GARCH modelling

[apanek]

Hello,

I am trying to model and forecast Brent and WTI crude oil volatility using very basic GARCH specifications. I compare the modelling and predictive power of the models with the actual volatility as measured by daily squared returns r2t.

Unfortunately, looking at the graphs, I can tell that the GARCH(1,1) does not do a good job at all. In fact the actual volatility is few times higher than the modelled volatility. Please find the graphs attached. This feeds to the forecasts made from such models.

I did not expext the model to exactly follow the actual volatility, but why the difference is so huge?

Is it because I have used the very basic models? Or is there something wrong with the specification? Or maybe squared returns are not a good proxy for volatility in this case?

[apanek]

Hi again,

I think I have an answer. The GARCH is not supposed to form a single-file line with the actual volatility, all it is expected to do is to track variations in the volatility.