Forecasting accuracy of stochastic volatility, GARCH and EWMA models under different volatility scenarios

The forecasting of the volatility of asset returns is a prerequisite for many risk management tasks in finance. The objective here is to identify the volatility scenarios that favour either Generalized Autoregressive Conditional Heteroscedasticity (GARCH) or Stochastic Volatility (SV) models. Scenarios are defined by the persistence of volatility (its robustness to shocks) and the volatility of volatility. A simulation experiment generates return series using both volatility models for a range of volatility scenarios representative of that observed in real assets. Forecasts are generated from SV, GARCH and Exponentially Weighted Moving Average (EWMA) volatility models. SV model forecasts are only noticeably more accurate than GARCH in scenarios with very high volatility of volatility and a stochastic volatility generating process. For scenarios with medium volatility of volatility, there is little penalty for using EWMA regardless of the volatility generating process. A set of return time series selected from FX rates, equity indices, equities and commodities is used to validate the simulation-based results. Broadly speaking, the real series come from the medium volatility of volatility scenarios where EWMA forecasts are reliably accurate. The robust structure of EWMA appears to contribute to its greater forecasting accuracy than more flexible GARCH model.