The Use of Judgment Versus Models: Forecasting Accuracy in Sveriges Riksbank 2010-2014

Abstract: In this study, we ask, how did the use of judgments and models affect the accuracy of the Riksbank's forecasts during the years 2010-2014? We examine the two components, judgment and models, that determine the Sveriges Riksbank's published forecasts for three macro indicators: GDP growth, CPIF inflation and the repo rate. The analysis compares the relative forecasting accuracy of the central bank's primary models, DSGE and BVAR, and the final published forecasts containing the judgmental adjustments of the Riksbank's Executive Board. The study draws on both graphical and econometric analyses. Our findings show that in the nowcast (one-two quarters), the published forecasts outperformed the two model forecasts. The forecasting accuracy of the published forecasts, however, progressively worsened at four to eight quarters for CPIF inflation and the repo rate, whereas the BVAR model, compared to the DSGE model, systematically improved in accuracy. By incorporating judgment in the Riksbank's nowcasts, we find an improvement in accuracy - more so than what would have been the case if the models were solely used. Judgmental adjustments are thus important for accuracy in the short term.