toward financial markets. Governors' preferences toward monetary policy objectives were better estimated using FOMC transcripts (1994-2016) whose narratives strictly match monetary policy practices and help ranking members on a partisanship scale, with a spectrum linked to the members' educational background. Though released with a five-year delay, the FOMC transcripts, as a proxy of internal communicaton, offer a better picture of the partisanship prevailing within monetary policy committees in the United States, that cannot be learned from Governors' addresses, but remain unable to capture non-conventional, but not less important topics.