A habilidade preditiva e os impactos dos choques na política monetária e no câmbio: comparações entre modelos DSGE, VAR e BVAR estimados para o Brasil

Abstract

This article compares the out-of-sample forecasting ability of a new Keynesian DSGE model, specified and estimated for Brazil, with those of a Vector Autoregression (VAR) and a Bayesian Vector Autoregression (BVAR). The article innovates in relation to other similar studies made for Brazil (Castro et al. (2011) e Caetano e Moura (2013)), by choosing a specification for the DSGE model that, allowing the use of an richer information set, made possible to compute the predictive ability of the DSGE from forecasts that are, truly, out-of-sample forecasts. Moreover, unlike other articles that used Brazilian data, it also verifies to what degree the responses of variables to a monetary and an exchange rate shock match those of a structural BVAR identified through sign restrictions imposed on impulse response functions. The estimated DSGE model is similar to the ones adopted by Justiniano e Preston (2010) and Alpanda (2010). The BVAR model was estimated using a Bayesian procedure similar to those proposed by Sims and Zha (1998), Waggoner and Zha (2003) and Ramíres, Waggoner and Zha (2007). The results shows that the estimated DSGE model is capable of making out-of-sample forecasts, for some variables, that are competitive when compared to those of a VAR and a BVAR model. Furthermore, the response of variables to a monetary and an exchange rate shock, in the DSGE and BVAR models, are very similar.