Cointegration and predictability of VECM approaches for Ibovespa

Abstract

The present research compares multivariate models applied to the IBovespa time series analysis. Macroeconomic variables, commodities and market indices are regressors suggested by the literature. The chosen approach uses a vector error correction model (VECM) alongside unit root and cointegration tests, robust under heteroskedasticity. The impact of national and international economic instability was controlled. To accomplish this, recessive cycles, in Brazil or in the United States, and the Brazilian electoral period were taken into account. In general, the evaluated models failed to meet the estimation’s assumptions, have low explanatory power and do not present significant relationship between IBOVESPA and dependent variables. However, evidence indicates that long-term relationships could exist, although this may not imply accuracy
in short-term predictions.