Density Forecasting with BVAR Models under Macroeconomic Data Uncertainty

Abstract

Macroeconomic data are subject to data revisions as later vintages are released. Yet, the usual way of generating real-time density forecasts from BVAR models makes no allowance for this form of data uncertainty. We evaluate two methods that consider data uncertainty when forecasting with BVAR models with/without stochastic volatility. First, the BVAR forecasting model is estimated on real-time vintages. Second, a model of data revisions is included, so that the BVAR is estimated on, and the forecasts conditioned on, estimates of the revised values. We show that both these methods improve the accuracy of density forecasts for US and UK output growth and inflation. We also investigate how the characteristics of the underlying data and revisions processes affect forecasting performance, and provide guidance that may benefit professional forecasters.