This paper considers an institutional investor who is implementing a long-term portfolio allocation strategy using forecasts of financial returns. We compare the performance of two competing macro-finance models, an unrestricted Vector AutoRegression (VAR) and a fully structural Dynamic Stochastic General Equilibrium (DSGE) model, at forecasting financial returns. We show that the DSGE model outperforms the unrestricted VAR at forecasting financial returns in the long term and generates substantially higher Sharpe ratios for mean-variance allocations. Even if it contains fewer unknown parameters, the DSGE model benefits from economically grounded restrictions that help anchor financial returns in the long term.