Recent empirical and theoretical studies have shown the necessity of including real estate in diversified portfolios. In this paper we present one of the latest developed Autoregressive. Conditoinal Heteroscedasticity models (QTARCH) and, using fifty-two years of Swiss data, we investigate the ability of this class of models to forecast conditional means and variances of portfolios containing only stocks ans bonds and portfolios containig also real estate. The analysis is done in a Markowitz efficient frontier framework. We find that if a simple QTARCH model had been used with our data, the portfolio's risk would have been significantly. It is also shown that even when the QTARCH methodology is applied, including real estate in diversified portfolios is still justified. Furthermore, three years of out-sample performances of the model are extremely convincing as well