Connecting the Dots: Assigning FOMC Members to Fed Dots Through Speech Quantification

Advisor: Carlos Viana de Carvalho

Co-advisor: Marcelo Medeiros

As (1) points out, monetary policy predictability can enhance a Central Bank stabilization policy efficacy. In this paper we aim to reduce uncertainty about one Federal Reserve forward guidance instrument by estimating full association probabilities distributions between members and the interest rate dot plot for each FOMC meeting. Our contribution to the literature is twofold: first, we propose a general Bayesian algorithm which estimates these association hypotheses between agents and actions whenever they are not observed. Second, we elaborate a novel and less subjective technique for quantifying text into data, using Latent Dirichlet Allocation (LDA) and shrinkage econometric tools. This method shows some desirable features such as positive correlation between the FOMC chair and the rest of the committee, and a policy stance ordering which partially reflects analysts and market participants views on this hawk-dove spectrum. Our tracking algorithm performs successfully in a simulated environment, in a sense that it on average considers the correct member-to-dot association as the most likely one. Using real data on speeches and Fed dots, it is also able to attribute the highest probability to the correct assignment hypothesis in the only meeting it is known for sure