Real-Time Forecast Density Combinations (Forecasting US GDP Growth Using Mixed-Frequency Data)

We combine the issues of dealing with variables sampled at mixed frequencies and the use ofreal-time data. In particular, the repeated observations forecasting (ROF) analysis of Stark andCroushore (2002) is extended to an autoregressive distributed lag setting in which the regressorsmay be sampled at higher frequencies than the regressand. For the US GDP quarterly growth rate, wecompare the forecasting performances of an AR model with several mixed-frequency models amongwhich the MIDAS approach. The additional dimension provided by different vintages allows us tocompute several forecasts for a given calendar date and use them to construct forecast densities.Scoring rules are employed to test for their equality and to construct combinations of them. Giventhe change of the implied weights over time, we propose time-varying ROF-based weights usingvintage data which present an alternative to traditional weighting schemes.