# Improving ARIMA forecasts

I am trying to use ARIMA to forecast stock returns. The problem is that variance is time-varying but ARIMA assumes that it is constant. As I understand, GARCH is only used for forecasting volatility (conditional variance), but not the conditional mean in which I am interested in. Is there any way to combine GARCH and ARIMA models?

Is it possible (maybe with some model specifications) to account for volatility clustering and to improve return forecasts? Thanks a lot!