Comparing Alternative Predictors Based on Large-Panel Factor Models