为什么都没有人回答呢？

(sorry I dont have CN input in the lab)
You should be clear about the concept you r studying.
Volatility itself is unobservable from the data with the same observation frequency. You cannot compare the GARCH prediction with the "REAL" volatility. That is why it is hard to evaluate the GARCH performance.
Many studies do suggest proxy for the true underlying volatility and method to evaluate the GARCH performance. Take a look at  Awartani and Corradi’s (2005) and the attached file.

Also you may want to compare it with the option implied or realized volatility from higher freq data. Prob you will find option implied vol is leading.