Abstract:
This paper aims to estimate output gap for Brazilian economy through di.erents methodolies. We use traditional univariate techniques and propose a new semi-estructural methodology that combines HP filtering and the production function approach. In order to compare the di.erent potential output estimates, we use a Phillips curve to predict free price inflation and a rolling forecast experiment as a test of forecast accuracy. Our results shows that the forecasts produced by the Local level and Watson models are even more inaccurate than those generated by the simplest univariate models. The main evidence is that the Beveridge-Nelson methodology outperforms all the models at all forecast horizons.