The Norwegian central bank currently uses a multi-model strategy to forecast GDP growth. Small individual models are weighted together in SAM (System for Averaging Models) to create a more accurate forecast. The SAM framework also contains 57 bivariate models that uses GDP growth, and one of the indicators that we will use to create models as the endogenous variables. This project adopts a multi-model strategy to cope with uncertainty prevailing about the best strategy for modeling and forecasting economic output, explores the properties of larger VAR models (VARs with three and four variables) and analyzes if they possess the necessary properties to be included inthe SAM framework. In other words - if larger-scale VARs can improve on the forecasts made by the bivariate models. This thesis addresses the following questions: Is it limiting to focus on bivariate vector autoregressive models when forecasting GDP growth? Can there be any gain in exploring the three and four variable VAR framework to forecast GDP growth? What combinations of survey indicators perform well in forecasting GDP growth at short horizons? How can one efficiently weight individual forecasting models to deal with model uncertainty?