MODEL SELECTION APPROACH TO REAL- TIME MACROECONOMIC FORECASTING

performance. (2) Model selection does not require the specification of a correct model for its valid application, as does the traditional hypothesis testing approach. (3) Finally, if properly designed, the probability of selecting the truly best model approaches I as the sample size increases, in contrast to the traditional hypothesis testing approach (see Swanson and White (1995». However, we note that it can sometimes be difficult to assess the type I error associated with testing the implicit model selection hypothesis that two models under consideration truly perform equally well based on observed differences in realized model selection criteria. Nevertheless, this is a defect of the same order of magnitude as using a traditional test whose size is known only asymptotically. Further, if there exists a true model among the alternatives, then if the size of a traditiona