Can Model-Based Forecasts Predict Stock Market Volatility Using Range-Based and Implied Volatility as Proxies?

This thesis attempts to evaluate the performance of parametric time series models and RiskMetrics methodology to predict volatility. Range-based price estimators and Model-free implied volatility are used as a proxy for actual ex-post volatility, with data collected from ten prominent global volatility indices. To better understand how volatility behaves, different models from the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) class were selected with Normal, Student-t and Generalized Error distribution (GED) innovations. A fixed rolling window methodology was used to estimate the models and predict the movements of volatility and, subsequently, their forecasting performances were evaluated using loss functions and regression analysis. The findings are not clear-cut; there does not seem to be a single best performing GARCH model. Depending on the indices chosen, for range-based estimator, APARCH (1,1) model with normal distribution overall outperforms the other models with the noticeable exception of HSI and KOSPI, where RiskMetrics seems to take the lead. When it comes to implied volatility prediction, GARCH (1,1) with Student-t performs relative well with the exception of UKX and SMI indices where GARCH (1,1) with Normal innovations and GED seem to do well respectively. Moreover, we also find evidence that all volatility forecasts are somewhat biased but they bear information about the future volatility.