Política monetária e regimes de inflação no Brasil: uma aplicação SET-VAR bayesiano

Abstract

This dissertation examines the hypothesis that the transmission of monetary policy shocks may differ between periods of lower and higher inflation. To investigate this possibility, a Bayesian Self-Exciting Threshold Vector Autoregression (SET-VAR) model is estimated, with thresholds determined endogenously using lagged inflation as the state variable. The dataset comprises monthly observations from February 2005 to December 2024, covering inflation, the Selic rate, the output gap, and changes in the nominal exchange rate. Monetary policy shocks are identified using an external proxy constructed from surprises in the Brazilian interbank deposit (DI) yield curve around the Monetary Policy Committee’s decisions. The impulse response functions are normalized to an initial 50-basis-point increase in the Selic rate. The results indicate clearer and more persistent responses in the regime where inflation exceeds the estimated threshold. In this regime, a contractionary shock produces a persistent increase in the Selic rate and a delayed decline in the output gap. Inflation initially rises, exhibiting the price puzzle, followed by negative responses at later horizons. The exchange rate response displays considerable uncertainty over most of the horizon examined. In the lower-inflation regime, credible intervals are wide, limiting the identification of persistent responses in inflation and economic activity. The results suggest that the inflation regime is relevant to the analysis of monetary policy transmission in Brazil, with distinct patterns and greater precision in the higher-inflation regime, although wide credible intervals that include zero limit the ability to draw more precise conclusions about monetary policy transmission across these regimes.