Abstract

In this paper we describe Norges Bank's system for averaging models (SAM) which produces model-based density forecasts for Norwegian Mainland GDP and inflation. We combine the forecasts from three main types of models typically used at central banks: Vector autoregressive models, leading indicator models and factor models. By combining models we hedge against uncertain instabilities. We update SAM several times during the quarter to highlight the importance of new data releases, and we show how the performance of SAM improves steadily as new information arrives. The framework is robust with regard to alternative vintages of data to evaluate against. We show that our chosen weighting scheme is superior or on a par with some common alternative weighting schemes, and, finally, that a strategy of trying to pick the best model, ex ante, is inferior to model combination.