In this paper we examine the real estate returns predictability employing US Real Estate Investment Trusts (REITs) and a set of possible predictors for the period January 1991 to December 2014. To this end we employ several forecasting models to test for REITs predictability under a flexible framework that captures parameter instability. Apart from the traditional factors examined in relevant studies, we also account for a series of sentiment and uncertainty indicators that may be significant predictors of REITs returns, especially during turbulent times when sentiment determines investment decisions to a greater extent. The empirical results indicate that the good predictors of REITs returns vary over time and over the forecast horizons. Our results suggest that economy-wide indicators, monetary policy instruments and sentiment indicators are among the most powerful predictors of REITs returns. In economic terms an investment strategy that is based on our forecasts outperforms a buy and hold strategy. The issue of the most suitable forecasting method is also discussed in detail. Our results might entail implications for investors and market authorities.