This paper empirically investigates the forecasting performances for the housing and stock returns of a series of SVAR models, including various combinations of the federal funds rate, term spread, external finance premium, TED spread, and GDP. Using US data 1975Q2-2008Q3, we find that, for both the in-sample-fitting and out-of-sample forecasting, the single-regime version always underperforms the regime-switching counterpart. The term spread and TED spread outperform other variables in predicting the stock returns. We also find preliminary evidence that the housing return may help predicting the stock return since 2006. None of the models we examine predict the 2008 downfall of housing returns.