A Comparison of Conditional Volatility Estimators for the ISE National 100 Index Returns

We compare more than 1000 different volatility models in terms of their fit to the historical ISE-100 Index data and their forecasting performance of the conditional variance in an out-of-sample setting. Exponential GARCH model of Nelson (1991) with “constant mean, t-distribution, one lag moving average term” specification achieves the best overall performance for modeling the ISE-100 return volatility. The t-distribution seems to characterize the distribution of the heavy tailed returns better than the Gaussian distribution or the generalized error distribution. In terms of forecasting performance, the best models are the ones that can accommodate a leverage effect. Results from fitting the selected exponential GARCH model to the historical ISE-100 return data indicates that the return volatility reacts to bad news 24% more than they react to good news as a result of a one standard deviation shock to the returns. As the magnitude of shock increases, the asymmetry becomes larger.