Why does the GARCH(1,1) model fail to provide sensible longer- horizon volatility forecasts?