Ensaios em econometria de séries temporais: núcleo e previsão da inflação no Brasil

Abstract

The objective of this thesis is to study the issues related to the core and the forecast of inflation in the Brazilian economy after the real plan by means of three independent tests that has as intersection the use of time series econometrics. In the first essay we compare the inflation forecasts obtained with common factors estimated from a large data set, with the predictions provided by the Central Bank of Brazil's Focus survey and by the integrated autoregressive moving average (ARIMA) and autoregressive vector (VAR ). The results show that the predictions of the Focus survey are the ones with the lowest mean square error between the models compared and that the gains of using large amounts of data to predict inflation are limited. The second essay investigates whether the core of inflation capturing the permanent price changes has any relation to the long-run inflation trajectory. For this, two core measurements are constructed with unobserved (UC) component models estimated by maximum likelihood and Kalman filter. The results show that the core measure constructed with a multivariate UC model is the only one among the measured measures that is unvented and a prior indicator of inflation, with the lowest error out-of-sample forecast. The third essay also proposes a new core measure of inflation following the notes of the second essay and advancing considering the possible non linearity and non-stationary inflation in Brazil. In order to estimate the new core, the empirical decomposition method (EMD) is used that does not use the assumptions of the series to be stationary or linear. The results indicate that the new core measure obtained has a better performance to predict out-of-sample inflation among the measures evaluated and has the statistical properties of being cointegrated, unbiased, attractive and strongly exogenous in relation to inflation, characteristics that are useful For monetary policy purposes.