Forecasting Volatility

[dacronje]

I have been reading quite a few papers on volatility forecasting using various GARCH and implied volatility methods. There is something that bothers.
It is s called volatility forecasting but then they forecast the unconditional variance. Then they test the accuracy of the so called volatility forecast by comparing the variance value with a "volatility proxy" (such as squared returns) that is in fact a variance value. Am I missing something here?

Help would be very much appreciated.

Thank you

 

[Keith Tan]

squared returns is not exactly variance. but their equations are related and they have strong autocorrelation.