Estimating Policy Uncertainty Within Monetary Policy Debates

Abstract

Studying policy uncertainty contained in collections of documents has been a major task for political researchers and economists, who aim at measuring this degree exclusively with wordlists and topic models to feed further econometric inferences or test hypotheses. Such bag-of-word applications constrain the analysis and cannot render a clear picture of uncertainty drivers and their persistence, even if semi-supervised strategies may offer coherent improvements at the topic level. This work proposes a semantic search strategy, using Top2vec, to identify sources of uncertainty, at the debate level, and uncover coherent topics whose representations will be used to get uncertainty prevalence within each debate. Unlike aggregate-level measurements, this strategy is suited to study per speaker contributions at central banks, where uncertainty is regarded as a forward guidance tool and a key strategy when devising monetary policy actions. Applied to FOMC transcripts (1994-2016), the resulting semantic space yields non-overlapping topic vectors indicating a dominance of economic discussions in uncertainty formation within committee meetings, while risks concerns are bounded to financial markets and investments using an investor jargon. Moreover, results demonstrate the importance of experts' contributions in steering the economic debate, hence coloring uncertainty with words not found in traditional uncertainty wordlists and diffusing a significant persistence to uncertainty prevalence during debates that exhibits fractal patterns.