This paper investigates time series methods for forecasting four Euro-area wide aggregate variables: real GDP, industrial production, price inflation, and the unemployment rate. We consider two empirical questions arising from this problem. First, is it better to build aggregate Euro-area wide forecasting models for these variables, or are there gains from aggregating country-specific forecasts for the component country variables? Second, are there gains from using information from additional predictors beyond simple univariate time series forecasts, and if so, how large are these gains, and how are these gains best achieved? It turns out that typically there are gains from forecasting these series at the country level, then pooling the forecasts, relative to forecasting at the aggregate level. This suggests that structural macroeconometric modeling of the Euro area is appropriately done at the country-specific level, rather than directly at the aggregate level. Moreover, our simulated out-of-sample forecast experiment provides little evidence that forecasts from multivariate models are more accurate than forecasts from univariate models. If we restrict attention to multivariate models, the forecasts obtained from a dynamic factor model appear to be somewhat more accurate than the other methods.
Author(s): Massimiliano Marcellino(Istituto di Economia Politica, Universita Bocconi IGIER), James H. Stock (Kennedy School of Government, Harvard University and the NBER) and Mark W. Watson (Department of Economics and Woodrow Wilson School, Princeton University and the NBER)