Classifying Brazilian Monetary Policy States with Interest Rate Gaps and Markov Models

This paper develops a monthly framework for classifying Brazilian monetary policy that distinguishes the level of monetary restraint, changes in that stance, and statistical regimes associated with Selic adjustments. The framework combines a seven-state deterministic classifier, an empirical Markov chain, a Gaussian hidden Markov model, and Markov-switching regressions. Project outputs provide 145 classified observations from May 2014 to May 2026, with a partial final month. Under the implemented nominal neutral rate convention, restrictive states account for 88.3% of observations. A four-state HMM achieves a lower BIC, whereas the three-state model records a lower temporal validation loss. Switching regressions select three Selic adjustment regimes after solutions with degenerate variances are excluded. An additional sensitivity exercise reduces the restrictive share to 51.0% when the neutral benchmark is interpreted as real, demonstrating how strongly the results depend on benchmark measurement. The contribution is an auditable comparative application for monetary monitoring. The findings describe historical patterns conditional on the data and specifications; they do not establish predictive superiority or causal identification.