Abstract

In 2006 the Norges Bank initiated a project to improve its short term forecasts. The current phase of the project is tasked with developing a system that provides model-based forecasts for gross domestic product and consumer price inflation excluding taxes and energy prices, for each policy round. Forecasts are recursively evaluated from 1999Q2 to the current quarter. The performance of the models over this period is then used to derive weights that are used to combine the forecasts. Our results indicate that model combination improves upon the forecasts from individual models. The ultimate goal of this phase is to provide density forecasts for GDP and CPI, to provide a formal characterisation of the uncertainty that surrounds the central forecasts. The system is flexible enough to incorporate new models as they are developed, refreshing the system of models.