Análise da taxa de juros e taxa de câmbio brasileira por meio de modelos de previsão

Abstract

The analysis of macro-economic variables through time-series models is widely used in the literature supporting economic theory, showing the actual behaviour of these variables. One of these macroeconomic variables have two variables that interfere with eou has relationships with other variables justifying the relevance in studying their behaviors. The first is the interest rate, which is very important in driving the economy, influencing the intention to spend and save of all economic agents, whether personal, commercial or industrial level (State or private). The second is the exchange rate, where its buoyancy determines the level of imports and exports affecting the trade balance. In this context the present research aims to describe the behavior of SELIC interest rates and Brazilian Exchange from January 1974 to June 2012 and January 1980 to may 2012, respectively. To this end, at first was used the Box-Jenkins model where the models showed through the analysis of residues which both had conditional heteroscedasticity in the waste of the models. Then joint modeling was used to the level of the process and the process variance (ARCH family models). The results showed that, for the SELIC interest rate series, the model selected was an ARIMA (1,1,1)-EGARCH (3,1) and, to the exchange rate, an ARIMA (0,1,1)-EGARCH (1,1). It is evidenced through these models that there is asymmetry of information, yet there was the leverage effect. In a second moment was chosen a model representing each one of the models of family ARCH (ARCH, GARCH, TARCH, EGARCH) and later held the combination of prediction by methods: ACP, middle and MMQO. The results obtained show that, in General, the performance measures MAPE, MSE and U-THEIL are superior to the combinations of prediction. In addition, the combination of forecast for different weights with ACP to check which of the types of weights provide better results. Therefore, it is concluded that the different weights allow the researcher to achieve greater accuracy in the choice of models combined, allowing aid managers in prior decision of the behavior of these variables that affect so scathing the health of the Brazilian economy.