An investigation of the relative performance of GARCH models versus simple rules in forecasting volatility