I want to ask why ARMA-GARCH is more and more popolar, and what's the advantage of this model.
You would use GARCH to account for stochastic volatility in a time series of returns. However, the returns time series may have components other than that can be explained by stochastic vol, such as trends or moving average. Therefore, ARMA or AR or some such series is used to de-trend. In risk managment it can be used for back testing, calculating VaR or ES. A good reference for this is Filtered Historical Simulation by Barone-Adesi(2000). Link for FHS. Also Kevin Dowd has good chapter explaining this - "Measuring Market Risk 2nd Ed, Chapter 4" -link Dowd.