Descobrindo modelos de previsão para a inflação brasileira: Uma análise a partir de uma gama ampla de indicadores ♦

Abstract

This work evaluates the prediction capabilities of econometric time series models based on macroeconomics indicators for Brazilian consumer price index (IPCA). We run a pseudo real time prediction exercise with twelve steps ahead horizon. Predictions are compared with univariate models such as the first order autoregressive model among others. The sample period goes from January 2000 to August 2015. We evaluated over 1170 different economic variable for each forecast period, searching for the best predictor set in each point in time using Autometrics algorithm as model selector. Models’ performance is compared using Model Confidence Set, developed by Hansen, Lunde and Nason (2010). Our results suggest possible gains in predictions that use a high number of indicators particularly at longer horizons.