This paper examines the combination forecast and multivariate regression approaches for equity premium predictability. We evaluate 27 specifications with a unique Canadian database to avoid the data mining inherent in using common U.S. data. We find significant predictive evidence for most models. In sample, multivariate regression predictions perform better than combination forecasts, although regression results display evidence of instability and overfitting. Out of sample, combination forecasts are superior when relying on many individual models, but imposing economic restrictions on multivariate regression predictions yields similar performance. Both approaches show that incorporating information from numerous variables improves forecasting precision and economic value.