Modeling exchange rate using symmetric and asymmetric GARCH models

Abstract: This paper attempts to study GARCH-type models, with emphasis on fitting GARCH models to exchange rate return series. The symmetric GARCH(1,1) model is compared with the asymmetric EGARCH(1,1) model. Both models are analysed with different conditional distributions, namely Normal, Student's t and skew Student's t for the return innovation. Parameter estimation is performed using a maximum-likelihood approximation. The model performance is assessed by looking at the lowest AIC and BIC. Four exchange rate returns are studied using daily data over the period from 2002 to 2015. Moreover, essential ideas of return time series and stylised facts will be analysed. Our results indicate that the asymmetric GARCH model improves generally estimation with fat-tailed densities in the conditional variance. Furthermore, persistence has found to be reduced with the use of heavy-tailed distributions. Asymmetry presence has been detected in the EGARCH model. Besides, we found that "good news" tend to increase volatility in comparison with "bad news".