This paper considers Bayesian variable selection in regressions with a large number of possibly highly correlated macroeconomic predictors. I show that acknowledging the correlation structure in the predictors can improve forecasts over existing popular Bayesian variable selection algorithms.