Modelling and Forecasting the Conditional Heteroscedasticity With Different Distribution Densities – Frontier Market Evidence

Abstract

This paper examines and compares alternative distribution density forecast methodology 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 methodology to investigate the effect of stock return volatility using Gaussian, Student-t and Generalised Error distribution densities. The evidence reveals that news impact is asymmetric leading to the existence of leverage effect in stock returns. The presence of leverage effect suggests that investors in these markets are to be rewarded for taking up additional leverage risks. This implies that by allocating portfolios, fund managers and /or investors should go beyond the mean-variance analysis and look into information about volatility, information asymmetry, correlation, skewness, and kurtosis. Furthermore, the evidence exhibits reverse volatility asymmetry, contradicting the widely accepted theory of volatility asymmetry. Regarding forecasting evaluation, the results reveal that the symmetric GARCH model coupled with fatter-tail distributions presents a better out-of-sample forecast.