Fiscal and Monetary Dominance in Brazil : A Machine Learning DSGE Approach

Abstract

This dissertation investigates the interactions between fiscal and mone- tary policies in Brazil using a hybrid approach that combines Markov-switching DSGE models (MS-DSGE) and supervised machine learning techniques. The structural model serves as a data-generating process to train classification al- gorithms, which are subsequently applied to real Brazilian time series. Among the tested models, XGBoost demonstrated superior performance, achieving an accuracy of 95% and a Log Loss of 0.17 on the test set. The results reveal a dynamic chronology: Brazil operated under fiscal dominance between 2013 and 2015, transitioning to monetary dominance in the 2015-2017 period fol- lowing the implementation of the expenditure ceiling. An original contribution of this work is the identification of an explosive regime during the pandemic (2020Q3-2021Q3), characterized by the simultaneously active stance of both fiscal and monetary authorities. Robustness analyses indicate that the identifi- cation of fiscal and explosive regimes is resilient to different filtering techniques, whereas monetary dominance appears more sensitive to modeling assumptions. The findings suggest a return to fiscal dominance in the most recent period.