Aggregate density forecasting from disaggregate components using Bayesian VARs

There is a considerable volume of literature concerned with point forecasting which aims to assess whether producing aggregate forecasts as the sum of the components' forecasts is better than alternative direct methods, whereas aggregate density forecasting from disaggregate components is still a relatively unexplored field. This paper develops an implementation of the bottom-up approach that is capable of producing well-performing and competitive density forecasts. This is achieved by accounting explicitly for the interaction between components, using Bayesian VARs to estimate the whole multivariate process and produce the aggregate forecasts. An empirical application using CPI and GDP data shows that the method can be used to produce aggregate density forecasts capable of accounting for the events resulting from the crisis. This suggests that it might be particularly useful for forecasting in turbulent times and therefore prove a valuable addition to the forecaster's toolkit.