Forecasting using a large number of predictors: is Bayesian regression a valid alternative to principal components? Discussion paper 32/2006: Christine De Mol, Domenico Giannone, Lucrezia Reichlin