I have to generate an out-of-sample forecast for an assignment at university. The question asks that I select an in-sample period, from 2008-2017, for the returns on my portfolio. I am curious as to what is the most appropriate time period to generate forecasts beyond here. I am unsure of whether the intuitions regarding my approach is to avoid the high volatility in 2008-2009, or even during 2011's sovereign debt crisis. Should I just model the entire period maybe? I am unsure. Any help would be appreciated.
Also, I have been asked to model the best-fit mean and variance time series. I have estimated the best ARMA model. Do I proceed by now estimating the best ARCH model or should I apply a GARCH model directly, assuming the stationarity implications run earlier indicate the mean is 0?
All help is appreciated.