This paper investigates when textual information from central bank communication improves forecasts of policy rate changes. Using minutes from the Brazilian Central Bank’s Monetary Policy Committee (Copom), we test whether the text helps predict changes in the Selic target rate between consecutive meetings. Minutes are encoded with sentence embeddings, and low-dimensional textual factors are extracted through principal component analysis estimated only on the training sample. Forecast performance is evaluated out of sample with an expanding-window backtest against persistence, random-walk, autoregressive, regularized, and state-space benchmarks. Text-only models perform poorly, but textual predictors improve forecasts when combined with short-run dynamics and regularization. The gains are economically meaningful and concentrated in periods of policy adjustment, while persistence remains hard to beat during rate-hold episodes. Overall, central bank communication contains useful forward-looking information, but its predictive signal is sparse and episodic.
Comissão Organizadora
Comissão Científica
Ferramenta completa de submissão, avaliação e publicação de anais para eventos científicos.
