A common problem in out-of-sample prediction is that there are potentially many relevant predictors that individually have only weak explanatory power. We propose bootstrap aggregation of pre-test predictors (or bagging for short) as a means of constructing forecasts from multiple regression...