Data Revisions and Real-time Probabilistic Forecasting of Macroeconomic Variables

Macroeconomic data are subject to revision over time as later vintages are released, yet the usual way of generating real-time out-of-sample forecasts from models effectively makes no allowance for this form of data uncertainty. We analyze a simple method which has been used in the context of point forecasting, and does make an allowance for data uncertainty. This method is applied to density forecasting in the presence of time-varying heteroscedasticity, and is shown in principle to improve real-time density forecasts. We show that the magnitude of the expected improvements depends on the nature of the data revisions.