Forecasting volatility: Evidence from the Macedonian stock exchange

This paper investigates the behavior of stock returns in an emerging stock market namely, the Macedonian Stock Exchange, focusing on the relationship between returns and conditional volatility. The conditional mean follows a GARCH-M model, while for the conditional variance one symmetric (GARCH) and four asymmetric GARCH types of models (EGARCH, GJR, TARCH and PGARCH) were tested. We examine how accurately these GARCH models forecast volatility under various error distributions. Three distributions were assumed, i.e. Gaussian, Student-t and Generalized Error Distribution. The empirical results show the following: (i) the Macedonian stock returns time series display stylized facts such as volatility clustering, high kurtosis, and low starting and slow-decaying autocorrelation function of squared returns; (ii) the asymmetric models show a little evidence on the existence of leverage effect; (iii) the estimated mean equation provide only a weak evidence on the existence of risk premium; (iv) the results are quite robust across different error distributions; and (v) GARCH models with non-Gaussian error distributions are superior to their counterparts estimated under normality in terms of their in-sample and out-of-sample forecasting accuracy.