This study reassesses the evidence and the practical relevance of long-horizon predictability of asset returns. We investigated whether predictability patterns can be profitably exploited and potentially replace the conventional approach used in the industry of extrapolating from historical samples. The analysis indicates that forward-looking models relying on steady-state equations for equities and initial yields to maturity for bonds are better predictors of the long-run direction of markets than naïve historical averages. Using a long-term U.S. sample (1926–2010) and relatively simple dynamic asset allocation strategies, we also found that predictability translates into significantly better risk-adjusted performance of strategies relying on forward-looking inputs.