Alternative Models for the Conditional Heteroscedasticity and the Predictive Accuracy of Variance Models – Empirical Evidence from East and North Africa Stock Markets

Abstract

Using empirical evidence from East and North Africa Stock Markets, this paper examines and compares alternative distribution density forecast methods of three generalised autoregressive conditional heteroscedasticity (GARCH) models. We employed the symmetric GARCH, Glosten Jagannathan and Runkle version of GARCH (GJR-GARCH) and Exponential GARCH methods to investigate the effect of stock return volatility using Gaussian, Student-t and Generalised Error distribution densities. The results show that the use of GJR and EGARCH with non-normal distribution densities appear justified to model the asymmetric characteristics of both indices. The evidence so far shows that in both markets, negative shocks would generally have a greater impact on future volatility than positive shocks, confirming the existence of leverage effect. The presence of leverage effect suggests that investors in these markets should be rewarded for taking up additional leverage risk as a fall in equity value (resulting from volatility) would mean a rise of debt to equity ratio and therefore, increase in financial distress risk. With respect to forecasting evaluation, the results indicate that clearly, symmetric GARCH model completely dominates the others in Kenya, while both GARCH and EGARCH best capture the Tunisian market.